diff --git a/.gitattributes b/.gitattributes index 486f63e4cd5325e72dc2c0eb1e4ab31e7a1ee717..9ad82bd4a54c79999407db5ec4592e1653f891de 100644 --- a/.gitattributes +++ b/.gitattributes @@ -15407,3 +15407,78 @@ parse/test/JVeM7uwDwK/JVeM7uwDwK_origin.pdf filter=lfs diff=lfs merge=lfs -text parse/test/JVeM7uwDwK/JVeM7uwDwK_span.pdf filter=lfs diff=lfs merge=lfs -text parse/test/JVeM7uwDwK/JVeM7uwDwK_layout.pdf filter=lfs diff=lfs merge=lfs -text parse/test/nKvGCUoiuW/nKvGCUoiuW_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/XSEBx0iSjFQ/XSEBx0iSjFQ_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/VYYf6S67pQc/VYYf6S67pQc_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/3FvF1db-bKT/3FvF1db-bKT_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/T8wHz4rnuGL/T8wHz4rnuGL_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/Loek7hfb46P/Loek7hfb46P_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/UKd6dpVGdu/UKd6dpVGdu_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/3RBY8fKjHeu/3RBY8fKjHeu_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/NHeAUKlTO8/NHeAUKlTO8_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/UKd6dpVGdu/UKd6dpVGdu_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/83LJRUzXWj/83LJRUzXWj_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/dev/NHeAUKlTO8/NHeAUKlTO8_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1WRibb0Z/S1WRibb0Z_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1WRibb0Z/S1WRibb0Z_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1WRibb0Z/S1WRibb0Z_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/lM2971LAwV/lM2971LAwV_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/lM2971LAwV/lM2971LAwV_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/lM2971LAwV/lM2971LAwV_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJxt0JHKvS/rJxt0JHKvS_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJxt0JHKvS/rJxt0JHKvS_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJxt0JHKvS/rJxt0JHKvS_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJWHIKqgl/HJWHIKqgl_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJWHIKqgl/HJWHIKqgl_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJWHIKqgl/HJWHIKqgl_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJqFGTslg/rJqFGTslg_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJqFGTslg/rJqFGTslg_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJqFGTslg/rJqFGTslg_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1vuO-bCW/S1vuO-bCW_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1vuO-bCW/S1vuO-bCW_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1vuO-bCW/S1vuO-bCW_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1c2cvqee/S1c2cvqee_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1c2cvqee/S1c2cvqee_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1c2cvqee/S1c2cvqee_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJe4_xSFDB/rJe4_xSFDB_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJe4_xSFDB/rJe4_xSFDB_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/rJe4_xSFDB/rJe4_xSFDB_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Skn9Shcxe/Skn9Shcxe_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Skn9Shcxe/Skn9Shcxe_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Skn9Shcxe/Skn9Shcxe_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJgcwh4FwS/BJgcwh4FwS_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJgcwh4FwS/BJgcwh4FwS_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/BJgcwh4FwS/BJgcwh4FwS_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Sk-oDY9ge/Sk-oDY9ge_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Sk-oDY9ge/Sk-oDY9ge_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/Sk-oDY9ge/Sk-oDY9ge_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/hm0i-cunzGW/hm0i-cunzGW_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/hm0i-cunzGW/hm0i-cunzGW_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/hm0i-cunzGW/hm0i-cunzGW_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1lZgyBYwS/r1lZgyBYwS_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1lZgyBYwS/r1lZgyBYwS_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/r1lZgyBYwS/r1lZgyBYwS_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJem3yHKwH/HJem3yHKwH_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJem3yHKwH/HJem3yHKwH_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/HJem3yHKwH/HJem3yHKwH_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/TR-Nj6nFx42/TR-Nj6nFx42_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/TR-Nj6nFx42/TR-Nj6nFx42_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/TR-Nj6nFx42/TR-Nj6nFx42_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/FsLTUzZlsgT/FsLTUzZlsgT_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/FsLTUzZlsgT/FsLTUzZlsgT_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/FsLTUzZlsgT/FsLTUzZlsgT_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/0lz69oI5iZP/0lz69oI5iZP_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/0lz69oI5iZP/0lz69oI5iZP_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/0lz69oI5iZP/0lz69oI5iZP_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/B1IzH7cxl/B1IzH7cxl_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/B1IzH7cxl/B1IzH7cxl_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/B1IzH7cxl/B1IzH7cxl_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/q8qLAbQBupm/q8qLAbQBupm_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/q8qLAbQBupm/q8qLAbQBupm_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/q8qLAbQBupm/q8qLAbQBupm_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkVhlh09tX/SkVhlh09tX_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkVhlh09tX/SkVhlh09tX_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/SkVhlh09tX/SkVhlh09tX_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1lvm305YQ/S1lvm305YQ_span.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1lvm305YQ/S1lvm305YQ_origin.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/S1lvm305YQ/S1lvm305YQ_layout.pdf filter=lfs diff=lfs merge=lfs -text +parse/train/YUEFlzlG_0c/YUEFlzlG_0c_origin.pdf filter=lfs diff=lfs merge=lfs -text diff --git a/parse/dev/09hVcSDkea/images/06a181c2f58d7ee657036e02167e76fee2211bdd65d0593da1ded42af9144f35.jpg b/parse/dev/09hVcSDkea/images/06a181c2f58d7ee657036e02167e76fee2211bdd65d0593da1ded42af9144f35.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a98cdbad8fb9ee30ce16b256f1bea1ba328ce0c1 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/06a181c2f58d7ee657036e02167e76fee2211bdd65d0593da1ded42af9144f35.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49f1168fc043d000a8730e0903fd8e65f5dbd42f96052f9754fd8187eca32b20 +size 109519 diff --git a/parse/dev/09hVcSDkea/images/0c8a620d9099ed9ff0ad3e2aaed16fcd2e9d97c08f253ca8ff92b89ace11e791.jpg b/parse/dev/09hVcSDkea/images/0c8a620d9099ed9ff0ad3e2aaed16fcd2e9d97c08f253ca8ff92b89ace11e791.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0df86ba65fb1a46eb6c0db2494ee802ff7d46bd5 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/0c8a620d9099ed9ff0ad3e2aaed16fcd2e9d97c08f253ca8ff92b89ace11e791.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:005d8d1b83648182fa4e7ea6406a40b4d7c077ac215ed2cd944f68413166aa28 +size 104527 diff --git a/parse/dev/09hVcSDkea/images/11722fd593ca78fd5d9b310b95d8c94bb026bc8ce78add68a2106de48bc05343.jpg b/parse/dev/09hVcSDkea/images/11722fd593ca78fd5d9b310b95d8c94bb026bc8ce78add68a2106de48bc05343.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3f7f48521d8b868f52994b9dc73ac681fde0d42b --- /dev/null +++ b/parse/dev/09hVcSDkea/images/11722fd593ca78fd5d9b310b95d8c94bb026bc8ce78add68a2106de48bc05343.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94b12be4634b70b15e669f58c4994f6845ec7591c89cf1f8ff572f94c60dd9e4 +size 29766 diff --git a/parse/dev/09hVcSDkea/images/18e6b78eddd4fd2046d76c403975aa716a6ba499d341db868418ae20e32e4835.jpg b/parse/dev/09hVcSDkea/images/18e6b78eddd4fd2046d76c403975aa716a6ba499d341db868418ae20e32e4835.jpg new file mode 100644 index 0000000000000000000000000000000000000000..920c99aee277d0c3df704b4f9ad58cb1c9d85360 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/18e6b78eddd4fd2046d76c403975aa716a6ba499d341db868418ae20e32e4835.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a06810a92cc0a05e85fe119898c7ecd5dd9e05bfaa1f3119d79e782308c8041e +size 9503 diff --git a/parse/dev/09hVcSDkea/images/1e98f4f2a2f34a6b30ac5a50bbf903b24ee53ebaff08915364c974e0ceeeea0e.jpg b/parse/dev/09hVcSDkea/images/1e98f4f2a2f34a6b30ac5a50bbf903b24ee53ebaff08915364c974e0ceeeea0e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5b5b769da7dcc3e1702041f525aa555002914597 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/1e98f4f2a2f34a6b30ac5a50bbf903b24ee53ebaff08915364c974e0ceeeea0e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9fee44e2a95443f71817ab5d32c4f566778f97abee2fb965b174c8fe12523d7c +size 22143 diff --git a/parse/dev/09hVcSDkea/images/257b2d046b6a80a0c4242edb91b80120eb4cc02c4fc1ff7be23b230fa7763efd.jpg b/parse/dev/09hVcSDkea/images/257b2d046b6a80a0c4242edb91b80120eb4cc02c4fc1ff7be23b230fa7763efd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..963aa513c641bd7d89fd75208a871734453c70f0 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/257b2d046b6a80a0c4242edb91b80120eb4cc02c4fc1ff7be23b230fa7763efd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e8af5f01c47704af6c929d5699f70a7e75a450d2f5e6e9f020f1ac01812d7b8 +size 24712 diff --git a/parse/dev/09hVcSDkea/images/2edaafd385f7816c24ee0c69f4e6edba89da99b65dc5010a1a3222e44ff964bb.jpg b/parse/dev/09hVcSDkea/images/2edaafd385f7816c24ee0c69f4e6edba89da99b65dc5010a1a3222e44ff964bb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8d30fc7280aa87b46989520a33670785fcd2e950 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/2edaafd385f7816c24ee0c69f4e6edba89da99b65dc5010a1a3222e44ff964bb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8dfa2eb91cc5da5430ddaba3a1a8c77e624ff5eab8e07bddfc6a3764a34e44cf +size 15773 diff --git a/parse/dev/09hVcSDkea/images/2f2175fb1329d3136feceda47e76b8d02304ba1093ecd6911c001590c997ed9b.jpg b/parse/dev/09hVcSDkea/images/2f2175fb1329d3136feceda47e76b8d02304ba1093ecd6911c001590c997ed9b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1267c6132b70c0b7b64ba40761ad2f2442a5e03f --- /dev/null +++ b/parse/dev/09hVcSDkea/images/2f2175fb1329d3136feceda47e76b8d02304ba1093ecd6911c001590c997ed9b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f02170075e47f5f772f75a45ce079a939a200bcdbc6c79192d9ecc4bc8b47254 +size 66322 diff --git a/parse/dev/09hVcSDkea/images/2f3f4d7c5fe049564a76505d36cdc91369e5917e24856f2a77ca68e436f7de97.jpg b/parse/dev/09hVcSDkea/images/2f3f4d7c5fe049564a76505d36cdc91369e5917e24856f2a77ca68e436f7de97.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c466a1fb8c09464d324ebe66a720ea1aa64d64c9 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/2f3f4d7c5fe049564a76505d36cdc91369e5917e24856f2a77ca68e436f7de97.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e21b3da1f11ac130f33820e815a1f6ad4728df449035d3ee24b5e72a562ebe0a +size 63101 diff --git a/parse/dev/09hVcSDkea/images/37fd9c891e3f63ff8b76f7ef5135f72ca6de00ce49141b1a69d3e6a1327cb30b.jpg b/parse/dev/09hVcSDkea/images/37fd9c891e3f63ff8b76f7ef5135f72ca6de00ce49141b1a69d3e6a1327cb30b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..045c85dfb780d929018524d7421613baf7654233 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/37fd9c891e3f63ff8b76f7ef5135f72ca6de00ce49141b1a69d3e6a1327cb30b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e975b71969ef2f136a57543d56cd9a71d6524f5f66074286e0c5715c25b07e4 +size 14971 diff --git a/parse/dev/09hVcSDkea/images/42e8fc53e8f4be30937cdb6ace8fd64d90e6685bfa047e152d28d0e0a10d8902.jpg b/parse/dev/09hVcSDkea/images/42e8fc53e8f4be30937cdb6ace8fd64d90e6685bfa047e152d28d0e0a10d8902.jpg new file mode 100644 index 0000000000000000000000000000000000000000..39da0fdc043a8ceff65b1d1aa31d86a375e6ab87 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/42e8fc53e8f4be30937cdb6ace8fd64d90e6685bfa047e152d28d0e0a10d8902.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:863df4b4cdc4c1cd8ccecf783f3c982e69f1faf5e7586954953db8cee6f9e031 +size 8287 diff --git a/parse/dev/09hVcSDkea/images/459f801e7ff81d17f3cfa23c0fd8ec3c9133f10c7e6196636b03f74ece08238a.jpg b/parse/dev/09hVcSDkea/images/459f801e7ff81d17f3cfa23c0fd8ec3c9133f10c7e6196636b03f74ece08238a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..df46d363629a5c5a2dfcf0a4dce7c2e3ba92e612 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/459f801e7ff81d17f3cfa23c0fd8ec3c9133f10c7e6196636b03f74ece08238a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0aa58d6345d62ee23c61f7f74fa995d9ac340adc1e5596e6ac3056cb84734043 +size 66646 diff --git a/parse/dev/09hVcSDkea/images/4807200fe7e925dbdc60b90edd5d9d60a1d6ba86cf5df067ecbafc9f4a16b409.jpg b/parse/dev/09hVcSDkea/images/4807200fe7e925dbdc60b90edd5d9d60a1d6ba86cf5df067ecbafc9f4a16b409.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6ab3b071d175ae88d672cc45258da2aca2a0b443 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/4807200fe7e925dbdc60b90edd5d9d60a1d6ba86cf5df067ecbafc9f4a16b409.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d1548b141785ba39bb2e197f9c2818dfff3c9dcd1d41c6aef803673cae17956 +size 110514 diff --git a/parse/dev/09hVcSDkea/images/4dd8395ea4882213cb68da1f6b25faa4ebd6cb6bd3705e0df1e770aeb9b9d374.jpg b/parse/dev/09hVcSDkea/images/4dd8395ea4882213cb68da1f6b25faa4ebd6cb6bd3705e0df1e770aeb9b9d374.jpg new file mode 100644 index 0000000000000000000000000000000000000000..de3ff4f9090478ab2d11e3572d3154c7a1017024 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/4dd8395ea4882213cb68da1f6b25faa4ebd6cb6bd3705e0df1e770aeb9b9d374.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c6ed7d48452c8a3fed7ab18d86f1ea6fcf9dd0c6f7ab066f182e5d3fd367c0f8 +size 65048 diff --git a/parse/dev/09hVcSDkea/images/5050220e448020d78790b1e83d4696485816144f9fb9557d3b542de7e5a3860c.jpg b/parse/dev/09hVcSDkea/images/5050220e448020d78790b1e83d4696485816144f9fb9557d3b542de7e5a3860c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2cec7c1383ffe1c2b15737a60736164ce5d21750 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/5050220e448020d78790b1e83d4696485816144f9fb9557d3b542de7e5a3860c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:57b5468b8394d1315711b0216e7cbcd0795214cf17082163fa13e15afa23463c +size 114481 diff --git a/parse/dev/09hVcSDkea/images/78b5233aa69e2b8710b84af8a93310904973cca210d8a5ee594e3a4fbe8b5102.jpg b/parse/dev/09hVcSDkea/images/78b5233aa69e2b8710b84af8a93310904973cca210d8a5ee594e3a4fbe8b5102.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5eae6fca7b462b5c37495832a5ad6dacd48f3278 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/78b5233aa69e2b8710b84af8a93310904973cca210d8a5ee594e3a4fbe8b5102.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:157a4005d87988ff90e59b8df4250b4e0673a6b66fde28beaaa3d77454cd2979 +size 15925 diff --git a/parse/dev/09hVcSDkea/images/7960520ef2506bb75f22ffff775afd2c62315d6c6cf140e221c32d08d23acb6c.jpg b/parse/dev/09hVcSDkea/images/7960520ef2506bb75f22ffff775afd2c62315d6c6cf140e221c32d08d23acb6c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fef416cca32f1e8838c33b30e37b90b61b3055ec --- /dev/null +++ b/parse/dev/09hVcSDkea/images/7960520ef2506bb75f22ffff775afd2c62315d6c6cf140e221c32d08d23acb6c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05c8545fc73d76284e309bdf833389a8d3683ecd78c13573e619369b8a8d9853 +size 108033 diff --git a/parse/dev/09hVcSDkea/images/83efd4ce5e9f67758c2f3c0126b6bd809a111992d57dd19dce7c41206828fde1.jpg b/parse/dev/09hVcSDkea/images/83efd4ce5e9f67758c2f3c0126b6bd809a111992d57dd19dce7c41206828fde1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9d9afb6b1193d495b7ca11f9e97b77c7ec3f3926 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/83efd4ce5e9f67758c2f3c0126b6bd809a111992d57dd19dce7c41206828fde1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e81a336612c2af940d6f1f4cdada5f358a948daff34337c4afcdb22614db877 +size 112050 diff --git a/parse/dev/09hVcSDkea/images/9f19c80d9892bf0eafbfada72334ba90039256ec6b69592291e8a2703246b3f1.jpg b/parse/dev/09hVcSDkea/images/9f19c80d9892bf0eafbfada72334ba90039256ec6b69592291e8a2703246b3f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..80c12831e4dc77c88b9a8fc1fb1dc2c0fa048c38 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/9f19c80d9892bf0eafbfada72334ba90039256ec6b69592291e8a2703246b3f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26e9c073e23e0f3f4deeca9d4021e3bc1d639f8e34af994f09fdc61ab49a2f87 +size 23741 diff --git a/parse/dev/09hVcSDkea/images/bff27fbe41eea6e635bbeb3ed3ffc0bcf18122b22ff97b3a0c695ab06f6813fe.jpg b/parse/dev/09hVcSDkea/images/bff27fbe41eea6e635bbeb3ed3ffc0bcf18122b22ff97b3a0c695ab06f6813fe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..909b8d9456a8678f7e2671b3ec9ba9f2cff75d52 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/bff27fbe41eea6e635bbeb3ed3ffc0bcf18122b22ff97b3a0c695ab06f6813fe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f546490c95d335e36bd6c23f2b56899cf00df574e942f563344a7d8ff15a7625 +size 44235 diff --git a/parse/dev/09hVcSDkea/images/c649970c6ab9f97e43291e9e16de40c65e8ad4cf1995b66201d058e9c24be0c1.jpg b/parse/dev/09hVcSDkea/images/c649970c6ab9f97e43291e9e16de40c65e8ad4cf1995b66201d058e9c24be0c1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9a28b2ea96191d8e9ec4671364c563b3f29eb8a8 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/c649970c6ab9f97e43291e9e16de40c65e8ad4cf1995b66201d058e9c24be0c1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8feb3a27ee7ca92bbcf05ae4839ab74cc49bbb3fac007b659bbc6ffa0a26c842 +size 113815 diff --git a/parse/dev/09hVcSDkea/images/c9f2e2d122de139b2a018c368ebcb283b9709621d2597891c248f8b565ae33bc.jpg b/parse/dev/09hVcSDkea/images/c9f2e2d122de139b2a018c368ebcb283b9709621d2597891c248f8b565ae33bc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bac7f7a48b0a038f15ccbd8b2542bdac868aa4f6 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/c9f2e2d122de139b2a018c368ebcb283b9709621d2597891c248f8b565ae33bc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:964e4b7362fd11f6fbe73cdf52e2a0500d890972950be3127c1daa590f956ff7 +size 15115 diff --git a/parse/dev/09hVcSDkea/images/cf62d90ea51b0125ac51771087f802e651e4c0fc7f7efb54455bf3cf8b69c76b.jpg b/parse/dev/09hVcSDkea/images/cf62d90ea51b0125ac51771087f802e651e4c0fc7f7efb54455bf3cf8b69c76b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8fef246d8eb401f6760f809125f256d7cd4e78f0 --- /dev/null +++ b/parse/dev/09hVcSDkea/images/cf62d90ea51b0125ac51771087f802e651e4c0fc7f7efb54455bf3cf8b69c76b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec6e1638d42809ca72b07a7e0fc887a001b3495594675ec3a07f7e1ce9c9c115 +size 48649 diff --git a/parse/dev/09hVcSDkea/images/ee5dc80eb05b3793abbd4054926c6210fd9d335cfb82360a4506a82308a74359.jpg b/parse/dev/09hVcSDkea/images/ee5dc80eb05b3793abbd4054926c6210fd9d335cfb82360a4506a82308a74359.jpg new file mode 100644 index 0000000000000000000000000000000000000000..60c48468915f7e4b91692477534365e09838e6ed --- /dev/null +++ b/parse/dev/09hVcSDkea/images/ee5dc80eb05b3793abbd4054926c6210fd9d335cfb82360a4506a82308a74359.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78ed5e4fbbe89b173b3e2eeb8b013707604f45bb186a2c48e3763e964fad74a3 +size 25511 diff --git a/parse/dev/3FvF1db-bKT/3FvF1db-bKT_layout.pdf b/parse/dev/3FvF1db-bKT/3FvF1db-bKT_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..09397646b43c62721896bfa87070f97cc0719ca6 --- /dev/null +++ b/parse/dev/3FvF1db-bKT/3FvF1db-bKT_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:143ba12f5d9ddf5e2eb3191f785d462b660faf64f77b155dc1c42c616dc3663b +size 743641 diff --git a/parse/dev/3RBY8fKjHeu/3RBY8fKjHeu_span.pdf b/parse/dev/3RBY8fKjHeu/3RBY8fKjHeu_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..543047bc8eec81fc43d9ac98b7bcf0c3be9232a9 --- /dev/null +++ b/parse/dev/3RBY8fKjHeu/3RBY8fKjHeu_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac3f154a4d5c59c8ed7f6e70ea7eb2c7f40626e36c450a35b7260a1d86b802c2 +size 3191163 diff --git a/parse/dev/42zs3qa2kpy/42zs3qa2kpy_model.json b/parse/dev/42zs3qa2kpy/42zs3qa2kpy_model.json new file mode 100644 index 0000000000000000000000000000000000000000..f5693b27c7a25063ed1ade19dc66028546dc71c1 --- /dev/null +++ b/parse/dev/42zs3qa2kpy/42zs3qa2kpy_model.json @@ -0,0 +1,31855 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 397, + 737, + 1303, + 737, + 1303, + 1319, + 397, + 1319 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1437, + 1403, + 1437, + 1403, + 1713, + 298, + 1713 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 299, + 1728, + 1404, + 1728, + 1404, + 1912, + 299, + 1912 + ], + "score": 0.977 + }, + { + "category_id": 0, + "poly": [ + 299, + 220, + 1406, + 220, + 1406, + 324, + 299, + 324 + ], + "score": 0.962 + }, + { + "category_id": 2, + "poly": [ + 337, + 2006, + 784, + 2006, + 784, + 2033, + 337, + 2033 + ], + "score": 0.906 + }, + { + "category_id": 1, + "poly": [ + 301, + 1927, + 1403, + 1927, + 1403, + 1988, + 301, + 1988 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 302, + 1369, + 573, + 1369, + 573, + 1404, + 302, + 1404 + ], + "score": 0.889 + }, + { + "category_id": 1, + "poly": [ + 315, + 374, + 1231, + 374, + 1231, + 594, + 315, + 594 + ], + "score": 0.884 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.881 + }, + { + "category_id": 0, + "poly": [ + 772, + 674, + 927, + 674, + 927, + 707, + 772, + 707 + ], + "score": 0.878 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 856, + 2088, + 856, + 2112, + 840, + 2112 + ], + "score": 0.743 + }, + { + "category_id": 13, + "poly": [ + 572, + 375, + 622, + 375, + 622, + 407, + 572, + 407 + ], + "score": 0.82, + "latex": "{ { \\bf L } } { \\bf u } ^ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 927, + 375, + 991, + 375, + 991, + 408, + 927, + 408 + ], + "score": 0.81, + "latex": "\\mathbf { S u } ^ { 1 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 775, + 375, + 844, + 375, + 844, + 409, + 775, + 409 + ], + "score": 0.72, + "latex": "\\mathbf { Y i n g ^ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 1002, + 374, + 1148, + 374, + 1148, + 408, + 1002, + 408 + ], + "score": 0.51, + "latex": " { \\mathbf { J u n } } { \\mathbf { Z } } { \\mathbf { h u } } ^ { 1 , 2 * }" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 217.0, + 1413.0, + 217.0, + 1413.0, + 271.0, + 297.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 273.0, + 1255.0, + 273.0, + 1255.0, + 328.0, + 293.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 2000.0, + 788.0, + 2000.0, + 788.0, + 2039.0, + 332.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1366.0, + 579.0, + 1366.0, + 579.0, + 1413.0, + 294.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 673.0, + 932.0, + 673.0, + 932.0, + 710.0, + 769.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 859.0, + 2087.0, + 859.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 739.0, + 1304.0, + 739.0, + 1304.0, + 772.0, + 395.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 767.0, + 1305.0, + 767.0, + 1305.0, + 806.0, + 392.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 800.0, + 1304.0, + 800.0, + 1304.0, + 833.0, + 394.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 831.0, + 1307.0, + 831.0, + 1307.0, + 863.0, + 394.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 861.0, + 1305.0, + 861.0, + 1305.0, + 894.0, + 393.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 891.0, + 1306.0, + 891.0, + 1306.0, + 927.0, + 393.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 924.0, + 1307.0, + 924.0, + 1307.0, + 954.0, + 393.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 950.0, + 1307.0, + 950.0, + 1307.0, + 988.0, + 392.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 981.0, + 1305.0, + 981.0, + 1305.0, + 1019.0, + 392.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1013.0, + 1307.0, + 1013.0, + 1307.0, + 1046.0, + 394.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1044.0, + 1306.0, + 1044.0, + 1306.0, + 1077.0, + 394.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1075.0, + 1307.0, + 1075.0, + 1307.0, + 1108.0, + 394.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1104.0, + 1305.0, + 1104.0, + 1305.0, + 1140.0, + 394.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1133.0, + 1306.0, + 1133.0, + 1306.0, + 1170.0, + 393.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1164.0, + 1306.0, + 1164.0, + 1306.0, + 1201.0, + 393.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1196.0, + 1305.0, + 1196.0, + 1305.0, + 1229.0, + 393.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1228.0, + 1303.0, + 1228.0, + 1303.0, + 1258.0, + 395.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1257.0, + 1304.0, + 1257.0, + 1304.0, + 1290.0, + 394.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1288.0, + 1239.0, + 1288.0, + 1239.0, + 1321.0, + 394.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1435.0, + 1403.0, + 1435.0, + 1403.0, + 1473.0, + 296.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1467.0, + 1404.0, + 1467.0, + 1404.0, + 1502.0, + 294.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1498.0, + 1405.0, + 1498.0, + 1405.0, + 1535.0, + 293.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1524.0, + 1407.0, + 1524.0, + 1407.0, + 1567.0, + 291.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1559.0, + 1408.0, + 1559.0, + 1408.0, + 1595.0, + 293.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1585.0, + 1407.0, + 1585.0, + 1407.0, + 1629.0, + 292.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1621.0, + 1407.0, + 1621.0, + 1407.0, + 1657.0, + 294.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1648.0, + 1408.0, + 1648.0, + 1408.0, + 1687.0, + 293.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1678.0, + 1147.0, + 1678.0, + 1147.0, + 1720.0, + 292.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1727.0, + 1405.0, + 1727.0, + 1405.0, + 1763.0, + 294.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1760.0, + 1405.0, + 1760.0, + 1405.0, + 1792.0, + 297.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1788.0, + 1408.0, + 1788.0, + 1408.0, + 1824.0, + 294.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1819.0, + 1406.0, + 1819.0, + 1406.0, + 1854.0, + 294.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1848.0, + 1408.0, + 1848.0, + 1408.0, + 1886.0, + 293.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1881.0, + 1217.0, + 1881.0, + 1217.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1925.0, + 1403.0, + 1925.0, + 1403.0, + 1963.0, + 296.0, + 1963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1955.0, + 1404.0, + 1955.0, + 1404.0, + 1993.0, + 295.0, + 1993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 373.0, + 571.0, + 373.0, + 571.0, + 414.0, + 308.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 373.0, + 774.0, + 373.0, + 774.0, + 414.0, + 623.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 373.0, + 926.0, + 373.0, + 926.0, + 414.0, + 845.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 373.0, + 1001.0, + 373.0, + 1001.0, + 414.0, + 992.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 373.0, + 1154.0, + 373.0, + 1154.0, + 414.0, + 1149.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 405.0, + 1232.0, + 405.0, + 1232.0, + 446.0, + 311.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 436.0, + 1058.0, + 436.0, + 1058.0, + 477.0, + 310.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 466.0, + 784.0, + 466.0, + 784.0, + 506.0, + 311.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 501.0, + 851.0, + 501.0, + 851.0, + 537.0, + 313.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 531.0, + 863.0, + 531.0, + 863.0, + 567.0, + 313.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 563.0, + 845.0, + 563.0, + 845.0, + 600.0, + 314.0, + 600.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 857, + 1405, + 857, + 1405, + 1224, + 298, + 1224 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 298, + 1240, + 1405, + 1240, + 1405, + 1455, + 298, + 1455 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 567, + 1404, + 567, + 1404, + 843, + 298, + 843 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1630, + 1404, + 1630, + 1404, + 1787, + 298, + 1787 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1402, + 228, + 1402, + 415, + 298, + 415 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 429, + 1404, + 429, + 1404, + 552, + 299, + 552 + ], + "score": 0.973 + }, + { + "category_id": 8, + "poly": [ + 550, + 1881, + 1148, + 1881, + 1148, + 1956, + 550, + 1956 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 293, + 1800, + 1403, + 1800, + 1403, + 1864, + 293, + 1864 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 296, + 1971, + 1408, + 1971, + 1408, + 2039, + 296, + 2039 + ], + "score": 0.953 + }, + { + "category_id": 0, + "poly": [ + 300, + 1501, + 557, + 1501, + 557, + 1538, + 300, + 1538 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 299, + 1572, + 931, + 1572, + 931, + 1606, + 299, + 1606 + ], + "score": 0.908 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.898 + }, + { + "category_id": 9, + "poly": [ + 1368, + 1902, + 1401, + 1902, + 1401, + 1932, + 1368, + 1932 + ], + "score": 0.878 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.725 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.289 + }, + { + "category_id": 13, + "poly": [ + 834, + 1663, + 916, + 1663, + 916, + 1697, + 834, + 1697 + ], + "score": 0.95, + "latex": "r ( s , a )" + }, + { + "category_id": 14, + "poly": [ + 551, + 1880, + 1150, + 1880, + 1150, + 1954, + 551, + 1954 + ], + "score": 0.94, + "latex": "\\nabla _ { \\theta } J ( \\pi _ { \\theta } ) = \\int _ { S } \\rho _ { \\pi } ( s ) \\int _ { A } \\nabla _ { \\theta } \\pi _ { \\theta } ( \\boldsymbol { a } | s ) Q ^ { \\pi } ( s , \\boldsymbol { a } ) \\mathrm { d } a \\ \\mathrm { d } s ." + }, + { + "category_id": 13, + "poly": [ + 793, + 2002, + 1239, + 2002, + 1239, + 2039, + 793, + 2039 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\eta ( \\bar { \\boldsymbol { \\pi } } _ { \\boldsymbol { \\theta } } ) : = \\int _ { S } \\rho _ { \\pi } ( \\pmb { \\mathscr { s } } ) \\int _ { \\mathcal { A } } \\pi _ { \\boldsymbol { \\theta } } ( \\pmb { a } | \\pmb { \\mathscr { s } } ) Q ^ { \\pi } ( \\pmb { \\mathscr { s } } , \\dot { \\pmb { a } } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 663, + 1663, + 782, + 1663, + 782, + 1697, + 663, + 1697 + ], + "score": 0.93, + "latex": "P ( \\pmb { s } ^ { \\prime } | \\pmb { s } , \\pmb { a } )" + }, + { + "category_id": 13, + "poly": [ + 886, + 1832, + 955, + 1832, + 955, + 1865, + 886, + 1865 + ], + "score": 0.93, + "latex": "J ( \\pi _ { \\theta } )" + }, + { + "category_id": 13, + "poly": [ + 990, + 1723, + 1334, + 1723, + 1334, + 1758, + 990, + 1758 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\rho _ { \\pi } ( \\pmb { \\mathscr { s } } ) = \\sum _ { n = 0 } ^ { \\infty } \\gamma ^ { n } \\bar { p } _ { \\pi } ( \\pmb { \\mathscr { s } } _ { n } = \\pmb { \\mathscr { s } } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1021, + 1632, + 1179, + 1632, + 1179, + 1665, + 1021, + 1665 + ], + "score": 0.92, + "latex": "\\langle S , \\mathcal { A } , P , r , \\gamma \\rangle" + }, + { + "category_id": 13, + "poly": [ + 517, + 1695, + 627, + 1695, + 627, + 1726, + 517, + 1726 + ], + "score": 0.91, + "latex": "\\gamma \\in ( 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 1180, + 1973, + 1247, + 1973, + 1247, + 2004, + 1180, + 2004 + ], + "score": 0.91, + "latex": "\\rho _ { \\pi } ( s )" + }, + { + "category_id": 13, + "poly": [ + 371, + 1725, + 774, + 1725, + 774, + 1760, + 371, + 1760 + ], + "score": 0.9, + "latex": "J ( \\pi ) = \\mathbb { E } _ { s \\sim \\rho _ { \\pi } ( s ) } \\mathbb { E } _ { \\pmb { a } \\sim \\pi ( \\cdot | s ) } \\left[ r ( s , \\pmb { a } ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 667, + 1834, + 705, + 1834, + 705, + 1864, + 667, + 1864 + ], + "score": 0.89, + "latex": "Q ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 736, + 1986, + 753, + 1986, + 753, + 2000, + 736, + 2000 + ], + "score": 0.86, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1315, + 1808, + 1346, + 1808, + 1346, + 1832, + 1315, + 1832 + ], + "score": 0.83, + "latex": "\\pi _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 414, + 1665, + 439, + 1665, + 439, + 1691, + 414, + 1691 + ], + "score": 0.82, + "latex": "\\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 972, + 1760, + 992, + 1760, + 992, + 1782, + 972, + 1782 + ], + "score": 0.78, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 884, + 1730, + 904, + 1730, + 904, + 1753, + 884, + 1753 + ], + "score": 0.77, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1192, + 1633, + 1215, + 1633, + 1215, + 1661, + 1192, + 1661 + ], + "score": 0.72, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 1085, + 1759, + 1109, + 1759, + 1109, + 1782, + 1085, + 1782 + ], + "score": 0.51, + "latex": "\\&" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1498.0, + 561.0, + 1498.0, + 561.0, + 1545.0, + 291.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1572.0, + 935.0, + 1572.0, + 935.0, + 1609.0, + 295.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 859.0, + 1406.0, + 859.0, + 1406.0, + 893.0, + 295.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 888.0, + 1410.0, + 888.0, + 1410.0, + 925.0, + 291.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 918.0, + 1406.0, + 918.0, + 1406.0, + 954.0, + 292.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 949.0, + 1405.0, + 949.0, + 1405.0, + 985.0, + 293.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 977.0, + 1406.0, + 977.0, + 1406.0, + 1020.0, + 292.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1011.0, + 1405.0, + 1011.0, + 1405.0, + 1044.0, + 292.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1043.0, + 1408.0, + 1043.0, + 1408.0, + 1077.0, + 293.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1068.0, + 1410.0, + 1068.0, + 1410.0, + 1110.0, + 292.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1104.0, + 1406.0, + 1104.0, + 1406.0, + 1138.0, + 295.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1132.0, + 1405.0, + 1132.0, + 1405.0, + 1169.0, + 293.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1162.0, + 1405.0, + 1162.0, + 1405.0, + 1200.0, + 293.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1192.0, + 941.0, + 1192.0, + 941.0, + 1231.0, + 293.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1235.0, + 1405.0, + 1235.0, + 1405.0, + 1277.0, + 292.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1270.0, + 1406.0, + 1270.0, + 1406.0, + 1308.0, + 293.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1302.0, + 1403.0, + 1302.0, + 1403.0, + 1336.0, + 293.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1328.0, + 1406.0, + 1328.0, + 1406.0, + 1370.0, + 292.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1363.0, + 1408.0, + 1363.0, + 1408.0, + 1398.0, + 295.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1392.0, + 1405.0, + 1392.0, + 1405.0, + 1427.0, + 293.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1423.0, + 1410.0, + 1423.0, + 1410.0, + 1458.0, + 292.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 568.0, + 1406.0, + 568.0, + 1406.0, + 604.0, + 296.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 594.0, + 1406.0, + 594.0, + 1406.0, + 640.0, + 291.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 629.0, + 1405.0, + 629.0, + 1405.0, + 666.0, + 294.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 658.0, + 1405.0, + 658.0, + 1405.0, + 695.0, + 294.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 688.0, + 1405.0, + 688.0, + 1405.0, + 725.0, + 294.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 721.0, + 1406.0, + 721.0, + 1406.0, + 757.0, + 294.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 749.0, + 1402.0, + 749.0, + 1402.0, + 786.0, + 294.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 781.0, + 1402.0, + 781.0, + 1402.0, + 817.0, + 294.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 809.0, + 1235.0, + 809.0, + 1235.0, + 847.0, + 293.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1632.0, + 1020.0, + 1632.0, + 1020.0, + 1666.0, + 297.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1632.0, + 1191.0, + 1632.0, + 1191.0, + 1666.0, + 1180.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1632.0, + 1405.0, + 1632.0, + 1405.0, + 1666.0, + 1216.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1660.0, + 413.0, + 1660.0, + 413.0, + 1699.0, + 292.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1660.0, + 662.0, + 1660.0, + 662.0, + 1699.0, + 440.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1660.0, + 833.0, + 1660.0, + 833.0, + 1699.0, + 783.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1660.0, + 1409.0, + 1660.0, + 1409.0, + 1699.0, + 917.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1694.0, + 516.0, + 1694.0, + 516.0, + 1727.0, + 296.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1694.0, + 1404.0, + 1694.0, + 1404.0, + 1727.0, + 628.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1712.0, + 370.0, + 1712.0, + 370.0, + 1771.0, + 288.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 1712.0, + 883.0, + 1712.0, + 883.0, + 1771.0, + 775.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 1712.0, + 989.0, + 1712.0, + 989.0, + 1771.0, + 905.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 1712.0, + 1410.0, + 1712.0, + 1410.0, + 1771.0, + 1335.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1753.0, + 971.0, + 1753.0, + 971.0, + 1790.0, + 294.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1753.0, + 1084.0, + 1753.0, + 1084.0, + 1790.0, + 993.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 1753.0, + 1267.0, + 1753.0, + 1267.0, + 1790.0, + 1110.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 228.0, + 1406.0, + 228.0, + 1406.0, + 266.0, + 293.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 261.0, + 1406.0, + 261.0, + 1406.0, + 297.0, + 294.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 288.0, + 1406.0, + 288.0, + 1406.0, + 328.0, + 292.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 322.0, + 1407.0, + 322.0, + 1407.0, + 357.0, + 294.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 352.0, + 1404.0, + 352.0, + 1404.0, + 388.0, + 294.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 384.0, + 1401.0, + 384.0, + 1401.0, + 416.0, + 297.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 429.0, + 1409.0, + 429.0, + 1409.0, + 465.0, + 294.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 460.0, + 1409.0, + 460.0, + 1409.0, + 496.0, + 293.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 490.0, + 1405.0, + 490.0, + 1405.0, + 525.0, + 294.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 521.0, + 1016.0, + 521.0, + 1016.0, + 556.0, + 294.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1799.0, + 1314.0, + 1799.0, + 1314.0, + 1837.0, + 295.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1799.0, + 1405.0, + 1799.0, + 1405.0, + 1837.0, + 1347.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1831.0, + 666.0, + 1831.0, + 666.0, + 1867.0, + 295.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1831.0, + 885.0, + 1831.0, + 885.0, + 1867.0, + 706.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1831.0, + 1164.0, + 1831.0, + 1164.0, + 1867.0, + 956.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1971.0, + 735.0, + 1971.0, + 735.0, + 2008.0, + 295.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 1971.0, + 1179.0, + 1971.0, + 1179.0, + 2008.0, + 754.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 1971.0, + 1407.0, + 1971.0, + 1407.0, + 2008.0, + 1248.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2000.0, + 792.0, + 2000.0, + 792.0, + 2040.0, + 293.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 2000.0, + 1406.0, + 2000.0, + 1406.0, + 2040.0, + 1240.0, + 2040.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1425, + 1405, + 1425, + 1405, + 1647, + 296, + 1647 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1210, + 1405, + 1210, + 1405, + 1335, + 297, + 1335 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 1776, + 1406, + 1776, + 1406, + 1902, + 296, + 1902 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 295, + 227, + 1406, + 227, + 1406, + 385, + 295, + 385 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 482, + 1405, + 482, + 1405, + 577, + 296, + 577 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 300, + 878, + 1404, + 878, + 1404, + 971, + 300, + 971 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 490, + 984, + 1204, + 984, + 1204, + 1120, + 490, + 1120 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 624, + 1915, + 1072, + 1915, + 1072, + 1990, + 624, + 1990 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 297, + 668, + 1402, + 668, + 1402, + 734, + 297, + 734 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 605, + 748, + 1094, + 748, + 1094, + 817, + 605, + 817 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 297, + 1132, + 1404, + 1132, + 1404, + 1196, + 297, + 1196 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 297, + 2001, + 1144, + 2001, + 1144, + 2036, + 297, + 2036 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 298, + 831, + 1213, + 831, + 1213, + 866, + 298, + 866 + ], + "score": 0.934 + }, + { + "category_id": 8, + "poly": [ + 355, + 398, + 1334, + 398, + 1334, + 472, + 355, + 472 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 297, + 611, + 995, + 611, + 995, + 645, + 297, + 645 + ], + "score": 0.918 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 815, + 74, + 815, + 105, + 299, + 105 + ], + "score": 0.915 + }, + { + "category_id": 0, + "poly": [ + 302, + 1370, + 786, + 1370, + 786, + 1402, + 302, + 1402 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1368, + 1066, + 1402, + 1066, + 1402, + 1096, + 1368, + 1096 + ], + "score": 0.879 + }, + { + "category_id": 9, + "poly": [ + 1367, + 766, + 1402, + 766, + 1402, + 795, + 1367, + 795 + ], + "score": 0.868 + }, + { + "category_id": 8, + "poly": [ + 630, + 1657, + 1069, + 1657, + 1069, + 1712, + 630, + 1712 + ], + "score": 0.867 + }, + { + "category_id": 9, + "poly": [ + 1368, + 417, + 1402, + 417, + 1402, + 448, + 1368, + 448 + ], + "score": 0.864 + }, + { + "category_id": 1, + "poly": [ + 296, + 1728, + 1043, + 1728, + 1043, + 1765, + 296, + 1765 + ], + "score": 0.805 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1664, + 1402, + 1664, + 1402, + 1693, + 1367, + 1693 + ], + "score": 0.789 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1938, + 1402, + 1938, + 1402, + 1967, + 1367, + 1967 + ], + "score": 0.712 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.671 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.516 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1938, + 1402, + 1938, + 1402, + 1967, + 1367, + 1967 + ], + "score": 0.239 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1664, + 1402, + 1664, + 1402, + 1693, + 1367, + 1693 + ], + "score": 0.182 + }, + { + "category_id": 13, + "poly": [ + 619, + 291, + 687, + 291, + 687, + 325, + 619, + 325 + ], + "score": 0.95, + "latex": "\\rho _ { \\pi } ( s )" + }, + { + "category_id": 14, + "poly": [ + 625, + 1914, + 1075, + 1914, + 1075, + 1991, + 625, + 1991 + ], + "score": 0.95, + "latex": "\\mathrm { d } { \\pmb x } _ { t } = \\bigg [ f ( { \\pmb x } _ { t } , t ) - \\frac { 1 } { 2 } g ^ { 2 } ( t ) { \\pmb s } _ { \\theta } ( { \\pmb x } _ { t } , t ) \\bigg ] \\mathrm { d } t ." + }, + { + "category_id": 14, + "poly": [ + 604, + 745, + 1094, + 745, + 1094, + 818, + 604, + 818 + ], + "score": 0.94, + "latex": "\\pi ^ { * } ( { \\pmb a } | { \\pmb s } ) = \\frac { 1 } { Z ( { \\pmb s } ) } \\mu ( { \\pmb a } | { \\pmb s } ) \\exp \\left( \\alpha Q _ { \\phi } ( { \\pmb s } , { \\pmb a } ) \\right) ," + }, + { + "category_id": 13, + "poly": [ + 965, + 1489, + 1115, + 1489, + 1115, + 1526, + 965, + 1526 + ], + "score": 0.94, + "latex": "\\{ \\pmb { x } ( t ) \\} _ { t \\in [ 0 , T ] }" + }, + { + "category_id": 13, + "poly": [ + 1166, + 483, + 1271, + 483, + 1271, + 518, + 1166, + 518 + ], + "score": 0.94, + "latex": "Q _ { \\phi } ( s , \\pmb { a } )" + }, + { + "category_id": 13, + "poly": [ + 732, + 1241, + 824, + 1241, + 824, + 1275, + 732, + 1275 + ], + "score": 0.93, + "latex": "\\pi _ { \\boldsymbol { \\theta } } ( \\pmb { a } | \\pmb { s } )" + }, + { + "category_id": 13, + "poly": [ + 806, + 230, + 972, + 230, + 972, + 264, + 806, + 264 + ], + "score": 0.93, + "latex": "\\{ ( s _ { n } , a _ { n } , r _ { n } ) \\}" + }, + { + "category_id": 14, + "poly": [ + 357, + 396, + 1339, + 396, + 1339, + 471, + 357, + 471 + ], + "score": 0.93, + "latex": "\\arg \\operatorname* { m a x } _ { \\pi } \\quad \\int _ { S } \\rho _ { \\mu } ( s ) \\int _ { \\mathcal { A } } \\pi ( a | s ) Q _ { \\phi } ( s , a ) \\mathrm { d } a \\mathrm { d } s - \\frac { 1 } { \\alpha } \\int _ { S } \\rho _ { \\mu } ( s ) D _ { \\mathrm { K L } } \\left( \\pi ( \\cdot | s ) \\right| | \\mu ( \\cdot | s ) ) \\mathrm { d } s ." + }, + { + "category_id": 13, + "poly": [ + 608, + 1522, + 931, + 1522, + 931, + 1557, + 608, + 1557 + ], + "score": 0.93, + "latex": "\\mathrm { d } \\pmb { x } _ { t } = f ( \\pmb { x } _ { t } , t ) \\mathrm { d } t + g ( t ) \\mathrm { d } \\pmb { w } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 954, + 1583, + 1221, + 1583, + 1221, + 1617, + 954, + 1617 + ], + "score": 0.93, + "latex": "p _ { T } ( \\pmb { x } _ { T } ) \\approx \\mathcal { N } ( \\pmb { x } _ { T } | 0 , \\pmb { I } )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1869, + 390, + 1869, + 390, + 1903, + 297, + 1903 + ], + "score": 0.93, + "latex": "\\mathcal { N } ( 0 , \\pmb { I } )" + }, + { + "category_id": 13, + "poly": [ + 1016, + 483, + 1083, + 483, + 1083, + 517, + 1016, + 517 + ], + "score": 0.92, + "latex": "\\hat { \\eta } ( \\pi _ { \\boldsymbol { \\theta } } )" + }, + { + "category_id": 13, + "poly": [ + 1153, + 261, + 1220, + 261, + 1220, + 295, + 1153, + 295 + ], + "score": 0.92, + "latex": "\\eta ( \\pi _ { \\theta } )" + }, + { + "category_id": 13, + "poly": [ + 777, + 1585, + 901, + 1585, + 901, + 1615, + 777, + 1615 + ], + "score": 0.92, + "latex": "\\alpha _ { t } , \\sigma _ { t } > 0" + }, + { + "category_id": 13, + "poly": [ + 404, + 292, + 471, + 292, + 471, + 324, + 404, + 324 + ], + "score": 0.92, + "latex": "\\hat { \\eta } ( \\pi _ { \\boldsymbol { \\theta } } )" + }, + { + "category_id": 14, + "poly": [ + 485, + 982, + 1211, + 982, + 1211, + 1123, + 485, + 1123 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } { \\underset { \\theta } { \\mathrm { a r g } \\mathrm { m i n } } } & { \\mathbb { E } _ { s \\sim \\mathcal { D } ^ { \\mu } } \\left[ D _ { \\mathrm { K L } } \\left( \\pi ^ { * } ( \\cdot | s ) | | \\pi _ { \\theta } ( \\cdot | s ) \\right) \\right] } \\\\ { \\underset { \\theta } { = \\mathrm { a r g } \\mathrm { m a x } } } & { \\mathbb { E } _ { ( s , a ) \\sim \\mathcal { D } ^ { \\mu } } \\left[ \\cfrac { 1 } { Z ( s ) } \\mathrm { l o g } \\pi _ { \\theta } ( a | s ) \\exp \\left( \\alpha Q _ { \\phi } ( s , a ) \\right) \\right] . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1489, + 378, + 1489, + 378, + 1523, + 297, + 1523 + ], + "score": 0.92, + "latex": "p _ { 0 } ( { \\pmb x } _ { 0 } )" + }, + { + "category_id": 13, + "poly": [ + 1317, + 230, + 1397, + 230, + 1397, + 264, + 1317, + 264 + ], + "score": 0.92, + "latex": "\\mu ( { \\pmb a } | { \\pmb s } )" + }, + { + "category_id": 13, + "poly": [ + 372, + 832, + 430, + 832, + 430, + 866, + 372, + 866 + ], + "score": 0.92, + "latex": "Z ( s )" + }, + { + "category_id": 13, + "poly": [ + 747, + 292, + 815, + 292, + 815, + 325, + 747, + 325 + ], + "score": 0.91, + "latex": "\\rho _ { \\mu } ( s )" + }, + { + "category_id": 13, + "poly": [ + 1022, + 1132, + 1210, + 1132, + 1210, + 1168, + 1022, + 1168 + ], + "score": 0.91, + "latex": "\\exp { ( \\alpha Q _ { \\phi } ( s , pmb { a } ) ) }" + }, + { + "category_id": 13, + "poly": [ + 696, + 547, + 735, + 547, + 735, + 572, + 696, + 572 + ], + "score": 0.89, + "latex": "{ \\mathcal { D } } ^ { \\mu }" + }, + { + "category_id": 14, + "poly": [ + 626, + 1657, + 1069, + 1657, + 1069, + 1716, + 626, + 1716 + ], + "score": 0.88, + "latex": "\\underset { \\theta } { \\arg \\operatorname* { m i n } } \\quad \\mathbb { E } _ { t , { \\boldsymbol { x } } _ { 0 } , { \\boldsymbol { \\epsilon } } } [ \\| { \\boldsymbol { \\sigma } } _ { t } { \\bf s } _ { \\theta } ( { \\boldsymbol { x } } _ { t } , t ) + { \\boldsymbol { \\epsilon } } \\| _ { 2 } ^ { 2 } ] ," + }, + { + "category_id": 13, + "poly": [ + 297, + 1583, + 664, + 1583, + 664, + 1617, + 297, + 1617 + ], + "score": 0.88, + "latex": "p _ { t 0 } ( \\pmb { x } _ { t } | \\pmb { x } _ { 0 } ) = \\mathcal { N } ( \\pmb { x } _ { t } | \\alpha _ { t } \\pmb { x } _ { 0 } , \\sigma _ { t } ^ { 2 } \\pmb { I } )" + }, + { + "category_id": 13, + "poly": [ + 1002, + 1620, + 1032, + 1620, + 1032, + 1645, + 1002, + 1645 + ], + "score": 0.88, + "latex": "\\scriptstyle { \\pmb { s } } \\theta" + }, + { + "category_id": 13, + "poly": [ + 1018, + 1529, + 1053, + 1529, + 1053, + 1554, + 1018, + 1554 + ], + "score": 0.87, + "latex": "{ \\pmb w } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 570, + 881, + 602, + 881, + 602, + 908, + 570, + 908 + ], + "score": 0.87, + "latex": "\\pi ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 554, + 2008, + 586, + 2008, + 586, + 2034, + 554, + 2034 + ], + "score": 0.87, + "latex": "\\scriptstyle { \\pmb x } _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 382, + 232, + 421, + 232, + 421, + 258, + 382, + 258 + ], + "score": 0.86, + "latex": "{ \\mathcal { D } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 985, + 942, + 1017, + 942, + 1017, + 968, + 985, + 968 + ], + "score": 0.86, + "latex": "\\pi ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 510, + 673, + 541, + 673, + 541, + 699, + 510, + 699 + ], + "score": 0.85, + "latex": "\\pi ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1336, + 944, + 1366, + 944, + 1366, + 971, + 1336, + 971 + ], + "score": 0.85, + "latex": "\\pi _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1332, + 1842, + 1404, + 1842, + 1404, + 1869, + 1332, + 1869 + ], + "score": 0.85, + "latex": "x _ { T } \\sim" + }, + { + "category_id": 13, + "poly": [ + 904, + 1873, + 926, + 1873, + 926, + 1897, + 904, + 1897 + ], + "score": 0.84, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 1339, + 1247, + 1370, + 1247, + 1370, + 1273, + 1339, + 1273 + ], + "score": 0.84, + "latex": "\\pi _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 795, + 548, + 816, + 548, + 816, + 572, + 795, + 572 + ], + "score": 0.81, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 725, + 915, + 746, + 915, + 746, + 941, + 725, + 941 + ], + "score": 0.81, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 1345, + 1616, + 1362, + 1616, + 1362, + 1642, + 1345, + 1642 + ], + "score": 0.8, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 371, + 1730, + 508, + 1730, + 508, + 1764, + 371, + 1764 + ], + "score": 0.8, + "latex": "t \\sim \\mathcal { U } ( 0 , T )" + }, + { + "category_id": 13, + "poly": [ + 659, + 358, + 678, + 358, + 678, + 385, + 659, + 385 + ], + "score": 0.79, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 1039, + 886, + 1059, + 886, + 1059, + 911, + 1039, + 911 + ], + "score": 0.79, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 461, + 358, + 482, + 358, + 482, + 380, + 461, + 380 + ], + "score": 0.77, + "latex": "\\pi" + }, + { + "category_id": 14, + "poly": [ + 490, + 1729, + 1046, + 1729, + 1046, + 1765, + 490, + 1765 + ], + "score": 0.76, + "latex": "\\begin{array} { r } { \\hat { \\mathbf { \\Phi } } ) , \\pmb { x } _ { 0 } \\sim p _ { 0 } ( \\pmb { x } _ { 0 } ) , \\pmb { \\epsilon } \\sim \\mathcal { N } ( 0 , \\pmb { I } ) , \\pmb { x } _ { t } = \\alpha _ { t } \\pmb { x } _ { 0 } + \\sigma _ { t } \\pmb { \\epsilon } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 656, + 520, + 675, + 520, + 675, + 542, + 656, + 542 + ], + "score": 0.75, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 346, + 1553, + 457, + 1553, + 457, + 1585, + 346, + 1585 + ], + "score": 0.71, + "latex": "f ( t ) , g ( t )" + }, + { + "category_id": 13, + "poly": [ + 520, + 1728, + 1039, + 1728, + 1039, + 1765, + 520, + 1765 + ], + "score": 0.54, + "latex": "\\begin{array} { r } { \\pmb { x } _ { 0 } \\sim p _ { 0 } ( \\pmb { x } _ { 0 } ) , \\epsilon \\sim \\mathcal { N } ( 0 , \\pmb { I } ) , \\pmb { x } _ { t } = \\alpha _ { t } \\pmb { x } _ { 0 } + \\sigma _ { t } \\epsilon . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 675, + 2009, + 690, + 2009, + 690, + 2030, + 675, + 2030 + ], + "score": 0.42, + "latex": "0" + }, + { + "category_id": 13, + "poly": [ + 346, + 1553, + 396, + 1553, + 396, + 1586, + 346, + 1586 + ], + "score": 0.41, + "latex": "f ( t )" + }, + { + "category_id": 13, + "poly": [ + 409, + 1554, + 457, + 1554, + 457, + 1585, + 409, + 1585 + ], + "score": 0.28, + "latex": "g ( t )" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 611.0, + 998.0, + 611.0, + 998.0, + 648.0, + 295.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1370.0, + 788.0, + 1370.0, + 788.0, + 1406.0, + 296.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1425.0, + 1405.0, + 1425.0, + 1405.0, + 1463.0, + 294.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1458.0, + 1406.0, + 1458.0, + 1406.0, + 1497.0, + 293.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1487.0, + 296.0, + 1487.0, + 296.0, + 1530.0, + 293.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1487.0, + 964.0, + 1487.0, + 964.0, + 1530.0, + 379.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 1487.0, + 1407.0, + 1487.0, + 1407.0, + 1530.0, + 1116.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1523.0, + 607.0, + 1523.0, + 607.0, + 1559.0, + 296.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 1523.0, + 1017.0, + 1523.0, + 1017.0, + 1559.0, + 932.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1523.0, + 1405.0, + 1523.0, + 1405.0, + 1559.0, + 1054.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1551.0, + 345.0, + 1551.0, + 345.0, + 1589.0, + 293.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 1551.0, + 1406.0, + 1551.0, + 1406.0, + 1589.0, + 458.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1580.0, + 296.0, + 1580.0, + 296.0, + 1623.0, + 293.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 1580.0, + 776.0, + 1580.0, + 776.0, + 1623.0, + 665.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 1580.0, + 953.0, + 1580.0, + 953.0, + 1623.0, + 902.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 1580.0, + 1407.0, + 1580.0, + 1407.0, + 1623.0, + 1222.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1611.0, + 1001.0, + 1611.0, + 1001.0, + 1650.0, + 294.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1611.0, + 1344.0, + 1611.0, + 1344.0, + 1650.0, + 1033.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 1611.0, + 1407.0, + 1611.0, + 1407.0, + 1650.0, + 1363.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1209.0, + 1405.0, + 1209.0, + 1405.0, + 1247.0, + 292.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1240.0, + 731.0, + 1240.0, + 731.0, + 1277.0, + 292.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1240.0, + 1338.0, + 1240.0, + 1338.0, + 1277.0, + 825.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1371.0, + 1240.0, + 1406.0, + 1240.0, + 1406.0, + 1277.0, + 1371.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1272.0, + 1405.0, + 1272.0, + 1405.0, + 1306.0, + 294.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1304.0, + 1218.0, + 1304.0, + 1218.0, + 1337.0, + 295.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1774.0, + 1406.0, + 1774.0, + 1406.0, + 1814.0, + 294.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1806.0, + 1408.0, + 1806.0, + 1408.0, + 1844.0, + 291.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1837.0, + 1331.0, + 1837.0, + 1331.0, + 1874.0, + 291.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1868.0, + 296.0, + 1868.0, + 296.0, + 1905.0, + 293.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1868.0, + 903.0, + 1868.0, + 903.0, + 1905.0, + 391.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1868.0, + 1346.0, + 1868.0, + 1346.0, + 1905.0, + 927.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 381.0, + 227.0, + 381.0, + 268.0, + 293.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 227.0, + 805.0, + 227.0, + 805.0, + 268.0, + 422.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 227.0, + 1316.0, + 227.0, + 1316.0, + 268.0, + 973.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 227.0, + 1408.0, + 227.0, + 1408.0, + 268.0, + 1398.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 258.0, + 1152.0, + 258.0, + 1152.0, + 299.0, + 292.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 258.0, + 1407.0, + 258.0, + 1407.0, + 299.0, + 1221.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 286.0, + 403.0, + 286.0, + 403.0, + 332.0, + 292.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 286.0, + 618.0, + 286.0, + 618.0, + 332.0, + 472.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 286.0, + 746.0, + 286.0, + 746.0, + 332.0, + 688.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 286.0, + 1407.0, + 286.0, + 1407.0, + 332.0, + 816.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 319.0, + 1406.0, + 319.0, + 1406.0, + 357.0, + 293.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 352.0, + 460.0, + 352.0, + 460.0, + 385.0, + 294.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 352.0, + 658.0, + 352.0, + 658.0, + 385.0, + 483.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 352.0, + 1325.0, + 352.0, + 1325.0, + 385.0, + 679.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 482.0, + 1015.0, + 482.0, + 1015.0, + 520.0, + 293.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 482.0, + 1165.0, + 482.0, + 1165.0, + 520.0, + 1084.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 482.0, + 1406.0, + 482.0, + 1406.0, + 520.0, + 1272.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 515.0, + 655.0, + 515.0, + 655.0, + 549.0, + 296.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 515.0, + 1405.0, + 515.0, + 1405.0, + 549.0, + 676.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 544.0, + 695.0, + 544.0, + 695.0, + 579.0, + 294.0, + 579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 544.0, + 794.0, + 544.0, + 794.0, + 579.0, + 736.0, + 579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 544.0, + 1059.0, + 544.0, + 1059.0, + 579.0, + 817.0, + 579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 876.0, + 569.0, + 876.0, + 569.0, + 916.0, + 294.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 876.0, + 1038.0, + 876.0, + 1038.0, + 916.0, + 603.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 876.0, + 1405.0, + 876.0, + 1405.0, + 916.0, + 1060.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 905.0, + 724.0, + 905.0, + 724.0, + 949.0, + 293.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 905.0, + 1405.0, + 905.0, + 1405.0, + 949.0, + 747.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 937.0, + 984.0, + 937.0, + 984.0, + 978.0, + 293.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 937.0, + 1335.0, + 937.0, + 1335.0, + 978.0, + 1018.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1367.0, + 937.0, + 1380.0, + 937.0, + 1380.0, + 978.0, + 1367.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 669.0, + 509.0, + 669.0, + 509.0, + 705.0, + 295.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 669.0, + 1407.0, + 669.0, + 1407.0, + 705.0, + 542.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 698.0, + 713.0, + 698.0, + 713.0, + 737.0, + 295.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1126.0, + 1021.0, + 1126.0, + 1021.0, + 1173.0, + 291.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1126.0, + 1409.0, + 1126.0, + 1409.0, + 1173.0, + 1211.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1163.0, + 452.0, + 1163.0, + 452.0, + 1200.0, + 294.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 553.0, + 2000.0, + 553.0, + 2039.0, + 294.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 2000.0, + 674.0, + 2000.0, + 674.0, + 2039.0, + 587.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 2000.0, + 1145.0, + 2000.0, + 1145.0, + 2039.0, + 691.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 824.0, + 371.0, + 824.0, + 371.0, + 876.0, + 293.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 824.0, + 1218.0, + 824.0, + 1218.0, + 876.0, + 431.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1725.0, + 370.0, + 1725.0, + 370.0, + 1770.0, + 294.0, + 1770.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 509, + 1405, + 509, + 1405, + 817, + 297, + 817 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 832, + 766, + 832, + 766, + 1259, + 297, + 1259 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 293, + 1405, + 293, + 1405, + 417, + 298, + 417 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1458, + 1405, + 1458, + 1405, + 1581, + 298, + 1581 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1831, + 1405, + 1831, + 1405, + 1925, + 298, + 1925 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 298, + 1723, + 1406, + 1723, + 1406, + 1818, + 298, + 1818 + ], + "score": 0.968 + }, + { + "category_id": 3, + "poly": [ + 788, + 841, + 1395, + 841, + 1395, + 1107, + 788, + 1107 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 298, + 1274, + 1401, + 1274, + 1401, + 1367, + 298, + 1367 + ], + "score": 0.958 + }, + { + "category_id": 4, + "poly": [ + 784, + 1128, + 1404, + 1128, + 1404, + 1221, + 784, + 1221 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 295, + 1972, + 1413, + 1972, + 1413, + 2035, + 295, + 2035 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 640, + 1931, + 1057, + 1931, + 1057, + 1967, + 640, + 1967 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 301, + 1646, + 1408, + 1646, + 1408, + 1710, + 301, + 1710 + ], + "score": 0.938 + }, + { + "category_id": 0, + "poly": [ + 298, + 1401, + 874, + 1401, + 874, + 1434, + 298, + 1434 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 299, + 225, + 481, + 225, + 481, + 261, + 299, + 261 + ], + "score": 0.906 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.897 + }, + { + "category_id": 0, + "poly": [ + 300, + 452, + 830, + 452, + 830, + 484, + 300, + 484 + ], + "score": 0.895 + }, + { + "category_id": 8, + "poly": [ + 313, + 1587, + 1332, + 1587, + 1332, + 1639, + 313, + 1639 + ], + "score": 0.883 + }, + { + "category_id": 9, + "poly": [ + 1369, + 1592, + 1401, + 1592, + 1401, + 1622, + 1369, + 1622 + ], + "score": 0.832 + }, + { + "category_id": 9, + "poly": [ + 1368, + 1935, + 1401, + 1935, + 1401, + 1964, + 1368, + 1964 + ], + "score": 0.826 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 857, + 2089, + 857, + 2111, + 840, + 2111 + ], + "score": 0.789 + }, + { + "category_id": 9, + "poly": [ + 1368, + 1935, + 1401, + 1935, + 1401, + 1964, + 1368, + 1964 + ], + "score": 0.109 + }, + { + "category_id": 13, + "poly": [ + 479, + 1755, + 571, + 1755, + 571, + 1789, + 479, + 1789 + ], + "score": 0.94, + "latex": "\\pi _ { \\boldsymbol { \\theta } } ( \\pmb { a } | \\pmb { s } )" + }, + { + "category_id": 14, + "poly": [ + 640, + 1931, + 1060, + 1931, + 1060, + 1968, + 640, + 1968 + ], + "score": 0.93, + "latex": "\\pi ( { \\pmb a } | { \\pmb s } ) \\propto \\mu _ { \\theta } ( { \\pmb a } | { \\pmb s } ) \\exp \\left( \\alpha Q _ { \\phi } ( { \\pmb s } , { \\pmb a } ) \\right) ." + }, + { + "category_id": 14, + "poly": [ + 313, + 1586, + 1346, + 1586, + 1346, + 1642, + 313, + 1642 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } { \\underset { \\theta } { \\arg \\operatorname* { m a x } } } & { \\mathbb { E } _ { s \\sim \\mathcal { D } ^ { \\mu } } \\mathbb { E } _ { a \\sim \\pi ^ { * } ( \\cdot | s ) } \\left[ \\log \\pi _ { \\theta } ( a | s ) \\right] \\Leftrightarrow \\underset { \\theta } { \\arg \\operatorname* { m i n } } \\quad \\mathbb { E } _ { s \\sim \\mathcal { D } ^ { \\mu } } \\left[ D _ { \\mathrm { K L } } \\left( \\pi ^ { * } ( \\cdot | s ) | | \\pi _ { \\theta } ( \\cdot | s ) \\right) \\right] . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 749, + 1649, + 782, + 1649, + 782, + 1676, + 749, + 1676 + ], + "score": 0.88, + "latex": "\\pi ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1288, + 2004, + 1329, + 2004, + 1329, + 2030, + 1288, + 2030 + ], + "score": 0.88, + "latex": "D ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 841, + 1730, + 872, + 1730, + 872, + 1755, + 841, + 1755 + ], + "score": 0.86, + "latex": "\\pi _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1207, + 1977, + 1239, + 1977, + 1239, + 2004, + 1207, + 2004 + ], + "score": 0.86, + "latex": "\\mu _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1120, + 573, + 1152, + 573, + 1152, + 600, + 1120, + 600 + ], + "score": 0.86, + "latex": "\\pi ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 672, + 2007, + 703, + 2007, + 703, + 2034, + 672, + 2034 + ], + "score": 0.86, + "latex": "\\pi _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 502, + 1492, + 535, + 1492, + 535, + 1519, + 502, + 1519 + ], + "score": 0.85, + "latex": "\\pi ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 787, + 1526, + 818, + 1526, + 818, + 1551, + 787, + 1551 + ], + "score": 0.84, + "latex": "\\pi _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1137, + 1979, + 1158, + 1979, + 1158, + 2004, + 1137, + 2004 + ], + "score": 0.82, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 982, + 2008, + 1003, + 2008, + 1003, + 2035, + 982, + 2035 + ], + "score": 0.82, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 1082, + 1869, + 1101, + 1869, + 1101, + 1891, + 1082, + 1891 + ], + "score": 0.76, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1199, + 1839, + 1220, + 1839, + 1220, + 1860, + 1199, + 1860 + ], + "score": 0.75, + "latex": "\\pi" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 854.0, + 1151.0, + 854.0, + 1151.0, + 877.0, + 1116.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 849.0, + 1194.0, + 849.0, + 1194.0, + 870.0, + 1156.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 855.0, + 1375.0, + 855.0, + 1375.0, + 879.0, + 1282.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 873.0, + 1386.0, + 873.0, + 1386.0, + 903.0, + 1281.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 899.0, + 1152.0, + 899.0, + 1152.0, + 922.0, + 1123.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 912.0, + 1123.0, + 912.0, + 1123.0, + 1025.0, + 1100.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 929.0, + 1042.0, + 929.0, + 1042.0, + 960.0, + 930.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 945.0, + 1150.0, + 945.0, + 1150.0, + 965.0, + 1124.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 950.0, + 1047.0, + 950.0, + 1047.0, + 981.0, + 930.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 953.0, + 813.0, + 953.0, + 813.0, + 1056.0, + 790.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 954.0, + 1230.0, + 954.0, + 1230.0, + 976.0, + 1195.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 971.0, + 1060.0, + 971.0, + 1060.0, + 1001.0, + 930.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 975.0, + 1269.0, + 975.0, + 1269.0, + 998.0, + 1234.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 989.0, + 1150.0, + 989.0, + 1150.0, + 1008.0, + 1124.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 996.0, + 1306.0, + 996.0, + 1306.0, + 1018.0, + 1271.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1030.0, + 1151.0, + 1030.0, + 1151.0, + 1054.0, + 1123.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1311.0, + 1035.0, + 1342.0, + 1035.0, + 1342.0, + 1054.0, + 1311.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 1079.0, + 1147.0, + 1079.0, + 1147.0, + 1094.0, + 1135.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 1070.0, + 1390.0, + 1070.0, + 1390.0, + 1109.0, + 1151.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 852.5, + 1072.0, + 852.5, + 1072.0, + 867.0, + 1050.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.75, + 1077.5, + 1074.75, + 1077.5, + 1074.75, + 1099.5, + 1068.75, + 1099.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1127.0, + 1404.0, + 1127.0, + 1404.0, + 1161.0, + 785.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1157.0, + 1403.0, + 1157.0, + 1403.0, + 1194.0, + 784.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1191.0, + 1194.0, + 1191.0, + 1194.0, + 1222.0, + 787.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1399.0, + 877.0, + 1399.0, + 877.0, + 1438.0, + 293.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 220.0, + 486.0, + 220.0, + 486.0, + 270.0, + 292.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 452.0, + 834.0, + 452.0, + 834.0, + 488.0, + 295.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 510.0, + 1405.0, + 510.0, + 1405.0, + 546.0, + 294.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 541.0, + 1406.0, + 541.0, + 1406.0, + 576.0, + 295.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 569.0, + 1119.0, + 569.0, + 1119.0, + 609.0, + 291.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 569.0, + 1408.0, + 569.0, + 1408.0, + 609.0, + 1153.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 601.0, + 1407.0, + 601.0, + 1407.0, + 637.0, + 291.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 631.0, + 1406.0, + 631.0, + 1406.0, + 671.0, + 292.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 663.0, + 1406.0, + 663.0, + 1406.0, + 699.0, + 295.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 691.0, + 1406.0, + 691.0, + 1406.0, + 728.0, + 292.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 725.0, + 1403.0, + 725.0, + 1403.0, + 757.0, + 296.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 755.0, + 1405.0, + 755.0, + 1405.0, + 791.0, + 295.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 785.0, + 1342.0, + 785.0, + 1342.0, + 821.0, + 295.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 830.0, + 771.0, + 830.0, + 771.0, + 865.0, + 296.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 860.0, + 770.0, + 860.0, + 770.0, + 894.0, + 295.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 891.0, + 768.0, + 891.0, + 768.0, + 926.0, + 295.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 923.0, + 771.0, + 923.0, + 771.0, + 955.0, + 295.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 954.0, + 768.0, + 954.0, + 768.0, + 986.0, + 295.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 984.0, + 770.0, + 984.0, + 770.0, + 1016.0, + 294.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1013.0, + 770.0, + 1013.0, + 770.0, + 1051.0, + 293.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1046.0, + 771.0, + 1046.0, + 771.0, + 1077.0, + 295.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1073.0, + 767.0, + 1073.0, + 767.0, + 1108.0, + 294.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1104.0, + 766.0, + 1104.0, + 766.0, + 1139.0, + 295.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1136.0, + 767.0, + 1136.0, + 767.0, + 1168.0, + 295.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1166.0, + 767.0, + 1166.0, + 767.0, + 1201.0, + 294.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1196.0, + 766.0, + 1196.0, + 766.0, + 1231.0, + 295.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1229.0, + 556.0, + 1229.0, + 556.0, + 1262.0, + 293.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 293.0, + 1408.0, + 293.0, + 1408.0, + 329.0, + 293.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 323.0, + 1406.0, + 323.0, + 1406.0, + 360.0, + 293.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 355.0, + 1406.0, + 355.0, + 1406.0, + 391.0, + 293.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 385.0, + 1019.0, + 385.0, + 1019.0, + 420.0, + 295.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1457.0, + 1406.0, + 1457.0, + 1406.0, + 1495.0, + 292.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1492.0, + 501.0, + 1492.0, + 501.0, + 1524.0, + 295.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1492.0, + 1403.0, + 1492.0, + 1403.0, + 1524.0, + 536.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1518.0, + 786.0, + 1518.0, + 786.0, + 1557.0, + 292.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 1518.0, + 1406.0, + 1518.0, + 1406.0, + 1557.0, + 819.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1548.0, + 736.0, + 1548.0, + 736.0, + 1588.0, + 292.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1828.0, + 1198.0, + 1828.0, + 1198.0, + 1871.0, + 292.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 1828.0, + 1409.0, + 1828.0, + 1409.0, + 1871.0, + 1221.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1861.0, + 1081.0, + 1861.0, + 1081.0, + 1901.0, + 292.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1861.0, + 1407.0, + 1861.0, + 1407.0, + 1901.0, + 1102.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1893.0, + 958.0, + 1893.0, + 958.0, + 1930.0, + 295.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1724.0, + 840.0, + 1724.0, + 840.0, + 1758.0, + 295.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1724.0, + 1406.0, + 1724.0, + 1406.0, + 1758.0, + 873.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1756.0, + 478.0, + 1756.0, + 478.0, + 1790.0, + 295.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1756.0, + 1406.0, + 1756.0, + 1406.0, + 1790.0, + 572.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1786.0, + 1407.0, + 1786.0, + 1407.0, + 1820.0, + 295.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1270.0, + 1405.0, + 1270.0, + 1405.0, + 1312.0, + 293.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1305.0, + 1403.0, + 1305.0, + 1403.0, + 1339.0, + 294.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1335.0, + 1035.0, + 1335.0, + 1035.0, + 1371.0, + 293.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1969.0, + 1136.0, + 1969.0, + 1136.0, + 2009.0, + 296.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 1969.0, + 1206.0, + 1969.0, + 1206.0, + 2009.0, + 1159.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1969.0, + 1406.0, + 1969.0, + 1406.0, + 2009.0, + 1240.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 671.0, + 2001.0, + 671.0, + 2039.0, + 293.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 2001.0, + 981.0, + 2001.0, + 981.0, + 2039.0, + 704.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 2001.0, + 1287.0, + 2001.0, + 1287.0, + 2039.0, + 1004.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 2001.0, + 1407.0, + 2001.0, + 1407.0, + 2039.0, + 1330.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1646.0, + 748.0, + 1646.0, + 748.0, + 1682.0, + 295.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1646.0, + 1406.0, + 1646.0, + 1406.0, + 1682.0, + 783.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1677.0, + 1306.0, + 1677.0, + 1306.0, + 1713.0, + 294.0, + 1713.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1006, + 1404, + 1006, + 1404, + 1254, + 298, + 1254 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 297, + 1788, + 1405, + 1788, + 1405, + 1974, + 297, + 1974 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1267, + 1405, + 1267, + 1405, + 1422, + 297, + 1422 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 583, + 1403, + 583, + 1403, + 738, + 298, + 738 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 414, + 1403, + 414, + 1403, + 569, + 298, + 569 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1506, + 1406, + 1506, + 1406, + 1602, + 298, + 1602 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 300, + 307, + 1400, + 307, + 1400, + 400, + 300, + 400 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 295, + 229, + 1402, + 229, + 1402, + 293, + 295, + 293 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 298, + 850, + 1404, + 850, + 1404, + 914, + 298, + 914 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 621, + 1988, + 1081, + 1988, + 1081, + 2041, + 621, + 2041 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 523, + 1740, + 1176, + 1740, + 1176, + 1779, + 523, + 1779 + ], + "score": 0.935 + }, + { + "category_id": 8, + "poly": [ + 504, + 1437, + 1191, + 1437, + 1191, + 1488, + 504, + 1488 + ], + "score": 0.932 + }, + { + "category_id": 9, + "poly": [ + 1355, + 1742, + 1401, + 1742, + 1401, + 1772, + 1355, + 1772 + ], + "score": 0.91 + }, + { + "category_id": 1, + "poly": [ + 298, + 1693, + 1151, + 1693, + 1151, + 1727, + 298, + 1727 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 302, + 783, + 783, + 783, + 783, + 818, + 302, + 818 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 104, + 300, + 104 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 299, + 1636, + 844, + 1636, + 844, + 1668, + 299, + 1668 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1993, + 1401, + 1993, + 1401, + 2024, + 1353, + 2024 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 297, + 951, + 854, + 951, + 854, + 982, + 297, + 982 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1368, + 1442, + 1401, + 1442, + 1401, + 1471, + 1368, + 1471 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.745 + }, + { + "category_id": 14, + "poly": [ + 619, + 1988, + 1079, + 1988, + 1079, + 2043, + 619, + 2043 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\mathcal { T } _ { \\mu } ^ { \\pi } Q ( s , \\pmb { a } ) : = \\underset { n \\geq 0 } { \\operatorname* { m a x } } \\{ ( \\mathcal { T } ^ { \\mu } ) ^ { n } \\mathcal { T } ^ { \\pi } Q ( s , \\pmb { a } ) \\} , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 662, + 1570, + 798, + 1570, + 798, + 1602, + 662, + 1602 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\pmb { a } \\sim \\mu _ { \\theta } ( \\cdot | \\pmb { s } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1231, + 230, + 1333, + 230, + 1333, + 265, + 1231, + 265 + ], + "score": 0.93, + "latex": "Q _ { \\phi } ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 1204, + 1390, + 1274, + 1390, + 1274, + 1424, + 1204, + 1424 + ], + "score": 0.92, + "latex": "\\mu ( \\cdot | s )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1568, + 390, + 1568, + 390, + 1602, + 298, + 1602 + ], + "score": 0.92, + "latex": "\\mathcal { N } ( 0 , \\pmb { I } )" + }, + { + "category_id": 13, + "poly": [ + 953, + 446, + 1036, + 446, + 1036, + 480, + 953, + 480 + ], + "score": 0.92, + "latex": "\\mu _ { \\boldsymbol { \\theta } } ( \\cdot | \\boldsymbol { s } )" + }, + { + "category_id": 13, + "poly": [ + 354, + 507, + 543, + 507, + 543, + 542, + 354, + 542 + ], + "score": 0.91, + "latex": "\\exp { ( \\alpha Q _ { \\phi } ( s , \\pmb { a } ) ) }" + }, + { + "category_id": 13, + "poly": [ + 492, + 371, + 605, + 371, + 605, + 398, + 492, + 398 + ], + "score": 0.91, + "latex": "\\alpha + \\infty" + }, + { + "category_id": 14, + "poly": [ + 506, + 1436, + 1189, + 1436, + 1189, + 1493, + 506, + 1493 + ], + "score": 0.91, + "latex": "\\theta = \\underset { \\theta } { \\arg \\operatorname* { m i n } } \\quad \\mathbb { E } _ { ( s , a ) \\sim D ^ { \\mu } , \\epsilon , t } [ \\| \\sigma _ { t } \\mathbf { s } _ { \\theta } ( \\alpha _ { t } \\pmb { a } + \\sigma _ { t } \\epsilon , s , t ) + \\epsilon \\| _ { 2 } ^ { 2 } ] ," + }, + { + "category_id": 14, + "poly": [ + 521, + 1737, + 1177, + 1737, + 1177, + 1779, + 521, + 1779 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\mathcal { T } ^ { \\pi } Q ( s , a ) = r ( s , a ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } \\sim P ( \\cdot \\mid s , a ) , a ^ { \\prime } \\sim \\pi ( \\cdot \\mid s ^ { \\prime } ) } Q ( s ^ { \\prime } , a ^ { \\prime } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 805, + 340, + 886, + 340, + 886, + 367, + 805, + 367 + ], + "score": 0.9, + "latex": "\\alpha 0" + }, + { + "category_id": 13, + "poly": [ + 689, + 1542, + 883, + 1542, + 883, + 1568, + 689, + 1568 + ], + "score": 0.9, + "latex": "\\mathbf { \\sigma } _ { \\mathbf { } } \\mathbf { a } _ { t } : = \\alpha _ { t } \\mathbf { \\pmb { a } } + \\sigma _ { t } \\mathbf { \\epsilon } \\mathbf { \\epsilon }" + }, + { + "category_id": 13, + "poly": [ + 787, + 477, + 826, + 477, + 826, + 511, + 787, + 511 + ], + "score": 0.89, + "latex": "Q _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 1358, + 1944, + 1396, + 1944, + 1396, + 1980, + 1358, + 1980 + ], + "score": 0.88, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 415, + 1790, + 456, + 1790, + 456, + 1819, + 415, + 1819 + ], + "score": 0.87, + "latex": "\\tau ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 1333, + 1541, + 1405, + 1541, + 1405, + 1568, + 1333, + 1568 + ], + "score": 0.86, + "latex": "a _ { T } \\sim" + }, + { + "category_id": 13, + "poly": [ + 421, + 1544, + 449, + 1544, + 449, + 1568, + 421, + 1568 + ], + "score": 0.85, + "latex": "s _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1262, + 1363, + 1291, + 1363, + 1291, + 1390, + 1262, + 1390 + ], + "score": 0.85, + "latex": "s _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 544, + 1507, + 692, + 1507, + 692, + 1541, + 544, + 1541 + ], + "score": 0.85, + "latex": "t \\sim \\mathcal { U } ( 0 , T )" + }, + { + "category_id": 13, + "poly": [ + 1226, + 1573, + 1254, + 1573, + 1254, + 1600, + 1226, + 1600 + ], + "score": 0.83, + "latex": "s _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 487, + 650, + 517, + 650, + 517, + 676, + 487, + 676 + ], + "score": 0.83, + "latex": "\\pi _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 802, + 1513, + 830, + 1513, + 830, + 1538, + 802, + 1538 + ], + "score": 0.83, + "latex": "\\sigma _ { t }" + }, + { + "category_id": 13, + "poly": [ + 713, + 1511, + 745, + 1511, + 745, + 1538, + 713, + 1538 + ], + "score": 0.82, + "latex": "\\alpha _ { t }" + }, + { + "category_id": 13, + "poly": [ + 376, + 1507, + 530, + 1507, + 530, + 1540, + 376, + 1540 + ], + "score": 0.81, + "latex": "\\epsilon \\sim \\mathcal { N } ( 0 , I )" + }, + { + "category_id": 13, + "poly": [ + 694, + 314, + 716, + 314, + 716, + 336, + 694, + 336 + ], + "score": 0.79, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 1114, + 1855, + 1135, + 1855, + 1135, + 1878, + 1114, + 1878 + ], + "score": 0.78, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 398, + 448, + 431, + 448, + 431, + 474, + 398, + 474 + ], + "score": 0.75, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 800, + 1397, + 820, + 1397, + 820, + 1418, + 800, + 1418 + ], + "score": 0.74, + "latex": "^ { a }" + }, + { + "category_id": 13, + "poly": [ + 1200, + 1543, + 1220, + 1543, + 1220, + 1565, + 1200, + 1565 + ], + "score": 0.74, + "latex": "^ { a }" + }, + { + "category_id": 13, + "poly": [ + 1244, + 478, + 1277, + 478, + 1277, + 505, + 1244, + 505 + ], + "score": 0.7, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 555, + 423, + 574, + 423, + 574, + 443, + 555, + 443 + ], + "score": 0.7, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1328, + 422, + 1345, + 422, + 1345, + 443, + 1328, + 443 + ], + "score": 0.67, + "latex": "\\pmb { s }" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 780.0, + 787.0, + 780.0, + 787.0, + 823.0, + 293.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1634.0, + 848.0, + 1634.0, + 848.0, + 1674.0, + 293.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 949.0, + 855.0, + 949.0, + 855.0, + 988.0, + 294.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1006.0, + 1404.0, + 1006.0, + 1404.0, + 1044.0, + 292.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1039.0, + 1405.0, + 1039.0, + 1405.0, + 1073.0, + 294.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1068.0, + 1405.0, + 1068.0, + 1405.0, + 1105.0, + 293.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1100.0, + 1405.0, + 1100.0, + 1405.0, + 1134.0, + 294.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1129.0, + 1406.0, + 1129.0, + 1406.0, + 1163.0, + 294.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1163.0, + 1406.0, + 1163.0, + 1406.0, + 1193.0, + 296.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1190.0, + 1406.0, + 1190.0, + 1406.0, + 1224.0, + 293.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1222.0, + 1362.0, + 1222.0, + 1362.0, + 1256.0, + 296.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1789.0, + 414.0, + 1789.0, + 414.0, + 1825.0, + 295.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1789.0, + 1405.0, + 1789.0, + 1405.0, + 1825.0, + 457.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1820.0, + 1406.0, + 1820.0, + 1406.0, + 1853.0, + 294.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1850.0, + 1113.0, + 1850.0, + 1113.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1850.0, + 1405.0, + 1850.0, + 1405.0, + 1886.0, + 1136.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1881.0, + 1405.0, + 1881.0, + 1405.0, + 1917.0, + 295.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1912.0, + 1403.0, + 1912.0, + 1403.0, + 1944.0, + 294.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1937.0, + 1357.0, + 1937.0, + 1357.0, + 1986.0, + 291.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1937.0, + 1410.0, + 1937.0, + 1410.0, + 1986.0, + 1397.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1269.0, + 1406.0, + 1269.0, + 1406.0, + 1302.0, + 296.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1295.0, + 1406.0, + 1295.0, + 1406.0, + 1334.0, + 294.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1329.0, + 1408.0, + 1329.0, + 1408.0, + 1366.0, + 294.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1358.0, + 1261.0, + 1358.0, + 1261.0, + 1395.0, + 295.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 1358.0, + 1407.0, + 1358.0, + 1407.0, + 1395.0, + 1292.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1386.0, + 799.0, + 1386.0, + 799.0, + 1430.0, + 292.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1386.0, + 1203.0, + 1386.0, + 1203.0, + 1430.0, + 821.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 1386.0, + 1289.0, + 1386.0, + 1289.0, + 1430.0, + 1275.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 585.0, + 1404.0, + 585.0, + 1404.0, + 618.0, + 296.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 614.0, + 1404.0, + 614.0, + 1404.0, + 647.0, + 294.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 645.0, + 486.0, + 645.0, + 486.0, + 682.0, + 294.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 645.0, + 1405.0, + 645.0, + 1405.0, + 682.0, + 518.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 678.0, + 1404.0, + 678.0, + 1404.0, + 711.0, + 296.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 706.0, + 1024.0, + 706.0, + 1024.0, + 743.0, + 294.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 414.0, + 554.0, + 414.0, + 554.0, + 448.0, + 296.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 414.0, + 1327.0, + 414.0, + 1327.0, + 448.0, + 575.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 414.0, + 1404.0, + 414.0, + 1404.0, + 448.0, + 1346.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 445.0, + 397.0, + 445.0, + 397.0, + 482.0, + 293.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 445.0, + 952.0, + 445.0, + 952.0, + 482.0, + 432.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 445.0, + 1405.0, + 445.0, + 1405.0, + 482.0, + 1037.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 476.0, + 786.0, + 476.0, + 786.0, + 513.0, + 294.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 476.0, + 1243.0, + 476.0, + 1243.0, + 513.0, + 827.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 476.0, + 1405.0, + 476.0, + 1405.0, + 513.0, + 1278.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 506.0, + 353.0, + 506.0, + 353.0, + 543.0, + 294.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 506.0, + 1404.0, + 506.0, + 1404.0, + 543.0, + 544.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 534.0, + 1340.0, + 534.0, + 1340.0, + 575.0, + 293.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1504.0, + 375.0, + 1504.0, + 375.0, + 1544.0, + 292.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1504.0, + 543.0, + 1504.0, + 543.0, + 1544.0, + 531.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1504.0, + 712.0, + 1504.0, + 712.0, + 1544.0, + 693.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1504.0, + 801.0, + 1504.0, + 801.0, + 1544.0, + 746.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1504.0, + 1409.0, + 1504.0, + 1409.0, + 1544.0, + 831.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1534.0, + 420.0, + 1534.0, + 420.0, + 1574.0, + 293.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1534.0, + 688.0, + 1534.0, + 688.0, + 1574.0, + 450.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 884.0, + 1534.0, + 1199.0, + 1534.0, + 1199.0, + 1574.0, + 884.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 1534.0, + 1332.0, + 1534.0, + 1332.0, + 1574.0, + 1221.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1565.0, + 297.0, + 1565.0, + 297.0, + 1604.0, + 292.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1565.0, + 661.0, + 1565.0, + 661.0, + 1604.0, + 391.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 1565.0, + 1225.0, + 1565.0, + 1225.0, + 1604.0, + 799.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1565.0, + 1408.0, + 1565.0, + 1408.0, + 1604.0, + 1255.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 309.0, + 693.0, + 309.0, + 693.0, + 343.0, + 295.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 309.0, + 1404.0, + 309.0, + 1404.0, + 343.0, + 717.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 338.0, + 804.0, + 338.0, + 804.0, + 373.0, + 294.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 338.0, + 1406.0, + 338.0, + 1406.0, + 373.0, + 887.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 368.0, + 491.0, + 368.0, + 491.0, + 405.0, + 295.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 368.0, + 1100.0, + 368.0, + 1100.0, + 405.0, + 606.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 1230.0, + 228.0, + 1230.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 228.0, + 1405.0, + 228.0, + 1405.0, + 267.0, + 1334.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 261.0, + 1027.0, + 261.0, + 1027.0, + 293.0, + 297.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 850.0, + 1406.0, + 850.0, + 1406.0, + 886.0, + 294.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 882.0, + 1301.0, + 882.0, + 1301.0, + 917.0, + 294.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1689.0, + 1151.0, + 1689.0, + 1151.0, + 1734.0, + 295.0, + 1734.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1498, + 1403, + 1498, + 1403, + 1926, + 298, + 1926 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 298, + 1055, + 1404, + 1055, + 1404, + 1248, + 298, + 1248 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1262, + 1405, + 1262, + 1405, + 1386, + 298, + 1386 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 300, + 1942, + 1402, + 1942, + 1402, + 2034, + 300, + 2034 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 305, + 231, + 1403, + 231, + 1403, + 503, + 305, + 503 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 299, + 689, + 1405, + 689, + 1405, + 784, + 299, + 784 + ], + "score": 0.969 + }, + { + "category_id": 4, + "poly": [ + 298, + 531, + 1407, + 531, + 1407, + 631, + 298, + 631 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 297, + 800, + 1401, + 800, + 1401, + 864, + 297, + 864 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 587, + 879, + 1000, + 879, + 1000, + 926, + 587, + 926 + ], + "score": 0.922 + }, + { + "category_id": 8, + "poly": [ + 513, + 977, + 1209, + 977, + 1209, + 1036, + 513, + 1036 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 301, + 1428, + 587, + 1428, + 587, + 1464, + 301, + 1464 + ], + "score": 0.912 + }, + { + "category_id": 8, + "poly": [ + 490, + 931, + 961, + 931, + 961, + 972, + 490, + 972 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.885 + }, + { + "category_id": 9, + "poly": [ + 1354, + 889, + 1401, + 889, + 1401, + 919, + 1354, + 919 + ], + "score": 0.884 + }, + { + "category_id": 9, + "poly": [ + 1354, + 936, + 1401, + 936, + 1401, + 966, + 1354, + 966 + ], + "score": 0.878 + }, + { + "category_id": 9, + "poly": [ + 1355, + 983, + 1401, + 983, + 1401, + 1013, + 1355, + 1013 + ], + "score": 0.875 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.774 + }, + { + "category_id": 14, + "poly": [ + 487, + 876, + 1212, + 876, + 1212, + 1040, + 487, + 1040 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { R _ { n } ^ { ( k ) } = r _ { n } + \\gamma \\operatorname* { m a x } ( R _ { n + 1 } ^ { ( k ) } , V _ { n + 1 } ^ { ( k - 1 ) } ) , } \\\\ { \\mathrm { w h e r e } \\quad V _ { n } ^ { ( k - 1 ) } : = \\mathbb { E } _ { a \\sim \\pi ( \\cdot \\vert s _ { n } ) } Q _ { \\phi } ( s _ { n } , a ) , } \\\\ { \\mathrm { a n d } \\quad \\phi = \\underset { \\phi } { \\arg \\operatorname* { m i n } } \\quad \\mathbb { E } _ { ( s _ { n } , a _ { n } ) \\sim \\mathcal { D } ^ { \\mu } } \\Vert Q _ { \\phi } ( s _ { n } , a _ { n } ) - R _ { n } ^ { ( k - 1 ) } \\Vert _ { 2 } ^ { 2 } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 379, + 1062, + 550, + 1062, + 550, + 1096, + 379, + 1096 + ], + "score": 0.93, + "latex": "k \\in \\{ 1 , 2 , \\dots \\}" + }, + { + "category_id": 13, + "poly": [ + 949, + 1055, + 1002, + 1055, + 1002, + 1093, + 949, + 1093 + ], + "score": 0.92, + "latex": "{ \\cal R } _ { n } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 984, + 530, + 1039, + 530, + 1039, + 568, + 984, + 568 + ], + "score": 0.92, + "latex": "R _ { n } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 599, + 801, + 835, + 801, + 835, + 835, + 599, + 835 + ], + "score": 0.92, + "latex": "\\mathcal { D } ^ { \\mu } = \\{ ( s _ { n } , a _ { n } , r _ { n } ) \\}" + }, + { + "category_id": 13, + "poly": [ + 1245, + 690, + 1319, + 690, + 1319, + 724, + 1245, + 724 + ], + "score": 0.92, + "latex": "( { \\mathcal { T } } ^ { \\mu } ) ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 556, + 1263, + 595, + 1263, + 595, + 1298, + 556, + 1298 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 584, + 753, + 623, + 753, + 623, + 789, + 584, + 789 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 1270, + 753, + 1309, + 753, + 1309, + 783, + 1270, + 783 + ], + "score": 0.89, + "latex": "Q ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 1264, + 1216, + 1303, + 1216, + 1303, + 1250, + 1264, + 1250 + ], + "score": 0.89, + "latex": "Q _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 1361, + 752, + 1397, + 752, + 1397, + 784, + 1361, + 784 + ], + "score": 0.89, + "latex": "Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 643, + 692, + 683, + 692, + 683, + 727, + 643, + 727 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 789, + 1217, + 827, + 1217, + 827, + 1246, + 789, + 1246 + ], + "score": 0.89, + "latex": "R _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1345, + 1324, + 1397, + 1324, + 1397, + 1354, + 1345, + 1354 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mathrm { v e m } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1354, + 337, + 1354, + 337, + 1391, + 298, + 1391 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 1155, + 1263, + 1208, + 1263, + 1208, + 1293, + 1155, + 1293 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mathrm { v e m } }" + }, + { + "category_id": 13, + "poly": [ + 1146, + 1294, + 1199, + 1294, + 1199, + 1324, + 1146, + 1324 + ], + "score": 0.88, + "latex": "\\mathcal { T } _ { \\mathrm { v e m } }" + }, + { + "category_id": 13, + "poly": [ + 791, + 1354, + 829, + 1354, + 829, + 1382, + 791, + 1382 + ], + "score": 0.87, + "latex": "\\tau ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 867, + 722, + 906, + 722, + 906, + 749, + 867, + 749 + ], + "score": 0.87, + "latex": "\\mathcal { T } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 992, + 1189, + 1012, + 1189, + 1012, + 1217, + 992, + 1217 + ], + "score": 0.83, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 370, + 696, + 391, + 696, + 391, + 723, + 370, + 723 + ], + "score": 0.81, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 1084, + 808, + 1104, + 808, + 1104, + 833, + 1084, + 833 + ], + "score": 0.81, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 1170, + 808, + 1190, + 808, + 1190, + 829, + 1170, + 829 + ], + "score": 0.78, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 921, + 1190, + 942, + 1190, + 942, + 1213, + 921, + 1213 + ], + "score": 0.77, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 510, + 753, + 533, + 753, + 533, + 779, + 510, + 779 + ], + "score": 0.36, + "latex": "\\textrm { C }" + }, + { + "category_id": 13, + "poly": [ + 870, + 538, + 896, + 538, + 896, + 569, + 870, + 569 + ], + "score": 0.34, + "latex": "\\mathrm { Q }" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 230.0, + 1397.0, + 230.0, + 1397.0, + 246.0, + 1378.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 271.0, + 1295.0, + 271.0, + 1295.0, + 282.0, + 1285.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 275.0, + 1398.0, + 275.0, + 1398.0, + 295.0, + 1372.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 313.0, + 341.0, + 313.0, + 341.0, + 324.0, + 332.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 302.0, + 453.0, + 302.0, + 453.0, + 324.0, + 441.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 306.0, + 712.0, + 306.0, + 712.0, + 323.0, + 703.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1371.0, + 324.0, + 1398.0, + 324.0, + 1398.0, + 344.0, + 1371.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 379.0, + 904.0, + 379.0, + 904.0, + 385.0, + 898.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 370.0, + 1162.0, + 370.0, + 1162.0, + 386.0, + 1149.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 372.0, + 1398.0, + 372.0, + 1398.0, + 392.0, + 1372.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 420.0, + 1398.0, + 420.0, + 1398.0, + 439.0, + 1372.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 468.0, + 1398.0, + 468.0, + 1398.0, + 486.0, + 1373.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 479.0, + 486.0, + 479.0, + 486.0, + 504.0, + 367.0, + 504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 477.0, + 748.0, + 477.0, + 748.0, + 505.0, + 627.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 477.0, + 1009.0, + 477.0, + 1009.0, + 505.0, + 888.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 477.0, + 1268.0, + 477.0, + 1268.0, + 505.0, + 1149.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 262.5, + 426.0, + 262.5, + 426.0, + 295.5, + 393.0, + 295.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 261.0, + 946.0, + 261.0, + 946.0, + 296.0, + 914.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.75, + 288.0, + 445.75, + 288.0, + 445.75, + 305.5, + 429.75, + 305.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 527.0, + 869.0, + 527.0, + 869.0, + 581.0, + 284.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 527.0, + 983.0, + 527.0, + 983.0, + 581.0, + 897.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 527.0, + 1404.0, + 527.0, + 1404.0, + 581.0, + 1040.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 566.0, + 1409.0, + 566.0, + 1409.0, + 604.0, + 293.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 600.0, + 1169.0, + 600.0, + 1169.0, + 635.0, + 295.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1426.0, + 593.0, + 1426.0, + 593.0, + 1468.0, + 291.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1499.0, + 1405.0, + 1499.0, + 1405.0, + 1534.0, + 294.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1530.0, + 1404.0, + 1530.0, + 1404.0, + 1562.0, + 294.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1558.0, + 1405.0, + 1558.0, + 1405.0, + 1597.0, + 293.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1588.0, + 1407.0, + 1588.0, + 1407.0, + 1628.0, + 292.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1619.0, + 1405.0, + 1619.0, + 1405.0, + 1657.0, + 293.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1651.0, + 1407.0, + 1651.0, + 1407.0, + 1687.0, + 294.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1680.0, + 1405.0, + 1680.0, + 1405.0, + 1717.0, + 293.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1713.0, + 1405.0, + 1713.0, + 1405.0, + 1749.0, + 294.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1742.0, + 1407.0, + 1742.0, + 1407.0, + 1778.0, + 294.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1772.0, + 1405.0, + 1772.0, + 1405.0, + 1810.0, + 293.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1802.0, + 1407.0, + 1802.0, + 1407.0, + 1841.0, + 293.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1833.0, + 1404.0, + 1833.0, + 1404.0, + 1869.0, + 294.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1867.0, + 1404.0, + 1867.0, + 1404.0, + 1899.0, + 294.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1896.0, + 955.0, + 1896.0, + 955.0, + 1930.0, + 293.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1044.0, + 378.0, + 1044.0, + 378.0, + 1110.0, + 286.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1044.0, + 948.0, + 1044.0, + 948.0, + 1110.0, + 551.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1044.0, + 1409.0, + 1044.0, + 1409.0, + 1110.0, + 1003.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1093.0, + 1405.0, + 1093.0, + 1405.0, + 1129.0, + 294.0, + 1129.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1123.0, + 1408.0, + 1123.0, + 1408.0, + 1159.0, + 294.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1151.0, + 1404.0, + 1151.0, + 1404.0, + 1191.0, + 293.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1182.0, + 920.0, + 1182.0, + 920.0, + 1221.0, + 293.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1182.0, + 991.0, + 1182.0, + 991.0, + 1221.0, + 943.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 1182.0, + 1406.0, + 1182.0, + 1406.0, + 1221.0, + 1013.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1216.0, + 788.0, + 1216.0, + 788.0, + 1252.0, + 294.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1216.0, + 1263.0, + 1216.0, + 1263.0, + 1252.0, + 828.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 1216.0, + 1315.0, + 1216.0, + 1315.0, + 1252.0, + 1304.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1260.0, + 555.0, + 1260.0, + 555.0, + 1300.0, + 292.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1260.0, + 1154.0, + 1260.0, + 1154.0, + 1300.0, + 596.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 1260.0, + 1407.0, + 1260.0, + 1407.0, + 1300.0, + 1209.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1290.0, + 1145.0, + 1290.0, + 1145.0, + 1329.0, + 292.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1290.0, + 1406.0, + 1290.0, + 1406.0, + 1329.0, + 1200.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1319.0, + 1344.0, + 1319.0, + 1344.0, + 1363.0, + 291.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1319.0, + 1410.0, + 1319.0, + 1410.0, + 1363.0, + 1398.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1351.0, + 790.0, + 1351.0, + 790.0, + 1390.0, + 338.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1351.0, + 1411.0, + 1351.0, + 1411.0, + 1390.0, + 830.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1942.0, + 1406.0, + 1942.0, + 1406.0, + 1980.0, + 295.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2010.0, + 295.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2004.0, + 1406.0, + 2004.0, + 1406.0, + 2037.0, + 296.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 688.0, + 369.0, + 688.0, + 369.0, + 726.0, + 293.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 688.0, + 642.0, + 688.0, + 642.0, + 726.0, + 392.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 688.0, + 1244.0, + 688.0, + 1244.0, + 726.0, + 684.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 688.0, + 1406.0, + 688.0, + 1406.0, + 726.0, + 1320.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 721.0, + 866.0, + 721.0, + 866.0, + 755.0, + 294.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 721.0, + 1405.0, + 721.0, + 1405.0, + 755.0, + 907.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 751.0, + 509.0, + 751.0, + 509.0, + 789.0, + 294.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 751.0, + 583.0, + 751.0, + 583.0, + 789.0, + 534.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 751.0, + 1269.0, + 751.0, + 1269.0, + 789.0, + 624.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 751.0, + 1360.0, + 751.0, + 1360.0, + 789.0, + 1310.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 751.0, + 1407.0, + 751.0, + 1407.0, + 789.0, + 1398.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 796.0, + 598.0, + 796.0, + 598.0, + 838.0, + 293.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 796.0, + 1083.0, + 796.0, + 1083.0, + 838.0, + 836.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 796.0, + 1169.0, + 796.0, + 1169.0, + 838.0, + 1105.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 796.0, + 1406.0, + 796.0, + 1406.0, + 838.0, + 1191.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 829.0, + 1407.0, + 829.0, + 1407.0, + 868.0, + 293.0, + 868.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 294, + 224, + 1400, + 224, + 1400, + 852, + 294, + 852 + ], + "score": 0.987, + "html": "
DatasetEnvironmentSfBC (Ours)IQLVEMAWRBAILBCQCQLDTDiffuser
Medium-ExpertHalfCheetah92.6±0.586.7-52.772.264.762.486.879.8
Medium-ExpertHopper108.6 ± 2.191.5-27.1106.2100.998.7107.6107.2
Medium-ExpertWalker109.8± 0.2109.6-53.8107.257.5111.0108.1108.4
MediumHalfCheetah45.9 ± 2.247.447.437.430.040.744.442.644.2
MediumHopper57.1 ± 4.166.356.635.962.254.558.067.658.5
MediumWalker77.9 ± 2.578.374.017.473.453.179.274.079.7
Medium-ReplayHalfCheetah37.1 ± 1.744.2140.340.338.246.236.642.2
Medium-ReplayHopper86.2 ±9.194.7128.494.733.148.682.796.8
Medium-ReplayWalker65.1±5.673.9-15.558.815.026.766.661.2
Average (Locomotion)75.676.9-34.371.651.963.974.775.3
DefaultAntMaze-umaze92.0±2.187.587.556.085.078.974.059.2-
DiverseAntMaze-umaze85.3±3.662.278.070.376.755.084.053.0-
PlayAntMaze-medium81.3± 2.671.278.00.015.00.061.20.0-
DiverseAntMaze-medium82.0±3.170.077.00.023.30.053.70.0-
PlayAntMaze-large59.3±14.339.657.00.00.06.715.80.0-
DiverseAntMaze-large45.5 ± 6.647.558.00.08.32.214.90.0-
Average (AntMaze)74.263.072.621.046.723.850.618.7-
Average (Maze2d)74.050.0-10.8-9.17.7-119.5
Average (FrankaKitchen)57.153.3-8.7-11.748.2--
Both-sideBidirectional-Car100.0±0.015.70.00.052.088.042.333.3-
Single-sideBidirectional-Car100.0±0.0100.0100.096.3100.0100.0100.0100.01
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1295, + 1406, + 1295, + 1406, + 1754, + 297, + 1754 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1034, + 1405, + 1034, + 1405, + 1280, + 298, + 1280 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 300, + 1942, + 1402, + 1942, + 1402, + 2034, + 300, + 2034 + ], + "score": 0.972 + }, + { + "category_id": 0, + "poly": [ + 300, + 1880, + 836, + 1880, + 836, + 1912, + 300, + 1912 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 814, + 76, + 814, + 104, + 300, + 104 + ], + "score": 0.895 + }, + { + "category_id": 0, + "poly": [ + 299, + 1806, + 557, + 1806, + 557, + 1842, + 299, + 1842 + ], + "score": 0.894 + }, + { + "category_id": 6, + "poly": [ + 296, + 875, + 1405, + 875, + 1405, + 969, + 296, + 969 + ], + "score": 0.8 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 857, + 2088, + 857, + 2111, + 842, + 2111 + ], + "score": 0.722 + }, + { + "category_id": 1, + "poly": [ + 296, + 875, + 1405, + 875, + 1405, + 969, + 296, + 969 + ], + "score": 0.179 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1880.0, + 840.0, + 1880.0, + 840.0, + 1916.0, + 296.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1804.0, + 561.0, + 1804.0, + 561.0, + 1847.0, + 293.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 872.0, + 1403.0, + 872.0, + 1403.0, + 911.0, + 293.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 906.0, + 1405.0, + 906.0, + 1405.0, + 940.0, + 295.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 938.0, + 1323.0, + 938.0, + 1323.0, + 972.0, + 295.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1295.0, + 1406.0, + 1295.0, + 1406.0, + 1330.0, + 296.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1326.0, + 1407.0, + 1326.0, + 1407.0, + 1360.0, + 292.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1354.0, + 1406.0, + 1354.0, + 1406.0, + 1391.0, + 294.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1387.0, + 1406.0, + 1387.0, + 1406.0, + 1422.0, + 294.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1417.0, + 1407.0, + 1417.0, + 1407.0, + 1452.0, + 292.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1448.0, + 1408.0, + 1448.0, + 1408.0, + 1483.0, + 294.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1477.0, + 1406.0, + 1477.0, + 1406.0, + 1513.0, + 294.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1508.0, + 1407.0, + 1508.0, + 1407.0, + 1543.0, + 295.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1536.0, + 1407.0, + 1536.0, + 1407.0, + 1577.0, + 292.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1571.0, + 1406.0, + 1571.0, + 1406.0, + 1604.0, + 294.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1601.0, + 1404.0, + 1601.0, + 1404.0, + 1636.0, + 295.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1627.0, + 1406.0, + 1627.0, + 1406.0, + 1669.0, + 292.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1662.0, + 1404.0, + 1662.0, + 1404.0, + 1697.0, + 295.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1689.0, + 1407.0, + 1689.0, + 1407.0, + 1729.0, + 291.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1720.0, + 1335.0, + 1720.0, + 1335.0, + 1760.0, + 292.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1036.0, + 1406.0, + 1036.0, + 1406.0, + 1069.0, + 295.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1066.0, + 1403.0, + 1066.0, + 1403.0, + 1099.0, + 295.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1096.0, + 1407.0, + 1096.0, + 1407.0, + 1133.0, + 293.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1125.0, + 1406.0, + 1125.0, + 1406.0, + 1165.0, + 292.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1158.0, + 1405.0, + 1158.0, + 1405.0, + 1192.0, + 295.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1185.0, + 1408.0, + 1185.0, + 1408.0, + 1224.0, + 292.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1217.0, + 1405.0, + 1217.0, + 1405.0, + 1256.0, + 293.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1250.0, + 1397.0, + 1250.0, + 1397.0, + 1283.0, + 295.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1974.0, + 1404.0, + 1974.0, + 1404.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 1406.0, + 2002.0, + 1406.0, + 2038.0, + 295.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 872.0, + 1403.0, + 872.0, + 1403.0, + 911.0, + 293.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 906.0, + 1405.0, + 906.0, + 1405.0, + 940.0, + 295.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 938.0, + 1323.0, + 938.0, + 1323.0, + 972.0, + 295.0, + 972.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1728, + 1405, + 1728, + 1405, + 2034, + 298, + 2034 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 825, + 1404, + 825, + 1404, + 1132, + 298, + 1132 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1147, + 1405, + 1147, + 1405, + 1394, + 298, + 1394 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1528, + 1404, + 1528, + 1404, + 1713, + 298, + 1713 + ], + "score": 0.978 + }, + { + "category_id": 3, + "poly": [ + 298, + 221, + 1400, + 221, + 1400, + 620, + 298, + 620 + ], + "score": 0.974 + }, + { + "category_id": 4, + "poly": [ + 297, + 648, + 1407, + 648, + 1407, + 742, + 297, + 742 + ], + "score": 0.962 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.884 + }, + { + "category_id": 0, + "poly": [ + 302, + 1460, + 828, + 1460, + 828, + 1490, + 302, + 1490 + ], + "score": 0.809 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2111, + 841, + 2111 + ], + "score": 0.792 + }, + { + "category_id": 13, + "poly": [ + 1367, + 950, + 1403, + 950, + 1403, + 979, + 1367, + 979 + ], + "score": 0.38, + "latex": "\\operatorname { F u }" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 226.0, + 526.0, + 226.0, + 526.0, + 252.0, + 458.0, + 252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 226.0, + 865.0, + 226.0, + 865.0, + 252.0, + 798.0, + 252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 224.0, + 1205.0, + 224.0, + 1205.0, + 253.0, + 1137.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 244.0, + 348.0, + 244.0, + 348.0, + 259.0, + 329.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 244.0, + 423.0, + 244.0, + 423.0, + 261.0, + 407.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 246.0, + 500.0, + 246.0, + 500.0, + 257.0, + 487.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 244.0, + 579.0, + 244.0, + 579.0, + 261.0, + 564.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 244.0, + 659.0, + 244.0, + 659.0, + 259.0, + 639.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 242.0, + 689.0, + 242.0, + 689.0, + 261.0, + 661.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 244.0, + 763.0, + 244.0, + 763.0, + 261.0, + 745.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 246.0, + 838.0, + 246.0, + 838.0, + 257.0, + 826.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 245.0, + 917.0, + 245.0, + 917.0, + 258.0, + 905.0, + 258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 244.0, + 997.0, + 244.0, + 997.0, + 259.0, + 979.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 242.0, + 1029.0, + 242.0, + 1029.0, + 261.0, + 1002.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 244.0, + 1102.0, + 244.0, + 1102.0, + 259.0, + 1086.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 246.0, + 1177.0, + 246.0, + 1177.0, + 257.0, + 1167.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 245.0, + 1255.0, + 245.0, + 1255.0, + 257.0, + 1245.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 245.0, + 1335.0, + 245.0, + 1335.0, + 259.0, + 1317.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1371.0, + 259.0, + 1391.0, + 259.0, + 1391.0, + 274.0, + 1371.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 311.0, + 343.0, + 311.0, + 343.0, + 375.0, + 296.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 320.0, + 332.0, + 320.0, + 332.0, + 332.0, + 321.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 437.0, + 334.0, + 437.0, + 334.0, + 448.0, + 323.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 419.0, + 543.0, + 419.0, + 543.0, + 444.0, + 441.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 419.0, + 883.0, + 419.0, + 883.0, + 445.0, + 783.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 419.0, + 1196.0, + 419.0, + 1196.0, + 445.0, + 1144.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 379.0, + 1395.0, + 379.0, + 1395.0, + 479.0, + 1370.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 488.0, + 343.0, + 488.0, + 343.0, + 552.0, + 296.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 496.0, + 334.0, + 496.0, + 334.0, + 513.0, + 319.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 582.0, + 829.0, + 582.0, + 829.0, + 588.0, + 823.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 596.0, + 518.0, + 596.0, + 518.0, + 622.0, + 466.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 593.0, + 856.0, + 593.0, + 856.0, + 624.0, + 808.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 596.0, + 1195.0, + 596.0, + 1195.0, + 623.0, + 1147.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 588.0, + 1397.0, + 588.0, + 1397.0, + 602.0, + 1372.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 647.0, + 1407.0, + 647.0, + 1407.0, + 682.0, + 294.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 679.0, + 1408.0, + 679.0, + 1408.0, + 713.0, + 292.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 706.0, + 1390.0, + 706.0, + 1390.0, + 746.0, + 294.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1459.0, + 833.0, + 1459.0, + 833.0, + 1494.0, + 296.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1726.0, + 1405.0, + 1726.0, + 1405.0, + 1763.0, + 293.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1755.0, + 1406.0, + 1755.0, + 1406.0, + 1797.0, + 293.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1787.0, + 1403.0, + 1787.0, + 1403.0, + 1824.0, + 295.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1821.0, + 1406.0, + 1821.0, + 1406.0, + 1853.0, + 296.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1851.0, + 1408.0, + 1851.0, + 1408.0, + 1886.0, + 295.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1881.0, + 1407.0, + 1881.0, + 1407.0, + 1917.0, + 293.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1908.0, + 1407.0, + 1908.0, + 1407.0, + 1948.0, + 291.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1942.0, + 1406.0, + 1942.0, + 1406.0, + 1979.0, + 292.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2008.0, + 295.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2004.0, + 1078.0, + 2004.0, + 1078.0, + 2036.0, + 293.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 826.0, + 1404.0, + 826.0, + 1404.0, + 861.0, + 294.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 858.0, + 1406.0, + 858.0, + 1406.0, + 896.0, + 292.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 887.0, + 1405.0, + 887.0, + 1405.0, + 924.0, + 293.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 919.0, + 1405.0, + 919.0, + 1405.0, + 954.0, + 294.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 949.0, + 1366.0, + 949.0, + 1366.0, + 985.0, + 294.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 980.0, + 1408.0, + 980.0, + 1408.0, + 1017.0, + 292.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1010.0, + 1405.0, + 1010.0, + 1405.0, + 1046.0, + 293.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1043.0, + 1405.0, + 1043.0, + 1405.0, + 1075.0, + 294.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1071.0, + 1404.0, + 1071.0, + 1404.0, + 1107.0, + 294.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1102.0, + 608.0, + 1102.0, + 608.0, + 1135.0, + 293.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1149.0, + 1405.0, + 1149.0, + 1405.0, + 1183.0, + 296.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1179.0, + 1405.0, + 1179.0, + 1405.0, + 1213.0, + 293.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1209.0, + 1405.0, + 1209.0, + 1405.0, + 1245.0, + 293.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1240.0, + 1407.0, + 1240.0, + 1407.0, + 1275.0, + 293.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1271.0, + 1408.0, + 1271.0, + 1408.0, + 1304.0, + 293.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1301.0, + 1406.0, + 1301.0, + 1406.0, + 1338.0, + 293.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1331.0, + 1406.0, + 1331.0, + 1406.0, + 1366.0, + 292.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1364.0, + 1131.0, + 1364.0, + 1131.0, + 1395.0, + 295.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1528.0, + 1405.0, + 1528.0, + 1405.0, + 1565.0, + 293.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1562.0, + 1402.0, + 1562.0, + 1402.0, + 1594.0, + 296.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1586.0, + 1405.0, + 1586.0, + 1405.0, + 1629.0, + 291.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1618.0, + 1405.0, + 1618.0, + 1405.0, + 1656.0, + 293.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1650.0, + 1406.0, + 1650.0, + 1406.0, + 1687.0, + 293.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1682.0, + 610.0, + 1682.0, + 610.0, + 1717.0, + 294.0, + 1717.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1169, + 1405, + 1169, + 1405, + 1385, + 297, + 1385 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 830, + 1404, + 830, + 1404, + 1076, + 298, + 1076 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1400, + 1405, + 1400, + 1405, + 1675, + 298, + 1675 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1788, + 1404, + 1788, + 1404, + 2035, + 299, + 2035 + ], + "score": 0.979 + }, + { + "category_id": 3, + "poly": [ + 298, + 226, + 1398, + 226, + 1398, + 644, + 298, + 644 + ], + "score": 0.968 + }, + { + "category_id": 4, + "poly": [ + 297, + 670, + 1404, + 670, + 1404, + 772, + 297, + 772 + ], + "score": 0.957 + }, + { + "category_id": 0, + "poly": [ + 300, + 1113, + 602, + 1113, + 602, + 1144, + 300, + 1144 + ], + "score": 0.899 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 300, + 1720, + 543, + 1720, + 543, + 1755, + 300, + 1755 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.794 + }, + { + "category_id": 13, + "poly": [ + 298, + 1583, + 447, + 1583, + 447, + 1616, + 298, + 1616 + ], + "score": 0.92, + "latex": "\\{ 1 , 2 , 3 , 4 , 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 983, + 701, + 1036, + 701, + 1036, + 739, + 983, + 739 + ], + "score": 0.92, + "latex": "{ \\cal R } _ { n } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 405, + 709, + 485, + 709, + 485, + 738, + 405, + 738 + ], + "score": 0.9, + "latex": "K = 1" + }, + { + "category_id": 13, + "poly": [ + 1228, + 1554, + 1257, + 1554, + 1257, + 1580, + 1228, + 1580 + ], + "score": 0.82, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 936, + 673, + 965, + 673, + 965, + 699, + 936, + 699 + ], + "score": 0.82, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 909, + 1523, + 938, + 1523, + 938, + 1550, + 909, + 1550 + ], + "score": 0.79, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1339, + 576, + 1380, + 576, + 1380, + 596, + 1339, + 596 + ], + "score": 0.33, + "latex": "\\mathsf { K } = 5" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 227.0, + 925.0, + 227.0, + 925.0, + 260.0, + 824.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 247.0, + 439.0, + 247.0, + 439.0, + 410.0, + 408.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 264.0, + 1272.0, + 264.0, + 1272.0, + 291.0, + 1226.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 278.0, + 466.0, + 278.0, + 466.0, + 308.0, + 433.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 278.0, + 900.0, + 278.0, + 900.0, + 288.0, + 889.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 287.0, + 1274.0, + 287.0, + 1274.0, + 318.0, + 1226.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 315.0, + 466.0, + 315.0, + 466.0, + 347.0, + 432.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 312.0, + 1275.0, + 312.0, + 1275.0, + 342.0, + 1224.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 338.0, + 1273.0, + 338.0, + 1273.0, + 364.0, + 1227.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 350.0, + 468.0, + 350.0, + 468.0, + 385.0, + 433.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 363.0, + 1273.0, + 363.0, + 1273.0, + 390.0, + 1227.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 396.0, + 462.0, + 396.0, + 462.0, + 415.0, + 447.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 411.0, + 581.0, + 411.0, + 581.0, + 433.0, + 515.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 406.0, + 710.0, + 406.0, + 710.0, + 434.0, + 617.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 405.0, + 815.0, + 405.0, + 815.0, + 438.0, + 744.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 409.0, + 942.0, + 409.0, + 942.0, + 433.0, + 848.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 409.0, + 1055.0, + 409.0, + 1055.0, + 433.0, + 967.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 408.0, + 1158.0, + 408.0, + 1158.0, + 436.0, + 1093.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 439.0, + 982.0, + 439.0, + 982.0, + 470.0, + 777.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 461.0, + 364.0, + 461.0, + 364.0, + 619.0, + 301.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 474.0, + 960.0, + 474.0, + 960.0, + 482.0, + 948.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 477.0, + 1383.0, + 477.0, + 1383.0, + 503.0, + 1337.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 517.0, + 409.0, + 517.0, + 409.0, + 527.0, + 397.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 515.0, + 443.0, + 515.0, + 443.0, + 526.0, + 431.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 499.0, + 1386.0, + 499.0, + 1386.0, + 530.0, + 1336.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 526.0, + 411.0, + 526.0, + 411.0, + 540.0, + 394.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 528.0, + 462.0, + 528.0, + 462.0, + 545.0, + 447.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 522.0, + 1386.0, + 522.0, + 1386.0, + 553.0, + 1334.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 542.0, + 364.0, + 542.0, + 364.0, + 566.0, + 334.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 550.0, + 513.0, + 550.0, + 513.0, + 571.0, + 494.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 548.0, + 1386.0, + 548.0, + 1386.0, + 577.0, + 1337.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1381.0, + 571.0, + 1386.0, + 571.0, + 1386.0, + 602.0, + 1381.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 604.0, + 363.0, + 604.0, + 363.0, + 627.0, + 344.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 617.0, + 1287.0, + 617.0, + 1287.0, + 644.0, + 388.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 668.0, + 935.0, + 668.0, + 935.0, + 706.0, + 294.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 668.0, + 1405.0, + 668.0, + 1405.0, + 706.0, + 966.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 692.0, + 404.0, + 692.0, + 404.0, + 755.0, + 286.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 692.0, + 982.0, + 692.0, + 982.0, + 755.0, + 486.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 692.0, + 1405.0, + 692.0, + 1405.0, + 755.0, + 1037.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 740.0, + 1269.0, + 740.0, + 1269.0, + 775.0, + 295.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1108.0, + 607.0, + 1108.0, + 607.0, + 1149.0, + 294.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1714.0, + 549.0, + 1714.0, + 549.0, + 1765.0, + 291.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1171.0, + 1405.0, + 1171.0, + 1405.0, + 1205.0, + 295.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1200.0, + 1407.0, + 1200.0, + 1407.0, + 1236.0, + 294.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1229.0, + 1405.0, + 1229.0, + 1405.0, + 1267.0, + 294.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1262.0, + 1405.0, + 1262.0, + 1405.0, + 1297.0, + 295.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1293.0, + 1405.0, + 1293.0, + 1405.0, + 1328.0, + 294.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1324.0, + 1407.0, + 1324.0, + 1407.0, + 1358.0, + 294.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1350.0, + 1157.0, + 1350.0, + 1157.0, + 1392.0, + 292.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 832.0, + 1405.0, + 832.0, + 1405.0, + 865.0, + 294.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 860.0, + 1405.0, + 860.0, + 1405.0, + 895.0, + 293.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 893.0, + 1405.0, + 893.0, + 1405.0, + 927.0, + 294.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 923.0, + 1407.0, + 923.0, + 1407.0, + 958.0, + 292.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 954.0, + 1404.0, + 954.0, + 1404.0, + 988.0, + 294.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 984.0, + 1404.0, + 984.0, + 1404.0, + 1018.0, + 294.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1013.0, + 1405.0, + 1013.0, + 1405.0, + 1048.0, + 292.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1042.0, + 1180.0, + 1042.0, + 1180.0, + 1083.0, + 292.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1402.0, + 1403.0, + 1402.0, + 1403.0, + 1435.0, + 296.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1432.0, + 1403.0, + 1432.0, + 1403.0, + 1465.0, + 295.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1462.0, + 1402.0, + 1462.0, + 1402.0, + 1495.0, + 296.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1493.0, + 1406.0, + 1493.0, + 1406.0, + 1526.0, + 295.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1522.0, + 908.0, + 1522.0, + 908.0, + 1558.0, + 295.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1522.0, + 1405.0, + 1522.0, + 1405.0, + 1558.0, + 939.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1549.0, + 1227.0, + 1549.0, + 1227.0, + 1590.0, + 292.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1549.0, + 1405.0, + 1549.0, + 1405.0, + 1590.0, + 1258.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1583.0, + 1406.0, + 1583.0, + 1406.0, + 1620.0, + 448.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1613.0, + 1405.0, + 1613.0, + 1405.0, + 1649.0, + 295.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1644.0, + 950.0, + 1644.0, + 950.0, + 1677.0, + 294.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1787.0, + 1405.0, + 1787.0, + 1405.0, + 1826.0, + 292.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1821.0, + 1408.0, + 1821.0, + 1408.0, + 1855.0, + 294.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1850.0, + 1406.0, + 1850.0, + 1406.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1883.0, + 1408.0, + 1883.0, + 1408.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1913.0, + 1405.0, + 1913.0, + 1405.0, + 1946.0, + 295.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1408.0, + 1942.0, + 1408.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2004.0, + 1151.0, + 2004.0, + 1151.0, + 2036.0, + 294.0, + 2036.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 297, + 1403, + 297, + 1403, + 390, + 299, + 390 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 299, + 509, + 1406, + 509, + 1406, + 663, + 299, + 663 + ], + "score": 0.963 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.885 + }, + { + "category_id": 0, + "poly": [ + 301, + 227, + 564, + 227, + 564, + 262, + 301, + 262 + ], + "score": 0.877 + }, + { + "category_id": 1, + "poly": [ + 299, + 764, + 1406, + 764, + 1406, + 856, + 299, + 856 + ], + "score": 0.826 + }, + { + "category_id": 1, + "poly": [ + 300, + 1678, + 1402, + 1678, + 1402, + 1740, + 300, + 1740 + ], + "score": 0.818 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.805 + }, + { + "category_id": 1, + "poly": [ + 292, + 1324, + 1401, + 1324, + 1401, + 1388, + 292, + 1388 + ], + "score": 0.787 + }, + { + "category_id": 1, + "poly": [ + 296, + 1589, + 1403, + 1589, + 1403, + 1652, + 296, + 1652 + ], + "score": 0.783 + }, + { + "category_id": 0, + "poly": [ + 301, + 440, + 609, + 440, + 609, + 472, + 301, + 472 + ], + "score": 0.781 + }, + { + "category_id": 1, + "poly": [ + 295, + 882, + 1401, + 882, + 1401, + 945, + 295, + 945 + ], + "score": 0.78 + }, + { + "category_id": 1, + "poly": [ + 301, + 1087, + 1404, + 1087, + 1404, + 1183, + 301, + 1183 + ], + "score": 0.753 + }, + { + "category_id": 1, + "poly": [ + 295, + 1765, + 1400, + 1765, + 1400, + 1829, + 295, + 1829 + ], + "score": 0.743 + }, + { + "category_id": 1, + "poly": [ + 293, + 1500, + 1401, + 1500, + 1401, + 1565, + 293, + 1565 + ], + "score": 0.732 + }, + { + "category_id": 1, + "poly": [ + 298, + 1205, + 1406, + 1205, + 1406, + 1300, + 298, + 1300 + ], + "score": 0.732 + }, + { + "category_id": 1, + "poly": [ + 301, + 969, + 1405, + 969, + 1405, + 1065, + 301, + 1065 + ], + "score": 0.709 + }, + { + "category_id": 1, + "poly": [ + 292, + 1412, + 1402, + 1412, + 1402, + 1477, + 292, + 1477 + ], + "score": 0.705 + }, + { + "category_id": 1, + "poly": [ + 296, + 1853, + 1401, + 1853, + 1401, + 1917, + 296, + 1917 + ], + "score": 0.701 + }, + { + "category_id": 0, + "poly": [ + 299, + 712, + 488, + 712, + 488, + 745, + 299, + 745 + ], + "score": 0.664 + }, + { + "category_id": 1, + "poly": [ + 301, + 1942, + 1403, + 1942, + 1403, + 2034, + 301, + 2034 + ], + "score": 0.651 + }, + { + "category_id": 1, + "poly": [ + 299, + 712, + 488, + 712, + 488, + 745, + 299, + 745 + ], + "score": 0.224 + }, + { + "category_id": 1, + "poly": [ + 301, + 440, + 609, + 440, + 609, + 472, + 301, + 472 + ], + "score": 0.101 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 569.0, + 229.0, + 569.0, + 269.0, + 295.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 868.0, + 2084.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 440.0, + 613.0, + 440.0, + 613.0, + 479.0, + 297.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 712.0, + 490.0, + 712.0, + 490.0, + 749.0, + 297.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 295.0, + 1404.0, + 295.0, + 1404.0, + 331.0, + 294.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 329.0, + 1405.0, + 329.0, + 1405.0, + 363.0, + 295.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 359.0, + 589.0, + 359.0, + 589.0, + 392.0, + 294.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 509.0, + 1404.0, + 509.0, + 1404.0, + 546.0, + 294.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 538.0, + 1408.0, + 538.0, + 1408.0, + 575.0, + 293.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 565.0, + 1410.0, + 565.0, + 1410.0, + 609.0, + 292.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 598.0, + 1406.0, + 598.0, + 1406.0, + 638.0, + 293.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 628.0, + 962.0, + 628.0, + 962.0, + 669.0, + 294.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 761.0, + 1407.0, + 761.0, + 1407.0, + 799.0, + 296.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 796.0, + 1404.0, + 796.0, + 1404.0, + 829.0, + 322.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 826.0, + 610.0, + 826.0, + 610.0, + 857.0, + 321.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1674.0, + 1404.0, + 1674.0, + 1404.0, + 1715.0, + 296.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1708.0, + 916.0, + 1708.0, + 916.0, + 1743.0, + 323.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1320.0, + 1406.0, + 1320.0, + 1406.0, + 1365.0, + 294.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1355.0, + 1047.0, + 1355.0, + 1047.0, + 1389.0, + 321.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1587.0, + 1406.0, + 1587.0, + 1406.0, + 1626.0, + 294.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1621.0, + 989.0, + 1621.0, + 989.0, + 1654.0, + 321.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 878.0, + 1406.0, + 878.0, + 1406.0, + 921.0, + 293.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 912.0, + 1294.0, + 912.0, + 1294.0, + 947.0, + 322.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1087.0, + 1405.0, + 1087.0, + 1405.0, + 1125.0, + 295.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1122.0, + 1405.0, + 1122.0, + 1405.0, + 1156.0, + 322.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1147.0, + 1366.0, + 1147.0, + 1366.0, + 1188.0, + 321.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1761.0, + 1405.0, + 1761.0, + 1405.0, + 1803.0, + 293.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1796.0, + 1176.0, + 1796.0, + 1176.0, + 1832.0, + 323.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1502.0, + 1403.0, + 1502.0, + 1403.0, + 1538.0, + 295.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1533.0, + 708.0, + 1533.0, + 708.0, + 1566.0, + 320.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1205.0, + 1408.0, + 1205.0, + 1408.0, + 1243.0, + 295.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1239.0, + 1404.0, + 1239.0, + 1404.0, + 1273.0, + 324.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1268.0, + 663.0, + 1268.0, + 663.0, + 1302.0, + 324.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 970.0, + 1407.0, + 970.0, + 1407.0, + 1005.0, + 295.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1000.0, + 1405.0, + 1000.0, + 1405.0, + 1038.0, + 321.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1033.0, + 1140.0, + 1033.0, + 1140.0, + 1067.0, + 324.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1414.0, + 1404.0, + 1414.0, + 1404.0, + 1450.0, + 295.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1443.0, + 987.0, + 1443.0, + 987.0, + 1478.0, + 319.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 1407.0, + 1849.0, + 1407.0, + 1893.0, + 294.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1884.0, + 1176.0, + 1884.0, + 1176.0, + 1918.0, + 318.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1941.0, + 1405.0, + 1941.0, + 1405.0, + 1978.0, + 295.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1971.0, + 1407.0, + 1971.0, + 1407.0, + 2007.0, + 320.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2001.0, + 1317.0, + 2001.0, + 1317.0, + 2038.0, + 322.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 712.0, + 490.0, + 712.0, + 490.0, + 749.0, + 297.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 440.0, + 613.0, + 440.0, + 613.0, + 479.0, + 297.0, + 479.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.882 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1403, + 228, + 1403, + 323, + 298, + 323 + ], + "score": 0.819 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 861, + 2088, + 861, + 2113, + 835, + 2113 + ], + "score": 0.818 + }, + { + "category_id": 1, + "poly": [ + 299, + 650, + 1400, + 650, + 1400, + 716, + 299, + 716 + ], + "score": 0.78 + }, + { + "category_id": 1, + "poly": [ + 291, + 1630, + 1402, + 1630, + 1402, + 1696, + 291, + 1696 + ], + "score": 0.765 + }, + { + "category_id": 1, + "poly": [ + 294, + 569, + 1400, + 569, + 1400, + 634, + 294, + 634 + ], + "score": 0.751 + }, + { + "category_id": 1, + "poly": [ + 295, + 1713, + 1403, + 1713, + 1403, + 1808, + 295, + 1808 + ], + "score": 0.745 + }, + { + "category_id": 1, + "poly": [ + 300, + 454, + 1403, + 454, + 1403, + 549, + 300, + 549 + ], + "score": 0.731 + }, + { + "category_id": 1, + "poly": [ + 303, + 342, + 1400, + 342, + 1400, + 437, + 303, + 437 + ], + "score": 0.724 + }, + { + "category_id": 1, + "poly": [ + 298, + 733, + 1399, + 733, + 1399, + 800, + 298, + 800 + ], + "score": 0.723 + }, + { + "category_id": 1, + "poly": [ + 297, + 1827, + 1403, + 1827, + 1403, + 1921, + 297, + 1921 + ], + "score": 0.723 + }, + { + "category_id": 1, + "poly": [ + 296, + 816, + 1400, + 816, + 1400, + 881, + 296, + 881 + ], + "score": 0.699 + }, + { + "category_id": 1, + "poly": [ + 297, + 1321, + 1401, + 1321, + 1401, + 1416, + 297, + 1416 + ], + "score": 0.696 + }, + { + "category_id": 1, + "poly": [ + 294, + 1238, + 1400, + 1238, + 1400, + 1305, + 294, + 1305 + ], + "score": 0.694 + }, + { + "category_id": 1, + "poly": [ + 290, + 1434, + 1404, + 1434, + 1404, + 1501, + 290, + 1501 + ], + "score": 0.692 + }, + { + "category_id": 1, + "poly": [ + 297, + 1517, + 1403, + 1517, + 1403, + 1612, + 297, + 1612 + ], + "score": 0.682 + }, + { + "category_id": 1, + "poly": [ + 298, + 899, + 1405, + 899, + 1405, + 995, + 298, + 995 + ], + "score": 0.658 + }, + { + "category_id": 1, + "poly": [ + 298, + 1012, + 1406, + 1012, + 1406, + 1107, + 298, + 1107 + ], + "score": 0.652 + }, + { + "category_id": 1, + "poly": [ + 300, + 1941, + 1404, + 1941, + 1404, + 2034, + 300, + 2034 + ], + "score": 0.647 + }, + { + "category_id": 1, + "poly": [ + 297, + 1125, + 1406, + 1125, + 1406, + 1221, + 297, + 1221 + ], + "score": 0.593 + }, + { + "category_id": 13, + "poly": [ + 459, + 1894, + 478, + 1894, + 478, + 1916, + 459, + 1916 + ], + "score": 0.49, + "latex": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 263.0, + 294.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 260.0, + 1408.0, + 260.0, + 1408.0, + 296.0, + 322.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 288.0, + 399.0, + 288.0, + 399.0, + 324.0, + 320.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 651.0, + 1405.0, + 651.0, + 1405.0, + 687.0, + 295.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 682.0, + 844.0, + 682.0, + 844.0, + 717.0, + 319.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1629.0, + 1405.0, + 1629.0, + 1405.0, + 1669.0, + 294.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1663.0, + 748.0, + 1663.0, + 748.0, + 1695.0, + 320.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 569.0, + 1405.0, + 569.0, + 1405.0, + 605.0, + 296.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 600.0, + 1067.0, + 600.0, + 1067.0, + 634.0, + 323.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1713.0, + 1406.0, + 1713.0, + 1406.0, + 1751.0, + 295.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1745.0, + 1404.0, + 1745.0, + 1404.0, + 1782.0, + 322.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1777.0, + 611.0, + 1777.0, + 611.0, + 1807.0, + 322.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 452.0, + 1408.0, + 452.0, + 1408.0, + 493.0, + 293.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 486.0, + 1404.0, + 486.0, + 1404.0, + 522.0, + 321.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 517.0, + 653.0, + 517.0, + 653.0, + 548.0, + 320.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 342.0, + 1404.0, + 342.0, + 1404.0, + 380.0, + 297.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 374.0, + 1405.0, + 374.0, + 1405.0, + 411.0, + 322.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 404.0, + 881.0, + 404.0, + 881.0, + 438.0, + 320.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 733.0, + 1404.0, + 733.0, + 1404.0, + 773.0, + 294.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 767.0, + 1243.0, + 767.0, + 1243.0, + 800.0, + 323.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1825.0, + 1407.0, + 1825.0, + 1407.0, + 1864.0, + 292.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1859.0, + 1405.0, + 1859.0, + 1405.0, + 1893.0, + 324.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1891.0, + 458.0, + 1891.0, + 458.0, + 1921.0, + 325.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1891.0, + 654.0, + 1891.0, + 654.0, + 1921.0, + 479.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 815.0, + 1406.0, + 815.0, + 1406.0, + 854.0, + 292.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 845.0, + 1274.0, + 845.0, + 1274.0, + 883.0, + 320.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1322.0, + 1405.0, + 1322.0, + 1405.0, + 1360.0, + 295.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1353.0, + 1406.0, + 1353.0, + 1406.0, + 1390.0, + 322.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1386.0, + 505.0, + 1386.0, + 505.0, + 1416.0, + 324.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1239.0, + 1405.0, + 1239.0, + 1405.0, + 1275.0, + 296.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1270.0, + 1181.0, + 1270.0, + 1181.0, + 1305.0, + 323.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1431.0, + 1405.0, + 1431.0, + 1405.0, + 1474.0, + 293.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1466.0, + 1044.0, + 1466.0, + 1044.0, + 1501.0, + 321.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1515.0, + 1408.0, + 1515.0, + 1408.0, + 1556.0, + 295.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1549.0, + 1405.0, + 1549.0, + 1405.0, + 1585.0, + 322.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1580.0, + 576.0, + 1580.0, + 576.0, + 1616.0, + 324.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 899.0, + 1407.0, + 899.0, + 1407.0, + 937.0, + 295.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 932.0, + 1406.0, + 932.0, + 1406.0, + 963.0, + 326.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 960.0, + 748.0, + 960.0, + 748.0, + 998.0, + 323.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1010.0, + 1408.0, + 1010.0, + 1408.0, + 1051.0, + 293.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1046.0, + 1403.0, + 1046.0, + 1403.0, + 1077.0, + 324.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1078.0, + 660.0, + 1078.0, + 660.0, + 1108.0, + 320.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1940.0, + 1405.0, + 1940.0, + 1405.0, + 1978.0, + 295.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2007.0, + 325.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 2002.0, + 861.0, + 2002.0, + 861.0, + 2036.0, + 324.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1124.0, + 1406.0, + 1124.0, + 1406.0, + 1164.0, + 294.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1155.0, + 1409.0, + 1155.0, + 1409.0, + 1196.0, + 320.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1189.0, + 622.0, + 1189.0, + 622.0, + 1220.0, + 324.0, + 1220.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2113, + 835, + 2113 + ], + "score": 0.835 + }, + { + "category_id": 1, + "poly": [ + 295, + 1014, + 1403, + 1014, + 1403, + 1079, + 295, + 1079 + ], + "score": 0.793 + }, + { + "category_id": 1, + "poly": [ + 296, + 818, + 1405, + 818, + 1405, + 884, + 296, + 884 + ], + "score": 0.781 + }, + { + "category_id": 1, + "poly": [ + 297, + 901, + 1407, + 901, + 1407, + 996, + 297, + 996 + ], + "score": 0.77 + }, + { + "category_id": 1, + "poly": [ + 294, + 477, + 1406, + 477, + 1406, + 544, + 294, + 544 + ], + "score": 0.758 + }, + { + "category_id": 1, + "poly": [ + 294, + 395, + 1403, + 395, + 1403, + 460, + 294, + 460 + ], + "score": 0.758 + }, + { + "category_id": 1, + "poly": [ + 294, + 1241, + 1404, + 1241, + 1404, + 1306, + 294, + 1306 + ], + "score": 0.755 + }, + { + "category_id": 1, + "poly": [ + 298, + 704, + 1407, + 704, + 1407, + 801, + 298, + 801 + ], + "score": 0.753 + }, + { + "category_id": 1, + "poly": [ + 296, + 1324, + 1400, + 1324, + 1400, + 1419, + 296, + 1419 + ], + "score": 0.749 + }, + { + "category_id": 1, + "poly": [ + 291, + 1437, + 1401, + 1437, + 1401, + 1504, + 291, + 1504 + ], + "score": 0.745 + }, + { + "category_id": 1, + "poly": [ + 293, + 229, + 1402, + 229, + 1402, + 295, + 293, + 295 + ], + "score": 0.729 + }, + { + "category_id": 1, + "poly": [ + 297, + 1097, + 1407, + 1097, + 1407, + 1223, + 297, + 1223 + ], + "score": 0.723 + }, + { + "category_id": 1, + "poly": [ + 297, + 560, + 1408, + 560, + 1408, + 685, + 297, + 685 + ], + "score": 0.72 + }, + { + "category_id": 1, + "poly": [ + 296, + 1696, + 1408, + 1696, + 1408, + 1821, + 296, + 1821 + ], + "score": 0.718 + }, + { + "category_id": 1, + "poly": [ + 296, + 1923, + 1401, + 1923, + 1401, + 1988, + 296, + 1988 + ], + "score": 0.7 + }, + { + "category_id": 1, + "poly": [ + 297, + 312, + 1402, + 312, + 1402, + 378, + 297, + 378 + ], + "score": 0.692 + }, + { + "category_id": 1, + "poly": [ + 298, + 1520, + 1406, + 1520, + 1406, + 1676, + 298, + 1676 + ], + "score": 0.688 + }, + { + "category_id": 1, + "poly": [ + 296, + 1840, + 1402, + 1840, + 1402, + 1904, + 296, + 1904 + ], + "score": 0.469 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1014.0, + 1408.0, + 1014.0, + 1408.0, + 1052.0, + 294.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1045.0, + 946.0, + 1045.0, + 946.0, + 1081.0, + 323.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 820.0, + 1408.0, + 820.0, + 1408.0, + 856.0, + 295.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 850.0, + 1407.0, + 850.0, + 1407.0, + 885.0, + 322.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 901.0, + 1404.0, + 901.0, + 1404.0, + 938.0, + 294.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 930.0, + 1410.0, + 930.0, + 1410.0, + 972.0, + 320.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 964.0, + 664.0, + 964.0, + 664.0, + 996.0, + 321.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 478.0, + 1405.0, + 478.0, + 1405.0, + 510.0, + 297.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 509.0, + 1298.0, + 509.0, + 1298.0, + 544.0, + 321.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 394.0, + 1405.0, + 394.0, + 1405.0, + 434.0, + 294.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 426.0, + 876.0, + 426.0, + 876.0, + 462.0, + 320.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1237.0, + 1409.0, + 1237.0, + 1409.0, + 1281.0, + 293.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1272.0, + 1015.0, + 1272.0, + 1015.0, + 1307.0, + 325.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 706.0, + 1408.0, + 706.0, + 1408.0, + 740.0, + 296.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 738.0, + 1405.0, + 738.0, + 1405.0, + 772.0, + 322.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 766.0, + 1282.0, + 766.0, + 1282.0, + 804.0, + 320.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1324.0, + 1404.0, + 1324.0, + 1404.0, + 1361.0, + 294.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1352.0, + 1405.0, + 1352.0, + 1405.0, + 1395.0, + 320.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1389.0, + 534.0, + 1389.0, + 534.0, + 1419.0, + 321.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1434.0, + 1404.0, + 1434.0, + 1404.0, + 1478.0, + 293.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1469.0, + 1223.0, + 1469.0, + 1223.0, + 1506.0, + 323.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 230.0, + 1405.0, + 230.0, + 1405.0, + 266.0, + 296.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 261.0, + 1273.0, + 261.0, + 1273.0, + 297.0, + 322.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1099.0, + 1404.0, + 1099.0, + 1404.0, + 1132.0, + 297.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1128.0, + 1405.0, + 1128.0, + 1405.0, + 1164.0, + 321.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1156.0, + 1408.0, + 1156.0, + 1408.0, + 1200.0, + 320.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1193.0, + 572.0, + 1193.0, + 572.0, + 1222.0, + 325.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 563.0, + 1405.0, + 563.0, + 1405.0, + 596.0, + 297.0, + 596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 592.0, + 1409.0, + 592.0, + 1409.0, + 628.0, + 321.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 622.0, + 1406.0, + 622.0, + 1406.0, + 659.0, + 322.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 653.0, + 609.0, + 653.0, + 609.0, + 686.0, + 321.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1693.0, + 1408.0, + 1693.0, + 1408.0, + 1735.0, + 293.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1724.0, + 1408.0, + 1724.0, + 1408.0, + 1763.0, + 321.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1760.0, + 1405.0, + 1760.0, + 1405.0, + 1792.0, + 320.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1790.0, + 539.0, + 1790.0, + 539.0, + 1818.0, + 322.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1923.0, + 1405.0, + 1923.0, + 1405.0, + 1959.0, + 297.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1953.0, + 1220.0, + 1953.0, + 1220.0, + 1990.0, + 321.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 313.0, + 1408.0, + 313.0, + 1408.0, + 349.0, + 296.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 344.0, + 1390.0, + 344.0, + 1390.0, + 380.0, + 324.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1521.0, + 1406.0, + 1521.0, + 1406.0, + 1558.0, + 295.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1550.0, + 1408.0, + 1550.0, + 1408.0, + 1589.0, + 320.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1581.0, + 1404.0, + 1581.0, + 1404.0, + 1619.0, + 321.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1611.0, + 1404.0, + 1611.0, + 1404.0, + 1651.0, + 320.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1645.0, + 580.0, + 1645.0, + 580.0, + 1678.0, + 322.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1835.0, + 1408.0, + 1835.0, + 1408.0, + 1880.0, + 295.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1871.0, + 775.0, + 1871.0, + 775.0, + 1905.0, + 323.0, + 1905.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 293, + 1396, + 1399, + 1396, + 1399, + 1808, + 293, + 1808 + ], + "score": 0.968 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2113, + 835, + 2113 + ], + "score": 0.855 + }, + { + "category_id": 0, + "poly": [ + 300, + 950, + 761, + 950, + 761, + 983, + 300, + 983 + ], + "score": 0.818 + }, + { + "category_id": 1, + "poly": [ + 318, + 991, + 1402, + 991, + 1402, + 1345, + 318, + 1345 + ], + "score": 0.801 + }, + { + "category_id": 1, + "poly": [ + 312, + 351, + 1384, + 351, + 1384, + 896, + 312, + 896 + ], + "score": 0.764 + }, + { + "category_id": 0, + "poly": [ + 302, + 227, + 702, + 227, + 702, + 262, + 302, + 262 + ], + "score": 0.641 + }, + { + "category_id": 4, + "poly": [ + 614, + 1834, + 1089, + 1834, + 1089, + 1867, + 614, + 1867 + ], + "score": 0.618 + }, + { + "category_id": 4, + "poly": [ + 614, + 1834, + 1088, + 1834, + 1088, + 1867, + 614, + 1867 + ], + "score": 0.402 + }, + { + "category_id": 6, + "poly": [ + 296, + 308, + 975, + 308, + 975, + 342, + 296, + 342 + ], + "score": 0.259 + }, + { + "category_id": 1, + "poly": [ + 302, + 227, + 702, + 227, + 702, + 262, + 302, + 262 + ], + "score": 0.243 + }, + { + "category_id": 5, + "poly": [ + 312, + 351, + 1384, + 351, + 1384, + 896, + 312, + 896 + ], + "score": 0.232, + "html": "
Initialize the score-based model sθ,the action evaluation model Q for every state-action pair in dataset Dμ
I/Training the behavior model for each gradient step do
Sample B data points (s,a) from Dμ,B Gaussian noises é from N(O,I) and B time t fromU(0,T)
Perturb a according to at :=αta +Ot∈ Updateθ←λsVθ∑[llσts𝑜(at,s,t)+∈l²2]
end for
//Training the action evaluationmodel iteratively
for iteration k =1 to K do
Initialize training parameters Φ of the action evaluation model Q for each gradient step do
Sample Bdata points (s,a,R(k-1)) from Dμ
Update 𝜙 ←-λq∀∑Q(s,a)-R(𝑘-1)|2]
end for
// Update the Q-training targets as in Algorithm 2
R(k)=Planning(Dμ,μe,Q)
end for
" + }, + { + "category_id": 13, + "poly": [ + 1008, + 476, + 1094, + 476, + 1094, + 507, + 1008, + 507 + ], + "score": 0.93, + "latex": "\\mathcal { N } ( 0 , \\pmb { I } )" + }, + { + "category_id": 13, + "poly": [ + 605, + 698, + 762, + 698, + 762, + 748, + 605, + 748 + ], + "score": 0.93, + "latex": "\\Big ( s , \\pmb { a } , R ^ { ( k - 1 ) } \\Big )" + }, + { + "category_id": 13, + "poly": [ + 729, + 384, + 782, + 384, + 782, + 422, + 729, + 422 + ], + "score": 0.92, + "latex": "{ \\cal R } _ { n } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 453, + 1228, + 565, + 1228, + 565, + 1258, + 453, + 1258 + ], + "score": 0.9, + "latex": "n = \\| \\mathcal { D } ^ { \\mu } \\|" + }, + { + "category_id": 13, + "poly": [ + 521, + 1051, + 577, + 1051, + 577, + 1081, + 521, + 1081 + ], + "score": 0.9, + "latex": "\\left\\{ s _ { n } \\right\\}" + }, + { + "category_id": 13, + "poly": [ + 1287, + 476, + 1371, + 476, + 1371, + 507, + 1287, + 507 + ], + "score": 0.9, + "latex": "\\mathcal { U } ( 0 , T )" + }, + { + "category_id": 13, + "poly": [ + 564, + 1113, + 1080, + 1113, + 1080, + 1178, + 564, + 1178 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { V _ { n } = \\sum _ { m } \\left[ \\exp \\left( \\alpha \\hat { R } _ { n } ^ { m } \\right) \\hat { R } _ { n } ^ { m } \\right] / \\sum _ { m } \\exp \\left( \\alpha \\hat { R } _ { n } ^ { m } \\right) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 577, + 477, + 639, + 477, + 639, + 507, + 577, + 507 + ], + "score": 0.89, + "latex": "( s , a )" + }, + { + "category_id": 13, + "poly": [ + 540, + 1082, + 595, + 1082, + 595, + 1112, + 540, + 1112 + ], + "score": 0.89, + "latex": "\\hat { \\pmb { a } } _ { n } ^ { 1 : M }" + }, + { + "category_id": 13, + "poly": [ + 452, + 617, + 516, + 617, + 516, + 641, + 452, + 641 + ], + "score": 0.89, + "latex": "k = 1" + }, + { + "category_id": 13, + "poly": [ + 977, + 644, + 1012, + 644, + 1012, + 674, + 977, + 674 + ], + "score": 0.88, + "latex": "Q _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 655, + 1082, + 746, + 1082, + 746, + 1114, + 655, + 1114 + ], + "score": 0.88, + "latex": "\\mu _ { \\boldsymbol { \\theta } } \\bigl ( \\cdot | \\boldsymbol { s } _ { n } \\bigr )" + }, + { + "category_id": 13, + "poly": [ + 459, + 748, + 916, + 748, + 916, + 781, + 459, + 781 + ], + "score": 0.87, + "latex": "\\begin{array} { r } { \\phi \\phi - \\lambda _ { Q } \\nabla _ { \\phi } \\sum [ \\| Q _ { \\phi } ( \\pmb { s } , \\pmb { a } ) - R ^ { ( k - 1 ) } \\| _ { 2 } ^ { 2 } ] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 989, + 1078, + 1231, + 1078, + 1231, + 1114, + 989, + 1114 + ], + "score": 0.87, + "latex": "\\hat { R } _ { n } ^ { 1 : M } = Q _ { \\phi } ( \\pmb { s } _ { n } , \\hat { \\pmb { a } } _ { n } ^ { 1 : M } )" + }, + { + "category_id": 13, + "poly": [ + 668, + 1310, + 729, + 1310, + 729, + 1342, + 668, + 1342 + ], + "score": 0.86, + "latex": "\\{ R _ { n } \\}" + }, + { + "category_id": 13, + "poly": [ + 1036, + 353, + 1076, + 353, + 1076, + 387, + 1036, + 387 + ], + "score": 0.86, + "latex": "Q _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 714, + 1051, + 751, + 1051, + 751, + 1076, + 714, + 1076 + ], + "score": 0.86, + "latex": "{ \\mathcal { D } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 353, + 1256, + 692, + 1256, + 692, + 1285, + 353, + 1285 + ], + "score": 0.86, + "latex": "R _ { n } = r _ { n } + \\gamma \\operatorname* { m a x } ( R _ { n + 1 } , V _ { n + 1 } )" + }, + { + "category_id": 13, + "poly": [ + 432, + 533, + 811, + 533, + 811, + 564, + 432, + 564 + ], + "score": 0.84, + "latex": "\\theta \\gets \\lambda _ { s } \\nabla _ { \\theta } \\sum [ \\| \\sigma _ { t } \\mathbf { s } _ { \\theta } ( { \\pmb a } _ { t } , { \\pmb s } , t ) + { \\pmb \\epsilon } \\| _ { 2 } ^ { 2 } ]" + }, + { + "category_id": 13, + "poly": [ + 647, + 645, + 665, + 645, + 665, + 672, + 647, + 672 + ], + "score": 0.84, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 585, + 508, + 767, + 508, + 767, + 532, + 585, + 532 + ], + "score": 0.84, + "latex": "\\mathbf { \\sigma } _ { \\mathbf { \\lambda } } \\mathbf { a } _ { t } : = \\alpha _ { t } \\mathbf { \\pmb { a } } + \\sigma _ { t } \\mathbf { \\epsilon } " + }, + { + "category_id": 13, + "poly": [ + 684, + 358, + 712, + 358, + 712, + 383, + 684, + 383 + ], + "score": 0.84, + "latex": "s _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 698, + 478, + 734, + 478, + 734, + 502, + 698, + 502 + ], + "score": 0.83, + "latex": "{ \\mathcal { D } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 353, + 833, + 665, + 833, + 665, + 867, + 353, + 867 + ], + "score": 0.82, + "latex": "R ^ { ( k ) } = \\operatorname { P l a n n i n g } ( { \\mathcal { D } } ^ { \\mu } , \\mu _ { \\theta } , Q _ { \\phi } )" + }, + { + "category_id": 13, + "poly": [ + 590, + 995, + 629, + 995, + 629, + 1021, + 590, + 1021 + ], + "score": 0.82, + "latex": "{ \\mathcal { D } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 1141, + 478, + 1164, + 478, + 1164, + 502, + 1141, + 502 + ], + "score": 0.8, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 1056, + 1260, + 1085, + 1260, + 1085, + 1283, + 1056, + 1283 + ], + "score": 0.78, + "latex": "r _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1189, + 392, + 1229, + 392, + 1229, + 419, + 1189, + 419 + ], + "score": 0.78, + "latex": "{ \\mathcal { D } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 546, + 617, + 572, + 617, + 572, + 641, + 546, + 641 + ], + "score": 0.78, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 745, + 478, + 768, + 478, + 768, + 502, + 745, + 502 + ], + "score": 0.77, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 434, + 478, + 457, + 478, + 457, + 502, + 434, + 502 + ], + "score": 0.76, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 1163, + 1000, + 1195, + 1000, + 1195, + 1026, + 1163, + 1026 + ], + "score": 0.75, + "latex": "\\mu _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 717, + 1260, + 735, + 1260, + 735, + 1280, + 717, + 1280 + ], + "score": 0.75, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 822, + 707, + 859, + 707, + 859, + 733, + 822, + 733 + ], + "score": 0.74, + "latex": "{ \\mathcal { D } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 937, + 482, + 952, + 482, + 952, + 501, + 937, + 501 + ], + "score": 0.74, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 462, + 709, + 485, + 709, + 485, + 733, + 462, + 733 + ], + "score": 0.74, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 433, + 510, + 452, + 510, + 452, + 529, + 433, + 529 + ], + "score": 0.72, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 1217, + 479, + 1231, + 479, + 1231, + 502, + 1217, + 502 + ], + "score": 0.7, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 1361, + 994, + 1400, + 994, + 1400, + 1028, + 1361, + 1028 + ], + "score": 0.45, + "latex": "Q _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 489, + 807, + 510, + 807, + 510, + 833, + 489, + 833 + ], + "score": 0.38, + "latex": "\\mathcal { Q }" + }, + { + "category_id": 13, + "poly": [ + 433, + 1084, + 460, + 1084, + 460, + 1109, + 433, + 1109 + ], + "score": 0.26, + "latex": "\\mathbf { M }" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1416.0, + 464.0, + 1416.0, + 464.0, + 1431.0, + 449.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 1425.0, + 717.0, + 1425.0, + 717.0, + 1448.0, + 640.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 1427.0, + 860.0, + 1427.0, + 860.0, + 1452.0, + 806.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1422.0, + 1121.0, + 1422.0, + 1121.0, + 1454.0, + 925.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 1451.0, + 468.0, + 1451.0, + 468.0, + 1468.0, + 454.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1450.0, + 909.0, + 1450.0, + 909.0, + 1582.0, + 555.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 1450.0, + 1311.0, + 1450.0, + 1311.0, + 1466.0, + 1296.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1501.0, + 477.0, + 1501.0, + 477.0, + 1524.0, + 455.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1501.0, + 1322.0, + 1501.0, + 1322.0, + 1523.0, + 1301.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1692.0, + 648.0, + 1692.0, + 648.0, + 1712.0, + 595.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 1699.0, + 766.0, + 1699.0, + 766.0, + 1728.0, + 687.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 1691.0, + 870.0, + 1691.0, + 870.0, + 1726.0, + 800.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 1746.0, + 763.0, + 1746.0, + 763.0, + 1773.0, + 690.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1745.0, + 1047.0, + 1745.0, + 1047.0, + 1773.0, + 978.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1783.0, + 512.0, + 1783.0, + 512.0, + 1806.0, + 330.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1780.0, + 867.0, + 1780.0, + 867.0, + 1808.0, + 588.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1781.0, + 1092.0, + 1781.0, + 1092.0, + 1805.0, + 925.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1779.0, + 1394.0, + 1779.0, + 1394.0, + 1809.0, + 1140.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1709.5, + 652.0, + 1709.5, + 652.0, + 1728.0, + 628.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 942.0, + 766.0, + 942.0, + 766.0, + 993.0, + 294.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 222.0, + 706.0, + 222.0, + 706.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1832.0, + 1089.0, + 1832.0, + 1089.0, + 1871.0, + 609.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1832.0, + 1088.0, + 1832.0, + 1088.0, + 1871.0, + 609.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 302.0, + 974.0, + 302.0, + 974.0, + 349.0, + 294.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 986.0, + 589.0, + 986.0, + 589.0, + 1032.0, + 320.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 986.0, + 1162.0, + 986.0, + 1162.0, + 1032.0, + 630.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 986.0, + 1360.0, + 986.0, + 1360.0, + 1032.0, + 1196.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 986.0, + 1405.0, + 986.0, + 1405.0, + 1032.0, + 1401.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1020.0, + 1362.0, + 1020.0, + 1362.0, + 1055.0, + 321.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1047.0, + 520.0, + 1047.0, + 520.0, + 1082.0, + 321.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1047.0, + 713.0, + 1047.0, + 713.0, + 1082.0, + 578.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 1047.0, + 790.0, + 1047.0, + 790.0, + 1082.0, + 752.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1071.0, + 432.0, + 1071.0, + 432.0, + 1124.0, + 343.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 1071.0, + 539.0, + 1071.0, + 539.0, + 1124.0, + 461.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1071.0, + 654.0, + 1071.0, + 654.0, + 1124.0, + 596.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1071.0, + 988.0, + 1071.0, + 988.0, + 1124.0, + 747.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 1071.0, + 1240.0, + 1071.0, + 1240.0, + 1124.0, + 1232.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1116.0, + 563.0, + 1116.0, + 563.0, + 1171.0, + 343.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1170.0, + 409.0, + 1170.0, + 409.0, + 1202.0, + 322.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1199.0, + 862.0, + 1199.0, + 862.0, + 1232.0, + 321.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1226.0, + 452.0, + 1226.0, + 452.0, + 1260.0, + 322.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 1226.0, + 652.0, + 1226.0, + 652.0, + 1260.0, + 566.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1253.0, + 352.0, + 1253.0, + 352.0, + 1290.0, + 348.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1253.0, + 716.0, + 1253.0, + 716.0, + 1290.0, + 693.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1253.0, + 1055.0, + 1253.0, + 1055.0, + 1290.0, + 736.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1281.0, + 411.0, + 1281.0, + 411.0, + 1312.0, + 322.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1309.0, + 667.0, + 1309.0, + 667.0, + 1345.0, + 322.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 349.0, + 683.0, + 349.0, + 683.0, + 388.0, + 322.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 349.0, + 1035.0, + 349.0, + 1035.0, + 388.0, + 713.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 349.0, + 1080.0, + 349.0, + 1080.0, + 388.0, + 1077.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 374.0, + 728.0, + 374.0, + 728.0, + 437.0, + 314.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 374.0, + 1188.0, + 374.0, + 1188.0, + 437.0, + 783.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 374.0, + 1240.0, + 374.0, + 1240.0, + 437.0, + 1230.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 423.0, + 636.0, + 423.0, + 636.0, + 451.0, + 323.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 450.0, + 583.0, + 450.0, + 583.0, + 481.0, + 323.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 474.0, + 433.0, + 474.0, + 433.0, + 509.0, + 350.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 474.0, + 576.0, + 474.0, + 576.0, + 509.0, + 458.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 474.0, + 697.0, + 474.0, + 697.0, + 509.0, + 640.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 474.0, + 744.0, + 474.0, + 744.0, + 509.0, + 735.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 474.0, + 936.0, + 474.0, + 936.0, + 509.0, + 769.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 474.0, + 1007.0, + 474.0, + 1007.0, + 509.0, + 953.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 474.0, + 1140.0, + 474.0, + 1140.0, + 509.0, + 1095.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 474.0, + 1216.0, + 474.0, + 1216.0, + 509.0, + 1165.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 474.0, + 1286.0, + 474.0, + 1286.0, + 509.0, + 1232.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 502.0, + 432.0, + 502.0, + 432.0, + 537.0, + 350.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 502.0, + 584.0, + 502.0, + 584.0, + 537.0, + 453.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 530.0, + 431.0, + 530.0, + 431.0, + 566.0, + 350.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 559.0, + 408.0, + 559.0, + 408.0, + 588.0, + 323.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 586.0, + 826.0, + 586.0, + 826.0, + 618.0, + 319.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 615.0, + 451.0, + 615.0, + 451.0, + 646.0, + 323.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 615.0, + 545.0, + 615.0, + 545.0, + 646.0, + 517.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 615.0, + 610.0, + 615.0, + 610.0, + 646.0, + 573.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 639.0, + 646.0, + 639.0, + 646.0, + 676.0, + 350.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 639.0, + 976.0, + 639.0, + 976.0, + 676.0, + 666.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 669.0, + 612.0, + 669.0, + 612.0, + 704.0, + 350.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 701.0, + 461.0, + 701.0, + 461.0, + 740.0, + 374.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 701.0, + 604.0, + 701.0, + 604.0, + 740.0, + 486.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 701.0, + 821.0, + 701.0, + 821.0, + 740.0, + 763.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 701.0, + 863.0, + 701.0, + 863.0, + 740.0, + 860.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 739.0, + 458.0, + 739.0, + 458.0, + 786.0, + 376.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 777.0, + 435.0, + 777.0, + 435.0, + 806.0, + 351.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 802.0, + 488.0, + 802.0, + 488.0, + 836.0, + 348.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 802.0, + 857.0, + 802.0, + 857.0, + 836.0, + 511.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 826.0, + 352.0, + 826.0, + 352.0, + 872.0, + 348.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 826.0, + 670.0, + 826.0, + 670.0, + 872.0, + 666.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 862.0, + 410.0, + 862.0, + 410.0, + 894.0, + 322.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 222.0, + 706.0, + 222.0, + 706.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 352, + 1407, + 352, + 1407, + 569, + 296, + 569 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 1005, + 1406, + 1005, + 1406, + 1174, + 296, + 1174 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 296, + 1343, + 1405, + 1343, + 1405, + 1536, + 296, + 1536 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1772, + 1405, + 1772, + 1405, + 1928, + 297, + 1928 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1941, + 1403, + 1941, + 1403, + 2035, + 299, + 2035 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 386, + 606, + 1314, + 606, + 1314, + 915, + 386, + 915 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 555, + 1552, + 1144, + 1552, + 1144, + 1759, + 555, + 1759 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 297, + 1265, + 1406, + 1265, + 1406, + 1330, + 297, + 1330 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 404, + 1191, + 1292, + 1191, + 1292, + 1237, + 404, + 1237 + ], + "score": 0.93 + }, + { + "category_id": 4, + "poly": [ + 522, + 940, + 1174, + 940, + 1174, + 973, + 522, + 973 + ], + "score": 0.929 + }, + { + "category_id": 9, + "poly": [ + 1354, + 1697, + 1402, + 1697, + 1402, + 1727, + 1354, + 1727 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1354, + 1200, + 1401, + 1200, + 1401, + 1232, + 1354, + 1232 + ], + "score": 0.878 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2112, + 836, + 2112 + ], + "score": 0.858 + }, + { + "category_id": 0, + "poly": [ + 301, + 226, + 711, + 226, + 711, + 262, + 301, + 262 + ], + "score": 0.841 + }, + { + "category_id": 0, + "poly": [ + 301, + 294, + 814, + 294, + 814, + 328, + 301, + 328 + ], + "score": 0.818 + }, + { + "category_id": 1, + "poly": [ + 301, + 294, + 814, + 294, + 814, + 328, + 301, + 328 + ], + "score": 0.103 + }, + { + "category_id": 13, + "poly": [ + 1131, + 1099, + 1354, + 1099, + 1354, + 1139, + 1131, + 1139 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { f ( { \\pmb x } , t ) = - \\frac { 1 } { 2 } \\beta ( t ) { \\pmb x } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 553, + 1549, + 1144, + 1549, + 1144, + 1762, + 553, + 1762 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { { { \\cal V } _ { n } ^ { ( k - 1 ) } = \\mathbb { E } _ { a \\sim \\pi ( \\cdot \\vert s _ { n } ) } Q _ { \\phi } ( s _ { n } , a ) } } \\\\ { { { \\mathrm { ~ } } } } \\\\ { { { \\mathrm { ~ } } } } \\\\ { { { \\displaystyle ~ \\approx \\sum _ { M } ( \\cdot \\vert s _ { n } ) \\frac { \\exp { \\left( \\alpha Q _ { \\phi } ( s _ { n } , a ) \\right) } } { Z ( s _ { n } ) } Q _ { \\phi } ( s _ { n } , a ) } } } \\\\ { { { \\mathrm { ~ } } } } \\\\ { { { \\displaystyle ~ \\approx \\sum _ { M } \\left[ \\frac { \\exp { \\left( \\alpha Q _ { \\phi } ( s _ { n } , a ) \\right) } } { \\sum _ { M } \\exp { \\left( \\alpha Q _ { \\phi } ( s _ { n } , a ) \\right) } } Q _ { \\phi } ( s _ { n } , a ) \\right] } , } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1136, + 459, + 1136, + 459, + 1177, + 297, + 1177 + ], + "score": 0.94, + "latex": "g ( t ) = \\sqrt { \\beta ( t ) }" + }, + { + "category_id": 13, + "poly": [ + 566, + 1098, + 919, + 1098, + 919, + 1139, + 566, + 1139 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\mathrm { d } \\pmb { x } = - \\frac { 1 } { 2 } \\beta ( t ) \\pmb { x } \\mathrm { d } t + \\sqrt { \\beta ( t ) } \\mathrm { d } \\mathbf { w } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1285, + 1466, + 1368, + 1466, + 1368, + 1503, + 1285, + 1503 + ], + "score": 0.92, + "latex": "V _ { n } ^ { ( k - 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 515, + 1298, + 576, + 1298, + 576, + 1330, + 515, + 1330 + ], + "score": 0.91, + "latex": "\\beta _ { \\mathrm { m a x } }" + }, + { + "category_id": 13, + "poly": [ + 979, + 1266, + 1336, + 1266, + 1336, + 1301, + 979, + 1301 + ], + "score": 0.91, + "latex": "\\beta ( t ) = ( \\beta _ { \\operatorname* { m a x } } - \\beta _ { \\operatorname* { m i n } } ) t + \\beta _ { \\operatorname* { m i n } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1298, + 354, + 1298, + 354, + 1330, + 298, + 1330 + ], + "score": 0.91, + "latex": "\\beta _ { \\mathrm { m i n } }" + }, + { + "category_id": 13, + "poly": [ + 589, + 2003, + 681, + 2003, + 681, + 2031, + 589, + 2031 + ], + "score": 0.9, + "latex": "D = 1 5" + }, + { + "category_id": 13, + "poly": [ + 925, + 1406, + 1012, + 1406, + 1012, + 1434, + 925, + 1434 + ], + "score": 0.9, + "latex": "K = 5" + }, + { + "category_id": 13, + "poly": [ + 1064, + 1376, + 1145, + 1376, + 1145, + 1404, + 1064, + 1404 + ], + "score": 0.89, + "latex": "K = 2" + }, + { + "category_id": 14, + "poly": [ + 404, + 1190, + 1293, + 1190, + 1293, + 1236, + 404, + 1236 + ], + "score": 0.89, + "latex": "p _ { t 0 } ( \\pmb { x } _ { t } | \\pmb { x } _ { 0 } ) = \\mathcal { N } ( \\pmb { x } _ { t } | \\alpha _ { t } \\pmb { x } _ { 0 } , \\sigma _ { t } ^ { 2 } I ) = \\mathcal { N } ( \\pmb { x } _ { t } | e ^ { - \\frac { 1 } { 2 } \\int _ { 0 } ^ { t } \\beta ( s ) \\mathrm { d } s } \\pmb { x } _ { 0 } , [ 1 - e ^ { - \\int _ { 0 } ^ { t } \\beta ( s ) \\mathrm { d } s } ] I ) ." + }, + { + "category_id": 13, + "poly": [ + 476, + 1406, + 563, + 1406, + 563, + 1434, + 476, + 1434 + ], + "score": 0.89, + "latex": "K = 4" + }, + { + "category_id": 13, + "poly": [ + 624, + 1780, + 646, + 1780, + 646, + 1801, + 624, + 1801 + ], + "score": 0.79, + "latex": "\\alpha" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 651.0, + 411.0, + 651.0, + 411.0, + 660.0, + 402.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 656.0, + 1083.0, + 656.0, + 1083.0, + 697.0, + 1065.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 665.0, + 1033.0, + 665.0, + 1033.0, + 673.0, + 1023.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 669.0, + 448.0, + 669.0, + 448.0, + 708.0, + 430.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 681.0, + 505.0, + 681.0, + 505.0, + 865.0, + 482.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 694.0, + 877.0, + 694.0, + 877.0, + 848.0, + 850.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 697.0, + 793.0, + 697.0, + 793.0, + 850.0, + 766.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 699.0, + 573.0, + 699.0, + 573.0, + 848.0, + 549.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 702.0, + 411.0, + 702.0, + 411.0, + 712.0, + 401.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 712.0, + 482.0, + 712.0, + 482.0, + 721.0, + 474.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 756.0, + 415.0, + 756.0, + 415.0, + 773.0, + 396.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 732.0, + 647.0, + 732.0, + 647.0, + 817.0, + 604.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 731.0, + 714.0, + 731.0, + 714.0, + 817.0, + 670.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 758.0, + 753.0, + 758.0, + 753.0, + 784.0, + 726.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 755.0, + 839.0, + 755.0, + 839.0, + 787.0, + 808.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 758.0, + 923.0, + 758.0, + 923.0, + 786.0, + 895.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 753.0, + 957.0, + 753.0, + 957.0, + 791.0, + 939.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 759.0, + 1032.0, + 759.0, + 1032.0, + 770.0, + 1023.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 753.0, + 1083.0, + 753.0, + 1083.0, + 792.0, + 1065.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 754.0, + 1134.0, + 754.0, + 1134.0, + 773.0, + 1098.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 771.0, + 1157.0, + 771.0, + 1157.0, + 780.0, + 1146.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 752.0, + 1202.0, + 752.0, + 1202.0, + 794.0, + 1181.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 754.0, + 1252.0, + 754.0, + 1252.0, + 773.0, + 1216.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 763.0, + 1280.0, + 763.0, + 1280.0, + 794.0, + 1256.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 843.0, + 1091.0, + 843.0, + 1091.0, + 897.0, + 1073.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 845.0, + 1077.0, + 845.0, + 1077.0, + 896.0, + 1059.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.0, + 865.0, + 1058.0, + 865.0, + 1058.0, + 876.0, + 1049.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 888.0, + 681.0, + 888.0, + 681.0, + 911.0, + 577.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 885.0, + 756.0, + 885.0, + 756.0, + 915.0, + 698.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 608.0, + 459.0, + 608.0, + 459.0, + 617.0, + 455.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 939.0, + 1174.0, + 939.0, + 1174.0, + 975.0, + 524.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 868.0, + 2084.0, + 868.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 224.0, + 714.0, + 224.0, + 714.0, + 266.0, + 293.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 294.0, + 814.0, + 294.0, + 814.0, + 330.0, + 294.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 352.0, + 1405.0, + 352.0, + 1405.0, + 387.0, + 294.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 385.0, + 1403.0, + 385.0, + 1403.0, + 416.0, + 295.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 415.0, + 1405.0, + 415.0, + 1405.0, + 449.0, + 295.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 444.0, + 1405.0, + 444.0, + 1405.0, + 479.0, + 293.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 474.0, + 1407.0, + 474.0, + 1407.0, + 510.0, + 293.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 508.0, + 1405.0, + 508.0, + 1405.0, + 539.0, + 296.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 537.0, + 1033.0, + 537.0, + 1033.0, + 568.0, + 294.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1007.0, + 1406.0, + 1007.0, + 1406.0, + 1040.0, + 295.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1037.0, + 1406.0, + 1037.0, + 1406.0, + 1070.0, + 295.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1067.0, + 1407.0, + 1067.0, + 1407.0, + 1103.0, + 294.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1099.0, + 565.0, + 1099.0, + 565.0, + 1139.0, + 291.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1099.0, + 1130.0, + 1099.0, + 1130.0, + 1139.0, + 920.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1099.0, + 1407.0, + 1099.0, + 1407.0, + 1139.0, + 1355.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1139.0, + 734.0, + 1139.0, + 734.0, + 1177.0, + 460.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1344.0, + 1406.0, + 1344.0, + 1406.0, + 1377.0, + 295.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1374.0, + 1063.0, + 1374.0, + 1063.0, + 1408.0, + 294.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1374.0, + 1405.0, + 1374.0, + 1405.0, + 1408.0, + 1146.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1403.0, + 475.0, + 1403.0, + 475.0, + 1439.0, + 294.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 1403.0, + 924.0, + 1403.0, + 924.0, + 1439.0, + 564.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 1403.0, + 1405.0, + 1403.0, + 1405.0, + 1439.0, + 1013.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1433.0, + 1406.0, + 1433.0, + 1406.0, + 1472.0, + 293.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1462.0, + 1284.0, + 1462.0, + 1284.0, + 1519.0, + 284.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1369.0, + 1462.0, + 1406.0, + 1462.0, + 1406.0, + 1519.0, + 1369.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1505.0, + 446.0, + 1505.0, + 446.0, + 1539.0, + 295.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1773.0, + 623.0, + 1773.0, + 623.0, + 1806.0, + 295.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 1773.0, + 1406.0, + 1773.0, + 1406.0, + 1806.0, + 647.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1805.0, + 1403.0, + 1805.0, + 1403.0, + 1838.0, + 295.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1833.0, + 1405.0, + 1833.0, + 1405.0, + 1870.0, + 294.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1862.0, + 1406.0, + 1862.0, + 1406.0, + 1901.0, + 292.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1890.0, + 1069.0, + 1890.0, + 1069.0, + 1933.0, + 294.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 1408.0, + 1972.0, + 1408.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2003.0, + 588.0, + 2003.0, + 588.0, + 2037.0, + 295.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 2003.0, + 1404.0, + 2003.0, + 1404.0, + 2037.0, + 682.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1264.0, + 978.0, + 1264.0, + 978.0, + 1304.0, + 294.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 1264.0, + 1406.0, + 1264.0, + 1406.0, + 1304.0, + 1337.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1296.0, + 514.0, + 1296.0, + 514.0, + 1332.0, + 355.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1296.0, + 688.0, + 1296.0, + 688.0, + 1332.0, + 577.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 294.0, + 814.0, + 294.0, + 814.0, + 330.0, + 294.0, + 330.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 628, + 1405, + 628, + 1405, + 812, + 298, + 812 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 367, + 1404, + 367, + 1404, + 614, + 297, + 614 + ], + "score": 0.979 + }, + { + "category_id": 5, + "poly": [ + 458, + 1066, + 1234, + 1066, + 1234, + 1246, + 458, + 1246 + ], + "score": 0.978, + "html": "
Diffusion Steps D5 steps10 steps15 steps25 steps
Performance (Locomotion)2.372.975.674.4
Performance (Antmaze)5.565.774.273.0
Runtime (1 episode, # envs=1)22.3 s38.0 s50.0 s93.0 s
Runtime (1 episode,# envs=20)1.5 s2.5 s3.2 s5.0s
" + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1403, + 228, + 1403, + 353, + 298, + 353 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 827, + 1405, + 827, + 1405, + 1013, + 297, + 1013 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 522, + 1438, + 1175, + 1438, + 1175, + 1921, + 522, + 1921 + ], + "score": 0.97 + }, + { + "category_id": 4, + "poly": [ + 633, + 1950, + 1065, + 1950, + 1065, + 1983, + 633, + 1983 + ], + "score": 0.906 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2113, + 835, + 2113 + ], + "score": 0.864 + }, + { + "category_id": 6, + "poly": [ + 295, + 1268, + 1409, + 1268, + 1409, + 1331, + 295, + 1331 + ], + "score": 0.483 + }, + { + "category_id": 13, + "poly": [ + 729, + 860, + 809, + 860, + 809, + 888, + 729, + 888 + ], + "score": 0.86, + "latex": "\\mathcal { D } = 1" + }, + { + "category_id": 13, + "poly": [ + 1237, + 261, + 1401, + 261, + 1401, + 295, + 1237, + 295 + ], + "score": 0.85, + "latex": "\\{ 5 , 1 0 , 1 5 , 2 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 1020, + 830, + 1047, + 830, + 1047, + 856, + 1020, + 856 + ], + "score": 0.78, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 827, + 860, + 855, + 860, + 855, + 887, + 827, + 887 + ], + "score": 0.75, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 878, + 584, + 911, + 584, + 911, + 610, + 878, + 610 + ], + "score": 0.74, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 503, + 752, + 566, + 752, + 566, + 780, + 503, + 780 + ], + "score": 0.69, + "latex": "\\scriptstyle ( \\mathrm { K } = 5 )" + }, + { + "category_id": 13, + "poly": [ + 1137, + 492, + 1170, + 492, + 1170, + 519, + 1137, + 519 + ], + "score": 0.68, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 735, + 1185, + 761, + 1185, + 761, + 1205, + 735, + 1205 + ], + "score": 0.3, + "latex": "{ \\mathfrak { s } } { = } 1" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1447.0, + 586.0, + 1447.0, + 586.0, + 1488.0, + 543.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1469.0, + 1060.0, + 1469.0, + 1060.0, + 1478.0, + 1051.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1487.0, + 973.0, + 1487.0, + 973.0, + 1499.0, + 961.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1504.0, + 885.0, + 1504.0, + 885.0, + 1517.0, + 873.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1540.0, + 554.0, + 1540.0, + 554.0, + 1770.0, + 519.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 1548.0, + 1149.0, + 1548.0, + 1149.0, + 1555.0, + 1141.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1560.0, + 1060.0, + 1560.0, + 1060.0, + 1570.0, + 1051.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1573.0, + 973.0, + 1573.0, + 973.0, + 1586.0, + 962.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1591.0, + 882.0, + 1591.0, + 882.0, + 1601.0, + 874.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1686.0, + 795.0, + 1686.0, + 795.0, + 1694.0, + 786.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1727.0, + 1060.0, + 1727.0, + 1060.0, + 1737.0, + 1051.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1752.0, + 881.0, + 1752.0, + 881.0, + 1757.0, + 876.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1748.0, + 1143.0, + 1748.0, + 1143.0, + 1778.0, + 1010.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1759.0, + 795.0, + 1759.0, + 795.0, + 1770.0, + 786.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 1774.0, + 1145.0, + 1774.0, + 1145.0, + 1806.0, + 1009.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 1799.0, + 1155.0, + 1799.0, + 1155.0, + 1835.0, + 1009.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1829.0, + 1161.0, + 1829.0, + 1161.0, + 1859.0, + 1012.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1870.0, + 712.0, + 1870.0, + 712.0, + 1899.0, + 691.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 1869.0, + 808.0, + 1869.0, + 808.0, + 1900.0, + 774.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 1868.0, + 899.0, + 1868.0, + 899.0, + 1900.0, + 861.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 1868.0, + 986.0, + 1868.0, + 986.0, + 1900.0, + 949.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1868.0, + 1075.0, + 1868.0, + 1075.0, + 1900.0, + 1039.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 1867.0, + 1164.0, + 1867.0, + 1164.0, + 1902.0, + 1125.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1891.0, + 946.0, + 1891.0, + 946.0, + 1924.0, + 812.0, + 1924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.75, + 1589.0, + 588.75, + 1589.0, + 588.75, + 1622.0, + 539.75, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.25, + 1641.5, + 692.25, + 1641.5, + 692.25, + 1683.0, + 629.25, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1725.5, + 591.0, + 1725.5, + 591.0, + 1757.0, + 537.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 1948.0, + 1066.0, + 1948.0, + 1066.0, + 1985.0, + 631.0, + 1985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1266.0, + 1408.0, + 1266.0, + 1408.0, + 1301.0, + 293.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1296.0, + 1123.0, + 1296.0, + 1123.0, + 1334.0, + 293.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 630.0, + 1403.0, + 630.0, + 1403.0, + 662.0, + 296.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 660.0, + 1406.0, + 660.0, + 1406.0, + 692.0, + 295.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 688.0, + 1408.0, + 688.0, + 1408.0, + 725.0, + 293.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 720.0, + 1407.0, + 720.0, + 1407.0, + 756.0, + 295.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 750.0, + 502.0, + 750.0, + 502.0, + 782.0, + 293.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 750.0, + 1407.0, + 750.0, + 1407.0, + 782.0, + 567.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 781.0, + 357.0, + 781.0, + 357.0, + 815.0, + 292.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 369.0, + 1405.0, + 369.0, + 1405.0, + 403.0, + 295.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 401.0, + 1404.0, + 401.0, + 1404.0, + 431.0, + 296.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 430.0, + 1406.0, + 430.0, + 1406.0, + 464.0, + 295.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 463.0, + 1405.0, + 463.0, + 1405.0, + 493.0, + 296.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 491.0, + 1136.0, + 491.0, + 1136.0, + 524.0, + 295.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 491.0, + 1405.0, + 491.0, + 1405.0, + 524.0, + 1171.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 521.0, + 1404.0, + 521.0, + 1404.0, + 554.0, + 295.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 552.0, + 1402.0, + 552.0, + 1402.0, + 586.0, + 295.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 583.0, + 877.0, + 583.0, + 877.0, + 617.0, + 296.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 583.0, + 1190.0, + 583.0, + 1190.0, + 617.0, + 912.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 230.0, + 1408.0, + 230.0, + 1408.0, + 266.0, + 294.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 260.0, + 1236.0, + 260.0, + 1236.0, + 296.0, + 293.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 293.0, + 1403.0, + 293.0, + 1403.0, + 326.0, + 294.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 322.0, + 1407.0, + 322.0, + 1407.0, + 355.0, + 294.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 825.0, + 1019.0, + 825.0, + 1019.0, + 864.0, + 294.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 825.0, + 1405.0, + 825.0, + 1405.0, + 864.0, + 1048.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 859.0, + 728.0, + 859.0, + 728.0, + 893.0, + 294.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 859.0, + 826.0, + 859.0, + 826.0, + 893.0, + 810.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 859.0, + 1405.0, + 859.0, + 1405.0, + 893.0, + 856.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 889.0, + 1408.0, + 889.0, + 1408.0, + 925.0, + 294.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 916.0, + 1406.0, + 916.0, + 1406.0, + 958.0, + 292.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 951.0, + 1405.0, + 951.0, + 1405.0, + 986.0, + 294.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 980.0, + 422.0, + 980.0, + 422.0, + 1017.0, + 294.0, + 1017.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 286, + 1405, + 286, + 1405, + 531, + 297, + 531 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1271, + 1405, + 1271, + 1405, + 1424, + 299, + 1424 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 1071, + 1406, + 1071, + 1406, + 1256, + 299, + 1256 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 299, + 734, + 1406, + 734, + 1406, + 887, + 299, + 887 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 299, + 902, + 1406, + 902, + 1406, + 1058, + 299, + 1058 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 299, + 547, + 1406, + 547, + 1406, + 639, + 299, + 639 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 299, + 1439, + 1404, + 1439, + 1404, + 1594, + 299, + 1594 + ], + "score": 0.967 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.883 + }, + { + "category_id": 2, + "poly": [ + 836, + 2089, + 864, + 2089, + 864, + 2113, + 836, + 2113 + ], + "score": 0.842 + }, + { + "category_id": 0, + "poly": [ + 298, + 678, + 945, + 678, + 945, + 708, + 298, + 708 + ], + "score": 0.743 + }, + { + "category_id": 1, + "poly": [ + 296, + 231, + 992, + 231, + 992, + 261, + 296, + 261 + ], + "score": 0.346 + }, + { + "category_id": 0, + "poly": [ + 296, + 231, + 992, + 231, + 992, + 261, + 296, + 261 + ], + "score": 0.248 + }, + { + "category_id": 1, + "poly": [ + 298, + 678, + 945, + 678, + 945, + 708, + 298, + 708 + ], + "score": 0.158 + }, + { + "category_id": 13, + "poly": [ + 619, + 1025, + 865, + 1025, + 865, + 1059, + 619, + 1059 + ], + "score": 0.93, + "latex": "\\tau \\in \\{ 0 . \\bar { 1 } , 0 . 2 , . . . , 0 . 9 \\}" + }, + { + "category_id": 13, + "poly": [ + 985, + 409, + 1064, + 409, + 1064, + 443, + 985, + 443 + ], + "score": 0.88, + "latex": "[ - 1 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 1354, + 1001, + 1373, + 1001, + 1373, + 1023, + 1354, + 1023 + ], + "score": 0.72, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 1236, + 319, + 1295, + 319, + 1295, + 348, + 1236, + 348 + ], + "score": 0.58, + "latex": "3 0 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 474, + 471, + 535, + 471, + 535, + 500, + 474, + 500 + ], + "score": 0.37, + "latex": "3 0 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 366, + 288, + 392, + 288, + 392, + 316, + 366, + 316 + ], + "score": 0.32, + "latex": "^ +" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 676.0, + 947.0, + 676.0, + 947.0, + 712.0, + 294.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 998.0, + 227.0, + 998.0, + 266.0, + 293.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 288.0, + 365.0, + 288.0, + 365.0, + 321.0, + 295.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 288.0, + 1405.0, + 288.0, + 1405.0, + 321.0, + 393.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 319.0, + 1235.0, + 319.0, + 1235.0, + 351.0, + 294.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 319.0, + 1405.0, + 319.0, + 1405.0, + 351.0, + 1296.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 349.0, + 1405.0, + 349.0, + 1405.0, + 381.0, + 294.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 379.0, + 1403.0, + 379.0, + 1403.0, + 412.0, + 295.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 408.0, + 984.0, + 408.0, + 984.0, + 446.0, + 294.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 408.0, + 1408.0, + 408.0, + 1408.0, + 446.0, + 1065.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 436.0, + 1406.0, + 436.0, + 1406.0, + 476.0, + 292.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 468.0, + 473.0, + 468.0, + 473.0, + 506.0, + 294.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 468.0, + 1406.0, + 468.0, + 1406.0, + 506.0, + 536.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 501.0, + 669.0, + 501.0, + 669.0, + 531.0, + 295.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1270.0, + 1404.0, + 1270.0, + 1404.0, + 1305.0, + 294.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1301.0, + 1406.0, + 1301.0, + 1406.0, + 1334.0, + 295.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1333.0, + 1405.0, + 1333.0, + 1405.0, + 1366.0, + 295.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1361.0, + 1406.0, + 1361.0, + 1406.0, + 1399.0, + 293.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1391.0, + 1160.0, + 1391.0, + 1160.0, + 1431.0, + 293.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1072.0, + 1404.0, + 1072.0, + 1404.0, + 1108.0, + 294.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1101.0, + 1404.0, + 1101.0, + 1404.0, + 1137.0, + 294.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1132.0, + 1411.0, + 1132.0, + 1411.0, + 1169.0, + 293.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1166.0, + 1403.0, + 1166.0, + 1403.0, + 1198.0, + 297.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1193.0, + 1407.0, + 1193.0, + 1407.0, + 1232.0, + 294.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1225.0, + 987.0, + 1225.0, + 987.0, + 1260.0, + 296.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 733.0, + 1408.0, + 733.0, + 1408.0, + 771.0, + 293.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 764.0, + 1406.0, + 764.0, + 1406.0, + 801.0, + 293.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 792.0, + 1411.0, + 792.0, + 1411.0, + 832.0, + 293.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 824.0, + 1409.0, + 824.0, + 1409.0, + 863.0, + 294.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 858.0, + 758.0, + 858.0, + 758.0, + 891.0, + 296.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 904.0, + 1404.0, + 904.0, + 1404.0, + 937.0, + 296.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 933.0, + 1406.0, + 933.0, + 1406.0, + 971.0, + 294.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 963.0, + 1411.0, + 963.0, + 1411.0, + 1001.0, + 293.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 995.0, + 1353.0, + 995.0, + 1353.0, + 1032.0, + 296.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 995.0, + 1406.0, + 995.0, + 1406.0, + 1032.0, + 1374.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1025.0, + 618.0, + 1025.0, + 618.0, + 1063.0, + 296.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1025.0, + 1299.0, + 1025.0, + 1299.0, + 1063.0, + 866.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 541.0, + 1406.0, + 541.0, + 1406.0, + 587.0, + 292.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 577.0, + 1408.0, + 577.0, + 1408.0, + 613.0, + 294.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 607.0, + 1255.0, + 607.0, + 1255.0, + 645.0, + 294.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1439.0, + 1406.0, + 1439.0, + 1406.0, + 1477.0, + 294.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1471.0, + 1404.0, + 1471.0, + 1404.0, + 1505.0, + 295.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1499.0, + 1404.0, + 1499.0, + 1404.0, + 1536.0, + 293.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1531.0, + 1405.0, + 1531.0, + 1405.0, + 1568.0, + 293.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1564.0, + 506.0, + 1564.0, + 506.0, + 1594.0, + 295.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 998.0, + 227.0, + 998.0, + 266.0, + 293.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 676.0, + 947.0, + 676.0, + 947.0, + 712.0, + 294.0, + 712.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 325, + 954, + 1328, + 954, + 1328, + 1183, + 325, + 1183 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 508, + 1521, + 1188, + 1521, + 1188, + 1611, + 508, + 1611 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 621, + 329, + 1076, + 329, + 1076, + 379, + 621, + 379 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 296, + 597, + 1404, + 597, + 1404, + 666, + 296, + 666 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 299, + 385, + 1409, + 385, + 1409, + 448, + 299, + 448 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 297, + 293, + 1237, + 293, + 1237, + 327, + 297, + 327 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 297, + 915, + 652, + 915, + 652, + 949, + 297, + 949 + ], + "score": 0.917 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 816, + 73, + 816, + 106, + 297, + 106 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 297, + 1665, + 535, + 1665, + 535, + 1696, + 297, + 1696 + ], + "score": 0.912 + }, + { + "category_id": 1, + "poly": [ + 300, + 1617, + 796, + 1617, + 796, + 1652, + 300, + 1652 + ], + "score": 0.91 + }, + { + "category_id": 1, + "poly": [ + 297, + 813, + 455, + 813, + 455, + 842, + 297, + 842 + ], + "score": 0.904 + }, + { + "category_id": 1, + "poly": [ + 299, + 1480, + 855, + 1480, + 855, + 1517, + 299, + 1517 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 299, + 1185, + 793, + 1185, + 793, + 1219, + 299, + 1219 + ], + "score": 0.902 + }, + { + "category_id": 1, + "poly": [ + 299, + 1237, + 1094, + 1237, + 1094, + 1273, + 299, + 1273 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1552, + 1403, + 1552, + 1403, + 1583, + 1353, + 1583 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1353, + 332, + 1402, + 332, + 1402, + 364, + 1353, + 364 + ], + "score": 0.89 + }, + { + "category_id": 1, + "poly": [ + 297, + 545, + 667, + 545, + 667, + 581, + 297, + 581 + ], + "score": 0.889 + }, + { + "category_id": 1, + "poly": [ + 296, + 452, + 920, + 452, + 920, + 486, + 296, + 486 + ], + "score": 0.888 + }, + { + "category_id": 1, + "poly": [ + 296, + 497, + 1099, + 497, + 1099, + 537, + 296, + 537 + ], + "score": 0.882 + }, + { + "category_id": 8, + "poly": [ + 470, + 1277, + 1229, + 1277, + 1229, + 1459, + 470, + 1459 + ], + "score": 0.881 + }, + { + "category_id": 1, + "poly": [ + 384, + 1950, + 986, + 1950, + 986, + 2000, + 384, + 2000 + ], + "score": 0.88 + }, + { + "category_id": 8, + "poly": [ + 324, + 842, + 1312, + 842, + 1312, + 895, + 324, + 895 + ], + "score": 0.879 + }, + { + "category_id": 9, + "poly": [ + 1353, + 770, + 1402, + 770, + 1402, + 802, + 1353, + 802 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1835, + 1403, + 1835, + 1403, + 1866, + 1353, + 1866 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1353, + 1403, + 1353, + 1403, + 1384, + 1353, + 1384 + ], + "score": 0.876 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1052, + 1403, + 1052, + 1403, + 1083, + 1353, + 1083 + ], + "score": 0.87 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 863, + 2087, + 863, + 2113, + 835, + 2113 + ], + "score": 0.865 + }, + { + "category_id": 9, + "poly": [ + 1354, + 855, + 1403, + 855, + 1403, + 885, + 1354, + 885 + ], + "score": 0.86 + }, + { + "category_id": 2, + "poly": [ + 1374, + 2003, + 1403, + 2003, + 1403, + 2032, + 1374, + 2032 + ], + "score": 0.841 + }, + { + "category_id": 8, + "poly": [ + 384, + 766, + 1303, + 766, + 1303, + 804, + 384, + 804 + ], + "score": 0.821 + }, + { + "category_id": 1, + "poly": [ + 297, + 699, + 370, + 699, + 370, + 728, + 297, + 728 + ], + "score": 0.81 + }, + { + "category_id": 0, + "poly": [ + 299, + 225, + 718, + 225, + 718, + 262, + 299, + 262 + ], + "score": 0.769 + }, + { + "category_id": 1, + "poly": [ + 300, + 728, + 770, + 728, + 770, + 761, + 300, + 761 + ], + "score": 0.54 + }, + { + "category_id": 8, + "poly": [ + 397, + 767, + 867, + 767, + 867, + 804, + 397, + 804 + ], + "score": 0.529 + }, + { + "category_id": 8, + "poly": [ + 385, + 1820, + 858, + 1820, + 858, + 1944, + 385, + 1944 + ], + "score": 0.364 + }, + { + "category_id": 8, + "poly": [ + 463, + 1701, + 1277, + 1701, + 1277, + 1817, + 463, + 1817 + ], + "score": 0.316 + }, + { + "category_id": 8, + "poly": [ + 575, + 1862, + 858, + 1862, + 858, + 1944, + 575, + 1944 + ], + "score": 0.223 + }, + { + "category_id": 1, + "poly": [ + 385, + 1820, + 858, + 1820, + 858, + 1944, + 385, + 1944 + ], + "score": 0.222 + }, + { + "category_id": 8, + "poly": [ + 551, + 1277, + 1226, + 1277, + 1226, + 1401, + 551, + 1401 + ], + "score": 0.161 + }, + { + "category_id": 9, + "poly": [ + 982, + 1864, + 1262, + 1864, + 1262, + 1900, + 982, + 1900 + ], + "score": 0.09 + }, + { + "category_id": 14, + "poly": [ + 408, + 1698, + 1317, + 1698, + 1317, + 2005, + 408, + 2005 + ], + "score": 0.94, + "latex": "\\begin{array} { r l r } & { \\begin{array} { l l } & { Q ^ { * } ( s , a ) \\geq T ^ { \\mu } Q ^ { * } ( s , a ) \\geq ( T ^ { \\mu } ) ^ { 2 } Q ^ { * } ( s , a ) } & { \\mathrm { ( b y ~ m o n o t o n i c i t y ~ o f ~ } T ^ { \\mu } ) } \\\\ & { \\qquad \\geq \\ldots \\geq ( T ^ { \\mu } ) ^ { n } Q ^ { * } ( s , a ) } \\\\ & { \\qquad \\geq ( T ^ { \\mu } ) ^ { n } { \\mathcal T } ^ { \\pi } Q ^ { * } ( s , a ) , } & { \\mathrm { ( b y ~ m o n o t o n i c i t y ~ o f ~ } ( T ^ { \\mu } ) ^ { n } ) } \\end{array} } \\\\ & { } \\\\ & { \\begin{array} { r l r } { \\mathrm { \\ h u s } } & { Q ^ { * } ( s , a ) \\geq { \\mathcal T } _ { \\mu } ^ { \\pi } Q ^ { * } ( s , a ) } \\\\ & { \\qquad \\geq { \\mathcal T } _ { \\mu } ^ { \\pi } { \\mathcal T } _ { \\mu } ^ { \\pi } Q ^ { * } ( s , a ) } & { \\mathrm { ( b y ~ m o n o t o n i c i t y ~ o f ~ } { \\mathcal T } _ { \\mu } ^ { \\pi } ) } \\\\ & { \\qquad \\geq \\ldots \\geq ( { \\mathcal T } _ { \\mu } ^ { \\pi } ) ^ { m } Q ^ { * } ( s , a ) , } & \\\\ { \\mathrm { \\ h u s } } & { Q ^ { * } ( s , a ) \\geq \\operatorname* { l i m } _ { m \\to \\infty } ( { \\mathcal T } _ { \\mu } ^ { \\pi } ) ^ { m } Q ^ { * } ( s , a ) = \\tilde { Q } ( s , a ) . } \\end{array} } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 324, + 950, + 1332, + 950, + 1332, + 1186, + 324, + 1186 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\left| { \\cal T } _ { \\mu } ^ { \\pi } Q _ { 1 } ( s , a ) - { \\cal T } _ { \\mu } ^ { \\pi } Q _ { 2 } ( s , a ) \\right| = \\displaystyle \\left| \\operatorname* { m a x } _ { n \\geq 0 } \\{ ( { \\cal T } ^ { \\mu } ) ^ { n } { \\cal T } ^ { \\pi } Q _ { 1 } ( s , a ) \\} - \\operatorname* { m a x } _ { n \\geq 0 } \\{ ( { \\cal T } ^ { \\mu } ) ^ { n } { \\cal T } ^ { \\pi } Q _ { 2 } ( s , a ) \\} \\right| } \\\\ & { \\qquad \\leq \\operatorname* { m a x } _ { n \\geq 0 } \\{ | ( { \\cal T } ^ { \\mu } ) ^ { n } { \\cal T } ^ { \\pi } Q _ { 1 } ( s , a ) - ( { \\cal T } ^ { \\mu } ) ^ { n } { \\cal T } ^ { \\pi } Q _ { 2 } ( s , a ) | \\} } \\\\ & { \\qquad \\leq \\operatorname* { m a x } _ { n \\geq 0 } \\{ \\gamma ^ { n + 1 } \\| Q _ { 1 } - Q _ { 2 } \\| _ { \\infty } \\} } \\\\ & { \\qquad = \\gamma \\| Q _ { 1 } - Q _ { 2 } \\| _ { \\infty } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 618, + 329, + 1079, + 329, + 1079, + 383, + 618, + 383 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\mathcal { T } _ { \\mu } ^ { \\pi } Q ( s , \\pmb { a } ) : = \\operatorname* { m a x } _ { n \\geq 0 } \\{ ( \\mathcal { T } ^ { \\mu } ) ^ { n } \\mathcal { T } ^ { \\pi } Q ( s , \\pmb { a } ) \\} . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 859, + 598, + 1227, + 598, + 1227, + 636, + 859, + 636 + ], + "score": 0.92, + "latex": "Q ^ { \\pi } ( s , a ) \\leq \\tilde { Q } ( s , a ) \\leq Q ^ { * } ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 515, + 1480, + 742, + 1480, + 742, + 1518, + 515, + 1518 + ], + "score": 0.92, + "latex": "\\tilde { Q } ( s , a ) \\leq Q ^ { * } ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 909, + 499, + 1090, + 499, + 1090, + 537, + 909, + 537 + ], + "score": 0.92, + "latex": "{ \\mathcal T } _ { \\mu } ^ { \\pi } Q _ { 1 } \\leq { \\mathcal T } _ { \\mu } ^ { \\pi } Q _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 604, + 1235, + 835, + 1235, + 835, + 1274, + 604, + 1274 + ], + "score": 0.92, + "latex": "Q ^ { \\pi } ( s , \\pmb { a } ) \\leq \\tilde { Q } ( s , \\pmb { a } )" + }, + { + "category_id": 13, + "poly": [ + 399, + 1666, + 525, + 1666, + 525, + 1697, + 399, + 1697 + ], + "score": 0.91, + "latex": "\\forall n , m \\in \\mathbb { N }" + }, + { + "category_id": 14, + "poly": [ + 507, + 1519, + 1191, + 1519, + 1191, + 1613, + 507, + 1613 + ], + "score": 0.91, + "latex": "\\begin{array} { r l } & { \\mathcal { T } ^ { \\pi } Q ^ { * } ( s , a ) = \\mathcal { R } ( s , { a } ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } } \\mathbb { E } _ { a ^ { \\prime } \\sim \\hat { \\pi } ( \\cdot \\vert s ^ { \\prime } ) } Q ^ { * } ( s ^ { \\prime } , { a } ^ { \\prime } ) } \\\\ & { \\qquad \\leq \\mathcal { R } ( s , a ) + \\gamma \\mathbb { E } _ { s ^ { \\prime } } \\operatorname* { m a x } _ { a ^ { \\prime } } Q ^ { * } ( s ^ { \\prime } , a ^ { \\prime } ) = Q ^ { * } ( s , a ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 676, + 499, + 799, + 499, + 799, + 532, + 676, + 532 + ], + "score": 0.91, + "latex": "\\forall Q _ { 1 } \\leq Q _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1198, + 387, + 1236, + 387, + 1236, + 423, + 1198, + 423 + ], + "score": 0.91, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 14, + "poly": [ + 469, + 1276, + 1232, + 1276, + 1232, + 1463, + 469, + 1463 + ], + "score": 0.9, + "latex": "\\begin{array} { r l } & { \\displaystyle { \\cal Q } ^ { \\pi } ( s , { \\boldsymbol a } ) = \\mathcal { T } ^ { \\pi } { \\cal Q } ^ { \\pi } ( s , { \\boldsymbol a } ) \\leq \\mathcal { T } _ { \\mu } ^ { \\pi } { \\cal Q } ^ { \\pi } ( s , { \\boldsymbol a } ) = \\mathcal { T } _ { \\mu } ^ { \\pi } \\mathcal { T } ^ { \\pi } { \\cal Q } ^ { \\pi } ( s , { \\boldsymbol a } ) } \\\\ & { \\leq \\mathcal { T } _ { \\mu } ^ { \\pi } \\mathcal { T } _ { \\mu } ^ { \\pi } { \\cal Q } ^ { \\pi } ( s , { \\boldsymbol a } ) \\quad \\mathrm { ( b y ~ m o n o t o n i c i t y ~ o f ~ } \\mathcal { T } _ { \\mu } ^ { \\pi } ) } \\\\ & { \\leq \\ldots \\leq ( \\mathcal { T } _ { \\mu } ^ { \\pi } ) ^ { m } { \\cal Q } ^ { \\pi } ( s , { \\boldsymbol a } ) , } \\\\ { \\mathrm { T h u s } \\quad { \\cal Q } ^ { \\pi } ( s , { \\boldsymbol a } ) \\leq \\displaystyle \\operatorname* { l i m } _ { m \\infty } ( \\mathcal { T } _ { \\mu } ^ { \\pi } ) ^ { m } { \\cal Q } ^ { \\pi } ( s , { \\boldsymbol a } ) = \\tilde { \\boldsymbol { Q } } ( s , { \\boldsymbol a } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 511, + 1618, + 787, + 1618, + 787, + 1652, + 511, + 1652 + ], + "score": 0.9, + "latex": "{ \\mathcal { T } } ^ { \\mu } Q ^ { * } ( s , a ) \\leq Q ^ { * } ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 369, + 729, + 665, + 729, + 665, + 761, + 369, + 761 + ], + "score": 0.9, + "latex": "\\forall Q _ { 1 } \\leq Q _ { 2 } , \\forall s , \\mathbf { a } , \\forall n \\in \\mathbb { N }" + }, + { + "category_id": 13, + "poly": [ + 875, + 454, + 912, + 454, + 912, + 490, + 875, + 490 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 462, + 1186, + 501, + 1186, + 501, + 1223, + 462, + 1223 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 372, + 916, + 544, + 916, + 544, + 949, + 372, + 949 + ], + "score": 0.9, + "latex": "\\forall Q _ { 1 } , Q _ { 2 } , \\forall s , a" + }, + { + "category_id": 14, + "poly": [ + 337, + 841, + 1318, + 841, + 1318, + 897, + 337, + 897 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi } Q _ { 1 } ( s , a ) = \\operatorname* { m a x } _ { n \\geq 0 } \\{ ( \\mathscr { T } ^ { \\mu } ) ^ { n } \\mathscr { T } ^ { \\pi } Q _ { 1 } ( s , a ) \\} \\leq \\operatorname* { m a x } _ { n \\geq 0 } \\{ ( \\mathscr { T } ^ { \\mu } ) ^ { n } \\mathscr { T } ^ { \\pi } Q _ { 2 } ( s , a ) \\} = \\mathscr { T } _ { \\mu } ^ { \\pi } Q _ { 2 } ( s , a ) ." + }, + { + "category_id": 13, + "poly": [ + 372, + 1242, + 435, + 1242, + 435, + 1272, + 372, + 1272 + ], + "score": 0.89, + "latex": "\\forall s , a" + }, + { + "category_id": 13, + "poly": [ + 892, + 1240, + 990, + 1240, + 990, + 1270, + 892, + 1270 + ], + "score": 0.88, + "latex": "\\forall m \\in \\mathbb { N }" + }, + { + "category_id": 13, + "poly": [ + 690, + 634, + 726, + 634, + 726, + 662, + 690, + 662 + ], + "score": 0.87, + "latex": "\\tau ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 779, + 634, + 819, + 634, + 819, + 662, + 779, + 662 + ], + "score": 0.86, + "latex": "\\mathcal { T } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 606, + 602, + 645, + 602, + 645, + 640, + 606, + 640 + ], + "score": 0.86, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 326, + 546, + 366, + 546, + 366, + 584, + 326, + 584 + ], + "score": 0.85, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 1334, + 604, + 1399, + 604, + 1399, + 634, + 1334, + 634 + ], + "score": 0.85, + "latex": "\\forall s , a" + }, + { + "category_id": 13, + "poly": [ + 353, + 633, + 439, + 633, + 439, + 665, + 353, + 665 + ], + "score": 0.85, + "latex": "Q ^ { \\pi } , Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 674, + 598, + 699, + 598, + 699, + 634, + 674, + 634 + ], + "score": 0.83, + "latex": "\\tilde { Q }" + }, + { + "category_id": 13, + "poly": [ + 635, + 1192, + 654, + 1192, + 654, + 1218, + 635, + 1218 + ], + "score": 0.82, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 325, + 500, + 364, + 500, + 364, + 537, + 325, + 537 + ], + "score": 0.81, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 1215, + 768, + 1297, + 768, + 1297, + 803, + 1215, + 803 + ], + "score": 0.74, + "latex": "( \\mathcal T ^ { \\mu } ) ^ { n } )" + }, + { + "category_id": 13, + "poly": [ + 504, + 553, + 523, + 553, + 523, + 579, + 504, + 579 + ], + "score": 0.74, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 399, + 767, + 860, + 767, + 860, + 804, + 399, + 804 + ], + "score": 0.73, + "latex": "( T ^ { \\mu } ) ^ { n } T ^ { \\pi } Q _ { 1 } ( s , a ) \\leq ( T ^ { \\mu } ) ^ { n } T ^ { \\pi } Q _ { 2 } ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 1123, + 770, + 1161, + 770, + 1161, + 798, + 1123, + 798 + ], + "score": 0.72, + "latex": "\\mathcal { T } ^ { \\pi }" + }, + { + "category_id": 14, + "poly": [ + 398, + 767, + 861, + 767, + 861, + 804, + 398, + 804 + ], + "score": 0.38, + "latex": "( T ^ { \\mu } ) ^ { n } T ^ { \\pi } Q _ { 1 } ( s , a ) \\leq ( T ^ { \\mu } ) ^ { n } T ^ { \\pi } Q _ { 2 } ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 295, + 605, + 313, + 605, + 313, + 631, + 295, + 631 + ], + "score": 0.3, + "latex": "^ 3" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 2007.0, + 1403.0, + 2007.0, + 1403.0, + 2033.0, + 1378.0, + 2033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 223.0, + 720.0, + 223.0, + 720.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 595.0, + 294.0, + 595.0, + 294.0, + 642.0, + 290.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 595.0, + 605.0, + 595.0, + 605.0, + 642.0, + 314.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 595.0, + 673.0, + 595.0, + 673.0, + 642.0, + 646.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 595.0, + 858.0, + 595.0, + 858.0, + 642.0, + 700.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 595.0, + 1333.0, + 595.0, + 1333.0, + 642.0, + 1228.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 595.0, + 1410.0, + 595.0, + 1410.0, + 642.0, + 1400.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 628.0, + 352.0, + 628.0, + 352.0, + 670.0, + 293.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 628.0, + 689.0, + 628.0, + 689.0, + 670.0, + 440.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 628.0, + 778.0, + 628.0, + 778.0, + 670.0, + 727.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 628.0, + 967.0, + 628.0, + 967.0, + 670.0, + 820.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 381.0, + 1197.0, + 381.0, + 1197.0, + 424.0, + 292.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 381.0, + 1406.0, + 381.0, + 1406.0, + 424.0, + 1237.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 415.0, + 494.0, + 415.0, + 494.0, + 452.0, + 294.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 290.0, + 1239.0, + 290.0, + 1239.0, + 332.0, + 294.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 913.0, + 371.0, + 913.0, + 371.0, + 953.0, + 295.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 913.0, + 654.0, + 913.0, + 654.0, + 953.0, + 545.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1662.0, + 398.0, + 1662.0, + 398.0, + 1701.0, + 295.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1662.0, + 536.0, + 1662.0, + 536.0, + 1701.0, + 526.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1612.0, + 510.0, + 1612.0, + 510.0, + 1658.0, + 295.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1612.0, + 799.0, + 1612.0, + 799.0, + 1658.0, + 788.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 808.0, + 458.0, + 808.0, + 458.0, + 846.0, + 294.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1476.0, + 514.0, + 1476.0, + 514.0, + 1523.0, + 290.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1476.0, + 860.0, + 1476.0, + 860.0, + 1523.0, + 743.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1182.0, + 461.0, + 1182.0, + 461.0, + 1225.0, + 297.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 1182.0, + 634.0, + 1182.0, + 634.0, + 1225.0, + 502.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 1182.0, + 798.0, + 1182.0, + 798.0, + 1225.0, + 655.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1235.0, + 371.0, + 1235.0, + 371.0, + 1278.0, + 294.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1235.0, + 603.0, + 1235.0, + 603.0, + 1278.0, + 436.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1235.0, + 891.0, + 1235.0, + 891.0, + 1278.0, + 836.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 1235.0, + 1098.0, + 1235.0, + 1098.0, + 1278.0, + 991.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 544.0, + 325.0, + 544.0, + 325.0, + 584.0, + 294.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 544.0, + 503.0, + 544.0, + 503.0, + 584.0, + 367.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 544.0, + 669.0, + 544.0, + 669.0, + 584.0, + 524.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 443.0, + 874.0, + 443.0, + 874.0, + 495.0, + 292.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 443.0, + 924.0, + 443.0, + 924.0, + 495.0, + 913.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 493.0, + 324.0, + 493.0, + 324.0, + 542.0, + 290.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 493.0, + 675.0, + 493.0, + 675.0, + 542.0, + 365.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 493.0, + 908.0, + 493.0, + 908.0, + 542.0, + 800.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 493.0, + 1100.0, + 493.0, + 1100.0, + 542.0, + 1091.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1946.0, + 407.0, + 1946.0, + 407.0, + 1998.0, + 379.0, + 1998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 693.0, + 377.0, + 693.0, + 377.0, + 739.0, + 290.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 727.0, + 368.0, + 727.0, + 368.0, + 763.0, + 297.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 727.0, + 773.0, + 727.0, + 773.0, + 763.0, + 666.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1815.0, + 407.0, + 1815.0, + 407.0, + 1859.0, + 381.0, + 1859.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 949, + 1405, + 949, + 1405, + 1118, + 297, + 1118 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 227, + 1404, + 227, + 1404, + 354, + 297, + 354 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 302, + 615, + 1398, + 615, + 1398, + 842, + 302, + 842 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 289, + 361, + 1405, + 361, + 1405, + 427, + 289, + 427 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 296, + 567, + 974, + 567, + 974, + 601, + 296, + 601 + ], + "score": 0.927 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 815, + 74, + 815, + 105, + 299, + 105 + ], + "score": 0.913 + }, + { + "category_id": 9, + "poly": [ + 1353, + 838, + 1402, + 838, + 1402, + 869, + 1353, + 869 + ], + "score": 0.901 + }, + { + "category_id": 8, + "poly": [ + 409, + 436, + 1291, + 436, + 1291, + 481, + 409, + 481 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1353, + 444, + 1402, + 444, + 1402, + 474, + 1353, + 474 + ], + "score": 0.862 + }, + { + "category_id": 1, + "poly": [ + 299, + 496, + 1397, + 496, + 1397, + 538, + 299, + 538 + ], + "score": 0.85 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.845 + }, + { + "category_id": 2, + "poly": [ + 1374, + 888, + 1402, + 888, + 1402, + 917, + 1374, + 917 + ], + "score": 0.808 + }, + { + "category_id": 13, + "poly": [ + 511, + 988, + 609, + 988, + 609, + 1023, + 511, + 1023 + ], + "score": 0.93, + "latex": "n ^ { * } ( s , { \\pmb a } )" + }, + { + "category_id": 13, + "poly": [ + 1138, + 262, + 1229, + 262, + 1229, + 295, + 1138, + 295 + ], + "score": 0.93, + "latex": "Q ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 1301, + 1050, + 1396, + 1050, + 1396, + 1086, + 1301, + 1086 + ], + "score": 0.92, + "latex": "\\gamma ^ { n ^ { * } ( s , \\pmb { a } ) }" + }, + { + "category_id": 13, + "poly": [ + 1215, + 498, + 1388, + 498, + 1388, + 535, + 1215, + 535 + ], + "score": 0.92, + "latex": "( \\mathcal { T } ^ { \\mu } ) ^ { n ^ { * } ( s , a ) } \\mathcal { T } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 299, + 986, + 457, + 986, + 457, + 1023, + 299, + 1023 + ], + "score": 0.92, + "latex": "\\| Q - { \\tilde { Q } } _ { n ^ { * } } \\| _ { \\infty }" + }, + { + "category_id": 13, + "poly": [ + 942, + 496, + 992, + 496, + 992, + 534, + 942, + 534 + ], + "score": 0.92, + "latex": "{ \\tilde { Q } } _ { n ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 756, + 363, + 1289, + 363, + 1289, + 399, + 756, + 399 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { { \\cal T } _ { \\mu } ^ { \\pi } Q ( s , a ) : = \\operatorname* { m a x } _ { 0 \\leq n \\leq N } \\{ ( { \\cal T } ^ { \\mu } ) ^ { n } { \\cal T } ^ { \\pi } Q ( s , a ) \\} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 352, + 498, + 890, + 498, + 890, + 539, + 352, + 539 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { n ^ { \\ast } ( \\pmb { \\mathscr { s } } , \\pmb { a } ) = \\arg \\operatorname* { m a x } _ { 0 \\leq n \\leq N } \\{ ( \\mathscr { T } ^ { \\mu } ) ^ { n } \\mathscr { T } ^ { \\pi } Q ( \\pmb { \\mathscr { s } } , \\pmb { a } ) \\} } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 310, + 610, + 1400, + 610, + 1400, + 846, + 310, + 846 + ], + "score": 0.9, + "latex": "\\begin{array} { r l } & { \\ | T _ { \\mu } ^ { \\pi } Q ( s , a ) - Q ^ { * } ( s , a ) | } \\\\ & { = | ( T ^ { \\mu } ) ^ { n ^ { * } ( s , a ) } T ^ { \\pi } Q ( s , a ) - Q ^ { * } ( s , a ) | } \\\\ & { \\leq | ( T ^ { \\mu } ) ^ { n ^ { * } ( s , a ) } T ^ { \\pi } Q ( s , a ) - ( T ^ { \\mu } ) ^ { n ^ { * } ( s , a ) } T ^ { \\pi } \\tilde { Q } _ { n ^ { * } } ( s , a ) | + | ( T ^ { \\mu } ) ^ { n ^ { * } ( s , a ) } T ^ { \\pi } \\tilde { Q } _ { n ^ { * } } ( s , a ) - Q ^ { * } ( s , a ) | } \\\\ & { = | ( T ^ { \\mu } ) ^ { n ^ { * } ( s , a ) } T ^ { \\pi } Q ( s , a ) - ( T ^ { \\mu } ) ^ { n ^ { * } ( s , a ) } T ^ { \\pi } \\tilde { Q } _ { n ^ { * } } ( s , a ) | + | \\tilde { Q } _ { n ^ { * } } ( s , a ) - Q ^ { * } ( s , a ) | } \\\\ & { \\leq \\gamma ^ { n ^ { * } ( s , a ) } \\| Q - \\tilde { Q } _ { n ^ { * } } \\| _ { \\infty } + \\| \\tilde { Q } _ { n ^ { * } } - Q ^ { * } \\| _ { \\infty } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 475, + 293, + 513, + 293, + 513, + 329, + 475, + 329 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 592, + 1054, + 631, + 1054, + 631, + 1090, + 592, + 1090 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 14, + "poly": [ + 409, + 435, + 1285, + 435, + 1285, + 481, + 409, + 481 + ], + "score": 0.84, + "latex": "\\begin{array} { r } { | { \\cal T } _ { \\mu } ^ { \\pi } Q ( s , a ) - Q ^ { * } ( s , a ) | \\leq { \\gamma } ^ { n ^ { * } ( s , a ) } \\| Q - \\tilde { Q } _ { n ^ { * } } \\| _ { \\infty } + \\| \\tilde { Q } _ { n ^ { * } } - Q ^ { * } \\| _ { \\infty } , \\quad \\forall s , a , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 373, + 956, + 399, + 956, + 399, + 986, + 373, + 986 + ], + "score": 0.83, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 828, + 950, + 1002, + 950, + 1002, + 989, + 828, + 989 + ], + "score": 0.82, + "latex": "\\| { \\tilde { Q } } _ { n ^ { * } } - Q ^ { * } \\| _ { \\infty }" + }, + { + "category_id": 13, + "poly": [ + 714, + 950, + 812, + 950, + 812, + 987, + 714, + 987 + ], + "score": 0.73, + "latex": "\\tilde { Q } _ { n ^ { * } } , Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 479, + 1023, + 505, + 1023, + 505, + 1052, + 479, + 1052 + ], + "score": 0.3, + "latex": "\\mathrm { \\bf Q }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 868.0, + 2084.0, + 868.0, + 2122.0, + 831.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 893.0, + 1401.0, + 893.0, + 1401.0, + 916.0, + 1379.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 951.0, + 372.0, + 951.0, + 372.0, + 991.0, + 292.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 951.0, + 713.0, + 951.0, + 713.0, + 991.0, + 400.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 951.0, + 827.0, + 951.0, + 827.0, + 991.0, + 813.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 951.0, + 1406.0, + 951.0, + 1406.0, + 991.0, + 1003.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 987.0, + 298.0, + 987.0, + 298.0, + 1027.0, + 295.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 987.0, + 510.0, + 987.0, + 510.0, + 1027.0, + 458.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 987.0, + 1406.0, + 987.0, + 1406.0, + 1027.0, + 610.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1019.0, + 478.0, + 1019.0, + 478.0, + 1055.0, + 294.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1019.0, + 1405.0, + 1019.0, + 1405.0, + 1055.0, + 506.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1050.0, + 591.0, + 1050.0, + 591.0, + 1089.0, + 290.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 1050.0, + 1300.0, + 1050.0, + 1300.0, + 1089.0, + 632.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1050.0, + 1408.0, + 1050.0, + 1408.0, + 1089.0, + 1397.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1085.0, + 1030.0, + 1085.0, + 1030.0, + 1119.0, + 295.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 229.0, + 1404.0, + 229.0, + 1404.0, + 265.0, + 294.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 262.0, + 1137.0, + 262.0, + 1137.0, + 295.0, + 295.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 262.0, + 1404.0, + 262.0, + 1404.0, + 295.0, + 1230.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 291.0, + 474.0, + 291.0, + 474.0, + 331.0, + 292.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 291.0, + 1407.0, + 291.0, + 1407.0, + 331.0, + 514.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 319.0, + 396.0, + 319.0, + 396.0, + 360.0, + 289.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 356.0, + 755.0, + 356.0, + 755.0, + 403.0, + 291.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 356.0, + 1406.0, + 356.0, + 1406.0, + 403.0, + 1290.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 388.0, + 370.0, + 388.0, + 370.0, + 431.0, + 291.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 565.0, + 974.0, + 565.0, + 974.0, + 605.0, + 296.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 494.0, + 351.0, + 494.0, + 351.0, + 542.0, + 288.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 494.0, + 941.0, + 494.0, + 941.0, + 542.0, + 891.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 494.0, + 1214.0, + 494.0, + 1214.0, + 542.0, + 993.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1389.0, + 494.0, + 1401.0, + 494.0, + 1401.0, + 542.0, + 1389.0, + 542.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 294, + 763, + 1400, + 763, + 1400, + 1358, + 294, + 1358 + ], + "score": 0.986, + "html": "
DatasetEnvironmentSfBCSfBC +GaussianSfBC + VAESfBC - Planning
Medium-ExpertHalfCheetah92.6 ± 0.579.4 ± 1.485.2 ± 2.991.4± 0.6
Medium-ExpertHopper108.6± 2.1107.8± 7.892.0 ± 7.3109.0 ±1.0
Medium-ExpertWalker109.8±0.271.5 ± 1.5109.3± 2.5109.4 ± 0.9
MediumHalfCheetah45.9 ± 2.242.0± 0.243.4± 0.142.4±0.2
MediumHopper57.1 ± 4.158.1 ± 1.565.6 ± 3.360.1 ± 4.2
MediumWalker77.9 ± 2.582.4± 1.179.1 ± 2.580.3 ± 0.9
Medium-ReplayHalfCheetah37.1 ± 1.736.2±1.242.4± 0.537.5 ± 0.6
Medium-ReplayHopper86.2 ± 9.167.8 ± 6.558.6± 4.858.6 ± 1.3
Medium-ReplayWalker65.1± 5.665.8± 4.462.2 ± 4.362.6 ± 2.2
Average75.667.970.972.3
DefaultAntMaze-umaze92.0± 2.193.3± 2.491.6± 2.496.7 ± 4.7
DiverseAntMaze-umaze85.3±3.688.3±2.478.3 ± 4.780.0 ± 10.8
PlayAntMaze-medium81.3 ± 2.680.0 ± 4.168.3± 2.435.0 ± 4.1
DiverseAntMaze-medium82.0±3.185.0 ± 7.165.0 ± 7.133.3 ± 6.2
PlayAntMaze-large59.3 ± 14.343.3 ± 7.135.0 ±8.28.3±8.5
DiverseAntMaze-large45.5±6.626.7±8.520.0±0.06.7 ± 4.7
Average74.269.459.743.3
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1610, + 1405, + 1610, + 1405, + 1979, + 297, + 1979 + ], + "score": 0.983 + }, + { + "category_id": 5, + "poly": [ + 294, + 302, + 1399, + 302, + 1399, + 599, + 294, + 599 + ], + "score": 0.983, + "html": "
DatasetEnvironmentSfBC (Ours)IQLVEMAWRBAILBCQCQLDTDiffuser
SparseMaze2d-umaze73.9 ± 6.647.4-1.0-12.85.7-113.9
SparseMaze2d-medium73.8±2.934.9■7.6-8.35.0121.5
SparseMaze2d-large74.4 ± 1.758.6=23.7=6.212.5123.0
Average (Maze2d)74.050.010.8■9.17.7=119.5
CompleteFrankaKitchen77.9 ± 0.662.50.0■8.1143.8=
PartialFrankaKitchen47.9 ± 4.146.315.4=18.949.8
MixedFrankaKitchen45.4 ± 1.651.010.6■8.151.0
Average (FrankaKitchen)57.153.3-8.7-11.748.2■■
" + }, + { + "category_id": 0, + "poly": [ + 299, + 1542, + 901, + 1542, + 901, + 1579, + 299, + 1579 + ], + "score": 0.914 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.898 + }, + { + "category_id": 6, + "poly": [ + 296, + 621, + 1404, + 621, + 1404, + 714, + 296, + 714 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.847 + }, + { + "category_id": 6, + "poly": [ + 298, + 226, + 867, + 226, + 867, + 263, + 298, + 263 + ], + "score": 0.669 + }, + { + "category_id": 6, + "poly": [ + 295, + 1380, + 1407, + 1380, + 1407, + 1474, + 295, + 1474 + ], + "score": 0.447 + }, + { + "category_id": 0, + "poly": [ + 298, + 226, + 867, + 226, + 867, + 263, + 298, + 263 + ], + "score": 0.332 + }, + { + "category_id": 13, + "poly": [ + 1367, + 653, + 1402, + 653, + 1402, + 681, + 1367, + 681 + ], + "score": 0.31, + "latex": "\\mathrm { F u }" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1539.0, + 905.0, + 1539.0, + 905.0, + 1584.0, + 292.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 618.0, + 1406.0, + 618.0, + 1406.0, + 658.0, + 291.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 652.0, + 1366.0, + 652.0, + 1366.0, + 686.0, + 295.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 682.0, + 447.0, + 682.0, + 447.0, + 714.0, + 292.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 225.0, + 868.0, + 225.0, + 868.0, + 266.0, + 295.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1378.0, + 1405.0, + 1378.0, + 1405.0, + 1416.0, + 293.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1412.0, + 1404.0, + 1412.0, + 1404.0, + 1443.0, + 297.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1442.0, + 1029.0, + 1442.0, + 1029.0, + 1475.0, + 294.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 225.0, + 868.0, + 225.0, + 868.0, + 266.0, + 295.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1608.0, + 1406.0, + 1608.0, + 1406.0, + 1649.0, + 292.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1642.0, + 1405.0, + 1642.0, + 1405.0, + 1678.0, + 292.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1674.0, + 1405.0, + 1674.0, + 1405.0, + 1708.0, + 295.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1700.0, + 1406.0, + 1700.0, + 1406.0, + 1741.0, + 292.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1735.0, + 1407.0, + 1735.0, + 1407.0, + 1769.0, + 295.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1762.0, + 1410.0, + 1762.0, + 1410.0, + 1801.0, + 292.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1792.0, + 1408.0, + 1792.0, + 1408.0, + 1830.0, + 294.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1824.0, + 1405.0, + 1824.0, + 1405.0, + 1860.0, + 294.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1854.0, + 1406.0, + 1854.0, + 1406.0, + 1892.0, + 294.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1885.0, + 1407.0, + 1885.0, + 1407.0, + 1922.0, + 291.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1917.0, + 1406.0, + 1917.0, + 1406.0, + 1951.0, + 295.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1947.0, + 1403.0, + 1947.0, + 1403.0, + 1984.0, + 292.0, + 1984.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 346, + 293, + 1350, + 293, + 1350, + 1669, + 346, + 1669 + ], + "score": 0.975 + }, + { + "category_id": 4, + "poly": [ + 297, + 1680, + 1405, + 1680, + 1405, + 1776, + 297, + 1776 + ], + "score": 0.953 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 866, + 2088, + 866, + 2113, + 834, + 2113 + ], + "score": 0.856 + }, + { + "category_id": 0, + "poly": [ + 299, + 227, + 624, + 227, + 624, + 261, + 299, + 261 + ], + "score": 0.663 + }, + { + "category_id": 2, + "poly": [ + 299, + 227, + 624, + 227, + 624, + 261, + 299, + 261 + ], + "score": 0.181 + }, + { + "category_id": 15, + "poly": [ + 395.0, + 308.0, + 654.0, + 308.0, + 654.0, + 332.0, + 395.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 314.0, + 973.0, + 314.0, + 973.0, + 338.0, + 759.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 314.0, + 1318.0, + 314.0, + 1318.0, + 338.0, + 1086.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 332.0, + 727.0, + 332.0, + 727.0, + 358.0, + 683.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 357.0, + 378.0, + 357.0, + 378.0, + 381.0, + 348.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 354.0, + 1060.0, + 354.0, + 1060.0, + 377.0, + 1022.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 368.0, + 723.0, + 368.0, + 723.0, + 391.0, + 685.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 398.0, + 376.0, + 398.0, + 376.0, + 421.0, + 349.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 395.0, + 668.0, + 395.0, + 668.0, + 425.0, + 598.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 401.0, + 721.0, + 401.0, + 721.0, + 424.0, + 693.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 399.0, + 1004.0, + 399.0, + 1004.0, + 428.0, + 935.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 389.0, + 1060.0, + 389.0, + 1060.0, + 412.0, + 1030.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 399.0, + 1341.0, + 399.0, + 1341.0, + 428.0, + 1272.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 413.0, + 667.0, + 413.0, + 667.0, + 443.0, + 597.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 418.0, + 1004.0, + 418.0, + 1004.0, + 446.0, + 935.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 424.0, + 1061.0, + 424.0, + 1061.0, + 447.0, + 1030.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 418.0, + 1341.0, + 418.0, + 1341.0, + 446.0, + 1272.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 441.0, + 376.0, + 441.0, + 376.0, + 460.0, + 349.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 437.0, + 666.0, + 437.0, + 666.0, + 461.0, + 597.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 435.0, + 723.0, + 435.0, + 723.0, + 458.0, + 693.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 440.0, + 1003.0, + 440.0, + 1003.0, + 464.0, + 936.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 440.0, + 1340.0, + 440.0, + 1340.0, + 464.0, + 1273.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 453.0, + 660.0, + 453.0, + 660.0, + 483.0, + 597.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 469.0, + 723.0, + 469.0, + 723.0, + 492.0, + 694.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 456.0, + 998.0, + 456.0, + 998.0, + 485.0, + 935.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 460.0, + 1061.0, + 460.0, + 1061.0, + 483.0, + 1030.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 456.0, + 1335.0, + 456.0, + 1335.0, + 486.0, + 1273.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 480.0, + 378.0, + 480.0, + 378.0, + 504.0, + 348.0, + 504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 503.0, + 723.0, + 503.0, + 723.0, + 527.0, + 694.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 495.0, + 1060.0, + 495.0, + 1060.0, + 518.0, + 1030.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 523.0, + 376.0, + 523.0, + 376.0, + 543.0, + 357.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 532.0, + 1057.0, + 532.0, + 1057.0, + 547.0, + 1042.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 538.0, + 723.0, + 538.0, + 723.0, + 560.0, + 704.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 554.0, + 384.0, + 554.0, + 384.0, + 569.0, + 370.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 549.0, + 467.0, + 549.0, + 467.0, + 574.0, + 437.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 550.0, + 544.0, + 550.0, + 544.0, + 574.0, + 508.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 549.0, + 619.0, + 549.0, + 619.0, + 572.0, + 581.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 554.0, + 729.0, + 554.0, + 729.0, + 569.0, + 715.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 550.0, + 809.0, + 550.0, + 809.0, + 574.0, + 779.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 550.0, + 884.0, + 550.0, + 884.0, + 574.0, + 848.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 550.0, + 956.0, + 550.0, + 956.0, + 572.0, + 920.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 554.0, + 1066.0, + 554.0, + 1066.0, + 569.0, + 1052.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 550.0, + 1146.0, + 550.0, + 1146.0, + 574.0, + 1116.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 550.0, + 1221.0, + 550.0, + 1221.0, + 574.0, + 1185.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 550.0, + 1293.0, + 550.0, + 1293.0, + 574.0, + 1257.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 580.0, + 624.0, + 580.0, + 624.0, + 600.0, + 426.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 578.0, + 944.0, + 578.0, + 944.0, + 603.0, + 781.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 577.0, + 1289.0, + 577.0, + 1289.0, + 601.0, + 1108.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 590.0, + 378.0, + 590.0, + 378.0, + 615.0, + 348.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 603.0, + 715.0, + 603.0, + 715.0, + 629.0, + 686.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 609.0, + 1053.0, + 609.0, + 1053.0, + 632.0, + 1022.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 634.0, + 376.0, + 634.0, + 376.0, + 652.0, + 349.0, + 652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 634.0, + 715.0, + 634.0, + 715.0, + 658.0, + 686.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 672.0, + 378.0, + 672.0, + 378.0, + 695.0, + 349.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 665.0, + 668.0, + 665.0, + 668.0, + 694.0, + 598.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 663.0, + 713.0, + 663.0, + 713.0, + 686.0, + 686.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 665.0, + 1004.0, + 665.0, + 1004.0, + 696.0, + 935.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 657.0, + 1052.0, + 657.0, + 1052.0, + 680.0, + 1022.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 665.0, + 1341.0, + 665.0, + 1341.0, + 694.0, + 1272.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 687.0, + 666.0, + 687.0, + 666.0, + 712.0, + 597.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 692.0, + 715.0, + 692.0, + 715.0, + 715.0, + 688.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 684.0, + 1004.0, + 684.0, + 1004.0, + 714.0, + 934.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 684.0, + 1341.0, + 684.0, + 1341.0, + 714.0, + 1271.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 712.0, + 378.0, + 712.0, + 378.0, + 735.0, + 348.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 704.0, + 1004.0, + 704.0, + 1004.0, + 732.0, + 934.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 704.0, + 1052.0, + 704.0, + 1052.0, + 727.0, + 1022.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 707.0, + 1340.0, + 707.0, + 1340.0, + 732.0, + 1271.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 726.0, + 660.0, + 726.0, + 660.0, + 752.0, + 597.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 723.0, + 713.0, + 723.0, + 713.0, + 746.0, + 686.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 726.0, + 995.0, + 726.0, + 995.0, + 752.0, + 934.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 726.0, + 1333.0, + 726.0, + 1333.0, + 752.0, + 1271.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 752.0, + 378.0, + 752.0, + 378.0, + 775.0, + 348.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 750.0, + 715.0, + 750.0, + 715.0, + 775.0, + 686.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 750.0, + 1052.0, + 750.0, + 1052.0, + 775.0, + 1022.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 780.0, + 715.0, + 780.0, + 715.0, + 803.0, + 686.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 780.0, + 997.0, + 780.0, + 997.0, + 790.0, + 983.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 793.0, + 376.0, + 793.0, + 376.0, + 813.0, + 357.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 801.0, + 1050.0, + 801.0, + 1050.0, + 820.0, + 1031.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 824.0, + 384.0, + 824.0, + 384.0, + 840.0, + 370.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 820.0, + 465.0, + 820.0, + 465.0, + 843.0, + 437.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 821.0, + 544.0, + 821.0, + 544.0, + 843.0, + 508.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 820.0, + 619.0, + 820.0, + 619.0, + 843.0, + 583.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 823.0, + 724.0, + 823.0, + 724.0, + 841.0, + 705.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 820.0, + 803.0, + 820.0, + 803.0, + 843.0, + 774.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 821.0, + 881.0, + 821.0, + 881.0, + 843.0, + 845.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 820.0, + 955.0, + 820.0, + 955.0, + 843.0, + 919.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 824.0, + 1058.0, + 824.0, + 1058.0, + 840.0, + 1044.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 820.0, + 1140.0, + 820.0, + 1140.0, + 843.0, + 1110.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 821.0, + 1218.0, + 821.0, + 1218.0, + 844.0, + 1182.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 821.0, + 1292.0, + 821.0, + 1292.0, + 843.0, + 1256.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 847.0, + 654.0, + 847.0, + 654.0, + 872.0, + 396.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 852.0, + 975.0, + 852.0, + 975.0, + 879.0, + 755.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 847.0, + 1315.0, + 847.0, + 1315.0, + 872.0, + 1082.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 870.0, + 378.0, + 870.0, + 378.0, + 893.0, + 348.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 889.0, + 671.0, + 889.0, + 671.0, + 901.0, + 657.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 866.0, + 721.0, + 866.0, + 721.0, + 889.0, + 685.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 878.0, + 1052.0, + 878.0, + 1052.0, + 901.0, + 1022.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 916.0, + 376.0, + 916.0, + 376.0, + 939.0, + 349.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 906.0, + 724.0, + 906.0, + 724.0, + 930.0, + 693.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 926.0, + 1052.0, + 926.0, + 1052.0, + 949.0, + 1022.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 936.0, + 668.0, + 936.0, + 668.0, + 966.0, + 598.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 947.0, + 723.0, + 947.0, + 723.0, + 972.0, + 693.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 963.0, + 378.0, + 963.0, + 378.0, + 1032.0, + 348.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 958.0, + 668.0, + 958.0, + 668.0, + 1003.0, + 596.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 989.0, + 724.0, + 989.0, + 724.0, + 1012.0, + 693.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1000.0, + 831.0, + 1000.0, + 831.0, + 1029.0, + 761.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 972.0, + 1052.0, + 972.0, + 1052.0, + 996.0, + 1024.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 997.0, + 1341.0, + 997.0, + 1341.0, + 1028.0, + 1272.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1030.0, + 723.0, + 1030.0, + 723.0, + 1055.0, + 693.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1019.0, + 831.0, + 1019.0, + 831.0, + 1068.0, + 761.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1021.0, + 1055.0, + 1021.0, + 1055.0, + 1044.0, + 1024.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1021.0, + 1341.0, + 1021.0, + 1341.0, + 1066.0, + 1271.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1056.0, + 376.0, + 1056.0, + 376.0, + 1075.0, + 357.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1072.0, + 730.0, + 1072.0, + 730.0, + 1095.0, + 702.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1070.0, + 1050.0, + 1070.0, + 1050.0, + 1090.0, + 1031.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1061.0, + 1333.0, + 1061.0, + 1333.0, + 1084.0, + 1271.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1095.0, + 384.0, + 1095.0, + 384.0, + 1110.0, + 370.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 1090.0, + 467.0, + 1090.0, + 467.0, + 1113.0, + 437.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1092.0, + 544.0, + 1092.0, + 544.0, + 1113.0, + 508.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1090.0, + 619.0, + 1090.0, + 619.0, + 1113.0, + 581.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1095.0, + 729.0, + 1095.0, + 729.0, + 1110.0, + 715.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1090.0, + 809.0, + 1090.0, + 809.0, + 1115.0, + 779.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1092.0, + 884.0, + 1092.0, + 884.0, + 1115.0, + 848.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1090.0, + 956.0, + 1090.0, + 956.0, + 1113.0, + 920.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1095.0, + 1058.0, + 1095.0, + 1058.0, + 1110.0, + 1044.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1090.0, + 1140.0, + 1090.0, + 1140.0, + 1113.0, + 1111.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 1090.0, + 1218.0, + 1090.0, + 1218.0, + 1113.0, + 1182.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 1090.0, + 1292.0, + 1090.0, + 1292.0, + 1113.0, + 1256.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1126.0, + 608.0, + 1126.0, + 608.0, + 1147.0, + 446.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1118.0, + 975.0, + 1118.0, + 975.0, + 1142.0, + 748.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 1124.0, + 1320.0, + 1124.0, + 1320.0, + 1149.0, + 1085.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1136.0, + 385.0, + 1136.0, + 385.0, + 1159.0, + 349.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1141.0, + 1060.0, + 1141.0, + 1060.0, + 1162.0, + 1024.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1162.0, + 715.0, + 1162.0, + 715.0, + 1185.0, + 685.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1176.0, + 384.0, + 1176.0, + 384.0, + 1195.0, + 359.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1179.0, + 1060.0, + 1179.0, + 1060.0, + 1202.0, + 1030.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1213.0, + 387.0, + 1213.0, + 387.0, + 1236.0, + 356.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 1213.0, + 495.0, + 1213.0, + 495.0, + 1258.0, + 426.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1204.0, + 723.0, + 1204.0, + 723.0, + 1225.0, + 686.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1207.0, + 916.0, + 1207.0, + 916.0, + 1253.0, + 846.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1219.0, + 1060.0, + 1219.0, + 1060.0, + 1242.0, + 1030.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 1210.0, + 1343.0, + 1210.0, + 1343.0, + 1258.0, + 1272.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1250.0, + 385.0, + 1250.0, + 385.0, + 1273.0, + 357.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 1248.0, + 494.0, + 1248.0, + 494.0, + 1295.0, + 426.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1245.0, + 716.0, + 1245.0, + 716.0, + 1269.0, + 685.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1247.0, + 915.0, + 1247.0, + 915.0, + 1293.0, + 845.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1259.0, + 1060.0, + 1259.0, + 1060.0, + 1282.0, + 1030.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 1248.0, + 1341.0, + 1248.0, + 1341.0, + 1293.0, + 1272.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1288.0, + 407.0, + 1288.0, + 407.0, + 1322.0, + 356.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 1295.0, + 432.0, + 1295.0, + 432.0, + 1313.0, + 412.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1289.0, + 668.0, + 1289.0, + 668.0, + 1316.0, + 567.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1287.0, + 716.0, + 1287.0, + 716.0, + 1310.0, + 685.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1308.0, + 931.0, + 1308.0, + 931.0, + 1324.0, + 898.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1299.0, + 1060.0, + 1299.0, + 1060.0, + 1322.0, + 1030.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1328.0, + 387.0, + 1328.0, + 387.0, + 1348.0, + 367.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1328.0, + 715.0, + 1328.0, + 715.0, + 1348.0, + 694.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1341.0, + 1058.0, + 1341.0, + 1058.0, + 1361.0, + 1039.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 1365.0, + 392.0, + 1365.0, + 392.0, + 1381.0, + 378.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1362.0, + 475.0, + 1362.0, + 475.0, + 1384.0, + 439.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1362.0, + 547.0, + 1362.0, + 547.0, + 1384.0, + 511.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1362.0, + 619.0, + 1362.0, + 619.0, + 1384.0, + 583.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1364.0, + 724.0, + 1364.0, + 724.0, + 1382.0, + 705.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1362.0, + 807.0, + 1362.0, + 807.0, + 1384.0, + 770.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1361.0, + 881.0, + 1361.0, + 881.0, + 1384.0, + 843.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1361.0, + 955.0, + 1361.0, + 955.0, + 1384.0, + 917.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1365.0, + 1066.0, + 1365.0, + 1066.0, + 1381.0, + 1052.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1362.0, + 1135.0, + 1362.0, + 1135.0, + 1384.0, + 1099.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 1362.0, + 1191.0, + 1362.0, + 1191.0, + 1384.0, + 1155.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1362.0, + 1249.0, + 1362.0, + 1249.0, + 1384.0, + 1213.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1362.0, + 1307.0, + 1362.0, + 1307.0, + 1384.0, + 1271.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1389.0, + 634.0, + 1389.0, + 634.0, + 1413.0, + 413.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 1388.0, + 969.0, + 1388.0, + 969.0, + 1413.0, + 754.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1387.0, + 1296.0, + 1387.0, + 1296.0, + 1414.0, + 1099.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1425.0, + 376.0, + 1425.0, + 376.0, + 1495.0, + 348.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 1468.0, + 668.0, + 1468.0, + 668.0, + 1506.0, + 597.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1427.0, + 715.0, + 1427.0, + 715.0, + 1511.0, + 685.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1414.0, + 827.0, + 1414.0, + 827.0, + 1501.0, + 755.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1445.0, + 1052.0, + 1445.0, + 1052.0, + 1470.0, + 1022.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1414.0, + 1164.0, + 1414.0, + 1164.0, + 1503.0, + 1092.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1498.0, + 1052.0, + 1498.0, + 1052.0, + 1521.0, + 1022.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1516.0, + 376.0, + 1516.0, + 376.0, + 1541.0, + 348.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1516.0, + 668.0, + 1516.0, + 668.0, + 1544.0, + 596.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 1521.0, + 715.0, + 1521.0, + 715.0, + 1539.0, + 688.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 1534.0, + 659.0, + 1534.0, + 659.0, + 1564.0, + 597.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 1551.0, + 713.0, + 1551.0, + 713.0, + 1570.0, + 688.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1550.0, + 1052.0, + 1550.0, + 1052.0, + 1573.0, + 1022.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1562.0, + 376.0, + 1562.0, + 376.0, + 1587.0, + 348.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 1582.0, + 713.0, + 1582.0, + 713.0, + 1601.0, + 688.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1610.0, + 376.0, + 1610.0, + 376.0, + 1630.0, + 357.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1611.0, + 713.0, + 1611.0, + 713.0, + 1631.0, + 694.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1605.0, + 1049.0, + 1605.0, + 1049.0, + 1625.0, + 1031.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1636.0, + 384.0, + 1636.0, + 384.0, + 1651.0, + 370.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1631.0, + 454.0, + 1631.0, + 454.0, + 1654.0, + 418.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1631.0, + 514.0, + 1631.0, + 514.0, + 1654.0, + 478.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1631.0, + 574.0, + 1631.0, + 574.0, + 1654.0, + 536.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1631.0, + 633.0, + 1631.0, + 633.0, + 1654.0, + 596.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1633.0, + 723.0, + 1633.0, + 723.0, + 1653.0, + 705.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 1631.0, + 792.0, + 1631.0, + 792.0, + 1654.0, + 754.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 1631.0, + 851.0, + 1631.0, + 851.0, + 1654.0, + 814.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1631.0, + 909.0, + 1631.0, + 909.0, + 1654.0, + 873.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 1631.0, + 969.0, + 1631.0, + 969.0, + 1654.0, + 933.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1636.0, + 1058.0, + 1636.0, + 1058.0, + 1651.0, + 1044.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1631.0, + 1129.0, + 1631.0, + 1129.0, + 1654.0, + 1093.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 1631.0, + 1188.0, + 1631.0, + 1188.0, + 1654.0, + 1151.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1631.0, + 1248.0, + 1631.0, + 1248.0, + 1654.0, + 1210.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1631.0, + 1306.0, + 1631.0, + 1306.0, + 1654.0, + 1268.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 647.5, + 746.0, + 647.5, + 746.0, + 664.5, + 718.0, + 664.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.75, + 659.5, + 745.75, + 659.5, + 745.75, + 676.5, + 716.75, + 676.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 705.0, + 666.0, + 705.0, + 666.0, + 731.0, + 597.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.25, + 995.5, + 658.25, + 995.5, + 658.25, + 1022.0, + 597.25, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1059.0, + 821.0, + 1059.0, + 821.0, + 1085.0, + 762.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1292.5, + 583.0, + 1292.5, + 583.0, + 1313.5, + 541.0, + 1313.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 1293.5, + 483.0, + 1293.5, + 483.0, + 1317.5, + 442.0, + 1317.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 1454.0, + 1162.0, + 1454.0, + 1162.0, + 1480.0, + 1094.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 1497.0, + 666.0, + 1497.0, + 666.0, + 1523.0, + 597.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1679.0, + 1405.0, + 1679.0, + 1405.0, + 1716.0, + 292.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1712.0, + 1405.0, + 1712.0, + 1405.0, + 1746.0, + 294.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1740.0, + 1381.0, + 1740.0, + 1381.0, + 1777.0, + 294.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 2083.0, + 872.0, + 2083.0, + 872.0, + 2123.0, + 828.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 222.0, + 629.0, + 222.0, + 629.0, + 268.0, + 292.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 222.0, + 629.0, + 222.0, + 629.0, + 268.0, + 292.0, + 268.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1142, + 1404, + 1142, + 1404, + 1266, + 297, + 1266 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 708, + 1405, + 708, + 1405, + 863, + 296, + 863 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 877, + 1406, + 877, + 1406, + 1002, + 296, + 1002 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1513, + 1406, + 1513, + 1406, + 1605, + 297, + 1605 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 493, + 1834, + 1207, + 1834, + 1207, + 2035, + 493, + 2035 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 298, + 1356, + 1403, + 1356, + 1403, + 1450, + 298, + 1450 + ], + "score": 0.96 + }, + { + "category_id": 8, + "poly": [ + 486, + 1453, + 1211, + 1453, + 1211, + 1507, + 486, + 1507 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 291, + 1685, + 1402, + 1685, + 1402, + 1754, + 291, + 1754 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 297, + 293, + 1400, + 293, + 1400, + 357, + 297, + 357 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 299, + 1015, + 1399, + 1015, + 1399, + 1080, + 299, + 1080 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 299, + 447, + 1402, + 447, + 1402, + 511, + 299, + 511 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 621, + 1085, + 1080, + 1085, + 1080, + 1136, + 621, + 1136 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 645, + 514, + 1053, + 514, + 1053, + 564, + 645, + 564 + ], + "score": 0.936 + }, + { + "category_id": 8, + "poly": [ + 477, + 1609, + 1221, + 1609, + 1221, + 1665, + 477, + 1665 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 292, + 1766, + 1402, + 1766, + 1402, + 1830, + 292, + 1830 + ], + "score": 0.934 + }, + { + "category_id": 0, + "poly": [ + 297, + 1299, + 458, + 1299, + 458, + 1332, + 297, + 1332 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 295, + 570, + 1215, + 570, + 1215, + 603, + 295, + 603 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 816, + 74, + 816, + 106, + 298, + 106 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 299, + 226, + 823, + 226, + 823, + 262, + 299, + 262 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 297, + 389, + 441, + 389, + 441, + 422, + 297, + 422 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1353, + 518, + 1402, + 518, + 1402, + 548, + 1353, + 548 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1354, + 1089, + 1401, + 1089, + 1401, + 1119, + 1354, + 1119 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1457, + 1401, + 1457, + 1401, + 1487, + 1353, + 1487 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1354, + 1612, + 1401, + 1612, + 1401, + 1643, + 1354, + 1643 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1355, + 682, + 1401, + 682, + 1401, + 709, + 1355, + 709 + ], + "score": 0.853 + }, + { + "category_id": 8, + "poly": [ + 288, + 611, + 1380, + 611, + 1380, + 682, + 288, + 682 + ], + "score": 0.842 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 863, + 2088, + 863, + 2113, + 834, + 2113 + ], + "score": 0.597 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 863, + 2088, + 863, + 2113, + 834, + 2113 + ], + "score": 0.244 + }, + { + "category_id": 14, + "poly": [ + 642, + 514, + 1055, + 514, + 1055, + 568, + 642, + 568 + ], + "score": 0.94, + "latex": "{ \\mathcal { T } } _ { \\mathrm { v e m } } V ( s ) : = \\operatorname* { m a x } _ { n \\geq 0 } \\{ ( { \\mathcal { T } } ^ { \\mu } ) ^ { n } { \\mathcal { T } } _ { \\mu } ^ { \\tau } V ( s ) \\} ," + }, + { + "category_id": 14, + "poly": [ + 487, + 1828, + 1208, + 1828, + 1208, + 2040, + 487, + 2040 + ], + "score": 0.94, + "latex": "\\begin{array} { r l r } { { \\mathcal { T } ^ { \\pi } Q ( s , a ) = r ( s , a ) + \\gamma \\mathbb { E } _ { a ^ { \\prime } \\sim \\pi ( \\cdot \\vert s ^ { \\prime } ) } Q ( s ^ { \\prime } , a ^ { \\prime } ) } } \\\\ & { } & { = r ( s , a ) + \\gamma \\mathbb { E } _ { a ^ { \\prime } \\sim \\mu ( \\cdot \\vert s ^ { \\prime } ) } \\frac { \\exp { ( \\alpha Q ( s ^ { \\prime } , a ^ { \\prime } ) ) } } { Z ( s ^ { \\prime } ) } Q ( s ^ { \\prime } , a ^ { \\prime } ) } \\\\ & { } & { \\approx r ( s , a ) + \\gamma \\sum _ { N } [ \\frac { \\exp { ( \\alpha Q ( s ^ { \\prime } , a ^ { \\prime } ) ) } } { \\sum _ { N } \\exp { ( \\alpha Q ( s ^ { \\prime } , a ^ { \\prime } ) ) } } Q ( s ^ { \\prime } , a ^ { \\prime } ) ] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1264, + 1544, + 1359, + 1544, + 1359, + 1579, + 1264, + 1579 + ], + "score": 0.93, + "latex": "\\xi _ { \\phi } ( \\pmb { s } , \\pmb { a } )" + }, + { + "category_id": 14, + "poly": [ + 619, + 1084, + 1079, + 1084, + 1079, + 1138, + 619, + 1138 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\mathcal { T } _ { \\mu } ^ { \\pi } Q ( s , \\pmb { a } ) : = \\underset { n \\geq 0 } { \\operatorname* { m a x } } \\{ ( \\mathcal { T } ^ { \\mu } ) ^ { n } \\mathcal { T } ^ { \\pi } Q ( s , \\pmb { a } ) \\} , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 386, + 971, + 467, + 971, + 467, + 1003, + 386, + 1003 + ], + "score": 0.93, + "latex": "r ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 519, + 970, + 625, + 970, + 625, + 1003, + 519, + 1003 + ], + "score": 0.93, + "latex": "P ( \\cdot | s , a )" + }, + { + "category_id": 13, + "poly": [ + 373, + 1513, + 468, + 1513, + 468, + 1549, + 373, + 1549 + ], + "score": 0.93, + "latex": "\\xi _ { \\phi } ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 876, + 1718, + 953, + 1718, + 953, + 1755, + 876, + 1755 + ], + "score": 0.93, + "latex": "\\{ { \\pmb a } _ { i } ^ { \\prime } \\} ^ { N }" + }, + { + "category_id": 13, + "poly": [ + 965, + 1173, + 1254, + 1173, + 1254, + 1205, + 965, + 1205 + ], + "score": 0.92, + "latex": "\\mathcal T ^ { \\pi } Q ( s , { \\pmb a } ) > \\mathcal T ^ { \\mu } Q ( s , { \\pmb a } )" + }, + { + "category_id": 13, + "poly": [ + 410, + 908, + 607, + 908, + 607, + 942, + 410, + 942 + ], + "score": 0.92, + "latex": "r ( s , a ) + \\gamma V ( s ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 1053, + 480, + 1105, + 480, + 1105, + 510, + 1053, + 510 + ], + "score": 0.91, + "latex": "\\mathcal { T } _ { \\mathrm { v e m } }" + }, + { + "category_id": 13, + "poly": [ + 695, + 908, + 756, + 908, + 756, + 942, + 695, + 942 + ], + "score": 0.91, + "latex": "V ( s )" + }, + { + "category_id": 13, + "poly": [ + 355, + 709, + 474, + 709, + 474, + 742, + 355, + 742 + ], + "score": 0.91, + "latex": "\\tau \\in [ 0 , 1 )" + }, + { + "category_id": 14, + "poly": [ + 310, + 610, + 1407, + 610, + 1407, + 685, + 310, + 685 + ], + "score": 0.91, + "latex": "\\Gamma _ { \\mu } ^ { \\tau } V ( s ) : = \\mathbb { E } _ { a \\sim \\mu ( \\cdot | s ) } \\left\\{ \\begin{array} { l l } { \\tau [ r ( s , a ) + \\gamma V ( s ^ { \\prime } ) ] + ( 1 - \\tau ) V ( s ) } & { \\mathrm { ~ i f ~ } \\quad r ( s , a ) + \\gamma V ( s ^ { \\prime } ) \\geq V ( s ) } \\\\ { ( 1 - \\tau ) [ r ( s , a ) + \\gamma V ( s ^ { \\prime } ) ] + \\tau V ( s ) } & { \\mathrm { ~ i f ~ } \\quad r ( s , a ) + \\gamma V ( s ^ { \\prime } ) < V ( s ) } \\end{array} \\right\\}" + }, + { + "category_id": 13, + "poly": [ + 973, + 909, + 1169, + 909, + 1169, + 942, + 973, + 942 + ], + "score": 0.91, + "latex": "r ( s , { \\pmb a } ) + \\gamma V ( { \\pmb s } ^ { \\prime } )" + }, + { + "category_id": 14, + "poly": [ + 476, + 1608, + 1223, + 1608, + 1223, + 1666, + 476, + 1666 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { \\mathrm { E M a Q } } Q ( s , a ) : = r ( s , a ) + \\gamma \\mathbb { E } _ { \\{ a _ { i } ^ { \\prime } \\} ^ { N } \\sim \\mu _ { \\theta } ( \\cdot \\vert s ^ { \\prime } ) } \\big [ \\operatorname* { m a x } _ { a _ { i } ^ { \\prime } \\in \\{ a _ { i } ^ { \\prime } \\} ^ { N } } Q ( s ^ { \\prime } , a _ { i } ^ { \\prime } ) \\big ] ." + }, + { + "category_id": 14, + "poly": [ + 486, + 1451, + 1210, + 1451, + 1210, + 1508, + 486, + 1508 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { \\mathrm { B C Q } } ^ { * } Q ( s , a ) : = r ( s , a ) + \\gamma \\operatorname* { m a x } _ { a ^ { \\prime } \\sim \\mu _ { \\theta } ( \\cdot \\vert s ^ { \\prime } ) } \\{ Q ( s ^ { \\prime } , a ^ { \\prime } + \\xi _ { \\phi } ( s ^ { \\prime } , a ^ { \\prime } ) ) \\} ," + }, + { + "category_id": 13, + "poly": [ + 1026, + 1052, + 1099, + 1052, + 1099, + 1080, + 1026, + 1080 + ], + "score": 0.9, + "latex": "\\pi > \\mu" + }, + { + "category_id": 13, + "poly": [ + 1000, + 1205, + 1039, + 1205, + 1039, + 1240, + 1000, + 1240 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 494, + 1049, + 531, + 1049, + 531, + 1085, + 494, + 1085 + ], + "score": 0.9, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\tau }" + }, + { + "category_id": 13, + "poly": [ + 298, + 908, + 358, + 908, + 358, + 942, + 298, + 942 + ], + "score": 0.9, + "latex": "V ( s )" + }, + { + "category_id": 13, + "poly": [ + 1272, + 1204, + 1311, + 1204, + 1311, + 1234, + 1272, + 1234 + ], + "score": 0.89, + "latex": "Q ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 360, + 941, + 524, + 941, + 524, + 972, + 360, + 972 + ], + "score": 0.89, + "latex": "\\pmb { s } ^ { \\prime } \\sim P ( \\cdot | \\pmb { s } , \\pmb { a } )" + }, + { + "category_id": 13, + "poly": [ + 997, + 1018, + 1036, + 1018, + 1036, + 1053, + 997, + 1053 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 397, + 480, + 449, + 480, + 449, + 510, + 397, + 510 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mathrm { v e m } }" + }, + { + "category_id": 13, + "poly": [ + 514, + 572, + 551, + 572, + 551, + 608, + 514, + 608 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\tau }" + }, + { + "category_id": 13, + "poly": [ + 783, + 450, + 823, + 450, + 823, + 485, + 783, + 485 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 1134, + 802, + 1172, + 802, + 1172, + 838, + 1134, + 838 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\tau }" + }, + { + "category_id": 13, + "poly": [ + 876, + 771, + 1131, + 771, + 1131, + 805, + 876, + 805 + ], + "score": 0.88, + "latex": "( \\mathcal T _ { \\mu } ^ { \\tau _ { 2 } } V ( \\pmb { s } ) > \\mathcal T _ { \\mu } ^ { \\tau _ { 1 } } V ( \\pmb { s } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 772, + 335, + 772, + 335, + 809, + 298, + 809 + ], + "score": 0.88, + "latex": "\\mathcal { T } _ { \\mu } ^ { \\tau }" + }, + { + "category_id": 13, + "poly": [ + 363, + 803, + 453, + 803, + 453, + 832, + 363, + 832 + ], + "score": 0.88, + "latex": "\\tau _ { 2 } > \\tau _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1245, + 1143, + 1284, + 1143, + 1284, + 1171, + 1245, + 1171 + ], + "score": 0.88, + "latex": "\\mathcal { T } ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 739, + 1721, + 776, + 1721, + 776, + 1749, + 739, + 1749 + ], + "score": 0.88, + "latex": "\\tau ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1032, + 800, + 1085, + 800, + 1085, + 832, + 1032, + 832 + ], + "score": 0.87, + "latex": "\\mathcal { \\tilde { T } } _ { \\mathrm { v e m } }" + }, + { + "category_id": 13, + "poly": [ + 299, + 1176, + 335, + 1176, + 335, + 1210, + 299, + 1210 + ], + "score": 0.87, + "latex": "\\smash { \\mathcal { T } _ { \\mu } ^ { \\tau } }" + }, + { + "category_id": 13, + "poly": [ + 1012, + 1688, + 1049, + 1688, + 1049, + 1716, + 1012, + 1716 + ], + "score": 0.87, + "latex": "\\tau ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 922, + 1688, + 961, + 1688, + 961, + 1716, + 922, + 1716 + ], + "score": 0.87, + "latex": "\\mathcal { T } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 1363, + 1203, + 1400, + 1203, + 1400, + 1234, + 1363, + 1234 + ], + "score": 0.87, + "latex": "Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 423, + 1721, + 522, + 1721, + 522, + 1749, + 423, + 1749 + ], + "score": 0.86, + "latex": "N \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 1226, + 1688, + 1297, + 1688, + 1297, + 1721, + 1226, + 1721 + ], + "score": 0.86, + "latex": "\\tau _ { \\mathrm { E M a Q } }" + }, + { + "category_id": 13, + "poly": [ + 1136, + 1688, + 1215, + 1688, + 1215, + 1716, + 1136, + 1716 + ], + "score": 0.85, + "latex": "N = 1" + }, + { + "category_id": 13, + "poly": [ + 1361, + 711, + 1400, + 711, + 1400, + 738, + 1361, + 738 + ], + "score": 0.85, + "latex": "\\mathcal { T } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 761, + 1769, + 789, + 1769, + 789, + 1795, + 761, + 1795 + ], + "score": 0.85, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 298, + 1721, + 336, + 1721, + 336, + 1749, + 298, + 1749 + ], + "score": 0.85, + "latex": "\\mathcal { T } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 830, + 741, + 867, + 741, + 867, + 769, + 830, + 769 + ], + "score": 0.84, + "latex": "\\tau ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1004, + 740, + 1058, + 740, + 1058, + 772, + 1004, + 772 + ], + "score": 0.84, + "latex": "V ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 304, + 742, + 403, + 742, + 403, + 770, + 304, + 770 + ], + "score": 0.83, + "latex": "\\tau = 0 . 5 )" + }, + { + "category_id": 13, + "poly": [ + 533, + 1720, + 604, + 1720, + 604, + 1754, + 533, + 1754 + ], + "score": 0.83, + "latex": "\\tau _ { \\mathrm { E M a Q } }" + }, + { + "category_id": 13, + "poly": [ + 430, + 1689, + 458, + 1689, + 458, + 1715, + 430, + 1715 + ], + "score": 0.81, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 477, + 1550, + 497, + 1550, + 497, + 1576, + 477, + 1576 + ], + "score": 0.8, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 384, + 1205, + 409, + 1205, + 409, + 1234, + 384, + 1234 + ], + "score": 0.78, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 1376, + 772, + 1401, + 772, + 1401, + 798, + 1376, + 798 + ], + "score": 0.77, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 880, + 741, + 983, + 741, + 983, + 770, + 880, + 770 + ], + "score": 0.76, + "latex": "\\tau 1 . 0" + }, + { + "category_id": 13, + "poly": [ + 944, + 914, + 960, + 914, + 960, + 936, + 944, + 936 + ], + "score": 0.57, + "latex": "\\pmb { s }" + }, + { + "category_id": 13, + "poly": [ + 483, + 1577, + 508, + 1577, + 508, + 1602, + 483, + 1602 + ], + "score": 0.4, + "latex": "_ \\mathrm { N }" + }, + { + "category_id": 13, + "poly": [ + 871, + 1018, + 895, + 1018, + 895, + 1048, + 871, + 1048 + ], + "score": 0.27, + "latex": "\\mathrm { Q }" + }, + { + "category_id": 13, + "poly": [ + 368, + 1049, + 390, + 1049, + 390, + 1075, + 368, + 1075 + ], + "score": 0.25, + "latex": "\\mathrm { v } ." + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1292.0, + 461.0, + 1292.0, + 461.0, + 1340.0, + 293.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 109.0, + 296.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 222.0, + 826.0, + 222.0, + 826.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 387.0, + 443.0, + 387.0, + 443.0, + 427.0, + 295.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1141.0, + 1244.0, + 1141.0, + 1244.0, + 1177.0, + 294.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 1141.0, + 1405.0, + 1141.0, + 1405.0, + 1177.0, + 1285.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1170.0, + 298.0, + 1170.0, + 298.0, + 1210.0, + 295.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1170.0, + 964.0, + 1170.0, + 964.0, + 1210.0, + 336.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1170.0, + 1406.0, + 1170.0, + 1406.0, + 1210.0, + 1255.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1199.0, + 383.0, + 1199.0, + 383.0, + 1242.0, + 291.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1199.0, + 999.0, + 1199.0, + 999.0, + 1242.0, + 410.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 1199.0, + 1271.0, + 1199.0, + 1271.0, + 1242.0, + 1040.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 1199.0, + 1362.0, + 1199.0, + 1362.0, + 1242.0, + 1312.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1199.0, + 1405.0, + 1199.0, + 1405.0, + 1242.0, + 1401.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1233.0, + 847.0, + 1233.0, + 847.0, + 1269.0, + 294.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 704.0, + 354.0, + 704.0, + 354.0, + 744.0, + 291.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 704.0, + 1360.0, + 704.0, + 1360.0, + 744.0, + 475.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 739.0, + 303.0, + 739.0, + 303.0, + 776.0, + 294.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 739.0, + 829.0, + 739.0, + 829.0, + 776.0, + 404.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 739.0, + 879.0, + 739.0, + 879.0, + 776.0, + 868.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 739.0, + 1003.0, + 739.0, + 1003.0, + 776.0, + 984.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 739.0, + 1407.0, + 739.0, + 1407.0, + 776.0, + 1059.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 764.0, + 297.0, + 764.0, + 297.0, + 809.0, + 291.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 764.0, + 875.0, + 764.0, + 875.0, + 809.0, + 336.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 764.0, + 1375.0, + 764.0, + 1375.0, + 809.0, + 1132.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 764.0, + 1407.0, + 764.0, + 1407.0, + 809.0, + 1402.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 798.0, + 362.0, + 798.0, + 362.0, + 839.0, + 290.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 798.0, + 1031.0, + 798.0, + 1031.0, + 839.0, + 454.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 798.0, + 1133.0, + 798.0, + 1133.0, + 839.0, + 1086.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 798.0, + 1408.0, + 798.0, + 1408.0, + 839.0, + 1173.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 834.0, + 368.0, + 834.0, + 368.0, + 865.0, + 291.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 874.0, + 1406.0, + 874.0, + 1406.0, + 915.0, + 294.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 907.0, + 409.0, + 907.0, + 409.0, + 943.0, + 359.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 907.0, + 694.0, + 907.0, + 694.0, + 943.0, + 608.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 907.0, + 943.0, + 907.0, + 943.0, + 943.0, + 757.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 907.0, + 972.0, + 907.0, + 972.0, + 943.0, + 961.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 907.0, + 1406.0, + 907.0, + 1406.0, + 943.0, + 1170.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 938.0, + 359.0, + 938.0, + 359.0, + 973.0, + 295.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 938.0, + 1408.0, + 938.0, + 1408.0, + 973.0, + 525.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 970.0, + 385.0, + 970.0, + 385.0, + 1003.0, + 295.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 970.0, + 518.0, + 970.0, + 518.0, + 1003.0, + 468.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 970.0, + 779.0, + 970.0, + 779.0, + 1003.0, + 626.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1512.0, + 372.0, + 1512.0, + 372.0, + 1549.0, + 294.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1512.0, + 1407.0, + 1512.0, + 1407.0, + 1549.0, + 469.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1542.0, + 476.0, + 1542.0, + 476.0, + 1582.0, + 292.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1542.0, + 1263.0, + 1542.0, + 1263.0, + 1582.0, + 498.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1360.0, + 1542.0, + 1404.0, + 1542.0, + 1404.0, + 1582.0, + 1360.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1575.0, + 482.0, + 1575.0, + 482.0, + 1608.0, + 296.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 1575.0, + 778.0, + 1575.0, + 778.0, + 1608.0, + 509.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1354.0, + 1407.0, + 1354.0, + 1407.0, + 1393.0, + 293.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1387.0, + 1407.0, + 1387.0, + 1407.0, + 1425.0, + 294.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1416.0, + 1341.0, + 1416.0, + 1341.0, + 1456.0, + 293.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1681.0, + 429.0, + 1681.0, + 429.0, + 1725.0, + 293.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1681.0, + 921.0, + 1681.0, + 921.0, + 1725.0, + 459.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1681.0, + 1011.0, + 1681.0, + 1011.0, + 1725.0, + 962.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 1681.0, + 1135.0, + 1681.0, + 1135.0, + 1725.0, + 1050.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1681.0, + 1225.0, + 1681.0, + 1225.0, + 1725.0, + 1216.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 1681.0, + 1405.0, + 1681.0, + 1405.0, + 1725.0, + 1298.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1717.0, + 422.0, + 1717.0, + 422.0, + 1758.0, + 337.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1717.0, + 532.0, + 1717.0, + 532.0, + 1758.0, + 523.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 1717.0, + 738.0, + 1717.0, + 738.0, + 1758.0, + 605.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 1717.0, + 875.0, + 1717.0, + 875.0, + 1758.0, + 777.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1717.0, + 1378.0, + 1717.0, + 1378.0, + 1758.0, + 954.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 289.0, + 1404.0, + 289.0, + 1404.0, + 331.0, + 292.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 321.0, + 1003.0, + 321.0, + 1003.0, + 360.0, + 293.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1011.0, + 870.0, + 1011.0, + 870.0, + 1059.0, + 293.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1011.0, + 996.0, + 1011.0, + 996.0, + 1059.0, + 896.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1011.0, + 1406.0, + 1011.0, + 1406.0, + 1059.0, + 1037.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1044.0, + 367.0, + 1044.0, + 367.0, + 1088.0, + 294.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1044.0, + 493.0, + 1044.0, + 493.0, + 1088.0, + 391.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1044.0, + 1025.0, + 1044.0, + 1025.0, + 1088.0, + 532.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1044.0, + 1370.0, + 1044.0, + 1370.0, + 1088.0, + 1100.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 443.0, + 782.0, + 443.0, + 782.0, + 485.0, + 294.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 443.0, + 1406.0, + 443.0, + 1406.0, + 485.0, + 824.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 478.0, + 396.0, + 478.0, + 396.0, + 515.0, + 294.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 478.0, + 1052.0, + 478.0, + 1052.0, + 515.0, + 450.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 478.0, + 1260.0, + 478.0, + 1260.0, + 515.0, + 1106.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1762.0, + 760.0, + 1762.0, + 760.0, + 1804.0, + 292.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1762.0, + 1405.0, + 1762.0, + 1405.0, + 1804.0, + 790.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1798.0, + 548.0, + 1798.0, + 548.0, + 1834.0, + 295.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 568.0, + 513.0, + 568.0, + 513.0, + 607.0, + 294.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 568.0, + 1217.0, + 568.0, + 1217.0, + 607.0, + 552.0, + 607.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/dev/4AZz9osqrar/4AZz9osqrar_model.json b/parse/dev/4AZz9osqrar/4AZz9osqrar_model.json new file mode 100644 index 0000000000000000000000000000000000000000..b8c228c390b9c0b39141732b8c290af454d71add --- /dev/null +++ b/parse/dev/4AZz9osqrar/4AZz9osqrar_model.json @@ -0,0 +1,37503 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 751, + 1304, + 751, + 1304, + 1332, + 398, + 1332 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 299, + 1452, + 1405, + 1452, + 1405, + 1666, + 299, + 1666 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1681, + 1404, + 1681, + 1404, + 1927, + 298, + 1927 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 301, + 1943, + 1402, + 1943, + 1402, + 2034, + 301, + 2034 + ], + "score": 0.967 + }, + { + "category_id": 0, + "poly": [ + 301, + 221, + 1401, + 221, + 1401, + 323, + 301, + 323 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 314, + 517, + 713, + 517, + 713, + 608, + 314, + 608 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 313, + 377, + 634, + 377, + 634, + 468, + 313, + 468 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 1036, + 517, + 1389, + 517, + 1389, + 609, + 1036, + 609 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 1032, + 378, + 1386, + 378, + 1386, + 469, + 1032, + 469 + ], + "score": 0.942 + }, + { + "category_id": 0, + "poly": [ + 302, + 1384, + 573, + 1384, + 573, + 1419, + 302, + 1419 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 104, + 299, + 104 + ], + "score": 0.877 + }, + { + "category_id": 0, + "poly": [ + 773, + 689, + 927, + 689, + 927, + 722, + 773, + 722 + ], + "score": 0.873 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 856, + 2088, + 856, + 2112, + 841, + 2112 + ], + "score": 0.715 + }, + { + "category_id": 15, + "poly": [ + 292.0, + 215.0, + 1404.0, + 215.0, + 1404.0, + 275.0, + 292.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 276.0, + 756.0, + 276.0, + 756.0, + 326.0, + 295.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1381.0, + 579.0, + 1381.0, + 579.0, + 1428.0, + 294.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 685.0, + 934.0, + 685.0, + 934.0, + 728.0, + 768.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2088.0, + 857.0, + 2088.0, + 857.0, + 2116.0, + 840.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 753.0, + 1308.0, + 753.0, + 1308.0, + 788.0, + 394.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 784.0, + 1305.0, + 784.0, + 1305.0, + 817.0, + 394.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 815.0, + 1305.0, + 815.0, + 1305.0, + 848.0, + 394.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 844.0, + 1305.0, + 844.0, + 1305.0, + 879.0, + 393.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 874.0, + 1305.0, + 874.0, + 1305.0, + 910.0, + 393.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 905.0, + 1306.0, + 905.0, + 1306.0, + 940.0, + 393.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 936.0, + 1309.0, + 936.0, + 1309.0, + 972.0, + 394.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 966.0, + 1305.0, + 966.0, + 1305.0, + 1002.0, + 393.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 999.0, + 1307.0, + 999.0, + 1307.0, + 1032.0, + 395.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1027.0, + 1307.0, + 1027.0, + 1307.0, + 1062.0, + 393.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1059.0, + 1305.0, + 1059.0, + 1305.0, + 1092.0, + 394.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1087.0, + 1305.0, + 1087.0, + 1305.0, + 1122.0, + 393.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1118.0, + 1308.0, + 1118.0, + 1308.0, + 1152.0, + 394.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1152.0, + 1304.0, + 1152.0, + 1304.0, + 1182.0, + 395.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1179.0, + 1308.0, + 1179.0, + 1308.0, + 1215.0, + 393.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1211.0, + 1306.0, + 1211.0, + 1306.0, + 1247.0, + 393.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1242.0, + 1305.0, + 1242.0, + 1305.0, + 1275.0, + 395.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1270.0, + 1305.0, + 1270.0, + 1305.0, + 1305.0, + 394.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1302.0, + 1071.0, + 1302.0, + 1071.0, + 1334.0, + 394.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1451.0, + 1407.0, + 1451.0, + 1407.0, + 1485.0, + 295.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1482.0, + 1406.0, + 1482.0, + 1406.0, + 1516.0, + 294.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1514.0, + 1405.0, + 1514.0, + 1405.0, + 1544.0, + 297.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1541.0, + 1407.0, + 1541.0, + 1407.0, + 1577.0, + 294.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1574.0, + 1403.0, + 1574.0, + 1403.0, + 1604.0, + 295.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1602.0, + 1405.0, + 1602.0, + 1405.0, + 1638.0, + 294.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1634.0, + 1244.0, + 1634.0, + 1244.0, + 1668.0, + 295.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1680.0, + 1404.0, + 1680.0, + 1404.0, + 1714.0, + 294.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1711.0, + 1405.0, + 1711.0, + 1405.0, + 1745.0, + 294.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1742.0, + 1405.0, + 1742.0, + 1405.0, + 1775.0, + 296.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1773.0, + 1404.0, + 1773.0, + 1404.0, + 1806.0, + 296.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1803.0, + 1406.0, + 1803.0, + 1406.0, + 1836.0, + 294.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1833.0, + 1404.0, + 1833.0, + 1404.0, + 1869.0, + 293.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1864.0, + 1406.0, + 1864.0, + 1406.0, + 1898.0, + 294.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1894.0, + 1053.0, + 1894.0, + 1053.0, + 1928.0, + 296.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1406.0, + 1942.0, + 1406.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2008.0, + 295.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2000.0, + 1404.0, + 2000.0, + 1404.0, + 2039.0, + 295.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 514.0, + 499.0, + 514.0, + 499.0, + 549.0, + 312.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 547.0, + 603.0, + 547.0, + 603.0, + 577.0, + 313.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 578.0, + 716.0, + 578.0, + 716.0, + 610.0, + 315.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 375.0, + 436.0, + 375.0, + 436.0, + 411.0, + 309.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 402.0, + 540.0, + 402.0, + 540.0, + 443.0, + 311.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 438.0, + 635.0, + 438.0, + 635.0, + 469.0, + 312.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 514.0, + 1178.0, + 514.0, + 1178.0, + 550.0, + 1035.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 543.0, + 1261.0, + 543.0, + 1261.0, + 581.0, + 1034.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 579.0, + 1391.0, + 579.0, + 1391.0, + 611.0, + 1032.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 377.0, + 1234.0, + 377.0, + 1234.0, + 409.0, + 1029.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 403.0, + 1257.0, + 403.0, + 1257.0, + 442.0, + 1029.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 438.0, + 1386.0, + 438.0, + 1386.0, + 469.0, + 1033.0, + 469.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1002, + 1404, + 1002, + 1404, + 1337, + 298, + 1337 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 741, + 1404, + 741, + 1404, + 987, + 297, + 987 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1738, + 1403, + 1738, + 1403, + 1861, + 299, + 1861 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 1353, + 1403, + 1353, + 1403, + 1475, + 299, + 1475 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1598, + 1404, + 1598, + 1404, + 1722, + 298, + 1722 + ], + "score": 0.97 + }, + { + "category_id": 4, + "poly": [ + 296, + 559, + 1406, + 559, + 1406, + 712, + 296, + 712 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 301, + 1492, + 1400, + 1492, + 1400, + 1584, + 301, + 1584 + ], + "score": 0.956 + }, + { + "category_id": 3, + "poly": [ + 303, + 239, + 1385, + 239, + 1385, + 550, + 303, + 550 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 297, + 1972, + 1397, + 1972, + 1397, + 2034, + 297, + 2034 + ], + "score": 0.947 + }, + { + "category_id": 0, + "poly": [ + 299, + 1906, + 1155, + 1906, + 1155, + 1943, + 299, + 1943 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.717 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.104 + }, + { + "category_id": 13, + "poly": [ + 593, + 618, + 643, + 618, + 643, + 649, + 593, + 649 + ], + "score": 0.87, + "latex": "\\Delta ^ { \\mathrm { S L } }" + }, + { + "category_id": 13, + "poly": [ + 1308, + 587, + 1371, + 587, + 1371, + 619, + 1308, + 619 + ], + "score": 0.85, + "latex": "\\Delta ^ { \\mathrm { s s L } }" + }, + { + "category_id": 13, + "poly": [ + 1377, + 627, + 1397, + 627, + 1397, + 649, + 1377, + 649 + ], + "score": 0.62, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 1279, + 256, + 1358, + 256, + 1358, + 275, + 1279, + 275 + ], + "score": 0.4, + "latex": "{ \\sf r } = 0 . 0 0 2 5" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 556.0, + 1409.0, + 556.0, + 1409.0, + 597.0, + 293.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 583.0, + 1307.0, + 583.0, + 1307.0, + 625.0, + 290.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 583.0, + 1408.0, + 583.0, + 1408.0, + 625.0, + 1372.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 614.0, + 592.0, + 614.0, + 592.0, + 658.0, + 291.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 614.0, + 1376.0, + 614.0, + 1376.0, + 658.0, + 644.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 614.0, + 1409.0, + 614.0, + 1409.0, + 658.0, + 1398.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 647.0, + 1407.0, + 647.0, + 1407.0, + 687.0, + 291.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 680.0, + 1016.0, + 680.0, + 1016.0, + 716.0, + 294.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 239.0, + 331.0, + 239.0, + 331.0, + 486.0, + 303.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 256.0, + 558.0, + 256.0, + 558.0, + 276.0, + 358.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 255.0, + 810.0, + 255.0, + 810.0, + 279.0, + 599.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 231.0, + 907.0, + 231.0, + 907.0, + 493.0, + 855.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 253.0, + 1110.0, + 253.0, + 1110.0, + 278.0, + 908.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 253.0, + 1278.0, + 253.0, + 1278.0, + 278.0, + 1149.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 265.0, + 344.0, + 265.0, + 344.0, + 287.0, + 322.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 281.0, + 443.0, + 281.0, + 443.0, + 316.0, + 401.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 281.0, + 998.0, + 281.0, + 998.0, + 315.0, + 951.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 305.0, + 344.0, + 305.0, + 344.0, + 328.0, + 323.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 306.0, + 435.0, + 306.0, + 435.0, + 338.0, + 402.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 344.0, + 345.0, + 344.0, + 345.0, + 366.0, + 323.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 339.0, + 893.0, + 339.0, + 893.0, + 368.0, + 881.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 384.0, + 345.0, + 384.0, + 345.0, + 406.0, + 323.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 382.0, + 890.0, + 382.0, + 890.0, + 410.0, + 876.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 427.0, + 344.0, + 427.0, + 344.0, + 445.0, + 331.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 468.0, + 345.0, + 468.0, + 345.0, + 484.0, + 332.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 474.0, + 396.0, + 474.0, + 396.0, + 492.0, + 365.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 474.0, + 438.0, + 474.0, + 438.0, + 492.0, + 406.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 474.0, + 480.0, + 474.0, + 480.0, + 492.0, + 446.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 474.0, + 523.0, + 474.0, + 523.0, + 492.0, + 487.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 475.0, + 566.0, + 475.0, + 566.0, + 492.0, + 525.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 477.0, + 645.0, + 477.0, + 645.0, + 491.0, + 616.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 475.0, + 689.0, + 475.0, + 689.0, + 492.0, + 656.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 475.0, + 729.0, + 475.0, + 729.0, + 492.0, + 696.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 475.0, + 771.0, + 475.0, + 771.0, + 492.0, + 738.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 474.0, + 812.0, + 474.0, + 812.0, + 492.0, + 779.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 471.0, + 1117.0, + 471.0, + 1117.0, + 495.0, + 911.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 474.0, + 1364.0, + 474.0, + 1364.0, + 495.0, + 1162.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 486.0, + 760.0, + 486.0, + 760.0, + 514.0, + 417.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 487.0, + 1312.0, + 487.0, + 1312.0, + 515.0, + 966.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 520.0, + 674.0, + 520.0, + 674.0, + 554.0, + 467.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 518.0, + 1266.0, + 518.0, + 1266.0, + 552.0, + 983.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.25, + 306.0, + 995.25, + 306.0, + 995.25, + 338.0, + 954.25, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1904.0, + 1158.0, + 1904.0, + 1158.0, + 1948.0, + 292.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1000.0, + 1405.0, + 1000.0, + 1405.0, + 1037.0, + 295.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1032.0, + 1409.0, + 1032.0, + 1409.0, + 1067.0, + 294.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1064.0, + 1404.0, + 1064.0, + 1404.0, + 1095.0, + 296.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1093.0, + 1405.0, + 1093.0, + 1405.0, + 1126.0, + 295.0, + 1126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1121.0, + 1405.0, + 1121.0, + 1405.0, + 1158.0, + 293.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1152.0, + 1404.0, + 1152.0, + 1404.0, + 1191.0, + 293.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1183.0, + 1404.0, + 1183.0, + 1404.0, + 1220.0, + 293.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1215.0, + 1405.0, + 1215.0, + 1405.0, + 1250.0, + 294.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1244.0, + 1404.0, + 1244.0, + 1404.0, + 1281.0, + 293.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1277.0, + 1405.0, + 1277.0, + 1405.0, + 1312.0, + 294.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1307.0, + 1271.0, + 1307.0, + 1271.0, + 1341.0, + 294.0, + 1341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 743.0, + 1404.0, + 743.0, + 1404.0, + 776.0, + 295.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 769.0, + 1405.0, + 769.0, + 1405.0, + 810.0, + 292.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 803.0, + 1406.0, + 803.0, + 1406.0, + 836.0, + 295.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 834.0, + 1406.0, + 834.0, + 1406.0, + 868.0, + 295.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 865.0, + 1404.0, + 865.0, + 1404.0, + 895.0, + 293.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 894.0, + 1406.0, + 894.0, + 1406.0, + 930.0, + 292.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 925.0, + 1405.0, + 925.0, + 1405.0, + 959.0, + 295.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 957.0, + 752.0, + 957.0, + 752.0, + 990.0, + 296.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1739.0, + 1404.0, + 1739.0, + 1404.0, + 1772.0, + 294.0, + 1772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1767.0, + 1407.0, + 1767.0, + 1407.0, + 1804.0, + 292.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1799.0, + 1407.0, + 1799.0, + 1407.0, + 1834.0, + 294.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1830.0, + 1080.0, + 1830.0, + 1080.0, + 1865.0, + 293.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1354.0, + 1403.0, + 1354.0, + 1403.0, + 1386.0, + 295.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1384.0, + 1403.0, + 1384.0, + 1403.0, + 1419.0, + 294.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1413.0, + 1404.0, + 1413.0, + 1404.0, + 1451.0, + 293.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1442.0, + 1124.0, + 1442.0, + 1124.0, + 1479.0, + 293.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1599.0, + 1404.0, + 1599.0, + 1404.0, + 1632.0, + 294.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1631.0, + 1405.0, + 1631.0, + 1405.0, + 1664.0, + 294.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1660.0, + 1406.0, + 1660.0, + 1406.0, + 1696.0, + 293.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1693.0, + 619.0, + 1693.0, + 619.0, + 1725.0, + 294.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1491.0, + 1405.0, + 1491.0, + 1405.0, + 1528.0, + 295.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1522.0, + 1403.0, + 1522.0, + 1403.0, + 1556.0, + 295.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1551.0, + 1406.0, + 1551.0, + 1406.0, + 1592.0, + 295.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2003.0, + 1402.0, + 2003.0, + 1402.0, + 2035.0, + 296.0, + 2035.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 373, + 1405, + 373, + 1405, + 736, + 297, + 736 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1029, + 1404, + 1029, + 1404, + 1259, + 297, + 1259 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1272, + 1403, + 1272, + 1403, + 1464, + 298, + 1464 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 751, + 1403, + 751, + 1403, + 907, + 298, + 907 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1683, + 1404, + 1683, + 1404, + 1868, + 297, + 1868 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1479, + 1404, + 1479, + 1404, + 1602, + 298, + 1602 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 301, + 921, + 1399, + 921, + 1399, + 1014, + 301, + 1014 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 295, + 229, + 1404, + 229, + 1404, + 293, + 295, + 293 + ], + "score": 0.951 + }, + { + "category_id": 0, + "poly": [ + 300, + 1631, + 641, + 1631, + 641, + 1662, + 300, + 1662 + ], + "score": 0.9 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 104, + 300, + 104 + ], + "score": 0.896 + }, + { + "category_id": 0, + "poly": [ + 300, + 322, + 669, + 322, + 669, + 353, + 300, + 353 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.661 + }, + { + "category_id": 2, + "poly": [ + 296, + 1891, + 1404, + 1891, + 1404, + 2033, + 296, + 2033 + ], + "score": 0.364 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.325 + }, + { + "category_id": 13, + "poly": [ + 1232, + 466, + 1338, + 466, + 1338, + 499, + 1232, + 499 + ], + "score": 0.94, + "latex": "\\mathbb { R } ^ { d } \\times [ C ]" + }, + { + "category_id": 13, + "poly": [ + 930, + 750, + 1114, + 750, + 1114, + 787, + 930, + 787 + ], + "score": 0.94, + "latex": "f _ { \\phi } : \\mathbb { R } ^ { d } \\mathbb { R } ^ { m }" + }, + { + "category_id": 13, + "poly": [ + 513, + 1191, + 623, + 1191, + 623, + 1226, + 513, + 1226 + ], + "score": 0.94, + "latex": "A _ { \\mathrm { I D } } ^ { \\mathrm { S L } } ( n , r )" + }, + { + "category_id": 13, + "poly": [ + 751, + 1191, + 862, + 1191, + 862, + 1226, + 751, + 1226 + ], + "score": 0.93, + "latex": "A _ { \\mathrm { I D } } ^ { \\mathrm { S L } } ( n , 1 )" + }, + { + "category_id": 13, + "poly": [ + 971, + 1223, + 1092, + 1223, + 1092, + 1260, + 971, + 1260 + ], + "score": 0.93, + "latex": "A _ { \\mathrm { I D } } ^ { \\mathrm { S S L } } ( n , r )" + }, + { + "category_id": 13, + "poly": [ + 618, + 526, + 896, + 526, + 896, + 578, + 618, + 578 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { r = \\frac { \\operatorname* { m i n } _ { j \\in [ C ] } \\mathcal { P } ( y = j ) } { \\operatorname* { m a x } _ { j \\in [ C ] } \\mathcal { P } ( y = j ) } \\le 1 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1010, + 1400, + 1140, + 1400, + 1140, + 1436, + 1010, + 1436 + ], + "score": 0.93, + "latex": "A _ { \\mathrm { O O D } } ^ { \\mathrm { S S L } } ( n , r )" + }, + { + "category_id": 13, + "poly": [ + 465, + 1805, + 617, + 1805, + 617, + 1839, + 465, + 1839 + ], + "score": 0.92, + "latex": "\\lbrace 1 , 0 . 1 , 0 . 0 1 \\rbrace" + }, + { + "category_id": 13, + "poly": [ + 827, + 1400, + 957, + 1400, + 957, + 1436, + 827, + 1436 + ], + "score": 0.92, + "latex": "A _ { \\mathrm { O O D } } ^ { \\mathrm { S L } } ( n , r )" + }, + { + "category_id": 13, + "poly": [ + 947, + 638, + 1033, + 638, + 1033, + 672, + 947, + 672 + ], + "score": 0.92, + "latex": "j \\in [ C ]" + }, + { + "category_id": 13, + "poly": [ + 298, + 812, + 471, + 812, + 471, + 845, + 298, + 845 + ], + "score": 0.91, + "latex": "g _ { \\theta } : \\mathbb { R } ^ { m } \\overset { \\cdot } { } \\mathbb { R } ^ { C }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1091, + 335, + 1091, + 335, + 1131, + 297, + 1131 + ], + "score": 0.91, + "latex": "\\widehat { \\mathcal { P } } _ { n } ^ { r }" + }, + { + "category_id": 13, + "poly": [ + 1153, + 1774, + 1362, + 1774, + 1362, + 1808, + 1153, + 1808 + ], + "score": 0.91, + "latex": "\\{ 1 , 0 . 0 0 4 , 0 . 0 0 2 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 298, + 669, + 442, + 669, + 442, + 702, + 298, + 702 + ], + "score": 0.9, + "latex": "\\mathcal { P } ^ { r } ( x | y = j )" + }, + { + "category_id": 13, + "poly": [ + 669, + 1835, + 933, + 1835, + 933, + 1869, + 669, + 1869 + ], + "score": 0.89, + "latex": "\\{ 0 . 7 5 , 0 . 5 , 0 . 2 5 , 0 . 1 2 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 973, + 1061, + 1003, + 1061, + 1003, + 1094, + 973, + 1094 + ], + "score": 0.89, + "latex": "f _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 1315, + 1098, + 1346, + 1098, + 1346, + 1132, + 1315, + 1132 + ], + "score": 0.89, + "latex": "f _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 298, + 702, + 335, + 702, + 335, + 739, + 298, + 739 + ], + "score": 0.89, + "latex": "\\widehat { \\mathcal { P } } _ { n } ^ { r }" + }, + { + "category_id": 13, + "poly": [ + 1150, + 576, + 1200, + 576, + 1200, + 606, + 1150, + 606 + ], + "score": 0.89, + "latex": "\\mathcal { P } ^ { \\mathrm { b a l } }" + }, + { + "category_id": 13, + "poly": [ + 807, + 814, + 838, + 814, + 838, + 848, + 807, + 848 + ], + "score": 0.89, + "latex": "f _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 733, + 1129, + 783, + 1129, + 783, + 1159, + 733, + 1159 + ], + "score": 0.89, + "latex": "\\mathcal { P } ^ { \\mathrm { b a l } }" + }, + { + "category_id": 13, + "poly": [ + 406, + 1334, + 437, + 1334, + 437, + 1368, + 406, + 1368 + ], + "score": 0.88, + "latex": "f _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 429, + 1059, + 479, + 1059, + 479, + 1088, + 429, + 1088 + ], + "score": 0.88, + "latex": "\\mathcal { P } ^ { \\mathrm { b a l } }" + }, + { + "category_id": 13, + "poly": [ + 915, + 1365, + 948, + 1365, + 948, + 1401, + 915, + 1401 + ], + "score": 0.88, + "latex": "\\widehat { \\mathcal { P } } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 298, + 609, + 364, + 609, + 364, + 636, + 298, + 636 + ], + "score": 0.87, + "latex": "r = 1" + }, + { + "category_id": 13, + "poly": [ + 1100, + 1305, + 1131, + 1305, + 1131, + 1334, + 1100, + 1334 + ], + "score": 0.87, + "latex": "\\mathcal { P } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1249, + 372, + 1284, + 372, + 1284, + 401, + 1249, + 401 + ], + "score": 0.87, + "latex": "\\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 691, + 1403, + 723, + 1403, + 723, + 1432, + 691, + 1432 + ], + "score": 0.87, + "latex": "\\mathcal { P } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 567, + 579, + 603, + 579, + 603, + 606, + 567, + 606 + ], + "score": 0.85, + "latex": "{ \\mathcal { P } } ^ { r }" + }, + { + "category_id": 13, + "poly": [ + 737, + 1161, + 786, + 1161, + 786, + 1190, + 737, + 1190 + ], + "score": 0.85, + "latex": "\\mathcal { P } ^ { \\mathrm { b a l } }" + }, + { + "category_id": 13, + "poly": [ + 702, + 706, + 738, + 706, + 738, + 734, + 702, + 734 + ], + "score": 0.85, + "latex": "{ \\mathcal { P } } ^ { r }" + }, + { + "category_id": 13, + "poly": [ + 835, + 1921, + 853, + 1921, + 853, + 1949, + 835, + 1949 + ], + "score": 0.85, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 532, + 784, + 552, + 784, + 552, + 815, + 532, + 815 + ], + "score": 0.84, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 583, + 1372, + 605, + 1372, + 605, + 1403, + 583, + 1403 + ], + "score": 0.84, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 1381, + 816, + 1401, + 816, + 1401, + 845, + 1381, + 845 + ], + "score": 0.82, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 923, + 1098, + 947, + 1098, + 947, + 1126, + 923, + 1126 + ], + "score": 0.82, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 1091, + 466, + 1116, + 466, + 1116, + 494, + 1091, + 494 + ], + "score": 0.82, + "latex": "\\mathcal { P }" + }, + { + "category_id": 13, + "poly": [ + 1191, + 875, + 1210, + 875, + 1210, + 907, + 1191, + 907 + ], + "score": 0.81, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 358, + 407, + 383, + 407, + 383, + 432, + 358, + 432 + ], + "score": 0.81, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 900, + 409, + 919, + 409, + 919, + 437, + 900, + 437 + ], + "score": 0.8, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 511, + 644, + 528, + 644, + 528, + 666, + 511, + 666 + ], + "score": 0.78, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 657, + 1372, + 675, + 1372, + 675, + 1399, + 657, + 1399 + ], + "score": 0.77, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 406, + 1373, + 423, + 1373, + 423, + 1400, + 406, + 1400 + ], + "score": 0.77, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 990, + 583, + 1006, + 583, + 1006, + 605, + 990, + 605 + ], + "score": 0.76, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 1185, + 1098, + 1204, + 1098, + 1204, + 1126, + 1185, + 1126 + ], + "score": 0.76, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 810, + 1339, + 830, + 1339, + 830, + 1361, + 810, + 1361 + ], + "score": 0.75, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 790, + 503, + 805, + 503, + 805, + 522, + 790, + 522 + ], + "score": 0.73, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 777, + 1105, + 794, + 1105, + 794, + 1125, + 777, + 1125 + ], + "score": 0.73, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 395, + 1103, + 416, + 1103, + 416, + 1125, + 395, + 1125 + ], + "score": 0.73, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 463, + 710, + 484, + 710, + 484, + 733, + 463, + 733 + ], + "score": 0.73, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 647, + 410, + 667, + 410, + 667, + 432, + 647, + 432 + ], + "score": 0.72, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1106, + 675, + 1122, + 675, + 1122, + 696, + 1106, + 696 + ], + "score": 0.7, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 1068, + 1340, + 1084, + 1340, + 1084, + 1361, + 1068, + 1361 + ], + "score": 0.69, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 1385, + 472, + 1401, + 472, + 1401, + 493, + 1385, + 493 + ], + "score": 0.69, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 1001, + 1546, + 1017, + 1546, + 1017, + 1568, + 1001, + 1568 + ], + "score": 0.67, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 432, + 1546, + 452, + 1546, + 452, + 1569, + 432, + 1569 + ], + "score": 0.64, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 1084, + 376, + 1098, + 376, + 1098, + 402, + 1084, + 402 + ], + "score": 0.27, + "latex": "/" + }, + { + "category_id": 13, + "poly": [ + 434, + 1030, + 486, + 1030, + 486, + 1059, + 434, + 1059 + ], + "score": 0.25, + "latex": "\\mathbf { \\Pi } ( \\mathbf { I D } )" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1628.0, + 644.0, + 1628.0, + 644.0, + 1667.0, + 293.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 321.0, + 670.0, + 321.0, + 670.0, + 357.0, + 295.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 1888.0, + 1405.0, + 1888.0, + 1405.0, + 1923.0, + 333.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1918.0, + 834.0, + 1918.0, + 834.0, + 1949.0, + 294.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1918.0, + 1405.0, + 1918.0, + 1405.0, + 1949.0, + 854.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1947.0, + 1244.0, + 1947.0, + 1244.0, + 1978.0, + 295.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2010.0, + 328.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 1405.0, + 2002.0, + 1405.0, + 2038.0, + 295.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 370.0, + 1083.0, + 370.0, + 1083.0, + 407.0, + 295.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 370.0, + 1248.0, + 370.0, + 1248.0, + 407.0, + 1099.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 370.0, + 1405.0, + 370.0, + 1405.0, + 407.0, + 1285.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 405.0, + 357.0, + 405.0, + 357.0, + 437.0, + 296.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 405.0, + 646.0, + 405.0, + 646.0, + 437.0, + 384.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 405.0, + 899.0, + 405.0, + 899.0, + 437.0, + 668.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 405.0, + 1405.0, + 405.0, + 1405.0, + 437.0, + 920.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 435.0, + 1406.0, + 435.0, + 1406.0, + 469.0, + 292.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 464.0, + 1090.0, + 464.0, + 1090.0, + 501.0, + 292.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 464.0, + 1231.0, + 464.0, + 1231.0, + 501.0, + 1117.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 464.0, + 1384.0, + 464.0, + 1384.0, + 501.0, + 1339.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 464.0, + 1405.0, + 464.0, + 1405.0, + 501.0, + 1402.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 497.0, + 789.0, + 497.0, + 789.0, + 530.0, + 296.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 497.0, + 1405.0, + 497.0, + 1405.0, + 530.0, + 806.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 523.0, + 617.0, + 523.0, + 617.0, + 581.0, + 294.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 523.0, + 1407.0, + 523.0, + 1407.0, + 581.0, + 897.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 572.0, + 566.0, + 572.0, + 566.0, + 612.0, + 294.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 572.0, + 989.0, + 572.0, + 989.0, + 612.0, + 604.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 572.0, + 1149.0, + 572.0, + 1149.0, + 612.0, + 1007.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 572.0, + 1406.0, + 572.0, + 1406.0, + 612.0, + 1201.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 606.0, + 297.0, + 606.0, + 297.0, + 641.0, + 291.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 606.0, + 1406.0, + 606.0, + 1406.0, + 641.0, + 365.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 637.0, + 510.0, + 637.0, + 510.0, + 674.0, + 295.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 637.0, + 946.0, + 637.0, + 946.0, + 674.0, + 529.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 637.0, + 1405.0, + 637.0, + 1405.0, + 674.0, + 1034.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 662.0, + 297.0, + 662.0, + 297.0, + 707.0, + 291.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 662.0, + 1105.0, + 662.0, + 1105.0, + 707.0, + 443.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 662.0, + 1408.0, + 662.0, + 1408.0, + 707.0, + 1123.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 694.0, + 297.0, + 694.0, + 297.0, + 749.0, + 291.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 694.0, + 462.0, + 694.0, + 462.0, + 749.0, + 336.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 694.0, + 701.0, + 694.0, + 701.0, + 749.0, + 485.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 694.0, + 752.0, + 694.0, + 752.0, + 749.0, + 739.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1027.0, + 433.0, + 1027.0, + 433.0, + 1064.0, + 294.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 1027.0, + 1405.0, + 1027.0, + 1405.0, + 1064.0, + 487.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1055.0, + 428.0, + 1055.0, + 428.0, + 1096.0, + 294.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 1055.0, + 972.0, + 1055.0, + 972.0, + 1096.0, + 480.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 1055.0, + 1406.0, + 1055.0, + 1406.0, + 1096.0, + 1004.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1091.0, + 394.0, + 1091.0, + 394.0, + 1135.0, + 336.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1091.0, + 776.0, + 1091.0, + 776.0, + 1135.0, + 417.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1091.0, + 922.0, + 1091.0, + 922.0, + 1135.0, + 795.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 1091.0, + 1184.0, + 1091.0, + 1184.0, + 1135.0, + 948.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1091.0, + 1314.0, + 1091.0, + 1314.0, + 1135.0, + 1205.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1091.0, + 1408.0, + 1091.0, + 1408.0, + 1135.0, + 1347.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1123.0, + 732.0, + 1123.0, + 732.0, + 1169.0, + 292.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1123.0, + 1406.0, + 1123.0, + 1406.0, + 1169.0, + 784.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1153.0, + 736.0, + 1153.0, + 736.0, + 1201.0, + 291.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1153.0, + 1409.0, + 1153.0, + 1409.0, + 1201.0, + 787.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1185.0, + 512.0, + 1185.0, + 512.0, + 1233.0, + 291.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 1185.0, + 750.0, + 1185.0, + 750.0, + 1233.0, + 624.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 1185.0, + 1406.0, + 1185.0, + 1406.0, + 1233.0, + 863.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1220.0, + 970.0, + 1220.0, + 970.0, + 1265.0, + 292.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1220.0, + 1106.0, + 1220.0, + 1106.0, + 1265.0, + 1093.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1269.0, + 1405.0, + 1269.0, + 1405.0, + 1308.0, + 294.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1304.0, + 1099.0, + 1304.0, + 1099.0, + 1337.0, + 296.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1304.0, + 1404.0, + 1304.0, + 1404.0, + 1337.0, + 1132.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1333.0, + 405.0, + 1333.0, + 405.0, + 1368.0, + 293.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1333.0, + 809.0, + 1333.0, + 809.0, + 1368.0, + 438.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1333.0, + 1067.0, + 1333.0, + 1067.0, + 1368.0, + 831.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 1333.0, + 1405.0, + 1333.0, + 1405.0, + 1368.0, + 1085.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1365.0, + 405.0, + 1365.0, + 405.0, + 1470.0, + 285.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1365.0, + 582.0, + 1365.0, + 582.0, + 1470.0, + 424.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 1365.0, + 656.0, + 1365.0, + 656.0, + 1470.0, + 606.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1365.0, + 690.0, + 1365.0, + 690.0, + 1470.0, + 676.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1365.0, + 826.0, + 1365.0, + 826.0, + 1470.0, + 724.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1365.0, + 1009.0, + 1365.0, + 1009.0, + 1470.0, + 958.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 1365.0, + 1414.0, + 1365.0, + 1414.0, + 1470.0, + 1141.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 749.0, + 929.0, + 749.0, + 929.0, + 786.0, + 294.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 749.0, + 1405.0, + 749.0, + 1405.0, + 786.0, + 1115.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 782.0, + 531.0, + 782.0, + 531.0, + 816.0, + 296.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 782.0, + 1404.0, + 782.0, + 1404.0, + 816.0, + 553.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 809.0, + 297.0, + 809.0, + 297.0, + 852.0, + 291.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 809.0, + 806.0, + 809.0, + 806.0, + 852.0, + 472.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 809.0, + 1380.0, + 809.0, + 1380.0, + 852.0, + 839.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 809.0, + 1409.0, + 809.0, + 1409.0, + 852.0, + 1402.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 844.0, + 1405.0, + 844.0, + 1405.0, + 877.0, + 296.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 872.0, + 1190.0, + 872.0, + 1190.0, + 909.0, + 294.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 872.0, + 1223.0, + 872.0, + 1223.0, + 909.0, + 1211.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1684.0, + 1405.0, + 1684.0, + 1405.0, + 1716.0, + 295.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1711.0, + 1407.0, + 1711.0, + 1407.0, + 1748.0, + 294.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1741.0, + 1405.0, + 1741.0, + 1405.0, + 1779.0, + 292.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1775.0, + 1152.0, + 1775.0, + 1152.0, + 1807.0, + 296.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 1775.0, + 1405.0, + 1775.0, + 1405.0, + 1807.0, + 1363.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1806.0, + 464.0, + 1806.0, + 464.0, + 1838.0, + 296.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 1806.0, + 1405.0, + 1806.0, + 1405.0, + 1838.0, + 618.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1837.0, + 668.0, + 1837.0, + 668.0, + 1868.0, + 296.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1837.0, + 1404.0, + 1837.0, + 1404.0, + 1868.0, + 934.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1478.0, + 1405.0, + 1478.0, + 1405.0, + 1514.0, + 294.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1509.0, + 1406.0, + 1509.0, + 1406.0, + 1545.0, + 293.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1542.0, + 431.0, + 1542.0, + 431.0, + 1574.0, + 293.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1542.0, + 1000.0, + 1542.0, + 1000.0, + 1574.0, + 453.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 1542.0, + 1407.0, + 1542.0, + 1407.0, + 1574.0, + 1018.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1570.0, + 1377.0, + 1570.0, + 1377.0, + 1604.0, + 293.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 918.0, + 1405.0, + 918.0, + 1405.0, + 957.0, + 292.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 952.0, + 1403.0, + 952.0, + 1403.0, + 983.0, + 293.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 983.0, + 876.0, + 983.0, + 876.0, + 1017.0, + 296.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 226.0, + 1409.0, + 226.0, + 1409.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 261.0, + 1337.0, + 261.0, + 1337.0, + 297.0, + 294.0, + 297.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1608, + 1406, + 1608, + 1406, + 1861, + 296, + 1861 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 1113, + 1405, + 1113, + 1405, + 1329, + 297, + 1329 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1342, + 1405, + 1342, + 1405, + 1497, + 298, + 1497 + ], + "score": 0.976 + }, + { + "category_id": 4, + "poly": [ + 296, + 943, + 1406, + 943, + 1406, + 1096, + 296, + 1096 + ], + "score": 0.967 + }, + { + "category_id": 3, + "poly": [ + 299, + 236, + 1388, + 236, + 1388, + 933, + 299, + 933 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 397, + 1875, + 1299, + 1875, + 1299, + 1954, + 397, + 1954 + ], + "score": 0.937 + }, + { + "category_id": 2, + "poly": [ + 293, + 1977, + 1412, + 1977, + 1412, + 2034, + 293, + 2034 + ], + "score": 0.933 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 295, + 1527, + 1288, + 1527, + 1288, + 1589, + 295, + 1589 + ], + "score": 0.857 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1901, + 1399, + 1901, + 1399, + 1930, + 1366, + 1930 + ], + "score": 0.853 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.795 + }, + { + "category_id": 14, + "poly": [ + 398, + 1875, + 1300, + 1875, + 1300, + 1953, + 398, + 1953 + ], + "score": 0.93, + "latex": "\\Delta ^ { \\mathrm { s s L } } ( n , r ) \\triangleq \\frac { A ^ { \\mathrm { s S L } } ( n , 1 ) - A ^ { \\mathrm { s S L } } ( n , r ) } { A ^ { \\mathrm { s S L } } ( n , 1 ) } \\ll \\Delta ^ { \\mathrm { s L } } ( n , r ) \\triangleq \\frac { A ^ { \\mathrm { s L } } ( n , 1 ) - A ^ { \\mathrm { s L } } ( n , r ) } { A ^ { \\mathrm { s L } } ( n , 1 ) } ." + }, + { + "category_id": 13, + "poly": [ + 900, + 1669, + 1177, + 1669, + 1177, + 1704, + 900, + 1704 + ], + "score": 0.92, + "latex": "A ^ { \\mathrm { S S L } } ( n , 1 ) - A ^ { \\mathrm { S S L } } ( n , r )" + }, + { + "category_id": 13, + "poly": [ + 750, + 1701, + 1007, + 1701, + 1007, + 1736, + 750, + 1736 + ], + "score": 0.92, + "latex": "A ^ { \\mathrm { { S L } } } ( n , 1 ) - A ^ { \\mathrm { { S L } } } ( \\dot { n } , r )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1765, + 563, + 1765, + 563, + 1801, + 298, + 1801 + ], + "score": 0.91, + "latex": "A ^ { \\mathrm { s s L } } ( n , r ) ) / A ^ { \\mathrm { s s L } } ( n , \\bar { 1 } )" + }, + { + "category_id": 13, + "poly": [ + 1081, + 1734, + 1406, + 1734, + 1406, + 1771, + 1081, + 1771 + ], + "score": 0.89, + "latex": "\\Delta ^ { \\mathrm { s s L } } ( n , r ) \\triangleq ( A ^ { \\mathrm { s s L } } ( n , 1 ) -" + }, + { + "category_id": 13, + "poly": [ + 834, + 1011, + 854, + 1011, + 854, + 1032, + 834, + 1032 + ], + "score": 0.75, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 942, + 1039, + 963, + 1039, + 963, + 1062, + 942, + 1062 + ], + "score": 0.72, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 492, + 1984, + 510, + 1984, + 510, + 2003, + 492, + 2003 + ], + "score": 0.7, + "latex": "_ n" + }, + { + "category_id": 13, + "poly": [ + 1259, + 482, + 1369, + 482, + 1369, + 506, + 1259, + 506 + ], + "score": 0.35, + "latex": "A _ { I D } ^ { S L } ( n , 0 . 0 0 2 5 )" + }, + { + "category_id": 13, + "poly": [ + 691, + 1611, + 725, + 1611, + 725, + 1639, + 691, + 1639 + ], + "score": 0.26, + "latex": "\\mathrm { I D }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 942.0, + 1411.0, + 942.0, + 1411.0, + 979.0, + 294.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 972.0, + 1406.0, + 972.0, + 1406.0, + 1010.0, + 294.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1004.0, + 833.0, + 1004.0, + 833.0, + 1041.0, + 295.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1004.0, + 1406.0, + 1004.0, + 1406.0, + 1041.0, + 855.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1034.0, + 941.0, + 1034.0, + 941.0, + 1070.0, + 293.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1034.0, + 1406.0, + 1034.0, + 1406.0, + 1070.0, + 964.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1067.0, + 866.0, + 1067.0, + 866.0, + 1097.0, + 294.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 236.0, + 948.0, + 236.0, + 948.0, + 264.0, + 915.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 259.0, + 359.0, + 259.0, + 359.0, + 514.0, + 287.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 262.0, + 348.0, + 262.0, + 348.0, + 286.0, + 321.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 262.0, + 951.0, + 262.0, + 951.0, + 312.0, + 894.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 289.0, + 950.0, + 289.0, + 950.0, + 329.0, + 896.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 316.0, + 348.0, + 316.0, + 348.0, + 340.0, + 321.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 310.0, + 929.0, + 310.0, + 929.0, + 403.0, + 895.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 330.0, + 945.0, + 330.0, + 945.0, + 355.0, + 918.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 370.0, + 348.0, + 370.0, + 348.0, + 446.0, + 322.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 350.0, + 783.0, + 350.0, + 783.0, + 464.0, + 670.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 362.0, + 945.0, + 362.0, + 945.0, + 499.0, + 901.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 348.0, + 1386.0, + 348.0, + 1386.0, + 462.0, + 1251.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 450.0, + 769.0, + 450.0, + 769.0, + 487.0, + 670.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 454.0, + 944.0, + 454.0, + 944.0, + 480.0, + 919.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 450.0, + 1369.0, + 450.0, + 1369.0, + 486.0, + 1252.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 476.0, + 777.0, + 476.0, + 777.0, + 512.0, + 670.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 486.0, + 944.0, + 486.0, + 944.0, + 510.0, + 920.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 477.0, + 1258.0, + 477.0, + 1258.0, + 510.0, + 1253.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 477.0, + 1376.0, + 477.0, + 1376.0, + 510.0, + 1370.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 515.0, + 362.0, + 515.0, + 362.0, + 532.0, + 346.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 512.0, + 478.0, + 512.0, + 478.0, + 536.0, + 382.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 512.0, + 784.0, + 512.0, + 784.0, + 536.0, + 479.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 512.0, + 1027.0, + 512.0, + 1027.0, + 533.0, + 975.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 514.0, + 1105.0, + 514.0, + 1105.0, + 533.0, + 1053.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 512.0, + 1183.0, + 512.0, + 1183.0, + 533.0, + 1132.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 512.0, + 1261.0, + 512.0, + 1261.0, + 533.0, + 1209.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 511.0, + 1345.0, + 511.0, + 1345.0, + 536.0, + 1282.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 525.0, + 726.0, + 525.0, + 726.0, + 553.0, + 408.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 523.0, + 1328.0, + 523.0, + 1328.0, + 553.0, + 1003.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 557.0, + 645.0, + 557.0, + 645.0, + 591.0, + 451.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 553.0, + 1243.0, + 553.0, + 1243.0, + 593.0, + 1050.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 601.0, + 364.0, + 601.0, + 364.0, + 854.0, + 298.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 649.0, + 941.0, + 649.0, + 941.0, + 670.0, + 919.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 695.0, + 762.0, + 695.0, + 762.0, + 733.0, + 666.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 694.0, + 943.0, + 694.0, + 943.0, + 718.0, + 919.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 693.0, + 1348.0, + 693.0, + 1348.0, + 734.0, + 1251.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 731.0, + 348.0, + 731.0, + 348.0, + 758.0, + 322.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 718.0, + 783.0, + 718.0, + 783.0, + 783.0, + 664.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 739.0, + 943.0, + 739.0, + 943.0, + 763.0, + 919.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 718.0, + 1380.0, + 718.0, + 1380.0, + 783.0, + 1251.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 787.0, + 347.0, + 787.0, + 347.0, + 809.0, + 323.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 770.0, + 762.0, + 770.0, + 762.0, + 807.0, + 666.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 784.0, + 943.0, + 784.0, + 943.0, + 807.0, + 919.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 770.0, + 1348.0, + 770.0, + 1348.0, + 807.0, + 1253.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 796.0, + 773.0, + 796.0, + 773.0, + 832.0, + 664.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 796.0, + 1378.0, + 796.0, + 1378.0, + 831.0, + 1251.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 821.0, + 783.0, + 821.0, + 783.0, + 857.0, + 663.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 829.0, + 941.0, + 829.0, + 941.0, + 852.0, + 918.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 821.0, + 1378.0, + 821.0, + 1378.0, + 856.0, + 1249.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 842.0, + 349.0, + 842.0, + 349.0, + 866.0, + 322.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 861.0, + 362.0, + 861.0, + 362.0, + 877.0, + 346.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 855.0, + 785.0, + 855.0, + 785.0, + 882.0, + 380.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 858.0, + 1024.0, + 858.0, + 1024.0, + 880.0, + 974.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 858.0, + 1102.0, + 858.0, + 1102.0, + 878.0, + 1050.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 858.0, + 1179.0, + 858.0, + 1179.0, + 878.0, + 1128.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 856.0, + 1258.0, + 856.0, + 1258.0, + 881.0, + 1204.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 853.0, + 1341.0, + 853.0, + 1341.0, + 882.0, + 1277.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 871.0, + 725.0, + 871.0, + 725.0, + 897.0, + 409.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 870.0, + 1322.0, + 870.0, + 1322.0, + 898.0, + 1002.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 901.0, + 659.0, + 901.0, + 659.0, + 934.0, + 437.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 900.0, + 1253.0, + 900.0, + 1253.0, + 937.0, + 1034.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 676.5, + 350.0, + 676.5, + 350.0, + 698.0, + 319.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1970.0, + 491.0, + 1970.0, + 491.0, + 2014.0, + 328.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1970.0, + 1406.0, + 1970.0, + 1406.0, + 2014.0, + 511.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 1344.0, + 2001.0, + 1344.0, + 2039.0, + 293.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1528.0, + 1290.0, + 1528.0, + 1290.0, + 1562.0, + 295.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1558.0, + 804.0, + 1558.0, + 804.0, + 1591.0, + 361.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 837.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1609.0, + 690.0, + 1609.0, + 690.0, + 1644.0, + 294.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 1609.0, + 1404.0, + 1609.0, + 1404.0, + 1644.0, + 726.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1641.0, + 1406.0, + 1641.0, + 1406.0, + 1676.0, + 295.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1667.0, + 899.0, + 1667.0, + 899.0, + 1706.0, + 291.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 1667.0, + 1407.0, + 1667.0, + 1407.0, + 1706.0, + 1178.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1698.0, + 749.0, + 1698.0, + 749.0, + 1741.0, + 290.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 1698.0, + 1412.0, + 1698.0, + 1412.0, + 1741.0, + 1008.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1731.0, + 1080.0, + 1731.0, + 1080.0, + 1776.0, + 290.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1760.0, + 297.0, + 1760.0, + 297.0, + 1806.0, + 291.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 1760.0, + 1409.0, + 1760.0, + 1409.0, + 1806.0, + 564.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1798.0, + 1404.0, + 1798.0, + 1404.0, + 1833.0, + 295.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1830.0, + 783.0, + 1830.0, + 783.0, + 1861.0, + 296.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1114.0, + 1405.0, + 1114.0, + 1405.0, + 1148.0, + 295.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1142.0, + 1405.0, + 1142.0, + 1405.0, + 1179.0, + 294.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1174.0, + 1407.0, + 1174.0, + 1407.0, + 1209.0, + 295.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1206.0, + 1406.0, + 1206.0, + 1406.0, + 1237.0, + 295.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1235.0, + 1402.0, + 1235.0, + 1402.0, + 1265.0, + 295.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1267.0, + 1403.0, + 1267.0, + 1403.0, + 1298.0, + 295.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1294.0, + 1408.0, + 1294.0, + 1408.0, + 1331.0, + 294.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1342.0, + 1403.0, + 1342.0, + 1403.0, + 1376.0, + 295.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1373.0, + 1407.0, + 1373.0, + 1407.0, + 1408.0, + 293.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1402.0, + 1407.0, + 1402.0, + 1407.0, + 1438.0, + 292.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1435.0, + 1405.0, + 1435.0, + 1405.0, + 1469.0, + 296.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1465.0, + 1270.0, + 1465.0, + 1270.0, + 1498.0, + 295.0, + 1498.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1349, + 1404, + 1349, + 1404, + 1655, + 298, + 1655 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 932, + 1404, + 932, + 1404, + 1117, + 298, + 1117 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1742, + 1403, + 1742, + 1403, + 1866, + 299, + 1866 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 1211, + 1403, + 1211, + 1403, + 1333, + 298, + 1333 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1881, + 1403, + 1881, + 1403, + 2036, + 299, + 2036 + ], + "score": 0.975 + }, + { + "category_id": 3, + "poly": [ + 468, + 229, + 1228, + 229, + 1228, + 502, + 468, + 502 + ], + "score": 0.968 + }, + { + "category_id": 4, + "poly": [ + 297, + 524, + 1405, + 524, + 1405, + 830, + 297, + 830 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 299, + 855, + 1400, + 855, + 1400, + 918, + 299, + 918 + ], + "score": 0.947 + }, + { + "category_id": 0, + "poly": [ + 299, + 1153, + 497, + 1153, + 497, + 1186, + 299, + 1186 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 302, + 1689, + 862, + 1689, + 862, + 1720, + 302, + 1720 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.704 + }, + { + "category_id": 13, + "poly": [ + 961, + 1973, + 1101, + 1973, + 1101, + 2006, + 961, + 2006 + ], + "score": 0.92, + "latex": "\\xi \\sim \\mathcal { N } ( 0 , I )" + }, + { + "category_id": 13, + "poly": [ + 841, + 1973, + 910, + 1973, + 910, + 2006, + 841, + 2006 + ], + "score": 0.92, + "latex": "\\{ 0 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 1112, + 1944, + 1180, + 1944, + 1180, + 1974, + 1112, + 1974 + ], + "score": 0.91, + "latex": "\\rho > 0" + }, + { + "category_id": 13, + "poly": [ + 667, + 2006, + 919, + 2006, + 919, + 2035, + 667, + 2035 + ], + "score": 0.9, + "latex": "x = e _ { 1 } - q \\tau e _ { 2 } + \\rho \\xi" + }, + { + "category_id": 13, + "poly": [ + 445, + 710, + 493, + 710, + 493, + 739, + 445, + 739 + ], + "score": 0.89, + "latex": "e _ { 3 : d }" + }, + { + "category_id": 13, + "poly": [ + 993, + 1945, + 1062, + 1945, + 1062, + 1971, + 993, + 1971 + ], + "score": 0.89, + "latex": "\\tau > 0" + }, + { + "category_id": 13, + "poly": [ + 562, + 1886, + 628, + 1886, + 628, + 1913, + 562, + 1913 + ], + "score": 0.88, + "latex": "e _ { 1 } , e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1334, + 556, + 1402, + 556, + 1402, + 585, + 1334, + 585 + ], + "score": 0.87, + "latex": "d - 2" + }, + { + "category_id": 13, + "poly": [ + 1050, + 560, + 1097, + 560, + 1097, + 587, + 1050, + 587 + ], + "score": 0.87, + "latex": "e _ { 3 : d }" + }, + { + "category_id": 13, + "poly": [ + 516, + 2004, + 605, + 2004, + 605, + 2035, + 516, + 2035 + ], + "score": 0.86, + "latex": "( y = 1 )" + }, + { + "category_id": 13, + "poly": [ + 1194, + 2004, + 1285, + 2004, + 1285, + 2035, + 1194, + 2035 + ], + "score": 0.86, + "latex": "( y \\ : = \\ : 2 )" + }, + { + "category_id": 13, + "poly": [ + 1291, + 742, + 1320, + 742, + 1320, + 768, + 1291, + 768 + ], + "score": 0.86, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 851, + 620, + 880, + 620, + 880, + 647, + 851, + 647 + ], + "score": 0.86, + "latex": "e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1120, + 711, + 1149, + 711, + 1149, + 738, + 1120, + 738 + ], + "score": 0.86, + "latex": "e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1198, + 711, + 1227, + 711, + 1227, + 738, + 1198, + 738 + ], + "score": 0.85, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 933, + 619, + 963, + 619, + 963, + 647, + 933, + 647 + ], + "score": 0.85, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 451, + 802, + 480, + 802, + 480, + 829, + 451, + 829 + ], + "score": 0.85, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 403, + 650, + 431, + 650, + 431, + 677, + 403, + 677 + ], + "score": 0.84, + "latex": "e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 861, + 772, + 890, + 772, + 890, + 799, + 861, + 799 + ], + "score": 0.84, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 951, + 530, + 980, + 530, + 980, + 555, + 951, + 555 + ], + "score": 0.84, + "latex": "e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1032, + 530, + 1061, + 530, + 1061, + 555, + 1032, + 555 + ], + "score": 0.84, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1212, + 743, + 1241, + 743, + 1241, + 768, + 1212, + 768 + ], + "score": 0.84, + "latex": "e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 914, + 1913, + 938, + 1913, + 938, + 1939, + 914, + 1939 + ], + "score": 0.83, + "latex": "\\mathcal { P }" + }, + { + "category_id": 13, + "poly": [ + 1346, + 2006, + 1404, + 2006, + 1404, + 2034, + 1346, + 2034 + ], + "score": 0.82, + "latex": "x =" + }, + { + "category_id": 13, + "poly": [ + 646, + 1978, + 663, + 1978, + 663, + 2005, + 646, + 2005 + ], + "score": 0.81, + "latex": "q" + }, + { + "category_id": 13, + "poly": [ + 1118, + 1883, + 1136, + 1883, + 1136, + 1909, + 1118, + 1909 + ], + "score": 0.81, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 662, + 1948, + 682, + 1948, + 682, + 1970, + 662, + 1970 + ], + "score": 0.79, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 371, + 557, + 389, + 557, + 389, + 584, + 371, + 584 + ], + "score": 0.77, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 457, + 1943, + 503, + 1943, + 503, + 1975, + 457, + 1975 + ], + "score": 0.74, + "latex": "y \\in" + }, + { + "category_id": 13, + "poly": [ + 1279, + 681, + 1298, + 681, + 1298, + 705, + 1279, + 705 + ], + "score": 0.69, + "latex": "v" + }, + { + "category_id": 13, + "poly": [ + 632, + 682, + 651, + 682, + 651, + 704, + 632, + 704 + ], + "score": 0.67, + "latex": "v" + }, + { + "category_id": 13, + "poly": [ + 1301, + 970, + 1321, + 970, + 1321, + 992, + 1301, + 992 + ], + "score": 0.66, + "latex": "n" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 247.0, + 574.0, + 247.0, + 574.0, + 276.0, + 490.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 257.0, + 1171.0, + 257.0, + 1171.0, + 284.0, + 1023.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 272.0, + 572.0, + 272.0, + 572.0, + 298.0, + 487.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 281.0, + 671.0, + 281.0, + 671.0, + 297.0, + 650.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 285.0, + 689.0, + 285.0, + 689.0, + 293.0, + 681.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 285.0, + 863.0, + 285.0, + 863.0, + 298.0, + 849.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 298.0, + 511.0, + 298.0, + 511.0, + 319.0, + 489.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 291.0, + 574.0, + 291.0, + 574.0, + 320.0, + 512.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 293.0, + 685.0, + 293.0, + 685.0, + 320.0, + 630.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 294.0, + 708.0, + 294.0, + 708.0, + 305.0, + 694.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 302.0, + 777.0, + 302.0, + 777.0, + 314.0, + 763.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 300.0, + 822.0, + 300.0, + 822.0, + 307.0, + 814.0, + 307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 316.0, + 628.0, + 316.0, + 628.0, + 329.0, + 617.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 325.0, + 645.0, + 325.0, + 645.0, + 332.0, + 636.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 327.0, + 701.0, + 327.0, + 701.0, + 332.0, + 696.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 319.0, + 1031.0, + 319.0, + 1031.0, + 355.0, + 987.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 332.0, + 667.0, + 332.0, + 667.0, + 339.0, + 658.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 332.0, + 687.0, + 332.0, + 687.0, + 340.0, + 678.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 332.0, + 793.0, + 332.0, + 793.0, + 344.0, + 781.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 332.0, + 834.0, + 332.0, + 834.0, + 339.0, + 823.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 335.0, + 619.0, + 335.0, + 619.0, + 344.0, + 610.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 341.0, + 654.0, + 341.0, + 654.0, + 350.0, + 645.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 343.0, + 853.0, + 343.0, + 853.0, + 350.0, + 845.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 352.0, + 691.0, + 352.0, + 691.0, + 362.0, + 683.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 355.0, + 890.0, + 355.0, + 890.0, + 367.0, + 877.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 362.0, + 1061.0, + 362.0, + 1061.0, + 369.0, + 1053.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 355.0, + 1209.0, + 355.0, + 1209.0, + 385.0, + 1080.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 369.0, + 913.0, + 369.0, + 913.0, + 381.0, + 901.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 366.0, + 1067.0, + 366.0, + 1067.0, + 396.0, + 1036.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 378.0, + 1210.0, + 378.0, + 1210.0, + 403.0, + 1112.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 410.0, + 501.0, + 410.0, + 501.0, + 423.0, + 487.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 414.0, + 535.0, + 414.0, + 535.0, + 423.0, + 526.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 405.0, + 553.0, + 405.0, + 553.0, + 414.0, + 543.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 417.0, + 568.0, + 417.0, + 568.0, + 425.0, + 559.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 417.0, + 608.0, + 417.0, + 608.0, + 429.0, + 596.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 413.0, + 737.0, + 413.0, + 737.0, + 430.0, + 721.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 403.0, + 753.0, + 403.0, + 753.0, + 415.0, + 741.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 410.0, + 795.0, + 410.0, + 795.0, + 420.0, + 786.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 403.0, + 874.0, + 403.0, + 874.0, + 415.0, + 861.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 406.0, + 923.0, + 406.0, + 923.0, + 422.0, + 907.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 416.0, + 1157.0, + 416.0, + 1157.0, + 432.0, + 1141.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 436.0, + 512.0, + 436.0, + 512.0, + 448.0, + 499.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 429.0, + 728.0, + 429.0, + 728.0, + 441.0, + 715.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 441.0, + 739.0, + 441.0, + 739.0, + 450.0, + 731.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 433.0, + 1037.0, + 433.0, + 1037.0, + 466.0, + 984.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 425.0, + 1170.0, + 425.0, + 1170.0, + 458.0, + 1137.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 446.0, + 599.0, + 446.0, + 599.0, + 456.0, + 591.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 452.0, + 706.0, + 452.0, + 706.0, + 465.0, + 692.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 447.0, + 731.0, + 447.0, + 731.0, + 459.0, + 719.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 453.0, + 802.0, + 453.0, + 802.0, + 460.0, + 794.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 445.0, + 1090.0, + 445.0, + 1090.0, + 479.0, + 1060.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 457.0, + 539.0, + 457.0, + 539.0, + 468.0, + 526.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 457.0, + 587.0, + 457.0, + 587.0, + 468.0, + 574.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 460.0, + 824.0, + 460.0, + 824.0, + 468.0, + 814.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 464.0, + 753.0, + 464.0, + 753.0, + 474.0, + 734.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 273.5, + 811.0, + 273.5, + 811.0, + 304.5, + 759.0, + 304.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 280.0, + 1143.0, + 280.0, + 1143.0, + 317.0, + 1089.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 313.5, + 667.0, + 313.5, + 667.0, + 328.0, + 647.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 404.5, + 902.0, + 404.5, + 902.0, + 427.5, + 881.0, + 427.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 523.0, + 950.0, + 523.0, + 950.0, + 559.0, + 295.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 523.0, + 1031.0, + 523.0, + 1031.0, + 559.0, + 981.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1062.0, + 523.0, + 1405.0, + 523.0, + 1405.0, + 559.0, + 1062.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 555.0, + 370.0, + 555.0, + 370.0, + 591.0, + 294.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 555.0, + 1049.0, + 555.0, + 1049.0, + 591.0, + 390.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 555.0, + 1333.0, + 555.0, + 1333.0, + 591.0, + 1098.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 555.0, + 1407.0, + 555.0, + 1407.0, + 591.0, + 1403.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 583.0, + 1406.0, + 583.0, + 1406.0, + 621.0, + 294.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 617.0, + 850.0, + 617.0, + 850.0, + 649.0, + 296.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 617.0, + 932.0, + 617.0, + 932.0, + 649.0, + 881.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 617.0, + 1405.0, + 617.0, + 1405.0, + 649.0, + 964.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 644.0, + 402.0, + 644.0, + 402.0, + 682.0, + 294.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 644.0, + 1407.0, + 644.0, + 1407.0, + 682.0, + 432.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 677.0, + 631.0, + 677.0, + 631.0, + 713.0, + 295.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 677.0, + 1278.0, + 677.0, + 1278.0, + 713.0, + 652.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 677.0, + 1405.0, + 677.0, + 1405.0, + 713.0, + 1299.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 709.0, + 444.0, + 709.0, + 444.0, + 740.0, + 296.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 709.0, + 1119.0, + 709.0, + 1119.0, + 740.0, + 494.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 709.0, + 1197.0, + 709.0, + 1197.0, + 740.0, + 1150.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 709.0, + 1402.0, + 709.0, + 1402.0, + 740.0, + 1228.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 735.0, + 1211.0, + 735.0, + 1211.0, + 773.0, + 294.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1242.0, + 735.0, + 1290.0, + 735.0, + 1290.0, + 773.0, + 1242.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 735.0, + 1405.0, + 735.0, + 1405.0, + 773.0, + 1321.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 766.0, + 860.0, + 766.0, + 860.0, + 804.0, + 294.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 766.0, + 1408.0, + 766.0, + 1408.0, + 804.0, + 891.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 796.0, + 450.0, + 796.0, + 450.0, + 834.0, + 292.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 796.0, + 1224.0, + 796.0, + 1224.0, + 834.0, + 481.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1148.0, + 503.0, + 1148.0, + 503.0, + 1194.0, + 291.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1688.0, + 865.0, + 1688.0, + 865.0, + 1724.0, + 294.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1350.0, + 1405.0, + 1350.0, + 1405.0, + 1381.0, + 296.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1381.0, + 1406.0, + 1381.0, + 1406.0, + 1413.0, + 296.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1409.0, + 1406.0, + 1409.0, + 1406.0, + 1445.0, + 294.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1441.0, + 1404.0, + 1441.0, + 1404.0, + 1473.0, + 296.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1470.0, + 1405.0, + 1470.0, + 1405.0, + 1505.0, + 293.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1501.0, + 1405.0, + 1501.0, + 1405.0, + 1535.0, + 294.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1530.0, + 1406.0, + 1530.0, + 1406.0, + 1568.0, + 292.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1560.0, + 1406.0, + 1560.0, + 1406.0, + 1597.0, + 293.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1592.0, + 1405.0, + 1592.0, + 1405.0, + 1628.0, + 294.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1623.0, + 1346.0, + 1623.0, + 1346.0, + 1658.0, + 294.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 928.0, + 1404.0, + 928.0, + 1404.0, + 970.0, + 293.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 963.0, + 1300.0, + 963.0, + 1300.0, + 996.0, + 293.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 963.0, + 1404.0, + 963.0, + 1404.0, + 996.0, + 1322.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 993.0, + 1406.0, + 993.0, + 1406.0, + 1030.0, + 293.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1017.0, + 1409.0, + 1017.0, + 1409.0, + 1063.0, + 292.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1051.0, + 1405.0, + 1051.0, + 1405.0, + 1089.0, + 295.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1086.0, + 1100.0, + 1086.0, + 1100.0, + 1119.0, + 293.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1740.0, + 1405.0, + 1740.0, + 1405.0, + 1780.0, + 292.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1774.0, + 1405.0, + 1774.0, + 1405.0, + 1808.0, + 293.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1804.0, + 1407.0, + 1804.0, + 1407.0, + 1840.0, + 294.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1831.0, + 583.0, + 1831.0, + 583.0, + 1873.0, + 293.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1208.0, + 1407.0, + 1208.0, + 1407.0, + 1247.0, + 292.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1240.0, + 1405.0, + 1240.0, + 1405.0, + 1277.0, + 293.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1272.0, + 1404.0, + 1272.0, + 1404.0, + 1308.0, + 293.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1301.0, + 687.0, + 1301.0, + 687.0, + 1338.0, + 295.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1879.0, + 561.0, + 1879.0, + 561.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1879.0, + 1117.0, + 1879.0, + 1117.0, + 1916.0, + 629.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1879.0, + 1404.0, + 1879.0, + 1404.0, + 1916.0, + 1137.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 913.0, + 1911.0, + 913.0, + 1947.0, + 294.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1911.0, + 1405.0, + 1911.0, + 1405.0, + 1947.0, + 939.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1941.0, + 456.0, + 1941.0, + 456.0, + 1978.0, + 294.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1941.0, + 661.0, + 1941.0, + 661.0, + 1978.0, + 504.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 1941.0, + 992.0, + 1941.0, + 992.0, + 1978.0, + 683.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1941.0, + 1111.0, + 1941.0, + 1111.0, + 1978.0, + 1063.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1941.0, + 1405.0, + 1941.0, + 1405.0, + 1978.0, + 1181.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1970.0, + 645.0, + 1970.0, + 645.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 1970.0, + 840.0, + 1970.0, + 840.0, + 2008.0, + 664.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 1970.0, + 960.0, + 1970.0, + 960.0, + 2008.0, + 911.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1970.0, + 1407.0, + 1970.0, + 1407.0, + 2008.0, + 1102.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2000.0, + 515.0, + 2000.0, + 515.0, + 2038.0, + 293.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 2000.0, + 666.0, + 2000.0, + 666.0, + 2038.0, + 606.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 2000.0, + 1193.0, + 2000.0, + 1193.0, + 2038.0, + 920.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 2000.0, + 1345.0, + 2000.0, + 1345.0, + 2038.0, + 1286.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 856.0, + 1405.0, + 856.0, + 1405.0, + 888.0, + 297.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 885.0, + 1304.0, + 885.0, + 1304.0, + 921.0, + 295.0, + 921.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 383, + 1405, + 383, + 1405, + 740, + 297, + 740 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 754, + 1404, + 754, + 1404, + 1031, + 297, + 1031 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1507, + 1404, + 1507, + 1404, + 1692, + 298, + 1692 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1706, + 1405, + 1706, + 1405, + 1953, + 297, + 1953 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1405, + 229, + 1405, + 365, + 298, + 365 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1302, + 1405, + 1302, + 1405, + 1427, + 298, + 1427 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 1043, + 1404, + 1043, + 1404, + 1159, + 298, + 1159 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 307, + 1172, + 1403, + 1172, + 1403, + 1278, + 307, + 1278 + ], + "score": 0.96 + }, + { + "category_id": 2, + "poly": [ + 299, + 1978, + 1403, + 1978, + 1403, + 2034, + 299, + 2034 + ], + "score": 0.94 + }, + { + "category_id": 0, + "poly": [ + 298, + 1455, + 933, + 1455, + 933, + 1488, + 298, + 1488 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.9 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.79 + }, + { + "category_id": 13, + "poly": [ + 417, + 512, + 786, + 512, + 786, + 547, + 417, + 547 + ], + "score": 0.93, + "latex": "f _ { W _ { 1 } , W _ { 2 } } ( x ) _ { y } \\geq f _ { W _ { 1 } , W _ { 2 } } ( x ) _ { y ^ { \\prime } } + 1" + }, + { + "category_id": 13, + "poly": [ + 647, + 415, + 797, + 415, + 797, + 449, + 647, + 449 + ], + "score": 0.93, + "latex": "W _ { 1 } \\in \\mathbb { R } ^ { m \\times d }" + }, + { + "category_id": 13, + "poly": [ + 870, + 478, + 1167, + 478, + 1167, + 514, + 870, + 514 + ], + "score": 0.93, + "latex": "\\| W _ { 1 } ^ { \\top } W _ { 1 } \\| _ { F } ^ { 2 } + \\| W _ { 2 } ^ { \\top } W _ { 2 } \\| _ { F } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 386, + 1112, + 548, + 1112, + 548, + 1158, + 386, + 1158 + ], + "score": 0.93, + "latex": "1 - O ( e ^ { - d ^ { \\frac { 1 } { 1 0 } } } )" + }, + { + "category_id": 13, + "poly": [ + 852, + 416, + 1000, + 416, + 1000, + 449, + 852, + 449 + ], + "score": 0.93, + "latex": "W _ { 2 } \\in \\mathbb { R } ^ { 3 \\times m }" + }, + { + "category_id": 13, + "poly": [ + 419, + 450, + 558, + 450, + 558, + 480, + 419, + 480 + ], + "score": 0.92, + "latex": "W _ { \\mathrm { S L } } = W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 378, + 606, + 544, + 606, + 544, + 640, + 378, + 640 + ], + "score": 0.92, + "latex": "W _ { \\mathrm { S S L } } \\in \\mathbb { R } ^ { m \\times d }" + }, + { + "category_id": 13, + "poly": [ + 1108, + 1076, + 1206, + 1076, + 1206, + 1112, + 1108, + 1112 + ], + "score": 0.92, + "latex": "n _ { 3 } \\leq d ^ { \\frac { 1 } { 5 } }" + }, + { + "category_id": 13, + "poly": [ + 1297, + 1040, + 1399, + 1040, + 1399, + 1078, + 1297, + 1078 + ], + "score": 0.92, + "latex": "\\rho = d ^ { - \\frac { 1 } { 5 } }" + }, + { + "category_id": 13, + "poly": [ + 817, + 1079, + 1055, + 1079, + 1055, + 1114, + 817, + 1114 + ], + "score": 0.91, + "latex": "n _ { 1 } , n _ { 2 } = \\Theta ( \\mathrm { p o l y } ( d ) )" + }, + { + "category_id": 13, + "poly": [ + 746, + 606, + 1278, + 606, + 1278, + 644, + 746, + 644 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { - \\mathbb { E } [ ( W ( x + \\xi ) ) ^ { T } ( W ( x + \\xi ^ { \\prime } ) ) ] + \\frac { 1 } { 2 } \\| W ^ { \\top } W \\| _ { F } ^ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 296, + 232, + 497, + 232, + 497, + 264, + 296, + 264 + ], + "score": 0.91, + "latex": "- e _ { 1 } - q \\tau e _ { 2 } + \\rho \\xi" + }, + { + "category_id": 13, + "poly": [ + 297, + 418, + 394, + 418, + 394, + 450, + 297, + 450 + ], + "score": 0.91, + "latex": "W _ { 2 } W _ { 1 } x" + }, + { + "category_id": 13, + "poly": [ + 847, + 232, + 990, + 232, + 990, + 263, + 847, + 263 + ], + "score": 0.91, + "latex": "x = e _ { 2 } + \\rho \\xi" + }, + { + "category_id": 13, + "poly": [ + 1240, + 544, + 1403, + 544, + 1403, + 577, + 1240, + 577 + ], + "score": 0.91, + "latex": "( x + \\xi , x + \\xi ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 1200, + 512, + 1280, + 512, + 1280, + 546, + 1200, + 546 + ], + "score": 0.9, + "latex": "y ^ { \\prime } \\ne y" + }, + { + "category_id": 13, + "poly": [ + 923, + 515, + 988, + 515, + 988, + 547, + 923, + 547 + ], + "score": 0.9, + "latex": "( x , y )" + }, + { + "category_id": 13, + "poly": [ + 346, + 1076, + 431, + 1076, + 431, + 1109, + 346, + 1109 + ], + "score": 0.9, + "latex": "\\tau = d ^ { \\frac { 1 } { 5 } }" + }, + { + "category_id": 13, + "poly": [ + 1070, + 1171, + 1377, + 1171, + 1377, + 1212, + 1070, + 1212 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\sum _ { i = 1 } ^ { m } \\langle e _ { 2 } , w _ { i } \\rangle ^ { 2 } \\leq O ( d ^ { - \\frac { 1 } { 1 0 } } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 725, + 1304, + 779, + 1304, + 779, + 1334, + 725, + 1334 + ], + "score": 0.9, + "latex": "W _ { \\mathrm { S L } }" + }, + { + "category_id": 13, + "poly": [ + 1184, + 1126, + 1261, + 1126, + 1261, + 1155, + 1184, + 1155 + ], + "score": 0.9, + "latex": "m \\geq 3" + }, + { + "category_id": 13, + "poly": [ + 1125, + 1210, + 1398, + 1210, + 1398, + 1248, + 1125, + 1248 + ], + "score": 0.89, + "latex": "\\| \\Pi e _ { 2 } \\| _ { 2 } \\ge 1 - O ( d ^ { - \\frac { 1 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 1115, + 419, + 1199, + 419, + 1199, + 449, + 1115, + 449 + ], + "score": 0.89, + "latex": "m \\geq 3" + }, + { + "category_id": 13, + "poly": [ + 505, + 1049, + 619, + 1049, + 619, + 1078, + 505, + 1078 + ], + "score": 0.89, + "latex": "n _ { 1 } , n _ { 2 } , n _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 378, + 1173, + 684, + 1173, + 684, + 1211, + 378, + 1211 + ], + "score": 0.89, + "latex": "W _ { \\mathrm { S L } } = [ w _ { 1 } , w _ { 2 } , \\cdot \\cdot \\cdot , w _ { m } ] ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 381, + 1213, + 703, + 1213, + 703, + 1248, + 381, + 1248 + ], + "score": 0.89, + "latex": "W _ { \\mathrm { S S L } } = [ \\tilde { w } _ { 1 } , \\tilde { w } _ { 2 } , \\cdot \\cdot \\cdot , \\tilde { w } _ { m } ] ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 1247, + 381, + 1405, + 381, + 1405, + 419, + 1247, + 419 + ], + "score": 0.88, + "latex": "f _ { W _ { 1 } , W _ { 2 } } ( x ) \\triangleq" + }, + { + "category_id": 13, + "poly": [ + 298, + 676, + 442, + 676, + 442, + 712, + 298, + 712 + ], + "score": 0.87, + "latex": "\\begin{array} { r } { \\frac { 1 } { 2 } \\lVert \\boldsymbol { W } ^ { \\top } \\boldsymbol { W } \\rVert _ { F } ^ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 749, + 850, + 778, + 850, + 778, + 878, + 749, + 878 + ], + "score": 0.86, + "latex": "e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 802, + 575, + 826, + 575, + 826, + 606, + 802, + 606 + ], + "score": 0.86, + "latex": "\\xi ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 299, + 712, + 377, + 712, + 377, + 740, + 299, + 740 + ], + "score": 0.86, + "latex": "W _ { \\mathrm { S S L } } x" + }, + { + "category_id": 13, + "poly": [ + 797, + 1246, + 859, + 1246, + 859, + 1277, + 797, + 1277 + ], + "score": 0.86, + "latex": "W _ { \\mathrm { S S L } }" + }, + { + "category_id": 13, + "poly": [ + 510, + 1860, + 629, + 1860, + 629, + 1889, + 510, + 1889 + ], + "score": 0.86, + "latex": "r = 0 . 0 0 2" + }, + { + "category_id": 13, + "poly": [ + 720, + 232, + 794, + 232, + 794, + 261, + 720, + 261 + ], + "score": 0.86, + "latex": "( y = 3 )" + }, + { + "category_id": 13, + "poly": [ + 691, + 261, + 772, + 261, + 772, + 307, + 691, + 307 + ], + "score": 0.86, + "latex": "\\frac { \\mathcal { P } ( y = 3 ) } { \\mathcal { P } ( y = 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 1099, + 648, + 1123, + 648, + 1123, + 680, + 1099, + 680 + ], + "score": 0.85, + "latex": "\\xi ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1034, + 942, + 1064, + 942, + 1064, + 969, + 1034, + 969 + ], + "score": 0.85, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 869, + 338, + 898, + 338, + 898, + 364, + 869, + 364 + ], + "score": 0.85, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 680, + 309, + 708, + 309, + 708, + 334, + 680, + 334 + ], + "score": 0.85, + "latex": "e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 759, + 310, + 787, + 310, + 787, + 334, + 759, + 334 + ], + "score": 0.85, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 432, + 643, + 454, + 643, + 454, + 675, + 432, + 675 + ], + "score": 0.84, + "latex": "\\hat { \\mathbb { E } }" + }, + { + "category_id": 13, + "poly": [ + 1032, + 649, + 1049, + 649, + 1049, + 680, + 1032, + 680 + ], + "score": 0.84, + "latex": "\\xi" + }, + { + "category_id": 13, + "poly": [ + 353, + 337, + 382, + 337, + 382, + 365, + 353, + 365 + ], + "score": 0.84, + "latex": "e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 379, + 1338, + 407, + 1338, + 407, + 1364, + 379, + 1364 + ], + "score": 0.84, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 381, + 1399, + 410, + 1399, + 410, + 1425, + 381, + 1425 + ], + "score": 0.84, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 732, + 576, + 750, + 576, + 750, + 607, + 732, + 607 + ], + "score": 0.83, + "latex": "\\xi" + }, + { + "category_id": 13, + "poly": [ + 379, + 912, + 408, + 912, + 408, + 939, + 379, + 939 + ], + "score": 0.83, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1246, + 1369, + 1275, + 1369, + 1275, + 1395, + 1246, + 1395 + ], + "score": 0.82, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 913, + 326, + 913, + 326, + 939, + 298, + 939 + ], + "score": 0.82, + "latex": "e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 786, + 263, + 954, + 263, + 954, + 308, + 786, + 308 + ], + "score": 0.8, + "latex": "\\frac { \\mathcal { P } ( y = 3 ) } { \\mathcal { P } ( y = 2 ) } = o ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 531, + 1251, + 559, + 1251, + 559, + 1276, + 531, + 1276 + ], + "score": 0.79, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 619, + 714, + 638, + 714, + 638, + 735, + 619, + 735 + ], + "score": 0.78, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 373, + 580, + 392, + 580, + 392, + 602, + 373, + 602 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 778, + 789, + 796, + 789, + 796, + 814, + 778, + 814 + ], + "score": 0.68, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 515, + 819, + 533, + 819, + 533, + 845, + 515, + 845 + ], + "score": 0.65, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 971, + 1177, + 1005, + 1177, + 1005, + 1206, + 971, + 1206 + ], + "score": 0.65, + "latex": "S L ," + }, + { + "category_id": 13, + "poly": [ + 1006, + 1215, + 1058, + 1215, + 1058, + 1245, + 1006, + 1245 + ], + "score": 0.63, + "latex": "S S L ," + }, + { + "category_id": 13, + "poly": [ + 408, + 1247, + 432, + 1247, + 432, + 1273, + 408, + 1273 + ], + "score": 0.34, + "latex": "\\Pi" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1972.0, + 1407.0, + 1972.0, + 1407.0, + 2011.0, + 329.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 2004.0, + 885.0, + 2004.0, + 885.0, + 2034.0, + 297.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1455.0, + 936.0, + 1455.0, + 936.0, + 1490.0, + 294.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 380.0, + 1246.0, + 380.0, + 1246.0, + 421.0, + 291.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 411.0, + 646.0, + 411.0, + 646.0, + 455.0, + 395.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 411.0, + 851.0, + 411.0, + 851.0, + 455.0, + 798.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 411.0, + 1114.0, + 411.0, + 1114.0, + 455.0, + 1001.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 411.0, + 1407.0, + 411.0, + 1407.0, + 455.0, + 1200.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 447.0, + 418.0, + 447.0, + 418.0, + 484.0, + 295.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 447.0, + 1406.0, + 447.0, + 1406.0, + 484.0, + 559.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 478.0, + 869.0, + 478.0, + 869.0, + 515.0, + 295.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 478.0, + 1403.0, + 478.0, + 1403.0, + 515.0, + 1168.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 511.0, + 416.0, + 511.0, + 416.0, + 548.0, + 292.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 511.0, + 922.0, + 511.0, + 922.0, + 548.0, + 787.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 511.0, + 1199.0, + 511.0, + 1199.0, + 548.0, + 989.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 511.0, + 1408.0, + 511.0, + 1408.0, + 548.0, + 1281.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 541.0, + 1239.0, + 541.0, + 1239.0, + 580.0, + 294.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 541.0, + 1407.0, + 541.0, + 1407.0, + 580.0, + 1404.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 569.0, + 372.0, + 569.0, + 372.0, + 612.0, + 292.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 569.0, + 731.0, + 569.0, + 731.0, + 612.0, + 393.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 569.0, + 801.0, + 569.0, + 801.0, + 612.0, + 751.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 569.0, + 1408.0, + 569.0, + 1408.0, + 612.0, + 827.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 603.0, + 377.0, + 603.0, + 377.0, + 647.0, + 292.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 603.0, + 745.0, + 603.0, + 745.0, + 647.0, + 545.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 603.0, + 1406.0, + 603.0, + 1406.0, + 647.0, + 1279.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 646.0, + 431.0, + 646.0, + 431.0, + 682.0, + 295.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 646.0, + 1031.0, + 646.0, + 1031.0, + 682.0, + 455.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 646.0, + 1098.0, + 646.0, + 1098.0, + 682.0, + 1050.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 646.0, + 1402.0, + 646.0, + 1402.0, + 682.0, + 1124.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 672.0, + 297.0, + 672.0, + 297.0, + 720.0, + 292.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 672.0, + 1408.0, + 672.0, + 1408.0, + 720.0, + 443.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 707.0, + 618.0, + 707.0, + 618.0, + 743.0, + 378.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 707.0, + 909.0, + 707.0, + 909.0, + 743.0, + 639.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 752.0, + 1406.0, + 752.0, + 1406.0, + 790.0, + 294.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 784.0, + 777.0, + 784.0, + 777.0, + 823.0, + 294.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 784.0, + 1408.0, + 784.0, + 1408.0, + 823.0, + 797.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 818.0, + 514.0, + 818.0, + 514.0, + 851.0, + 296.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 818.0, + 1405.0, + 818.0, + 1405.0, + 851.0, + 534.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 847.0, + 748.0, + 847.0, + 748.0, + 883.0, + 295.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 847.0, + 1406.0, + 847.0, + 1406.0, + 883.0, + 779.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 877.0, + 1406.0, + 877.0, + 1406.0, + 912.0, + 292.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 906.0, + 297.0, + 906.0, + 297.0, + 947.0, + 291.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 906.0, + 378.0, + 906.0, + 378.0, + 947.0, + 327.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 906.0, + 1406.0, + 906.0, + 1406.0, + 947.0, + 409.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 936.0, + 1033.0, + 936.0, + 1033.0, + 975.0, + 292.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 936.0, + 1406.0, + 936.0, + 1406.0, + 975.0, + 1065.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 970.0, + 1405.0, + 970.0, + 1405.0, + 1002.0, + 294.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1001.0, + 589.0, + 1001.0, + 589.0, + 1034.0, + 296.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1509.0, + 1408.0, + 1509.0, + 1408.0, + 1541.0, + 296.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1537.0, + 1405.0, + 1537.0, + 1405.0, + 1572.0, + 295.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1569.0, + 1406.0, + 1569.0, + 1406.0, + 1605.0, + 294.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1597.0, + 1405.0, + 1597.0, + 1405.0, + 1634.0, + 293.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1627.0, + 1405.0, + 1627.0, + 1405.0, + 1666.0, + 292.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1659.0, + 1081.0, + 1659.0, + 1081.0, + 1695.0, + 294.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1709.0, + 1403.0, + 1709.0, + 1403.0, + 1739.0, + 296.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1740.0, + 1405.0, + 1740.0, + 1405.0, + 1770.0, + 296.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1769.0, + 1406.0, + 1769.0, + 1406.0, + 1803.0, + 295.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1798.0, + 1405.0, + 1798.0, + 1405.0, + 1833.0, + 294.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1828.0, + 1403.0, + 1828.0, + 1403.0, + 1862.0, + 295.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1858.0, + 509.0, + 1858.0, + 509.0, + 1892.0, + 292.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 1858.0, + 1408.0, + 1858.0, + 1408.0, + 1892.0, + 630.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1889.0, + 1407.0, + 1889.0, + 1407.0, + 1925.0, + 294.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1921.0, + 1405.0, + 1921.0, + 1405.0, + 1955.0, + 295.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 229.0, + 719.0, + 229.0, + 719.0, + 266.0, + 498.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 229.0, + 846.0, + 229.0, + 846.0, + 266.0, + 795.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 229.0, + 1410.0, + 229.0, + 1410.0, + 266.0, + 991.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 258.0, + 785.0, + 258.0, + 785.0, + 304.0, + 773.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 258.0, + 1406.0, + 258.0, + 1406.0, + 304.0, + 955.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 303.0, + 679.0, + 303.0, + 679.0, + 338.0, + 295.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 303.0, + 758.0, + 303.0, + 758.0, + 338.0, + 709.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 303.0, + 1408.0, + 303.0, + 1408.0, + 338.0, + 788.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 334.0, + 352.0, + 334.0, + 352.0, + 368.0, + 296.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 334.0, + 868.0, + 334.0, + 868.0, + 368.0, + 383.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 334.0, + 1318.0, + 334.0, + 1318.0, + 368.0, + 899.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.25, + 263.5, + 873.25, + 263.5, + 873.25, + 309.5, + 290.25, + 309.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1302.0, + 724.0, + 1302.0, + 724.0, + 1339.0, + 295.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1302.0, + 1405.0, + 1302.0, + 1405.0, + 1339.0, + 780.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1332.0, + 378.0, + 1332.0, + 378.0, + 1369.0, + 292.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1332.0, + 1405.0, + 1332.0, + 1405.0, + 1369.0, + 408.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1363.0, + 1245.0, + 1363.0, + 1245.0, + 1401.0, + 295.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 1363.0, + 1403.0, + 1363.0, + 1403.0, + 1401.0, + 1276.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1394.0, + 380.0, + 1394.0, + 380.0, + 1430.0, + 295.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1394.0, + 1380.0, + 1394.0, + 1380.0, + 1430.0, + 411.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1040.0, + 504.0, + 1040.0, + 504.0, + 1081.0, + 293.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 1040.0, + 1296.0, + 1040.0, + 1296.0, + 1081.0, + 620.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1040.0, + 1404.0, + 1040.0, + 1404.0, + 1081.0, + 1400.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1077.0, + 345.0, + 1077.0, + 345.0, + 1119.0, + 294.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1077.0, + 816.0, + 1077.0, + 816.0, + 1119.0, + 432.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1077.0, + 1107.0, + 1077.0, + 1107.0, + 1119.0, + 1056.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 1077.0, + 1404.0, + 1077.0, + 1404.0, + 1119.0, + 1207.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1115.0, + 385.0, + 1115.0, + 385.0, + 1164.0, + 292.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1115.0, + 1183.0, + 1115.0, + 1183.0, + 1164.0, + 549.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 1115.0, + 1279.0, + 1115.0, + 1279.0, + 1164.0, + 1262.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1163.0, + 377.0, + 1163.0, + 377.0, + 1220.0, + 301.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1163.0, + 970.0, + 1163.0, + 970.0, + 1220.0, + 685.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 1163.0, + 1069.0, + 1163.0, + 1069.0, + 1220.0, + 1006.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1163.0, + 1392.0, + 1163.0, + 1392.0, + 1220.0, + 1378.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1206.0, + 380.0, + 1206.0, + 380.0, + 1253.0, + 304.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1206.0, + 1005.0, + 1206.0, + 1005.0, + 1253.0, + 704.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1206.0, + 1124.0, + 1206.0, + 1124.0, + 1253.0, + 1059.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1206.0, + 1408.0, + 1206.0, + 1408.0, + 1253.0, + 1399.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1240.0, + 407.0, + 1240.0, + 407.0, + 1283.0, + 331.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1240.0, + 530.0, + 1240.0, + 530.0, + 1283.0, + 433.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1240.0, + 796.0, + 1240.0, + 796.0, + 1283.0, + 560.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1240.0, + 871.0, + 1240.0, + 871.0, + 1283.0, + 860.0, + 1283.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1306, + 1404, + 1306, + 1404, + 1702, + 297, + 1702 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1046, + 1404, + 1046, + 1404, + 1291, + 297, + 1291 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 845, + 1404, + 845, + 1404, + 1030, + 298, + 1030 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1794, + 1404, + 1794, + 1404, + 1948, + 298, + 1948 + ], + "score": 0.975 + }, + { + "category_id": 3, + "poly": [ + 325, + 246, + 1374, + 246, + 1374, + 524, + 325, + 524 + ], + "score": 0.968 + }, + { + "category_id": 4, + "poly": [ + 296, + 551, + 1406, + 551, + 1406, + 827, + 296, + 827 + ], + "score": 0.956 + }, + { + "category_id": 2, + "poly": [ + 296, + 1978, + 1406, + 1978, + 1406, + 2034, + 296, + 2034 + ], + "score": 0.941 + }, + { + "category_id": 0, + "poly": [ + 294, + 1737, + 1346, + 1737, + 1346, + 1774, + 294, + 1774 + ], + "score": 0.91 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.879 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.774 + }, + { + "category_id": 13, + "poly": [ + 874, + 1397, + 950, + 1397, + 950, + 1428, + 874, + 1428 + ], + "score": 0.89, + "latex": "3 6 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 634, + 1550, + 711, + 1550, + 711, + 1580, + 634, + 1580 + ], + "score": 0.87, + "latex": "4 4 . 3 \\%" + }, + { + "category_id": 13, + "poly": [ + 518, + 1549, + 598, + 1549, + 598, + 1580, + 518, + 1580 + ], + "score": 0.84, + "latex": "( 7 0 . { \\bar { 1 } } \\%" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 243.0, + 1364.0, + 243.0, + 1364.0, + 266.0, + 1248.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 249.0, + 768.0, + 249.0, + 768.0, + 277.0, + 587.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 249.0, + 1097.0, + 249.0, + 1097.0, + 279.0, + 917.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 249.0, + 1216.0, + 249.0, + 1216.0, + 273.0, + 1189.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 250.0, + 573.0, + 250.0, + 573.0, + 279.0, + 517.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1163.0, + 252.0, + 1198.0, + 252.0, + 1198.0, + 508.0, + 1163.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 253.0, + 899.0, + 253.0, + 899.0, + 278.0, + 846.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 254.0, + 462.0, + 254.0, + 462.0, + 282.0, + 340.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 261.0, + 1361.0, + 261.0, + 1361.0, + 287.0, + 1246.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 299.0, + 675.0, + 299.0, + 675.0, + 368.0, + 602.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 288.0, + 773.0, + 288.0, + 773.0, + 379.0, + 693.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 289.0, + 1216.0, + 289.0, + 1216.0, + 315.0, + 1191.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 281.0, + 1329.0, + 281.0, + 1329.0, + 322.0, + 1246.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 337.0, + 382.0, + 337.0, + 382.0, + 365.0, + 331.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 335.0, + 512.0, + 335.0, + 512.0, + 483.0, + 481.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 338.0, + 984.0, + 338.0, + 984.0, + 357.0, + 962.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 313.0, + 1094.0, + 313.0, + 1094.0, + 365.0, + 1035.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 331.0, + 1216.0, + 331.0, + 1216.0, + 355.0, + 1189.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 357.0, + 396.0, + 357.0, + 396.0, + 384.0, + 329.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 351.0, + 470.0, + 351.0, + 470.0, + 378.0, + 403.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 357.0, + 677.0, + 357.0, + 677.0, + 446.0, + 602.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 350.0, + 778.0, + 350.0, + 778.0, + 523.0, + 686.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 351.0, + 841.0, + 351.0, + 841.0, + 464.0, + 814.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 351.0, + 1110.0, + 351.0, + 1110.0, + 464.0, + 933.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 373.0, + 1216.0, + 373.0, + 1216.0, + 397.0, + 1188.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 393.0, + 446.0, + 393.0, + 446.0, + 419.0, + 354.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 415.0, + 1215.0, + 415.0, + 1215.0, + 438.0, + 1188.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 429.0, + 684.0, + 429.0, + 684.0, + 518.0, + 595.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 433.0, + 1102.0, + 433.0, + 1102.0, + 511.0, + 1021.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 456.0, + 1215.0, + 456.0, + 1215.0, + 478.0, + 1192.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 477.0, + 452.0, + 477.0, + 452.0, + 502.0, + 402.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 488.0, + 384.0, + 488.0, + 384.0, + 513.0, + 335.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 497.0, + 467.0, + 497.0, + 467.0, + 521.0, + 402.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 498.0, + 1215.0, + 498.0, + 1215.0, + 516.0, + 1192.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.75, + 428.0, + 1002.75, + 428.0, + 1002.75, + 505.0, + 941.75, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 552.0, + 1404.0, + 552.0, + 1404.0, + 585.0, + 295.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 581.0, + 1404.0, + 581.0, + 1404.0, + 617.0, + 293.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 612.0, + 1407.0, + 612.0, + 1407.0, + 645.0, + 295.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 645.0, + 1407.0, + 645.0, + 1407.0, + 677.0, + 295.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 675.0, + 1407.0, + 675.0, + 1407.0, + 707.0, + 295.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 706.0, + 1404.0, + 706.0, + 1404.0, + 739.0, + 295.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 736.0, + 1404.0, + 736.0, + 1404.0, + 769.0, + 294.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 763.0, + 1406.0, + 763.0, + 1406.0, + 801.0, + 293.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 797.0, + 1408.0, + 797.0, + 1408.0, + 830.0, + 295.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1971.0, + 1407.0, + 1971.0, + 1407.0, + 2011.0, + 329.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2004.0, + 1355.0, + 2004.0, + 1355.0, + 2036.0, + 294.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1735.0, + 1352.0, + 1735.0, + 1352.0, + 1777.0, + 292.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1308.0, + 1404.0, + 1308.0, + 1404.0, + 1340.0, + 296.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1333.0, + 1406.0, + 1333.0, + 1406.0, + 1374.0, + 292.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1367.0, + 1405.0, + 1367.0, + 1405.0, + 1402.0, + 294.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1397.0, + 873.0, + 1397.0, + 873.0, + 1433.0, + 294.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 1397.0, + 1406.0, + 1397.0, + 1406.0, + 1433.0, + 951.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1429.0, + 1405.0, + 1429.0, + 1405.0, + 1461.0, + 295.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1460.0, + 1404.0, + 1460.0, + 1404.0, + 1492.0, + 295.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1488.0, + 1405.0, + 1488.0, + 1405.0, + 1525.0, + 294.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1520.0, + 1404.0, + 1520.0, + 1404.0, + 1556.0, + 294.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1545.0, + 517.0, + 1545.0, + 517.0, + 1589.0, + 291.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 1545.0, + 633.0, + 1545.0, + 633.0, + 1589.0, + 599.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 1545.0, + 1408.0, + 1545.0, + 1408.0, + 1589.0, + 712.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1576.0, + 1406.0, + 1576.0, + 1406.0, + 1618.0, + 294.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1610.0, + 1405.0, + 1610.0, + 1405.0, + 1646.0, + 295.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1642.0, + 1407.0, + 1642.0, + 1407.0, + 1678.0, + 295.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1672.0, + 838.0, + 1672.0, + 838.0, + 1708.0, + 295.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1045.0, + 1406.0, + 1045.0, + 1406.0, + 1082.0, + 294.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1077.0, + 1405.0, + 1077.0, + 1405.0, + 1111.0, + 295.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1109.0, + 1404.0, + 1109.0, + 1404.0, + 1139.0, + 296.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1139.0, + 1404.0, + 1139.0, + 1404.0, + 1172.0, + 294.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1167.0, + 1405.0, + 1167.0, + 1405.0, + 1201.0, + 295.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1197.0, + 1404.0, + 1197.0, + 1404.0, + 1233.0, + 294.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1227.0, + 1406.0, + 1227.0, + 1406.0, + 1263.0, + 292.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1256.0, + 1209.0, + 1256.0, + 1209.0, + 1293.0, + 291.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 847.0, + 1405.0, + 847.0, + 1405.0, + 879.0, + 294.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 878.0, + 1405.0, + 878.0, + 1405.0, + 910.0, + 294.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 908.0, + 1405.0, + 908.0, + 1405.0, + 940.0, + 296.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 937.0, + 1405.0, + 937.0, + 1405.0, + 973.0, + 294.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 968.0, + 1405.0, + 968.0, + 1405.0, + 1004.0, + 293.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1000.0, + 1081.0, + 1000.0, + 1081.0, + 1034.0, + 293.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1794.0, + 1404.0, + 1794.0, + 1404.0, + 1827.0, + 296.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1822.0, + 1404.0, + 1822.0, + 1404.0, + 1860.0, + 295.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1854.0, + 1406.0, + 1854.0, + 1406.0, + 1891.0, + 292.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1883.0, + 1405.0, + 1883.0, + 1405.0, + 1923.0, + 293.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1915.0, + 1408.0, + 1915.0, + 1408.0, + 1953.0, + 293.0, + 1953.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1514, + 1405, + 1514, + 1405, + 1730, + 298, + 1730 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1283, + 1406, + 1283, + 1406, + 1501, + 297, + 1501 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1002, + 1404, + 1002, + 1404, + 1208, + 297, + 1208 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 726, + 1404, + 726, + 1404, + 900, + 297, + 900 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1819, + 1404, + 1819, + 1404, + 2035, + 297, + 2035 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 367, + 1405, + 367, + 1405, + 523, + 297, + 523 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 536, + 1404, + 536, + 1404, + 640, + 299, + 640 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1403, + 228, + 1403, + 353, + 298, + 353 + ], + "score": 0.976 + }, + { + "category_id": 8, + "poly": [ + 524, + 650, + 1175, + 650, + 1175, + 716, + 524, + 716 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 496, + 911, + 1201, + 911, + 1201, + 976, + 496, + 976 + ], + "score": 0.946 + }, + { + "category_id": 0, + "poly": [ + 300, + 1761, + 588, + 1761, + 588, + 1798, + 300, + 1798 + ], + "score": 0.913 + }, + { + "category_id": 0, + "poly": [ + 298, + 1232, + 538, + 1232, + 538, + 1264, + 298, + 1264 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1366, + 919, + 1400, + 919, + 1400, + 950, + 1366, + 950 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1366, + 659, + 1400, + 659, + 1400, + 689, + 1366, + 689 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.796 + }, + { + "category_id": 13, + "poly": [ + 815, + 1096, + 900, + 1096, + 900, + 1129, + 815, + 1129 + ], + "score": 0.93, + "latex": "K ( \\cdot , h )" + }, + { + "category_id": 13, + "poly": [ + 978, + 567, + 1262, + 567, + 1262, + 612, + 978, + 612 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\widehat { L } ( \\phi ) = \\frac { 1 } { n } \\sum _ { j = 1 } ^ { n } \\ell ( x _ { j } , \\phi ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 1165, + 462, + 1165, + 462, + 1209, + 298, + 1209 + ], + "score": 0.93, + "latex": "f _ { \\phi } ( x _ { j } ) , h ) \\big ) ^ { - \\alpha }" + }, + { + "category_id": 14, + "poly": [ + 521, + 648, + 1176, + 648, + 1176, + 718, + 521, + 718 + ], + "score": 0.93, + "latex": "\\operatorname* { m i n } _ { \\phi } \\widehat { L } ( \\phi + \\epsilon ( \\phi ) ) , \\quad \\mathrm { w h e r e } \\quad \\epsilon ( \\phi ) = \\underset { \\| \\epsilon \\| < \\rho } { \\arg \\operatorname* { m a x } } \\epsilon ^ { \\top } \\nabla _ { \\phi } \\widehat { L } ( \\phi ) ." + }, + { + "category_id": 13, + "poly": [ + 865, + 567, + 923, + 567, + 923, + 607, + 865, + 607 + ], + "score": 0.92, + "latex": "\\widehat { L } ( \\phi )" + }, + { + "category_id": 13, + "poly": [ + 1075, + 1126, + 1404, + 1126, + 1404, + 1169, + 1075, + 1169 + ], + "score": 0.92, + "latex": "\\begin{array}{c} \\begin{array} { r } { w _ { i } = \\left( \\frac { 1 } { n } \\sum _ { j = 1 } ^ { n } K ( f _ { \\phi } ( x _ { i } ) - \\right.} \\end{array} \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 495, + 907, + 1202, + 907, + 1202, + 976, + 495, + 976 + ], + "score": 0.92, + "latex": "\\operatorname* { m i n } _ { \\phi } \\widehat { L } ( \\phi + \\epsilon _ { w } ( \\phi ) ) , \\quad \\mathrm { w h e r e } \\quad \\epsilon _ { w } ( \\phi ) = \\underset { \\| \\epsilon \\| < \\rho } { \\arg \\operatorname* { m a x } } \\epsilon ^ { \\top } \\nabla _ { \\phi } \\widehat { L } _ { w } ( \\phi ) ." + }, + { + "category_id": 13, + "poly": [ + 800, + 819, + 1137, + 819, + 1137, + 864, + 800, + 864 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\hat { L } _ { w } ( \\bar { \\phi } ) = \\frac { 1 } { n } \\sum _ { j = 1 } ^ { n } \\dot { w } _ { j } \\ell ( x _ { j } , \\phi ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 805, + 573, + 836, + 573, + 836, + 607, + 805, + 607 + ], + "score": 0.89, + "latex": "f _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 610, + 1175, + 683, + 1175, + 683, + 1202, + 610, + 1202 + ], + "score": 0.89, + "latex": "\\alpha > 0" + }, + { + "category_id": 13, + "poly": [ + 517, + 1096, + 547, + 1096, + 547, + 1130, + 517, + 1130 + ], + "score": 0.89, + "latex": "f _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 1016, + 1545, + 1056, + 1545, + 1056, + 1575, + 1016, + 1575 + ], + "score": 0.87, + "latex": "1 \\%" + }, + { + "category_id": 13, + "poly": [ + 1253, + 1010, + 1284, + 1010, + 1284, + 1038, + 1253, + 1038 + ], + "score": 0.87, + "latex": "x _ { j }" + }, + { + "category_id": 13, + "poly": [ + 1047, + 1009, + 1081, + 1009, + 1081, + 1038, + 1047, + 1038 + ], + "score": 0.86, + "latex": "w _ { j }" + }, + { + "category_id": 13, + "poly": [ + 693, + 822, + 788, + 822, + 788, + 854, + 693, + 854 + ], + "score": 0.86, + "latex": "w \\in \\mathbb { R } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1039, + 1040, + 1066, + 1040, + 1066, + 1069, + 1039, + 1069 + ], + "score": 0.86, + "latex": "y _ { j }" + }, + { + "category_id": 13, + "poly": [ + 376, + 1134, + 409, + 1134, + 409, + 1161, + 376, + 1161 + ], + "score": 0.85, + "latex": "w _ { i }" + }, + { + "category_id": 13, + "poly": [ + 945, + 871, + 978, + 871, + 978, + 898, + 945, + 898 + ], + "score": 0.83, + "latex": "\\epsilon _ { w }" + }, + { + "category_id": 13, + "poly": [ + 1259, + 861, + 1280, + 861, + 1280, + 894, + 1259, + 894 + ], + "score": 0.82, + "latex": "\\widehat { L }" + }, + { + "category_id": 13, + "poly": [ + 540, + 1174, + 561, + 1174, + 561, + 1201, + 540, + 1201 + ], + "score": 0.81, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 1329, + 1097, + 1347, + 1097, + 1347, + 1123, + 1329, + 1123 + ], + "score": 0.77, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 1060, + 1377, + 1189, + 1377, + 1189, + 1407, + 1060, + 1407 + ], + "score": 0.38, + "latex": "\\mathbf { S S L + S A M }" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1759.0, + 593.0, + 1759.0, + 593.0, + 1803.0, + 292.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1230.0, + 541.0, + 1230.0, + 541.0, + 1267.0, + 295.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 2085.0, + 859.0, + 2085.0, + 859.0, + 2116.0, + 836.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1516.0, + 1403.0, + 1516.0, + 1403.0, + 1550.0, + 296.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1544.0, + 1015.0, + 1544.0, + 1015.0, + 1581.0, + 293.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1544.0, + 1407.0, + 1544.0, + 1407.0, + 1581.0, + 1057.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1574.0, + 1405.0, + 1574.0, + 1405.0, + 1611.0, + 293.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1605.0, + 1407.0, + 1605.0, + 1407.0, + 1642.0, + 292.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1636.0, + 1405.0, + 1636.0, + 1405.0, + 1670.0, + 295.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1666.0, + 1407.0, + 1666.0, + 1407.0, + 1705.0, + 293.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1699.0, + 1279.0, + 1699.0, + 1279.0, + 1733.0, + 295.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1284.0, + 1407.0, + 1284.0, + 1407.0, + 1319.0, + 295.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1312.0, + 1406.0, + 1312.0, + 1406.0, + 1350.0, + 292.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1346.0, + 1406.0, + 1346.0, + 1406.0, + 1381.0, + 295.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1376.0, + 1059.0, + 1376.0, + 1059.0, + 1411.0, + 295.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1190.0, + 1376.0, + 1408.0, + 1376.0, + 1408.0, + 1411.0, + 1190.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1406.0, + 1408.0, + 1406.0, + 1408.0, + 1442.0, + 294.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1437.0, + 1406.0, + 1437.0, + 1406.0, + 1470.0, + 294.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1468.0, + 1053.0, + 1468.0, + 1053.0, + 1503.0, + 295.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1005.0, + 1046.0, + 1005.0, + 1046.0, + 1038.0, + 295.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 1005.0, + 1252.0, + 1005.0, + 1252.0, + 1038.0, + 1082.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 1005.0, + 1404.0, + 1005.0, + 1404.0, + 1038.0, + 1285.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1036.0, + 1038.0, + 1036.0, + 1038.0, + 1069.0, + 295.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1036.0, + 1406.0, + 1036.0, + 1406.0, + 1069.0, + 1067.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1067.0, + 1406.0, + 1067.0, + 1406.0, + 1100.0, + 295.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1094.0, + 516.0, + 1094.0, + 516.0, + 1130.0, + 294.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1094.0, + 814.0, + 1094.0, + 814.0, + 1130.0, + 548.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 1094.0, + 1328.0, + 1094.0, + 1328.0, + 1130.0, + 901.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1094.0, + 1404.0, + 1094.0, + 1404.0, + 1130.0, + 1348.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1124.0, + 375.0, + 1124.0, + 375.0, + 1177.0, + 288.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1124.0, + 1074.0, + 1124.0, + 1074.0, + 1177.0, + 410.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1170.0, + 297.0, + 1170.0, + 297.0, + 1210.0, + 293.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1170.0, + 539.0, + 1170.0, + 539.0, + 1210.0, + 463.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1170.0, + 609.0, + 1170.0, + 609.0, + 1210.0, + 562.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 1170.0, + 1238.0, + 1170.0, + 1238.0, + 1210.0, + 684.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 726.0, + 1405.0, + 726.0, + 1405.0, + 763.0, + 294.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 760.0, + 1405.0, + 760.0, + 1405.0, + 791.0, + 296.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 786.0, + 1405.0, + 786.0, + 1405.0, + 826.0, + 292.0, + 826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 818.0, + 692.0, + 818.0, + 692.0, + 870.0, + 287.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 818.0, + 799.0, + 818.0, + 799.0, + 870.0, + 789.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 818.0, + 1410.0, + 818.0, + 1410.0, + 870.0, + 1138.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 863.0, + 944.0, + 863.0, + 944.0, + 902.0, + 293.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 863.0, + 1258.0, + 863.0, + 1258.0, + 902.0, + 979.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 863.0, + 1293.0, + 863.0, + 1293.0, + 902.0, + 1281.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1817.0, + 1406.0, + 1817.0, + 1406.0, + 1855.0, + 295.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1848.0, + 1404.0, + 1848.0, + 1404.0, + 1883.0, + 295.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1878.0, + 1404.0, + 1878.0, + 1404.0, + 1917.0, + 294.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 1406.0, + 1911.0, + 1406.0, + 1944.0, + 294.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1941.0, + 1407.0, + 1941.0, + 1407.0, + 1975.0, + 295.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2005.0, + 295.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 2004.0, + 1404.0, + 2004.0, + 1404.0, + 2034.0, + 297.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 368.0, + 1407.0, + 368.0, + 1407.0, + 401.0, + 296.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 398.0, + 1407.0, + 398.0, + 1407.0, + 432.0, + 295.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 429.0, + 1406.0, + 429.0, + 1406.0, + 462.0, + 296.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 456.0, + 1406.0, + 456.0, + 1406.0, + 496.0, + 292.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 489.0, + 818.0, + 489.0, + 818.0, + 522.0, + 294.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 533.0, + 1406.0, + 533.0, + 1406.0, + 574.0, + 295.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 560.0, + 804.0, + 560.0, + 804.0, + 625.0, + 281.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 560.0, + 864.0, + 560.0, + 864.0, + 625.0, + 837.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 560.0, + 977.0, + 560.0, + 977.0, + 625.0, + 924.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 560.0, + 1415.0, + 560.0, + 1415.0, + 625.0, + 1263.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 606.0, + 1052.0, + 606.0, + 1052.0, + 645.0, + 294.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 231.0, + 1405.0, + 231.0, + 1405.0, + 264.0, + 294.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 262.0, + 1405.0, + 262.0, + 1405.0, + 295.0, + 294.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 293.0, + 1404.0, + 293.0, + 1404.0, + 326.0, + 296.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 321.0, + 1383.0, + 321.0, + 1383.0, + 354.0, + 296.0, + 354.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 425, + 224, + 1266, + 224, + 1266, + 791, + 425, + 791 + ], + "score": 0.983, + "html": "
(a) CIFAR, IDr = 0.01,stepr = 0.01,exp
MethodAcc.(%)Gap Freq.Gap RareAcc.(%)
SimSiam84.3± 0.20.0350.08181.4±0.3
SimSiam+SAM84.7 ± 0.30.0440.07582.1 ± 0.4
SimSiam,balanced85.8±0.20.0380.03782.0± 0.4
SimSiam+rwSAM85.6 ± 0.40.0370.06682.7 ± 0.5
(b) ImageNet,OODTarget dataset
MethodCUBCarsAircraftsPetsAvg.
MoCo v269.9±0.788.4±0.482.9 ± 0.680.1 ± 0.680.3
MoCo v2+SAM69.9 ± 0.588.8 ± 0.583.4± 0.481.5 ± 0.880.9
MoCo v2,balanced69.8 ± 0.588.6± 0.482.7 ± 0.580.0± 0.480.2
MoCo v2+rwSAM70.3 ± 0.788.7 ± 0.384.9 ± 0.681.7 ± 0.481.4
SimSiam70.0±0.387.0± 0.681.5± 0.783.8 ± 0.580.6
SimSiam, balanced70.5 ± 0.887.9±0.781.8 ± 0.782.7 ±0.480.7
SimSiam+rwSAM70.7 ± 0.888.4 ± 0.682.6 ± 0.684.0 ± 0.481.4
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1317, + 1405, + 1317, + 1405, + 1682, + 298, + 1682 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1819, + 1404, + 1819, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1008, + 1405, + 1008, + 1405, + 1133, + 298, + 1133 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1148, + 1404, + 1148, + 1404, + 1301, + 298, + 1301 + ], + "score": 0.976 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 105, + 299, + 105 + ], + "score": 0.894 + }, + { + "category_id": 0, + "poly": [ + 299, + 1745, + 543, + 1745, + 543, + 1780, + 299, + 1780 + ], + "score": 0.893 + }, + { + "category_id": 6, + "poly": [ + 296, + 810, + 1405, + 810, + 1405, + 965, + 296, + 965 + ], + "score": 0.859 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.793 + }, + { + "category_id": 15, + "poly": [ + 294.0, + 70.0, + 817.0, + 70.0, + 817.0, + 110.0, + 294.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1741.0, + 548.0, + 1741.0, + 548.0, + 1789.0, + 292.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 812.0, + 1403.0, + 812.0, + 1403.0, + 845.0, + 295.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 838.0, + 1405.0, + 838.0, + 1405.0, + 878.0, + 293.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 869.0, + 1406.0, + 869.0, + 1406.0, + 909.0, + 294.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 901.0, + 1406.0, + 901.0, + 1406.0, + 939.0, + 293.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 935.0, + 1361.0, + 935.0, + 1361.0, + 969.0, + 295.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1317.0, + 1408.0, + 1317.0, + 1408.0, + 1351.0, + 295.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1346.0, + 1405.0, + 1346.0, + 1405.0, + 1380.0, + 295.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1377.0, + 1406.0, + 1377.0, + 1406.0, + 1411.0, + 295.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1407.0, + 1403.0, + 1407.0, + 1403.0, + 1440.0, + 296.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1438.0, + 1407.0, + 1438.0, + 1407.0, + 1472.0, + 295.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1469.0, + 1405.0, + 1469.0, + 1405.0, + 1503.0, + 295.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1497.0, + 1405.0, + 1497.0, + 1405.0, + 1533.0, + 293.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1530.0, + 1405.0, + 1530.0, + 1405.0, + 1564.0, + 296.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1561.0, + 1405.0, + 1561.0, + 1405.0, + 1592.0, + 296.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1590.0, + 1405.0, + 1590.0, + 1405.0, + 1623.0, + 293.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1621.0, + 1403.0, + 1621.0, + 1403.0, + 1653.0, + 293.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1650.0, + 443.0, + 1650.0, + 443.0, + 1685.0, + 293.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1855.0, + 296.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1851.0, + 1406.0, + 1851.0, + 1406.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1408.0, + 1882.0, + 1408.0, + 1917.0, + 294.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 1405.0, + 1912.0, + 1405.0, + 1946.0, + 293.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1974.0, + 1406.0, + 1974.0, + 1406.0, + 2005.0, + 296.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2004.0, + 1147.0, + 2004.0, + 1147.0, + 2038.0, + 296.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1008.0, + 1408.0, + 1008.0, + 1408.0, + 1041.0, + 295.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1039.0, + 1407.0, + 1039.0, + 1407.0, + 1075.0, + 295.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1070.0, + 1406.0, + 1070.0, + 1406.0, + 1103.0, + 295.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1100.0, + 1006.0, + 1100.0, + 1006.0, + 1136.0, + 295.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1148.0, + 1406.0, + 1148.0, + 1406.0, + 1181.0, + 297.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1174.0, + 1404.0, + 1174.0, + 1404.0, + 1213.0, + 295.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1206.0, + 1405.0, + 1206.0, + 1405.0, + 1244.0, + 293.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1238.0, + 1406.0, + 1238.0, + 1406.0, + 1275.0, + 296.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1268.0, + 1178.0, + 1268.0, + 1178.0, + 1306.0, + 293.0, + 1306.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 732, + 1405, + 732, + 1405, + 886, + 299, + 886 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 901, + 1404, + 901, + 1404, + 1026, + 299, + 1026 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 499, + 1406, + 499, + 1406, + 622, + 299, + 622 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1041, + 1404, + 1041, + 1404, + 1163, + 298, + 1163 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 299, + 293, + 1407, + 293, + 1407, + 387, + 299, + 387 + ], + "score": 0.968 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.887 + }, + { + "category_id": 0, + "poly": [ + 299, + 1209, + 488, + 1209, + 488, + 1242, + 299, + 1242 + ], + "score": 0.853 + }, + { + "category_id": 0, + "poly": [ + 300, + 665, + 587, + 665, + 587, + 701, + 300, + 701 + ], + "score": 0.845 + }, + { + "category_id": 0, + "poly": [ + 301, + 431, + 745, + 431, + 745, + 465, + 301, + 465 + ], + "score": 0.838 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.816 + }, + { + "category_id": 0, + "poly": [ + 302, + 228, + 625, + 228, + 625, + 261, + 302, + 261 + ], + "score": 0.785 + }, + { + "category_id": 1, + "poly": [ + 299, + 1258, + 1406, + 1258, + 1406, + 1351, + 299, + 1351 + ], + "score": 0.75 + }, + { + "category_id": 1, + "poly": [ + 294, + 1478, + 1397, + 1478, + 1397, + 1542, + 294, + 1542 + ], + "score": 0.658 + }, + { + "category_id": 1, + "poly": [ + 296, + 1971, + 1398, + 1971, + 1398, + 2035, + 296, + 2035 + ], + "score": 0.656 + }, + { + "category_id": 1, + "poly": [ + 296, + 1558, + 1399, + 1558, + 1399, + 1625, + 296, + 1625 + ], + "score": 0.629 + }, + { + "category_id": 1, + "poly": [ + 298, + 1638, + 1402, + 1638, + 1402, + 1734, + 298, + 1734 + ], + "score": 0.579 + }, + { + "category_id": 1, + "poly": [ + 298, + 1368, + 1406, + 1368, + 1406, + 1462, + 298, + 1462 + ], + "score": 0.573 + }, + { + "category_id": 1, + "poly": [ + 297, + 1749, + 1402, + 1749, + 1402, + 1844, + 297, + 1844 + ], + "score": 0.57 + }, + { + "category_id": 1, + "poly": [ + 297, + 1860, + 1401, + 1860, + 1401, + 1954, + 297, + 1954 + ], + "score": 0.545 + }, + { + "category_id": 1, + "poly": [ + 302, + 228, + 625, + 228, + 625, + 261, + 302, + 261 + ], + "score": 0.096 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1209.0, + 490.0, + 1209.0, + 490.0, + 1246.0, + 297.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 664.0, + 591.0, + 664.0, + 591.0, + 704.0, + 295.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 429.0, + 749.0, + 429.0, + 749.0, + 469.0, + 296.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 868.0, + 2084.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 228.0, + 629.0, + 228.0, + 629.0, + 267.0, + 298.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 732.0, + 1410.0, + 732.0, + 1410.0, + 769.0, + 295.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 761.0, + 1409.0, + 761.0, + 1409.0, + 801.0, + 293.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 796.0, + 1405.0, + 796.0, + 1405.0, + 829.0, + 295.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 824.0, + 1405.0, + 824.0, + 1405.0, + 861.0, + 292.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 855.0, + 803.0, + 855.0, + 803.0, + 891.0, + 293.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 898.0, + 1408.0, + 898.0, + 1408.0, + 939.0, + 293.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 931.0, + 1409.0, + 931.0, + 1409.0, + 971.0, + 293.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 965.0, + 1406.0, + 965.0, + 1406.0, + 997.0, + 294.0, + 997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 995.0, + 1320.0, + 995.0, + 1320.0, + 1027.0, + 294.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 497.0, + 1409.0, + 497.0, + 1409.0, + 534.0, + 293.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 528.0, + 1407.0, + 528.0, + 1407.0, + 563.0, + 294.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 559.0, + 1404.0, + 559.0, + 1404.0, + 594.0, + 294.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 592.0, + 1161.0, + 592.0, + 1161.0, + 624.0, + 296.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1038.0, + 1403.0, + 1038.0, + 1403.0, + 1075.0, + 295.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1071.0, + 1405.0, + 1071.0, + 1405.0, + 1105.0, + 292.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1102.0, + 1406.0, + 1102.0, + 1406.0, + 1137.0, + 292.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1132.0, + 763.0, + 1132.0, + 763.0, + 1167.0, + 293.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 291.0, + 1408.0, + 291.0, + 1408.0, + 330.0, + 293.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 325.0, + 1407.0, + 325.0, + 1407.0, + 359.0, + 296.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 356.0, + 994.0, + 356.0, + 994.0, + 390.0, + 296.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1257.0, + 1407.0, + 1257.0, + 1407.0, + 1295.0, + 294.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1285.0, + 1410.0, + 1285.0, + 1410.0, + 1329.0, + 320.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1320.0, + 503.0, + 1320.0, + 503.0, + 1352.0, + 322.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1478.0, + 1403.0, + 1478.0, + 1403.0, + 1515.0, + 294.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1510.0, + 1235.0, + 1510.0, + 1235.0, + 1543.0, + 322.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1971.0, + 1403.0, + 1971.0, + 1403.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2003.0, + 1400.0, + 2003.0, + 1400.0, + 2036.0, + 322.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1558.0, + 1403.0, + 1558.0, + 1403.0, + 1594.0, + 295.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1590.0, + 1176.0, + 1590.0, + 1176.0, + 1625.0, + 321.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1636.0, + 1406.0, + 1636.0, + 1406.0, + 1676.0, + 293.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1666.0, + 1406.0, + 1666.0, + 1406.0, + 1709.0, + 321.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1701.0, + 1136.0, + 1701.0, + 1136.0, + 1735.0, + 322.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1369.0, + 1403.0, + 1369.0, + 1403.0, + 1403.0, + 296.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1399.0, + 1406.0, + 1399.0, + 1406.0, + 1434.0, + 324.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1428.0, + 782.0, + 1428.0, + 782.0, + 1466.0, + 324.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1749.0, + 1406.0, + 1749.0, + 1406.0, + 1786.0, + 295.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1782.0, + 1404.0, + 1782.0, + 1404.0, + 1816.0, + 324.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1812.0, + 696.0, + 1812.0, + 696.0, + 1845.0, + 322.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1861.0, + 1403.0, + 1861.0, + 1403.0, + 1895.0, + 296.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1891.0, + 1405.0, + 1891.0, + 1405.0, + 1928.0, + 324.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1922.0, + 781.0, + 1922.0, + 781.0, + 1954.0, + 323.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 228.0, + 629.0, + 228.0, + 629.0, + 267.0, + 298.0, + 267.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.887 + }, + { + "category_id": 1, + "poly": [ + 297, + 371, + 1404, + 371, + 1404, + 464, + 297, + 464 + ], + "score": 0.875 + }, + { + "category_id": 1, + "poly": [ + 296, + 482, + 1405, + 482, + 1405, + 575, + 296, + 575 + ], + "score": 0.866 + }, + { + "category_id": 1, + "poly": [ + 299, + 228, + 1405, + 228, + 1405, + 353, + 299, + 353 + ], + "score": 0.855 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 861, + 2088, + 861, + 2113, + 835, + 2113 + ], + "score": 0.822 + }, + { + "category_id": 1, + "poly": [ + 298, + 593, + 1400, + 593, + 1400, + 658, + 298, + 658 + ], + "score": 0.804 + }, + { + "category_id": 1, + "poly": [ + 296, + 1536, + 1402, + 1536, + 1402, + 1601, + 296, + 1601 + ], + "score": 0.776 + }, + { + "category_id": 1, + "poly": [ + 295, + 1617, + 1402, + 1617, + 1402, + 1681, + 295, + 1681 + ], + "score": 0.768 + }, + { + "category_id": 1, + "poly": [ + 299, + 1008, + 1399, + 1008, + 1399, + 1074, + 299, + 1074 + ], + "score": 0.753 + }, + { + "category_id": 1, + "poly": [ + 299, + 898, + 1405, + 898, + 1405, + 991, + 299, + 991 + ], + "score": 0.752 + }, + { + "category_id": 1, + "poly": [ + 298, + 1089, + 1406, + 1089, + 1406, + 1183, + 298, + 1183 + ], + "score": 0.744 + }, + { + "category_id": 1, + "poly": [ + 297, + 1200, + 1401, + 1200, + 1401, + 1266, + 297, + 1266 + ], + "score": 0.738 + }, + { + "category_id": 1, + "poly": [ + 295, + 1698, + 1402, + 1698, + 1402, + 1763, + 295, + 1763 + ], + "score": 0.729 + }, + { + "category_id": 1, + "poly": [ + 300, + 674, + 1402, + 674, + 1402, + 771, + 300, + 771 + ], + "score": 0.727 + }, + { + "category_id": 1, + "poly": [ + 297, + 1393, + 1405, + 1393, + 1405, + 1518, + 297, + 1518 + ], + "score": 0.718 + }, + { + "category_id": 1, + "poly": [ + 296, + 1779, + 1402, + 1779, + 1402, + 1843, + 296, + 1843 + ], + "score": 0.65 + }, + { + "category_id": 1, + "poly": [ + 298, + 785, + 1405, + 785, + 1405, + 881, + 298, + 881 + ], + "score": 0.633 + }, + { + "category_id": 1, + "poly": [ + 296, + 1860, + 1399, + 1860, + 1399, + 1924, + 296, + 1924 + ], + "score": 0.625 + }, + { + "category_id": 1, + "poly": [ + 299, + 1942, + 1403, + 1942, + 1403, + 2035, + 299, + 2035 + ], + "score": 0.514 + }, + { + "category_id": 1, + "poly": [ + 300, + 1282, + 1402, + 1282, + 1402, + 1378, + 300, + 1378 + ], + "score": 0.392 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 369.0, + 1409.0, + 369.0, + 1409.0, + 409.0, + 294.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 403.0, + 1407.0, + 403.0, + 1407.0, + 437.0, + 322.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 433.0, + 454.0, + 433.0, + 454.0, + 465.0, + 321.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 483.0, + 1405.0, + 483.0, + 1405.0, + 516.0, + 296.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 514.0, + 1406.0, + 514.0, + 1406.0, + 548.0, + 322.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 542.0, + 400.0, + 542.0, + 400.0, + 578.0, + 320.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 230.0, + 1406.0, + 230.0, + 1406.0, + 266.0, + 295.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 259.0, + 1409.0, + 259.0, + 1409.0, + 299.0, + 321.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 291.0, + 1406.0, + 291.0, + 1406.0, + 326.0, + 321.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 322.0, + 432.0, + 322.0, + 432.0, + 353.0, + 322.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 596.0, + 1402.0, + 596.0, + 1402.0, + 628.0, + 297.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 626.0, + 1317.0, + 626.0, + 1317.0, + 659.0, + 323.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1536.0, + 1406.0, + 1536.0, + 1406.0, + 1572.0, + 295.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1567.0, + 1404.0, + 1567.0, + 1404.0, + 1602.0, + 321.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1617.0, + 1404.0, + 1617.0, + 1404.0, + 1653.0, + 294.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1649.0, + 1097.0, + 1649.0, + 1097.0, + 1681.0, + 323.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1009.0, + 1403.0, + 1009.0, + 1403.0, + 1044.0, + 295.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1040.0, + 1228.0, + 1040.0, + 1228.0, + 1075.0, + 321.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 895.0, + 1405.0, + 895.0, + 1405.0, + 937.0, + 294.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 926.0, + 1410.0, + 926.0, + 1410.0, + 969.0, + 319.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 960.0, + 530.0, + 960.0, + 530.0, + 992.0, + 323.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1091.0, + 1404.0, + 1091.0, + 1404.0, + 1125.0, + 295.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1120.0, + 1408.0, + 1120.0, + 1408.0, + 1157.0, + 321.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1151.0, + 396.0, + 1151.0, + 396.0, + 1185.0, + 321.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1198.0, + 1405.0, + 1198.0, + 1405.0, + 1241.0, + 295.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1234.0, + 1083.0, + 1234.0, + 1083.0, + 1267.0, + 321.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1698.0, + 1404.0, + 1698.0, + 1404.0, + 1734.0, + 294.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1728.0, + 1268.0, + 1728.0, + 1268.0, + 1764.0, + 322.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 676.0, + 1404.0, + 676.0, + 1404.0, + 710.0, + 296.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 704.0, + 1407.0, + 704.0, + 1407.0, + 743.0, + 321.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 736.0, + 1150.0, + 736.0, + 1150.0, + 774.0, + 324.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1392.0, + 1406.0, + 1392.0, + 1406.0, + 1430.0, + 294.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1424.0, + 1403.0, + 1424.0, + 1403.0, + 1459.0, + 321.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1454.0, + 1406.0, + 1454.0, + 1406.0, + 1493.0, + 320.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1487.0, + 811.0, + 1487.0, + 811.0, + 1519.0, + 321.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1775.0, + 1408.0, + 1775.0, + 1408.0, + 1819.0, + 293.0, + 1819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1811.0, + 569.0, + 1811.0, + 569.0, + 1843.0, + 323.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 785.0, + 1405.0, + 785.0, + 1405.0, + 823.0, + 296.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 818.0, + 1405.0, + 818.0, + 1405.0, + 853.0, + 323.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 850.0, + 800.0, + 850.0, + 800.0, + 882.0, + 321.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1856.0, + 1404.0, + 1856.0, + 1404.0, + 1898.0, + 292.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1891.0, + 850.0, + 1891.0, + 850.0, + 1925.0, + 322.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1939.0, + 1404.0, + 1939.0, + 1404.0, + 1979.0, + 293.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2008.0, + 322.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2002.0, + 1251.0, + 2002.0, + 1251.0, + 2039.0, + 322.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1281.0, + 1406.0, + 1281.0, + 1406.0, + 1319.0, + 295.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1313.0, + 1404.0, + 1313.0, + 1404.0, + 1350.0, + 323.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1345.0, + 1005.0, + 1345.0, + 1005.0, + 1379.0, + 323.0, + 1379.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.881 + }, + { + "category_id": 1, + "poly": [ + 294, + 1534, + 1404, + 1534, + 1404, + 1600, + 294, + 1600 + ], + "score": 0.859 + }, + { + "category_id": 1, + "poly": [ + 297, + 958, + 1401, + 958, + 1401, + 1023, + 297, + 1023 + ], + "score": 0.849 + }, + { + "category_id": 1, + "poly": [ + 295, + 1394, + 1400, + 1394, + 1400, + 1458, + 295, + 1458 + ], + "score": 0.837 + }, + { + "category_id": 1, + "poly": [ + 299, + 229, + 1403, + 229, + 1403, + 323, + 299, + 323 + ], + "score": 0.835 + }, + { + "category_id": 1, + "poly": [ + 295, + 1276, + 1406, + 1276, + 1406, + 1372, + 295, + 1372 + ], + "score": 0.834 + }, + { + "category_id": 1, + "poly": [ + 292, + 1479, + 1402, + 1479, + 1402, + 1515, + 292, + 1515 + ], + "score": 0.831 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2113, + 836, + 2113 + ], + "score": 0.827 + }, + { + "category_id": 1, + "poly": [ + 300, + 1160, + 1405, + 1160, + 1405, + 1256, + 300, + 1256 + ], + "score": 0.825 + }, + { + "category_id": 1, + "poly": [ + 299, + 1941, + 1404, + 1941, + 1404, + 2035, + 299, + 2035 + ], + "score": 0.82 + }, + { + "category_id": 1, + "poly": [ + 290, + 1737, + 1403, + 1737, + 1403, + 1803, + 290, + 1803 + ], + "score": 0.812 + }, + { + "category_id": 1, + "poly": [ + 297, + 810, + 1405, + 810, + 1405, + 936, + 297, + 936 + ], + "score": 0.811 + }, + { + "category_id": 1, + "poly": [ + 299, + 1823, + 1404, + 1823, + 1404, + 1917, + 299, + 1917 + ], + "score": 0.799 + }, + { + "category_id": 1, + "poly": [ + 297, + 1621, + 1407, + 1621, + 1407, + 1716, + 297, + 1716 + ], + "score": 0.779 + }, + { + "category_id": 1, + "poly": [ + 297, + 1043, + 1406, + 1043, + 1406, + 1138, + 297, + 1138 + ], + "score": 0.778 + }, + { + "category_id": 1, + "poly": [ + 299, + 462, + 1406, + 462, + 1406, + 557, + 299, + 557 + ], + "score": 0.748 + }, + { + "category_id": 1, + "poly": [ + 298, + 578, + 1406, + 578, + 1406, + 673, + 298, + 673 + ], + "score": 0.742 + }, + { + "category_id": 1, + "poly": [ + 299, + 694, + 1403, + 694, + 1403, + 790, + 299, + 790 + ], + "score": 0.729 + }, + { + "category_id": 1, + "poly": [ + 299, + 345, + 1405, + 345, + 1405, + 440, + 299, + 440 + ], + "score": 0.706 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1535.0, + 1405.0, + 1535.0, + 1405.0, + 1571.0, + 296.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1566.0, + 990.0, + 1566.0, + 990.0, + 1601.0, + 321.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 954.0, + 1403.0, + 954.0, + 1403.0, + 996.0, + 295.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 991.0, + 706.0, + 991.0, + 706.0, + 1024.0, + 320.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1391.0, + 1405.0, + 1391.0, + 1405.0, + 1432.0, + 294.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1425.0, + 836.0, + 1425.0, + 836.0, + 1459.0, + 322.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 224.0, + 1405.0, + 224.0, + 1405.0, + 270.0, + 292.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 260.0, + 1408.0, + 260.0, + 1408.0, + 297.0, + 321.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 291.0, + 548.0, + 291.0, + 548.0, + 324.0, + 319.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1276.0, + 1406.0, + 1276.0, + 1406.0, + 1314.0, + 294.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1307.0, + 1408.0, + 1307.0, + 1408.0, + 1347.0, + 321.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1340.0, + 547.0, + 1340.0, + 547.0, + 1372.0, + 320.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1479.0, + 1407.0, + 1479.0, + 1407.0, + 1518.0, + 296.0, + 1518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1155.0, + 1406.0, + 1155.0, + 1406.0, + 1202.0, + 293.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1192.0, + 1406.0, + 1192.0, + 1406.0, + 1228.0, + 320.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1224.0, + 1096.0, + 1224.0, + 1096.0, + 1258.0, + 324.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1938.0, + 1405.0, + 1938.0, + 1405.0, + 1980.0, + 293.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1973.0, + 1404.0, + 1973.0, + 1404.0, + 2007.0, + 323.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 2003.0, + 1344.0, + 2003.0, + 1344.0, + 2037.0, + 323.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1739.0, + 1406.0, + 1739.0, + 1406.0, + 1775.0, + 294.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1767.0, + 1174.0, + 1767.0, + 1174.0, + 1804.0, + 323.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 805.0, + 1410.0, + 805.0, + 1410.0, + 852.0, + 291.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 842.0, + 1403.0, + 842.0, + 1403.0, + 878.0, + 321.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 874.0, + 1405.0, + 874.0, + 1405.0, + 909.0, + 321.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 905.0, + 692.0, + 905.0, + 692.0, + 938.0, + 321.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1823.0, + 1405.0, + 1823.0, + 1405.0, + 1860.0, + 295.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1852.0, + 1408.0, + 1852.0, + 1408.0, + 1891.0, + 322.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1880.0, + 400.0, + 1880.0, + 400.0, + 1921.0, + 319.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1621.0, + 1405.0, + 1621.0, + 1405.0, + 1659.0, + 295.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1654.0, + 1409.0, + 1654.0, + 1409.0, + 1688.0, + 323.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1684.0, + 1056.0, + 1684.0, + 1056.0, + 1718.0, + 323.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1043.0, + 1403.0, + 1043.0, + 1403.0, + 1081.0, + 295.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1075.0, + 1404.0, + 1075.0, + 1404.0, + 1110.0, + 323.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1106.0, + 859.0, + 1106.0, + 859.0, + 1140.0, + 325.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 460.0, + 1408.0, + 460.0, + 1408.0, + 501.0, + 296.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 492.0, + 1407.0, + 492.0, + 1407.0, + 530.0, + 322.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 523.0, + 1220.0, + 523.0, + 1220.0, + 561.0, + 322.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 574.0, + 1406.0, + 574.0, + 1406.0, + 619.0, + 292.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 610.0, + 1406.0, + 610.0, + 1406.0, + 647.0, + 321.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 639.0, + 749.0, + 639.0, + 749.0, + 677.0, + 322.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 694.0, + 1407.0, + 694.0, + 1407.0, + 732.0, + 295.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 727.0, + 1404.0, + 727.0, + 1404.0, + 762.0, + 323.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 757.0, + 757.0, + 757.0, + 757.0, + 791.0, + 324.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 347.0, + 1405.0, + 347.0, + 1405.0, + 382.0, + 297.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 378.0, + 1405.0, + 378.0, + 1405.0, + 412.0, + 324.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 405.0, + 1108.0, + 405.0, + 1108.0, + 446.0, + 323.0, + 446.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 816, + 75, + 816, + 105, + 300, + 105 + ], + "score": 0.89 + }, + { + "category_id": 1, + "poly": [ + 297, + 1226, + 1406, + 1226, + 1406, + 1322, + 297, + 1322 + ], + "score": 0.888 + }, + { + "category_id": 1, + "poly": [ + 298, + 1562, + 1403, + 1562, + 1403, + 1628, + 298, + 1628 + ], + "score": 0.877 + }, + { + "category_id": 1, + "poly": [ + 298, + 1339, + 1406, + 1339, + 1406, + 1434, + 298, + 1434 + ], + "score": 0.871 + }, + { + "category_id": 1, + "poly": [ + 297, + 1084, + 1407, + 1084, + 1407, + 1208, + 297, + 1208 + ], + "score": 0.87 + }, + { + "category_id": 1, + "poly": [ + 294, + 727, + 1401, + 727, + 1401, + 793, + 294, + 793 + ], + "score": 0.845 + }, + { + "category_id": 1, + "poly": [ + 300, + 810, + 1351, + 810, + 1351, + 845, + 300, + 845 + ], + "score": 0.844 + }, + { + "category_id": 1, + "poly": [ + 296, + 1644, + 1401, + 1644, + 1401, + 1709, + 296, + 1709 + ], + "score": 0.842 + }, + { + "category_id": 1, + "poly": [ + 292, + 647, + 1402, + 647, + 1402, + 712, + 292, + 712 + ], + "score": 0.841 + }, + { + "category_id": 1, + "poly": [ + 293, + 861, + 1401, + 861, + 1401, + 926, + 293, + 926 + ], + "score": 0.838 + }, + { + "category_id": 1, + "poly": [ + 296, + 1726, + 1401, + 1726, + 1401, + 1791, + 296, + 1791 + ], + "score": 0.834 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2113, + 835, + 2113 + ], + "score": 0.833 + }, + { + "category_id": 1, + "poly": [ + 298, + 941, + 1405, + 941, + 1405, + 1067, + 298, + 1067 + ], + "score": 0.83 + }, + { + "category_id": 1, + "poly": [ + 298, + 1451, + 1405, + 1451, + 1405, + 1546, + 298, + 1546 + ], + "score": 0.827 + }, + { + "category_id": 1, + "poly": [ + 297, + 534, + 1402, + 534, + 1402, + 628, + 297, + 628 + ], + "score": 0.822 + }, + { + "category_id": 1, + "poly": [ + 301, + 229, + 1400, + 229, + 1400, + 324, + 301, + 324 + ], + "score": 0.783 + }, + { + "category_id": 1, + "poly": [ + 294, + 452, + 1403, + 452, + 1403, + 517, + 294, + 517 + ], + "score": 0.774 + }, + { + "category_id": 1, + "poly": [ + 300, + 340, + 1399, + 340, + 1399, + 435, + 300, + 435 + ], + "score": 0.771 + }, + { + "category_id": 1, + "poly": [ + 297, + 1808, + 1402, + 1808, + 1402, + 1873, + 297, + 1873 + ], + "score": 0.747 + }, + { + "category_id": 1, + "poly": [ + 299, + 1971, + 1399, + 1971, + 1399, + 2034, + 299, + 2034 + ], + "score": 0.728 + }, + { + "category_id": 1, + "poly": [ + 300, + 1889, + 1401, + 1889, + 1401, + 1955, + 300, + 1955 + ], + "score": 0.687 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1226.0, + 1407.0, + 1226.0, + 1407.0, + 1264.0, + 294.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1261.0, + 1404.0, + 1261.0, + 1404.0, + 1295.0, + 323.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1289.0, + 1027.0, + 1289.0, + 1027.0, + 1323.0, + 323.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1563.0, + 1405.0, + 1563.0, + 1405.0, + 1599.0, + 294.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1594.0, + 1054.0, + 1594.0, + 1054.0, + 1630.0, + 323.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1339.0, + 1407.0, + 1339.0, + 1407.0, + 1377.0, + 295.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1373.0, + 1403.0, + 1373.0, + 1403.0, + 1407.0, + 322.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1401.0, + 809.0, + 1401.0, + 809.0, + 1435.0, + 322.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1086.0, + 1407.0, + 1086.0, + 1407.0, + 1119.0, + 295.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1117.0, + 1409.0, + 1117.0, + 1409.0, + 1150.0, + 324.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1147.0, + 1409.0, + 1147.0, + 1409.0, + 1183.0, + 320.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1175.0, + 400.0, + 1175.0, + 400.0, + 1210.0, + 320.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 728.0, + 1404.0, + 728.0, + 1404.0, + 764.0, + 296.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 759.0, + 1264.0, + 759.0, + 1264.0, + 795.0, + 322.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 809.0, + 1353.0, + 809.0, + 1353.0, + 848.0, + 296.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1643.0, + 1403.0, + 1643.0, + 1403.0, + 1682.0, + 296.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1677.0, + 1064.0, + 1677.0, + 1064.0, + 1710.0, + 323.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 645.0, + 1404.0, + 645.0, + 1404.0, + 685.0, + 294.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 678.0, + 1225.0, + 678.0, + 1225.0, + 713.0, + 321.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 861.0, + 1406.0, + 861.0, + 1406.0, + 897.0, + 295.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 890.0, + 1387.0, + 890.0, + 1387.0, + 928.0, + 320.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1722.0, + 1405.0, + 1722.0, + 1405.0, + 1765.0, + 295.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1757.0, + 1049.0, + 1757.0, + 1049.0, + 1793.0, + 320.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 939.0, + 1409.0, + 939.0, + 1409.0, + 982.0, + 293.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 975.0, + 1406.0, + 975.0, + 1406.0, + 1008.0, + 321.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1004.0, + 1408.0, + 1004.0, + 1408.0, + 1040.0, + 320.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1036.0, + 504.0, + 1036.0, + 504.0, + 1066.0, + 323.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1451.0, + 1406.0, + 1451.0, + 1406.0, + 1489.0, + 295.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1480.0, + 1405.0, + 1480.0, + 1405.0, + 1522.0, + 321.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1513.0, + 1077.0, + 1513.0, + 1077.0, + 1547.0, + 322.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 536.0, + 1404.0, + 536.0, + 1404.0, + 570.0, + 296.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 564.0, + 1406.0, + 564.0, + 1406.0, + 604.0, + 320.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 596.0, + 585.0, + 596.0, + 585.0, + 629.0, + 321.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 227.0, + 1404.0, + 227.0, + 1404.0, + 268.0, + 295.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 262.0, + 1404.0, + 262.0, + 1404.0, + 296.0, + 323.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 289.0, + 1024.0, + 289.0, + 1024.0, + 328.0, + 323.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 455.0, + 1404.0, + 455.0, + 1404.0, + 487.0, + 297.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 483.0, + 1113.0, + 483.0, + 1113.0, + 519.0, + 323.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 339.0, + 1404.0, + 339.0, + 1404.0, + 379.0, + 295.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 369.0, + 1404.0, + 369.0, + 1404.0, + 412.0, + 320.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 403.0, + 583.0, + 403.0, + 583.0, + 439.0, + 322.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1807.0, + 1407.0, + 1807.0, + 1407.0, + 1844.0, + 296.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1837.0, + 1326.0, + 1837.0, + 1326.0, + 1875.0, + 321.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1967.0, + 1405.0, + 1967.0, + 1405.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 2004.0, + 664.0, + 2004.0, + 664.0, + 2032.0, + 325.0, + 2032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1884.0, + 1409.0, + 1884.0, + 1409.0, + 1930.0, + 293.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1921.0, + 845.0, + 1921.0, + 845.0, + 1954.0, + 321.0, + 1954.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 816, + 75, + 816, + 105, + 300, + 105 + ], + "score": 0.892 + }, + { + "category_id": 1, + "poly": [ + 298, + 765, + 1403, + 765, + 1403, + 830, + 298, + 830 + ], + "score": 0.844 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2112, + 835, + 2112 + ], + "score": 0.836 + }, + { + "category_id": 1, + "poly": [ + 296, + 848, + 1402, + 848, + 1402, + 913, + 296, + 913 + ], + "score": 0.833 + }, + { + "category_id": 1, + "poly": [ + 298, + 651, + 1407, + 651, + 1407, + 746, + 298, + 746 + ], + "score": 0.83 + }, + { + "category_id": 1, + "poly": [ + 298, + 931, + 1406, + 931, + 1406, + 1026, + 298, + 1026 + ], + "score": 0.82 + }, + { + "category_id": 1, + "poly": [ + 300, + 425, + 1402, + 425, + 1402, + 519, + 300, + 519 + ], + "score": 0.814 + }, + { + "category_id": 1, + "poly": [ + 297, + 538, + 1409, + 538, + 1409, + 631, + 297, + 631 + ], + "score": 0.778 + }, + { + "category_id": 1, + "poly": [ + 294, + 228, + 1405, + 228, + 1405, + 294, + 294, + 294 + ], + "score": 0.761 + }, + { + "category_id": 1, + "poly": [ + 298, + 1159, + 1401, + 1159, + 1401, + 1255, + 298, + 1255 + ], + "score": 0.752 + }, + { + "category_id": 1, + "poly": [ + 307, + 311, + 1400, + 311, + 1400, + 407, + 307, + 407 + ], + "score": 0.739 + }, + { + "category_id": 1, + "poly": [ + 298, + 1043, + 1407, + 1043, + 1407, + 1141, + 298, + 1141 + ], + "score": 0.735 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 767.0, + 1405.0, + 767.0, + 1405.0, + 803.0, + 296.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 795.0, + 1408.0, + 795.0, + 1408.0, + 834.0, + 323.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 844.0, + 1406.0, + 844.0, + 1406.0, + 890.0, + 293.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 879.0, + 1154.0, + 879.0, + 1154.0, + 914.0, + 321.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 650.0, + 1405.0, + 650.0, + 1405.0, + 690.0, + 296.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 683.0, + 1409.0, + 683.0, + 1409.0, + 720.0, + 321.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 711.0, + 410.0, + 711.0, + 410.0, + 747.0, + 320.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 927.0, + 1408.0, + 927.0, + 1408.0, + 971.0, + 295.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 961.0, + 1404.0, + 961.0, + 1404.0, + 998.0, + 321.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 993.0, + 581.0, + 993.0, + 581.0, + 1027.0, + 322.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 427.0, + 1404.0, + 427.0, + 1404.0, + 461.0, + 296.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 456.0, + 1409.0, + 456.0, + 1409.0, + 495.0, + 320.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 487.0, + 589.0, + 487.0, + 589.0, + 520.0, + 320.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 537.0, + 1409.0, + 537.0, + 1409.0, + 576.0, + 292.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 570.0, + 1409.0, + 570.0, + 1409.0, + 606.0, + 320.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 600.0, + 397.0, + 600.0, + 397.0, + 634.0, + 321.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 260.0, + 1408.0, + 260.0, + 1408.0, + 295.0, + 321.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1158.0, + 1406.0, + 1158.0, + 1406.0, + 1195.0, + 293.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1190.0, + 1406.0, + 1190.0, + 1406.0, + 1228.0, + 323.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1220.0, + 1086.0, + 1220.0, + 1086.0, + 1258.0, + 323.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 309.0, + 1404.0, + 309.0, + 1404.0, + 350.0, + 298.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 343.0, + 1404.0, + 343.0, + 1404.0, + 380.0, + 322.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 372.0, + 1013.0, + 372.0, + 1013.0, + 409.0, + 323.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1046.0, + 1407.0, + 1046.0, + 1407.0, + 1081.0, + 298.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1076.0, + 1409.0, + 1076.0, + 1409.0, + 1113.0, + 322.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1108.0, + 1385.0, + 1108.0, + 1385.0, + 1143.0, + 325.0, + 1143.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1668, + 1405, + 1668, + 1405, + 2034, + 297, + 2034 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 961, + 1405, + 961, + 1405, + 1361, + 297, + 1361 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1376, + 1406, + 1376, + 1406, + 1651, + 298, + 1651 + ], + "score": 0.982 + }, + { + "category_id": 3, + "poly": [ + 314, + 240, + 1378, + 240, + 1378, + 493, + 314, + 493 + ], + "score": 0.959 + }, + { + "category_id": 4, + "poly": [ + 295, + 527, + 1409, + 527, + 1409, + 652, + 295, + 652 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2112, + 835, + 2112 + ], + "score": 0.846 + }, + { + "category_id": 0, + "poly": [ + 302, + 762, + 691, + 762, + 691, + 799, + 302, + 799 + ], + "score": 0.823 + }, + { + "category_id": 0, + "poly": [ + 301, + 876, + 698, + 876, + 698, + 906, + 301, + 906 + ], + "score": 0.802 + }, + { + "category_id": 13, + "poly": [ + 297, + 1022, + 470, + 1022, + 470, + 1055, + 297, + 1055 + ], + "score": 0.92, + "latex": "5 0 0 0 \\times e ^ { \\beta ( \\overline { { { c } } } - 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 795, + 1026, + 969, + 1026, + 969, + 1059, + 795, + 1059 + ], + "score": 0.91, + "latex": "r \\in \\{ 0 . 1 , 0 . 0 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 883, + 1208, + 1148, + 1208, + 1148, + 1242, + 883, + 1242 + ], + "score": 0.89, + "latex": "\\{ 0 . 7 \\mathrm { { 5 } , 0 . 5 , 0 . 2 5 , 0 . 1 2 5 } \\}" + }, + { + "category_id": 13, + "poly": [ + 533, + 1182, + 565, + 1182, + 565, + 1209, + 533, + 1209 + ], + "score": 0.87, + "latex": "n _ { s }" + }, + { + "category_id": 13, + "poly": [ + 741, + 1060, + 773, + 1060, + 773, + 1087, + 741, + 1087 + ], + "score": 0.86, + "latex": "n _ { s }" + }, + { + "category_id": 13, + "poly": [ + 1060, + 565, + 1078, + 565, + 1078, + 587, + 1060, + 587 + ], + "score": 0.72, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 1078, + 999, + 1094, + 999, + 1094, + 1020, + 1078, + 1020 + ], + "score": 0.57, + "latex": "c" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 246.0, + 371.0, + 246.0, + 371.0, + 272.0, + 332.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 253.0, + 673.0, + 253.0, + 673.0, + 279.0, + 520.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 249.0, + 816.0, + 249.0, + 816.0, + 273.0, + 754.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 244.0, + 918.0, + 244.0, + 918.0, + 270.0, + 869.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 250.0, + 1234.0, + 250.0, + 1234.0, + 281.0, + 1048.0, + 281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 248.0, + 1366.0, + 248.0, + 1366.0, + 271.0, + 1290.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 273.0, + 371.0, + 273.0, + 371.0, + 299.0, + 331.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 266.0, + 810.0, + 266.0, + 810.0, + 292.0, + 751.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 271.0, + 918.0, + 271.0, + 918.0, + 295.0, + 869.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 265.0, + 1360.0, + 265.0, + 1360.0, + 289.0, + 1289.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 299.0, + 372.0, + 299.0, + 372.0, + 326.0, + 331.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 295.0, + 937.0, + 295.0, + 937.0, + 423.0, + 835.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 312.0, + 371.0, + 312.0, + 371.0, + 406.0, + 314.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 324.0, + 918.0, + 324.0, + 918.0, + 349.0, + 870.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 353.0, + 372.0, + 353.0, + 372.0, + 379.0, + 331.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 352.0, + 918.0, + 352.0, + 918.0, + 376.0, + 869.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 379.0, + 371.0, + 379.0, + 371.0, + 406.0, + 331.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 406.0, + 372.0, + 406.0, + 372.0, + 431.0, + 331.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 406.0, + 918.0, + 406.0, + 918.0, + 430.0, + 870.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 433.0, + 371.0, + 433.0, + 371.0, + 458.0, + 332.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 433.0, + 916.0, + 433.0, + 916.0, + 457.0, + 870.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 461.0, + 371.0, + 461.0, + 371.0, + 483.0, + 332.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 472.0, + 446.0, + 472.0, + 446.0, + 488.0, + 430.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 472.0, + 528.0, + 472.0, + 528.0, + 488.0, + 514.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 473.0, + 611.0, + 473.0, + 611.0, + 488.0, + 596.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 472.0, + 694.0, + 472.0, + 694.0, + 488.0, + 677.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 471.0, + 781.0, + 471.0, + 781.0, + 489.0, + 757.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 460.0, + 918.0, + 460.0, + 918.0, + 484.0, + 869.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 471.0, + 942.0, + 471.0, + 942.0, + 488.0, + 926.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 471.0, + 1032.0, + 471.0, + 1032.0, + 489.0, + 1001.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 471.0, + 1115.0, + 471.0, + 1115.0, + 488.0, + 1085.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 469.0, + 1200.0, + 469.0, + 1200.0, + 490.0, + 1166.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 469.0, + 1282.0, + 469.0, + 1282.0, + 490.0, + 1249.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 468.0, + 1370.0, + 468.0, + 1370.0, + 489.0, + 1330.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 527.0, + 1410.0, + 527.0, + 1410.0, + 563.0, + 293.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 561.0, + 1059.0, + 561.0, + 1059.0, + 591.0, + 297.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 561.0, + 1405.0, + 561.0, + 1405.0, + 591.0, + 1079.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 588.0, + 1411.0, + 588.0, + 1411.0, + 628.0, + 293.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 618.0, + 357.0, + 618.0, + 357.0, + 656.0, + 293.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 761.0, + 695.0, + 761.0, + 695.0, + 802.0, + 296.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 874.0, + 701.0, + 874.0, + 701.0, + 911.0, + 295.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1668.0, + 1408.0, + 1668.0, + 1408.0, + 1702.0, + 295.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1698.0, + 1403.0, + 1698.0, + 1403.0, + 1728.0, + 296.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1728.0, + 1405.0, + 1728.0, + 1405.0, + 1762.0, + 296.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1761.0, + 1405.0, + 1761.0, + 1405.0, + 1792.0, + 296.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1790.0, + 1405.0, + 1790.0, + 1405.0, + 1821.0, + 296.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1818.0, + 1406.0, + 1818.0, + 1406.0, + 1852.0, + 294.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 1405.0, + 1849.0, + 1405.0, + 1884.0, + 294.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1876.0, + 1406.0, + 1876.0, + 1406.0, + 1919.0, + 292.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1909.0, + 1405.0, + 1909.0, + 1405.0, + 1946.0, + 294.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1944.0, + 1402.0, + 1944.0, + 1402.0, + 1975.0, + 296.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1974.0, + 1403.0, + 1974.0, + 1403.0, + 2004.0, + 295.0, + 2004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2003.0, + 540.0, + 2003.0, + 540.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 962.0, + 1403.0, + 962.0, + 1403.0, + 994.0, + 297.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 990.0, + 1077.0, + 990.0, + 1077.0, + 1028.0, + 294.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 990.0, + 1406.0, + 990.0, + 1406.0, + 1028.0, + 1095.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1016.0, + 296.0, + 1016.0, + 296.0, + 1066.0, + 290.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 1016.0, + 794.0, + 1016.0, + 794.0, + 1066.0, + 471.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 1016.0, + 1407.0, + 1016.0, + 1407.0, + 1066.0, + 970.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1057.0, + 740.0, + 1057.0, + 740.0, + 1089.0, + 295.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 1057.0, + 1405.0, + 1057.0, + 1405.0, + 1089.0, + 774.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1087.0, + 1406.0, + 1087.0, + 1406.0, + 1119.0, + 295.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1118.0, + 1403.0, + 1118.0, + 1403.0, + 1151.0, + 295.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1148.0, + 1405.0, + 1148.0, + 1405.0, + 1181.0, + 296.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1177.0, + 532.0, + 1177.0, + 532.0, + 1213.0, + 295.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 1177.0, + 1406.0, + 1177.0, + 1406.0, + 1213.0, + 566.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1208.0, + 882.0, + 1208.0, + 882.0, + 1244.0, + 295.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1208.0, + 1405.0, + 1208.0, + 1405.0, + 1244.0, + 1149.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1237.0, + 1405.0, + 1237.0, + 1405.0, + 1274.0, + 294.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1272.0, + 1403.0, + 1272.0, + 1403.0, + 1300.0, + 297.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1299.0, + 1403.0, + 1299.0, + 1403.0, + 1332.0, + 295.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1329.0, + 405.0, + 1329.0, + 405.0, + 1365.0, + 291.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1375.0, + 1404.0, + 1375.0, + 1404.0, + 1408.0, + 296.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1403.0, + 1407.0, + 1403.0, + 1407.0, + 1445.0, + 291.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1437.0, + 1409.0, + 1437.0, + 1409.0, + 1473.0, + 295.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1468.0, + 1404.0, + 1468.0, + 1404.0, + 1500.0, + 296.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1499.0, + 1404.0, + 1499.0, + 1404.0, + 1532.0, + 296.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1529.0, + 1404.0, + 1529.0, + 1404.0, + 1562.0, + 296.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1562.0, + 1404.0, + 1562.0, + 1404.0, + 1592.0, + 293.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1589.0, + 1409.0, + 1589.0, + 1409.0, + 1625.0, + 293.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1623.0, + 662.0, + 1623.0, + 662.0, + 1652.0, + 298.0, + 1652.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1412, + 1406, + 1412, + 1406, + 1659, + 298, + 1659 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 298, + 757, + 1402, + 757, + 1402, + 998, + 298, + 998 + ], + "score": 0.98, + "html": "
Imbalanced Ratio r|r =1,balancedr=0.004r=0.0025
Data Quantity n|116K(87K58K29K14K|[116K 87K 58K 29K 14K|80K60K40K20K10K
MoCo V2,ID50.443.540.937.030.849.543.239.536.630.540.6 38.835.531.927.2
MoCo V2,OOD80.379.879.777.477.080.280.179.577.8 77.379.2 78.877.775.674.4
Supervised, ID54.351.6 46.140.5 26.352.949.644.037.3 24.946.1 42.036.327.520.3
Supervised, OOD76.674.7 71.967.4 59.175.573.370.465.8 57.871.8 69.165.960.354.3
SimSiam,OOD80.780.479.978.7 77.280.679.979.678.8 76.979.8 79.378.8 77.5 76.0
" + }, + { + "category_id": 1, + "poly": [ + 299, + 1119, + 1404, + 1119, + 1404, + 1244, + 299, + 1244 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1839, + 1405, + 1839, + 1405, + 2043, + 297, + 2043 + ], + "score": 0.978 + }, + { + "category_id": 3, + "poly": [ + 302, + 240, + 1385, + 240, + 1385, + 576, + 302, + 576 + ], + "score": 0.945 + }, + { + "category_id": 4, + "poly": [ + 296, + 585, + 1403, + 585, + 1403, + 680, + 296, + 680 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 298, + 1258, + 1021, + 1258, + 1021, + 1291, + 298, + 1291 + ], + "score": 0.921 + }, + { + "category_id": 6, + "poly": [ + 606, + 724, + 1091, + 724, + 1091, + 755, + 606, + 755 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 300, + 1061, + 644, + 1061, + 644, + 1093, + 300, + 1093 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 301, + 1781, + 696, + 1781, + 696, + 1813, + 301, + 1813 + ], + "score": 0.902 + }, + { + "category_id": 0, + "poly": [ + 303, + 1339, + 717, + 1339, + 717, + 1376, + 303, + 1376 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 302, + 1707, + 690, + 1707, + 690, + 1744, + 302, + 1744 + ], + "score": 0.871 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2113, + 836, + 2113 + ], + "score": 0.851 + }, + { + "category_id": 13, + "poly": [ + 998, + 1993, + 1128, + 1993, + 1128, + 2039, + 998, + 2039 + ], + "score": 0.93, + "latex": "\\textstyle { \\frac { 1 } { p } } + { \\frac { 1 } { q } } = 1" + }, + { + "category_id": 13, + "poly": [ + 997, + 1901, + 1348, + 1901, + 1348, + 1946, + 997, + 1946 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\widehat { L } _ { w } ( \\phi ) \\ = \\ \\frac { 1 } { n } \\sum _ { j = 1 } ^ { n } w _ { j } \\ell ( x _ { j } , \\phi ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 757, + 1942, + 873, + 1942, + 873, + 1982, + 757, + 1982 + ], + "score": 0.92, + "latex": "\\nabla _ { \\phi } \\widehat { L } _ { w } ( \\phi )" + }, + { + "category_id": 13, + "poly": [ + 295, + 1982, + 906, + 1982, + 906, + 2045, + 295, + 2045 + ], + "score": 0.89, + "latex": "\\rho \\mathrm { s g n } ( \\nabla _ { \\phi } \\widehat { L } _ { w } ( \\phi ) ) \\left| \\nabla _ { \\phi } \\widehat { L } _ { w } ( \\phi ) \\right| ^ { q - 1 } / \\left( \\| \\nabla _ { \\phi } \\widehat { L } _ { w } ( \\phi ) \\| _ { q } ^ { q } \\right) ^ { 1 / p }" + }, + { + "category_id": 13, + "poly": [ + 1199, + 1948, + 1252, + 1948, + 1252, + 1982, + 1199, + 1982 + ], + "score": 0.89, + "latex": "\\epsilon ( \\phi )" + }, + { + "category_id": 13, + "poly": [ + 1303, + 1947, + 1402, + 1947, + 1402, + 1983, + 1303, + 1983 + ], + "score": 0.88, + "latex": "\\begin{array} { r l } { \\epsilon ( \\phi ) } & { { } = } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 648, + 1949, + 669, + 1949, + 669, + 1981, + 648, + 1981 + ], + "score": 0.8, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 1278, + 258, + 1364, + 258, + 1364, + 278, + 1278, + 278 + ], + "score": 0.53, + "latex": "{ \\sf r } = 0 . 0 0 2 5" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 237.0, + 335.0, + 237.0, + 335.0, + 518.0, + 300.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 237.0, + 844.0, + 237.0, + 844.0, + 512.0, + 809.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 243.0, + 846.0, + 243.0, + 846.0, + 262.0, + 831.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 258.0, + 346.0, + 258.0, + 346.0, + 284.0, + 321.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 268.0, + 845.0, + 268.0, + 845.0, + 289.0, + 832.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 252.0, + 1095.0, + 252.0, + 1095.0, + 280.0, + 857.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 254.0, + 1277.0, + 254.0, + 1277.0, + 282.0, + 1126.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1365.0, + 254.0, + 1370.0, + 254.0, + 1370.0, + 282.0, + 1365.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 303.0, + 346.0, + 303.0, + 346.0, + 324.0, + 323.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 297.0, + 846.0, + 297.0, + 846.0, + 349.0, + 832.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 283.0, + 963.0, + 283.0, + 963.0, + 350.0, + 906.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 346.0, + 346.0, + 346.0, + 346.0, + 367.0, + 323.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 345.0, + 735.0, + 345.0, + 735.0, + 378.0, + 648.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 356.0, + 846.0, + 356.0, + 846.0, + 377.0, + 832.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 388.0, + 346.0, + 388.0, + 346.0, + 453.0, + 323.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 368.0, + 767.0, + 368.0, + 767.0, + 452.0, + 643.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 383.0, + 846.0, + 383.0, + 846.0, + 434.0, + 832.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 444.0, + 766.0, + 444.0, + 766.0, + 476.0, + 648.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 445.0, + 845.0, + 445.0, + 845.0, + 461.0, + 835.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 471.0, + 346.0, + 471.0, + 346.0, + 498.0, + 322.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 469.0, + 765.0, + 469.0, + 765.0, + 498.0, + 645.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 472.0, + 846.0, + 472.0, + 846.0, + 491.0, + 831.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 502.0, + 425.0, + 502.0, + 425.0, + 523.0, + 376.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 503.0, + 500.0, + 503.0, + 500.0, + 522.0, + 452.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 502.0, + 575.0, + 502.0, + 575.0, + 523.0, + 526.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 502.0, + 650.0, + 502.0, + 650.0, + 523.0, + 600.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 502.0, + 728.0, + 502.0, + 728.0, + 523.0, + 672.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 496.0, + 1094.0, + 496.0, + 1094.0, + 519.0, + 866.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 496.0, + 1364.0, + 496.0, + 1364.0, + 519.0, + 1142.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 516.0, + 710.0, + 516.0, + 710.0, + 539.0, + 406.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 511.0, + 1303.0, + 511.0, + 1303.0, + 539.0, + 925.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 543.0, + 645.0, + 543.0, + 645.0, + 582.0, + 436.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 543.0, + 1192.0, + 543.0, + 1192.0, + 582.0, + 1005.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 586.0, + 1404.0, + 586.0, + 1404.0, + 620.0, + 295.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 617.0, + 1404.0, + 617.0, + 1404.0, + 651.0, + 295.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 648.0, + 1037.0, + 648.0, + 1037.0, + 682.0, + 296.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 718.0, + 1095.0, + 718.0, + 1095.0, + 763.0, + 602.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1056.0, + 646.0, + 1056.0, + 646.0, + 1098.0, + 295.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1779.0, + 701.0, + 1779.0, + 701.0, + 1816.0, + 295.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1336.0, + 722.0, + 1336.0, + 722.0, + 1380.0, + 294.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1703.0, + 696.0, + 1703.0, + 696.0, + 1750.0, + 294.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1413.0, + 1407.0, + 1413.0, + 1407.0, + 1446.0, + 295.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1444.0, + 1406.0, + 1444.0, + 1406.0, + 1476.0, + 293.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1476.0, + 1403.0, + 1476.0, + 1403.0, + 1507.0, + 296.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1505.0, + 1406.0, + 1505.0, + 1406.0, + 1539.0, + 296.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1536.0, + 1406.0, + 1536.0, + 1406.0, + 1569.0, + 295.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1564.0, + 1407.0, + 1564.0, + 1407.0, + 1601.0, + 292.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1597.0, + 1404.0, + 1597.0, + 1404.0, + 1631.0, + 296.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1628.0, + 946.0, + 1628.0, + 946.0, + 1658.0, + 297.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1115.0, + 1406.0, + 1115.0, + 1406.0, + 1159.0, + 291.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1151.0, + 1405.0, + 1151.0, + 1405.0, + 1186.0, + 293.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1183.0, + 1404.0, + 1183.0, + 1404.0, + 1215.0, + 294.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1213.0, + 971.0, + 1213.0, + 971.0, + 1246.0, + 295.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1834.0, + 1406.0, + 1834.0, + 1406.0, + 1875.0, + 294.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1870.0, + 1406.0, + 1870.0, + 1406.0, + 1903.0, + 295.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1889.0, + 996.0, + 1889.0, + 996.0, + 1956.0, + 285.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 1889.0, + 1416.0, + 1889.0, + 1416.0, + 1956.0, + 1349.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1941.0, + 647.0, + 1941.0, + 647.0, + 1987.0, + 292.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1941.0, + 756.0, + 1941.0, + 756.0, + 1987.0, + 670.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1941.0, + 1198.0, + 1941.0, + 1198.0, + 1987.0, + 874.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 1941.0, + 1302.0, + 1941.0, + 1302.0, + 1987.0, + 1253.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1941.0, + 1406.0, + 1941.0, + 1406.0, + 1987.0, + 1403.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1973.0, + 294.0, + 1973.0, + 294.0, + 2058.0, + 284.0, + 2058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 1973.0, + 997.0, + 1973.0, + 997.0, + 2058.0, + 907.0, + 2058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1973.0, + 1416.0, + 1973.0, + 1416.0, + 2058.0, + 1129.0, + 2058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1982.5, + 911.0, + 1982.5, + 911.0, + 2003.5, + 870.0, + 2003.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1256.0, + 1025.0, + 1256.0, + 1025.0, + 1295.0, + 294.0, + 1295.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 899, + 1749, + 1377, + 1749, + 1377, + 2017, + 899, + 2017 + ], + "score": 0.981, + "html": "
Method|Backbone|Acc.
SupervisedCRT (Kang et al., 2020)LADE (Hong et al., 2021)RIDE (Wang et al., 2021)RIDE (Wang et al., 2021)ResNet-50ResNet-50ResNeXt-50ResNet-50ResNeXt-5049.3
52.053.054.9
56.4
MoCo V2MoCo V2+rwSAMResNet-5055.0
ResNet-5055.5
" + }, + { + "category_id": 1, + "poly": [ + 299, + 1758, + 865, + 1758, + 865, + 2034, + 299, + 2034 + ], + "score": 0.981 + }, + { + "category_id": 3, + "poly": [ + 421, + 223, + 1280, + 223, + 1280, + 518, + 421, + 518 + ], + "score": 0.972 + }, + { + "category_id": 4, + "poly": [ + 296, + 558, + 1405, + 558, + 1405, + 713, + 296, + 713 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 297, + 1598, + 1403, + 1598, + 1403, + 1660, + 297, + 1660 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 299, + 766, + 1401, + 766, + 1401, + 836, + 299, + 836 + ], + "score": 0.944 + }, + { + "category_id": 6, + "poly": [ + 928, + 1718, + 1350, + 1718, + 1350, + 1747, + 928, + 1747 + ], + "score": 0.9 + }, + { + "category_id": 0, + "poly": [ + 300, + 1701, + 641, + 1701, + 641, + 1732, + 300, + 1732 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.883 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2113, + 835, + 2113 + ], + "score": 0.854 + }, + { + "category_id": 0, + "poly": [ + 297, + 870, + 1055, + 870, + 1055, + 905, + 297, + 905 + ], + "score": 0.804 + }, + { + "category_id": 8, + "poly": [ + 770, + 1110, + 1001, + 1110, + 1001, + 1150, + 770, + 1150 + ], + "score": 0.73 + }, + { + "category_id": 8, + "poly": [ + 639, + 1189, + 1108, + 1189, + 1108, + 1263, + 639, + 1263 + ], + "score": 0.707 + }, + { + "category_id": 8, + "poly": [ + 729, + 1483, + 1041, + 1483, + 1041, + 1523, + 729, + 1523 + ], + "score": 0.587 + }, + { + "category_id": 1, + "poly": [ + 308, + 914, + 1065, + 914, + 1065, + 1107, + 308, + 1107 + ], + "score": 0.565 + }, + { + "category_id": 1, + "poly": [ + 310, + 1157, + 920, + 1157, + 920, + 1188, + 310, + 1188 + ], + "score": 0.56 + }, + { + "category_id": 0, + "poly": [ + 300, + 1526, + 434, + 1526, + 434, + 1557, + 300, + 1557 + ], + "score": 0.406 + }, + { + "category_id": 1, + "poly": [ + 300, + 1526, + 434, + 1526, + 434, + 1557, + 300, + 1557 + ], + "score": 0.384 + }, + { + "category_id": 8, + "poly": [ + 544, + 1388, + 1215, + 1388, + 1215, + 1443, + 544, + 1443 + ], + "score": 0.265 + }, + { + "category_id": 0, + "poly": [ + 311, + 1126, + 433, + 1126, + 433, + 1155, + 311, + 1155 + ], + "score": 0.231 + }, + { + "category_id": 1, + "poly": [ + 299, + 1446, + 1368, + 1446, + 1368, + 1479, + 299, + 1479 + ], + "score": 0.214 + }, + { + "category_id": 1, + "poly": [ + 308, + 1253, + 648, + 1253, + 648, + 1284, + 308, + 1284 + ], + "score": 0.148 + }, + { + "category_id": 1, + "poly": [ + 311, + 1126, + 433, + 1126, + 433, + 1155, + 311, + 1155 + ], + "score": 0.118 + }, + { + "category_id": 1, + "poly": [ + 299, + 1287, + 1009, + 1287, + 1009, + 1349, + 299, + 1349 + ], + "score": 0.104 + }, + { + "category_id": 1, + "poly": [ + 297, + 870, + 1055, + 870, + 1055, + 905, + 297, + 905 + ], + "score": 0.1 + }, + { + "category_id": 14, + "poly": [ + 636, + 1183, + 1109, + 1183, + 1109, + 1269, + 636, + 1269 + ], + "score": 0.93, + "latex": "w _ { i } = \\big ( \\frac { 1 } { n } \\sum _ { j = 1 } ^ { n ^ { * } } K ( f _ { \\phi } ( x _ { i } ) - f _ { \\phi } ( x _ { j } ) , h ) \\big ) ^ { - \\alpha } ." + }, + { + "category_id": 13, + "poly": [ + 733, + 1070, + 825, + 1070, + 825, + 1106, + 733, + 1106 + ], + "score": 0.93, + "latex": "\\{ x _ { i } \\} _ { i = 1 } ^ { b }" + }, + { + "category_id": 13, + "poly": [ + 818, + 766, + 1115, + 766, + 1115, + 807, + 818, + 807 + ], + "score": 0.93, + "latex": "\\phi = \\phi - \\eta \\nabla _ { \\phi } \\widehat { L } ( \\phi + \\epsilon ( \\phi ) )" + }, + { + "category_id": 13, + "poly": [ + 807, + 1035, + 899, + 1035, + 899, + 1073, + 807, + 1073 + ], + "score": 0.92, + "latex": "\\{ x _ { i } \\} _ { i = 1 } ^ { b }" + }, + { + "category_id": 13, + "poly": [ + 484, + 1350, + 535, + 1350, + 535, + 1383, + 484, + 1383 + ], + "score": 0.92, + "latex": "\\epsilon ( \\phi )" + }, + { + "category_id": 13, + "poly": [ + 721, + 766, + 780, + 766, + 780, + 806, + 721, + 806 + ], + "score": 0.92, + "latex": "\\widehat { L } ( \\phi )" + }, + { + "category_id": 13, + "poly": [ + 807, + 1310, + 899, + 1310, + 899, + 1347, + 807, + 1347 + ], + "score": 0.92, + "latex": "\\{ x _ { i } \\} _ { i = 1 } ^ { b }" + }, + { + "category_id": 13, + "poly": [ + 865, + 1347, + 941, + 1347, + 941, + 1384, + 865, + 1384 + ], + "score": 0.91, + "latex": "\\widehat { L } _ { w } ( \\phi )" + }, + { + "category_id": 13, + "poly": [ + 659, + 1630, + 727, + 1630, + 727, + 1661, + 659, + 1661 + ], + "score": 0.9, + "latex": "\\rho = 5" + }, + { + "category_id": 13, + "poly": [ + 778, + 1630, + 871, + 1630, + 871, + 1657, + 778, + 1657 + ], + "score": 0.9, + "latex": "\\alpha = 1 . 2" + }, + { + "category_id": 13, + "poly": [ + 385, + 1009, + 450, + 1009, + 450, + 1037, + 385, + 1037 + ], + "score": 0.89, + "latex": "i = 0" + }, + { + "category_id": 13, + "poly": [ + 385, + 1285, + 450, + 1285, + 450, + 1312, + 385, + 1312 + ], + "score": 0.89, + "latex": "i = 0" + }, + { + "category_id": 13, + "poly": [ + 964, + 1032, + 1000, + 1032, + 1000, + 1070, + 964, + 1070 + ], + "score": 0.89, + "latex": "\\widehat { \\mathcal { D } } _ { s }" + }, + { + "category_id": 13, + "poly": [ + 1333, + 1600, + 1402, + 1600, + 1402, + 1631, + 1333, + 1631 + ], + "score": 0.89, + "latex": "\\rho = 2" + }, + { + "category_id": 13, + "poly": [ + 344, + 1629, + 437, + 1629, + 437, + 1658, + 344, + 1658 + ], + "score": 0.89, + "latex": "\\alpha = 0 . 5" + }, + { + "category_id": 14, + "poly": [ + 770, + 1107, + 1000, + 1107, + 1000, + 1150, + 770, + 1150 + ], + "score": 0.89, + "latex": "\\phi \\phi - \\eta \\nabla _ { \\phi } \\widehat { L } ( \\phi ) ." + }, + { + "category_id": 13, + "poly": [ + 690, + 911, + 726, + 911, + 726, + 948, + 690, + 948 + ], + "score": 0.88, + "latex": "\\widehat { \\mathcal { D } } _ { s }" + }, + { + "category_id": 13, + "poly": [ + 674, + 1074, + 695, + 1074, + 695, + 1105, + 674, + 1105 + ], + "score": 0.85, + "latex": "\\phi" + }, + { + "category_id": 14, + "poly": [ + 729, + 1478, + 1040, + 1478, + 1040, + 1522, + 729, + 1522 + ], + "score": 0.85, + "latex": "\\phi \\phi - \\eta \\nabla _ { \\phi } \\widehat { L } ( \\phi + \\epsilon ( \\phi ) ) ." + }, + { + "category_id": 13, + "poly": [ + 964, + 1307, + 1000, + 1307, + 1000, + 1344, + 964, + 1344 + ], + "score": 0.84, + "latex": "\\widehat { \\mathcal { D } } _ { s }" + }, + { + "category_id": 13, + "poly": [ + 711, + 950, + 731, + 950, + 731, + 980, + 711, + 980 + ], + "score": 0.83, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 733, + 1441, + 825, + 1441, + 825, + 1479, + 733, + 1479 + ], + "score": 0.83, + "latex": "\\{ x _ { i } \\} _ { i = 1 } ^ { b }" + }, + { + "category_id": 13, + "poly": [ + 634, + 1605, + 652, + 1605, + 652, + 1630, + 634, + 1630 + ], + "score": 0.82, + "latex": "\\rho" + }, + { + "category_id": 14, + "poly": [ + 547, + 1388, + 1223, + 1388, + 1223, + 1448, + 547, + 1448 + ], + "score": 0.81, + "latex": "\\epsilon _ { \\phi } = \\rho \\mathrm { s g n } ( \\nabla _ { \\phi } \\widehat { L } _ { w } ( \\phi ) ) \\left| \\nabla _ { \\phi } \\widehat { L } _ { w } ( \\phi ) \\right| ^ { q - 1 } / \\left( \\lVert \\nabla _ { \\phi } \\widehat { L } _ { w } ( \\phi ) \\rVert _ { q } ^ { q } \\right) ^ { 1 / p }" + }, + { + "category_id": 13, + "poly": [ + 701, + 1605, + 722, + 1605, + 722, + 1626, + 701, + 1626 + ], + "score": 0.8, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 674, + 1446, + 695, + 1446, + 695, + 1477, + 674, + 1477 + ], + "score": 0.76, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 673, + 983, + 696, + 983, + 696, + 1005, + 673, + 1005 + ], + "score": 0.75, + "latex": "w" + }, + { + "category_id": 13, + "poly": [ + 396, + 1851, + 438, + 1851, + 438, + 1879, + 396, + 1879 + ], + "score": 0.37, + "latex": "\\mathrm { X u }" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 227.0, + 703.0, + 227.0, + 703.0, + 248.0, + 624.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 226.0, + 899.0, + 226.0, + 899.0, + 249.0, + 794.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 227.0, + 1061.0, + 227.0, + 1061.0, + 249.0, + 982.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 226.0, + 1256.0, + 226.0, + 1256.0, + 249.0, + 1152.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 238.0, + 548.0, + 238.0, + 548.0, + 261.0, + 434.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 245.0, + 738.0, + 245.0, + 738.0, + 268.0, + 626.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 244.0, + 906.0, + 244.0, + 906.0, + 269.0, + 794.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 245.0, + 1095.0, + 245.0, + 1095.0, + 268.0, + 983.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 245.0, + 1263.0, + 245.0, + 1263.0, + 268.0, + 1152.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 278.0, + 748.0, + 278.0, + 748.0, + 347.0, + 612.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 268.0, + 926.0, + 268.0, + 926.0, + 353.0, + 842.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 273.0, + 1113.0, + 273.0, + 1113.0, + 350.0, + 1035.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 268.0, + 1284.0, + 268.0, + 1284.0, + 355.0, + 1172.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 332.0, + 562.0, + 332.0, + 562.0, + 409.0, + 421.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 325.0, + 609.0, + 325.0, + 609.0, + 444.0, + 583.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 355.0, + 755.0, + 355.0, + 755.0, + 434.0, + 614.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 355.0, + 927.0, + 355.0, + 927.0, + 439.0, + 840.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 349.0, + 960.0, + 349.0, + 960.0, + 434.0, + 940.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 355.0, + 1112.0, + 355.0, + 1112.0, + 437.0, + 1002.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 352.0, + 1285.0, + 352.0, + 1285.0, + 440.0, + 1194.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 390.0, + 649.0, + 390.0, + 649.0, + 402.0, + 636.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 425.0, + 522.0, + 425.0, + 522.0, + 446.0, + 498.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 486.0, + 469.0, + 486.0, + 469.0, + 507.0, + 426.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 480.0, + 549.0, + 480.0, + 549.0, + 501.0, + 503.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 486.0, + 637.0, + 486.0, + 637.0, + 502.0, + 616.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 440.0, + 922.0, + 440.0, + 922.0, + 519.0, + 774.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 436.0, + 1115.0, + 436.0, + 1115.0, + 521.0, + 1028.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 436.0, + 1283.0, + 436.0, + 1283.0, + 520.0, + 1172.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 497.0, + 563.0, + 497.0, + 563.0, + 521.0, + 503.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 375.5, + 1218.0, + 375.5, + 1218.0, + 415.5, + 1164.0, + 415.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 560.0, + 1403.0, + 560.0, + 1403.0, + 593.0, + 295.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 590.0, + 1403.0, + 590.0, + 1403.0, + 624.0, + 295.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 620.0, + 1405.0, + 620.0, + 1405.0, + 657.0, + 295.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 650.0, + 1405.0, + 650.0, + 1405.0, + 687.0, + 295.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 678.0, + 1157.0, + 678.0, + 1157.0, + 718.0, + 295.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1714.0, + 1351.0, + 1714.0, + 1351.0, + 1752.0, + 928.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1698.0, + 645.0, + 1698.0, + 645.0, + 1737.0, + 295.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 867.0, + 1057.0, + 867.0, + 1057.0, + 911.0, + 296.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1524.0, + 437.0, + 1524.0, + 437.0, + 1558.0, + 298.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1119.0, + 440.0, + 1119.0, + 440.0, + 1161.0, + 306.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1758.0, + 864.0, + 1758.0, + 864.0, + 1791.0, + 297.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1790.0, + 867.0, + 1790.0, + 867.0, + 1821.0, + 298.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1820.0, + 866.0, + 1820.0, + 866.0, + 1853.0, + 296.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1850.0, + 395.0, + 1850.0, + 395.0, + 1883.0, + 297.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1850.0, + 868.0, + 1850.0, + 868.0, + 1883.0, + 439.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1882.0, + 868.0, + 1882.0, + 868.0, + 1913.0, + 297.0, + 1913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1912.0, + 864.0, + 1912.0, + 864.0, + 1942.0, + 297.0, + 1942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1941.0, + 869.0, + 1941.0, + 869.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1971.0, + 865.0, + 1971.0, + 865.0, + 2005.0, + 296.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2001.0, + 868.0, + 2001.0, + 868.0, + 2037.0, + 295.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1595.0, + 633.0, + 1595.0, + 633.0, + 1634.0, + 293.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1595.0, + 700.0, + 1595.0, + 700.0, + 1634.0, + 653.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1595.0, + 1332.0, + 1595.0, + 1332.0, + 1634.0, + 723.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1595.0, + 1408.0, + 1595.0, + 1408.0, + 1634.0, + 1403.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1626.0, + 343.0, + 1626.0, + 343.0, + 1662.0, + 295.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1626.0, + 658.0, + 1626.0, + 658.0, + 1662.0, + 438.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1626.0, + 777.0, + 1626.0, + 777.0, + 1662.0, + 728.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1626.0, + 883.0, + 1626.0, + 883.0, + 1662.0, + 872.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 765.0, + 720.0, + 765.0, + 720.0, + 810.0, + 293.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 765.0, + 817.0, + 765.0, + 817.0, + 810.0, + 781.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 765.0, + 1406.0, + 765.0, + 1406.0, + 810.0, + 1116.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 801.0, + 595.0, + 801.0, + 595.0, + 838.0, + 295.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 914.0, + 689.0, + 914.0, + 689.0, + 949.0, + 313.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 914.0, + 734.0, + 914.0, + 734.0, + 949.0, + 727.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 950.0, + 337.0, + 950.0, + 337.0, + 977.0, + 312.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 945.0, + 710.0, + 945.0, + 710.0, + 982.0, + 339.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 945.0, + 738.0, + 945.0, + 738.0, + 982.0, + 732.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 981.0, + 337.0, + 981.0, + 337.0, + 1006.0, + 312.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 979.0, + 672.0, + 979.0, + 672.0, + 1012.0, + 341.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 979.0, + 707.0, + 979.0, + 707.0, + 1012.0, + 697.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1009.0, + 338.0, + 1009.0, + 338.0, + 1038.0, + 310.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1006.0, + 384.0, + 1006.0, + 384.0, + 1041.0, + 340.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1006.0, + 644.0, + 1006.0, + 644.0, + 1041.0, + 451.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 1040.0, + 338.0, + 1040.0, + 338.0, + 1071.0, + 309.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1033.0, + 806.0, + 1033.0, + 806.0, + 1075.0, + 364.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 1033.0, + 963.0, + 1033.0, + 963.0, + 1075.0, + 900.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1033.0, + 1012.0, + 1033.0, + 1012.0, + 1075.0, + 1001.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1075.0, + 338.0, + 1075.0, + 338.0, + 1106.0, + 310.0, + 1106.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1068.0, + 673.0, + 1068.0, + 673.0, + 1110.0, + 368.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1068.0, + 732.0, + 1068.0, + 732.0, + 1110.0, + 696.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1068.0, + 1066.0, + 1068.0, + 1066.0, + 1110.0, + 826.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1153.0, + 922.0, + 1153.0, + 922.0, + 1193.0, + 306.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1524.0, + 437.0, + 1524.0, + 437.0, + 1558.0, + 298.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1433.0, + 673.0, + 1433.0, + 673.0, + 1487.0, + 292.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1433.0, + 732.0, + 1433.0, + 732.0, + 1487.0, + 696.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1433.0, + 1372.0, + 1433.0, + 1372.0, + 1487.0, + 826.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 1250.0, + 649.0, + 1250.0, + 649.0, + 1289.0, + 309.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1119.0, + 440.0, + 1119.0, + 440.0, + 1161.0, + 306.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1283.0, + 384.0, + 1283.0, + 384.0, + 1314.0, + 298.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1283.0, + 643.0, + 1283.0, + 643.0, + 1314.0, + 451.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1310.0, + 806.0, + 1310.0, + 806.0, + 1350.0, + 291.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 1310.0, + 963.0, + 1310.0, + 963.0, + 1350.0, + 900.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1310.0, + 1011.0, + 1310.0, + 1011.0, + 1350.0, + 1001.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 867.0, + 1057.0, + 867.0, + 1057.0, + 911.0, + 296.0, + 911.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 796, + 1404, + 796, + 1404, + 1132, + 298, + 1132 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1404, + 229, + 1404, + 505, + 298, + 505 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1788, + 1404, + 1788, + 1404, + 2035, + 298, + 2035 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1546, + 1404, + 1546, + 1404, + 1699, + 298, + 1699 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1148, + 1402, + 1148, + 1402, + 1270, + 299, + 1270 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 299, + 521, + 1404, + 521, + 1404, + 642, + 299, + 642 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 1286, + 1404, + 1286, + 1404, + 1531, + 298, + 1531 + ], + "score": 0.968 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 302, + 1734, + 723, + 1734, + 723, + 1765, + 302, + 1765 + ], + "score": 0.868 + }, + { + "category_id": 0, + "poly": [ + 299, + 744, + 995, + 744, + 995, + 776, + 299, + 776 + ], + "score": 0.862 + }, + { + "category_id": 2, + "poly": [ + 836, + 2089, + 863, + 2089, + 863, + 2112, + 836, + 2112 + ], + "score": 0.846 + }, + { + "category_id": 0, + "poly": [ + 301, + 681, + 788, + 681, + 788, + 715, + 301, + 715 + ], + "score": 0.836 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1733.0, + 727.0, + 1733.0, + 727.0, + 1769.0, + 296.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 745.0, + 997.0, + 745.0, + 997.0, + 780.0, + 296.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2122.0, + 832.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 678.0, + 794.0, + 678.0, + 794.0, + 721.0, + 294.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 794.0, + 1408.0, + 794.0, + 1408.0, + 833.0, + 293.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 825.0, + 1405.0, + 825.0, + 1405.0, + 861.0, + 295.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 856.0, + 1405.0, + 856.0, + 1405.0, + 891.0, + 296.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 883.0, + 1408.0, + 883.0, + 1408.0, + 923.0, + 292.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 916.0, + 1404.0, + 916.0, + 1404.0, + 952.0, + 293.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 947.0, + 1405.0, + 947.0, + 1405.0, + 984.0, + 293.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 980.0, + 1402.0, + 980.0, + 1402.0, + 1011.0, + 297.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1007.0, + 1408.0, + 1007.0, + 1408.0, + 1043.0, + 295.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1040.0, + 1406.0, + 1040.0, + 1406.0, + 1074.0, + 296.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1067.0, + 1405.0, + 1067.0, + 1405.0, + 1108.0, + 293.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1103.0, + 1291.0, + 1103.0, + 1291.0, + 1134.0, + 296.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 230.0, + 1406.0, + 230.0, + 1406.0, + 263.0, + 296.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 261.0, + 1404.0, + 261.0, + 1404.0, + 294.0, + 296.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 290.0, + 1408.0, + 290.0, + 1408.0, + 327.0, + 294.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 321.0, + 1405.0, + 321.0, + 1405.0, + 354.0, + 296.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 353.0, + 1406.0, + 353.0, + 1406.0, + 385.0, + 296.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 383.0, + 1405.0, + 383.0, + 1405.0, + 415.0, + 296.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 413.0, + 1406.0, + 413.0, + 1406.0, + 447.0, + 293.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 444.0, + 1406.0, + 444.0, + 1406.0, + 477.0, + 294.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 471.0, + 366.0, + 471.0, + 366.0, + 507.0, + 291.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1786.0, + 1408.0, + 1786.0, + 1408.0, + 1822.0, + 293.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1820.0, + 1405.0, + 1820.0, + 1405.0, + 1854.0, + 294.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1849.0, + 1408.0, + 1849.0, + 1408.0, + 1886.0, + 293.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1881.0, + 1405.0, + 1881.0, + 1405.0, + 1915.0, + 294.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1912.0, + 1404.0, + 1912.0, + 1404.0, + 1945.0, + 294.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1975.0, + 294.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2005.0, + 293.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2001.0, + 1405.0, + 2001.0, + 1405.0, + 2037.0, + 295.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1546.0, + 1406.0, + 1546.0, + 1406.0, + 1579.0, + 297.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1575.0, + 1402.0, + 1575.0, + 1402.0, + 1608.0, + 297.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1604.0, + 1405.0, + 1604.0, + 1405.0, + 1642.0, + 293.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1636.0, + 1404.0, + 1636.0, + 1404.0, + 1674.0, + 295.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1666.0, + 1179.0, + 1666.0, + 1179.0, + 1704.0, + 295.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1144.0, + 1406.0, + 1144.0, + 1406.0, + 1182.0, + 293.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1179.0, + 1406.0, + 1179.0, + 1406.0, + 1211.0, + 295.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1209.0, + 1404.0, + 1209.0, + 1404.0, + 1245.0, + 294.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1240.0, + 933.0, + 1240.0, + 933.0, + 1273.0, + 295.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 520.0, + 1404.0, + 520.0, + 1404.0, + 556.0, + 294.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 554.0, + 1405.0, + 554.0, + 1405.0, + 583.0, + 297.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 584.0, + 1404.0, + 584.0, + 1404.0, + 616.0, + 295.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 612.0, + 520.0, + 612.0, + 520.0, + 642.0, + 293.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1284.0, + 1409.0, + 1284.0, + 1409.0, + 1321.0, + 293.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1315.0, + 1404.0, + 1315.0, + 1404.0, + 1349.0, + 294.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1345.0, + 1404.0, + 1345.0, + 1404.0, + 1381.0, + 293.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1375.0, + 1405.0, + 1375.0, + 1405.0, + 1415.0, + 292.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1407.0, + 1408.0, + 1407.0, + 1408.0, + 1441.0, + 293.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1436.0, + 1406.0, + 1436.0, + 1406.0, + 1476.0, + 292.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1468.0, + 1404.0, + 1468.0, + 1404.0, + 1503.0, + 293.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1498.0, + 665.0, + 1498.0, + 665.0, + 1532.0, + 294.0, + 1532.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 229, + 1404, + 229, + 1404, + 506, + 298, + 506 + ], + "score": 0.955 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 104, + 300, + 104 + ], + "score": 0.852 + }, + { + "category_id": 2, + "poly": [ + 836, + 2089, + 864, + 2089, + 864, + 2112, + 836, + 2112 + ], + "score": 0.834 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2086.0, + 868.0, + 2086.0, + 868.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 229.0, + 1404.0, + 229.0, + 1404.0, + 261.0, + 294.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 259.0, + 1406.0, + 259.0, + 1406.0, + 296.0, + 296.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 293.0, + 1404.0, + 293.0, + 1404.0, + 322.0, + 297.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 320.0, + 1404.0, + 320.0, + 1404.0, + 353.0, + 296.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 350.0, + 1405.0, + 350.0, + 1405.0, + 386.0, + 293.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 381.0, + 1404.0, + 381.0, + 1404.0, + 416.0, + 294.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 413.0, + 1405.0, + 413.0, + 1405.0, + 446.0, + 294.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 439.0, + 1405.0, + 439.0, + 1405.0, + 477.0, + 293.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 472.0, + 1046.0, + 472.0, + 1046.0, + 509.0, + 294.0, + 509.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1224, + 1406, + 1224, + 1406, + 1414, + 296, + 1414 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 284, + 1407, + 284, + 1407, + 448, + 296, + 448 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 296, + 1517, + 1407, + 1517, + 1407, + 1675, + 296, + 1675 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 635, + 816, + 1061, + 816, + 1061, + 880, + 635, + 880 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 652, + 935, + 1045, + 935, + 1045, + 993, + 652, + 993 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 298, + 1002, + 1403, + 1002, + 1403, + 1067, + 298, + 1067 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 295, + 1703, + 1402, + 1703, + 1402, + 1777, + 295, + 1777 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 397, + 1923, + 1300, + 1923, + 1300, + 2031, + 397, + 2031 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 295, + 1442, + 1402, + 1442, + 1402, + 1507, + 295, + 1507 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 296, + 740, + 1403, + 740, + 1403, + 805, + 296, + 805 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 617, + 1855, + 1083, + 1855, + 1083, + 1902, + 617, + 1902 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 295, + 458, + 1400, + 458, + 1400, + 522, + 295, + 522 + ], + "score": 0.933 + }, + { + "category_id": 8, + "poly": [ + 637, + 1786, + 1062, + 1786, + 1062, + 1833, + 637, + 1833 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 300, + 889, + 799, + 889, + 799, + 924, + 300, + 924 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 299, + 224, + 702, + 224, + 702, + 263, + 299, + 263 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 297, + 528, + 1399, + 528, + 1399, + 606, + 297, + 606 + ], + "score": 0.919 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 816, + 73, + 816, + 106, + 298, + 106 + ], + "score": 0.917 + }, + { + "category_id": 9, + "poly": [ + 1365, + 948, + 1401, + 948, + 1401, + 979, + 1365, + 979 + ], + "score": 0.881 + }, + { + "category_id": 9, + "poly": [ + 1365, + 836, + 1401, + 836, + 1401, + 866, + 1365, + 866 + ], + "score": 0.88 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1866, + 1401, + 1866, + 1401, + 1896, + 1365, + 1896 + ], + "score": 0.869 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1797, + 1401, + 1797, + 1401, + 1827, + 1365, + 1827 + ], + "score": 0.868 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 867, + 2087, + 867, + 2114, + 834, + 2114 + ], + "score": 0.864 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1167, + 1401, + 1167, + 1401, + 1197, + 1365, + 1197 + ], + "score": 0.86 + }, + { + "category_id": 8, + "poly": [ + 304, + 1081, + 1393, + 1081, + 1393, + 1172, + 304, + 1172 + ], + "score": 0.859 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1960, + 1401, + 1960, + 1401, + 1991, + 1365, + 1991 + ], + "score": 0.845 + }, + { + "category_id": 8, + "poly": [ + 306, + 611, + 1128, + 611, + 1128, + 713, + 306, + 713 + ], + "score": 0.638 + }, + { + "category_id": 1, + "poly": [ + 306, + 611, + 1128, + 611, + 1128, + 713, + 306, + 713 + ], + "score": 0.273 + }, + { + "category_id": 13, + "poly": [ + 380, + 1259, + 729, + 1259, + 729, + 1332, + 380, + 1332 + ], + "score": 0.95, + "latex": "\\begin{array} { r } { 1 - ( n ^ { 2 } + 2 n ) ( \\frac { e ^ { - \\frac { d ^ { \\frac { 1 } { 5 } } } { 2 } } } { d ^ { \\frac { 1 } { 1 0 } } } + 2 e ^ { - \\sqrt { d } } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 920, + 1735, + 1021, + 1735, + 1021, + 1777, + 920, + 1777 + ], + "score": 0.94, + "latex": "x = x _ { i } ^ { ( 1 ) }" + }, + { + "category_id": 14, + "poly": [ + 633, + 810, + 1066, + 810, + 1066, + 882, + 633, + 882 + ], + "score": 0.94, + "latex": "\\Pr \\left( | \\langle \\xi , e _ { 1 } \\rangle | \\geq d ^ { \\frac { 1 } { 1 0 } } \\right) \\leq d ^ { - { \\frac { 1 } { 1 0 } } } \\cdot e ^ { - { \\frac { d ^ { \\frac { 1 } { 5 } } } { 2 } } } ." + }, + { + "category_id": 13, + "poly": [ + 711, + 286, + 1030, + 286, + 1030, + 329, + 711, + 329 + ], + "score": 0.94, + "latex": "x _ { i } ^ { ( 1 ) } = e _ { 1 } - q _ { i } ^ { ( 1 ) } \\tau e _ { 2 } + \\rho \\xi _ { i } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 1252, + 287, + 1398, + 287, + 1398, + 329, + 1252, + 329 + ], + "score": 0.93, + "latex": "q _ { i } ^ { ( 1 ) } \\in \\{ 0 , 1 \\}" + }, + { + "category_id": 14, + "poly": [ + 652, + 933, + 1047, + 933, + 1047, + 994, + 652, + 994 + ], + "score": 0.93, + "latex": "\\operatorname* { P r } \\left( | \\| \\xi \\| _ { 2 } ^ { 2 } - d | \\geq 4 d ^ { \\frac { 3 } { 4 } } \\right) \\leq 2 e ^ { - \\sqrt { d } } ." + }, + { + "category_id": 13, + "poly": [ + 1085, + 1598, + 1236, + 1598, + 1236, + 1637, + 1085, + 1637 + ], + "score": 0.93, + "latex": "1 - O ( d ^ { - { \\frac { 1 } { 1 0 } } } )" + }, + { + "category_id": 13, + "poly": [ + 396, + 1634, + 599, + 1634, + 599, + 1674, + 396, + 1674 + ], + "score": 0.93, + "latex": "\\lVert \\boldsymbol { w } _ { 3 } ^ { * } \\rVert _ { 2 } ^ { 2 } \\leq O ( d ^ { - \\frac { 1 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 586, + 742, + 760, + 742, + 760, + 775, + 586, + 775 + ], + "score": 0.93, + "latex": "\\xi , \\xi ^ { \\prime } \\sim \\mathcal { N } ( 0 , I )" + }, + { + "category_id": 13, + "poly": [ + 748, + 412, + 839, + 412, + 839, + 447, + 748, + 447 + ], + "score": 0.93, + "latex": "\\mathcal { N } ( 0 , I )" + }, + { + "category_id": 13, + "poly": [ + 810, + 1603, + 964, + 1603, + 964, + 1637, + 810, + 1637 + ], + "score": 0.93, + "latex": "\\{ w _ { 1 } ^ { * } , w _ { 2 } ^ { * } , w _ { 3 } ^ { * } \\}" + }, + { + "category_id": 14, + "poly": [ + 401, + 1919, + 1299, + 1919, + 1299, + 2032, + 401, + 2032 + ], + "score": 0.93, + "latex": "{ w _ { 3 } ^ { * } } ^ { \\top } x = \\frac { 1 } { \\rho d } \\left( e _ { 1 } - q _ { i } ^ { ( 1 ) } \\tau e _ { 2 } + \\rho \\xi _ { i } ^ { ( 1 ) } \\right) ^ { \\top } \\left( \\sum _ { j = 1 } ^ { n _ { 3 } } \\xi _ { j } ^ { ( 3 ) } \\right) \\leq \\frac { n _ { 3 } ( \\tau + 1 ) } { \\rho d } d ^ { \\frac { 1 } { 1 0 } } + \\frac { 3 n _ { 3 } } { d } d ^ { \\frac { 3 } { 5 } } ." + }, + { + "category_id": 13, + "poly": [ + 295, + 560, + 413, + 560, + 413, + 603, + 295, + 603 + ], + "score": 0.92, + "latex": "1 - e ^ { - d ^ { \\frac { 1 } { 1 0 } } }" + }, + { + "category_id": 13, + "poly": [ + 1169, + 373, + 1260, + 373, + 1260, + 408, + 1169, + 408 + ], + "score": 0.92, + "latex": "i \\in [ n _ { 3 } ]" + }, + { + "category_id": 13, + "poly": [ + 1263, + 333, + 1354, + 333, + 1354, + 368, + 1263, + 368 + ], + "score": 0.92, + "latex": "i \\in [ n _ { 2 } ]" + }, + { + "category_id": 13, + "poly": [ + 612, + 1328, + 788, + 1328, + 788, + 1369, + 612, + 1369 + ], + "score": 0.92, + "latex": "| \\langle \\xi _ { i } , \\xi _ { j } \\rangle | \\leq 3 d ^ { \\frac { 3 } { 5 } }" + }, + { + "category_id": 13, + "poly": [ + 888, + 369, + 1089, + 369, + 1089, + 408, + 888, + 408 + ], + "score": 0.92, + "latex": "x _ { i } ^ { ( 3 ) } = e _ { 2 } + \\rho \\xi _ { i } ^ { ( 3 ) }" + }, + { + "category_id": 13, + "poly": [ + 504, + 529, + 658, + 529, + 658, + 564, + 504, + 564 + ], + "score": 0.92, + "latex": "\\xi _ { i } \\sim \\mathcal { N } ( 0 , I )" + }, + { + "category_id": 13, + "poly": [ + 298, + 366, + 444, + 366, + 444, + 408, + 298, + 408 + ], + "score": 0.92, + "latex": "q _ { i } ^ { ( 1 ) } \\in \\{ 0 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 423, + 1334, + 501, + 1334, + 501, + 1368, + 423, + 1368 + ], + "score": 0.92, + "latex": "i \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 846, + 328, + 1186, + 328, + 1186, + 369, + 846, + 369 + ], + "score": 0.92, + "latex": "x _ { i } ^ { ( 2 ) } = - e _ { 1 } - q _ { i } ^ { ( 2 ) } \\tau e _ { 2 } + \\rho \\xi _ { i } ^ { ( 2 ) }" + }, + { + "category_id": 13, + "poly": [ + 606, + 1558, + 653, + 1558, + 653, + 1601, + 606, + 1601 + ], + "score": 0.91, + "latex": "\\xi _ { i } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 867, + 1566, + 959, + 1566, + 959, + 1600, + 867, + 1600 + ], + "score": 0.91, + "latex": "i \\in [ n _ { k } ]" + }, + { + "category_id": 13, + "poly": [ + 332, + 408, + 379, + 408, + 379, + 448, + 332, + 448 + ], + "score": 0.91, + "latex": "\\xi _ { i } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 1091, + 1514, + 1314, + 1514, + 1314, + 1562, + 1091, + 1562 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { w _ { 3 } ^ { * } = \\frac { 1 } { \\rho d } \\sum _ { i = 1 } ^ { n _ { 3 } } \\xi _ { i } ^ { ( 3 ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 777, + 1228, + 840, + 1228, + 840, + 1260, + 777, + 1260 + ], + "score": 0.91, + "latex": "i \\neq j" + }, + { + "category_id": 13, + "poly": [ + 336, + 670, + 513, + 670, + 513, + 711, + 336, + 711 + ], + "score": 0.91, + "latex": "| \\langle \\xi _ { i } , \\xi _ { j } \\rangle | \\leq 3 d ^ { \\frac { 3 } { 5 } }" + }, + { + "category_id": 14, + "poly": [ + 310, + 1073, + 1393, + 1073, + 1393, + 1171, + 310, + 1171 + ], + "score": 0.91, + "latex": "\\operatorname* { P r } \\left( | \\langle \\xi , \\xi ^ { \\prime } \\rangle | \\ge 3 d ^ { \\frac 3 5 } \\right) \\le \\operatorname* { P r } \\left( \\| \\xi \\| _ { 2 } \\ge \\sqrt { d } + 2 d ^ { \\frac 3 8 } \\right) + \\operatorname* { P r } \\left( | \\langle \\xi ^ { \\prime } , \\frac { \\xi } { \\| \\xi \\| } \\rangle | \\ge d ^ { \\frac 1 { 1 0 } } \\right) \\le \\frac { e ^ { - \\frac { d ^ { \\frac { 1 } { 2 } } } { 2 } } } { d ^ { \\frac 1 { 1 0 } } } + 2 e ^ { - \\sqrt { d } } ." + }, + { + "category_id": 13, + "poly": [ + 526, + 890, + 560, + 890, + 560, + 926, + 526, + 926 + ], + "score": 0.91, + "latex": "\\chi _ { d } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1108, + 293, + 1200, + 293, + 1200, + 328, + 1108, + 328 + ], + "score": 0.9, + "latex": "i \\in [ n _ { 1 } ]" + }, + { + "category_id": 13, + "poly": [ + 970, + 1366, + 1040, + 1366, + 1040, + 1407, + 970, + 1407 + ], + "score": 0.9, + "latex": "e ^ { - d ^ { \\frac { 1 } { 1 0 } } }" + }, + { + "category_id": 13, + "poly": [ + 966, + 531, + 1111, + 531, + 1111, + 564, + 966, + 564 + ], + "score": 0.89, + "latex": "n \\leq { \\mathrm { p o l y } } ( d )" + }, + { + "category_id": 13, + "poly": [ + 754, + 613, + 960, + 613, + 960, + 653, + 754, + 653 + ], + "score": 0.89, + "latex": "| \\| \\xi _ { i } \\| _ { 2 } ^ { 2 } - d | \\le 4 d ^ { \\frac { 3 } { 4 } }" + }, + { + "category_id": 13, + "poly": [ + 869, + 1334, + 931, + 1334, + 931, + 1367, + 869, + 1367 + ], + "score": 0.89, + "latex": "i \\neq j" + }, + { + "category_id": 13, + "poly": [ + 296, + 1328, + 343, + 1328, + 343, + 1363, + 296, + 1363 + ], + "score": 0.89, + "latex": "4 d ^ { \\frac { 3 } { 4 } }" + }, + { + "category_id": 13, + "poly": [ + 704, + 530, + 786, + 530, + 786, + 564, + 704, + 564 + ], + "score": 0.89, + "latex": "i \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 1249, + 1282, + 1405, + 1282, + 1405, + 1321, + 1249, + 1321 + ], + "score": 0.88, + "latex": "| | \\xi _ { i } | | _ { 2 } ^ { 2 } - d | \\leq" + }, + { + "category_id": 13, + "poly": [ + 503, + 1229, + 530, + 1229, + 530, + 1261, + 503, + 1261 + ], + "score": 0.88, + "latex": "\\xi _ { j }" + }, + { + "category_id": 13, + "poly": [ + 1042, + 617, + 1119, + 617, + 1119, + 652, + 1042, + 652 + ], + "score": 0.88, + "latex": "i \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 594, + 675, + 656, + 675, + 656, + 710, + 594, + 710 + ], + "score": 0.88, + "latex": "i \\neq j" + }, + { + "category_id": 13, + "poly": [ + 840, + 1523, + 942, + 1523, + 942, + 1555, + 840, + 1555 + ], + "score": 0.87, + "latex": "w _ { 1 } ^ { * } = e _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 430, + 1229, + 454, + 1229, + 454, + 1259, + 430, + 1259 + ], + "score": 0.87, + "latex": "\\xi _ { i }" + }, + { + "category_id": 13, + "poly": [ + 554, + 1035, + 579, + 1035, + 579, + 1066, + 554, + 1066 + ], + "score": 0.87, + "latex": "\\xi ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 622, + 1004, + 647, + 1004, + 647, + 1035, + 622, + 1035 + ], + "score": 0.87, + "latex": "\\xi ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 335, + 613, + 509, + 613, + 509, + 653, + 335, + 653 + ], + "score": 0.86, + "latex": "| \\langle \\xi _ { i } , e _ { 1 } \\rangle | \\leq d ^ { \\frac { 1 } { 1 0 } }" + }, + { + "category_id": 13, + "poly": [ + 955, + 1521, + 1077, + 1521, + 1077, + 1556, + 955, + 1556 + ], + "score": 0.86, + "latex": "w _ { 2 } ^ { \\ast } = - e _ { 1 }" + }, + { + "category_id": 14, + "poly": [ + 616, + 1852, + 1081, + 1852, + 1081, + 1899, + 616, + 1899 + ], + "score": 0.84, + "latex": "\\begin{array} { r } { w _ { 2 } ^ { * ^ { \\top } } x = - 1 + \\langle \\xi _ { i } ^ { ( 1 ) } , e _ { 1 } \\rangle \\rho \\leq - 1 + \\rho d ^ { \\frac { 1 } { 1 0 } } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 523, + 613, + 698, + 613, + 698, + 653, + 523, + 653 + ], + "score": 0.84, + "latex": "| \\langle \\xi _ { i } , e _ { 2 } \\rangle | \\leq d ^ { \\frac { 1 } { 1 0 } }" + }, + { + "category_id": 13, + "poly": [ + 638, + 1036, + 655, + 1036, + 655, + 1066, + 638, + 1066 + ], + "score": 0.83, + "latex": "\\xi" + }, + { + "category_id": 13, + "poly": [ + 555, + 1005, + 572, + 1005, + 572, + 1035, + 555, + 1035 + ], + "score": 0.83, + "latex": "\\xi" + }, + { + "category_id": 14, + "poly": [ + 636, + 1785, + 1060, + 1785, + 1060, + 1832, + 636, + 1832 + ], + "score": 0.83, + "latex": "w _ { 1 } ^ { * } { } ^ { \\top } x = 1 + \\langle \\xi _ { i } ^ { ( 1 ) } , e _ { 1 } \\rangle \\rho \\geq 1 - \\rho d ^ { \\frac { 1 } { 1 0 } } ," + }, + { + "category_id": 13, + "poly": [ + 1384, + 1005, + 1401, + 1005, + 1401, + 1036, + 1384, + 1036 + ], + "score": 0.81, + "latex": "\\xi" + }, + { + "category_id": 13, + "poly": [ + 1023, + 1281, + 1197, + 1281, + 1197, + 1320, + 1023, + 1320 + ], + "score": 0.8, + "latex": "| \\langle \\xi _ { i } , e _ { 2 } \\rangle | \\leq d ^ { \\frac { 1 } { 1 0 } }" + }, + { + "category_id": 13, + "poly": [ + 326, + 1380, + 344, + 1380, + 344, + 1406, + 326, + 1406 + ], + "score": 0.79, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 544, + 1380, + 561, + 1380, + 561, + 1406, + 544, + 1406 + ], + "score": 0.79, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 836, + 1281, + 1009, + 1281, + 1009, + 1320, + 836, + 1320 + ], + "score": 0.74, + "latex": "| \\langle \\xi _ { i } , e _ { 1 } \\rangle | \\leq d ^ { \\frac { 1 } { 1 0 } }" + }, + { + "category_id": 13, + "poly": [ + 621, + 575, + 640, + 575, + 640, + 601, + 621, + 601 + ], + "score": 0.6, + "latex": "d _ { \\mathrm { { z } } }" + }, + { + "category_id": 13, + "poly": [ + 732, + 1564, + 814, + 1564, + 814, + 1600, + 732, + 1600 + ], + "score": 0.59, + "latex": "k \\in [ 3 ]" + }, + { + "category_id": 13, + "poly": [ + 482, + 1707, + 523, + 1707, + 523, + 1735, + 482, + 1735 + ], + "score": 0.42, + "latex": "E . 2" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 224.0, + 706.0, + 224.0, + 706.0, + 265.0, + 295.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1223.0, + 429.0, + 1223.0, + 429.0, + 1263.0, + 296.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1223.0, + 502.0, + 1223.0, + 502.0, + 1263.0, + 455.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1223.0, + 776.0, + 1223.0, + 776.0, + 1263.0, + 531.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 1223.0, + 1404.0, + 1223.0, + 1404.0, + 1263.0, + 841.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1283.0, + 379.0, + 1283.0, + 379.0, + 1324.0, + 295.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1259.0, + 835.0, + 1259.0, + 835.0, + 1325.0, + 730.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1259.0, + 1022.0, + 1259.0, + 1022.0, + 1325.0, + 1010.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1259.0, + 1248.0, + 1259.0, + 1248.0, + 1325.0, + 1198.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1328.0, + 422.0, + 1328.0, + 422.0, + 1372.0, + 344.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 1328.0, + 611.0, + 1328.0, + 611.0, + 1372.0, + 502.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 1328.0, + 868.0, + 1328.0, + 868.0, + 1372.0, + 789.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 1328.0, + 1406.0, + 1328.0, + 1406.0, + 1372.0, + 932.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1360.0, + 325.0, + 1360.0, + 325.0, + 1421.0, + 288.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1360.0, + 543.0, + 1360.0, + 543.0, + 1421.0, + 345.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1360.0, + 969.0, + 1360.0, + 969.0, + 1421.0, + 562.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 1360.0, + 1413.0, + 1360.0, + 1413.0, + 1421.0, + 1041.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.25, + 1298.5, + 607.25, + 1298.5, + 607.25, + 1335.5, + 550.25, + 1335.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 282.0, + 710.0, + 282.0, + 710.0, + 386.0, + 291.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 282.0, + 1251.0, + 282.0, + 1251.0, + 386.0, + 1201.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 282.0, + 1416.0, + 282.0, + 1416.0, + 386.0, + 1399.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 360.0, + 887.0, + 360.0, + 887.0, + 416.0, + 445.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 366.0, + 1168.0, + 366.0, + 1168.0, + 412.0, + 1090.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 366.0, + 1407.0, + 366.0, + 1407.0, + 412.0, + 1261.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 410.0, + 747.0, + 410.0, + 747.0, + 449.0, + 380.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 410.0, + 849.0, + 410.0, + 849.0, + 449.0, + 840.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 324.5, + 886.0, + 324.5, + 886.0, + 376.0, + 290.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 333.0, + 988.0, + 333.0, + 988.0, + 367.5, + 896.0, + 367.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 357.5, + 1109.0, + 357.5, + 1109.0, + 425.0, + 922.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 398.0, + 398.0, + 398.0, + 398.0, + 462.5, + 278.0, + 462.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 1506.0, + 839.0, + 1506.0, + 839.0, + 1576.0, + 281.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1506.0, + 954.0, + 1506.0, + 954.0, + 1576.0, + 943.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1506.0, + 1090.0, + 1506.0, + 1090.0, + 1576.0, + 1078.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 1506.0, + 1417.0, + 1506.0, + 1417.0, + 1576.0, + 1315.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1550.0, + 605.0, + 1550.0, + 605.0, + 1608.0, + 288.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 1550.0, + 731.0, + 1550.0, + 731.0, + 1608.0, + 654.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 1550.0, + 866.0, + 1550.0, + 866.0, + 1608.0, + 815.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1550.0, + 1412.0, + 1550.0, + 1412.0, + 1608.0, + 960.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1600.0, + 809.0, + 1600.0, + 809.0, + 1642.0, + 293.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1600.0, + 1084.0, + 1600.0, + 1084.0, + 1642.0, + 965.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1600.0, + 1409.0, + 1600.0, + 1409.0, + 1642.0, + 1237.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1636.0, + 395.0, + 1636.0, + 395.0, + 1675.0, + 292.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1636.0, + 610.0, + 1636.0, + 610.0, + 1675.0, + 600.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 997.0, + 554.0, + 997.0, + 554.0, + 1040.0, + 295.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 997.0, + 621.0, + 997.0, + 621.0, + 1040.0, + 573.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 997.0, + 1383.0, + 997.0, + 1383.0, + 1040.0, + 648.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 997.0, + 1405.0, + 997.0, + 1405.0, + 1040.0, + 1402.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1027.0, + 553.0, + 1027.0, + 553.0, + 1072.0, + 293.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1027.0, + 637.0, + 1027.0, + 637.0, + 1072.0, + 580.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1027.0, + 669.0, + 1027.0, + 669.0, + 1072.0, + 656.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1705.0, + 481.0, + 1705.0, + 481.0, + 1739.0, + 295.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1705.0, + 1404.0, + 1705.0, + 1404.0, + 1739.0, + 524.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 1728.0, + 919.0, + 1728.0, + 919.0, + 1795.0, + 281.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1728.0, + 1241.0, + 1728.0, + 1241.0, + 1795.0, + 1022.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1444.0, + 1405.0, + 1444.0, + 1405.0, + 1480.0, + 297.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1476.0, + 991.0, + 1476.0, + 991.0, + 1508.0, + 297.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 739.0, + 585.0, + 739.0, + 585.0, + 779.0, + 294.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 739.0, + 1405.0, + 739.0, + 1405.0, + 779.0, + 761.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 770.0, + 653.0, + 770.0, + 653.0, + 807.0, + 293.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 457.0, + 1404.0, + 457.0, + 1404.0, + 495.0, + 294.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 489.0, + 608.0, + 489.0, + 608.0, + 523.0, + 297.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 886.0, + 525.0, + 886.0, + 525.0, + 930.0, + 294.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 886.0, + 803.0, + 886.0, + 803.0, + 930.0, + 561.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 524.0, + 503.0, + 524.0, + 503.0, + 570.0, + 292.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 524.0, + 703.0, + 524.0, + 703.0, + 570.0, + 659.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 524.0, + 965.0, + 524.0, + 965.0, + 570.0, + 787.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 524.0, + 1405.0, + 524.0, + 1405.0, + 570.0, + 1112.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 557.0, + 294.0, + 557.0, + 294.0, + 614.0, + 289.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 557.0, + 620.0, + 557.0, + 620.0, + 614.0, + 414.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 557.0, + 759.0, + 557.0, + 759.0, + 614.0, + 641.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 606.0, + 334.0, + 606.0, + 334.0, + 660.0, + 306.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 606.0, + 522.0, + 606.0, + 522.0, + 660.0, + 510.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 606.0, + 753.0, + 606.0, + 753.0, + 660.0, + 699.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 606.0, + 1041.0, + 606.0, + 1041.0, + 660.0, + 961.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 606.0, + 1130.0, + 606.0, + 1130.0, + 660.0, + 1120.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 667.0, + 335.0, + 667.0, + 335.0, + 717.0, + 308.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 667.0, + 593.0, + 667.0, + 593.0, + 717.0, + 514.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 667.0, + 667.0, + 667.0, + 667.0, + 717.0, + 657.0, + 717.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1571, + 1408, + 1571, + 1408, + 1772, + 296, + 1772 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1868, + 1406, + 1868, + 1406, + 2036, + 297, + 2036 + ], + "score": 0.978 + }, + { + "category_id": 8, + "poly": [ + 318, + 473, + 1328, + 473, + 1328, + 586, + 318, + 586 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 295, + 1136, + 1402, + 1136, + 1402, + 1203, + 295, + 1203 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 676, + 716, + 1023, + 716, + 1023, + 763, + 676, + 763 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 552, + 1056, + 1147, + 1056, + 1147, + 1122, + 552, + 1122 + ], + "score": 0.937 + }, + { + "category_id": 8, + "poly": [ + 438, + 277, + 1259, + 277, + 1259, + 350, + 438, + 350 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 296, + 891, + 1035, + 891, + 1035, + 940, + 296, + 940 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 298, + 1236, + 631, + 1236, + 631, + 1268, + 298, + 1268 + ], + "score": 0.93 + }, + { + "category_id": 8, + "poly": [ + 645, + 1445, + 1138, + 1445, + 1138, + 1512, + 645, + 1512 + ], + "score": 0.929 + }, + { + "category_id": 8, + "poly": [ + 482, + 815, + 1212, + 815, + 1212, + 864, + 482, + 864 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 297, + 772, + 483, + 772, + 483, + 804, + 297, + 804 + ], + "score": 0.928 + }, + { + "category_id": 8, + "poly": [ + 685, + 641, + 1013, + 641, + 1013, + 690, + 685, + 690 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 296, + 598, + 508, + 598, + 508, + 630, + 296, + 630 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 225, + 672, + 225, + 672, + 265, + 297, + 265 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 296, + 364, + 1072, + 364, + 1072, + 405, + 296, + 405 + ], + "score": 0.916 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 817, + 73, + 817, + 106, + 297, + 106 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 297, + 970, + 817, + 970, + 817, + 1003, + 297, + 1003 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 296, + 1524, + 1105, + 1524, + 1105, + 1561, + 296, + 1561 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 300, + 1009, + 1145, + 1009, + 1145, + 1044, + 300, + 1044 + ], + "score": 0.913 + }, + { + "category_id": 1, + "poly": [ + 297, + 419, + 714, + 419, + 714, + 461, + 297, + 461 + ], + "score": 0.911 + }, + { + "category_id": 8, + "poly": [ + 645, + 1385, + 1018, + 1385, + 1018, + 1438, + 645, + 1438 + ], + "score": 0.91 + }, + { + "category_id": 8, + "poly": [ + 645, + 1326, + 1019, + 1326, + 1019, + 1378, + 645, + 1378 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1073, + 1401, + 1073, + 1401, + 1105, + 1351, + 1105 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1351, + 726, + 1401, + 726, + 1401, + 757, + 1351, + 757 + ], + "score": 0.887 + }, + { + "category_id": 1, + "poly": [ + 300, + 1802, + 1100, + 1802, + 1100, + 1838, + 300, + 1838 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1350, + 298, + 1401, + 298, + 1401, + 330, + 1350, + 330 + ], + "score": 0.886 + }, + { + "category_id": 9, + "poly": [ + 1351, + 652, + 1401, + 652, + 1401, + 683, + 1351, + 683 + ], + "score": 0.883 + }, + { + "category_id": 9, + "poly": [ + 1351, + 826, + 1401, + 826, + 1401, + 857, + 1351, + 857 + ], + "score": 0.883 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1461, + 1401, + 1461, + 1401, + 1493, + 1351, + 1493 + ], + "score": 0.88 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1287, + 1401, + 1287, + 1401, + 1318, + 1351, + 1318 + ], + "score": 0.879 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1332, + 1401, + 1332, + 1401, + 1364, + 1351, + 1364 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1391, + 1401, + 1391, + 1401, + 1422, + 1351, + 1422 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1352, + 517, + 1401, + 517, + 1401, + 549, + 1352, + 549 + ], + "score": 0.859 + }, + { + "category_id": 2, + "poly": [ + 1374, + 900, + 1403, + 900, + 1403, + 929, + 1374, + 929 + ], + "score": 0.815 + }, + { + "category_id": 8, + "poly": [ + 561, + 1280, + 813, + 1280, + 813, + 1321, + 561, + 1321 + ], + "score": 0.52 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 863, + 2088, + 863, + 2113, + 834, + 2113 + ], + "score": 0.503 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 863, + 2087, + 863, + 2113, + 834, + 2113 + ], + "score": 0.337 + }, + { + "category_id": 8, + "poly": [ + 561, + 1280, + 813, + 1280, + 813, + 1322, + 561, + 1322 + ], + "score": 0.283 + }, + { + "category_id": 8, + "poly": [ + 561, + 1280, + 1019, + 1280, + 1019, + 1381, + 561, + 1381 + ], + "score": 0.11 + }, + { + "category_id": 13, + "poly": [ + 535, + 892, + 826, + 892, + 826, + 941, + 535, + 941 + ], + "score": 0.95, + "latex": "\\begin{array} { r } { \\| w _ { 3 } ^ { * } \\| _ { 2 } \\leq \\frac { 2 n _ { 3 } \\sqrt { d } } { \\rho d } \\leq 2 d ^ { - \\frac { 1 } { 1 0 } } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 558, + 1276, + 1138, + 1276, + 1138, + 1515, + 558, + 1515 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\| \\boldsymbol { W } \\| _ { F } ^ { 2 } = T r ( \\boldsymbol { W } \\boldsymbol { W } ^ { \\top } ) } \\\\ & { \\qquad = \\underset { W _ { 2 } W _ { 1 } = W } { \\operatorname* { m i n } } T r ( \\boldsymbol { W } _ { 2 } \\boldsymbol { W } _ { 1 } \\boldsymbol { W } _ { 1 } ^ { \\top } \\boldsymbol { W } _ { 2 } ^ { \\top } ) } \\\\ & { \\qquad = \\underset { W _ { 2 } W _ { 1 } = W } { \\operatorname* { m i n } } T r ( \\boldsymbol { W } _ { 1 } \\boldsymbol { W } _ { 1 } ^ { \\top } \\boldsymbol { W } _ { 2 } ^ { \\top } \\boldsymbol { W } _ { 2 } ) } \\\\ & { \\qquad \\le \\frac { 1 } { 2 } \\underset { W _ { 2 } W _ { 1 } = W } { \\operatorname* { m i n } } ( \\| \\boldsymbol { W } _ { 1 } \\boldsymbol { W } _ { 1 } ^ { \\top } \\| _ { F } ^ { 2 } + \\| \\boldsymbol { W } _ { 2 } ^ { \\top } \\boldsymbol { W } _ { 2 } \\| _ { F } ^ { 2 } ) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 881, + 1524, + 1094, + 1524, + 1094, + 1562, + 881, + 1562 + ], + "score": 0.94, + "latex": "W _ { 1 } W _ { 1 } ^ { \\top } = W _ { 2 } ^ { \\top } W _ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 437, + 276, + 1261, + 276, + 1261, + 351, + 437, + 351 + ], + "score": 0.94, + "latex": "\\boldsymbol { w _ { 1 } ^ { * } } ^ { \\top } \\boldsymbol { x } - \\boldsymbol { w _ { 3 } ^ { * } } ^ { \\top } \\boldsymbol { x } \\ge 1 - \\rho d ^ { \\frac { 1 } { 1 0 } } - \\frac { n _ { 3 } ( \\tau + 1 ) } { \\rho d } d ^ { \\frac { 1 } { 1 0 } } - \\frac { 3 n _ { 3 } } { d } d ^ { \\frac { 3 } { 5 } } \\ge 1 - O ( d ^ { - \\frac { 1 } { 1 0 } } ) ." + }, + { + "category_id": 13, + "poly": [ + 1183, + 1651, + 1353, + 1651, + 1353, + 1699, + 1183, + 1699 + ], + "score": 0.94, + "latex": "W _ { 1 } = \\Sigma _ { m \\times d } ^ { \\frac { 1 } { 2 } } V" + }, + { + "category_id": 14, + "poly": [ + 320, + 470, + 1333, + 470, + 1333, + 589, + 320, + 589 + ], + "score": 0.93, + "latex": "{ w _ { 3 } ^ { * } } ^ { \\top } \\boldsymbol { x } = \\frac { 1 } { \\rho d } \\left( \\sum _ { j = 1 } ^ { n _ { 3 } } \\xi _ { j } ^ { ( 3 ) } \\right) ^ { \\top } \\left( e _ { 2 } + \\rho \\xi _ { i } ^ { ( 3 ) } \\right) \\ge \\frac { 1 } { d } \\| \\xi _ { i } ^ { ( 3 ) } \\| _ { 2 } ^ { 2 } - \\frac { 3 n _ { 3 } } { d } d ^ { \\frac { 3 } { 5 } } - \\frac { n _ { 3 } d ^ { \\frac { 1 } { 1 0 } } } { \\rho d } \\ge 1 - O ( d ^ { - \\frac { 1 } { 5 } } ) ." + }, + { + "category_id": 13, + "poly": [ + 1014, + 1899, + 1212, + 1899, + 1212, + 1939, + 1014, + 1939 + ], + "score": 0.93, + "latex": "\\tilde { w } _ { y } ^ { \\top } x \\ge w _ { y ^ { \\prime } } ^ { \\top } x + 1" + }, + { + "category_id": 13, + "poly": [ + 1044, + 1604, + 1108, + 1604, + 1108, + 1655, + 1044, + 1655 + ], + "score": 0.93, + "latex": "\\Sigma _ { p \\times q } ^ { \\frac { 1 } { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 585, + 1136, + 735, + 1136, + 735, + 1170, + 585, + 1170 + ], + "score": 0.93, + "latex": "W _ { 2 } \\in \\mathbb { R } ^ { 3 \\times m }" + }, + { + "category_id": 13, + "poly": [ + 1136, + 1571, + 1254, + 1571, + 1254, + 1602, + 1136, + 1602 + ], + "score": 0.92, + "latex": "\\Sigma \\in \\mathbb { R } ^ { 3 \\times d }" + }, + { + "category_id": 13, + "poly": [ + 548, + 224, + 642, + 224, + 642, + 266, + 548, + 266 + ], + "score": 0.92, + "latex": "( x _ { i } ^ { ( 1 ) } , 1 )" + }, + { + "category_id": 13, + "poly": [ + 910, + 365, + 1061, + 365, + 1061, + 405, + 910, + 405 + ], + "score": 0.92, + "latex": "1 - O ( d ^ { - { \\frac { 1 } { 1 0 } } } )" + }, + { + "category_id": 13, + "poly": [ + 361, + 1900, + 647, + 1900, + 647, + 1935, + 361, + 1935 + ], + "score": 0.92, + "latex": "\\| w _ { 1 } \\| _ { 2 } ^ { 2 } + \\| w _ { 2 } \\| _ { 2 } ^ { 2 } + \\| w _ { 3 } \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 395, + 418, + 496, + 418, + 496, + 461, + 395, + 461 + ], + "score": 0.92, + "latex": "x = x _ { i } ^ { ( 3 ) }" + }, + { + "category_id": 13, + "poly": [ + 374, + 1136, + 527, + 1136, + 527, + 1170, + 374, + 1170 + ], + "score": 0.92, + "latex": "W _ { 1 } \\in \\mathbb { R } ^ { m \\times d }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1698, + 466, + 1698, + 466, + 1743, + 297, + 1743 + ], + "score": 0.92, + "latex": "W _ { 2 } = U \\Sigma _ { 3 \\times m } ^ { \\frac { 1 } { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 955, + 1705, + 1397, + 1705, + 1397, + 1744, + 955, + 1744 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\| \\boldsymbol { W } \\| _ { F } ^ { 2 } = \\frac { 1 } { 2 } ( \\| \\boldsymbol { W } _ { 1 } \\boldsymbol { W } _ { 1 } ^ { \\top } \\| _ { F } ^ { 2 } + \\| \\boldsymbol { W } _ { 2 } ^ { \\top } \\boldsymbol { W } _ { 2 } \\| _ { F } ^ { 2 } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 759, + 1708, + 905, + 1708, + 905, + 1740, + 759, + 1740 + ], + "score": 0.92, + "latex": "W = W _ { 2 } W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1168, + 386, + 1168, + 386, + 1203, + 298, + 1203 + ], + "score": 0.92, + "latex": "W _ { 2 } ^ { \\top } W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 505, + 363, + 553, + 363, + 553, + 405, + 505, + 405 + ], + "score": 0.91, + "latex": "x _ { i } ^ { ( 2 ) }" + }, + { + "category_id": 14, + "poly": [ + 552, + 1054, + 1143, + 1054, + 1143, + 1123, + 552, + 1123 + ], + "score": 0.91, + "latex": "\\| \\boldsymbol { W } \\| _ { F } ^ { 2 } = \\frac { 1 } { 2 } \\operatorname* { m i n } _ { W _ { 2 } W _ { 1 } = W } \\left( \\| \\boldsymbol { W } _ { 2 } ^ { \\top } \\boldsymbol { W } _ { 2 } \\| _ { F } ^ { 2 } + \\| \\boldsymbol { W } _ { 1 } \\boldsymbol { W } _ { 1 } ^ { \\top } \\| _ { F } ^ { 2 } \\right) ," + }, + { + "category_id": 13, + "poly": [ + 896, + 1870, + 1047, + 1870, + 1047, + 1903, + 896, + 1903 + ], + "score": 0.91, + "latex": "\\{ \\hat { w } _ { 1 } , \\hat { w } _ { 2 } , \\hat { w } _ { 3 } \\}" + }, + { + "category_id": 13, + "poly": [ + 858, + 1617, + 952, + 1617, + 952, + 1649, + 858, + 1649 + ], + "score": 0.91, + "latex": "p , q \\geq 3" + }, + { + "category_id": 13, + "poly": [ + 354, + 1938, + 419, + 1938, + 419, + 1970, + 354, + 1970 + ], + "score": 0.91, + "latex": "( x , y )" + }, + { + "category_id": 13, + "poly": [ + 1278, + 1136, + 1404, + 1136, + 1404, + 1174, + 1278, + 1174 + ], + "score": 0.91, + "latex": "W _ { 1 } W _ { 1 } ^ { \\top } =" + }, + { + "category_id": 13, + "poly": [ + 497, + 1008, + 626, + 1008, + 626, + 1040, + 497, + 1040 + ], + "score": 0.91, + "latex": "W \\in \\mathbb { R } ^ { 3 \\times d }" + }, + { + "category_id": 13, + "poly": [ + 472, + 1937, + 553, + 1937, + 553, + 1969, + 472, + 1969 + ], + "score": 0.9, + "latex": "y ^ { \\prime } \\ne y" + }, + { + "category_id": 13, + "poly": [ + 891, + 1013, + 970, + 1013, + 970, + 1042, + 891, + 1042 + ], + "score": 0.9, + "latex": "m \\geq 3 ." + }, + { + "category_id": 13, + "poly": [ + 544, + 1575, + 679, + 1575, + 679, + 1603, + 544, + 1603 + ], + "score": 0.9, + "latex": "W = U \\Sigma V" + }, + { + "category_id": 13, + "poly": [ + 430, + 1622, + 543, + 1622, + 543, + 1649, + 430, + 1649 + ], + "score": 0.89, + "latex": "\\sigma _ { 1 } , \\sigma _ { 2 } , \\sigma _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1259, + 1619, + 1324, + 1619, + 1324, + 1649, + 1259, + 1649 + ], + "score": 0.89, + "latex": "p \\times q" + }, + { + "category_id": 13, + "poly": [ + 355, + 1662, + 535, + 1662, + 535, + 1698, + 355, + 1698 + ], + "score": 0.88, + "latex": "{ \\sqrt { \\sigma _ { 1 } } } , { \\sqrt { \\sigma _ { 2 } } } , { \\sqrt { \\sigma _ { 3 } } }" + }, + { + "category_id": 13, + "poly": [ + 347, + 1968, + 867, + 1968, + 867, + 2006, + 347, + 2006 + ], + "score": 0.88, + "latex": "\\langle \\hat { w } _ { 1 } , e _ { 1 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 2 } , e _ { 1 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 3 } , e _ { 1 } \\rangle ^ { 2 } \\leq O ( d ^ { - \\frac { 1 } { 1 0 } } )" + }, + { + "category_id": 13, + "poly": [ + 1256, + 1868, + 1292, + 1868, + 1292, + 1898, + 1256, + 1898 + ], + "score": 0.88, + "latex": "\\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 713, + 2006, + 765, + 2006, + 765, + 2035, + 713, + 2035 + ], + "score": 0.87, + "latex": "W _ { \\mathrm { S L } }" + }, + { + "category_id": 14, + "poly": [ + 486, + 815, + 1212, + 815, + 1212, + 861, + 486, + 861 + ], + "score": 0.87, + "latex": "\\begin{array} { r } { w _ { 3 } ^ { * ^ { \\top } } x - \\operatorname* { m a x } \\{ { w _ { 1 } ^ { * ^ { \\top } } } x , { w _ { 2 } ^ { * ^ { \\top } } } x \\} \\geq { w _ { 3 } ^ { * ^ { \\top } } } x - \\rho d ^ { \\frac { 1 } { 1 0 } } \\geq 1 - O ( d ^ { - \\frac { 1 } { 1 0 } } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 633, + 2009, + 661, + 2009, + 661, + 2034, + 633, + 2034 + ], + "score": 0.84, + "latex": "e _ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 683, + 640, + 1012, + 640, + 1012, + 687, + 683, + 687 + ], + "score": 0.8, + "latex": "{ w _ { 1 } ^ { * } } ^ { \\top } x = \\langle \\rho \\xi _ { i } ^ { ( 3 ) } , e _ { 1 } \\rangle \\le \\rho d ^ { \\frac { 1 } { 1 0 } } ," + }, + { + "category_id": 13, + "poly": [ + 1018, + 1575, + 1050, + 1575, + 1050, + 1602, + 1018, + 1602 + ], + "score": 0.77, + "latex": "W" + }, + { + "category_id": 14, + "poly": [ + 674, + 714, + 1025, + 714, + 1025, + 761, + 674, + 761 + ], + "score": 0.75, + "latex": "{ w _ { 2 } ^ { * } } ^ { \\top } x = \\langle \\rho \\xi _ { i } ^ { ( 3 ) } , - e _ { 1 } \\rangle \\le \\rho d ^ { \\frac { 1 } { 1 0 } } ." + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 905.0, + 1402.0, + 905.0, + 1402.0, + 928.0, + 1379.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 868.0, + 2084.0, + 868.0, + 2125.0, + 830.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1569.0, + 543.0, + 1569.0, + 543.0, + 1607.0, + 294.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1569.0, + 1017.0, + 1569.0, + 1017.0, + 1607.0, + 680.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1569.0, + 1135.0, + 1569.0, + 1135.0, + 1607.0, + 1051.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1569.0, + 1405.0, + 1569.0, + 1405.0, + 1607.0, + 1255.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1615.0, + 1258.0, + 1615.0, + 1258.0, + 1649.0, + 1109.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1615.0, + 1404.0, + 1615.0, + 1404.0, + 1649.0, + 1325.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1689.0, + 296.0, + 1689.0, + 296.0, + 1756.0, + 288.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1689.0, + 758.0, + 1689.0, + 758.0, + 1756.0, + 467.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 1689.0, + 954.0, + 1689.0, + 954.0, + 1756.0, + 906.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1689.0, + 1416.0, + 1689.0, + 1416.0, + 1756.0, + 1398.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1737.0, + 1081.0, + 1737.0, + 1081.0, + 1771.0, + 295.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1740.0, + 1404.0, + 1740.0, + 1404.0, + 1768.0, + 1376.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1866.0, + 895.0, + 1866.0, + 895.0, + 1905.0, + 294.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 1866.0, + 1255.0, + 1866.0, + 1255.0, + 1905.0, + 1048.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 1866.0, + 1408.0, + 1866.0, + 1408.0, + 1905.0, + 1293.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1892.0, + 360.0, + 1892.0, + 360.0, + 1952.0, + 286.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 1892.0, + 1013.0, + 1892.0, + 1013.0, + 1952.0, + 648.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1892.0, + 1414.0, + 1892.0, + 1414.0, + 1952.0, + 1213.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1933.0, + 353.0, + 1933.0, + 353.0, + 1971.0, + 292.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 1933.0, + 471.0, + 1933.0, + 471.0, + 1971.0, + 420.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 1933.0, + 1406.0, + 1933.0, + 1406.0, + 1971.0, + 554.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1970.0, + 346.0, + 1970.0, + 346.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 1970.0, + 1406.0, + 1970.0, + 1406.0, + 2007.0, + 868.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2000.0, + 632.0, + 2000.0, + 632.0, + 2039.0, + 292.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 2000.0, + 712.0, + 2000.0, + 712.0, + 2039.0, + 662.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 2000.0, + 776.0, + 2000.0, + 776.0, + 2039.0, + 766.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1129.0, + 373.0, + 1129.0, + 373.0, + 1176.0, + 292.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 1129.0, + 584.0, + 1129.0, + 584.0, + 1176.0, + 528.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1129.0, + 1277.0, + 1129.0, + 1277.0, + 1176.0, + 736.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1160.0, + 297.0, + 1160.0, + 297.0, + 1212.0, + 291.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 1160.0, + 405.0, + 1160.0, + 405.0, + 1212.0, + 387.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 883.0, + 534.0, + 883.0, + 534.0, + 942.0, + 286.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 883.0, + 1046.0, + 883.0, + 1046.0, + 942.0, + 827.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1234.0, + 632.0, + 1234.0, + 632.0, + 1271.0, + 296.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 768.0, + 486.0, + 768.0, + 486.0, + 808.0, + 295.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 596.0, + 510.0, + 596.0, + 510.0, + 633.0, + 296.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 221.0, + 547.0, + 221.0, + 547.0, + 272.0, + 293.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 221.0, + 676.0, + 221.0, + 676.0, + 272.0, + 643.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 355.0, + 504.0, + 355.0, + 504.0, + 418.0, + 292.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 355.0, + 909.0, + 355.0, + 909.0, + 418.0, + 554.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1062.0, + 355.0, + 1072.0, + 355.0, + 1072.0, + 418.0, + 1062.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 966.0, + 823.0, + 966.0, + 823.0, + 1009.0, + 294.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1518.0, + 880.0, + 1518.0, + 880.0, + 1569.0, + 292.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1518.0, + 1106.0, + 1518.0, + 1106.0, + 1569.0, + 1095.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1005.0, + 496.0, + 1005.0, + 496.0, + 1049.0, + 294.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 1005.0, + 890.0, + 1005.0, + 890.0, + 1049.0, + 627.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1005.0, + 1150.0, + 1005.0, + 1150.0, + 1049.0, + 971.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 417.0, + 717.0, + 417.0, + 717.0, + 465.0, + 497.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 420.0, + 485.0, + 420.0, + 485.0, + 466.5, + 292.0, + 466.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1801.0, + 1103.0, + 1801.0, + 1103.0, + 1840.0, + 293.0, + 1840.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 1690, + 1405, + 1690, + 1405, + 1947, + 295, + 1947 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 477, + 1406, + 477, + 1406, + 593, + 296, + 593 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 296, + 223, + 1407, + 223, + 1407, + 372, + 296, + 372 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 678, + 1405, + 678, + 1405, + 795, + 297, + 795 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 976, + 1405, + 976, + 1405, + 1046, + 298, + 1046 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 652, + 604, + 1049, + 604, + 1049, + 664, + 652, + 664 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 296, + 1959, + 1403, + 1959, + 1403, + 2040, + 296, + 2040 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 646, + 907, + 1051, + 907, + 1051, + 952, + 646, + 952 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 296, + 1134, + 1405, + 1134, + 1405, + 1206, + 296, + 1206 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 399, + 1060, + 1296, + 1060, + 1296, + 1122, + 399, + 1122 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 485, + 382, + 1212, + 382, + 1212, + 455, + 485, + 455 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 299, + 861, + 773, + 861, + 773, + 894, + 299, + 894 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 296, + 1427, + 426, + 1427, + 426, + 1458, + 296, + 1458 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 549, + 1258, + 1156, + 1258, + 1156, + 1320, + 549, + 1320 + ], + "score": 0.931 + }, + { + "category_id": 8, + "poly": [ + 514, + 808, + 1182, + 808, + 1182, + 852, + 514, + 852 + ], + "score": 0.929 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 816, + 73, + 816, + 106, + 298, + 106 + ], + "score": 0.919 + }, + { + "category_id": 8, + "poly": [ + 548, + 1212, + 929, + 1212, + 929, + 1253, + 548, + 1253 + ], + "score": 0.918 + }, + { + "category_id": 8, + "poly": [ + 481, + 1467, + 865, + 1467, + 865, + 1507, + 481, + 1507 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1351, + 616, + 1400, + 616, + 1400, + 648, + 1351, + 648 + ], + "score": 0.902 + }, + { + "category_id": 9, + "poly": [ + 1351, + 914, + 1400, + 914, + 1400, + 946, + 1351, + 946 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1218, + 1400, + 1218, + 1400, + 1249, + 1352, + 1249 + ], + "score": 0.882 + }, + { + "category_id": 8, + "poly": [ + 481, + 1511, + 1221, + 1511, + 1221, + 1553, + 481, + 1553 + ], + "score": 0.88 + }, + { + "category_id": 8, + "poly": [ + 552, + 1327, + 807, + 1327, + 807, + 1365, + 552, + 1365 + ], + "score": 0.88 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1626, + 1401, + 1626, + 1401, + 1657, + 1351, + 1657 + ], + "score": 0.879 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1276, + 1401, + 1276, + 1401, + 1306, + 1351, + 1306 + ], + "score": 0.878 + }, + { + "category_id": 9, + "poly": [ + 1351, + 815, + 1400, + 815, + 1400, + 847, + 1351, + 847 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1351, + 403, + 1400, + 403, + 1400, + 436, + 1351, + 436 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1473, + 1400, + 1473, + 1400, + 1503, + 1352, + 1503 + ], + "score": 0.876 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1332, + 1401, + 1332, + 1401, + 1363, + 1351, + 1363 + ], + "score": 0.875 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1569, + 1401, + 1569, + 1401, + 1600, + 1351, + 1600 + ], + "score": 0.873 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1381, + 1400, + 1381, + 1400, + 1411, + 1352, + 1411 + ], + "score": 0.87 + }, + { + "category_id": 8, + "poly": [ + 551, + 1375, + 740, + 1375, + 740, + 1413, + 551, + 1413 + ], + "score": 0.869 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.866 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1077, + 1401, + 1077, + 1401, + 1109, + 1351, + 1109 + ], + "score": 0.865 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1517, + 1400, + 1517, + 1400, + 1548, + 1352, + 1548 + ], + "score": 0.859 + }, + { + "category_id": 8, + "poly": [ + 489, + 1556, + 918, + 1556, + 918, + 1613, + 489, + 1613 + ], + "score": 0.537 + }, + { + "category_id": 8, + "poly": [ + 487, + 1619, + 629, + 1619, + 629, + 1659, + 487, + 1659 + ], + "score": 0.283 + }, + { + "category_id": 8, + "poly": [ + 485, + 1556, + 922, + 1556, + 922, + 1659, + 485, + 1659 + ], + "score": 0.27 + }, + { + "category_id": 14, + "poly": [ + 542, + 1206, + 1156, + 1206, + 1156, + 1419, + 542, + 1419 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\langle \\hat { w } _ { 1 } , e _ { 1 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 2 } , e _ { 1 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 3 } , e _ { 1 } \\rangle ^ { 2 } } \\\\ & { \\geq \\Big ( 1 + t - O ( d ^ { - \\frac { 1 } { 1 0 } } ) \\Big ) ^ { 2 } + \\Big ( 1 - t - O ( d ^ { - \\frac { 1 } { 1 0 } } ) \\Big ) ^ { 2 } + t ^ { 2 } } \\\\ & { \\geq 2 + 3 t ^ { 2 } - O ( d ^ { - \\frac { 1 } { 1 0 } } ) } \\\\ & { \\geq 2 - O ( d ^ { - \\frac { 1 } { 1 0 } } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 898, + 1992, + 1120, + 1992, + 1120, + 2040, + 898, + 2040 + ], + "score": 0.94, + "latex": "W _ { 1 } = U ^ { \\top } \\Sigma _ { m \\times d } ^ { \\frac { 1 } { 2 } } V _ { 1 }" + }, + { + "category_id": 14, + "poly": [ + 649, + 602, + 1047, + 602, + 1047, + 663, + 649, + 663 + ], + "score": 0.94, + "latex": "\\Big | \\mathbb { E } _ { i \\in [ n _ { 1 } ] , q _ { i } ^ { ( 1 ) } = 0 } [ x _ { i } ^ { ( 1 ) } ] - e _ { 1 } \\Big | \\le d ^ { - \\frac { 1 } { 1 0 } } ," + }, + { + "category_id": 13, + "poly": [ + 908, + 303, + 1062, + 303, + 1062, + 338, + 908, + 338 + ], + "score": 0.93, + "latex": "\\{ w _ { 1 } ^ { * } , w _ { 2 } ^ { * } , w _ { 3 } ^ { * } \\}" + }, + { + "category_id": 13, + "poly": [ + 296, + 1754, + 482, + 1754, + 482, + 1789, + 296, + 1789 + ], + "score": 0.93, + "latex": "( W _ { 2 } W _ { 1 } x ) _ { y ^ { \\prime } } + 1" + }, + { + "category_id": 13, + "poly": [ + 1061, + 675, + 1161, + 675, + 1161, + 718, + 1061, + 718 + ], + "score": 0.93, + "latex": "q _ { i } ^ { ( 1 ) } = 0" + }, + { + "category_id": 13, + "poly": [ + 570, + 333, + 786, + 333, + 786, + 372, + 570, + 372 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\left\\{ \\frac { 1 } { \\alpha } w _ { 1 } ^ { * } , \\frac { 1 } { \\alpha } w _ { 2 } ^ { * } , \\frac { 1 } { \\alpha } w _ { 3 } ^ { * } \\right\\} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 407, + 755, + 504, + 755, + 504, + 795, + 407, + 795 + ], + "score": 0.93, + "latex": "q _ { i } ^ { ( 1 ) } = 0" + }, + { + "category_id": 13, + "poly": [ + 566, + 714, + 808, + 714, + 808, + 756, + 566, + 756 + ], + "score": 0.93, + "latex": "( \\hat { w } _ { 1 } - \\hat { w } _ { 3 } ) ^ { \\top } x _ { i } ^ { ( 1 ) } \\geq 1" + }, + { + "category_id": 13, + "poly": [ + 546, + 1722, + 853, + 1722, + 853, + 1758, + 546, + 1758 + ], + "score": 0.93, + "latex": "\\lVert \\dot { W } _ { 1 } ^ { \\top } W _ { 1 } \\rVert _ { F } ^ { 2 } + \\lVert W _ { 2 } ^ { \\top } W _ { 2 } \\rVert _ { F } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1125, + 1691, + 1275, + 1691, + 1275, + 1724, + 1125, + 1724 + ], + "score": 0.93, + "latex": "W _ { 1 } \\in \\mathbb { R } ^ { m \\times d }" + }, + { + "category_id": 13, + "poly": [ + 524, + 979, + 647, + 979, + 647, + 1009, + 524, + 1009 + ], + "score": 0.92, + "latex": "\\hat { w } _ { 1 } , \\hat { w } _ { 2 } , \\hat { w } _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 889, + 720, + 983, + 720, + 983, + 755, + 889, + 755 + ], + "score": 0.92, + "latex": "i \\in [ n _ { 1 } ]" + }, + { + "category_id": 13, + "poly": [ + 344, + 479, + 494, + 479, + 494, + 514, + 344, + 514 + ], + "score": 0.92, + "latex": "\\{ \\hat { w } _ { 1 } , \\hat { w } _ { 2 } , \\hat { w } _ { 3 } \\}" + }, + { + "category_id": 14, + "poly": [ + 476, + 1464, + 1221, + 1464, + 1221, + 1664, + 476, + 1664 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { \\langle \\hat { w } _ { 1 } , e _ { 2 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 2 } , e _ { 2 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 3 } , e _ { 2 } \\rangle ^ { 2 } } \\\\ & { \\leq \\| \\hat { w } _ { 1 } \\| _ { 2 } ^ { 2 } + \\| \\hat { w } _ { 2 } \\| _ { 2 } ^ { 2 } + \\| \\hat { w } _ { 3 } \\| _ { 2 } ^ { 2 } - \\langle \\hat { w } _ { 1 } , e _ { 1 } \\rangle ^ { 2 } - \\langle \\hat { w } _ { 2 } , e _ { 1 } \\rangle ^ { 2 } - \\langle \\hat { w } _ { 3 } , e _ { 1 } \\rangle ^ { 2 } } \\\\ & { \\leq \\Big ( 2 + O \\big ( d ^ { - \\frac { 1 } { 1 0 } } \\big ) \\Big ) - \\Big ( 2 - O \\big ( d ^ { - \\frac { 1 } { 1 0 } } \\big ) \\Big ) } \\\\ & { \\leq O \\big ( d ^ { - \\frac { 1 } { 1 0 } } \\big ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1329, + 1137, + 1396, + 1137, + 1396, + 1177, + 1329, + 1177 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { t \\leq \\frac { 1 } { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 343, + 1962, + 416, + 1962, + 416, + 1994, + 343, + 1994 + ], + "score": 0.92, + "latex": "\\Sigma _ { m \\times d }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1820, + 645, + 1820, + 645, + 1856, + 298, + 1856 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { W _ { 2 } ^ { \\top } W _ { 2 } = W _ { 1 } W _ { 1 } ^ { \\top } = U ^ { \\top } \\Sigma U } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 468, + 1962, + 540, + 1962, + 540, + 1994, + 468, + 1994 + ], + "score": 0.92, + "latex": "\\Sigma _ { 3 \\times m }" + }, + { + "category_id": 13, + "poly": [ + 761, + 224, + 807, + 224, + 807, + 266, + 761, + 266 + ], + "score": 0.92, + "latex": "\\xi _ { i } ^ { ( k ) }" + }, + { + "category_id": 14, + "poly": [ + 403, + 1057, + 1294, + 1057, + 1294, + 1123, + 403, + 1123 + ], + "score": 0.92, + "latex": "\\langle \\hat { w } _ { 1 } , e _ { 1 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 2 } , e _ { 1 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 3 } , e _ { 1 } \\rangle ^ { 2 } \\geq \\Big ( t + 1 - O \\big ( d ^ { - \\frac { 1 } { 1 0 } } \\big ) \\Big ) ^ { 2 } > 2 . 2 5 - O \\big ( d ^ { - \\frac { 1 } { 1 0 } } \\big ) ," + }, + { + "category_id": 13, + "poly": [ + 1017, + 231, + 1109, + 231, + 1109, + 265, + 1017, + 265 + ], + "score": 0.92, + "latex": "i \\in [ n _ { k } ]" + }, + { + "category_id": 13, + "poly": [ + 1055, + 1824, + 1196, + 1824, + 1196, + 1851, + 1055, + 1851 + ], + "score": 0.92, + "latex": "\\Sigma \\in \\mathbb { R } ^ { m \\times m }" + }, + { + "category_id": 13, + "poly": [ + 561, + 1134, + 1035, + 1134, + 1035, + 1175, + 561, + 1175 + ], + "score": 0.92, + "latex": "\\| \\hat { w } _ { 1 } \\| _ { 2 } ^ { 2 } + \\| \\hat { w } _ { 2 } \\| _ { 2 } ^ { 2 } + \\| \\hat { w } _ { 3 } \\| _ { 2 } ^ { 2 } \\leq 2 + O ( d ^ { - \\frac { 1 } { 1 0 } } )" + }, + { + "category_id": 14, + "poly": [ + 484, + 380, + 1214, + 380, + 1214, + 452, + 484, + 452 + ], + "score": 0.91, + "latex": "\\| \\frac { 1 } { \\alpha } w _ { 1 } ^ { * } \\| _ { 2 } ^ { 2 } + \\| \\frac { 1 } { \\alpha } w _ { 2 } ^ { * } \\| _ { 2 } ^ { 2 } + \\| \\frac { 1 } { \\alpha } w _ { 3 } ^ { * } \\| _ { 2 } ^ { 2 } = \\frac { 2 + \\| w _ { 3 } ^ { * } \\| _ { 2 } ^ { 2 } } { \\alpha ^ { 2 } } \\le 2 + O ( d ^ { - \\frac { 1 } { 1 0 } } ) ." + }, + { + "category_id": 13, + "poly": [ + 589, + 1009, + 655, + 1009, + 655, + 1047, + 589, + 1047 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { t \\geq \\frac { 1 } { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 645, + 558, + 798, + 558, + 798, + 593, + 645, + 593 + ], + "score": 0.91, + "latex": "n _ { 1 } \\geq \\mathrm { p o l y } ( d )" + }, + { + "category_id": 13, + "poly": [ + 1122, + 1788, + 1341, + 1788, + 1341, + 1825, + 1122, + 1825 + ], + "score": 0.91, + "latex": "W _ { 2 } ^ { \\top } W _ { 2 } = W _ { 1 } W _ { 1 } ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 387, + 1008, + 551, + 1008, + 551, + 1045, + 387, + 1045 + ], + "score": 0.91, + "latex": "\\hat { w } _ { 3 } ^ { \\top } e _ { 1 } \\triangleq t \\ge 0" + }, + { + "category_id": 14, + "poly": [ + 645, + 904, + 1055, + 904, + 1055, + 950, + 645, + 950 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { ( \\hat { w } _ { 2 } - \\hat { w } _ { 3 } ) ^ { \\top } ( - e _ { 1 } ) \\ge 1 - O ( d ^ { - \\frac { 1 } { 1 0 } } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1197, + 299, + 1404, + 299, + 1404, + 338, + 1197, + 338 + ], + "score": 0.91, + "latex": "\\alpha \\geq 1 - O ( d ^ { - { \\frac { 1 } { 1 0 } } } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 720, + 449, + 720, + 449, + 755, + 298, + 755 + ], + "score": 0.91, + "latex": "\\{ \\hat { w } _ { 1 } , \\hat { w } _ { 2 } , \\hat { w } _ { 3 } \\}" + }, + { + "category_id": 13, + "poly": [ + 910, + 1755, + 1042, + 1755, + 1042, + 1786, + 910, + 1786 + ], + "score": 0.91, + "latex": "W _ { \\mathrm { S L } } = W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 788, + 1787, + 1066, + 1787, + 1066, + 1826, + 788, + 1826 + ], + "score": 0.91, + "latex": "W _ { 2 } W _ { 1 } = \\left[ \\hat { w } _ { 1 } , \\hat { w } _ { 2 } , \\hat { w } _ { 3 } \\right] ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 1079, + 549, + 1194, + 549, + 1194, + 589, + 1079, + 589 + ], + "score": 0.9, + "latex": "1 - e ^ { d ^ { - \\frac { 1 } { 1 0 } } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1722, + 369, + 1722, + 369, + 1751, + 298, + 1751 + ], + "score": 0.9, + "latex": "\\mathbb { R } ^ { 3 \\times m }" + }, + { + "category_id": 14, + "poly": [ + 514, + 805, + 1184, + 805, + 1184, + 851, + 514, + 851 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\big ( \\hat { w } _ { 1 } - \\hat { w } _ { 3 } \\big ) ^ { \\top } e _ { 1 } \\geq 1 - \\| \\hat { w } _ { 1 } - \\hat { w } _ { 3 } \\| _ { 2 } \\cdot d ^ { - \\frac { 1 } { 1 0 } } \\geq 1 - O \\big ( d ^ { - \\frac { 1 } { 1 0 } } \\big ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 764, + 510, + 1245, + 510, + 1245, + 550, + 764, + 550 + ], + "score": 0.89, + "latex": "\\| \\hat { w } _ { 1 } \\| _ { 2 } ^ { 2 } + \\| \\hat { w } _ { 2 } \\| _ { 2 } ^ { 2 } + \\| \\hat { w } _ { 3 } \\| _ { 2 } ^ { 2 } \\leq 2 + O ( d ^ { - \\frac { 1 } { 1 0 } } )" + }, + { + "category_id": 13, + "poly": [ + 658, + 1792, + 700, + 1792, + 700, + 1823, + 658, + 1823 + ], + "score": 0.89, + "latex": "W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1237, + 1724, + 1403, + 1724, + 1403, + 1758, + 1237, + 1758 + ], + "score": 0.89, + "latex": "( W _ { 2 } W _ { 1 } x ) _ { y } \\ \\geq" + }, + { + "category_id": 13, + "poly": [ + 819, + 2002, + 860, + 2002, + 860, + 2034, + 819, + 2034 + ], + "score": 0.89, + "latex": "W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 832, + 1854, + 873, + 1854, + 873, + 1884, + 832, + 1884 + ], + "score": 0.88, + "latex": "W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 295, + 292, + 420, + 292, + 420, + 336, + 295, + 336 + ], + "score": 0.88, + "latex": "1 - e ^ { - d ^ { \\frac { 1 } { 1 0 } } } \\Big \\qquad " + }, + { + "category_id": 13, + "poly": [ + 1330, + 1693, + 1403, + 1693, + 1403, + 1724, + 1330, + 1724 + ], + "score": 0.88, + "latex": "W _ { 2 } \\in" + }, + { + "category_id": 13, + "poly": [ + 661, + 1694, + 713, + 1694, + 713, + 1723, + 661, + 1723 + ], + "score": 0.87, + "latex": "W _ { \\mathrm { S L } }" + }, + { + "category_id": 13, + "poly": [ + 561, + 1793, + 603, + 1793, + 603, + 1822, + 561, + 1822 + ], + "score": 0.85, + "latex": "W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1009, + 983, + 1036, + 983, + 1036, + 1008, + 1009, + 1008 + ], + "score": 0.85, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 877, + 1963, + 900, + 1963, + 900, + 1989, + 877, + 1989 + ], + "score": 0.84, + "latex": "\\Sigma" + }, + { + "category_id": 13, + "poly": [ + 1308, + 1885, + 1331, + 1885, + 1331, + 1911, + 1308, + 1911 + ], + "score": 0.82, + "latex": "\\Sigma" + }, + { + "category_id": 13, + "poly": [ + 521, + 1856, + 547, + 1856, + 547, + 1881, + 521, + 1881 + ], + "score": 0.81, + "latex": "U" + }, + { + "category_id": 13, + "poly": [ + 297, + 1884, + 320, + 1884, + 320, + 1911, + 297, + 1911 + ], + "score": 0.8, + "latex": "\\Sigma" + }, + { + "category_id": 13, + "poly": [ + 884, + 229, + 967, + 229, + 967, + 265, + 884, + 265 + ], + "score": 0.65, + "latex": "k \\in [ 3 ]" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2126.0, + 829.0, + 2126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1686.0, + 660.0, + 1686.0, + 660.0, + 1732.0, + 290.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1686.0, + 1124.0, + 1686.0, + 1124.0, + 1732.0, + 714.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 1686.0, + 1329.0, + 1686.0, + 1329.0, + 1732.0, + 1276.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1686.0, + 1410.0, + 1686.0, + 1410.0, + 1732.0, + 1404.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1714.0, + 297.0, + 1714.0, + 297.0, + 1765.0, + 289.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1714.0, + 545.0, + 1714.0, + 545.0, + 1765.0, + 370.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1714.0, + 1236.0, + 1714.0, + 1236.0, + 1765.0, + 854.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1714.0, + 1407.0, + 1714.0, + 1407.0, + 1765.0, + 1404.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 1753.0, + 909.0, + 1753.0, + 909.0, + 1791.0, + 483.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1753.0, + 1406.0, + 1753.0, + 1406.0, + 1791.0, + 1043.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1788.0, + 560.0, + 1788.0, + 560.0, + 1830.0, + 290.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 1788.0, + 657.0, + 1788.0, + 657.0, + 1830.0, + 604.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1788.0, + 787.0, + 1788.0, + 787.0, + 1830.0, + 701.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1788.0, + 1121.0, + 1788.0, + 1121.0, + 1830.0, + 1067.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1788.0, + 1408.0, + 1788.0, + 1408.0, + 1830.0, + 1342.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1815.0, + 297.0, + 1815.0, + 297.0, + 1861.0, + 292.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 1815.0, + 1054.0, + 1815.0, + 1054.0, + 1861.0, + 646.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1815.0, + 1408.0, + 1815.0, + 1408.0, + 1861.0, + 1197.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1851.0, + 520.0, + 1851.0, + 520.0, + 1889.0, + 294.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1851.0, + 831.0, + 1851.0, + 831.0, + 1889.0, + 548.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1851.0, + 1407.0, + 1851.0, + 1407.0, + 1889.0, + 874.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1879.0, + 296.0, + 1879.0, + 296.0, + 1919.0, + 290.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1879.0, + 1307.0, + 1879.0, + 1307.0, + 1919.0, + 321.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 1879.0, + 1407.0, + 1879.0, + 1407.0, + 1919.0, + 1332.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1911.0, + 478.0, + 1911.0, + 478.0, + 1948.0, + 295.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 477.0, + 343.0, + 477.0, + 343.0, + 515.0, + 294.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 477.0, + 1404.0, + 477.0, + 1404.0, + 515.0, + 495.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 511.0, + 763.0, + 511.0, + 763.0, + 551.0, + 293.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 511.0, + 1406.0, + 511.0, + 1406.0, + 551.0, + 1246.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 543.0, + 644.0, + 543.0, + 644.0, + 601.0, + 288.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 543.0, + 1078.0, + 543.0, + 1078.0, + 601.0, + 799.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 543.0, + 1311.0, + 543.0, + 1311.0, + 601.0, + 1195.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 221.0, + 760.0, + 221.0, + 760.0, + 269.0, + 289.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 221.0, + 883.0, + 221.0, + 883.0, + 269.0, + 808.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 221.0, + 1016.0, + 221.0, + 1016.0, + 269.0, + 968.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 221.0, + 1405.0, + 221.0, + 1405.0, + 269.0, + 1110.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 262.0, + 1404.0, + 262.0, + 1404.0, + 295.0, + 296.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 294.0, + 294.0, + 294.0, + 294.0, + 344.0, + 288.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 294.0, + 907.0, + 294.0, + 907.0, + 344.0, + 421.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 294.0, + 1196.0, + 294.0, + 1196.0, + 344.0, + 1063.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 294.0, + 1410.0, + 294.0, + 1410.0, + 344.0, + 1405.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 319.0, + 569.0, + 319.0, + 569.0, + 386.0, + 287.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 319.0, + 1356.0, + 319.0, + 1356.0, + 386.0, + 787.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 670.0, + 1060.0, + 670.0, + 1060.0, + 725.0, + 285.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 670.0, + 1414.0, + 670.0, + 1414.0, + 725.0, + 1162.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 703.0, + 297.0, + 703.0, + 297.0, + 766.0, + 286.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 703.0, + 565.0, + 703.0, + 565.0, + 766.0, + 450.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 703.0, + 888.0, + 703.0, + 888.0, + 766.0, + 809.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 703.0, + 1418.0, + 703.0, + 1418.0, + 766.0, + 984.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 757.0, + 949.0, + 757.0, + 949.0, + 798.0, + 505.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 744.0, + 469.0, + 744.0, + 469.0, + 812.0, + 283.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 971.0, + 523.0, + 971.0, + 523.0, + 1015.0, + 295.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 971.0, + 1008.0, + 971.0, + 1008.0, + 1015.0, + 648.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 971.0, + 1403.0, + 971.0, + 1403.0, + 1015.0, + 1037.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1007.0, + 386.0, + 1007.0, + 386.0, + 1048.0, + 295.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1007.0, + 588.0, + 1007.0, + 588.0, + 1048.0, + 552.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1007.0, + 766.0, + 1007.0, + 766.0, + 1048.0, + 656.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1957.0, + 342.0, + 1957.0, + 342.0, + 1997.0, + 295.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1957.0, + 467.0, + 1957.0, + 467.0, + 1997.0, + 417.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1957.0, + 876.0, + 1957.0, + 876.0, + 1997.0, + 541.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 1957.0, + 1404.0, + 1957.0, + 1404.0, + 1997.0, + 901.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1994.0, + 818.0, + 1994.0, + 818.0, + 2048.0, + 293.0, + 2048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 1994.0, + 897.0, + 1994.0, + 897.0, + 2048.0, + 861.0, + 2048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1994.0, + 1409.0, + 1994.0, + 1409.0, + 2048.0, + 1121.0, + 2048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1126.0, + 560.0, + 1126.0, + 560.0, + 1185.0, + 288.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1126.0, + 1328.0, + 1126.0, + 1328.0, + 1185.0, + 1036.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1126.0, + 1414.0, + 1126.0, + 1414.0, + 1185.0, + 1397.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1165.0, + 373.0, + 1165.0, + 373.0, + 1208.0, + 292.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 857.0, + 772.0, + 857.0, + 772.0, + 898.0, + 294.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1422.0, + 430.0, + 1422.0, + 430.0, + 1461.0, + 294.0, + 1461.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 21, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 834, + 1405, + 834, + 1405, + 1028, + 296, + 1028 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 296, + 1487, + 1407, + 1487, + 1407, + 1693, + 296, + 1693 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 295, + 1706, + 1407, + 1706, + 1407, + 1896, + 295, + 1896 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1112, + 1404, + 1112, + 1404, + 1235, + 298, + 1235 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 687, + 1406, + 687, + 1406, + 805, + 296, + 805 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1907, + 1405, + 1907, + 1405, + 2038, + 297, + 2038 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 223, + 1405, + 223, + 1405, + 305, + 297, + 305 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 295, + 614, + 1403, + 614, + 1403, + 679, + 295, + 679 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 331, + 1249, + 1317, + 1249, + 1317, + 1347, + 331, + 1347 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 442, + 1041, + 1254, + 1041, + 1254, + 1086, + 442, + 1086 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 816, + 73, + 816, + 106, + 298, + 106 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 291, + 1423, + 1269, + 1423, + 1269, + 1459, + 291, + 1459 + ], + "score": 0.909 + }, + { + "category_id": 1, + "poly": [ + 294, + 315, + 1081, + 315, + 1081, + 353, + 294, + 353 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1048, + 1401, + 1048, + 1401, + 1080, + 1351, + 1080 + ], + "score": 0.879 + }, + { + "category_id": 9, + "poly": [ + 1351, + 453, + 1401, + 453, + 1401, + 484, + 1351, + 484 + ], + "score": 0.878 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.87 + }, + { + "category_id": 9, + "poly": [ + 1351, + 501, + 1401, + 501, + 1401, + 532, + 1351, + 532 + ], + "score": 0.858 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1282, + 1401, + 1282, + 1401, + 1315, + 1352, + 1315 + ], + "score": 0.853 + }, + { + "category_id": 8, + "poly": [ + 463, + 493, + 1210, + 493, + 1210, + 539, + 463, + 539 + ], + "score": 0.832 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1359, + 1403, + 1359, + 1403, + 1388, + 1374, + 1388 + ], + "score": 0.808 + }, + { + "category_id": 8, + "poly": [ + 304, + 377, + 1378, + 377, + 1378, + 450, + 304, + 450 + ], + "score": 0.805 + }, + { + "category_id": 2, + "poly": [ + 1374, + 550, + 1402, + 550, + 1402, + 579, + 1374, + 579 + ], + "score": 0.787 + }, + { + "category_id": 8, + "poly": [ + 319, + 373, + 1377, + 373, + 1377, + 540, + 319, + 540 + ], + "score": 0.277 + }, + { + "category_id": 13, + "poly": [ + 621, + 1190, + 717, + 1190, + 717, + 1232, + 621, + 1232 + ], + "score": 0.95, + "latex": "q _ { i } ^ { ( 2 ) } = 1" + }, + { + "category_id": 13, + "poly": [ + 623, + 1585, + 808, + 1585, + 808, + 1615, + 623, + 1615 + ], + "score": 0.95, + "latex": "\\tilde { w } _ { 1 } , \\tilde { w } _ { 2 } , \\cdots , \\tilde { w } _ { m }" + }, + { + "category_id": 14, + "poly": [ + 329, + 1246, + 1321, + 1246, + 1321, + 1347, + 329, + 1347 + ], + "score": 0.94, + "latex": "e _ { 2 } ^ { \\top } M e _ { 2 } = \\mathbb { E } _ { x } [ ( e _ { 2 } ^ { \\top } x ) ^ { 2 } ] \\geq ( \\mathbb { E } _ { x } [ e _ { 2 } ^ { \\top } x ] ) ^ { 2 } \\geq \\left( { \\frac { 1 } { 3 } } \\tau - e _ { 2 } ^ { \\top } \\left( { \\frac { 1 } { n } } \\sum _ { i = 1 } ^ { n } \\xi _ { i } \\right) \\right) ^ { 2 } \\geq \\Omega ( \\tau ^ { 2 } ) = \\Omega ( d ^ { \\frac { 2 } { 5 } } ) ." + }, + { + "category_id": 13, + "poly": [ + 617, + 956, + 787, + 956, + 787, + 994, + 617, + 994 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { | M _ { i , i } ^ { \\prime } - 1 | \\leq \\frac { 1 } { d } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 393, + 1613, + 617, + 1613, + 617, + 1655, + 393, + 1655 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { M = \\sum _ { i = 1 } ^ { d } \\bar { \\lambda _ { i } } v _ { i } v _ { i } ^ { \\top } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 715, + 1144, + 849, + 1144, + 849, + 1193, + 715, + 1193 + ], + "score": 0.94, + "latex": "1 - { \\textstyle \\frac { 1 } { 2 } } e ^ { - d ^ { \\frac { 1 } { 1 0 } } }" + }, + { + "category_id": 13, + "poly": [ + 300, + 1949, + 467, + 1949, + 467, + 2000, + 300, + 2000 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\frac { \\boldsymbol { u } ^ { \\top } M \\boldsymbol { u } } { \\| \\boldsymbol { u } \\| _ { 2 } ^ { 2 } } \\leq O ( 1 ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 932, + 1489, + 1104, + 1489, + 1104, + 1524, + 932, + 1524 + ], + "score": 0.94, + "latex": "M = \\mathbb { E } _ { x } [ x x ^ { \\top } ]" + }, + { + "category_id": 13, + "poly": [ + 1151, + 1913, + 1352, + 1913, + 1352, + 1953, + 1151, + 1953 + ], + "score": 0.94, + "latex": "e _ { 2 } ^ { \\top } M e _ { 2 } \\geq \\Omega ( d ^ { \\frac { 2 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 1107, + 909, + 1242, + 909, + 1242, + 958, + 1107, + 958 + ], + "score": 0.93, + "latex": "1 - { \\textstyle \\frac { 1 } { 2 } } e ^ { - d ^ { \\frac { 1 } { 1 0 } } }" + }, + { + "category_id": 13, + "poly": [ + 869, + 1551, + 1050, + 1551, + 1050, + 1587, + 869, + 1587 + ], + "score": 0.93, + "latex": "\\| M - W ^ { \\top } W \\| _ { F } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 299, + 954, + 423, + 954, + 423, + 994, + 299, + 994 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { | M _ { i , j } ^ { \\prime } | \\leq \\frac { 1 } { d } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 353, + 1191, + 450, + 1191, + 450, + 1232, + 353, + 1232 + ], + "score": 0.93, + "latex": "q _ { i } ^ { ( 1 ) } = 1" + }, + { + "category_id": 13, + "poly": [ + 559, + 1997, + 765, + 1997, + 765, + 2036, + 559, + 2036 + ], + "score": 0.93, + "latex": "\\zeta _ { 1 } ^ { 2 } \\geq 1 - O ( d ^ { - \\frac { 1 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 863, + 871, + 1090, + 871, + 1090, + 913, + 863, + 913 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { M ^ { \\prime } = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\xi _ { i } \\xi _ { i } ^ { \\top } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1165, + 226, + 1361, + 226, + 1361, + 271, + 1165, + 271 + ], + "score": 0.93, + "latex": "W _ { 2 } = V _ { 2 } \\Sigma _ { 3 \\times m } ^ { \\frac { 1 } { 2 } } U" + }, + { + "category_id": 13, + "poly": [ + 973, + 1651, + 1153, + 1651, + 1153, + 1693, + 973, + 1693 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { e _ { 2 } = \\sum _ { i = 1 } ^ { d } \\zeta _ { i } v _ { i } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 347, + 996, + 456, + 996, + 456, + 1028, + 347, + 1028 + ], + "score": 0.93, + "latex": "\\| u \\| _ { 2 } = 1" + }, + { + "category_id": 13, + "poly": [ + 509, + 223, + 583, + 223, + 583, + 271, + 509, + 271 + ], + "score": 0.93, + "latex": "{ \\boldsymbol { \\Sigma } } _ { m \\times d } ^ { \\frac { 1 } { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 502, + 1706, + 648, + 1706, + 648, + 1744, + 502, + 1744 + ], + "score": 0.93, + "latex": "\\lambda _ { 1 } \\geq \\Omega ( d ^ { \\frac { 2 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 919, + 1905, + 1037, + 1905, + 1037, + 1949, + 919, + 1949 + ], + "score": 0.92, + "latex": "1 - e ^ { - d ^ { \\frac { 1 } { 1 0 } } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 765, + 497, + 765, + 497, + 804, + 298, + 804 + ], + "score": 0.92, + "latex": "e _ { 2 } ^ { \\top } M e _ { 2 } \\geq \\Omega ( d ^ { \\frac { 2 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 1217, + 769, + 1324, + 769, + 1324, + 804, + 1217, + 804 + ], + "score": 0.92, + "latex": "\\| u \\| _ { 2 } = 1" + }, + { + "category_id": 13, + "poly": [ + 1050, + 767, + 1164, + 767, + 1164, + 802, + 1050, + 802 + ], + "score": 0.92, + "latex": "u ^ { \\top } e _ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 865, + 687, + 1046, + 687, + 1046, + 725, + 865, + 725 + ], + "score": 0.92, + "latex": "M \\ \\triangleq \\mathbb { E } _ { x } [ x x ^ { \\top } ]" + }, + { + "category_id": 13, + "poly": [ + 297, + 231, + 424, + 231, + 424, + 267, + 297, + 267 + ], + "score": 0.92, + "latex": "V _ { 1 } \\in \\mathbb { R } ^ { d \\times d }" + }, + { + "category_id": 13, + "poly": [ + 507, + 992, + 619, + 992, + 619, + 1026, + 507, + 1026 + ], + "score": 0.92, + "latex": "u ^ { \\top } e _ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 527, + 269, + 653, + 269, + 653, + 302, + 527, + 302 + ], + "score": 0.92, + "latex": "V _ { 2 } \\in \\mathbb { R } ^ { 3 \\times 3 }" + }, + { + "category_id": 13, + "poly": [ + 866, + 957, + 941, + 957, + 941, + 991, + 866, + 991 + ], + "score": 0.92, + "latex": "i \\in [ d ]" + }, + { + "category_id": 13, + "poly": [ + 777, + 1195, + 1035, + 1195, + 1035, + 1235, + 777, + 1235 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\| \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\xi _ { i } \\| _ { 2 } \\leq O ( \\frac { 1 } { d } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1118, + 722, + 1237, + 722, + 1237, + 764, + 1118, + 764 + ], + "score": 0.91, + "latex": "1 - e ^ { - d ^ { \\frac { 1 } { 1 0 } } }" + }, + { + "category_id": 13, + "poly": [ + 946, + 271, + 1179, + 271, + 1179, + 304, + 946, + 304 + ], + "score": 0.91, + "latex": "W _ { 2 } W _ { 1 } = V _ { 2 } \\Sigma _ { 3 \\times d } V _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1257, + 955, + 1342, + 955, + 1342, + 986, + 1257, + 986 + ], + "score": 0.91, + "latex": "u \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 624, + 734, + 823, + 734, + 823, + 767, + 624, + 767 + ], + "score": 0.91, + "latex": "n _ { 1 } , n _ { 2 } \\geq \\mathrm { p o l y } ( d )" + }, + { + "category_id": 13, + "poly": [ + 853, + 768, + 938, + 768, + 938, + 800, + 853, + 800 + ], + "score": 0.91, + "latex": "u \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1743, + 664, + 1743, + 664, + 1781, + 298, + 1781 + ], + "score": 0.91, + "latex": "d ^ { - \\frac { 2 } { 5 } } , d ^ { \\frac { 2 } { 5 } } + d ^ { - \\frac { 2 } { 5 } } , d ^ { - \\frac { 2 } { 5 } } , \\cdot \\cdot \\cdot , d ^ { - \\frac { 2 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 965, + 237, + 1037, + 237, + 1037, + 268, + 965, + 268 + ], + "score": 0.91, + "latex": "\\Sigma _ { m \\times d }" + }, + { + "category_id": 13, + "poly": [ + 298, + 868, + 344, + 868, + 344, + 910, + 298, + 910 + ], + "score": 0.91, + "latex": "\\xi _ { i } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 586, + 841, + 796, + 841, + 796, + 869, + 586, + 869 + ], + "score": 0.91, + "latex": "n = n _ { 1 } + n _ { 2 } + n _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 703, + 1710, + 924, + 1710, + 924, + 1746, + 703, + 1746 + ], + "score": 0.91, + "latex": "\\operatorname* { m a x } _ { i \\neq 1 } \\lambda _ { i } \\leq O ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 502, + 957, + 565, + 957, + 565, + 990, + 502, + 990 + ], + "score": 0.91, + "latex": "i \\neq j" + }, + { + "category_id": 13, + "poly": [ + 678, + 921, + 820, + 921, + 820, + 954, + 678, + 954 + ], + "score": 0.91, + "latex": "n \\geq { \\mathrm { p o l y } } ( d )" + }, + { + "category_id": 13, + "poly": [ + 935, + 1779, + 1052, + 1779, + 1052, + 1824, + 935, + 1824 + ], + "score": 0.9, + "latex": "1 - e ^ { - d ^ { \\frac { 1 } { 1 0 } } }" + }, + { + "category_id": 13, + "poly": [ + 1196, + 1711, + 1409, + 1711, + 1409, + 1745, + 1196, + 1745 + ], + "score": 0.9, + "latex": "\\mathbb { E } [ M ] = \\mathbb { d i a g } ( 1 +" + }, + { + "category_id": 13, + "poly": [ + 298, + 1859, + 357, + 1859, + 357, + 1892, + 298, + 1892 + ], + "score": 0.89, + "latex": "O ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 1059, + 1825, + 1200, + 1825, + 1200, + 1862, + 1059, + 1862 + ], + "score": 0.89, + "latex": "\\lambda _ { 1 } \\geq \\Omega ( d ^ { \\frac { 2 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 596, + 768, + 774, + 768, + 774, + 804, + 596, + 804 + ], + "score": 0.89, + "latex": "u ^ { \\top } M u \\leq O ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 342, + 315, + 725, + 315, + 725, + 352, + 342, + 352 + ], + "score": 0.89, + "latex": "W _ { \\mathrm { S L } } = W _ { 1 } = [ w _ { 1 } , w _ { 2 } , \\cdot \\cdot \\cdot , w _ { m } ] ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1155, + 438, + 1155, + 438, + 1190, + 298, + 1190 + ], + "score": 0.89, + "latex": "n \\geq { \\mathrm { p o l y } } ( d )" + }, + { + "category_id": 13, + "poly": [ + 698, + 1827, + 1017, + 1827, + 1017, + 1862, + 698, + 1862 + ], + "score": 0.89, + "latex": "_ i \\mid \\lambda _ { i } ( T ) - \\lambda _ { i } ( S ) \\mid \\leq \\| S - T \\|" + }, + { + "category_id": 13, + "poly": [ + 1251, + 1829, + 1405, + 1829, + 1405, + 1863, + 1251, + 1863 + ], + "score": 0.89, + "latex": "\\operatorname* { m a x } _ { i \\neq 1 } \\lambda _ { i } \\leq" + }, + { + "category_id": 13, + "poly": [ + 700, + 1620, + 730, + 1620, + 730, + 1650, + 700, + 1650 + ], + "score": 0.88, + "latex": "\\lambda _ { i }" + }, + { + "category_id": 14, + "poly": [ + 444, + 1040, + 1257, + 1040, + 1257, + 1084, + 444, + 1084 + ], + "score": 0.87, + "latex": "u ^ { \\top } M u \\leq 2 \\| u \\| _ { 2 } ^ { 2 } + 2 u ^ { \\top } ( \\rho ^ { 2 } M ^ { \\prime } ) u \\leq 2 + 2 \\rho ^ { 2 } + 2 \\rho ^ { 2 } \\| M ^ { \\prime } - I \\| _ { F } \\leq O ( 1 ) ." + }, + { + "category_id": 13, + "poly": [ + 1094, + 1153, + 1113, + 1153, + 1113, + 1194, + 1094, + 1194 + ], + "score": 0.86, + "latex": "\\frac 1 3" + }, + { + "category_id": 13, + "poly": [ + 432, + 1663, + 457, + 1663, + 457, + 1688, + 432, + 1688 + ], + "score": 0.85, + "latex": "v _ { i }" + }, + { + "category_id": 13, + "poly": [ + 644, + 1662, + 673, + 1662, + 673, + 1688, + 644, + 1688 + ], + "score": 0.84, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1066, + 1782, + 1353, + 1782, + 1353, + 1826, + 1066, + 1826 + ], + "score": 0.84, + "latex": "\\| M - \\mathbb { E } [ M ] \\| \\leq O ( d ^ { - \\frac { 2 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 722, + 1963, + 749, + 1963, + 749, + 1988, + 722, + 1988 + ], + "score": 0.82, + "latex": "e _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1097, + 1620, + 1130, + 1620, + 1130, + 1647, + 1097, + 1647 + ], + "score": 0.81, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1100, + 839, + 1124, + 839, + 1124, + 870, + 1100, + 870 + ], + "score": 0.77, + "latex": "\\xi _ { i }" + }, + { + "category_id": 13, + "poly": [ + 544, + 1964, + 564, + 1964, + 564, + 1985, + 544, + 1985 + ], + "score": 0.76, + "latex": "u" + }, + { + "category_id": 14, + "poly": [ + 316, + 377, + 1379, + 377, + 1379, + 435, + 316, + 435 + ], + "score": 0.75, + "latex": "\\sum ^ { m } \\langle w _ { i } , e _ { 2 } \\rangle ^ { 2 } = \\| W _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } = \\| U ^ { \\top } \\Sigma _ { m \\times d } ^ { \\frac { 1 } { 2 } } V _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } \\leq \\| V _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } \\cdot \\| \\Sigma _ { d \\times d } V _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } = \\| V _ { 2 } \\Sigma _ { 3 \\times d } V _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1135, + 838, + 1222, + 838, + 1222, + 871, + 1135, + 871 + ], + "score": 0.74, + "latex": "( i \\in [ n ] )" + }, + { + "category_id": 13, + "poly": [ + 1125, + 1589, + 1154, + 1589, + 1154, + 1612, + 1125, + 1612 + ], + "score": 0.74, + "latex": "m" + }, + { + "category_id": 14, + "poly": [ + 468, + 491, + 1212, + 491, + 1212, + 536, + 468, + 536 + ], + "score": 0.74, + "latex": "\\begin{array} { r } { = \\| \\ b { W _ { 2 } W _ { 1 } e _ { 2 } } \\| _ { 2 } ^ { 2 } = \\langle \\hat { w } _ { 1 } , e _ { 2 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 2 } , e _ { 2 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 3 } , e _ { 2 } \\rangle ^ { 2 } \\leq O ( d ^ { - \\frac { 1 } { 1 0 } } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 804, + 1622, + 817, + 1622, + 817, + 1647, + 804, + 1647 + ], + "score": 0.72, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 297, + 1619, + 330, + 1619, + 330, + 1648, + 297, + 1648 + ], + "score": 0.71, + "latex": "M" + }, + { + "category_id": 14, + "poly": [ + 318, + 372, + 1374, + 372, + 1374, + 538, + 318, + 538 + ], + "score": 0.59, + "latex": "\\begin{array} { r l r } { { \\sum _ { i = 1 } ^ { m } \\langle w _ { i } , e _ { 2 } \\rangle ^ { 2 } = \\| W _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } = \\| U ^ { \\top } \\Sigma _ { m \\times d } ^ { \\frac { 1 } { 2 } } V _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } \\le \\| V _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } \\cdot \\| \\Sigma _ { d \\times d } V _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } = \\| V _ { 2 } \\Sigma _ { 3 \\times d } V _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } } } \\\\ & { } & { \\quad \\quad \\quad \\quad ( 3 \\times \\quad \\quad \\quad \\quad ( 3 - \\quad \\quad \\quad \\quad \\quad \\quad ( 3 ) } \\\\ & { } & { \\quad \\quad \\quad \\quad = \\| W _ { 2 } W _ { 1 } e _ { 2 } \\| _ { 2 } ^ { 2 } = \\langle \\hat { w } _ { 1 } , e _ { 2 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 2 } , e _ { 2 } \\rangle ^ { 2 } + \\langle \\hat { w } _ { 3 } , e _ { 2 } \\rangle ^ { 2 } \\le O ( d ^ { - \\frac { 1 } { 1 0 } } ) . \\qquad \\quad \\quad ( 3 - \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1366, + 735, + 1403, + 735, + 1403, + 766, + 1366, + 766 + ], + "score": 0.55, + "latex": "( l )" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2123.0, + 830.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1363.0, + 1403.0, + 1363.0, + 1403.0, + 1390.0, + 1377.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 555.0, + 1401.0, + 555.0, + 1401.0, + 577.0, + 1380.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 835.0, + 585.0, + 835.0, + 585.0, + 872.0, + 295.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 835.0, + 1099.0, + 835.0, + 1099.0, + 872.0, + 797.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 835.0, + 1134.0, + 835.0, + 1134.0, + 872.0, + 1125.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 835.0, + 1405.0, + 835.0, + 1405.0, + 872.0, + 1223.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 854.0, + 297.0, + 854.0, + 297.0, + 928.0, + 283.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 854.0, + 862.0, + 854.0, + 862.0, + 928.0, + 345.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 854.0, + 1416.0, + 854.0, + 1416.0, + 928.0, + 1091.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 909.0, + 677.0, + 909.0, + 677.0, + 968.0, + 285.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 909.0, + 1106.0, + 909.0, + 1106.0, + 968.0, + 821.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 909.0, + 1405.0, + 909.0, + 1405.0, + 968.0, + 1243.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 953.0, + 298.0, + 953.0, + 298.0, + 993.0, + 295.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 953.0, + 501.0, + 953.0, + 501.0, + 993.0, + 424.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 953.0, + 616.0, + 953.0, + 616.0, + 993.0, + 566.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 953.0, + 865.0, + 953.0, + 865.0, + 993.0, + 788.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 953.0, + 1256.0, + 953.0, + 1256.0, + 993.0, + 942.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.0, + 953.0, + 1406.0, + 953.0, + 1406.0, + 993.0, + 1343.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 989.0, + 346.0, + 989.0, + 346.0, + 1032.0, + 295.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 989.0, + 506.0, + 989.0, + 506.0, + 1032.0, + 457.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 989.0, + 730.0, + 989.0, + 730.0, + 1032.0, + 620.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1490.0, + 931.0, + 1490.0, + 931.0, + 1527.0, + 294.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1490.0, + 1410.0, + 1490.0, + 1410.0, + 1527.0, + 1105.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1523.0, + 1403.0, + 1523.0, + 1403.0, + 1553.0, + 296.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1549.0, + 868.0, + 1549.0, + 868.0, + 1590.0, + 291.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1549.0, + 1409.0, + 1549.0, + 1409.0, + 1590.0, + 1051.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1583.0, + 622.0, + 1583.0, + 622.0, + 1620.0, + 293.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1583.0, + 1124.0, + 1583.0, + 1124.0, + 1620.0, + 809.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 1583.0, + 1407.0, + 1583.0, + 1407.0, + 1620.0, + 1155.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1613.0, + 296.0, + 1613.0, + 296.0, + 1664.0, + 291.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1613.0, + 392.0, + 1613.0, + 392.0, + 1664.0, + 331.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 1613.0, + 699.0, + 1613.0, + 699.0, + 1664.0, + 618.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 1613.0, + 803.0, + 1613.0, + 803.0, + 1664.0, + 731.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1613.0, + 1096.0, + 1613.0, + 1096.0, + 1664.0, + 818.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1613.0, + 1409.0, + 1613.0, + 1409.0, + 1664.0, + 1131.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1653.0, + 431.0, + 1653.0, + 431.0, + 1700.0, + 294.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 1653.0, + 643.0, + 1653.0, + 643.0, + 1700.0, + 458.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1653.0, + 972.0, + 1653.0, + 972.0, + 1700.0, + 674.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 1653.0, + 1168.0, + 1653.0, + 1168.0, + 1700.0, + 1154.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1703.0, + 501.0, + 1703.0, + 501.0, + 1751.0, + 293.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 1703.0, + 702.0, + 1703.0, + 702.0, + 1751.0, + 649.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1703.0, + 1195.0, + 1703.0, + 1195.0, + 1751.0, + 925.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1737.0, + 297.0, + 1737.0, + 297.0, + 1788.0, + 293.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 1737.0, + 1409.0, + 1737.0, + 1409.0, + 1788.0, + 665.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1780.0, + 934.0, + 1780.0, + 934.0, + 1831.0, + 290.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1780.0, + 1065.0, + 1780.0, + 1065.0, + 1831.0, + 1053.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1780.0, + 1408.0, + 1780.0, + 1408.0, + 1831.0, + 1354.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1823.0, + 697.0, + 1823.0, + 697.0, + 1866.0, + 292.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 1823.0, + 1058.0, + 1823.0, + 1058.0, + 1866.0, + 1018.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 1823.0, + 1250.0, + 1823.0, + 1250.0, + 1866.0, + 1201.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1848.0, + 297.0, + 1848.0, + 297.0, + 1899.0, + 291.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1848.0, + 375.0, + 1848.0, + 375.0, + 1899.0, + 358.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1111.0, + 1405.0, + 1111.0, + 1405.0, + 1149.0, + 295.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1151.0, + 297.0, + 1151.0, + 297.0, + 1198.0, + 285.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1151.0, + 714.0, + 1151.0, + 714.0, + 1198.0, + 439.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1151.0, + 1093.0, + 1151.0, + 1093.0, + 1198.0, + 850.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 1151.0, + 1405.0, + 1151.0, + 1405.0, + 1198.0, + 1114.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1177.0, + 352.0, + 1177.0, + 352.0, + 1244.0, + 283.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1177.0, + 620.0, + 1177.0, + 620.0, + 1244.0, + 451.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1177.0, + 776.0, + 1177.0, + 776.0, + 1244.0, + 718.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1177.0, + 1197.0, + 1177.0, + 1197.0, + 1244.0, + 1036.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 686.0, + 864.0, + 686.0, + 864.0, + 728.0, + 294.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 686.0, + 1407.0, + 686.0, + 1407.0, + 728.0, + 1047.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 720.0, + 623.0, + 720.0, + 623.0, + 776.0, + 290.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 720.0, + 1117.0, + 720.0, + 1117.0, + 776.0, + 824.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 720.0, + 1365.0, + 720.0, + 1365.0, + 776.0, + 1238.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 720.0, + 1411.0, + 720.0, + 1411.0, + 776.0, + 1404.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 764.0, + 297.0, + 764.0, + 297.0, + 810.0, + 294.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 764.0, + 595.0, + 764.0, + 595.0, + 810.0, + 498.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 764.0, + 852.0, + 764.0, + 852.0, + 810.0, + 775.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 764.0, + 1049.0, + 764.0, + 1049.0, + 810.0, + 939.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 764.0, + 1216.0, + 764.0, + 1216.0, + 810.0, + 1165.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 764.0, + 1336.0, + 764.0, + 1336.0, + 810.0, + 1325.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1904.0, + 918.0, + 1904.0, + 918.0, + 1957.0, + 289.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 1904.0, + 1150.0, + 1904.0, + 1150.0, + 1957.0, + 1038.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 1904.0, + 1408.0, + 1904.0, + 1408.0, + 1957.0, + 1353.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1998.0, + 558.0, + 1998.0, + 558.0, + 2039.0, + 294.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1998.0, + 908.0, + 1998.0, + 908.0, + 2039.0, + 766.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.25, + 1942.5, + 1410.25, + 1942.5, + 1410.25, + 2004.0, + 285.25, + 2004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 263.0, + 526.0, + 263.0, + 526.0, + 311.0, + 291.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 263.0, + 945.0, + 263.0, + 945.0, + 311.0, + 654.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 263.0, + 1195.0, + 263.0, + 1195.0, + 311.0, + 1180.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 612.0, + 1405.0, + 612.0, + 1405.0, + 652.0, + 294.0, + 652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 645.0, + 1024.0, + 645.0, + 1024.0, + 681.0, + 295.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1422.0, + 1273.0, + 1422.0, + 1273.0, + 1461.0, + 295.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 306.0, + 341.0, + 306.0, + 341.0, + 360.0, + 292.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 306.0, + 1084.0, + 306.0, + 1084.0, + 360.0, + 726.0, + 360.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 22, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 714, + 781, + 985, + 781, + 985, + 871, + 714, + 871 + ], + "score": 0.959 + }, + { + "category_id": 8, + "poly": [ + 658, + 284, + 1037, + 284, + 1037, + 367, + 658, + 367 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 537, + 945, + 1160, + 945, + 1160, + 1033, + 537, + 1033 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 289, + 884, + 1320, + 884, + 1320, + 927, + 289, + 927 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 295, + 381, + 881, + 381, + 881, + 420, + 295, + 420 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 292, + 224, + 1194, + 224, + 1194, + 267, + 292, + 267 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 297, + 72, + 817, + 72, + 817, + 106, + 297, + 106 + ], + "score": 0.921 + }, + { + "category_id": 8, + "poly": [ + 626, + 436, + 1071, + 436, + 1071, + 516, + 626, + 516 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 297, + 715, + 1099, + 715, + 1099, + 759, + 297, + 759 + ], + "score": 0.915 + }, + { + "category_id": 9, + "poly": [ + 1351, + 808, + 1401, + 808, + 1401, + 840, + 1351, + 840 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1351, + 972, + 1401, + 972, + 1401, + 1004, + 1351, + 1004 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 295, + 651, + 1024, + 651, + 1024, + 703, + 295, + 703 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1351, + 600, + 1401, + 600, + 1401, + 631, + 1351, + 631 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1351, + 462, + 1401, + 462, + 1401, + 493, + 1351, + 493 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1351, + 308, + 1402, + 308, + 1402, + 340, + 1351, + 340 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1351, + 536, + 1401, + 536, + 1401, + 568, + 1351, + 568 + ], + "score": 0.88 + }, + { + "category_id": 1, + "poly": [ + 297, + 1050, + 1410, + 1050, + 1410, + 1139, + 297, + 1139 + ], + "score": 0.866 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2113, + 834, + 2113 + ], + "score": 0.857 + }, + { + "category_id": 8, + "poly": [ + 719, + 521, + 904, + 521, + 904, + 584, + 719, + 584 + ], + "score": 0.834 + }, + { + "category_id": 8, + "poly": [ + 720, + 593, + 834, + 593, + 834, + 635, + 720, + 635 + ], + "score": 0.736 + }, + { + "category_id": 13, + "poly": [ + 774, + 716, + 900, + 716, + 900, + 762, + 774, + 762 + ], + "score": 0.95, + "latex": "\\begin{array} { r } { t = \\frac { 1 } { \\zeta _ { 1 } ^ { 2 } } - 1 } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 623, + 434, + 1074, + 434, + 1074, + 640, + 623, + 640 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { \\displaystyle \\frac { u ^ { \\top } M u } { \\| u \\| _ { 2 } ^ { 2 } } = ( 1 - \\zeta _ { 1 } ^ { 2 } ) \\lambda _ { 1 } - \\frac { \\zeta _ { 1 } ^ { 2 } \\sum _ { i = 2 } ^ { d } \\zeta _ { i } ^ { 2 } \\lambda _ { i } } { 1 - \\zeta _ { 1 } ^ { 2 } } } \\\\ { \\displaystyle \\qquad \\geq \\frac { \\lambda _ { 1 } } { 2 } - \\operatorname* { m a x } _ { i \\neq 1 } \\lambda _ { i } } \\\\ { \\displaystyle \\qquad \\geq \\Omega ( d ^ { \\frac { 2 } { 5 } } ) , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 714, + 775, + 985, + 775, + 985, + 872, + 714, + 872 + ], + "score": 0.94, + "latex": "\\boldsymbol { u } = - t \\zeta _ { 1 } \\boldsymbol { v } _ { 1 } + \\sum _ { i = 2 } ^ { d } \\zeta _ { i } \\boldsymbol { v } _ { i } ," + }, + { + "category_id": 14, + "poly": [ + 659, + 279, + 1040, + 279, + 1040, + 368, + 659, + 368 + ], + "score": 0.94, + "latex": "u = \\sqrt { 1 - \\zeta _ { 1 } ^ { 2 } } v _ { 1 } - \\frac { \\zeta _ { 1 } \\sum _ { i = 2 } ^ { d } \\zeta _ { i } v _ { i } } { \\sqrt { 1 - \\zeta _ { 1 } ^ { 2 } } } ." + }, + { + "category_id": 14, + "poly": [ + 536, + 938, + 1161, + 938, + 1161, + 1035, + 536, + 1035 + ], + "score": 0.94, + "latex": "\\frac { u ^ { \\top } M u } { \\| u \\| _ { 2 } ^ { 2 } } \\geq t ^ { 2 } \\lambda _ { 1 } \\zeta _ { 1 } ^ { 2 } + \\sum _ { i = 2 } ^ { d } \\lambda _ { i } \\zeta _ { i } ^ { 2 } \\geq t ^ { 2 } e _ { 2 } ^ { \\top } M e _ { 2 } \\geq \\Omega ( t ^ { 2 } d ^ { \\frac { 2 } { 5 } } ) ." + }, + { + "category_id": 13, + "poly": [ + 942, + 227, + 1026, + 227, + 1026, + 267, + 942, + 267 + ], + "score": 0.93, + "latex": "\\zeta _ { 1 } ^ { 2 } \\leq \\frac { 1 } { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1242, + 1052, + 1396, + 1052, + 1396, + 1092, + 1242, + 1092 + ], + "score": 0.93, + "latex": "t ^ { 2 } \\leq O ( d ^ { - { \\frac { 1 } { 5 } } } )" + }, + { + "category_id": 13, + "poly": [ + 660, + 886, + 743, + 886, + 743, + 927, + 660, + 927 + ], + "score": 0.93, + "latex": "\\zeta _ { 1 } ^ { 2 } \\geq \\frac { 1 } { 2 }" + }, + { + "category_id": 13, + "poly": [ + 503, + 227, + 586, + 227, + 586, + 267, + 503, + 267 + ], + "score": 0.93, + "latex": "\\zeta _ { 1 } ^ { 2 } \\geq \\frac { 1 } { 2 }" + }, + { + "category_id": 13, + "poly": [ + 530, + 651, + 697, + 651, + 697, + 703, + 530, + 703 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\frac { \\boldsymbol { u } ^ { \\top } M \\boldsymbol { u } } { \\| \\boldsymbol { u } \\| _ { 2 } ^ { 2 } } \\leq O ( 1 ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 466, + 382, + 580, + 382, + 580, + 416, + 466, + 416 + ], + "score": 0.93, + "latex": "u ^ { \\top } e _ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 653, + 1050, + 820, + 1050, + 820, + 1100, + 653, + 1100 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\frac { \\boldsymbol { u } ^ { \\top } M \\boldsymbol { u } } { \\| \\boldsymbol { u } \\| _ { 2 } ^ { 2 } } \\leq O ( 1 ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 930, + 657, + 1013, + 657, + 1013, + 697, + 930, + 697 + ], + "score": 0.93, + "latex": "\\zeta _ { 1 } ^ { 2 } \\geq \\frac { 1 } { 2 }" + }, + { + "category_id": 13, + "poly": [ + 968, + 887, + 1077, + 887, + 1077, + 924, + 968, + 924 + ], + "score": 0.92, + "latex": "\\| u \\| _ { 2 } ^ { 2 } \\leq 1" + }, + { + "category_id": 13, + "poly": [ + 449, + 1099, + 655, + 1099, + 655, + 1137, + 449, + 1137 + ], + "score": 0.92, + "latex": "\\zeta _ { 1 } ^ { 2 } \\geq 1 - O ( d ^ { - \\frac { 1 } { 5 } } )" + }, + { + "category_id": 13, + "poly": [ + 631, + 383, + 740, + 383, + 740, + 419, + 631, + 419 + ], + "score": 0.92, + "latex": "\\| u \\| _ { 2 } ^ { 2 } = 1" + }, + { + "category_id": 13, + "poly": [ + 466, + 886, + 579, + 886, + 579, + 920, + 466, + 920 + ], + "score": 0.92, + "latex": "u ^ { \\top } e _ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 852, + 890, + 917, + 890, + 917, + 921, + 852, + 921 + ], + "score": 0.92, + "latex": "t \\leq 1" + }, + { + "category_id": 13, + "poly": [ + 450, + 718, + 479, + 718, + 479, + 753, + 450, + 753 + ], + "score": 0.89, + "latex": "\\zeta _ { 1 } ^ { 2 }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 884.0, + 465.0, + 884.0, + 465.0, + 927.0, + 293.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 884.0, + 659.0, + 884.0, + 659.0, + 927.0, + 580.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 884.0, + 851.0, + 884.0, + 851.0, + 927.0, + 744.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 884.0, + 967.0, + 884.0, + 967.0, + 927.0, + 918.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 884.0, + 1321.0, + 884.0, + 1321.0, + 927.0, + 1078.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 380.0, + 465.0, + 380.0, + 465.0, + 421.0, + 296.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 380.0, + 630.0, + 380.0, + 630.0, + 421.0, + 581.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 380.0, + 883.0, + 380.0, + 883.0, + 421.0, + 741.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 221.0, + 502.0, + 221.0, + 502.0, + 272.0, + 291.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 221.0, + 941.0, + 221.0, + 941.0, + 272.0, + 587.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 221.0, + 1198.0, + 221.0, + 1198.0, + 272.0, + 1027.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 709.0, + 449.0, + 709.0, + 449.0, + 765.0, + 290.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 709.0, + 773.0, + 709.0, + 773.0, + 765.0, + 480.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 709.0, + 1105.0, + 709.0, + 1105.0, + 765.0, + 901.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 650.0, + 529.0, + 650.0, + 529.0, + 710.0, + 293.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 650.0, + 929.0, + 650.0, + 929.0, + 710.0, + 698.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 650.0, + 1028.0, + 650.0, + 1028.0, + 710.0, + 1014.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1046.0, + 652.0, + 1046.0, + 652.0, + 1107.0, + 296.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1046.0, + 1241.0, + 1046.0, + 1241.0, + 1107.0, + 821.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1046.0, + 1407.0, + 1046.0, + 1407.0, + 1107.0, + 1397.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1099.0, + 448.0, + 1099.0, + 448.0, + 1138.0, + 294.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1099.0, + 665.0, + 1099.0, + 665.0, + 1138.0, + 656.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 1109.0, + 1399.0, + 1109.0, + 1399.0, + 1128.0, + 1379.0, + 1128.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 23, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/dev/5uIL1E8h1E/5uIL1E8h1E.md b/parse/dev/5uIL1E8h1E/5uIL1E8h1E.md new file mode 100644 index 0000000000000000000000000000000000000000..88e834d13e5d74bd18fa1cc3d6599cdea5f19199 --- /dev/null +++ b/parse/dev/5uIL1E8h1E/5uIL1E8h1E.md @@ -0,0 +1,539 @@ +# Residual Scheduling: A New Reinforcement Learning Approach to Solving Job Shop Scheduling Problem + +Anonymous Author(s) +Affiliation +Address +email + +# Abstract + +1 Job-shop scheduling problem (JSP) is a mathematical optimization problem widely +2 used in industries like manufacturing, and flexible JSP (FJSP) is also a common +3 variant. Since they are NP-hard, it is intractable to find the optimal solution for +4 all cases within reasonable times. Thus, it becomes important to develop efficient +5 heuristics to solve JSP/FJSP. A kind of method of solving scheduling problems +6 is construction heuristics, which constructs scheduling solutions via heuristics. +7 Recently, many methods for construction heuristics leverage deep reinforcement +8 learning (DRL) with graph neural networks (GNN). In this paper, we propose a new +9 approach, named residual scheduling, to solving JSP/FJSP. In this new approach, +10 we remove irrelevant machines and jobs such as those finished, such that the states +11 include the remaining (or relevant) machines and jobs only. Our experiments show +12 that our approach reaches state-of-the-art (SOTA) among all known construction +13 heuristics on most well-known open JSP and FJSP benchmarks. In addition, we +14 also observe that even though our model is trained for scheduling problems of +15 smaller sizes, our method still performs well for scheduling problems of large sizes. +16 Interestingly in our experiments, our approach even reaches zero gap for 49 among +17 50 JSP instances whose job numbers are more than 150 on 20 machines. + +# 18 1 Introduction + +19 The job-shop scheduling problem (JSP) is a mathematical optimization (MO) problem widely used in +20 many industries, like manufacturing (Zhang et al., 2020; Waschneck et al., 2016). For example, a +21 semiconductor manufacturing process can be viewed as a complex JSP problem (Waschneck et al., +22 2016), where a set of given jobs are assigned to a set of machines under some constraints to achieve +23 some expected goals such as minimizing makespan which is focused on in this paper. While there are +24 many variants of JSP (Abdolrazzagh-Nezhad and Abdullah, 2017), we also consider an extension +25 called flexible JSP (FJSP) where job operations can be done on designated machines. +26 A generic approach to solving MO problems is to use mathematical programming, such as mixed +27 integer linear programming (MILP) and constraint satisfaction problem (CSP). Two popular generic +28 MO solvers for solving MO are OR-Tools (Perron and Furnon, 2019) and IBM ILOG CPLEX +29 Optimizer (abbr. CPLEX) (Cplex, 2009). However, both JSP and FJSP, as well as many other MO +30 problems, have been shown to be NP-hard (Garey and Johnson, 1979; Lageweg et al., 1977). That +31 said, it is unrealistic and intractable to find the optimal solution for all cases within reasonable times. +32 These tools can obtain the optimal solutions if sufficient time (or unlimited time) is given; otherwise, +33 return best-effort solutions during the limited time, which usually have gaps to the optimum. When +34 problems are scaled up, the gaps usually grow significantly. +35 In practice, some heuristics (Gupta and Sivakumar, 2006; Haupt, 1989) or approximate methods +36 (Jansen et al., 2000) were used to cope with the issue of intractability. A simple greedy approach is to +37 use the heuristics following the so-called priority dispatching rule (PDR) (Haupt, 1989) to construct +38 solutions. These can also be viewed as a kind of solution construction heuristics or construction +39 heuristics. Some of PDR examples are First In First Out (FIFO), Shortest Processing Time (SPT), +40 Most WorK Remaining (MWKR), and Most Operation Remaining (MOR). Although these heuristics +41 are usually computationally fast, it is hard to design generally effective rules to minimize the gap to +42 the optimum, and the derived results are usually far from the optimum. +43 Furthermore, a generic approach to automating the design of heuristics is called metaheuristics, such +44 as tabu search (Dell’Amico and Trubian, 1993; Saidi-Mehrabad and Fattahi, 2007) , genetic algorithm +45 (GA) (Pezzella et al., 2008; Ren and Wang, 2012), and PSO algorithms (Lian et al., 2006; Liu et al., +46 2011). However, metaheuristics still take a high computation time, and it is not ensured to obtain the +47 optimal solution either. +48 Recently, deep reinforcement learning (DRL) has made several significant successes for some +49 applications, such as AlphaGo (Silver et al., 2016), AlphaStar (Vinyals et al., 2019), AlphaTensor +50 (Fawzi et al., 2022), and thus it also attracted much attention in the MO problems, including chip +51 design (Mirhoseini et al., 2021) and scheduling problems (Zhang et al., 2023). In the past, several +52 researchers used DRL methods as construction heuristics, and their methods did improve scheduling +53 performance, illustrated as follows. Park et al. (2020) proposed a method based on DQN (Mnih et al., +54 2015) for JSP in semiconductor manufacturing and showed that their DQN model outperformed GA +55 in terms of both scheduling performance (namely gap to the optimum on makespan) and computation +56 time. Lin et al. (2019) and Luo (2020) proposed different DQN models to decide the scheduling action +57 among the heuristic rules and improved the makespan and the tardiness over PDRs, respectively. +58 A recent DRL-based approach to solving JSP/FJSP problems is to leverage graph neural networks +59 (GNN) to design a size-agnostic representation (Zhang et al., 2020; Park et al., 2021b,a; Song et al., +60 2023). In this approach, graph representation has better generalization ability in larger instances +61 and provides a holistic view of scheduling states. Zhang et al. (2020) proposed a DRL method +62 with disjunctive graph representation for JSP, called L2D (Learning to Dispatch), and used GNN +63 to encode the graph for scheduling decision. Besides, Song et al. (2023) extended their methods +64 to FJSP. Park et al. (2021b) used a similar strategy of (Zhang et al., 2020) but with different state +65 features and model structure. Park et al. (2021a) proposed a new approach to solving JSP, called +66 ScheduleNet, by using a different graph representation and a DRL model with the graph attention for +67 scheduling decision. Most of the experiments above showed that their models trained from small +68 instances still worked reasonably well for large test instances, and generally better than PDRs. Among +69 these methods, ScheduleNet achieved state-of-the-art (SOTA) performance. There are still other +70 DRL-based approaches to solving JSP/FJSP problems, but not construction heuristics. Zhang et al. +71 (2022) proposes another approach, called Learning to Search (L2S), a kind of search-based heuristics. +72 In this paper, we propose a new approach to solving JSP/FJSP, a kind of construction heuristics, also +73 based on GNN. In this new approach, we remove irrelevant machines and jobs, such as those finished, +74 such that the states include the remaining machines and jobs only. This approach is named residual +75 scheduling in this paper to indicate to work on the remaining graph. +76 Without irrelevant information, our experiments show that our approach reaches SOTA by outper +77 forming the above mentioned construction methods on some well-known open benchmarks, seven +78 for JSP and two for FJSP, as described in Section 4. We also observe that even though our model +79 is trained for scheduling problems of smaller sizes, our method still performs well for scheduling +80 problems of large sizes. Interestingly in our experiments, our approach even reaches zero gap for 49 +81 among 50 JSP instances whose job numbers are more than 150 on 20 machines. + +# 82 2 Problem Formulation + +# 2.1 JSP and FJSP + +A $n \times m$ JSP instance contains $n$ jobs and $m$ machines. Each job $J _ { j }$ consists of a sequence of $k _ { j }$ operations $\{ O _ { j , 1 } , \dotsc , O _ { j , k _ { j } } \}$ , where operation $O _ { j , i }$ must be started after $O _ { j , i - 1 }$ is finished. One machine can process at most one operation at a time, and preemption is not allowed upon processing operations. In JSP, one operation $O _ { j , i }$ is allowed to be processed on one designated machine, denoted by $M _ { j , i }$ , with a processing time, denoted by $T _ { j , i } ^ { ( o p ) }$ . Table 1 (a) illustrates a $3 \times 3$ JSP instance, where the three jobs have 3, 3, 2 operations respectively, each of which is designated to be processed on + +90 one of the three machines $\{ M _ { 1 } , M _ { 2 } , M _ { 3 } \}$ in the table. A solution of a JSP instance is to dispatch all +91 operations Oj,i to the corresponding machine Mj,i at time τ (s)j,i , such that the above constraints are +92 +93 While there are different expected goals, such as makespan, tardiness, etc., this paper focuses on +94 makespan. Let the first osolution is defined to be $\tau = 0$ in a JSP soluti all operations nitially. , where +95 $T ^ { ( m k s p ) } = \operatorname* { m a x } ( \tau _ { j , i } ^ { ( c ) } )$ $O _ { j , i }$ $\tau _ { j , i } ^ { ( c ) } = \tau _ { j , i } ^ { ( s ) } + T _ { j , i } ^ { ( o p ) }$ +96 denotes the completion time of $O _ { j , i }$ . The makespans for the two solutions illustrated in Figure 1 (a) +97 and (b) are 12 and 15 respectively. The objective is to derive a solution that minimizes the makespan +98 $T ^ { ( m k s p ) }$ , and the solution of Figure 1 (a) reaches the optimal. +99 A $n \times m$ FJSP instance is also a $n \times m$ JSP instance with the following difference. In FJSP, +100 all operations $O _ { j , i }$ are allowed to be dispatched to multiple designated machines with designated +101 processing times. Table 1 (b) illustrates a $3 \times 3$ FJSP instance, where multiple machines can be +102 designated to be processed for one operation. Figure 1 (c) illustrates a solution of an FJSP instance, +103 which takes a shorter time than that in Figure 1 (d). + +Table 1: JSP and FJSP instances +(b) A $3 \times 3$ FJSP instance + +
JobOperationM1M2M3
Job 1O1,132
O1.235
O1.343
Job 2O2,12
02.24
02.33
Job 3O3,134
03.222
+ +(a) A $3 \times 3$ JSP instance + +
JobOperationM1M2M3
Job 1O1,13
O1.25
O1.34
Job 2O2,12
02,24
O2,33
Job 3O3,13
03.22
+ +![](images/54e0e1086515bc311b9d4a17fdee1c89c424b15d95726a4af33fc4a68fbe5ad1.jpg) +Figure 1: Both (a) and (b) are solutions of the $3 \mathrm { x } 3$ JSP instance in Table 1 (a), and the former has the minimal makespan, 12. Both (c) and (d) are solutions of the $3 \mathrm { x } 3$ FJSP instance in Table 1 (b), and the former has the minimal makespan, 9. + +# 104 2.2 Construction Heuristics + +105 An approach to solving these scheduling problems is to construct solutions step by step in a greedy +106 manner, and the heuristics based on this approach is called construction heuristics in this paper. In +107 the approach of construction heuristics, a scheduling solution is constructed through a sequence of +108 partial solutions in a chronicle order of dispatching operations step by step, defined as follows. The +109 $t$ -th partial solution $S _ { t }$ associates with a dispatching time $\tau _ { t }$ and includes a partial set of operations +110 that have been dispatched by $\tau _ { t }$ (inclusive) while satisfying the above JSP constraints, and all the +111 remaining operations must be dispatched after $\tau _ { t }$ (inclusive). The whole construction starts with $S _ { 0 }$ +112 where none of operations have been dispatched and the dispatching time is $\tau _ { 0 } = 0$ . For each $S _ { t }$ , a set +113 of operations to be chosen for dispatching form a set of pairs of $( M , O )$ , called candidates $C _ { t }$ , where +114 operations $O$ are allowed to be dispatched on machines $M$ at $\tau _ { t }$ . An agent (or a heuristic algorithm) +115 chooses one from candidates $C _ { t }$ for dispatching, and transits the partial solution to the next $S _ { t + 1 }$ . If +116 there exists no operations for dispatching, the whole solution construction process is done and the +117 partial solution is a solution, since no further operations are to be dispatched. +118 Figure 2 illustrates a solution construction process for the 3x3 JSP instance in Table 1(a), constructed +119 through nine partial solutions step by step. The initial partial solution $S _ { 0 }$ starts without any operations +120 dispatched as in Figure 2 (a). The initial candidates $C _ { 0 }$ are $\{ ( M _ { 1 } , O _ { 1 , 1 } ) , ( M _ { 3 } , O _ { 2 , 1 } ) , ( \bar { M _ { 1 } } , O _ { 3 , 1 } ) \}$ +121 Following some heuristic, construct a solution from partial solution $S _ { 0 }$ to $S _ { 9 }$ step by step as in the +122 Figure, where the dashed line in red indicate the time $\tau _ { t }$ . The last one $S _ { 9 }$ , the same as the one in +123 Figure 1 (a), is a solution, since all operations have been dispatched, and the last operation ends at +124 time 12, the makespan of the solution. +125 For FJSP, the process of solution construction is almost the same except for that one operation have +126 multiple choices from candidates. Besides, an approach based on solution construction can be also +127 viewed as the so-called Markov decision process $( M D P )$ , and the MDP formulation for solution +128 construction is described in more detail in the appendix. + +![](images/cb96d32bb1f5dcecc876eaf54a9e7682b46451e9161fb74084ead5c2b39e2e49.jpg) +Figure 2: Solution construction, a sequence of partial solutions from $S _ { 0 }$ to $S _ { 8 }$ . + +# 129 3 Our Approach + +130 In this section, we present a new approach, called residual scheduling, to solving scheduling problems. +131 We introduce the residual scheduling in Subsection 3.1, describe the design of the graph representation +132 in Subsection 3.2, propose a model architecture based on graph neural network in Subsection 3.3 and +133 present a method to train this model in Subsection 3.4; + +# 34 3.1 Residual Scheduling + +135 In our approach, the key is to remove irrelevant information, particularly for operations, from states +136 (including partial solutions). An important benefit from this is that we do not need to include all +137 irrelevant information while training to minimize the makespan. Let us illustrate by the state for the +138 partial solution $S _ { 3 }$ at time $\tau _ { 3 } = 3$ in Figure 2 (d). All processing by $\tau _ { 3 }$ are irrelevant to the remaining +139 scheduling. Since operations $O _ { 1 , 1 }$ and $O _ { 2 , 1 }$ are both finished and irrelevant the rest of scheduling, +140 they can be removed from the state of $S _ { 3 }$ . In addition, operation $O _ { 2 , 2 }$ is dispatched at time 2 (before +141 $\tau _ { 3 } = 3$ ) and its processing time is $T _ { 2 , 1 } ^ { ( o p ) } = 4$ , so the operation is marked as ongoing. Thus, the +142 operation can be modified to start at $\tau _ { 3 } = 3$ with a processing time . Thus, the modified +143 state for $S _ { 3 }$ do not contain both $O _ { 1 , 1 }$ and $O _ { 2 , 1 }$ , and modify $O _ { 2 , 2 }$ as above. Let us consider two more +144 examples. For $S _ { 4 }$ , one more operation $O _ { 2 , 2 }$ is dispatched and thus marked as ongoing, however, the +145 time $\tau _ { 4 }$ remains unchanged and no more operations are removed. In this case, the state is almost the +146 same except for including one more ongoing operation $O _ { 2 , 2 }$ . Then, for $S _ { 5 }$ , two more operations $O _ { 3 , 1 }$ +147 and $O _ { 2 , 2 }$ are removed and the ongoing operation $O _ { 1 , 2 }$ changes its processing time to the remaining +148 time (5-3). +149 For residual scheduling, we also reset the dispatching time $\tau = 0$ for all states with partial solutions +150 modified as above, so we derive makespans which is also irrelevant to the earlier operations. Given +151 a scheduling policy $\pi$ , $T _ { \pi } ^ { ( m k s p ) } ( S )$ is defined to be the makespan derived from an episode starting +152 from states $S$ by following $\pi$ , and $T _ { \pi } ^ { ( m k s p ) } ( S , a )$ the makespan by taking action $a$ on $S$ . + +# 153 3.2 Residual Graph Representation + +154 In this paper, our model design is based on graph neural network (GNN), and leverage GNN to +155 extract the scheduling decision from the relationship in graph. In this subsection, we present the +156 graph representation. Like many other researchers such as Park et al. (2021a), we formulate a partial +157 solution into a graph $\mathcal { G } = ( \nu , \mathcal { E } )$ , where $\nu$ is a set of nodes and $\mathcal { E }$ is a set of edges. A node is either a +158 machine node $M$ or an operation node $O$ . An edge connects two nodes to represent the relationship +159 between two nodes, basically including three kinds of edges, namely operation-to-operation $( O \to O$ ), +160 machine-to-operation $M O$ ) and operation-to-machine $( O \to M$ ). All operations in the same +161 job are fully connected as $O O$ edges. If an operation $O$ is able to be performed on a machine +162 $M$ , there exists both $O \to M$ and $M O$ directed edges. In (Park et al., 2021a), they also let all +163 machines be fully connected as $M \to M$ edges. However, our experiments in section 4 show that +164 mutual $M \to M$ edges do not help much based on our Residual Scheduling. An illustration for graph +165 representation of $S _ { 3 }$ is depicted in Figure 3 (a). +166 In the graph representation, all nodes need to include some attributes so that a partial solution $S$ at +167 the dispatching time $\tau$ can be supported in the MDP formulation (in the appendix). Note that many of +168 the attributes below are normalized to reduce variance. For nodes corresponding to operations $O _ { j , i }$ , +169 we have the following attributes: +170 Status $\phi _ { j , i }$ : The operation status $\phi _ { j , i }$ is completed if the operation has been finished by $\tau$ , ongoing if +171 the operation is ongoing (i.e., has been dispatched to some machine by $\tau$ and is still being processed +172 at $\tau$ ), ready if the operation designated to the machine which is idle has not been dispatched yet and +173 its precedent operation has been finished, and unready otherwise. For example, in Figure 3 (a), the +174 gray nodes are completed, the red ongoing, the yellow ready and the white unready. In our residual +175 scheduling, there exists no completed operations in all partial solutions, since they are removes for +176 irrelevance of the rest of scheduling. The attribute is a one-hot vector to represent the current status +177 of the operation, which is one of ongoing, ready and unready. Illustration for all states $S _ { 0 }$ to $S _ { 8 }$ are +178 shown in the appendix. +179 Normalized processing time $\bar { T } _ { j , i } ^ { ( o p ) }$ : Let the maximal processing time be $T _ { m a x } ^ { ( o p ) } = \operatorname* { m a x } _ { \forall j , i } ( T _ { j , i } ^ { ( o p ) } )$ +180 Then, T¯(op)j,i $\bar { T } _ { j , i } ^ { ( o p ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }$ . In our residual scheduling, the operations that have been finished are +181 removed in partial solutions and therefore their processing time can be ignored; the operations that +182 has not been dispatched yet still keep their processing times the same; the operations that are ongoing +183 change their processing times to the remaining times after the dispatching time $\tau _ { t }$ . As for FJSP, the +184 operations that has not been dispatched yet may have several processing times on different machines, +185 and thus we can simply choose the average of these processing times. + +![](images/a3a8af5ab1851b807add19a4a2a54f81ec9c7979dcede6e1a0cd5113aacd0d36.jpg) +Figure 3: Graph representation and networks. + +ob remaining time , and let the process $\bar { T } _ { j , i } ^ { ( j o b ) }$ : Let the rest of pre for the whole job $J _ { j }$ $T _ { j , i } ^ { ( j o b ) } =$ $\sum _ { \forall i ^ { \prime } \geq i } T _ { j , i ^ { \prime } } ^ { ( o p ) }$ aced by th ing timocessing time for the original jo $j$ $\begin{array} { r } { T _ { j } ^ { ( j o b ) } = \sum _ { \forall i ^ { \prime } } T _ { j , i ^ { \prime } } ^ { ( o p ) } } \end{array}$ . In practice, T ( j ob )j is $j$ $\bar { T } _ { j , i } ^ { ( j o b ) } = T _ { j , i } ^ { ( j o b ) } / T _ { j } ^ { ( j o b ) }$ . For FJSP, since operations $O _ { j , i }$ can be dispatched to different designated machines $M _ { l }$ , say with the processing time T (op)j,i,l , we simply let T (op)j,i be the average of $T _ { j , i , l } ^ { ( o p ) }$ for all $M _ { l }$ . + +For machine nodes corresponding to machines $M _ { l }$ , we have the following attributes: + +Machine status $\phi _ { l }$ : The machine status $\phi _ { l }$ is processing if some operation has been dispatched to and is being processed by $M _ { l }$ at $\tau$ , and idle otherwise (no operation is being processed at $\tau$ ). The attribute is a one-hot vector to represent the current status, which is one of processing and idle. + +Normalized operation processing time $\bar { T } _ { l } ^ { ( m a c ) }$ : On the machine $M _ { l }$ , the processing time $T _ { l } ^ { ( m a c ) }$ is $T _ { j , i } ^ { ( o p ) }$ (the same as the normalized processing time for node $O _ { j , i }$ ) if the machine status is processing, i.e., some ongoing operation $O _ { j , i }$ is being processed but not finished yet, is zero if the machine status is idle. Then, this attribute is normalized to $T _ { m a x } ^ { ( o p ) }$ and thus $\bar { T } _ { l } ^ { ( m a c ) } = T _ { l } ^ { ( m a c ) } / T _ { m a x } ^ { ( o p ) }$ . + +Now, consider edges in a residual scheduling graph. As described above, there exists three relationship sets for edges, $O O$ , $O \to M$ and $M O$ . First, for the same job, say $J _ { j }$ , all of its operation nodes for $O _ { j , i }$ are fully connected. Note that for residual scheduling the operations finished by the dispatching time $\tau$ are removed and thus have no edges to them. Second, a machine node for $M _ { l }$ is connected to an operation node for $O _ { j , i }$ , if the operation $O _ { j , i }$ is designated to be processed on the machine $M _ { l }$ , which forms two edges $\dot { O } M$ and $M O$ . Both contains the following attribute. + +Normalized operation processing tim e T¯(edge): The attribute is $\bar { T } _ { j , i , l } ^ { ( e d g e ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }$ . Here, $T _ { j , i } ^ { ( o p ) } = T _ { j , i , l } ^ { ( o p ) }$ $O _ { j , i }$ $T _ { j , i } ^ { ( o p ) }$ is the remaining time as described above. + +# 3.3 Graph Neural Network + +In this subsection, we present our model based on graph neural network (GNN). GNN are a family of deep neural networks (Battaglia et al., 2018) that can learn representation of graph-structured data, widely used in many applications (Lv et al., 2021; Zhou et al., 2020). A GNN aggregates information from node itself and its neighboring nodes and then update the data itself, which allows the GNN to capture the complex relationships within the data graph. For GNN, we choose Graph Isomorphism Network (GIN), which was shown to have strong discriminative power ( $\mathrm { { X u } }$ et al., 2019) and summarily reviewed as follows. Given a graph $\mathcal { G } = ( \nu , \mathcal { E } )$ and $K$ GNN layers $K$ iterations), GIN performs the $k$ -th iterations of updating feature embedding $h ^ { ( k ) }$ for each node $v \in \mathcal V$ : + +$$ +h _ { v } ^ { ( k ) } = M L P ^ { ( k ) } ( ( 1 + \epsilon ^ { ( k ) } ) h _ { v } ^ { ( k - 1 ) } + \sum _ { u \in N _ { b } ( v ) } h _ { u } ^ { ( k - 1 ) } ) , +$$ + +217 where $h _ { v } ^ { ( k ) }$ is the embedding of node $v$ at the $k$ -th layer, $\epsilon ^ { ( k ) }$ is an arbitrary number that can be +218 learned, and $N _ { b } ( v )$ is the neighbors of $v$ via edges in $\mathcal { E }$ . Note that $h _ { v } ^ { ( 0 ) }$ refers to its raw features for +219 input. $M L P ^ { ( k ) }$ is a Multi-Layer Perceptron $( M L P )$ for the $k$ -th layer with a batch normalization +220 (Ioffe and Szegedy, 2015). +221 Furthermore, we actually use heterogeneous GIN, also called HGIN, since there are two types of +222 nodes, machine and operation nodes, and three relations, $O O$ , $O \to M$ and $M O$ in the +223 graph representation. Although we do not have cross machine relations $M \to M$ as described above, +224 updating machine nodes requires to include the update from itself as in (1), that is, there is also one +225 more relation $M \to M$ . Thus, HGIN encodes graph information between all relations by using the +226 four MLPs as follows, + +$$ +h _ { v } ^ { ( k + 1 ) } = \sum _ { \mathcal { R } } M L P _ { \mathcal { R } } ^ { ( k + 1 ) } ( ( 1 + \epsilon _ { \mathcal { R } } ^ { ( k + 1 ) } ) h _ { v } ^ { ( k ) } + \sum _ { u \in N _ { \mathcal { R } } ( v ) } h _ { u } ^ { ( k ) } ) +$$ + +where 227 $\mathcal { R }$ is one of the above fo elatio $M L P _ { \mathcal { R } } ^ { ( k ) }$ is the MLP for $\mathcal { R }$ . For example, for $S _ { 0 }$ in Figure 2 (a), the embedding of $M _ { 1 }$ $( k + 1 )$ -st iteration can be derived as follows. + +$$ +h _ { M _ { 1 } } ^ { ( k + 1 ) } = M L P _ { M M } ^ { ( k + 1 ) } ( ( 1 + \epsilon _ { M M } ^ { ( k + 1 ) } ) h _ { M _ { 1 } } ^ { ( k ) } ) + M L P _ { O M } ^ { ( k + 1 ) } ( h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } ) +$$ + +229 Similarly, the embedding of $O _ { 1 , 1 }$ in the $( k + 1 )$ -st iteration is: + +$$ +h _ { O _ { 1 , 1 } } ^ { ( k + 1 ) } = M L P _ { O O } ^ { ( k + 1 ) } ( ( 1 + \epsilon _ { O O } ^ { ( k + 1 ) } ) h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } ) + M L P _ { M O } ^ { ( k + 1 ) } ( h _ { M _ { 1 } } ^ { ( k ) } ) +$$ + +230 In our approach, an action includes the two phases, graph embedding phase and action selection phase. +231 Let $h _ { \mathcal { G } } ^ { ( k ) }$ denote the whole embedding of the graphs $\mathcal { G }$ , a summation of the embeddings of all nodes, +232 $h _ { v } ^ { ( k + 1 ) }$ . In the graph embedding phase, we use an HGIN to encode node and graph embeddings as +233 described above. An example with three HGIN layers is illustrated in Figure 3 (b). + +In the action selection phase, we select an action based on a policy, after node and graph embedding are encoded in the graph embedding phase. The policy is described as follows. First, collect all ready operations $O$ to be dispatched to machines $M$ . Then, for all pairs $( M , O )$ , feed their node embeddings $( h _ { M } ^ { ( k ) } , h _ { O } ^ { ( k ) } )$ into a MLP $S c o r e ( M , O )$ to calculate their scores as shown in Figure 3 (c). The probability of selecting $( M , O )$ is calculated based on a softmax function of all scores, which also serves as the model policy $\pi$ for the current state. + +# 3.4 Policy-Based RL Training + +41 In this paper, we propose to use a policy-based RL training mechanism that follows REINFORCE +42 (Sutton and Barto, 2018) to update our model by policy gradient with a normalized advantage +43 makespan with respect to a baseline policy $\pi _ { b }$ as follows. + +$$ +A _ { \pi } ( S , a ) = { \frac { T _ { \pi _ { b } } ^ { ( m k s p ) } ( S , a ) - T _ { \pi } ^ { ( m k s p ) } ( S , a ) } { T _ { \pi _ { b } } ^ { ( m k s p ) } ( S , a ) } } +$$ + +244 In this paper, we choose a lightweight PDR, MWKR, as baseline $\pi _ { b }$ , which performed best for +245 makespan among all PDRs reported from the previous work (Zhang et al., 2020). In fact, our +246 experiment also shows that using MWKR is better than the other PDRs shown in the appendix. The +247 model for policy $\pi$ is parametrized by $\theta$ , which is updated by $\nabla _ { \boldsymbol { \theta } } l o g \pi _ { \boldsymbol { \theta } } A _ { \pi _ { \boldsymbol { \theta } } } \bigl ( S _ { t } , a _ { t } \bigr )$ . Our algorithm +248 based on REINFORCE is listed in the appendix. + +# 4 Experiments + +# 4.1 Experimental Settings and Evaluation Benchmarks + +In our experiments, the settings of our model are described as follows. All embedding and hidden vectors in our model have a dimension of 256. The model contains three HGIN layers for graph embedding, and an MLP for the score function, as shown in Figure 3 (b) and (c). All MLP networks including those in HGIN and for score contain two hidden layers. The parameters of our model, such as MLP, generally follow the default settings in PyTorch (Paszke et al., 2019) and PyTorch Geometric (Fey and Lenssen, 2019). More settings are in the appendix. + +Each of our models is trained with one million episodes, each with one scheduling instance. Each instance is generated by following the procedure which is used to generate the TA dataset (Taillard, 1993). Given $( N , M )$ , we use the procedure to generate an $n \times m$ JSP instance by conforming to the following distribution, $n \sim \mathcal { U } ( \bar { 3 } , N )$ , $m \sim \mathcal { U } ( 3 , n )$ , and operation count $k _ { j } = m$ , where $\boldsymbol { \mathcal { U } } ( \boldsymbol { x } , \boldsymbol { y } )$ represents a distribution that uniformly samples an integer in a close interval $[ x , y ]$ at random. The details of designation for machines and processing times refer to (Taillard, 1993) and thus are omitted here. We choose (10,10) for all experiments, since (10,10) generally performs better than the other two as described in the appendix. Following the method described in Subsection 3.4, the model is updated from the above randomly generated instances. For testing our models for JSP and FJSP, seven JSP open benchmarks and two FJSP open benchmarks are used, as listed in the appendix. + +67 The performance for a given policy method $\pi$ on an instance is measured by the makespan gap $G$ +68 defined as + +$$ +G = \frac { T _ { \pi } ^ { \left( m k s p \right) } - T _ { \pi * } ^ { \left( m k s p \right) } } { T _ { \pi * } ^ { \left( m k s p \right) } } +$$ + +where 269 $T _ { \pi * } ^ { ( m k s p ) }$ is the optimal makespan or the best-effort makespan, from a mathematical optimization 270 tool, OR-Tools, serving as . By the best-effort makespan, we mean the makespan derived with a + +Table 2: Average makespan gaps for TA benchmarks. + +
Size15×1520×1520×2030×1530×2050×1550×20100×20Aug.
RS0.1480.1650.1690.1440.1770.0670.1000.0260.0500.1250.150
RS+op0.1430.1930.1590.1920.2130.1230.126
MWKR0.1910.2050.2580.2330.2350.3280.2180.2170.2770.2390.2280.3520.2510.2490.3440.1680.1730.2410.2060.1790.1760.2550.2390.0830.0910.1440.1350.1950.1970.2750.254
MORSPT
FIFO0.2390.3140.2730.3110.311
L2DParkSchN0.2590.2010.1520.3000.3160.3290.3360.2230.2650.1360.0920.0660.2700.2210.161
0.2490.1940.2920.1720.2460.1900.3190.2370.1590.1380.2120.135
+ +271 sufficiently large time limitation, namely half a day with OR-Tools. For comparison in experiments, +272 we use a server with Intel Xeon E5-2683 CPU and a single NVIDIA GeForce GTX 1080 Ti GPU. +273 Our method uses a CPU thread and a GPU to train and evaluate, while OR-Tools uses eight threads +274 to find the solution. + +# 4.2 Experiments for JSP + +For JSP, we first train a model based on residual scheduling, named RS. For ablation testing, we also train a model, named $\mathrm { R S + o p }$ , by following the same training method but without removing irrelevant operations. When using these models to solve testing instances, action selection is based on the greedy policy that simply chooses the action $( M , O )$ with the highest score deterministically, obtained from the score network as in Figure 3 (c). + +281 For comparison, we consider the three DRL construction heuristics, respectively developed in (Zhang +282 et al., 2020) called L2D, (Park et al., 2021b) by Park et al., and (Park et al., 2021a), called ScheduleNet. +283 We directly use the performance results of these methods for open benchmarks from their articles. +284 For simplicity, they are named L2D, Park and SchN respectively in this paper. We also include some +285 construction heuristics based PDR, such as MWKR, MOR, SPT and FIFO. Besides, to derive the +286 gaps to the optimum in all cases, OR-Tools serve as $\pi *$ as described in (6). +287 Now, let us analyze the performances of RS as follows. Table 2 shows the average makespan gaps +288 for each collection of JSP TA benchmarks with sizes, $1 5 { \times } 1 5$ , $2 0 \times 1 5$ , $2 0 \times 2 0$ , $3 0 \times 1 5$ , $3 0 \times 2 0$ , $5 0 \times 1 5$ , +289 $5 0 \times 2 0$ and $1 0 0 \times 2 0$ , where the best performances (the smallest gaps) are marked in bold. In general, +290 RS performs the best, and generally outperforms the other methods for all collections by large +291 margins, except for that it has slightly higher gaps than ${ \mathrm { R S + o p } }$ for the two collections, $1 5 \times 1 5$ and +292 $2 0 \times 2 0$ . In fact, ${ \mathrm { R S + o p } }$ also generally outperforms the rest of methods, except for that it is very +293 close to SchN for two collections. For the other six open benchmarks, ABZ, FT, ORB, YN, SWV +294 and LA, the performances are similar and thus presented in the appendix. It is concluded that RS +295 generally performs better than other construction heuristics by large margins. + +# 296 4.3 Experiments for FJSP + +Table 3: Average makespan gaps for FJSP open benchmarks + +
MethodMKLA(rdata)LA(edata)LA(vdata)
RS0.2320.0990.1460.031
RS+op0.2540.1130.1680.029
DRL-G0.2540.1110.1500.040
MWKR MOR0.282 0.296 0.4570.125 0.1470.149 0.1790.051 0.061
+ +297 For FJSP, we also train a model based on residual scheduling, named RS, and a ablation version, +298 named $\mathrm { R S + o p }$ , without removing irrelevant operations. We compares ours with one DRL construction +299 heuristics developed by (Song et al., 2023), called DRL-G, and four PDR-based heuristics, MOR, +00 MWKR, SPT and FIFO. We directly use the performance results of these methods for open datasets +01 according to the reports from (Song et al., 2023). +02 Table 3 shows the average makespan gaps in the four open benchmarks, MK, LA(rdata), LA(edata) +03 and LA(vdata). From the table, RS generally outperforms all the other methods for all benchmarks +04 by large margins, except for that ${ \mathsf { R S + o p } }$ is slightly better for the benchmark LA(vdata). + +# 05 5 Discussions + +In this paper, we propose a new approach, called residual scheduling, to solving JSP an FJSP problems, and the experiments show that our approach reaches SOTA among DRL-based construction heuristics on the above open JSP and FJSP benchmarks. We further discusses three issues: large instances, computation times and further improvement. + +![](images/1b83fd2ef9921abe986163a176919527de1940ffe5a722430f846b273b3a20da.jpg) +Figure 4: Average makespan gaps of JSP instances with different problem sizes. + +310 First, from the above experiments particularly for TA benchmark for JSP, we observe that the average +311 gaps gets smaller as the number of jobs increases, even if we use the same model trained with +312 $\bar { ( \cal N , M ) } = ( 1 0 , 1 0 )$ . In order to investigate size-agnostics, we further generate 13 collections of JSP +313 instances of sizes for testing, from $1 5 \times 1 5$ to $2 0 0 \times 2 0$ , and generate 10 instances for each collection +314 by using the procedure above. Figure 4 shows the average gaps for these collections for RS and L2D, +315 and these collections are listed in the order of sizes in the x-axis. Note that we only show the results +316 of L2D in addition to our RS, since L2D is the only open-source among the above DRL heuristics. +317 Interestingly, using RS, the average gaps are nearly zero for the collections with sizes larger than 100 +318 $\times 1 5$ , namely, $1 0 0 \times 1 5$ , $1 0 0 \times 2 0$ , $1 5 0 \times 1 5$ , $2 0 0 \times 1 5$ and $2 0 0 \times 2 0$ . Among the 50 JSP instances +319 in the five collections, 49 reaches zero gaps. A strong implication is that our RS approach can be +320 scaled up for job sizes and even reach the optimal for sufficient large job count. +321 Second, the computation times for RS are relatively small and has low variance like most of other +322 construction heuristics. Here, we just use the collection of TA $1 0 0 \mathrm { x } 2 0$ for illustration. It takes about +323 30 seconds on average for both RS and $\mathrm { R S + o p }$ , about 28 for L2D and about 444 for SchN. In contrast, +324 it takes about 4000 seconds with high variance for OR-tools. The times for other collections are listed +325 in more detail in the appendix. +326 Third, as proposed by Song et al. (2023), construction heuristics can further improve the gap by +327 constructing multiple solutions based on the softmax policy, in addition to the greedy policy. They +328 had a version constructing 100 solutions for FJSP, called $\mathrm { \ D R L + 1 0 0 }$ in this paper. In this paper, we +329 also implement a RS version for FJSP based on the softmax policy, as described in Subsection 3.3, +330 and then use the version, called ${ \mathrm { R S } } { + } 1 0 0 \ $ , to constructing 100 solutions. In Table 4, the experimental +331 results show that ${ \mathrm { R S } } { + } 1 0 0 \ $ performs the best, much better than RS, DRL-G and $\mathrm { \ D R L + 1 0 0 }$ . An +332 important property for such an improvement is that constructing multiple solutions can be done in +333 parallel. That is, for construction heuristics, the solution quality can be improved by adding more +334 computation powers. + +Table 4: Average makespan gaps for FJSP open benchmark. + +
MethodMKLA(rdata)LA(edata)LA(vdata)
RSRS+1000.2320.1540.0990.0470.1460.0790.0310.007
DRL-GDRL+1000.2540.1900.1110.0580.1500.0820.0400.014
+ +# References + +Majid Abdolrazzagh-Nezhad and Salwani Abdullah. 2017. Job Shop Scheduling: Classification, Constraints and Objective Functions. International Journal of Computer and Information Engineering 11, 4 (2017), 429–434. + +Joseph William Adams, Egon Balas, and Daniel J. Zawack. 1988. The Shifting Bottleneck Procedure for Job Shop Scheduling. Management science 34, 3 (1988), 391–401. + +David L. Applegate and William J. Cook. 1991. A Computational Study of the Job-Shop Scheduling Problem. INFORMS Journal on Computing 3, 2 (1991), 149–156. https://doi.org/10.1287/ ijoc.3.2.149 + +Peter W. Battaglia, Jessica B. Hamrick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinícius Flores Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner, Çaglar Gülçehre, H. Francis Song, Andrew J. Ballard, Justin Gilmer, George E. Dahl, Ashish Vaswani, Kelsey R. Allen, Charles Nash, Victoria Langston, Chris Dyer, Nicolas Heess, Daan Wierstra, Pushmeet Kohli, Matthew M. Botvinick, Oriol Vinyals, Yujia Li, and Razvan Pascanu. 2018. Relational inductive biases, deep learning, and graph networks. CoRR abs/1806.01261 (2018). arXiv:1806.01261 http://arxiv.org/abs/1806.01261 + +Dennis Behnke and Martin Josef Geiger. 2012. Test instances for the flexible job shop scheduling problem with work centers. Arbeitspapier/Research Paper/Helmut-Schmidt-Universität, Lehrstuhl für Betriebswirtschaftslehre, insbes. Logistik-Management (2012). + +Paolo Brandimarte. 1993. Routing and scheduling in a flexible job shop by tabu search. Ann. Oper. Res. 41, 3 (1993), 157–183. https://doi.org/10.1007/BF02023073 + +IBM ILOG Cplex. 2009. V12. 1: User’s Manual for CPLEX. International Business Machines Corporation 46, 53 (2009), 157. + +Mauro Dell’Amico and Marco Trubian. 1993. Applying tabu search to the job-shop scheduling problem. Annals of Operations Research 41, 3 (1993), 231–252. https://doi.org/10.1007/ BF02023076 + +Alhussein Fawzi, Matej Balog, Aja Huang, Thomas Hubert, Bernardino Romera-Paredes, Mohammadamin Barekatain, Alexander Novikov, Francisco J R Ruiz, Julian Schrittwieser, Grzegorz Swirszcz, et al. 2022. Discovering faster matrix multiplication algorithms with reinforcement learning. Nature 610, 7930 (2022), 47–53. + +Matthias Fey and Jan Eric Lenssen. 2019. Fast Graph Representation Learning with PyTorch Geometric, In ICLR Workshop on Representation Learning on Graphs and Manifolds. CoRR abs/1903.02428. arXiv:1903.02428 http://arxiv.org/abs/1903.02428 + +M. R. Garey and David S. Johnson. 1979. Computers and Intractability: A Guide to the Theory of NP-Completeness. W. H. Freeman, USA. + +Amit Kumar Gupta and Appa Iyer Sivakumar. 2006. Job shop scheduling techniques in semiconductor manufacturing. The International Journal of Advanced Manufacturing Technology 27, 11 (2006), 1163–1169. + +Reinhard Haupt. 1989. A survey of priority rule-based scheduling. Operations-Research-Spektrum 11, 1 (1989), 3–16. + +Johann Hurink, Bernd Jurisch, and Monika Thole. 1994. Tabu search for the job-shop scheduling problem with multi-purpose machines. Operations-Research-Spektrum 15 (1994), 205–215. + +Sergey Ioffe and Christian Szegedy. 2015. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. In International Conference on Machine Learning (ICML) (JMLR Workshop and Conference Proceedings, Vol. 37), Francis R. Bach and David M. Blei (Eds.). JMLR.org, 448–456. http://proceedings.mlr.press/v37/ioffe15.html + +381 Klaus Jansen, Monaldo Mastrolilli, and Roberto Solis-Oba. 2000. Approximation Algorithms for +382 Flexible Job Shop Problems. In Latin American Symposium on Theoretical Informatics (Lecture +383 Notes in Computer Science, Vol. 1776), Gaston H. Gonnet, Daniel Panario, and Alfredo Viola +384 (Eds.). Springer, 68–77. https://doi.org/10.1007/10719839_7 +385 BJ Lageweg, JK Lenstra, and AHG Rinnooy Kan. 1977. Job-shop scheduling by implicit enumeration. +386 Management Science 24, 4 (1977), 441–450. + +Stephen Lawrence. 1984. Resouce constrained project scheduling: An experimental investigation of heuristic scheduling techniques (Supplement). Graduate School of Industrial Administration, Carnegie-Mellon University (1984). + +90 Zhigang Lian, Bin Jiao, and Xingsheng Gu. 2006. A similar particle swarm optimization algorithm +91 for job-shop scheduling to minimize makespan. Appl. Math. Comput. 183, 2 (2006), 1008–1017. +92 https://doi.org/10.1016/j.amc.2006.05.168 +93 Chun-Cheng Lin, Der-Jiunn Deng, Yen-Ling Chih, and Hsin-Ting Chiu. 2019. Smart Manufacturing +394 Scheduling With Edge Computing Using Multiclass Deep Q Network. IEEE Trans. Ind. Informatics +395 15, 7 (2019), 4276–4284. https://doi.org/10.1109/TII.2019.2908210 +96 Min Liu, Zhi-jiang Sun, Junwei Yan, and Jing-song Kang. 2011. An adaptive annealing genetic +97 algorithm for the job-shop planning and scheduling problem. Expert Systems with Applications 38, +98 8 (2011), 9248–9255. https://doi.org/10.1016/j.eswa.2011.01.136 +99 Shu Luo. 2020. Dynamic scheduling for flexible job shop with new job insertions by deep reinforce +00 ment learning. Applied Soft Computing 91 (2020), 106208. https://doi.org/10.1016/j. +01 asoc.2020.106208 +402 Mingqi Lv, Zhaoxiong Hong, Ling Chen, Tieming Chen, Tiantian Zhu, and Shouling Ji. 2021. +403 Temporal Multi-Graph Convolutional Network for Traffic Flow Prediction. IEEE Transactions +404 on Intelligent Transportation Systems 22, 6 (2021), 3337–3348. https://doi.org/10.1109/ +405 TITS.2020.2983763 +406 Azalia Mirhoseini, Anna Goldie, Mustafa Yazgan, Joe Wenjie Jiang, Ebrahim M. Songhori, Shen +407 Wang, Young-Joon Lee, Eric Johnson, Omkar Pathak, Azade Nazi, Jiwoo Pak, Andy Tong, Kavya +408 Srinivasa, Will Hang, Emre Tuncer, Quoc V. Le, James Laudon, Richard Ho, Roger Carpenter, and +409 Jeff Dean. 2021. A graph placement methodology for fast chip design. Nature 594, 7862 (2021), +410 207–212. +411 Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Belle +412 mare, Alex Graves, Martin A. Riedmiller, Andreas Fidjeland, Georg Ostrovski, Stig Petersen, +413 Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, +414 Shane Legg, and Demis Hassabis. 2015. Human-level control through deep reinforcement learning. +415 Nature 518, 7540 (2015), 529–533. https://doi.org/10.1038/nature14236 + +416 J.F. Muth and G.L. Thompson. 1963. Industrial Scheduling. Prentice-Hall. + +417 In-Beom Park, Jaeseok Huh, Joongkyun Kim, and Jonghun Park. 2020. A Reinforcement Learning +418 Approach to Robust Scheduling of Semiconductor Manufacturing Facilities. IEEE Transactions on +419 Automation Science and Engineering 17, 3 (2020), 1420–1431. https://doi.org/10.1109/ +420 TASE.2019.2956762 +421 Junyoung Park, Sanjar Bakhtiyar, and Jinkyoo Park. 2021a. ScheduleNet: Learn to solve multi-agent +422 scheduling problems with reinforcement learning. CoRR abs/2106.03051 (2021). arXiv:2106.03051 +423 https://arxiv.org/abs/2106.03051 +424 Junyoung Park, Jaehyeong Chun, Sang Hun Kim, Youngkook Kim, and Jinkyoo Park. 2021b. +425 Learning to schedule job-shop problems: representation and policy learning using graph neural +426 network and reinforcement learning. International Journal of Production Research 59, 11 (2021), +427 3360–3377. https://doi.org/10.1080/00207543.2020.1870013 +429 Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Ed +430 ward Z. Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit +431 Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. 2019. PyTorch: An Imperative Style, High +432 Performance Deep Learning Library. In Neural Information Processing Systems (NeurIPS), +433 Hanna M. Wallach, Hugo Larochelle, Alina Beygelzimer, Florence d’Alché-Buc, Emily B. Fox, +434 and Roman Garnett (Eds.). 8024–8035. https://proceedings.neurips.cc/paper/2019/ +435 hash/bdbca288fee7f92f2bfa9f7012727740-Abstract.html +436 Laurent Perron and Vincent Furnon. 2019. OR-Tools. Google. https://developers.google. +437 com/optimization/ +438 Ferdinando Pezzella, Gianluca Morganti, and Giampiero Ciaschetti. 2008. A genetic algorithm for +439 the Flexible Job-shop Scheduling Problem. Computers and Operations Research 35, 10 (2008), +440 3202–3212. https://doi.org/10.1016/j.cor.2007.02.014 +441 Qing-dao-er-ji Ren and Yuping Wang. 2012. A new hybrid genetic algorithm for job shop scheduling +442 problem. Computers and Operations Research 39, 10 (2012), 2291–2299. https://doi.org/ +443 10.1016/j.cor.2011.12.005 +444 Mohammad Saidi-Mehrabad and Parviz Fattahi. 2007. Flexible job shop scheduling with tabu search +445 algorithms. The International Journal of Advanced Manufacturing Technology 32, 5 (2007), +446 563–570. +447 David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driess +448 che, Julian Schrittwieser, Ioannis Antonoglou, Vedavyas Panneershelvam, Marc Lanctot, Sander +449 Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya Sutskever, Timothy P. Lillicrap, +450 Madeleine Leach, Koray Kavukcuoglu, Thore Graepel, and Demis Hassabis. 2016. Mastering +451 the game of Go with deep neural networks and tree search. Nature 529, 7587 (2016), 484–489. +452 https://doi.org/10.1038/nature16961 +453 Wen Song, Xinyang Chen, Qiqiang Li, and Zhiguang Cao. 2023. Flexible Job-Shop Scheduling via +454 Graph Neural Network and Deep Reinforcement Learning. IEEE Trans. Ind. Informatics 19, 2 +455 (2023), 1600–1610. https://doi.org/10.1109/TII.2022.3189725 +456 Robert H. Storer, S. David Wu, and Renzo Vaccari. 1992. New search spaces for sequencing problems +457 with application to job shop scheduling. Management science 38, 10 (1992), 1495–1509. +458 Richard S. Sutton and Andrew G. Barto. 2018. Reinforcement Learning: An Introduction (second +459 ed.). The MIT Press. http://incompleteideas.net/book/the-book-2nd.html +460 Éric D. Taillard. 1993. Benchmarks for basic scheduling problems. european journal of operational +461 research 64, 2 (1993), 278–285. +462 Oriol Vinyals, Igor Babuschkin, Wojciech M. Czarnecki, Michaël Mathieu, Andrew Dudzik, Junyoung +463 Chung, David H. Choi, Richard Powell, Timo Ewalds, Petko Georgiev, Junhyuk Oh, Dan Horgan, +464 Manuel Kroiss, Ivo Danihelka, Aja Huang, Laurent Sifre, Trevor Cai, John P. Agapiou, Max +465 Jaderberg, Alexander Sasha Vezhnevets, Rémi Leblond, Tobias Pohlen, Valentin Dalibard, David +466 Budden, Yury Sulsky, James Molloy, Tom Le Paine, Çaglar Gülçehre, Ziyu Wang, Tobias Pfaff, +467 Yuhuai Wu, Roman Ring, Dani Yogatama, Dario Wünsch, Katrina McKinney, Oliver Smith, Tom +468 Schaul, Timothy P. Lillicrap, Koray Kavukcuoglu, Demis Hassabis, Chris Apps, and David Silver. +469 2019. Grandmaster level in StarCraft II using multi-agent reinforcement learning. Nature 575, +470 7782 (2019), 350–354. https://doi.org/10.1038/s41586-019-1724-z +471 Bernd Waschneck, Thomas Altenmüller, Thomas Bauernhansl, and Andreas Kyek. 2016. Production +472 Scheduling in Complex Job Shops from an Industry 4.0 Perspective: A Review and Challenges +473 in the Semiconductor Industry. In Proceedings of the 1st International Workshop on Science, +474 Application and Methods in Industry 4.0 (i-KNOW) (CEUR Workshop Proceedings, Vol. 1793), +475 Roman Kern, Gerald Reiner, and Olivia Bluder (Eds.). CEUR-WS.org. http://ceur-ws.org/ +476 Vol-1793/paper3.pdf +477 Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. 2019. How Powerful are Graph Neural +478 Networks? (2019). https://openreview.net/forum?id=ryGs6iA5Km +479 Takeshi Yamada and Ryohei Nakano. 1992. A Genetic Algorithm Applicable to Large-Scale Job +480 Shop Problems. In Parallel Problem Solving from Nature 2, (PPSN-II), Reinhard Männer and +481 Bernard Manderick (Eds.). Elsevier, 283–292. +482 Cong Zhang, Wen Song, Zhiguang Cao, Jie Zhang, Puay Siew Tan, and Chi Xu. 2020. Learning +483 to Dispatch for Job Shop Scheduling via Deep Reinforcement Learning. In Neural Information +484 Processing Systems (NeurIPS), Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria +485 Florina Balcan, and Hsuan-Tien Lin (Eds.). https://proceedings.neurips.cc/paper/ +486 2020/hash/11958dfee29b6709f48a9ba0387a2431-Abstract.html +487 Cong Zhang, Wen Song, Zhiguang Cao, Jie Zhang, Puay Siew Tan, and Chi Xu. 2022. Learning to +488 Search for Job Shop Scheduling via Deep Reinforcement Learning. CoRR abs/2211.10936 (2022). +489 https://doi.org/10.48550/arXiv.2211.10936 arXiv:2211.10936 +490 Cong Zhang, Yaoxin Wu, Yining Ma, Wen Song, Zhang Le, Zhiguang Cao, and Jie Zhang. 2023. A +491 review on learning to solve combinatorial optimisation problems in manufacturing. IET Collabora +492 tive Intelligent Manufacturing 5, 1 (2023), e12072. https://doi.org/10.1049/cim2.12072 +493 arXiv:https://ietresearch.onlinelibrary.wiley.com/doi/pdf/10.1049/cim2.12072 +494 Jie Zhou, Ganqu Cui, Shengding Hu, Zhengyan Zhang, Cheng Yang, Zhiyuan Liu, Lifeng Wang, +495 Changcheng Li, and Maosong Sun. 2020. Graph neural networks: A review of methods and +496 applications. AI Open 1 (2020), 57–81. https://doi.org/10.1016/j.aiopen.2021.01.001 \ No newline at end of file diff --git a/parse/dev/5uIL1E8h1E/5uIL1E8h1E_content_list.json b/parse/dev/5uIL1E8h1E/5uIL1E8h1E_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..8332a0d45b85bd62280374cd8e0440575d06f59c --- /dev/null +++ b/parse/dev/5uIL1E8h1E/5uIL1E8h1E_content_list.json @@ -0,0 +1,1382 @@ +[ + { + "type": "text", + "text": "Residual Scheduling: A New Reinforcement Learning Approach to Solving Job Shop Scheduling Problem ", + "text_level": 1, + "bbox": [ + 176, + 122, + 823, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Anonymous Author(s) \nAffiliation \nAddress \nemail ", + "bbox": [ + 423, + 226, + 578, + 281 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 318, + 535, + 334 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 Job-shop scheduling problem (JSP) is a mathematical optimization problem widely \n2 used in industries like manufacturing, and flexible JSP (FJSP) is also a common \n3 variant. Since they are NP-hard, it is intractable to find the optimal solution for \n4 all cases within reasonable times. Thus, it becomes important to develop efficient \n5 heuristics to solve JSP/FJSP. A kind of method of solving scheduling problems \n6 is construction heuristics, which constructs scheduling solutions via heuristics. \n7 Recently, many methods for construction heuristics leverage deep reinforcement \n8 learning (DRL) with graph neural networks (GNN). In this paper, we propose a new \n9 approach, named residual scheduling, to solving JSP/FJSP. In this new approach, \n10 we remove irrelevant machines and jobs such as those finished, such that the states \n11 include the remaining (or relevant) machines and jobs only. Our experiments show \n12 that our approach reaches state-of-the-art (SOTA) among all known construction \n13 heuristics on most well-known open JSP and FJSP benchmarks. In addition, we \n14 also observe that even though our model is trained for scheduling problems of \n15 smaller sizes, our method still performs well for scheduling problems of large sizes. \n16 Interestingly in our experiments, our approach even reaches zero gap for 49 among \n17 50 JSP instances whose job numbers are more than 150 on 20 machines. ", + "bbox": [ + 148, + 348, + 766, + 583 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "18 1 Introduction ", + "text_level": 1, + "bbox": [ + 148, + 608, + 312, + 626 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "19 The job-shop scheduling problem (JSP) is a mathematical optimization (MO) problem widely used in \n20 many industries, like manufacturing (Zhang et al., 2020; Waschneck et al., 2016). For example, a \n21 semiconductor manufacturing process can be viewed as a complex JSP problem (Waschneck et al., \n22 2016), where a set of given jobs are assigned to a set of machines under some constraints to achieve \n23 some expected goals such as minimizing makespan which is focused on in this paper. While there are \n24 many variants of JSP (Abdolrazzagh-Nezhad and Abdullah, 2017), we also consider an extension \n25 called flexible JSP (FJSP) where job operations can be done on designated machines. \n26 A generic approach to solving MO problems is to use mathematical programming, such as mixed \n27 integer linear programming (MILP) and constraint satisfaction problem (CSP). Two popular generic \n28 MO solvers for solving MO are OR-Tools (Perron and Furnon, 2019) and IBM ILOG CPLEX \n29 Optimizer (abbr. CPLEX) (Cplex, 2009). However, both JSP and FJSP, as well as many other MO \n30 problems, have been shown to be NP-hard (Garey and Johnson, 1979; Lageweg et al., 1977). That \n31 said, it is unrealistic and intractable to find the optimal solution for all cases within reasonable times. \n32 These tools can obtain the optimal solutions if sufficient time (or unlimited time) is given; otherwise, \n33 return best-effort solutions during the limited time, which usually have gaps to the optimum. When \n34 problems are scaled up, the gaps usually grow significantly. \n35 In practice, some heuristics (Gupta and Sivakumar, 2006; Haupt, 1989) or approximate methods \n36 (Jansen et al., 2000) were used to cope with the issue of intractability. A simple greedy approach is to \n37 use the heuristics following the so-called priority dispatching rule (PDR) (Haupt, 1989) to construct \n38 solutions. These can also be viewed as a kind of solution construction heuristics or construction \n39 heuristics. Some of PDR examples are First In First Out (FIFO), Shortest Processing Time (SPT), \n40 Most WorK Remaining (MWKR), and Most Operation Remaining (MOR). Although these heuristics \n41 are usually computationally fast, it is hard to design generally effective rules to minimize the gap to \n42 the optimum, and the derived results are usually far from the optimum. \n43 Furthermore, a generic approach to automating the design of heuristics is called metaheuristics, such \n44 as tabu search (Dell’Amico and Trubian, 1993; Saidi-Mehrabad and Fattahi, 2007) , genetic algorithm \n45 (GA) (Pezzella et al., 2008; Ren and Wang, 2012), and PSO algorithms (Lian et al., 2006; Liu et al., \n46 2011). However, metaheuristics still take a high computation time, and it is not ensured to obtain the \n47 optimal solution either. \n48 Recently, deep reinforcement learning (DRL) has made several significant successes for some \n49 applications, such as AlphaGo (Silver et al., 2016), AlphaStar (Vinyals et al., 2019), AlphaTensor \n50 (Fawzi et al., 2022), and thus it also attracted much attention in the MO problems, including chip \n51 design (Mirhoseini et al., 2021) and scheduling problems (Zhang et al., 2023). In the past, several \n52 researchers used DRL methods as construction heuristics, and their methods did improve scheduling \n53 performance, illustrated as follows. Park et al. (2020) proposed a method based on DQN (Mnih et al., \n54 2015) for JSP in semiconductor manufacturing and showed that their DQN model outperformed GA \n55 in terms of both scheduling performance (namely gap to the optimum on makespan) and computation \n56 time. Lin et al. (2019) and Luo (2020) proposed different DQN models to decide the scheduling action \n57 among the heuristic rules and improved the makespan and the tardiness over PDRs, respectively. \n58 A recent DRL-based approach to solving JSP/FJSP problems is to leverage graph neural networks \n59 (GNN) to design a size-agnostic representation (Zhang et al., 2020; Park et al., 2021b,a; Song et al., \n60 2023). In this approach, graph representation has better generalization ability in larger instances \n61 and provides a holistic view of scheduling states. Zhang et al. (2020) proposed a DRL method \n62 with disjunctive graph representation for JSP, called L2D (Learning to Dispatch), and used GNN \n63 to encode the graph for scheduling decision. Besides, Song et al. (2023) extended their methods \n64 to FJSP. Park et al. (2021b) used a similar strategy of (Zhang et al., 2020) but with different state \n65 features and model structure. Park et al. (2021a) proposed a new approach to solving JSP, called \n66 ScheduleNet, by using a different graph representation and a DRL model with the graph attention for \n67 scheduling decision. Most of the experiments above showed that their models trained from small \n68 instances still worked reasonably well for large test instances, and generally better than PDRs. Among \n69 these methods, ScheduleNet achieved state-of-the-art (SOTA) performance. There are still other \n70 DRL-based approaches to solving JSP/FJSP problems, but not construction heuristics. Zhang et al. \n71 (2022) proposes another approach, called Learning to Search (L2S), a kind of search-based heuristics. \n72 In this paper, we propose a new approach to solving JSP/FJSP, a kind of construction heuristics, also \n73 based on GNN. In this new approach, we remove irrelevant machines and jobs, such as those finished, \n74 such that the states include the remaining machines and jobs only. This approach is named residual \n75 scheduling in this paper to indicate to work on the remaining graph. \n76 Without irrelevant information, our experiments show that our approach reaches SOTA by outper \n77 forming the above mentioned construction methods on some well-known open benchmarks, seven \n78 for JSP and two for FJSP, as described in Section 4. We also observe that even though our model \n79 is trained for scheduling problems of smaller sizes, our method still performs well for scheduling \n80 problems of large sizes. Interestingly in our experiments, our approach even reaches zero gap for 49 \n81 among 50 JSP instances whose job numbers are more than 150 on 20 machines. ", + "bbox": [ + 147, + 638, + 825, + 736 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 742, + 825, + 867 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 148, + 873, + 825, + 901 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 92, + 825, + 174 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 181, + 825, + 250 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 145, + 256, + 825, + 395 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 145, + 401, + 825, + 594 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 601, + 825, + 656 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 662, + 825, + 746 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "82 2 Problem Formulation ", + "text_level": 1, + "bbox": [ + 153, + 766, + 385, + 784 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2.1 JSP and FJSP ", + "text_level": 1, + "bbox": [ + 169, + 797, + 312, + 813 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "A $n \\times m$ JSP instance contains $n$ jobs and $m$ machines. Each job $J _ { j }$ consists of a sequence of $k _ { j }$ operations $\\{ O _ { j , 1 } , \\dotsc , O _ { j , k _ { j } } \\}$ , where operation $O _ { j , i }$ must be started after $O _ { j , i - 1 }$ is finished. One machine can process at most one operation at a time, and preemption is not allowed upon processing operations. In JSP, one operation $O _ { j , i }$ is allowed to be processed on one designated machine, denoted by $M _ { j , i }$ , with a processing time, denoted by $T _ { j , i } ^ { ( o p ) }$ . Table 1 (a) illustrates a $3 \\times 3$ JSP instance, where the three jobs have 3, 3, 2 operations respectively, each of which is designated to be processed on ", + "bbox": [ + 161, + 824, + 823, + 911 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "90 one of the three machines $\\{ M _ { 1 } , M _ { 2 } , M _ { 3 } \\}$ in the table. A solution of a JSP instance is to dispatch all \n91 operations Oj,i to the corresponding machine Mj,i at time τ (s)j,i , such that the above constraints are \n92 \n93 While there are different expected goals, such as makespan, tardiness, etc., this paper focuses on \n94 makespan. Let the first osolution is defined to be $\\tau = 0$ in a JSP soluti all operations nitially. , where \n95 $T ^ { ( m k s p ) } = \\operatorname* { m a x } ( \\tau _ { j , i } ^ { ( c ) } )$ $O _ { j , i }$ $\\tau _ { j , i } ^ { ( c ) } = \\tau _ { j , i } ^ { ( s ) } + T _ { j , i } ^ { ( o p ) }$ \n96 denotes the completion time of $O _ { j , i }$ . The makespans for the two solutions illustrated in Figure 1 (a) \n97 and (b) are 12 and 15 respectively. The objective is to derive a solution that minimizes the makespan \n98 $T ^ { ( m k s p ) }$ , and the solution of Figure 1 (a) reaches the optimal. \n99 A $n \\times m$ FJSP instance is also a $n \\times m$ JSP instance with the following difference. In FJSP, \n100 all operations $O _ { j , i }$ are allowed to be dispatched to multiple designated machines with designated \n101 processing times. Table 1 (b) illustrates a $3 \\times 3$ FJSP instance, where multiple machines can be \n102 designated to be processed for one operation. Figure 1 (c) illustrates a solution of an FJSP instance, \n103 which takes a shorter time than that in Figure 1 (d). ", + "bbox": [ + 145, + 90, + 825, + 137 + ], + "page_idx": 2 + }, + { + "type": "table", + "img_path": "images/4261351e53a18de76aff84d9645da1b0e5c1ee65d6c98a9dfa3509a2e1823974.jpg", + "table_caption": [ + "Table 1: JSP and FJSP instances ", + "(b) A $3 \\times 3$ FJSP instance " + ], + "table_footnote": [], + "table_body": "
JobOperationM1M2M3
Job 1O1,132
O1.235
O1.343
Job 2O2,12
02.24
02.33
Job 3O3,134
03.222
", + "bbox": [ + 501, + 191, + 774, + 323 + ], + "page_idx": 2 + }, + { + "type": "table", + "img_path": "images/ee2298f6d1a4eedfbf6f7eea3fd20b3dd269d7f61439042e8181cd727720a23e.jpg", + "table_caption": [ + "(a) A $3 \\times 3$ JSP instance " + ], + "table_footnote": [], + "table_body": "
JobOperationM1M2M3
Job 1O1,13
O1.25
O1.34
Job 2O2,12
02,24
O2,33
Job 3O3,13
03.22
", + "bbox": [ + 220, + 193, + 490, + 323 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/54e0e1086515bc311b9d4a17fdee1c89c424b15d95726a4af33fc4a68fbe5ad1.jpg", + "image_caption": [ + "Figure 1: Both (a) and (b) are solutions of the $3 \\mathrm { x } 3$ JSP instance in Table 1 (a), and the former has the minimal makespan, 12. Both (c) and (d) are solutions of the $3 \\mathrm { x } 3$ FJSP instance in Table 1 (b), and the former has the minimal makespan, 9. " + ], + "image_footnote": [], + "bbox": [ + 178, + 364, + 813, + 445 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 517, + 825, + 608 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 614, + 825, + 684 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "104 2.2 Construction Heuristics ", + "text_level": 1, + "bbox": [ + 148, + 704, + 379, + 719 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "105 An approach to solving these scheduling problems is to construct solutions step by step in a greedy \n106 manner, and the heuristics based on this approach is called construction heuristics in this paper. In \n107 the approach of construction heuristics, a scheduling solution is constructed through a sequence of \n108 partial solutions in a chronicle order of dispatching operations step by step, defined as follows. The \n109 $t$ -th partial solution $S _ { t }$ associates with a dispatching time $\\tau _ { t }$ and includes a partial set of operations \n110 that have been dispatched by $\\tau _ { t }$ (inclusive) while satisfying the above JSP constraints, and all the \n111 remaining operations must be dispatched after $\\tau _ { t }$ (inclusive). The whole construction starts with $S _ { 0 }$ \n112 where none of operations have been dispatched and the dispatching time is $\\tau _ { 0 } = 0$ . For each $S _ { t }$ , a set \n113 of operations to be chosen for dispatching form a set of pairs of $( M , O )$ , called candidates $C _ { t }$ , where \n114 operations $O$ are allowed to be dispatched on machines $M$ at $\\tau _ { t }$ . An agent (or a heuristic algorithm) \n115 chooses one from candidates $C _ { t }$ for dispatching, and transits the partial solution to the next $S _ { t + 1 }$ . If \n116 there exists no operations for dispatching, the whole solution construction process is done and the \n117 partial solution is a solution, since no further operations are to be dispatched. \n118 Figure 2 illustrates a solution construction process for the 3x3 JSP instance in Table 1(a), constructed \n119 through nine partial solutions step by step. The initial partial solution $S _ { 0 }$ starts without any operations \n120 dispatched as in Figure 2 (a). The initial candidates $C _ { 0 }$ are $\\{ ( M _ { 1 } , O _ { 1 , 1 } ) , ( M _ { 3 } , O _ { 2 , 1 } ) , ( \\bar { M _ { 1 } } , O _ { 3 , 1 } ) \\}$ \n121 Following some heuristic, construct a solution from partial solution $S _ { 0 }$ to $S _ { 9 }$ step by step as in the \n122 Figure, where the dashed line in red indicate the time $\\tau _ { t }$ . The last one $S _ { 9 }$ , the same as the one in \n123 Figure 1 (a), is a solution, since all operations have been dispatched, and the last operation ends at \n124 time 12, the makespan of the solution. \n125 For FJSP, the process of solution construction is almost the same except for that one operation have \n126 multiple choices from candidates. Besides, an approach based on solution construction can be also \n127 viewed as the so-called Markov decision process $( M D P )$ , and the MDP formulation for solution \n128 construction is described in more detail in the appendix. ", + "bbox": [ + 140, + 731, + 825, + 911 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/cb96d32bb1f5dcecc876eaf54a9e7682b46451e9161fb74084ead5c2b39e2e49.jpg", + "image_caption": [ + "Figure 2: Solution construction, a sequence of partial solutions from $S _ { 0 }$ to $S _ { 8 }$ . " + ], + "image_footnote": [], + "bbox": [ + 194, + 90, + 787, + 367 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 421, + 825, + 518 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 525, + 825, + 582 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "129 3 Our Approach ", + "text_level": 1, + "bbox": [ + 145, + 603, + 326, + 621 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "130 In this section, we present a new approach, called residual scheduling, to solving scheduling problems. \n131 We introduce the residual scheduling in Subsection 3.1, describe the design of the graph representation \n132 in Subsection 3.2, propose a model architecture based on graph neural network in Subsection 3.3 and \n133 present a method to train this model in Subsection 3.4; ", + "bbox": [ + 142, + 637, + 825, + 693 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "34 3.1 Residual Scheduling ", + "text_level": 1, + "bbox": [ + 153, + 713, + 354, + 728 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "135 In our approach, the key is to remove irrelevant information, particularly for operations, from states \n136 (including partial solutions). An important benefit from this is that we do not need to include all \n137 irrelevant information while training to minimize the makespan. Let us illustrate by the state for the \n138 partial solution $S _ { 3 }$ at time $\\tau _ { 3 } = 3$ in Figure 2 (d). All processing by $\\tau _ { 3 }$ are irrelevant to the remaining \n139 scheduling. Since operations $O _ { 1 , 1 }$ and $O _ { 2 , 1 }$ are both finished and irrelevant the rest of scheduling, \n140 they can be removed from the state of $S _ { 3 }$ . In addition, operation $O _ { 2 , 2 }$ is dispatched at time 2 (before \n141 $\\tau _ { 3 } = 3$ ) and its processing time is $T _ { 2 , 1 } ^ { ( o p ) } = 4$ , so the operation is marked as ongoing. Thus, the \n142 operation can be modified to start at $\\tau _ { 3 } = 3$ with a processing time . Thus, the modified \n143 state for $S _ { 3 }$ do not contain both $O _ { 1 , 1 }$ and $O _ { 2 , 1 }$ , and modify $O _ { 2 , 2 }$ as above. Let us consider two more \n144 examples. For $S _ { 4 }$ , one more operation $O _ { 2 , 2 }$ is dispatched and thus marked as ongoing, however, the \n145 time $\\tau _ { 4 }$ remains unchanged and no more operations are removed. In this case, the state is almost the \n146 same except for including one more ongoing operation $O _ { 2 , 2 }$ . Then, for $S _ { 5 }$ , two more operations $O _ { 3 , 1 }$ \n147 and $O _ { 2 , 2 }$ are removed and the ongoing operation $O _ { 1 , 2 }$ changes its processing time to the remaining \n148 time (5-3). \n149 For residual scheduling, we also reset the dispatching time $\\tau = 0$ for all states with partial solutions \n150 modified as above, so we derive makespans which is also irrelevant to the earlier operations. Given \n151 a scheduling policy $\\pi$ , $T _ { \\pi } ^ { ( m k s p ) } ( S )$ is defined to be the makespan derived from an episode starting \n152 from states $S$ by following $\\pi$ , and $T _ { \\pi } ^ { ( m k s p ) } ( S , a )$ the makespan by taking action $a$ on $S$ . ", + "bbox": [ + 140, + 739, + 825, + 911 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 150, + 92, + 825, + 119 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 126, + 825, + 189 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "153 3.2 Residual Graph Representation ", + "text_level": 1, + "bbox": [ + 148, + 215, + 433, + 231 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "154 In this paper, our model design is based on graph neural network (GNN), and leverage GNN to \n155 extract the scheduling decision from the relationship in graph. In this subsection, we present the \n156 graph representation. Like many other researchers such as Park et al. (2021a), we formulate a partial \n157 solution into a graph $\\mathcal { G } = ( \\nu , \\mathcal { E } )$ , where $\\nu$ is a set of nodes and $\\mathcal { E }$ is a set of edges. A node is either a \n158 machine node $M$ or an operation node $O$ . An edge connects two nodes to represent the relationship \n159 between two nodes, basically including three kinds of edges, namely operation-to-operation $( O \\to O$ ), \n160 machine-to-operation $M O$ ) and operation-to-machine $( O \\to M$ ). All operations in the same \n161 job are fully connected as $O O$ edges. If an operation $O$ is able to be performed on a machine \n162 $M$ , there exists both $O \\to M$ and $M O$ directed edges. In (Park et al., 2021a), they also let all \n163 machines be fully connected as $M \\to M$ edges. However, our experiments in section 4 show that \n164 mutual $M \\to M$ edges do not help much based on our Residual Scheduling. An illustration for graph \n165 representation of $S _ { 3 }$ is depicted in Figure 3 (a). \n166 In the graph representation, all nodes need to include some attributes so that a partial solution $S$ at \n167 the dispatching time $\\tau$ can be supported in the MDP formulation (in the appendix). Note that many of \n168 the attributes below are normalized to reduce variance. For nodes corresponding to operations $O _ { j , i }$ , \n169 we have the following attributes: \n170 Status $\\phi _ { j , i }$ : The operation status $\\phi _ { j , i }$ is completed if the operation has been finished by $\\tau$ , ongoing if \n171 the operation is ongoing (i.e., has been dispatched to some machine by $\\tau$ and is still being processed \n172 at $\\tau$ ), ready if the operation designated to the machine which is idle has not been dispatched yet and \n173 its precedent operation has been finished, and unready otherwise. For example, in Figure 3 (a), the \n174 gray nodes are completed, the red ongoing, the yellow ready and the white unready. In our residual \n175 scheduling, there exists no completed operations in all partial solutions, since they are removes for \n176 irrelevance of the rest of scheduling. The attribute is a one-hot vector to represent the current status \n177 of the operation, which is one of ongoing, ready and unready. Illustration for all states $S _ { 0 }$ to $S _ { 8 }$ are \n178 shown in the appendix. \n179 Normalized processing time $\\bar { T } _ { j , i } ^ { ( o p ) }$ : Let the maximal processing time be $T _ { m a x } ^ { ( o p ) } = \\operatorname* { m a x } _ { \\forall j , i } ( T _ { j , i } ^ { ( o p ) } )$ \n180 Then, T¯(op)j,i $\\bar { T } _ { j , i } ^ { ( o p ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }$ . In our residual scheduling, the operations that have been finished are \n181 removed in partial solutions and therefore their processing time can be ignored; the operations that \n182 has not been dispatched yet still keep their processing times the same; the operations that are ongoing \n183 change their processing times to the remaining times after the dispatching time $\\tau _ { t }$ . As for FJSP, the \n184 operations that has not been dispatched yet may have several processing times on different machines, \n185 and thus we can simply choose the average of these processing times. ", + "bbox": [ + 140, + 246, + 826, + 412 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/a3a8af5ab1851b807add19a4a2a54f81ec9c7979dcede6e1a0cd5113aacd0d36.jpg", + "image_caption": [ + "Figure 3: Graph representation and networks. " + ], + "image_footnote": [], + "bbox": [ + 174, + 435, + 820, + 556 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 612, + 826, + 667 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 674, + 825, + 799 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 805, + 826, + 912 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "ob remaining time , and let the process $\\bar { T } _ { j , i } ^ { ( j o b ) }$ : Let the rest of pre for the whole job $J _ { j }$ $T _ { j , i } ^ { ( j o b ) } =$ $\\sum _ { \\forall i ^ { \\prime } \\geq i } T _ { j , i ^ { \\prime } } ^ { ( o p ) }$ aced by th ing timocessing time for the original jo $j$ $\\begin{array} { r } { T _ { j } ^ { ( j o b ) } = \\sum _ { \\forall i ^ { \\prime } } T _ { j , i ^ { \\prime } } ^ { ( o p ) } } \\end{array}$ . In practice, T ( j ob )j is $j$ $\\bar { T } _ { j , i } ^ { ( j o b ) } = T _ { j , i } ^ { ( j o b ) } / T _ { j } ^ { ( j o b ) }$ . For FJSP, since operations $O _ { j , i }$ can be dispatched to different designated machines $M _ { l }$ , say with the processing time T (op)j,i,l , we simply let T (op)j,i be the average of $T _ { j , i , l } ^ { ( o p ) }$ for all $M _ { l }$ . ", + "bbox": [ + 148, + 88, + 826, + 183 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "For machine nodes corresponding to machines $M _ { l }$ , we have the following attributes: ", + "bbox": [ + 166, + 185, + 725, + 202 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Machine status $\\phi _ { l }$ : The machine status $\\phi _ { l }$ is processing if some operation has been dispatched to and is being processed by $M _ { l }$ at $\\tau$ , and idle otherwise (no operation is being processed at $\\tau$ ). The attribute is a one-hot vector to represent the current status, which is one of processing and idle. ", + "bbox": [ + 173, + 207, + 825, + 250 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Normalized operation processing time $\\bar { T } _ { l } ^ { ( m a c ) }$ : On the machine $M _ { l }$ , the processing time $T _ { l } ^ { ( m a c ) }$ is $T _ { j , i } ^ { ( o p ) }$ (the same as the normalized processing time for node $O _ { j , i }$ ) if the machine status is processing, i.e., some ongoing operation $O _ { j , i }$ is being processed but not finished yet, is zero if the machine status is idle. Then, this attribute is normalized to $T _ { m a x } ^ { ( o p ) }$ and thus $\\bar { T } _ { l } ^ { ( m a c ) } = T _ { l } ^ { ( m a c ) } / T _ { m a x } ^ { ( o p ) }$ . ", + "bbox": [ + 173, + 256, + 826, + 327 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Now, consider edges in a residual scheduling graph. As described above, there exists three relationship sets for edges, $O O$ , $O \\to M$ and $M O$ . First, for the same job, say $J _ { j }$ , all of its operation nodes for $O _ { j , i }$ are fully connected. Note that for residual scheduling the operations finished by the dispatching time $\\tau$ are removed and thus have no edges to them. Second, a machine node for $M _ { l }$ is connected to an operation node for $O _ { j , i }$ , if the operation $O _ { j , i }$ is designated to be processed on the machine $M _ { l }$ , which forms two edges $\\dot { O } M$ and $M O$ . Both contains the following attribute. ", + "bbox": [ + 173, + 330, + 825, + 415 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Normalized operation processing tim e T¯(edge): The attribute is $\\bar { T } _ { j , i , l } ^ { ( e d g e ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }$ . Here, $T _ { j , i } ^ { ( o p ) } = T _ { j , i , l } ^ { ( o p ) }$ $O _ { j , i }$ $T _ { j , i } ^ { ( o p ) }$ is the remaining time as described above. ", + "bbox": [ + 173, + 420, + 826, + 474 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "3.3 Graph Neural Network ", + "text_level": 1, + "bbox": [ + 174, + 488, + 375, + 503 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In this subsection, we present our model based on graph neural network (GNN). GNN are a family of deep neural networks (Battaglia et al., 2018) that can learn representation of graph-structured data, widely used in many applications (Lv et al., 2021; Zhou et al., 2020). A GNN aggregates information from node itself and its neighboring nodes and then update the data itself, which allows the GNN to capture the complex relationships within the data graph. For GNN, we choose Graph Isomorphism Network (GIN), which was shown to have strong discriminative power ( $\\mathrm { { X u } }$ et al., 2019) and summarily reviewed as follows. Given a graph $\\mathcal { G } = ( \\nu , \\mathcal { E } )$ and $K$ GNN layers $K$ iterations), GIN performs the $k$ -th iterations of updating feature embedding $h ^ { ( k ) }$ for each node $v \\in \\mathcal V$ : ", + "bbox": [ + 153, + 513, + 826, + 627 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/2ff8cf48f7fc89b4b2026b4e2f6ba5a18a536573b1a5ce94368e07ebd887b7e6.jpg", + "text": "$$\nh _ { v } ^ { ( k ) } = M L P ^ { ( k ) } ( ( 1 + \\epsilon ^ { ( k ) } ) h _ { v } ^ { ( k - 1 ) } + \\sum _ { u \\in N _ { b } ( v ) } h _ { u } ^ { ( k - 1 ) } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 320, + 628, + 676, + 664 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "217 where $h _ { v } ^ { ( k ) }$ is the embedding of node $v$ at the $k$ -th layer, $\\epsilon ^ { ( k ) }$ is an arbitrary number that can be \n218 learned, and $N _ { b } ( v )$ is the neighbors of $v$ via edges in $\\mathcal { E }$ . Note that $h _ { v } ^ { ( 0 ) }$ refers to its raw features for \n219 input. $M L P ^ { ( k ) }$ is a Multi-Layer Perceptron $( M L P )$ for the $k$ -th layer with a batch normalization \n220 (Ioffe and Szegedy, 2015). \n221 Furthermore, we actually use heterogeneous GIN, also called HGIN, since there are two types of \n222 nodes, machine and operation nodes, and three relations, $O O$ , $O \\to M$ and $M O$ in the \n223 graph representation. Although we do not have cross machine relations $M \\to M$ as described above, \n224 updating machine nodes requires to include the update from itself as in (1), that is, there is also one \n225 more relation $M \\to M$ . Thus, HGIN encodes graph information between all relations by using the \n226 four MLPs as follows, ", + "bbox": [ + 140, + 667, + 825, + 731 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 138, + 736, + 826, + 819 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/392e5b294a0810c8146e8d439eb968e34c1d3cbeae97269e1e396ed5a736bbd6.jpg", + "text": "$$\nh _ { v } ^ { ( k + 1 ) } = \\sum _ { \\mathcal { R } } M L P _ { \\mathcal { R } } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { \\mathcal { R } } ^ { ( k + 1 ) } ) h _ { v } ^ { ( k ) } + \\sum _ { u \\in N _ { \\mathcal { R } } ( v ) } h _ { u } ^ { ( k ) } )\n$$", + "text_format": "latex", + "bbox": [ + 299, + 819, + 699, + 857 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where 227 $\\mathcal { R }$ is one of the above fo elatio $M L P _ { \\mathcal { R } } ^ { ( k ) }$ is the MLP for $\\mathcal { R }$ . For example, for $S _ { 0 }$ in Figure 2 (a), the embedding of $M _ { 1 }$ $( k + 1 )$ -st iteration can be derived as follows. ", + "bbox": [ + 135, + 859, + 828, + 891 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/3d184bbed84e82a3cd83570384051eeed279a8eb4a46bf70afb8b0f65157e291.jpg", + "text": "$$\nh _ { M _ { 1 } } ^ { ( k + 1 ) } = M L P _ { M M } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { M M } ^ { ( k + 1 ) } ) h _ { M _ { 1 } } ^ { ( k ) } ) + M L P _ { O M } ^ { ( k + 1 ) } ( h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } )\n$$", + "text_format": "latex", + "bbox": [ + 235, + 892, + 761, + 916 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "229 Similarly, the embedding of $O _ { 1 , 1 }$ in the $( k + 1 )$ -st iteration is: ", + "bbox": [ + 140, + 90, + 580, + 106 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/3a318d233ecc5b46d854239d39f00d2fdff89b88a439536327c31a2b3834eceb.jpg", + "text": "$$\nh _ { O _ { 1 , 1 } } ^ { ( k + 1 ) } = M L P _ { O O } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { O O } ^ { ( k + 1 ) } ) h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } ) + M L P _ { M O } ^ { ( k + 1 ) } ( h _ { M _ { 1 } } ^ { ( k ) } )\n$$", + "text_format": "latex", + "bbox": [ + 235, + 109, + 761, + 133 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "230 In our approach, an action includes the two phases, graph embedding phase and action selection phase. \n231 Let $h _ { \\mathcal { G } } ^ { ( k ) }$ denote the whole embedding of the graphs $\\mathcal { G }$ , a summation of the embeddings of all nodes, \n232 $h _ { v } ^ { ( k + 1 ) }$ . In the graph embedding phase, we use an HGIN to encode node and graph embeddings as \n233 described above. An example with three HGIN layers is illustrated in Figure 3 (b). ", + "bbox": [ + 140, + 142, + 825, + 208 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In the action selection phase, we select an action based on a policy, after node and graph embedding are encoded in the graph embedding phase. The policy is described as follows. First, collect all ready operations $O$ to be dispatched to machines $M$ . Then, for all pairs $( M , O )$ , feed their node embeddings $( h _ { M } ^ { ( k ) } , h _ { O } ^ { ( k ) } )$ into a MLP $S c o r e ( M , O )$ to calculate their scores as shown in Figure 3 (c). The probability of selecting $( M , O )$ is calculated based on a softmax function of all scores, which also serves as the model policy $\\pi$ for the current state. ", + "bbox": [ + 158, + 213, + 825, + 301 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "3.4 Policy-Based RL Training ", + "text_level": 1, + "bbox": [ + 168, + 315, + 393, + 330 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "41 In this paper, we propose to use a policy-based RL training mechanism that follows REINFORCE \n42 (Sutton and Barto, 2018) to update our model by policy gradient with a normalized advantage \n43 makespan with respect to a baseline policy $\\pi _ { b }$ as follows. ", + "bbox": [ + 150, + 340, + 825, + 383 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/b06d82d1c07484ee7e7eab67883f499d4e532d443dc6370d26bccc60092f3d81.jpg", + "text": "$$\nA _ { \\pi } ( S , a ) = { \\frac { T _ { \\pi _ { b } } ^ { ( m k s p ) } ( S , a ) - T _ { \\pi } ^ { ( m k s p ) } ( S , a ) } { T _ { \\pi _ { b } } ^ { ( m k s p ) } ( S , a ) } }\n$$", + "text_format": "latex", + "bbox": [ + 351, + 386, + 647, + 428 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "244 In this paper, we choose a lightweight PDR, MWKR, as baseline $\\pi _ { b }$ , which performed best for \n245 makespan among all PDRs reported from the previous work (Zhang et al., 2020). In fact, our \n246 experiment also shows that using MWKR is better than the other PDRs shown in the appendix. The \n247 model for policy $\\pi$ is parametrized by $\\theta$ , which is updated by $\\nabla _ { \\boldsymbol { \\theta } } l o g \\pi _ { \\boldsymbol { \\theta } } A _ { \\pi _ { \\boldsymbol { \\theta } } } \\bigl ( S _ { t } , a _ { t } \\bigr )$ . Our algorithm \n248 based on REINFORCE is listed in the appendix. ", + "bbox": [ + 140, + 435, + 825, + 507 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4 Experiments ", + "text_level": 1, + "bbox": [ + 166, + 525, + 312, + 542 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.1 Experimental Settings and Evaluation Benchmarks ", + "text_level": 1, + "bbox": [ + 160, + 555, + 568, + 570 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In our experiments, the settings of our model are described as follows. All embedding and hidden vectors in our model have a dimension of 256. The model contains three HGIN layers for graph embedding, and an MLP for the score function, as shown in Figure 3 (b) and (c). All MLP networks including those in HGIN and for score contain two hidden layers. The parameters of our model, such as MLP, generally follow the default settings in PyTorch (Paszke et al., 2019) and PyTorch Geometric (Fey and Lenssen, 2019). More settings are in the appendix. ", + "bbox": [ + 169, + 580, + 825, + 665 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Each of our models is trained with one million episodes, each with one scheduling instance. Each instance is generated by following the procedure which is used to generate the TA dataset (Taillard, 1993). Given $( N , M )$ , we use the procedure to generate an $n \\times m$ JSP instance by conforming to the following distribution, $n \\sim \\mathcal { U } ( \\bar { 3 } , N )$ , $m \\sim \\mathcal { U } ( 3 , n )$ , and operation count $k _ { j } = m$ , where $\\boldsymbol { \\mathcal { U } } ( \\boldsymbol { x } , \\boldsymbol { y } )$ represents a distribution that uniformly samples an integer in a close interval $[ x , y ]$ at random. The details of designation for machines and processing times refer to (Taillard, 1993) and thus are omitted here. We choose (10,10) for all experiments, since (10,10) generally performs better than the other two as described in the appendix. Following the method described in Subsection 3.4, the model is updated from the above randomly generated instances. For testing our models for JSP and FJSP, seven JSP open benchmarks and two FJSP open benchmarks are used, as listed in the appendix. ", + "bbox": [ + 161, + 670, + 825, + 809 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "67 The performance for a given policy method $\\pi$ on an instance is measured by the makespan gap $G$ \n68 defined as ", + "bbox": [ + 155, + 815, + 823, + 842 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/f381b1f4ca34a456d19d426492d9502ae08f10b6d9c37038fd76c7e426b573a4.jpg", + "text": "$$\nG = \\frac { T _ { \\pi } ^ { \\left( m k s p \\right) } - T _ { \\pi * } ^ { \\left( m k s p \\right) } } { T _ { \\pi * } ^ { \\left( m k s p \\right) } }\n$$", + "text_format": "latex", + "bbox": [ + 413, + 839, + 583, + 878 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where 269 $T _ { \\pi * } ^ { ( m k s p ) }$ is the optimal makespan or the best-effort makespan, from a mathematical optimization 270 tool, OR-Tools, serving as . By the best-effort makespan, we mean the makespan derived with a ", + "bbox": [ + 143, + 881, + 826, + 912 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/d571f08c06cb4d62d220b992e5665dea662faf647c622b0c5bde8cc35206e59a.jpg", + "table_caption": [ + "Table 2: Average makespan gaps for TA benchmarks. " + ], + "table_footnote": [], + "table_body": "
Size15×1520×1520×2030×1530×2050×1550×20100×20Aug.
RS0.1480.1650.1690.1440.1770.0670.1000.0260.0500.1250.150
RS+op0.1430.1930.1590.1920.2130.1230.126
MWKR0.1910.2050.2580.2330.2350.3280.2180.2170.2770.2390.2280.3520.2510.2490.3440.1680.1730.2410.2060.1790.1760.2550.2390.0830.0910.1440.1350.1950.1970.2750.254
MORSPT
FIFO0.2390.3140.2730.3110.311
L2DParkSchN0.2590.2010.1520.3000.3160.3290.3360.2230.2650.1360.0920.0660.2700.2210.161
0.2490.1940.2920.1720.2460.1900.3190.2370.1590.1380.2120.135
", + "bbox": [ + 173, + 109, + 834, + 257 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "271 sufficiently large time limitation, namely half a day with OR-Tools. For comparison in experiments, \n272 we use a server with Intel Xeon E5-2683 CPU and a single NVIDIA GeForce GTX 1080 Ti GPU. \n273 Our method uses a CPU thread and a GPU to train and evaluate, while OR-Tools uses eight threads \n274 to find the solution. ", + "bbox": [ + 142, + 281, + 826, + 337 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.2 Experiments for JSP ", + "text_level": 1, + "bbox": [ + 158, + 353, + 357, + 369 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "For JSP, we first train a model based on residual scheduling, named RS. For ablation testing, we also train a model, named $\\mathrm { R S + o p }$ , by following the same training method but without removing irrelevant operations. When using these models to solve testing instances, action selection is based on the greedy policy that simply chooses the action $( M , O )$ with the highest score deterministically, obtained from the score network as in Figure 3 (c). ", + "bbox": [ + 171, + 378, + 825, + 449 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "281 For comparison, we consider the three DRL construction heuristics, respectively developed in (Zhang \n282 et al., 2020) called L2D, (Park et al., 2021b) by Park et al., and (Park et al., 2021a), called ScheduleNet. \n283 We directly use the performance results of these methods for open benchmarks from their articles. \n284 For simplicity, they are named L2D, Park and SchN respectively in this paper. We also include some \n285 construction heuristics based PDR, such as MWKR, MOR, SPT and FIFO. Besides, to derive the \n286 gaps to the optimum in all cases, OR-Tools serve as $\\pi *$ as described in (6). \n287 Now, let us analyze the performances of RS as follows. Table 2 shows the average makespan gaps \n288 for each collection of JSP TA benchmarks with sizes, $1 5 { \\times } 1 5$ , $2 0 \\times 1 5$ , $2 0 \\times 2 0$ , $3 0 \\times 1 5$ , $3 0 \\times 2 0$ , $5 0 \\times 1 5$ , \n289 $5 0 \\times 2 0$ and $1 0 0 \\times 2 0$ , where the best performances (the smallest gaps) are marked in bold. In general, \n290 RS performs the best, and generally outperforms the other methods for all collections by large \n291 margins, except for that it has slightly higher gaps than ${ \\mathrm { R S + o p } }$ for the two collections, $1 5 \\times 1 5$ and \n292 $2 0 \\times 2 0$ . In fact, ${ \\mathrm { R S + o p } }$ also generally outperforms the rest of methods, except for that it is very \n293 close to SchN for two collections. For the other six open benchmarks, ABZ, FT, ORB, YN, SWV \n294 and LA, the performances are similar and thus presented in the appendix. It is concluded that RS \n295 generally performs better than other construction heuristics by large margins. ", + "bbox": [ + 142, + 454, + 825, + 537 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 143, + 544, + 825, + 670 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "296 4.3 Experiments for FJSP ", + "text_level": 1, + "bbox": [ + 142, + 685, + 367, + 700 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/c99b51216be96430166143b65903ab65a9561dbec698a97fd465ba2717ab21ec.jpg", + "table_caption": [ + "Table 3: Average makespan gaps for FJSP open benchmarks " + ], + "table_footnote": [], + "table_body": "
MethodMKLA(rdata)LA(edata)LA(vdata)
RS0.2320.0990.1460.031
RS+op0.2540.1130.1680.029
DRL-G0.2540.1110.1500.040
MWKR MOR0.282 0.296 0.4570.125 0.1470.149 0.1790.051 0.061
", + "bbox": [ + 303, + 738, + 694, + 856 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "297 For FJSP, we also train a model based on residual scheduling, named RS, and a ablation version, \n298 named $\\mathrm { R S + o p }$ , without removing irrelevant operations. We compares ours with one DRL construction \n299 heuristics developed by (Song et al., 2023), called DRL-G, and four PDR-based heuristics, MOR, \n00 MWKR, SPT and FIFO. We directly use the performance results of these methods for open datasets \n01 according to the reports from (Song et al., 2023). \n02 Table 3 shows the average makespan gaps in the four open benchmarks, MK, LA(rdata), LA(edata) \n03 and LA(vdata). From the table, RS generally outperforms all the other methods for all benchmarks \n04 by large margins, except for that ${ \\mathsf { R S + o p } }$ is slightly better for the benchmark LA(vdata). ", + "bbox": [ + 143, + 869, + 825, + 912 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 151, + 90, + 825, + 119 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 150, + 126, + 828, + 167 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "05 5 Discussions ", + "text_level": 1, + "bbox": [ + 156, + 190, + 300, + 207 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In this paper, we propose a new approach, called residual scheduling, to solving JSP an FJSP problems, and the experiments show that our approach reaches SOTA among DRL-based construction heuristics on the above open JSP and FJSP benchmarks. We further discusses three issues: large instances, computation times and further improvement. ", + "bbox": [ + 173, + 223, + 826, + 279 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/1b83fd2ef9921abe986163a176919527de1940ffe5a722430f846b273b3a20da.jpg", + "image_caption": [ + "Figure 4: Average makespan gaps of JSP instances with different problem sizes. " + ], + "image_footnote": [], + "bbox": [ + 174, + 292, + 825, + 388 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "310 First, from the above experiments particularly for TA benchmark for JSP, we observe that the average \n311 gaps gets smaller as the number of jobs increases, even if we use the same model trained with \n312 $\\bar { ( \\cal N , M ) } = ( 1 0 , 1 0 )$ . In order to investigate size-agnostics, we further generate 13 collections of JSP \n313 instances of sizes for testing, from $1 5 \\times 1 5$ to $2 0 0 \\times 2 0$ , and generate 10 instances for each collection \n314 by using the procedure above. Figure 4 shows the average gaps for these collections for RS and L2D, \n315 and these collections are listed in the order of sizes in the x-axis. Note that we only show the results \n316 of L2D in addition to our RS, since L2D is the only open-source among the above DRL heuristics. \n317 Interestingly, using RS, the average gaps are nearly zero for the collections with sizes larger than 100 \n318 $\\times 1 5$ , namely, $1 0 0 \\times 1 5$ , $1 0 0 \\times 2 0$ , $1 5 0 \\times 1 5$ , $2 0 0 \\times 1 5$ and $2 0 0 \\times 2 0$ . Among the 50 JSP instances \n319 in the five collections, 49 reaches zero gaps. A strong implication is that our RS approach can be \n320 scaled up for job sizes and even reach the optimal for sufficient large job count. \n321 Second, the computation times for RS are relatively small and has low variance like most of other \n322 construction heuristics. Here, we just use the collection of TA $1 0 0 \\mathrm { x } 2 0$ for illustration. It takes about \n323 30 seconds on average for both RS and $\\mathrm { R S + o p }$ , about 28 for L2D and about 444 for SchN. In contrast, \n324 it takes about 4000 seconds with high variance for OR-tools. The times for other collections are listed \n325 in more detail in the appendix. \n326 Third, as proposed by Song et al. (2023), construction heuristics can further improve the gap by \n327 constructing multiple solutions based on the softmax policy, in addition to the greedy policy. They \n328 had a version constructing 100 solutions for FJSP, called $\\mathrm { \\ D R L + 1 0 0 }$ in this paper. In this paper, we \n329 also implement a RS version for FJSP based on the softmax policy, as described in Subsection 3.3, \n330 and then use the version, called ${ \\mathrm { R S } } { + } 1 0 0 \\ $ , to constructing 100 solutions. In Table 4, the experimental \n331 results show that ${ \\mathrm { R S } } { + } 1 0 0 \\ $ performs the best, much better than RS, DRL-G and $\\mathrm { \\ D R L + 1 0 0 }$ . An \n332 important property for such an improvement is that constructing multiple solutions can be done in \n333 parallel. That is, for construction heuristics, the solution quality can be improved by adding more \n334 computation powers. ", + "bbox": [ + 138, + 429, + 826, + 582 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 587, + 825, + 657 + ], + "page_idx": 8 + }, + { + "type": "table", + "img_path": "images/89ca0c42d2768ed4edd4d57820fcff3afe69503db3148f90ca2752b714a6be6a.jpg", + "table_caption": [ + "Table 4: Average makespan gaps for FJSP open benchmark. " + ], + "table_footnote": [], + "table_body": "
MethodMKLA(rdata)LA(edata)LA(vdata)
RSRS+1000.2320.1540.0990.0470.1460.0790.0310.007
DRL-GDRL+1000.2540.1900.1110.0580.1500.0820.0400.014
", + "bbox": [ + 297, + 694, + 699, + 770 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 786, + 825, + 912 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "References ", + "text_level": 1, + "bbox": [ + 161, + 90, + 266, + 106 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Majid Abdolrazzagh-Nezhad and Salwani Abdullah. 2017. Job Shop Scheduling: Classification, Constraints and Objective Functions. International Journal of Computer and Information Engineering 11, 4 (2017), 429–434. ", + "bbox": [ + 171, + 114, + 825, + 156 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Joseph William Adams, Egon Balas, and Daniel J. Zawack. 1988. The Shifting Bottleneck Procedure for Job Shop Scheduling. Management science 34, 3 (1988), 391–401. ", + "bbox": [ + 165, + 167, + 825, + 196 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "David L. Applegate and William J. Cook. 1991. A Computational Study of the Job-Shop Scheduling Problem. INFORMS Journal on Computing 3, 2 (1991), 149–156. https://doi.org/10.1287/ ijoc.3.2.149 ", + "bbox": [ + 163, + 208, + 825, + 251 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Peter W. Battaglia, Jessica B. Hamrick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinícius Flores Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner, Çaglar Gülçehre, H. Francis Song, Andrew J. Ballard, Justin Gilmer, George E. Dahl, Ashish Vaswani, Kelsey R. Allen, Charles Nash, Victoria Langston, Chris Dyer, Nicolas Heess, Daan Wierstra, Pushmeet Kohli, Matthew M. Botvinick, Oriol Vinyals, Yujia Li, and Razvan Pascanu. 2018. Relational inductive biases, deep learning, and graph networks. CoRR abs/1806.01261 (2018). arXiv:1806.01261 http://arxiv.org/abs/1806.01261 ", + "bbox": [ + 161, + 261, + 825, + 359 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Dennis Behnke and Martin Josef Geiger. 2012. Test instances for the flexible job shop scheduling problem with work centers. Arbeitspapier/Research Paper/Helmut-Schmidt-Universität, Lehrstuhl für Betriebswirtschaftslehre, insbes. Logistik-Management (2012). ", + "bbox": [ + 156, + 371, + 821, + 414 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Paolo Brandimarte. 1993. Routing and scheduling in a flexible job shop by tabu search. Ann. Oper. Res. 41, 3 (1993), 157–183. https://doi.org/10.1007/BF02023073 ", + "bbox": [ + 166, + 425, + 825, + 454 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "IBM ILOG Cplex. 2009. V12. 1: User’s Manual for CPLEX. International Business Machines Corporation 46, 53 (2009), 157. ", + "bbox": [ + 161, + 464, + 825, + 494 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Mauro Dell’Amico and Marco Trubian. 1993. Applying tabu search to the job-shop scheduling problem. Annals of Operations Research 41, 3 (1993), 231–252. https://doi.org/10.1007/ BF02023076 ", + "bbox": [ + 165, + 505, + 823, + 547 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Alhussein Fawzi, Matej Balog, Aja Huang, Thomas Hubert, Bernardino Romera-Paredes, Mohammadamin Barekatain, Alexander Novikov, Francisco J R Ruiz, Julian Schrittwieser, Grzegorz Swirszcz, et al. 2022. Discovering faster matrix multiplication algorithms with reinforcement learning. Nature 610, 7930 (2022), 47–53. ", + "bbox": [ + 171, + 559, + 826, + 616 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Matthias Fey and Jan Eric Lenssen. 2019. Fast Graph Representation Learning with PyTorch Geometric, In ICLR Workshop on Representation Learning on Graphs and Manifolds. CoRR abs/1903.02428. arXiv:1903.02428 http://arxiv.org/abs/1903.02428 ", + "bbox": [ + 160, + 627, + 823, + 670 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "M. R. Garey and David S. Johnson. 1979. Computers and Intractability: A Guide to the Theory of NP-Completeness. W. H. Freeman, USA. ", + "bbox": [ + 160, + 680, + 826, + 710 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Amit Kumar Gupta and Appa Iyer Sivakumar. 2006. Job shop scheduling techniques in semiconductor manufacturing. The International Journal of Advanced Manufacturing Technology 27, 11 (2006), 1163–1169. ", + "bbox": [ + 168, + 720, + 823, + 763 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Reinhard Haupt. 1989. A survey of priority rule-based scheduling. Operations-Research-Spektrum 11, 1 (1989), 3–16. ", + "bbox": [ + 166, + 775, + 823, + 804 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Johann Hurink, Bernd Jurisch, and Monika Thole. 1994. Tabu search for the job-shop scheduling problem with multi-purpose machines. Operations-Research-Spektrum 15 (1994), 205–215. ", + "bbox": [ + 160, + 814, + 826, + 844 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Sergey Ioffe and Christian Szegedy. 2015. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. In International Conference on Machine Learning (ICML) (JMLR Workshop and Conference Proceedings, Vol. 37), Francis R. Bach and David M. Blei (Eds.). JMLR.org, 448–456. http://proceedings.mlr.press/v37/ioffe15.html ", + "bbox": [ + 171, + 856, + 826, + 912 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "381 Klaus Jansen, Monaldo Mastrolilli, and Roberto Solis-Oba. 2000. Approximation Algorithms for \n382 Flexible Job Shop Problems. In Latin American Symposium on Theoretical Informatics (Lecture \n383 Notes in Computer Science, Vol. 1776), Gaston H. Gonnet, Daniel Panario, and Alfredo Viola \n384 (Eds.). Springer, 68–77. https://doi.org/10.1007/10719839_7 \n385 BJ Lageweg, JK Lenstra, and AHG Rinnooy Kan. 1977. Job-shop scheduling by implicit enumeration. \n386 Management Science 24, 4 (1977), 441–450. ", + "bbox": [ + 142, + 90, + 826, + 147 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 148, + 159, + 825, + 188 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Stephen Lawrence. 1984. Resouce constrained project scheduling: An experimental investigation of heuristic scheduling techniques (Supplement). Graduate School of Industrial Administration, Carnegie-Mellon University (1984). ", + "bbox": [ + 163, + 200, + 823, + 242 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "90 Zhigang Lian, Bin Jiao, and Xingsheng Gu. 2006. A similar particle swarm optimization algorithm \n91 for job-shop scheduling to minimize makespan. Appl. Math. Comput. 183, 2 (2006), 1008–1017. \n92 https://doi.org/10.1016/j.amc.2006.05.168 \n93 Chun-Cheng Lin, Der-Jiunn Deng, Yen-Ling Chih, and Hsin-Ting Chiu. 2019. Smart Manufacturing \n394 Scheduling With Edge Computing Using Multiclass Deep Q Network. IEEE Trans. Ind. Informatics \n395 15, 7 (2019), 4276–4284. https://doi.org/10.1109/TII.2019.2908210 \n96 Min Liu, Zhi-jiang Sun, Junwei Yan, and Jing-song Kang. 2011. An adaptive annealing genetic \n97 algorithm for the job-shop planning and scheduling problem. Expert Systems with Applications 38, \n98 8 (2011), 9248–9255. https://doi.org/10.1016/j.eswa.2011.01.136 \n99 Shu Luo. 2020. Dynamic scheduling for flexible job shop with new job insertions by deep reinforce \n00 ment learning. Applied Soft Computing 91 (2020), 106208. https://doi.org/10.1016/j. \n01 asoc.2020.106208 \n402 Mingqi Lv, Zhaoxiong Hong, Ling Chen, Tieming Chen, Tiantian Zhu, and Shouling Ji. 2021. \n403 Temporal Multi-Graph Convolutional Network for Traffic Flow Prediction. IEEE Transactions \n404 on Intelligent Transportation Systems 22, 6 (2021), 3337–3348. https://doi.org/10.1109/ \n405 TITS.2020.2983763 \n406 Azalia Mirhoseini, Anna Goldie, Mustafa Yazgan, Joe Wenjie Jiang, Ebrahim M. Songhori, Shen \n407 Wang, Young-Joon Lee, Eric Johnson, Omkar Pathak, Azade Nazi, Jiwoo Pak, Andy Tong, Kavya \n408 Srinivasa, Will Hang, Emre Tuncer, Quoc V. Le, James Laudon, Richard Ho, Roger Carpenter, and \n409 Jeff Dean. 2021. A graph placement methodology for fast chip design. Nature 594, 7862 (2021), \n410 207–212. \n411 Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Belle \n412 mare, Alex Graves, Martin A. Riedmiller, Andreas Fidjeland, Georg Ostrovski, Stig Petersen, \n413 Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, \n414 Shane Legg, and Demis Hassabis. 2015. Human-level control through deep reinforcement learning. \n415 Nature 518, 7540 (2015), 529–533. https://doi.org/10.1038/nature14236 ", + "bbox": [ + 151, + 255, + 826, + 297 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 150, + 309, + 826, + 352 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 151, + 364, + 826, + 406 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 151, + 417, + 828, + 460 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 473, + 826, + 529 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 541, + 826, + 611 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 623, + 828, + 694 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "416 J.F. Muth and G.L. Thompson. 1963. Industrial Scheduling. Prentice-Hall. ", + "bbox": [ + 142, + 705, + 663, + 720 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "417 In-Beom Park, Jaeseok Huh, Joongkyun Kim, and Jonghun Park. 2020. A Reinforcement Learning \n418 Approach to Robust Scheduling of Semiconductor Manufacturing Facilities. IEEE Transactions on \n419 Automation Science and Engineering 17, 3 (2020), 1420–1431. https://doi.org/10.1109/ \n420 TASE.2019.2956762 \n421 Junyoung Park, Sanjar Bakhtiyar, and Jinkyoo Park. 2021a. ScheduleNet: Learn to solve multi-agent \n422 scheduling problems with reinforcement learning. CoRR abs/2106.03051 (2021). arXiv:2106.03051 \n423 https://arxiv.org/abs/2106.03051 \n424 Junyoung Park, Jaehyeong Chun, Sang Hun Kim, Youngkook Kim, and Jinkyoo Park. 2021b. \n425 Learning to schedule job-shop problems: representation and policy learning using graph neural \n426 network and reinforcement learning. International Journal of Production Research 59, 11 (2021), \n427 3360–3377. https://doi.org/10.1080/00207543.2020.1870013 \n429 Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Ed \n430 ward Z. Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit \n431 Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. 2019. PyTorch: An Imperative Style, High \n432 Performance Deep Learning Library. In Neural Information Processing Systems (NeurIPS), \n433 Hanna M. Wallach, Hugo Larochelle, Alina Beygelzimer, Florence d’Alché-Buc, Emily B. Fox, \n434 and Roman Garnett (Eds.). 8024–8035. https://proceedings.neurips.cc/paper/2019/ \n435 hash/bdbca288fee7f92f2bfa9f7012727740-Abstract.html \n436 Laurent Perron and Vincent Furnon. 2019. OR-Tools. Google. https://developers.google. \n437 com/optimization/ \n438 Ferdinando Pezzella, Gianluca Morganti, and Giampiero Ciaschetti. 2008. A genetic algorithm for \n439 the Flexible Job-shop Scheduling Problem. Computers and Operations Research 35, 10 (2008), \n440 3202–3212. https://doi.org/10.1016/j.cor.2007.02.014 \n441 Qing-dao-er-ji Ren and Yuping Wang. 2012. A new hybrid genetic algorithm for job shop scheduling \n442 problem. Computers and Operations Research 39, 10 (2012), 2291–2299. https://doi.org/ \n443 10.1016/j.cor.2011.12.005 \n444 Mohammad Saidi-Mehrabad and Parviz Fattahi. 2007. Flexible job shop scheduling with tabu search \n445 algorithms. The International Journal of Advanced Manufacturing Technology 32, 5 (2007), \n446 563–570. \n447 David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driess \n448 che, Julian Schrittwieser, Ioannis Antonoglou, Vedavyas Panneershelvam, Marc Lanctot, Sander \n449 Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya Sutskever, Timothy P. Lillicrap, \n450 Madeleine Leach, Koray Kavukcuoglu, Thore Graepel, and Demis Hassabis. 2016. Mastering \n451 the game of Go with deep neural networks and tree search. Nature 529, 7587 (2016), 484–489. \n452 https://doi.org/10.1038/nature16961 \n453 Wen Song, Xinyang Chen, Qiqiang Li, and Zhiguang Cao. 2023. Flexible Job-Shop Scheduling via \n454 Graph Neural Network and Deep Reinforcement Learning. IEEE Trans. Ind. Informatics 19, 2 \n455 (2023), 1600–1610. https://doi.org/10.1109/TII.2022.3189725 \n456 Robert H. Storer, S. David Wu, and Renzo Vaccari. 1992. New search spaces for sequencing problems \n457 with application to job shop scheduling. Management science 38, 10 (1992), 1495–1509. \n458 Richard S. Sutton and Andrew G. Barto. 2018. Reinforcement Learning: An Introduction (second \n459 ed.). The MIT Press. http://incompleteideas.net/book/the-book-2nd.html \n460 Éric D. Taillard. 1993. Benchmarks for basic scheduling problems. european journal of operational \n461 research 64, 2 (1993), 278–285. \n462 Oriol Vinyals, Igor Babuschkin, Wojciech M. Czarnecki, Michaël Mathieu, Andrew Dudzik, Junyoung \n463 Chung, David H. Choi, Richard Powell, Timo Ewalds, Petko Georgiev, Junhyuk Oh, Dan Horgan, \n464 Manuel Kroiss, Ivo Danihelka, Aja Huang, Laurent Sifre, Trevor Cai, John P. Agapiou, Max \n465 Jaderberg, Alexander Sasha Vezhnevets, Rémi Leblond, Tobias Pohlen, Valentin Dalibard, David \n466 Budden, Yury Sulsky, James Molloy, Tom Le Paine, Çaglar Gülçehre, Ziyu Wang, Tobias Pfaff, \n467 Yuhuai Wu, Roman Ring, Dani Yogatama, Dario Wünsch, Katrina McKinney, Oliver Smith, Tom \n468 Schaul, Timothy P. Lillicrap, Koray Kavukcuoglu, Demis Hassabis, Chris Apps, and David Silver. \n469 2019. Grandmaster level in StarCraft II using multi-agent reinforcement learning. Nature 575, \n470 7782 (2019), 350–354. https://doi.org/10.1038/s41586-019-1724-z \n471 Bernd Waschneck, Thomas Altenmüller, Thomas Bauernhansl, and Andreas Kyek. 2016. Production \n472 Scheduling in Complex Job Shops from an Industry 4.0 Perspective: A Review and Challenges \n473 in the Semiconductor Industry. In Proceedings of the 1st International Workshop on Science, \n474 Application and Methods in Industry 4.0 (i-KNOW) (CEUR Workshop Proceedings, Vol. 1793), \n475 Roman Kern, Gerald Reiner, and Olivia Bluder (Eds.). CEUR-WS.org. http://ceur-ws.org/ \n476 Vol-1793/paper3.pdf \n477 Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. 2019. How Powerful are Graph Neural \n478 Networks? (2019). https://openreview.net/forum?id=ryGs6iA5Km \n479 Takeshi Yamada and Ryohei Nakano. 1992. A Genetic Algorithm Applicable to Large-Scale Job \n480 Shop Problems. In Parallel Problem Solving from Nature 2, (PPSN-II), Reinhard Männer and \n481 Bernard Manderick (Eds.). Elsevier, 283–292. \n482 Cong Zhang, Wen Song, Zhiguang Cao, Jie Zhang, Puay Siew Tan, and Chi Xu. 2020. Learning \n483 to Dispatch for Job Shop Scheduling via Deep Reinforcement Learning. In Neural Information \n484 Processing Systems (NeurIPS), Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria \n485 Florina Balcan, and Hsuan-Tien Lin (Eds.). https://proceedings.neurips.cc/paper/ \n486 2020/hash/11958dfee29b6709f48a9ba0387a2431-Abstract.html \n487 Cong Zhang, Wen Song, Zhiguang Cao, Jie Zhang, Puay Siew Tan, and Chi Xu. 2022. Learning to \n488 Search for Job Shop Scheduling via Deep Reinforcement Learning. CoRR abs/2211.10936 (2022). \n489 https://doi.org/10.48550/arXiv.2211.10936 arXiv:2211.10936 \n490 Cong Zhang, Yaoxin Wu, Yining Ma, Wen Song, Zhang Le, Zhiguang Cao, and Jie Zhang. 2023. A \n491 review on learning to solve combinatorial optimisation problems in manufacturing. IET Collabora \n492 tive Intelligent Manufacturing 5, 1 (2023), e12072. https://doi.org/10.1049/cim2.12072 \n493 arXiv:https://ietresearch.onlinelibrary.wiley.com/doi/pdf/10.1049/cim2.12072 \n494 Jie Zhou, Ganqu Cui, Shengding Hu, Zhengyan Zhang, Cheng Yang, Zhiyuan Liu, Lifeng Wang, \n495 Changcheng Li, and Maosong Sun. 2020. Graph neural networks: A review of methods and \n496 applications. AI Open 1 (2020), 57–81. https://doi.org/10.1016/j.aiopen.2021.01.001 ", + "bbox": [ + 142, + 733, + 826, + 787 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 800, + 825, + 843 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 856, + 826, + 911 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 137, + 104, + 828, + 921 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "", + "bbox": [ + 137, + 90, + 828, + 382 + ], + "page_idx": 12 + } +] \ No newline at end of file diff --git a/parse/dev/5uIL1E8h1E/5uIL1E8h1E_middle.json b/parse/dev/5uIL1E8h1E/5uIL1E8h1E_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..aa4d51d0b7516a6de149c49d79550446e65a1365 --- /dev/null +++ b/parse/dev/5uIL1E8h1E/5uIL1E8h1E_middle.json @@ -0,0 +1,48989 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 97, + 504, + 137 + ], + "lines": [ + { + "bbox": [ + 106, + 95, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 106, + 95, + 505, + 120 + ], + "score": 1.0, + "content": "Residual Scheduling: A New Reinforcement Learning", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 118, + 495, + 139 + ], + "spans": [ + { + "bbox": [ + 116, + 118, + 495, + 139 + ], + "score": 1.0, + "content": "Approach to Solving Job Shop Scheduling Problem", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 259, + 179, + 354, + 223 + ], + "lines": [ + { + "bbox": [ + 258, + 178, + 356, + 191 + ], + "spans": [ + { + "bbox": [ + 258, + 178, + 356, + 191 + ], + "score": 1.0, + "content": "Anonymous Author(s)", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 283, + 189, + 328, + 202 + ], + "spans": [ + { + "bbox": [ + 283, + 189, + 328, + 202 + ], + "score": 1.0, + "content": "Affiliation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 286, + 200, + 324, + 213 + ], + "spans": [ + { + "bbox": [ + 286, + 200, + 324, + 213 + ], + "score": 1.0, + "content": "Address", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 290, + 213, + 320, + 222 + ], + "spans": [ + { + "bbox": [ + 290, + 213, + 320, + 222 + ], + "score": 1.0, + "content": "email", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 283, + 252, + 328, + 265 + ], + "lines": [ + { + "bbox": [ + 281, + 251, + 331, + 267 + ], + "spans": [ + { + "bbox": [ + 281, + 251, + 331, + 267 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 91, + 276, + 469, + 462 + ], + "lines": [ + { + "bbox": [ + 92, + 276, + 470, + 290 + ], + "spans": [ + { + "bbox": [ + 92, + 280, + 99, + 288 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 141, + 276, + 470, + 290 + ], + "score": 1.0, + "content": "Job-shop scheduling problem (JSP) is a mathematical optimization problem widely", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 92, + 287, + 470, + 299 + ], + "spans": [ + { + "bbox": [ + 92, + 290, + 99, + 298 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 142, + 287, + 470, + 299 + ], + "score": 1.0, + "content": "used in industries like manufacturing, and flexible JSP (FJSP) is also a common", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 93, + 298, + 469, + 311 + ], + "spans": [ + { + "bbox": [ + 93, + 301, + 99, + 309 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 142, + 298, + 469, + 311 + ], + "score": 1.0, + "content": "variant. Since they are NP-hard, it is intractable to find the optimal solution for", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 92, + 308, + 470, + 322 + ], + "spans": [ + { + "bbox": [ + 92, + 312, + 99, + 321 + ], + "score": 1.0, + "content": "4", + "type": "text" + }, + { + "bbox": [ + 141, + 308, + 470, + 322 + ], + "score": 1.0, + "content": "all cases within reasonable times. Thus, it becomes important to develop efficient", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 93, + 320, + 470, + 333 + ], + "spans": [ + { + "bbox": [ + 93, + 322, + 99, + 331 + ], + "score": 1.0, + "content": "5", + "type": "text" + }, + { + "bbox": [ + 141, + 320, + 470, + 333 + ], + "score": 1.0, + "content": "heuristics to solve JSP/FJSP. A kind of method of solving scheduling problems", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 92, + 331, + 470, + 343 + ], + "spans": [ + { + "bbox": [ + 92, + 333, + 99, + 343 + ], + "score": 1.0, + "content": "6", + "type": "text" + }, + { + "bbox": [ + 141, + 331, + 470, + 343 + ], + "score": 1.0, + "content": "is construction heuristics, which constructs scheduling solutions via heuristics.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 92, + 342, + 470, + 355 + ], + "spans": [ + { + "bbox": [ + 92, + 345, + 99, + 353 + ], + "score": 1.0, + "content": "7", + "type": "text" + }, + { + "bbox": [ + 141, + 342, + 470, + 355 + ], + "score": 1.0, + "content": "Recently, many methods for construction heuristics leverage deep reinforcement", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 92, + 352, + 470, + 366 + ], + "spans": [ + { + "bbox": [ + 92, + 356, + 99, + 364 + ], + "score": 1.0, + "content": "8", + "type": "text" + }, + { + "bbox": [ + 141, + 352, + 470, + 366 + ], + "score": 1.0, + "content": "learning (DRL) with graph neural networks (GNN). In this paper, we propose a new", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 92, + 364, + 470, + 377 + ], + "spans": [ + { + "bbox": [ + 92, + 366, + 99, + 375 + ], + "score": 1.0, + "content": "9", + "type": "text" + }, + { + "bbox": [ + 141, + 364, + 470, + 377 + ], + "score": 1.0, + "content": "approach, named residual scheduling, to solving JSP/FJSP. In this new approach,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 375, + 469, + 387 + ], + "spans": [ + { + "bbox": [ + 90, + 377, + 99, + 386 + ], + "score": 1.0, + "content": "10", + "type": "text" + }, + { + "bbox": [ + 141, + 375, + 469, + 387 + ], + "score": 1.0, + "content": "we remove irrelevant machines and jobs such as those finished, such that the states", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 90, + 385, + 470, + 398 + ], + "spans": [ + { + "bbox": [ + 90, + 388, + 99, + 397 + ], + "score": 1.0, + "content": "11", + "type": "text" + }, + { + "bbox": [ + 141, + 385, + 470, + 398 + ], + "score": 1.0, + "content": "include the remaining (or relevant) machines and jobs only. Our experiments show", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 396, + 470, + 409 + ], + "spans": [ + { + "bbox": [ + 90, + 399, + 99, + 408 + ], + "score": 1.0, + "content": "12", + "type": "text" + }, + { + "bbox": [ + 141, + 396, + 470, + 409 + ], + "score": 1.0, + "content": "that our approach reaches state-of-the-art (SOTA) among all known construction", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 408, + 470, + 420 + ], + "spans": [ + { + "bbox": [ + 90, + 410, + 99, + 419 + ], + "score": 1.0, + "content": "13", + "type": "text" + }, + { + "bbox": [ + 141, + 408, + 470, + 420 + ], + "score": 1.0, + "content": "heuristics on most well-known open JSP and FJSP benchmarks. In addition, we", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 419, + 470, + 430 + ], + "spans": [ + { + "bbox": [ + 89, + 421, + 99, + 430 + ], + "score": 1.0, + "content": "14", + "type": "text" + }, + { + "bbox": [ + 142, + 419, + 470, + 430 + ], + "score": 1.0, + "content": "also observe that even though our model is trained for scheduling problems of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 88, + 429, + 471, + 442 + ], + "spans": [ + { + "bbox": [ + 88, + 431, + 100, + 442 + ], + "score": 1.0, + "content": "15", + "type": "text" + }, + { + "bbox": [ + 141, + 429, + 471, + 442 + ], + "score": 1.0, + "content": "smaller sizes, our method still performs well for scheduling problems of large sizes.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 90, + 439, + 470, + 453 + ], + "spans": [ + { + "bbox": [ + 90, + 442, + 99, + 451 + ], + "score": 1.0, + "content": "16", + "type": "text" + }, + { + "bbox": [ + 141, + 439, + 470, + 453 + ], + "score": 1.0, + "content": "Interestingly in our experiments, our approach even reaches zero gap for 49 among", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 90, + 452, + 433, + 463 + ], + "spans": [ + { + "bbox": [ + 90, + 453, + 99, + 463 + ], + "score": 1.0, + "content": "17", + "type": "text" + }, + { + "bbox": [ + 142, + 452, + 433, + 463 + ], + "score": 1.0, + "content": "50 JSP instances whose job numbers are more than 150 on 20 machines.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 91, + 482, + 191, + 496 + ], + "lines": [ + { + "bbox": [ + 87, + 481, + 192, + 498 + ], + "spans": [ + { + "bbox": [ + 87, + 481, + 192, + 498 + ], + "score": 1.0, + "content": "18 1 Introduction", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 90, + 506, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 90, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 90, + 509, + 100, + 518 + ], + "score": 1.0, + "content": "19", + "type": "text" + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "The job-shop scheduling problem (JSP) is a mathematical optimization (MO) problem widely used in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 518, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 89, + 520, + 100, + 529 + ], + "score": 1.0, + "content": "20", + "type": "text" + }, + { + "bbox": [ + 105, + 518, + 506, + 531 + ], + "score": 1.0, + "content": "many industries, like manufacturing (Zhang et al., 2020; Waschneck et al., 2016). For example, a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 529, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 89, + 531, + 100, + 540 + ], + "score": 1.0, + "content": "21", + "type": "text" + }, + { + "bbox": [ + 105, + 529, + 506, + 541 + ], + "score": 1.0, + "content": "semiconductor manufacturing process can be viewed as a complex JSP problem (Waschneck et al.,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 540, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 89, + 542, + 100, + 551 + ], + "score": 1.0, + "content": "22", + "type": "text" + }, + { + "bbox": [ + 106, + 540, + 506, + 552 + ], + "score": 1.0, + "content": "2016), where a set of given jobs are assigned to a set of machines under some constraints to achieve", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 551, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 89, + 553, + 100, + 562 + ], + "score": 1.0, + "content": "23", + "type": "text" + }, + { + "bbox": [ + 106, + 551, + 506, + 563 + ], + "score": 1.0, + "content": "some expected goals such as minimizing makespan which is focused on in this paper. While there are", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 89, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 89, + 564, + 100, + 572 + ], + "score": 1.0, + "content": "24", + "type": "text" + }, + { + "bbox": [ + 106, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "many variants of JSP (Abdolrazzagh-Nezhad and Abdullah, 2017), we also consider an extension", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 89, + 572, + 449, + 585 + ], + "spans": [ + { + "bbox": [ + 89, + 574, + 100, + 583 + ], + "score": 1.0, + "content": "25", + "type": "text" + }, + { + "bbox": [ + 106, + 572, + 449, + 585 + ], + "score": 1.0, + "content": "called flexible JSP (FJSP) where job operations can be done on designated machines.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 90, + 588, + 505, + 687 + ], + "lines": [ + { + "bbox": [ + 89, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 89, + 591, + 99, + 600 + ], + "score": 1.0, + "content": "26", + "type": "text" + }, + { + "bbox": [ + 105, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "A generic approach to solving MO problems is to use mathematical programming, such as mixed", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 89, + 601, + 99, + 611 + ], + "score": 1.0, + "content": "27", + "type": "text" + }, + { + "bbox": [ + 105, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "integer linear programming (MILP) and constraint satisfaction problem (CSP). Two popular generic", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 89, + 610, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 89, + 612, + 99, + 622 + ], + "score": 1.0, + "content": "28", + "type": "text" + }, + { + "bbox": [ + 105, + 610, + 506, + 623 + ], + "score": 1.0, + "content": "MO solvers for solving MO are OR-Tools (Perron and Furnon, 2019) and IBM ILOG CPLEX", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 88, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 88, + 623, + 100, + 634 + ], + "score": 1.0, + "content": "29", + "type": "text" + }, + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "Optimizer (abbr. CPLEX) (Cplex, 2009). However, both JSP and FJSP, as well as many other MO", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 89, + 634, + 100, + 644 + ], + "score": 1.0, + "content": "30", + "type": "text" + }, + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "score": 1.0, + "content": "problems, have been shown to be NP-hard (Garey and Johnson, 1979; Lageweg et al., 1977). That", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 88, + 641, + 507, + 657 + ], + "spans": [ + { + "bbox": [ + 88, + 645, + 100, + 655 + ], + "score": 1.0, + "content": "31", + "type": "text" + }, + { + "bbox": [ + 105, + 641, + 507, + 657 + ], + "score": 1.0, + "content": "said, it is unrealistic and intractable to find the optimal solution for all cases within reasonable times.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 89, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 89, + 656, + 100, + 666 + ], + "score": 1.0, + "content": "32", + "type": "text" + }, + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "These tools can obtain the optimal solutions if sufficient time (or unlimited time) is given; otherwise,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 89, + 664, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 89, + 667, + 100, + 677 + ], + "score": 1.0, + "content": "33", + "type": "text" + }, + { + "bbox": [ + 105, + 664, + 506, + 678 + ], + "score": 1.0, + "content": "return best-effort solutions during the limited time, which usually have gaps to the optimum. When", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 89, + 676, + 346, + 689 + ], + "spans": [ + { + "bbox": [ + 89, + 678, + 100, + 688 + ], + "score": 1.0, + "content": "34", + "type": "text" + }, + { + "bbox": [ + 105, + 676, + 346, + 689 + ], + "score": 1.0, + "content": "problems are scaled up, the gaps usually grow significantly.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 91, + 692, + 505, + 714 + ], + "lines": [ + { + "bbox": [ + 87, + 691, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 87, + 691, + 505, + 705 + ], + "score": 1.0, + "content": "35 In practice, some heuristics (Gupta and Sivakumar, 2006; Haupt, 1989) or approximate methods", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 87, + 703, + 506, + 716 + ], + "spans": [ + { + "bbox": [ + 87, + 703, + 506, + 716 + ], + "score": 1.0, + "content": "36 (Jansen et al., 2000) were used to cope with the issue of intractability. A simple greedy approach is to", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 102, + 731, + 499, + 742 + ], + "lines": [ + { + "bbox": [ + 106, + 730, + 500, + 743 + ], + "spans": [ + { + "bbox": [ + 106, + 730, + 500, + 743 + ], + "score": 1.0, + "content": "Submitted to 37th Conference on Neural Information Processing Systems (NeurIPS 2023). Do not distribute.", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 97, + 504, + 137 + ], + "lines": [ + { + "bbox": [ + 106, + 95, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 106, + 95, + 505, + 120 + ], + "score": 1.0, + "content": "Residual Scheduling: A New Reinforcement Learning", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 118, + 495, + 139 + ], + "spans": [ + { + "bbox": [ + 116, + 118, + 495, + 139 + ], + "score": 1.0, + "content": "Approach to Solving Job Shop Scheduling Problem", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "list", + "bbox": [ + 259, + 179, + 354, + 223 + ], + "lines": [ + { + "bbox": [ + 258, + 178, + 356, + 191 + ], + "spans": [ + { + "bbox": [ + 258, + 178, + 356, + 191 + ], + "score": 1.0, + "content": "Anonymous Author(s)", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 283, + 189, + 328, + 202 + ], + "spans": [ + { + "bbox": [ + 283, + 189, + 328, + 202 + ], + "score": 1.0, + "content": "Affiliation", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 286, + 200, + 324, + 213 + ], + "spans": [ + { + "bbox": [ + 286, + 200, + 324, + 213 + ], + "score": 1.0, + "content": "Address", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 290, + 213, + 320, + 222 + ], + "spans": [ + { + "bbox": [ + 290, + 213, + 320, + 222 + ], + "score": 1.0, + "content": "email", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + } + ], + "index": 3.5, + "bbox_fs": [ + 258, + 178, + 356, + 222 + ] + }, + { + "type": "title", + "bbox": [ + 283, + 252, + 328, + 265 + ], + "lines": [ + { + "bbox": [ + 281, + 251, + 331, + 267 + ], + "spans": [ + { + "bbox": [ + 281, + 251, + 331, + 267 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "index", + "bbox": [ + 91, + 276, + 469, + 462 + ], + "lines": [ + { + "bbox": [ + 92, + 276, + 470, + 290 + ], + "spans": [ + { + "bbox": [ + 92, + 280, + 99, + 288 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 141, + 276, + 470, + 290 + ], + "score": 1.0, + "content": "Job-shop scheduling problem (JSP) is a mathematical optimization problem widely", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 287, + 470, + 299 + ], + "spans": [ + { + "bbox": [ + 92, + 290, + 99, + 298 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 142, + 287, + 470, + 299 + ], + "score": 1.0, + "content": "used in industries like manufacturing, and flexible JSP (FJSP) is also a common", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 298, + 469, + 311 + ], + "spans": [ + { + "bbox": [ + 93, + 301, + 99, + 309 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 142, + 298, + 469, + 311 + ], + "score": 1.0, + "content": "variant. Since they are NP-hard, it is intractable to find the optimal solution for", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 308, + 470, + 322 + ], + "spans": [ + { + "bbox": [ + 92, + 312, + 99, + 321 + ], + "score": 1.0, + "content": "4", + "type": "text" + }, + { + "bbox": [ + 141, + 308, + 470, + 322 + ], + "score": 1.0, + "content": "all cases within reasonable times. Thus, it becomes important to develop efficient", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 320, + 470, + 333 + ], + "spans": [ + { + "bbox": [ + 93, + 322, + 99, + 331 + ], + "score": 1.0, + "content": "5", + "type": "text" + }, + { + "bbox": [ + 141, + 320, + 470, + 333 + ], + "score": 1.0, + "content": "heuristics to solve JSP/FJSP. A kind of method of solving scheduling problems", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 331, + 470, + 343 + ], + "spans": [ + { + "bbox": [ + 92, + 333, + 99, + 343 + ], + "score": 1.0, + "content": "6", + "type": "text" + }, + { + "bbox": [ + 141, + 331, + 470, + 343 + ], + "score": 1.0, + "content": "is construction heuristics, which constructs scheduling solutions via heuristics.", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 342, + 470, + 355 + ], + "spans": [ + { + "bbox": [ + 92, + 345, + 99, + 353 + ], + "score": 1.0, + "content": "7", + "type": "text" + }, + { + "bbox": [ + 141, + 342, + 470, + 355 + ], + "score": 1.0, + "content": "Recently, many methods for construction heuristics leverage deep reinforcement", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 352, + 470, + 366 + ], + "spans": [ + { + "bbox": [ + 92, + 356, + 99, + 364 + ], + "score": 1.0, + "content": "8", + "type": "text" + }, + { + "bbox": [ + 141, + 352, + 470, + 366 + ], + "score": 1.0, + "content": "learning (DRL) with graph neural networks (GNN). In this paper, we propose a new", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 364, + 470, + 377 + ], + "spans": [ + { + "bbox": [ + 92, + 366, + 99, + 375 + ], + "score": 1.0, + "content": "9", + "type": "text" + }, + { + "bbox": [ + 141, + 364, + 470, + 377 + ], + "score": 1.0, + "content": "approach, named residual scheduling, to solving JSP/FJSP. In this new approach,", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 375, + 469, + 387 + ], + "spans": [ + { + "bbox": [ + 90, + 377, + 99, + 386 + ], + "score": 1.0, + "content": "10", + "type": "text" + }, + { + "bbox": [ + 141, + 375, + 469, + 387 + ], + "score": 1.0, + "content": "we remove irrelevant machines and jobs such as those finished, such that the states", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 385, + 470, + 398 + ], + "spans": [ + { + "bbox": [ + 90, + 388, + 99, + 397 + ], + "score": 1.0, + "content": "11", + "type": "text" + }, + { + "bbox": [ + 141, + 385, + 470, + 398 + ], + "score": 1.0, + "content": "include the remaining (or relevant) machines and jobs only. Our experiments show", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 396, + 470, + 409 + ], + "spans": [ + { + "bbox": [ + 90, + 399, + 99, + 408 + ], + "score": 1.0, + "content": "12", + "type": "text" + }, + { + "bbox": [ + 141, + 396, + 470, + 409 + ], + "score": 1.0, + "content": "that our approach reaches state-of-the-art (SOTA) among all known construction", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 408, + 470, + 420 + ], + "spans": [ + { + "bbox": [ + 90, + 410, + 99, + 419 + ], + "score": 1.0, + "content": "13", + "type": "text" + }, + { + "bbox": [ + 141, + 408, + 470, + 420 + ], + "score": 1.0, + "content": "heuristics on most well-known open JSP and FJSP benchmarks. In addition, we", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 419, + 470, + 430 + ], + "spans": [ + { + "bbox": [ + 89, + 421, + 99, + 430 + ], + "score": 1.0, + "content": "14", + "type": "text" + }, + { + "bbox": [ + 142, + 419, + 470, + 430 + ], + "score": 1.0, + "content": "also observe that even though our model is trained for scheduling problems of", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 429, + 471, + 442 + ], + "spans": [ + { + "bbox": [ + 88, + 431, + 100, + 442 + ], + "score": 1.0, + "content": "15", + "type": "text" + }, + { + "bbox": [ + 141, + 429, + 471, + 442 + ], + "score": 1.0, + "content": "smaller sizes, our method still performs well for scheduling problems of large sizes.", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 439, + 470, + 453 + ], + "spans": [ + { + "bbox": [ + 90, + 442, + 99, + 451 + ], + "score": 1.0, + "content": "16", + "type": "text" + }, + { + "bbox": [ + 141, + 439, + 470, + 453 + ], + "score": 1.0, + "content": "Interestingly in our experiments, our approach even reaches zero gap for 49 among", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 452, + 433, + 463 + ], + "spans": [ + { + "bbox": [ + 90, + 453, + 99, + 463 + ], + "score": 1.0, + "content": "17", + "type": "text" + }, + { + "bbox": [ + 142, + 452, + 433, + 463 + ], + "score": 1.0, + "content": "50 JSP instances whose job numbers are more than 150 on 20 machines.", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + } + ], + "index": 15, + "bbox_fs": [ + 88, + 276, + 471, + 463 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 482, + 191, + 496 + ], + "lines": [ + { + "bbox": [ + 87, + 481, + 192, + 498 + ], + "spans": [ + { + "bbox": [ + 87, + 481, + 192, + 498 + ], + "score": 1.0, + "content": "18 1 Introduction", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "index", + "bbox": [ + 90, + 506, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 90, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 90, + 509, + 100, + 518 + ], + "score": 1.0, + "content": "19", + "type": "text" + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "The job-shop scheduling problem (JSP) is a mathematical optimization (MO) problem widely used in", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 518, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 89, + 520, + 100, + 529 + ], + "score": 1.0, + "content": "20", + "type": "text" + }, + { + "bbox": [ + 105, + 518, + 506, + 531 + ], + "score": 1.0, + "content": "many industries, like manufacturing (Zhang et al., 2020; Waschneck et al., 2016). For example, a", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 529, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 89, + 531, + 100, + 540 + ], + "score": 1.0, + "content": "21", + "type": "text" + }, + { + "bbox": [ + 105, + 529, + 506, + 541 + ], + "score": 1.0, + "content": "semiconductor manufacturing process can be viewed as a complex JSP problem (Waschneck et al.,", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 540, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 89, + 542, + 100, + 551 + ], + "score": 1.0, + "content": "22", + "type": "text" + }, + { + "bbox": [ + 106, + 540, + 506, + 552 + ], + "score": 1.0, + "content": "2016), where a set of given jobs are assigned to a set of machines under some constraints to achieve", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 551, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 89, + 553, + 100, + 562 + ], + "score": 1.0, + "content": "23", + "type": "text" + }, + { + "bbox": [ + 106, + 551, + 506, + 563 + ], + "score": 1.0, + "content": "some expected goals such as minimizing makespan which is focused on in this paper. While there are", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 89, + 564, + 100, + 572 + ], + "score": 1.0, + "content": "24", + "type": "text" + }, + { + "bbox": [ + 106, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "many variants of JSP (Abdolrazzagh-Nezhad and Abdullah, 2017), we also consider an extension", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 572, + 449, + 585 + ], + "spans": [ + { + "bbox": [ + 89, + 574, + 100, + 583 + ], + "score": 1.0, + "content": "25", + "type": "text" + }, + { + "bbox": [ + 106, + 572, + 449, + 585 + ], + "score": 1.0, + "content": "called flexible JSP (FJSP) where job operations can be done on designated machines.", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 89, + 591, + 99, + 600 + ], + "score": 1.0, + "content": "26", + "type": "text" + }, + { + "bbox": [ + 105, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "A generic approach to solving MO problems is to use mathematical programming, such as mixed", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 89, + 601, + 99, + 611 + ], + "score": 1.0, + "content": "27", + "type": "text" + }, + { + "bbox": [ + 105, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "integer linear programming (MILP) and constraint satisfaction problem (CSP). Two popular generic", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 610, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 89, + 612, + 99, + 622 + ], + "score": 1.0, + "content": "28", + "type": "text" + }, + { + "bbox": [ + 105, + 610, + 506, + 623 + ], + "score": 1.0, + "content": "MO solvers for solving MO are OR-Tools (Perron and Furnon, 2019) and IBM ILOG CPLEX", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 88, + 623, + 100, + 634 + ], + "score": 1.0, + "content": "29", + "type": "text" + }, + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "Optimizer (abbr. CPLEX) (Cplex, 2009). However, both JSP and FJSP, as well as many other MO", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 89, + 634, + 100, + 644 + ], + "score": 1.0, + "content": "30", + "type": "text" + }, + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "score": 1.0, + "content": "problems, have been shown to be NP-hard (Garey and Johnson, 1979; Lageweg et al., 1977). That", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 641, + 507, + 657 + ], + "spans": [ + { + "bbox": [ + 88, + 645, + 100, + 655 + ], + "score": 1.0, + "content": "31", + "type": "text" + }, + { + "bbox": [ + 105, + 641, + 507, + 657 + ], + "score": 1.0, + "content": "said, it is unrealistic and intractable to find the optimal solution for all cases within reasonable times.", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 89, + 656, + 100, + 666 + ], + "score": 1.0, + "content": "32", + "type": "text" + }, + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "These tools can obtain the optimal solutions if sufficient time (or unlimited time) is given; otherwise,", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 664, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 89, + 667, + 100, + 677 + ], + "score": 1.0, + "content": "33", + "type": "text" + }, + { + "bbox": [ + 105, + 664, + 506, + 678 + ], + "score": 1.0, + "content": "return best-effort solutions during the limited time, which usually have gaps to the optimum. When", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 676, + 346, + 689 + ], + "spans": [ + { + "bbox": [ + 89, + 678, + 100, + 688 + ], + "score": 1.0, + "content": "34", + "type": "text" + }, + { + "bbox": [ + 105, + 676, + 346, + 689 + ], + "score": 1.0, + "content": "problems are scaled up, the gaps usually grow significantly.", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 87, + 691, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 87, + 691, + 505, + 705 + ], + "score": 1.0, + "content": "35 In practice, some heuristics (Gupta and Sivakumar, 2006; Haupt, 1989) or approximate methods", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 87, + 703, + 506, + 716 + ], + "spans": [ + { + "bbox": [ + 87, + 703, + 506, + 716 + ], + "score": 1.0, + "content": "36 (Jansen et al., 2000) were used to cope with the issue of intractability. A simple greedy approach is to", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 73, + 506, + 84 + ], + "spans": [ + { + "bbox": [ + 89, + 75, + 99, + 84 + ], + "score": 1.0, + "content": "37", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 73, + 506, + 84 + ], + "score": 1.0, + "content": "use the heuristics following the so-called priority dispatching rule (PDR) (Haupt, 1989) to construct", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 84, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 89, + 86, + 99, + 95 + ], + "score": 1.0, + "content": "38", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 84, + 506, + 96 + ], + "score": 1.0, + "content": "solutions. These can also be viewed as a kind of solution construction heuristics or construction", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 94, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 89, + 96, + 99, + 106 + ], + "score": 1.0, + "content": "39", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "score": 1.0, + "content": "heuristics. Some of PDR examples are First In First Out (FIFO), Shortest Processing Time (SPT),", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 106, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 89, + 108, + 100, + 117 + ], + "score": 1.0, + "content": "40", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 106, + 506, + 118 + ], + "score": 1.0, + "content": "Most WorK Remaining (MWKR), and Most Operation Remaining (MOR). Although these heuristics", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 116, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 89, + 119, + 99, + 127 + ], + "score": 1.0, + "content": "41", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 116, + 506, + 129 + ], + "score": 1.0, + "content": "are usually computationally fast, it is hard to design generally effective rules to minimize the gap to", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 127, + 391, + 140 + ], + "spans": [ + { + "bbox": [ + 89, + 129, + 100, + 139 + ], + "score": 1.0, + "content": "42", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 127, + 391, + 140 + ], + "score": 1.0, + "content": "the optimum, and the derived results are usually far from the optimum.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 89, + 145, + 100, + 155 + ], + "score": 1.0, + "content": "43", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "Furthermore, a generic approach to automating the design of heuristics is called metaheuristics, such", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 89, + 157, + 100, + 166 + ], + "score": 1.0, + "content": "44", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 154, + 505, + 166 + ], + "score": 1.0, + "content": "as tabu search (Dell’Amico and Trubian, 1993; Saidi-Mehrabad and Fattahi, 2007) , genetic algorithm", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 89, + 167, + 100, + 177 + ], + "score": 1.0, + "content": "45", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "(GA) (Pezzella et al., 2008; Ren and Wang, 2012), and PSO algorithms (Lian et al., 2006; Liu et al.,", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 175, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 89, + 178, + 100, + 188 + ], + "score": 1.0, + "content": "46", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 175, + 506, + 189 + ], + "score": 1.0, + "content": "2011). However, metaheuristics still take a high computation time, and it is not ensured to obtain the", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 187, + 201, + 199 + ], + "spans": [ + { + "bbox": [ + 89, + 189, + 100, + 199 + ], + "score": 1.0, + "content": "47", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 187, + 201, + 199 + ], + "score": 1.0, + "content": "optimal solution either.", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 203, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 89, + 206, + 100, + 215 + ], + "score": 1.0, + "content": "48", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 203, + 506, + 216 + ], + "score": 1.0, + "content": "Recently, deep reinforcement learning (DRL) has made several significant successes for some", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 88, + 217, + 99, + 226 + ], + "score": 1.0, + "content": "49", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "applications, such as AlphaGo (Silver et al., 2016), AlphaStar (Vinyals et al., 2019), AlphaTensor", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 224, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 88, + 227, + 100, + 238 + ], + "score": 1.0, + "content": "50", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 224, + 506, + 239 + ], + "score": 1.0, + "content": "(Fawzi et al., 2022), and thus it also attracted much attention in the MO problems, including chip", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 88, + 238, + 100, + 249 + ], + "score": 1.0, + "content": "51", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 237, + 505, + 248 + ], + "score": 1.0, + "content": "design (Mirhoseini et al., 2021) and scheduling problems (Zhang et al., 2023). In the past, several", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 246, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 88, + 249, + 100, + 260 + ], + "score": 1.0, + "content": "52", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 246, + 506, + 261 + ], + "score": 1.0, + "content": "researchers used DRL methods as construction heuristics, and their methods did improve scheduling", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 258, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 88, + 261, + 99, + 270 + ], + "score": 1.0, + "content": "53", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 258, + 506, + 271 + ], + "score": 1.0, + "content": "performance, illustrated as follows. Park et al. (2020) proposed a method based on DQN (Mnih et al.,", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 269, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 88, + 271, + 100, + 281 + ], + "score": 1.0, + "content": "54", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 269, + 506, + 281 + ], + "score": 1.0, + "content": "2015) for JSP in semiconductor manufacturing and showed that their DQN model outperformed GA", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 280, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 88, + 282, + 99, + 292 + ], + "score": 1.0, + "content": "55", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 280, + 506, + 294 + ], + "score": 1.0, + "content": "in terms of both scheduling performance (namely gap to the optimum on makespan) and computation", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 291, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 88, + 292, + 100, + 303 + ], + "score": 1.0, + "content": "56", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 291, + 505, + 303 + ], + "score": 1.0, + "content": "time. Lin et al. (2019) and Luo (2020) proposed different DQN models to decide the scheduling action", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 302, + 494, + 315 + ], + "spans": [ + { + "bbox": [ + 89, + 304, + 100, + 313 + ], + "score": 1.0, + "content": "57", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 302, + 494, + 315 + ], + "score": 1.0, + "content": "among the heuristic rules and improved the makespan and the tardiness over PDRs, respectively.", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 88, + 321, + 99, + 330 + ], + "score": 1.0, + "content": "58", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "A recent DRL-based approach to solving JSP/FJSP problems is to leverage graph neural networks", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 328, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 88, + 331, + 99, + 341 + ], + "score": 1.0, + "content": "59", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 328, + 506, + 343 + ], + "score": 1.0, + "content": "(GNN) to design a size-agnostic representation (Zhang et al., 2020; Park et al., 2021b,a; Song et al.,", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 89, + 342, + 99, + 352 + ], + "score": 1.0, + "content": "60", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "2023). In this approach, graph representation has better generalization ability in larger instances", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 350, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 88, + 353, + 99, + 362 + ], + "score": 1.0, + "content": "61", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 350, + 506, + 364 + ], + "score": 1.0, + "content": "and provides a holistic view of scheduling states. Zhang et al. (2020) proposed a DRL method", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 88, + 364, + 99, + 374 + ], + "score": 1.0, + "content": "62", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "with disjunctive graph representation for JSP, called L2D (Learning to Dispatch), and used GNN", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 372, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 88, + 374, + 100, + 385 + ], + "score": 1.0, + "content": "63", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 372, + 506, + 385 + ], + "score": 1.0, + "content": "to encode the graph for scheduling decision. Besides, Song et al. (2023) extended their methods", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 88, + 386, + 100, + 395 + ], + "score": 1.0, + "content": "64", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "to FJSP. Park et al. (2021b) used a similar strategy of (Zhang et al., 2020) but with different state", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 394, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 88, + 396, + 100, + 407 + ], + "score": 1.0, + "content": "65", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 394, + 506, + 408 + ], + "score": 1.0, + "content": "features and model structure. Park et al. (2021a) proposed a new approach to solving JSP, called", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 405, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 88, + 407, + 99, + 417 + ], + "score": 1.0, + "content": "66", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "score": 1.0, + "content": "ScheduleNet, by using a different graph representation and a DRL model with the graph attention for", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 88, + 419, + 100, + 428 + ], + "score": 1.0, + "content": "67", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "scheduling decision. Most of the experiments above showed that their models trained from small", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 425, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 88, + 429, + 99, + 439 + ], + "score": 1.0, + "content": "68", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 425, + 506, + 441 + ], + "score": 1.0, + "content": "instances still worked reasonably well for large test instances, and generally better than PDRs. Among", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 438, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 88, + 440, + 100, + 450 + ], + "score": 1.0, + "content": "69", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 438, + 506, + 450 + ], + "score": 1.0, + "content": "these methods, ScheduleNet achieved state-of-the-art (SOTA) performance. There are still other", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 450, + 507, + 461 + ], + "spans": [ + { + "bbox": [ + 89, + 451, + 99, + 460 + ], + "score": 1.0, + "content": "70", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 450, + 507, + 461 + ], + "score": 1.0, + "content": "DRL-based approaches to solving JSP/FJSP problems, but not construction heuristics. Zhang et al.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 460, + 507, + 472 + ], + "spans": [ + { + "bbox": [ + 89, + 462, + 99, + 472 + ], + "score": 1.0, + "content": "71", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 460, + 507, + 472 + ], + "score": 1.0, + "content": "(2022) proposes another approach, called Learning to Search (L2S), a kind of search-based heuristics.", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 89, + 478, + 100, + 488 + ], + "score": 1.0, + "content": "72", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "In this paper, we propose a new approach to solving JSP/FJSP, a kind of construction heuristics, also", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 486, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 89, + 489, + 100, + 499 + ], + "score": 1.0, + "content": "73", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 486, + 506, + 500 + ], + "score": 1.0, + "content": "based on GNN. In this new approach, we remove irrelevant machines and jobs, such as those finished,", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 498, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 89, + 500, + 100, + 510 + ], + "score": 1.0, + "content": "74", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 498, + 505, + 511 + ], + "score": 1.0, + "content": "such that the states include the remaining machines and jobs only. This approach is named residual", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 508, + 378, + 522 + ], + "spans": [ + { + "bbox": [ + 89, + 511, + 100, + 520 + ], + "score": 1.0, + "content": "75", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 508, + 378, + 522 + ], + "score": 1.0, + "content": "scheduling in this paper to indicate to work on the remaining graph.", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 525, + 507, + 538 + ], + "spans": [ + { + "bbox": [ + 89, + 528, + 100, + 537 + ], + "score": 1.0, + "content": "76", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 525, + 507, + 538 + ], + "score": 1.0, + "content": "Without irrelevant information, our experiments show that our approach reaches SOTA by outper-", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 536, + 506, + 549 + ], + "spans": [ + { + "bbox": [ + 89, + 538, + 100, + 547 + ], + "score": 1.0, + "content": "77", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 536, + 506, + 549 + ], + "score": 1.0, + "content": "forming the above mentioned construction methods on some well-known open benchmarks, seven", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 547, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 89, + 549, + 100, + 559 + ], + "score": 1.0, + "content": "78", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "score": 1.0, + "content": "for JSP and two for FJSP, as described in Section 4. We also observe that even though our model", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 557, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 89, + 560, + 100, + 570 + ], + "score": 1.0, + "content": "79", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 557, + 506, + 572 + ], + "score": 1.0, + "content": "is trained for scheduling problems of smaller sizes, our method still performs well for scheduling", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 569, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 89, + 571, + 100, + 581 + ], + "score": 1.0, + "content": "80", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 569, + 506, + 581 + ], + "score": 1.0, + "content": "problems of large sizes. Interestingly in our experiments, our approach even reaches zero gap for 49", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 580, + 427, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 582, + 100, + 592 + ], + "score": 1.0, + "content": "81", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 580, + 427, + 592 + ], + "score": 1.0, + "content": "among 50 JSP instances whose job numbers are more than 150 on 20 machines.", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_start_line": true + } + ], + "index": 28, + "bbox_fs": [ + 89, + 506, + 506, + 585 + ] + }, + { + "type": "index", + "bbox": [ + 90, + 588, + 505, + 687 + ], + "lines": [], + "index": 36, + "bbox_fs": [ + 88, + 589, + 507, + 689 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 91, + 692, + 505, + 714 + ], + "lines": [], + "index": 41.5, + "bbox_fs": [ + 87, + 691, + 506, + 716 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 90, + 73, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 89, + 73, + 506, + 84 + ], + "spans": [ + { + "bbox": [ + 89, + 75, + 99, + 84 + ], + "score": 1.0, + "content": "37", + "type": "text" + }, + { + "bbox": [ + 105, + 73, + 506, + 84 + ], + "score": 1.0, + "content": "use the heuristics following the so-called priority dispatching rule (PDR) (Haupt, 1989) to construct", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 84, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 89, + 86, + 99, + 95 + ], + "score": 1.0, + "content": "38", + "type": "text" + }, + { + "bbox": [ + 106, + 84, + 506, + 96 + ], + "score": 1.0, + "content": "solutions. These can also be viewed as a kind of solution construction heuristics or construction", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 89, + 94, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 89, + 96, + 99, + 106 + ], + "score": 1.0, + "content": "39", + "type": "text" + }, + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "score": 1.0, + "content": "heuristics. Some of PDR examples are First In First Out (FIFO), Shortest Processing Time (SPT),", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 89, + 106, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 89, + 108, + 100, + 117 + ], + "score": 1.0, + "content": "40", + "type": "text" + }, + { + "bbox": [ + 105, + 106, + 506, + 118 + ], + "score": 1.0, + "content": "Most WorK Remaining (MWKR), and Most Operation Remaining (MOR). Although these heuristics", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 89, + 116, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 89, + 119, + 99, + 127 + ], + "score": 1.0, + "content": "41", + "type": "text" + }, + { + "bbox": [ + 104, + 116, + 506, + 129 + ], + "score": 1.0, + "content": "are usually computationally fast, it is hard to design generally effective rules to minimize the gap to", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 127, + 391, + 140 + ], + "spans": [ + { + "bbox": [ + 89, + 129, + 100, + 139 + ], + "score": 1.0, + "content": "42", + "type": "text" + }, + { + "bbox": [ + 104, + 127, + 391, + 140 + ], + "score": 1.0, + "content": "the optimum, and the derived results are usually far from the optimum.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 90, + 144, + 505, + 198 + ], + "lines": [ + { + "bbox": [ + 89, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 89, + 145, + 100, + 155 + ], + "score": 1.0, + "content": "43", + "type": "text" + }, + { + "bbox": [ + 104, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "Furthermore, a generic approach to automating the design of heuristics is called metaheuristics, such", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 89, + 157, + 100, + 166 + ], + "score": 1.0, + "content": "44", + "type": "text" + }, + { + "bbox": [ + 105, + 154, + 505, + 166 + ], + "score": 1.0, + "content": "as tabu search (Dell’Amico and Trubian, 1993; Saidi-Mehrabad and Fattahi, 2007) , genetic algorithm", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 89, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 89, + 167, + 100, + 177 + ], + "score": 1.0, + "content": "45", + "type": "text" + }, + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "(GA) (Pezzella et al., 2008; Ren and Wang, 2012), and PSO algorithms (Lian et al., 2006; Liu et al.,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 175, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 89, + 178, + 100, + 188 + ], + "score": 1.0, + "content": "46", + "type": "text" + }, + { + "bbox": [ + 105, + 175, + 506, + 189 + ], + "score": 1.0, + "content": "2011). However, metaheuristics still take a high computation time, and it is not ensured to obtain the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 187, + 201, + 199 + ], + "spans": [ + { + "bbox": [ + 89, + 189, + 100, + 199 + ], + "score": 1.0, + "content": "47", + "type": "text" + }, + { + "bbox": [ + 105, + 187, + 201, + 199 + ], + "score": 1.0, + "content": "optimal solution either.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 89, + 203, + 505, + 313 + ], + "lines": [ + { + "bbox": [ + 89, + 203, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 89, + 206, + 100, + 215 + ], + "score": 1.0, + "content": "48", + "type": "text" + }, + { + "bbox": [ + 105, + 203, + 506, + 216 + ], + "score": 1.0, + "content": "Recently, deep reinforcement learning (DRL) has made several significant successes for some", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 88, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 88, + 217, + 99, + 226 + ], + "score": 1.0, + "content": "49", + "type": "text" + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "applications, such as AlphaGo (Silver et al., 2016), AlphaStar (Vinyals et al., 2019), AlphaTensor", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 88, + 224, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 88, + 227, + 100, + 238 + ], + "score": 1.0, + "content": "50", + "type": "text" + }, + { + "bbox": [ + 104, + 224, + 506, + 239 + ], + "score": 1.0, + "content": "(Fawzi et al., 2022), and thus it also attracted much attention in the MO problems, including chip", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 88, + 238, + 100, + 249 + ], + "score": 1.0, + "content": "51", + "type": "text" + }, + { + "bbox": [ + 106, + 237, + 505, + 248 + ], + "score": 1.0, + "content": "design (Mirhoseini et al., 2021) and scheduling problems (Zhang et al., 2023). In the past, several", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 88, + 246, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 88, + 249, + 100, + 260 + ], + "score": 1.0, + "content": "52", + "type": "text" + }, + { + "bbox": [ + 105, + 246, + 506, + 261 + ], + "score": 1.0, + "content": "researchers used DRL methods as construction heuristics, and their methods did improve scheduling", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 88, + 258, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 88, + 261, + 99, + 270 + ], + "score": 1.0, + "content": "53", + "type": "text" + }, + { + "bbox": [ + 105, + 258, + 506, + 271 + ], + "score": 1.0, + "content": "performance, illustrated as follows. Park et al. (2020) proposed a method based on DQN (Mnih et al.,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 88, + 269, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 88, + 271, + 100, + 281 + ], + "score": 1.0, + "content": "54", + "type": "text" + }, + { + "bbox": [ + 105, + 269, + 506, + 281 + ], + "score": 1.0, + "content": "2015) for JSP in semiconductor manufacturing and showed that their DQN model outperformed GA", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 88, + 280, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 88, + 282, + 99, + 292 + ], + "score": 1.0, + "content": "55", + "type": "text" + }, + { + "bbox": [ + 104, + 280, + 506, + 294 + ], + "score": 1.0, + "content": "in terms of both scheduling performance (namely gap to the optimum on makespan) and computation", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 88, + 291, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 88, + 292, + 100, + 303 + ], + "score": 1.0, + "content": "56", + "type": "text" + }, + { + "bbox": [ + 105, + 291, + 505, + 303 + ], + "score": 1.0, + "content": "time. Lin et al. (2019) and Luo (2020) proposed different DQN models to decide the scheduling action", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 302, + 494, + 315 + ], + "spans": [ + { + "bbox": [ + 89, + 304, + 100, + 313 + ], + "score": 1.0, + "content": "57", + "type": "text" + }, + { + "bbox": [ + 105, + 302, + 494, + 315 + ], + "score": 1.0, + "content": "among the heuristic rules and improved the makespan and the tardiness over PDRs, respectively.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 89, + 318, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 88, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 88, + 321, + 99, + 330 + ], + "score": 1.0, + "content": "58", + "type": "text" + }, + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "A recent DRL-based approach to solving JSP/FJSP problems is to leverage graph neural networks", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 88, + 328, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 88, + 331, + 99, + 341 + ], + "score": 1.0, + "content": "59", + "type": "text" + }, + { + "bbox": [ + 104, + 328, + 506, + 343 + ], + "score": 1.0, + "content": "(GNN) to design a size-agnostic representation (Zhang et al., 2020; Park et al., 2021b,a; Song et al.,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 89, + 342, + 99, + 352 + ], + "score": 1.0, + "content": "60", + "type": "text" + }, + { + "bbox": [ + 105, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "2023). In this approach, graph representation has better generalization ability in larger instances", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 88, + 350, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 88, + 353, + 99, + 362 + ], + "score": 1.0, + "content": "61", + "type": "text" + }, + { + "bbox": [ + 105, + 350, + 506, + 364 + ], + "score": 1.0, + "content": "and provides a holistic view of scheduling states. Zhang et al. (2020) proposed a DRL method", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 88, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 88, + 364, + 99, + 374 + ], + "score": 1.0, + "content": "62", + "type": "text" + }, + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "with disjunctive graph representation for JSP, called L2D (Learning to Dispatch), and used GNN", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 88, + 372, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 88, + 374, + 100, + 385 + ], + "score": 1.0, + "content": "63", + "type": "text" + }, + { + "bbox": [ + 104, + 372, + 506, + 385 + ], + "score": 1.0, + "content": "to encode the graph for scheduling decision. Besides, Song et al. (2023) extended their methods", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 88, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 88, + 386, + 100, + 395 + ], + "score": 1.0, + "content": "64", + "type": "text" + }, + { + "bbox": [ + 105, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "to FJSP. Park et al. (2021b) used a similar strategy of (Zhang et al., 2020) but with different state", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 394, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 88, + 396, + 100, + 407 + ], + "score": 1.0, + "content": "65", + "type": "text" + }, + { + "bbox": [ + 105, + 394, + 506, + 408 + ], + "score": 1.0, + "content": "features and model structure. Park et al. (2021a) proposed a new approach to solving JSP, called", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 88, + 405, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 88, + 407, + 99, + 417 + ], + "score": 1.0, + "content": "66", + "type": "text" + }, + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "score": 1.0, + "content": "ScheduleNet, by using a different graph representation and a DRL model with the graph attention for", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 88, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 88, + 419, + 100, + 428 + ], + "score": 1.0, + "content": "67", + "type": "text" + }, + { + "bbox": [ + 105, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "scheduling decision. Most of the experiments above showed that their models trained from small", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 88, + 425, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 88, + 429, + 99, + 439 + ], + "score": 1.0, + "content": "68", + "type": "text" + }, + { + "bbox": [ + 104, + 425, + 506, + 441 + ], + "score": 1.0, + "content": "instances still worked reasonably well for large test instances, and generally better than PDRs. Among", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 88, + 438, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 88, + 440, + 100, + 450 + ], + "score": 1.0, + "content": "69", + "type": "text" + }, + { + "bbox": [ + 106, + 438, + 506, + 450 + ], + "score": 1.0, + "content": "these methods, ScheduleNet achieved state-of-the-art (SOTA) performance. There are still other", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 450, + 507, + 461 + ], + "spans": [ + { + "bbox": [ + 89, + 451, + 99, + 460 + ], + "score": 1.0, + "content": "70", + "type": "text" + }, + { + "bbox": [ + 106, + 450, + 507, + 461 + ], + "score": 1.0, + "content": "DRL-based approaches to solving JSP/FJSP problems, but not construction heuristics. Zhang et al.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 89, + 460, + 507, + 472 + ], + "spans": [ + { + "bbox": [ + 89, + 462, + 99, + 472 + ], + "score": 1.0, + "content": "71", + "type": "text" + }, + { + "bbox": [ + 105, + 460, + 507, + 472 + ], + "score": 1.0, + "content": "(2022) proposes another approach, called Learning to Search (L2S), a kind of search-based heuristics.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 90, + 476, + 505, + 520 + ], + "lines": [ + { + "bbox": [ + 89, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 89, + 478, + 100, + 488 + ], + "score": 1.0, + "content": "72", + "type": "text" + }, + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "In this paper, we propose a new approach to solving JSP/FJSP, a kind of construction heuristics, also", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 486, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 89, + 489, + 100, + 499 + ], + "score": 1.0, + "content": "73", + "type": "text" + }, + { + "bbox": [ + 104, + 486, + 506, + 500 + ], + "score": 1.0, + "content": "based on GNN. In this new approach, we remove irrelevant machines and jobs, such as those finished,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 498, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 89, + 500, + 100, + 510 + ], + "score": 1.0, + "content": "74", + "type": "text" + }, + { + "bbox": [ + 105, + 498, + 505, + 511 + ], + "score": 1.0, + "content": "such that the states include the remaining machines and jobs only. This approach is named residual", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 89, + 508, + 378, + 522 + ], + "spans": [ + { + "bbox": [ + 89, + 511, + 100, + 520 + ], + "score": 1.0, + "content": "75", + "type": "text" + }, + { + "bbox": [ + 104, + 508, + 378, + 522 + ], + "score": 1.0, + "content": "scheduling in this paper to indicate to work on the remaining graph.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 90, + 525, + 505, + 591 + ], + "lines": [ + { + "bbox": [ + 89, + 525, + 507, + 538 + ], + "spans": [ + { + "bbox": [ + 89, + 528, + 100, + 537 + ], + "score": 1.0, + "content": "76", + "type": "text" + }, + { + "bbox": [ + 105, + 525, + 507, + 538 + ], + "score": 1.0, + "content": "Without irrelevant information, our experiments show that our approach reaches SOTA by outper-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 89, + 536, + 506, + 549 + ], + "spans": [ + { + "bbox": [ + 89, + 538, + 100, + 547 + ], + "score": 1.0, + "content": "77", + "type": "text" + }, + { + "bbox": [ + 105, + 536, + 506, + 549 + ], + "score": 1.0, + "content": "forming the above mentioned construction methods on some well-known open benchmarks, seven", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 89, + 547, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 89, + 549, + 100, + 559 + ], + "score": 1.0, + "content": "78", + "type": "text" + }, + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "score": 1.0, + "content": "for JSP and two for FJSP, as described in Section 4. We also observe that even though our model", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 89, + 557, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 89, + 560, + 100, + 570 + ], + "score": 1.0, + "content": "79", + "type": "text" + }, + { + "bbox": [ + 104, + 557, + 506, + 572 + ], + "score": 1.0, + "content": "is trained for scheduling problems of smaller sizes, our method still performs well for scheduling", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 89, + 569, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 89, + 571, + 100, + 581 + ], + "score": 1.0, + "content": "80", + "type": "text" + }, + { + "bbox": [ + 105, + 569, + 506, + 581 + ], + "score": 1.0, + "content": "problems of large sizes. Interestingly in our experiments, our approach even reaches zero gap for 49", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 89, + 580, + 427, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 582, + 100, + 592 + ], + "score": 1.0, + "content": "81", + "type": "text" + }, + { + "bbox": [ + 105, + 580, + 427, + 592 + ], + "score": 1.0, + "content": "among 50 JSP instances whose job numbers are more than 150 on 20 machines.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41.5 + }, + { + "type": "title", + "bbox": [ + 94, + 607, + 236, + 621 + ], + "lines": [ + { + "bbox": [ + 90, + 606, + 238, + 622 + ], + "spans": [ + { + "bbox": [ + 90, + 606, + 238, + 622 + ], + "score": 1.0, + "content": "82 2 Problem Formulation", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + }, + { + "type": "title", + "bbox": [ + 104, + 632, + 191, + 644 + ], + "lines": [ + { + "bbox": [ + 105, + 631, + 192, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 192, + 646 + ], + "score": 1.0, + "content": "2.1 JSP and FJSP", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 99, + 653, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 651, + 504, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 116, + 666 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 116, + 654, + 145, + 663 + ], + "score": 0.87, + "content": "n \\times m", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 651, + 236, + 666 + ], + "score": 1.0, + "content": "JSP instance contains", + "type": "text" + }, + { + "bbox": [ + 237, + 656, + 244, + 663 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 651, + 282, + 666 + ], + "score": 1.0, + "content": "jobs and", + "type": "text" + }, + { + "bbox": [ + 282, + 655, + 292, + 663 + ], + "score": 0.73, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 651, + 376, + 666 + ], + "score": 1.0, + "content": "machines. Each job", + "type": "text" + }, + { + "bbox": [ + 377, + 653, + 387, + 665 + ], + "score": 0.88, + "content": "J _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 651, + 493, + 666 + ], + "score": 1.0, + "content": "consists of a sequence of", + "type": "text" + }, + { + "bbox": [ + 493, + 653, + 504, + 665 + ], + "score": 0.87, + "content": "k _ { j }", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 663, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 152, + 677 + ], + "score": 1.0, + "content": "operations", + "type": "text" + }, + { + "bbox": [ + 152, + 664, + 225, + 677 + ], + "score": 0.93, + "content": "\\{ O _ { j , 1 } , \\dotsc , O _ { j , k _ { j } } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 663, + 299, + 677 + ], + "score": 1.0, + "content": ", where operation", + "type": "text" + }, + { + "bbox": [ + 299, + 664, + 316, + 676 + ], + "score": 0.9, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 663, + 406, + 677 + ], + "score": 1.0, + "content": "must be started after", + "type": "text" + }, + { + "bbox": [ + 406, + 664, + 434, + 676 + ], + "score": 0.91, + "content": "O _ { j , i - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 663, + 506, + 677 + ], + "score": 1.0, + "content": "is finished. One", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 674, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 674, + 505, + 689 + ], + "score": 1.0, + "content": "machine can process at most one operation at a time, and preemption is not allowed upon processing", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 104, + 685, + 506, + 698 + ], + "spans": [ + { + "bbox": [ + 104, + 685, + 237, + 698 + ], + "score": 1.0, + "content": "operations. In JSP, one operation", + "type": "text" + }, + { + "bbox": [ + 238, + 687, + 255, + 698 + ], + "score": 0.9, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 685, + 506, + 698 + ], + "score": 1.0, + "content": "is allowed to be processed on one designated machine, denoted", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 101, + 693, + 509, + 716 + ], + "spans": [ + { + "bbox": [ + 101, + 693, + 118, + 716 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 119, + 700, + 138, + 712 + ], + "score": 0.91, + "content": "M _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 693, + 281, + 716 + ], + "score": 1.0, + "content": ", with a processing time, denoted by", + "type": "text" + }, + { + "bbox": [ + 281, + 698, + 304, + 714 + ], + "score": 0.92, + "content": "T _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 693, + 400, + 716 + ], + "score": 1.0, + "content": ". Table 1 (a) illustrates a", + "type": "text" + }, + { + "bbox": [ + 400, + 700, + 424, + 711 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 693, + 509, + 716 + ], + "score": 1.0, + "content": "JSP instance, where", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "the three jobs have 3, 3, 2 operations respectively, each of which is designated to be processed on", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 49.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 90, + 73, + 505, + 138 + ], + "lines": [], + "index": 2.5, + "bbox_fs": [ + 89, + 73, + 506, + 140 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 90, + 144, + 505, + 198 + ], + "lines": [], + "index": 8, + "bbox_fs": [ + 89, + 143, + 506, + 199 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 89, + 203, + 505, + 313 + ], + "lines": [], + "index": 15.5, + "bbox_fs": [ + 88, + 203, + 506, + 315 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 89, + 318, + 505, + 471 + ], + "lines": [], + "index": 27.5, + "bbox_fs": [ + 88, + 318, + 507, + 472 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 90, + 476, + 505, + 520 + ], + "lines": [], + "index": 36.5, + "bbox_fs": [ + 89, + 476, + 506, + 522 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 90, + 525, + 505, + 591 + ], + "lines": [], + "index": 41.5, + "bbox_fs": [ + 89, + 525, + 507, + 592 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 94, + 607, + 236, + 621 + ], + "lines": [ + { + "bbox": [ + 90, + 606, + 238, + 622 + ], + "spans": [ + { + "bbox": [ + 90, + 606, + 238, + 622 + ], + "score": 1.0, + "content": "82 2 Problem Formulation", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + }, + { + "type": "title", + "bbox": [ + 104, + 632, + 191, + 644 + ], + "lines": [ + { + "bbox": [ + 105, + 631, + 192, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 192, + 646 + ], + "score": 1.0, + "content": "2.1 JSP and FJSP", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 99, + 653, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 651, + 504, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 116, + 666 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 116, + 654, + 145, + 663 + ], + "score": 0.87, + "content": "n \\times m", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 651, + 236, + 666 + ], + "score": 1.0, + "content": "JSP instance contains", + "type": "text" + }, + { + "bbox": [ + 237, + 656, + 244, + 663 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 651, + 282, + 666 + ], + "score": 1.0, + "content": "jobs and", + "type": "text" + }, + { + "bbox": [ + 282, + 655, + 292, + 663 + ], + "score": 0.73, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 651, + 376, + 666 + ], + "score": 1.0, + "content": "machines. Each job", + "type": "text" + }, + { + "bbox": [ + 377, + 653, + 387, + 665 + ], + "score": 0.88, + "content": "J _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 651, + 493, + 666 + ], + "score": 1.0, + "content": "consists of a sequence of", + "type": "text" + }, + { + "bbox": [ + 493, + 653, + 504, + 665 + ], + "score": 0.87, + "content": "k _ { j }", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 663, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 152, + 677 + ], + "score": 1.0, + "content": "operations", + "type": "text" + }, + { + "bbox": [ + 152, + 664, + 225, + 677 + ], + "score": 0.93, + "content": "\\{ O _ { j , 1 } , \\dotsc , O _ { j , k _ { j } } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 663, + 299, + 677 + ], + "score": 1.0, + "content": ", where operation", + "type": "text" + }, + { + "bbox": [ + 299, + 664, + 316, + 676 + ], + "score": 0.9, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 663, + 406, + 677 + ], + "score": 1.0, + "content": "must be started after", + "type": "text" + }, + { + "bbox": [ + 406, + 664, + 434, + 676 + ], + "score": 0.91, + "content": "O _ { j , i - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 663, + 506, + 677 + ], + "score": 1.0, + "content": "is finished. One", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 674, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 674, + 505, + 689 + ], + "score": 1.0, + "content": "machine can process at most one operation at a time, and preemption is not allowed upon processing", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 104, + 685, + 506, + 698 + ], + "spans": [ + { + "bbox": [ + 104, + 685, + 237, + 698 + ], + "score": 1.0, + "content": "operations. In JSP, one operation", + "type": "text" + }, + { + "bbox": [ + 238, + 687, + 255, + 698 + ], + "score": 0.9, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 685, + 506, + 698 + ], + "score": 1.0, + "content": "is allowed to be processed on one designated machine, denoted", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 101, + 693, + 509, + 716 + ], + "spans": [ + { + "bbox": [ + 101, + 693, + 118, + 716 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 119, + 700, + 138, + 712 + ], + "score": 0.91, + "content": "M _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 693, + 281, + 716 + ], + "score": 1.0, + "content": ", with a processing time, denoted by", + "type": "text" + }, + { + "bbox": [ + 281, + 698, + 304, + 714 + ], + "score": 0.92, + "content": "T _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 693, + 400, + 716 + ], + "score": 1.0, + "content": ". Table 1 (a) illustrates a", + "type": "text" + }, + { + "bbox": [ + 400, + 700, + 424, + 711 + ], + "score": 0.88, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 693, + 509, + 716 + ], + "score": 1.0, + "content": "JSP instance, where", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "the three jobs have 3, 3, 2 operations respectively, each of which is designated to be processed on", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 49.5, + "bbox_fs": [ + 101, + 651, + 509, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 89, + 72, + 505, + 109 + ], + "lines": [ + { + "bbox": [ + 90, + 72, + 505, + 85 + ], + "spans": [ + { + "bbox": [ + 90, + 72, + 211, + 85 + ], + "score": 1.0, + "content": "90 one of the three machines", + "type": "text" + }, + { + "bbox": [ + 211, + 73, + 273, + 85 + ], + "score": 0.94, + "content": "\\{ M _ { 1 } , M _ { 2 } , M _ { 3 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 72, + 505, + 85 + ], + "score": 1.0, + "content": "in the table. A solution of a JSP instance is to dispatch all", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 78, + 511, + 108 + ], + "spans": [ + { + "bbox": [ + 89, + 88, + 100, + 97 + ], + "score": 1.0, + "content": "91", + "type": "text" + }, + { + "bbox": [ + 98, + 78, + 511, + 108 + ], + "score": 1.0, + "content": "operations Oj,i to the corresponding machine Mj,i at time τ (s)j,i , such that the above constraints are", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 89, + 100, + 100, + 109 + ], + "spans": [ + { + "bbox": [ + 89, + 100, + 100, + 109 + ], + "score": 1.0, + "content": "92", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table", + "bbox": [ + 307, + 152, + 474, + 256 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 240, + 121, + 371, + 133 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 240, + 120, + 371, + 133 + ], + "spans": [ + { + "bbox": [ + 240, + 120, + 371, + 133 + ], + "score": 1.0, + "content": "Table 1: JSP and FJSP instances", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "table_caption", + "bbox": [ + 342, + 142, + 439, + 152 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 342, + 141, + 440, + 153 + ], + "spans": [ + { + "bbox": [ + 342, + 141, + 365, + 153 + ], + "score": 1.0, + "content": "(b) A", + "type": "text" + }, + { + "bbox": [ + 365, + 142, + 387, + 151 + ], + "score": 0.73, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 141, + 440, + 153 + ], + "score": 1.0, + "content": "FJSP instance", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "table_body", + "bbox": [ + 307, + 152, + 474, + 256 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 307, + 152, + 474, + 256 + ], + "spans": [ + { + "bbox": [ + 307, + 152, + 474, + 256 + ], + "score": 0.776, + "html": "
JobOperationM1M2M3
Job 1O1,132
O1.235
O1.343
Job 2O2,12
02.24
02.33
Job 3O3,134
03.222
", + "type": "table", + "image_path": "4261351e53a18de76aff84d9645da1b0e5c1ee65d6c98a9dfa3509a2e1823974.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 307, + 152, + 474, + 165.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 307, + 165.0, + 474, + 178.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 307, + 178.0, + 474, + 191.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 307, + 191.0, + 474, + 204.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 307, + 204.0, + 474, + 217.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 307, + 217.0, + 474, + 230.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 307, + 230.0, + 474, + 243.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 307, + 243.0, + 474, + 256.0 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 5 + }, + { + "type": "table", + "bbox": [ + 135, + 153, + 300, + 256 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 172, + 142, + 263, + 152 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 171, + 141, + 264, + 153 + ], + "spans": [ + { + "bbox": [ + 171, + 141, + 194, + 153 + ], + "score": 1.0, + "content": "(a) A", + "type": "text" + }, + { + "bbox": [ + 194, + 142, + 216, + 151 + ], + "score": 0.75, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 141, + 264, + 153 + ], + "score": 1.0, + "content": "JSP instance", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "table_body", + "bbox": [ + 135, + 153, + 300, + 256 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 135, + 153, + 300, + 256 + ], + "spans": [ + { + "bbox": [ + 135, + 153, + 300, + 256 + ], + "score": 0.881, + "html": "
JobOperationM1M2M3
Job 1O1,13
O1.25
O1.34
Job 2O2,12
02,24
O2,33
Job 3O3,13
03.22
", + "type": "table", + "image_path": "ee2298f6d1a4eedfbf6f7eea3fd20b3dd269d7f61439042e8181cd727720a23e.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 135, + 153, + 300, + 167.71428571428572 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 135, + 167.71428571428572, + 300, + 182.42857142857144 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 135, + 182.42857142857144, + 300, + 197.14285714285717 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 135, + 197.14285714285717, + 300, + 211.8571428571429 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 135, + 211.8571428571429, + 300, + 226.5714285714286 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 135, + 226.5714285714286, + 300, + 241.28571428571433 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 135, + 241.28571428571433, + 300, + 256.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 6.5 + }, + { + "type": "image", + "bbox": [ + 109, + 289, + 498, + 353 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 289, + 498, + 353 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 289, + 498, + 353 + ], + "spans": [ + { + "bbox": [ + 109, + 289, + 498, + 353 + ], + "score": 0.952, + "type": "image", + "image_path": "54e0e1086515bc311b9d4a17fdee1c89c424b15d95726a4af33fc4a68fbe5ad1.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 109, + 289, + 498, + 310.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 109, + 310.3333333333333, + 498, + 331.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 109, + 331.66666666666663, + 498, + 352.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 362, + 504, + 397 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 288, + 375 + ], + "score": 1.0, + "content": "Figure 1: Both (a) and (b) are solutions of the", + "type": "text" + }, + { + "bbox": [ + 288, + 363, + 305, + 374 + ], + "score": 0.38, + "content": "3 \\mathrm { x } 3", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "JSP instance in Table 1 (a), and the former has the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 342, + 386 + ], + "score": 1.0, + "content": "minimal makespan, 12. Both (c) and (d) are solutions of the", + "type": "text" + }, + { + "bbox": [ + 343, + 374, + 359, + 384 + ], + "score": 0.43, + "content": "3 \\mathrm { x } 3", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "FJSP instance in Table 1 (b), and the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 384, + 258, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 258, + 397 + ], + "score": 1.0, + "content": "former has the minimal makespan, 9.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 90, + 410, + 505, + 482 + ], + "lines": [ + { + "bbox": [ + 89, + 411, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 89, + 413, + 100, + 422 + ], + "score": 1.0, + "content": "93", + "type": "text" + }, + { + "bbox": [ + 106, + 411, + 506, + 423 + ], + "score": 1.0, + "content": "While there are different expected goals, such as makespan, tardiness, etc., this paper focuses on", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 421, + 422, + 456 + ], + "spans": [ + { + "bbox": [ + 89, + 424, + 99, + 434 + ], + "score": 1.0, + "content": "94", + "type": "text" + }, + { + "bbox": [ + 101, + 421, + 207, + 456 + ], + "score": 1.0, + "content": "makespan. Let the first osolution is defined to be", + "type": "text" + }, + { + "bbox": [ + 289, + 423, + 314, + 432 + ], + "score": 0.9, + "content": "\\tau = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 421, + 372, + 456 + ], + "score": 1.0, + "content": "in a JSP soluti all operations", + "type": "text" + }, + { + "bbox": [ + 390, + 421, + 422, + 456 + ], + "score": 1.0, + "content": "nitially. , where", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 433, + 504, + 449 + ], + "spans": [ + { + "bbox": [ + 89, + 437, + 99, + 448 + ], + "score": 1.0, + "content": "95", + "type": "text" + }, + { + "bbox": [ + 208, + 433, + 298, + 448 + ], + "score": 0.93, + "content": "T ^ { ( m k s p ) } = \\operatorname* { m a x } ( \\tau _ { j , i } ^ { ( c ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 435, + 390, + 448 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 433, + 504, + 449 + ], + "score": 0.93, + "content": "\\tau _ { j , i } ^ { ( c ) } = \\tau _ { j , i } ^ { ( s ) } + T _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 89, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 89, + 450, + 100, + 459 + ], + "score": 1.0, + "content": "96", + "type": "text" + }, + { + "bbox": [ + 105, + 447, + 232, + 461 + ], + "score": 1.0, + "content": "denotes the completion time of", + "type": "text" + }, + { + "bbox": [ + 233, + 448, + 250, + 460 + ], + "score": 0.9, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 447, + 506, + 461 + ], + "score": 1.0, + "content": ". The makespans for the two solutions illustrated in Figure 1 (a)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 88, + 457, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 88, + 460, + 100, + 471 + ], + "score": 1.0, + "content": "97", + "type": "text" + }, + { + "bbox": [ + 104, + 457, + 506, + 471 + ], + "score": 1.0, + "content": "and (b) are 12 and 15 respectively. The objective is to derive a solution that minimizes the makespan", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 467, + 354, + 484 + ], + "spans": [ + { + "bbox": [ + 89, + 472, + 100, + 482 + ], + "score": 1.0, + "content": "98", + "type": "text" + }, + { + "bbox": [ + 106, + 469, + 141, + 480 + ], + "score": 0.9, + "content": "T ^ { ( m k s p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 467, + 354, + 484 + ], + "score": 1.0, + "content": ", and the solution of Figure 1 (a) reaches the optimal.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 87, + 487, + 505, + 542 + ], + "lines": [ + { + "bbox": [ + 89, + 487, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 89, + 489, + 100, + 498 + ], + "score": 1.0, + "content": "99", + "type": "text" + }, + { + "bbox": [ + 105, + 487, + 117, + 498 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 117, + 488, + 147, + 497 + ], + "score": 0.88, + "content": "n \\times m", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 487, + 250, + 498 + ], + "score": 1.0, + "content": "FJSP instance is also a", + "type": "text" + }, + { + "bbox": [ + 250, + 488, + 280, + 497 + ], + "score": 0.89, + "content": "n \\times m", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 487, + 506, + 498 + ], + "score": 1.0, + "content": "JSP instance with the following difference. In FJSP,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 86, + 497, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 86, + 499, + 100, + 510 + ], + "score": 1.0, + "content": "100", + "type": "text" + }, + { + "bbox": [ + 105, + 497, + 164, + 510 + ], + "score": 1.0, + "content": "all operations", + "type": "text" + }, + { + "bbox": [ + 165, + 498, + 182, + 510 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 497, + 506, + 510 + ], + "score": 1.0, + "content": "are allowed to be dispatched to multiple designated machines with designated", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 86, + 508, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 86, + 510, + 99, + 520 + ], + "score": 1.0, + "content": "101", + "type": "text" + }, + { + "bbox": [ + 105, + 508, + 280, + 521 + ], + "score": 1.0, + "content": "processing times. Table 1 (b) illustrates a", + "type": "text" + }, + { + "bbox": [ + 281, + 509, + 305, + 519 + ], + "score": 0.89, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 508, + 506, + 521 + ], + "score": 1.0, + "content": "FJSP instance, where multiple machines can be", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 86, + 518, + 508, + 533 + ], + "spans": [ + { + "bbox": [ + 86, + 521, + 100, + 532 + ], + "score": 1.0, + "content": "102", + "type": "text" + }, + { + "bbox": [ + 104, + 518, + 508, + 533 + ], + "score": 1.0, + "content": "designated to be processed for one operation. Figure 1 (c) illustrates a solution of an FJSP instance,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 86, + 531, + 312, + 542 + ], + "spans": [ + { + "bbox": [ + 86, + 532, + 100, + 541 + ], + "score": 1.0, + "content": "103", + "type": "text" + }, + { + "bbox": [ + 105, + 531, + 312, + 542 + ], + "score": 1.0, + "content": "which takes a shorter time than that in Figure 1 (d).", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 91, + 558, + 232, + 570 + ], + "lines": [ + { + "bbox": [ + 87, + 557, + 234, + 571 + ], + "spans": [ + { + "bbox": [ + 87, + 557, + 234, + 571 + ], + "score": 1.0, + "content": "104 2.2 Construction Heuristics", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 86, + 579, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 86, + 579, + 505, + 593 + ], + "spans": [ + { + "bbox": [ + 86, + 582, + 99, + 591 + ], + "score": 1.0, + "content": "105", + "type": "text" + }, + { + "bbox": [ + 104, + 579, + 505, + 593 + ], + "score": 1.0, + "content": "An approach to solving these scheduling problems is to construct solutions step by step in a greedy", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 86, + 590, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 86, + 593, + 100, + 602 + ], + "score": 1.0, + "content": "106", + "type": "text" + }, + { + "bbox": [ + 105, + 590, + 506, + 603 + ], + "score": 1.0, + "content": "manner, and the heuristics based on this approach is called construction heuristics in this paper. In", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 86, + 601, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 86, + 603, + 100, + 614 + ], + "score": 1.0, + "content": "107", + "type": "text" + }, + { + "bbox": [ + 105, + 601, + 506, + 615 + ], + "score": 1.0, + "content": "the approach of construction heuristics, a scheduling solution is constructed through a sequence of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 86, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 86, + 614, + 100, + 625 + ], + "score": 1.0, + "content": "108", + "type": "text" + }, + { + "bbox": [ + 104, + 612, + 506, + 625 + ], + "score": 1.0, + "content": "partial solutions in a chronicle order of dispatching operations step by step, defined as follows. The", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 86, + 623, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 86, + 625, + 100, + 636 + ], + "score": 1.0, + "content": "109", + "type": "text" + }, + { + "bbox": [ + 106, + 624, + 111, + 633 + ], + "score": 0.72, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 112, + 623, + 186, + 636 + ], + "score": 1.0, + "content": "-th partial solution", + "type": "text" + }, + { + "bbox": [ + 187, + 624, + 198, + 635 + ], + "score": 0.87, + "content": "S _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 623, + 337, + 636 + ], + "score": 1.0, + "content": "associates with a dispatching time", + "type": "text" + }, + { + "bbox": [ + 338, + 625, + 347, + 635 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 623, + 505, + 636 + ], + "score": 1.0, + "content": "and includes a partial set of operations", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 86, + 634, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 86, + 636, + 100, + 646 + ], + "score": 1.0, + "content": "110", + "type": "text" + }, + { + "bbox": [ + 105, + 634, + 226, + 647 + ], + "score": 1.0, + "content": "that have been dispatched by", + "type": "text" + }, + { + "bbox": [ + 227, + 636, + 236, + 645 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 634, + 505, + 647 + ], + "score": 1.0, + "content": "(inclusive) while satisfying the above JSP constraints, and all the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 86, + 645, + 504, + 658 + ], + "spans": [ + { + "bbox": [ + 86, + 648, + 99, + 657 + ], + "score": 1.0, + "content": "111", + "type": "text" + }, + { + "bbox": [ + 104, + 645, + 293, + 658 + ], + "score": 1.0, + "content": "remaining operations must be dispatched after", + "type": "text" + }, + { + "bbox": [ + 293, + 647, + 302, + 656 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 645, + 492, + 658 + ], + "score": 1.0, + "content": "(inclusive). The whole construction starts with", + "type": "text" + }, + { + "bbox": [ + 492, + 645, + 504, + 657 + ], + "score": 0.87, + "content": "S _ { 0 }", + "type": "inline_equation" + } + ], + "index": 45 + }, + { + "bbox": [ + 86, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 86, + 658, + 100, + 668 + ], + "score": 1.0, + "content": "112", + "type": "text" + }, + { + "bbox": [ + 105, + 655, + 401, + 669 + ], + "score": 1.0, + "content": "where none of operations have been dispatched and the dispatching time is", + "type": "text" + }, + { + "bbox": [ + 401, + 657, + 430, + 667 + ], + "score": 0.91, + "content": "\\tau _ { 0 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 655, + 470, + 669 + ], + "score": 1.0, + "content": ". For each", + "type": "text" + }, + { + "bbox": [ + 470, + 657, + 481, + 667 + ], + "score": 0.87, + "content": "S _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 655, + 506, + 669 + ], + "score": 1.0, + "content": ", a set", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 86, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 100, + 678 + ], + "score": 1.0, + "content": "113", + "type": "text" + }, + { + "bbox": [ + 105, + 667, + 358, + 680 + ], + "score": 1.0, + "content": "of operations to be chosen for dispatching form a set of pairs of", + "type": "text" + }, + { + "bbox": [ + 358, + 667, + 389, + 678 + ], + "score": 0.88, + "content": "( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 667, + 463, + 680 + ], + "score": 1.0, + "content": ", called candidates", + "type": "text" + }, + { + "bbox": [ + 463, + 668, + 475, + 678 + ], + "score": 0.87, + "content": "C _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 667, + 506, + 680 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 86, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 100, + 690 + ], + "score": 1.0, + "content": "114", + "type": "text" + }, + { + "bbox": [ + 105, + 677, + 150, + 691 + ], + "score": 1.0, + "content": "operations", + "type": "text" + }, + { + "bbox": [ + 150, + 678, + 160, + 688 + ], + "score": 0.8, + "content": "O", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 677, + 329, + 691 + ], + "score": 1.0, + "content": "are allowed to be dispatched on machines", + "type": "text" + }, + { + "bbox": [ + 329, + 678, + 341, + 688 + ], + "score": 0.81, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 677, + 351, + 691 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 352, + 680, + 360, + 689 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 677, + 506, + 691 + ], + "score": 1.0, + "content": ". An agent (or a heuristic algorithm)", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 86, + 688, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 100, + 700 + ], + "score": 1.0, + "content": "115", + "type": "text" + }, + { + "bbox": [ + 104, + 688, + 223, + 702 + ], + "score": 1.0, + "content": "chooses one from candidates", + "type": "text" + }, + { + "bbox": [ + 224, + 690, + 235, + 700 + ], + "score": 0.87, + "content": "C _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 688, + 470, + 702 + ], + "score": 1.0, + "content": "for dispatching, and transits the partial solution to the next", + "type": "text" + }, + { + "bbox": [ + 471, + 689, + 491, + 701 + ], + "score": 0.91, + "content": "S _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 688, + 507, + 702 + ], + "score": 1.0, + "content": ". If", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 86, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 100, + 711 + ], + "score": 1.0, + "content": "116", + "type": "text" + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "there exists no operations for dispatching, the whole solution construction process is done and the", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 86, + 711, + 416, + 723 + ], + "spans": [ + { + "bbox": [ + 86, + 712, + 100, + 723 + ], + "score": 1.0, + "content": "117", + "type": "text" + }, + { + "bbox": [ + 104, + 711, + 416, + 723 + ], + "score": 1.0, + "content": "partial solution is a solution, since no further operations are to be dispatched.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 45 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 89, + 72, + 505, + 109 + ], + "lines": [ + { + "bbox": [ + 90, + 72, + 505, + 85 + ], + "spans": [ + { + "bbox": [ + 90, + 72, + 211, + 85 + ], + "score": 1.0, + "content": "90 one of the three machines", + "type": "text" + }, + { + "bbox": [ + 211, + 73, + 273, + 85 + ], + "score": 0.94, + "content": "\\{ M _ { 1 } , M _ { 2 } , M _ { 3 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 72, + 505, + 85 + ], + "score": 1.0, + "content": "in the table. A solution of a JSP instance is to dispatch all", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 78, + 511, + 108 + ], + "spans": [ + { + "bbox": [ + 89, + 88, + 100, + 97 + ], + "score": 1.0, + "content": "91", + "type": "text" + }, + { + "bbox": [ + 98, + 78, + 511, + 108 + ], + "score": 1.0, + "content": "operations Oj,i to the corresponding machine Mj,i at time τ (s)j,i , such that the above constraints are", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 100, + 100, + 109 + ], + "spans": [ + { + "bbox": [ + 89, + 100, + 100, + 109 + ], + "score": 1.0, + "content": "92", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 411, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 89, + 413, + 100, + 422 + ], + "score": 1.0, + "content": "93", + "type": "text" + }, + { + "bbox": [ + 106, + 411, + 506, + 423 + ], + "score": 1.0, + "content": "While there are different expected goals, such as makespan, tardiness, etc., this paper focuses on", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 421, + 422, + 456 + ], + "spans": [ + { + "bbox": [ + 89, + 424, + 99, + 434 + ], + "score": 1.0, + "content": "94", + "type": "text" + }, + { + "bbox": [ + 101, + 421, + 207, + 456 + ], + "score": 1.0, + "content": "makespan. Let the first osolution is defined to be", + "type": "text" + }, + { + "bbox": [ + 289, + 423, + 314, + 432 + ], + "score": 0.9, + "content": "\\tau = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 421, + 372, + 456 + ], + "score": 1.0, + "content": "in a JSP soluti all operations", + "type": "text" + }, + { + "bbox": [ + 390, + 421, + 422, + 456 + ], + "score": 1.0, + "content": "nitially. , where", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 433, + 504, + 449 + ], + "spans": [ + { + "bbox": [ + 89, + 437, + 99, + 448 + ], + "score": 1.0, + "content": "95", + "type": "text" + }, + { + "bbox": [ + 208, + 433, + 298, + 448 + ], + "score": 0.93, + "content": "T ^ { ( m k s p ) } = \\operatorname* { m a x } ( \\tau _ { j , i } ^ { ( c ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 435, + 390, + 448 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 433, + 504, + 449 + ], + "score": 0.93, + "content": "\\tau _ { j , i } ^ { ( c ) } = \\tau _ { j , i } ^ { ( s ) } + T _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 89, + 450, + 100, + 459 + ], + "score": 1.0, + "content": "96", + "type": "text" + }, + { + "bbox": [ + 105, + 447, + 232, + 461 + ], + "score": 1.0, + "content": "denotes the completion time of", + "type": "text" + }, + { + "bbox": [ + 233, + 448, + 250, + 460 + ], + "score": 0.9, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 447, + 506, + 461 + ], + "score": 1.0, + "content": ". The makespans for the two solutions illustrated in Figure 1 (a)", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 457, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 88, + 460, + 100, + 471 + ], + "score": 1.0, + "content": "97", + "type": "text" + }, + { + "bbox": [ + 104, + 457, + 506, + 471 + ], + "score": 1.0, + "content": "and (b) are 12 and 15 respectively. The objective is to derive a solution that minimizes the makespan", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 467, + 354, + 484 + ], + "spans": [ + { + "bbox": [ + 89, + 472, + 100, + 482 + ], + "score": 1.0, + "content": "98", + "type": "text" + }, + { + "bbox": [ + 106, + 469, + 141, + 480 + ], + "score": 0.9, + "content": "T ^ { ( m k s p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 467, + 354, + 484 + ], + "score": 1.0, + "content": ", and the solution of Figure 1 (a) reaches the optimal.", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 487, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 89, + 489, + 100, + 498 + ], + "score": 1.0, + "content": "99", + "type": "text" + }, + { + "bbox": [ + 105, + 487, + 117, + 498 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 117, + 488, + 147, + 497 + ], + "score": 0.88, + "content": "n \\times m", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 487, + 250, + 498 + ], + "score": 1.0, + "content": "FJSP instance is also a", + "type": "text" + }, + { + "bbox": [ + 250, + 488, + 280, + 497 + ], + "score": 0.89, + "content": "n \\times m", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 487, + 506, + 498 + ], + "score": 1.0, + "content": "JSP instance with the following difference. In FJSP,", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 497, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 86, + 499, + 100, + 510 + ], + "score": 1.0, + "content": "100", + "type": "text" + }, + { + "bbox": [ + 105, + 497, + 164, + 510 + ], + "score": 1.0, + "content": "all operations", + "type": "text" + }, + { + "bbox": [ + 165, + 498, + 182, + 510 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 497, + 506, + 510 + ], + "score": 1.0, + "content": "are allowed to be dispatched to multiple designated machines with designated", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 508, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 86, + 510, + 99, + 520 + ], + "score": 1.0, + "content": "101", + "type": "text" + }, + { + "bbox": [ + 105, + 508, + 280, + 521 + ], + "score": 1.0, + "content": "processing times. Table 1 (b) illustrates a", + "type": "text" + }, + { + "bbox": [ + 281, + 509, + 305, + 519 + ], + "score": 0.89, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 508, + 506, + 521 + ], + "score": 1.0, + "content": "FJSP instance, where multiple machines can be", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 518, + 508, + 533 + ], + "spans": [ + { + "bbox": [ + 86, + 521, + 100, + 532 + ], + "score": 1.0, + "content": "102", + "type": "text" + }, + { + "bbox": [ + 104, + 518, + 508, + 533 + ], + "score": 1.0, + "content": "designated to be processed for one operation. Figure 1 (c) illustrates a solution of an FJSP instance,", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 531, + 312, + 542 + ], + "spans": [ + { + "bbox": [ + 86, + 532, + 100, + 541 + ], + "score": 1.0, + "content": "103", + "type": "text" + }, + { + "bbox": [ + 105, + 531, + 312, + 542 + ], + "score": 1.0, + "content": "which takes a shorter time than that in Figure 1 (d).", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + } + ], + "index": 1, + "bbox_fs": [ + 89, + 72, + 511, + 109 + ] + }, + { + "type": "table", + "bbox": [ + 307, + 152, + 474, + 256 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 240, + 121, + 371, + 133 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 240, + 120, + 371, + 133 + ], + "spans": [ + { + "bbox": [ + 240, + 120, + 371, + 133 + ], + "score": 1.0, + "content": "Table 1: JSP and FJSP instances", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "table_caption", + "bbox": [ + 342, + 142, + 439, + 152 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 342, + 141, + 440, + 153 + ], + "spans": [ + { + "bbox": [ + 342, + 141, + 365, + 153 + ], + "score": 1.0, + "content": "(b) A", + "type": "text" + }, + { + "bbox": [ + 365, + 142, + 387, + 151 + ], + "score": 0.73, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 141, + 440, + 153 + ], + "score": 1.0, + "content": "FJSP instance", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "table_body", + "bbox": [ + 307, + 152, + 474, + 256 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 307, + 152, + 474, + 256 + ], + "spans": [ + { + "bbox": [ + 307, + 152, + 474, + 256 + ], + "score": 0.776, + "html": "
JobOperationM1M2M3
Job 1O1,132
O1.235
O1.343
Job 2O2,12
02.24
02.33
Job 3O3,134
03.222
", + "type": "table", + "image_path": "4261351e53a18de76aff84d9645da1b0e5c1ee65d6c98a9dfa3509a2e1823974.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 307, + 152, + 474, + 165.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 307, + 165.0, + 474, + 178.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 307, + 178.0, + 474, + 191.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 307, + 191.0, + 474, + 204.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 307, + 204.0, + 474, + 217.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 307, + 217.0, + 474, + 230.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 307, + 230.0, + 474, + 243.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 307, + 243.0, + 474, + 256.0 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 5 + }, + { + "type": "table", + "bbox": [ + 135, + 153, + 300, + 256 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 172, + 142, + 263, + 152 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 171, + 141, + 264, + 153 + ], + "spans": [ + { + "bbox": [ + 171, + 141, + 194, + 153 + ], + "score": 1.0, + "content": "(a) A", + "type": "text" + }, + { + "bbox": [ + 194, + 142, + 216, + 151 + ], + "score": 0.75, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 141, + 264, + 153 + ], + "score": 1.0, + "content": "JSP instance", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "table_body", + "bbox": [ + 135, + 153, + 300, + 256 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 135, + 153, + 300, + 256 + ], + "spans": [ + { + "bbox": [ + 135, + 153, + 300, + 256 + ], + "score": 0.881, + "html": "
JobOperationM1M2M3
Job 1O1,13
O1.25
O1.34
Job 2O2,12
02,24
O2,33
Job 3O3,13
03.22
", + "type": "table", + "image_path": "ee2298f6d1a4eedfbf6f7eea3fd20b3dd269d7f61439042e8181cd727720a23e.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 135, + 153, + 300, + 167.71428571428572 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 135, + 167.71428571428572, + 300, + 182.42857142857144 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 135, + 182.42857142857144, + 300, + 197.14285714285717 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 135, + 197.14285714285717, + 300, + 211.8571428571429 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 135, + 211.8571428571429, + 300, + 226.5714285714286 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 135, + 226.5714285714286, + 300, + 241.28571428571433 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 135, + 241.28571428571433, + 300, + 256.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 6.5 + }, + { + "type": "image", + "bbox": [ + 109, + 289, + 498, + 353 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 289, + 498, + 353 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 289, + 498, + 353 + ], + "spans": [ + { + "bbox": [ + 109, + 289, + 498, + 353 + ], + "score": 0.952, + "type": "image", + "image_path": "54e0e1086515bc311b9d4a17fdee1c89c424b15d95726a4af33fc4a68fbe5ad1.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 109, + 289, + 498, + 310.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 109, + 310.3333333333333, + 498, + 331.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 109, + 331.66666666666663, + 498, + 352.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 362, + 504, + 397 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 288, + 375 + ], + "score": 1.0, + "content": "Figure 1: Both (a) and (b) are solutions of the", + "type": "text" + }, + { + "bbox": [ + 288, + 363, + 305, + 374 + ], + "score": 0.38, + "content": "3 \\mathrm { x } 3", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "JSP instance in Table 1 (a), and the former has the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 342, + 386 + ], + "score": 1.0, + "content": "minimal makespan, 12. Both (c) and (d) are solutions of the", + "type": "text" + }, + { + "bbox": [ + 343, + 374, + 359, + 384 + ], + "score": 0.43, + "content": "3 \\mathrm { x } 3", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "FJSP instance in Table 1 (b), and the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 384, + 258, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 258, + 397 + ], + "score": 1.0, + "content": "former has the minimal makespan, 9.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + } + ], + "index": 23.5 + }, + { + "type": "index", + "bbox": [ + 90, + 410, + 505, + 482 + ], + "lines": [], + "index": 29.5, + "bbox_fs": [ + 88, + 411, + 506, + 484 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 87, + 487, + 505, + 542 + ], + "lines": [], + "index": 35, + "bbox_fs": [ + 86, + 487, + 508, + 542 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 91, + 558, + 232, + 570 + ], + "lines": [ + { + "bbox": [ + 87, + 557, + 234, + 571 + ], + "spans": [ + { + "bbox": [ + 87, + 557, + 234, + 571 + ], + "score": 1.0, + "content": "104 2.2 Construction Heuristics", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38 + }, + { + "type": "index", + "bbox": [ + 86, + 579, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 86, + 579, + 505, + 593 + ], + "spans": [ + { + "bbox": [ + 86, + 582, + 99, + 591 + ], + "score": 1.0, + "content": "105", + "type": "text" + }, + { + "bbox": [ + 104, + 579, + 505, + 593 + ], + "score": 1.0, + "content": "An approach to solving these scheduling problems is to construct solutions step by step in a greedy", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 590, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 86, + 593, + 100, + 602 + ], + "score": 1.0, + "content": "106", + "type": "text" + }, + { + "bbox": [ + 105, + 590, + 506, + 603 + ], + "score": 1.0, + "content": "manner, and the heuristics based on this approach is called construction heuristics in this paper. In", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 601, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 86, + 603, + 100, + 614 + ], + "score": 1.0, + "content": "107", + "type": "text" + }, + { + "bbox": [ + 105, + 601, + 506, + 615 + ], + "score": 1.0, + "content": "the approach of construction heuristics, a scheduling solution is constructed through a sequence of", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 86, + 614, + 100, + 625 + ], + "score": 1.0, + "content": "108", + "type": "text" + }, + { + "bbox": [ + 104, + 612, + 506, + 625 + ], + "score": 1.0, + "content": "partial solutions in a chronicle order of dispatching operations step by step, defined as follows. The", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 623, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 86, + 625, + 100, + 636 + ], + "score": 1.0, + "content": "109", + "type": "text" + }, + { + "bbox": [ + 106, + 624, + 111, + 633 + ], + "score": 0.72, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 112, + 623, + 186, + 636 + ], + "score": 1.0, + "content": "-th partial solution", + "type": "text" + }, + { + "bbox": [ + 187, + 624, + 198, + 635 + ], + "score": 0.87, + "content": "S _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 623, + 337, + 636 + ], + "score": 1.0, + "content": "associates with a dispatching time", + "type": "text" + }, + { + "bbox": [ + 338, + 625, + 347, + 635 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 623, + 505, + 636 + ], + "score": 1.0, + "content": "and includes a partial set of operations", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 634, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 86, + 636, + 100, + 646 + ], + "score": 1.0, + "content": "110", + "type": "text" + }, + { + "bbox": [ + 105, + 634, + 226, + 647 + ], + "score": 1.0, + "content": "that have been dispatched by", + "type": "text" + }, + { + "bbox": [ + 227, + 636, + 236, + 645 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 634, + 505, + 647 + ], + "score": 1.0, + "content": "(inclusive) while satisfying the above JSP constraints, and all the", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 645, + 504, + 658 + ], + "spans": [ + { + "bbox": [ + 86, + 648, + 99, + 657 + ], + "score": 1.0, + "content": "111", + "type": "text" + }, + { + "bbox": [ + 104, + 645, + 293, + 658 + ], + "score": 1.0, + "content": "remaining operations must be dispatched after", + "type": "text" + }, + { + "bbox": [ + 293, + 647, + 302, + 656 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 645, + 492, + 658 + ], + "score": 1.0, + "content": "(inclusive). The whole construction starts with", + "type": "text" + }, + { + "bbox": [ + 492, + 645, + 504, + 657 + ], + "score": 0.87, + "content": "S _ { 0 }", + "type": "inline_equation" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 86, + 658, + 100, + 668 + ], + "score": 1.0, + "content": "112", + "type": "text" + }, + { + "bbox": [ + 105, + 655, + 401, + 669 + ], + "score": 1.0, + "content": "where none of operations have been dispatched and the dispatching time is", + "type": "text" + }, + { + "bbox": [ + 401, + 657, + 430, + 667 + ], + "score": 0.91, + "content": "\\tau _ { 0 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 655, + 470, + 669 + ], + "score": 1.0, + "content": ". For each", + "type": "text" + }, + { + "bbox": [ + 470, + 657, + 481, + 667 + ], + "score": 0.87, + "content": "S _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 655, + 506, + 669 + ], + "score": 1.0, + "content": ", a set", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 100, + 678 + ], + "score": 1.0, + "content": "113", + "type": "text" + }, + { + "bbox": [ + 105, + 667, + 358, + 680 + ], + "score": 1.0, + "content": "of operations to be chosen for dispatching form a set of pairs of", + "type": "text" + }, + { + "bbox": [ + 358, + 667, + 389, + 678 + ], + "score": 0.88, + "content": "( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 667, + 463, + 680 + ], + "score": 1.0, + "content": ", called candidates", + "type": "text" + }, + { + "bbox": [ + 463, + 668, + 475, + 678 + ], + "score": 0.87, + "content": "C _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 667, + 506, + 680 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 47, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 100, + 690 + ], + "score": 1.0, + "content": "114", + "type": "text" + }, + { + "bbox": [ + 105, + 677, + 150, + 691 + ], + "score": 1.0, + "content": "operations", + "type": "text" + }, + { + "bbox": [ + 150, + 678, + 160, + 688 + ], + "score": 0.8, + "content": "O", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 677, + 329, + 691 + ], + "score": 1.0, + "content": "are allowed to be dispatched on machines", + "type": "text" + }, + { + "bbox": [ + 329, + 678, + 341, + 688 + ], + "score": 0.81, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 677, + 351, + 691 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 352, + 680, + 360, + 689 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 677, + 506, + 691 + ], + "score": 1.0, + "content": ". An agent (or a heuristic algorithm)", + "type": "text" + } + ], + "index": 48, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 688, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 100, + 700 + ], + "score": 1.0, + "content": "115", + "type": "text" + }, + { + "bbox": [ + 104, + 688, + 223, + 702 + ], + "score": 1.0, + "content": "chooses one from candidates", + "type": "text" + }, + { + "bbox": [ + 224, + 690, + 235, + 700 + ], + "score": 0.87, + "content": "C _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 688, + 470, + 702 + ], + "score": 1.0, + "content": "for dispatching, and transits the partial solution to the next", + "type": "text" + }, + { + "bbox": [ + 471, + 689, + 491, + 701 + ], + "score": 0.91, + "content": "S _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 688, + 507, + 702 + ], + "score": 1.0, + "content": ". If", + "type": "text" + } + ], + "index": 49, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 100, + 711 + ], + "score": 1.0, + "content": "116", + "type": "text" + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "there exists no operations for dispatching, the whole solution construction process is done and the", + "type": "text" + } + ], + "index": 50, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 711, + 416, + 723 + ], + "spans": [ + { + "bbox": [ + 86, + 712, + 100, + 723 + ], + "score": 1.0, + "content": "117", + "type": "text" + }, + { + "bbox": [ + 104, + 711, + 416, + 723 + ], + "score": 1.0, + "content": "partial solution is a solution, since no further operations are to be dispatched.", + "type": "text" + } + ], + "index": 51, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 334, + 506, + 348 + ], + "spans": [ + { + "bbox": [ + 85, + 336, + 100, + 347 + ], + "score": 1.0, + "content": "118", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 334, + 506, + 348 + ], + "score": 1.0, + "content": "Figure 2 illustrates a solution construction process for the 3x3 JSP instance in Table 1(a), constructed", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 345, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 86, + 348, + 100, + 357 + ], + "score": 1.0, + "content": "119", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 345, + 378, + 358 + ], + "score": 1.0, + "content": "through nine partial solutions step by step. The initial partial solution", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 378, + 346, + 389, + 356 + ], + "score": 0.87, + "content": "S _ { 0 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 390, + 345, + 505, + 358 + ], + "score": 1.0, + "content": "starts without any operations", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 355, + 503, + 370 + ], + "spans": [ + { + "bbox": [ + 85, + 358, + 100, + 369 + ], + "score": 1.0, + "content": "120", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 355, + 319, + 370 + ], + "score": 1.0, + "content": "dispatched as in Figure 2 (a). The initial candidates", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 320, + 357, + 332, + 367 + ], + "score": 0.9, + "content": "C _ { 0 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 333, + 355, + 349, + 370 + ], + "score": 1.0, + "content": "are", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 349, + 356, + 503, + 369 + ], + "score": 0.83, + "content": "\\{ ( M _ { 1 } , O _ { 1 , 1 } ) , ( M _ { 3 } , O _ { 2 , 1 } ) , ( \\bar { M _ { 1 } } , O _ { 3 , 1 } ) \\}", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 367, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 86, + 369, + 99, + 379 + ], + "score": 1.0, + "content": "121", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 367, + 382, + 380 + ], + "score": 1.0, + "content": "Following some heuristic, construct a solution from partial solution", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 382, + 368, + 394, + 379 + ], + "score": 0.87, + "content": "S _ { 0 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 394, + 367, + 405, + 380 + ], + "score": 1.0, + "content": "to", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 406, + 368, + 417, + 379 + ], + "score": 0.87, + "content": "S _ { 9 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 417, + 367, + 506, + 380 + ], + "score": 1.0, + "content": "step by step as in the", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 378, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 86, + 380, + 100, + 390 + ], + "score": 1.0, + "content": "122", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 378, + 327, + 391 + ], + "score": 1.0, + "content": "Figure, where the dashed line in red indicate the time", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 328, + 380, + 337, + 389 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 337, + 378, + 395, + 391 + ], + "score": 1.0, + "content": ". The last one", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 396, + 379, + 407, + 389 + ], + "score": 0.88, + "content": "S _ { 9 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 407, + 378, + 505, + 391 + ], + "score": 1.0, + "content": ", the same as the one in", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 388, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 85, + 390, + 100, + 402 + ], + "score": 1.0, + "content": "123", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 388, + 506, + 403 + ], + "score": 1.0, + "content": "Figure 1 (a), is a solution, since all operations have been dispatched, and the last operation ends at", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 400, + 261, + 413 + ], + "spans": [ + { + "bbox": [ + 86, + 402, + 100, + 411 + ], + "score": 1.0, + "content": "124", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 400, + 261, + 413 + ], + "score": 1.0, + "content": "time 12, the makespan of the solution.", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 86, + 419, + 99, + 428 + ], + "score": 1.0, + "content": "125", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "For FJSP, the process of solution construction is almost the same except for that one operation have", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 426, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 86, + 429, + 100, + 439 + ], + "score": 1.0, + "content": "126", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 426, + 505, + 440 + ], + "score": 1.0, + "content": "multiple choices from candidates. Besides, an approach based on solution construction can be also", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 86, + 440, + 99, + 450 + ], + "score": 1.0, + "content": "127", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 438, + 311, + 450 + ], + "score": 1.0, + "content": "viewed as the so-called Markov decision process", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 311, + 438, + 340, + 450 + ], + "score": 0.37, + "content": "( M D P )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 340, + 438, + 505, + 450 + ], + "score": 1.0, + "content": ", and the MDP formulation for solution", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 449, + 332, + 462 + ], + "spans": [ + { + "bbox": [ + 86, + 451, + 100, + 460 + ], + "score": 1.0, + "content": "128", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 449, + 332, + 462 + ], + "score": 1.0, + "content": "construction is described in more detail in the appendix.", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + } + ], + "index": 45, + "bbox_fs": [ + 86, + 579, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 72, + 482, + 291 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 72, + 482, + 291 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 72, + 482, + 291 + ], + "spans": [ + { + "bbox": [ + 119, + 72, + 482, + 291 + ], + "score": 0.965, + "type": "image", + "image_path": "cb96d32bb1f5dcecc876eaf54a9e7682b46451e9161fb74084ead5c2b39e2e49.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 72, + 482, + 145.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 145.0, + 482, + 218.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 218.0, + 482, + 291.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 148, + 300, + 462, + 312 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 298, + 462, + 314 + ], + "spans": [ + { + "bbox": [ + 147, + 298, + 424, + 314 + ], + "score": 1.0, + "content": "Figure 2: Solution construction, a sequence of partial solutions from", + "type": "text" + }, + { + "bbox": [ + 424, + 301, + 435, + 311 + ], + "score": 0.88, + "content": "S _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 298, + 447, + 314 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 447, + 301, + 459, + 311 + ], + "score": 0.87, + "content": "S _ { 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 298, + 462, + 314 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 86, + 334, + 505, + 411 + ], + "lines": [ + { + "bbox": [ + 85, + 334, + 506, + 348 + ], + "spans": [ + { + "bbox": [ + 85, + 336, + 100, + 347 + ], + "score": 1.0, + "content": "118", + "type": "text" + }, + { + "bbox": [ + 105, + 334, + 506, + 348 + ], + "score": 1.0, + "content": "Figure 2 illustrates a solution construction process for the 3x3 JSP instance in Table 1(a), constructed", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 86, + 345, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 86, + 348, + 100, + 357 + ], + "score": 1.0, + "content": "119", + "type": "text" + }, + { + "bbox": [ + 106, + 345, + 378, + 358 + ], + "score": 1.0, + "content": "through nine partial solutions step by step. The initial partial solution", + "type": "text" + }, + { + "bbox": [ + 378, + 346, + 389, + 356 + ], + "score": 0.87, + "content": "S _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 345, + 505, + 358 + ], + "score": 1.0, + "content": "starts without any operations", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 85, + 355, + 503, + 370 + ], + "spans": [ + { + "bbox": [ + 85, + 358, + 100, + 369 + ], + "score": 1.0, + "content": "120", + "type": "text" + }, + { + "bbox": [ + 105, + 355, + 319, + 370 + ], + "score": 1.0, + "content": "dispatched as in Figure 2 (a). The initial candidates", + "type": "text" + }, + { + "bbox": [ + 320, + 357, + 332, + 367 + ], + "score": 0.9, + "content": "C _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 355, + 349, + 370 + ], + "score": 1.0, + "content": "are", + "type": "text" + }, + { + "bbox": [ + 349, + 356, + 503, + 369 + ], + "score": 0.83, + "content": "\\{ ( M _ { 1 } , O _ { 1 , 1 } ) , ( M _ { 3 } , O _ { 2 , 1 } ) , ( \\bar { M _ { 1 } } , O _ { 3 , 1 } ) \\}", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 86, + 367, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 86, + 369, + 99, + 379 + ], + "score": 1.0, + "content": "121", + "type": "text" + }, + { + "bbox": [ + 106, + 367, + 382, + 380 + ], + "score": 1.0, + "content": "Following some heuristic, construct a solution from partial solution", + "type": "text" + }, + { + "bbox": [ + 382, + 368, + 394, + 379 + ], + "score": 0.87, + "content": "S _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 367, + 405, + 380 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 406, + 368, + 417, + 379 + ], + "score": 0.87, + "content": "S _ { 9 }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 367, + 506, + 380 + ], + "score": 1.0, + "content": "step by step as in the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 86, + 378, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 86, + 380, + 100, + 390 + ], + "score": 1.0, + "content": "122", + "type": "text" + }, + { + "bbox": [ + 106, + 378, + 327, + 391 + ], + "score": 1.0, + "content": "Figure, where the dashed line in red indicate the time", + "type": "text" + }, + { + "bbox": [ + 328, + 380, + 337, + 389 + ], + "score": 0.84, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 378, + 395, + 391 + ], + "score": 1.0, + "content": ". The last one", + "type": "text" + }, + { + "bbox": [ + 396, + 379, + 407, + 389 + ], + "score": 0.88, + "content": "S _ { 9 }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 378, + 505, + 391 + ], + "score": 1.0, + "content": ", the same as the one in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 85, + 388, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 85, + 390, + 100, + 402 + ], + "score": 1.0, + "content": "123", + "type": "text" + }, + { + "bbox": [ + 105, + 388, + 506, + 403 + ], + "score": 1.0, + "content": "Figure 1 (a), is a solution, since all operations have been dispatched, and the last operation ends at", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 86, + 400, + 261, + 413 + ], + "spans": [ + { + "bbox": [ + 86, + 402, + 100, + 411 + ], + "score": 1.0, + "content": "124", + "type": "text" + }, + { + "bbox": [ + 106, + 400, + 261, + 413 + ], + "score": 1.0, + "content": "time 12, the makespan of the solution.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 87, + 416, + 505, + 461 + ], + "lines": [ + { + "bbox": [ + 86, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 86, + 419, + 99, + 428 + ], + "score": 1.0, + "content": "125", + "type": "text" + }, + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "For FJSP, the process of solution construction is almost the same except for that one operation have", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 86, + 426, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 86, + 429, + 100, + 439 + ], + "score": 1.0, + "content": "126", + "type": "text" + }, + { + "bbox": [ + 105, + 426, + 505, + 440 + ], + "score": 1.0, + "content": "multiple choices from candidates. Besides, an approach based on solution construction can be also", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 86, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 86, + 440, + 99, + 450 + ], + "score": 1.0, + "content": "127", + "type": "text" + }, + { + "bbox": [ + 105, + 438, + 311, + 450 + ], + "score": 1.0, + "content": "viewed as the so-called Markov decision process", + "type": "text" + }, + { + "bbox": [ + 311, + 438, + 340, + 450 + ], + "score": 0.37, + "content": "( M D P )", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 438, + 505, + 450 + ], + "score": 1.0, + "content": ", and the MDP formulation for solution", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 86, + 449, + 332, + 462 + ], + "spans": [ + { + "bbox": [ + 86, + 451, + 100, + 460 + ], + "score": 1.0, + "content": "128", + "type": "text" + }, + { + "bbox": [ + 105, + 449, + 332, + 462 + ], + "score": 1.0, + "content": "construction is described in more detail in the appendix.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "title", + "bbox": [ + 89, + 478, + 200, + 492 + ], + "lines": [ + { + "bbox": [ + 85, + 477, + 202, + 496 + ], + "spans": [ + { + "bbox": [ + 85, + 477, + 202, + 496 + ], + "score": 1.0, + "content": "129 3 Our Approach", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 87, + 505, + 505, + 549 + ], + "lines": [ + { + "bbox": [ + 86, + 504, + 507, + 518 + ], + "spans": [ + { + "bbox": [ + 86, + 507, + 99, + 516 + ], + "score": 1.0, + "content": "130", + "type": "text" + }, + { + "bbox": [ + 105, + 504, + 507, + 518 + ], + "score": 1.0, + "content": "In this section, we present a new approach, called residual scheduling, to solving scheduling problems.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 86, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 86, + 518, + 99, + 527 + ], + "score": 1.0, + "content": "131", + "type": "text" + }, + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "We introduce the residual scheduling in Subsection 3.1, describe the design of the graph representation", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 86, + 526, + 505, + 539 + ], + "spans": [ + { + "bbox": [ + 86, + 528, + 100, + 539 + ], + "score": 1.0, + "content": "132", + "type": "text" + }, + { + "bbox": [ + 105, + 526, + 505, + 539 + ], + "score": 1.0, + "content": "in Subsection 3.2, propose a model architecture based on graph neural network in Subsection 3.3 and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 86, + 537, + 327, + 550 + ], + "spans": [ + { + "bbox": [ + 86, + 539, + 99, + 549 + ], + "score": 1.0, + "content": "133", + "type": "text" + }, + { + "bbox": [ + 105, + 537, + 327, + 550 + ], + "score": 1.0, + "content": "present a method to train this model in Subsection 3.4;", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 94, + 565, + 217, + 577 + ], + "lines": [ + { + "bbox": [ + 90, + 563, + 219, + 579 + ], + "spans": [ + { + "bbox": [ + 90, + 563, + 219, + 579 + ], + "score": 1.0, + "content": "34 3.1 Residual Scheduling", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 86, + 586, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 86, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 86, + 587, + 100, + 598 + ], + "score": 1.0, + "content": "135", + "type": "text" + }, + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "score": 1.0, + "content": "In our approach, the key is to remove irrelevant information, particularly for operations, from states", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 86, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 86, + 598, + 100, + 609 + ], + "score": 1.0, + "content": "136", + "type": "text" + }, + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "(including partial solutions). An important benefit from this is that we do not need to include all", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 86, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 86, + 609, + 100, + 620 + ], + "score": 1.0, + "content": "137", + "type": "text" + }, + { + "bbox": [ + 106, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "irrelevant information while training to minimize the makespan. Let us illustrate by the state for the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 86, + 618, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 86, + 621, + 99, + 630 + ], + "score": 1.0, + "content": "138", + "type": "text" + }, + { + "bbox": [ + 105, + 618, + 168, + 632 + ], + "score": 1.0, + "content": "partial solution", + "type": "text" + }, + { + "bbox": [ + 168, + 619, + 180, + 630 + ], + "score": 0.89, + "content": "S _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 618, + 210, + 632 + ], + "score": 1.0, + "content": "at time", + "type": "text" + }, + { + "bbox": [ + 210, + 619, + 239, + 630 + ], + "score": 0.91, + "content": "\\tau _ { 3 } = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 618, + 373, + 632 + ], + "score": 1.0, + "content": "in Figure 2 (d). All processing by", + "type": "text" + }, + { + "bbox": [ + 373, + 620, + 384, + 630 + ], + "score": 0.85, + "content": "\\tau _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 618, + 506, + 632 + ], + "score": 1.0, + "content": "are irrelevant to the remaining", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 86, + 627, + 507, + 644 + ], + "spans": [ + { + "bbox": [ + 86, + 631, + 100, + 642 + ], + "score": 1.0, + "content": "139", + "type": "text" + }, + { + "bbox": [ + 104, + 627, + 226, + 644 + ], + "score": 1.0, + "content": "scheduling. Since operations", + "type": "text" + }, + { + "bbox": [ + 226, + 630, + 246, + 641 + ], + "score": 0.93, + "content": "O _ { 1 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 627, + 264, + 644 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 264, + 630, + 284, + 641 + ], + "score": 0.91, + "content": "O _ { 2 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 627, + 507, + 644 + ], + "score": 1.0, + "content": "are both finished and irrelevant the rest of scheduling,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 86, + 640, + 506, + 654 + ], + "spans": [ + { + "bbox": [ + 86, + 642, + 100, + 653 + ], + "score": 1.0, + "content": "140", + "type": "text" + }, + { + "bbox": [ + 105, + 640, + 258, + 654 + ], + "score": 1.0, + "content": "they can be removed from the state of", + "type": "text" + }, + { + "bbox": [ + 258, + 641, + 270, + 651 + ], + "score": 0.88, + "content": "S _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 640, + 362, + 654 + ], + "score": 1.0, + "content": ". In addition, operation", + "type": "text" + }, + { + "bbox": [ + 362, + 641, + 382, + 653 + ], + "score": 0.92, + "content": "O _ { 2 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 640, + 506, + 654 + ], + "score": 1.0, + "content": "is dispatched at time 2 (before", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 86, + 647, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 86, + 656, + 99, + 667 + ], + "score": 1.0, + "content": "141", + "type": "text" + }, + { + "bbox": [ + 106, + 654, + 138, + 666 + ], + "score": 0.88, + "content": "\\tau _ { 3 } = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 647, + 250, + 675 + ], + "score": 1.0, + "content": ") and its processing time is", + "type": "text" + }, + { + "bbox": [ + 251, + 652, + 294, + 667 + ], + "score": 0.93, + "content": "T _ { 2 , 1 } ^ { ( o p ) } = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 647, + 506, + 675 + ], + "score": 1.0, + "content": ", so the operation is marked as ongoing. Thus, the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 86, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 100, + 680 + ], + "score": 1.0, + "content": "142", + "type": "text" + }, + { + "bbox": [ + 105, + 667, + 253, + 680 + ], + "score": 1.0, + "content": "operation can be modified to start at", + "type": "text" + }, + { + "bbox": [ + 253, + 668, + 281, + 678 + ], + "score": 0.9, + "content": "\\tau _ { 3 } = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 667, + 375, + 680 + ], + "score": 1.0, + "content": "with a processing time", + "type": "text" + }, + { + "bbox": [ + 423, + 667, + 506, + 680 + ], + "score": 1.0, + "content": ". Thus, the modified", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 86, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 86, + 681, + 99, + 690 + ], + "score": 1.0, + "content": "143", + "type": "text" + }, + { + "bbox": [ + 105, + 677, + 141, + 691 + ], + "score": 1.0, + "content": "state for", + "type": "text" + }, + { + "bbox": [ + 141, + 678, + 153, + 689 + ], + "score": 0.89, + "content": "S _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 677, + 233, + 691 + ], + "score": 1.0, + "content": "do not contain both", + "type": "text" + }, + { + "bbox": [ + 234, + 678, + 253, + 690 + ], + "score": 0.91, + "content": "O _ { 1 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 677, + 271, + 691 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 271, + 679, + 290, + 690 + ], + "score": 0.89, + "content": "O _ { 2 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 677, + 342, + 691 + ], + "score": 1.0, + "content": ", and modify", + "type": "text" + }, + { + "bbox": [ + 342, + 679, + 361, + 690 + ], + "score": 0.9, + "content": "O _ { 2 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "as above. Let us consider two more", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 86, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 99, + 700 + ], + "score": 1.0, + "content": "144", + "type": "text" + }, + { + "bbox": [ + 105, + 689, + 165, + 701 + ], + "score": 1.0, + "content": "examples. For", + "type": "text" + }, + { + "bbox": [ + 165, + 689, + 177, + 700 + ], + "score": 0.89, + "content": "S _ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 689, + 261, + 701 + ], + "score": 1.0, + "content": ", one more operation", + "type": "text" + }, + { + "bbox": [ + 261, + 690, + 280, + 701 + ], + "score": 0.91, + "content": "O _ { 2 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "is dispatched and thus marked as ongoing, however, the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 86, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 99, + 712 + ], + "score": 1.0, + "content": "145", + "type": "text" + }, + { + "bbox": [ + 106, + 700, + 126, + 712 + ], + "score": 1.0, + "content": "time", + "type": "text" + }, + { + "bbox": [ + 126, + 701, + 137, + 711 + ], + "score": 0.85, + "content": "\\tau _ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "remains unchanged and no more operations are removed. In this case, the state is almost the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 86, + 710, + 505, + 725 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 99, + 722 + ], + "score": 1.0, + "content": "146", + "type": "text" + }, + { + "bbox": [ + 105, + 710, + 324, + 725 + ], + "score": 1.0, + "content": "same except for including one more ongoing operation", + "type": "text" + }, + { + "bbox": [ + 324, + 711, + 343, + 723 + ], + "score": 0.89, + "content": "O _ { 2 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 710, + 387, + 725 + ], + "score": 1.0, + "content": ". Then, for", + "type": "text" + }, + { + "bbox": [ + 387, + 712, + 398, + 722 + ], + "score": 0.87, + "content": "S _ { 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 710, + 485, + 725 + ], + "score": 1.0, + "content": ", two more operations", + "type": "text" + }, + { + "bbox": [ + 485, + 711, + 505, + 723 + ], + "score": 0.92, + "content": "O _ { 3 , 1 }", + "type": "inline_equation" + } + ], + "index": 32 + } + ], + "index": 26.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 11, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 72, + 482, + 291 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 72, + 482, + 291 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 72, + 482, + 291 + ], + "spans": [ + { + "bbox": [ + 119, + 72, + 482, + 291 + ], + "score": 0.965, + "type": "image", + "image_path": "cb96d32bb1f5dcecc876eaf54a9e7682b46451e9161fb74084ead5c2b39e2e49.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 72, + 482, + 145.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 145.0, + 482, + 218.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 218.0, + 482, + 291.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 148, + 300, + 462, + 312 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 298, + 462, + 314 + ], + "spans": [ + { + "bbox": [ + 147, + 298, + 424, + 314 + ], + "score": 1.0, + "content": "Figure 2: Solution construction, a sequence of partial solutions from", + "type": "text" + }, + { + "bbox": [ + 424, + 301, + 435, + 311 + ], + "score": 0.88, + "content": "S _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 298, + 447, + 314 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 447, + 301, + 459, + 311 + ], + "score": 0.87, + "content": "S _ { 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 298, + 462, + 314 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "index", + "bbox": [ + 86, + 334, + 505, + 411 + ], + "lines": [], + "index": 7, + "bbox_fs": [ + 85, + 334, + 506, + 413 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 87, + 416, + 505, + 461 + ], + "lines": [], + "index": 12.5, + "bbox_fs": [ + 86, + 416, + 505, + 462 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 89, + 478, + 200, + 492 + ], + "lines": [ + { + "bbox": [ + 85, + 477, + 202, + 496 + ], + "spans": [ + { + "bbox": [ + 85, + 477, + 202, + 496 + ], + "score": 1.0, + "content": "129 3 Our Approach", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "index", + "bbox": [ + 87, + 505, + 505, + 549 + ], + "lines": [ + { + "bbox": [ + 86, + 504, + 507, + 518 + ], + "spans": [ + { + "bbox": [ + 86, + 507, + 99, + 516 + ], + "score": 1.0, + "content": "130", + "type": "text" + }, + { + "bbox": [ + 105, + 504, + 507, + 518 + ], + "score": 1.0, + "content": "In this section, we present a new approach, called residual scheduling, to solving scheduling problems.", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 86, + 518, + 99, + 527 + ], + "score": 1.0, + "content": "131", + "type": "text" + }, + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "We introduce the residual scheduling in Subsection 3.1, describe the design of the graph representation", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 526, + 505, + 539 + ], + "spans": [ + { + "bbox": [ + 86, + 528, + 100, + 539 + ], + "score": 1.0, + "content": "132", + "type": "text" + }, + { + "bbox": [ + 105, + 526, + 505, + 539 + ], + "score": 1.0, + "content": "in Subsection 3.2, propose a model architecture based on graph neural network in Subsection 3.3 and", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 537, + 327, + 550 + ], + "spans": [ + { + "bbox": [ + 86, + 539, + 99, + 549 + ], + "score": 1.0, + "content": "133", + "type": "text" + }, + { + "bbox": [ + 105, + 537, + 327, + 550 + ], + "score": 1.0, + "content": "present a method to train this model in Subsection 3.4;", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + } + ], + "index": 17.5, + "bbox_fs": [ + 86, + 504, + 507, + 550 + ] + }, + { + "type": "title", + "bbox": [ + 94, + 565, + 217, + 577 + ], + "lines": [ + { + "bbox": [ + 90, + 563, + 219, + 579 + ], + "spans": [ + { + "bbox": [ + 90, + 563, + 219, + 579 + ], + "score": 1.0, + "content": "34 3.1 Residual Scheduling", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "index", + "bbox": [ + 86, + 586, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 86, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 86, + 587, + 100, + 598 + ], + "score": 1.0, + "content": "135", + "type": "text" + }, + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "score": 1.0, + "content": "In our approach, the key is to remove irrelevant information, particularly for operations, from states", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 86, + 598, + 100, + 609 + ], + "score": 1.0, + "content": "136", + "type": "text" + }, + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "(including partial solutions). An important benefit from this is that we do not need to include all", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 86, + 609, + 100, + 620 + ], + "score": 1.0, + "content": "137", + "type": "text" + }, + { + "bbox": [ + 106, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "irrelevant information while training to minimize the makespan. Let us illustrate by the state for the", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 618, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 86, + 621, + 99, + 630 + ], + "score": 1.0, + "content": "138", + "type": "text" + }, + { + "bbox": [ + 105, + 618, + 168, + 632 + ], + "score": 1.0, + "content": "partial solution", + "type": "text" + }, + { + "bbox": [ + 168, + 619, + 180, + 630 + ], + "score": 0.89, + "content": "S _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 618, + 210, + 632 + ], + "score": 1.0, + "content": "at time", + "type": "text" + }, + { + "bbox": [ + 210, + 619, + 239, + 630 + ], + "score": 0.91, + "content": "\\tau _ { 3 } = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 618, + 373, + 632 + ], + "score": 1.0, + "content": "in Figure 2 (d). All processing by", + "type": "text" + }, + { + "bbox": [ + 373, + 620, + 384, + 630 + ], + "score": 0.85, + "content": "\\tau _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 618, + 506, + 632 + ], + "score": 1.0, + "content": "are irrelevant to the remaining", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 627, + 507, + 644 + ], + "spans": [ + { + "bbox": [ + 86, + 631, + 100, + 642 + ], + "score": 1.0, + "content": "139", + "type": "text" + }, + { + "bbox": [ + 104, + 627, + 226, + 644 + ], + "score": 1.0, + "content": "scheduling. Since operations", + "type": "text" + }, + { + "bbox": [ + 226, + 630, + 246, + 641 + ], + "score": 0.93, + "content": "O _ { 1 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 627, + 264, + 644 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 264, + 630, + 284, + 641 + ], + "score": 0.91, + "content": "O _ { 2 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 627, + 507, + 644 + ], + "score": 1.0, + "content": "are both finished and irrelevant the rest of scheduling,", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 640, + 506, + 654 + ], + "spans": [ + { + "bbox": [ + 86, + 642, + 100, + 653 + ], + "score": 1.0, + "content": "140", + "type": "text" + }, + { + "bbox": [ + 105, + 640, + 258, + 654 + ], + "score": 1.0, + "content": "they can be removed from the state of", + "type": "text" + }, + { + "bbox": [ + 258, + 641, + 270, + 651 + ], + "score": 0.88, + "content": "S _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 640, + 362, + 654 + ], + "score": 1.0, + "content": ". In addition, operation", + "type": "text" + }, + { + "bbox": [ + 362, + 641, + 382, + 653 + ], + "score": 0.92, + "content": "O _ { 2 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 640, + 506, + 654 + ], + "score": 1.0, + "content": "is dispatched at time 2 (before", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 647, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 86, + 656, + 99, + 667 + ], + "score": 1.0, + "content": "141", + "type": "text" + }, + { + "bbox": [ + 106, + 654, + 138, + 666 + ], + "score": 0.88, + "content": "\\tau _ { 3 } = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 647, + 250, + 675 + ], + "score": 1.0, + "content": ") and its processing time is", + "type": "text" + }, + { + "bbox": [ + 251, + 652, + 294, + 667 + ], + "score": 0.93, + "content": "T _ { 2 , 1 } ^ { ( o p ) } = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 647, + 506, + 675 + ], + "score": 1.0, + "content": ", so the operation is marked as ongoing. Thus, the", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 100, + 680 + ], + "score": 1.0, + "content": "142", + "type": "text" + }, + { + "bbox": [ + 105, + 667, + 253, + 680 + ], + "score": 1.0, + "content": "operation can be modified to start at", + "type": "text" + }, + { + "bbox": [ + 253, + 668, + 281, + 678 + ], + "score": 0.9, + "content": "\\tau _ { 3 } = 3", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 667, + 375, + 680 + ], + "score": 1.0, + "content": "with a processing time", + "type": "text" + }, + { + "bbox": [ + 423, + 667, + 506, + 680 + ], + "score": 1.0, + "content": ". Thus, the modified", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 86, + 681, + 99, + 690 + ], + "score": 1.0, + "content": "143", + "type": "text" + }, + { + "bbox": [ + 105, + 677, + 141, + 691 + ], + "score": 1.0, + "content": "state for", + "type": "text" + }, + { + "bbox": [ + 141, + 678, + 153, + 689 + ], + "score": 0.89, + "content": "S _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 677, + 233, + 691 + ], + "score": 1.0, + "content": "do not contain both", + "type": "text" + }, + { + "bbox": [ + 234, + 678, + 253, + 690 + ], + "score": 0.91, + "content": "O _ { 1 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 677, + 271, + 691 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 271, + 679, + 290, + 690 + ], + "score": 0.89, + "content": "O _ { 2 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 677, + 342, + 691 + ], + "score": 1.0, + "content": ", and modify", + "type": "text" + }, + { + "bbox": [ + 342, + 679, + 361, + 690 + ], + "score": 0.9, + "content": "O _ { 2 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "as above. Let us consider two more", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 99, + 700 + ], + "score": 1.0, + "content": "144", + "type": "text" + }, + { + "bbox": [ + 105, + 689, + 165, + 701 + ], + "score": 1.0, + "content": "examples. For", + "type": "text" + }, + { + "bbox": [ + 165, + 689, + 177, + 700 + ], + "score": 0.89, + "content": "S _ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 689, + 261, + 701 + ], + "score": 1.0, + "content": ", one more operation", + "type": "text" + }, + { + "bbox": [ + 261, + 690, + 280, + 701 + ], + "score": 0.91, + "content": "O _ { 2 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "is dispatched and thus marked as ongoing, however, the", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 99, + 712 + ], + "score": 1.0, + "content": "145", + "type": "text" + }, + { + "bbox": [ + 106, + 700, + 126, + 712 + ], + "score": 1.0, + "content": "time", + "type": "text" + }, + { + "bbox": [ + 126, + 701, + 137, + 711 + ], + "score": 0.85, + "content": "\\tau _ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "remains unchanged and no more operations are removed. In this case, the state is almost the", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 710, + 505, + 725 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 99, + 722 + ], + "score": 1.0, + "content": "146", + "type": "text" + }, + { + "bbox": [ + 105, + 710, + 324, + 725 + ], + "score": 1.0, + "content": "same except for including one more ongoing operation", + "type": "text" + }, + { + "bbox": [ + 324, + 711, + 343, + 723 + ], + "score": 0.89, + "content": "O _ { 2 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 710, + 387, + 725 + ], + "score": 1.0, + "content": ". Then, for", + "type": "text" + }, + { + "bbox": [ + 387, + 712, + 398, + 722 + ], + "score": 0.87, + "content": "S _ { 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 710, + 485, + 725 + ], + "score": 1.0, + "content": ", two more operations", + "type": "text" + }, + { + "bbox": [ + 485, + 711, + 505, + 723 + ], + "score": 0.92, + "content": "O _ { 3 , 1 }", + "type": "inline_equation" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 87, + 71, + 506, + 87 + ], + "spans": [ + { + "bbox": [ + 87, + 71, + 123, + 87 + ], + "score": 1.0, + "content": "147 and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 124, + 73, + 143, + 84 + ], + "score": 0.9, + "content": "O _ { 2 , 2 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 143, + 71, + 303, + 87 + ], + "score": 1.0, + "content": "are removed and the ongoing operation", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 304, + 73, + 323, + 85 + ], + "score": 0.91, + "content": "O _ { 1 , 2 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 324, + 71, + 506, + 87 + ], + "score": 1.0, + "content": "changes its processing time to the remaining", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 87, + 83, + 152, + 96 + ], + "spans": [ + { + "bbox": [ + 87, + 83, + 152, + 96 + ], + "score": 1.0, + "content": "148 time (5-3).", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 100, + 505, + 112 + ], + "spans": [ + { + "bbox": [ + 86, + 101, + 100, + 111 + ], + "score": 1.0, + "content": "149", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 100, + 341, + 112 + ], + "score": 1.0, + "content": "For residual scheduling, we also reset the dispatching time", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 342, + 101, + 366, + 110 + ], + "score": 0.9, + "content": "\\tau = 0", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 367, + 100, + 505, + 112 + ], + "score": 1.0, + "content": "for all states with partial solutions", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 86, + 113, + 100, + 122 + ], + "score": 1.0, + "content": "150", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 110, + 506, + 123 + ], + "score": 1.0, + "content": "modified as above, so we derive makespans which is also irrelevant to the earlier operations. Given", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 117, + 510, + 141 + ], + "spans": [ + { + "bbox": [ + 86, + 126, + 100, + 136 + ], + "score": 1.0, + "content": "151", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 100, + 117, + 188, + 141 + ], + "score": 1.0, + "content": "a scheduling policy", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 188, + 127, + 195, + 135 + ], + "score": 0.46, + "content": "\\pi", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 195, + 117, + 199, + 141 + ], + "score": 1.0, + "content": ",", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 199, + 122, + 248, + 136 + ], + "score": 0.92, + "content": "T _ { \\pi } ^ { ( m k s p ) } ( S )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 248, + 117, + 510, + 141 + ], + "score": 1.0, + "content": "is defined to be the makespan derived from an episode starting", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 133, + 460, + 153 + ], + "spans": [ + { + "bbox": [ + 86, + 140, + 100, + 150 + ], + "score": 1.0, + "content": "152", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 102, + 133, + 153, + 153 + ], + "score": 1.0, + "content": "from states", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 153, + 138, + 161, + 148 + ], + "score": 0.79, + "content": "S", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 162, + 133, + 215, + 153 + ], + "score": 1.0, + "content": "by following", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 216, + 140, + 223, + 149 + ], + "score": 0.73, + "content": "\\pi", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 223, + 133, + 244, + 153 + ], + "score": 1.0, + "content": ", and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 244, + 136, + 302, + 150 + ], + "score": 0.94, + "content": "T _ { \\pi } ^ { ( m k s p ) } ( S , a )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 302, + 133, + 426, + 153 + ], + "score": 1.0, + "content": "the makespan by taking action", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 426, + 140, + 433, + 148 + ], + "score": 0.7, + "content": "a", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 433, + 133, + 447, + 153 + ], + "score": 1.0, + "content": "on", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 447, + 138, + 455, + 148 + ], + "score": 0.75, + "content": "S", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 455, + 133, + 460, + 153 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + } + ], + "index": 26.5, + "bbox_fs": [ + 86, + 585, + 507, + 725 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 92, + 73, + 505, + 95 + ], + "lines": [ + { + "bbox": [ + 87, + 71, + 506, + 87 + ], + "spans": [ + { + "bbox": [ + 87, + 71, + 123, + 87 + ], + "score": 1.0, + "content": "147 and", + "type": "text" + }, + { + "bbox": [ + 124, + 73, + 143, + 84 + ], + "score": 0.9, + "content": "O _ { 2 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 71, + 303, + 87 + ], + "score": 1.0, + "content": "are removed and the ongoing operation", + "type": "text" + }, + { + "bbox": [ + 304, + 73, + 323, + 85 + ], + "score": 0.91, + "content": "O _ { 1 , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 71, + 506, + 87 + ], + "score": 1.0, + "content": "changes its processing time to the remaining", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 87, + 83, + 152, + 96 + ], + "spans": [ + { + "bbox": [ + 87, + 83, + 152, + 96 + ], + "score": 1.0, + "content": "148 time (5-3).", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 87, + 100, + 505, + 150 + ], + "lines": [ + { + "bbox": [ + 86, + 100, + 505, + 112 + ], + "spans": [ + { + "bbox": [ + 86, + 101, + 100, + 111 + ], + "score": 1.0, + "content": "149", + "type": "text" + }, + { + "bbox": [ + 105, + 100, + 341, + 112 + ], + "score": 1.0, + "content": "For residual scheduling, we also reset the dispatching time", + "type": "text" + }, + { + "bbox": [ + 342, + 101, + 366, + 110 + ], + "score": 0.9, + "content": "\\tau = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 100, + 505, + 112 + ], + "score": 1.0, + "content": "for all states with partial solutions", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 86, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 86, + 113, + 100, + 122 + ], + "score": 1.0, + "content": "150", + "type": "text" + }, + { + "bbox": [ + 105, + 110, + 506, + 123 + ], + "score": 1.0, + "content": "modified as above, so we derive makespans which is also irrelevant to the earlier operations. Given", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 86, + 117, + 510, + 141 + ], + "spans": [ + { + "bbox": [ + 86, + 126, + 100, + 136 + ], + "score": 1.0, + "content": "151", + "type": "text" + }, + { + "bbox": [ + 100, + 117, + 188, + 141 + ], + "score": 1.0, + "content": "a scheduling policy", + "type": "text" + }, + { + "bbox": [ + 188, + 127, + 195, + 135 + ], + "score": 0.46, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 117, + 199, + 141 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 199, + 122, + 248, + 136 + ], + "score": 0.92, + "content": "T _ { \\pi } ^ { ( m k s p ) } ( S )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 117, + 510, + 141 + ], + "score": 1.0, + "content": "is defined to be the makespan derived from an episode starting", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 86, + 133, + 460, + 153 + ], + "spans": [ + { + "bbox": [ + 86, + 140, + 100, + 150 + ], + "score": 1.0, + "content": "152", + "type": "text" + }, + { + "bbox": [ + 102, + 133, + 153, + 153 + ], + "score": 1.0, + "content": "from states", + "type": "text" + }, + { + "bbox": [ + 153, + 138, + 161, + 148 + ], + "score": 0.79, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 133, + 215, + 153 + ], + "score": 1.0, + "content": "by following", + "type": "text" + }, + { + "bbox": [ + 216, + 140, + 223, + 149 + ], + "score": 0.73, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 133, + 244, + 153 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 244, + 136, + 302, + 150 + ], + "score": 0.94, + "content": "T _ { \\pi } ^ { ( m k s p ) } ( S , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 133, + 426, + 153 + ], + "score": 1.0, + "content": "the makespan by taking action", + "type": "text" + }, + { + "bbox": [ + 426, + 140, + 433, + 148 + ], + "score": 0.7, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 133, + 447, + 153 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 447, + 138, + 455, + 148 + ], + "score": 0.75, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 133, + 460, + 153 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 91, + 171, + 265, + 183 + ], + "lines": [ + { + "bbox": [ + 87, + 169, + 266, + 186 + ], + "spans": [ + { + "bbox": [ + 87, + 169, + 266, + 186 + ], + "score": 1.0, + "content": "153 3.2 Residual Graph Representation", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 86, + 195, + 506, + 327 + ], + "lines": [ + { + "bbox": [ + 86, + 196, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 86, + 198, + 99, + 207 + ], + "score": 1.0, + "content": "154", + "type": "text" + }, + { + "bbox": [ + 106, + 196, + 506, + 208 + ], + "score": 1.0, + "content": "In this paper, our model design is based on graph neural network (GNN), and leverage GNN to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 86, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 86, + 209, + 99, + 218 + ], + "score": 1.0, + "content": "155", + "type": "text" + }, + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "extract the scheduling decision from the relationship in graph. In this subsection, we present the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 86, + 217, + 506, + 230 + ], + "spans": [ + { + "bbox": [ + 86, + 219, + 100, + 230 + ], + "score": 1.0, + "content": "156", + "type": "text" + }, + { + "bbox": [ + 105, + 217, + 506, + 230 + ], + "score": 1.0, + "content": "graph representation. Like many other researchers such as Park et al. (2021a), we formulate a partial", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 86, + 228, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 86, + 230, + 100, + 240 + ], + "score": 1.0, + "content": "157", + "type": "text" + }, + { + "bbox": [ + 105, + 228, + 190, + 241 + ], + "score": 1.0, + "content": "solution into a graph", + "type": "text" + }, + { + "bbox": [ + 190, + 228, + 236, + 240 + ], + "score": 0.93, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 228, + 266, + 241 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 266, + 229, + 275, + 238 + ], + "score": 0.82, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 228, + 357, + 241 + ], + "score": 1.0, + "content": "is a set of nodes and", + "type": "text" + }, + { + "bbox": [ + 357, + 229, + 365, + 238 + ], + "score": 0.78, + "content": "\\mathcal { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 228, + 506, + 241 + ], + "score": 1.0, + "content": "is a set of edges. A node is either a", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 86, + 239, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 86, + 241, + 99, + 251 + ], + "score": 1.0, + "content": "158", + "type": "text" + }, + { + "bbox": [ + 105, + 239, + 164, + 252 + ], + "score": 1.0, + "content": "machine node", + "type": "text" + }, + { + "bbox": [ + 165, + 240, + 177, + 249 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 239, + 262, + 252 + ], + "score": 1.0, + "content": "or an operation node", + "type": "text" + }, + { + "bbox": [ + 262, + 240, + 271, + 249 + ], + "score": 0.79, + "content": "O", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 239, + 506, + 252 + ], + "score": 1.0, + "content": ". An edge connects two nodes to represent the relationship", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 86, + 250, + 506, + 263 + ], + "spans": [ + { + "bbox": [ + 86, + 252, + 100, + 262 + ], + "score": 1.0, + "content": "159", + "type": "text" + }, + { + "bbox": [ + 105, + 250, + 466, + 263 + ], + "score": 1.0, + "content": "between two nodes, basically including three kinds of edges, namely operation-to-operation", + "type": "text" + }, + { + "bbox": [ + 466, + 250, + 500, + 261 + ], + "score": 0.87, + "content": "( O \\to O", + "type": "inline_equation" + }, + { + "bbox": [ + 500, + 250, + 506, + 263 + ], + "score": 1.0, + "content": "),", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 86, + 260, + 506, + 274 + ], + "spans": [ + { + "bbox": [ + 86, + 263, + 100, + 273 + ], + "score": 1.0, + "content": "160", + "type": "text" + }, + { + "bbox": [ + 105, + 260, + 199, + 274 + ], + "score": 1.0, + "content": "machine-to-operation", + "type": "text" + }, + { + "bbox": [ + 200, + 261, + 237, + 271 + ], + "score": 0.87, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 260, + 351, + 274 + ], + "score": 1.0, + "content": ") and operation-to-machine", + "type": "text" + }, + { + "bbox": [ + 351, + 261, + 388, + 272 + ], + "score": 0.88, + "content": "( O \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 260, + 506, + 274 + ], + "score": 1.0, + "content": "). All operations in the same", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 86, + 272, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 86, + 274, + 99, + 283 + ], + "score": 1.0, + "content": "161", + "type": "text" + }, + { + "bbox": [ + 105, + 272, + 214, + 284 + ], + "score": 1.0, + "content": "job are fully connected as", + "type": "text" + }, + { + "bbox": [ + 214, + 272, + 248, + 282 + ], + "score": 0.9, + "content": "O O", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 272, + 340, + 284 + ], + "score": 1.0, + "content": "edges. If an operation", + "type": "text" + }, + { + "bbox": [ + 341, + 272, + 350, + 282 + ], + "score": 0.83, + "content": "O", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 272, + 506, + 284 + ], + "score": 1.0, + "content": "is able to be performed on a machine", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 86, + 283, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 86, + 284, + 100, + 295 + ], + "score": 1.0, + "content": "162", + "type": "text" + }, + { + "bbox": [ + 107, + 283, + 118, + 293 + ], + "score": 0.7, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 283, + 191, + 295 + ], + "score": 1.0, + "content": ", there exists both", + "type": "text" + }, + { + "bbox": [ + 192, + 283, + 227, + 293 + ], + "score": 0.91, + "content": "O \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 283, + 245, + 295 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 246, + 283, + 281, + 293 + ], + "score": 0.91, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 283, + 505, + 295 + ], + "score": 1.0, + "content": "directed edges. In (Park et al., 2021a), they also let all", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 86, + 293, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 86, + 295, + 100, + 306 + ], + "score": 1.0, + "content": "163", + "type": "text" + }, + { + "bbox": [ + 105, + 293, + 236, + 307 + ], + "score": 1.0, + "content": "machines be fully connected as", + "type": "text" + }, + { + "bbox": [ + 236, + 294, + 275, + 304 + ], + "score": 0.9, + "content": "M \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 293, + 506, + 307 + ], + "score": 1.0, + "content": "edges. However, our experiments in section 4 show that", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 86, + 304, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 86, + 306, + 100, + 317 + ], + "score": 1.0, + "content": "164", + "type": "text" + }, + { + "bbox": [ + 105, + 304, + 136, + 318 + ], + "score": 1.0, + "content": "mutual", + "type": "text" + }, + { + "bbox": [ + 136, + 305, + 174, + 315 + ], + "score": 0.91, + "content": "M \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 304, + 506, + 318 + ], + "score": 1.0, + "content": "edges do not help much based on our Residual Scheduling. An illustration for graph", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 86, + 316, + 297, + 328 + ], + "spans": [ + { + "bbox": [ + 86, + 317, + 100, + 328 + ], + "score": 1.0, + "content": "165", + "type": "text" + }, + { + "bbox": [ + 106, + 316, + 176, + 328 + ], + "score": 1.0, + "content": "representation of", + "type": "text" + }, + { + "bbox": [ + 176, + 316, + 188, + 327 + ], + "score": 0.89, + "content": "S _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 316, + 297, + 328 + ], + "score": 1.0, + "content": "is depicted in Figure 3 (a).", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 12.5 + }, + { + "type": "image", + "bbox": [ + 107, + 345, + 502, + 441 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 345, + 502, + 441 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 345, + 502, + 441 + ], + "spans": [ + { + "bbox": [ + 107, + 345, + 502, + 441 + ], + "score": 0.964, + "type": "image", + "image_path": "a3a8af5ab1851b807add19a4a2a54f81ec9c7979dcede6e1a0cd5113aacd0d36.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 107, + 345, + 502, + 377.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 107, + 377.0, + 502, + 409.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 107, + 409.0, + 502, + 441.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 213, + 454, + 397, + 466 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 213, + 453, + 397, + 468 + ], + "spans": [ + { + "bbox": [ + 213, + 453, + 397, + 468 + ], + "score": 1.0, + "content": "Figure 3: Graph representation and networks.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + } + ], + "index": 21.0 + }, + { + "type": "text", + "bbox": [ + 87, + 485, + 506, + 529 + ], + "lines": [ + { + "bbox": [ + 86, + 485, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 86, + 487, + 99, + 497 + ], + "score": 1.0, + "content": "166", + "type": "text" + }, + { + "bbox": [ + 105, + 485, + 486, + 498 + ], + "score": 1.0, + "content": "In the graph representation, all nodes need to include some attributes so that a partial solution", + "type": "text" + }, + { + "bbox": [ + 486, + 486, + 494, + 495 + ], + "score": 0.8, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 485, + 506, + 498 + ], + "score": 1.0, + "content": "at", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 86, + 496, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 86, + 498, + 99, + 507 + ], + "score": 1.0, + "content": "167", + "type": "text" + }, + { + "bbox": [ + 105, + 496, + 187, + 508 + ], + "score": 1.0, + "content": "the dispatching time", + "type": "text" + }, + { + "bbox": [ + 188, + 498, + 195, + 506 + ], + "score": 0.73, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 496, + 505, + 508 + ], + "score": 1.0, + "content": "can be supported in the MDP formulation (in the appendix). Note that many of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 86, + 504, + 507, + 522 + ], + "spans": [ + { + "bbox": [ + 86, + 508, + 100, + 519 + ], + "score": 1.0, + "content": "168", + "type": "text" + }, + { + "bbox": [ + 104, + 504, + 484, + 522 + ], + "score": 1.0, + "content": "the attributes below are normalized to reduce variance. For nodes corresponding to operations", + "type": "text" + }, + { + "bbox": [ + 485, + 507, + 502, + 519 + ], + "score": 0.9, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 504, + 507, + 522 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 86, + 518, + 239, + 531 + ], + "spans": [ + { + "bbox": [ + 86, + 520, + 99, + 529 + ], + "score": 1.0, + "content": "169", + "type": "text" + }, + { + "bbox": [ + 105, + 518, + 239, + 531 + ], + "score": 1.0, + "content": "we have the following attributes:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 86, + 534, + 505, + 633 + ], + "lines": [ + { + "bbox": [ + 86, + 534, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 86, + 536, + 100, + 546 + ], + "score": 1.0, + "content": "170", + "type": "text" + }, + { + "bbox": [ + 105, + 534, + 133, + 547 + ], + "score": 1.0, + "content": "Status", + "type": "text" + }, + { + "bbox": [ + 134, + 534, + 150, + 547 + ], + "score": 0.93, + "content": "\\phi _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 534, + 236, + 547 + ], + "score": 1.0, + "content": ": The operation status", + "type": "text" + }, + { + "bbox": [ + 237, + 535, + 253, + 546 + ], + "score": 0.91, + "content": "\\phi _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 534, + 451, + 547 + ], + "score": 1.0, + "content": "is completed if the operation has been finished by", + "type": "text" + }, + { + "bbox": [ + 451, + 537, + 458, + 544 + ], + "score": 0.67, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 534, + 506, + 547 + ], + "score": 1.0, + "content": ", ongoing if", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 86, + 545, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 86, + 547, + 99, + 557 + ], + "score": 1.0, + "content": "171", + "type": "text" + }, + { + "bbox": [ + 105, + 545, + 388, + 558 + ], + "score": 1.0, + "content": "the operation is ongoing (i.e., has been dispatched to some machine by", + "type": "text" + }, + { + "bbox": [ + 388, + 547, + 395, + 555 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 545, + 506, + 558 + ], + "score": 1.0, + "content": "and is still being processed", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 86, + 557, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 86, + 558, + 100, + 568 + ], + "score": 1.0, + "content": "172", + "type": "text" + }, + { + "bbox": [ + 105, + 557, + 116, + 568 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 116, + 558, + 123, + 567 + ], + "score": 0.6, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 557, + 505, + 568 + ], + "score": 1.0, + "content": "), ready if the operation designated to the machine which is idle has not been dispatched yet and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 86, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 86, + 568, + 100, + 579 + ], + "score": 1.0, + "content": "173", + "type": "text" + }, + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "its precedent operation has been finished, and unready otherwise. For example, in Figure 3 (a), the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 86, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 86, + 580, + 100, + 589 + ], + "score": 1.0, + "content": "174", + "type": "text" + }, + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "gray nodes are completed, the red ongoing, the yellow ready and the white unready. In our residual", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 86, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 86, + 591, + 100, + 601 + ], + "score": 1.0, + "content": "175", + "type": "text" + }, + { + "bbox": [ + 105, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "scheduling, there exists no completed operations in all partial solutions, since they are removes for", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 86, + 601, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 86, + 602, + 100, + 611 + ], + "score": 1.0, + "content": "176", + "type": "text" + }, + { + "bbox": [ + 106, + 601, + 505, + 611 + ], + "score": 1.0, + "content": "irrelevance of the rest of scheduling. The attribute is a one-hot vector to represent the current status", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 86, + 610, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 86, + 612, + 100, + 622 + ], + "score": 1.0, + "content": "177", + "type": "text" + }, + { + "bbox": [ + 105, + 610, + 454, + 623 + ], + "score": 1.0, + "content": "of the operation, which is one of ongoing, ready and unready. Illustration for all states", + "type": "text" + }, + { + "bbox": [ + 454, + 611, + 466, + 622 + ], + "score": 0.88, + "content": "S _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 610, + 477, + 623 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 477, + 611, + 489, + 622 + ], + "score": 0.89, + "content": "S _ { 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 610, + 506, + 623 + ], + "score": 1.0, + "content": "are", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 86, + 622, + 201, + 634 + ], + "spans": [ + { + "bbox": [ + 86, + 624, + 100, + 633 + ], + "score": 1.0, + "content": "178", + "type": "text" + }, + { + "bbox": [ + 105, + 622, + 201, + 634 + ], + "score": 1.0, + "content": "shown in the appendix.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 87, + 638, + 506, + 723 + ], + "lines": [ + { + "bbox": [ + 86, + 631, + 503, + 660 + ], + "spans": [ + { + "bbox": [ + 86, + 642, + 100, + 653 + ], + "score": 1.0, + "content": "179", + "type": "text" + }, + { + "bbox": [ + 104, + 631, + 225, + 660 + ], + "score": 1.0, + "content": "Normalized processing time", + "type": "text" + }, + { + "bbox": [ + 226, + 638, + 248, + 654 + ], + "score": 0.92, + "content": "\\bar { T } _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 631, + 403, + 660 + ], + "score": 1.0, + "content": ": Let the maximal processing time be", + "type": "text" + }, + { + "bbox": [ + 403, + 638, + 503, + 654 + ], + "score": 0.93, + "content": "T _ { m a x } ^ { ( o p ) } = \\operatorname* { m a x } _ { \\forall j , i } ( T _ { j , i } ^ { ( o p ) } )", + "type": "inline_equation" + } + ], + "index": 36 + }, + { + "bbox": [ + 86, + 651, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 86, + 658, + 100, + 668 + ], + "score": 1.0, + "content": "180", + "type": "text" + }, + { + "bbox": [ + 103, + 651, + 160, + 670 + ], + "score": 1.0, + "content": "Then, T¯(op)j,i", + "type": "text" + }, + { + "bbox": [ + 132, + 653, + 218, + 670 + ], + "score": 0.93, + "content": "\\bar { T } _ { j , i } ^ { ( o p ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 653, + 506, + 670 + ], + "score": 1.0, + "content": ". In our residual scheduling, the operations that have been finished are", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 86, + 668, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 99, + 678 + ], + "score": 1.0, + "content": "181", + "type": "text" + }, + { + "bbox": [ + 105, + 668, + 506, + 679 + ], + "score": 1.0, + "content": "removed in partial solutions and therefore their processing time can be ignored; the operations that", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 85, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 85, + 679, + 100, + 690 + ], + "score": 1.0, + "content": "182", + "type": "text" + }, + { + "bbox": [ + 104, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "has not been dispatched yet still keep their processing times the same; the operations that are ongoing", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 85, + 689, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 85, + 690, + 100, + 701 + ], + "score": 1.0, + "content": "183", + "type": "text" + }, + { + "bbox": [ + 106, + 689, + 424, + 701 + ], + "score": 1.0, + "content": "change their processing times to the remaining times after the dispatching time", + "type": "text" + }, + { + "bbox": [ + 424, + 690, + 433, + 700 + ], + "score": 0.83, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 689, + 505, + 701 + ], + "score": 1.0, + "content": ". As for FJSP, the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 86, + 700, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 100, + 712 + ], + "score": 1.0, + "content": "184", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 507, + 712 + ], + "score": 1.0, + "content": "operations that has not been dispatched yet may have several processing times on different machines,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 86, + 711, + 386, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 100, + 722 + ], + "score": 1.0, + "content": "185", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 386, + 724 + ], + "score": 1.0, + "content": "and thus we can simply choose the average of these processing times.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 92, + 73, + 505, + 95 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 87, + 71, + 506, + 96 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 87, + 100, + 505, + 150 + ], + "lines": [], + "index": 3.5, + "bbox_fs": [ + 86, + 100, + 510, + 153 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 91, + 171, + 265, + 183 + ], + "lines": [ + { + "bbox": [ + 87, + 169, + 266, + 186 + ], + "spans": [ + { + "bbox": [ + 87, + 169, + 266, + 186 + ], + "score": 1.0, + "content": "153 3.2 Residual Graph Representation", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "index", + "bbox": [ + 86, + 195, + 506, + 327 + ], + "lines": [ + { + "bbox": [ + 86, + 196, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 86, + 198, + 99, + 207 + ], + "score": 1.0, + "content": "154", + "type": "text" + }, + { + "bbox": [ + 106, + 196, + 506, + 208 + ], + "score": 1.0, + "content": "In this paper, our model design is based on graph neural network (GNN), and leverage GNN to", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 86, + 209, + 99, + 218 + ], + "score": 1.0, + "content": "155", + "type": "text" + }, + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "extract the scheduling decision from the relationship in graph. In this subsection, we present the", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 217, + 506, + 230 + ], + "spans": [ + { + "bbox": [ + 86, + 219, + 100, + 230 + ], + "score": 1.0, + "content": "156", + "type": "text" + }, + { + "bbox": [ + 105, + 217, + 506, + 230 + ], + "score": 1.0, + "content": "graph representation. Like many other researchers such as Park et al. (2021a), we formulate a partial", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 228, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 86, + 230, + 100, + 240 + ], + "score": 1.0, + "content": "157", + "type": "text" + }, + { + "bbox": [ + 105, + 228, + 190, + 241 + ], + "score": 1.0, + "content": "solution into a graph", + "type": "text" + }, + { + "bbox": [ + 190, + 228, + 236, + 240 + ], + "score": 0.93, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 228, + 266, + 241 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 266, + 229, + 275, + 238 + ], + "score": 0.82, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 228, + 357, + 241 + ], + "score": 1.0, + "content": "is a set of nodes and", + "type": "text" + }, + { + "bbox": [ + 357, + 229, + 365, + 238 + ], + "score": 0.78, + "content": "\\mathcal { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 228, + 506, + 241 + ], + "score": 1.0, + "content": "is a set of edges. A node is either a", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 239, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 86, + 241, + 99, + 251 + ], + "score": 1.0, + "content": "158", + "type": "text" + }, + { + "bbox": [ + 105, + 239, + 164, + 252 + ], + "score": 1.0, + "content": "machine node", + "type": "text" + }, + { + "bbox": [ + 165, + 240, + 177, + 249 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 239, + 262, + 252 + ], + "score": 1.0, + "content": "or an operation node", + "type": "text" + }, + { + "bbox": [ + 262, + 240, + 271, + 249 + ], + "score": 0.79, + "content": "O", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 239, + 506, + 252 + ], + "score": 1.0, + "content": ". An edge connects two nodes to represent the relationship", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 250, + 506, + 263 + ], + "spans": [ + { + "bbox": [ + 86, + 252, + 100, + 262 + ], + "score": 1.0, + "content": "159", + "type": "text" + }, + { + "bbox": [ + 105, + 250, + 466, + 263 + ], + "score": 1.0, + "content": "between two nodes, basically including three kinds of edges, namely operation-to-operation", + "type": "text" + }, + { + "bbox": [ + 466, + 250, + 500, + 261 + ], + "score": 0.87, + "content": "( O \\to O", + "type": "inline_equation" + }, + { + "bbox": [ + 500, + 250, + 506, + 263 + ], + "score": 1.0, + "content": "),", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 260, + 506, + 274 + ], + "spans": [ + { + "bbox": [ + 86, + 263, + 100, + 273 + ], + "score": 1.0, + "content": "160", + "type": "text" + }, + { + "bbox": [ + 105, + 260, + 199, + 274 + ], + "score": 1.0, + "content": "machine-to-operation", + "type": "text" + }, + { + "bbox": [ + 200, + 261, + 237, + 271 + ], + "score": 0.87, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 260, + 351, + 274 + ], + "score": 1.0, + "content": ") and operation-to-machine", + "type": "text" + }, + { + "bbox": [ + 351, + 261, + 388, + 272 + ], + "score": 0.88, + "content": "( O \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 260, + 506, + 274 + ], + "score": 1.0, + "content": "). All operations in the same", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 272, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 86, + 274, + 99, + 283 + ], + "score": 1.0, + "content": "161", + "type": "text" + }, + { + "bbox": [ + 105, + 272, + 214, + 284 + ], + "score": 1.0, + "content": "job are fully connected as", + "type": "text" + }, + { + "bbox": [ + 214, + 272, + 248, + 282 + ], + "score": 0.9, + "content": "O O", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 272, + 340, + 284 + ], + "score": 1.0, + "content": "edges. If an operation", + "type": "text" + }, + { + "bbox": [ + 341, + 272, + 350, + 282 + ], + "score": 0.83, + "content": "O", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 272, + 506, + 284 + ], + "score": 1.0, + "content": "is able to be performed on a machine", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 283, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 86, + 284, + 100, + 295 + ], + "score": 1.0, + "content": "162", + "type": "text" + }, + { + "bbox": [ + 107, + 283, + 118, + 293 + ], + "score": 0.7, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 283, + 191, + 295 + ], + "score": 1.0, + "content": ", there exists both", + "type": "text" + }, + { + "bbox": [ + 192, + 283, + 227, + 293 + ], + "score": 0.91, + "content": "O \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 283, + 245, + 295 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 246, + 283, + 281, + 293 + ], + "score": 0.91, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 283, + 505, + 295 + ], + "score": 1.0, + "content": "directed edges. In (Park et al., 2021a), they also let all", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 293, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 86, + 295, + 100, + 306 + ], + "score": 1.0, + "content": "163", + "type": "text" + }, + { + "bbox": [ + 105, + 293, + 236, + 307 + ], + "score": 1.0, + "content": "machines be fully connected as", + "type": "text" + }, + { + "bbox": [ + 236, + 294, + 275, + 304 + ], + "score": 0.9, + "content": "M \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 293, + 506, + 307 + ], + "score": 1.0, + "content": "edges. However, our experiments in section 4 show that", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 304, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 86, + 306, + 100, + 317 + ], + "score": 1.0, + "content": "164", + "type": "text" + }, + { + "bbox": [ + 105, + 304, + 136, + 318 + ], + "score": 1.0, + "content": "mutual", + "type": "text" + }, + { + "bbox": [ + 136, + 305, + 174, + 315 + ], + "score": 0.91, + "content": "M \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 304, + 506, + 318 + ], + "score": 1.0, + "content": "edges do not help much based on our Residual Scheduling. An illustration for graph", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 316, + 297, + 328 + ], + "spans": [ + { + "bbox": [ + 86, + 317, + 100, + 328 + ], + "score": 1.0, + "content": "165", + "type": "text" + }, + { + "bbox": [ + 106, + 316, + 176, + 328 + ], + "score": 1.0, + "content": "representation of", + "type": "text" + }, + { + "bbox": [ + 176, + 316, + 188, + 327 + ], + "score": 0.89, + "content": "S _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 316, + 297, + 328 + ], + "score": 1.0, + "content": "is depicted in Figure 3 (a).", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 485, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 86, + 487, + 99, + 497 + ], + "score": 1.0, + "content": "166", + "type": "text" + }, + { + "bbox": [ + 105, + 485, + 486, + 498 + ], + "score": 1.0, + "content": "In the graph representation, all nodes need to include some attributes so that a partial solution", + "type": "text" + }, + { + "bbox": [ + 486, + 486, + 494, + 495 + ], + "score": 0.8, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 485, + 506, + 498 + ], + "score": 1.0, + "content": "at", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 496, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 86, + 498, + 99, + 507 + ], + "score": 1.0, + "content": "167", + "type": "text" + }, + { + "bbox": [ + 105, + 496, + 187, + 508 + ], + "score": 1.0, + "content": "the dispatching time", + "type": "text" + }, + { + "bbox": [ + 188, + 498, + 195, + 506 + ], + "score": 0.73, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 496, + 505, + 508 + ], + "score": 1.0, + "content": "can be supported in the MDP formulation (in the appendix). Note that many of", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 504, + 507, + 522 + ], + "spans": [ + { + "bbox": [ + 86, + 508, + 100, + 519 + ], + "score": 1.0, + "content": "168", + "type": "text" + }, + { + "bbox": [ + 104, + 504, + 484, + 522 + ], + "score": 1.0, + "content": "the attributes below are normalized to reduce variance. For nodes corresponding to operations", + "type": "text" + }, + { + "bbox": [ + 485, + 507, + 502, + 519 + ], + "score": 0.9, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 504, + 507, + 522 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 518, + 239, + 531 + ], + "spans": [ + { + "bbox": [ + 86, + 520, + 99, + 529 + ], + "score": 1.0, + "content": "169", + "type": "text" + }, + { + "bbox": [ + 105, + 518, + 239, + 531 + ], + "score": 1.0, + "content": "we have the following attributes:", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 534, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 86, + 536, + 100, + 546 + ], + "score": 1.0, + "content": "170", + "type": "text" + }, + { + "bbox": [ + 105, + 534, + 133, + 547 + ], + "score": 1.0, + "content": "Status", + "type": "text" + }, + { + "bbox": [ + 134, + 534, + 150, + 547 + ], + "score": 0.93, + "content": "\\phi _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 534, + 236, + 547 + ], + "score": 1.0, + "content": ": The operation status", + "type": "text" + }, + { + "bbox": [ + 237, + 535, + 253, + 546 + ], + "score": 0.91, + "content": "\\phi _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 534, + 451, + 547 + ], + "score": 1.0, + "content": "is completed if the operation has been finished by", + "type": "text" + }, + { + "bbox": [ + 451, + 537, + 458, + 544 + ], + "score": 0.67, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 534, + 506, + 547 + ], + "score": 1.0, + "content": ", ongoing if", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 545, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 86, + 547, + 99, + 557 + ], + "score": 1.0, + "content": "171", + "type": "text" + }, + { + "bbox": [ + 105, + 545, + 388, + 558 + ], + "score": 1.0, + "content": "the operation is ongoing (i.e., has been dispatched to some machine by", + "type": "text" + }, + { + "bbox": [ + 388, + 547, + 395, + 555 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 545, + 506, + 558 + ], + "score": 1.0, + "content": "and is still being processed", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 557, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 86, + 558, + 100, + 568 + ], + "score": 1.0, + "content": "172", + "type": "text" + }, + { + "bbox": [ + 105, + 557, + 116, + 568 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 116, + 558, + 123, + 567 + ], + "score": 0.6, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 557, + 505, + 568 + ], + "score": 1.0, + "content": "), ready if the operation designated to the machine which is idle has not been dispatched yet and", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 86, + 568, + 100, + 579 + ], + "score": 1.0, + "content": "173", + "type": "text" + }, + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "its precedent operation has been finished, and unready otherwise. For example, in Figure 3 (a), the", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 86, + 580, + 100, + 589 + ], + "score": 1.0, + "content": "174", + "type": "text" + }, + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "gray nodes are completed, the red ongoing, the yellow ready and the white unready. In our residual", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 86, + 591, + 100, + 601 + ], + "score": 1.0, + "content": "175", + "type": "text" + }, + { + "bbox": [ + 105, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "scheduling, there exists no completed operations in all partial solutions, since they are removes for", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 601, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 86, + 602, + 100, + 611 + ], + "score": 1.0, + "content": "176", + "type": "text" + }, + { + "bbox": [ + 106, + 601, + 505, + 611 + ], + "score": 1.0, + "content": "irrelevance of the rest of scheduling. The attribute is a one-hot vector to represent the current status", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 610, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 86, + 612, + 100, + 622 + ], + "score": 1.0, + "content": "177", + "type": "text" + }, + { + "bbox": [ + 105, + 610, + 454, + 623 + ], + "score": 1.0, + "content": "of the operation, which is one of ongoing, ready and unready. Illustration for all states", + "type": "text" + }, + { + "bbox": [ + 454, + 611, + 466, + 622 + ], + "score": 0.88, + "content": "S _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 610, + 477, + 623 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 477, + 611, + 489, + 622 + ], + "score": 0.89, + "content": "S _ { 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 610, + 506, + 623 + ], + "score": 1.0, + "content": "are", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 622, + 201, + 634 + ], + "spans": [ + { + "bbox": [ + 86, + 624, + 100, + 633 + ], + "score": 1.0, + "content": "178", + "type": "text" + }, + { + "bbox": [ + 105, + 622, + 201, + 634 + ], + "score": 1.0, + "content": "shown in the appendix.", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 631, + 503, + 660 + ], + "spans": [ + { + "bbox": [ + 86, + 642, + 100, + 653 + ], + "score": 1.0, + "content": "179", + "type": "text" + }, + { + "bbox": [ + 104, + 631, + 225, + 660 + ], + "score": 1.0, + "content": "Normalized processing time", + "type": "text" + }, + { + "bbox": [ + 226, + 638, + 248, + 654 + ], + "score": 0.92, + "content": "\\bar { T } _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 631, + 403, + 660 + ], + "score": 1.0, + "content": ": Let the maximal processing time be", + "type": "text" + }, + { + "bbox": [ + 403, + 638, + 503, + 654 + ], + "score": 0.93, + "content": "T _ { m a x } ^ { ( o p ) } = \\operatorname* { m a x } _ { \\forall j , i } ( T _ { j , i } ^ { ( o p ) } )", + "type": "inline_equation" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 651, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 86, + 658, + 100, + 668 + ], + "score": 1.0, + "content": "180", + "type": "text" + }, + { + "bbox": [ + 103, + 651, + 160, + 670 + ], + "score": 1.0, + "content": "Then, T¯(op)j,i", + "type": "text" + }, + { + "bbox": [ + 132, + 653, + 218, + 670 + ], + "score": 0.93, + "content": "\\bar { T } _ { j , i } ^ { ( o p ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 653, + 506, + 670 + ], + "score": 1.0, + "content": ". In our residual scheduling, the operations that have been finished are", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 668, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 99, + 678 + ], + "score": 1.0, + "content": "181", + "type": "text" + }, + { + "bbox": [ + 105, + 668, + 506, + 679 + ], + "score": 1.0, + "content": "removed in partial solutions and therefore their processing time can be ignored; the operations that", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 85, + 679, + 100, + 690 + ], + "score": 1.0, + "content": "182", + "type": "text" + }, + { + "bbox": [ + 104, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "has not been dispatched yet still keep their processing times the same; the operations that are ongoing", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 689, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 85, + 690, + 100, + 701 + ], + "score": 1.0, + "content": "183", + "type": "text" + }, + { + "bbox": [ + 106, + 689, + 424, + 701 + ], + "score": 1.0, + "content": "change their processing times to the remaining times after the dispatching time", + "type": "text" + }, + { + "bbox": [ + 424, + 690, + 433, + 700 + ], + "score": 0.83, + "content": "\\tau _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 689, + 505, + 701 + ], + "score": 1.0, + "content": ". As for FJSP, the", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 700, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 100, + 712 + ], + "score": 1.0, + "content": "184", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 507, + 712 + ], + "score": 1.0, + "content": "operations that has not been dispatched yet may have several processing times on different machines,", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 711, + 386, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 100, + 722 + ], + "score": 1.0, + "content": "185", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 386, + 724 + ], + "score": 1.0, + "content": "and thus we can simply choose the average of these processing times.", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + } + ], + "index": 12.5, + "bbox_fs": [ + 86, + 196, + 506, + 328 + ] + }, + { + "type": "image", + "bbox": [ + 107, + 345, + 502, + 441 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 345, + 502, + 441 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 345, + 502, + 441 + ], + "spans": [ + { + "bbox": [ + 107, + 345, + 502, + 441 + ], + "score": 0.964, + "type": "image", + "image_path": "a3a8af5ab1851b807add19a4a2a54f81ec9c7979dcede6e1a0cd5113aacd0d36.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 107, + 345, + 502, + 377.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 107, + 377.0, + 502, + 409.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 107, + 409.0, + 502, + 441.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 213, + 454, + 397, + 466 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 213, + 453, + 397, + 468 + ], + "spans": [ + { + "bbox": [ + 213, + 453, + 397, + 468 + ], + "score": 1.0, + "content": "Figure 3: Graph representation and networks.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + } + ], + "index": 21.0 + }, + { + "type": "index", + "bbox": [ + 87, + 485, + 506, + 529 + ], + "lines": [], + "index": 24.5, + "bbox_fs": [ + 86, + 485, + 507, + 531 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 86, + 534, + 505, + 633 + ], + "lines": [], + "index": 31, + "bbox_fs": [ + 86, + 534, + 506, + 634 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 87, + 638, + 506, + 723 + ], + "lines": [], + "index": 39, + "bbox_fs": [ + 85, + 631, + 507, + 724 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 70, + 506, + 145 + ], + "lines": [ + { + "bbox": [ + 162, + 65, + 505, + 111 + ], + "spans": [ + { + "bbox": [ + 162, + 65, + 243, + 111 + ], + "score": 1.0, + "content": "ob remaining time , and let the process", + "type": "text" + }, + { + "bbox": [ + 243, + 70, + 269, + 87 + ], + "score": 0.93, + "content": "\\bar { T } _ { j , i } ^ { ( j o b ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 65, + 349, + 111 + ], + "score": 1.0, + "content": ": Let the rest of pre for the whole job", + "type": "text" + }, + { + "bbox": [ + 439, + 73, + 450, + 86 + ], + "score": 0.87, + "content": "J _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 70, + 505, + 87 + ], + "score": 0.91, + "content": "T _ { j , i } ^ { ( j o b ) } =", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 84, + 509, + 128 + ], + "spans": [ + { + "bbox": [ + 107, + 86, + 161, + 102 + ], + "score": 0.91, + "content": "\\sum _ { \\forall i ^ { \\prime } \\geq i } T _ { j , i ^ { \\prime } } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 85, + 201, + 128 + ], + "score": 1.0, + "content": "aced by th", + "type": "text" + }, + { + "bbox": [ + 219, + 85, + 349, + 128 + ], + "score": 1.0, + "content": "ing timocessing time for the original jo", + "type": "text" + }, + { + "bbox": [ + 349, + 89, + 356, + 101 + ], + "score": 0.76, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 86, + 454, + 102 + ], + "score": 0.93, + "content": "\\begin{array} { r } { T _ { j } ^ { ( j o b ) } = \\sum _ { \\forall i ^ { \\prime } } T _ { j , i ^ { \\prime } } ^ { ( o p ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 84, + 509, + 104 + ], + "score": 1.0, + "content": ". In practice,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 99, + 96, + 507, + 122 + ], + "spans": [ + { + "bbox": [ + 99, + 96, + 141, + 121 + ], + "score": 1.0, + "content": "T ( j ob )j is", + "type": "text" + }, + { + "bbox": [ + 353, + 105, + 360, + 116 + ], + "score": 0.77, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 102, + 484, + 118 + ], + "score": 0.91, + "content": "\\bar { T } _ { j , i } ^ { ( j o b ) } = T _ { j , i } ^ { ( j o b ) } / T _ { j } ^ { ( j o b ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 98, + 507, + 122 + ], + "score": 1.0, + "content": ". For", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 115, + 507, + 130 + ], + "spans": [ + { + "bbox": [ + 104, + 115, + 201, + 130 + ], + "score": 1.0, + "content": "FJSP, since operations", + "type": "text" + }, + { + "bbox": [ + 201, + 117, + 219, + 129 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 115, + 434, + 130 + ], + "score": 1.0, + "content": "can be dispatched to different designated machines", + "type": "text" + }, + { + "bbox": [ + 434, + 118, + 447, + 127 + ], + "score": 0.86, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 115, + 507, + 130 + ], + "score": 1.0, + "content": ", say with the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 420, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 171, + 142 + ], + "score": 1.0, + "content": "processing time", + "type": "text" + }, + { + "bbox": [ + 172, + 126, + 282, + 147 + ], + "score": 1.0, + "content": "T (op)j,i,l , we simply let T (op)j,i", + "type": "text" + }, + { + "bbox": [ + 278, + 127, + 349, + 146 + ], + "score": 1.0, + "content": "be the average of", + "type": "text" + }, + { + "bbox": [ + 350, + 128, + 372, + 145 + ], + "score": 0.93, + "content": "T _ { j , i , l } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 127, + 401, + 146 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 401, + 130, + 415, + 142 + ], + "score": 0.87, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 127, + 420, + 146 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 102, + 147, + 444, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 146, + 446, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 294, + 162 + ], + "score": 1.0, + "content": "For machine nodes corresponding to machines", + "type": "text" + }, + { + "bbox": [ + 295, + 148, + 308, + 159 + ], + "score": 0.89, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 146, + 446, + 162 + ], + "score": 1.0, + "content": ", we have the following attributes:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 505, + 198 + ], + "lines": [ + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 172, + 178 + ], + "score": 1.0, + "content": "Machine status", + "type": "text" + }, + { + "bbox": [ + 173, + 165, + 183, + 176 + ], + "score": 0.86, + "content": "\\phi _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 164, + 269, + 178 + ], + "score": 1.0, + "content": ": The machine status", + "type": "text" + }, + { + "bbox": [ + 270, + 165, + 279, + 176 + ], + "score": 0.87, + "content": "\\phi _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "is processing if some operation has been dispatched to", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 214, + 188 + ], + "score": 1.0, + "content": "and is being processed by", + "type": "text" + }, + { + "bbox": [ + 214, + 176, + 228, + 186 + ], + "score": 0.88, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 176, + 239, + 188 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 239, + 177, + 246, + 185 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 176, + 471, + 188 + ], + "score": 1.0, + "content": ", and idle otherwise (no operation is being processed at", + "type": "text" + }, + { + "bbox": [ + 472, + 178, + 478, + 185 + ], + "score": 0.7, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 176, + 505, + 188 + ], + "score": 1.0, + "content": "). The", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 186, + 486, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 486, + 199 + ], + "score": 1.0, + "content": "attribute is a one-hot vector to represent the current status, which is one of processing and idle.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 203, + 506, + 259 + ], + "lines": [ + { + "bbox": [ + 102, + 200, + 505, + 220 + ], + "spans": [ + { + "bbox": [ + 102, + 200, + 265, + 220 + ], + "score": 1.0, + "content": "Normalized operation processing time", + "type": "text" + }, + { + "bbox": [ + 266, + 203, + 295, + 218 + ], + "score": 0.92, + "content": "\\bar { T } _ { l } ^ { ( m a c ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 200, + 366, + 220 + ], + "score": 1.0, + "content": ": On the machine", + "type": "text" + }, + { + "bbox": [ + 366, + 205, + 380, + 217 + ], + "score": 0.89, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 200, + 464, + 220 + ], + "score": 1.0, + "content": ", the processing time", + "type": "text" + }, + { + "bbox": [ + 464, + 203, + 494, + 218 + ], + "score": 0.93, + "content": "T _ { l } ^ { ( m a c ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 200, + 505, + 220 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 214, + 507, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 129, + 234 + ], + "score": 0.91, + "content": "T _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 214, + 343, + 235 + ], + "score": 1.0, + "content": "(the same as the normalized processing time for node", + "type": "text" + }, + { + "bbox": [ + 344, + 220, + 362, + 232 + ], + "score": 0.89, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 214, + 507, + 235 + ], + "score": 1.0, + "content": ") if the machine status is processing,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 231, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 220, + 245 + ], + "score": 1.0, + "content": "i.e., some ongoing operation", + "type": "text" + }, + { + "bbox": [ + 221, + 232, + 238, + 244 + ], + "score": 0.93, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 231, + 506, + 245 + ], + "score": 1.0, + "content": "is being processed but not finished yet, is zero if the machine status", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 101, + 237, + 449, + 264 + ], + "spans": [ + { + "bbox": [ + 101, + 237, + 281, + 264 + ], + "score": 1.0, + "content": "is idle. Then, this attribute is normalized to", + "type": "text" + }, + { + "bbox": [ + 281, + 243, + 304, + 257 + ], + "score": 0.91, + "content": "T _ { m a x } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 237, + 342, + 264 + ], + "score": 1.0, + "content": "and thus", + "type": "text" + }, + { + "bbox": [ + 342, + 243, + 441, + 259 + ], + "score": 0.93, + "content": "\\bar { T } _ { l } ^ { ( m a c ) } = T _ { l } ^ { ( m a c ) } / T _ { m a x } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 237, + 449, + 264 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 262, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "score": 1.0, + "content": "Now, consider edges in a residual scheduling graph. As described above, there exists three relationship", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 273, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 167, + 286 + ], + "score": 1.0, + "content": "sets for edges,", + "type": "text" + }, + { + "bbox": [ + 167, + 273, + 200, + 284 + ], + "score": 0.88, + "content": "O O", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 273, + 205, + 286 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 205, + 273, + 241, + 284 + ], + "score": 0.88, + "content": "O \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 273, + 260, + 286 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 260, + 273, + 296, + 284 + ], + "score": 0.9, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 273, + 412, + 286 + ], + "score": 1.0, + "content": ". First, for the same job, say", + "type": "text" + }, + { + "bbox": [ + 413, + 274, + 424, + 286 + ], + "score": 0.88, + "content": "J _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 273, + 505, + 286 + ], + "score": 1.0, + "content": ", all of its operation", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 283, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 147, + 296 + ], + "score": 1.0, + "content": "nodes for", + "type": "text" + }, + { + "bbox": [ + 147, + 285, + 165, + 297 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 283, + 505, + 296 + ], + "score": 1.0, + "content": "are fully connected. Note that for residual scheduling the operations finished by the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 175, + 308 + ], + "score": 1.0, + "content": "dispatching time", + "type": "text" + }, + { + "bbox": [ + 175, + 298, + 182, + 305 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 295, + 480, + 308 + ], + "score": 1.0, + "content": "are removed and thus have no edges to them. Second, a machine node for", + "type": "text" + }, + { + "bbox": [ + 481, + 296, + 495, + 306 + ], + "score": 0.89, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 306, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 251, + 319 + ], + "score": 1.0, + "content": "connected to an operation node for", + "type": "text" + }, + { + "bbox": [ + 251, + 306, + 268, + 318 + ], + "score": 0.89, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 306, + 337, + 319 + ], + "score": 1.0, + "content": ", if the operation", + "type": "text" + }, + { + "bbox": [ + 338, + 306, + 355, + 318 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 306, + 505, + 319 + ], + "score": 1.0, + "content": "is designated to be processed on the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 317, + 500, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 142, + 329 + ], + "score": 1.0, + "content": "machine", + "type": "text" + }, + { + "bbox": [ + 143, + 317, + 156, + 328 + ], + "score": 0.87, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 317, + 255, + 329 + ], + "score": 1.0, + "content": ", which forms two edges", + "type": "text" + }, + { + "bbox": [ + 256, + 317, + 291, + 327 + ], + "score": 0.9, + "content": "\\dot { O } M", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 317, + 309, + 329 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 309, + 317, + 344, + 327 + ], + "score": 0.9, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 317, + 500, + 329 + ], + "score": 1.0, + "content": ". Both contains the following attribute.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 506, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 328, + 507, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 271, + 348 + ], + "score": 1.0, + "content": "Normalized operation processing tim", + "type": "text" + }, + { + "bbox": [ + 267, + 329, + 312, + 349 + ], + "score": 1.0, + "content": "e T¯(edge):", + "type": "text" + }, + { + "bbox": [ + 306, + 329, + 376, + 354 + ], + "score": 1.0, + "content": "The attribute is", + "type": "text" + }, + { + "bbox": [ + 376, + 333, + 474, + 350 + ], + "score": 0.93, + "content": "\\bar { T } _ { j , i , l } ^ { ( e d g e ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 328, + 482, + 352 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 480, + 335, + 507, + 348 + ], + "score": 1.0, + "content": "Here,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 348, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 107, + 348, + 165, + 366 + ], + "score": 0.93, + "content": "T _ { j , i } ^ { ( o p ) } = T _ { j , i , l } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 352, + 319, + 364 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 349, + 479, + 366 + ], + "score": 0.92, + "content": "T _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 351, + 506, + 364 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 364, + 249, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 249, + 376 + ], + "score": 1.0, + "content": "remaining time as described above.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 107, + 387, + 230, + 399 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 231, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 231, + 400 + ], + "score": 1.0, + "content": "3.3 Graph Neural Network", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 94, + 407, + 506, + 497 + ], + "lines": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "score": 1.0, + "content": "In this subsection, we present our model based on graph neural network (GNN). GNN are a family", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "score": 1.0, + "content": "of deep neural networks (Battaglia et al., 2018) that can learn representation of graph-structured", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "data, widely used in many applications (Lv et al., 2021; Zhou et al., 2020). A GNN aggregates", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 441, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 505, + 453 + ], + "score": 1.0, + "content": "information from node itself and its neighboring nodes and then update the data itself, which allows", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "the GNN to capture the complex relationships within the data graph. For GNN, we choose Graph", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 461, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 442, + 475 + ], + "score": 1.0, + "content": "Isomorphism Network (GIN), which was shown to have strong discriminative power (", + "type": "text" + }, + { + "bbox": [ + 442, + 462, + 456, + 472 + ], + "score": 0.28, + "content": "\\mathrm { { X u } }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 461, + 506, + 475 + ], + "score": 1.0, + "content": "et al., 2019)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 472, + 507, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 316, + 486 + ], + "score": 1.0, + "content": "and summarily reviewed as follows. Given a graph", + "type": "text" + }, + { + "bbox": [ + 316, + 473, + 363, + 485 + ], + "score": 0.93, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 472, + 381, + 486 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 382, + 473, + 392, + 483 + ], + "score": 0.8, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 472, + 448, + 486 + ], + "score": 1.0, + "content": "GNN layers", + "type": "text" + }, + { + "bbox": [ + 448, + 474, + 459, + 483 + ], + "score": 0.79, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 472, + 507, + 486 + ], + "score": 1.0, + "content": "iterations),", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 484, + 470, + 498 + ], + "spans": [ + { + "bbox": [ + 104, + 484, + 180, + 498 + ], + "score": 1.0, + "content": "GIN performs the", + "type": "text" + }, + { + "bbox": [ + 180, + 486, + 186, + 495 + ], + "score": 0.83, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 484, + 363, + 498 + ], + "score": 1.0, + "content": "-th iterations of updating feature embedding", + "type": "text" + }, + { + "bbox": [ + 363, + 484, + 381, + 496 + ], + "score": 0.9, + "content": "h ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 484, + 439, + 498 + ], + "score": 1.0, + "content": "for each node", + "type": "text" + }, + { + "bbox": [ + 440, + 486, + 465, + 496 + ], + "score": 0.9, + "content": "v \\in \\mathcal V", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 484, + 470, + 498 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 26.5 + }, + { + "type": "interline_equation", + "bbox": [ + 196, + 498, + 414, + 526 + ], + "lines": [ + { + "bbox": [ + 196, + 498, + 414, + 526 + ], + "spans": [ + { + "bbox": [ + 196, + 498, + 414, + 526 + ], + "score": 0.92, + "content": "h _ { v } ^ { ( k ) } = M L P ^ { ( k ) } ( ( 1 + \\epsilon ^ { ( k ) } ) h _ { v } ^ { ( k - 1 ) } + \\sum _ { u \\in N _ { b } ( v ) } h _ { u } ^ { ( k - 1 ) } ) ,", + "type": "interline_equation", + "image_path": "2ff8cf48f7fc89b4b2026b4e2f6ba5a18a536573b1a5ce94368e07ebd887b7e6.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 196, + 498, + 414, + 526 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 86, + 529, + 505, + 579 + ], + "lines": [ + { + "bbox": [ + 86, + 528, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 86, + 532, + 100, + 542 + ], + "score": 1.0, + "content": "217", + "type": "text" + }, + { + "bbox": [ + 103, + 528, + 134, + 544 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 528, + 152, + 542 + ], + "score": 0.92, + "content": "h _ { v } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 528, + 263, + 544 + ], + "score": 1.0, + "content": "is the embedding of node", + "type": "text" + }, + { + "bbox": [ + 263, + 533, + 270, + 541 + ], + "score": 0.78, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 528, + 297, + 544 + ], + "score": 1.0, + "content": "at the", + "type": "text" + }, + { + "bbox": [ + 298, + 531, + 304, + 541 + ], + "score": 0.82, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 528, + 343, + 544 + ], + "score": 1.0, + "content": "-th layer,", + "type": "text" + }, + { + "bbox": [ + 343, + 529, + 360, + 541 + ], + "score": 0.9, + "content": "\\epsilon ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 528, + 506, + 544 + ], + "score": 1.0, + "content": "is an arbitrary number that can be", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 86, + 540, + 508, + 559 + ], + "spans": [ + { + "bbox": [ + 86, + 546, + 100, + 556 + ], + "score": 1.0, + "content": "218", + "type": "text" + }, + { + "bbox": [ + 102, + 540, + 158, + 559 + ], + "score": 1.0, + "content": "learned, and", + "type": "text" + }, + { + "bbox": [ + 158, + 544, + 184, + 556 + ], + "score": 0.91, + "content": "N _ { b } ( v )", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 540, + 263, + 559 + ], + "score": 1.0, + "content": "is the neighbors of", + "type": "text" + }, + { + "bbox": [ + 263, + 547, + 270, + 554 + ], + "score": 0.76, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 540, + 321, + 559 + ], + "score": 1.0, + "content": "via edges in", + "type": "text" + }, + { + "bbox": [ + 321, + 545, + 328, + 554 + ], + "score": 0.75, + "content": "\\mathcal { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 540, + 372, + 559 + ], + "score": 1.0, + "content": ". Note that", + "type": "text" + }, + { + "bbox": [ + 372, + 541, + 390, + 555 + ], + "score": 0.91, + "content": "h _ { v } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 540, + 508, + 559 + ], + "score": 1.0, + "content": "refers to its raw features for", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 85, + 554, + 507, + 571 + ], + "spans": [ + { + "bbox": [ + 85, + 558, + 100, + 568 + ], + "score": 1.0, + "content": "219", + "type": "text" + }, + { + "bbox": [ + 104, + 554, + 134, + 571 + ], + "score": 1.0, + "content": "input.", + "type": "text" + }, + { + "bbox": [ + 135, + 556, + 172, + 567 + ], + "score": 0.9, + "content": "M L P ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 554, + 290, + 571 + ], + "score": 1.0, + "content": "is a Multi-Layer Perceptron", + "type": "text" + }, + { + "bbox": [ + 291, + 557, + 318, + 568 + ], + "score": 0.66, + "content": "( M L P )", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 554, + 349, + 571 + ], + "score": 1.0, + "content": "for the", + "type": "text" + }, + { + "bbox": [ + 350, + 557, + 357, + 567 + ], + "score": 0.81, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 554, + 507, + 571 + ], + "score": 1.0, + "content": "-th layer with a batch normalization", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 86, + 567, + 214, + 580 + ], + "spans": [ + { + "bbox": [ + 86, + 569, + 100, + 579 + ], + "score": 1.0, + "content": "220", + "type": "text" + }, + { + "bbox": [ + 105, + 567, + 214, + 580 + ], + "score": 1.0, + "content": "(Ioffe and Szegedy, 2015).", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 85, + 583, + 506, + 649 + ], + "lines": [ + { + "bbox": [ + 85, + 584, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 85, + 586, + 99, + 596 + ], + "score": 1.0, + "content": "221", + "type": "text" + }, + { + "bbox": [ + 105, + 584, + 506, + 596 + ], + "score": 1.0, + "content": "Furthermore, we actually use heterogeneous GIN, also called HGIN, since there are two types of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 86, + 595, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 86, + 597, + 99, + 606 + ], + "score": 1.0, + "content": "222", + "type": "text" + }, + { + "bbox": [ + 106, + 595, + 344, + 606 + ], + "score": 1.0, + "content": "nodes, machine and operation nodes, and three relations,", + "type": "text" + }, + { + "bbox": [ + 344, + 595, + 378, + 605 + ], + "score": 0.83, + "content": "O O", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 595, + 383, + 606 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 383, + 595, + 420, + 605 + ], + "score": 0.85, + "content": "O \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 595, + 440, + 606 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 440, + 595, + 477, + 605 + ], + "score": 0.91, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 595, + 505, + 606 + ], + "score": 1.0, + "content": "in the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 85, + 605, + 507, + 619 + ], + "spans": [ + { + "bbox": [ + 85, + 608, + 100, + 618 + ], + "score": 1.0, + "content": "223", + "type": "text" + }, + { + "bbox": [ + 105, + 605, + 388, + 619 + ], + "score": 1.0, + "content": "graph representation. Although we do not have cross machine relations", + "type": "text" + }, + { + "bbox": [ + 388, + 606, + 426, + 616 + ], + "score": 0.9, + "content": "M \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 605, + 507, + 619 + ], + "score": 1.0, + "content": "as described above,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 86, + 616, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 86, + 619, + 99, + 628 + ], + "score": 1.0, + "content": "224", + "type": "text" + }, + { + "bbox": [ + 105, + 616, + 506, + 630 + ], + "score": 1.0, + "content": "updating machine nodes requires to include the update from itself as in (1), that is, there is also one", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 86, + 628, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 86, + 630, + 99, + 639 + ], + "score": 1.0, + "content": "225", + "type": "text" + }, + { + "bbox": [ + 106, + 628, + 163, + 640 + ], + "score": 1.0, + "content": "more relation", + "type": "text" + }, + { + "bbox": [ + 163, + 628, + 201, + 637 + ], + "score": 0.91, + "content": "M \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 628, + 505, + 640 + ], + "score": 1.0, + "content": ". Thus, HGIN encodes graph information between all relations by using the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 85, + 637, + 199, + 650 + ], + "spans": [ + { + "bbox": [ + 85, + 640, + 100, + 650 + ], + "score": 1.0, + "content": "226", + "type": "text" + }, + { + "bbox": [ + 105, + 637, + 199, + 650 + ], + "score": 1.0, + "content": "four MLPs as follows,", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38.5 + }, + { + "type": "interline_equation", + "bbox": [ + 183, + 649, + 428, + 679 + ], + "lines": [ + { + "bbox": [ + 183, + 649, + 428, + 679 + ], + "spans": [ + { + "bbox": [ + 183, + 649, + 428, + 679 + ], + "score": 0.94, + "content": "h _ { v } ^ { ( k + 1 ) } = \\sum _ { \\mathcal { R } } M L P _ { \\mathcal { R } } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { \\mathcal { R } } ^ { ( k + 1 ) } ) h _ { v } ^ { ( k ) } + \\sum _ { u \\in N _ { \\mathcal { R } } ( v ) } h _ { u } ^ { ( k ) } )", + "type": "interline_equation", + "image_path": "392e5b294a0810c8146e8d439eb968e34c1d3cbeae97269e1e396ed5a736bbd6.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 183, + 649, + 428, + 659.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 183, + 659.0, + 428, + 669.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 183, + 669.0, + 428, + 679.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 83, + 681, + 507, + 706 + ], + "lines": [ + { + "bbox": [ + 87, + 674, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 87, + 674, + 133, + 704 + ], + "score": 1.0, + "content": "where 227", + "type": "text" + }, + { + "bbox": [ + 133, + 684, + 143, + 693 + ], + "score": 0.83, + "content": "\\mathcal { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 674, + 231, + 704 + ], + "score": 1.0, + "content": "is one of the above fo", + "type": "text" + }, + { + "bbox": [ + 246, + 674, + 272, + 704 + ], + "score": 1.0, + "content": "elatio", + "type": "text" + }, + { + "bbox": [ + 297, + 681, + 334, + 696 + ], + "score": 0.93, + "content": "M L P _ { \\mathcal { R } } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 674, + 397, + 704 + ], + "score": 1.0, + "content": "is the MLP for", + "type": "text" + }, + { + "bbox": [ + 398, + 684, + 407, + 693 + ], + "score": 0.82, + "content": "\\mathcal { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 674, + 482, + 704 + ], + "score": 1.0, + "content": ". For example, for", + "type": "text" + }, + { + "bbox": [ + 482, + 684, + 493, + 694 + ], + "score": 0.87, + "content": "S _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 674, + 505, + 704 + ], + "score": 1.0, + "content": "in", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 694, + 456, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 231, + 706 + ], + "score": 1.0, + "content": "Figure 2 (a), the embedding of", + "type": "text" + }, + { + "bbox": [ + 231, + 695, + 246, + 705 + ], + "score": 0.87, + "content": "M _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 695, + 304, + 706 + ], + "score": 0.9, + "content": "( k + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 694, + 456, + 706 + ], + "score": 1.0, + "content": "-st iteration can be derived as follows.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 707, + 466, + 726 + ], + "lines": [ + { + "bbox": [ + 144, + 707, + 466, + 726 + ], + "spans": [ + { + "bbox": [ + 144, + 707, + 466, + 726 + ], + "score": 0.89, + "content": "h _ { M _ { 1 } } ^ { ( k + 1 ) } = M L P _ { M M } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { M M } ^ { ( k + 1 ) } ) h _ { M _ { 1 } } ^ { ( k ) } ) + M L P _ { O M } ^ { ( k + 1 ) } ( h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } )", + "type": "interline_equation", + "image_path": "3d184bbed84e82a3cd83570384051eeed279a8eb4a46bf70afb8b0f65157e291.jpg" + } + ] + } + ], + "index": 47, + "virtual_lines": [ + { + "bbox": [ + 144, + 707, + 466, + 726 + ], + "spans": [], + "index": 47 + } + ] + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 84, + 63, + 101, + 507 + ], + "lines": [ + { + "bbox": [ + 85, + 74, + 100, + 86 + ], + "spans": [ + { + "bbox": [ + 85, + 74, + 100, + 86 + ], + "score": 1.0, + "content": "186", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 90, + 100, + 102 + ], + "spans": [ + { + "bbox": [ + 85, + 90, + 100, + 102 + ], + "score": 1.0, + "content": "187", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 106, + 100, + 118 + ], + "spans": [ + { + "bbox": [ + 85, + 106, + 100, + 118 + ], + "score": 1.0, + "content": "188", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 118, + 100, + 130 + ], + "spans": [ + { + "bbox": [ + 85, + 118, + 100, + 130 + ], + "score": 1.0, + "content": "189", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 132, + 100, + 146 + ], + "spans": [ + { + "bbox": [ + 85, + 132, + 100, + 146 + ], + "score": 1.0, + "content": "190", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 149, + 100, + 162 + ], + "spans": [ + { + "bbox": [ + 85, + 149, + 100, + 162 + ], + "score": 1.0, + "content": "191", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 165, + 100, + 178 + ], + "spans": [ + { + "bbox": [ + 85, + 165, + 100, + 178 + ], + "score": 1.0, + "content": "192", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 177, + 100, + 189 + ], + "spans": [ + { + "bbox": [ + 85, + 177, + 100, + 189 + ], + "score": 1.0, + "content": "193", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 187, + 101, + 202 + ], + "spans": [ + { + "bbox": [ + 84, + 187, + 101, + 202 + ], + "score": 1.0, + "content": "194", + "type": "text" + } + ] + }, + { + "bbox": [ + 86, + 207, + 100, + 220 + ], + "spans": [ + { + "bbox": [ + 86, + 207, + 100, + 220 + ], + "score": 1.0, + "content": "195", + "type": "text" + } + ] + }, + { + "bbox": [ + 86, + 221, + 100, + 234 + ], + "spans": [ + { + "bbox": [ + 86, + 221, + 100, + 234 + ], + "score": 1.0, + "content": "196", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 233, + 100, + 246 + ], + "spans": [ + { + "bbox": [ + 85, + 233, + 100, + 246 + ], + "score": 1.0, + "content": "197", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 246, + 101, + 261 + ], + "spans": [ + { + "bbox": [ + 84, + 246, + 101, + 261 + ], + "score": 1.0, + "content": "198", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 263, + 100, + 276 + ], + "spans": [ + { + "bbox": [ + 84, + 263, + 100, + 276 + ], + "score": 1.0, + "content": "199", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 274, + 100, + 287 + ], + "spans": [ + { + "bbox": [ + 84, + 274, + 100, + 287 + ], + "score": 1.0, + "content": "200", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 285, + 100, + 298 + ], + "spans": [ + { + "bbox": [ + 84, + 285, + 100, + 298 + ], + "score": 1.0, + "content": "201", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 296, + 100, + 309 + ], + "spans": [ + { + "bbox": [ + 84, + 296, + 100, + 309 + ], + "score": 1.0, + "content": "202", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 307, + 100, + 320 + ], + "spans": [ + { + "bbox": [ + 85, + 307, + 100, + 320 + ], + "score": 1.0, + "content": "203", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 318, + 100, + 332 + ], + "spans": [ + { + "bbox": [ + 84, + 318, + 100, + 332 + ], + "score": 1.0, + "content": "204", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 338, + 100, + 350 + ], + "spans": [ + { + "bbox": [ + 85, + 338, + 100, + 350 + ], + "score": 1.0, + "content": "205", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 353, + 100, + 366 + ], + "spans": [ + { + "bbox": [ + 85, + 353, + 100, + 366 + ], + "score": 1.0, + "content": "206", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 365, + 100, + 379 + ], + "spans": [ + { + "bbox": [ + 84, + 365, + 100, + 379 + ], + "score": 1.0, + "content": "207", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 389, + 101, + 402 + ], + "spans": [ + { + "bbox": [ + 84, + 389, + 101, + 402 + ], + "score": 1.0, + "content": "208", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 408, + 100, + 421 + ], + "spans": [ + { + "bbox": [ + 84, + 408, + 100, + 421 + ], + "score": 1.0, + "content": "209", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 419, + 100, + 432 + ], + "spans": [ + { + "bbox": [ + 84, + 419, + 100, + 432 + ], + "score": 1.0, + "content": "210", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 430, + 100, + 443 + ], + "spans": [ + { + "bbox": [ + 84, + 430, + 100, + 443 + ], + "score": 1.0, + "content": "211", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 442, + 100, + 454 + ], + "spans": [ + { + "bbox": [ + 84, + 442, + 100, + 454 + ], + "score": 1.0, + "content": "212", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 453, + 100, + 465 + ], + "spans": [ + { + "bbox": [ + 84, + 453, + 100, + 465 + ], + "score": 1.0, + "content": "213", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 463, + 101, + 476 + ], + "spans": [ + { + "bbox": [ + 84, + 463, + 101, + 476 + ], + "score": 1.0, + "content": "214", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 474, + 101, + 488 + ], + "spans": [ + { + "bbox": [ + 84, + 474, + 101, + 488 + ], + "score": 1.0, + "content": "215", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 486, + 101, + 500 + ], + "spans": [ + { + "bbox": [ + 84, + 486, + 101, + 500 + ], + "score": 1.0, + "content": "216", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 70, + 506, + 145 + ], + "lines": [ + { + "bbox": [ + 162, + 65, + 505, + 111 + ], + "spans": [ + { + "bbox": [ + 162, + 65, + 243, + 111 + ], + "score": 1.0, + "content": "ob remaining time , and let the process", + "type": "text" + }, + { + "bbox": [ + 243, + 70, + 269, + 87 + ], + "score": 0.93, + "content": "\\bar { T } _ { j , i } ^ { ( j o b ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 65, + 349, + 111 + ], + "score": 1.0, + "content": ": Let the rest of pre for the whole job", + "type": "text" + }, + { + "bbox": [ + 439, + 73, + 450, + 86 + ], + "score": 0.87, + "content": "J _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 70, + 505, + 87 + ], + "score": 0.91, + "content": "T _ { j , i } ^ { ( j o b ) } =", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 84, + 509, + 128 + ], + "spans": [ + { + "bbox": [ + 107, + 86, + 161, + 102 + ], + "score": 0.91, + "content": "\\sum _ { \\forall i ^ { \\prime } \\geq i } T _ { j , i ^ { \\prime } } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 85, + 201, + 128 + ], + "score": 1.0, + "content": "aced by th", + "type": "text" + }, + { + "bbox": [ + 219, + 85, + 349, + 128 + ], + "score": 1.0, + "content": "ing timocessing time for the original jo", + "type": "text" + }, + { + "bbox": [ + 349, + 89, + 356, + 101 + ], + "score": 0.76, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 86, + 454, + 102 + ], + "score": 0.93, + "content": "\\begin{array} { r } { T _ { j } ^ { ( j o b ) } = \\sum _ { \\forall i ^ { \\prime } } T _ { j , i ^ { \\prime } } ^ { ( o p ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 84, + 509, + 104 + ], + "score": 1.0, + "content": ". In practice,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 99, + 96, + 507, + 122 + ], + "spans": [ + { + "bbox": [ + 99, + 96, + 141, + 121 + ], + "score": 1.0, + "content": "T ( j ob )j is", + "type": "text" + }, + { + "bbox": [ + 353, + 105, + 360, + 116 + ], + "score": 0.77, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 102, + 484, + 118 + ], + "score": 0.91, + "content": "\\bar { T } _ { j , i } ^ { ( j o b ) } = T _ { j , i } ^ { ( j o b ) } / T _ { j } ^ { ( j o b ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 98, + 507, + 122 + ], + "score": 1.0, + "content": ". For", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 115, + 507, + 130 + ], + "spans": [ + { + "bbox": [ + 104, + 115, + 201, + 130 + ], + "score": 1.0, + "content": "FJSP, since operations", + "type": "text" + }, + { + "bbox": [ + 201, + 117, + 219, + 129 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 115, + 434, + 130 + ], + "score": 1.0, + "content": "can be dispatched to different designated machines", + "type": "text" + }, + { + "bbox": [ + 434, + 118, + 447, + 127 + ], + "score": 0.86, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 115, + 507, + 130 + ], + "score": 1.0, + "content": ", say with the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 420, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 171, + 142 + ], + "score": 1.0, + "content": "processing time", + "type": "text" + }, + { + "bbox": [ + 172, + 126, + 282, + 147 + ], + "score": 1.0, + "content": "T (op)j,i,l , we simply let T (op)j,i", + "type": "text" + }, + { + "bbox": [ + 278, + 127, + 349, + 146 + ], + "score": 1.0, + "content": "be the average of", + "type": "text" + }, + { + "bbox": [ + 350, + 128, + 372, + 145 + ], + "score": 0.93, + "content": "T _ { j , i , l } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 127, + 401, + 146 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 401, + 130, + 415, + 142 + ], + "score": 0.87, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 127, + 420, + 146 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 99, + 65, + 509, + 147 + ] + }, + { + "type": "text", + "bbox": [ + 102, + 147, + 444, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 146, + 446, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 294, + 162 + ], + "score": 1.0, + "content": "For machine nodes corresponding to machines", + "type": "text" + }, + { + "bbox": [ + 295, + 148, + 308, + 159 + ], + "score": 0.89, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 146, + 446, + 162 + ], + "score": 1.0, + "content": ", we have the following attributes:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 146, + 446, + 162 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 505, + 198 + ], + "lines": [ + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 172, + 178 + ], + "score": 1.0, + "content": "Machine status", + "type": "text" + }, + { + "bbox": [ + 173, + 165, + 183, + 176 + ], + "score": 0.86, + "content": "\\phi _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 164, + 269, + 178 + ], + "score": 1.0, + "content": ": The machine status", + "type": "text" + }, + { + "bbox": [ + 270, + 165, + 279, + 176 + ], + "score": 0.87, + "content": "\\phi _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "is processing if some operation has been dispatched to", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 214, + 188 + ], + "score": 1.0, + "content": "and is being processed by", + "type": "text" + }, + { + "bbox": [ + 214, + 176, + 228, + 186 + ], + "score": 0.88, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 176, + 239, + 188 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 239, + 177, + 246, + 185 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 176, + 471, + 188 + ], + "score": 1.0, + "content": ", and idle otherwise (no operation is being processed at", + "type": "text" + }, + { + "bbox": [ + 472, + 178, + 478, + 185 + ], + "score": 0.7, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 176, + 505, + 188 + ], + "score": 1.0, + "content": "). The", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 186, + 486, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 486, + 199 + ], + "score": 1.0, + "content": "attribute is a one-hot vector to represent the current status, which is one of processing and idle.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 164, + 506, + 199 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 203, + 506, + 259 + ], + "lines": [ + { + "bbox": [ + 102, + 200, + 505, + 220 + ], + "spans": [ + { + "bbox": [ + 102, + 200, + 265, + 220 + ], + "score": 1.0, + "content": "Normalized operation processing time", + "type": "text" + }, + { + "bbox": [ + 266, + 203, + 295, + 218 + ], + "score": 0.92, + "content": "\\bar { T } _ { l } ^ { ( m a c ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 200, + 366, + 220 + ], + "score": 1.0, + "content": ": On the machine", + "type": "text" + }, + { + "bbox": [ + 366, + 205, + 380, + 217 + ], + "score": 0.89, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 200, + 464, + 220 + ], + "score": 1.0, + "content": ", the processing time", + "type": "text" + }, + { + "bbox": [ + 464, + 203, + 494, + 218 + ], + "score": 0.93, + "content": "T _ { l } ^ { ( m a c ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 200, + 505, + 220 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 214, + 507, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 217, + 129, + 234 + ], + "score": 0.91, + "content": "T _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 214, + 343, + 235 + ], + "score": 1.0, + "content": "(the same as the normalized processing time for node", + "type": "text" + }, + { + "bbox": [ + 344, + 220, + 362, + 232 + ], + "score": 0.89, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 214, + 507, + 235 + ], + "score": 1.0, + "content": ") if the machine status is processing,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 231, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 220, + 245 + ], + "score": 1.0, + "content": "i.e., some ongoing operation", + "type": "text" + }, + { + "bbox": [ + 221, + 232, + 238, + 244 + ], + "score": 0.93, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 231, + 506, + 245 + ], + "score": 1.0, + "content": "is being processed but not finished yet, is zero if the machine status", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 101, + 237, + 449, + 264 + ], + "spans": [ + { + "bbox": [ + 101, + 237, + 281, + 264 + ], + "score": 1.0, + "content": "is idle. Then, this attribute is normalized to", + "type": "text" + }, + { + "bbox": [ + 281, + 243, + 304, + 257 + ], + "score": 0.91, + "content": "T _ { m a x } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 237, + 342, + 264 + ], + "score": 1.0, + "content": "and thus", + "type": "text" + }, + { + "bbox": [ + 342, + 243, + 441, + 259 + ], + "score": 0.93, + "content": "\\bar { T } _ { l } ^ { ( m a c ) } = T _ { l } ^ { ( m a c ) } / T _ { m a x } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 237, + 449, + 264 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 101, + 200, + 507, + 264 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 262, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "score": 1.0, + "content": "Now, consider edges in a residual scheduling graph. As described above, there exists three relationship", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 273, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 167, + 286 + ], + "score": 1.0, + "content": "sets for edges,", + "type": "text" + }, + { + "bbox": [ + 167, + 273, + 200, + 284 + ], + "score": 0.88, + "content": "O O", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 273, + 205, + 286 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 205, + 273, + 241, + 284 + ], + "score": 0.88, + "content": "O \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 273, + 260, + 286 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 260, + 273, + 296, + 284 + ], + "score": 0.9, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 273, + 412, + 286 + ], + "score": 1.0, + "content": ". First, for the same job, say", + "type": "text" + }, + { + "bbox": [ + 413, + 274, + 424, + 286 + ], + "score": 0.88, + "content": "J _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 273, + 505, + 286 + ], + "score": 1.0, + "content": ", all of its operation", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 283, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 147, + 296 + ], + "score": 1.0, + "content": "nodes for", + "type": "text" + }, + { + "bbox": [ + 147, + 285, + 165, + 297 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 283, + 505, + 296 + ], + "score": 1.0, + "content": "are fully connected. Note that for residual scheduling the operations finished by the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 175, + 308 + ], + "score": 1.0, + "content": "dispatching time", + "type": "text" + }, + { + "bbox": [ + 175, + 298, + 182, + 305 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 295, + 480, + 308 + ], + "score": 1.0, + "content": "are removed and thus have no edges to them. Second, a machine node for", + "type": "text" + }, + { + "bbox": [ + 481, + 296, + 495, + 306 + ], + "score": 0.89, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 306, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 251, + 319 + ], + "score": 1.0, + "content": "connected to an operation node for", + "type": "text" + }, + { + "bbox": [ + 251, + 306, + 268, + 318 + ], + "score": 0.89, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 306, + 337, + 319 + ], + "score": 1.0, + "content": ", if the operation", + "type": "text" + }, + { + "bbox": [ + 338, + 306, + 355, + 318 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 306, + 505, + 319 + ], + "score": 1.0, + "content": "is designated to be processed on the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 317, + 500, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 142, + 329 + ], + "score": 1.0, + "content": "machine", + "type": "text" + }, + { + "bbox": [ + 143, + 317, + 156, + 328 + ], + "score": 0.87, + "content": "M _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 317, + 255, + 329 + ], + "score": 1.0, + "content": ", which forms two edges", + "type": "text" + }, + { + "bbox": [ + 256, + 317, + 291, + 327 + ], + "score": 0.9, + "content": "\\dot { O } M", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 317, + 309, + 329 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 309, + 317, + 344, + 327 + ], + "score": 0.9, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 317, + 500, + 329 + ], + "score": 1.0, + "content": ". Both contains the following attribute.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 261, + 506, + 329 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 506, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 328, + 507, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 271, + 348 + ], + "score": 1.0, + "content": "Normalized operation processing tim", + "type": "text" + }, + { + "bbox": [ + 267, + 329, + 312, + 349 + ], + "score": 1.0, + "content": "e T¯(edge):", + "type": "text" + }, + { + "bbox": [ + 306, + 329, + 376, + 354 + ], + "score": 1.0, + "content": "The attribute is", + "type": "text" + }, + { + "bbox": [ + 376, + 333, + 474, + 350 + ], + "score": 0.93, + "content": "\\bar { T } _ { j , i , l } ^ { ( e d g e ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 328, + 482, + 352 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 480, + 335, + 507, + 348 + ], + "score": 1.0, + "content": "Here,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 348, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 107, + 348, + 165, + 366 + ], + "score": 0.93, + "content": "T _ { j , i } ^ { ( o p ) } = T _ { j , i , l } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 352, + 319, + 364 + ], + "score": 0.91, + "content": "O _ { j , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 349, + 479, + 366 + ], + "score": 0.92, + "content": "T _ { j , i } ^ { ( o p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 351, + 506, + 364 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 364, + 249, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 249, + 376 + ], + "score": 1.0, + "content": "remaining time as described above.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 328, + 507, + 376 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 387, + 230, + 399 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 231, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 231, + 400 + ], + "score": 1.0, + "content": "3.3 Graph Neural Network", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 94, + 407, + 506, + 497 + ], + "lines": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 505, + 421 + ], + "score": 1.0, + "content": "In this subsection, we present our model based on graph neural network (GNN). GNN are a family", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "score": 1.0, + "content": "of deep neural networks (Battaglia et al., 2018) that can learn representation of graph-structured", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "data, widely used in many applications (Lv et al., 2021; Zhou et al., 2020). A GNN aggregates", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 441, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 505, + 453 + ], + "score": 1.0, + "content": "information from node itself and its neighboring nodes and then update the data itself, which allows", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "the GNN to capture the complex relationships within the data graph. For GNN, we choose Graph", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 461, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 442, + 475 + ], + "score": 1.0, + "content": "Isomorphism Network (GIN), which was shown to have strong discriminative power (", + "type": "text" + }, + { + "bbox": [ + 442, + 462, + 456, + 472 + ], + "score": 0.28, + "content": "\\mathrm { { X u } }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 461, + 506, + 475 + ], + "score": 1.0, + "content": "et al., 2019)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 472, + 507, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 316, + 486 + ], + "score": 1.0, + "content": "and summarily reviewed as follows. Given a graph", + "type": "text" + }, + { + "bbox": [ + 316, + 473, + 363, + 485 + ], + "score": 0.93, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 472, + 381, + 486 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 382, + 473, + 392, + 483 + ], + "score": 0.8, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 472, + 448, + 486 + ], + "score": 1.0, + "content": "GNN layers", + "type": "text" + }, + { + "bbox": [ + 448, + 474, + 459, + 483 + ], + "score": 0.79, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 472, + 507, + 486 + ], + "score": 1.0, + "content": "iterations),", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 484, + 470, + 498 + ], + "spans": [ + { + "bbox": [ + 104, + 484, + 180, + 498 + ], + "score": 1.0, + "content": "GIN performs the", + "type": "text" + }, + { + "bbox": [ + 180, + 486, + 186, + 495 + ], + "score": 0.83, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 484, + 363, + 498 + ], + "score": 1.0, + "content": "-th iterations of updating feature embedding", + "type": "text" + }, + { + "bbox": [ + 363, + 484, + 381, + 496 + ], + "score": 0.9, + "content": "h ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 484, + 439, + 498 + ], + "score": 1.0, + "content": "for each node", + "type": "text" + }, + { + "bbox": [ + 440, + 486, + 465, + 496 + ], + "score": 0.9, + "content": "v \\in \\mathcal V", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 484, + 470, + 498 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 26.5, + "bbox_fs": [ + 104, + 407, + 507, + 498 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 196, + 498, + 414, + 526 + ], + "lines": [ + { + "bbox": [ + 196, + 498, + 414, + 526 + ], + "spans": [ + { + "bbox": [ + 196, + 498, + 414, + 526 + ], + "score": 0.92, + "content": "h _ { v } ^ { ( k ) } = M L P ^ { ( k ) } ( ( 1 + \\epsilon ^ { ( k ) } ) h _ { v } ^ { ( k - 1 ) } + \\sum _ { u \\in N _ { b } ( v ) } h _ { u } ^ { ( k - 1 ) } ) ,", + "type": "interline_equation", + "image_path": "2ff8cf48f7fc89b4b2026b4e2f6ba5a18a536573b1a5ce94368e07ebd887b7e6.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 196, + 498, + 414, + 526 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "index", + "bbox": [ + 86, + 529, + 505, + 579 + ], + "lines": [ + { + "bbox": [ + 86, + 528, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 86, + 532, + 100, + 542 + ], + "score": 1.0, + "content": "217", + "type": "text" + }, + { + "bbox": [ + 103, + 528, + 134, + 544 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 528, + 152, + 542 + ], + "score": 0.92, + "content": "h _ { v } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 528, + 263, + 544 + ], + "score": 1.0, + "content": "is the embedding of node", + "type": "text" + }, + { + "bbox": [ + 263, + 533, + 270, + 541 + ], + "score": 0.78, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 528, + 297, + 544 + ], + "score": 1.0, + "content": "at the", + "type": "text" + }, + { + "bbox": [ + 298, + 531, + 304, + 541 + ], + "score": 0.82, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 528, + 343, + 544 + ], + "score": 1.0, + "content": "-th layer,", + "type": "text" + }, + { + "bbox": [ + 343, + 529, + 360, + 541 + ], + "score": 0.9, + "content": "\\epsilon ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 528, + 506, + 544 + ], + "score": 1.0, + "content": "is an arbitrary number that can be", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 540, + 508, + 559 + ], + "spans": [ + { + "bbox": [ + 86, + 546, + 100, + 556 + ], + "score": 1.0, + "content": "218", + "type": "text" + }, + { + "bbox": [ + 102, + 540, + 158, + 559 + ], + "score": 1.0, + "content": "learned, and", + "type": "text" + }, + { + "bbox": [ + 158, + 544, + 184, + 556 + ], + "score": 0.91, + "content": "N _ { b } ( v )", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 540, + 263, + 559 + ], + "score": 1.0, + "content": "is the neighbors of", + "type": "text" + }, + { + "bbox": [ + 263, + 547, + 270, + 554 + ], + "score": 0.76, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 540, + 321, + 559 + ], + "score": 1.0, + "content": "via edges in", + "type": "text" + }, + { + "bbox": [ + 321, + 545, + 328, + 554 + ], + "score": 0.75, + "content": "\\mathcal { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 540, + 372, + 559 + ], + "score": 1.0, + "content": ". Note that", + "type": "text" + }, + { + "bbox": [ + 372, + 541, + 390, + 555 + ], + "score": 0.91, + "content": "h _ { v } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 540, + 508, + 559 + ], + "score": 1.0, + "content": "refers to its raw features for", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 554, + 507, + 571 + ], + "spans": [ + { + "bbox": [ + 85, + 558, + 100, + 568 + ], + "score": 1.0, + "content": "219", + "type": "text" + }, + { + "bbox": [ + 104, + 554, + 134, + 571 + ], + "score": 1.0, + "content": "input.", + "type": "text" + }, + { + "bbox": [ + 135, + 556, + 172, + 567 + ], + "score": 0.9, + "content": "M L P ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 554, + 290, + 571 + ], + "score": 1.0, + "content": "is a Multi-Layer Perceptron", + "type": "text" + }, + { + "bbox": [ + 291, + 557, + 318, + 568 + ], + "score": 0.66, + "content": "( M L P )", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 554, + 349, + 571 + ], + "score": 1.0, + "content": "for the", + "type": "text" + }, + { + "bbox": [ + 350, + 557, + 357, + 567 + ], + "score": 0.81, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 554, + 507, + 571 + ], + "score": 1.0, + "content": "-th layer with a batch normalization", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 567, + 214, + 580 + ], + "spans": [ + { + "bbox": [ + 86, + 569, + 100, + 579 + ], + "score": 1.0, + "content": "220", + "type": "text" + }, + { + "bbox": [ + 105, + 567, + 214, + 580 + ], + "score": 1.0, + "content": "(Ioffe and Szegedy, 2015).", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 584, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 85, + 586, + 99, + 596 + ], + "score": 1.0, + "content": "221", + "type": "text" + }, + { + "bbox": [ + 105, + 584, + 506, + 596 + ], + "score": 1.0, + "content": "Furthermore, we actually use heterogeneous GIN, also called HGIN, since there are two types of", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 595, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 86, + 597, + 99, + 606 + ], + "score": 1.0, + "content": "222", + "type": "text" + }, + { + "bbox": [ + 106, + 595, + 344, + 606 + ], + "score": 1.0, + "content": "nodes, machine and operation nodes, and three relations,", + "type": "text" + }, + { + "bbox": [ + 344, + 595, + 378, + 605 + ], + "score": 0.83, + "content": "O O", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 595, + 383, + 606 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 383, + 595, + 420, + 605 + ], + "score": 0.85, + "content": "O \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 595, + 440, + 606 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 440, + 595, + 477, + 605 + ], + "score": 0.91, + "content": "M O", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 595, + 505, + 606 + ], + "score": 1.0, + "content": "in the", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 605, + 507, + 619 + ], + "spans": [ + { + "bbox": [ + 85, + 608, + 100, + 618 + ], + "score": 1.0, + "content": "223", + "type": "text" + }, + { + "bbox": [ + 105, + 605, + 388, + 619 + ], + "score": 1.0, + "content": "graph representation. Although we do not have cross machine relations", + "type": "text" + }, + { + "bbox": [ + 388, + 606, + 426, + 616 + ], + "score": 0.9, + "content": "M \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 605, + 507, + 619 + ], + "score": 1.0, + "content": "as described above,", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 616, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 86, + 619, + 99, + 628 + ], + "score": 1.0, + "content": "224", + "type": "text" + }, + { + "bbox": [ + 105, + 616, + 506, + 630 + ], + "score": 1.0, + "content": "updating machine nodes requires to include the update from itself as in (1), that is, there is also one", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 628, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 86, + 630, + 99, + 639 + ], + "score": 1.0, + "content": "225", + "type": "text" + }, + { + "bbox": [ + 106, + 628, + 163, + 640 + ], + "score": 1.0, + "content": "more relation", + "type": "text" + }, + { + "bbox": [ + 163, + 628, + 201, + 637 + ], + "score": 0.91, + "content": "M \\to M", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 628, + 505, + 640 + ], + "score": 1.0, + "content": ". Thus, HGIN encodes graph information between all relations by using the", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 637, + 199, + 650 + ], + "spans": [ + { + "bbox": [ + 85, + 640, + 100, + 650 + ], + "score": 1.0, + "content": "226", + "type": "text" + }, + { + "bbox": [ + 105, + 637, + 199, + 650 + ], + "score": 1.0, + "content": "four MLPs as follows,", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + } + ], + "index": 33.5, + "bbox_fs": [ + 85, + 528, + 508, + 580 + ] + }, + { + "type": "index", + "bbox": [ + 85, + 583, + 506, + 649 + ], + "lines": [], + "index": 38.5, + "bbox_fs": [ + 85, + 584, + 507, + 650 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 183, + 649, + 428, + 679 + ], + "lines": [ + { + "bbox": [ + 183, + 649, + 428, + 679 + ], + "spans": [ + { + "bbox": [ + 183, + 649, + 428, + 679 + ], + "score": 0.94, + "content": "h _ { v } ^ { ( k + 1 ) } = \\sum _ { \\mathcal { R } } M L P _ { \\mathcal { R } } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { \\mathcal { R } } ^ { ( k + 1 ) } ) h _ { v } ^ { ( k ) } + \\sum _ { u \\in N _ { \\mathcal { R } } ( v ) } h _ { u } ^ { ( k ) } )", + "type": "interline_equation", + "image_path": "392e5b294a0810c8146e8d439eb968e34c1d3cbeae97269e1e396ed5a736bbd6.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 183, + 649, + 428, + 659.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 183, + 659.0, + 428, + 669.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 183, + 669.0, + 428, + 679.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 83, + 681, + 507, + 706 + ], + "lines": [ + { + "bbox": [ + 87, + 674, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 87, + 674, + 133, + 704 + ], + "score": 1.0, + "content": "where 227", + "type": "text" + }, + { + "bbox": [ + 133, + 684, + 143, + 693 + ], + "score": 0.83, + "content": "\\mathcal { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 674, + 231, + 704 + ], + "score": 1.0, + "content": "is one of the above fo", + "type": "text" + }, + { + "bbox": [ + 246, + 674, + 272, + 704 + ], + "score": 1.0, + "content": "elatio", + "type": "text" + }, + { + "bbox": [ + 297, + 681, + 334, + 696 + ], + "score": 0.93, + "content": "M L P _ { \\mathcal { R } } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 674, + 397, + 704 + ], + "score": 1.0, + "content": "is the MLP for", + "type": "text" + }, + { + "bbox": [ + 398, + 684, + 407, + 693 + ], + "score": 0.82, + "content": "\\mathcal { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 674, + 482, + 704 + ], + "score": 1.0, + "content": ". For example, for", + "type": "text" + }, + { + "bbox": [ + 482, + 684, + 493, + 694 + ], + "score": 0.87, + "content": "S _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 674, + 505, + 704 + ], + "score": 1.0, + "content": "in", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 694, + 456, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 231, + 706 + ], + "score": 1.0, + "content": "Figure 2 (a), the embedding of", + "type": "text" + }, + { + "bbox": [ + 231, + 695, + 246, + 705 + ], + "score": 0.87, + "content": "M _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 695, + 304, + 706 + ], + "score": 0.9, + "content": "( k + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 694, + 456, + 706 + ], + "score": 1.0, + "content": "-st iteration can be derived as follows.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 87, + 674, + 505, + 706 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 707, + 466, + 726 + ], + "lines": [ + { + "bbox": [ + 144, + 707, + 466, + 726 + ], + "spans": [ + { + "bbox": [ + 144, + 707, + 466, + 726 + ], + "score": 0.89, + "content": "h _ { M _ { 1 } } ^ { ( k + 1 ) } = M L P _ { M M } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { M M } ^ { ( k + 1 ) } ) h _ { M _ { 1 } } ^ { ( k ) } ) + M L P _ { O M } ^ { ( k + 1 ) } ( h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } )", + "type": "interline_equation", + "image_path": "3d184bbed84e82a3cd83570384051eeed279a8eb4a46bf70afb8b0f65157e291.jpg" + } + ] + } + ], + "index": 47, + "virtual_lines": [ + { + "bbox": [ + 144, + 707, + 466, + 726 + ], + "spans": [], + "index": 47 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 86, + 72, + 355, + 84 + ], + "lines": [ + { + "bbox": [ + 84, + 72, + 356, + 86 + ], + "spans": [ + { + "bbox": [ + 84, + 72, + 220, + 86 + ], + "score": 1.0, + "content": "229 Similarly, the embedding of", + "type": "text" + }, + { + "bbox": [ + 221, + 73, + 240, + 85 + ], + "score": 0.92, + "content": "O _ { 1 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 72, + 266, + 86 + ], + "score": 1.0, + "content": "in the", + "type": "text" + }, + { + "bbox": [ + 267, + 73, + 297, + 85 + ], + "score": 0.92, + "content": "( k + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 72, + 356, + 86 + ], + "score": 1.0, + "content": "-st iteration is:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 87, + 466, + 106 + ], + "lines": [ + { + "bbox": [ + 144, + 87, + 466, + 106 + ], + "spans": [ + { + "bbox": [ + 144, + 87, + 466, + 106 + ], + "score": 0.91, + "content": "h _ { O _ { 1 , 1 } } ^ { ( k + 1 ) } = M L P _ { O O } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { O O } ^ { ( k + 1 ) } ) h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } ) + M L P _ { M O } ^ { ( k + 1 ) } ( h _ { M _ { 1 } } ^ { ( k ) } )", + "type": "interline_equation", + "image_path": "3a318d233ecc5b46d854239d39f00d2fdff89b88a439536327c31a2b3834eceb.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 144, + 87, + 466, + 106 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 86, + 113, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 86, + 114, + 506, + 126 + ], + "spans": [ + { + "bbox": [ + 86, + 115, + 100, + 126 + ], + "score": 1.0, + "content": "230", + "type": "text" + }, + { + "bbox": [ + 104, + 114, + 506, + 126 + ], + "score": 1.0, + "content": "In our approach, an action includes the two phases, graph embedding phase and action selection phase.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 86, + 120, + 507, + 146 + ], + "spans": [ + { + "bbox": [ + 86, + 128, + 99, + 139 + ], + "score": 1.0, + "content": "231", + "type": "text" + }, + { + "bbox": [ + 100, + 120, + 122, + 146 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 122, + 125, + 140, + 140 + ], + "score": 0.9, + "content": "h _ { \\mathcal { G } } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 120, + 313, + 146 + ], + "score": 1.0, + "content": "denote the whole embedding of the graphs", + "type": "text" + }, + { + "bbox": [ + 313, + 127, + 321, + 138 + ], + "score": 0.83, + "content": "\\mathcal { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 120, + 507, + 146 + ], + "score": 1.0, + "content": ", a summation of the embeddings of all nodes,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 86, + 134, + 510, + 160 + ], + "spans": [ + { + "bbox": [ + 86, + 144, + 100, + 154 + ], + "score": 1.0, + "content": "232", + "type": "text" + }, + { + "bbox": [ + 106, + 140, + 134, + 153 + ], + "score": 0.9, + "content": "h _ { v } ^ { ( k + 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 134, + 510, + 160 + ], + "score": 1.0, + "content": ". In the graph embedding phase, we use an HGIN to encode node and graph embeddings as", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 86, + 153, + 438, + 165 + ], + "spans": [ + { + "bbox": [ + 86, + 155, + 100, + 164 + ], + "score": 1.0, + "content": "233", + "type": "text" + }, + { + "bbox": [ + 105, + 153, + 438, + 165 + ], + "score": 1.0, + "content": "described above. An example with three HGIN layers is illustrated in Figure 3 (b).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 97, + 169, + 505, + 239 + ], + "lines": [ + { + "bbox": [ + 105, + 168, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 505, + 182 + ], + "score": 1.0, + "content": "In the action selection phase, we select an action based on a policy, after node and graph embedding", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 179, + 505, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 505, + 192 + ], + "score": 1.0, + "content": "are encoded in the graph embedding phase. The policy is described as follows. First, collect all", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 191, + 506, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 177, + 203 + ], + "score": 1.0, + "content": "ready operations", + "type": "text" + }, + { + "bbox": [ + 177, + 192, + 187, + 201 + ], + "score": 0.8, + "content": "O", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 191, + 310, + 203 + ], + "score": 1.0, + "content": "to be dispatched to machines", + "type": "text" + }, + { + "bbox": [ + 310, + 192, + 322, + 201 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 191, + 405, + 203 + ], + "score": 1.0, + "content": ". Then, for all pairs", + "type": "text" + }, + { + "bbox": [ + 405, + 191, + 436, + 203 + ], + "score": 0.59, + "content": "( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 191, + 506, + 203 + ], + "score": 1.0, + "content": ", feed their node", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 102, + 198, + 510, + 223 + ], + "spans": [ + { + "bbox": [ + 102, + 198, + 157, + 223 + ], + "score": 1.0, + "content": "embeddings", + "type": "text" + }, + { + "bbox": [ + 157, + 202, + 203, + 217 + ], + "score": 0.8, + "content": "( h _ { M } ^ { ( k ) } , h _ { O } ^ { ( k ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 198, + 252, + 223 + ], + "score": 1.0, + "content": "into a MLP", + "type": "text" + }, + { + "bbox": [ + 253, + 204, + 309, + 217 + ], + "score": 0.86, + "content": "S c o r e ( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 198, + 510, + 223 + ], + "score": 1.0, + "content": "to calculate their scores as shown in Figure 3 (c).", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 104, + 214, + 221, + 228 + ], + "score": 1.0, + "content": "The probability of selecting", + "type": "text" + }, + { + "bbox": [ + 222, + 216, + 253, + 227 + ], + "score": 0.87, + "content": "( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "is calculated based on a softmax function of all scores, which", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 226, + 324, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 232, + 239 + ], + "score": 1.0, + "content": "also serves as the model policy", + "type": "text" + }, + { + "bbox": [ + 232, + 229, + 240, + 237 + ], + "score": 0.78, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 226, + 324, + 239 + ], + "score": 1.0, + "content": "for the current state.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 103, + 250, + 241, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 243, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 243, + 266 + ], + "score": 1.0, + "content": "3.4 Policy-Based RL Training", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 92, + 270, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 91, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 91, + 271, + 505, + 282 + ], + "score": 1.0, + "content": "41 In this paper, we propose to use a policy-based RL training mechanism that follows REINFORCE", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 91, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 91, + 285, + 99, + 292 + ], + "score": 1.0, + "content": "42", + "type": "text" + }, + { + "bbox": [ + 105, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "(Sutton and Barto, 2018) to update our model by policy gradient with a normalized advantage", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 91, + 293, + 337, + 305 + ], + "spans": [ + { + "bbox": [ + 91, + 296, + 99, + 303 + ], + "score": 1.0, + "content": "43", + "type": "text" + }, + { + "bbox": [ + 105, + 293, + 279, + 305 + ], + "score": 1.0, + "content": "makespan with respect to a baseline policy", + "type": "text" + }, + { + "bbox": [ + 279, + 294, + 290, + 303 + ], + "score": 0.86, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 293, + 337, + 305 + ], + "score": 1.0, + "content": "as follows.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 306, + 396, + 339 + ], + "lines": [ + { + "bbox": [ + 215, + 306, + 396, + 339 + ], + "spans": [ + { + "bbox": [ + 215, + 306, + 396, + 339 + ], + "score": 0.94, + "content": "A _ { \\pi } ( S , a ) = { \\frac { T _ { \\pi _ { b } } ^ { ( m k s p ) } ( S , a ) - T _ { \\pi } ^ { ( m k s p ) } ( S , a ) } { T _ { \\pi _ { b } } ^ { ( m k s p ) } ( S , a ) } }", + "type": "interline_equation", + "image_path": "b06d82d1c07484ee7e7eab67883f499d4e532d443dc6370d26bccc60092f3d81.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 215, + 306, + 396, + 322.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 215, + 322.5, + 396, + 339.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 86, + 345, + 505, + 402 + ], + "lines": [ + { + "bbox": [ + 86, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 86, + 348, + 100, + 357 + ], + "score": 1.0, + "content": "244", + "type": "text" + }, + { + "bbox": [ + 105, + 346, + 381, + 358 + ], + "score": 1.0, + "content": "In this paper, we choose a lightweight PDR, MWKR, as baseline", + "type": "text" + }, + { + "bbox": [ + 381, + 348, + 392, + 357 + ], + "score": 0.85, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 346, + 505, + 358 + ], + "score": 1.0, + "content": ", which performed best for", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 86, + 357, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 86, + 360, + 100, + 369 + ], + "score": 1.0, + "content": "245", + "type": "text" + }, + { + "bbox": [ + 105, + 357, + 505, + 370 + ], + "score": 1.0, + "content": "makespan among all PDRs reported from the previous work (Zhang et al., 2020). In fact, our", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 85, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 85, + 370, + 100, + 380 + ], + "score": 1.0, + "content": "246", + "type": "text" + }, + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "experiment also shows that using MWKR is better than the other PDRs shown in the appendix. The", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 85, + 378, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 85, + 380, + 100, + 391 + ], + "score": 1.0, + "content": "247", + "type": "text" + }, + { + "bbox": [ + 105, + 378, + 176, + 393 + ], + "score": 1.0, + "content": "model for policy", + "type": "text" + }, + { + "bbox": [ + 176, + 381, + 183, + 389 + ], + "score": 0.76, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 378, + 262, + 393 + ], + "score": 1.0, + "content": "is parametrized by", + "type": "text" + }, + { + "bbox": [ + 262, + 380, + 268, + 389 + ], + "score": 0.8, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 378, + 356, + 393 + ], + "score": 1.0, + "content": ", which is updated by", + "type": "text" + }, + { + "bbox": [ + 356, + 379, + 440, + 391 + ], + "score": 0.92, + "content": "\\nabla _ { \\boldsymbol { \\theta } } l o g \\pi _ { \\boldsymbol { \\theta } } A _ { \\pi _ { \\boldsymbol { \\theta } } } \\bigl ( S _ { t } , a _ { t } \\bigr )", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 378, + 505, + 393 + ], + "score": 1.0, + "content": ". Our algorithm", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 85, + 389, + 301, + 403 + ], + "spans": [ + { + "bbox": [ + 85, + 392, + 100, + 402 + ], + "score": 1.0, + "content": "248", + "type": "text" + }, + { + "bbox": [ + 105, + 389, + 301, + 403 + ], + "score": 1.0, + "content": "based on REINFORCE is listed in the appendix.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 102, + 416, + 191, + 430 + ], + "lines": [ + { + "bbox": [ + 104, + 414, + 193, + 434 + ], + "spans": [ + { + "bbox": [ + 104, + 414, + 193, + 434 + ], + "score": 1.0, + "content": "4 Experiments", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 98, + 440, + 348, + 452 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 350, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 350, + 454 + ], + "score": 1.0, + "content": "4.1 Experimental Settings and Evaluation Benchmarks", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 104, + 460, + 505, + 527 + ], + "lines": [ + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "score": 1.0, + "content": "In our experiments, the settings of our model are described as follows. All embedding and hidden", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "vectors in our model have a dimension of 256. The model contains three HGIN layers for graph", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "embedding, and an MLP for the score function, as shown in Figure 3 (b) and (c). All MLP networks", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 494, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 505, + 505 + ], + "score": 1.0, + "content": "including those in HGIN and for score contain two hidden layers. The parameters of our model, such", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 504, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 505, + 517 + ], + "score": 1.0, + "content": "as MLP, generally follow the default settings in PyTorch (Paszke et al., 2019) and PyTorch Geometric", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 515, + 348, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 348, + 528 + ], + "score": 1.0, + "content": "(Fey and Lenssen, 2019). More settings are in the appendix.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 99, + 531, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 543 + ], + "score": 1.0, + "content": "Each of our models is trained with one million episodes, each with one scheduling instance. Each", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 543, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 506, + 555 + ], + "score": 1.0, + "content": "instance is generated by following the procedure which is used to generate the TA dataset (Taillard,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 553, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 163, + 567 + ], + "score": 1.0, + "content": "1993). Given", + "type": "text" + }, + { + "bbox": [ + 163, + 554, + 195, + 564 + ], + "score": 0.77, + "content": "( N , M )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 553, + 348, + 567 + ], + "score": 1.0, + "content": ", we use the procedure to generate an", + "type": "text" + }, + { + "bbox": [ + 348, + 554, + 377, + 564 + ], + "score": 0.9, + "content": "n \\times m", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 553, + 506, + 567 + ], + "score": 1.0, + "content": "JSP instance by conforming to", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 563, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 213, + 577 + ], + "score": 1.0, + "content": "the following distribution,", + "type": "text" + }, + { + "bbox": [ + 213, + 564, + 267, + 576 + ], + "score": 0.89, + "content": "n \\sim \\mathcal { U } ( \\bar { 3 } , N )", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 563, + 271, + 577 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 271, + 564, + 324, + 576 + ], + "score": 0.91, + "content": "m \\sim \\mathcal { U } ( 3 , n )", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 563, + 410, + 577 + ], + "score": 1.0, + "content": ", and operation count", + "type": "text" + }, + { + "bbox": [ + 410, + 565, + 443, + 576 + ], + "score": 0.91, + "content": "k _ { j } = m", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 563, + 473, + 577 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 473, + 564, + 505, + 576 + ], + "score": 0.92, + "content": "\\boldsymbol { \\mathcal { U } } ( \\boldsymbol { x } , \\boldsymbol { y } )", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 575, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 417, + 587 + ], + "score": 1.0, + "content": "represents a distribution that uniformly samples an integer in a close interval", + "type": "text" + }, + { + "bbox": [ + 418, + 576, + 439, + 587 + ], + "score": 0.88, + "content": "[ x , y ]", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 575, + 506, + 587 + ], + "score": 1.0, + "content": "at random. The", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "score": 1.0, + "content": "details of designation for machines and processing times refer to (Taillard, 1993) and thus are omitted", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 596, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 506, + 609 + ], + "score": 1.0, + "content": "here. We choose (10,10) for all experiments, since (10,10) generally performs better than the other", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 609, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 620 + ], + "score": 1.0, + "content": "two as described in the appendix. Following the method described in Subsection 3.4, the model is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 618, + 507, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 507, + 632 + ], + "score": 1.0, + "content": "updated from the above randomly generated instances. For testing our models for JSP and FJSP,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 630, + 489, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 630, + 489, + 642 + ], + "score": 1.0, + "content": "seven JSP open benchmarks and two FJSP open benchmarks are used, as listed in the appendix.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 95, + 646, + 504, + 667 + ], + "lines": [ + { + "bbox": [ + 91, + 645, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 91, + 645, + 286, + 660 + ], + "score": 1.0, + "content": "67 The performance for a given policy method", + "type": "text" + }, + { + "bbox": [ + 286, + 649, + 293, + 656 + ], + "score": 0.78, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 645, + 494, + 660 + ], + "score": 1.0, + "content": "on an instance is measured by the makespan gap", + "type": "text" + }, + { + "bbox": [ + 495, + 646, + 504, + 657 + ], + "score": 0.76, + "content": "G", + "type": "inline_equation" + } + ], + "index": 41 + }, + { + "bbox": [ + 91, + 657, + 150, + 669 + ], + "spans": [ + { + "bbox": [ + 91, + 657, + 150, + 669 + ], + "score": 1.0, + "content": "68 defined as", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 665, + 357, + 696 + ], + "lines": [ + { + "bbox": [ + 253, + 665, + 357, + 696 + ], + "spans": [ + { + "bbox": [ + 253, + 665, + 357, + 696 + ], + "score": 0.94, + "content": "G = \\frac { T _ { \\pi } ^ { \\left( m k s p \\right) } - T _ { \\pi * } ^ { \\left( m k s p \\right) } } { T _ { \\pi * } ^ { \\left( m k s p \\right) } }", + "type": "interline_equation", + "image_path": "f381b1f4ca34a456d19d426492d9502ae08f10b6d9c37038fd76c7e426b573a4.jpg" + } + ] + } + ], + "index": 43.5, + "virtual_lines": [ + { + "bbox": [ + 253, + 665, + 357, + 680.5 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 253, + 680.5, + 357, + 696.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 88, + 698, + 506, + 723 + ], + "lines": [ + { + "bbox": [ + 86, + 692, + 510, + 718 + ], + "spans": [ + { + "bbox": [ + 86, + 692, + 132, + 718 + ], + "score": 1.0, + "content": "where 269", + "type": "text" + }, + { + "bbox": [ + 132, + 697, + 166, + 711 + ], + "score": 0.93, + "content": "T _ { \\pi * } ^ { ( m k s p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 692, + 510, + 718 + ], + "score": 1.0, + "content": "is the optimal makespan or the best-effort makespan, from a mathematical optimization", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 84, + 709, + 507, + 724 + ], + "spans": [ + { + "bbox": [ + 84, + 709, + 214, + 724 + ], + "score": 1.0, + "content": "270 tool, OR-Tools, serving as", + "type": "text" + }, + { + "bbox": [ + 227, + 709, + 507, + 724 + ], + "score": 1.0, + "content": ". By the best-effort makespan, we mean the makespan derived with a", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 86, + 72, + 355, + 84 + ], + "lines": [ + { + "bbox": [ + 84, + 72, + 356, + 86 + ], + "spans": [ + { + "bbox": [ + 84, + 72, + 220, + 86 + ], + "score": 1.0, + "content": "229 Similarly, the embedding of", + "type": "text" + }, + { + "bbox": [ + 221, + 73, + 240, + 85 + ], + "score": 0.92, + "content": "O _ { 1 , 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 72, + 266, + 86 + ], + "score": 1.0, + "content": "in the", + "type": "text" + }, + { + "bbox": [ + 267, + 73, + 297, + 85 + ], + "score": 0.92, + "content": "( k + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 72, + 356, + 86 + ], + "score": 1.0, + "content": "-st iteration is:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 84, + 72, + 356, + 86 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 87, + 466, + 106 + ], + "lines": [ + { + "bbox": [ + 144, + 87, + 466, + 106 + ], + "spans": [ + { + "bbox": [ + 144, + 87, + 466, + 106 + ], + "score": 0.91, + "content": "h _ { O _ { 1 , 1 } } ^ { ( k + 1 ) } = M L P _ { O O } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { O O } ^ { ( k + 1 ) } ) h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } ) + M L P _ { M O } ^ { ( k + 1 ) } ( h _ { M _ { 1 } } ^ { ( k ) } )", + "type": "interline_equation", + "image_path": "3a318d233ecc5b46d854239d39f00d2fdff89b88a439536327c31a2b3834eceb.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 144, + 87, + 466, + 106 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "index", + "bbox": [ + 86, + 113, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 86, + 114, + 506, + 126 + ], + "spans": [ + { + "bbox": [ + 86, + 115, + 100, + 126 + ], + "score": 1.0, + "content": "230", + "type": "text" + }, + { + "bbox": [ + 104, + 114, + 506, + 126 + ], + "score": 1.0, + "content": "In our approach, an action includes the two phases, graph embedding phase and action selection phase.", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 120, + 507, + 146 + ], + "spans": [ + { + "bbox": [ + 86, + 128, + 99, + 139 + ], + "score": 1.0, + "content": "231", + "type": "text" + }, + { + "bbox": [ + 100, + 120, + 122, + 146 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 122, + 125, + 140, + 140 + ], + "score": 0.9, + "content": "h _ { \\mathcal { G } } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 120, + 313, + 146 + ], + "score": 1.0, + "content": "denote the whole embedding of the graphs", + "type": "text" + }, + { + "bbox": [ + 313, + 127, + 321, + 138 + ], + "score": 0.83, + "content": "\\mathcal { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 120, + 507, + 146 + ], + "score": 1.0, + "content": ", a summation of the embeddings of all nodes,", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 134, + 510, + 160 + ], + "spans": [ + { + "bbox": [ + 86, + 144, + 100, + 154 + ], + "score": 1.0, + "content": "232", + "type": "text" + }, + { + "bbox": [ + 106, + 140, + 134, + 153 + ], + "score": 0.9, + "content": "h _ { v } ^ { ( k + 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 134, + 510, + 160 + ], + "score": 1.0, + "content": ". In the graph embedding phase, we use an HGIN to encode node and graph embeddings as", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 153, + 438, + 165 + ], + "spans": [ + { + "bbox": [ + 86, + 155, + 100, + 164 + ], + "score": 1.0, + "content": "233", + "type": "text" + }, + { + "bbox": [ + 105, + 153, + 438, + 165 + ], + "score": 1.0, + "content": "described above. An example with three HGIN layers is illustrated in Figure 3 (b).", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + } + ], + "index": 3.5, + "bbox_fs": [ + 86, + 114, + 510, + 165 + ] + }, + { + "type": "text", + "bbox": [ + 97, + 169, + 505, + 239 + ], + "lines": [ + { + "bbox": [ + 105, + 168, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 505, + 182 + ], + "score": 1.0, + "content": "In the action selection phase, we select an action based on a policy, after node and graph embedding", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 179, + 505, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 505, + 192 + ], + "score": 1.0, + "content": "are encoded in the graph embedding phase. The policy is described as follows. First, collect all", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 191, + 506, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 177, + 203 + ], + "score": 1.0, + "content": "ready operations", + "type": "text" + }, + { + "bbox": [ + 177, + 192, + 187, + 201 + ], + "score": 0.8, + "content": "O", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 191, + 310, + 203 + ], + "score": 1.0, + "content": "to be dispatched to machines", + "type": "text" + }, + { + "bbox": [ + 310, + 192, + 322, + 201 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 191, + 405, + 203 + ], + "score": 1.0, + "content": ". Then, for all pairs", + "type": "text" + }, + { + "bbox": [ + 405, + 191, + 436, + 203 + ], + "score": 0.59, + "content": "( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 191, + 506, + 203 + ], + "score": 1.0, + "content": ", feed their node", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 102, + 198, + 510, + 223 + ], + "spans": [ + { + "bbox": [ + 102, + 198, + 157, + 223 + ], + "score": 1.0, + "content": "embeddings", + "type": "text" + }, + { + "bbox": [ + 157, + 202, + 203, + 217 + ], + "score": 0.8, + "content": "( h _ { M } ^ { ( k ) } , h _ { O } ^ { ( k ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 198, + 252, + 223 + ], + "score": 1.0, + "content": "into a MLP", + "type": "text" + }, + { + "bbox": [ + 253, + 204, + 309, + 217 + ], + "score": 0.86, + "content": "S c o r e ( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 198, + 510, + 223 + ], + "score": 1.0, + "content": "to calculate their scores as shown in Figure 3 (c).", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 104, + 214, + 221, + 228 + ], + "score": 1.0, + "content": "The probability of selecting", + "type": "text" + }, + { + "bbox": [ + 222, + 216, + 253, + 227 + ], + "score": 0.87, + "content": "( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "is calculated based on a softmax function of all scores, which", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 226, + 324, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 232, + 239 + ], + "score": 1.0, + "content": "also serves as the model policy", + "type": "text" + }, + { + "bbox": [ + 232, + 229, + 240, + 237 + ], + "score": 0.78, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 226, + 324, + 239 + ], + "score": 1.0, + "content": "for the current state.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8.5, + "bbox_fs": [ + 102, + 168, + 510, + 239 + ] + }, + { + "type": "title", + "bbox": [ + 103, + 250, + 241, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 243, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 243, + 266 + ], + "score": 1.0, + "content": "3.4 Policy-Based RL Training", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "index", + "bbox": [ + 92, + 270, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 91, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 91, + 271, + 505, + 282 + ], + "score": 1.0, + "content": "41 In this paper, we propose to use a policy-based RL training mechanism that follows REINFORCE", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 91, + 285, + 99, + 292 + ], + "score": 1.0, + "content": "42", + "type": "text" + }, + { + "bbox": [ + 105, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "(Sutton and Barto, 2018) to update our model by policy gradient with a normalized advantage", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 293, + 337, + 305 + ], + "spans": [ + { + "bbox": [ + 91, + 296, + 99, + 303 + ], + "score": 1.0, + "content": "43", + "type": "text" + }, + { + "bbox": [ + 105, + 293, + 279, + 305 + ], + "score": 1.0, + "content": "makespan with respect to a baseline policy", + "type": "text" + }, + { + "bbox": [ + 279, + 294, + 290, + 303 + ], + "score": 0.86, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 293, + 337, + 305 + ], + "score": 1.0, + "content": "as follows.", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + } + ], + "index": 14, + "bbox_fs": [ + 91, + 271, + 505, + 305 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 306, + 396, + 339 + ], + "lines": [ + { + "bbox": [ + 215, + 306, + 396, + 339 + ], + "spans": [ + { + "bbox": [ + 215, + 306, + 396, + 339 + ], + "score": 0.94, + "content": "A _ { \\pi } ( S , a ) = { \\frac { T _ { \\pi _ { b } } ^ { ( m k s p ) } ( S , a ) - T _ { \\pi } ^ { ( m k s p ) } ( S , a ) } { T _ { \\pi _ { b } } ^ { ( m k s p ) } ( S , a ) } }", + "type": "interline_equation", + "image_path": "b06d82d1c07484ee7e7eab67883f499d4e532d443dc6370d26bccc60092f3d81.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 215, + 306, + 396, + 322.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 215, + 322.5, + 396, + 339.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "index", + "bbox": [ + 86, + 345, + 505, + 402 + ], + "lines": [ + { + "bbox": [ + 86, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 86, + 348, + 100, + 357 + ], + "score": 1.0, + "content": "244", + "type": "text" + }, + { + "bbox": [ + 105, + 346, + 381, + 358 + ], + "score": 1.0, + "content": "In this paper, we choose a lightweight PDR, MWKR, as baseline", + "type": "text" + }, + { + "bbox": [ + 381, + 348, + 392, + 357 + ], + "score": 0.85, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 346, + 505, + 358 + ], + "score": 1.0, + "content": ", which performed best for", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 357, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 86, + 360, + 100, + 369 + ], + "score": 1.0, + "content": "245", + "type": "text" + }, + { + "bbox": [ + 105, + 357, + 505, + 370 + ], + "score": 1.0, + "content": "makespan among all PDRs reported from the previous work (Zhang et al., 2020). In fact, our", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 85, + 370, + 100, + 380 + ], + "score": 1.0, + "content": "246", + "type": "text" + }, + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "experiment also shows that using MWKR is better than the other PDRs shown in the appendix. The", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 378, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 85, + 380, + 100, + 391 + ], + "score": 1.0, + "content": "247", + "type": "text" + }, + { + "bbox": [ + 105, + 378, + 176, + 393 + ], + "score": 1.0, + "content": "model for policy", + "type": "text" + }, + { + "bbox": [ + 176, + 381, + 183, + 389 + ], + "score": 0.76, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 378, + 262, + 393 + ], + "score": 1.0, + "content": "is parametrized by", + "type": "text" + }, + { + "bbox": [ + 262, + 380, + 268, + 389 + ], + "score": 0.8, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 378, + 356, + 393 + ], + "score": 1.0, + "content": ", which is updated by", + "type": "text" + }, + { + "bbox": [ + 356, + 379, + 440, + 391 + ], + "score": 0.92, + "content": "\\nabla _ { \\boldsymbol { \\theta } } l o g \\pi _ { \\boldsymbol { \\theta } } A _ { \\pi _ { \\boldsymbol { \\theta } } } \\bigl ( S _ { t } , a _ { t } \\bigr )", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 378, + 505, + 393 + ], + "score": 1.0, + "content": ". Our algorithm", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 389, + 301, + 403 + ], + "spans": [ + { + "bbox": [ + 85, + 392, + 100, + 402 + ], + "score": 1.0, + "content": "248", + "type": "text" + }, + { + "bbox": [ + 105, + 389, + 301, + 403 + ], + "score": 1.0, + "content": "based on REINFORCE is listed in the appendix.", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + } + ], + "index": 20, + "bbox_fs": [ + 85, + 346, + 505, + 403 + ] + }, + { + "type": "title", + "bbox": [ + 102, + 416, + 191, + 430 + ], + "lines": [ + { + "bbox": [ + 104, + 414, + 193, + 434 + ], + "spans": [ + { + "bbox": [ + 104, + 414, + 193, + 434 + ], + "score": 1.0, + "content": "4 Experiments", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 98, + 440, + 348, + 452 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 350, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 350, + 454 + ], + "score": 1.0, + "content": "4.1 Experimental Settings and Evaluation Benchmarks", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 104, + 460, + 505, + 527 + ], + "lines": [ + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "score": 1.0, + "content": "In our experiments, the settings of our model are described as follows. All embedding and hidden", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "vectors in our model have a dimension of 256. The model contains three HGIN layers for graph", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "embedding, and an MLP for the score function, as shown in Figure 3 (b) and (c). All MLP networks", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 494, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 505, + 505 + ], + "score": 1.0, + "content": "including those in HGIN and for score contain two hidden layers. The parameters of our model, such", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 504, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 505, + 517 + ], + "score": 1.0, + "content": "as MLP, generally follow the default settings in PyTorch (Paszke et al., 2019) and PyTorch Geometric", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 515, + 348, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 348, + 528 + ], + "score": 1.0, + "content": "(Fey and Lenssen, 2019). More settings are in the appendix.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 460, + 505, + 528 + ] + }, + { + "type": "text", + "bbox": [ + 99, + 531, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 543 + ], + "score": 1.0, + "content": "Each of our models is trained with one million episodes, each with one scheduling instance. Each", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 543, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 506, + 555 + ], + "score": 1.0, + "content": "instance is generated by following the procedure which is used to generate the TA dataset (Taillard,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 553, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 163, + 567 + ], + "score": 1.0, + "content": "1993). Given", + "type": "text" + }, + { + "bbox": [ + 163, + 554, + 195, + 564 + ], + "score": 0.77, + "content": "( N , M )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 553, + 348, + 567 + ], + "score": 1.0, + "content": ", we use the procedure to generate an", + "type": "text" + }, + { + "bbox": [ + 348, + 554, + 377, + 564 + ], + "score": 0.9, + "content": "n \\times m", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 553, + 506, + 567 + ], + "score": 1.0, + "content": "JSP instance by conforming to", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 563, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 213, + 577 + ], + "score": 1.0, + "content": "the following distribution,", + "type": "text" + }, + { + "bbox": [ + 213, + 564, + 267, + 576 + ], + "score": 0.89, + "content": "n \\sim \\mathcal { U } ( \\bar { 3 } , N )", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 563, + 271, + 577 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 271, + 564, + 324, + 576 + ], + "score": 0.91, + "content": "m \\sim \\mathcal { U } ( 3 , n )", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 563, + 410, + 577 + ], + "score": 1.0, + "content": ", and operation count", + "type": "text" + }, + { + "bbox": [ + 410, + 565, + 443, + 576 + ], + "score": 0.91, + "content": "k _ { j } = m", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 563, + 473, + 577 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 473, + 564, + 505, + 576 + ], + "score": 0.92, + "content": "\\boldsymbol { \\mathcal { U } } ( \\boldsymbol { x } , \\boldsymbol { y } )", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 575, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 417, + 587 + ], + "score": 1.0, + "content": "represents a distribution that uniformly samples an integer in a close interval", + "type": "text" + }, + { + "bbox": [ + 418, + 576, + 439, + 587 + ], + "score": 0.88, + "content": "[ x , y ]", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 575, + 506, + 587 + ], + "score": 1.0, + "content": "at random. The", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 599 + ], + "score": 1.0, + "content": "details of designation for machines and processing times refer to (Taillard, 1993) and thus are omitted", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 596, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 506, + 609 + ], + "score": 1.0, + "content": "here. We choose (10,10) for all experiments, since (10,10) generally performs better than the other", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 609, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 620 + ], + "score": 1.0, + "content": "two as described in the appendix. Following the method described in Subsection 3.4, the model is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 618, + 507, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 507, + 632 + ], + "score": 1.0, + "content": "updated from the above randomly generated instances. For testing our models for JSP and FJSP,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 630, + 489, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 630, + 489, + 642 + ], + "score": 1.0, + "content": "seven JSP open benchmarks and two FJSP open benchmarks are used, as listed in the appendix.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 532, + 507, + 642 + ] + }, + { + "type": "index", + "bbox": [ + 95, + 646, + 504, + 667 + ], + "lines": [ + { + "bbox": [ + 91, + 645, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 91, + 645, + 286, + 660 + ], + "score": 1.0, + "content": "67 The performance for a given policy method", + "type": "text" + }, + { + "bbox": [ + 286, + 649, + 293, + 656 + ], + "score": 0.78, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 645, + 494, + 660 + ], + "score": 1.0, + "content": "on an instance is measured by the makespan gap", + "type": "text" + }, + { + "bbox": [ + 495, + 646, + 504, + 657 + ], + "score": 0.76, + "content": "G", + "type": "inline_equation" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 657, + 150, + 669 + ], + "spans": [ + { + "bbox": [ + 91, + 657, + 150, + 669 + ], + "score": 1.0, + "content": "68 defined as", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + } + ], + "index": 41.5, + "bbox_fs": [ + 91, + 645, + 504, + 669 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 665, + 357, + 696 + ], + "lines": [ + { + "bbox": [ + 253, + 665, + 357, + 696 + ], + "spans": [ + { + "bbox": [ + 253, + 665, + 357, + 696 + ], + "score": 0.94, + "content": "G = \\frac { T _ { \\pi } ^ { \\left( m k s p \\right) } - T _ { \\pi * } ^ { \\left( m k s p \\right) } } { T _ { \\pi * } ^ { \\left( m k s p \\right) } }", + "type": "interline_equation", + "image_path": "f381b1f4ca34a456d19d426492d9502ae08f10b6d9c37038fd76c7e426b573a4.jpg" + } + ] + } + ], + "index": 43.5, + "virtual_lines": [ + { + "bbox": [ + 253, + 665, + 357, + 680.5 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 253, + 680.5, + 357, + 696.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 88, + 698, + 506, + 723 + ], + "lines": [ + { + "bbox": [ + 86, + 692, + 510, + 718 + ], + "spans": [ + { + "bbox": [ + 86, + 692, + 132, + 718 + ], + "score": 1.0, + "content": "where 269", + "type": "text" + }, + { + "bbox": [ + 132, + 697, + 166, + 711 + ], + "score": 0.93, + "content": "T _ { \\pi * } ^ { ( m k s p ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 692, + 510, + 718 + ], + "score": 1.0, + "content": "is the optimal makespan or the best-effort makespan, from a mathematical optimization", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 84, + 709, + 507, + 724 + ], + "spans": [ + { + "bbox": [ + 84, + 709, + 214, + 724 + ], + "score": 1.0, + "content": "270 tool, OR-Tools, serving as", + "type": "text" + }, + { + "bbox": [ + 227, + 709, + 507, + 724 + ], + "score": 1.0, + "content": ". By the best-effort makespan, we mean the makespan derived with a", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 84, + 692, + 510, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 87, + 511, + 204 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 197, + 77, + 413, + 87 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 76, + 413, + 89 + ], + "spans": [ + { + "bbox": [ + 198, + 76, + 413, + 89 + ], + "score": 1.0, + "content": "Table 2: Average makespan gaps for TA benchmarks.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 106, + 87, + 511, + 204 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 87, + 511, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 87, + 511, + 204 + ], + "score": 0.983, + "html": "
Size15×1520×1520×2030×1530×2050×1550×20100×20Aug.
RS0.1480.1650.1690.1440.1770.0670.1000.0260.0500.1250.150
RS+op0.1430.1930.1590.1920.2130.1230.126
MWKR0.1910.2050.2580.2330.2350.3280.2180.2170.2770.2390.2280.3520.2510.2490.3440.1680.1730.2410.2060.1790.1760.2550.2390.0830.0910.1440.1350.1950.1970.2750.254
MORSPT
FIFO0.2390.3140.2730.3110.311
L2DParkSchN0.2590.2010.1520.3000.3160.3290.3360.2230.2650.1360.0920.0660.2700.2210.161
0.2490.1940.2920.1720.2460.1900.3190.2370.1590.1380.2120.135
", + "type": "table", + "image_path": "d571f08c06cb4d62d220b992e5665dea662faf647c622b0c5bde8cc35206e59a.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 87, + 511, + 126.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 126.0, + 511, + 165.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 165.0, + 511, + 204.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 87, + 223, + 506, + 267 + ], + "lines": [ + { + "bbox": [ + 86, + 223, + 507, + 237 + ], + "spans": [ + { + "bbox": [ + 86, + 226, + 99, + 235 + ], + "score": 1.0, + "content": "271", + "type": "text" + }, + { + "bbox": [ + 105, + 223, + 507, + 237 + ], + "score": 1.0, + "content": "sufficiently large time limitation, namely half a day with OR-Tools. For comparison in experiments,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 86, + 235, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 86, + 237, + 100, + 246 + ], + "score": 1.0, + "content": "272", + "type": "text" + }, + { + "bbox": [ + 105, + 235, + 506, + 247 + ], + "score": 1.0, + "content": "we use a server with Intel Xeon E5-2683 CPU and a single NVIDIA GeForce GTX 1080 Ti GPU.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 85, + 244, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 85, + 247, + 100, + 258 + ], + "score": 1.0, + "content": "273", + "type": "text" + }, + { + "bbox": [ + 105, + 244, + 505, + 259 + ], + "score": 1.0, + "content": "Our method uses a CPU thread and a GPU to train and evaluate, while OR-Tools uses eight threads", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 85, + 256, + 187, + 269 + ], + "spans": [ + { + "bbox": [ + 85, + 258, + 100, + 268 + ], + "score": 1.0, + "content": "274", + "type": "text" + }, + { + "bbox": [ + 105, + 256, + 187, + 269 + ], + "score": 1.0, + "content": "to find the solution.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5.5 + }, + { + "type": "title", + "bbox": [ + 97, + 280, + 219, + 293 + ], + "lines": [ + { + "bbox": [ + 104, + 279, + 220, + 295 + ], + "spans": [ + { + "bbox": [ + 104, + 279, + 220, + 295 + ], + "score": 1.0, + "content": "4.2 Experiments for JSP", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 105, + 300, + 505, + 356 + ], + "lines": [ + { + "bbox": [ + 106, + 301, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 505, + 313 + ], + "score": 1.0, + "content": "For JSP, we first train a model based on residual scheduling, named RS. For ablation testing, we", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 310, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 216, + 325 + ], + "score": 1.0, + "content": "also train a model, named", + "type": "text" + }, + { + "bbox": [ + 216, + 312, + 246, + 323 + ], + "score": 0.6, + "content": "\\mathrm { R S + o p }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 310, + 506, + 325 + ], + "score": 1.0, + "content": ", by following the same training method but without removing", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "irrelevant operations. When using these models to solve testing instances, action selection is based", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 333, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 313, + 347 + ], + "score": 1.0, + "content": "on the greedy policy that simply chooses the action", + "type": "text" + }, + { + "bbox": [ + 313, + 334, + 344, + 345 + ], + "score": 0.92, + "content": "( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 333, + 506, + 347 + ], + "score": 1.0, + "content": "with the highest score deterministically,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 344, + 311, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 311, + 357 + ], + "score": 1.0, + "content": "obtained from the score network as in Figure 3 (c).", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 87, + 360, + 505, + 426 + ], + "lines": [ + { + "bbox": [ + 86, + 360, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 86, + 362, + 99, + 372 + ], + "score": 1.0, + "content": "281", + "type": "text" + }, + { + "bbox": [ + 104, + 360, + 505, + 374 + ], + "score": 1.0, + "content": "For comparison, we consider the three DRL construction heuristics, respectively developed in (Zhang", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 86, + 372, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 86, + 373, + 100, + 383 + ], + "score": 1.0, + "content": "282", + "type": "text" + }, + { + "bbox": [ + 105, + 372, + 506, + 383 + ], + "score": 1.0, + "content": "et al., 2020) called L2D, (Park et al., 2021b) by Park et al., and (Park et al., 2021a), called ScheduleNet.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 86, + 381, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 86, + 384, + 100, + 394 + ], + "score": 1.0, + "content": "283", + "type": "text" + }, + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "score": 1.0, + "content": "We directly use the performance results of these methods for open benchmarks from their articles.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 86, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 86, + 396, + 99, + 405 + ], + "score": 1.0, + "content": "284", + "type": "text" + }, + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "For simplicity, they are named L2D, Park and SchN respectively in this paper. We also include some", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 86, + 404, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 86, + 406, + 100, + 416 + ], + "score": 1.0, + "content": "285", + "type": "text" + }, + { + "bbox": [ + 105, + 404, + 505, + 417 + ], + "score": 1.0, + "content": "construction heuristics based PDR, such as MWKR, MOR, SPT and FIFO. Besides, to derive the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 86, + 415, + 407, + 427 + ], + "spans": [ + { + "bbox": [ + 86, + 417, + 100, + 427 + ], + "score": 1.0, + "content": "286", + "type": "text" + }, + { + "bbox": [ + 105, + 415, + 315, + 427 + ], + "score": 1.0, + "content": "gaps to the optimum in all cases, OR-Tools serve as", + "type": "text" + }, + { + "bbox": [ + 315, + 417, + 328, + 425 + ], + "score": 0.82, + "content": "\\pi *", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 415, + 407, + 427 + ], + "score": 1.0, + "content": "as described in (6).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 88, + 431, + 505, + 531 + ], + "lines": [ + { + "bbox": [ + 86, + 430, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 86, + 434, + 99, + 443 + ], + "score": 1.0, + "content": "287", + "type": "text" + }, + { + "bbox": [ + 105, + 430, + 506, + 445 + ], + "score": 1.0, + "content": "Now, let us analyze the performances of RS as follows. Table 2 shows the average makespan gaps", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 86, + 442, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 86, + 444, + 99, + 453 + ], + "score": 1.0, + "content": "288", + "type": "text" + }, + { + "bbox": [ + 105, + 442, + 322, + 454 + ], + "score": 1.0, + "content": "for each collection of JSP TA benchmarks with sizes,", + "type": "text" + }, + { + "bbox": [ + 323, + 443, + 349, + 453 + ], + "score": 0.79, + "content": "1 5 { \\times } 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 442, + 352, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 353, + 443, + 380, + 453 + ], + "score": 0.73, + "content": "2 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 442, + 383, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 383, + 443, + 410, + 453 + ], + "score": 0.83, + "content": "2 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 442, + 414, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 414, + 443, + 441, + 453 + ], + "score": 0.81, + "content": "3 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 442, + 444, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 445, + 443, + 472, + 453 + ], + "score": 0.8, + "content": "3 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 442, + 475, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 475, + 443, + 502, + 453 + ], + "score": 0.81, + "content": "5 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 442, + 506, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 86, + 452, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 86, + 456, + 99, + 464 + ], + "score": 1.0, + "content": "289", + "type": "text" + }, + { + "bbox": [ + 106, + 453, + 134, + 464 + ], + "score": 0.84, + "content": "5 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 452, + 151, + 466 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 152, + 453, + 183, + 464 + ], + "score": 0.87, + "content": "1 0 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 452, + 506, + 466 + ], + "score": 1.0, + "content": ", where the best performances (the smallest gaps) are marked in bold. In general,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 86, + 464, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 86, + 466, + 99, + 477 + ], + "score": 1.0, + "content": "290", + "type": "text" + }, + { + "bbox": [ + 105, + 464, + 506, + 478 + ], + "score": 1.0, + "content": "RS performs the best, and generally outperforms the other methods for all collections by large", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 85, + 475, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 85, + 477, + 99, + 487 + ], + "score": 1.0, + "content": "291", + "type": "text" + }, + { + "bbox": [ + 105, + 475, + 328, + 487 + ], + "score": 1.0, + "content": "margins, except for that it has slightly higher gaps than", + "type": "text" + }, + { + "bbox": [ + 328, + 476, + 357, + 487 + ], + "score": 0.7, + "content": "{ \\mathrm { R S + o p } }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 475, + 453, + 487 + ], + "score": 1.0, + "content": "for the two collections,", + "type": "text" + }, + { + "bbox": [ + 453, + 475, + 487, + 486 + ], + "score": 0.9, + "content": "1 5 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 475, + 506, + 487 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 86, + 485, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 86, + 488, + 99, + 497 + ], + "score": 1.0, + "content": "292", + "type": "text" + }, + { + "bbox": [ + 106, + 486, + 141, + 497 + ], + "score": 0.89, + "content": "2 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 485, + 178, + 499 + ], + "score": 1.0, + "content": ". In fact,", + "type": "text" + }, + { + "bbox": [ + 178, + 486, + 208, + 498 + ], + "score": 0.73, + "content": "{ \\mathrm { R S + o p } }", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 485, + 505, + 499 + ], + "score": 1.0, + "content": "also generally outperforms the rest of methods, except for that it is very", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 86, + 497, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 86, + 499, + 100, + 509 + ], + "score": 1.0, + "content": "293", + "type": "text" + }, + { + "bbox": [ + 105, + 497, + 505, + 509 + ], + "score": 1.0, + "content": "close to SchN for two collections. For the other six open benchmarks, ABZ, FT, ORB, YN, SWV", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 85, + 507, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 85, + 510, + 100, + 520 + ], + "score": 1.0, + "content": "294", + "type": "text" + }, + { + "bbox": [ + 105, + 507, + 505, + 520 + ], + "score": 1.0, + "content": "and LA, the performances are similar and thus presented in the appendix. It is concluded that RS", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 86, + 519, + 416, + 532 + ], + "spans": [ + { + "bbox": [ + 86, + 521, + 99, + 530 + ], + "score": 1.0, + "content": "295", + "type": "text" + }, + { + "bbox": [ + 105, + 519, + 416, + 532 + ], + "score": 1.0, + "content": "generally performs better than other construction heuristics by large margins.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 87, + 543, + 225, + 555 + ], + "lines": [ + { + "bbox": [ + 83, + 542, + 226, + 557 + ], + "spans": [ + { + "bbox": [ + 83, + 542, + 226, + 557 + ], + "score": 1.0, + "content": "296 4.3 Experiments for FJSP", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "table", + "bbox": [ + 186, + 585, + 425, + 678 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 184, + 574, + 427, + 585 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 182, + 572, + 428, + 588 + ], + "spans": [ + { + "bbox": [ + 182, + 572, + 428, + 588 + ], + "score": 1.0, + "content": "Table 3: Average makespan gaps for FJSP open benchmarks", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "table_body", + "bbox": [ + 186, + 585, + 425, + 678 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 186, + 585, + 425, + 678 + ], + "spans": [ + { + "bbox": [ + 186, + 585, + 425, + 678 + ], + "score": 0.977, + "html": "
MethodMKLA(rdata)LA(edata)LA(vdata)
RS0.2320.0990.1460.031
RS+op0.2540.1130.1680.029
DRL-G0.2540.1110.1500.040
MWKR MOR0.282 0.296 0.4570.125 0.1470.149 0.1790.051 0.061
", + "type": "table", + "image_path": "c99b51216be96430166143b65903ab65a9561dbec698a97fd465ba2717ab21ec.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 186, + 585, + 425, + 598.2857142857143 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 186, + 598.2857142857143, + 425, + 611.5714285714287 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 186, + 611.5714285714287, + 425, + 624.857142857143 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 186, + 624.857142857143, + 425, + 638.1428571428573 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 186, + 638.1428571428573, + 425, + 651.4285714285717 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 186, + 651.4285714285717, + 425, + 664.714285714286 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 186, + 664.714285714286, + 425, + 678.0000000000003 + ], + "spans": [], + "index": 37 + } + ] + } + ], + "index": 32.0 + }, + { + "type": "text", + "bbox": [ + 88, + 689, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 86, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 100, + 700 + ], + "score": 1.0, + "content": "297", + "type": "text" + }, + { + "bbox": [ + 104, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "For FJSP, we also train a model based on residual scheduling, named RS, and a ablation version,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 86, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 703, + 100, + 711 + ], + "score": 1.0, + "content": "298", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 134, + 712 + ], + "score": 1.0, + "content": "named", + "type": "text" + }, + { + "bbox": [ + 135, + 700, + 163, + 711 + ], + "score": 0.78, + "content": "\\mathrm { R S + o p }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 700, + 505, + 712 + ], + "score": 1.0, + "content": ", without removing irrelevant operations. We compares ours with one DRL construction", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 86, + 711, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 100, + 722 + ], + "score": 1.0, + "content": "299", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 506, + 723 + ], + "score": 1.0, + "content": "heuristics developed by (Song et al., 2023), called DRL-G, and four PDR-based heuristics, MOR,", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 87, + 511, + 204 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 197, + 77, + 413, + 87 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 76, + 413, + 89 + ], + "spans": [ + { + "bbox": [ + 198, + 76, + 413, + 89 + ], + "score": 1.0, + "content": "Table 2: Average makespan gaps for TA benchmarks.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 106, + 87, + 511, + 204 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 87, + 511, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 87, + 511, + 204 + ], + "score": 0.983, + "html": "
Size15×1520×1520×2030×1530×2050×1550×20100×20Aug.
RS0.1480.1650.1690.1440.1770.0670.1000.0260.0500.1250.150
RS+op0.1430.1930.1590.1920.2130.1230.126
MWKR0.1910.2050.2580.2330.2350.3280.2180.2170.2770.2390.2280.3520.2510.2490.3440.1680.1730.2410.2060.1790.1760.2550.2390.0830.0910.1440.1350.1950.1970.2750.254
MORSPT
FIFO0.2390.3140.2730.3110.311
L2DParkSchN0.2590.2010.1520.3000.3160.3290.3360.2230.2650.1360.0920.0660.2700.2210.161
0.2490.1940.2920.1720.2460.1900.3190.2370.1590.1380.2120.135
", + "type": "table", + "image_path": "d571f08c06cb4d62d220b992e5665dea662faf647c622b0c5bde8cc35206e59a.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 87, + 511, + 126.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 126.0, + 511, + 165.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 165.0, + 511, + 204.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "index", + "bbox": [ + 87, + 223, + 506, + 267 + ], + "lines": [ + { + "bbox": [ + 86, + 223, + 507, + 237 + ], + "spans": [ + { + "bbox": [ + 86, + 226, + 99, + 235 + ], + "score": 1.0, + "content": "271", + "type": "text" + }, + { + "bbox": [ + 105, + 223, + 507, + 237 + ], + "score": 1.0, + "content": "sufficiently large time limitation, namely half a day with OR-Tools. For comparison in experiments,", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 235, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 86, + 237, + 100, + 246 + ], + "score": 1.0, + "content": "272", + "type": "text" + }, + { + "bbox": [ + 105, + 235, + 506, + 247 + ], + "score": 1.0, + "content": "we use a server with Intel Xeon E5-2683 CPU and a single NVIDIA GeForce GTX 1080 Ti GPU.", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 244, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 85, + 247, + 100, + 258 + ], + "score": 1.0, + "content": "273", + "type": "text" + }, + { + "bbox": [ + 105, + 244, + 505, + 259 + ], + "score": 1.0, + "content": "Our method uses a CPU thread and a GPU to train and evaluate, while OR-Tools uses eight threads", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 256, + 187, + 269 + ], + "spans": [ + { + "bbox": [ + 85, + 258, + 100, + 268 + ], + "score": 1.0, + "content": "274", + "type": "text" + }, + { + "bbox": [ + 105, + 256, + 187, + 269 + ], + "score": 1.0, + "content": "to find the solution.", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + } + ], + "index": 5.5, + "bbox_fs": [ + 85, + 223, + 507, + 269 + ] + }, + { + "type": "title", + "bbox": [ + 97, + 280, + 219, + 293 + ], + "lines": [ + { + "bbox": [ + 104, + 279, + 220, + 295 + ], + "spans": [ + { + "bbox": [ + 104, + 279, + 220, + 295 + ], + "score": 1.0, + "content": "4.2 Experiments for JSP", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 105, + 300, + 505, + 356 + ], + "lines": [ + { + "bbox": [ + 106, + 301, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 505, + 313 + ], + "score": 1.0, + "content": "For JSP, we first train a model based on residual scheduling, named RS. For ablation testing, we", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 310, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 216, + 325 + ], + "score": 1.0, + "content": "also train a model, named", + "type": "text" + }, + { + "bbox": [ + 216, + 312, + 246, + 323 + ], + "score": 0.6, + "content": "\\mathrm { R S + o p }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 310, + 506, + 325 + ], + "score": 1.0, + "content": ", by following the same training method but without removing", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "irrelevant operations. When using these models to solve testing instances, action selection is based", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 333, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 313, + 347 + ], + "score": 1.0, + "content": "on the greedy policy that simply chooses the action", + "type": "text" + }, + { + "bbox": [ + 313, + 334, + 344, + 345 + ], + "score": 0.92, + "content": "( M , O )", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 333, + 506, + 347 + ], + "score": 1.0, + "content": "with the highest score deterministically,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 344, + 311, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 311, + 357 + ], + "score": 1.0, + "content": "obtained from the score network as in Figure 3 (c).", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 301, + 506, + 357 + ] + }, + { + "type": "index", + "bbox": [ + 87, + 360, + 505, + 426 + ], + "lines": [ + { + "bbox": [ + 86, + 360, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 86, + 362, + 99, + 372 + ], + "score": 1.0, + "content": "281", + "type": "text" + }, + { + "bbox": [ + 104, + 360, + 505, + 374 + ], + "score": 1.0, + "content": "For comparison, we consider the three DRL construction heuristics, respectively developed in (Zhang", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 372, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 86, + 373, + 100, + 383 + ], + "score": 1.0, + "content": "282", + "type": "text" + }, + { + "bbox": [ + 105, + 372, + 506, + 383 + ], + "score": 1.0, + "content": "et al., 2020) called L2D, (Park et al., 2021b) by Park et al., and (Park et al., 2021a), called ScheduleNet.", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 381, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 86, + 384, + 100, + 394 + ], + "score": 1.0, + "content": "283", + "type": "text" + }, + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "score": 1.0, + "content": "We directly use the performance results of these methods for open benchmarks from their articles.", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 86, + 396, + 99, + 405 + ], + "score": 1.0, + "content": "284", + "type": "text" + }, + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "For simplicity, they are named L2D, Park and SchN respectively in this paper. We also include some", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 404, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 86, + 406, + 100, + 416 + ], + "score": 1.0, + "content": "285", + "type": "text" + }, + { + "bbox": [ + 105, + 404, + 505, + 417 + ], + "score": 1.0, + "content": "construction heuristics based PDR, such as MWKR, MOR, SPT and FIFO. Besides, to derive the", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 415, + 407, + 427 + ], + "spans": [ + { + "bbox": [ + 86, + 417, + 100, + 427 + ], + "score": 1.0, + "content": "286", + "type": "text" + }, + { + "bbox": [ + 105, + 415, + 315, + 427 + ], + "score": 1.0, + "content": "gaps to the optimum in all cases, OR-Tools serve as", + "type": "text" + }, + { + "bbox": [ + 315, + 417, + 328, + 425 + ], + "score": 0.82, + "content": "\\pi *", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 415, + 407, + 427 + ], + "score": 1.0, + "content": "as described in (6).", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 430, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 86, + 434, + 99, + 443 + ], + "score": 1.0, + "content": "287", + "type": "text" + }, + { + "bbox": [ + 105, + 430, + 506, + 445 + ], + "score": 1.0, + "content": "Now, let us analyze the performances of RS as follows. Table 2 shows the average makespan gaps", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 442, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 86, + 444, + 99, + 453 + ], + "score": 1.0, + "content": "288", + "type": "text" + }, + { + "bbox": [ + 105, + 442, + 322, + 454 + ], + "score": 1.0, + "content": "for each collection of JSP TA benchmarks with sizes,", + "type": "text" + }, + { + "bbox": [ + 323, + 443, + 349, + 453 + ], + "score": 0.79, + "content": "1 5 { \\times } 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 442, + 352, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 353, + 443, + 380, + 453 + ], + "score": 0.73, + "content": "2 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 442, + 383, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 383, + 443, + 410, + 453 + ], + "score": 0.83, + "content": "2 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 442, + 414, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 414, + 443, + 441, + 453 + ], + "score": 0.81, + "content": "3 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 442, + 444, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 445, + 443, + 472, + 453 + ], + "score": 0.8, + "content": "3 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 442, + 475, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 475, + 443, + 502, + 453 + ], + "score": 0.81, + "content": "5 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 442, + 506, + 454 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 452, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 86, + 456, + 99, + 464 + ], + "score": 1.0, + "content": "289", + "type": "text" + }, + { + "bbox": [ + 106, + 453, + 134, + 464 + ], + "score": 0.84, + "content": "5 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 452, + 151, + 466 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 152, + 453, + 183, + 464 + ], + "score": 0.87, + "content": "1 0 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 452, + 506, + 466 + ], + "score": 1.0, + "content": ", where the best performances (the smallest gaps) are marked in bold. In general,", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 464, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 86, + 466, + 99, + 477 + ], + "score": 1.0, + "content": "290", + "type": "text" + }, + { + "bbox": [ + 105, + 464, + 506, + 478 + ], + "score": 1.0, + "content": "RS performs the best, and generally outperforms the other methods for all collections by large", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 475, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 85, + 477, + 99, + 487 + ], + "score": 1.0, + "content": "291", + "type": "text" + }, + { + "bbox": [ + 105, + 475, + 328, + 487 + ], + "score": 1.0, + "content": "margins, except for that it has slightly higher gaps than", + "type": "text" + }, + { + "bbox": [ + 328, + 476, + 357, + 487 + ], + "score": 0.7, + "content": "{ \\mathrm { R S + o p } }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 475, + 453, + 487 + ], + "score": 1.0, + "content": "for the two collections,", + "type": "text" + }, + { + "bbox": [ + 453, + 475, + 487, + 486 + ], + "score": 0.9, + "content": "1 5 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 475, + 506, + 487 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 485, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 86, + 488, + 99, + 497 + ], + "score": 1.0, + "content": "292", + "type": "text" + }, + { + "bbox": [ + 106, + 486, + 141, + 497 + ], + "score": 0.89, + "content": "2 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 485, + 178, + 499 + ], + "score": 1.0, + "content": ". In fact,", + "type": "text" + }, + { + "bbox": [ + 178, + 486, + 208, + 498 + ], + "score": 0.73, + "content": "{ \\mathrm { R S + o p } }", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 485, + 505, + 499 + ], + "score": 1.0, + "content": "also generally outperforms the rest of methods, except for that it is very", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 497, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 86, + 499, + 100, + 509 + ], + "score": 1.0, + "content": "293", + "type": "text" + }, + { + "bbox": [ + 105, + 497, + 505, + 509 + ], + "score": 1.0, + "content": "close to SchN for two collections. For the other six open benchmarks, ABZ, FT, ORB, YN, SWV", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 507, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 85, + 510, + 100, + 520 + ], + "score": 1.0, + "content": "294", + "type": "text" + }, + { + "bbox": [ + 105, + 507, + 505, + 520 + ], + "score": 1.0, + "content": "and LA, the performances are similar and thus presented in the appendix. It is concluded that RS", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 519, + 416, + 532 + ], + "spans": [ + { + "bbox": [ + 86, + 521, + 99, + 530 + ], + "score": 1.0, + "content": "295", + "type": "text" + }, + { + "bbox": [ + 105, + 519, + 416, + 532 + ], + "score": 1.0, + "content": "generally performs better than other construction heuristics by large margins.", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + } + ], + "index": 16.5, + "bbox_fs": [ + 86, + 360, + 506, + 427 + ] + }, + { + "type": "index", + "bbox": [ + 88, + 431, + 505, + 531 + ], + "lines": [], + "index": 24, + "bbox_fs": [ + 85, + 430, + 506, + 532 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 87, + 543, + 225, + 555 + ], + "lines": [ + { + "bbox": [ + 83, + 542, + 226, + 557 + ], + "spans": [ + { + "bbox": [ + 83, + 542, + 226, + 557 + ], + "score": 1.0, + "content": "296 4.3 Experiments for FJSP", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "table", + "bbox": [ + 186, + 585, + 425, + 678 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 184, + 574, + 427, + 585 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 182, + 572, + 428, + 588 + ], + "spans": [ + { + "bbox": [ + 182, + 572, + 428, + 588 + ], + "score": 1.0, + "content": "Table 3: Average makespan gaps for FJSP open benchmarks", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "table_body", + "bbox": [ + 186, + 585, + 425, + 678 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 186, + 585, + 425, + 678 + ], + "spans": [ + { + "bbox": [ + 186, + 585, + 425, + 678 + ], + "score": 0.977, + "html": "
MethodMKLA(rdata)LA(edata)LA(vdata)
RS0.2320.0990.1460.031
RS+op0.2540.1130.1680.029
DRL-G0.2540.1110.1500.040
MWKR MOR0.282 0.296 0.4570.125 0.1470.149 0.1790.051 0.061
", + "type": "table", + "image_path": "c99b51216be96430166143b65903ab65a9561dbec698a97fd465ba2717ab21ec.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 186, + 585, + 425, + 598.2857142857143 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 186, + 598.2857142857143, + 425, + 611.5714285714287 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 186, + 611.5714285714287, + 425, + 624.857142857143 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 186, + 624.857142857143, + 425, + 638.1428571428573 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 186, + 638.1428571428573, + 425, + 651.4285714285717 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 186, + 651.4285714285717, + 425, + 664.714285714286 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 186, + 664.714285714286, + 425, + 678.0000000000003 + ], + "spans": [], + "index": 37 + } + ] + } + ], + "index": 32.0 + }, + { + "type": "index", + "bbox": [ + 88, + 689, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 86, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 100, + 700 + ], + "score": 1.0, + "content": "297", + "type": "text" + }, + { + "bbox": [ + 104, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "For FJSP, we also train a model based on residual scheduling, named RS, and a ablation version,", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 703, + 100, + 711 + ], + "score": 1.0, + "content": "298", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 134, + 712 + ], + "score": 1.0, + "content": "named", + "type": "text" + }, + { + "bbox": [ + 135, + 700, + 163, + 711 + ], + "score": 0.78, + "content": "\\mathrm { R S + o p }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 700, + 505, + 712 + ], + "score": 1.0, + "content": ", without removing irrelevant operations. We compares ours with one DRL construction", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 711, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 100, + 722 + ], + "score": 1.0, + "content": "299", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 506, + 723 + ], + "score": 1.0, + "content": "heuristics developed by (Song et al., 2023), called DRL-G, and four PDR-based heuristics, MOR,", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 71, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 91, + 76, + 100, + 83 + ], + "score": 1.0, + "content": "00", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 102, + 71, + 506, + 86 + ], + "score": 1.0, + "content": "MWKR, SPT and FIFO. We directly use the performance results of these methods for open datasets", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 83, + 304, + 96 + ], + "spans": [ + { + "bbox": [ + 91, + 87, + 99, + 94 + ], + "score": 1.0, + "content": "01", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 103, + 83, + 304, + 96 + ], + "score": 1.0, + "content": "according to the reports from (Song et al., 2023).", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 99, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 91, + 103, + 100, + 111 + ], + "score": 1.0, + "content": "02", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 102, + 99, + 506, + 113 + ], + "score": 1.0, + "content": "Table 3 shows the average makespan gaps in the four open benchmarks, MK, LA(rdata), LA(edata)", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 111, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 91, + 114, + 99, + 122 + ], + "score": 1.0, + "content": "03", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 111, + 505, + 123 + ], + "score": 1.0, + "content": "and LA(vdata). From the table, RS generally outperforms all the other methods for all benchmarks", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 122, + 456, + 134 + ], + "spans": [ + { + "bbox": [ + 92, + 124, + 100, + 132 + ], + "score": 1.0, + "content": "04", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 122, + 238, + 134 + ], + "score": 1.0, + "content": "by large margins, except for that", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 238, + 122, + 267, + 133 + ], + "score": 0.3, + "content": "{ \\mathsf { R S + o p } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 268, + 122, + 456, + 134 + ], + "score": 1.0, + "content": "is slightly better for the benchmark LA(vdata).", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + } + ], + "index": 39, + "bbox_fs": [ + 86, + 688, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 93, + 72, + 505, + 95 + ], + "lines": [ + { + "bbox": [ + 91, + 71, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 91, + 76, + 100, + 83 + ], + "score": 1.0, + "content": "00", + "type": "text" + }, + { + "bbox": [ + 102, + 71, + 506, + 86 + ], + "score": 1.0, + "content": "MWKR, SPT and FIFO. We directly use the performance results of these methods for open datasets", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 91, + 83, + 304, + 96 + ], + "spans": [ + { + "bbox": [ + 91, + 87, + 99, + 94 + ], + "score": 1.0, + "content": "01", + "type": "text" + }, + { + "bbox": [ + 103, + 83, + 304, + 96 + ], + "score": 1.0, + "content": "according to the reports from (Song et al., 2023).", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 92, + 100, + 507, + 133 + ], + "lines": [ + { + "bbox": [ + 91, + 99, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 91, + 103, + 100, + 111 + ], + "score": 1.0, + "content": "02", + "type": "text" + }, + { + "bbox": [ + 102, + 99, + 506, + 113 + ], + "score": 1.0, + "content": "Table 3 shows the average makespan gaps in the four open benchmarks, MK, LA(rdata), LA(edata)", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 91, + 111, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 91, + 114, + 99, + 122 + ], + "score": 1.0, + "content": "03", + "type": "text" + }, + { + "bbox": [ + 104, + 111, + 505, + 123 + ], + "score": 1.0, + "content": "and LA(vdata). From the table, RS generally outperforms all the other methods for all benchmarks", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 92, + 122, + 456, + 134 + ], + "spans": [ + { + "bbox": [ + 92, + 124, + 100, + 132 + ], + "score": 1.0, + "content": "04", + "type": "text" + }, + { + "bbox": [ + 105, + 122, + 238, + 134 + ], + "score": 1.0, + "content": "by large margins, except for that", + "type": "text" + }, + { + "bbox": [ + 238, + 122, + 267, + 133 + ], + "score": 0.3, + "content": "{ \\mathsf { R S + o p } }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 122, + 456, + 134 + ], + "score": 1.0, + "content": "is slightly better for the benchmark LA(vdata).", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 96, + 151, + 184, + 164 + ], + "lines": [ + { + "bbox": [ + 93, + 150, + 185, + 165 + ], + "spans": [ + { + "bbox": [ + 93, + 150, + 185, + 165 + ], + "score": 1.0, + "content": "05 5 Discussions", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 177, + 506, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "score": 1.0, + "content": "In this paper, we propose a new approach, called residual scheduling, to solving JSP an FJSP problems,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "and the experiments show that our approach reaches SOTA among DRL-based construction heuristics", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 199, + 507, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 507, + 211 + ], + "score": 1.0, + "content": "on the above open JSP and FJSP benchmarks. We further discusses three issues: large instances,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 210, + 287, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 287, + 222 + ], + "score": 1.0, + "content": "computation times and further improvement.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5 + }, + { + "type": "image", + "bbox": [ + 107, + 232, + 505, + 308 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 232, + 505, + 308 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 232, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 107, + 232, + 505, + 308 + ], + "score": 0.961, + "type": "image", + "image_path": "1b83fd2ef9921abe986163a176919527de1940ffe5a722430f846b273b3a20da.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 107, + 232, + 505, + 257.3333333333333 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 107, + 257.3333333333333, + 505, + 282.66666666666663 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 107, + 282.66666666666663, + 505, + 307.99999999999994 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 141, + 315, + 464, + 328 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 142, + 314, + 465, + 330 + ], + "spans": [ + { + "bbox": [ + 142, + 314, + 465, + 330 + ], + "score": 1.0, + "content": "Figure 4: Average makespan gaps of JSP instances with different problem sizes.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + } + ], + "index": 12.0 + }, + { + "type": "text", + "bbox": [ + 85, + 340, + 506, + 461 + ], + "lines": [ + { + "bbox": [ + 85, + 340, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 85, + 342, + 100, + 352 + ], + "score": 1.0, + "content": "310", + "type": "text" + }, + { + "bbox": [ + 104, + 340, + 506, + 354 + ], + "score": 1.0, + "content": "First, from the above experiments particularly for TA benchmark for JSP, we observe that the average", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 85, + 351, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 85, + 353, + 100, + 363 + ], + "score": 1.0, + "content": "311", + "type": "text" + }, + { + "bbox": [ + 105, + 351, + 506, + 363 + ], + "score": 1.0, + "content": "gaps gets smaller as the number of jobs increases, even if we use the same model trained with", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 85, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 85, + 363, + 100, + 374 + ], + "score": 1.0, + "content": "312", + "type": "text" + }, + { + "bbox": [ + 106, + 362, + 183, + 374 + ], + "score": 0.91, + "content": "\\bar { ( \\cal N , M ) } = ( 1 0 , 1 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 362, + 506, + 375 + ], + "score": 1.0, + "content": ". In order to investigate size-agnostics, we further generate 13 collections of JSP", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 85, + 373, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 85, + 375, + 100, + 386 + ], + "score": 1.0, + "content": "313", + "type": "text" + }, + { + "bbox": [ + 104, + 373, + 242, + 386 + ], + "score": 1.0, + "content": "instances of sizes for testing, from", + "type": "text" + }, + { + "bbox": [ + 243, + 374, + 275, + 384 + ], + "score": 0.89, + "content": "1 5 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 373, + 286, + 386 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 286, + 374, + 325, + 384 + ], + "score": 0.91, + "content": "2 0 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 373, + 506, + 386 + ], + "score": 1.0, + "content": ", and generate 10 instances for each collection", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 85, + 384, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 85, + 386, + 100, + 397 + ], + "score": 1.0, + "content": "314", + "type": "text" + }, + { + "bbox": [ + 105, + 384, + 506, + 397 + ], + "score": 1.0, + "content": "by using the procedure above. Figure 4 shows the average gaps for these collections for RS and L2D,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 85, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 85, + 397, + 100, + 407 + ], + "score": 1.0, + "content": "315", + "type": "text" + }, + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "and these collections are listed in the order of sizes in the x-axis. Note that we only show the results", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 85, + 405, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 85, + 408, + 100, + 419 + ], + "score": 1.0, + "content": "316", + "type": "text" + }, + { + "bbox": [ + 105, + 405, + 506, + 419 + ], + "score": 1.0, + "content": "of L2D in addition to our RS, since L2D is the only open-source among the above DRL heuristics.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 86, + 416, + 506, + 431 + ], + "spans": [ + { + "bbox": [ + 86, + 420, + 99, + 428 + ], + "score": 1.0, + "content": "317", + "type": "text" + }, + { + "bbox": [ + 104, + 416, + 506, + 431 + ], + "score": 1.0, + "content": "Interestingly, using RS, the average gaps are nearly zero for the collections with sizes larger than 100", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 85, + 428, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 85, + 430, + 100, + 439 + ], + "score": 1.0, + "content": "318", + "type": "text" + }, + { + "bbox": [ + 106, + 428, + 128, + 439 + ], + "score": 0.84, + "content": "\\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 428, + 165, + 440 + ], + "score": 1.0, + "content": ", namely,", + "type": "text" + }, + { + "bbox": [ + 165, + 428, + 203, + 439 + ], + "score": 0.85, + "content": "1 0 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 428, + 207, + 440 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 207, + 428, + 245, + 439 + ], + "score": 0.86, + "content": "1 0 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 428, + 249, + 440 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 249, + 428, + 288, + 439 + ], + "score": 0.84, + "content": "1 5 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 428, + 291, + 440 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 291, + 428, + 330, + 439 + ], + "score": 0.86, + "content": "2 0 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 428, + 347, + 440 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 348, + 428, + 387, + 438 + ], + "score": 0.9, + "content": "2 0 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 428, + 506, + 440 + ], + "score": 1.0, + "content": ". Among the 50 JSP instances", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 86, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 86, + 441, + 99, + 451 + ], + "score": 1.0, + "content": "319", + "type": "text" + }, + { + "bbox": [ + 104, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "in the five collections, 49 reaches zero gaps. A strong implication is that our RS approach can be", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 86, + 449, + 425, + 463 + ], + "spans": [ + { + "bbox": [ + 86, + 452, + 99, + 461 + ], + "score": 1.0, + "content": "320", + "type": "text" + }, + { + "bbox": [ + 104, + 449, + 425, + 463 + ], + "score": 1.0, + "content": "scaled up for job sizes and even reach the optimal for sufficient large job count.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 86, + 465, + 505, + 521 + ], + "lines": [ + { + "bbox": [ + 86, + 466, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 86, + 468, + 99, + 477 + ], + "score": 1.0, + "content": "321", + "type": "text" + }, + { + "bbox": [ + 106, + 466, + 506, + 478 + ], + "score": 1.0, + "content": "Second, the computation times for RS are relatively small and has low variance like most of other", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 86, + 477, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 86, + 479, + 99, + 488 + ], + "score": 1.0, + "content": "322", + "type": "text" + }, + { + "bbox": [ + 105, + 477, + 354, + 488 + ], + "score": 1.0, + "content": "construction heuristics. Here, we just use the collection of TA", + "type": "text" + }, + { + "bbox": [ + 355, + 477, + 386, + 488 + ], + "score": 0.62, + "content": "1 0 0 \\mathrm { x } 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 477, + 505, + 488 + ], + "score": 1.0, + "content": "for illustration. It takes about", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 86, + 488, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 86, + 490, + 99, + 499 + ], + "score": 1.0, + "content": "323", + "type": "text" + }, + { + "bbox": [ + 105, + 488, + 260, + 500 + ], + "score": 1.0, + "content": "30 seconds on average for both RS and", + "type": "text" + }, + { + "bbox": [ + 260, + 488, + 289, + 499 + ], + "score": 0.67, + "content": "\\mathrm { R S + o p }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 488, + 506, + 500 + ], + "score": 1.0, + "content": ", about 28 for L2D and about 444 for SchN. In contrast,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 85, + 498, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 85, + 500, + 100, + 510 + ], + "score": 1.0, + "content": "324", + "type": "text" + }, + { + "bbox": [ + 105, + 498, + 506, + 510 + ], + "score": 1.0, + "content": "it takes about 4000 seconds with high variance for OR-tools. The times for other collections are listed", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 85, + 510, + 230, + 522 + ], + "spans": [ + { + "bbox": [ + 85, + 511, + 100, + 522 + ], + "score": 1.0, + "content": "325", + "type": "text" + }, + { + "bbox": [ + 105, + 510, + 230, + 522 + ], + "score": 1.0, + "content": "in more detail in the appendix.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27 + }, + { + "type": "table", + "bbox": [ + 182, + 550, + 428, + 610 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 184, + 540, + 427, + 551 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 184, + 537, + 426, + 553 + ], + "spans": [ + { + "bbox": [ + 184, + 537, + 426, + 553 + ], + "score": 1.0, + "content": "Table 4: Average makespan gaps for FJSP open benchmark.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "table_body", + "bbox": [ + 182, + 550, + 428, + 610 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 182, + 550, + 428, + 610 + ], + "spans": [ + { + "bbox": [ + 182, + 550, + 428, + 610 + ], + "score": 0.978, + "html": "
MethodMKLA(rdata)LA(edata)LA(vdata)
RSRS+1000.2320.1540.0990.0470.1460.0790.0310.007
DRL-GDRL+1000.2540.1900.1110.0580.1500.0820.0400.014
", + "type": "table", + "image_path": "89ca0c42d2768ed4edd4d57820fcff3afe69503db3148f90ca2752b714a6be6a.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 182, + 550, + 428, + 570.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 182, + 570.0, + 428, + 590.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 182, + 590.0, + 428, + 610.0 + ], + "spans": [], + "index": 33 + } + ] + } + ], + "index": 31.0 + }, + { + "type": "text", + "bbox": [ + 86, + 623, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 86, + 623, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 86, + 626, + 99, + 635 + ], + "score": 1.0, + "content": "326", + "type": "text" + }, + { + "bbox": [ + 105, + 623, + 505, + 636 + ], + "score": 1.0, + "content": "Third, as proposed by Song et al. (2023), construction heuristics can further improve the gap by", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 86, + 633, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 86, + 637, + 99, + 646 + ], + "score": 1.0, + "content": "327", + "type": "text" + }, + { + "bbox": [ + 104, + 633, + 506, + 648 + ], + "score": 1.0, + "content": "constructing multiple solutions based on the softmax policy, in addition to the greedy policy. They", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 86, + 644, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 86, + 648, + 99, + 657 + ], + "score": 1.0, + "content": "328", + "type": "text" + }, + { + "bbox": [ + 105, + 644, + 337, + 658 + ], + "score": 1.0, + "content": "had a version constructing 100 solutions for FJSP, called", + "type": "text" + }, + { + "bbox": [ + 337, + 646, + 380, + 656 + ], + "score": 0.6, + "content": "\\mathrm { \\ D R L + 1 0 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 644, + 506, + 658 + ], + "score": 1.0, + "content": "in this paper. In this paper, we", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 86, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 86, + 658, + 99, + 667 + ], + "score": 1.0, + "content": "329", + "type": "text" + }, + { + "bbox": [ + 106, + 655, + 506, + 669 + ], + "score": 1.0, + "content": "also implement a RS version for FJSP based on the softmax policy, as described in Subsection 3.3,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 86, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 99, + 678 + ], + "score": 1.0, + "content": "330", + "type": "text" + }, + { + "bbox": [ + 105, + 667, + 233, + 680 + ], + "score": 1.0, + "content": "and then use the version, called", + "type": "text" + }, + { + "bbox": [ + 233, + 667, + 267, + 678 + ], + "score": 0.86, + "content": "{ \\mathrm { R S } } { + } 1 0 0 \\ ", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 667, + 506, + 680 + ], + "score": 1.0, + "content": ", to constructing 100 solutions. In Table 4, the experimental", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 85, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 85, + 680, + 100, + 690 + ], + "score": 1.0, + "content": "331", + "type": "text" + }, + { + "bbox": [ + 106, + 678, + 179, + 690 + ], + "score": 1.0, + "content": "results show that", + "type": "text" + }, + { + "bbox": [ + 180, + 678, + 215, + 689 + ], + "score": 0.85, + "content": "{ \\mathrm { R S } } { + } 1 0 0 \\ ", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 678, + 440, + 690 + ], + "score": 1.0, + "content": "performs the best, much better than RS, DRL-G and", + "type": "text" + }, + { + "bbox": [ + 440, + 678, + 483, + 689 + ], + "score": 0.78, + "content": "\\mathrm { \\ D R L + 1 0 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 678, + 505, + 690 + ], + "score": 1.0, + "content": ". An", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 85, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 85, + 690, + 100, + 701 + ], + "score": 1.0, + "content": "332", + "type": "text" + }, + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "important property for such an improvement is that constructing multiple solutions can be done in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 85, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 85, + 702, + 100, + 712 + ], + "score": 1.0, + "content": "333", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "parallel. That is, for construction heuristics, the solution quality can be improved by adding more", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 86, + 712, + 192, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 99, + 722 + ], + "score": 1.0, + "content": "334", + "type": "text" + }, + { + "bbox": [ + 106, + 712, + 192, + 724 + ], + "score": 1.0, + "content": "computation powers.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 38 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 93, + 72, + 505, + 95 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 91, + 71, + 506, + 96 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 92, + 100, + 507, + 133 + ], + "lines": [], + "index": 3, + "bbox_fs": [ + 91, + 99, + 506, + 134 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 96, + 151, + 184, + 164 + ], + "lines": [ + { + "bbox": [ + 93, + 150, + 185, + 165 + ], + "spans": [ + { + "bbox": [ + 93, + 150, + 185, + 165 + ], + "score": 1.0, + "content": "05 5 Discussions", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 177, + 506, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "score": 1.0, + "content": "In this paper, we propose a new approach, called residual scheduling, to solving JSP an FJSP problems,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "and the experiments show that our approach reaches SOTA among DRL-based construction heuristics", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 199, + 507, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 507, + 211 + ], + "score": 1.0, + "content": "on the above open JSP and FJSP benchmarks. We further discusses three issues: large instances,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 210, + 287, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 287, + 222 + ], + "score": 1.0, + "content": "computation times and further improvement.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5, + "bbox_fs": [ + 104, + 177, + 507, + 222 + ] + }, + { + "type": "image", + "bbox": [ + 107, + 232, + 505, + 308 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 232, + 505, + 308 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 232, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 107, + 232, + 505, + 308 + ], + "score": 0.961, + "type": "image", + "image_path": "1b83fd2ef9921abe986163a176919527de1940ffe5a722430f846b273b3a20da.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 107, + 232, + 505, + 257.3333333333333 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 107, + 257.3333333333333, + 505, + 282.66666666666663 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 107, + 282.66666666666663, + 505, + 307.99999999999994 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 141, + 315, + 464, + 328 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 142, + 314, + 465, + 330 + ], + "spans": [ + { + "bbox": [ + 142, + 314, + 465, + 330 + ], + "score": 1.0, + "content": "Figure 4: Average makespan gaps of JSP instances with different problem sizes.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + } + ], + "index": 12.0 + }, + { + "type": "index", + "bbox": [ + 85, + 340, + 506, + 461 + ], + "lines": [ + { + "bbox": [ + 85, + 340, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 85, + 342, + 100, + 352 + ], + "score": 1.0, + "content": "310", + "type": "text" + }, + { + "bbox": [ + 104, + 340, + 506, + 354 + ], + "score": 1.0, + "content": "First, from the above experiments particularly for TA benchmark for JSP, we observe that the average", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 351, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 85, + 353, + 100, + 363 + ], + "score": 1.0, + "content": "311", + "type": "text" + }, + { + "bbox": [ + 105, + 351, + 506, + 363 + ], + "score": 1.0, + "content": "gaps gets smaller as the number of jobs increases, even if we use the same model trained with", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 85, + 363, + 100, + 374 + ], + "score": 1.0, + "content": "312", + "type": "text" + }, + { + "bbox": [ + 106, + 362, + 183, + 374 + ], + "score": 0.91, + "content": "\\bar { ( \\cal N , M ) } = ( 1 0 , 1 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 362, + 506, + 375 + ], + "score": 1.0, + "content": ". In order to investigate size-agnostics, we further generate 13 collections of JSP", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 373, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 85, + 375, + 100, + 386 + ], + "score": 1.0, + "content": "313", + "type": "text" + }, + { + "bbox": [ + 104, + 373, + 242, + 386 + ], + "score": 1.0, + "content": "instances of sizes for testing, from", + "type": "text" + }, + { + "bbox": [ + 243, + 374, + 275, + 384 + ], + "score": 0.89, + "content": "1 5 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 373, + 286, + 386 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 286, + 374, + 325, + 384 + ], + "score": 0.91, + "content": "2 0 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 373, + 506, + 386 + ], + "score": 1.0, + "content": ", and generate 10 instances for each collection", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 384, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 85, + 386, + 100, + 397 + ], + "score": 1.0, + "content": "314", + "type": "text" + }, + { + "bbox": [ + 105, + 384, + 506, + 397 + ], + "score": 1.0, + "content": "by using the procedure above. Figure 4 shows the average gaps for these collections for RS and L2D,", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 85, + 397, + 100, + 407 + ], + "score": 1.0, + "content": "315", + "type": "text" + }, + { + "bbox": [ + 106, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "and these collections are listed in the order of sizes in the x-axis. Note that we only show the results", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 405, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 85, + 408, + 100, + 419 + ], + "score": 1.0, + "content": "316", + "type": "text" + }, + { + "bbox": [ + 105, + 405, + 506, + 419 + ], + "score": 1.0, + "content": "of L2D in addition to our RS, since L2D is the only open-source among the above DRL heuristics.", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 416, + 506, + 431 + ], + "spans": [ + { + "bbox": [ + 86, + 420, + 99, + 428 + ], + "score": 1.0, + "content": "317", + "type": "text" + }, + { + "bbox": [ + 104, + 416, + 506, + 431 + ], + "score": 1.0, + "content": "Interestingly, using RS, the average gaps are nearly zero for the collections with sizes larger than 100", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 428, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 85, + 430, + 100, + 439 + ], + "score": 1.0, + "content": "318", + "type": "text" + }, + { + "bbox": [ + 106, + 428, + 128, + 439 + ], + "score": 0.84, + "content": "\\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 428, + 165, + 440 + ], + "score": 1.0, + "content": ", namely,", + "type": "text" + }, + { + "bbox": [ + 165, + 428, + 203, + 439 + ], + "score": 0.85, + "content": "1 0 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 428, + 207, + 440 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 207, + 428, + 245, + 439 + ], + "score": 0.86, + "content": "1 0 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 428, + 249, + 440 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 249, + 428, + 288, + 439 + ], + "score": 0.84, + "content": "1 5 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 428, + 291, + 440 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 291, + 428, + 330, + 439 + ], + "score": 0.86, + "content": "2 0 0 \\times 1 5", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 428, + 347, + 440 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 348, + 428, + 387, + 438 + ], + "score": 0.9, + "content": "2 0 0 \\times 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 428, + 506, + 440 + ], + "score": 1.0, + "content": ". Among the 50 JSP instances", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 86, + 441, + 99, + 451 + ], + "score": 1.0, + "content": "319", + "type": "text" + }, + { + "bbox": [ + 104, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "in the five collections, 49 reaches zero gaps. A strong implication is that our RS approach can be", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 449, + 425, + 463 + ], + "spans": [ + { + "bbox": [ + 86, + 452, + 99, + 461 + ], + "score": 1.0, + "content": "320", + "type": "text" + }, + { + "bbox": [ + 104, + 449, + 425, + 463 + ], + "score": 1.0, + "content": "scaled up for job sizes and even reach the optimal for sufficient large job count.", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 466, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 86, + 468, + 99, + 477 + ], + "score": 1.0, + "content": "321", + "type": "text" + }, + { + "bbox": [ + 106, + 466, + 506, + 478 + ], + "score": 1.0, + "content": "Second, the computation times for RS are relatively small and has low variance like most of other", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 477, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 86, + 479, + 99, + 488 + ], + "score": 1.0, + "content": "322", + "type": "text" + }, + { + "bbox": [ + 105, + 477, + 354, + 488 + ], + "score": 1.0, + "content": "construction heuristics. Here, we just use the collection of TA", + "type": "text" + }, + { + "bbox": [ + 355, + 477, + 386, + 488 + ], + "score": 0.62, + "content": "1 0 0 \\mathrm { x } 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 477, + 505, + 488 + ], + "score": 1.0, + "content": "for illustration. It takes about", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 488, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 86, + 490, + 99, + 499 + ], + "score": 1.0, + "content": "323", + "type": "text" + }, + { + "bbox": [ + 105, + 488, + 260, + 500 + ], + "score": 1.0, + "content": "30 seconds on average for both RS and", + "type": "text" + }, + { + "bbox": [ + 260, + 488, + 289, + 499 + ], + "score": 0.67, + "content": "\\mathrm { R S + o p }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 488, + 506, + 500 + ], + "score": 1.0, + "content": ", about 28 for L2D and about 444 for SchN. In contrast,", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 498, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 85, + 500, + 100, + 510 + ], + "score": 1.0, + "content": "324", + "type": "text" + }, + { + "bbox": [ + 105, + 498, + 506, + 510 + ], + "score": 1.0, + "content": "it takes about 4000 seconds with high variance for OR-tools. The times for other collections are listed", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 510, + 230, + 522 + ], + "spans": [ + { + "bbox": [ + 85, + 511, + 100, + 522 + ], + "score": 1.0, + "content": "325", + "type": "text" + }, + { + "bbox": [ + 105, + 510, + 230, + 522 + ], + "score": 1.0, + "content": "in more detail in the appendix.", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 623, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 86, + 626, + 99, + 635 + ], + "score": 1.0, + "content": "326", + "type": "text" + }, + { + "bbox": [ + 105, + 623, + 505, + 636 + ], + "score": 1.0, + "content": "Third, as proposed by Song et al. (2023), construction heuristics can further improve the gap by", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 633, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 86, + 637, + 99, + 646 + ], + "score": 1.0, + "content": "327", + "type": "text" + }, + { + "bbox": [ + 104, + 633, + 506, + 648 + ], + "score": 1.0, + "content": "constructing multiple solutions based on the softmax policy, in addition to the greedy policy. They", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 644, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 86, + 648, + 99, + 657 + ], + "score": 1.0, + "content": "328", + "type": "text" + }, + { + "bbox": [ + 105, + 644, + 337, + 658 + ], + "score": 1.0, + "content": "had a version constructing 100 solutions for FJSP, called", + "type": "text" + }, + { + "bbox": [ + 337, + 646, + 380, + 656 + ], + "score": 0.6, + "content": "\\mathrm { \\ D R L + 1 0 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 644, + 506, + 658 + ], + "score": 1.0, + "content": "in this paper. In this paper, we", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 86, + 658, + 99, + 667 + ], + "score": 1.0, + "content": "329", + "type": "text" + }, + { + "bbox": [ + 106, + 655, + 506, + 669 + ], + "score": 1.0, + "content": "also implement a RS version for FJSP based on the softmax policy, as described in Subsection 3.3,", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 99, + 678 + ], + "score": 1.0, + "content": "330", + "type": "text" + }, + { + "bbox": [ + 105, + 667, + 233, + 680 + ], + "score": 1.0, + "content": "and then use the version, called", + "type": "text" + }, + { + "bbox": [ + 233, + 667, + 267, + 678 + ], + "score": 0.86, + "content": "{ \\mathrm { R S } } { + } 1 0 0 \\ ", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 667, + 506, + 680 + ], + "score": 1.0, + "content": ", to constructing 100 solutions. In Table 4, the experimental", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 85, + 680, + 100, + 690 + ], + "score": 1.0, + "content": "331", + "type": "text" + }, + { + "bbox": [ + 106, + 678, + 179, + 690 + ], + "score": 1.0, + "content": "results show that", + "type": "text" + }, + { + "bbox": [ + 180, + 678, + 215, + 689 + ], + "score": 0.85, + "content": "{ \\mathrm { R S } } { + } 1 0 0 \\ ", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 678, + 440, + 690 + ], + "score": 1.0, + "content": "performs the best, much better than RS, DRL-G and", + "type": "text" + }, + { + "bbox": [ + 440, + 678, + 483, + 689 + ], + "score": 0.78, + "content": "\\mathrm { \\ D R L + 1 0 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 678, + 505, + 690 + ], + "score": 1.0, + "content": ". An", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 85, + 690, + 100, + 701 + ], + "score": 1.0, + "content": "332", + "type": "text" + }, + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "important property for such an improvement is that constructing multiple solutions can be done in", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 85, + 702, + 100, + 712 + ], + "score": 1.0, + "content": "333", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "parallel. That is, for construction heuristics, the solution quality can be improved by adding more", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 712, + 192, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 99, + 722 + ], + "score": 1.0, + "content": "334", + "type": "text" + }, + { + "bbox": [ + 106, + 712, + 192, + 724 + ], + "score": 1.0, + "content": "computation powers.", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + } + ], + "index": 19, + "bbox_fs": [ + 85, + 340, + 506, + 463 + ] + }, + { + "type": "index", + "bbox": [ + 86, + 465, + 505, + 521 + ], + "lines": [], + "index": 27, + "bbox_fs": [ + 85, + 466, + 506, + 522 + ], + "lines_deleted": true + }, + { + "type": "table", + "bbox": [ + 182, + 550, + 428, + 610 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 184, + 540, + 427, + 551 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 184, + 537, + 426, + 553 + ], + "spans": [ + { + "bbox": [ + 184, + 537, + 426, + 553 + ], + "score": 1.0, + "content": "Table 4: Average makespan gaps for FJSP open benchmark.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "table_body", + "bbox": [ + 182, + 550, + 428, + 610 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 182, + 550, + 428, + 610 + ], + "spans": [ + { + "bbox": [ + 182, + 550, + 428, + 610 + ], + "score": 0.978, + "html": "
MethodMKLA(rdata)LA(edata)LA(vdata)
RSRS+1000.2320.1540.0990.0470.1460.0790.0310.007
DRL-GDRL+1000.2540.1900.1110.0580.1500.0820.0400.014
", + "type": "table", + "image_path": "89ca0c42d2768ed4edd4d57820fcff3afe69503db3148f90ca2752b714a6be6a.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 182, + 550, + 428, + 570.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 182, + 570.0, + 428, + 590.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 182, + 590.0, + 428, + 610.0 + ], + "spans": [], + "index": 33 + } + ] + } + ], + "index": 31.0 + }, + { + "type": "index", + "bbox": [ + 86, + 623, + 505, + 723 + ], + "lines": [], + "index": 38, + "bbox_fs": [ + 85, + 623, + 506, + 724 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 99, + 72, + 163, + 84 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 91, + 505, + 124 + ], + "lines": [ + { + "bbox": [ + 106, + 90, + 506, + 104 + ], + "spans": [ + { + "bbox": [ + 106, + 90, + 506, + 104 + ], + "score": 1.0, + "content": "Majid Abdolrazzagh-Nezhad and Salwani Abdullah. 2017. Job Shop Scheduling: Classification, Con-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 101, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 101, + 505, + 116 + ], + "score": 1.0, + "content": "straints and Objective Functions. International Journal of Computer and Information Engineering", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 113, + 210, + 124 + ], + "spans": [ + { + "bbox": [ + 115, + 113, + 210, + 124 + ], + "score": 1.0, + "content": "11, 4 (2017), 429–434.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 101, + 133, + 505, + 156 + ], + "lines": [ + { + "bbox": [ + 105, + 133, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 505, + 146 + ], + "score": 1.0, + "content": "Joseph William Adams, Egon Balas, and Daniel J. Zawack. 1988. The Shifting Bottleneck Procedure", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 144, + 401, + 157 + ], + "spans": [ + { + "bbox": [ + 116, + 144, + 401, + 157 + ], + "score": 1.0, + "content": "for Job Shop Scheduling. Management science 34, 3 (1988), 391–401.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 100, + 165, + 505, + 199 + ], + "lines": [ + { + "bbox": [ + 105, + 164, + 505, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 505, + 179 + ], + "score": 1.0, + "content": "David L. Applegate and William J. Cook. 1991. A Computational Study of the Job-Shop Scheduling", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 505, + 189 + ], + "score": 1.0, + "content": "Problem. INFORMS Journal on Computing 3, 2 (1991), 149–156. https://doi.org/10.1287/", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 187, + 182, + 199 + ], + "spans": [ + { + "bbox": [ + 115, + 187, + 182, + 199 + ], + "score": 1.0, + "content": "ijoc.3.2.149", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 99, + 207, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 506, + 221 + ], + "score": 1.0, + "content": "Peter W. Battaglia, Jessica B. Hamrick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinícius Flores", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 218, + 506, + 231 + ], + "spans": [ + { + "bbox": [ + 115, + 218, + 506, + 231 + ], + "score": 1.0, + "content": "Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 230, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 116, + 230, + 506, + 243 + ], + "score": 1.0, + "content": "Çaglar Gülçehre, H. Francis Song, Andrew J. Ballard, Justin Gilmer, George E. Dahl, Ashish", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 241, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 115, + 241, + 506, + 253 + ], + "score": 1.0, + "content": "Vaswani, Kelsey R. Allen, Charles Nash, Victoria Langston, Chris Dyer, Nicolas Heess, Daan", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 252, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 115, + 252, + 506, + 264 + ], + "score": 1.0, + "content": "Wierstra, Pushmeet Kohli, Matthew M. Botvinick, Oriol Vinyals, Yujia Li, and Razvan Pascanu.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 263, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 115, + 263, + 506, + 275 + ], + "score": 1.0, + "content": "2018. Relational inductive biases, deep learning, and graph networks. CoRR abs/1806.01261", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 274, + 390, + 286 + ], + "spans": [ + { + "bbox": [ + 116, + 274, + 390, + 286 + ], + "score": 1.0, + "content": "(2018). arXiv:1806.01261 http://arxiv.org/abs/1806.01261", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 96, + 294, + 503, + 328 + ], + "lines": [ + { + "bbox": [ + 104, + 293, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 104, + 293, + 505, + 308 + ], + "score": 1.0, + "content": "Dennis Behnke and Martin Josef Geiger. 2012. Test instances for the flexible job shop scheduling", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 306, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 115, + 306, + 505, + 317 + ], + "score": 1.0, + "content": "problem with work centers. Arbeitspapier/Research Paper/Helmut-Schmidt-Universität, Lehrstuhl", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 317, + 383, + 329 + ], + "spans": [ + { + "bbox": [ + 115, + 317, + 383, + 329 + ], + "score": 1.0, + "content": "für Betriebswirtschaftslehre, insbes. Logistik-Management (2012).", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 102, + 337, + 505, + 360 + ], + "lines": [ + { + "bbox": [ + 105, + 335, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 506, + 351 + ], + "score": 1.0, + "content": "Paolo Brandimarte. 1993. Routing and scheduling in a flexible job shop by tabu search. Ann. Oper.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 348, + 413, + 361 + ], + "spans": [ + { + "bbox": [ + 115, + 348, + 413, + 361 + ], + "score": 1.0, + "content": "Res. 41, 3 (1993), 157–183. https://doi.org/10.1007/BF02023073", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 99, + 368, + 505, + 392 + ], + "lines": [ + { + "bbox": [ + 105, + 367, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 506, + 382 + ], + "score": 1.0, + "content": "IBM ILOG Cplex. 2009. V12. 1: User’s Manual for CPLEX. International Business Machines", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 379, + 248, + 392 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 248, + 392 + ], + "score": 1.0, + "content": "Corporation 46, 53 (2009), 157.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 101, + 400, + 504, + 434 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "score": 1.0, + "content": "Mauro Dell’Amico and Marco Trubian. 1993. Applying tabu search to the job-shop scheduling", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 411, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 115, + 411, + 505, + 425 + ], + "score": 1.0, + "content": "problem. Annals of Operations Research 41, 3 (1993), 231–252. https://doi.org/10.1007/", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 423, + 171, + 433 + ], + "spans": [ + { + "bbox": [ + 116, + 423, + 171, + 433 + ], + "score": 1.0, + "content": "BF02023076", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 105, + 443, + 506, + 488 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 506, + 456 + ], + "score": 1.0, + "content": "Alhussein Fawzi, Matej Balog, Aja Huang, Thomas Hubert, Bernardino Romera-Paredes, Moham-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 453, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 116, + 453, + 506, + 467 + ], + "score": 1.0, + "content": "madamin Barekatain, Alexander Novikov, Francisco J R Ruiz, Julian Schrittwieser, Grzegorz", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 115, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "Swirszcz, et al. 2022. Discovering faster matrix multiplication algorithms with reinforcement", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 476, + 288, + 488 + ], + "spans": [ + { + "bbox": [ + 116, + 476, + 288, + 488 + ], + "score": 1.0, + "content": "learning. Nature 610, 7930 (2022), 47–53.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 98, + 497, + 504, + 531 + ], + "lines": [ + { + "bbox": [ + 106, + 497, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 506, + 510 + ], + "score": 1.0, + "content": "Matthias Fey and Jan Eric Lenssen. 2019. Fast Graph Representation Learning with PyTorch", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 508, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 116, + 508, + 506, + 520 + ], + "score": 1.0, + "content": "Geometric, In ICLR Workshop on Representation Learning on Graphs and Manifolds. CoRR", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 518, + 428, + 531 + ], + "spans": [ + { + "bbox": [ + 115, + 518, + 428, + 531 + ], + "score": 1.0, + "content": "abs/1903.02428. arXiv:1903.02428 http://arxiv.org/abs/1903.02428", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 98, + 539, + 506, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 538, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 553 + ], + "score": 1.0, + "content": "M. R. Garey and David S. Johnson. 1979. Computers and Intractability: A Guide to the Theory of", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 551, + 284, + 563 + ], + "spans": [ + { + "bbox": [ + 115, + 551, + 284, + 563 + ], + "score": 1.0, + "content": "NP-Completeness. W. H. Freeman, USA.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 103, + 571, + 504, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 506, + 585 + ], + "score": 1.0, + "content": "Amit Kumar Gupta and Appa Iyer Sivakumar. 2006. Job shop scheduling techniques in semiconductor", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 583, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 116, + 583, + 506, + 595 + ], + "score": 1.0, + "content": "manufacturing. The International Journal of Advanced Manufacturing Technology 27, 11 (2006),", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 592, + 165, + 605 + ], + "spans": [ + { + "bbox": [ + 115, + 592, + 165, + 605 + ], + "score": 1.0, + "content": "1163–1169.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 102, + 614, + 504, + 637 + ], + "lines": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "Reinhard Haupt. 1989. A survey of priority rule-based scheduling. Operations-Research-Spektrum", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 625, + 195, + 637 + ], + "spans": [ + { + "bbox": [ + 115, + 625, + 195, + 637 + ], + "score": 1.0, + "content": "11, 1 (1989), 3–16.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 98, + 645, + 506, + 669 + ], + "lines": [ + { + "bbox": [ + 105, + 645, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 505, + 660 + ], + "score": 1.0, + "content": "Johann Hurink, Bernd Jurisch, and Monika Thole. 1994. Tabu search for the job-shop scheduling", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 657, + 486, + 669 + ], + "spans": [ + { + "bbox": [ + 115, + 657, + 486, + 669 + ], + "score": 1.0, + "content": "problem with multi-purpose machines. Operations-Research-Spektrum 15 (1994), 205–215.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 105, + 678, + 506, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "score": 1.0, + "content": "Sergey Ioffe and Christian Szegedy. 2015. Batch Normalization: Accelerating Deep Network Training", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 689, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 115, + 689, + 506, + 702 + ], + "score": 1.0, + "content": "by Reducing Internal Covariate Shift. In International Conference on Machine Learning (ICML)", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 700, + 507, + 713 + ], + "spans": [ + { + "bbox": [ + 115, + 700, + 507, + 713 + ], + "score": 1.0, + "content": "(JMLR Workshop and Conference Proceedings, Vol. 37), Francis R. Bach and David M. Blei (Eds.).", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 711, + 443, + 724 + ], + "spans": [ + { + "bbox": [ + 115, + 711, + 443, + 724 + ], + "score": 1.0, + "content": "JMLR.org, 448–456. http://proceedings.mlr.press/v37/ioffe15.html", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 85, + 84, + 101, + 737 + ], + "lines": [ + { + "bbox": [ + 83, + 92, + 101, + 105 + ], + "spans": [ + { + "bbox": [ + 83, + 92, + 101, + 105 + ], + "score": 1.0, + "content": "336", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 104, + 100, + 115 + ], + "spans": [ + { + "bbox": [ + 84, + 104, + 100, + 115 + ], + "score": 1.0, + "content": "337", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 114, + 101, + 128 + ], + "spans": [ + { + "bbox": [ + 83, + 114, + 101, + 128 + ], + "score": 1.0, + "content": "338", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 135, + 101, + 148 + ], + "spans": [ + { + "bbox": [ + 83, + 135, + 101, + 148 + ], + "score": 1.0, + "content": "339", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 145, + 101, + 160 + ], + "spans": [ + { + "bbox": [ + 83, + 145, + 101, + 160 + ], + "score": 1.0, + "content": "340", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 166, + 101, + 180 + ], + "spans": [ + { + "bbox": [ + 83, + 166, + 101, + 180 + ], + "score": 1.0, + "content": "341", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 178, + 101, + 190 + ], + "spans": [ + { + "bbox": [ + 84, + 178, + 101, + 190 + ], + "score": 1.0, + "content": "342", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 189, + 101, + 202 + ], + "spans": [ + { + "bbox": [ + 83, + 189, + 101, + 202 + ], + "score": 1.0, + "content": "343", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 210, + 101, + 221 + ], + "spans": [ + { + "bbox": [ + 84, + 210, + 101, + 221 + ], + "score": 1.0, + "content": "344", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 220, + 101, + 234 + ], + "spans": [ + { + "bbox": [ + 83, + 220, + 101, + 234 + ], + "score": 1.0, + "content": "345", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 231, + 101, + 244 + ], + "spans": [ + { + "bbox": [ + 83, + 231, + 101, + 244 + ], + "score": 1.0, + "content": "346", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 242, + 101, + 254 + ], + "spans": [ + { + "bbox": [ + 85, + 242, + 101, + 254 + ], + "score": 1.0, + "content": "347", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 252, + 101, + 266 + ], + "spans": [ + { + "bbox": [ + 83, + 252, + 101, + 266 + ], + "score": 1.0, + "content": "348", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 264, + 101, + 276 + ], + "spans": [ + { + "bbox": [ + 84, + 264, + 101, + 276 + ], + "score": 1.0, + "content": "349", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 275, + 101, + 291 + ], + "spans": [ + { + "bbox": [ + 84, + 275, + 101, + 291 + ], + "score": 1.0, + "content": "350", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 296, + 100, + 308 + ], + "spans": [ + { + "bbox": [ + 84, + 296, + 100, + 308 + ], + "score": 1.0, + "content": "351", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 306, + 101, + 320 + ], + "spans": [ + { + "bbox": [ + 83, + 306, + 101, + 320 + ], + "score": 1.0, + "content": "352", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 317, + 101, + 333 + ], + "spans": [ + { + "bbox": [ + 84, + 317, + 101, + 333 + ], + "score": 1.0, + "content": "353", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 339, + 101, + 351 + ], + "spans": [ + { + "bbox": [ + 85, + 339, + 101, + 351 + ], + "score": 1.0, + "content": "354", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 350, + 101, + 364 + ], + "spans": [ + { + "bbox": [ + 84, + 350, + 101, + 364 + ], + "score": 1.0, + "content": "355", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 371, + 101, + 382 + ], + "spans": [ + { + "bbox": [ + 84, + 371, + 101, + 382 + ], + "score": 1.0, + "content": "356", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 381, + 101, + 397 + ], + "spans": [ + { + "bbox": [ + 84, + 381, + 101, + 397 + ], + "score": 1.0, + "content": "357", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 402, + 101, + 415 + ], + "spans": [ + { + "bbox": [ + 83, + 402, + 101, + 415 + ], + "score": 1.0, + "content": "358", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 413, + 101, + 426 + ], + "spans": [ + { + "bbox": [ + 83, + 413, + 101, + 426 + ], + "score": 1.0, + "content": "359", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 424, + 101, + 440 + ], + "spans": [ + { + "bbox": [ + 84, + 424, + 101, + 440 + ], + "score": 1.0, + "content": "360", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 445, + 100, + 457 + ], + "spans": [ + { + "bbox": [ + 84, + 445, + 100, + 457 + ], + "score": 1.0, + "content": "361", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 455, + 101, + 469 + ], + "spans": [ + { + "bbox": [ + 83, + 455, + 101, + 469 + ], + "score": 1.0, + "content": "362", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 466, + 101, + 480 + ], + "spans": [ + { + "bbox": [ + 83, + 466, + 101, + 480 + ], + "score": 1.0, + "content": "363", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 478, + 101, + 491 + ], + "spans": [ + { + "bbox": [ + 85, + 478, + 101, + 491 + ], + "score": 1.0, + "content": "364", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 498, + 101, + 511 + ], + "spans": [ + { + "bbox": [ + 83, + 498, + 101, + 511 + ], + "score": 1.0, + "content": "365", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 509, + 101, + 523 + ], + "spans": [ + { + "bbox": [ + 83, + 509, + 101, + 523 + ], + "score": 1.0, + "content": "366", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 520, + 101, + 534 + ], + "spans": [ + { + "bbox": [ + 85, + 520, + 101, + 534 + ], + "score": 1.0, + "content": "367", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 541, + 101, + 553 + ], + "spans": [ + { + "bbox": [ + 84, + 541, + 101, + 553 + ], + "score": 1.0, + "content": "368", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 553, + 101, + 567 + ], + "spans": [ + { + "bbox": [ + 85, + 553, + 101, + 567 + ], + "score": 1.0, + "content": "369", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 572, + 101, + 587 + ], + "spans": [ + { + "bbox": [ + 83, + 572, + 101, + 587 + ], + "score": 1.0, + "content": "370", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 585, + 101, + 596 + ], + "spans": [ + { + "bbox": [ + 85, + 585, + 101, + 596 + ], + "score": 1.0, + "content": "371", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 595, + 101, + 609 + ], + "spans": [ + { + "bbox": [ + 84, + 595, + 101, + 609 + ], + "score": 1.0, + "content": "372", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 615, + 101, + 629 + ], + "spans": [ + { + "bbox": [ + 83, + 615, + 101, + 629 + ], + "score": 1.0, + "content": "373", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 627, + 101, + 639 + ], + "spans": [ + { + "bbox": [ + 85, + 627, + 101, + 639 + ], + "score": 1.0, + "content": "374", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 647, + 101, + 660 + ], + "spans": [ + { + "bbox": [ + 83, + 647, + 101, + 660 + ], + "score": 1.0, + "content": "375", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 658, + 101, + 673 + ], + "spans": [ + { + "bbox": [ + 84, + 658, + 101, + 673 + ], + "score": 1.0, + "content": "376", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 678, + 101, + 692 + ], + "spans": [ + { + "bbox": [ + 83, + 678, + 101, + 692 + ], + "score": 1.0, + "content": "377", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 690, + 101, + 704 + ], + "spans": [ + { + "bbox": [ + 83, + 690, + 101, + 704 + ], + "score": 1.0, + "content": "378", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 702, + 101, + 713 + ], + "spans": [ + { + "bbox": [ + 84, + 702, + 101, + 713 + ], + "score": 1.0, + "content": "379", + "type": "text" + } + ] + }, + { + "bbox": [ + 83, + 711, + 101, + 725 + ], + "spans": [ + { + "bbox": [ + 83, + 711, + 101, + 725 + ], + "score": 1.0, + "content": "380", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 99, + 72, + 163, + 84 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 91, + 505, + 124 + ], + "lines": [ + { + "bbox": [ + 106, + 90, + 506, + 104 + ], + "spans": [ + { + "bbox": [ + 106, + 90, + 506, + 104 + ], + "score": 1.0, + "content": "Majid Abdolrazzagh-Nezhad and Salwani Abdullah. 2017. Job Shop Scheduling: Classification, Con-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 101, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 101, + 505, + 116 + ], + "score": 1.0, + "content": "straints and Objective Functions. International Journal of Computer and Information Engineering", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 113, + 210, + 124 + ], + "spans": [ + { + "bbox": [ + 115, + 113, + 210, + 124 + ], + "score": 1.0, + "content": "11, 4 (2017), 429–434.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 106, + 90, + 506, + 124 + ] + }, + { + "type": "text", + "bbox": [ + 101, + 133, + 505, + 156 + ], + "lines": [ + { + "bbox": [ + 105, + 133, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 505, + 146 + ], + "score": 1.0, + "content": "Joseph William Adams, Egon Balas, and Daniel J. Zawack. 1988. The Shifting Bottleneck Procedure", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 144, + 401, + 157 + ], + "spans": [ + { + "bbox": [ + 116, + 144, + 401, + 157 + ], + "score": 1.0, + "content": "for Job Shop Scheduling. Management science 34, 3 (1988), 391–401.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 133, + 505, + 157 + ] + }, + { + "type": "text", + "bbox": [ + 100, + 165, + 505, + 199 + ], + "lines": [ + { + "bbox": [ + 105, + 164, + 505, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 505, + 179 + ], + "score": 1.0, + "content": "David L. Applegate and William J. Cook. 1991. A Computational Study of the Job-Shop Scheduling", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 505, + 189 + ], + "score": 1.0, + "content": "Problem. INFORMS Journal on Computing 3, 2 (1991), 149–156. https://doi.org/10.1287/", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 187, + 182, + 199 + ], + "spans": [ + { + "bbox": [ + 115, + 187, + 182, + 199 + ], + "score": 1.0, + "content": "ijoc.3.2.149", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 164, + 505, + 199 + ] + }, + { + "type": "text", + "bbox": [ + 99, + 207, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 506, + 221 + ], + "score": 1.0, + "content": "Peter W. Battaglia, Jessica B. Hamrick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinícius Flores", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 218, + 506, + 231 + ], + "spans": [ + { + "bbox": [ + 115, + 218, + 506, + 231 + ], + "score": 1.0, + "content": "Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 230, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 116, + 230, + 506, + 243 + ], + "score": 1.0, + "content": "Çaglar Gülçehre, H. Francis Song, Andrew J. Ballard, Justin Gilmer, George E. Dahl, Ashish", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 241, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 115, + 241, + 506, + 253 + ], + "score": 1.0, + "content": "Vaswani, Kelsey R. Allen, Charles Nash, Victoria Langston, Chris Dyer, Nicolas Heess, Daan", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 252, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 115, + 252, + 506, + 264 + ], + "score": 1.0, + "content": "Wierstra, Pushmeet Kohli, Matthew M. Botvinick, Oriol Vinyals, Yujia Li, and Razvan Pascanu.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 263, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 115, + 263, + 506, + 275 + ], + "score": 1.0, + "content": "2018. Relational inductive biases, deep learning, and graph networks. CoRR abs/1806.01261", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 274, + 390, + 286 + ], + "spans": [ + { + "bbox": [ + 116, + 274, + 390, + 286 + ], + "score": 1.0, + "content": "(2018). arXiv:1806.01261 http://arxiv.org/abs/1806.01261", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 207, + 506, + 286 + ] + }, + { + "type": "text", + "bbox": [ + 96, + 294, + 503, + 328 + ], + "lines": [ + { + "bbox": [ + 104, + 293, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 104, + 293, + 505, + 308 + ], + "score": 1.0, + "content": "Dennis Behnke and Martin Josef Geiger. 2012. Test instances for the flexible job shop scheduling", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 306, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 115, + 306, + 505, + 317 + ], + "score": 1.0, + "content": "problem with work centers. Arbeitspapier/Research Paper/Helmut-Schmidt-Universität, Lehrstuhl", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 317, + 383, + 329 + ], + "spans": [ + { + "bbox": [ + 115, + 317, + 383, + 329 + ], + "score": 1.0, + "content": "für Betriebswirtschaftslehre, insbes. Logistik-Management (2012).", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17, + "bbox_fs": [ + 104, + 293, + 505, + 329 + ] + }, + { + "type": "text", + "bbox": [ + 102, + 337, + 505, + 360 + ], + "lines": [ + { + "bbox": [ + 105, + 335, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 506, + 351 + ], + "score": 1.0, + "content": "Paolo Brandimarte. 1993. Routing and scheduling in a flexible job shop by tabu search. Ann. Oper.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 348, + 413, + 361 + ], + "spans": [ + { + "bbox": [ + 115, + 348, + 413, + 361 + ], + "score": 1.0, + "content": "Res. 41, 3 (1993), 157–183. https://doi.org/10.1007/BF02023073", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 335, + 506, + 361 + ] + }, + { + "type": "text", + "bbox": [ + 99, + 368, + 505, + 392 + ], + "lines": [ + { + "bbox": [ + 105, + 367, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 506, + 382 + ], + "score": 1.0, + "content": "IBM ILOG Cplex. 2009. V12. 1: User’s Manual for CPLEX. International Business Machines", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 379, + 248, + 392 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 248, + 392 + ], + "score": 1.0, + "content": "Corporation 46, 53 (2009), 157.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 367, + 506, + 392 + ] + }, + { + "type": "text", + "bbox": [ + 101, + 400, + 504, + 434 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "score": 1.0, + "content": "Mauro Dell’Amico and Marco Trubian. 1993. Applying tabu search to the job-shop scheduling", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 411, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 115, + 411, + 505, + 425 + ], + "score": 1.0, + "content": "problem. Annals of Operations Research 41, 3 (1993), 231–252. https://doi.org/10.1007/", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 423, + 171, + 433 + ], + "spans": [ + { + "bbox": [ + 116, + 423, + 171, + 433 + ], + "score": 1.0, + "content": "BF02023076", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 399, + 505, + 433 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 443, + 506, + 488 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 506, + 456 + ], + "score": 1.0, + "content": "Alhussein Fawzi, Matej Balog, Aja Huang, Thomas Hubert, Bernardino Romera-Paredes, Moham-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 453, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 116, + 453, + 506, + 467 + ], + "score": 1.0, + "content": "madamin Barekatain, Alexander Novikov, Francisco J R Ruiz, Julian Schrittwieser, Grzegorz", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 115, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "Swirszcz, et al. 2022. Discovering faster matrix multiplication algorithms with reinforcement", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 476, + 288, + 488 + ], + "spans": [ + { + "bbox": [ + 116, + 476, + 288, + 488 + ], + "score": 1.0, + "content": "learning. Nature 610, 7930 (2022), 47–53.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 443, + 506, + 488 + ] + }, + { + "type": "text", + "bbox": [ + 98, + 497, + 504, + 531 + ], + "lines": [ + { + "bbox": [ + 106, + 497, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 506, + 510 + ], + "score": 1.0, + "content": "Matthias Fey and Jan Eric Lenssen. 2019. Fast Graph Representation Learning with PyTorch", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 508, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 116, + 508, + 506, + 520 + ], + "score": 1.0, + "content": "Geometric, In ICLR Workshop on Representation Learning on Graphs and Manifolds. CoRR", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 518, + 428, + 531 + ], + "spans": [ + { + "bbox": [ + 115, + 518, + 428, + 531 + ], + "score": 1.0, + "content": "abs/1903.02428. arXiv:1903.02428 http://arxiv.org/abs/1903.02428", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 106, + 497, + 506, + 531 + ] + }, + { + "type": "text", + "bbox": [ + 98, + 539, + 506, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 538, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 553 + ], + "score": 1.0, + "content": "M. R. Garey and David S. Johnson. 1979. Computers and Intractability: A Guide to the Theory of", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 551, + 284, + 563 + ], + "spans": [ + { + "bbox": [ + 115, + 551, + 284, + 563 + ], + "score": 1.0, + "content": "NP-Completeness. W. H. Freeman, USA.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 538, + 506, + 563 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 571, + 504, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 506, + 585 + ], + "score": 1.0, + "content": "Amit Kumar Gupta and Appa Iyer Sivakumar. 2006. Job shop scheduling techniques in semiconductor", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 583, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 116, + 583, + 506, + 595 + ], + "score": 1.0, + "content": "manufacturing. The International Journal of Advanced Manufacturing Technology 27, 11 (2006),", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 592, + 165, + 605 + ], + "spans": [ + { + "bbox": [ + 115, + 592, + 165, + 605 + ], + "score": 1.0, + "content": "1163–1169.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 570, + 506, + 605 + ] + }, + { + "type": "text", + "bbox": [ + 102, + 614, + 504, + 637 + ], + "lines": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "Reinhard Haupt. 1989. A survey of priority rule-based scheduling. Operations-Research-Spektrum", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 625, + 195, + 637 + ], + "spans": [ + { + "bbox": [ + 115, + 625, + 195, + 637 + ], + "score": 1.0, + "content": "11, 1 (1989), 3–16.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 613, + 505, + 637 + ] + }, + { + "type": "text", + "bbox": [ + 98, + 645, + 506, + 669 + ], + "lines": [ + { + "bbox": [ + 105, + 645, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 505, + 660 + ], + "score": 1.0, + "content": "Johann Hurink, Bernd Jurisch, and Monika Thole. 1994. Tabu search for the job-shop scheduling", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 657, + 486, + 669 + ], + "spans": [ + { + "bbox": [ + 115, + 657, + 486, + 669 + ], + "score": 1.0, + "content": "problem with multi-purpose machines. Operations-Research-Spektrum 15 (1994), 205–215.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 645, + 505, + 669 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 678, + 506, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "score": 1.0, + "content": "Sergey Ioffe and Christian Szegedy. 2015. Batch Normalization: Accelerating Deep Network Training", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 689, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 115, + 689, + 506, + 702 + ], + "score": 1.0, + "content": "by Reducing Internal Covariate Shift. In International Conference on Machine Learning (ICML)", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 700, + 507, + 713 + ], + "spans": [ + { + "bbox": [ + 115, + 700, + 507, + 713 + ], + "score": 1.0, + "content": "(JMLR Workshop and Conference Proceedings, Vol. 37), Francis R. Bach and David M. Blei (Eds.).", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 711, + 443, + 724 + ], + "spans": [ + { + "bbox": [ + 115, + 711, + 443, + 724 + ], + "score": 1.0, + "content": "JMLR.org, 448–456. http://proceedings.mlr.press/v37/ioffe15.html", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 676, + 507, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 87, + 72, + 506, + 117 + ], + "lines": [ + { + "bbox": [ + 86, + 71, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 74, + 99, + 84 + ], + "score": 1.0, + "content": "381", + "type": "text" + }, + { + "bbox": [ + 105, + 71, + 506, + 86 + ], + "score": 1.0, + "content": "Klaus Jansen, Monaldo Mastrolilli, and Roberto Solis-Oba. 2000. Approximation Algorithms for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 85, + 82, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 85, + 85, + 100, + 96 + ], + "score": 1.0, + "content": "382", + "type": "text" + }, + { + "bbox": [ + 115, + 82, + 506, + 97 + ], + "score": 1.0, + "content": "Flexible Job Shop Problems. In Latin American Symposium on Theoretical Informatics (Lecture", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 86, + 93, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 86, + 96, + 99, + 106 + ], + "score": 1.0, + "content": "383", + "type": "text" + }, + { + "bbox": [ + 115, + 93, + 506, + 108 + ], + "score": 1.0, + "content": "Notes in Computer Science, Vol. 1776), Gaston H. Gonnet, Daniel Panario, and Alfredo Viola", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 85, + 105, + 397, + 118 + ], + "spans": [ + { + "bbox": [ + 85, + 107, + 100, + 118 + ], + "score": 1.0, + "content": "384", + "type": "text" + }, + { + "bbox": [ + 116, + 105, + 397, + 118 + ], + "score": 1.0, + "content": "(Eds.). Springer, 68–77. https://doi.org/10.1007/10719839_7", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 91, + 126, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 89, + 126, + 507, + 140 + ], + "spans": [ + { + "bbox": [ + 89, + 129, + 100, + 138 + ], + "score": 1.0, + "content": "385", + "type": "text" + }, + { + "bbox": [ + 105, + 126, + 507, + 140 + ], + "score": 1.0, + "content": "BJ Lageweg, JK Lenstra, and AHG Rinnooy Kan. 1977. Job-shop scheduling by implicit enumeration.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 88, + 137, + 297, + 150 + ], + "spans": [ + { + "bbox": [ + 88, + 139, + 101, + 150 + ], + "score": 1.0, + "content": "386", + "type": "text" + }, + { + "bbox": [ + 115, + 137, + 297, + 150 + ], + "score": 1.0, + "content": "Management Science 24, 4 (1977), 441–450.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 100, + 159, + 504, + 192 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "score": 1.0, + "content": "Stephen Lawrence. 1984. Resouce constrained project scheduling: An experimental investigation", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 169, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 115, + 169, + 506, + 182 + ], + "score": 1.0, + "content": "of heuristic scheduling techniques (Supplement). Graduate School of Industrial Administration,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 117, + 180, + 263, + 194 + ], + "spans": [ + { + "bbox": [ + 117, + 180, + 263, + 194 + ], + "score": 1.0, + "content": "Carnegie-Mellon University (1984).", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 93, + 202, + 506, + 236 + ], + "lines": [ + { + "bbox": [ + 92, + 202, + 504, + 215 + ], + "spans": [ + { + "bbox": [ + 92, + 206, + 98, + 212 + ], + "score": 1.0, + "content": "90", + "type": "text" + }, + { + "bbox": [ + 104, + 202, + 504, + 215 + ], + "score": 1.0, + "content": "Zhigang Lian, Bin Jiao, and Xingsheng Gu. 2006. A similar particle swarm optimization algorithm", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 91, + 213, + 507, + 226 + ], + "spans": [ + { + "bbox": [ + 91, + 217, + 99, + 223 + ], + "score": 1.0, + "content": "91", + "type": "text" + }, + { + "bbox": [ + 115, + 213, + 507, + 226 + ], + "score": 1.0, + "content": "for job-shop scheduling to minimize makespan. Appl. Math. Comput. 183, 2 (2006), 1008–1017.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 91, + 224, + 333, + 236 + ], + "spans": [ + { + "bbox": [ + 91, + 227, + 99, + 235 + ], + "score": 1.0, + "content": "92", + "type": "text" + }, + { + "bbox": [ + 115, + 224, + 333, + 236 + ], + "score": 1.0, + "content": "https://doi.org/10.1016/j.amc.2006.05.168", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 92, + 245, + 506, + 279 + ], + "lines": [ + { + "bbox": [ + 90, + 244, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 90, + 248, + 100, + 257 + ], + "score": 1.0, + "content": "93", + "type": "text" + }, + { + "bbox": [ + 105, + 244, + 505, + 259 + ], + "score": 1.0, + "content": "Chun-Cheng Lin, Der-Jiunn Deng, Yen-Ling Chih, and Hsin-Ting Chiu. 2019. Smart Manufacturing", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 256, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 89, + 259, + 100, + 268 + ], + "score": 1.0, + "content": "394", + "type": "text" + }, + { + "bbox": [ + 116, + 256, + 505, + 268 + ], + "score": 1.0, + "content": "Scheduling With Edge Computing Using Multiclass Deep Q Network. IEEE Trans. Ind. Informatics", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 266, + 435, + 280 + ], + "spans": [ + { + "bbox": [ + 88, + 270, + 99, + 279 + ], + "score": 1.0, + "content": "395", + "type": "text" + }, + { + "bbox": [ + 115, + 266, + 435, + 280 + ], + "score": 1.0, + "content": "15, 7 (2019), 4276–4284. https://doi.org/10.1109/TII.2019.2908210", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 93, + 289, + 506, + 322 + ], + "lines": [ + { + "bbox": [ + 92, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 92, + 291, + 99, + 299 + ], + "score": 1.0, + "content": "96", + "type": "text" + }, + { + "bbox": [ + 103, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "Min Liu, Zhi-jiang Sun, Junwei Yan, and Jing-song Kang. 2011. An adaptive annealing genetic", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 91, + 299, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 91, + 303, + 99, + 310 + ], + "score": 1.0, + "content": "97", + "type": "text" + }, + { + "bbox": [ + 115, + 299, + 506, + 313 + ], + "score": 1.0, + "content": "algorithm for the job-shop planning and scheduling problem. Expert Systems with Applications 38,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 91, + 309, + 431, + 324 + ], + "spans": [ + { + "bbox": [ + 91, + 313, + 99, + 321 + ], + "score": 1.0, + "content": "98", + "type": "text" + }, + { + "bbox": [ + 114, + 309, + 431, + 324 + ], + "score": 1.0, + "content": "8 (2011), 9248–9255. https://doi.org/10.1016/j.eswa.2011.01.136", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 93, + 331, + 507, + 365 + ], + "lines": [ + { + "bbox": [ + 91, + 330, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 91, + 335, + 100, + 342 + ], + "score": 1.0, + "content": "99", + "type": "text" + }, + { + "bbox": [ + 102, + 330, + 506, + 345 + ], + "score": 1.0, + "content": "Shu Luo. 2020. Dynamic scheduling for flexible job shop with new job insertions by deep reinforce-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 91, + 341, + 507, + 357 + ], + "spans": [ + { + "bbox": [ + 91, + 346, + 99, + 353 + ], + "score": 1.0, + "content": "00", + "type": "text" + }, + { + "bbox": [ + 114, + 341, + 507, + 357 + ], + "score": 1.0, + "content": "ment learning. Applied Soft Computing 91 (2020), 106208. https://doi.org/10.1016/j.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 91, + 354, + 203, + 365 + ], + "spans": [ + { + "bbox": [ + 91, + 356, + 99, + 365 + ], + "score": 1.0, + "content": "01", + "type": "text" + }, + { + "bbox": [ + 115, + 354, + 203, + 364 + ], + "score": 1.0, + "content": "asoc.2020.106208", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 86, + 375, + 506, + 419 + ], + "lines": [ + { + "bbox": [ + 86, + 375, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 86, + 377, + 99, + 387 + ], + "score": 1.0, + "content": "402", + "type": "text" + }, + { + "bbox": [ + 105, + 375, + 506, + 388 + ], + "score": 1.0, + "content": "Mingqi Lv, Zhaoxiong Hong, Ling Chen, Tieming Chen, Tiantian Zhu, and Shouling Ji. 2021.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 85, + 386, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 85, + 387, + 100, + 398 + ], + "score": 1.0, + "content": "403", + "type": "text" + }, + { + "bbox": [ + 116, + 386, + 506, + 398 + ], + "score": 1.0, + "content": "Temporal Multi-Graph Convolutional Network for Traffic Flow Prediction. IEEE Transactions", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 85, + 397, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 85, + 398, + 100, + 409 + ], + "score": 1.0, + "content": "404", + "type": "text" + }, + { + "bbox": [ + 115, + 397, + 505, + 409 + ], + "score": 1.0, + "content": "on Intelligent Transportation Systems 22, 6 (2021), 3337–3348. https://doi.org/10.1109/", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 85, + 408, + 207, + 420 + ], + "spans": [ + { + "bbox": [ + 85, + 409, + 100, + 420 + ], + "score": 1.0, + "content": "405", + "type": "text" + }, + { + "bbox": [ + 116, + 408, + 207, + 419 + ], + "score": 1.0, + "content": "TITS.2020.2983763", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 86, + 429, + 506, + 484 + ], + "lines": [ + { + "bbox": [ + 86, + 429, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 86, + 431, + 100, + 440 + ], + "score": 1.0, + "content": "406", + "type": "text" + }, + { + "bbox": [ + 105, + 429, + 506, + 441 + ], + "score": 1.0, + "content": "Azalia Mirhoseini, Anna Goldie, Mustafa Yazgan, Joe Wenjie Jiang, Ebrahim M. Songhori, Shen", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 85, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 85, + 442, + 100, + 452 + ], + "score": 1.0, + "content": "407", + "type": "text" + }, + { + "bbox": [ + 115, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "Wang, Young-Joon Lee, Eric Johnson, Omkar Pathak, Azade Nazi, Jiwoo Pak, Andy Tong, Kavya", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 86, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 86, + 453, + 99, + 462 + ], + "score": 1.0, + "content": "408", + "type": "text" + }, + { + "bbox": [ + 116, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "Srinivasa, Will Hang, Emre Tuncer, Quoc V. Le, James Laudon, Richard Ho, Roger Carpenter, and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 85, + 461, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 85, + 463, + 100, + 474 + ], + "score": 1.0, + "content": "409", + "type": "text" + }, + { + "bbox": [ + 114, + 461, + 506, + 475 + ], + "score": 1.0, + "content": "Jeff Dean. 2021. A graph placement methodology for fast chip design. Nature 594, 7862 (2021),", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 85, + 472, + 157, + 485 + ], + "spans": [ + { + "bbox": [ + 85, + 474, + 100, + 485 + ], + "score": 1.0, + "content": "410", + "type": "text" + }, + { + "bbox": [ + 115, + 472, + 157, + 485 + ], + "score": 1.0, + "content": "207–212.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 86, + 494, + 507, + 550 + ], + "lines": [ + { + "bbox": [ + 86, + 495, + 507, + 506 + ], + "spans": [ + { + "bbox": [ + 86, + 496, + 99, + 505 + ], + "score": 1.0, + "content": "411", + "type": "text" + }, + { + "bbox": [ + 106, + 495, + 507, + 506 + ], + "score": 1.0, + "content": "Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Belle-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 85, + 504, + 507, + 518 + ], + "spans": [ + { + "bbox": [ + 85, + 507, + 100, + 517 + ], + "score": 1.0, + "content": "412", + "type": "text" + }, + { + "bbox": [ + 115, + 504, + 507, + 518 + ], + "score": 1.0, + "content": "mare, Alex Graves, Martin A. Riedmiller, Andreas Fidjeland, Georg Ostrovski, Stig Petersen,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 86, + 515, + 507, + 529 + ], + "spans": [ + { + "bbox": [ + 86, + 518, + 99, + 527 + ], + "score": 1.0, + "content": "413", + "type": "text" + }, + { + "bbox": [ + 115, + 515, + 507, + 529 + ], + "score": 1.0, + "content": "Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 85, + 525, + 507, + 541 + ], + "spans": [ + { + "bbox": [ + 85, + 529, + 100, + 539 + ], + "score": 1.0, + "content": "414", + "type": "text" + }, + { + "bbox": [ + 115, + 525, + 507, + 541 + ], + "score": 1.0, + "content": "Shane Legg, and Demis Hassabis. 2015. Human-level control through deep reinforcement learning.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 85, + 538, + 448, + 550 + ], + "spans": [ + { + "bbox": [ + 85, + 540, + 100, + 550 + ], + "score": 1.0, + "content": "415", + "type": "text" + }, + { + "bbox": [ + 115, + 538, + 448, + 550 + ], + "score": 1.0, + "content": "Nature 518, 7540 (2015), 529–533. https://doi.org/10.1038/nature14236", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 87, + 559, + 406, + 571 + ], + "lines": [ + { + "bbox": [ + 84, + 558, + 408, + 572 + ], + "spans": [ + { + "bbox": [ + 84, + 558, + 408, + 572 + ], + "score": 1.0, + "content": "416 J.F. Muth and G.L. Thompson. 1963. Industrial Scheduling. Prentice-Hall.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 87, + 581, + 506, + 624 + ], + "lines": [ + { + "bbox": [ + 86, + 579, + 506, + 594 + ], + "spans": [ + { + "bbox": [ + 86, + 582, + 100, + 592 + ], + "score": 1.0, + "content": "417", + "type": "text" + }, + { + "bbox": [ + 104, + 579, + 506, + 594 + ], + "score": 1.0, + "content": "In-Beom Park, Jaeseok Huh, Joongkyun Kim, and Jonghun Park. 2020. A Reinforcement Learning", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 85, + 590, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 85, + 593, + 100, + 604 + ], + "score": 1.0, + "content": "418", + "type": "text" + }, + { + "bbox": [ + 115, + 590, + 506, + 604 + ], + "score": 1.0, + "content": "Approach to Robust Scheduling of Semiconductor Manufacturing Facilities. IEEE Transactions on", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 85, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 85, + 604, + 100, + 615 + ], + "score": 1.0, + "content": "419", + "type": "text" + }, + { + "bbox": [ + 115, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "Automation Science and Engineering 17, 3 (2020), 1420–1431. https://doi.org/10.1109/", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 86, + 613, + 208, + 625 + ], + "spans": [ + { + "bbox": [ + 86, + 615, + 100, + 625 + ], + "score": 1.0, + "content": "420", + "type": "text" + }, + { + "bbox": [ + 115, + 613, + 208, + 625 + ], + "score": 1.0, + "content": "TASE.2019.2956762", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 90, + 634, + 505, + 668 + ], + "lines": [ + { + "bbox": [ + 89, + 633, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 89, + 637, + 99, + 646 + ], + "score": 1.0, + "content": "421", + "type": "text" + }, + { + "bbox": [ + 104, + 633, + 506, + 647 + ], + "score": 1.0, + "content": "Junyoung Park, Sanjar Bakhtiyar, and Jinkyoo Park. 2021a. ScheduleNet: Learn to solve multi-agent", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 89, + 645, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 89, + 648, + 100, + 657 + ], + "score": 1.0, + "content": "422", + "type": "text" + }, + { + "bbox": [ + 116, + 645, + 505, + 657 + ], + "score": 1.0, + "content": "scheduling problems with reinforcement learning. CoRR abs/2106.03051 (2021). arXiv:2106.03051", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 88, + 657, + 286, + 669 + ], + "spans": [ + { + "bbox": [ + 88, + 659, + 100, + 669 + ], + "score": 1.0, + "content": "423", + "type": "text" + }, + { + "bbox": [ + 116, + 657, + 286, + 669 + ], + "score": 1.0, + "content": "https://arxiv.org/abs/2106.03051", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 87, + 678, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 86, + 678, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 100, + 690 + ], + "score": 1.0, + "content": "424", + "type": "text" + }, + { + "bbox": [ + 105, + 678, + 506, + 690 + ], + "score": 1.0, + "content": "Junyoung Park, Jaehyeong Chun, Sang Hun Kim, Youngkook Kim, and Jinkyoo Park. 2021b.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 85, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 85, + 690, + 100, + 701 + ], + "score": 1.0, + "content": "425", + "type": "text" + }, + { + "bbox": [ + 115, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Learning to schedule job-shop problems: representation and policy learning using graph neural", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 85, + 700, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 85, + 701, + 100, + 712 + ], + "score": 1.0, + "content": "426", + "type": "text" + }, + { + "bbox": [ + 116, + 700, + 506, + 712 + ], + "score": 1.0, + "content": "network and reinforcement learning. International Journal of Production Research 59, 11 (2021),", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 85, + 711, + 408, + 723 + ], + "spans": [ + { + "bbox": [ + 85, + 712, + 100, + 723 + ], + "score": 1.0, + "content": "427", + "type": "text" + }, + { + "bbox": [ + 116, + 711, + 408, + 723 + ], + "score": 1.0, + "content": "3360–3377. https://doi.org/10.1080/00207543.2020.1870013", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 310, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 87, + 72, + 506, + 117 + ], + "lines": [ + { + "bbox": [ + 86, + 71, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 74, + 99, + 84 + ], + "score": 1.0, + "content": "381", + "type": "text" + }, + { + "bbox": [ + 105, + 71, + 506, + 86 + ], + "score": 1.0, + "content": "Klaus Jansen, Monaldo Mastrolilli, and Roberto Solis-Oba. 2000. Approximation Algorithms for", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 82, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 85, + 85, + 100, + 96 + ], + "score": 1.0, + "content": "382", + "type": "text" + }, + { + "bbox": [ + 115, + 82, + 506, + 97 + ], + "score": 1.0, + "content": "Flexible Job Shop Problems. In Latin American Symposium on Theoretical Informatics (Lecture", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 93, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 86, + 96, + 99, + 106 + ], + "score": 1.0, + "content": "383", + "type": "text" + }, + { + "bbox": [ + 115, + 93, + 506, + 108 + ], + "score": 1.0, + "content": "Notes in Computer Science, Vol. 1776), Gaston H. Gonnet, Daniel Panario, and Alfredo Viola", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 105, + 397, + 118 + ], + "spans": [ + { + "bbox": [ + 85, + 107, + 100, + 118 + ], + "score": 1.0, + "content": "384", + "type": "text" + }, + { + "bbox": [ + 116, + 105, + 397, + 118 + ], + "score": 1.0, + "content": "(Eds.). Springer, 68–77. https://doi.org/10.1007/10719839_7", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 126, + 507, + 140 + ], + "spans": [ + { + "bbox": [ + 89, + 129, + 100, + 138 + ], + "score": 1.0, + "content": "385", + "type": "text" + }, + { + "bbox": [ + 105, + 126, + 507, + 140 + ], + "score": 1.0, + "content": "BJ Lageweg, JK Lenstra, and AHG Rinnooy Kan. 1977. Job-shop scheduling by implicit enumeration.", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 137, + 297, + 150 + ], + "spans": [ + { + "bbox": [ + 88, + 139, + 101, + 150 + ], + "score": 1.0, + "content": "386", + "type": "text" + }, + { + "bbox": [ + 115, + 137, + 297, + 150 + ], + "score": 1.0, + "content": "Management Science 24, 4 (1977), 441–450.", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + } + ], + "index": 1.5, + "bbox_fs": [ + 85, + 71, + 506, + 118 + ] + }, + { + "type": "index", + "bbox": [ + 91, + 126, + 505, + 149 + ], + "lines": [], + "index": 4.5, + "bbox_fs": [ + 88, + 126, + 507, + 150 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 100, + 159, + 504, + 192 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "score": 1.0, + "content": "Stephen Lawrence. 1984. Resouce constrained project scheduling: An experimental investigation", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 169, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 115, + 169, + 506, + 182 + ], + "score": 1.0, + "content": "of heuristic scheduling techniques (Supplement). Graduate School of Industrial Administration,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 117, + 180, + 263, + 194 + ], + "spans": [ + { + "bbox": [ + 117, + 180, + 263, + 194 + ], + "score": 1.0, + "content": "Carnegie-Mellon University (1984).", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 158, + 506, + 194 + ] + }, + { + "type": "index", + "bbox": [ + 93, + 202, + 506, + 236 + ], + "lines": [ + { + "bbox": [ + 92, + 202, + 504, + 215 + ], + "spans": [ + { + "bbox": [ + 92, + 206, + 98, + 212 + ], + "score": 1.0, + "content": "90", + "type": "text" + }, + { + "bbox": [ + 104, + 202, + 504, + 215 + ], + "score": 1.0, + "content": "Zhigang Lian, Bin Jiao, and Xingsheng Gu. 2006. A similar particle swarm optimization algorithm", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 213, + 507, + 226 + ], + "spans": [ + { + "bbox": [ + 91, + 217, + 99, + 223 + ], + "score": 1.0, + "content": "91", + "type": "text" + }, + { + "bbox": [ + 115, + 213, + 507, + 226 + ], + "score": 1.0, + "content": "for job-shop scheduling to minimize makespan. Appl. Math. Comput. 183, 2 (2006), 1008–1017.", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 224, + 333, + 236 + ], + "spans": [ + { + "bbox": [ + 91, + 227, + 99, + 235 + ], + "score": 1.0, + "content": "92", + "type": "text" + }, + { + "bbox": [ + 115, + 224, + 333, + 236 + ], + "score": 1.0, + "content": "https://doi.org/10.1016/j.amc.2006.05.168", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 244, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 90, + 248, + 100, + 257 + ], + "score": 1.0, + "content": "93", + "type": "text" + }, + { + "bbox": [ + 105, + 244, + 505, + 259 + ], + "score": 1.0, + "content": "Chun-Cheng Lin, Der-Jiunn Deng, Yen-Ling Chih, and Hsin-Ting Chiu. 2019. Smart Manufacturing", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 256, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 89, + 259, + 100, + 268 + ], + "score": 1.0, + "content": "394", + "type": "text" + }, + { + "bbox": [ + 116, + 256, + 505, + 268 + ], + "score": 1.0, + "content": "Scheduling With Edge Computing Using Multiclass Deep Q Network. IEEE Trans. Ind. Informatics", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 266, + 435, + 280 + ], + "spans": [ + { + "bbox": [ + 88, + 270, + 99, + 279 + ], + "score": 1.0, + "content": "395", + "type": "text" + }, + { + "bbox": [ + 115, + 266, + 435, + 280 + ], + "score": 1.0, + "content": "15, 7 (2019), 4276–4284. https://doi.org/10.1109/TII.2019.2908210", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 92, + 291, + 99, + 299 + ], + "score": 1.0, + "content": "96", + "type": "text" + }, + { + "bbox": [ + 103, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "Min Liu, Zhi-jiang Sun, Junwei Yan, and Jing-song Kang. 2011. An adaptive annealing genetic", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 299, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 91, + 303, + 99, + 310 + ], + "score": 1.0, + "content": "97", + "type": "text" + }, + { + "bbox": [ + 115, + 299, + 506, + 313 + ], + "score": 1.0, + "content": "algorithm for the job-shop planning and scheduling problem. Expert Systems with Applications 38,", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 309, + 431, + 324 + ], + "spans": [ + { + "bbox": [ + 91, + 313, + 99, + 321 + ], + "score": 1.0, + "content": "98", + "type": "text" + }, + { + "bbox": [ + 114, + 309, + 431, + 324 + ], + "score": 1.0, + "content": "8 (2011), 9248–9255. https://doi.org/10.1016/j.eswa.2011.01.136", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 330, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 91, + 335, + 100, + 342 + ], + "score": 1.0, + "content": "99", + "type": "text" + }, + { + "bbox": [ + 102, + 330, + 506, + 345 + ], + "score": 1.0, + "content": "Shu Luo. 2020. Dynamic scheduling for flexible job shop with new job insertions by deep reinforce-", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 341, + 507, + 357 + ], + "spans": [ + { + "bbox": [ + 91, + 346, + 99, + 353 + ], + "score": 1.0, + "content": "00", + "type": "text" + }, + { + "bbox": [ + 114, + 341, + 507, + 357 + ], + "score": 1.0, + "content": "ment learning. Applied Soft Computing 91 (2020), 106208. https://doi.org/10.1016/j.", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 354, + 203, + 365 + ], + "spans": [ + { + "bbox": [ + 91, + 356, + 99, + 365 + ], + "score": 1.0, + "content": "01", + "type": "text" + }, + { + "bbox": [ + 115, + 354, + 203, + 364 + ], + "score": 1.0, + "content": "asoc.2020.106208", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 375, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 86, + 377, + 99, + 387 + ], + "score": 1.0, + "content": "402", + "type": "text" + }, + { + "bbox": [ + 105, + 375, + 506, + 388 + ], + "score": 1.0, + "content": "Mingqi Lv, Zhaoxiong Hong, Ling Chen, Tieming Chen, Tiantian Zhu, and Shouling Ji. 2021.", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 386, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 85, + 387, + 100, + 398 + ], + "score": 1.0, + "content": "403", + "type": "text" + }, + { + "bbox": [ + 116, + 386, + 506, + 398 + ], + "score": 1.0, + "content": "Temporal Multi-Graph Convolutional Network for Traffic Flow Prediction. IEEE Transactions", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 397, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 85, + 398, + 100, + 409 + ], + "score": 1.0, + "content": "404", + "type": "text" + }, + { + "bbox": [ + 115, + 397, + 505, + 409 + ], + "score": 1.0, + "content": "on Intelligent Transportation Systems 22, 6 (2021), 3337–3348. https://doi.org/10.1109/", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 408, + 207, + 420 + ], + "spans": [ + { + "bbox": [ + 85, + 409, + 100, + 420 + ], + "score": 1.0, + "content": "405", + "type": "text" + }, + { + "bbox": [ + 116, + 408, + 207, + 419 + ], + "score": 1.0, + "content": "TITS.2020.2983763", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 429, + 506, + 441 + ], + "spans": [ + { + "bbox": [ + 86, + 431, + 100, + 440 + ], + "score": 1.0, + "content": "406", + "type": "text" + }, + { + "bbox": [ + 105, + 429, + 506, + 441 + ], + "score": 1.0, + "content": "Azalia Mirhoseini, Anna Goldie, Mustafa Yazgan, Joe Wenjie Jiang, Ebrahim M. Songhori, Shen", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 85, + 442, + 100, + 452 + ], + "score": 1.0, + "content": "407", + "type": "text" + }, + { + "bbox": [ + 115, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "Wang, Young-Joon Lee, Eric Johnson, Omkar Pathak, Azade Nazi, Jiwoo Pak, Andy Tong, Kavya", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 86, + 453, + 99, + 462 + ], + "score": 1.0, + "content": "408", + "type": "text" + }, + { + "bbox": [ + 116, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "Srinivasa, Will Hang, Emre Tuncer, Quoc V. Le, James Laudon, Richard Ho, Roger Carpenter, and", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 461, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 85, + 463, + 100, + 474 + ], + "score": 1.0, + "content": "409", + "type": "text" + }, + { + "bbox": [ + 114, + 461, + 506, + 475 + ], + "score": 1.0, + "content": "Jeff Dean. 2021. A graph placement methodology for fast chip design. Nature 594, 7862 (2021),", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 472, + 157, + 485 + ], + "spans": [ + { + "bbox": [ + 85, + 474, + 100, + 485 + ], + "score": 1.0, + "content": "410", + "type": "text" + }, + { + "bbox": [ + 115, + 472, + 157, + 485 + ], + "score": 1.0, + "content": "207–212.", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 495, + 507, + 506 + ], + "spans": [ + { + "bbox": [ + 86, + 496, + 99, + 505 + ], + "score": 1.0, + "content": "411", + "type": "text" + }, + { + "bbox": [ + 106, + 495, + 507, + 506 + ], + "score": 1.0, + "content": "Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Belle-", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 504, + 507, + 518 + ], + "spans": [ + { + "bbox": [ + 85, + 507, + 100, + 517 + ], + "score": 1.0, + "content": "412", + "type": "text" + }, + { + "bbox": [ + 115, + 504, + 507, + 518 + ], + "score": 1.0, + "content": "mare, Alex Graves, Martin A. Riedmiller, Andreas Fidjeland, Georg Ostrovski, Stig Petersen,", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 515, + 507, + 529 + ], + "spans": [ + { + "bbox": [ + 86, + 518, + 99, + 527 + ], + "score": 1.0, + "content": "413", + "type": "text" + }, + { + "bbox": [ + 115, + 515, + 507, + 529 + ], + "score": 1.0, + "content": "Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra,", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 525, + 507, + 541 + ], + "spans": [ + { + "bbox": [ + 85, + 529, + 100, + 539 + ], + "score": 1.0, + "content": "414", + "type": "text" + }, + { + "bbox": [ + 115, + 525, + 507, + 541 + ], + "score": 1.0, + "content": "Shane Legg, and Demis Hassabis. 2015. Human-level control through deep reinforcement learning.", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 538, + 448, + 550 + ], + "spans": [ + { + "bbox": [ + 85, + 540, + 100, + 550 + ], + "score": 1.0, + "content": "415", + "type": "text" + }, + { + "bbox": [ + 115, + 538, + 448, + 550 + ], + "score": 1.0, + "content": "Nature 518, 7540 (2015), 529–533. https://doi.org/10.1038/nature14236", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + } + ], + "index": 10, + "bbox_fs": [ + 91, + 202, + 507, + 236 + ] + }, + { + "type": "index", + "bbox": [ + 92, + 245, + 506, + 279 + ], + "lines": [], + "index": 13, + "bbox_fs": [ + 88, + 244, + 505, + 280 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 93, + 289, + 506, + 322 + ], + "lines": [], + "index": 16, + "bbox_fs": [ + 91, + 288, + 506, + 324 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 93, + 331, + 507, + 365 + ], + "lines": [], + "index": 19, + "bbox_fs": [ + 91, + 330, + 507, + 365 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 86, + 375, + 506, + 419 + ], + "lines": [], + "index": 22.5, + "bbox_fs": [ + 85, + 375, + 506, + 420 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 86, + 429, + 506, + 484 + ], + "lines": [], + "index": 27, + "bbox_fs": [ + 85, + 429, + 506, + 485 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 86, + 494, + 507, + 550 + ], + "lines": [], + "index": 32, + "bbox_fs": [ + 85, + 495, + 507, + 550 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 87, + 559, + 406, + 571 + ], + "lines": [ + { + "bbox": [ + 84, + 558, + 408, + 572 + ], + "spans": [ + { + "bbox": [ + 84, + 558, + 408, + 572 + ], + "score": 1.0, + "content": "416 J.F. Muth and G.L. Thompson. 1963. Industrial Scheduling. Prentice-Hall.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 84, + 558, + 408, + 572 + ] + }, + { + "type": "index", + "bbox": [ + 87, + 581, + 506, + 624 + ], + "lines": [ + { + "bbox": [ + 86, + 579, + 506, + 594 + ], + "spans": [ + { + "bbox": [ + 86, + 582, + 100, + 592 + ], + "score": 1.0, + "content": "417", + "type": "text" + }, + { + "bbox": [ + 104, + 579, + 506, + 594 + ], + "score": 1.0, + "content": "In-Beom Park, Jaeseok Huh, Joongkyun Kim, and Jonghun Park. 2020. A Reinforcement Learning", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 590, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 85, + 593, + 100, + 604 + ], + "score": 1.0, + "content": "418", + "type": "text" + }, + { + "bbox": [ + 115, + 590, + 506, + 604 + ], + "score": 1.0, + "content": "Approach to Robust Scheduling of Semiconductor Manufacturing Facilities. IEEE Transactions on", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 85, + 604, + 100, + 615 + ], + "score": 1.0, + "content": "419", + "type": "text" + }, + { + "bbox": [ + 115, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "Automation Science and Engineering 17, 3 (2020), 1420–1431. https://doi.org/10.1109/", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 613, + 208, + 625 + ], + "spans": [ + { + "bbox": [ + 86, + 615, + 100, + 625 + ], + "score": 1.0, + "content": "420", + "type": "text" + }, + { + "bbox": [ + 115, + 613, + 208, + 625 + ], + "score": 1.0, + "content": "TASE.2019.2956762", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 633, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 89, + 637, + 99, + 646 + ], + "score": 1.0, + "content": "421", + "type": "text" + }, + { + "bbox": [ + 104, + 633, + 506, + 647 + ], + "score": 1.0, + "content": "Junyoung Park, Sanjar Bakhtiyar, and Jinkyoo Park. 2021a. ScheduleNet: Learn to solve multi-agent", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 645, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 89, + 648, + 100, + 657 + ], + "score": 1.0, + "content": "422", + "type": "text" + }, + { + "bbox": [ + 116, + 645, + 505, + 657 + ], + "score": 1.0, + "content": "scheduling problems with reinforcement learning. CoRR abs/2106.03051 (2021). arXiv:2106.03051", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 657, + 286, + 669 + ], + "spans": [ + { + "bbox": [ + 88, + 659, + 100, + 669 + ], + "score": 1.0, + "content": "423", + "type": "text" + }, + { + "bbox": [ + 116, + 657, + 286, + 669 + ], + "score": 1.0, + "content": "https://arxiv.org/abs/2106.03051", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 678, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 100, + 690 + ], + "score": 1.0, + "content": "424", + "type": "text" + }, + { + "bbox": [ + 105, + 678, + 506, + 690 + ], + "score": 1.0, + "content": "Junyoung Park, Jaehyeong Chun, Sang Hun Kim, Youngkook Kim, and Jinkyoo Park. 2021b.", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 85, + 690, + 100, + 701 + ], + "score": 1.0, + "content": "425", + "type": "text" + }, + { + "bbox": [ + 115, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Learning to schedule job-shop problems: representation and policy learning using graph neural", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 700, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 85, + 701, + 100, + 712 + ], + "score": 1.0, + "content": "426", + "type": "text" + }, + { + "bbox": [ + 116, + 700, + 506, + 712 + ], + "score": 1.0, + "content": "network and reinforcement learning. International Journal of Production Research 59, 11 (2021),", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 711, + 408, + 723 + ], + "spans": [ + { + "bbox": [ + 85, + 712, + 100, + 723 + ], + "score": 1.0, + "content": "427", + "type": "text" + }, + { + "bbox": [ + 116, + 711, + 408, + 723 + ], + "score": 1.0, + "content": "3360–3377. https://doi.org/10.1080/00207543.2020.1870013", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 82, + 507, + 97 + ], + "spans": [ + { + "bbox": [ + 86, + 85, + 101, + 96 + ], + "score": 1.0, + "content": "429", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 82, + 507, + 97 + ], + "score": 1.0, + "content": "Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Ed-", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 85, + 96, + 100, + 108 + ], + "score": 1.0, + "content": "430", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "ward Z. Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 105, + 507, + 119 + ], + "spans": [ + { + "bbox": [ + 86, + 108, + 100, + 118 + ], + "score": 1.0, + "content": "431", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 105, + 507, + 119 + ], + "score": 1.0, + "content": "Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. 2019. PyTorch: An Imperative Style, High-", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 115, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 85, + 118, + 101, + 129 + ], + "score": 1.0, + "content": "432", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 115, + 506, + 129 + ], + "score": 1.0, + "content": "Performance Deep Learning Library. In Neural Information Processing Systems (NeurIPS),", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 125, + 507, + 141 + ], + "spans": [ + { + "bbox": [ + 85, + 129, + 100, + 139 + ], + "score": 1.0, + "content": "433", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 125, + 507, + 141 + ], + "score": 1.0, + "content": "Hanna M. Wallach, Hugo Larochelle, Alina Beygelzimer, Florence d’Alché-Buc, Emily B. Fox,", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 138, + 504, + 151 + ], + "spans": [ + { + "bbox": [ + 84, + 138, + 101, + 151 + ], + "score": 1.0, + "content": "434", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 138, + 504, + 151 + ], + "score": 1.0, + "content": "and Roman Garnett (Eds.). 8024–8035. https://proceedings.neurips.cc/paper/2019/", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 150, + 386, + 163 + ], + "spans": [ + { + "bbox": [ + 84, + 150, + 101, + 163 + ], + "score": 1.0, + "content": "435", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 150, + 386, + 162 + ], + "score": 1.0, + "content": "hash/bdbca288fee7f92f2bfa9f7012727740-Abstract.html", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 165, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 85, + 169, + 101, + 180 + ], + "score": 1.0, + "content": "436", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 165, + 506, + 182 + ], + "score": 1.0, + "content": "Laurent Perron and Vincent Furnon. 2019. OR-Tools. Google. https://developers.google.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 179, + 207, + 193 + ], + "spans": [ + { + "bbox": [ + 84, + 179, + 101, + 193 + ], + "score": 1.0, + "content": "437", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 117, + 180, + 207, + 191 + ], + "score": 1.0, + "content": "com/optimization/", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 197, + 506, + 210 + ], + "spans": [ + { + "bbox": [ + 86, + 199, + 101, + 210 + ], + "score": 1.0, + "content": "438", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 197, + 506, + 210 + ], + "score": 1.0, + "content": "Ferdinando Pezzella, Gianluca Morganti, and Giampiero Ciaschetti. 2008. A genetic algorithm for", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 208, + 507, + 221 + ], + "spans": [ + { + "bbox": [ + 85, + 210, + 101, + 221 + ], + "score": 1.0, + "content": "439", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 208, + 507, + 221 + ], + "score": 1.0, + "content": "the Flexible Job-shop Scheduling Problem. Computers and Operations Research 35, 10 (2008),", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 219, + 387, + 233 + ], + "spans": [ + { + "bbox": [ + 85, + 219, + 101, + 233 + ], + "score": 1.0, + "content": "440", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 219, + 387, + 232 + ], + "score": 1.0, + "content": "3202–3212. https://doi.org/10.1016/j.cor.2007.02.014", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 236, + 507, + 252 + ], + "spans": [ + { + "bbox": [ + 85, + 239, + 100, + 250 + ], + "score": 1.0, + "content": "441", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 236, + 507, + 252 + ], + "score": 1.0, + "content": "Qing-dao-er-ji Ren and Yuping Wang. 2012. A new hybrid genetic algorithm for job shop scheduling", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 245, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 84, + 249, + 101, + 262 + ], + "score": 1.0, + "content": "442", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 113, + 245, + 506, + 264 + ], + "score": 1.0, + "content": "problem. Computers and Operations Research 39, 10 (2012), 2291–2299. https://doi.org/", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 258, + 249, + 272 + ], + "spans": [ + { + "bbox": [ + 86, + 261, + 100, + 272 + ], + "score": 1.0, + "content": "443", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 258, + 249, + 272 + ], + "score": 1.0, + "content": "10.1016/j.cor.2011.12.005", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 277, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 85, + 279, + 101, + 289 + ], + "score": 1.0, + "content": "444", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 277, + 505, + 291 + ], + "score": 1.0, + "content": "Mohammad Saidi-Mehrabad and Parviz Fattahi. 2007. Flexible job shop scheduling with tabu search", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 289, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 84, + 289, + 101, + 302 + ], + "score": 1.0, + "content": "445", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 289, + 506, + 302 + ], + "score": 1.0, + "content": "algorithms. The International Journal of Advanced Manufacturing Technology 32, 5 (2007),", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 297, + 158, + 313 + ], + "spans": [ + { + "bbox": [ + 84, + 301, + 101, + 313 + ], + "score": 1.0, + "content": "446", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 297, + 158, + 313 + ], + "score": 1.0, + "content": "563–570.", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 86, + 319, + 101, + 330 + ], + "score": 1.0, + "content": "447", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driess-", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 329, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 85, + 331, + 101, + 341 + ], + "score": 1.0, + "content": "448", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 329, + 506, + 342 + ], + "score": 1.0, + "content": "che, Julian Schrittwieser, Ioannis Antonoglou, Vedavyas Panneershelvam, Marc Lanctot, Sander", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 338, + 507, + 355 + ], + "spans": [ + { + "bbox": [ + 84, + 340, + 101, + 353 + ], + "score": 1.0, + "content": "449", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 338, + 507, + 355 + ], + "score": 1.0, + "content": "Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya Sutskever, Timothy P. Lillicrap,", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 348, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 85, + 353, + 100, + 364 + ], + "score": 1.0, + "content": "450", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 348, + 506, + 366 + ], + "score": 1.0, + "content": "Madeleine Leach, Koray Kavukcuoglu, Thore Graepel, and Demis Hassabis. 2016. Mastering", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 361, + 507, + 376 + ], + "spans": [ + { + "bbox": [ + 86, + 363, + 100, + 374 + ], + "score": 1.0, + "content": "451", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 361, + 507, + 376 + ], + "score": 1.0, + "content": "the game of Go with deep neural networks and tree search. Nature 529, 7587 (2016), 484–489.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 372, + 303, + 387 + ], + "spans": [ + { + "bbox": [ + 86, + 374, + 100, + 385 + ], + "score": 1.0, + "content": "452", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 372, + 303, + 387 + ], + "score": 1.0, + "content": "https://doi.org/10.1038/nature16961", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 391, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 86, + 392, + 101, + 404 + ], + "score": 1.0, + "content": "453", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 391, + 506, + 405 + ], + "score": 1.0, + "content": "Wen Song, Xinyang Chen, Qiqiang Li, and Zhiguang Cao. 2023. Flexible Job-Shop Scheduling via", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 402, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 84, + 403, + 101, + 416 + ], + "score": 1.0, + "content": "454", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 402, + 506, + 415 + ], + "score": 1.0, + "content": "Graph Neural Network and Deep Reinforcement Learning. IEEE Trans. Ind. Informatics 19, 2", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 413, + 412, + 426 + ], + "spans": [ + { + "bbox": [ + 85, + 415, + 101, + 426 + ], + "score": 1.0, + "content": "455", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 413, + 412, + 426 + ], + "score": 1.0, + "content": "(2023), 1600–1610. https://doi.org/10.1109/TII.2022.3189725", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 429, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 86, + 433, + 101, + 444 + ], + "score": 1.0, + "content": "456", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 429, + 506, + 446 + ], + "score": 1.0, + "content": "Robert H. Storer, S. David Wu, and Renzo Vaccari. 1992. New search spaces for sequencing problems", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 442, + 473, + 456 + ], + "spans": [ + { + "bbox": [ + 86, + 444, + 100, + 455 + ], + "score": 1.0, + "content": "457", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 442, + 473, + 456 + ], + "score": 1.0, + "content": "with application to job shop scheduling. Management science 38, 10 (1992), 1495–1509.", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 86, + 462, + 101, + 473 + ], + "score": 1.0, + "content": "458", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "Richard S. Sutton and Andrew G. Barto. 2018. Reinforcement Learning: An Introduction (second", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 472, + 465, + 485 + ], + "spans": [ + { + "bbox": [ + 85, + 474, + 100, + 485 + ], + "score": 1.0, + "content": "459", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 472, + 465, + 485 + ], + "score": 1.0, + "content": "ed.). The MIT Press. http://incompleteideas.net/book/the-book-2nd.html", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 489, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 86, + 493, + 101, + 504 + ], + "score": 1.0, + "content": "460", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 489, + 506, + 506 + ], + "score": 1.0, + "content": "Éric D. Taillard. 1993. Benchmarks for basic scheduling problems. european journal of operational", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 504, + 247, + 515 + ], + "spans": [ + { + "bbox": [ + 85, + 504, + 100, + 515 + ], + "score": 1.0, + "content": "461", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 504, + 247, + 515 + ], + "score": 1.0, + "content": "research 64, 2 (1993), 278–285.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 518, + 507, + 536 + ], + "spans": [ + { + "bbox": [ + 85, + 523, + 101, + 533 + ], + "score": 1.0, + "content": "462", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 518, + 507, + 536 + ], + "score": 1.0, + "content": "Oriol Vinyals, Igor Babuschkin, Wojciech M. Czarnecki, Michaël Mathieu, Andrew Dudzik, Junyoung", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 532, + 507, + 545 + ], + "spans": [ + { + "bbox": [ + 85, + 533, + 100, + 545 + ], + "score": 1.0, + "content": "463", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 532, + 507, + 545 + ], + "score": 1.0, + "content": "Chung, David H. Choi, Richard Powell, Timo Ewalds, Petko Georgiev, Junhyuk Oh, Dan Horgan,", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 542, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 85, + 544, + 100, + 555 + ], + "score": 1.0, + "content": "464", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 542, + 505, + 556 + ], + "score": 1.0, + "content": "Manuel Kroiss, Ivo Danihelka, Aja Huang, Laurent Sifre, Trevor Cai, John P. Agapiou, Max", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 85, + 555, + 100, + 567 + ], + "score": 1.0, + "content": "465", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 554, + 506, + 567 + ], + "score": 1.0, + "content": "Jaderberg, Alexander Sasha Vezhnevets, Rémi Leblond, Tobias Pohlen, Valentin Dalibard, David", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 564, + 507, + 578 + ], + "spans": [ + { + "bbox": [ + 84, + 565, + 101, + 578 + ], + "score": 1.0, + "content": "466", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 564, + 507, + 578 + ], + "score": 1.0, + "content": "Budden, Yury Sulsky, James Molloy, Tom Le Paine, Çaglar Gülçehre, Ziyu Wang, Tobias Pfaff,", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 575, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 85, + 577, + 100, + 588 + ], + "score": 1.0, + "content": "467", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 575, + 506, + 589 + ], + "score": 1.0, + "content": "Yuhuai Wu, Roman Ring, Dani Yogatama, Dario Wünsch, Katrina McKinney, Oliver Smith, Tom", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 587, + 507, + 600 + ], + "spans": [ + { + "bbox": [ + 85, + 588, + 101, + 599 + ], + "score": 1.0, + "content": "468", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 587, + 507, + 600 + ], + "score": 1.0, + "content": "Schaul, Timothy P. Lillicrap, Koray Kavukcuoglu, Demis Hassabis, Chris Apps, and David Silver.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 86, + 599, + 101, + 610 + ], + "score": 1.0, + "content": "469", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "2019. Grandmaster level in StarCraft II using multi-agent reinforcement learning. Nature 575,", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 608, + 432, + 621 + ], + "spans": [ + { + "bbox": [ + 85, + 610, + 100, + 621 + ], + "score": 1.0, + "content": "470", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 608, + 432, + 621 + ], + "score": 1.0, + "content": "7782 (2019), 350–354. https://doi.org/10.1038/s41586-019-1724-z", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 85, + 628, + 101, + 639 + ], + "score": 1.0, + "content": "471", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "Bernd Waschneck, Thomas Altenmüller, Thomas Bauernhansl, and Andreas Kyek. 2016. Production", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 638, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 85, + 639, + 101, + 650 + ], + "score": 1.0, + "content": "472", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 638, + 505, + 651 + ], + "score": 1.0, + "content": "Scheduling in Complex Job Shops from an Industry 4.0 Perspective: A Review and Challenges", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 84, + 649, + 101, + 662 + ], + "score": 1.0, + "content": "473", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "in the Semiconductor Industry. In Proceedings of the 1st International Workshop on Science,", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 85, + 661, + 101, + 672 + ], + "score": 1.0, + "content": "474", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "Application and Methods in Industry 4.0 (i-KNOW) (CEUR Workshop Proceedings, Vol. 1793),", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 668, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 84, + 671, + 101, + 684 + ], + "score": 1.0, + "content": "475", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 668, + 505, + 685 + ], + "score": 1.0, + "content": "Roman Kern, Gerald Reiner, and Olivia Bluder (Eds.). CEUR-WS.org. http://ceur-ws.org/", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 680, + 220, + 697 + ], + "spans": [ + { + "bbox": [ + 85, + 683, + 100, + 694 + ], + "score": 1.0, + "content": "476", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 680, + 220, + 697 + ], + "score": 1.0, + "content": "Vol-1793/paper3.pdf", + "type": "text", + "cross_page": true + } + ], + "index": 47, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 85, + 700, + 102, + 713 + ], + "score": 1.0, + "content": "477", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. 2019. How Powerful are Graph Neural", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 709, + 421, + 725 + ], + "spans": [ + { + "bbox": [ + 84, + 709, + 102, + 725 + ], + "score": 1.0, + "content": "478", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 710, + 421, + 724 + ], + "score": 1.0, + "content": "Networks? (2019). https://openreview.net/forum?id=ryGs6iA5Km", + "type": "text", + "cross_page": true + } + ], + "index": 49, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 73, + 506, + 87 + ], + "spans": [ + { + "bbox": [ + 86, + 75, + 100, + 84 + ], + "score": 1.0, + "content": "479", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 73, + 506, + 87 + ], + "score": 1.0, + "content": "Takeshi Yamada and Ryohei Nakano. 1992. A Genetic Algorithm Applicable to Large-Scale Job-", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 84, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 85, + 84, + 100, + 97 + ], + "score": 1.0, + "content": "480", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 84, + 506, + 96 + ], + "score": 1.0, + "content": "Shop Problems. In Parallel Problem Solving from Nature 2, (PPSN-II), Reinhard Männer and", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 95, + 302, + 107 + ], + "spans": [ + { + "bbox": [ + 85, + 96, + 100, + 107 + ], + "score": 1.0, + "content": "481", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 95, + 302, + 107 + ], + "score": 1.0, + "content": "Bernard Manderick (Eds.). Elsevier, 283–292.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 111, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 85, + 115, + 100, + 126 + ], + "score": 1.0, + "content": "482", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 111, + 506, + 129 + ], + "score": 1.0, + "content": "Cong Zhang, Wen Song, Zhiguang Cao, Jie Zhang, Puay Siew Tan, and Chi Xu. 2020. Learning", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 124, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 85, + 126, + 100, + 137 + ], + "score": 1.0, + "content": "483", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 124, + 506, + 138 + ], + "score": 1.0, + "content": "to Dispatch for Job Shop Scheduling via Deep Reinforcement Learning. In Neural Information", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 85, + 137, + 100, + 148 + ], + "score": 1.0, + "content": "484", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 145, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 85, + 148, + 100, + 159 + ], + "score": 1.0, + "content": "485", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 145, + 506, + 160 + ], + "score": 1.0, + "content": "Florina Balcan, and Hsuan-Tien Lin (Eds.). https://proceedings.neurips.cc/paper/", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 157, + 411, + 170 + ], + "spans": [ + { + "bbox": [ + 85, + 159, + 100, + 170 + ], + "score": 1.0, + "content": "486", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 157, + 411, + 170 + ], + "score": 1.0, + "content": "2020/hash/11958dfee29b6709f48a9ba0387a2431-Abstract.html", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 176, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 85, + 177, + 100, + 188 + ], + "score": 1.0, + "content": "487", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 176, + 506, + 190 + ], + "score": 1.0, + "content": "Cong Zhang, Wen Song, Zhiguang Cao, Jie Zhang, Puay Siew Tan, and Chi Xu. 2022. Learning to", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 186, + 507, + 201 + ], + "spans": [ + { + "bbox": [ + 85, + 189, + 100, + 200 + ], + "score": 1.0, + "content": "488", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 186, + 507, + 201 + ], + "score": 1.0, + "content": "Search for Job Shop Scheduling via Deep Reinforcement Learning. CoRR abs/2211.10936 (2022).", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 197, + 410, + 212 + ], + "spans": [ + { + "bbox": [ + 85, + 200, + 100, + 210 + ], + "score": 1.0, + "content": "489", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 197, + 410, + 212 + ], + "score": 1.0, + "content": "https://doi.org/10.48550/arXiv.2211.10936 arXiv:2211.10936", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 216, + 506, + 230 + ], + "spans": [ + { + "bbox": [ + 85, + 218, + 100, + 229 + ], + "score": 1.0, + "content": "490", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 216, + 506, + 230 + ], + "score": 1.0, + "content": "Cong Zhang, Yaoxin Wu, Yining Ma, Wen Song, Zhang Le, Zhiguang Cao, and Jie Zhang. 2023. A", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 227, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 85, + 230, + 100, + 240 + ], + "score": 1.0, + "content": "491", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 227, + 506, + 241 + ], + "score": 1.0, + "content": "review on learning to solve combinatorial optimisation problems in manufacturing. IET Collabora-", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 238, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 85, + 240, + 100, + 251 + ], + "score": 1.0, + "content": "492", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 238, + 506, + 252 + ], + "score": 1.0, + "content": "tive Intelligent Manufacturing 5, 1 (2023), e12072. https://doi.org/10.1049/cim2.12072", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 249, + 429, + 263 + ], + "spans": [ + { + "bbox": [ + 85, + 251, + 100, + 262 + ], + "score": 1.0, + "content": "493", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 249, + 429, + 263 + ], + "score": 1.0, + "content": "arXiv:https://ietresearch.onlinelibrary.wiley.com/doi/pdf/10.1049/cim2.12072", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 268, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 86, + 271, + 99, + 280 + ], + "score": 1.0, + "content": "494", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 268, + 506, + 282 + ], + "score": 1.0, + "content": "Jie Zhou, Ganqu Cui, Shengding Hu, Zhengyan Zhang, Cheng Yang, Zhiyuan Liu, Lifeng Wang,", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 279, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 85, + 281, + 100, + 292 + ], + "score": 1.0, + "content": "495", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 279, + 505, + 293 + ], + "score": 1.0, + "content": "Changcheng Li, and Maosong Sun. 2020. Graph neural networks: A review of methods and", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 290, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 85, + 292, + 100, + 303 + ], + "score": 1.0, + "content": "496", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 290, + 506, + 305 + ], + "score": 1.0, + "content": "applications. AI Open 1 (2020), 57–81. https://doi.org/10.1016/j.aiopen.2021.01.001", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + } + ], + "index": 37.5, + "bbox_fs": [ + 85, + 579, + 506, + 625 + ] + }, + { + "type": "index", + "bbox": [ + 90, + 634, + 505, + 668 + ], + "lines": [], + "index": 41, + "bbox_fs": [ + 88, + 633, + 506, + 669 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 87, + 678, + 506, + 722 + ], + "lines": [], + "index": 44.5, + "bbox_fs": [ + 85, + 678, + 506, + 723 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 84, + 83, + 507, + 730 + ], + "lines": [ + { + "bbox": [ + 86, + 82, + 507, + 97 + ], + "spans": [ + { + "bbox": [ + 86, + 85, + 101, + 96 + ], + "score": 1.0, + "content": "429", + "type": "text" + }, + { + "bbox": [ + 114, + 82, + 507, + 97 + ], + "score": 1.0, + "content": "Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Ed-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 85, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 85, + 96, + 100, + 108 + ], + "score": 1.0, + "content": "430", + "type": "text" + }, + { + "bbox": [ + 115, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "ward Z. Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 86, + 105, + 507, + 119 + ], + "spans": [ + { + "bbox": [ + 86, + 108, + 100, + 118 + ], + "score": 1.0, + "content": "431", + "type": "text" + }, + { + "bbox": [ + 115, + 105, + 507, + 119 + ], + "score": 1.0, + "content": "Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. 2019. PyTorch: An Imperative Style, High-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 85, + 115, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 85, + 118, + 101, + 129 + ], + "score": 1.0, + "content": "432", + "type": "text" + }, + { + "bbox": [ + 115, + 115, + 506, + 129 + ], + "score": 1.0, + "content": "Performance Deep Learning Library. In Neural Information Processing Systems (NeurIPS),", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 85, + 125, + 507, + 141 + ], + "spans": [ + { + "bbox": [ + 85, + 129, + 100, + 139 + ], + "score": 1.0, + "content": "433", + "type": "text" + }, + { + "bbox": [ + 114, + 125, + 507, + 141 + ], + "score": 1.0, + "content": "Hanna M. Wallach, Hugo Larochelle, Alina Beygelzimer, Florence d’Alché-Buc, Emily B. Fox,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 84, + 138, + 504, + 151 + ], + "spans": [ + { + "bbox": [ + 84, + 138, + 101, + 151 + ], + "score": 1.0, + "content": "434", + "type": "text" + }, + { + "bbox": [ + 115, + 138, + 504, + 151 + ], + "score": 1.0, + "content": "and Roman Garnett (Eds.). 8024–8035. https://proceedings.neurips.cc/paper/2019/", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 84, + 150, + 386, + 163 + ], + "spans": [ + { + "bbox": [ + 84, + 150, + 101, + 163 + ], + "score": 1.0, + "content": "435", + "type": "text" + }, + { + "bbox": [ + 115, + 150, + 386, + 162 + ], + "score": 1.0, + "content": "hash/bdbca288fee7f92f2bfa9f7012727740-Abstract.html", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 85, + 165, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 85, + 169, + 101, + 180 + ], + "score": 1.0, + "content": "436", + "type": "text" + }, + { + "bbox": [ + 104, + 165, + 506, + 182 + ], + "score": 1.0, + "content": "Laurent Perron and Vincent Furnon. 2019. OR-Tools. Google. https://developers.google.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 84, + 179, + 207, + 193 + ], + "spans": [ + { + "bbox": [ + 84, + 179, + 101, + 193 + ], + "score": 1.0, + "content": "437", + "type": "text" + }, + { + "bbox": [ + 117, + 180, + 207, + 191 + ], + "score": 1.0, + "content": "com/optimization/", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 86, + 197, + 506, + 210 + ], + "spans": [ + { + "bbox": [ + 86, + 199, + 101, + 210 + ], + "score": 1.0, + "content": "438", + "type": "text" + }, + { + "bbox": [ + 105, + 197, + 506, + 210 + ], + "score": 1.0, + "content": "Ferdinando Pezzella, Gianluca Morganti, and Giampiero Ciaschetti. 2008. A genetic algorithm for", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 85, + 208, + 507, + 221 + ], + "spans": [ + { + "bbox": [ + 85, + 210, + 101, + 221 + ], + "score": 1.0, + "content": "439", + "type": "text" + }, + { + "bbox": [ + 115, + 208, + 507, + 221 + ], + "score": 1.0, + "content": "the Flexible Job-shop Scheduling Problem. Computers and Operations Research 35, 10 (2008),", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 85, + 219, + 387, + 233 + ], + "spans": [ + { + "bbox": [ + 85, + 219, + 101, + 233 + ], + "score": 1.0, + "content": "440", + "type": "text" + }, + { + "bbox": [ + 115, + 219, + 387, + 232 + ], + "score": 1.0, + "content": "3202–3212. https://doi.org/10.1016/j.cor.2007.02.014", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 85, + 236, + 507, + 252 + ], + "spans": [ + { + "bbox": [ + 85, + 239, + 100, + 250 + ], + "score": 1.0, + "content": "441", + "type": "text" + }, + { + "bbox": [ + 105, + 236, + 507, + 252 + ], + "score": 1.0, + "content": "Qing-dao-er-ji Ren and Yuping Wang. 2012. A new hybrid genetic algorithm for job shop scheduling", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 84, + 245, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 84, + 249, + 101, + 262 + ], + "score": 1.0, + "content": "442", + "type": "text" + }, + { + "bbox": [ + 113, + 245, + 506, + 264 + ], + "score": 1.0, + "content": "problem. Computers and Operations Research 39, 10 (2012), 2291–2299. https://doi.org/", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 86, + 258, + 249, + 272 + ], + "spans": [ + { + "bbox": [ + 86, + 261, + 100, + 272 + ], + "score": 1.0, + "content": "443", + "type": "text" + }, + { + "bbox": [ + 115, + 258, + 249, + 272 + ], + "score": 1.0, + "content": "10.1016/j.cor.2011.12.005", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 85, + 277, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 85, + 279, + 101, + 289 + ], + "score": 1.0, + "content": "444", + "type": "text" + }, + { + "bbox": [ + 104, + 277, + 505, + 291 + ], + "score": 1.0, + "content": "Mohammad Saidi-Mehrabad and Parviz Fattahi. 2007. Flexible job shop scheduling with tabu search", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 84, + 289, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 84, + 289, + 101, + 302 + ], + "score": 1.0, + "content": "445", + "type": "text" + }, + { + "bbox": [ + 115, + 289, + 506, + 302 + ], + "score": 1.0, + "content": "algorithms. The International Journal of Advanced Manufacturing Technology 32, 5 (2007),", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 84, + 297, + 158, + 313 + ], + "spans": [ + { + "bbox": [ + 84, + 301, + 101, + 313 + ], + "score": 1.0, + "content": "446", + "type": "text" + }, + { + "bbox": [ + 114, + 297, + 158, + 313 + ], + "score": 1.0, + "content": "563–570.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 86, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 86, + 319, + 101, + 330 + ], + "score": 1.0, + "content": "447", + "type": "text" + }, + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driess-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 85, + 329, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 85, + 331, + 101, + 341 + ], + "score": 1.0, + "content": "448", + "type": "text" + }, + { + "bbox": [ + 115, + 329, + 506, + 342 + ], + "score": 1.0, + "content": "che, Julian Schrittwieser, Ioannis Antonoglou, Vedavyas Panneershelvam, Marc Lanctot, Sander", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 84, + 338, + 507, + 355 + ], + "spans": [ + { + "bbox": [ + 84, + 340, + 101, + 353 + ], + "score": 1.0, + "content": "449", + "type": "text" + }, + { + "bbox": [ + 114, + 338, + 507, + 355 + ], + "score": 1.0, + "content": "Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya Sutskever, Timothy P. Lillicrap,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 85, + 348, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 85, + 353, + 100, + 364 + ], + "score": 1.0, + "content": "450", + "type": "text" + }, + { + "bbox": [ + 114, + 348, + 506, + 366 + ], + "score": 1.0, + "content": "Madeleine Leach, Koray Kavukcuoglu, Thore Graepel, and Demis Hassabis. 2016. Mastering", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 86, + 361, + 507, + 376 + ], + "spans": [ + { + "bbox": [ + 86, + 363, + 100, + 374 + ], + "score": 1.0, + "content": "451", + "type": "text" + }, + { + "bbox": [ + 115, + 361, + 507, + 376 + ], + "score": 1.0, + "content": "the game of Go with deep neural networks and tree search. Nature 529, 7587 (2016), 484–489.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 86, + 372, + 303, + 387 + ], + "spans": [ + { + "bbox": [ + 86, + 374, + 100, + 385 + ], + "score": 1.0, + "content": "452", + "type": "text" + }, + { + "bbox": [ + 114, + 372, + 303, + 387 + ], + "score": 1.0, + "content": "https://doi.org/10.1038/nature16961", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 86, + 391, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 86, + 392, + 101, + 404 + ], + "score": 1.0, + "content": "453", + "type": "text" + }, + { + "bbox": [ + 105, + 391, + 506, + 405 + ], + "score": 1.0, + "content": "Wen Song, Xinyang Chen, Qiqiang Li, and Zhiguang Cao. 2023. Flexible Job-Shop Scheduling via", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 84, + 402, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 84, + 403, + 101, + 416 + ], + "score": 1.0, + "content": "454", + "type": "text" + }, + { + "bbox": [ + 115, + 402, + 506, + 415 + ], + "score": 1.0, + "content": "Graph Neural Network and Deep Reinforcement Learning. IEEE Trans. Ind. Informatics 19, 2", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 85, + 413, + 412, + 426 + ], + "spans": [ + { + "bbox": [ + 85, + 415, + 101, + 426 + ], + "score": 1.0, + "content": "455", + "type": "text" + }, + { + "bbox": [ + 115, + 413, + 412, + 426 + ], + "score": 1.0, + "content": "(2023), 1600–1610. https://doi.org/10.1109/TII.2022.3189725", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 86, + 429, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 86, + 433, + 101, + 444 + ], + "score": 1.0, + "content": "456", + "type": "text" + }, + { + "bbox": [ + 104, + 429, + 506, + 446 + ], + "score": 1.0, + "content": "Robert H. Storer, S. David Wu, and Renzo Vaccari. 1992. New search spaces for sequencing problems", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 86, + 442, + 473, + 456 + ], + "spans": [ + { + "bbox": [ + 86, + 444, + 100, + 455 + ], + "score": 1.0, + "content": "457", + "type": "text" + }, + { + "bbox": [ + 115, + 442, + 473, + 456 + ], + "score": 1.0, + "content": "with application to job shop scheduling. Management science 38, 10 (1992), 1495–1509.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 86, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 86, + 462, + 101, + 473 + ], + "score": 1.0, + "content": "458", + "type": "text" + }, + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "Richard S. Sutton and Andrew G. Barto. 2018. Reinforcement Learning: An Introduction (second", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 85, + 472, + 465, + 485 + ], + "spans": [ + { + "bbox": [ + 85, + 474, + 100, + 485 + ], + "score": 1.0, + "content": "459", + "type": "text" + }, + { + "bbox": [ + 115, + 472, + 465, + 485 + ], + "score": 1.0, + "content": "ed.). The MIT Press. http://incompleteideas.net/book/the-book-2nd.html", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 86, + 489, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 86, + 493, + 101, + 504 + ], + "score": 1.0, + "content": "460", + "type": "text" + }, + { + "bbox": [ + 104, + 489, + 506, + 506 + ], + "score": 1.0, + "content": "Éric D. Taillard. 1993. Benchmarks for basic scheduling problems. european journal of operational", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 85, + 504, + 247, + 515 + ], + "spans": [ + { + "bbox": [ + 85, + 504, + 100, + 515 + ], + "score": 1.0, + "content": "461", + "type": "text" + }, + { + "bbox": [ + 115, + 504, + 247, + 515 + ], + "score": 1.0, + "content": "research 64, 2 (1993), 278–285.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 85, + 518, + 507, + 536 + ], + "spans": [ + { + "bbox": [ + 85, + 523, + 101, + 533 + ], + "score": 1.0, + "content": "462", + "type": "text" + }, + { + "bbox": [ + 104, + 518, + 507, + 536 + ], + "score": 1.0, + "content": "Oriol Vinyals, Igor Babuschkin, Wojciech M. Czarnecki, Michaël Mathieu, Andrew Dudzik, Junyoung", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 85, + 532, + 507, + 545 + ], + "spans": [ + { + "bbox": [ + 85, + 533, + 100, + 545 + ], + "score": 1.0, + "content": "463", + "type": "text" + }, + { + "bbox": [ + 115, + 532, + 507, + 545 + ], + "score": 1.0, + "content": "Chung, David H. Choi, Richard Powell, Timo Ewalds, Petko Georgiev, Junhyuk Oh, Dan Horgan,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 85, + 542, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 85, + 544, + 100, + 555 + ], + "score": 1.0, + "content": "464", + "type": "text" + }, + { + "bbox": [ + 115, + 542, + 505, + 556 + ], + "score": 1.0, + "content": "Manuel Kroiss, Ivo Danihelka, Aja Huang, Laurent Sifre, Trevor Cai, John P. Agapiou, Max", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 85, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 85, + 555, + 100, + 567 + ], + "score": 1.0, + "content": "465", + "type": "text" + }, + { + "bbox": [ + 115, + 554, + 506, + 567 + ], + "score": 1.0, + "content": "Jaderberg, Alexander Sasha Vezhnevets, Rémi Leblond, Tobias Pohlen, Valentin Dalibard, David", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 84, + 564, + 507, + 578 + ], + "spans": [ + { + "bbox": [ + 84, + 565, + 101, + 578 + ], + "score": 1.0, + "content": "466", + "type": "text" + }, + { + "bbox": [ + 115, + 564, + 507, + 578 + ], + "score": 1.0, + "content": "Budden, Yury Sulsky, James Molloy, Tom Le Paine, Çaglar Gülçehre, Ziyu Wang, Tobias Pfaff,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 85, + 575, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 85, + 577, + 100, + 588 + ], + "score": 1.0, + "content": "467", + "type": "text" + }, + { + "bbox": [ + 115, + 575, + 506, + 589 + ], + "score": 1.0, + "content": "Yuhuai Wu, Roman Ring, Dani Yogatama, Dario Wünsch, Katrina McKinney, Oliver Smith, Tom", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 85, + 587, + 507, + 600 + ], + "spans": [ + { + "bbox": [ + 85, + 588, + 101, + 599 + ], + "score": 1.0, + "content": "468", + "type": "text" + }, + { + "bbox": [ + 115, + 587, + 507, + 600 + ], + "score": 1.0, + "content": "Schaul, Timothy P. Lillicrap, Koray Kavukcuoglu, Demis Hassabis, Chris Apps, and David Silver.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 86, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 86, + 599, + 101, + 610 + ], + "score": 1.0, + "content": "469", + "type": "text" + }, + { + "bbox": [ + 115, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "2019. Grandmaster level in StarCraft II using multi-agent reinforcement learning. Nature 575,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 85, + 608, + 432, + 621 + ], + "spans": [ + { + "bbox": [ + 85, + 610, + 100, + 621 + ], + "score": 1.0, + "content": "470", + "type": "text" + }, + { + "bbox": [ + 115, + 608, + 432, + 621 + ], + "score": 1.0, + "content": "7782 (2019), 350–354. https://doi.org/10.1038/s41586-019-1724-z", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 85, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 85, + 628, + 101, + 639 + ], + "score": 1.0, + "content": "471", + "type": "text" + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "Bernd Waschneck, Thomas Altenmüller, Thomas Bauernhansl, and Andreas Kyek. 2016. Production", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 85, + 638, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 85, + 639, + 101, + 650 + ], + "score": 1.0, + "content": "472", + "type": "text" + }, + { + "bbox": [ + 115, + 638, + 505, + 651 + ], + "score": 1.0, + "content": "Scheduling in Complex Job Shops from an Industry 4.0 Perspective: A Review and Challenges", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 84, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 84, + 649, + 101, + 662 + ], + "score": 1.0, + "content": "473", + "type": "text" + }, + { + "bbox": [ + 115, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "in the Semiconductor Industry. In Proceedings of the 1st International Workshop on Science,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 85, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 85, + 661, + 101, + 672 + ], + "score": 1.0, + "content": "474", + "type": "text" + }, + { + "bbox": [ + 114, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "Application and Methods in Industry 4.0 (i-KNOW) (CEUR Workshop Proceedings, Vol. 1793),", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 84, + 668, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 84, + 671, + 101, + 684 + ], + "score": 1.0, + "content": "475", + "type": "text" + }, + { + "bbox": [ + 114, + 668, + 505, + 685 + ], + "score": 1.0, + "content": "Roman Kern, Gerald Reiner, and Olivia Bluder (Eds.). CEUR-WS.org. http://ceur-ws.org/", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 85, + 680, + 220, + 697 + ], + "spans": [ + { + "bbox": [ + 85, + 683, + 100, + 694 + ], + "score": 1.0, + "content": "476", + "type": "text" + }, + { + "bbox": [ + 114, + 680, + 220, + 697 + ], + "score": 1.0, + "content": "Vol-1793/paper3.pdf", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 85, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 85, + 700, + 102, + 713 + ], + "score": 1.0, + "content": "477", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. 2019. How Powerful are Graph Neural", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 84, + 709, + 421, + 725 + ], + "spans": [ + { + "bbox": [ + 84, + 709, + 102, + 725 + ], + "score": 1.0, + "content": "478", + "type": "text" + }, + { + "bbox": [ + 115, + 710, + 421, + 724 + ], + "score": 1.0, + "content": "Networks? (2019). https://openreview.net/forum?id=ryGs6iA5Km", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 24.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 84, + 83, + 507, + 730 + ], + "lines": [], + "index": 24.5, + "bbox_fs": [ + 84, + 82, + 507, + 725 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 84, + 72, + 507, + 303 + ], + "lines": [ + { + "bbox": [ + 86, + 73, + 506, + 87 + ], + "spans": [ + { + "bbox": [ + 86, + 75, + 100, + 84 + ], + "score": 1.0, + "content": "479", + "type": "text" + }, + { + "bbox": [ + 105, + 73, + 506, + 87 + ], + "score": 1.0, + "content": "Takeshi Yamada and Ryohei Nakano. 1992. A Genetic Algorithm Applicable to Large-Scale Job-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 85, + 84, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 85, + 84, + 100, + 97 + ], + "score": 1.0, + "content": "480", + "type": "text" + }, + { + "bbox": [ + 116, + 84, + 506, + 96 + ], + "score": 1.0, + "content": "Shop Problems. In Parallel Problem Solving from Nature 2, (PPSN-II), Reinhard Männer and", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 85, + 95, + 302, + 107 + ], + "spans": [ + { + "bbox": [ + 85, + 96, + 100, + 107 + ], + "score": 1.0, + "content": "481", + "type": "text" + }, + { + "bbox": [ + 116, + 95, + 302, + 107 + ], + "score": 1.0, + "content": "Bernard Manderick (Eds.). Elsevier, 283–292.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 85, + 111, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 85, + 115, + 100, + 126 + ], + "score": 1.0, + "content": "482", + "type": "text" + }, + { + "bbox": [ + 105, + 111, + 506, + 129 + ], + "score": 1.0, + "content": "Cong Zhang, Wen Song, Zhiguang Cao, Jie Zhang, Puay Siew Tan, and Chi Xu. 2020. Learning", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 85, + 124, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 85, + 126, + 100, + 137 + ], + "score": 1.0, + "content": "483", + "type": "text" + }, + { + "bbox": [ + 115, + 124, + 506, + 138 + ], + "score": 1.0, + "content": "to Dispatch for Job Shop Scheduling via Deep Reinforcement Learning. In Neural Information", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 85, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 85, + 137, + 100, + 148 + ], + "score": 1.0, + "content": "484", + "type": "text" + }, + { + "bbox": [ + 115, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 85, + 145, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 85, + 148, + 100, + 159 + ], + "score": 1.0, + "content": "485", + "type": "text" + }, + { + "bbox": [ + 115, + 145, + 506, + 160 + ], + "score": 1.0, + "content": "Florina Balcan, and Hsuan-Tien Lin (Eds.). https://proceedings.neurips.cc/paper/", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 85, + 157, + 411, + 170 + ], + "spans": [ + { + "bbox": [ + 85, + 159, + 100, + 170 + ], + "score": 1.0, + "content": "486", + "type": "text" + }, + { + "bbox": [ + 115, + 157, + 411, + 170 + ], + "score": 1.0, + "content": "2020/hash/11958dfee29b6709f48a9ba0387a2431-Abstract.html", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 85, + 176, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 85, + 177, + 100, + 188 + ], + "score": 1.0, + "content": "487", + "type": "text" + }, + { + "bbox": [ + 106, + 176, + 506, + 190 + ], + "score": 1.0, + "content": "Cong Zhang, Wen Song, Zhiguang Cao, Jie Zhang, Puay Siew Tan, and Chi Xu. 2022. Learning to", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 85, + 186, + 507, + 201 + ], + "spans": [ + { + "bbox": [ + 85, + 189, + 100, + 200 + ], + "score": 1.0, + "content": "488", + "type": "text" + }, + { + "bbox": [ + 115, + 186, + 507, + 201 + ], + "score": 1.0, + "content": "Search for Job Shop Scheduling via Deep Reinforcement Learning. CoRR abs/2211.10936 (2022).", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 85, + 197, + 410, + 212 + ], + "spans": [ + { + "bbox": [ + 85, + 200, + 100, + 210 + ], + "score": 1.0, + "content": "489", + "type": "text" + }, + { + "bbox": [ + 115, + 197, + 410, + 212 + ], + "score": 1.0, + "content": "https://doi.org/10.48550/arXiv.2211.10936 arXiv:2211.10936", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 85, + 216, + 506, + 230 + ], + "spans": [ + { + "bbox": [ + 85, + 218, + 100, + 229 + ], + "score": 1.0, + "content": "490", + "type": "text" + }, + { + "bbox": [ + 106, + 216, + 506, + 230 + ], + "score": 1.0, + "content": "Cong Zhang, Yaoxin Wu, Yining Ma, Wen Song, Zhang Le, Zhiguang Cao, and Jie Zhang. 2023. A", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 85, + 227, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 85, + 230, + 100, + 240 + ], + "score": 1.0, + "content": "491", + "type": "text" + }, + { + "bbox": [ + 115, + 227, + 506, + 241 + ], + "score": 1.0, + "content": "review on learning to solve combinatorial optimisation problems in manufacturing. IET Collabora-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 85, + 238, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 85, + 240, + 100, + 251 + ], + "score": 1.0, + "content": "492", + "type": "text" + }, + { + "bbox": [ + 114, + 238, + 506, + 252 + ], + "score": 1.0, + "content": "tive Intelligent Manufacturing 5, 1 (2023), e12072. https://doi.org/10.1049/cim2.12072", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 85, + 249, + 429, + 263 + ], + "spans": [ + { + "bbox": [ + 85, + 251, + 100, + 262 + ], + "score": 1.0, + "content": "493", + "type": "text" + }, + { + "bbox": [ + 116, + 249, + 429, + 263 + ], + "score": 1.0, + "content": "arXiv:https://ietresearch.onlinelibrary.wiley.com/doi/pdf/10.1049/cim2.12072", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 86, + 268, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 86, + 271, + 99, + 280 + ], + "score": 1.0, + "content": "494", + "type": "text" + }, + { + "bbox": [ + 105, + 268, + 506, + 282 + ], + "score": 1.0, + "content": "Jie Zhou, Ganqu Cui, Shengding Hu, Zhengyan Zhang, Cheng Yang, Zhiyuan Liu, Lifeng Wang,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 85, + 279, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 85, + 281, + 100, + 292 + ], + "score": 1.0, + "content": "495", + "type": "text" + }, + { + "bbox": [ + 116, + 279, + 505, + 293 + ], + "score": 1.0, + "content": "Changcheng Li, and Maosong Sun. 2020. Graph neural networks: A review of methods and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 85, + 290, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 85, + 292, + 100, + 303 + ], + "score": 1.0, + "content": "496", + "type": "text" + }, + { + "bbox": [ + 116, + 290, + 506, + 305 + ], + "score": 1.0, + "content": "applications. AI Open 1 (2020), 57–81. https://doi.org/10.1016/j.aiopen.2021.01.001", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 8.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 84, + 72, + 507, + 303 + ], + "lines": [], + "index": 8.5, + "bbox_fs": [ + 85, + 73, + 507, + 305 + ], + "lines_deleted": true + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/dev/5uIL1E8h1E/5uIL1E8h1E_model.json b/parse/dev/5uIL1E8h1E/5uIL1E8h1E_model.json new file mode 100644 index 0000000000000000000000000000000000000000..6885338cabb039fbf533902eb99d559406dabfdc --- /dev/null +++ b/parse/dev/5uIL1E8h1E/5uIL1E8h1E_model.json @@ -0,0 +1,28944 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 253, + 769, + 1305, + 769, + 1305, + 1286, + 253, + 1286 + ], + "score": 0.975 + }, + { + "category_id": 0, + "poly": [ + 300, + 271, + 1401, + 271, + 1401, + 383, + 300, + 383 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 250, + 1636, + 1405, + 1636, + 1405, + 1911, + 250, + 1911 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 251, + 1408, + 1404, + 1408, + 1404, + 1622, + 251, + 1622 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 720, + 499, + 986, + 499, + 986, + 620, + 720, + 620 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 255, + 1924, + 1404, + 1924, + 1404, + 1986, + 255, + 1986 + ], + "score": 0.892 + }, + { + "category_id": 0, + "poly": [ + 788, + 701, + 913, + 701, + 913, + 737, + 788, + 737 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 254, + 1340, + 531, + 1340, + 531, + 1378, + 254, + 1378 + ], + "score": 0.886 + }, + { + "category_id": 2, + "poly": [ + 285, + 2033, + 1388, + 2033, + 1388, + 2062, + 285, + 2062 + ], + "score": 0.707 + }, + { + "category_id": 1, + "poly": [ + 285, + 2033, + 1388, + 2033, + 1388, + 2062, + 285, + 2062 + ], + "score": 0.126 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 266.0, + 1404.0, + 266.0, + 1404.0, + 334.0, + 295.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 328.0, + 1375.0, + 328.0, + 1375.0, + 387.0, + 324.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 698.0, + 920.0, + 698.0, + 920.0, + 742.0, + 782.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 1338.0, + 535.0, + 1338.0, + 535.0, + 1384.0, + 243.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2029.0, + 1390.0, + 2029.0, + 1390.0, + 2066.0, + 295.0, + 2066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 778.0, + 277.0, + 778.0, + 277.0, + 801.0, + 258.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 767.0, + 1306.0, + 767.0, + 1306.0, + 806.0, + 393.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 807.0, + 277.0, + 807.0, + 277.0, + 830.0, + 258.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 799.0, + 1306.0, + 799.0, + 1306.0, + 832.0, + 395.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 837.0, + 277.0, + 837.0, + 277.0, + 861.0, + 259.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 830.0, + 1305.0, + 830.0, + 1305.0, + 864.0, + 395.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 868.0, + 277.0, + 868.0, + 277.0, + 893.0, + 258.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 858.0, + 1307.0, + 858.0, + 1307.0, + 897.0, + 393.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 897.0, + 277.0, + 897.0, + 277.0, + 921.0, + 259.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 891.0, + 1306.0, + 891.0, + 1306.0, + 925.0, + 394.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 927.0, + 277.0, + 927.0, + 277.0, + 953.0, + 258.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 920.0, + 1306.0, + 920.0, + 1306.0, + 954.0, + 393.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 959.0, + 277.0, + 959.0, + 277.0, + 983.0, + 258.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 952.0, + 1306.0, + 952.0, + 1306.0, + 987.0, + 394.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 989.0, + 277.0, + 989.0, + 277.0, + 1013.0, + 258.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 979.0, + 1307.0, + 979.0, + 1307.0, + 1019.0, + 393.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 1019.0, + 277.0, + 1019.0, + 277.0, + 1043.0, + 258.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1012.0, + 1308.0, + 1012.0, + 1308.0, + 1048.0, + 393.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1049.0, + 277.0, + 1049.0, + 277.0, + 1074.0, + 250.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1042.0, + 1305.0, + 1042.0, + 1305.0, + 1076.0, + 394.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1079.0, + 276.0, + 1079.0, + 276.0, + 1105.0, + 250.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1072.0, + 1306.0, + 1072.0, + 1306.0, + 1108.0, + 393.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1109.0, + 277.0, + 1109.0, + 277.0, + 1135.0, + 250.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1102.0, + 1306.0, + 1102.0, + 1306.0, + 1137.0, + 394.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1141.0, + 277.0, + 1141.0, + 277.0, + 1166.0, + 250.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1134.0, + 1306.0, + 1134.0, + 1306.0, + 1167.0, + 394.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1171.0, + 277.0, + 1171.0, + 277.0, + 1195.0, + 249.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1164.0, + 1306.0, + 1164.0, + 1306.0, + 1197.0, + 395.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1199.0, + 280.0, + 1199.0, + 280.0, + 1228.0, + 247.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1193.0, + 1311.0, + 1193.0, + 1311.0, + 1230.0, + 394.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1230.0, + 277.0, + 1230.0, + 277.0, + 1255.0, + 250.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1221.0, + 1307.0, + 1221.0, + 1307.0, + 1261.0, + 393.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1261.0, + 277.0, + 1261.0, + 277.0, + 1287.0, + 250.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1256.0, + 1205.0, + 1256.0, + 1205.0, + 1287.0, + 396.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1642.0, + 277.0, + 1642.0, + 277.0, + 1668.0, + 248.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1638.0, + 1404.0, + 1638.0, + 1404.0, + 1671.0, + 294.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1672.0, + 277.0, + 1672.0, + 277.0, + 1699.0, + 248.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1665.0, + 1405.0, + 1665.0, + 1405.0, + 1702.0, + 293.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1702.0, + 277.0, + 1702.0, + 277.0, + 1728.0, + 248.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1695.0, + 1407.0, + 1695.0, + 1407.0, + 1732.0, + 294.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1732.0, + 280.0, + 1732.0, + 280.0, + 1762.0, + 246.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1727.0, + 1404.0, + 1727.0, + 1404.0, + 1763.0, + 295.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1763.0, + 278.0, + 1763.0, + 278.0, + 1790.0, + 248.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1757.0, + 1407.0, + 1757.0, + 1407.0, + 1792.0, + 293.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1792.0, + 280.0, + 1792.0, + 280.0, + 1822.0, + 246.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1783.0, + 1411.0, + 1783.0, + 1411.0, + 1826.0, + 292.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1824.0, + 278.0, + 1824.0, + 278.0, + 1850.0, + 248.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1815.0, + 1408.0, + 1815.0, + 1408.0, + 1856.0, + 293.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1854.0, + 278.0, + 1854.0, + 278.0, + 1882.0, + 248.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1846.0, + 1407.0, + 1846.0, + 1407.0, + 1884.0, + 292.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1884.0, + 278.0, + 1884.0, + 278.0, + 1912.0, + 248.0, + 1912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1879.0, + 962.0, + 1879.0, + 962.0, + 1915.0, + 294.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1414.0, + 278.0, + 1414.0, + 278.0, + 1440.0, + 251.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1408.0, + 1404.0, + 1408.0, + 1404.0, + 1443.0, + 294.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1446.0, + 278.0, + 1446.0, + 278.0, + 1471.0, + 249.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1440.0, + 1407.0, + 1440.0, + 1407.0, + 1475.0, + 294.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1476.0, + 278.0, + 1476.0, + 278.0, + 1502.0, + 248.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1470.0, + 1407.0, + 1470.0, + 1407.0, + 1504.0, + 294.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1506.0, + 278.0, + 1506.0, + 278.0, + 1531.0, + 248.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1500.0, + 1406.0, + 1500.0, + 1406.0, + 1535.0, + 295.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1537.0, + 278.0, + 1537.0, + 278.0, + 1563.0, + 248.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1531.0, + 1406.0, + 1531.0, + 1406.0, + 1565.0, + 295.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1568.0, + 278.0, + 1568.0, + 278.0, + 1591.0, + 248.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1560.0, + 1406.0, + 1560.0, + 1406.0, + 1595.0, + 295.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1597.0, + 278.0, + 1597.0, + 278.0, + 1622.0, + 249.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1591.0, + 1249.0, + 1591.0, + 1249.0, + 1625.0, + 295.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 496.0, + 989.0, + 496.0, + 989.0, + 533.0, + 717.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 527.0, + 913.0, + 527.0, + 913.0, + 563.0, + 787.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 557.0, + 902.0, + 557.0, + 902.0, + 592.0, + 797.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 593.0, + 890.0, + 593.0, + 890.0, + 619.0, + 808.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 1922.0, + 1405.0, + 1922.0, + 1405.0, + 1960.0, + 243.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 1954.0, + 1406.0, + 1954.0, + 1406.0, + 1989.0, + 243.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2029.0, + 1390.0, + 2029.0, + 1390.0, + 2066.0, + 295.0, + 2066.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 249, + 885, + 1405, + 885, + 1405, + 1311, + 249, + 1311 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 249, + 566, + 1404, + 566, + 1404, + 871, + 249, + 871 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 251, + 1461, + 1404, + 1461, + 1404, + 1644, + 251, + 1644 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 250, + 203, + 1404, + 203, + 1404, + 386, + 250, + 386 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 250, + 400, + 1404, + 400, + 1404, + 552, + 250, + 552 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 276, + 1814, + 1402, + 1814, + 1402, + 2008, + 276, + 2008 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 251, + 1324, + 1404, + 1324, + 1404, + 1447, + 251, + 1447 + ], + "score": 0.957 + }, + { + "category_id": 0, + "poly": [ + 263, + 1688, + 658, + 1688, + 658, + 1725, + 263, + 1725 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 290, + 1757, + 531, + 1757, + 531, + 1790, + 290, + 1790 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.726 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.097 + }, + { + "category_id": 13, + "poly": [ + 424, + 1845, + 625, + 1845, + 625, + 1881, + 424, + 1881 + ], + "score": 0.93, + "latex": "\\{ O _ { j , 1 } , \\dotsc , O _ { j , k _ { j } } \\}" + }, + { + "category_id": 13, + "poly": [ + 783, + 1941, + 845, + 1941, + 845, + 1984, + 783, + 1984 + ], + "score": 0.92, + "latex": "T _ { j , i } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 1130, + 1846, + 1207, + 1846, + 1207, + 1879, + 1130, + 1879 + ], + "score": 0.91, + "latex": "O _ { j , i - 1 }" + }, + { + "category_id": 13, + "poly": [ + 331, + 1945, + 386, + 1945, + 386, + 1980, + 331, + 1980 + ], + "score": 0.91, + "latex": "M _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 662, + 1909, + 709, + 1909, + 709, + 1940, + 662, + 1940 + ], + "score": 0.9, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 832, + 1847, + 880, + 1847, + 880, + 1880, + 832, + 1880 + ], + "score": 0.9, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 1048, + 1816, + 1077, + 1816, + 1077, + 1848, + 1048, + 1848 + ], + "score": 0.88, + "latex": "J _ { j }" + }, + { + "category_id": 13, + "poly": [ + 1113, + 1946, + 1178, + 1946, + 1178, + 1975, + 1113, + 1975 + ], + "score": 0.88, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 1371, + 1816, + 1400, + 1816, + 1400, + 1848, + 1371, + 1848 + ], + "score": 0.87, + "latex": "k _ { j }" + }, + { + "category_id": 13, + "poly": [ + 324, + 1817, + 403, + 1817, + 403, + 1844, + 324, + 1844 + ], + "score": 0.87, + "latex": "n \\times m" + }, + { + "category_id": 13, + "poly": [ + 659, + 1823, + 678, + 1823, + 678, + 1842, + 659, + 1842 + ], + "score": 0.76, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 785, + 1822, + 812, + 1822, + 812, + 1842, + 785, + 1842 + ], + "score": 0.73, + "latex": "m" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1685.0, + 662.0, + 1685.0, + 662.0, + 1730.0, + 251.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1754.0, + 535.0, + 1754.0, + 535.0, + 1796.0, + 293.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 892.0, + 277.0, + 892.0, + 277.0, + 917.0, + 247.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 884.0, + 1405.0, + 884.0, + 1405.0, + 920.0, + 292.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 921.0, + 277.0, + 921.0, + 277.0, + 948.0, + 247.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 912.0, + 1408.0, + 912.0, + 1408.0, + 953.0, + 291.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 951.0, + 277.0, + 951.0, + 277.0, + 978.0, + 249.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 944.0, + 1405.0, + 944.0, + 1405.0, + 979.0, + 293.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 983.0, + 277.0, + 983.0, + 277.0, + 1008.0, + 247.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 974.0, + 1407.0, + 974.0, + 1407.0, + 1013.0, + 293.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1012.0, + 277.0, + 1012.0, + 277.0, + 1039.0, + 247.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1006.0, + 1405.0, + 1006.0, + 1405.0, + 1040.0, + 292.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1041.0, + 280.0, + 1041.0, + 280.0, + 1070.0, + 246.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1036.0, + 1407.0, + 1036.0, + 1407.0, + 1070.0, + 291.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1074.0, + 279.0, + 1074.0, + 279.0, + 1099.0, + 247.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1066.0, + 1407.0, + 1066.0, + 1407.0, + 1102.0, + 292.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1102.0, + 280.0, + 1102.0, + 280.0, + 1131.0, + 246.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1095.0, + 1407.0, + 1095.0, + 1407.0, + 1135.0, + 292.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1133.0, + 277.0, + 1133.0, + 277.0, + 1159.0, + 247.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1127.0, + 1407.0, + 1127.0, + 1407.0, + 1162.0, + 292.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1164.0, + 279.0, + 1164.0, + 279.0, + 1189.0, + 247.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1156.0, + 1405.0, + 1156.0, + 1405.0, + 1190.0, + 293.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1194.0, + 277.0, + 1194.0, + 277.0, + 1221.0, + 247.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1183.0, + 1408.0, + 1183.0, + 1408.0, + 1227.0, + 291.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1223.0, + 280.0, + 1223.0, + 280.0, + 1252.0, + 246.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1219.0, + 1407.0, + 1219.0, + 1407.0, + 1251.0, + 296.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1255.0, + 277.0, + 1255.0, + 277.0, + 1280.0, + 249.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1250.0, + 1409.0, + 1250.0, + 1409.0, + 1281.0, + 295.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1285.0, + 276.0, + 1285.0, + 276.0, + 1312.0, + 249.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1279.0, + 1410.0, + 1279.0, + 1410.0, + 1313.0, + 293.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 573.0, + 279.0, + 573.0, + 279.0, + 598.0, + 249.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 565.0, + 1406.0, + 565.0, + 1406.0, + 602.0, + 292.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 603.0, + 277.0, + 603.0, + 277.0, + 630.0, + 247.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 596.0, + 1407.0, + 596.0, + 1407.0, + 634.0, + 293.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 631.0, + 279.0, + 631.0, + 279.0, + 662.0, + 246.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 624.0, + 1406.0, + 624.0, + 1406.0, + 665.0, + 291.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 662.0, + 279.0, + 662.0, + 279.0, + 692.0, + 246.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 659.0, + 1404.0, + 659.0, + 1404.0, + 691.0, + 296.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 693.0, + 279.0, + 693.0, + 279.0, + 724.0, + 246.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 686.0, + 1406.0, + 686.0, + 1406.0, + 725.0, + 292.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 725.0, + 277.0, + 725.0, + 277.0, + 751.0, + 247.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 718.0, + 1408.0, + 718.0, + 1408.0, + 754.0, + 293.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 753.0, + 280.0, + 753.0, + 280.0, + 782.0, + 246.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 748.0, + 1406.0, + 748.0, + 1406.0, + 783.0, + 294.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 786.0, + 277.0, + 786.0, + 277.0, + 812.0, + 247.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 778.0, + 1407.0, + 778.0, + 1407.0, + 817.0, + 291.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 813.0, + 280.0, + 813.0, + 280.0, + 843.0, + 246.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 811.0, + 1404.0, + 811.0, + 1404.0, + 843.0, + 294.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 845.0, + 279.0, + 845.0, + 279.0, + 872.0, + 249.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 840.0, + 1373.0, + 840.0, + 1373.0, + 875.0, + 293.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1467.0, + 278.0, + 1467.0, + 278.0, + 1492.0, + 249.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1459.0, + 1410.0, + 1459.0, + 1410.0, + 1496.0, + 292.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1496.0, + 278.0, + 1496.0, + 278.0, + 1522.0, + 248.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1489.0, + 1407.0, + 1489.0, + 1407.0, + 1526.0, + 294.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1527.0, + 278.0, + 1527.0, + 278.0, + 1553.0, + 248.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1522.0, + 1404.0, + 1522.0, + 1404.0, + 1554.0, + 295.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1558.0, + 278.0, + 1558.0, + 278.0, + 1584.0, + 248.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1549.0, + 1407.0, + 1549.0, + 1407.0, + 1589.0, + 290.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1588.0, + 278.0, + 1588.0, + 278.0, + 1614.0, + 248.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1582.0, + 1406.0, + 1582.0, + 1406.0, + 1616.0, + 292.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1618.0, + 278.0, + 1618.0, + 278.0, + 1645.0, + 248.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1613.0, + 1188.0, + 1613.0, + 1188.0, + 1645.0, + 294.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 210.0, + 277.0, + 210.0, + 277.0, + 235.0, + 248.0, + 235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 205.0, + 1406.0, + 205.0, + 1406.0, + 236.0, + 294.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 239.0, + 277.0, + 239.0, + 277.0, + 265.0, + 248.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 235.0, + 1406.0, + 235.0, + 1406.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 269.0, + 277.0, + 269.0, + 277.0, + 296.0, + 248.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 263.0, + 1408.0, + 263.0, + 1408.0, + 298.0, + 293.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 301.0, + 278.0, + 301.0, + 278.0, + 326.0, + 248.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 295.0, + 1406.0, + 295.0, + 1406.0, + 330.0, + 293.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 331.0, + 277.0, + 331.0, + 277.0, + 355.0, + 248.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 324.0, + 1407.0, + 324.0, + 1407.0, + 360.0, + 291.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 361.0, + 278.0, + 361.0, + 278.0, + 387.0, + 248.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 354.0, + 1087.0, + 354.0, + 1087.0, + 391.0, + 291.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 405.0, + 278.0, + 405.0, + 278.0, + 431.0, + 248.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 399.0, + 1404.0, + 399.0, + 1404.0, + 434.0, + 291.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 437.0, + 278.0, + 437.0, + 278.0, + 462.0, + 248.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 430.0, + 1404.0, + 430.0, + 1404.0, + 463.0, + 294.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 466.0, + 278.0, + 466.0, + 278.0, + 492.0, + 248.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 458.0, + 1407.0, + 458.0, + 1407.0, + 496.0, + 293.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 497.0, + 278.0, + 497.0, + 278.0, + 523.0, + 248.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 488.0, + 1407.0, + 488.0, + 1407.0, + 526.0, + 293.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 527.0, + 278.0, + 527.0, + 278.0, + 553.0, + 248.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 522.0, + 559.0, + 522.0, + 559.0, + 553.0, + 293.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1811.0, + 323.0, + 1811.0, + 323.0, + 1851.0, + 292.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1811.0, + 658.0, + 1811.0, + 658.0, + 1851.0, + 404.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1811.0, + 784.0, + 1811.0, + 784.0, + 1851.0, + 679.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1811.0, + 1047.0, + 1811.0, + 1047.0, + 1851.0, + 813.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1811.0, + 1370.0, + 1811.0, + 1370.0, + 1851.0, + 1078.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1811.0, + 1405.0, + 1811.0, + 1405.0, + 1851.0, + 1401.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1843.0, + 423.0, + 1843.0, + 423.0, + 1883.0, + 294.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 1843.0, + 831.0, + 1843.0, + 831.0, + 1883.0, + 626.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1843.0, + 1129.0, + 1843.0, + 1129.0, + 1883.0, + 881.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 1843.0, + 1406.0, + 1843.0, + 1406.0, + 1883.0, + 1208.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1873.0, + 1405.0, + 1873.0, + 1405.0, + 1914.0, + 292.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1904.0, + 661.0, + 1904.0, + 661.0, + 1941.0, + 291.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1904.0, + 1406.0, + 1904.0, + 1406.0, + 1941.0, + 710.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1926.0, + 330.0, + 1926.0, + 330.0, + 1990.0, + 283.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 1926.0, + 782.0, + 1926.0, + 782.0, + 1990.0, + 387.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1926.0, + 1112.0, + 1926.0, + 1112.0, + 1990.0, + 846.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1926.0, + 1415.0, + 1926.0, + 1415.0, + 1990.0, + 1179.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2012.0, + 294.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1328.0, + 278.0, + 1328.0, + 278.0, + 1358.0, + 249.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1323.0, + 1403.0, + 1323.0, + 1403.0, + 1358.0, + 295.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1359.0, + 279.0, + 1359.0, + 279.0, + 1387.0, + 249.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1351.0, + 1407.0, + 1351.0, + 1407.0, + 1391.0, + 291.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1390.0, + 279.0, + 1390.0, + 279.0, + 1418.0, + 248.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1386.0, + 1404.0, + 1386.0, + 1404.0, + 1420.0, + 294.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1420.0, + 278.0, + 1420.0, + 278.0, + 1447.0, + 249.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1413.0, + 1051.0, + 1413.0, + 1051.0, + 1451.0, + 291.0, + 1451.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 240, + 1611, + 1405, + 1611, + 1405, + 2008, + 240, + 2008 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 244, + 1353, + 1405, + 1353, + 1405, + 1507, + 244, + 1507 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 249, + 201, + 1405, + 201, + 1405, + 305, + 249, + 305 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 250, + 1141, + 1404, + 1141, + 1404, + 1340, + 250, + 1340 + ], + "score": 0.97 + }, + { + "category_id": 4, + "poly": [ + 297, + 1008, + 1402, + 1008, + 1402, + 1103, + 297, + 1103 + ], + "score": 0.963 + }, + { + "category_id": 3, + "poly": [ + 305, + 803, + 1386, + 803, + 1386, + 983, + 305, + 983 + ], + "score": 0.952 + }, + { + "category_id": 6, + "poly": [ + 667, + 338, + 1033, + 338, + 1033, + 370, + 667, + 370 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 255, + 1551, + 646, + 1551, + 646, + 1584, + 255, + 1584 + ], + "score": 0.891 + }, + { + "category_id": 5, + "poly": [ + 376, + 427, + 834, + 427, + 834, + 712, + 376, + 712 + ], + "score": 0.881, + "html": "
JobOperationM1M2M3
Job 1O1,13
O1.25
O1.34
Job 2O2,12
02,24
O2,33
Job 3O3,13
03.22
" + }, + { + "category_id": 5, + "poly": [ + 855, + 423, + 1317, + 423, + 1317, + 712, + 855, + 712 + ], + "score": 0.776, + "html": "
JobOperationM1M2M3
Job 1O1,132
O1.235
O1.343
Job 2O2,12
02.24
02.33
Job 3O3,134
03.222
" + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.645 + }, + { + "category_id": 6, + "poly": [ + 478, + 395, + 733, + 395, + 733, + 423, + 478, + 423 + ], + "score": 0.608 + }, + { + "category_id": 6, + "poly": [ + 950, + 395, + 1221, + 395, + 1221, + 423, + 950, + 423 + ], + "score": 0.558 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.484 + }, + { + "category_id": 13, + "poly": [ + 588, + 203, + 760, + 203, + 760, + 238, + 588, + 238 + ], + "score": 0.94, + "latex": "\\{ M _ { 1 } , M _ { 2 } , M _ { 3 } \\}" + }, + { + "category_id": 13, + "poly": [ + 1174, + 1203, + 1400, + 1203, + 1400, + 1249, + 1174, + 1249 + ], + "score": 0.93, + "latex": "\\tau _ { j , i } ^ { ( c ) } = \\tau _ { j , i } ^ { ( s ) } + T _ { j , i } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 578, + 1204, + 829, + 1204, + 829, + 1247, + 578, + 1247 + ], + "score": 0.93, + "latex": "T ^ { ( m k s p ) } = \\operatorname* { m a x } ( \\tau _ { j , i } ^ { ( c ) } )" + }, + { + "category_id": 13, + "poly": [ + 516, + 527, + 568, + 527, + 568, + 556, + 516, + 556 + ], + "score": 0.92, + "latex": "\\underline { { O _ { 1 , 3 } } }" + }, + { + "category_id": 13, + "poly": [ + 998, + 527, + 1049, + 527, + 1049, + 556, + 998, + 556 + ], + "score": 0.92, + "latex": "\\underline { { O _ { 1 , 3 } } }" + }, + { + "category_id": 13, + "poly": [ + 516, + 619, + 568, + 619, + 568, + 648, + 516, + 648 + ], + "score": 0.91, + "latex": "\\underline { { O _ { 2 , 3 } } }" + }, + { + "category_id": 13, + "poly": [ + 517, + 652, + 568, + 652, + 568, + 680, + 517, + 680 + ], + "score": 0.91, + "latex": "\\underline { { O _ { 3 , 1 } } }" + }, + { + "category_id": 13, + "poly": [ + 813, + 242, + 868, + 242, + 868, + 275, + 813, + 275 + ], + "score": 0.91, + "latex": "M _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 955, + 234, + 1001, + 234, + 1001, + 280, + 955, + 280 + ], + "score": 0.91, + "latex": "\\tau _ { j , i } ^ { ( s ) }" + }, + { + "category_id": 13, + "poly": [ + 1116, + 1825, + 1195, + 1825, + 1195, + 1855, + 1116, + 1855 + ], + "score": 0.91, + "latex": "\\tau _ { 0 } = 0" + }, + { + "category_id": 13, + "poly": [ + 459, + 1384, + 508, + 1384, + 508, + 1417, + 459, + 1417 + ], + "score": 0.91, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 998, + 619, + 1049, + 619, + 1049, + 648, + 998, + 648 + ], + "score": 0.91, + "latex": "\\underline { { O _ { 2 , 3 } } }" + }, + { + "category_id": 13, + "poly": [ + 1309, + 1916, + 1366, + 1916, + 1366, + 1948, + 1309, + 1948 + ], + "score": 0.91, + "latex": "S _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 517, + 560, + 568, + 560, + 568, + 587, + 517, + 587 + ], + "score": 0.91, + "latex": "O _ { 2 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 1036, + 1210, + 1084, + 1210, + 1084, + 1246, + 1036, + 1246 + ], + "score": 0.91, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 1112, + 435, + 1151, + 435, + 1151, + 461, + 1112, + 461 + ], + "score": 0.91, + "latex": "M _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 517, + 681, + 568, + 681, + 568, + 710, + 517, + 710 + ], + "score": 0.91, + "latex": "\\underline { { O _ { 3 , 2 } } }" + }, + { + "category_id": 13, + "poly": [ + 421, + 242, + 470, + 242, + 470, + 276, + 421, + 276 + ], + "score": 0.9, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 1185, + 434, + 1224, + 434, + 1224, + 461, + 1185, + 461 + ], + "score": 0.9, + "latex": "M _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 631, + 434, + 670, + 434, + 670, + 461, + 631, + 461 + ], + "score": 0.9, + "latex": "M _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 516, + 589, + 568, + 589, + 568, + 617, + 516, + 617 + ], + "score": 0.9, + "latex": "O _ { 2 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1304, + 392, + 1304, + 392, + 1335, + 297, + 1335 + ], + "score": 0.9, + "latex": "T ^ { ( m k s p ) }" + }, + { + "category_id": 13, + "poly": [ + 998, + 560, + 1049, + 560, + 1049, + 586, + 998, + 586 + ], + "score": 0.9, + "latex": "O _ { 2 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 648, + 1246, + 696, + 1246, + 696, + 1278, + 648, + 1278 + ], + "score": 0.9, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 998, + 652, + 1049, + 652, + 1049, + 679, + 998, + 679 + ], + "score": 0.9, + "latex": "O _ { 3 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 804, + 1175, + 874, + 1175, + 874, + 1202, + 804, + 1202 + ], + "score": 0.9, + "latex": "\\tau = 0" + }, + { + "category_id": 13, + "poly": [ + 516, + 497, + 568, + 497, + 568, + 526, + 516, + 526 + ], + "score": 0.89, + "latex": "\\underline { { O _ { 1 , 2 } } }" + }, + { + "category_id": 13, + "poly": [ + 1257, + 435, + 1297, + 435, + 1297, + 461, + 1257, + 461 + ], + "score": 0.89, + "latex": "M _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 998, + 681, + 1049, + 681, + 1049, + 710, + 998, + 710 + ], + "score": 0.89, + "latex": "\\underline { { O _ { 3 , 2 } } }" + }, + { + "category_id": 13, + "poly": [ + 781, + 1414, + 849, + 1414, + 849, + 1443, + 781, + 1443 + ], + "score": 0.89, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 998, + 589, + 1049, + 589, + 1049, + 618, + 998, + 618 + ], + "score": 0.89, + "latex": "\\underline { { O _ { 2 , 2 } } }" + }, + { + "category_id": 13, + "poly": [ + 703, + 434, + 742, + 434, + 742, + 461, + 703, + 461 + ], + "score": 0.89, + "latex": "M _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 517, + 466, + 568, + 466, + 568, + 495, + 517, + 495 + ], + "score": 0.89, + "latex": "O _ { 1 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 696, + 1357, + 779, + 1357, + 779, + 1383, + 696, + 1383 + ], + "score": 0.89, + "latex": "n \\times m" + }, + { + "category_id": 13, + "poly": [ + 326, + 1356, + 411, + 1356, + 411, + 1383, + 326, + 1383 + ], + "score": 0.88, + "latex": "n \\times m" + }, + { + "category_id": 13, + "poly": [ + 996, + 1855, + 1081, + 1855, + 1081, + 1886, + 996, + 1886 + ], + "score": 0.88, + "latex": "( M , O )" + }, + { + "category_id": 13, + "poly": [ + 998, + 497, + 1049, + 497, + 1049, + 526, + 998, + 526 + ], + "score": 0.88, + "latex": "\\underline { { O _ { 1 , 2 } } }" + }, + { + "category_id": 13, + "poly": [ + 623, + 1917, + 654, + 1917, + 654, + 1946, + 623, + 1946 + ], + "score": 0.87, + "latex": "C _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1288, + 1856, + 1320, + 1856, + 1320, + 1885, + 1288, + 1885 + ], + "score": 0.87, + "latex": "C _ { t }" + }, + { + "category_id": 13, + "poly": [ + 998, + 466, + 1049, + 466, + 1049, + 495, + 998, + 495 + ], + "score": 0.87, + "latex": "O _ { 1 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 520, + 1734, + 550, + 1734, + 550, + 1764, + 520, + 1764 + ], + "score": 0.87, + "latex": "S _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1369, + 1794, + 1401, + 1794, + 1401, + 1825, + 1369, + 1825 + ], + "score": 0.87, + "latex": "S _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1308, + 1825, + 1337, + 1825, + 1337, + 1855, + 1308, + 1855 + ], + "score": 0.87, + "latex": "S _ { t }" + }, + { + "category_id": 13, + "poly": [ + 776, + 434, + 815, + 434, + 815, + 461, + 776, + 461 + ], + "score": 0.86, + "latex": "M _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 978, + 1890, + 1002, + 1890, + 1002, + 1915, + 978, + 1915 + ], + "score": 0.84, + "latex": "\\tau _ { t }" + }, + { + "category_id": 13, + "poly": [ + 939, + 1738, + 965, + 1738, + 965, + 1764, + 939, + 1764 + ], + "score": 0.84, + "latex": "\\tau _ { t }" + }, + { + "category_id": 13, + "poly": [ + 816, + 1798, + 841, + 1798, + 841, + 1824, + 816, + 1824 + ], + "score": 0.84, + "latex": "\\tau _ { t }" + }, + { + "category_id": 13, + "poly": [ + 631, + 1768, + 657, + 1768, + 657, + 1794, + 631, + 1794 + ], + "score": 0.84, + "latex": "\\tau _ { t }" + }, + { + "category_id": 13, + "poly": [ + 915, + 1886, + 948, + 1886, + 948, + 1912, + 915, + 1912 + ], + "score": 0.81, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 419, + 1886, + 445, + 1886, + 445, + 1912, + 419, + 1912 + ], + "score": 0.8, + "latex": "O" + }, + { + "category_id": 13, + "poly": [ + 540, + 396, + 600, + 396, + 600, + 422, + 540, + 422 + ], + "score": 0.75, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 1015, + 395, + 1075, + 395, + 1075, + 422, + 1015, + 422 + ], + "score": 0.73, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 297, + 1736, + 311, + 1736, + 311, + 1761, + 297, + 1761 + ], + "score": 0.72, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 393, + 590, + 454, + 590, + 454, + 611, + 393, + 611 + ], + "score": 0.49, + "latex": "\\mathrm { J o b } 2" + }, + { + "category_id": 13, + "poly": [ + 1285, + 859, + 1319, + 859, + 1319, + 884, + 1285, + 884 + ], + "score": 0.48, + "latex": "O _ { 1 , 3 }" + }, + { + "category_id": 13, + "poly": [ + 953, + 1041, + 998, + 1041, + 998, + 1069, + 953, + 1069 + ], + "score": 0.43, + "latex": "3 \\mathrm { x } 3" + }, + { + "category_id": 13, + "poly": [ + 708, + 274, + 754, + 274, + 754, + 301, + 708, + 301 + ], + "score": 0.43, + "latex": "3 { \\tt X } 3" + }, + { + "category_id": 13, + "poly": [ + 780, + 864, + 814, + 864, + 814, + 885, + 780, + 885 + ], + "score": 0.43, + "latex": "O _ { 1 , 3 }" + }, + { + "category_id": 13, + "poly": [ + 802, + 1010, + 848, + 1010, + 848, + 1039, + 802, + 1039 + ], + "score": 0.38, + "latex": "3 \\mathrm { x } 3" + }, + { + "category_id": 13, + "poly": [ + 644, + 620, + 657, + 620, + 657, + 640, + 644, + 640 + ], + "score": 0.29, + "latex": "^ 3" + }, + { + "category_id": 13, + "poly": [ + 393, + 667, + 454, + 667, + 454, + 688, + 393, + 688 + ], + "score": 0.28, + "latex": "\\operatorname { J o b } 3" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1007.0, + 801.0, + 1007.0, + 801.0, + 1043.0, + 295.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1007.0, + 1404.0, + 1007.0, + 1404.0, + 1043.0, + 849.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1040.0, + 952.0, + 1040.0, + 952.0, + 1073.0, + 293.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1040.0, + 1404.0, + 1040.0, + 1404.0, + 1073.0, + 999.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1068.0, + 717.0, + 1068.0, + 717.0, + 1104.0, + 295.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 819.0, + 475.0, + 819.0, + 475.0, + 858.0, + 307.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 821.0, + 747.0, + 821.0, + 747.0, + 861.0, + 582.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 815.0, + 1022.0, + 815.0, + 1022.0, + 855.0, + 857.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 824.0, + 1159.0, + 824.0, + 1159.0, + 854.0, + 1128.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 822.0, + 1209.0, + 822.0, + 1209.0, + 855.0, + 1165.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 829.0, + 1244.0, + 829.0, + 1244.0, + 848.0, + 1219.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 858.0, + 335.0, + 858.0, + 335.0, + 883.0, + 308.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 856.0, + 508.0, + 856.0, + 508.0, + 888.0, + 463.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 858.0, + 612.0, + 858.0, + 612.0, + 887.0, + 580.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 854.0, + 699.0, + 854.0, + 699.0, + 892.0, + 648.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 853.0, + 972.0, + 853.0, + 972.0, + 882.0, + 857.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 859.0, + 1158.0, + 859.0, + 1158.0, + 884.0, + 1130.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 860.0, + 1203.0, + 860.0, + 1203.0, + 879.0, + 1173.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 867.0, + 1266.0, + 867.0, + 1266.0, + 879.0, + 1245.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 857.0, + 1284.0, + 857.0, + 1284.0, + 887.0, + 1281.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 857.0, + 1326.0, + 857.0, + 1326.0, + 887.0, + 1320.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 893.0, + 334.0, + 893.0, + 334.0, + 913.0, + 310.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 886.0, + 374.0, + 886.0, + 374.0, + 915.0, + 339.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 883.0, + 448.0, + 883.0, + 448.0, + 920.0, + 399.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 892.0, + 611.0, + 892.0, + 611.0, + 917.0, + 582.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 889.0, + 650.0, + 889.0, + 650.0, + 918.0, + 612.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 886.0, + 757.0, + 886.0, + 757.0, + 919.0, + 713.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 889.0, + 914.0, + 889.0, + 914.0, + 908.0, + 890.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 889.0, + 964.0, + 889.0, + 964.0, + 908.0, + 934.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 882.0, + 1018.0, + 882.0, + 1018.0, + 912.0, + 971.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 893.0, + 1154.0, + 893.0, + 1154.0, + 913.0, + 1132.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 924.0, + 346.0, + 924.0, + 346.0, + 936.0, + 335.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 918.0, + 394.0, + 918.0, + 394.0, + 942.0, + 362.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 922.0, + 435.0, + 922.0, + 435.0, + 938.0, + 420.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 917.0, + 526.0, + 917.0, + 526.0, + 942.0, + 444.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 924.0, + 622.0, + 924.0, + 622.0, + 942.0, + 607.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 920.0, + 667.0, + 920.0, + 667.0, + 944.0, + 634.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 918.0, + 710.0, + 918.0, + 710.0, + 945.0, + 672.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 924.0, + 748.0, + 924.0, + 748.0, + 942.0, + 732.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 922.0, + 782.0, + 922.0, + 782.0, + 942.0, + 759.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 919.0, + 841.0, + 919.0, + 841.0, + 944.0, + 814.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 924.0, + 895.0, + 924.0, + 895.0, + 941.0, + 880.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 918.0, + 991.0, + 918.0, + 991.0, + 944.0, + 905.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 923.0, + 1024.0, + 923.0, + 1024.0, + 941.0, + 1009.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 924.0, + 1169.0, + 924.0, + 1169.0, + 941.0, + 1154.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 919.0, + 1231.0, + 919.0, + 1231.0, + 944.0, + 1180.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 923.0, + 1258.0, + 923.0, + 1258.0, + 942.0, + 1237.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 924.0, + 1280.0, + 924.0, + 1280.0, + 941.0, + 1264.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 918.0, + 1349.0, + 918.0, + 1349.0, + 947.0, + 1301.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 948.0, + 454.0, + 948.0, + 454.0, + 988.0, + 408.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 948.0, + 729.0, + 948.0, + 729.0, + 987.0, + 682.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 950.0, + 1001.0, + 950.0, + 1001.0, + 985.0, + 959.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 948.0, + 1276.0, + 948.0, + 1276.0, + 987.0, + 1231.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1274.0, + 824.0, + 1318.0, + 824.0, + 1318.0, + 855.5, + 1274.0, + 855.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 854.0, + 423.0, + 854.0, + 423.0, + 889.0, + 374.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.75, + 857.5, + 816.75, + 857.5, + 816.75, + 889.5, + 774.75, + 889.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.25, + 888.0, + 484.25, + 888.0, + 484.25, + 916.0, + 453.25, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.25, + 889.0, + 689.25, + 889.0, + 689.25, + 916.0, + 651.25, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.75, + 891.5, + 877.75, + 891.5, + 877.75, + 909.5, + 858.75, + 909.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 891.0, + 1258.0, + 891.0, + 1258.0, + 913.5, + 1220.0, + 913.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 336.0, + 1033.0, + 336.0, + 1033.0, + 372.0, + 667.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1548.0, + 650.0, + 1548.0, + 650.0, + 1588.0, + 244.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 394.0, + 539.0, + 394.0, + 539.0, + 425.0, + 476.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 394.0, + 735.0, + 394.0, + 735.0, + 425.0, + 601.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 394.0, + 1014.0, + 394.0, + 1014.0, + 425.0, + 950.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 394.0, + 1224.0, + 394.0, + 1224.0, + 425.0, + 1076.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1618.0, + 277.0, + 1618.0, + 277.0, + 1643.0, + 241.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1609.0, + 1405.0, + 1609.0, + 1405.0, + 1648.0, + 290.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1649.0, + 278.0, + 1649.0, + 278.0, + 1674.0, + 240.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1641.0, + 1406.0, + 1641.0, + 1406.0, + 1677.0, + 293.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1677.0, + 278.0, + 1677.0, + 278.0, + 1706.0, + 240.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1672.0, + 1406.0, + 1672.0, + 1406.0, + 1709.0, + 294.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1707.0, + 278.0, + 1707.0, + 278.0, + 1737.0, + 240.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1700.0, + 1408.0, + 1700.0, + 1408.0, + 1738.0, + 290.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1738.0, + 278.0, + 1738.0, + 278.0, + 1767.0, + 239.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1732.0, + 296.0, + 1732.0, + 296.0, + 1768.0, + 293.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1732.0, + 519.0, + 1732.0, + 519.0, + 1768.0, + 312.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1732.0, + 938.0, + 1732.0, + 938.0, + 1768.0, + 551.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 1732.0, + 1405.0, + 1732.0, + 1405.0, + 1768.0, + 966.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1768.0, + 278.0, + 1768.0, + 278.0, + 1797.0, + 239.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1762.0, + 630.0, + 1762.0, + 630.0, + 1798.0, + 294.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 1762.0, + 1405.0, + 1762.0, + 1405.0, + 1798.0, + 658.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1800.0, + 277.0, + 1800.0, + 277.0, + 1825.0, + 240.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1793.0, + 815.0, + 1793.0, + 815.0, + 1829.0, + 291.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 1793.0, + 1368.0, + 1793.0, + 1368.0, + 1829.0, + 842.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1793.0, + 1405.0, + 1793.0, + 1405.0, + 1829.0, + 1402.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1828.0, + 280.0, + 1828.0, + 280.0, + 1857.0, + 239.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1822.0, + 1115.0, + 1822.0, + 1115.0, + 1859.0, + 293.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 1822.0, + 1307.0, + 1822.0, + 1307.0, + 1859.0, + 1196.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1822.0, + 1406.0, + 1822.0, + 1406.0, + 1859.0, + 1338.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1861.0, + 278.0, + 1861.0, + 278.0, + 1886.0, + 240.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1854.0, + 995.0, + 1854.0, + 995.0, + 1890.0, + 294.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 1854.0, + 1287.0, + 1854.0, + 1287.0, + 1890.0, + 1082.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 1854.0, + 1406.0, + 1854.0, + 1406.0, + 1890.0, + 1321.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1890.0, + 280.0, + 1890.0, + 280.0, + 1918.0, + 239.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1883.0, + 418.0, + 1883.0, + 418.0, + 1920.0, + 293.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1883.0, + 914.0, + 1883.0, + 914.0, + 1920.0, + 446.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 1883.0, + 977.0, + 1883.0, + 977.0, + 1920.0, + 949.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1883.0, + 1406.0, + 1883.0, + 1406.0, + 1920.0, + 1003.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1921.0, + 278.0, + 1921.0, + 278.0, + 1945.0, + 240.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1912.0, + 622.0, + 1912.0, + 622.0, + 1951.0, + 291.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 1912.0, + 1308.0, + 1912.0, + 1308.0, + 1951.0, + 655.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1367.0, + 1912.0, + 1409.0, + 1912.0, + 1409.0, + 1951.0, + 1367.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1952.0, + 278.0, + 1952.0, + 278.0, + 1976.0, + 240.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1944.0, + 1406.0, + 1944.0, + 1406.0, + 1979.0, + 293.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1980.0, + 278.0, + 1980.0, + 278.0, + 2009.0, + 239.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1975.0, + 1157.0, + 1975.0, + 1157.0, + 2009.0, + 291.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1359.0, + 278.0, + 1359.0, + 278.0, + 1385.0, + 248.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1353.0, + 325.0, + 1353.0, + 325.0, + 1386.0, + 294.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 1353.0, + 695.0, + 1353.0, + 695.0, + 1386.0, + 412.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1353.0, + 1406.0, + 1353.0, + 1406.0, + 1386.0, + 780.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1388.0, + 278.0, + 1388.0, + 278.0, + 1417.0, + 240.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1382.0, + 458.0, + 1382.0, + 458.0, + 1419.0, + 294.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 1382.0, + 1406.0, + 1382.0, + 1406.0, + 1419.0, + 509.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1419.0, + 277.0, + 1419.0, + 277.0, + 1445.0, + 241.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1413.0, + 780.0, + 1413.0, + 780.0, + 1449.0, + 292.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1413.0, + 1406.0, + 1413.0, + 1406.0, + 1449.0, + 850.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1449.0, + 278.0, + 1449.0, + 278.0, + 1478.0, + 240.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1441.0, + 1412.0, + 1441.0, + 1412.0, + 1483.0, + 291.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1480.0, + 278.0, + 1480.0, + 278.0, + 1505.0, + 241.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1475.0, + 869.0, + 1475.0, + 869.0, + 1508.0, + 294.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 202.0, + 587.0, + 202.0, + 587.0, + 238.0, + 250.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 202.0, + 1405.0, + 202.0, + 1405.0, + 238.0, + 761.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 247.0, + 279.0, + 247.0, + 279.0, + 272.0, + 249.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 279.0, + 279.0, + 279.0, + 279.0, + 304.0, + 249.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 272.0, + 707.0, + 272.0, + 707.0, + 308.0, + 292.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 272.0, + 1271.0, + 272.0, + 1271.0, + 308.0, + 755.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 274.25, + 217.5, + 1421.25, + 217.5, + 1421.25, + 301.5, + 274.25, + 301.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1148.0, + 278.0, + 1148.0, + 278.0, + 1174.0, + 248.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1143.0, + 1406.0, + 1143.0, + 1406.0, + 1176.0, + 295.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1179.0, + 277.0, + 1179.0, + 277.0, + 1206.0, + 248.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 1171.0, + 577.0, + 1171.0, + 577.0, + 1269.0, + 281.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1171.0, + 1035.0, + 1171.0, + 1035.0, + 1269.0, + 875.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 1171.0, + 1173.0, + 1171.0, + 1173.0, + 1269.0, + 1085.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1171.0, + 1406.0, + 1171.0, + 1406.0, + 1269.0, + 1401.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1216.0, + 277.0, + 1216.0, + 277.0, + 1245.0, + 248.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1250.0, + 278.0, + 1250.0, + 278.0, + 1277.0, + 248.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1243.0, + 647.0, + 1243.0, + 647.0, + 1281.0, + 294.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1243.0, + 1407.0, + 1243.0, + 1407.0, + 1281.0, + 697.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1279.0, + 280.0, + 1279.0, + 280.0, + 1310.0, + 246.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1271.0, + 1406.0, + 1271.0, + 1406.0, + 1311.0, + 291.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1313.0, + 278.0, + 1313.0, + 278.0, + 1339.0, + 248.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1299.0, + 296.0, + 1299.0, + 296.0, + 1347.0, + 290.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1299.0, + 985.0, + 1299.0, + 985.0, + 1347.0, + 393.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 1196.5, + 1409.0, + 1196.5, + 1409.0, + 1240.5, + 1291.0, + 1240.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1224.0, + 1253.0, + 1224.0, + 1253.0, + 1246.5, + 1213.0, + 1246.5 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 243, + 1403, + 1405, + 1403, + 1405, + 1526, + 243, + 1526 + ], + "score": 0.972 + }, + { + "category_id": 3, + "poly": [ + 331, + 201, + 1341, + 201, + 1341, + 809, + 331, + 809 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 243, + 1158, + 1403, + 1158, + 1403, + 1282, + 243, + 1282 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 241, + 930, + 1405, + 930, + 1405, + 1144, + 241, + 1144 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 239, + 1628, + 1405, + 1628, + 1405, + 2008, + 239, + 2008 + ], + "score": 0.937 + }, + { + "category_id": 4, + "poly": [ + 412, + 835, + 1285, + 835, + 1285, + 869, + 412, + 869 + ], + "score": 0.9 + }, + { + "category_id": 0, + "poly": [ + 248, + 1330, + 558, + 1330, + 558, + 1369, + 248, + 1369 + ], + "score": 0.898 + }, + { + "category_id": 0, + "poly": [ + 262, + 1570, + 605, + 1570, + 605, + 1603, + 262, + 1603 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 841, + 2062, + 858, + 2062, + 858, + 2084, + 841, + 2084 + ], + "score": 0.758 + }, + { + "category_id": 13, + "poly": [ + 630, + 1751, + 684, + 1751, + 684, + 1783, + 630, + 1783 + ], + "score": 0.93, + "latex": "O _ { 1 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 698, + 1813, + 819, + 1813, + 819, + 1855, + 698, + 1855 + ], + "score": 0.93, + "latex": "T _ { 2 , 1 } ^ { ( o p ) } = 4" + }, + { + "category_id": 13, + "poly": [ + 1349, + 1977, + 1403, + 1977, + 1403, + 2010, + 1349, + 2010 + ], + "score": 0.92, + "latex": "O _ { 3 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 1008, + 1781, + 1062, + 1781, + 1062, + 1815, + 1008, + 1815 + ], + "score": 0.92, + "latex": "O _ { 2 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 736, + 1751, + 790, + 1751, + 790, + 1783, + 736, + 1783 + ], + "score": 0.91, + "latex": "O _ { 2 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 727, + 1917, + 779, + 1917, + 779, + 1949, + 727, + 1949 + ], + "score": 0.91, + "latex": "O _ { 2 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 650, + 1886, + 703, + 1886, + 703, + 1919, + 650, + 1919 + ], + "score": 0.91, + "latex": "O _ { 1 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 585, + 1721, + 664, + 1721, + 664, + 1750, + 585, + 1750 + ], + "score": 0.91, + "latex": "\\tau _ { 3 } = 3" + }, + { + "category_id": 13, + "poly": [ + 704, + 1857, + 782, + 1857, + 782, + 1885, + 704, + 1885 + ], + "score": 0.9, + "latex": "\\tau _ { 3 } = 3" + }, + { + "category_id": 13, + "poly": [ + 1043, + 1854, + 1175, + 1854, + 1175, + 1888, + 1043, + 1888 + ], + "score": 0.9, + "latex": "4 - ( 3 - 2 )" + }, + { + "category_id": 13, + "poly": [ + 952, + 1887, + 1004, + 1887, + 1004, + 1918, + 952, + 1918 + ], + "score": 0.9, + "latex": "O _ { 2 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 889, + 993, + 924, + 993, + 924, + 1022, + 889, + 1022 + ], + "score": 0.9, + "latex": "C _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 901, + 1977, + 954, + 1977, + 954, + 2010, + 901, + 2010 + ], + "score": 0.89, + "latex": "O _ { 2 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 461, + 1916, + 493, + 1916, + 493, + 1946, + 461, + 1946 + ], + "score": 0.89, + "latex": "S _ { 4 }" + }, + { + "category_id": 13, + "poly": [ + 754, + 1887, + 807, + 1887, + 807, + 1918, + 754, + 1918 + ], + "score": 0.89, + "latex": "O _ { 2 , 1 }" + }, + { + "category_id": 13, + "poly": [ + 468, + 1720, + 500, + 1720, + 500, + 1750, + 468, + 1750 + ], + "score": 0.89, + "latex": "S _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 393, + 1886, + 426, + 1886, + 426, + 1915, + 393, + 1915 + ], + "score": 0.89, + "latex": "S _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 719, + 1783, + 750, + 1783, + 750, + 1811, + 719, + 1811 + ], + "score": 0.88, + "latex": "S _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1819, + 385, + 1819, + 385, + 1851, + 297, + 1851 + ], + "score": 0.88, + "latex": "\\tau _ { 3 } = 3" + }, + { + "category_id": 13, + "poly": [ + 1179, + 837, + 1211, + 837, + 1211, + 866, + 1179, + 866 + ], + "score": 0.88, + "latex": "S _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1100, + 1054, + 1132, + 1054, + 1132, + 1083, + 1100, + 1083 + ], + "score": 0.88, + "latex": "S _ { 9 }" + }, + { + "category_id": 13, + "poly": [ + 1243, + 837, + 1275, + 837, + 1275, + 866, + 1243, + 866 + ], + "score": 0.87, + "latex": "S _ { 8 }" + }, + { + "category_id": 13, + "poly": [ + 1076, + 1978, + 1107, + 1978, + 1107, + 2006, + 1076, + 2006 + ], + "score": 0.87, + "latex": "S _ { 5 }" + }, + { + "category_id": 13, + "poly": [ + 1051, + 963, + 1083, + 963, + 1083, + 991, + 1051, + 991 + ], + "score": 0.87, + "latex": "S _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1063, + 1024, + 1095, + 1024, + 1095, + 1053, + 1063, + 1053 + ], + "score": 0.87, + "latex": "S _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1128, + 1024, + 1160, + 1024, + 1160, + 1053, + 1128, + 1053 + ], + "score": 0.87, + "latex": "S _ { 9 }" + }, + { + "category_id": 13, + "poly": [ + 1038, + 1724, + 1067, + 1724, + 1067, + 1750, + 1038, + 1750 + ], + "score": 0.85, + "latex": "\\tau _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 352, + 1949, + 381, + 1949, + 381, + 1976, + 352, + 1976 + ], + "score": 0.85, + "latex": "\\tau _ { 4 }" + }, + { + "category_id": 13, + "poly": [ + 912, + 1057, + 937, + 1057, + 937, + 1083, + 912, + 1083 + ], + "score": 0.84, + "latex": "\\tau _ { t }" + }, + { + "category_id": 13, + "poly": [ + 971, + 991, + 1399, + 991, + 1399, + 1025, + 971, + 1025 + ], + "score": 0.83, + "latex": "\\{ ( M _ { 1 } , O _ { 1 , 1 } ) , ( M _ { 3 } , O _ { 2 , 1 } ) , ( \\bar { M _ { 1 } } , O _ { 3 , 1 } ) \\}" + }, + { + "category_id": 13, + "poly": [ + 866, + 1219, + 946, + 1219, + 946, + 1250, + 866, + 1250 + ], + "score": 0.37, + "latex": "( M D P )" + }, + { + "category_id": 13, + "poly": [ + 1204, + 368, + 1232, + 368, + 1232, + 391, + 1204, + 391 + ], + "score": 0.36, + "latex": "S _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 971, + 992, + 1110, + 992, + 1110, + 1025, + 971, + 1025 + ], + "score": 0.32, + "latex": "\\{ ( M _ { 1 } , O _ { 1 , 1 } )" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 229.0, + 389.0, + 229.0, + 389.0, + 250.0, + 357.0, + 250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 229.0, + 740.0, + 229.0, + 740.0, + 250.0, + 709.0, + 250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 229.0, + 1089.0, + 229.0, + 1089.0, + 250.0, + 1059.0, + 250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 225.0, + 1145.0, + 225.0, + 1145.0, + 256.0, + 1096.0, + 256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 256.0, + 389.0, + 256.0, + 389.0, + 277.0, + 358.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 255.0, + 739.0, + 255.0, + 739.0, + 277.0, + 709.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 256.0, + 1092.0, + 256.0, + 1092.0, + 278.0, + 1059.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 281.0, + 387.0, + 281.0, + 387.0, + 305.0, + 358.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 280.0, + 738.0, + 280.0, + 738.0, + 304.0, + 710.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 279.0, + 782.0, + 279.0, + 782.0, + 306.0, + 743.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 281.0, + 1089.0, + 281.0, + 1089.0, + 305.0, + 1060.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 278.0, + 1135.0, + 278.0, + 1135.0, + 308.0, + 1092.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 307.0, + 400.0, + 307.0, + 400.0, + 322.0, + 386.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 308.0, + 1103.0, + 308.0, + 1103.0, + 322.0, + 1087.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 322.0, + 415.0, + 322.0, + 415.0, + 344.0, + 333.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 321.0, + 496.0, + 321.0, + 496.0, + 350.0, + 416.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 318.0, + 585.0, + 318.0, + 585.0, + 352.0, + 504.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 323.0, + 622.0, + 323.0, + 622.0, + 349.0, + 587.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 322.0, + 765.0, + 322.0, + 765.0, + 344.0, + 683.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 318.0, + 847.0, + 318.0, + 847.0, + 352.0, + 767.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 319.0, + 935.0, + 319.0, + 935.0, + 352.0, + 856.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 322.0, + 1116.0, + 322.0, + 1116.0, + 345.0, + 1035.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 323.0, + 1155.0, + 323.0, + 1155.0, + 349.0, + 1119.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 319.0, + 1209.0, + 319.0, + 1209.0, + 352.0, + 1156.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 356.0, + 541.0, + 356.0, + 541.0, + 400.0, + 457.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 356.0, + 890.0, + 356.0, + 890.0, + 400.0, + 807.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 358.0, + 1203.0, + 358.0, + 1203.0, + 399.0, + 1160.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 358.0, + 1241.0, + 358.0, + 1241.0, + 399.0, + 1233.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 436.0, + 389.0, + 436.0, + 389.0, + 458.0, + 357.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 431.0, + 448.0, + 431.0, + 448.0, + 462.0, + 394.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 436.0, + 740.0, + 436.0, + 740.0, + 458.0, + 709.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 432.0, + 843.0, + 432.0, + 843.0, + 462.0, + 748.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 435.0, + 1092.0, + 435.0, + 1092.0, + 461.0, + 1058.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 433.0, + 1193.0, + 433.0, + 1193.0, + 464.0, + 1098.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 461.0, + 390.0, + 461.0, + 390.0, + 486.0, + 356.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 462.0, + 740.0, + 462.0, + 740.0, + 484.0, + 709.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 458.0, + 835.0, + 458.0, + 835.0, + 489.0, + 783.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 461.0, + 1092.0, + 461.0, + 1092.0, + 488.0, + 1058.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 490.0, + 387.0, + 490.0, + 387.0, + 511.0, + 360.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 490.0, + 739.0, + 490.0, + 739.0, + 511.0, + 710.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 490.0, + 1090.0, + 490.0, + 1090.0, + 512.0, + 1059.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 486.0, + 1134.0, + 486.0, + 1134.0, + 512.0, + 1093.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 483.0, + 1207.0, + 483.0, + 1207.0, + 518.0, + 1158.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 514.0, + 400.0, + 514.0, + 400.0, + 531.0, + 386.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 512.0, + 450.0, + 512.0, + 450.0, + 528.0, + 421.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 517.0, + 494.0, + 517.0, + 494.0, + 526.0, + 482.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 517.0, + 750.0, + 517.0, + 750.0, + 527.0, + 739.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 512.0, + 801.0, + 512.0, + 801.0, + 528.0, + 771.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 514.0, + 846.0, + 514.0, + 846.0, + 528.0, + 833.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 515.0, + 1102.0, + 515.0, + 1102.0, + 531.0, + 1087.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 514.0, + 1152.0, + 514.0, + 1152.0, + 532.0, + 1123.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 517.0, + 1194.0, + 517.0, + 1194.0, + 528.0, + 1185.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 514.0, + 1230.0, + 514.0, + 1230.0, + 533.0, + 1211.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 528.0, + 450.0, + 528.0, + 450.0, + 556.0, + 333.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 529.0, + 613.0, + 529.0, + 613.0, + 557.0, + 532.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 528.0, + 804.0, + 528.0, + 804.0, + 559.0, + 683.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 529.0, + 1200.0, + 529.0, + 1200.0, + 559.0, + 1035.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 566.0, + 540.0, + 566.0, + 540.0, + 606.0, + 458.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 566.0, + 889.0, + 566.0, + 889.0, + 605.0, + 809.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 564.0, + 1238.0, + 564.0, + 1238.0, + 605.0, + 1161.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 644.0, + 384.0, + 644.0, + 384.0, + 665.0, + 354.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 636.0, + 536.0, + 636.0, + 536.0, + 673.0, + 389.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 640.0, + 737.0, + 640.0, + 737.0, + 666.0, + 704.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 635.0, + 887.0, + 635.0, + 887.0, + 671.0, + 742.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 644.0, + 1086.0, + 644.0, + 1086.0, + 666.0, + 1057.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 639.0, + 1237.0, + 639.0, + 1237.0, + 672.0, + 1096.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 668.0, + 386.0, + 668.0, + 386.0, + 694.0, + 353.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 668.0, + 478.0, + 668.0, + 478.0, + 695.0, + 430.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 667.0, + 737.0, + 667.0, + 737.0, + 694.0, + 704.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 668.0, + 1088.0, + 668.0, + 1088.0, + 694.0, + 1055.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 665.0, + 1183.0, + 665.0, + 1183.0, + 699.0, + 1130.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 669.0, + 1273.0, + 669.0, + 1273.0, + 694.0, + 1227.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 694.0, + 385.0, + 694.0, + 385.0, + 720.0, + 353.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 689.0, + 432.0, + 689.0, + 432.0, + 722.0, + 388.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 692.0, + 503.0, + 692.0, + 503.0, + 720.0, + 453.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 695.0, + 734.0, + 695.0, + 734.0, + 717.0, + 705.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 692.0, + 780.0, + 692.0, + 780.0, + 718.0, + 740.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 690.0, + 854.0, + 690.0, + 854.0, + 721.0, + 804.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 696.0, + 1088.0, + 696.0, + 1088.0, + 722.0, + 1055.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 689.0, + 1133.0, + 689.0, + 1133.0, + 724.0, + 1090.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 692.0, + 1206.0, + 692.0, + 1206.0, + 722.0, + 1155.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 692.0, + 1256.0, + 692.0, + 1256.0, + 723.0, + 1211.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 722.0, + 397.0, + 722.0, + 397.0, + 737.0, + 382.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 718.0, + 449.0, + 718.0, + 449.0, + 739.0, + 415.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 719.0, + 494.0, + 719.0, + 494.0, + 738.0, + 475.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 720.0, + 539.0, + 720.0, + 539.0, + 736.0, + 509.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 720.0, + 747.0, + 720.0, + 747.0, + 736.0, + 733.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 718.0, + 799.0, + 718.0, + 799.0, + 739.0, + 765.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 720.0, + 843.0, + 720.0, + 843.0, + 736.0, + 828.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 714.0, + 897.0, + 714.0, + 897.0, + 741.0, + 856.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 718.0, + 946.0, + 718.0, + 946.0, + 739.0, + 918.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 720.0, + 1101.0, + 720.0, + 1101.0, + 740.0, + 1082.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 719.0, + 1151.0, + 719.0, + 1151.0, + 741.0, + 1117.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 720.0, + 1197.0, + 720.0, + 1197.0, + 740.0, + 1176.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 718.0, + 1292.0, + 718.0, + 1292.0, + 741.0, + 1207.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 737.0, + 409.0, + 737.0, + 409.0, + 759.0, + 330.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 738.0, + 448.0, + 738.0, + 448.0, + 764.0, + 412.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 736.0, + 501.0, + 736.0, + 501.0, + 765.0, + 450.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 734.0, + 590.0, + 734.0, + 590.0, + 766.0, + 517.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 736.0, + 762.0, + 736.0, + 762.0, + 758.0, + 680.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 731.0, + 838.0, + 731.0, + 838.0, + 765.0, + 767.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 735.0, + 1114.0, + 735.0, + 1114.0, + 761.0, + 1031.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 772.0, + 536.0, + 772.0, + 536.0, + 812.0, + 451.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 772.0, + 887.0, + 772.0, + 887.0, + 812.0, + 805.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 772.0, + 1236.0, + 772.0, + 1236.0, + 812.0, + 1159.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.25, + 308.0, + 747.25, + 308.0, + 747.25, + 322.0, + 739.25, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 323.0, + 652.0, + 323.0, + 652.0, + 348.0, + 613.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.25, + 459.5, + 481.25, + 459.5, + 481.25, + 487.5, + 432.25, + 487.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.75, + 461.5, + 1183.75, + 461.5, + 1183.75, + 489.0, + 1135.75, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.25, + 485.0, + 430.25, + 485.0, + 430.25, + 511.0, + 394.25, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.25, + 485.5, + 781.25, + 485.5, + 781.25, + 511.5, + 745.25, + 511.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.25, + 528.5, + 509.25, + 528.5, + 509.25, + 556.5, + 461.25, + 556.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 527.5, + 860.0, + 527.5, + 860.0, + 556.5, + 813.0, + 556.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.25, + 666.5, + 828.25, + 666.5, + 828.25, + 694.5, + 780.25, + 694.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.75, + 665.5, + 921.75, + 665.5, + 921.75, + 693.5, + 874.75, + 693.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 830.0, + 1178.0, + 830.0, + 1178.0, + 874.0, + 411.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 830.0, + 1242.0, + 830.0, + 1242.0, + 874.0, + 1212.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 830.0, + 1286.0, + 830.0, + 1286.0, + 874.0, + 1276.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1326.0, + 563.0, + 1326.0, + 563.0, + 1378.0, + 237.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1564.0, + 609.0, + 1564.0, + 609.0, + 1611.0, + 252.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1409.0, + 277.0, + 1409.0, + 277.0, + 1435.0, + 240.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1402.0, + 1409.0, + 1402.0, + 1409.0, + 1441.0, + 292.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1439.0, + 277.0, + 1439.0, + 277.0, + 1466.0, + 240.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1434.0, + 1405.0, + 1434.0, + 1405.0, + 1469.0, + 294.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1467.0, + 278.0, + 1467.0, + 278.0, + 1498.0, + 239.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1463.0, + 1405.0, + 1463.0, + 1405.0, + 1498.0, + 294.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1499.0, + 277.0, + 1499.0, + 277.0, + 1526.0, + 240.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1494.0, + 909.0, + 1494.0, + 909.0, + 1529.0, + 292.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1164.0, + 277.0, + 1164.0, + 277.0, + 1191.0, + 240.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1158.0, + 1404.0, + 1158.0, + 1404.0, + 1193.0, + 293.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1193.0, + 278.0, + 1193.0, + 278.0, + 1220.0, + 240.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1186.0, + 1404.0, + 1186.0, + 1404.0, + 1223.0, + 293.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1224.0, + 277.0, + 1224.0, + 277.0, + 1251.0, + 240.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1217.0, + 865.0, + 1217.0, + 865.0, + 1252.0, + 294.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 1217.0, + 1404.0, + 1217.0, + 1404.0, + 1252.0, + 947.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1255.0, + 278.0, + 1255.0, + 278.0, + 1280.0, + 240.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1249.0, + 923.0, + 1249.0, + 923.0, + 1284.0, + 293.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 935.0, + 279.0, + 935.0, + 279.0, + 966.0, + 238.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 929.0, + 1406.0, + 929.0, + 1406.0, + 968.0, + 294.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 967.0, + 278.0, + 967.0, + 278.0, + 994.0, + 240.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 961.0, + 1050.0, + 961.0, + 1050.0, + 996.0, + 295.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 961.0, + 1405.0, + 961.0, + 1405.0, + 996.0, + 1084.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 995.0, + 278.0, + 995.0, + 278.0, + 1026.0, + 238.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 988.0, + 888.0, + 988.0, + 888.0, + 1029.0, + 294.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 988.0, + 970.0, + 988.0, + 970.0, + 1029.0, + 925.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 988.0, + 1408.0, + 988.0, + 1408.0, + 1029.0, + 1400.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1027.0, + 275.0, + 1027.0, + 275.0, + 1054.0, + 240.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1022.0, + 1062.0, + 1022.0, + 1062.0, + 1057.0, + 295.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1022.0, + 1127.0, + 1022.0, + 1127.0, + 1057.0, + 1096.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 1022.0, + 1406.0, + 1022.0, + 1406.0, + 1057.0, + 1161.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1058.0, + 278.0, + 1058.0, + 278.0, + 1085.0, + 240.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1052.0, + 911.0, + 1052.0, + 911.0, + 1087.0, + 295.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1052.0, + 1099.0, + 1052.0, + 1099.0, + 1087.0, + 938.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1052.0, + 1405.0, + 1052.0, + 1405.0, + 1087.0, + 1133.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1086.0, + 279.0, + 1086.0, + 279.0, + 1117.0, + 238.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1079.0, + 1408.0, + 1079.0, + 1408.0, + 1120.0, + 292.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1118.0, + 278.0, + 1118.0, + 278.0, + 1144.0, + 240.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1113.0, + 726.0, + 1113.0, + 726.0, + 1148.0, + 295.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1633.0, + 279.0, + 1633.0, + 279.0, + 1663.0, + 239.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1626.0, + 1408.0, + 1626.0, + 1408.0, + 1665.0, + 292.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1663.0, + 279.0, + 1663.0, + 279.0, + 1693.0, + 239.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1658.0, + 1405.0, + 1658.0, + 1405.0, + 1693.0, + 295.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1694.0, + 279.0, + 1694.0, + 279.0, + 1724.0, + 239.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1690.0, + 1405.0, + 1690.0, + 1405.0, + 1724.0, + 296.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1725.0, + 277.0, + 1725.0, + 277.0, + 1751.0, + 240.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1717.0, + 467.0, + 1717.0, + 467.0, + 1758.0, + 292.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1717.0, + 584.0, + 1717.0, + 584.0, + 1758.0, + 501.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 1717.0, + 1037.0, + 1717.0, + 1037.0, + 1758.0, + 665.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1717.0, + 1408.0, + 1717.0, + 1408.0, + 1758.0, + 1068.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1754.0, + 279.0, + 1754.0, + 279.0, + 1784.0, + 239.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1744.0, + 629.0, + 1744.0, + 629.0, + 1789.0, + 291.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1744.0, + 735.0, + 1744.0, + 735.0, + 1789.0, + 685.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 1744.0, + 1409.0, + 1744.0, + 1409.0, + 1789.0, + 791.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1785.0, + 279.0, + 1785.0, + 279.0, + 1815.0, + 239.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1780.0, + 718.0, + 1780.0, + 718.0, + 1818.0, + 293.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1780.0, + 1007.0, + 1780.0, + 1007.0, + 1818.0, + 751.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1780.0, + 1406.0, + 1780.0, + 1406.0, + 1818.0, + 1063.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1823.0, + 277.0, + 1823.0, + 277.0, + 1853.0, + 239.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1798.0, + 296.0, + 1798.0, + 296.0, + 1875.0, + 283.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 1798.0, + 697.0, + 1798.0, + 697.0, + 1875.0, + 386.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1798.0, + 1406.0, + 1798.0, + 1406.0, + 1875.0, + 820.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1859.0, + 279.0, + 1859.0, + 279.0, + 1889.0, + 239.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1853.0, + 703.0, + 1853.0, + 703.0, + 1889.0, + 292.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1853.0, + 1042.0, + 1853.0, + 1042.0, + 1889.0, + 783.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 1853.0, + 1406.0, + 1853.0, + 1406.0, + 1889.0, + 1176.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1892.0, + 277.0, + 1892.0, + 277.0, + 1918.0, + 240.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1883.0, + 392.0, + 1883.0, + 392.0, + 1922.0, + 293.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 1883.0, + 649.0, + 1883.0, + 649.0, + 1922.0, + 427.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1883.0, + 753.0, + 1883.0, + 753.0, + 1922.0, + 704.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1883.0, + 951.0, + 1883.0, + 951.0, + 1922.0, + 808.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1883.0, + 1406.0, + 1883.0, + 1406.0, + 1922.0, + 1005.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1922.0, + 277.0, + 1922.0, + 277.0, + 1946.0, + 240.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1915.0, + 460.0, + 1915.0, + 460.0, + 1949.0, + 293.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1915.0, + 726.0, + 1915.0, + 726.0, + 1949.0, + 494.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1915.0, + 1406.0, + 1915.0, + 1406.0, + 1949.0, + 780.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1952.0, + 277.0, + 1952.0, + 277.0, + 1978.0, + 240.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1946.0, + 351.0, + 1946.0, + 351.0, + 1980.0, + 296.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1946.0, + 1405.0, + 1946.0, + 1405.0, + 1980.0, + 382.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1982.0, + 277.0, + 1982.0, + 277.0, + 2008.0, + 240.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1973.0, + 900.0, + 1973.0, + 900.0, + 2014.0, + 292.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 1973.0, + 1075.0, + 1973.0, + 1075.0, + 2014.0, + 955.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1973.0, + 1348.0, + 1973.0, + 1348.0, + 2014.0, + 1108.0, + 2014.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 239, + 543, + 1406, + 543, + 1406, + 911, + 239, + 911 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 242, + 1773, + 1406, + 1773, + 1406, + 2009, + 242, + 2009 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 240, + 1484, + 1405, + 1484, + 1405, + 1760, + 240, + 1760 + ], + "score": 0.969 + }, + { + "category_id": 3, + "poly": [ + 299, + 961, + 1397, + 961, + 1397, + 1227, + 299, + 1227 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 243, + 1349, + 1406, + 1349, + 1406, + 1471, + 243, + 1471 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 242, + 279, + 1405, + 279, + 1405, + 419, + 242, + 419 + ], + "score": 0.94 + }, + { + "category_id": 4, + "poly": [ + 594, + 1263, + 1104, + 1263, + 1104, + 1297, + 594, + 1297 + ], + "score": 0.918 + }, + { + "category_id": 0, + "poly": [ + 253, + 477, + 738, + 477, + 738, + 511, + 253, + 511 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 256, + 203, + 1404, + 203, + 1404, + 266, + 256, + 266 + ], + "score": 0.875 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.706 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.098 + }, + { + "category_id": 13, + "poly": [ + 679, + 378, + 839, + 378, + 839, + 419, + 679, + 419 + ], + "score": 0.94, + "latex": "T _ { \\pi } ^ { ( m k s p ) } ( S , a )" + }, + { + "category_id": 13, + "poly": [ + 369, + 1815, + 608, + 1815, + 608, + 1862, + 369, + 1862 + ], + "score": 0.93, + "latex": "\\bar { T } _ { j , i } ^ { ( o p ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 1122, + 1773, + 1398, + 1773, + 1398, + 1819, + 1122, + 1819 + ], + "score": 0.93, + "latex": "T _ { m a x } ^ { ( o p ) } = \\operatorname* { m a x } _ { \\forall j , i } ( T _ { j , i } ^ { ( o p ) } )" + }, + { + "category_id": 13, + "poly": [ + 529, + 636, + 656, + 636, + 656, + 669, + 529, + 669 + ], + "score": 0.93, + "latex": "\\mathcal { G } = ( \\nu , \\mathcal { E } )" + }, + { + "category_id": 13, + "poly": [ + 373, + 1486, + 417, + 1486, + 417, + 1520, + 373, + 1520 + ], + "score": 0.93, + "latex": "\\phi _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 628, + 1773, + 691, + 1773, + 691, + 1819, + 628, + 1819 + ], + "score": 0.92, + "latex": "\\bar { T } _ { j , i } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 554, + 340, + 689, + 340, + 689, + 380, + 554, + 380 + ], + "score": 0.92, + "latex": "T _ { \\pi } ^ { ( m k s p ) } ( S )" + }, + { + "category_id": 13, + "poly": [ + 534, + 787, + 632, + 787, + 632, + 815, + 534, + 815 + ], + "score": 0.91, + "latex": "O \\to M" + }, + { + "category_id": 13, + "poly": [ + 380, + 848, + 485, + 848, + 485, + 876, + 380, + 876 + ], + "score": 0.91, + "latex": "M \\to M" + }, + { + "category_id": 13, + "poly": [ + 684, + 787, + 783, + 787, + 783, + 815, + 684, + 815 + ], + "score": 0.91, + "latex": "M O" + }, + { + "category_id": 13, + "poly": [ + 845, + 205, + 899, + 205, + 899, + 238, + 845, + 238 + ], + "score": 0.91, + "latex": "O _ { 1 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 659, + 1487, + 703, + 1487, + 703, + 1519, + 659, + 1519 + ], + "score": 0.91, + "latex": "\\phi _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 657, + 818, + 764, + 818, + 764, + 846, + 657, + 846 + ], + "score": 0.9, + "latex": "M \\to M" + }, + { + "category_id": 13, + "poly": [ + 345, + 205, + 398, + 205, + 398, + 236, + 345, + 236 + ], + "score": 0.9, + "latex": "O _ { 2 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 596, + 758, + 689, + 758, + 689, + 785, + 596, + 785 + ], + "score": 0.9, + "latex": "O O" + }, + { + "category_id": 13, + "poly": [ + 1348, + 1410, + 1396, + 1410, + 1396, + 1444, + 1348, + 1444 + ], + "score": 0.9, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 950, + 281, + 1019, + 281, + 1019, + 308, + 950, + 308 + ], + "score": 0.9, + "latex": "\\tau = 0" + }, + { + "category_id": 13, + "poly": [ + 1327, + 1698, + 1360, + 1698, + 1360, + 1728, + 1327, + 1728 + ], + "score": 0.89, + "latex": "S _ { 8 }" + }, + { + "category_id": 13, + "poly": [ + 491, + 879, + 523, + 879, + 523, + 909, + 491, + 909 + ], + "score": 0.89, + "latex": "S _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1263, + 1698, + 1295, + 1698, + 1295, + 1728, + 1263, + 1728 + ], + "score": 0.88, + "latex": "S _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 976, + 727, + 1078, + 727, + 1078, + 756, + 976, + 756 + ], + "score": 0.88, + "latex": "( O \\to M" + }, + { + "category_id": 13, + "poly": [ + 556, + 727, + 659, + 727, + 659, + 755, + 556, + 755 + ], + "score": 0.87, + "latex": "M O" + }, + { + "category_id": 13, + "poly": [ + 1296, + 697, + 1390, + 697, + 1390, + 726, + 1296, + 726 + ], + "score": 0.87, + "latex": "( O \\to O" + }, + { + "category_id": 13, + "poly": [ + 1180, + 1919, + 1205, + 1919, + 1205, + 1946, + 1180, + 1946 + ], + "score": 0.83, + "latex": "\\tau _ { t }" + }, + { + "category_id": 13, + "poly": [ + 948, + 758, + 973, + 758, + 973, + 784, + 948, + 784 + ], + "score": 0.83, + "latex": "O" + }, + { + "category_id": 13, + "poly": [ + 740, + 637, + 764, + 637, + 764, + 663, + 740, + 663 + ], + "score": 0.82, + "latex": "\\nu" + }, + { + "category_id": 13, + "poly": [ + 1352, + 1351, + 1374, + 1351, + 1374, + 1377, + 1352, + 1377 + ], + "score": 0.8, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 427, + 386, + 449, + 386, + 449, + 413, + 427, + 413 + ], + "score": 0.79, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 459, + 667, + 492, + 667, + 492, + 693, + 459, + 693 + ], + "score": 0.79, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 730, + 667, + 755, + 667, + 755, + 693, + 730, + 693 + ], + "score": 0.79, + "latex": "O" + }, + { + "category_id": 13, + "poly": [ + 993, + 637, + 1014, + 637, + 1014, + 663, + 993, + 663 + ], + "score": 0.78, + "latex": "\\mathcal { E }" + }, + { + "category_id": 13, + "poly": [ + 1243, + 386, + 1264, + 386, + 1264, + 413, + 1243, + 413 + ], + "score": 0.75, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 1079, + 1521, + 1099, + 1521, + 1099, + 1543, + 1079, + 1543 + ], + "score": 0.75, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 523, + 1385, + 542, + 1385, + 542, + 1407, + 523, + 1407 + ], + "score": 0.73, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 600, + 391, + 620, + 391, + 620, + 414, + 600, + 414 + ], + "score": 0.73, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 298, + 788, + 330, + 788, + 330, + 815, + 298, + 815 + ], + "score": 0.7, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1186, + 391, + 1204, + 391, + 1204, + 413, + 1186, + 413 + ], + "score": 0.7, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 1255, + 1493, + 1273, + 1493, + 1273, + 1513, + 1255, + 1513 + ], + "score": 0.67, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 324, + 1551, + 343, + 1551, + 343, + 1575, + 324, + 1575 + ], + "score": 0.6, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 524, + 353, + 543, + 353, + 543, + 375, + 524, + 375 + ], + "score": 0.46, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 499, + 1194, + 529, + 1194, + 529, + 1222, + 499, + 1222 + ], + "score": 0.4, + "latex": "S _ { 3 }" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 987.0, + 422.0, + 987.0, + 422.0, + 1022.0, + 376.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 999.0, + 349.0, + 999.0, + 349.0, + 1041.0, + 304.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1057.0, + 351.0, + 1057.0, + 351.0, + 1103.0, + 301.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1076.0, + 366.0, + 1076.0, + 366.0, + 1086.0, + 358.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1066.0, + 424.0, + 1066.0, + 424.0, + 1104.0, + 373.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1032.0, + 784.0, + 1032.0, + 784.0, + 1114.0, + 595.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 1030.0, + 908.0, + 1030.0, + 908.0, + 1085.0, + 858.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1057.0, + 1051.0, + 1057.0, + 1051.0, + 1114.0, + 980.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1121.0, + 348.0, + 1121.0, + 348.0, + 1161.0, + 304.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1135.0, + 422.0, + 1135.0, + 422.0, + 1175.0, + 374.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 1080.0, + 648.0, + 1080.0, + 648.0, + 1097.0, + 635.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1083.0, + 776.0, + 1083.0, + 776.0, + 1091.0, + 766.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1086.0, + 1130.0, + 1086.0, + 1130.0, + 1156.0, + 1059.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1189.0, + 498.0, + 1189.0, + 498.0, + 1226.0, + 352.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 1189.0, + 533.0, + 1189.0, + 533.0, + 1226.0, + 530.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1187.0, + 939.0, + 1187.0, + 939.0, + 1230.0, + 707.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1189.0, + 1326.0, + 1189.0, + 1326.0, + 1223.0, + 1129.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.25, + 991.5, + 489.25, + 991.5, + 489.25, + 1020.5, + 467.25, + 1020.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.75, + 997.5, + 574.75, + 997.5, + 574.75, + 1015.0, + 549.75, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.75, + 1008.5, + 1149.75, + 1008.5, + 1149.75, + 1091.5, + 1046.75, + 1091.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 1068.5, + 502.0, + 1068.5, + 502.0, + 1106.5, + 458.0, + 1106.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1072.0, + 572.0, + 1072.0, + 572.0, + 1100.0, + 543.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 1061.5, + 867.0, + 1061.5, + 867.0, + 1101.0, + 769.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.75, + 1060.5, + 993.75, + 1060.5, + 993.75, + 1103.5, + 892.75, + 1103.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 1065.0, + 1395.0, + 1065.0, + 1395.0, + 1110.0, + 1156.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 1261.0, + 1105.0, + 1261.0, + 1105.0, + 1301.0, + 593.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 472.0, + 741.0, + 472.0, + 741.0, + 518.0, + 244.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2092.0, + 838.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2092.0, + 838.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 550.0, + 277.0, + 550.0, + 277.0, + 575.0, + 240.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 545.0, + 1406.0, + 545.0, + 1406.0, + 578.0, + 295.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 581.0, + 277.0, + 581.0, + 277.0, + 606.0, + 240.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 575.0, + 1405.0, + 575.0, + 1405.0, + 609.0, + 295.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 610.0, + 279.0, + 610.0, + 279.0, + 639.0, + 239.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 605.0, + 1406.0, + 605.0, + 1406.0, + 641.0, + 293.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 639.0, + 279.0, + 639.0, + 279.0, + 669.0, + 239.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 634.0, + 528.0, + 634.0, + 528.0, + 671.0, + 292.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 634.0, + 739.0, + 634.0, + 739.0, + 671.0, + 657.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 634.0, + 992.0, + 634.0, + 992.0, + 671.0, + 765.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 634.0, + 1406.0, + 634.0, + 1406.0, + 671.0, + 1015.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 672.0, + 277.0, + 672.0, + 277.0, + 698.0, + 240.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 665.0, + 458.0, + 665.0, + 458.0, + 702.0, + 293.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 665.0, + 729.0, + 665.0, + 729.0, + 702.0, + 493.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 665.0, + 1407.0, + 665.0, + 1407.0, + 702.0, + 756.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 700.0, + 279.0, + 700.0, + 279.0, + 730.0, + 239.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 695.0, + 1295.0, + 695.0, + 1295.0, + 732.0, + 293.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1391.0, + 695.0, + 1408.0, + 695.0, + 1408.0, + 732.0, + 1391.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 731.0, + 279.0, + 731.0, + 279.0, + 760.0, + 239.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 724.0, + 555.0, + 724.0, + 555.0, + 763.0, + 292.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 724.0, + 975.0, + 724.0, + 975.0, + 763.0, + 660.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 724.0, + 1407.0, + 724.0, + 1407.0, + 763.0, + 1079.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 763.0, + 276.0, + 763.0, + 276.0, + 788.0, + 239.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 758.0, + 595.0, + 758.0, + 595.0, + 791.0, + 293.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 758.0, + 947.0, + 758.0, + 947.0, + 791.0, + 690.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 758.0, + 1406.0, + 758.0, + 1406.0, + 791.0, + 974.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 791.0, + 279.0, + 791.0, + 279.0, + 820.0, + 239.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 787.0, + 533.0, + 787.0, + 533.0, + 820.0, + 331.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 787.0, + 683.0, + 787.0, + 683.0, + 820.0, + 633.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 787.0, + 1405.0, + 787.0, + 1405.0, + 820.0, + 784.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 821.0, + 279.0, + 821.0, + 279.0, + 851.0, + 239.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 816.0, + 656.0, + 816.0, + 656.0, + 853.0, + 293.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 816.0, + 1407.0, + 816.0, + 1407.0, + 853.0, + 765.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 852.0, + 279.0, + 852.0, + 279.0, + 881.0, + 239.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 845.0, + 379.0, + 845.0, + 379.0, + 884.0, + 292.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 845.0, + 1406.0, + 845.0, + 1406.0, + 884.0, + 486.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 883.0, + 279.0, + 883.0, + 279.0, + 912.0, + 239.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 879.0, + 490.0, + 879.0, + 490.0, + 912.0, + 295.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 879.0, + 825.0, + 879.0, + 825.0, + 912.0, + 524.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1785.0, + 278.0, + 1785.0, + 278.0, + 1814.0, + 241.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1754.0, + 627.0, + 1754.0, + 627.0, + 1836.0, + 289.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 1754.0, + 1121.0, + 1754.0, + 1121.0, + 1836.0, + 692.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1754.0, + 1416.0, + 1754.0, + 1416.0, + 1836.0, + 1399.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1828.0, + 278.0, + 1828.0, + 278.0, + 1857.0, + 239.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1815.0, + 1407.0, + 1815.0, + 1407.0, + 1862.0, + 609.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1861.0, + 276.0, + 1861.0, + 276.0, + 1886.0, + 239.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1856.0, + 1406.0, + 1856.0, + 1406.0, + 1888.0, + 292.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1888.0, + 279.0, + 1888.0, + 279.0, + 1917.0, + 238.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1882.0, + 1405.0, + 1882.0, + 1405.0, + 1921.0, + 291.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1919.0, + 279.0, + 1919.0, + 279.0, + 1949.0, + 238.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1916.0, + 1179.0, + 1916.0, + 1179.0, + 1949.0, + 295.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1916.0, + 1403.0, + 1916.0, + 1403.0, + 1949.0, + 1206.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1950.0, + 279.0, + 1950.0, + 279.0, + 1979.0, + 239.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1945.0, + 1409.0, + 1945.0, + 1409.0, + 1980.0, + 293.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1982.0, + 279.0, + 1982.0, + 279.0, + 2007.0, + 241.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1975.0, + 1073.0, + 1975.0, + 1073.0, + 2012.0, + 293.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.25, + 1811.0, + 445.25, + 1811.0, + 445.25, + 1863.5, + 286.25, + 1863.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1805.0, + 549.0, + 1805.0, + 549.0, + 1856.5, + 449.0, + 1856.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1490.0, + 278.0, + 1490.0, + 278.0, + 1519.0, + 240.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1485.0, + 372.0, + 1485.0, + 372.0, + 1521.0, + 293.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1485.0, + 658.0, + 1485.0, + 658.0, + 1521.0, + 418.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1485.0, + 1254.0, + 1485.0, + 1254.0, + 1521.0, + 704.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1274.0, + 1485.0, + 1406.0, + 1485.0, + 1406.0, + 1521.0, + 1274.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1520.0, + 277.0, + 1520.0, + 277.0, + 1549.0, + 240.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1515.0, + 1078.0, + 1515.0, + 1078.0, + 1552.0, + 294.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1515.0, + 1406.0, + 1515.0, + 1406.0, + 1552.0, + 1100.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1552.0, + 278.0, + 1552.0, + 278.0, + 1580.0, + 240.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1548.0, + 323.0, + 1548.0, + 323.0, + 1580.0, + 294.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1548.0, + 1405.0, + 1548.0, + 1405.0, + 1580.0, + 344.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1580.0, + 278.0, + 1580.0, + 278.0, + 1610.0, + 240.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1576.0, + 1406.0, + 1576.0, + 1406.0, + 1613.0, + 294.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1613.0, + 278.0, + 1613.0, + 278.0, + 1638.0, + 240.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1606.0, + 1406.0, + 1606.0, + 1406.0, + 1643.0, + 293.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1642.0, + 278.0, + 1642.0, + 278.0, + 1670.0, + 239.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1639.0, + 1405.0, + 1639.0, + 1405.0, + 1672.0, + 294.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1674.0, + 278.0, + 1674.0, + 278.0, + 1699.0, + 240.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1670.0, + 1404.0, + 1670.0, + 1404.0, + 1699.0, + 295.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1702.0, + 278.0, + 1702.0, + 278.0, + 1730.0, + 239.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1696.0, + 1262.0, + 1696.0, + 1262.0, + 1733.0, + 294.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 1696.0, + 1326.0, + 1696.0, + 1326.0, + 1733.0, + 1296.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 1696.0, + 1406.0, + 1696.0, + 1406.0, + 1733.0, + 1361.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1734.0, + 278.0, + 1734.0, + 278.0, + 1759.0, + 241.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1729.0, + 560.0, + 1729.0, + 560.0, + 1762.0, + 294.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1355.0, + 277.0, + 1355.0, + 277.0, + 1381.0, + 240.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1349.0, + 1351.0, + 1349.0, + 1351.0, + 1384.0, + 293.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1349.0, + 1406.0, + 1349.0, + 1406.0, + 1384.0, + 1375.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1385.0, + 277.0, + 1385.0, + 277.0, + 1411.0, + 240.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1378.0, + 522.0, + 1378.0, + 522.0, + 1413.0, + 294.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1378.0, + 1405.0, + 1378.0, + 1405.0, + 1413.0, + 543.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1413.0, + 279.0, + 1413.0, + 279.0, + 1443.0, + 239.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1402.0, + 1347.0, + 1402.0, + 1347.0, + 1452.0, + 290.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1402.0, + 1410.0, + 1402.0, + 1410.0, + 1452.0, + 1397.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1446.0, + 277.0, + 1446.0, + 277.0, + 1472.0, + 240.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1441.0, + 665.0, + 1441.0, + 665.0, + 1475.0, + 293.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 283.0, + 279.0, + 283.0, + 279.0, + 310.0, + 239.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 278.0, + 949.0, + 278.0, + 949.0, + 313.0, + 293.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 278.0, + 1405.0, + 278.0, + 1405.0, + 313.0, + 1020.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 314.0, + 279.0, + 314.0, + 279.0, + 341.0, + 239.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 308.0, + 1406.0, + 308.0, + 1406.0, + 344.0, + 292.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 352.0, + 278.0, + 352.0, + 278.0, + 380.0, + 239.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 280.0, + 325.0, + 523.0, + 325.0, + 523.0, + 394.0, + 280.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 325.0, + 553.0, + 325.0, + 553.0, + 394.0, + 544.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 325.0, + 1417.0, + 325.0, + 1417.0, + 394.0, + 690.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 390.0, + 279.0, + 390.0, + 279.0, + 418.0, + 241.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 372.0, + 426.0, + 372.0, + 426.0, + 425.0, + 285.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 372.0, + 599.0, + 372.0, + 599.0, + 425.0, + 450.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 372.0, + 678.0, + 372.0, + 678.0, + 425.0, + 621.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 372.0, + 1185.0, + 372.0, + 1185.0, + 425.0, + 840.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 372.0, + 1242.0, + 372.0, + 1242.0, + 425.0, + 1205.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 372.0, + 1280.0, + 372.0, + 1280.0, + 425.0, + 1265.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 199.0, + 344.0, + 199.0, + 344.0, + 242.0, + 242.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 199.0, + 844.0, + 199.0, + 844.0, + 242.0, + 399.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 199.0, + 1406.0, + 199.0, + 1406.0, + 242.0, + 900.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 231.0, + 424.0, + 231.0, + 424.0, + 268.0, + 244.0, + 268.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 238, + 1621, + 1406, + 1621, + 1406, + 1805, + 238, + 1805 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 729, + 1404, + 729, + 1404, + 914, + 297, + 914 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 240, + 1470, + 1405, + 1470, + 1405, + 1611, + 240, + 1611 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 295, + 926, + 1406, + 926, + 1406, + 1045, + 295, + 1045 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 262, + 1131, + 1406, + 1131, + 1406, + 1382, + 262, + 1382 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 295, + 565, + 1406, + 565, + 1406, + 720, + 295, + 720 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 295, + 457, + 1405, + 457, + 1405, + 552, + 295, + 552 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 254, + 195, + 1406, + 195, + 1406, + 403, + 254, + 403 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 507, + 1807, + 1188, + 1807, + 1188, + 1884, + 507, + 1884 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 543, + 1386, + 1151, + 1386, + 1151, + 1460, + 543, + 1460 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 233, + 1893, + 1411, + 1893, + 1411, + 1962, + 233, + 1962 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 394, + 1965, + 1300, + 1965, + 1300, + 2016, + 394, + 2016 + ], + "score": 0.911 + }, + { + "category_id": 0, + "poly": [ + 299, + 1076, + 641, + 1076, + 641, + 1110, + 299, + 1110 + ], + "score": 0.909 + }, + { + "category_id": 1, + "poly": [ + 285, + 411, + 1236, + 411, + 1236, + 445, + 285, + 445 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1395, + 1402, + 1395, + 1402, + 1426, + 1367, + 1426 + ], + "score": 0.874 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1819, + 1402, + 1819, + 1402, + 1849, + 1367, + 1849 + ], + "score": 0.861 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1976, + 1402, + 1976, + 1402, + 2007, + 1367, + 2007 + ], + "score": 0.791 + }, + { + "category_id": 2, + "poly": [ + 839, + 2061, + 859, + 2061, + 859, + 2086, + 839, + 2086 + ], + "score": 0.599 + }, + { + "category_id": 2, + "poly": [ + 839, + 2061, + 859, + 2061, + 859, + 2086, + 839, + 2086 + ], + "score": 0.31 + }, + { + "category_id": 2, + "poly": [ + 239, + 176, + 282, + 176, + 282, + 1409, + 239, + 1409 + ], + "score": 0.228 + }, + { + "category_id": 2, + "poly": [ + 235, + 315, + 281, + 315, + 281, + 1310, + 235, + 1310 + ], + "score": 0.119 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1977, + 1402, + 1977, + 1402, + 2007, + 1367, + 2007 + ], + "score": 0.101 + }, + { + "category_id": 14, + "poly": [ + 509, + 1805, + 1191, + 1805, + 1191, + 1887, + 509, + 1887 + ], + "score": 0.94, + "latex": "h _ { v } ^ { ( k + 1 ) } = \\sum _ { \\mathcal { R } } M L P _ { \\mathcal { R } } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { \\mathcal { R } } ^ { ( k + 1 ) } ) h _ { v } ^ { ( k ) } + \\sum _ { u \\in N _ { \\mathcal { R } } ( v ) } h _ { u } ^ { ( k ) } )" + }, + { + "category_id": 13, + "poly": [ + 951, + 676, + 1226, + 676, + 1226, + 720, + 951, + 720 + ], + "score": 0.93, + "latex": "\\bar { T } _ { l } ^ { ( m a c ) } = T _ { l } ^ { ( m a c ) } / T _ { m a x } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 1047, + 926, + 1317, + 926, + 1317, + 973, + 1047, + 973 + ], + "score": 0.93, + "latex": "\\bar { T } _ { j , i , l } ^ { ( e d g e ) } = T _ { j , i } ^ { ( o p ) } / T _ { m a x } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 676, + 197, + 748, + 197, + 748, + 244, + 676, + 244 + ], + "score": 0.93, + "latex": "\\bar { T } _ { j , i } ^ { ( j o b ) }" + }, + { + "category_id": 13, + "poly": [ + 1291, + 564, + 1374, + 564, + 1374, + 607, + 1291, + 607 + ], + "score": 0.93, + "latex": "T _ { l } ^ { ( m a c ) }" + }, + { + "category_id": 13, + "poly": [ + 614, + 646, + 662, + 646, + 662, + 680, + 614, + 680 + ], + "score": 0.93, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 973, + 356, + 1036, + 356, + 1036, + 404, + 973, + 404 + ], + "score": 0.93, + "latex": "T _ { j , i , l } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 1026, + 239, + 1262, + 239, + 1262, + 285, + 1026, + 285 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { T _ { j } ^ { ( j o b ) } = \\sum _ { \\forall i ^ { \\prime } } T _ { j , i ^ { \\prime } } ^ { ( o p ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 478, + 356, + 541, + 356, + 541, + 404, + 478, + 404 + ], + "score": 0.93, + "latex": "T _ { j , i , l } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 756, + 927, + 841, + 927, + 841, + 973, + 756, + 973 + ], + "score": 0.93, + "latex": "\\bar { T } _ { j , i , l } ^ { ( e d g e ) }" + }, + { + "category_id": 13, + "poly": [ + 298, + 969, + 461, + 969, + 461, + 1017, + 298, + 1017 + ], + "score": 0.93, + "latex": "T _ { j , i } ^ { ( o p ) } = T _ { j , i , l } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 827, + 1892, + 929, + 1892, + 929, + 1934, + 827, + 1934 + ], + "score": 0.93, + "latex": "M L P _ { \\mathcal { R } } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 880, + 1315, + 1009, + 1315, + 1009, + 1349, + 880, + 1349 + ], + "score": 0.93, + "latex": "\\mathcal { G } = ( \\nu , \\mathcal { E } )" + }, + { + "category_id": 13, + "poly": [ + 709, + 356, + 773, + 356, + 773, + 403, + 709, + 403 + ], + "score": 0.93, + "latex": "T _ { j , i } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 739, + 564, + 822, + 564, + 822, + 607, + 739, + 607 + ], + "score": 0.92, + "latex": "\\bar { T } _ { l } ^ { ( m a c ) }" + }, + { + "category_id": 13, + "poly": [ + 374, + 1468, + 424, + 1468, + 424, + 1506, + 374, + 1506 + ], + "score": 0.92, + "latex": "h _ { v } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 1267, + 970, + 1331, + 970, + 1331, + 1017, + 1267, + 1017 + ], + "score": 0.92, + "latex": "T _ { j , i } ^ { ( o p ) }" + }, + { + "category_id": 14, + "poly": [ + 546, + 1384, + 1151, + 1384, + 1151, + 1462, + 546, + 1462 + ], + "score": 0.92, + "latex": "h _ { v } ^ { ( k ) } = M L P ^ { ( k ) } ( ( 1 + \\epsilon ^ { ( k ) } ) h _ { v } ^ { ( k - 1 ) } + \\sum _ { u \\in N _ { b } ( v ) } h _ { u } ^ { ( k - 1 ) } ) ," + }, + { + "category_id": 13, + "poly": [ + 297, + 604, + 360, + 604, + 360, + 650, + 297, + 650 + ], + "score": 0.91, + "latex": "T _ { j , i } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 441, + 1512, + 513, + 1512, + 513, + 1546, + 441, + 1546 + ], + "score": 0.91, + "latex": "N _ { b } ( v )" + }, + { + "category_id": 13, + "poly": [ + 1036, + 1505, + 1084, + 1505, + 1084, + 1543, + 1036, + 1543 + ], + "score": 0.91, + "latex": "h _ { v } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 782, + 677, + 845, + 677, + 845, + 715, + 782, + 715 + ], + "score": 0.91, + "latex": "T _ { m a x } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 839, + 979, + 888, + 979, + 888, + 1013, + 839, + 1013 + ], + "score": 0.91, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 1084, + 285, + 1346, + 285, + 1346, + 328, + 1084, + 328 + ], + "score": 0.91, + "latex": "\\bar { T } _ { j , i } ^ { ( j o b ) } = T _ { j , i } ^ { ( j o b ) } / T _ { j } ^ { ( j o b ) }" + }, + { + "category_id": 13, + "poly": [ + 1294, + 197, + 1403, + 197, + 1403, + 243, + 1294, + 243 + ], + "score": 0.91, + "latex": "T _ { j , i } ^ { ( j o b ) } =" + }, + { + "category_id": 13, + "poly": [ + 410, + 792, + 459, + 792, + 459, + 825, + 410, + 825 + ], + "score": 0.91, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 298, + 240, + 449, + 240, + 449, + 286, + 298, + 286 + ], + "score": 0.91, + "latex": "\\sum _ { \\forall i ^ { \\prime } \\geq i } T _ { j , i ^ { \\prime } } ^ { ( o p ) }" + }, + { + "category_id": 13, + "poly": [ + 297, + 285, + 369, + 285, + 369, + 328, + 297, + 328 + ], + "score": 0.91, + "latex": "T _ { j } ^ { ( j o b ) }" + }, + { + "category_id": 13, + "poly": [ + 939, + 852, + 987, + 852, + 987, + 885, + 939, + 885 + ], + "score": 0.91, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 1224, + 1654, + 1327, + 1654, + 1327, + 1682, + 1224, + 1682 + ], + "score": 0.91, + "latex": "M O" + }, + { + "category_id": 13, + "poly": [ + 454, + 1745, + 560, + 1745, + 560, + 1772, + 454, + 1772 + ], + "score": 0.91, + "latex": "M \\to M" + }, + { + "category_id": 13, + "poly": [ + 561, + 325, + 610, + 325, + 610, + 359, + 561, + 359 + ], + "score": 0.91, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 860, + 883, + 957, + 883, + 957, + 909, + 860, + 909 + ], + "score": 0.9, + "latex": "M O" + }, + { + "category_id": 13, + "poly": [ + 1223, + 1350, + 1293, + 1350, + 1293, + 1378, + 1223, + 1378 + ], + "score": 0.9, + "latex": "v \\in \\mathcal V" + }, + { + "category_id": 13, + "poly": [ + 1011, + 1346, + 1060, + 1346, + 1060, + 1378, + 1011, + 1378 + ], + "score": 0.9, + "latex": "h ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 375, + 1545, + 478, + 1545, + 478, + 1576, + 375, + 1576 + ], + "score": 0.9, + "latex": "M L P ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 955, + 1472, + 1000, + 1472, + 1000, + 1504, + 955, + 1504 + ], + "score": 0.9, + "latex": "\\epsilon ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 712, + 883, + 809, + 883, + 809, + 909, + 712, + 909 + ], + "score": 0.9, + "latex": "\\dot { O } M" + }, + { + "category_id": 13, + "poly": [ + 1080, + 1685, + 1186, + 1685, + 1186, + 1713, + 1080, + 1713 + ], + "score": 0.9, + "latex": "M \\to M" + }, + { + "category_id": 13, + "poly": [ + 759, + 1931, + 845, + 1931, + 845, + 1963, + 759, + 1963 + ], + "score": 0.9, + "latex": "( k + 1 )" + }, + { + "category_id": 13, + "poly": [ + 724, + 761, + 823, + 761, + 823, + 789, + 724, + 789 + ], + "score": 0.9, + "latex": "M O" + }, + { + "category_id": 13, + "poly": [ + 1337, + 823, + 1375, + 823, + 1375, + 852, + 1337, + 852 + ], + "score": 0.89, + "latex": "M _ { l }" + }, + { + "category_id": 13, + "poly": [ + 699, + 852, + 747, + 852, + 747, + 884, + 699, + 884 + ], + "score": 0.89, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 1019, + 572, + 1056, + 572, + 1056, + 603, + 1019, + 603 + ], + "score": 0.89, + "latex": "M _ { l }" + }, + { + "category_id": 13, + "poly": [ + 956, + 612, + 1006, + 612, + 1006, + 647, + 956, + 647 + ], + "score": 0.89, + "latex": "O _ { j , i }" + }, + { + "category_id": 13, + "poly": [ + 820, + 413, + 857, + 413, + 857, + 444, + 820, + 444 + ], + "score": 0.89, + "latex": "M _ { l }" + }, + { + "category_id": 14, + "poly": [ + 402, + 1965, + 1295, + 1965, + 1295, + 2017, + 402, + 2017 + ], + "score": 0.89, + "latex": "h _ { M _ { 1 } } ^ { ( k + 1 ) } = M L P _ { M M } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { M M } ^ { ( k + 1 ) } ) h _ { M _ { 1 } } ^ { ( k ) } ) + M L P _ { O M } ^ { ( k + 1 ) } ( h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } )" + }, + { + "category_id": 13, + "poly": [ + 596, + 490, + 634, + 490, + 634, + 519, + 596, + 519 + ], + "score": 0.88, + "latex": "M _ { l }" + }, + { + "category_id": 13, + "poly": [ + 571, + 761, + 671, + 761, + 671, + 789, + 571, + 789 + ], + "score": 0.88, + "latex": "O \\to M" + }, + { + "category_id": 13, + "poly": [ + 1148, + 762, + 1178, + 762, + 1178, + 795, + 1148, + 795 + ], + "score": 0.88, + "latex": "J _ { j }" + }, + { + "category_id": 13, + "poly": [ + 466, + 761, + 558, + 761, + 558, + 789, + 466, + 789 + ], + "score": 0.88, + "latex": "O O" + }, + { + "category_id": 13, + "poly": [ + 1116, + 363, + 1153, + 363, + 1153, + 395, + 1116, + 395 + ], + "score": 0.87, + "latex": "M _ { l }" + }, + { + "category_id": 13, + "poly": [ + 1340, + 1901, + 1371, + 1901, + 1371, + 1930, + 1340, + 1930 + ], + "score": 0.87, + "latex": "S _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 750, + 460, + 777, + 460, + 777, + 491, + 750, + 491 + ], + "score": 0.87, + "latex": "\\phi _ { l }" + }, + { + "category_id": 13, + "poly": [ + 398, + 883, + 436, + 883, + 436, + 912, + 398, + 912 + ], + "score": 0.87, + "latex": "M _ { l }" + }, + { + "category_id": 13, + "poly": [ + 643, + 1931, + 684, + 1931, + 684, + 1960, + 643, + 1960 + ], + "score": 0.87, + "latex": "M _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1221, + 205, + 1250, + 205, + 1250, + 239, + 1221, + 239 + ], + "score": 0.87, + "latex": "J _ { j }" + }, + { + "category_id": 13, + "poly": [ + 1207, + 328, + 1244, + 328, + 1244, + 355, + 1207, + 355 + ], + "score": 0.86, + "latex": "M _ { l }" + }, + { + "category_id": 13, + "poly": [ + 481, + 460, + 509, + 460, + 509, + 491, + 481, + 491 + ], + "score": 0.86, + "latex": "\\phi _ { l }" + }, + { + "category_id": 13, + "poly": [ + 1065, + 1654, + 1169, + 1654, + 1169, + 1682, + 1065, + 1682 + ], + "score": 0.85, + "latex": "O \\to M" + }, + { + "category_id": 13, + "poly": [ + 372, + 1901, + 399, + 1901, + 399, + 1927, + 372, + 1927 + ], + "score": 0.83, + "latex": "\\mathcal { R }" + }, + { + "category_id": 13, + "poly": [ + 957, + 1654, + 1052, + 1654, + 1052, + 1682, + 957, + 1682 + ], + "score": 0.83, + "latex": "O O" + }, + { + "category_id": 13, + "poly": [ + 501, + 1351, + 519, + 1351, + 519, + 1377, + 501, + 1377 + ], + "score": 0.83, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 828, + 1476, + 846, + 1476, + 846, + 1503, + 828, + 1503 + ], + "score": 0.82, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1106, + 1901, + 1132, + 1901, + 1132, + 1927, + 1106, + 1927 + ], + "score": 0.82, + "latex": "\\mathcal { R }" + }, + { + "category_id": 13, + "poly": [ + 973, + 1548, + 992, + 1548, + 992, + 1575, + 973, + 1575 + ], + "score": 0.81, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1062, + 1316, + 1091, + 1316, + 1091, + 1344, + 1062, + 1344 + ], + "score": 0.8, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1247, + 1317, + 1276, + 1317, + 1276, + 1343, + 1247, + 1343 + ], + "score": 0.79, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 732, + 1481, + 750, + 1481, + 750, + 1503, + 732, + 1503 + ], + "score": 0.78, + "latex": "v" + }, + { + "category_id": 13, + "poly": [ + 983, + 293, + 1000, + 293, + 1000, + 324, + 983, + 324 + ], + "score": 0.77, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 666, + 494, + 684, + 494, + 684, + 516, + 666, + 516 + ], + "score": 0.77, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 488, + 828, + 508, + 828, + 508, + 849, + 488, + 849 + ], + "score": 0.77, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 732, + 1520, + 750, + 1520, + 750, + 1541, + 732, + 1541 + ], + "score": 0.76, + "latex": "v" + }, + { + "category_id": 13, + "poly": [ + 972, + 249, + 989, + 249, + 989, + 281, + 972, + 281 + ], + "score": 0.76, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 894, + 1514, + 913, + 1514, + 913, + 1541, + 894, + 1541 + ], + "score": 0.75, + "latex": "\\mathcal { E }" + }, + { + "category_id": 13, + "poly": [ + 1312, + 495, + 1330, + 495, + 1330, + 516, + 1312, + 516 + ], + "score": 0.7, + "latex": "\\tau" + }, + { + "category_id": 13, + "poly": [ + 809, + 1548, + 884, + 1548, + 884, + 1578, + 809, + 1578 + ], + "score": 0.66, + "latex": "( M L P )" + }, + { + "category_id": 13, + "poly": [ + 1229, + 1286, + 1267, + 1286, + 1267, + 1313, + 1229, + 1313 + ], + "score": 0.28, + "latex": "\\mathrm { { X u } }" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1074.0, + 642.0, + 1074.0, + 642.0, + 1112.0, + 293.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 840.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 840.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 208.0, + 280.0, + 208.0, + 280.0, + 240.0, + 238.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 251.0, + 278.0, + 251.0, + 278.0, + 284.0, + 238.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 295.0, + 280.0, + 295.0, + 280.0, + 329.0, + 238.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 330.0, + 278.0, + 330.0, + 278.0, + 363.0, + 238.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 368.0, + 280.0, + 368.0, + 280.0, + 408.0, + 237.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 418.0, + 277.0, + 418.0, + 277.0, + 451.0, + 240.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 462.0, + 278.0, + 462.0, + 278.0, + 494.0, + 240.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 494.0, + 280.0, + 494.0, + 280.0, + 526.0, + 238.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 525.0, + 280.0, + 525.0, + 280.0, + 559.0, + 240.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 576.0, + 278.0, + 576.0, + 278.0, + 612.0, + 240.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 616.0, + 278.0, + 616.0, + 278.0, + 651.0, + 240.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 651.0, + 278.0, + 651.0, + 278.0, + 684.0, + 240.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 690.0, + 278.0, + 690.0, + 278.0, + 723.0, + 240.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 734.0, + 278.0, + 734.0, + 278.0, + 768.0, + 238.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 765.0, + 280.0, + 765.0, + 280.0, + 798.0, + 237.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 795.0, + 278.0, + 795.0, + 278.0, + 829.0, + 237.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 826.0, + 280.0, + 826.0, + 280.0, + 859.0, + 237.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 856.0, + 280.0, + 856.0, + 280.0, + 888.0, + 237.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 886.0, + 280.0, + 886.0, + 280.0, + 924.0, + 235.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 940.0, + 278.0, + 940.0, + 278.0, + 973.0, + 237.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 983.0, + 278.0, + 983.0, + 278.0, + 1017.0, + 237.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1016.0, + 280.0, + 1016.0, + 280.0, + 1055.0, + 235.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1081.0, + 281.0, + 1081.0, + 281.0, + 1119.0, + 235.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1137.0, + 278.0, + 1137.0, + 278.0, + 1170.0, + 237.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1167.0, + 280.0, + 1167.0, + 280.0, + 1201.0, + 237.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1199.0, + 278.0, + 1199.0, + 278.0, + 1231.0, + 237.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1229.0, + 280.0, + 1229.0, + 280.0, + 1262.0, + 235.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1259.0, + 280.0, + 1259.0, + 280.0, + 1292.0, + 235.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1290.0, + 280.0, + 1290.0, + 280.0, + 1323.0, + 237.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1319.0, + 281.0, + 1319.0, + 281.0, + 1356.0, + 235.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1352.0, + 281.0, + 1352.0, + 281.0, + 1390.0, + 235.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 330.0, + 279.0, + 330.0, + 279.0, + 361.0, + 240.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 367.0, + 280.0, + 367.0, + 280.0, + 404.0, + 237.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 416.0, + 279.0, + 416.0, + 279.0, + 452.0, + 237.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 461.0, + 280.0, + 461.0, + 280.0, + 495.0, + 237.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 492.0, + 280.0, + 492.0, + 280.0, + 527.0, + 237.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 522.0, + 281.0, + 522.0, + 281.0, + 563.0, + 236.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 575.0, + 280.0, + 575.0, + 280.0, + 612.0, + 239.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 614.0, + 280.0, + 614.0, + 280.0, + 651.0, + 239.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 649.0, + 280.0, + 649.0, + 280.0, + 686.0, + 237.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 686.0, + 281.0, + 686.0, + 281.0, + 727.0, + 236.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 733.0, + 280.0, + 733.0, + 280.0, + 767.0, + 236.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 763.0, + 280.0, + 763.0, + 280.0, + 799.0, + 236.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 794.0, + 279.0, + 794.0, + 279.0, + 830.0, + 236.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 824.0, + 280.0, + 824.0, + 280.0, + 859.0, + 236.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 855.0, + 280.0, + 855.0, + 280.0, + 890.0, + 237.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 885.0, + 280.0, + 885.0, + 280.0, + 922.0, + 237.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 939.0, + 279.0, + 939.0, + 279.0, + 977.0, + 237.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 981.0, + 280.0, + 981.0, + 280.0, + 1018.0, + 237.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1014.0, + 281.0, + 1014.0, + 281.0, + 1055.0, + 236.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1079.0, + 281.0, + 1079.0, + 281.0, + 1119.0, + 235.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1136.0, + 280.0, + 1136.0, + 280.0, + 1171.0, + 236.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1166.0, + 280.0, + 1166.0, + 280.0, + 1201.0, + 236.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1197.0, + 280.0, + 1197.0, + 280.0, + 1232.0, + 236.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1228.0, + 281.0, + 1228.0, + 281.0, + 1262.0, + 236.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1260.0, + 280.0, + 1260.0, + 280.0, + 1290.0, + 237.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1288.0, + 281.0, + 1288.0, + 281.0, + 1323.0, + 236.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1628.0, + 276.0, + 1628.0, + 276.0, + 1656.0, + 238.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1623.0, + 1407.0, + 1623.0, + 1407.0, + 1658.0, + 292.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1660.0, + 276.0, + 1660.0, + 276.0, + 1684.0, + 240.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1655.0, + 956.0, + 1655.0, + 956.0, + 1686.0, + 296.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1655.0, + 1064.0, + 1655.0, + 1064.0, + 1686.0, + 1053.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 1655.0, + 1223.0, + 1655.0, + 1223.0, + 1686.0, + 1170.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 1655.0, + 1404.0, + 1655.0, + 1404.0, + 1686.0, + 1328.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1689.0, + 278.0, + 1689.0, + 278.0, + 1717.0, + 238.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1683.0, + 1079.0, + 1683.0, + 1079.0, + 1722.0, + 294.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 1683.0, + 1410.0, + 1683.0, + 1410.0, + 1722.0, + 1187.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1722.0, + 276.0, + 1722.0, + 276.0, + 1746.0, + 241.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1713.0, + 1407.0, + 1713.0, + 1407.0, + 1751.0, + 292.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1751.0, + 276.0, + 1751.0, + 276.0, + 1775.0, + 240.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1746.0, + 453.0, + 1746.0, + 453.0, + 1778.0, + 295.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1746.0, + 1404.0, + 1746.0, + 1404.0, + 1778.0, + 561.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1779.0, + 278.0, + 1779.0, + 278.0, + 1807.0, + 238.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1771.0, + 553.0, + 1771.0, + 553.0, + 1808.0, + 293.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 727.0, + 1405.0, + 727.0, + 1405.0, + 766.0, + 294.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 759.0, + 465.0, + 759.0, + 465.0, + 795.0, + 295.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 759.0, + 570.0, + 759.0, + 570.0, + 795.0, + 559.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 759.0, + 723.0, + 759.0, + 723.0, + 795.0, + 672.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 759.0, + 1147.0, + 759.0, + 1147.0, + 795.0, + 824.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 759.0, + 1405.0, + 759.0, + 1405.0, + 795.0, + 1179.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 788.0, + 409.0, + 788.0, + 409.0, + 824.0, + 295.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 788.0, + 1405.0, + 788.0, + 1405.0, + 824.0, + 460.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 820.0, + 487.0, + 820.0, + 487.0, + 856.0, + 295.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 820.0, + 1336.0, + 820.0, + 1336.0, + 856.0, + 509.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 820.0, + 1406.0, + 820.0, + 1406.0, + 856.0, + 1376.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 852.0, + 698.0, + 852.0, + 698.0, + 888.0, + 295.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 852.0, + 938.0, + 852.0, + 938.0, + 888.0, + 748.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 852.0, + 1405.0, + 852.0, + 1405.0, + 888.0, + 988.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 883.0, + 397.0, + 883.0, + 397.0, + 914.0, + 295.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 883.0, + 711.0, + 883.0, + 711.0, + 914.0, + 437.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 883.0, + 859.0, + 883.0, + 859.0, + 914.0, + 810.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 883.0, + 1391.0, + 883.0, + 1391.0, + 914.0, + 958.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1478.0, + 278.0, + 1478.0, + 278.0, + 1507.0, + 240.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1469.0, + 373.0, + 1469.0, + 373.0, + 1512.0, + 287.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1469.0, + 731.0, + 1469.0, + 731.0, + 1512.0, + 425.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1469.0, + 827.0, + 1469.0, + 827.0, + 1512.0, + 751.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1469.0, + 954.0, + 1469.0, + 954.0, + 1512.0, + 847.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1469.0, + 1406.0, + 1469.0, + 1406.0, + 1512.0, + 1001.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1518.0, + 278.0, + 1518.0, + 278.0, + 1546.0, + 239.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1502.0, + 440.0, + 1502.0, + 440.0, + 1553.0, + 285.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1502.0, + 731.0, + 1502.0, + 731.0, + 1553.0, + 514.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1502.0, + 893.0, + 1502.0, + 893.0, + 1553.0, + 751.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1502.0, + 1035.0, + 1502.0, + 1035.0, + 1553.0, + 914.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 1502.0, + 1413.0, + 1502.0, + 1413.0, + 1553.0, + 1085.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1552.0, + 280.0, + 1552.0, + 280.0, + 1579.0, + 237.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1540.0, + 374.0, + 1540.0, + 374.0, + 1587.0, + 289.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1540.0, + 808.0, + 1540.0, + 808.0, + 1587.0, + 479.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 1540.0, + 972.0, + 1540.0, + 972.0, + 1587.0, + 885.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1540.0, + 1409.0, + 1540.0, + 1409.0, + 1587.0, + 993.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1582.0, + 280.0, + 1582.0, + 280.0, + 1610.0, + 239.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1577.0, + 597.0, + 1577.0, + 597.0, + 1612.0, + 293.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 933.0, + 755.0, + 933.0, + 755.0, + 968.0, + 293.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 914.0, + 1046.0, + 914.0, + 1046.0, + 986.0, + 851.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 913.0, + 1340.0, + 913.0, + 1340.0, + 980.0, + 1318.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 975.0, + 1408.0, + 975.0, + 1408.0, + 1012.0, + 1332.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1013.0, + 694.0, + 1013.0, + 694.0, + 1047.0, + 293.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 915.0, + 868.0, + 915.0, + 868.0, + 969.5, + 743.0, + 969.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.75, + 917.0, + 1254.75, + 917.0, + 1254.75, + 966.0, + 1161.75, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.75, + 931.5, + 1409.75, + 931.5, + 1409.75, + 968.5, + 1334.75, + 968.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.25, + 961.0, + 374.25, + 961.0, + 374.25, + 1009.0, + 284.25, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 964.0, + 474.0, + 964.0, + 474.0, + 1009.0, + 383.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 260.0, + 1143.0, + 275.0, + 1143.0, + 275.0, + 1162.0, + 260.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1132.0, + 1404.0, + 1132.0, + 1404.0, + 1170.0, + 292.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 260.0, + 1173.0, + 275.0, + 1173.0, + 275.0, + 1192.0, + 260.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1163.0, + 1404.0, + 1163.0, + 1404.0, + 1198.0, + 294.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1192.0, + 1405.0, + 1192.0, + 1405.0, + 1230.0, + 292.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 260.0, + 1235.0, + 275.0, + 1235.0, + 275.0, + 1254.0, + 260.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1225.0, + 1404.0, + 1225.0, + 1404.0, + 1259.0, + 294.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 260.0, + 1263.0, + 275.0, + 1263.0, + 275.0, + 1283.0, + 260.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1255.0, + 1405.0, + 1255.0, + 1405.0, + 1289.0, + 295.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 261.0, + 1295.0, + 275.0, + 1295.0, + 275.0, + 1313.0, + 261.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1283.0, + 1228.0, + 1283.0, + 1228.0, + 1321.0, + 292.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1283.0, + 1408.0, + 1283.0, + 1408.0, + 1321.0, + 1268.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 260.0, + 1325.0, + 277.0, + 1325.0, + 277.0, + 1344.0, + 260.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1313.0, + 879.0, + 1313.0, + 879.0, + 1351.0, + 292.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1313.0, + 1061.0, + 1313.0, + 1061.0, + 1351.0, + 1010.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1313.0, + 1246.0, + 1313.0, + 1246.0, + 1351.0, + 1092.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1313.0, + 1409.0, + 1313.0, + 1409.0, + 1351.0, + 1277.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 1357.0, + 278.0, + 1357.0, + 278.0, + 1380.0, + 259.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1346.0, + 500.0, + 1346.0, + 500.0, + 1385.0, + 290.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 1346.0, + 1010.0, + 1346.0, + 1010.0, + 1385.0, + 520.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 1346.0, + 1222.0, + 1346.0, + 1222.0, + 1385.0, + 1061.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 1346.0, + 1307.0, + 1346.0, + 1307.0, + 1385.0, + 1294.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 556.0, + 738.0, + 556.0, + 738.0, + 612.0, + 285.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 556.0, + 1018.0, + 556.0, + 1018.0, + 612.0, + 823.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 556.0, + 1290.0, + 556.0, + 1290.0, + 612.0, + 1057.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 556.0, + 1403.0, + 556.0, + 1403.0, + 612.0, + 1375.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 595.0, + 296.0, + 595.0, + 296.0, + 654.0, + 288.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 595.0, + 955.0, + 595.0, + 955.0, + 654.0, + 361.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 595.0, + 1411.0, + 595.0, + 1411.0, + 654.0, + 1007.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 644.0, + 613.0, + 644.0, + 613.0, + 681.0, + 292.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 644.0, + 1406.0, + 644.0, + 1406.0, + 681.0, + 663.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 661.0, + 781.0, + 661.0, + 781.0, + 735.0, + 282.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 661.0, + 950.0, + 661.0, + 950.0, + 735.0, + 846.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 661.0, + 1248.0, + 661.0, + 1248.0, + 735.0, + 1227.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 457.0, + 480.0, + 457.0, + 480.0, + 495.0, + 293.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 457.0, + 749.0, + 457.0, + 749.0, + 495.0, + 510.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 457.0, + 1406.0, + 457.0, + 1406.0, + 495.0, + 778.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 489.0, + 595.0, + 489.0, + 595.0, + 523.0, + 295.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 489.0, + 665.0, + 489.0, + 665.0, + 523.0, + 635.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 489.0, + 1311.0, + 489.0, + 1311.0, + 523.0, + 685.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1331.0, + 489.0, + 1405.0, + 489.0, + 1405.0, + 523.0, + 1331.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 519.0, + 1352.0, + 519.0, + 1352.0, + 553.0, + 294.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 210.0, + 278.0, + 210.0, + 278.0, + 236.0, + 247.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 181.0, + 297.0, + 181.0, + 297.0, + 311.0, + 289.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 181.0, + 675.0, + 181.0, + 675.0, + 311.0, + 450.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 181.0, + 971.0, + 181.0, + 971.0, + 311.0, + 749.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 181.0, + 1025.0, + 181.0, + 1025.0, + 311.0, + 990.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 181.0, + 1293.0, + 181.0, + 1293.0, + 311.0, + 1263.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 249.0, + 280.0, + 249.0, + 280.0, + 282.0, + 246.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 237.0, + 296.0, + 237.0, + 296.0, + 358.0, + 291.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 237.0, + 560.0, + 237.0, + 560.0, + 358.0, + 450.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 237.0, + 971.0, + 237.0, + 971.0, + 358.0, + 611.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 237.0, + 1025.0, + 237.0, + 1025.0, + 358.0, + 1001.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 236.0, + 1414.0, + 236.0, + 1414.0, + 290.0, + 1263.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 294.0, + 281.0, + 294.0, + 281.0, + 327.0, + 246.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 273.0, + 1410.0, + 273.0, + 1410.0, + 340.0, + 1347.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 328.0, + 280.0, + 328.0, + 280.0, + 360.0, + 246.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 320.0, + 560.0, + 320.0, + 560.0, + 362.0, + 290.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 320.0, + 1206.0, + 320.0, + 1206.0, + 362.0, + 611.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 320.0, + 1409.0, + 320.0, + 1409.0, + 362.0, + 1245.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 369.0, + 278.0, + 369.0, + 278.0, + 397.0, + 247.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 365.0, + 477.0, + 365.0, + 477.0, + 396.0, + 292.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 355.0, + 972.0, + 355.0, + 972.0, + 406.0, + 774.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 355.0, + 1115.0, + 355.0, + 1115.0, + 406.0, + 1037.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 355.0, + 1168.0, + 355.0, + 1168.0, + 406.0, + 1154.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.25, + 213.5, + 1397.25, + 213.5, + 1397.25, + 228.5, + 1380.25, + 228.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 240.0, + 1200.0, + 240.0, + 1200.0, + 291.0, + 1097.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.25, + 230.0, + 1274.25, + 230.0, + 1274.25, + 285.5, + 1180.25, + 285.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 267.0, + 392.0, + 267.0, + 392.0, + 337.5, + 277.0, + 337.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.25, + 351.0, + 552.25, + 351.0, + 552.25, + 392.5, + 466.25, + 392.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.25, + 352.5, + 785.25, + 352.5, + 785.25, + 410.5, + 479.25, + 410.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1873.0, + 371.0, + 1873.0, + 371.0, + 1957.0, + 242.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1873.0, + 642.0, + 1873.0, + 642.0, + 1957.0, + 400.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1873.0, + 758.0, + 1873.0, + 758.0, + 1957.0, + 685.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1873.0, + 1105.0, + 1873.0, + 1105.0, + 1957.0, + 930.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1873.0, + 1339.0, + 1873.0, + 1339.0, + 1957.0, + 1133.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 1873.0, + 1405.0, + 1873.0, + 1405.0, + 1957.0, + 1372.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1935.0, + 278.0, + 1935.0, + 278.0, + 1959.0, + 240.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1930.0, + 642.0, + 1930.0, + 642.0, + 1963.0, + 293.0, + 1963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1930.0, + 758.0, + 1930.0, + 758.0, + 1963.0, + 685.0, + 1963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1930.0, + 1269.0, + 1930.0, + 1269.0, + 1963.0, + 846.0, + 1963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 408.0, + 819.0, + 408.0, + 819.0, + 450.0, + 293.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 408.0, + 1239.0, + 408.0, + 1239.0, + 450.0, + 858.0, + 450.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 277, + 1476, + 1405, + 1476, + 1405, + 1783, + 277, + 1783 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 241, + 961, + 1404, + 961, + 1404, + 1118, + 241, + 1118 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 290, + 1278, + 1403, + 1278, + 1403, + 1465, + 290, + 1465 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 271, + 471, + 1405, + 471, + 1405, + 664, + 271, + 664 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 258, + 752, + 1403, + 752, + 1403, + 845, + 258, + 845 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 600, + 855, + 1098, + 855, + 1098, + 943, + 600, + 943 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 240, + 316, + 1405, + 316, + 1405, + 459, + 240, + 459 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 703, + 1851, + 992, + 1851, + 992, + 1934, + 703, + 1934 + ], + "score": 0.952 + }, + { + "category_id": 0, + "poly": [ + 274, + 1223, + 969, + 1223, + 969, + 1258, + 274, + 1258 + ], + "score": 0.937 + }, + { + "category_id": 1, + "poly": [ + 264, + 1795, + 1401, + 1795, + 1401, + 1854, + 264, + 1854 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 246, + 1941, + 1406, + 1941, + 1406, + 2009, + 246, + 2009 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 241, + 201, + 988, + 201, + 988, + 236, + 241, + 236 + ], + "score": 0.905 + }, + { + "category_id": 0, + "poly": [ + 288, + 695, + 672, + 695, + 672, + 730, + 288, + 730 + ], + "score": 0.904 + }, + { + "category_id": 0, + "poly": [ + 286, + 1157, + 532, + 1157, + 532, + 1196, + 286, + 1196 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1368, + 883, + 1402, + 883, + 1402, + 914, + 1368, + 914 + ], + "score": 0.878 + }, + { + "category_id": 9, + "poly": [ + 1368, + 1879, + 1401, + 1879, + 1401, + 1908, + 1368, + 1908 + ], + "score": 0.834 + }, + { + "category_id": 9, + "poly": [ + 1368, + 255, + 1402, + 255, + 1402, + 285, + 1368, + 285 + ], + "score": 0.823 + }, + { + "category_id": 2, + "poly": [ + 841, + 2060, + 858, + 2060, + 858, + 2084, + 841, + 2084 + ], + "score": 0.741 + }, + { + "category_id": 8, + "poly": [ + 399, + 243, + 1295, + 243, + 1295, + 296, + 399, + 296 + ], + "score": 0.706 + }, + { + "category_id": 14, + "poly": [ + 704, + 1848, + 994, + 1848, + 994, + 1934, + 704, + 1934 + ], + "score": 0.94, + "latex": "G = \\frac { T _ { \\pi } ^ { \\left( m k s p \\right) } - T _ { \\pi * } ^ { \\left( m k s p \\right) } } { T _ { \\pi * } ^ { \\left( m k s p \\right) } }" + }, + { + "category_id": 14, + "poly": [ + 599, + 852, + 1100, + 852, + 1100, + 943, + 599, + 943 + ], + "score": 0.94, + "latex": "A _ { \\pi } ( S , a ) = { \\frac { T _ { \\pi _ { b } } ^ { ( m k s p ) } ( S , a ) - T _ { \\pi } ^ { ( m k s p ) } ( S , a ) } { T _ { \\pi _ { b } } ^ { ( m k s p ) } ( S , a ) } }" + }, + { + "category_id": 13, + "poly": [ + 368, + 1937, + 462, + 1937, + 462, + 1976, + 368, + 1976 + ], + "score": 0.93, + "latex": "T _ { \\pi * } ^ { ( m k s p ) }" + }, + { + "category_id": 13, + "poly": [ + 742, + 203, + 827, + 203, + 827, + 237, + 742, + 237 + ], + "score": 0.92, + "latex": "( k + 1 )" + }, + { + "category_id": 13, + "poly": [ + 991, + 1054, + 1224, + 1054, + 1224, + 1088, + 991, + 1088 + ], + "score": 0.92, + "latex": "\\nabla _ { \\boldsymbol { \\theta } } l o g \\pi _ { \\boldsymbol { \\theta } } A _ { \\pi _ { \\boldsymbol { \\theta } } } \\bigl ( S _ { t } , a _ { t } \\bigr )" + }, + { + "category_id": 13, + "poly": [ + 1315, + 1569, + 1403, + 1569, + 1403, + 1601, + 1315, + 1601 + ], + "score": 0.92, + "latex": "\\boldsymbol { \\mathcal { U } } ( \\boldsymbol { x } , \\boldsymbol { y } )" + }, + { + "category_id": 13, + "poly": [ + 614, + 205, + 667, + 205, + 667, + 238, + 614, + 238 + ], + "score": 0.92, + "latex": "O _ { 1 , 1 }" + }, + { + "category_id": 14, + "poly": [ + 402, + 244, + 1297, + 244, + 1297, + 297, + 402, + 297 + ], + "score": 0.91, + "latex": "h _ { O _ { 1 , 1 } } ^ { ( k + 1 ) } = M L P _ { O O } ^ { ( k + 1 ) } ( ( 1 + \\epsilon _ { O O } ^ { ( k + 1 ) } ) h _ { O _ { 1 , 1 } } ^ { ( k ) } + h _ { O _ { 1 , 2 } } ^ { ( k ) } + h _ { O _ { 1 , 3 } } ^ { ( k ) } ) + M L P _ { M O } ^ { ( k + 1 ) } ( h _ { M _ { 1 } } ^ { ( k ) } )" + }, + { + "category_id": 13, + "poly": [ + 754, + 1569, + 902, + 1569, + 902, + 1602, + 754, + 1602 + ], + "score": 0.91, + "latex": "m \\sim \\mathcal { U } ( 3 , n )" + }, + { + "category_id": 13, + "poly": [ + 1141, + 1570, + 1231, + 1570, + 1231, + 1602, + 1141, + 1602 + ], + "score": 0.91, + "latex": "k _ { j } = m" + }, + { + "category_id": 13, + "poly": [ + 341, + 348, + 390, + 348, + 390, + 390, + 341, + 390 + ], + "score": 0.9, + "latex": "h _ { \\mathcal { G } } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 297, + 389, + 374, + 389, + 374, + 426, + 297, + 426 + ], + "score": 0.9, + "latex": "h _ { v } ^ { ( k + 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 969, + 1541, + 1048, + 1541, + 1048, + 1568, + 969, + 1568 + ], + "score": 0.9, + "latex": "n \\times m" + }, + { + "category_id": 13, + "poly": [ + 594, + 1568, + 742, + 1568, + 742, + 1602, + 594, + 1602 + ], + "score": 0.89, + "latex": "n \\sim \\mathcal { U } ( \\bar { 3 } , N )" + }, + { + "category_id": 13, + "poly": [ + 1162, + 1602, + 1221, + 1602, + 1221, + 1632, + 1162, + 1632 + ], + "score": 0.88, + "latex": "[ x , y ]" + }, + { + "category_id": 13, + "poly": [ + 617, + 601, + 703, + 601, + 703, + 632, + 617, + 632 + ], + "score": 0.87, + "latex": "( M , O )" + }, + { + "category_id": 13, + "poly": [ + 703, + 568, + 859, + 568, + 859, + 603, + 703, + 603 + ], + "score": 0.86, + "latex": "S c o r e ( M , O )" + }, + { + "category_id": 13, + "poly": [ + 777, + 818, + 807, + 818, + 807, + 844, + 777, + 844 + ], + "score": 0.86, + "latex": "\\pi _ { b }" + }, + { + "category_id": 13, + "poly": [ + 1060, + 969, + 1091, + 969, + 1091, + 994, + 1060, + 994 + ], + "score": 0.85, + "latex": "\\pi _ { b }" + }, + { + "category_id": 13, + "poly": [ + 872, + 355, + 893, + 355, + 893, + 384, + 872, + 384 + ], + "score": 0.83, + "latex": "\\mathcal { G }" + }, + { + "category_id": 13, + "poly": [ + 598, + 1980, + 632, + 1980, + 632, + 2003, + 598, + 2003 + ], + "score": 0.81, + "latex": "\\pi *" + }, + { + "category_id": 13, + "poly": [ + 494, + 534, + 520, + 534, + 520, + 560, + 494, + 560 + ], + "score": 0.8, + "latex": "O" + }, + { + "category_id": 13, + "poly": [ + 438, + 563, + 566, + 563, + 566, + 605, + 438, + 605 + ], + "score": 0.8, + "latex": "( h _ { M } ^ { ( k ) } , h _ { O } ^ { ( k ) } )" + }, + { + "category_id": 13, + "poly": [ + 730, + 1056, + 747, + 1056, + 747, + 1082, + 730, + 1082 + ], + "score": 0.8, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 863, + 534, + 895, + 534, + 895, + 560, + 863, + 560 + ], + "score": 0.79, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 647, + 637, + 667, + 637, + 667, + 659, + 647, + 659 + ], + "score": 0.78, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 796, + 1803, + 816, + 1803, + 816, + 1824, + 796, + 1824 + ], + "score": 0.78, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 455, + 1539, + 544, + 1539, + 544, + 1569, + 455, + 1569 + ], + "score": 0.77, + "latex": "( N , M )" + }, + { + "category_id": 13, + "poly": [ + 1375, + 1797, + 1401, + 1797, + 1401, + 1825, + 1375, + 1825 + ], + "score": 0.76, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 491, + 1060, + 511, + 1060, + 511, + 1082, + 491, + 1082 + ], + "score": 0.76, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1126, + 533, + 1213, + 533, + 1213, + 565, + 1126, + 565 + ], + "score": 0.59, + "latex": "( M , O )" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1221.0, + 974.0, + 1221.0, + 974.0, + 1263.0, + 292.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 689.0, + 676.0, + 689.0, + 676.0, + 740.0, + 292.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1150.0, + 537.0, + 1150.0, + 537.0, + 1207.0, + 290.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1478.0, + 1405.0, + 1478.0, + 1405.0, + 1510.0, + 296.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1510.0, + 1406.0, + 1510.0, + 1406.0, + 1542.0, + 295.0, + 1542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1537.0, + 454.0, + 1537.0, + 454.0, + 1575.0, + 294.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1537.0, + 968.0, + 1537.0, + 968.0, + 1575.0, + 545.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.0, + 1537.0, + 1406.0, + 1537.0, + 1406.0, + 1575.0, + 1049.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1566.0, + 593.0, + 1566.0, + 593.0, + 1605.0, + 294.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1566.0, + 753.0, + 1566.0, + 753.0, + 1605.0, + 743.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 1566.0, + 1140.0, + 1566.0, + 1140.0, + 1605.0, + 903.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 1566.0, + 1314.0, + 1566.0, + 1314.0, + 1605.0, + 1232.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1599.0, + 1161.0, + 1599.0, + 1161.0, + 1633.0, + 292.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 1599.0, + 1406.0, + 1599.0, + 1406.0, + 1633.0, + 1222.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1628.0, + 1406.0, + 1628.0, + 1406.0, + 1664.0, + 294.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1657.0, + 1408.0, + 1657.0, + 1408.0, + 1694.0, + 292.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1692.0, + 1405.0, + 1692.0, + 1405.0, + 1724.0, + 295.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1719.0, + 1409.0, + 1719.0, + 1409.0, + 1757.0, + 293.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1750.0, + 1360.0, + 1750.0, + 1360.0, + 1786.0, + 295.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 969.0, + 278.0, + 969.0, + 278.0, + 994.0, + 240.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 963.0, + 1059.0, + 963.0, + 1059.0, + 997.0, + 294.0, + 997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 963.0, + 1405.0, + 963.0, + 1405.0, + 997.0, + 1092.0, + 997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1001.0, + 278.0, + 1001.0, + 278.0, + 1026.0, + 240.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 993.0, + 1405.0, + 993.0, + 1405.0, + 1030.0, + 294.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1029.0, + 279.0, + 1029.0, + 279.0, + 1057.0, + 237.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1023.0, + 1405.0, + 1023.0, + 1405.0, + 1060.0, + 292.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1058.0, + 279.0, + 1058.0, + 279.0, + 1088.0, + 237.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1052.0, + 490.0, + 1052.0, + 490.0, + 1092.0, + 292.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1052.0, + 729.0, + 1052.0, + 729.0, + 1092.0, + 512.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1052.0, + 990.0, + 1052.0, + 990.0, + 1092.0, + 748.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 1052.0, + 1405.0, + 1052.0, + 1405.0, + 1092.0, + 1225.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1089.0, + 279.0, + 1089.0, + 279.0, + 1117.0, + 237.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1081.0, + 837.0, + 1081.0, + 837.0, + 1120.0, + 294.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1279.0, + 1405.0, + 1279.0, + 1405.0, + 1314.0, + 293.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1310.0, + 1405.0, + 1310.0, + 1405.0, + 1347.0, + 293.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1341.0, + 1405.0, + 1341.0, + 1405.0, + 1375.0, + 293.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1373.0, + 1404.0, + 1373.0, + 1404.0, + 1405.0, + 296.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1402.0, + 1405.0, + 1402.0, + 1405.0, + 1438.0, + 294.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1433.0, + 968.0, + 1433.0, + 968.0, + 1469.0, + 294.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 468.0, + 1405.0, + 468.0, + 1405.0, + 507.0, + 292.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 499.0, + 1404.0, + 499.0, + 1404.0, + 536.0, + 293.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 531.0, + 493.0, + 531.0, + 493.0, + 566.0, + 292.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 531.0, + 862.0, + 531.0, + 862.0, + 566.0, + 521.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 531.0, + 1125.0, + 531.0, + 1125.0, + 566.0, + 896.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 531.0, + 1406.0, + 531.0, + 1406.0, + 566.0, + 1214.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 552.0, + 437.0, + 552.0, + 437.0, + 621.0, + 284.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 552.0, + 702.0, + 552.0, + 702.0, + 621.0, + 567.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 552.0, + 1418.0, + 552.0, + 1418.0, + 621.0, + 860.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 596.0, + 616.0, + 596.0, + 616.0, + 634.0, + 290.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 596.0, + 1406.0, + 596.0, + 1406.0, + 634.0, + 704.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 630.0, + 646.0, + 630.0, + 646.0, + 666.0, + 293.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 630.0, + 900.0, + 630.0, + 900.0, + 666.0, + 668.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 586.5, + 541.0, + 586.5, + 541.0, + 596.5, + 534.0, + 596.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 753.0, + 1405.0, + 753.0, + 1405.0, + 786.0, + 255.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 792.0, + 277.0, + 792.0, + 277.0, + 813.0, + 255.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 784.0, + 1405.0, + 784.0, + 1405.0, + 818.0, + 294.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 823.0, + 277.0, + 823.0, + 277.0, + 843.0, + 255.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 814.0, + 776.0, + 814.0, + 776.0, + 848.0, + 292.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 814.0, + 938.0, + 814.0, + 938.0, + 848.0, + 808.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 322.0, + 278.0, + 322.0, + 278.0, + 350.0, + 239.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 317.0, + 1408.0, + 317.0, + 1408.0, + 352.0, + 291.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 357.0, + 277.0, + 357.0, + 277.0, + 388.0, + 239.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 280.0, + 336.0, + 340.0, + 336.0, + 340.0, + 408.0, + 280.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 336.0, + 871.0, + 336.0, + 871.0, + 408.0, + 391.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 336.0, + 1410.0, + 336.0, + 1410.0, + 408.0, + 894.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 400.0, + 280.0, + 400.0, + 280.0, + 429.0, + 239.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 374.0, + 296.0, + 374.0, + 296.0, + 446.0, + 281.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 374.0, + 1418.0, + 374.0, + 1418.0, + 446.0, + 375.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 431.0, + 278.0, + 431.0, + 278.0, + 458.0, + 239.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 425.0, + 1217.0, + 425.0, + 1217.0, + 460.0, + 293.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 1792.0, + 795.0, + 1792.0, + 795.0, + 1834.0, + 254.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1792.0, + 1374.0, + 1792.0, + 1374.0, + 1834.0, + 817.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1792.0, + 1406.0, + 1792.0, + 1406.0, + 1834.0, + 1402.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1825.0, + 419.0, + 1825.0, + 419.0, + 1861.0, + 253.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1923.0, + 367.0, + 1923.0, + 367.0, + 1997.0, + 241.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1923.0, + 1418.0, + 1923.0, + 1418.0, + 1997.0, + 463.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1972.0, + 597.0, + 1972.0, + 597.0, + 2013.0, + 234.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 1972.0, + 1409.0, + 1972.0, + 1409.0, + 2013.0, + 633.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 201.0, + 613.0, + 201.0, + 613.0, + 240.0, + 236.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 201.0, + 741.0, + 201.0, + 741.0, + 240.0, + 668.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 201.0, + 991.0, + 201.0, + 991.0, + 240.0, + 828.0, + 240.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 295, + 243, + 1421, + 243, + 1421, + 569, + 295, + 569 + ], + "score": 0.983, + "html": "
Size15×1520×1520×2030×1530×2050×1550×20100×20Aug.
RS0.1480.1650.1690.1440.1770.0670.1000.0260.0500.1250.150
RS+op0.1430.1930.1590.1920.2130.1230.126
MWKR0.1910.2050.2580.2330.2350.3280.2180.2170.2770.2390.2280.3520.2510.2490.3440.1680.1730.2410.2060.1790.1760.2550.2390.0830.0910.1440.1350.1950.1970.2750.254
MORSPT
FIFO0.2390.3140.2730.3110.311
L2DParkSchN0.2590.2010.1520.3000.3160.3290.3360.2230.2650.1360.0920.0660.2700.2210.161
0.2490.1940.2920.1720.2460.1900.3190.2370.1590.1380.2120.135
" + }, + { + "category_id": 1, + "poly": [ + 246, + 1199, + 1405, + 1199, + 1405, + 1475, + 246, + 1475 + ], + "score": 0.977 + }, + { + "category_id": 5, + "poly": [ + 518, + 1626, + 1183, + 1626, + 1183, + 1885, + 518, + 1885 + ], + "score": 0.977, + "html": "
MethodMKLA(rdata)LA(edata)LA(vdata)
RS0.2320.0990.1460.031
RS+op0.2540.1130.1680.029
DRL-G0.2540.1110.1500.040
MWKR MOR0.282 0.296 0.4570.125 0.1470.149 0.1790.051 0.061
" + }, + { + "category_id": 1, + "poly": [ + 293, + 835, + 1404, + 835, + 1404, + 989, + 293, + 989 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 244, + 1002, + 1404, + 1002, + 1404, + 1186, + 244, + 1186 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 242, + 621, + 1407, + 621, + 1407, + 744, + 242, + 744 + ], + "score": 0.956 + }, + { + "category_id": 6, + "poly": [ + 549, + 214, + 1149, + 214, + 1149, + 243, + 549, + 243 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 242, + 1509, + 625, + 1509, + 625, + 1542, + 242, + 1542 + ], + "score": 0.91 + }, + { + "category_id": 6, + "poly": [ + 513, + 1596, + 1187, + 1596, + 1187, + 1627, + 513, + 1627 + ], + "score": 0.91 + }, + { + "category_id": 1, + "poly": [ + 245, + 1914, + 1404, + 1914, + 1404, + 2009, + 245, + 2009 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 271, + 780, + 609, + 780, + 609, + 814, + 271, + 814 + ], + "score": 0.875 + }, + { + "category_id": 2, + "poly": [ + 839, + 2061, + 859, + 2061, + 859, + 2086, + 839, + 2086 + ], + "score": 0.82 + }, + { + "category_id": 7, + "poly": [ + 245, + 1914, + 1404, + 1914, + 1404, + 2009, + 245, + 2009 + ], + "score": 0.131 + }, + { + "category_id": 13, + "poly": [ + 872, + 928, + 957, + 928, + 957, + 961, + 872, + 961 + ], + "score": 0.92, + "latex": "( M , O )" + }, + { + "category_id": 13, + "poly": [ + 1260, + 1322, + 1354, + 1322, + 1354, + 1351, + 1260, + 1351 + ], + "score": 0.9, + "latex": "1 5 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 296, + 1352, + 394, + 1352, + 394, + 1381, + 296, + 1381 + ], + "score": 0.89, + "latex": "2 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 423, + 1261, + 510, + 1261, + 510, + 1290, + 423, + 1290 + ], + "score": 0.87, + "latex": "1 0 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 876, + 244, + 958, + 244, + 958, + 274, + 876, + 274 + ], + "score": 0.86, + "latex": "3 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 656, + 244, + 738, + 244, + 738, + 274, + 656, + 274 + ], + "score": 0.86, + "latex": "2 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 766, + 244, + 847, + 244, + 847, + 274, + 766, + 274 + ], + "score": 0.85, + "latex": "3 0 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 986, + 243, + 1067, + 243, + 1067, + 274, + 986, + 274 + ], + "score": 0.84, + "latex": "5 0 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 297, + 1261, + 373, + 1261, + 373, + 1290, + 297, + 1290 + ], + "score": 0.84, + "latex": "5 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 1096, + 244, + 1178, + 244, + 1178, + 274, + 1096, + 274 + ], + "score": 0.84, + "latex": "5 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 547, + 244, + 627, + 244, + 627, + 274, + 547, + 274 + ], + "score": 0.83, + "latex": "2 0 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 1066, + 1231, + 1141, + 1231, + 1141, + 1260, + 1066, + 1260 + ], + "score": 0.83, + "latex": "2 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 876, + 1159, + 912, + 1159, + 912, + 1183, + 876, + 1183 + ], + "score": 0.82, + "latex": "\\pi *" + }, + { + "category_id": 13, + "poly": [ + 1322, + 1231, + 1397, + 1231, + 1397, + 1260, + 1322, + 1260 + ], + "score": 0.81, + "latex": "5 0 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 1151, + 1231, + 1227, + 1231, + 1227, + 1260, + 1151, + 1260 + ], + "score": 0.81, + "latex": "3 0 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 1237, + 1231, + 1312, + 1231, + 1312, + 1260, + 1237, + 1260 + ], + "score": 0.8, + "latex": "3 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 898, + 1231, + 971, + 1231, + 971, + 1261, + 898, + 1261 + ], + "score": 0.79, + "latex": "1 5 { \\times } 1 5" + }, + { + "category_id": 13, + "poly": [ + 375, + 1946, + 454, + 1946, + 454, + 1977, + 375, + 1977 + ], + "score": 0.78, + "latex": "\\mathrm { R S + o p }" + }, + { + "category_id": 13, + "poly": [ + 1208, + 246, + 1301, + 246, + 1301, + 273, + 1208, + 273 + ], + "score": 0.73, + "latex": "1 0 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 981, + 1231, + 1056, + 1231, + 1056, + 1260, + 981, + 1260 + ], + "score": 0.73, + "latex": "2 0 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 496, + 1352, + 580, + 1352, + 580, + 1384, + 496, + 1384 + ], + "score": 0.73, + "latex": "{ \\mathrm { R S + o p } }" + }, + { + "category_id": 13, + "poly": [ + 913, + 1323, + 994, + 1323, + 994, + 1353, + 913, + 1353 + ], + "score": 0.7, + "latex": "{ \\mathrm { R S + o p } }" + }, + { + "category_id": 13, + "poly": [ + 438, + 250, + 516, + 250, + 516, + 273, + 438, + 273 + ], + "score": 0.68, + "latex": "1 5 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 602, + 868, + 685, + 868, + 685, + 898, + 602, + 898 + ], + "score": 0.6, + "latex": "\\mathrm { R S + o p }" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 212.0, + 1148.0, + 212.0, + 1148.0, + 248.0, + 551.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1507.0, + 630.0, + 1507.0, + 630.0, + 1548.0, + 233.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1590.0, + 1189.0, + 1590.0, + 1189.0, + 1634.0, + 508.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 777.0, + 612.0, + 777.0, + 612.0, + 820.0, + 289.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2089.0, + 839.0, + 2089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1920.0, + 278.0, + 1920.0, + 278.0, + 1947.0, + 240.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1912.0, + 1407.0, + 1912.0, + 1407.0, + 1952.0, + 291.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1953.0, + 278.0, + 1953.0, + 278.0, + 1976.0, + 241.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1946.0, + 374.0, + 1946.0, + 374.0, + 1980.0, + 292.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1946.0, + 1405.0, + 1946.0, + 1405.0, + 1980.0, + 455.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1983.0, + 278.0, + 1983.0, + 278.0, + 2007.0, + 241.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1976.0, + 1408.0, + 1976.0, + 1408.0, + 2009.0, + 292.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1206.0, + 276.0, + 1206.0, + 276.0, + 1231.0, + 241.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1197.0, + 1407.0, + 1197.0, + 1407.0, + 1237.0, + 292.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1236.0, + 277.0, + 1236.0, + 277.0, + 1261.0, + 241.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1230.0, + 897.0, + 1230.0, + 897.0, + 1263.0, + 294.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 1230.0, + 980.0, + 1230.0, + 980.0, + 1263.0, + 972.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1230.0, + 1065.0, + 1230.0, + 1065.0, + 1263.0, + 1057.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1230.0, + 1150.0, + 1230.0, + 1150.0, + 1263.0, + 1142.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 1230.0, + 1236.0, + 1230.0, + 1236.0, + 1263.0, + 1228.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 1230.0, + 1321.0, + 1230.0, + 1321.0, + 1263.0, + 1313.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1230.0, + 1406.0, + 1230.0, + 1406.0, + 1263.0, + 1398.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1267.0, + 276.0, + 1267.0, + 276.0, + 1291.0, + 241.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1257.0, + 296.0, + 1257.0, + 296.0, + 1297.0, + 293.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1257.0, + 422.0, + 1257.0, + 422.0, + 1297.0, + 374.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1257.0, + 1408.0, + 1257.0, + 1408.0, + 1297.0, + 511.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1297.0, + 277.0, + 1297.0, + 277.0, + 1325.0, + 239.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1289.0, + 1406.0, + 1289.0, + 1406.0, + 1328.0, + 293.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1327.0, + 277.0, + 1327.0, + 277.0, + 1355.0, + 238.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1321.0, + 912.0, + 1321.0, + 912.0, + 1355.0, + 292.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 1321.0, + 1259.0, + 1321.0, + 1259.0, + 1355.0, + 995.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1321.0, + 1406.0, + 1321.0, + 1406.0, + 1355.0, + 1355.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1358.0, + 277.0, + 1358.0, + 277.0, + 1383.0, + 239.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1349.0, + 495.0, + 1349.0, + 495.0, + 1388.0, + 395.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1349.0, + 1405.0, + 1349.0, + 1405.0, + 1388.0, + 581.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1387.0, + 279.0, + 1387.0, + 279.0, + 1415.0, + 239.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1383.0, + 1404.0, + 1383.0, + 1404.0, + 1415.0, + 294.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1417.0, + 279.0, + 1417.0, + 279.0, + 1445.0, + 238.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1411.0, + 1405.0, + 1411.0, + 1405.0, + 1447.0, + 292.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1449.0, + 277.0, + 1449.0, + 277.0, + 1474.0, + 241.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1443.0, + 1157.0, + 1443.0, + 1157.0, + 1479.0, + 293.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 837.0, + 1405.0, + 837.0, + 1405.0, + 870.0, + 295.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 863.0, + 601.0, + 863.0, + 601.0, + 903.0, + 294.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 863.0, + 1406.0, + 863.0, + 1406.0, + 903.0, + 686.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 899.0, + 1405.0, + 899.0, + 1405.0, + 932.0, + 296.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 927.0, + 871.0, + 927.0, + 871.0, + 964.0, + 295.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 927.0, + 1407.0, + 927.0, + 1407.0, + 964.0, + 958.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 957.0, + 866.0, + 957.0, + 866.0, + 994.0, + 295.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1008.0, + 277.0, + 1008.0, + 277.0, + 1036.0, + 239.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1000.0, + 1405.0, + 1000.0, + 1405.0, + 1040.0, + 291.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1038.0, + 278.0, + 1038.0, + 278.0, + 1066.0, + 239.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1034.0, + 1408.0, + 1034.0, + 1408.0, + 1066.0, + 294.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1069.0, + 278.0, + 1069.0, + 278.0, + 1097.0, + 239.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1061.0, + 1408.0, + 1061.0, + 1408.0, + 1097.0, + 292.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1102.0, + 277.0, + 1102.0, + 277.0, + 1126.0, + 240.0, + 1126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1094.0, + 1405.0, + 1094.0, + 1405.0, + 1130.0, + 294.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1130.0, + 278.0, + 1130.0, + 278.0, + 1157.0, + 239.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1123.0, + 1405.0, + 1123.0, + 1405.0, + 1159.0, + 292.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1160.0, + 278.0, + 1160.0, + 278.0, + 1188.0, + 239.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1155.0, + 875.0, + 1155.0, + 875.0, + 1188.0, + 292.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 1155.0, + 1132.0, + 1155.0, + 1132.0, + 1188.0, + 913.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 628.0, + 276.0, + 628.0, + 276.0, + 655.0, + 239.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 621.0, + 1410.0, + 621.0, + 1410.0, + 660.0, + 292.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 659.0, + 278.0, + 659.0, + 278.0, + 685.0, + 239.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 653.0, + 1407.0, + 653.0, + 1407.0, + 687.0, + 292.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 687.0, + 279.0, + 687.0, + 279.0, + 717.0, + 238.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 678.0, + 1403.0, + 678.0, + 1403.0, + 720.0, + 294.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 717.0, + 280.0, + 717.0, + 280.0, + 747.0, + 238.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 713.0, + 520.0, + 713.0, + 520.0, + 748.0, + 292.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1920.0, + 278.0, + 1920.0, + 278.0, + 1947.0, + 240.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1912.0, + 1407.0, + 1912.0, + 1407.0, + 1952.0, + 291.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1953.0, + 278.0, + 1953.0, + 278.0, + 1976.0, + 241.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1946.0, + 374.0, + 1946.0, + 374.0, + 1980.0, + 292.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1946.0, + 1405.0, + 1946.0, + 1405.0, + 1980.0, + 455.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1983.0, + 278.0, + 1983.0, + 278.0, + 2007.0, + 241.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1976.0, + 1408.0, + 1976.0, + 1408.0, + 2009.0, + 292.0, + 2009.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 507, + 1530, + 1191, + 1530, + 1191, + 1696, + 507, + 1696 + ], + "score": 0.978, + "html": "
MethodMKLA(rdata)LA(edata)LA(vdata)
RSRS+1000.2320.1540.0990.0470.1460.0790.0310.007
DRL-GDRL+1000.2540.1900.1110.0580.1500.0820.0400.014
" + }, + { + "category_id": 1, + "poly": [ + 238, + 947, + 1406, + 947, + 1406, + 1283, + 238, + 1283 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 492, + 1407, + 492, + 1407, + 615, + 296, + 615 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 240, + 1294, + 1405, + 1294, + 1405, + 1449, + 240, + 1449 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 239, + 1732, + 1405, + 1732, + 1405, + 2009, + 239, + 2009 + ], + "score": 0.962 + }, + { + "category_id": 3, + "poly": [ + 298, + 646, + 1404, + 646, + 1404, + 856, + 298, + 856 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 259, + 202, + 1404, + 202, + 1404, + 266, + 259, + 266 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 258, + 279, + 1411, + 279, + 1411, + 372, + 258, + 372 + ], + "score": 0.922 + }, + { + "category_id": 4, + "poly": [ + 394, + 877, + 1290, + 877, + 1290, + 912, + 394, + 912 + ], + "score": 0.92 + }, + { + "category_id": 6, + "poly": [ + 512, + 1500, + 1187, + 1500, + 1187, + 1531, + 512, + 1531 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 269, + 420, + 513, + 420, + 513, + 456, + 269, + 456 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 859, + 2061, + 859, + 2084, + 840, + 2084 + ], + "score": 0.798 + }, + { + "category_id": 13, + "poly": [ + 296, + 1007, + 510, + 1007, + 510, + 1041, + 296, + 1041 + ], + "score": 0.91, + "latex": "\\bar { ( \\cal N , M ) } = ( 1 0 , 1 0 )" + }, + { + "category_id": 13, + "poly": [ + 797, + 1039, + 903, + 1039, + 903, + 1068, + 797, + 1068 + ], + "score": 0.91, + "latex": "2 0 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 967, + 1190, + 1075, + 1190, + 1075, + 1219, + 967, + 1219 + ], + "score": 0.9, + "latex": "2 0 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 675, + 1039, + 766, + 1039, + 766, + 1068, + 675, + 1068 + ], + "score": 0.89, + "latex": "1 5 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 810, + 1190, + 918, + 1190, + 918, + 1220, + 810, + 1220 + ], + "score": 0.86, + "latex": "2 0 0 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 649, + 1855, + 743, + 1855, + 743, + 1884, + 649, + 1884 + ], + "score": 0.86, + "latex": "{ \\mathrm { R S } } { + } 1 0 0 \\ " + }, + { + "category_id": 13, + "poly": [ + 577, + 1190, + 683, + 1190, + 683, + 1220, + 577, + 1220 + ], + "score": 0.86, + "latex": "1 0 0 \\times 2 0" + }, + { + "category_id": 13, + "poly": [ + 500, + 1885, + 599, + 1885, + 599, + 1914, + 500, + 1914 + ], + "score": 0.85, + "latex": "{ \\mathrm { R S } } { + } 1 0 0 \\ " + }, + { + "category_id": 13, + "poly": [ + 460, + 1190, + 565, + 1190, + 565, + 1220, + 460, + 1220 + ], + "score": 0.85, + "latex": "1 0 0 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 694, + 1190, + 800, + 1190, + 800, + 1220, + 694, + 1220 + ], + "score": 0.84, + "latex": "1 5 0 \\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 297, + 1190, + 356, + 1190, + 356, + 1220, + 297, + 1220 + ], + "score": 0.84, + "latex": "\\times 1 5" + }, + { + "category_id": 13, + "poly": [ + 1224, + 1885, + 1342, + 1885, + 1342, + 1914, + 1224, + 1914 + ], + "score": 0.78, + "latex": "\\mathrm { \\ D R L + 1 0 0 }" + }, + { + "category_id": 13, + "poly": [ + 724, + 1358, + 803, + 1358, + 803, + 1388, + 724, + 1388 + ], + "score": 0.67, + "latex": "\\mathrm { R S + o p }" + }, + { + "category_id": 13, + "poly": [ + 987, + 1327, + 1073, + 1327, + 1073, + 1356, + 987, + 1356 + ], + "score": 0.62, + "latex": "1 0 0 \\mathrm { x } 2 0" + }, + { + "category_id": 13, + "poly": [ + 938, + 1795, + 1057, + 1795, + 1057, + 1823, + 938, + 1823 + ], + "score": 0.6, + "latex": "\\mathrm { \\ D R L + 1 0 0 }" + }, + { + "category_id": 13, + "poly": [ + 663, + 341, + 744, + 341, + 744, + 371, + 663, + 371 + ], + "score": 0.3, + "latex": "{ \\mathsf { R S + o p } }" + }, + { + "category_id": 13, + "poly": [ + 580, + 1660, + 641, + 1660, + 641, + 1689, + 580, + 1689 + ], + "score": 0.29, + "latex": "+ 1 0 0" + }, + { + "category_id": 15, + "poly": [ + 273.0, + 632.0, + 399.0, + 632.0, + 399.0, + 767.0, + 273.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 665.0, + 1371.0, + 665.0, + 1371.0, + 705.0, + 1316.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 703.0, + 1385.0, + 703.0, + 1385.0, + 741.0, + 1317.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 732.0, + 376.0, + 732.0, + 376.0, + 767.0, + 318.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 751.0, + 377.0, + 751.0, + 377.0, + 785.0, + 318.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 817.0, + 969.0, + 817.0, + 969.0, + 857.0, + 800.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 781.0, + 532.0, + 781.0, + 532.0, + 814.0, + 444.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 780.0, + 605.0, + 780.0, + 605.0, + 814.0, + 514.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 778.5, + 678.0, + 778.5, + 678.0, + 816.0, + 585.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.25, + 779.5, + 748.25, + 779.5, + 748.25, + 814.0, + 658.25, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.25, + 777.5, + 822.25, + 777.5, + 822.25, + 815.5, + 729.25, + 815.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.75, + 779.5, + 893.75, + 779.5, + 893.75, + 814.5, + 800.75, + 814.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.25, + 777.5, + 966.25, + 777.5, + 966.25, + 816.5, + 871.25, + 816.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 781.5, + 1043.0, + 781.5, + 1043.0, + 819.0, + 939.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 778.5, + 1116.0, + 778.5, + 1116.0, + 819.0, + 1010.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 783.5, + 1185.0, + 783.5, + 1185.0, + 817.5, + 1084.0, + 817.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.25, + 780.5, + 1257.25, + 780.5, + 1257.25, + 819.5, + 1154.25, + 819.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.75, + 783.5, + 1327.75, + 783.5, + 1327.75, + 817.5, + 1227.75, + 817.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.75, + 782.5, + 1400.75, + 782.5, + 1400.75, + 818.5, + 1299.75, + 818.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 873.0, + 1294.0, + 873.0, + 1294.0, + 918.0, + 396.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1494.0, + 1185.0, + 1494.0, + 1185.0, + 1538.0, + 512.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 260.0, + 417.0, + 516.0, + 417.0, + 516.0, + 461.0, + 260.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 951.0, + 279.0, + 951.0, + 279.0, + 980.0, + 237.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 945.0, + 1406.0, + 945.0, + 1406.0, + 984.0, + 291.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 981.0, + 278.0, + 981.0, + 278.0, + 1011.0, + 237.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 977.0, + 1406.0, + 977.0, + 1406.0, + 1011.0, + 292.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1011.0, + 279.0, + 1011.0, + 279.0, + 1041.0, + 237.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1006.0, + 1407.0, + 1006.0, + 1407.0, + 1044.0, + 511.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1043.0, + 279.0, + 1043.0, + 279.0, + 1073.0, + 237.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1037.0, + 674.0, + 1037.0, + 674.0, + 1073.0, + 291.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1037.0, + 796.0, + 1037.0, + 796.0, + 1073.0, + 767.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1037.0, + 1407.0, + 1037.0, + 1407.0, + 1073.0, + 904.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1073.0, + 279.0, + 1073.0, + 279.0, + 1103.0, + 237.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1067.0, + 1407.0, + 1067.0, + 1407.0, + 1104.0, + 294.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1103.0, + 279.0, + 1103.0, + 279.0, + 1133.0, + 237.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1099.0, + 1404.0, + 1099.0, + 1404.0, + 1133.0, + 295.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1134.0, + 279.0, + 1134.0, + 279.0, + 1164.0, + 237.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1127.0, + 1407.0, + 1127.0, + 1407.0, + 1164.0, + 294.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1167.0, + 276.0, + 1167.0, + 276.0, + 1191.0, + 240.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1157.0, + 1408.0, + 1157.0, + 1408.0, + 1199.0, + 291.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1197.0, + 278.0, + 1197.0, + 278.0, + 1221.0, + 238.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1190.0, + 459.0, + 1190.0, + 459.0, + 1224.0, + 357.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 1190.0, + 576.0, + 1190.0, + 576.0, + 1224.0, + 566.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 1190.0, + 693.0, + 1190.0, + 693.0, + 1224.0, + 684.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1190.0, + 809.0, + 1190.0, + 809.0, + 1224.0, + 801.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1190.0, + 966.0, + 1190.0, + 966.0, + 1224.0, + 919.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 1190.0, + 1406.0, + 1190.0, + 1406.0, + 1224.0, + 1076.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1227.0, + 276.0, + 1227.0, + 276.0, + 1253.0, + 240.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1217.0, + 1406.0, + 1217.0, + 1406.0, + 1257.0, + 291.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1257.0, + 276.0, + 1257.0, + 276.0, + 1283.0, + 240.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1249.0, + 1181.0, + 1249.0, + 1181.0, + 1288.0, + 291.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 492.0, + 1408.0, + 492.0, + 1408.0, + 528.0, + 293.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 522.0, + 1407.0, + 522.0, + 1407.0, + 557.0, + 291.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 553.0, + 1409.0, + 553.0, + 1409.0, + 588.0, + 294.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 585.0, + 798.0, + 585.0, + 798.0, + 617.0, + 294.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1301.0, + 276.0, + 1301.0, + 276.0, + 1326.0, + 239.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1296.0, + 1406.0, + 1296.0, + 1406.0, + 1329.0, + 295.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1333.0, + 277.0, + 1333.0, + 277.0, + 1357.0, + 240.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1326.0, + 986.0, + 1326.0, + 986.0, + 1358.0, + 293.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1326.0, + 1405.0, + 1326.0, + 1405.0, + 1358.0, + 1074.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1362.0, + 277.0, + 1362.0, + 277.0, + 1387.0, + 239.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1356.0, + 723.0, + 1356.0, + 723.0, + 1389.0, + 294.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1356.0, + 1408.0, + 1356.0, + 1408.0, + 1389.0, + 804.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1391.0, + 280.0, + 1391.0, + 280.0, + 1419.0, + 237.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1386.0, + 1406.0, + 1386.0, + 1406.0, + 1419.0, + 293.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1422.0, + 280.0, + 1422.0, + 280.0, + 1450.0, + 237.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1417.0, + 639.0, + 1417.0, + 639.0, + 1450.0, + 293.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1740.0, + 277.0, + 1740.0, + 277.0, + 1764.0, + 239.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1732.0, + 1404.0, + 1732.0, + 1404.0, + 1768.0, + 293.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1770.0, + 276.0, + 1770.0, + 276.0, + 1795.0, + 239.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1759.0, + 1408.0, + 1759.0, + 1408.0, + 1802.0, + 291.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1800.0, + 277.0, + 1800.0, + 277.0, + 1825.0, + 239.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1789.0, + 937.0, + 1789.0, + 937.0, + 1830.0, + 292.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1789.0, + 1408.0, + 1789.0, + 1408.0, + 1830.0, + 1058.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1830.0, + 277.0, + 1830.0, + 277.0, + 1855.0, + 239.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1822.0, + 1408.0, + 1822.0, + 1408.0, + 1859.0, + 295.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1860.0, + 277.0, + 1860.0, + 277.0, + 1885.0, + 239.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1853.0, + 648.0, + 1853.0, + 648.0, + 1890.0, + 293.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1853.0, + 1408.0, + 1853.0, + 1408.0, + 1890.0, + 744.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1889.0, + 279.0, + 1889.0, + 279.0, + 1918.0, + 238.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1885.0, + 499.0, + 1885.0, + 499.0, + 1918.0, + 295.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1885.0, + 1223.0, + 1885.0, + 1223.0, + 1918.0, + 600.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.0, + 1885.0, + 1405.0, + 1885.0, + 1405.0, + 1918.0, + 1343.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1919.0, + 279.0, + 1919.0, + 279.0, + 1948.0, + 238.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1914.0, + 1406.0, + 1914.0, + 1406.0, + 1949.0, + 292.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1950.0, + 279.0, + 1950.0, + 279.0, + 1979.0, + 238.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1945.0, + 1406.0, + 1945.0, + 1406.0, + 1982.0, + 293.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1983.0, + 277.0, + 1983.0, + 277.0, + 2008.0, + 239.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1978.0, + 535.0, + 1978.0, + 535.0, + 2012.0, + 295.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 212.0, + 278.0, + 212.0, + 278.0, + 233.0, + 254.0, + 233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 198.0, + 1406.0, + 198.0, + 1406.0, + 240.0, + 286.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 243.0, + 276.0, + 243.0, + 276.0, + 262.0, + 254.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 233.0, + 846.0, + 233.0, + 846.0, + 267.0, + 287.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 287.0, + 278.0, + 287.0, + 278.0, + 310.0, + 254.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 277.0, + 1407.0, + 277.0, + 1407.0, + 315.0, + 286.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 319.0, + 277.0, + 319.0, + 277.0, + 339.0, + 255.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 310.0, + 1404.0, + 310.0, + 1404.0, + 344.0, + 289.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 256.0, + 347.0, + 278.0, + 347.0, + 278.0, + 369.0, + 256.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 340.0, + 662.0, + 340.0, + 662.0, + 374.0, + 292.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 340.0, + 1269.0, + 340.0, + 1269.0, + 374.0, + 745.0, + 374.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 0, + "poly": [ + 275, + 200, + 455, + 200, + 455, + 235, + 275, + 235 + ], + "score": 0.742 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 866, + 2061, + 866, + 2086, + 836, + 2086 + ], + "score": 0.704 + }, + { + "category_id": 1, + "poly": [ + 283, + 1113, + 1401, + 1113, + 1401, + 1207, + 283, + 1207 + ], + "score": 0.524 + }, + { + "category_id": 1, + "poly": [ + 287, + 1588, + 1402, + 1588, + 1402, + 1682, + 287, + 1682 + ], + "score": 0.502 + }, + { + "category_id": 1, + "poly": [ + 273, + 1381, + 1402, + 1381, + 1402, + 1476, + 273, + 1476 + ], + "score": 0.493 + }, + { + "category_id": 1, + "poly": [ + 293, + 253, + 1403, + 253, + 1403, + 345, + 293, + 345 + ], + "score": 0.491 + }, + { + "category_id": 1, + "poly": [ + 273, + 1498, + 1407, + 1498, + 1407, + 1565, + 273, + 1565 + ], + "score": 0.477 + }, + { + "category_id": 1, + "poly": [ + 286, + 937, + 1403, + 937, + 1403, + 1001, + 286, + 1001 + ], + "score": 0.461 + }, + { + "category_id": 1, + "poly": [ + 278, + 460, + 1404, + 460, + 1404, + 553, + 278, + 553 + ], + "score": 0.457 + }, + { + "category_id": 1, + "poly": [ + 277, + 1023, + 1404, + 1023, + 1404, + 1089, + 277, + 1089 + ], + "score": 0.452 + }, + { + "category_id": 1, + "poly": [ + 293, + 1232, + 1406, + 1232, + 1406, + 1356, + 293, + 1356 + ], + "score": 0.443 + }, + { + "category_id": 1, + "poly": [ + 268, + 817, + 1399, + 817, + 1399, + 913, + 268, + 913 + ], + "score": 0.437 + }, + { + "category_id": 1, + "poly": [ + 285, + 1706, + 1402, + 1706, + 1402, + 1772, + 285, + 1772 + ], + "score": 0.432 + }, + { + "category_id": 1, + "poly": [ + 282, + 371, + 1403, + 371, + 1403, + 435, + 282, + 435 + ], + "score": 0.432 + }, + { + "category_id": 1, + "poly": [ + 274, + 1794, + 1407, + 1794, + 1407, + 1860, + 274, + 1860 + ], + "score": 0.42 + }, + { + "category_id": 1, + "poly": [ + 294, + 1884, + 1406, + 1884, + 1406, + 2009, + 294, + 2009 + ], + "score": 0.367 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 866, + 2061, + 866, + 2086, + 836, + 2086 + ], + "score": 0.332 + }, + { + "category_id": 1, + "poly": [ + 275, + 576, + 1403, + 576, + 1403, + 794, + 275, + 794 + ], + "score": 0.176 + }, + { + "category_id": 2, + "poly": [ + 237, + 236, + 283, + 236, + 283, + 2049, + 237, + 2049 + ], + "score": 0.121 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 196.0, + 460.0, + 196.0, + 460.0, + 241.0, + 296.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 256.0, + 283.0, + 256.0, + 283.0, + 294.0, + 233.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 290.0, + 278.0, + 290.0, + 278.0, + 322.0, + 235.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 318.0, + 283.0, + 318.0, + 283.0, + 357.0, + 233.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 375.0, + 283.0, + 375.0, + 283.0, + 413.0, + 233.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 405.0, + 283.0, + 405.0, + 283.0, + 445.0, + 233.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 463.0, + 283.0, + 463.0, + 283.0, + 501.0, + 233.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 497.0, + 281.0, + 497.0, + 281.0, + 529.0, + 235.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 525.0, + 283.0, + 525.0, + 283.0, + 563.0, + 233.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 585.0, + 281.0, + 585.0, + 281.0, + 616.0, + 235.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 612.0, + 283.0, + 612.0, + 283.0, + 650.0, + 233.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 642.0, + 283.0, + 642.0, + 283.0, + 680.0, + 233.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 674.0, + 281.0, + 674.0, + 281.0, + 708.0, + 237.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 702.0, + 283.0, + 702.0, + 283.0, + 740.0, + 233.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 736.0, + 281.0, + 736.0, + 281.0, + 768.0, + 235.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 764.0, + 283.0, + 764.0, + 283.0, + 810.0, + 235.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 824.0, + 278.0, + 824.0, + 278.0, + 858.0, + 235.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 852.0, + 283.0, + 852.0, + 283.0, + 889.0, + 233.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 883.0, + 283.0, + 883.0, + 283.0, + 925.0, + 235.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 943.0, + 281.0, + 943.0, + 281.0, + 975.0, + 237.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 973.0, + 283.0, + 973.0, + 283.0, + 1013.0, + 235.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1031.0, + 281.0, + 1031.0, + 281.0, + 1063.0, + 235.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1059.0, + 281.0, + 1059.0, + 281.0, + 1103.0, + 235.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1117.0, + 283.0, + 1117.0, + 283.0, + 1154.0, + 233.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1148.0, + 283.0, + 1148.0, + 283.0, + 1186.0, + 233.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1178.0, + 283.0, + 1178.0, + 283.0, + 1224.0, + 235.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1238.0, + 278.0, + 1238.0, + 278.0, + 1272.0, + 235.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1266.0, + 283.0, + 1266.0, + 283.0, + 1304.0, + 233.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1296.0, + 283.0, + 1296.0, + 283.0, + 1334.0, + 233.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1330.0, + 281.0, + 1330.0, + 281.0, + 1364.0, + 237.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1384.0, + 283.0, + 1384.0, + 283.0, + 1421.0, + 233.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1416.0, + 283.0, + 1416.0, + 283.0, + 1453.0, + 233.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1447.0, + 281.0, + 1447.0, + 281.0, + 1485.0, + 237.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1505.0, + 281.0, + 1505.0, + 281.0, + 1537.0, + 235.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1537.0, + 281.0, + 1537.0, + 281.0, + 1575.0, + 237.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1591.0, + 283.0, + 1591.0, + 283.0, + 1631.0, + 233.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1625.0, + 281.0, + 1625.0, + 281.0, + 1657.0, + 237.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1653.0, + 283.0, + 1653.0, + 283.0, + 1694.0, + 235.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1710.0, + 283.0, + 1710.0, + 283.0, + 1748.0, + 233.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1744.0, + 281.0, + 1744.0, + 281.0, + 1776.0, + 237.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1798.0, + 283.0, + 1798.0, + 283.0, + 1836.0, + 233.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1830.0, + 283.0, + 1830.0, + 283.0, + 1870.0, + 235.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1886.0, + 283.0, + 1886.0, + 283.0, + 1924.0, + 233.0, + 1924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1918.0, + 283.0, + 1918.0, + 283.0, + 1956.0, + 233.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1950.0, + 281.0, + 1950.0, + 281.0, + 1981.0, + 235.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1977.0, + 283.0, + 1977.0, + 283.0, + 2015.0, + 233.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1109.0, + 1404.0, + 1109.0, + 1404.0, + 1152.0, + 293.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1144.0, + 1405.0, + 1144.0, + 1405.0, + 1181.0, + 321.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1177.0, + 477.0, + 1177.0, + 477.0, + 1205.0, + 324.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1586.0, + 1406.0, + 1586.0, + 1406.0, + 1625.0, + 293.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1620.0, + 1408.0, + 1620.0, + 1408.0, + 1654.0, + 323.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1647.0, + 461.0, + 1647.0, + 461.0, + 1683.0, + 322.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1381.0, + 1406.0, + 1381.0, + 1406.0, + 1419.0, + 295.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1413.0, + 1406.0, + 1413.0, + 1406.0, + 1447.0, + 324.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1441.0, + 1190.0, + 1441.0, + 1190.0, + 1477.0, + 322.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 252.0, + 1406.0, + 252.0, + 1406.0, + 289.0, + 295.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 281.0, + 1405.0, + 281.0, + 1405.0, + 323.0, + 320.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 314.0, + 586.0, + 314.0, + 586.0, + 346.0, + 321.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1495.0, + 1408.0, + 1495.0, + 1408.0, + 1538.0, + 292.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1532.0, + 791.0, + 1532.0, + 791.0, + 1564.0, + 322.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 933.0, + 1407.0, + 933.0, + 1407.0, + 975.0, + 293.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 968.0, + 1149.0, + 968.0, + 1149.0, + 1004.0, + 322.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 457.0, + 1404.0, + 457.0, + 1404.0, + 498.0, + 293.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 490.0, + 1404.0, + 490.0, + 1404.0, + 526.0, + 320.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 522.0, + 506.0, + 522.0, + 506.0, + 553.0, + 321.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1022.0, + 1407.0, + 1022.0, + 1407.0, + 1062.0, + 293.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1055.0, + 690.0, + 1055.0, + 690.0, + 1091.0, + 324.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1232.0, + 1408.0, + 1232.0, + 1408.0, + 1267.0, + 295.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1260.0, + 1406.0, + 1260.0, + 1406.0, + 1299.0, + 323.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1293.0, + 1405.0, + 1293.0, + 1405.0, + 1330.0, + 321.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1324.0, + 801.0, + 1324.0, + 801.0, + 1357.0, + 323.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 814.0, + 1404.0, + 814.0, + 1404.0, + 856.0, + 289.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 263.0, + 859.0, + 277.0, + 859.0, + 277.0, + 876.0, + 263.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 850.0, + 1404.0, + 850.0, + 1404.0, + 882.0, + 322.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 264.0, + 892.0, + 274.0, + 892.0, + 274.0, + 905.0, + 264.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 881.0, + 1066.0, + 881.0, + 1066.0, + 915.0, + 321.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1703.0, + 1405.0, + 1703.0, + 1405.0, + 1744.0, + 292.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1738.0, + 543.0, + 1738.0, + 543.0, + 1771.0, + 322.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 371.0, + 1405.0, + 371.0, + 1405.0, + 407.0, + 294.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 402.0, + 1115.0, + 402.0, + 1115.0, + 438.0, + 324.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1792.0, + 1404.0, + 1792.0, + 1404.0, + 1834.0, + 293.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1827.0, + 1352.0, + 1827.0, + 1352.0, + 1861.0, + 322.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1880.0, + 1407.0, + 1880.0, + 1407.0, + 1923.0, + 293.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1914.0, + 1408.0, + 1914.0, + 1408.0, + 1950.0, + 321.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1945.0, + 1411.0, + 1945.0, + 1411.0, + 1982.0, + 320.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1976.0, + 1233.0, + 1976.0, + 1233.0, + 2012.0, + 321.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 576.0, + 1406.0, + 576.0, + 1406.0, + 614.0, + 294.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 608.0, + 1408.0, + 608.0, + 1408.0, + 644.0, + 321.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 640.0, + 1406.0, + 640.0, + 1406.0, + 675.0, + 323.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 670.0, + 1406.0, + 670.0, + 1406.0, + 705.0, + 322.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 700.0, + 1408.0, + 700.0, + 1408.0, + 735.0, + 322.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 731.0, + 1406.0, + 731.0, + 1406.0, + 766.0, + 322.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 762.0, + 1086.0, + 762.0, + 1086.0, + 797.0, + 323.0, + 797.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 835, + 2061, + 863, + 2061, + 863, + 2086, + 835, + 2086 + ], + "score": 0.721 + }, + { + "category_id": 1, + "poly": [ + 259, + 922, + 1409, + 922, + 1409, + 1015, + 259, + 1015 + ], + "score": 0.705 + }, + { + "category_id": 1, + "poly": [ + 240, + 1042, + 1407, + 1042, + 1407, + 1165, + 240, + 1165 + ], + "score": 0.68 + }, + { + "category_id": 1, + "poly": [ + 240, + 1192, + 1407, + 1192, + 1407, + 1345, + 240, + 1345 + ], + "score": 0.674 + }, + { + "category_id": 1, + "poly": [ + 240, + 1373, + 1409, + 1373, + 1409, + 1528, + 240, + 1528 + ], + "score": 0.672 + }, + { + "category_id": 1, + "poly": [ + 256, + 682, + 1406, + 682, + 1406, + 776, + 256, + 776 + ], + "score": 0.663 + }, + { + "category_id": 1, + "poly": [ + 260, + 803, + 1406, + 803, + 1406, + 897, + 260, + 897 + ], + "score": 0.66 + }, + { + "category_id": 1, + "poly": [ + 260, + 562, + 1407, + 562, + 1407, + 656, + 260, + 656 + ], + "score": 0.622 + }, + { + "category_id": 1, + "poly": [ + 254, + 352, + 1403, + 352, + 1403, + 415, + 254, + 415 + ], + "score": 0.61 + }, + { + "category_id": 1, + "poly": [ + 278, + 443, + 1402, + 443, + 1402, + 535, + 278, + 535 + ], + "score": 0.568 + }, + { + "category_id": 1, + "poly": [ + 242, + 1615, + 1406, + 1615, + 1406, + 1736, + 242, + 1736 + ], + "score": 0.566 + }, + { + "category_id": 1, + "poly": [ + 252, + 1763, + 1405, + 1763, + 1405, + 1857, + 252, + 1857 + ], + "score": 0.56 + }, + { + "category_id": 1, + "poly": [ + 244, + 202, + 1406, + 202, + 1406, + 327, + 244, + 327 + ], + "score": 0.555 + }, + { + "category_id": 1, + "poly": [ + 243, + 1553, + 1130, + 1553, + 1130, + 1588, + 243, + 1588 + ], + "score": 0.534 + }, + { + "category_id": 1, + "poly": [ + 242, + 1884, + 1408, + 1884, + 1408, + 2008, + 242, + 2008 + ], + "score": 0.504 + }, + { + "category_id": 2, + "poly": [ + 835, + 2061, + 863, + 2061, + 863, + 2086, + 835, + 2086 + ], + "score": 0.352 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 868.0, + 2058.0, + 868.0, + 2098.0, + 832.0, + 2098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 868.0, + 2058.0, + 868.0, + 2098.0, + 832.0, + 2098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 931.0, + 278.0, + 931.0, + 278.0, + 952.0, + 255.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 919.0, + 1408.0, + 919.0, + 1408.0, + 960.0, + 286.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 962.0, + 277.0, + 962.0, + 277.0, + 983.0, + 253.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 949.0, + 1411.0, + 949.0, + 1411.0, + 993.0, + 317.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 991.0, + 275.0, + 991.0, + 275.0, + 1014.0, + 253.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 986.0, + 564.0, + 986.0, + 564.0, + 1013.0, + 322.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1049.0, + 277.0, + 1049.0, + 277.0, + 1076.0, + 240.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1044.0, + 1408.0, + 1044.0, + 1408.0, + 1078.0, + 294.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1077.0, + 280.0, + 1077.0, + 280.0, + 1108.0, + 238.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1073.0, + 1406.0, + 1073.0, + 1406.0, + 1108.0, + 323.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1108.0, + 280.0, + 1108.0, + 280.0, + 1137.0, + 238.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1103.0, + 1403.0, + 1103.0, + 1403.0, + 1138.0, + 322.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1138.0, + 280.0, + 1138.0, + 280.0, + 1169.0, + 238.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1136.0, + 577.0, + 1136.0, + 577.0, + 1166.0, + 323.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1199.0, + 278.0, + 1199.0, + 278.0, + 1224.0, + 240.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1193.0, + 1406.0, + 1193.0, + 1406.0, + 1226.0, + 294.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1228.0, + 278.0, + 1228.0, + 278.0, + 1257.0, + 238.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1221.0, + 1406.0, + 1221.0, + 1406.0, + 1259.0, + 322.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1261.0, + 277.0, + 1261.0, + 277.0, + 1286.0, + 239.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1255.0, + 1404.0, + 1255.0, + 1404.0, + 1288.0, + 323.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1288.0, + 280.0, + 1288.0, + 280.0, + 1317.0, + 238.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1283.0, + 1408.0, + 1283.0, + 1408.0, + 1320.0, + 319.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1319.0, + 280.0, + 1319.0, + 280.0, + 1348.0, + 238.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1313.0, + 437.0, + 1313.0, + 437.0, + 1348.0, + 321.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1380.0, + 276.0, + 1380.0, + 276.0, + 1405.0, + 240.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1375.0, + 1409.0, + 1375.0, + 1409.0, + 1408.0, + 296.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1409.0, + 280.0, + 1409.0, + 280.0, + 1437.0, + 238.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1401.0, + 1409.0, + 1401.0, + 1409.0, + 1441.0, + 320.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1441.0, + 277.0, + 1441.0, + 277.0, + 1466.0, + 239.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1431.0, + 1410.0, + 1431.0, + 1410.0, + 1472.0, + 322.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1470.0, + 280.0, + 1470.0, + 280.0, + 1498.0, + 238.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1460.0, + 1411.0, + 1460.0, + 1411.0, + 1505.0, + 320.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1500.0, + 280.0, + 1500.0, + 280.0, + 1529.0, + 238.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1496.0, + 1247.0, + 1496.0, + 1247.0, + 1529.0, + 322.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 690.0, + 278.0, + 690.0, + 278.0, + 714.0, + 250.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 680.0, + 1405.0, + 680.0, + 1405.0, + 721.0, + 293.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 721.0, + 278.0, + 721.0, + 278.0, + 745.0, + 249.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 713.0, + 1403.0, + 713.0, + 1403.0, + 747.0, + 323.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 751.0, + 276.0, + 751.0, + 276.0, + 775.0, + 245.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 741.0, + 1209.0, + 741.0, + 1209.0, + 780.0, + 321.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 257.0, + 811.0, + 276.0, + 811.0, + 276.0, + 832.0, + 257.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 801.0, + 1404.0, + 801.0, + 1404.0, + 839.0, + 288.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 842.0, + 276.0, + 842.0, + 276.0, + 863.0, + 255.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 833.0, + 1408.0, + 833.0, + 1408.0, + 871.0, + 320.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 872.0, + 276.0, + 872.0, + 276.0, + 894.0, + 254.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 861.0, + 1199.0, + 861.0, + 1199.0, + 900.0, + 319.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 573.0, + 274.0, + 573.0, + 274.0, + 591.0, + 258.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 562.0, + 1402.0, + 562.0, + 1402.0, + 599.0, + 289.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 603.0, + 275.0, + 603.0, + 275.0, + 622.0, + 254.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 592.0, + 1410.0, + 592.0, + 1410.0, + 630.0, + 322.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 632.0, + 276.0, + 632.0, + 276.0, + 655.0, + 254.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 624.0, + 927.0, + 624.0, + 927.0, + 657.0, + 320.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 359.0, + 279.0, + 359.0, + 279.0, + 385.0, + 248.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 352.0, + 1409.0, + 352.0, + 1409.0, + 389.0, + 292.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 387.0, + 281.0, + 387.0, + 281.0, + 417.0, + 247.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 382.0, + 827.0, + 382.0, + 827.0, + 418.0, + 320.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 441.0, + 1405.0, + 441.0, + 1405.0, + 479.0, + 294.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 472.0, + 1407.0, + 472.0, + 1407.0, + 508.0, + 322.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 502.0, + 732.0, + 502.0, + 732.0, + 539.0, + 325.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1619.0, + 278.0, + 1619.0, + 278.0, + 1645.0, + 239.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1609.0, + 1406.0, + 1609.0, + 1406.0, + 1652.0, + 291.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1648.0, + 279.0, + 1648.0, + 279.0, + 1678.0, + 237.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1641.0, + 1406.0, + 1641.0, + 1406.0, + 1680.0, + 321.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1678.0, + 279.0, + 1678.0, + 279.0, + 1709.0, + 238.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1674.0, + 1403.0, + 1674.0, + 1403.0, + 1709.0, + 320.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1711.0, + 278.0, + 1711.0, + 278.0, + 1738.0, + 239.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1704.0, + 578.0, + 1704.0, + 578.0, + 1738.0, + 321.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1770.0, + 276.0, + 1770.0, + 276.0, + 1795.0, + 248.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1761.0, + 1407.0, + 1761.0, + 1407.0, + 1799.0, + 291.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1800.0, + 278.0, + 1800.0, + 278.0, + 1826.0, + 248.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1793.0, + 1404.0, + 1793.0, + 1404.0, + 1827.0, + 323.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1831.0, + 279.0, + 1831.0, + 279.0, + 1860.0, + 246.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1826.0, + 796.0, + 1826.0, + 796.0, + 1860.0, + 323.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 208.0, + 276.0, + 208.0, + 276.0, + 235.0, + 240.0, + 235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 198.0, + 1406.0, + 198.0, + 1406.0, + 241.0, + 293.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 238.0, + 279.0, + 238.0, + 279.0, + 267.0, + 237.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 229.0, + 1406.0, + 229.0, + 1406.0, + 272.0, + 320.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 269.0, + 277.0, + 269.0, + 277.0, + 295.0, + 239.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 259.0, + 1407.0, + 259.0, + 1407.0, + 300.0, + 320.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 298.0, + 279.0, + 298.0, + 279.0, + 328.0, + 237.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 294.0, + 1104.0, + 294.0, + 1104.0, + 329.0, + 323.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1552.0, + 1134.0, + 1552.0, + 1134.0, + 1591.0, + 236.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1890.0, + 278.0, + 1890.0, + 278.0, + 1917.0, + 239.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1884.0, + 1408.0, + 1884.0, + 1408.0, + 1919.0, + 294.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1918.0, + 279.0, + 1918.0, + 279.0, + 1949.0, + 238.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1912.0, + 1407.0, + 1912.0, + 1407.0, + 1950.0, + 320.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1949.0, + 279.0, + 1949.0, + 279.0, + 1980.0, + 238.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1945.0, + 1408.0, + 1945.0, + 1408.0, + 1980.0, + 323.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1980.0, + 279.0, + 1980.0, + 279.0, + 2009.0, + 237.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1975.0, + 1134.0, + 1975.0, + 1134.0, + 2010.0, + 323.0, + 2010.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.819 + }, + { + "category_id": 1, + "poly": [ + 235, + 233, + 1409, + 233, + 1409, + 2029, + 235, + 2029 + ], + "score": 0.724 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 238.0, + 281.0, + 238.0, + 281.0, + 268.0, + 239.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 228.0, + 1409.0, + 228.0, + 1409.0, + 270.0, + 318.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 268.0, + 279.0, + 268.0, + 279.0, + 300.0, + 237.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 262.0, + 1407.0, + 262.0, + 1407.0, + 300.0, + 320.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 300.0, + 279.0, + 300.0, + 279.0, + 329.0, + 239.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 294.0, + 1409.0, + 294.0, + 1409.0, + 331.0, + 322.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 329.0, + 281.0, + 329.0, + 281.0, + 359.0, + 237.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 321.0, + 1407.0, + 321.0, + 1407.0, + 361.0, + 320.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 359.0, + 279.0, + 359.0, + 279.0, + 388.0, + 237.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 349.0, + 1409.0, + 349.0, + 1409.0, + 394.0, + 318.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 386.0, + 283.0, + 386.0, + 283.0, + 422.0, + 235.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 384.0, + 1401.0, + 384.0, + 1401.0, + 422.0, + 322.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 418.0, + 283.0, + 418.0, + 283.0, + 454.0, + 235.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 418.0, + 1073.0, + 418.0, + 1073.0, + 450.0, + 322.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 471.0, + 281.0, + 471.0, + 281.0, + 501.0, + 237.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 461.0, + 1407.0, + 461.0, + 1407.0, + 507.0, + 291.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 499.0, + 281.0, + 499.0, + 281.0, + 537.0, + 235.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 501.0, + 577.0, + 501.0, + 577.0, + 533.0, + 326.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 554.0, + 281.0, + 554.0, + 281.0, + 584.0, + 239.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 548.0, + 1407.0, + 548.0, + 1407.0, + 586.0, + 292.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 584.0, + 281.0, + 584.0, + 281.0, + 614.0, + 237.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 578.0, + 1409.0, + 578.0, + 1409.0, + 616.0, + 322.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 610.0, + 282.0, + 610.0, + 282.0, + 648.0, + 238.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 610.0, + 1075.0, + 610.0, + 1075.0, + 647.0, + 322.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 665.0, + 279.0, + 665.0, + 279.0, + 695.0, + 237.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 657.0, + 1409.0, + 657.0, + 1409.0, + 700.0, + 292.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 693.0, + 283.0, + 693.0, + 283.0, + 728.0, + 235.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 683.0, + 1407.0, + 683.0, + 1407.0, + 734.0, + 316.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 726.0, + 279.0, + 726.0, + 279.0, + 758.0, + 239.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 718.0, + 693.0, + 718.0, + 693.0, + 756.0, + 320.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 777.0, + 283.0, + 777.0, + 283.0, + 805.0, + 237.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 772.0, + 1405.0, + 772.0, + 1405.0, + 809.0, + 291.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 805.0, + 283.0, + 805.0, + 283.0, + 841.0, + 235.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 803.0, + 1407.0, + 803.0, + 1407.0, + 841.0, + 320.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 837.0, + 283.0, + 837.0, + 283.0, + 872.0, + 235.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 826.0, + 440.0, + 826.0, + 440.0, + 871.0, + 319.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 888.0, + 281.0, + 888.0, + 281.0, + 918.0, + 239.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 884.0, + 1405.0, + 884.0, + 1405.0, + 922.0, + 292.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 920.0, + 281.0, + 920.0, + 281.0, + 949.0, + 237.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 914.0, + 1407.0, + 914.0, + 1407.0, + 951.0, + 320.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 947.0, + 283.0, + 947.0, + 283.0, + 983.0, + 235.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 939.0, + 1411.0, + 939.0, + 1411.0, + 987.0, + 318.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 981.0, + 279.0, + 981.0, + 279.0, + 1013.0, + 237.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 969.0, + 1407.0, + 969.0, + 1407.0, + 1017.0, + 318.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1011.0, + 279.0, + 1011.0, + 279.0, + 1040.0, + 239.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1003.0, + 1411.0, + 1003.0, + 1411.0, + 1046.0, + 320.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1040.0, + 279.0, + 1040.0, + 279.0, + 1072.0, + 239.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1034.0, + 842.0, + 1034.0, + 842.0, + 1076.0, + 318.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1091.0, + 281.0, + 1091.0, + 281.0, + 1123.0, + 239.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1088.0, + 1407.0, + 1088.0, + 1407.0, + 1125.0, + 292.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1121.0, + 283.0, + 1121.0, + 283.0, + 1157.0, + 235.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1117.0, + 1407.0, + 1117.0, + 1407.0, + 1155.0, + 322.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1155.0, + 281.0, + 1155.0, + 281.0, + 1184.0, + 237.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1149.0, + 1146.0, + 1149.0, + 1146.0, + 1186.0, + 322.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1204.0, + 281.0, + 1204.0, + 281.0, + 1234.0, + 239.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1194.0, + 1407.0, + 1194.0, + 1407.0, + 1240.0, + 289.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1234.0, + 279.0, + 1234.0, + 279.0, + 1265.0, + 239.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1230.0, + 1316.0, + 1230.0, + 1316.0, + 1267.0, + 322.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1285.0, + 281.0, + 1285.0, + 281.0, + 1315.0, + 239.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1281.0, + 1405.0, + 1281.0, + 1405.0, + 1319.0, + 292.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1317.0, + 279.0, + 1317.0, + 279.0, + 1348.0, + 237.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1313.0, + 1294.0, + 1313.0, + 1294.0, + 1348.0, + 320.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1370.0, + 281.0, + 1370.0, + 281.0, + 1400.0, + 239.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1360.0, + 1407.0, + 1360.0, + 1407.0, + 1408.0, + 289.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1400.0, + 279.0, + 1400.0, + 279.0, + 1431.0, + 237.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1400.0, + 687.0, + 1400.0, + 687.0, + 1431.0, + 322.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1453.0, + 281.0, + 1453.0, + 281.0, + 1483.0, + 237.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1441.0, + 1409.0, + 1441.0, + 1409.0, + 1491.0, + 289.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1483.0, + 279.0, + 1483.0, + 279.0, + 1514.0, + 237.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1479.0, + 1409.0, + 1479.0, + 1409.0, + 1516.0, + 322.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1512.0, + 279.0, + 1512.0, + 279.0, + 1544.0, + 237.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1508.0, + 1405.0, + 1508.0, + 1405.0, + 1546.0, + 322.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1544.0, + 279.0, + 1544.0, + 279.0, + 1575.0, + 237.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1540.0, + 1407.0, + 1540.0, + 1407.0, + 1577.0, + 320.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1571.0, + 283.0, + 1571.0, + 283.0, + 1607.0, + 235.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1569.0, + 1409.0, + 1569.0, + 1409.0, + 1607.0, + 322.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1603.0, + 279.0, + 1603.0, + 279.0, + 1635.0, + 237.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1599.0, + 1407.0, + 1599.0, + 1407.0, + 1637.0, + 322.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1635.0, + 281.0, + 1635.0, + 281.0, + 1664.0, + 237.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1631.0, + 1409.0, + 1631.0, + 1409.0, + 1668.0, + 322.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1664.0, + 281.0, + 1664.0, + 281.0, + 1696.0, + 239.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1660.0, + 1407.0, + 1660.0, + 1407.0, + 1698.0, + 322.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1696.0, + 279.0, + 1696.0, + 279.0, + 1727.0, + 237.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1690.0, + 1202.0, + 1690.0, + 1202.0, + 1727.0, + 322.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1747.0, + 281.0, + 1747.0, + 281.0, + 1777.0, + 237.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1741.0, + 1405.0, + 1741.0, + 1405.0, + 1779.0, + 292.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1777.0, + 281.0, + 1777.0, + 281.0, + 1806.0, + 237.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1773.0, + 1405.0, + 1773.0, + 1405.0, + 1810.0, + 322.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1804.0, + 283.0, + 1804.0, + 283.0, + 1840.0, + 235.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1803.0, + 1405.0, + 1803.0, + 1405.0, + 1840.0, + 320.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1838.0, + 281.0, + 1838.0, + 281.0, + 1868.0, + 237.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1834.0, + 1405.0, + 1834.0, + 1405.0, + 1872.0, + 318.0, + 1872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1866.0, + 283.0, + 1866.0, + 283.0, + 1901.0, + 235.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1858.0, + 1403.0, + 1858.0, + 1403.0, + 1905.0, + 318.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1899.0, + 279.0, + 1899.0, + 279.0, + 1929.0, + 237.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1889.0, + 613.0, + 1889.0, + 613.0, + 1937.0, + 317.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1947.0, + 285.0, + 1947.0, + 285.0, + 1982.0, + 237.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1945.0, + 1405.0, + 1945.0, + 1405.0, + 1982.0, + 292.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1972.0, + 284.0, + 1972.0, + 284.0, + 2015.0, + 236.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1974.0, + 1170.0, + 1974.0, + 1170.0, + 2012.0, + 320.0, + 2012.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 236, + 200, + 1410, + 200, + 1410, + 843, + 236, + 843 + ], + "score": 0.846 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.829 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 870.0, + 2057.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 210.0, + 278.0, + 210.0, + 278.0, + 236.0, + 240.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 203.0, + 1408.0, + 203.0, + 1408.0, + 242.0, + 293.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 235.0, + 280.0, + 235.0, + 280.0, + 271.0, + 238.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 235.0, + 1406.0, + 235.0, + 1406.0, + 269.0, + 324.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 269.0, + 278.0, + 269.0, + 278.0, + 299.0, + 238.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 265.0, + 840.0, + 265.0, + 840.0, + 299.0, + 323.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 322.0, + 279.0, + 322.0, + 279.0, + 352.0, + 238.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 311.0, + 1407.0, + 311.0, + 1407.0, + 359.0, + 292.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 352.0, + 279.0, + 352.0, + 279.0, + 382.0, + 238.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 347.0, + 1406.0, + 347.0, + 1406.0, + 385.0, + 321.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 382.0, + 279.0, + 382.0, + 279.0, + 412.0, + 238.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 377.0, + 1408.0, + 377.0, + 1408.0, + 415.0, + 320.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 412.0, + 279.0, + 412.0, + 279.0, + 442.0, + 238.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 405.0, + 1406.0, + 405.0, + 1406.0, + 446.0, + 320.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 444.0, + 279.0, + 444.0, + 279.0, + 474.0, + 238.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 437.0, + 1144.0, + 437.0, + 1144.0, + 474.0, + 321.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 494.0, + 278.0, + 494.0, + 278.0, + 524.0, + 238.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 490.0, + 1407.0, + 490.0, + 1407.0, + 528.0, + 295.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 526.0, + 279.0, + 526.0, + 279.0, + 556.0, + 238.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 519.0, + 1411.0, + 519.0, + 1411.0, + 561.0, + 321.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 556.0, + 279.0, + 556.0, + 279.0, + 586.0, + 238.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 549.0, + 1139.0, + 549.0, + 1139.0, + 591.0, + 320.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 608.0, + 278.0, + 608.0, + 278.0, + 638.0, + 238.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 602.0, + 1406.0, + 602.0, + 1406.0, + 640.0, + 297.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 639.0, + 278.0, + 639.0, + 278.0, + 669.0, + 238.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 633.0, + 1407.0, + 633.0, + 1407.0, + 672.0, + 320.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 669.0, + 279.0, + 669.0, + 279.0, + 699.0, + 238.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 662.0, + 1407.0, + 662.0, + 1407.0, + 702.0, + 319.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 699.0, + 278.0, + 699.0, + 278.0, + 730.0, + 238.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 694.0, + 1193.0, + 694.0, + 1193.0, + 732.0, + 323.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 754.0, + 276.0, + 754.0, + 276.0, + 780.0, + 239.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 747.0, + 1407.0, + 747.0, + 1407.0, + 785.0, + 292.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 782.0, + 279.0, + 782.0, + 279.0, + 812.0, + 238.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 777.0, + 1404.0, + 777.0, + 1404.0, + 815.0, + 324.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 812.0, + 279.0, + 812.0, + 279.0, + 842.0, + 238.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 806.0, + 1406.0, + 806.0, + 1406.0, + 848.0, + 323.0, + 848.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/dev/83LJRUzXWj/83LJRUzXWj_origin.pdf b/parse/dev/83LJRUzXWj/83LJRUzXWj_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..2998ede56be38f958f0a9aa99a4a5207da89ab19 --- /dev/null +++ b/parse/dev/83LJRUzXWj/83LJRUzXWj_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42fcec027856616cdbc4c22372fc39f004b73dc4c68e1c7eebdad89357bc0d5f +size 2765181 diff --git a/parse/dev/99RpBVpLiX/images/00329b6d87cc4a0de3cf02a9d8f2c4d8c0f68dfffea4c73f3e9c66568ffb67b0.jpg b/parse/dev/99RpBVpLiX/images/00329b6d87cc4a0de3cf02a9d8f2c4d8c0f68dfffea4c73f3e9c66568ffb67b0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..91900dc55a13b2e23cc153db22e1ff88b54f20bf --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/00329b6d87cc4a0de3cf02a9d8f2c4d8c0f68dfffea4c73f3e9c66568ffb67b0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8fe4d3fbf7c781fb2083e579ab226194c72630fe7e39bac2c9a1a91913045bd5 +size 37353 diff --git a/parse/dev/99RpBVpLiX/images/01a9cd5fe45a7a1c3ebe714d16f014c0e9e8c733d0efdbc31c5eb8c7d4fa7229.jpg b/parse/dev/99RpBVpLiX/images/01a9cd5fe45a7a1c3ebe714d16f014c0e9e8c733d0efdbc31c5eb8c7d4fa7229.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b605f40111a3475af3a39a4458ec2b84a66771f3 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/01a9cd5fe45a7a1c3ebe714d16f014c0e9e8c733d0efdbc31c5eb8c7d4fa7229.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b57f13a88cda2aeeb6e457978bd75de142678447ada310b50df0f378ddf8e7a7 +size 26320 diff --git a/parse/dev/99RpBVpLiX/images/06e9b679d9ac75e3bfb88595664412630cad8789e06e36abbd471059ba6fb3e2.jpg b/parse/dev/99RpBVpLiX/images/06e9b679d9ac75e3bfb88595664412630cad8789e06e36abbd471059ba6fb3e2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..107bec02b8f5debee38c8657e0a3c2eae28d8cbe --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/06e9b679d9ac75e3bfb88595664412630cad8789e06e36abbd471059ba6fb3e2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:37d462fc219d270d787a2f356798ec741fff1b89347ad992495d44e5e85f3389 +size 21393 diff --git a/parse/dev/99RpBVpLiX/images/074803d1e837740d9ab89559c55dcd530a5e5607166a2326ebfb22f16997bd81.jpg b/parse/dev/99RpBVpLiX/images/074803d1e837740d9ab89559c55dcd530a5e5607166a2326ebfb22f16997bd81.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a545de9e82a886e1d0933715063a202b1c6b1ae4 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/074803d1e837740d9ab89559c55dcd530a5e5607166a2326ebfb22f16997bd81.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f26754046c2047ab4b7c42aa1d6ded0560ba66f64d75478b56e4daf3858a3815 +size 18754 diff --git a/parse/dev/99RpBVpLiX/images/0d7bcb5b933f53d40dd69f6c615b1888feb82cafbdaab98f22f7a7d5222762f1.jpg b/parse/dev/99RpBVpLiX/images/0d7bcb5b933f53d40dd69f6c615b1888feb82cafbdaab98f22f7a7d5222762f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6006d6f0446afc38ea2773a8d22a1764e228ca01 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/0d7bcb5b933f53d40dd69f6c615b1888feb82cafbdaab98f22f7a7d5222762f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7570a3f1606694722a8ff6404012bd9472f2bf80a28818e618a440eff40fadc +size 25099 diff --git a/parse/dev/99RpBVpLiX/images/0e7973b63b7577d7391e1907fc14f1f067120ba87250308c8d5f00b30d840c9c.jpg b/parse/dev/99RpBVpLiX/images/0e7973b63b7577d7391e1907fc14f1f067120ba87250308c8d5f00b30d840c9c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e7798910c6e038b38ee37ab673975eb412691a5c --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/0e7973b63b7577d7391e1907fc14f1f067120ba87250308c8d5f00b30d840c9c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5663234dd3da97b3a5a4e98471bbf6ce06197c8c4a7460dc54a5f8693253ef47 +size 54058 diff --git a/parse/dev/99RpBVpLiX/images/17563a4d75f10debc921b83564fdddea010a5b531ab9b60af9c90af801d9f639.jpg b/parse/dev/99RpBVpLiX/images/17563a4d75f10debc921b83564fdddea010a5b531ab9b60af9c90af801d9f639.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e636c7289871e8a50ffdbfa59bfc03ab9c638c4a --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/17563a4d75f10debc921b83564fdddea010a5b531ab9b60af9c90af801d9f639.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0ddc706e8f3234661852d8b1f9f25ad7d2c7e21eeb13fe6284f0dbe6d1974da +size 125449 diff --git a/parse/dev/99RpBVpLiX/images/1878e5bbd313919072f44f907d84cdc185dbdaf41690c61851ae500f0f1e4e52.jpg b/parse/dev/99RpBVpLiX/images/1878e5bbd313919072f44f907d84cdc185dbdaf41690c61851ae500f0f1e4e52.jpg new file mode 100644 index 0000000000000000000000000000000000000000..53b8b37b504fbfd3e02ea84989ceba5f2a15050e --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/1878e5bbd313919072f44f907d84cdc185dbdaf41690c61851ae500f0f1e4e52.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:082c3879073892d9179840996c6c03d8c343acf20a65d52b6377f1165470c1c2 +size 18703 diff --git a/parse/dev/99RpBVpLiX/images/1b411506496aab471bdbc5439187a8464438207f597e41c6d730f99e5aa08733.jpg b/parse/dev/99RpBVpLiX/images/1b411506496aab471bdbc5439187a8464438207f597e41c6d730f99e5aa08733.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89a925b7858b24a02e931682b8241a282b4aa157 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/1b411506496aab471bdbc5439187a8464438207f597e41c6d730f99e5aa08733.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fac5769213b5582e8115ac95c5bfb8b3ac0bfb553ede3e6f095e4cf846564113 +size 45873 diff --git a/parse/dev/99RpBVpLiX/images/1b8fd87cee4930c380b11d96a90f2231d1282852e14b1a9868b06d5ce71f4c07.jpg b/parse/dev/99RpBVpLiX/images/1b8fd87cee4930c380b11d96a90f2231d1282852e14b1a9868b06d5ce71f4c07.jpg new file mode 100644 index 0000000000000000000000000000000000000000..07695c10aab5ecbc04577721a118141a443819a9 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/1b8fd87cee4930c380b11d96a90f2231d1282852e14b1a9868b06d5ce71f4c07.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85a9be4698ce606b80ba0ca2d822135c3b4a0533550744d34965e796e8014eb2 +size 84309 diff --git a/parse/dev/99RpBVpLiX/images/1f6aca1d41fd1b8ca4e83f1de5a5e2ede1c2d8f7ac11b59ca95a71c1b57bbae9.jpg b/parse/dev/99RpBVpLiX/images/1f6aca1d41fd1b8ca4e83f1de5a5e2ede1c2d8f7ac11b59ca95a71c1b57bbae9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3f4412d1c60c0750e1de06ca9d49256603b280f8 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/1f6aca1d41fd1b8ca4e83f1de5a5e2ede1c2d8f7ac11b59ca95a71c1b57bbae9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:770f61c369b39590fd70e082e7cfa5d115668d38593935d8a95d3cb360e334dc +size 125199 diff --git a/parse/dev/99RpBVpLiX/images/26e6c4d8daa7f0a09cd4324569fffb0b87dddbf00420d113f1fb987fb3679230.jpg b/parse/dev/99RpBVpLiX/images/26e6c4d8daa7f0a09cd4324569fffb0b87dddbf00420d113f1fb987fb3679230.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e4dbf6794316e0f9d5c9ca26bc751451b5f37a14 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/26e6c4d8daa7f0a09cd4324569fffb0b87dddbf00420d113f1fb987fb3679230.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f5857abea577587017673b6dfdf667ca010782ad871a25e0aafa943bcb4755b +size 15515 diff --git a/parse/dev/99RpBVpLiX/images/2a5df0128fcba1b30e2ad1f57e6d224faef524652a30a074be462f89d1a82fbd.jpg b/parse/dev/99RpBVpLiX/images/2a5df0128fcba1b30e2ad1f57e6d224faef524652a30a074be462f89d1a82fbd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6ab126b22012dae20ad4ad9e33244a3d1d210b38 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/2a5df0128fcba1b30e2ad1f57e6d224faef524652a30a074be462f89d1a82fbd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:948643cb4a77f1dbc439175c3b540a911e5205ee3c7e6b200da5cb804654bc41 +size 13767 diff --git a/parse/dev/99RpBVpLiX/images/318ac6a2dbbc193d0adb6745c476aa73483ff838cb05af06d096faf6955ba5b7.jpg b/parse/dev/99RpBVpLiX/images/318ac6a2dbbc193d0adb6745c476aa73483ff838cb05af06d096faf6955ba5b7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b10bb0443c04b316835736228f93b5171f4a7b57 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/318ac6a2dbbc193d0adb6745c476aa73483ff838cb05af06d096faf6955ba5b7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec5768711ce73393ce94b04ae72d377deeb62a730be4215645b75c15986c5251 +size 19375 diff --git a/parse/dev/99RpBVpLiX/images/350b608278c64eb6e0c935958b69d2d447dcdae48300495f45793fc782f353be.jpg b/parse/dev/99RpBVpLiX/images/350b608278c64eb6e0c935958b69d2d447dcdae48300495f45793fc782f353be.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c66de00efdf22e6e6622f406ce003488dc12987f --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/350b608278c64eb6e0c935958b69d2d447dcdae48300495f45793fc782f353be.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7c31fbe58613fff568a69b9308554237d32aab0b3d034c4f8745040937b8400 +size 34347 diff --git a/parse/dev/99RpBVpLiX/images/3b016dedd4c9b726c17c6dc4e1d8a20e99f04c3c70dd489edabe2a3917c1c879.jpg b/parse/dev/99RpBVpLiX/images/3b016dedd4c9b726c17c6dc4e1d8a20e99f04c3c70dd489edabe2a3917c1c879.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6ef31a9a7b20ff2f1403006ff55dcbd75f673407 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/3b016dedd4c9b726c17c6dc4e1d8a20e99f04c3c70dd489edabe2a3917c1c879.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:817425d659761ce1a5d531b5109e95002b9fb8ddf802cec7e685aa22786bd5e5 +size 41320 diff --git a/parse/dev/99RpBVpLiX/images/3bb3c615fd54712ac5daabe45274cbbdbdb36feae115f5013c0f31a9e30ba786.jpg b/parse/dev/99RpBVpLiX/images/3bb3c615fd54712ac5daabe45274cbbdbdb36feae115f5013c0f31a9e30ba786.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b9a16a67f345884207054312fa1dbc1ec601a64b --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/3bb3c615fd54712ac5daabe45274cbbdbdb36feae115f5013c0f31a9e30ba786.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52a7e7194ffd693c853d41714ef17e06bf3c54d43418cb98bc7d1e6c86f28939 +size 41555 diff --git a/parse/dev/99RpBVpLiX/images/3cac46e4ee8904c17dcf23f84f659d94d0b5de274d545f1145ce4d534cdb2870.jpg b/parse/dev/99RpBVpLiX/images/3cac46e4ee8904c17dcf23f84f659d94d0b5de274d545f1145ce4d534cdb2870.jpg new file mode 100644 index 0000000000000000000000000000000000000000..efdd6cba6893dc5b11ac0af2696f20edfba3c9e9 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/3cac46e4ee8904c17dcf23f84f659d94d0b5de274d545f1145ce4d534cdb2870.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30c93c6c9b94fe43268459e38ed0614b241d79fb2f1711d0da3d697a3c1cb9ab +size 125372 diff --git a/parse/dev/99RpBVpLiX/images/3dfb9099250d4c0b21ce775c2abf155d6b63bf7527c6e2d53805ebc94ad9aec4.jpg b/parse/dev/99RpBVpLiX/images/3dfb9099250d4c0b21ce775c2abf155d6b63bf7527c6e2d53805ebc94ad9aec4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d344eae6a3c3fc75eb91260b2415573086ff2035 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/3dfb9099250d4c0b21ce775c2abf155d6b63bf7527c6e2d53805ebc94ad9aec4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b13799f55fa62471fa6473db2f0c5abca383d27c932a80753e76a6b45483323 +size 66863 diff --git a/parse/dev/99RpBVpLiX/images/3e0dff1d2b4b2a992d655fe00cde9643e08722f6b93e2ed00d0961c0bda1e5da.jpg b/parse/dev/99RpBVpLiX/images/3e0dff1d2b4b2a992d655fe00cde9643e08722f6b93e2ed00d0961c0bda1e5da.jpg new file mode 100644 index 0000000000000000000000000000000000000000..38ce2fe431320122bd336750ecd3c840edc68027 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/3e0dff1d2b4b2a992d655fe00cde9643e08722f6b93e2ed00d0961c0bda1e5da.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c25e6a0d085ca536a130e2bbbc765b4c79b9f86dae5574fa5a557eec218ed76c +size 8592 diff --git a/parse/dev/99RpBVpLiX/images/3ed57abeadf135801547608ae79994e9361c70a0c1e4f57aa435a54ab9aea319.jpg b/parse/dev/99RpBVpLiX/images/3ed57abeadf135801547608ae79994e9361c70a0c1e4f57aa435a54ab9aea319.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f485c5ed04579454762e1918861ebed8178f82e --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/3ed57abeadf135801547608ae79994e9361c70a0c1e4f57aa435a54ab9aea319.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:debb2e0c7146149c882ddfb13be1d97f4e2c4325c5ccfb281ce3067041361a02 +size 25516 diff --git a/parse/dev/99RpBVpLiX/images/403a90eff393c8b225bf8dcc60868b970ef2231fc36ba30a32add19d6d3248d3.jpg b/parse/dev/99RpBVpLiX/images/403a90eff393c8b225bf8dcc60868b970ef2231fc36ba30a32add19d6d3248d3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c25bb3823c66a9ef0bb6201a4908622f38887a03 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/403a90eff393c8b225bf8dcc60868b970ef2231fc36ba30a32add19d6d3248d3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a974bc954a29ef9d88f4ae4b269dd9a47648a60dd2122d3682d26099632969f4 +size 33578 diff --git a/parse/dev/99RpBVpLiX/images/47165b983f5b129fc9225ef80b60e620e7ce04a72710e60a44c4933ce37f2da4.jpg b/parse/dev/99RpBVpLiX/images/47165b983f5b129fc9225ef80b60e620e7ce04a72710e60a44c4933ce37f2da4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9dc29edd81c9d682c98f5bfb8485567720d0a6d0 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/47165b983f5b129fc9225ef80b60e620e7ce04a72710e60a44c4933ce37f2da4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d11fa226cc154aa877874dbe965965404e68a54a6e06f5a975610e08d6b91851 +size 47651 diff --git a/parse/dev/99RpBVpLiX/images/47b1b09bdce03bb35fc427ee6125c3d427310176cfb02aff38d6ba1870ed9ec9.jpg b/parse/dev/99RpBVpLiX/images/47b1b09bdce03bb35fc427ee6125c3d427310176cfb02aff38d6ba1870ed9ec9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8444607baa371d82f498f9516de50c77bfeab64c --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/47b1b09bdce03bb35fc427ee6125c3d427310176cfb02aff38d6ba1870ed9ec9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d280aa703682effabaf2a1d6c54801a3293c7c7bd83c03c8ef13d9b0383f145 +size 30369 diff --git a/parse/dev/99RpBVpLiX/images/4bf6f42daacad2045c717910842f6cf10a52741a02e9098ac6b6f88fee04cd71.jpg b/parse/dev/99RpBVpLiX/images/4bf6f42daacad2045c717910842f6cf10a52741a02e9098ac6b6f88fee04cd71.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7ad85a8424c418d129230d9ed83272db1f28b097 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/4bf6f42daacad2045c717910842f6cf10a52741a02e9098ac6b6f88fee04cd71.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d78e355a7e53645ac75aaf17226269792177262b8ad399ad2220a0c93fad983 +size 70371 diff --git a/parse/dev/99RpBVpLiX/images/4e0578bce689429f2ad634fafc6cba20f47c74d91a95081d4b4d743cdb044280.jpg b/parse/dev/99RpBVpLiX/images/4e0578bce689429f2ad634fafc6cba20f47c74d91a95081d4b4d743cdb044280.jpg new file mode 100644 index 0000000000000000000000000000000000000000..904c2e373d6c1a0e54708672f5f25cc4b6faf2c2 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/4e0578bce689429f2ad634fafc6cba20f47c74d91a95081d4b4d743cdb044280.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ec4b83e4cf30a53b180174207c6eb1999ff9b6ada2d96c1847b0d66dd05a074 +size 61371 diff --git a/parse/dev/99RpBVpLiX/images/511b2ac6b000ba4276b0b9a9ce1c193154bba88e4db555e722d2fcd288705ff2.jpg b/parse/dev/99RpBVpLiX/images/511b2ac6b000ba4276b0b9a9ce1c193154bba88e4db555e722d2fcd288705ff2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..193287198b0016c3ba2015f5f6bf6157bba855a9 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/511b2ac6b000ba4276b0b9a9ce1c193154bba88e4db555e722d2fcd288705ff2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d10b5d819b91ae48ea931d9cf281a73fd4505a5cce8f3fc4df2178b12a493802 +size 5631 diff --git a/parse/dev/99RpBVpLiX/images/54930b7cc7f4d548a9f444368d1c39e587800c212fc68474d1d1f6f5704caad4.jpg b/parse/dev/99RpBVpLiX/images/54930b7cc7f4d548a9f444368d1c39e587800c212fc68474d1d1f6f5704caad4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4f58448baafa5f78c677f4ce2ad8ecc9a02dee1d --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/54930b7cc7f4d548a9f444368d1c39e587800c212fc68474d1d1f6f5704caad4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:840b68d08e4b69f73dbe5df93a5ff2e38c1e0ccdcf740de53ddb6e3b0050ee91 +size 31663 diff --git a/parse/dev/99RpBVpLiX/images/588955a2fe6a273262af9e2b5cc2d8268e4181be399ba1ffec5a199ee0c8c9da.jpg b/parse/dev/99RpBVpLiX/images/588955a2fe6a273262af9e2b5cc2d8268e4181be399ba1ffec5a199ee0c8c9da.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d376823b341d56dbcd181f08cbadb3cde431ef15 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/588955a2fe6a273262af9e2b5cc2d8268e4181be399ba1ffec5a199ee0c8c9da.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:256f3875bc36706b5cb05e7fd78ba4d0a96a66585bb32b714f2e129f0894194c +size 20498 diff --git a/parse/dev/99RpBVpLiX/images/59b68d27db606f3f6c3762aa2addb7bcd9709fa5f7b2f38ae509978afb4ea60a.jpg b/parse/dev/99RpBVpLiX/images/59b68d27db606f3f6c3762aa2addb7bcd9709fa5f7b2f38ae509978afb4ea60a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1fb2ea6a1789c7d0e0f42823f956d6597fbfed38 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/59b68d27db606f3f6c3762aa2addb7bcd9709fa5f7b2f38ae509978afb4ea60a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d48278223c406bdcbfc3c939792a82c1845dc7637081220d8c827cbb248a827 +size 204427 diff --git a/parse/dev/99RpBVpLiX/images/5e3fe77c335b7d9474c1c6063ad1ad8f61c591409d55dd2aaaef019afedaa5b5.jpg b/parse/dev/99RpBVpLiX/images/5e3fe77c335b7d9474c1c6063ad1ad8f61c591409d55dd2aaaef019afedaa5b5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..07085917b19116e51fe7c805bf7e25c6463a4ac8 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/5e3fe77c335b7d9474c1c6063ad1ad8f61c591409d55dd2aaaef019afedaa5b5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42888f63ab58db3cff10056a253d9d7d508475f31cb86e4744ff670a8007ef3d +size 16969 diff --git a/parse/dev/99RpBVpLiX/images/608d9de46dabe3b2af4a8c0e41c1bd6550dfc2a682ba1c164a17563fba817c53.jpg b/parse/dev/99RpBVpLiX/images/608d9de46dabe3b2af4a8c0e41c1bd6550dfc2a682ba1c164a17563fba817c53.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cc284650f72d987c2afc4eba871a53a2531bce10 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/608d9de46dabe3b2af4a8c0e41c1bd6550dfc2a682ba1c164a17563fba817c53.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:040d38e81d1ca64832f6161dc0ac43e617971ccb5ccf492e4c3737243e82c678 +size 55315 diff --git a/parse/dev/99RpBVpLiX/images/6c634472cc5fc7b387cab09e3c5bb32b60358ce31e272d5e98d5b121f2a6a96e.jpg b/parse/dev/99RpBVpLiX/images/6c634472cc5fc7b387cab09e3c5bb32b60358ce31e272d5e98d5b121f2a6a96e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f96131d3015f6a3aa6851c4f12c5ee1505e137dd --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/6c634472cc5fc7b387cab09e3c5bb32b60358ce31e272d5e98d5b121f2a6a96e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b463f8b68131ba82ad5ff921463ca743a7b9a13d70522335a84fc80e08a3c27d +size 10652 diff --git a/parse/dev/99RpBVpLiX/images/6e256c18b99981f696dd1a282f929470264abb8b593e8c7478110648dbb29fe8.jpg b/parse/dev/99RpBVpLiX/images/6e256c18b99981f696dd1a282f929470264abb8b593e8c7478110648dbb29fe8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2011c81e8d152c55077487fa1998eaeb122690a5 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/6e256c18b99981f696dd1a282f929470264abb8b593e8c7478110648dbb29fe8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a455356d5caf4003d6d2604a30025bbdb61f8177bea63a344391959fa9ed7d8d +size 14317 diff --git a/parse/dev/99RpBVpLiX/images/747e26130d74268e0824d8c98d687cb97a5ef1cb2967af2fb9584b80db519432.jpg b/parse/dev/99RpBVpLiX/images/747e26130d74268e0824d8c98d687cb97a5ef1cb2967af2fb9584b80db519432.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1b3d8137b0c9ef03ef5770c2c279a131a5aacb28 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/747e26130d74268e0824d8c98d687cb97a5ef1cb2967af2fb9584b80db519432.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:616a588215cc44804e5a903ed8b17cfba2478df9314c7ab89fb36a20fd8a93f3 +size 7032 diff --git a/parse/dev/99RpBVpLiX/images/7b3cb796e7b6ddbdb8f70bad0e216bfe6a7690598126edb1dfc2c2eb7bdf0a54.jpg b/parse/dev/99RpBVpLiX/images/7b3cb796e7b6ddbdb8f70bad0e216bfe6a7690598126edb1dfc2c2eb7bdf0a54.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c6cbe4ea03776e443096c75fec82c57eb60094ac --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/7b3cb796e7b6ddbdb8f70bad0e216bfe6a7690598126edb1dfc2c2eb7bdf0a54.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1ee225ecb252ee175c210f8709da7078ef3fa73c6ae8811b0910b1718236178 +size 11756 diff --git a/parse/dev/99RpBVpLiX/images/861d7f94dfed99b1e9924a05a46badddac6421fa84099bed9e5cc3a90ff59f12.jpg b/parse/dev/99RpBVpLiX/images/861d7f94dfed99b1e9924a05a46badddac6421fa84099bed9e5cc3a90ff59f12.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f6de92e3aeedae38a69a568fec2c632c7c1b3814 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/861d7f94dfed99b1e9924a05a46badddac6421fa84099bed9e5cc3a90ff59f12.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b04d10792b579619442b25dd48f6fc6f27c1905f025da9deff5f5a60d478e1c +size 2286 diff --git a/parse/dev/99RpBVpLiX/images/862af189be8aaafc63b7e01f055cb14a15477964ebf1c509f5eb0002f5430abe.jpg b/parse/dev/99RpBVpLiX/images/862af189be8aaafc63b7e01f055cb14a15477964ebf1c509f5eb0002f5430abe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..69a4fb5679b95160b94ba14cfff03fc741842ecd --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/862af189be8aaafc63b7e01f055cb14a15477964ebf1c509f5eb0002f5430abe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d24535b44ed40154bc4140c0b5012bf0f6df0c7ae8cad46a25b5d86077d22003 +size 14702 diff --git a/parse/dev/99RpBVpLiX/images/8f244a9cd33bcdebdbce7d2160b2e84e21494245a23ee7235bdea9ea386c2931.jpg b/parse/dev/99RpBVpLiX/images/8f244a9cd33bcdebdbce7d2160b2e84e21494245a23ee7235bdea9ea386c2931.jpg new file mode 100644 index 0000000000000000000000000000000000000000..445515f6204da32528d8466ab2bfb6dfe66c38c0 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/8f244a9cd33bcdebdbce7d2160b2e84e21494245a23ee7235bdea9ea386c2931.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99a85c7919ff134070637bbc9dd5bbca004401bf2ae3c32e60b0041b8ef642ba +size 111309 diff --git a/parse/dev/99RpBVpLiX/images/9122a730c6bce2f52fda1cda1ca3702c70e0ec06a578675f90e17f79f9e98fa6.jpg b/parse/dev/99RpBVpLiX/images/9122a730c6bce2f52fda1cda1ca3702c70e0ec06a578675f90e17f79f9e98fa6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5665bc0e16b24ad0a3283f3b1e9717520e0839cb --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/9122a730c6bce2f52fda1cda1ca3702c70e0ec06a578675f90e17f79f9e98fa6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa94b66337bc7cae8fa0d63a2adb65dac37d4b70b25d6cb8b7b8a953c15c69ca +size 15811 diff --git a/parse/dev/99RpBVpLiX/images/9883cc48af9c82da9e62599b2bd1408309ebf51eacb6b2dc6aaf5933b13ddf52.jpg b/parse/dev/99RpBVpLiX/images/9883cc48af9c82da9e62599b2bd1408309ebf51eacb6b2dc6aaf5933b13ddf52.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9bb9ab0fd7a57dc79762fd5237d9cd3f6cdae4f0 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/9883cc48af9c82da9e62599b2bd1408309ebf51eacb6b2dc6aaf5933b13ddf52.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9f674c0d714b05177d89d0ab07eefdcc542f21969318064b6e69933b991afb4 +size 139401 diff --git a/parse/dev/99RpBVpLiX/images/9a1160cf033b4219214d79be18a4f6fc1e623b34dacc9decacec9461308391d0.jpg b/parse/dev/99RpBVpLiX/images/9a1160cf033b4219214d79be18a4f6fc1e623b34dacc9decacec9461308391d0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5fdd83a3d70923c2568de755888b380a8ac64851 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/9a1160cf033b4219214d79be18a4f6fc1e623b34dacc9decacec9461308391d0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:139c748acca4195e240a684d8271c7ff5d7aa7c57bf456b84b80aa5c4754b71d +size 9601 diff --git a/parse/dev/99RpBVpLiX/images/a1aca3a24867c1fb4e63e5611ceed6175d2b325e874f9caa9012ac1a7b7728c3.jpg b/parse/dev/99RpBVpLiX/images/a1aca3a24867c1fb4e63e5611ceed6175d2b325e874f9caa9012ac1a7b7728c3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ff958df9e2e386e793b8c077211321f22baba23a --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/a1aca3a24867c1fb4e63e5611ceed6175d2b325e874f9caa9012ac1a7b7728c3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:909af44773d7f897fbb804af6771e4cf2fd34b95f37ad5c4270e942208b15e5d +size 57769 diff --git a/parse/dev/99RpBVpLiX/images/a58335e9603011e4fef2d3d63a6e24784eda7c48b426bb697f3d016e2a1374fb.jpg b/parse/dev/99RpBVpLiX/images/a58335e9603011e4fef2d3d63a6e24784eda7c48b426bb697f3d016e2a1374fb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bb74cee72ec9bd3fd167e7358c756e02adf51f29 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/a58335e9603011e4fef2d3d63a6e24784eda7c48b426bb697f3d016e2a1374fb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de3128c68542ab92fefa0b45d62aec27d5615b754241a4e2de34c27ac0d06e29 +size 48550 diff --git a/parse/dev/99RpBVpLiX/images/a7c749ad0f07bd1009f74b405d55648a8842d2885b70d705a1e681eeb9d5f60f.jpg b/parse/dev/99RpBVpLiX/images/a7c749ad0f07bd1009f74b405d55648a8842d2885b70d705a1e681eeb9d5f60f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c5fa29918a36e9ba50a642defcd37acd8df4f205 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/a7c749ad0f07bd1009f74b405d55648a8842d2885b70d705a1e681eeb9d5f60f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f11ab99f65d979770175f412d45ff888b6ad5f5b28b62eb4bb590ce5f61fdaba +size 24661 diff --git a/parse/dev/99RpBVpLiX/images/ab27d8a37388f408723238404a49bd589c165d01e3141059c830f958105e57f7.jpg b/parse/dev/99RpBVpLiX/images/ab27d8a37388f408723238404a49bd589c165d01e3141059c830f958105e57f7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1353528e6e754d42d6c33fcf638783e8ed706822 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/ab27d8a37388f408723238404a49bd589c165d01e3141059c830f958105e57f7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f325ced360bfbbf9475be6e500a4a975fa65a85b93563651c601f02cdafe4d4 +size 4500 diff --git a/parse/dev/99RpBVpLiX/images/aed1ae2ef5823b4f0901d2eda3d6706032a2b3d2d6c8436c8f44e79a7513f897.jpg b/parse/dev/99RpBVpLiX/images/aed1ae2ef5823b4f0901d2eda3d6706032a2b3d2d6c8436c8f44e79a7513f897.jpg new file mode 100644 index 0000000000000000000000000000000000000000..18bdbff1577235581897641d53169411231e7af8 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/aed1ae2ef5823b4f0901d2eda3d6706032a2b3d2d6c8436c8f44e79a7513f897.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe58310215a9b8a1807cad79d68f8f0f848d2a909f563fa7563e86b4761d3244 +size 11053 diff --git a/parse/dev/99RpBVpLiX/images/b634cf9f834a82ee8196e8517f095dac26e968b764032f27debefc43b4a8c0f2.jpg b/parse/dev/99RpBVpLiX/images/b634cf9f834a82ee8196e8517f095dac26e968b764032f27debefc43b4a8c0f2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..adbd1440e4a5ea9a2fb127f1c3e87ea9d8b07d42 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/b634cf9f834a82ee8196e8517f095dac26e968b764032f27debefc43b4a8c0f2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4db97fbe6d03b6751f2b12a1205ef7e8f368a5357fd534f23ed8a8ce34863286 +size 6945 diff --git a/parse/dev/99RpBVpLiX/images/b72f97d41b64917604ba6ad071a1d53f3108dbbad8342e74af2fbf20ddf369cd.jpg b/parse/dev/99RpBVpLiX/images/b72f97d41b64917604ba6ad071a1d53f3108dbbad8342e74af2fbf20ddf369cd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fc1fffed96a82badaffe396a3e7996152c3d7bcc --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/b72f97d41b64917604ba6ad071a1d53f3108dbbad8342e74af2fbf20ddf369cd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d5520ab9b1f61d55bbc80567c6cdaa360570b940e845cf352d456958f5586598 +size 65731 diff --git a/parse/dev/99RpBVpLiX/images/bbbb26b31e6f11aac454be69eabacf494707fb4cce072061ada4adc9c7f53ec4.jpg b/parse/dev/99RpBVpLiX/images/bbbb26b31e6f11aac454be69eabacf494707fb4cce072061ada4adc9c7f53ec4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eabc2bcfed8fbf183d5f6c341673c6b961f66ebd --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/bbbb26b31e6f11aac454be69eabacf494707fb4cce072061ada4adc9c7f53ec4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e647d23e56341c49f4abda1fd2381bae302cf317cb264835b6a325fb4a16963 +size 19243 diff --git a/parse/dev/99RpBVpLiX/images/c2c592f6b5b1298f4cf26d3c98ab74dbf5dcda349362d5f6b237f470f6959555.jpg b/parse/dev/99RpBVpLiX/images/c2c592f6b5b1298f4cf26d3c98ab74dbf5dcda349362d5f6b237f470f6959555.jpg new file mode 100644 index 0000000000000000000000000000000000000000..703a407c3769284c8a5a30a678a2e31404134899 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/c2c592f6b5b1298f4cf26d3c98ab74dbf5dcda349362d5f6b237f470f6959555.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a76123810cea6c016f248312a0107135d229c89c2b1d45437715a8459707667 +size 15210 diff --git a/parse/dev/99RpBVpLiX/images/c33229ad4541f9909b2a8bf34a43bf586390a9d831267e40ca98a2aa20916c9c.jpg b/parse/dev/99RpBVpLiX/images/c33229ad4541f9909b2a8bf34a43bf586390a9d831267e40ca98a2aa20916c9c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..68ee502ab557ea646213bc379bcd0708abb8166e --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/c33229ad4541f9909b2a8bf34a43bf586390a9d831267e40ca98a2aa20916c9c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6284f53bebe020ec9694033cbdcd638e446b6747f2cfd76a5a8c19126a16d1ff +size 45618 diff --git a/parse/dev/99RpBVpLiX/images/c40d007b7223cb13f03c3a39c1a84167b49ec116489ad7a1d57388bade3ffc05.jpg b/parse/dev/99RpBVpLiX/images/c40d007b7223cb13f03c3a39c1a84167b49ec116489ad7a1d57388bade3ffc05.jpg new file mode 100644 index 0000000000000000000000000000000000000000..39f69e922f6d84facb741becebc4273b9cdbfbdb --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/c40d007b7223cb13f03c3a39c1a84167b49ec116489ad7a1d57388bade3ffc05.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01dce98a884ba768d8b5ba2897ebe099b7da87701d9f3301469a6e76837c9792 +size 14739 diff --git a/parse/dev/99RpBVpLiX/images/c972c28f8f0145dbb50c2223064757f1c6c386d7165bc1897c8ee2baed2747f3.jpg b/parse/dev/99RpBVpLiX/images/c972c28f8f0145dbb50c2223064757f1c6c386d7165bc1897c8ee2baed2747f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8d15e31e32931cc548052d7048203459d4177105 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/c972c28f8f0145dbb50c2223064757f1c6c386d7165bc1897c8ee2baed2747f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6ee5c3a66cf8d7c5f8b46e24271ab7350790017d331cd2058652a7752ebd908 +size 12490 diff --git a/parse/dev/99RpBVpLiX/images/cc180606452926980fb55217dd94ad5d6f200c2f1228efff2b837642f54b8a12.jpg b/parse/dev/99RpBVpLiX/images/cc180606452926980fb55217dd94ad5d6f200c2f1228efff2b837642f54b8a12.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f7a0b73d3397c569daae88e5c2b371b2f669f802 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/cc180606452926980fb55217dd94ad5d6f200c2f1228efff2b837642f54b8a12.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c06baf59eaae964c9e43e9f9e6c9e747ce4308c294c44468476a76b726f6e67e +size 10479 diff --git a/parse/dev/99RpBVpLiX/images/d012f125866d2c62770124700bba28d127971f0cfb60f82c6d657aea6e4d6dec.jpg b/parse/dev/99RpBVpLiX/images/d012f125866d2c62770124700bba28d127971f0cfb60f82c6d657aea6e4d6dec.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ea391806e6bd038ddf9169ea71958bd7117e6ea5 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/d012f125866d2c62770124700bba28d127971f0cfb60f82c6d657aea6e4d6dec.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d9aad30eb88bf2b0652a9bce09014c7f01a99c175255df1c566101165f59d55 +size 175662 diff --git a/parse/dev/99RpBVpLiX/images/d31ee5996b5e5abe92c8bb4b9551dd6b045100d37b6dfd9ad7e9fa6fc0cf5a90.jpg b/parse/dev/99RpBVpLiX/images/d31ee5996b5e5abe92c8bb4b9551dd6b045100d37b6dfd9ad7e9fa6fc0cf5a90.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a25758c8ce50e8428147f7022353a0c83506e3c6 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/d31ee5996b5e5abe92c8bb4b9551dd6b045100d37b6dfd9ad7e9fa6fc0cf5a90.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e75f28fbc4116069d7e340b06a09eafa9380e011e82661a1355c03f2fcda9a8 +size 15259 diff --git a/parse/dev/99RpBVpLiX/images/d3352273e007de6889202d50908a835f6207a6d7e8a7bd424847222ebd9ae37a.jpg b/parse/dev/99RpBVpLiX/images/d3352273e007de6889202d50908a835f6207a6d7e8a7bd424847222ebd9ae37a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..af7b83e88162dfaa6a31c51b02c4ded4a222fa08 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/d3352273e007de6889202d50908a835f6207a6d7e8a7bd424847222ebd9ae37a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94e765ce4bf1a6fc63a632fe0f8a03f7fc522f5ec039c64807f3b41c77def3f1 +size 12878 diff --git a/parse/dev/99RpBVpLiX/images/dfdb9f671d8717985c4d8efe9205d05480baa3a96afd00a4317e89df5bf120e4.jpg b/parse/dev/99RpBVpLiX/images/dfdb9f671d8717985c4d8efe9205d05480baa3a96afd00a4317e89df5bf120e4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6f61674ff18e1bafc92acaa49d82d562e5a1c25f --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/dfdb9f671d8717985c4d8efe9205d05480baa3a96afd00a4317e89df5bf120e4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53ac5ba7fef04dbf326ab5f43819fc7568b580f41ec5e158038eaa2f5256c14d +size 24448 diff --git a/parse/dev/99RpBVpLiX/images/e529e103f2f38663d7448a35d681ce8627d3e1f7225f41b98cb4c67b82a512a1.jpg b/parse/dev/99RpBVpLiX/images/e529e103f2f38663d7448a35d681ce8627d3e1f7225f41b98cb4c67b82a512a1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6c49d7d81b54d68a6aeb02a230576574aebb02df --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/e529e103f2f38663d7448a35d681ce8627d3e1f7225f41b98cb4c67b82a512a1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3011e17dd4e026d1b50fdb7ba2a4e614cc3b357d4923771e5ea4d5fe6c02ede2 +size 41771 diff --git a/parse/dev/99RpBVpLiX/images/e5b45a4c54a685a22a261b74dee482c78fc174f1f2f3ec57cbad0a7fd54e99e4.jpg b/parse/dev/99RpBVpLiX/images/e5b45a4c54a685a22a261b74dee482c78fc174f1f2f3ec57cbad0a7fd54e99e4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..41206b4ea0f6031d211fb0c291acc5f275e814c7 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/e5b45a4c54a685a22a261b74dee482c78fc174f1f2f3ec57cbad0a7fd54e99e4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f380e68d7a56a118904cc037a4cfb05da2abcffd29cb089828566362120901a6 +size 14925 diff --git a/parse/dev/99RpBVpLiX/images/ea887c444405296cea4a2f931a042c9ecd1bc02f478d5b4d724653325dd78966.jpg b/parse/dev/99RpBVpLiX/images/ea887c444405296cea4a2f931a042c9ecd1bc02f478d5b4d724653325dd78966.jpg new file mode 100644 index 0000000000000000000000000000000000000000..58382fd7e9cc800b802e09e625d7ae3d0b28ea27 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/ea887c444405296cea4a2f931a042c9ecd1bc02f478d5b4d724653325dd78966.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c176d1962322391d9489e025dea2d34397105c23bfe55472228396461d934544 +size 15124 diff --git a/parse/dev/99RpBVpLiX/images/eb692e180a2bbe80aa4aa786cd5887600accae652fa2496ab3dd73bae8f079bd.jpg b/parse/dev/99RpBVpLiX/images/eb692e180a2bbe80aa4aa786cd5887600accae652fa2496ab3dd73bae8f079bd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e7fa0f57270baee5d7f80f9bf30ce18b264fae8 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/eb692e180a2bbe80aa4aa786cd5887600accae652fa2496ab3dd73bae8f079bd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21c74ace336feae518fc48d419e93d839190f5414c79daad7ab5c95b238d636c +size 50705 diff --git a/parse/dev/99RpBVpLiX/images/f4a573f7aeefb6d95cd0ae19fdbe65c32c00c34e195027ae0053d50bd94dc789.jpg b/parse/dev/99RpBVpLiX/images/f4a573f7aeefb6d95cd0ae19fdbe65c32c00c34e195027ae0053d50bd94dc789.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d5ef7fd2fa42eaa4f62af87f1df72716ff78c9fc --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/f4a573f7aeefb6d95cd0ae19fdbe65c32c00c34e195027ae0053d50bd94dc789.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d8ee41b1b6b6b65424c7ba1107f26621fdc0053d17da7b422c427871f67a837 +size 19206 diff --git a/parse/dev/99RpBVpLiX/images/fb10f1f203088e8de434ec919feaabeeb82f049ed39c1aa612affa89ce19c0a0.jpg b/parse/dev/99RpBVpLiX/images/fb10f1f203088e8de434ec919feaabeeb82f049ed39c1aa612affa89ce19c0a0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..08425a6c69aeb8bafe08966550422ce8f2efd594 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/fb10f1f203088e8de434ec919feaabeeb82f049ed39c1aa612affa89ce19c0a0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f196b7b755b45b4ce37ee97baf5f76c4a1f89eb2cc81b29baf5080c4a0193c3 +size 34057 diff --git a/parse/dev/99RpBVpLiX/images/fc2132ad4f9f269af61073f0633416192dbb95d7bcf5a4deff4d8e3be1c42246.jpg b/parse/dev/99RpBVpLiX/images/fc2132ad4f9f269af61073f0633416192dbb95d7bcf5a4deff4d8e3be1c42246.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eaead093d1a2e0ce6d5bfbe8b5fea1937619d556 --- /dev/null +++ b/parse/dev/99RpBVpLiX/images/fc2132ad4f9f269af61073f0633416192dbb95d7bcf5a4deff4d8e3be1c42246.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:939539eb4ef7a9cb52825113b1d21f24ac881572fab76d9e0a8304f5fb99a40f +size 45051 diff --git a/parse/dev/BIpTWmO_BY/images/0627ef1cfea2ffc92ba698190f901d395c9988f3911c76caebe044d99f9e2441.jpg b/parse/dev/BIpTWmO_BY/images/0627ef1cfea2ffc92ba698190f901d395c9988f3911c76caebe044d99f9e2441.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89d322b6f162c5bc88fdae61ae468731623f88bb --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/0627ef1cfea2ffc92ba698190f901d395c9988f3911c76caebe044d99f9e2441.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e06ead4b87c49454bc235b373fa73572138ecc971af3be72a493fb5517195572 +size 35206 diff --git a/parse/dev/BIpTWmO_BY/images/32cae721a4baef29787c923b91b8b83c44364eb813d0093387924577fe8d75e0.jpg b/parse/dev/BIpTWmO_BY/images/32cae721a4baef29787c923b91b8b83c44364eb813d0093387924577fe8d75e0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8bc95662bb624bb2e0d14725a87e7498526b3eaa --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/32cae721a4baef29787c923b91b8b83c44364eb813d0093387924577fe8d75e0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6ad825d7a651135c12e4669a389072351df3e922362e0b71d26939716d46fe8 +size 49874 diff --git a/parse/dev/BIpTWmO_BY/images/4384aa14b9980f43388b03d921e3553294b730c3273314c5103c7b8295a62a35.jpg b/parse/dev/BIpTWmO_BY/images/4384aa14b9980f43388b03d921e3553294b730c3273314c5103c7b8295a62a35.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6e5f9037468785ca74fdd437444120d75a70dcd5 --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/4384aa14b9980f43388b03d921e3553294b730c3273314c5103c7b8295a62a35.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:31941fc44f5de4fe688f39bfad059e2ae4340145c4963e7106691a9d60a1cd9a +size 7223 diff --git a/parse/dev/BIpTWmO_BY/images/5455e4382fa10d63e00bda5b2abf80eec6d444e16bb2fb128e99e6fdd869263f.jpg b/parse/dev/BIpTWmO_BY/images/5455e4382fa10d63e00bda5b2abf80eec6d444e16bb2fb128e99e6fdd869263f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..edda9a9173e07f032c2fc3ac6dd8d9087af67cf5 --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/5455e4382fa10d63e00bda5b2abf80eec6d444e16bb2fb128e99e6fdd869263f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2de918c6be4cba08ef6fbe34d3a48daee6949615acf92705e162cd5d71435d75 +size 53391 diff --git a/parse/dev/BIpTWmO_BY/images/54f16d2d4d6a2d5606837650e2f0e9cf71b10b2ac174eafb482d5aca55ce2976.jpg b/parse/dev/BIpTWmO_BY/images/54f16d2d4d6a2d5606837650e2f0e9cf71b10b2ac174eafb482d5aca55ce2976.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dc0aad0f31d65eb805532de5d120e6cc434f9459 --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/54f16d2d4d6a2d5606837650e2f0e9cf71b10b2ac174eafb482d5aca55ce2976.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50ed9e038a173b4ff275bae3636a9556c3dfc18eab11803618776de595121572 +size 6684 diff --git a/parse/dev/BIpTWmO_BY/images/667edf2dc6165b5575283e8887bab05c21f4cbbbb63f84e41edbfbd74efe03a5.jpg b/parse/dev/BIpTWmO_BY/images/667edf2dc6165b5575283e8887bab05c21f4cbbbb63f84e41edbfbd74efe03a5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7b9cb476433c7107b1f8d1c0ffda5d97414be9bd --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/667edf2dc6165b5575283e8887bab05c21f4cbbbb63f84e41edbfbd74efe03a5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:933035f04876d9d5aaae7c2470deecdb0b32141f7c70a654b770de081acee5f7 +size 48370 diff --git a/parse/dev/BIpTWmO_BY/images/6e1beadf8a294fb2612cff44e69fef782695c8aa9bd7eccd0d621999d72076d3.jpg b/parse/dev/BIpTWmO_BY/images/6e1beadf8a294fb2612cff44e69fef782695c8aa9bd7eccd0d621999d72076d3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d049d3bcb6f7ed8247e3f57a8d1db93cebf7a204 --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/6e1beadf8a294fb2612cff44e69fef782695c8aa9bd7eccd0d621999d72076d3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bdc3cc48007c93f290fdaf8e44f6ab60bd40fbf13d6282e0fe14cf973006028a +size 31296 diff --git a/parse/dev/BIpTWmO_BY/images/70312f209f976a747db4540c9abe2d4c162c1d1c40fdd8a411e48bf0f4b08231.jpg b/parse/dev/BIpTWmO_BY/images/70312f209f976a747db4540c9abe2d4c162c1d1c40fdd8a411e48bf0f4b08231.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b2763ebce949e8edd3c2cb40f4cd10d5c4bcbd35 --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/70312f209f976a747db4540c9abe2d4c162c1d1c40fdd8a411e48bf0f4b08231.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33eac794a3ea805ddf514b5cb18388dfb6049727323378e6b20c74d051ba69a9 +size 51627 diff --git a/parse/dev/BIpTWmO_BY/images/ebbc7e9112e038b12920b0ffd639ad94c4abbbfd03535278a8fecfae4350f837.jpg b/parse/dev/BIpTWmO_BY/images/ebbc7e9112e038b12920b0ffd639ad94c4abbbfd03535278a8fecfae4350f837.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e85395933b8383000b8b842e1441a2660d694001 --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/ebbc7e9112e038b12920b0ffd639ad94c4abbbfd03535278a8fecfae4350f837.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a7ebbfadb2449bf2f3c547f4045eb09ed5d808d1211dcab3708fbbc277d29a2 +size 17000 diff --git a/parse/dev/BIpTWmO_BY/images/f6077e6c5d0398d92411af95da71dee8ba10348c7df87eee8e1179ea6b998eed.jpg b/parse/dev/BIpTWmO_BY/images/f6077e6c5d0398d92411af95da71dee8ba10348c7df87eee8e1179ea6b998eed.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aa76e64700403837b0225fc8c6c49bb6809ca098 --- /dev/null +++ b/parse/dev/BIpTWmO_BY/images/f6077e6c5d0398d92411af95da71dee8ba10348c7df87eee8e1179ea6b998eed.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff014db553892d9db41dda1aadb1ee557503cc0d60aa8d74c7cb06d992bd890d +size 5561 diff --git a/parse/dev/BqxE86ufTzq/images/024ed6eb6812529d108ea8c7d7b05b039008e1f6105202c8881ffb66d793baf9.jpg b/parse/dev/BqxE86ufTzq/images/024ed6eb6812529d108ea8c7d7b05b039008e1f6105202c8881ffb66d793baf9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eb2fb5fc316cca26e1b310451732717c3912502a --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/024ed6eb6812529d108ea8c7d7b05b039008e1f6105202c8881ffb66d793baf9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:714d2d195e3f9693d1dd95562d3999cc157b408acd5e767d1e8903e20fe66bdc +size 34433 diff --git a/parse/dev/BqxE86ufTzq/images/04cb3381b3e4479ef9f8eb923ff18a6ff9319ddfa75cab2ee1dd4f5ead6f4446.jpg b/parse/dev/BqxE86ufTzq/images/04cb3381b3e4479ef9f8eb923ff18a6ff9319ddfa75cab2ee1dd4f5ead6f4446.jpg new file mode 100644 index 0000000000000000000000000000000000000000..17f261ddc5f4c3fc552b961cbf9fe63a70632d24 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/04cb3381b3e4479ef9f8eb923ff18a6ff9319ddfa75cab2ee1dd4f5ead6f4446.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8cd127ee6146da59f0922949ccc8a6440ed68ebc11f137a7f2f9e0601abcb89a +size 5558 diff --git a/parse/dev/BqxE86ufTzq/images/65b0748aea10980890c055c5b25ba147392ad48925c02862b210ec7008dfaffb.jpg b/parse/dev/BqxE86ufTzq/images/65b0748aea10980890c055c5b25ba147392ad48925c02862b210ec7008dfaffb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..155223567af17d357fead4702f21ff689c321cfe --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/65b0748aea10980890c055c5b25ba147392ad48925c02862b210ec7008dfaffb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cfbdf195a0b783eefabfa780dcd40acbd67cca1b66d4932584a7d6bcbbf41571 +size 151913 diff --git a/parse/dev/BqxE86ufTzq/images/757a366380ae64b67a39d38e5e372ed8fa30032ce6e99ca5e363f4442873c978.jpg b/parse/dev/BqxE86ufTzq/images/757a366380ae64b67a39d38e5e372ed8fa30032ce6e99ca5e363f4442873c978.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6e5ca187315243949b14366498a6345ea623b489 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/757a366380ae64b67a39d38e5e372ed8fa30032ce6e99ca5e363f4442873c978.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9497e7b17b4f719a6f099308e34335d69216621cedfd64f3cf80d41aeae6d299 +size 22015 diff --git a/parse/dev/BqxE86ufTzq/images/7c32d3bb5231860f6864fbd6f54ad51df8c6ad6c75f69a7b6c620036c8f70480.jpg b/parse/dev/BqxE86ufTzq/images/7c32d3bb5231860f6864fbd6f54ad51df8c6ad6c75f69a7b6c620036c8f70480.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0b5e467b65b2eefa083121382b10c1b85d87b418 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/7c32d3bb5231860f6864fbd6f54ad51df8c6ad6c75f69a7b6c620036c8f70480.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a8de41887610829b749ba3e5f95bf9bab9965a8dd7f920046563b12cf46d8bb +size 77438 diff --git a/parse/dev/BqxE86ufTzq/images/875017c918d22ef8ee2637b30752f06b3e267e44d9b06f8ffcb0b7ce2d2731d9.jpg b/parse/dev/BqxE86ufTzq/images/875017c918d22ef8ee2637b30752f06b3e267e44d9b06f8ffcb0b7ce2d2731d9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ab93a2e7a3e0383745b63012bf1c7d9317b5e714 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/875017c918d22ef8ee2637b30752f06b3e267e44d9b06f8ffcb0b7ce2d2731d9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d30e8aa00b800224adb2fc20fead4e042c59ce3c6858a7654149d5c9545f14a +size 82730 diff --git a/parse/dev/BqxE86ufTzq/images/90b4e7dbc77db37f85fd5ef282242e41507decf888f9bad780d6f321e946b15f.jpg b/parse/dev/BqxE86ufTzq/images/90b4e7dbc77db37f85fd5ef282242e41507decf888f9bad780d6f321e946b15f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5d6c897d44971151afa9312e794079829437c2e9 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/90b4e7dbc77db37f85fd5ef282242e41507decf888f9bad780d6f321e946b15f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c73e4086683a6dd6cc23441436de877cc976b12fe5a68b1406b27823efce805b +size 42732 diff --git a/parse/dev/BqxE86ufTzq/images/96da1649f527531a17f8b257d50be54de178294c3491baf9dee569c24763f0dd.jpg b/parse/dev/BqxE86ufTzq/images/96da1649f527531a17f8b257d50be54de178294c3491baf9dee569c24763f0dd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2ecac49c539fd6f25b17f94e40ed2c79c404a1f2 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/96da1649f527531a17f8b257d50be54de178294c3491baf9dee569c24763f0dd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7258860e0595b6c6bf3d55f609db68634792e9deca786997bc6b140275600893 +size 43696 diff --git a/parse/dev/BqxE86ufTzq/images/9c494efb45d9cec38d880301e105c8105116b19086f192a43782b1e7ac3706a9.jpg b/parse/dev/BqxE86ufTzq/images/9c494efb45d9cec38d880301e105c8105116b19086f192a43782b1e7ac3706a9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e55af6f3d0d033572e14ec70d45c526b15ec333b --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/9c494efb45d9cec38d880301e105c8105116b19086f192a43782b1e7ac3706a9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:840fcfa1180a2b229057f7f80186918f8a91d4f45abb3e02b43d4b15d150fbe8 +size 2282 diff --git a/parse/dev/BqxE86ufTzq/images/ba093351df76a0a952804a417a05a4534f2b0bc293c8e6a6f688eb9ca0853d5f.jpg b/parse/dev/BqxE86ufTzq/images/ba093351df76a0a952804a417a05a4534f2b0bc293c8e6a6f688eb9ca0853d5f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d513340f29e2d0fd7ad436167d66004f5c9594b9 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/ba093351df76a0a952804a417a05a4534f2b0bc293c8e6a6f688eb9ca0853d5f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7732f8f51de0b0f09bfe8334d9cb8b7961bea69f43fd364a89e96e4de98adcbb +size 2071 diff --git a/parse/dev/BqxE86ufTzq/images/bbd2a6c4060de3ec8b62db689716a5c038d724158d530e4407130b4dac912290.jpg b/parse/dev/BqxE86ufTzq/images/bbd2a6c4060de3ec8b62db689716a5c038d724158d530e4407130b4dac912290.jpg new file mode 100644 index 0000000000000000000000000000000000000000..04a975bf9f307d2b606ecb87832fa6b9580b2b50 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/bbd2a6c4060de3ec8b62db689716a5c038d724158d530e4407130b4dac912290.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:babb35e81e7d401f14140e679afa385ab8262c2c561ec9ef7c112587374962cb +size 26904 diff --git a/parse/dev/BqxE86ufTzq/images/c2375990894eb4fad9be86c7d72efb700db506ae000d691d8a1bad0b40c66def.jpg b/parse/dev/BqxE86ufTzq/images/c2375990894eb4fad9be86c7d72efb700db506ae000d691d8a1bad0b40c66def.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a6ceddd9cea3e22c3db675bb017cea7fab342760 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/c2375990894eb4fad9be86c7d72efb700db506ae000d691d8a1bad0b40c66def.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da62c5d609fadbc7ccbb907a79ff2eae5dcdbddda3ae874c521f549e2ab9c607 +size 44914 diff --git a/parse/dev/BqxE86ufTzq/images/c36af697c7bf441bc9834d63d8310675929be67b3f9b5a9fdc570ff99456e65b.jpg b/parse/dev/BqxE86ufTzq/images/c36af697c7bf441bc9834d63d8310675929be67b3f9b5a9fdc570ff99456e65b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bf4d117669e8e7de1f0824a465ca3ebae11e7718 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/c36af697c7bf441bc9834d63d8310675929be67b3f9b5a9fdc570ff99456e65b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c2985f76d96c317b9deff06a5c0b44696f4d106ec964783c2a9983c4ef4a2c3 +size 238777 diff --git a/parse/dev/BqxE86ufTzq/images/c7c2c4e907bdd854f61b249767c9743a3034890d7a35af961755aaa794923c1f.jpg b/parse/dev/BqxE86ufTzq/images/c7c2c4e907bdd854f61b249767c9743a3034890d7a35af961755aaa794923c1f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b826beed212aacacb9f0a394761800a4423869d4 --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/c7c2c4e907bdd854f61b249767c9743a3034890d7a35af961755aaa794923c1f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8961fa5bf2d23af86bc6b70af62c5bac7a615dc8f98f6dee16d1f14f8bd9c276 +size 11345 diff --git a/parse/dev/BqxE86ufTzq/images/d31e71048c6a3dab26c8bd44ac7979721cc47c1ea40a09d3f42c033511fd15e9.jpg b/parse/dev/BqxE86ufTzq/images/d31e71048c6a3dab26c8bd44ac7979721cc47c1ea40a09d3f42c033511fd15e9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..622531713e5d7fb1076ab6663b80c29763334d4d --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/d31e71048c6a3dab26c8bd44ac7979721cc47c1ea40a09d3f42c033511fd15e9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ed105d916eabda8eb845edf127132c99b66a217c535baabcf2c943132e7a9fd +size 4141 diff --git a/parse/dev/BqxE86ufTzq/images/d68c3970eba997b59ff1eae04695820183e10fc5912775364ec372a1a89129d8.jpg b/parse/dev/BqxE86ufTzq/images/d68c3970eba997b59ff1eae04695820183e10fc5912775364ec372a1a89129d8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..97382a0ad34a81cff26207491990ab0056e3221d --- /dev/null +++ b/parse/dev/BqxE86ufTzq/images/d68c3970eba997b59ff1eae04695820183e10fc5912775364ec372a1a89129d8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9666cf7067a2728fae46eec08c3a6a76f7d634e3cc7bb0aae4013fccf3cb5565 +size 291850 diff --git a/parse/dev/F72ximsx7C1/F72ximsx7C1.md b/parse/dev/F72ximsx7C1/F72ximsx7C1.md new file mode 100644 index 0000000000000000000000000000000000000000..47833baeb6995293ee4cd2d6b83250bfe401d202 --- /dev/null +++ b/parse/dev/F72ximsx7C1/F72ximsx7C1.md @@ -0,0 +1,833 @@ +# HOW ATTENTIVE ARE GRAPH ATTENTION NETWORKS? + +Shaked Brody +Technion +shakedbr@cs.technion.ac.il + +Uri Alon Language Technologies Institute Carnegie Mellon University ualon@cs.cmu.edu + +# Eran Yahav + +yahave@cs.technion.ac.il + +# ABSTRACT + +Graph Attention Networks (GATs) are one of the most popular GNN architectures and are considered as the state-of-the-art architecture for representation learning with graphs. In GAT, every node attends to its neighbors given its own representation as the query. However, in this paper we show that GAT computes a very limited kind of attention: the ranking of the attention scores is unconditioned on the query node. We formally define this restricted kind of attention as static attention and distinguish it from a strictly more expressive dynamic attention. Because GATs use a static attention mechanism, there are simple graph problems that GAT cannot express: in a controlled problem, we show that static attention hinders GAT from even fitting the training data. To remove this limitation, we introduce a simple fix by modifying the order of operations and propose GATv2: a dynamic graph attention variant that is strictly more expressive than GAT. We perform an extensive evaluation and show that GATv2 outperforms GAT across 12 OGB and other benchmarks while we match their parametric costs. Our code is available at https://github.com/tech-srl/how_attentive_are_ gats.1 GATv2 is available as part of the PyTorch Geometric library,2 the Deep Graph Library,3 and the TensorFlow GNN library.4 + +# 1 INTRODUCTION + +Graph neural networks (GNNs; Gori et al., 2005; Scarselli et al., 2008) have seen increasing popularity over the past few years (Duvenaud et al., 2015; Atwood and Towsley, 2016; Bronstein et al., 2017; Monti et al., 2017). GNNs provide a general and efficient framework to learn from graph-structured data. Thus, GNNs are easily applicable in domains where the data can be represented as a set of nodes and the prediction depends on the relationships (edges) between the nodes. Such domains include molecules, social networks, product recommendation, computer programs and more. + +In a GNN, each node iteratively updates its state by interacting with its neighbors. GNN variants (Wu et al., 2019; Xu et al., 2019; Li et al., 2016) mostly differ in how each node aggregates and combines the representations of its neighbors with its own. Velickovi ˇ c et al. (2018) pioneered the use ´ of attention-based neighborhood aggregation, in one of the most common GNN variants – Graph Attention Network (GAT). In GAT, every node updates its representation by attending to its neighbors using its own representation as the query. This generalizes the standard averaging or max-pooling of neighbors (Kipf and Welling, 2017; Hamilton et al., 2017), by allowing every node to compute a weighted average of its neighbors, and (softly) select its most relevant neighbors. The work of + +![](images/3fd2d6f1a250485c1844b504947dfd994af31b1f76fbbbea17a0ae62e23a556a.jpg) +Figure 1: In a complete bipartite graph of “query nodes” $\{ q 0 , . . . , q 9 \}$ and “key nodes” $\{ k 0 , . . . , k 9 \}$ : standard GAT (Figure 1a) computes static attention – the ranking of attention coefficients is global for all nodes in the graph, and is unconditioned on the query node. For example, all queries $_ { g 0 }$ to $q 9 ,$ ) attend mostly to the 8th key $( k 8 )$ . In contrast, GATv2 (Figure 1b) can actually compute dynamic attention, where every query has a different ranking of attention coefficients of the keys. + +Velickovi ˇ c et al. also generalizes the Transformer’s (Vaswani et al., 2017) self-attention mechanism, ´ from sequences to graphs (Joshi, 2020). + +Nowadays, GAT is one of the most popular GNN architectures (Bronstein et al., 2021) and is considered as the state-of-the-art neural architecture for learning with graphs (Wang et al., 2019a). Nevertheless, in this paper we show that GAT does not actually compute the expressive, well known, type of attention (Bahdanau et al., 2014), which we call dynamic attention. Instead, we show that GAT computes only a restricted “static” form of attention: for any query node, the attention function is monotonic with respect to the neighbor (key) scores. That is, the ranking (the argsort) of attention coefficients is identical for all nodes in the graph, and is unconditioned on the query node. This fact severely hurts the expressiveness of GAT, and is demonstrated in Figure 1a. + +Supposedly, the conceptual idea of attention as the form of interaction between GNN nodes is orthogonal to the specific choice of attention function. However, Velickovi ˇ c et al.’s original design of ´ GAT has spread to a variety of domains (Yang et al., 2020; Wang et al., 2019c; Huang and Carley, 2019; Ma et al., 2020; Kosaraju et al., 2019; Nathani et al., 2019; Wu et al., 2020; Zhang et al., 2020) and has become the default implementation of “graph attention network” in all popular GNN libraries such as PyTorch Geometric (Fey and Lenssen, 2019), DGL (Wang et al., 2019b), and others (Dwivedi et al., 2020; Gordic, 2020; Brockschmidt, 2020). Many other works employed GNNs with attention ´ mechanisms other than the standard GAT’s (Yun et al. (2019), see also Appendix A). However, none of these works identified the monotonicity of GAT’s attention mechanism, the theoretical differences between attention types, nor empirically compared their performance. + +To overcome the limitation we identified in GAT, we introduce a simple fix to its attention function by only modifying the order of internal operations. The result is GATv2 – a graph attention variant that has a universal approximator attention function, and is thus strictly more expressive than GAT. The effect of fixing the attention function in GATv2 is demonstrated in Figure 1b. + +In summary, our main contribution is identifying that one of the most popular GNN types, the graph attention network, does not compute dynamic attention, the kind of attention that it seems to compute. We introduce formal definitions for analyzing the expressive power of graph attention mechanisms (Definitions 3.1 and 3.2), and derive our claims theoretically (Theorem 1) from the equations of Velickovi ˇ c et al. (2018). Empirically, we use a synthetic problem to show that standard GAT ´ cannot express problems that require dynamic attention (Section 4.1). We introduce a simple fix by switching the order of internal operations in GAT, and propose GATv2, which does compute dynamic attention (Theorem 2). We further conduct a thorough empirical comparison of GAT and GATv2 and find that GATv2 outperforms GAT across 12 benchmarks of node-, link-, and graph-prediction. For example, GATv2 outperforms extensively tuned GNNs by over $1 . 4 \%$ in the difficult “UnseenProj Test” set of the VarMisuse task (Allamanis et al., 2018), without any hyperparameter tuning; and GATv2 improves over an extensively-tuned GAT by $1 1 . 5 \%$ in 13 prediction objectives in QM9. In node-prediction benchmarks from OGB (Hu et al., 2020), not only that GATv2 outperforms GAT with respect to accuracy – we find that dynamic attention provided a much better robustness to noise. + +# 2 PRELIMINARIES + +A directed graph $\mathcal { G } = ( \nu , \mathcal { E } )$ contains nodes $\mathcal { V } = \{ 1 , . . . , n \}$ and edges $\mathcal { E } \subseteq \mathcal { V } \times \mathcal { V }$ , where $( j , i ) \in \mathcal { E }$ denotes an edge from a node $j$ to a node $i$ . We assume that every node $\textit { i } \in \mathcal { V }$ has an initial representation $\bar { \boldsymbol { h } } _ { i } ^ { ( 0 ) } \in \mathbb { R } ^ { d _ { 0 } }$ . An undirected graph can be represented with bidirectional edges. + +# 2.1 GRAPH NEURAL NETWORKS + +A graph neural network (GNN) layer updates every node representation by aggregating its neighbors’ representations. A layer’s input is a set of node representations $\{ h _ { i } \in \bar { \mathbb { R } } ^ { d } \bar { | } \ i \stackrel { - } { \in } \mathcal { V } \}$ and the set of edges $\mathcal { E }$ . A layer outputs a new set of node representations $\{ h _ { i } ^ { \prime } \in \mathbb { R } ^ { d ^ { \prime } } \mid i \in \mathcal { V } \}$ , where the same parametric function is applied to every node given its neighbors $\mathcal { N } _ { i } = \{ j \in \mathcal { V } \mid ( j , i ) \in \mathcal { E } \}$ : + +$$ +\begin{array} { r } { \pmb { h } _ { i } ^ { \prime } = f _ { \theta } \left( \pmb { h } _ { i } , \mathrm { A G G R E G A T E } \left( \{ \pmb { h } _ { j } \ | \ j \in \mathcal { N } _ { i } \} \right) \right) } \end{array} +$$ + +The design of $f$ and AGGREGATE is what mostly distinguishes one type of GNN from the other. For example, a common variant of GraphSAGE (Hamilton et al., 2017) performs an element-wise mean as AGGREGATE, followed by concatenation with $\boldsymbol { h } _ { i }$ , a linear layer and a ReLU as $f$ . + +# 2.2 GRAPH ATTENTION NETWORKS + +GraphSAGE and many other popular GNN architectures $\mathrm { { X u } }$ et al., 2019; Duvenaud et al., 2015) weigh all neighbors $j \in \mathcal N _ { i }$ with equal importance (e.g., mean or max-pooling as AGGREGATE). To address this limitation, GAT (Velickovi ˇ c et al., 2018) instantiates Equation (1) by computing a learned ´ weighted average of the representations of ${ \mathcal { N } } _ { i }$ . A scoring function $e : \mathbb { R } ^ { d } \times \mathbb { R } ^ { d } \mathbb { R }$ computes a score for every edge $( j , i )$ , which indicates the importance of the features of the neighbor $j$ to the node $i$ : + +$$ +e ( \pmb { h } _ { i } , \pmb { h } _ { j } ) = \mathrm { L e a k y R e L U } ( \pmb { a } ^ { \top } \cdot [ \pmb { W } \pmb { h } _ { i } | | \pmb { W } \pmb { h } _ { j } | ) +$$ + +where $\pmb { a } \in \mathbb { R } ^ { 2 d ^ { \prime } }$ , $W \in \mathbb { R } ^ { d ^ { \prime } \times d }$ are learned, and $\parallel$ denotes vector concatenation. These attention scores are normalized across all neighbors $j \in \mathcal N _ { i }$ using softmax, and the attention function is defined as: + +$$ +\alpha _ { i j } = \mathrm { s o f t m a x } _ { j } \left( e \left( h _ { i } , h _ { j } \right) \right) = \frac { \exp { \left( e \left( h _ { i } , h _ { j } \right) \right) } } { \sum _ { j ^ { \prime } \in { \mathcal { N } } _ { i } } \exp { \left( e \left( h _ { i } , h _ { j ^ { \prime } } \right) \right) } } +$$ + +Then, GAT computes a weighted average of the transformed features of the neighbor nodes (followed by a nonlinearity $\sigma$ ) as the new representation of $i$ , using the normalized attention coefficients: + +$$ +\begin{array} { r } { \pmb { h } _ { i } ^ { \prime } = \sigma \left( \sum _ { j \in \mathcal { N } _ { i } } \pmb { \alpha } _ { i j } \cdot \pmb { W } \pmb { h } _ { j } \right) } \end{array} +$$ + +From now on, we will refer to Equations (2) to (4) as the definition of GAT. + +# 3 THE EXPRESSIVE POWER OF GRAPH ATTENTION MECHANISMS + +In this section, we explain why attention is limited when it is not dynamic (Section 3.1). We then show that GAT is severely constrained, because it can only compute static attention (Section 3.2). Next, we show how GAT can be fixed (Section 3.3), by simply modifying the order of operations. + +We refer to a neural architecture (e.g., the scoring or the attention function of GAT) as a family of functions, parameterized by the learned parameters. An element in the family is a concrete function with specific trained weights. In the following, we use $[ n ]$ to denote the set $[ n ] = \{ 1 , 2 , . . . , n \} \subset \mathbb { N }$ . + +# 3.1 THE IMPORTANCE OF DYNAMIC WEIGHTING + +Attention is a mechanism for computing a distribution over a set of input key vectors, given an additional query vector. If the attention function always weighs one key at least as much as any other key, unconditioned on the query, we say that this attention function is static: + +Definition 3.1 (Static attention). A (possibly infinite) family of scoring functions $\mathcal { F } \subseteq$ $\left( \mathbb { R } ^ { d } \times \mathbb { R } ^ { d } \to \mathbb { R } \right)$ computes static scoring for a given set of key vectors $\mathbb { K } = \{ k _ { 1 } , . . . , k _ { n } \} \subset \mathbb { R } ^ { d }$ and query vectors $\mathbb { Q } = \{ \ b { q } _ { 1 } , . . . , \ b { q } _ { m } \} \subset \mathbb { R } ^ { d }$ , if for every $f \in { \mathcal { F } }$ there exists a “highest scoring” key $j _ { f } \in [ n ]$ such that for every query $i \in [ m ]$ and key $j \in [ n ]$ it holds that $f \left( \boldsymbol { q } _ { i } , \boldsymbol { k } _ { j _ { f } } \right) \geq f \left( \boldsymbol { q } _ { i } , \boldsymbol { k } _ { j } \right)$ . We say that a family of attention functions computes static attention given $\Bbbk$ and $\mathbb { Q }$ , if its scoring function computes static scoring, possibly followed by monotonic normalization such as softmax. + +Static attention is very limited because every function $f \in { \mathcal { F } }$ has a key that is always selected, regardless of the query. Such functions cannot model situations where different keys have different relevance to different queries. Static attention is demonstrated in Figure 1a. + +The general and powerful form of attention is dynamic attention: + +Definition 3.2 (Dynamic attention). A (possibly infinite) family of scoring functions $\mathcal { F } \subseteq$ $\left( \mathbb { R } ^ { d } \times \mathbb { R } ^ { d } \to \mathbb { R } \right)$ computes dynamic scoring for a given set of key vectors $\mathbb { K } = \{ \boldsymbol { k } _ { 1 } , . . . , \boldsymbol { k } _ { n } \} \subset \mathbb { R } ^ { d }$ and query vectors $\mathbb { Q } = \{ \ b { q } _ { 1 } , . . . , \ b { q } _ { m } \} \subset \mathbb { R } ^ { d }$ , if for any mapping $\varphi$ : $[ m ] [ n ]$ there exists $f \in { \mathcal { F } }$ such that for any query $i \in [ m ]$ and any key $j _ { \neq \varphi ( i ) } \in [ n ] \colon f \left( q _ { i } , k _ { \varphi ( i ) } \right) > f \left( q _ { i } , k _ { j } \right) .$ . We say that a family of attention functions computes dynamic attention for $\mathbb { K }$ and $\mathbb { Q }$ , if its scoring function computes dynamic scoring, possibly followed by monotonic normalization such as softmax. + +That is, dynamic attention can select every key $\varphi \left( i \right)$ using the query $i$ , by making $f \left( q _ { i } , k _ { \varphi ( i ) } \right)$ the maximal in $\left\{ f \left( \pmb { q } _ { i } , \pmb { k } _ { j } \right) \mid j \in [ n ] \right\}$ . Note that dynamic and static attention are exclusive properties, but they are not complementary. Further, every dynamic attention family has strict subsets of static attention families with respect to the same $\mathbb { K }$ and $\mathbb { Q }$ . Dynamic attention is demonstrated in Figure 1b. + +Attending by decaying Another way to think about attention is the ability to “focus” on the most relevant inputs, given a query. Focusing is only possible by decaying other inputs, i.e., giving these decayed inputs lower scores than others. If one key is always given an equal or greater attention score than other keys (as in static attention), no query can ignore this key or decay this key’s score. + +# 3.2 THE LIMITED EXPRESSIVITY OF GAT + +Although the scoring function $e$ can be defined in various ways, the original definition of Velickovi ˇ c´ et al. (2018) (Equation (2)) has become the de facto practice: it has spread to a variety of domains and is now the standard implementation of “graph attention network” in all popular GNN libraries (Fey and Lenssen, 2019; Wang et al., 2019b; Dwivedi et al., 2020; Gordic, 2020; Brockschmidt, 2020). ´ + +The motivation of GAT is to compute a representation for every node as a weighted average of its neighbors. Statedly, GAT is inspired by the attention mechanism of Bahdanau et al. (2014) and the self-attention mechanism of the Transformer (Vaswani et al., 2017). Nonetheless: + +Theorem 1. A GAT layer computes only static attention, for any set of node representations $\mathbb { K } =$ $\mathbb { Q } = \{ h _ { 1 } , . . . , h _ { n } \}$ . In particular, for $n > 1$ , a GAT layer does not compute dynamic attention. + +Proof. Let $\mathcal { G } = ( \nu , \mathcal { E } )$ be a graph modeled by a GAT layer with some $\textbf { \em a }$ and $W$ values (Equations (2) and (3)), and having node representations $\{ h _ { 1 } , . . . , h _ { n } \}$ . The learned parameter $\textbf { \em a }$ can be written as a concatenation $\pmb { a } = [ \pmb { a } _ { 1 } | | \pmb { a } _ { 2 } ] \in \mathbb { R } ^ { 2 d ^ { \prime } }$ such that $\mathbf { \boldsymbol { a } } _ { 1 } , \mathbf { \boldsymbol { a } } _ { 2 } \in \mathbb { R } ^ { d ^ { \prime } }$ , and Equation (2) can be re-written as: + +$$ +e \left( \boldsymbol { h } _ { i } , \boldsymbol { h } _ { j } \right) = \mathrm { L e a k y R e L U } \left( \boldsymbol { a } _ { 1 } ^ { \top } \boldsymbol { W } \boldsymbol { h } _ { i } + \boldsymbol { a } _ { 2 } ^ { \top } \boldsymbol { W } \boldsymbol { h } _ { j } \right) +$$ + +Since $\nu$ is finite, there exists a node $j _ { m a x } \in \mathcal { V }$ such that ${ \pmb a } _ { 2 } ^ { \top } { \pmb W } { \pmb h } _ { j _ { m a x } }$ is maximal among all nodes $j \in \mathcal { V } \left( j _ { m a x } \right.$ is the $j _ { f }$ required by Definition 3.1). Due to the monotonicity of LeakyReLU and softmax, for every query node $i \in \mathcal V$ , the node $j _ { m a x }$ also leads to the maximal value of its attention distribution $\{ \alpha _ { i j } \mid j \in \mathcal { V } \}$ . Thus, from Definition 3.1 directly, $\alpha$ computes only static attention. This also implies that $\alpha$ does not compute dynamic attention, because in GAT, Definition 3.2 holds only for constant mappings $\varphi$ that map all inputs to the same output. □ + +The consequence of Theorem 1 is that for any set of nodes $\nu$ and a trained GAT layer, the attention function $\alpha$ defines a constant ranking (argsort) of the nodes, unconditioned on the query nodes $i$ . That is, we can denote $s _ { j } = { a } _ { 2 } ^ { \top } \bar { W } h _ { j }$ and get that for any choice of $\boldsymbol { h } _ { i }$ , $\alpha$ is monotonic with respect to the per-node scores $\{ s _ { j } \mid j \in \mathcal { V } \}$ . This global ranking induces the local ranking of every neighborhood ${ \mathcal { N } } _ { i }$ . The only effect of $\boldsymbol { h } _ { i }$ is in the “sharpness” of the produced attention distribution. This is demonstrated in Figure 1a (bottom), where different curves denote different queries $( h _ { i } )$ . A discussion regarding the generalization to multi-head attention can be found in Appendix C. + +# 3.3 BUILDING DYNAMIC GRAPH ATTENTION NETWORKS + +To create a dynamic graph attention network, we modify the order of internal operations in GAT and introduce GATv2 – a simple fix of GAT that has a strictly more expressive attention mechanism. + +GATv2 The main problem in the standard GAT scoring function (Equation (2)) is that the learned layers $W$ and $^ { a }$ are applied consecutively, and thus can be collapsed into a single linear layer. To fix this limitation, we simply apply the $\textbf { \em a }$ layer after the nonlinearity (LeakyReLU), and the $W$ layer after the concatenation, 5 effectively applying an MLP to compute the score for each query-key pair: + +$$ +\begin{array} { r l } & { e \left( \boldsymbol { h } _ { i } , \boldsymbol { h } _ { j } \right) = \mathrm { L e a k y R e L U } \left( \boldsymbol { a } ^ { \top } \cdot \left[ \boldsymbol { W } \boldsymbol { h } _ { i } \right| \left| \boldsymbol { W } \boldsymbol { h } _ { j } \right| \right) \right. } \\ & { e \left. \left( \boldsymbol { h } _ { i } , \boldsymbol { h } _ { j } \right) = \boldsymbol { a } ^ { \top } \mathrm { L e a k y R e L U } \left( \boldsymbol { W } \cdot \left[ \boldsymbol { h } _ { i } \right| \left| \boldsymbol { h } _ { j } \right] \right) } \end{array} +$$ + +The simple modification makes a significant difference in the expressiveness of the attention function: + +Theorem 2. A GATv2 layer computes dynamic attention for any set of node representations $\mathbb { K } =$ $\mathbb { Q } = \{ h _ { 1 } , . . . , h _ { n } \}$ . + +We prove Theorem 2 in Appendix B. The main idea is that we can define an appropriate function that GATv2 will be a universal approximator (Cybenko, 1989; Hornik, 1991) of. In contrast, GAT (Equation (52)) cannot approximate any such desired function (Theorem 1). + +Complexity GATv2 has the same time-complexity as GAT’s declared complexity: $\mathcal { O } \left( | \mathcal { V } | d d ^ { \prime } + | \mathcal { E } | d ^ { \prime } \right)$ . However, by merging its linear layers, GAT can be computed faster than stated by Velickovi ˇ c et al. ´ (2018). For a detailed time- and parametric-complexity analysis, see Appendix I. + +# 4 EVALUATION + +First, we demonstrate the weakness of GAT using a simple synthetic problem that GAT cannot even fit (cannot even achieve high training accuracy), but is easily solvable by GATv2 (Section 4.1). Second, we show that GATv2 is much more robust to edge noise, because its dynamic attention mechanisms allow it to decay noisy (false) edges, while GAT’s performance severely decreases as noise increases (Section 4.2). Finally, we compare GAT and GATv2 across 12 benchmarks overall. (Sections 4.3 to 4.6 and appendix F.4). We find that GAT is inferior to GATv2 across all examined benchmarks. + +Setup When previous results exist, we take hyperparameters that were tuned for GAT and use them in GATv2, without any additional tuning. Self-supervision (Kim and Oh, 2021; Rong et al., 2020a), graph regularization (Zhao and Akoglu, 2020; Rong et al., 2020b), and other tricks (Wang, 2021; Huang et al., 2021) are orthogonal to the contribution of the GNN layer itself, and may further improve all GNNs. In all experiments of GATv2, we constrain the learned matrix by setting $W = [ W ^ { \prime } | | W ^ { \prime } ]$ , to rule out the increased number of parameters over GAT as the source of empirical difference (see Appendix I.2). Training details, statistics, and code are provided in Appendix D. + +![](images/79aab347d93a4c3bb0bb39665869459e72b57008bc185548856fefb238200d5b.jpg) +Figure 2: The DICTIONARYLOOKUP problem of size $k { = } 4$ : every node in the bottom row has an alphabetic attribute $( \{ { \mathrm { A } } , { \mathrm { B } } , { \mathrm { C } } , \ldots \} )$ and a numeric value $( \{ 1 , 2 , 3 , \ldots \} )$ ; every node in the upper row has only an attribute; the goal is to predict the value for each node in the upper row, using its attribute. + +![](images/eee3bdfe02443e30b1eb1a486a618dd6fcf3d8c448dc16f59b4318aa5c305de0.jpg) +Figure 3: The DICTIONARYLOOKUP problem: GATv2 easily achieves $100 \%$ train and test accuracies even for $k { = } 1 0 0$ and using only a single head. + +Our main goal is to compare dynamic and static graph attention mechanisms. However, for reference, we also include non-attentive baselines such as GCN (Kipf and Welling, 2017), GIN (Xu et al., 2019) and GraphSAGE (Hamilton et al., 2017). These non-attentive GNNs can be thought of as a special case of attention, where every node gives all its neighbors the same attention score. Additional comparison to a Transformer-style scaled dot-product attention (“DPGAT”), which is strictly weaker than our proposed GATv2 (see a proof in Appendix G.1), is shown in Appendix G. + +# 4.1 SYNTHETIC BENCHMARK: DICTIONARYLOOKUP + +The DICTIONARYLOOKUP problem is a contrived problem that we designed to test the ability of a GNN architecture to perform dynamic attention. Here, we demonstrate that GAT cannot learn this simple problem. Figure 2 shows a complete bipartite graph of $2 k$ nodes. Each “key node” in the bottom row has an attribute $( \{ \mathrm { A } , \mathrm { B } , \mathrm { C } , \bar { \ldots } \} )$ and a value $( \{ 1 , 2 , 3 , \ldots \} )$ . Each “query node” in the upper row has only an attribute $( \{ \mathrm { A } , \mathrm { B } , \mathrm { C } , \ldots \} )$ . The goal is to predict the value of every query node (upper row), according to its attribute. Each graph in the dataset has a different mapping from attributes to values. We created a separate dataset for each $k = \{ 1 , 2 , 3 , \ldots \}$ , for which we trained a different model, and measured per-node accuracy. + +Although this is a contrived problem, it is relevant to any subgraph with keys that share more than one query, and each query needs to attend to the keys differently. Such subgraphs are very common in a variety of real-world domains. This problem tests the layer itself because it can be solved using a single GNN layer, without suffering from multi-layer side-effects such as over-smoothing (Li et al., 2018), over-squashing (Alon and Yahav, 2021), or vanishing gradients (Li et al., 2019). Our code will be made publicly available, to serve as a testbed for future graph attention mechanisms. + +Results Figure 3 shows the following surprising results: GAT with a single head $( \mathrm { G A T _ { 1 } } _ { h } )$ ) failed to fit the training set for any value of $k$ , no matter for how many iterations it was trained, and after trying various training methods. Thus, it expectedly fails to generalize (resulting in low test accuracy). Using 8 heads, $\mathrm { G A T } _ { 8 h }$ successfully fits the training set, but generalizes poorly to the test set. In contrast, GATv2 easily achieves $100 \%$ training and $100 \%$ test accuracies for any value of $k$ , and even for $k { = } 1 0 0$ (not shown) and using a single head, thanks to its ability to perform dynamic attention. These results clearly show the limitations of GAT, which are easily solved by GATv2. An additional comparison to GIN, which could not fit this dataset, is provided in Figure 5 in Appendix F.1. + +Visualization Figure 1a (top) shows a heatmap of GAT’s attention scores in this DICTIONARYLOOKUP problem. As shown, all query nodes $q 0$ to $q 9$ attend mostly to the eighth key $( k 8 )$ , and have the same ranking of attention coefficients (Figure 1a (bottom)). In contrast, Figure 1b shows how GATv2 can select a different key node for every query node, because it computes dynamic attention. + +The role of multi-head attention Velickovi ˇ c et al. (2018) found the role of multi-head attention to ´ be stabilizing the learning process. Nevertheless, Figure 3 shows that increasing the number of heads strictly increases training accuracy, and thus, the expressivity. Thus, GAT depends on having multiple attention heads. In contrast, even a single GATv2 head generalizes better than a multi-head GAT. + +# 4.2 ROBUSTNESS TO NOISE + +We examine the robustness of dynamic and static attention to noise. In particular, we focus on structural noise: given an input graph $\mathcal { G } = ( \nu , \mathcal { E } )$ and a noise ratio $0 \leq p \leq 1$ , we randomly sample $| { \mathcal { E } } | \times p$ non-existing edges $\mathcal { E } ^ { \prime }$ from $\mathcal { V } \times \mathcal { V } \backslash \mathcal { E }$ . We then train the GNN on the noisy graph $\mathcal { G } ^ { \prime } { = } ( \dot { \mathcal { V } } , \mathcal { E } \cup \bar { \mathcal { E } } ^ { \prime } )$ . + +![](images/77bbac2b2bb1e1b15f9fdfb4dc28a2c29279f545cc08fbe57d963cffd3d7f569.jpg) +Figure 4: Test accuracy compared to the noise ratio: GATv2 is more robust to structural noise compared to GAT. Each point is an average of 10 runs, error bars show standard deviation. + +Results Figure 8 shows the accuracy on two node-prediction datasets from the Open Graph Benchmark (OGB; Hu et al., 2020) as a function of the noise ratio $p$ . As $p$ increases, all models show a natural decline in test accuracy in both datasets. Yet, thanks to their ability to compute dynamic attention, GATv2 shows a milder degradation in accuracy compared to GAT, which shows a steeper descent. We hypothesize that the ability to perform dynamic attention helps the models distinguishing between given data edges $( \mathcal { E } )$ and noise edges $( \mathcal { E } ^ { \prime } )$ ; in contrast, GAT cannot distinguish between edges, because it scores the source and target nodes separately. These results clearly demonstrate the robustness of dynamic attention over static attention in noisy settings, which are common in reality. + +# 4.3 PROGRAMS: VARMISUSE + +Setup VARMISUSE (Allamanis et al., 2018) is an inductive node-pointing problem that depends on 11 types of syntactic and semantic interactions between elements in computer programs. + +We used the framework of Brockschmidt (2020), who performed an extensive hyperparameter tuning by searching over 30 configurations for every GNN type. We took their best GAT hyperparameters and used them to train GATv2, without further tuning. + +Results As shown in Table 1, GATv2 is more accurate than GAT and other GNNs in the SeenProj test sets. Furthermore, GATv2 achieves an even higher improvement in the UnseenProj test set. Overall, these results demonstrate the power of GATv2 in modeling complex relational problems, especially + +Table 1: Accuracy (5 runs $\pm$ stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both test sets, using GAT’s hyperparameters. $\dagger$ previously reported by Brockschmidt (2020). + +
ModelSeenProjUnseenProj
No-GCN+87.2±1.581.4±2.3
AttentionGIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
+ +since it outperforms extensively tuned models, without any further tuning by us. + +# 4.4 NODE-PREDICTION + +We further compare GATv2, GAT, and other GNNs on four node-prediction datasets from OGB. + +Table 2: Average accuracy (Table 2a) and ROC-AUC (Table 2b) in node-prediction datasets (10 runs±std). In all datasets, GATv2 outperforms GAT. † – previously reported by Hu et al. (2020). + +
(a)(b)
ModelAttn. Headsogbn-arxivogbn-productsogbn-magogbn-proteins
GCN+071.74±0.2978.97±0.3330.43±0.2572.51±0.35
GraphSAGE+071.49±0.2778.70±0.3631.53±0.1577.68±0.20
GAT171.59±0.3879.04±1.5432.20±1.4670.77±5.79
871.54±0.3077.23±2.3731.75±1.6078.63±1.62
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.4477.23±3.32
871.87±0.2578.46±2.4532.52±0.3979.52±0.55
+ +Results Results are shown in Table 2. In all settings and all datasets, GATv2 is more accurate than GAT and the non-attentive GNNs. Interestingly, in the datasets of Table 2a, even a single head of $G A T \nu 2$ outperforms GAT with 8 heads. In Table 2b (ogbn-proteins), increasing the number of heads results in a major improvement for GAT (from 70.77 to 78.63), while GATv2 already gets most of the benefit using a single attention head. These results demonstrate the superiority of GATv2 over GAT in node prediction (and even with a single head), thanks to GATv2’s dynamic attention. + +# 4.5 GRAPH-PREDICTION: QM9 + +Setup In the QM9 dataset (Ramakrishnan et al., 2014; Gilmer et al., 2017), each graph is a molecule and the goal is to regress each graph to 13 real-valued quantum chemical properties. We used the implementation of Brockschmidt (2020) who performed an extensive hyperparameter search over 500 configurations; we took their best-found configuration of GAT to implement GATv2. + +Table 3: Average error rates (lower is better), 5 runs for each property, on the QM9 dataset. The best result among GAT and GATv2 is marked in bold; the globally best result among all GNNs is marked in bold and underline. $\dagger$ was previously tuned and reported by Brockschmidt (2020). + +
Predicted Property
Model12345678910111213Rel. to GAT
GCNt3.214.221.451.622.4216.38 17.40 7.828.249.057.003.931.02-1.5%
GINt2.644.671.421.502.2715.63 12.93 5.8818.715.625.383.531.05-2.3%
GATt2.684.651.481.532.3152.39 14.87 7.6116.867.646.544.111.48+0%
GATv22.654.281.411.472.2916.37 14.03 6.076.286.605.973.571.59-11.5%
+ +Results Table 3 shows the main results: GATv2 achieves a lower (better) average error than GAT, by $1 1 . 5 \%$ relatively. GAT achieves the overall highest average error. In some properties, the non-attentive GNNs, GCN and GIN, perform best. We hypothesize that attention is not needed in modeling these properties. Generally, GATv2 achieves the lowest overall average relative error (rightmost column). + +# 4.6 LINK-PREDICTION + +We compare GATv2, GAT, and other GNNs in link-prediction datasets from OGB. + +Results Table 8 shows that in all datasets, GATv2 achieves a higher MRR than GAT, which achieves the lowest MRR. However, the non-attentive GraphSAGE performs better than all attentive GNNs. We hypothesize that attention might not be needed in these datasets. Another possibility is that dynamic attention is especially useful in graphs that have high node degrees: in ogbn-products and ogbn-proteins (Table 2) the average node degrees are 50.5 and 597, respectively (see Table 5 in + +Appendix E). ogbl-collab and ogbl-citation2 (Table 8), however, have much lower average node degrees – of 8.2 and 20.7. We hypothesize that a dynamic attention mechanism is especially useful to select the most relevant neighbors when the total number of neighbors is high. We leave the study of the effect of the datasets’s average node degrees on the optimal GNN architecture for future work. + +# 4.7 DISCUSSION + +In all examined benchmarks, we found that $G A T \nu 2$ is more accurate than GAT. Further, we found that GATv2 is significantly more robust to noise than GAT. In the synthetic DICTIONARYLOOKUP benchmark (Section 4.1), GAT fails to express the data, and thus achieves even poor training accuracy. + +In few of the benchmarks (Table 8 and some of the properties in Table 3) – a non-attentive model such as GCN or GIN achieved a higher accuracy than all GNNs that do use attention. We hypothesize that attention is not needed in these datasets. + +Which graph attention mechanism should I use? It is usually impossible to determine in advance which architecture would perform best. A theoretically weaker model may perform better in practice, because a stronger model might overfit the training data if the task is “too simple” and does not require such expressiveness. Intuitively, we believe that the more complex the interactions between nodes are – the more benefit a GNN can take from theoretically stronger graph attention mechanisms such as GATv2. The main question is whether the problem has a global ranking of “influential” nodes (GAT is sufficient), or do different nodes have different rankings of neighbors (use GATv2). + +Velickovi ˇ c, the author of GAT, has confirmed on Twitter ´ 6 that GAT was designed to work in the “easy-to-overfit” datasets of the time (2017), such as Cora, Citeseer and Pubmed (Sen et al., 2008), where the data might had an underlying static ranking of “globally important” nodes. Velickovi ˇ c´ agreed that newer and more challenging benchmarks may demand stronger attention mechanisms such as GATv2. that depends on the representation of the query node as well. In this paper, we revisit the traditional assumptions and show that many modern graph benchmarks and datasets contain more complex interactions, and thus require dynamic attention. + +# 5 CONCLUSION + +In this paper, we identify that the popular and widespread Graph Attention Network does not compute dynamic attention. Instead, the attention mechanism in the standard definition and implementations of GAT is only static: for any query, its neighbor-scoring is monotonic with respect to per-node scores. Further, the ranking (the argsort) of attention coefficients is identical for all nodes in the graph, and is unconditioned on the query node. As a result, GAT cannot even express simple alignment problems. To address this limitation, we introduce a simple fix and propose GATv2: by modifying the order of operations in GAT, GATv2 achieves a universal approximator attention function and is thus strictly more powerful than GAT. + +We demonstrate the empirical advantage of GATv2 over GAT in a synthetic problem that requires dynamic selection of nodes, and in 12 benchmarks from OGB and other public datasets. Our experiments show that GATv2 outperforms GAT in all benchmarks while having the same parametric cost. + +We encourage the community to use GATv2 instead or in addition to GAT whenever comparing new GNN architectures to the common strong baselines. In complex tasks and domains and in challenging datasets, a model that uses GAT as an internal component can replace it with GATv2 to benefit from a strictly more powerful model. To this end, we make our code publicly available at https: //github.com/tech-srl/how_attentive_are_gats ,7 and GATv2 is available as part of the PyTorch Geometric library,8 the Deep Graph Library,9 and the TensorFlow GNN library.10 + +# ACKNOWLEDGMENTS + +We thank Gail Weiss for the helpful discussions, thorough feedback, and inspirational paper (Weiss et al., 2018). We also thank Petar Velickovi ˇ c for the useful initial discussion about the complexity ´ and implementation of GAT, Lucio Dery, and the Graph Representation Learning Reading Group at Mila11 for their useful comments and questions. + +# A RELATED WORK + +Attention in GNNs Modeling pairwise interactions between elements in graph-structured data goes back to interaction networks (Battaglia et al., 2016; Hoshen, 2017) and relational networks (Santoro et al., 2017). The GAT formulation of Velickovi ˇ c et al. (2018) rose as the most popular framework ´ for attentional GNNs, thanks to its simplicity, generality, and applicability beyond reinforcement learning (Denil et al., 2017; Duan et al., 2017). Nevertheless, in this work, we show that the popular and widespread definition of GAT is severely constrained to static attention only. + +Other graph attention mechanisms Many works employed GNNs with attention mechanisms other than the standard GAT’s (Zhang et al., 2018; Thekumparampil et al., 2018; Gao and Ji, 2019; Lukovnikov and Fischer, 2021; Shi et al., 2020; Dwivedi and Bresson, 2020; Busbridge et al., 2019; Rong et al., 2020a; Velickovi ˇ c et al., 2020), and Lee et al. (2018) conducted an extensive survey ´ of attention types in GNNs. However, none of these works identified the monotonicity of GAT’s attention mechanism, the theoretical differences between attention types, nor empirically compared their performance. Kim and Oh (2021) compared two graph attention mechanisms empirically, but in a specific self-supervised scenario, without observing the theoretical difference in their expressiveness. + +The static attention of GAT Qiu et al. (2018) recognized the order-preserving property of GAT, but did not identify the severe theoretical constraint that this property implies: the inability to perform dynamic attention (Theorem 1). Furthermore, they presented GAT’s monotonicity as a desired trait (!) To the best of our knowledge, our work is the first work to recognize the inability of GAT to perform dynamic attention and its practical harmful consequences. + +# B PROOF FOR THEOREM 2 + +For brevity, we repeat our definition of dynamic attention (Definition 3.2): + +Definition 3.2 (Dynamic attention). A (possibly infinite) family of scoring functions $\mathcal { F } \subseteq$ $\left( \mathbb { R } ^ { d } \times \mathbb { R } ^ { d } \to \mathbb { R } \right)$ computes dynamic scoring for a given set of key vectors $\mathbb { K } = \{ \boldsymbol { k } _ { 1 } , . . . , \boldsymbol { k } _ { n } \} \subset \mathbb { R } ^ { d }$ and query vectors $\mathbb { Q } = \{ \ b { q } _ { 1 } , . . . , \ b { q } _ { m } \} \subset \mathbb { R } ^ { d }$ , if for any mapping $\varphi$ : $[ m ] [ n ]$ there exists $f \in { \mathcal { F } }$ such that for any query $i \in [ m ]$ and any key $j _ { \neq \varphi ( i ) } \in [ n ] \colon f \left( q _ { i } , k _ { \varphi ( i ) } \right) > f \left( q _ { i } , k _ { j } \right) .$ . We say that a family of attention functions computes dynamic attention for $\mathbb { K }$ and $\mathbb { Q }$ , if its scoring function computes dynamic scoring, possibly followed by monotonic normalization such as softmax. + +Theorem 2. A GATv2 layer computes dynamic attention for any set of node representations $\mathbb { K } =$ $\mathbb { Q } = \{ h _ { 1 } , . . . , h _ { n } \}$ . + +Proof. Let $\mathcal { G } ~ = ~ ( \nu , \mathcal { E } )$ be a graph modeled by a GATv2 layer, having node representations $\{ h _ { 1 } , . . . , h _ { n } \}$ , and let $\varphi : [ n ] \to [ n ]$ be any node mapping $[ n ] [ n ]$ . We define $\dot { g } : \mathbb { R } ^ { 2 d } \mathbb { R }$ as follows: + +$$ +g \left( \pmb { x } \right) = \left. { 1 } \pmb { \mathrm { \rbrace } } \ i : \pmb { x } = \left[ \pmb { h } _ { i } \right| \left| \pmb { h } _ { \varphi ( i ) } \right] \right. +$$ + +Next, we define a continues function $\widetilde { g } : { \mathbb { R } } ^ { 2 d } \to { \mathbb { R } }$ that equals to $g$ in only specific $n ^ { 2 }$ inputs: + +$$ +\widetilde { g } ( [ h _ { i } \| h _ { j } ] ) = g ( [ h _ { i } \| h _ { j } ] ) , \forall i , j \in [ n ] +$$ + +For all other inputs $\boldsymbol { x } \in \mathbb { R } ^ { 2 d }$ , $\widetilde g ( x )$ realizes to any values that maintain the continuity of $\widetilde g$ (this is epossible because we fixed the values of $\widetilde g$ for only a finite set of points). 12 + +Thus, for every node $i \in \nu$ and $j _ { \neq \varphi ( i ) } \in \mathcal { V }$ : + +$$ +1 = \widetilde { g } \left( \left[ h _ { i } \Vert h _ { \varphi ( i ) } \right] \right) > \widetilde { g } \left( \left[ h _ { i } \Vert h _ { j } \right] \right) = 0 +$$ + +If we concatenate the two input vectors, and define the scoring function $e$ of GATv2 (Equation (7)) as a function of the concatenated vector $[ h _ { i } | | h _ { j } ]$ , from the universal approximation theorem (Hornik et al., 1989; Cybenko, 1989; Funahashi, 1989; Hornik, 1991), $e$ can approximate $\widetilde g$ for any compact subset of R2d. + +Thus, for any sufficiently small $\epsilon$ (any $0 < \epsilon < 1 / 2$ ) there exist parameters $W$ and $\textbf { \em a }$ such that for every node $i \in \nu$ and every $\mathscr { j } _ { \neq \varphi ( i ) }$ : + +$$ +{ e _ { W , a } } \left( { h _ { i } } , { h _ { \varphi ( i ) } } \right) > 1 - \epsilon > 0 + \epsilon > { e _ { W , a } } \left( { h _ { i } } , { h _ { j } } \right) +$$ + +and due to the increasing monotonicity of softmax: + +$$ +\alpha _ { i , \varphi ( i ) } > \alpha _ { i , j } +$$ + +The choice of nonlinearity In general, these results hold if GATv2 had used any common nonpolynomial activation function (such as ReLU, sigmoid, or the hyperbolic tangent function). The LeakyReLU activation function of GATv2 does not change its universal approximation ability (Leshno et al., 1993; Pinkus, 1999; Park et al., 2021), and it was chosen only for consistency with the original definition of GAT. + +# C GENERALIZATION TO MULTI-HEAD ATTENTION + +Velickovi ˇ c et al. (2018) found it beneficial to employ ´ $H$ separate attention heads and concatenate their outputs, similarly to Transformers. In this case, Theorem 1 holds for each head separately: every head $h \in [ H ]$ has a (possibly different) node that maximizes $\{ s _ { j } ^ { ( h ) } \mid j \in \mathcal { V } \}$ , and the output is the concatenation of $H$ static attention heads. + +# D TRAINING DETAILS + +In this section we elaborate on the training details of all of our experiments. All models use residual connections as in Velickovi ˇ c et al. (2018). All used code and data are publicly available under the ´ MIT license. + +# D.1 NODE- AND LINK-PREDICTION + +We used the provided splits of OGB (Hu et al., 2020) and the Adam optimizer. We tuned the following hyperparameters: number of layers $\in \{ 2 , 3 , 6 \}$ , hidden size $\in \{ 6 4 , 1 2 8 , 2 5 6 \}$ , learning rate $\in \{ 0 . 0 0 0 5 , 0 . 0 0 1 , 0 . 0 0 5 , 0 . 0 1 \}$ and sampling method – full batch, GraphSAINT (Zeng et al., 2019) and NeighborSampling (Hamilton et al., 2017). We tuned hyperparameters according to validation score and early stopping. The final hyperparameters are detailed in Table 4. + +Table 4: Training details of node- and link-prediction datasets. + +
Dataset# layersHidden sizeLearning rateSampling method
ogbn-arxiv32560.01GraphSAINT
ogbn-products31280.001NeighborSampling
ogbn-mag22560.01NeighborSampling
ogbn-proteins6640.01NeighborSampling
ogbl-collab3640.001Full Batch
ogbl-citation232560.0005NeighborSampling
+ +# D.2 ROBUSTNESS TO NOISE + +In these experiments, we used the same best-found hyperparameters in node-prediction, with 8 attention heads in ogbn-arxiv and 1 head in ogbn-mag. Each point is an average of 10 runs. + +# D.3 SYNTHETIC BENCHMARK: DICTIONARYLOOKUP + +In all experiments, we used a learning rate decay of 0.5, a hidden size of $d = 1 2 8$ , a batch size of 1024, and the Adam optimizer. + +We created a separate dataset for every graph size $( k )$ , and we split each such dataset to train and test with a ratio of 80:20. Since this is a contrived problem, we did not use a validation set, and the reported test results can be thought of as validation results. Every model was trained on a fixed value of $k$ . Every key node (bottom row in Figure 2) was encoded as a sum of learned attribute embedding and a value embedding, followed by ReLU. + +We experimented with layer normalization, batch normalization, dropout, various activation functions and various learning rates. None of these changed the general trend, so the experiments in Figure 3 were conducted without any normalization, without dropout and a learning rate of 0.001. + +# D.4 PROGRAMS: VARMISUSE + +We used the code, splits, and the same best-found configurations as Brockschmidt (2020), who performed an extensive hyperparameter tuning by searching over 30 configurations for each GNN type. We trained each model five times. + +We took the best-found hyperparameters of Brockschmidt (2020) for GAT and used them to train GATv2, without any further tuning. + +# D.5 GRAPH-PREDICTION: QM9 + +We used the code and splits of Brockschmidt (2020) who performed an extensive hyperparameter search over 500 configurations. We took the best-found hyperparameters of Brockschmidt (2020) for GAT and used them to train GATv2. The only minor change from GAT is placing a residual connection after every layer, rather than after every other layer, which is within the experimented hyperparameter search that was reported by Brockschmidt (2020). + +# D.6 COMPUTE AND RESOURCES + +Our experiments consumed approximately 100 days of GPU in total. We used cloud GPUs of type V100, and we used RTX 3080 and 3090 in local GPU machines. + +# E DATA STATISTICS + +# E.1 NODE- AND LINK-PREDICTION DATASETS + +Statistics of the OGB datasets we used for node- and link-prediction are shown in Table 5. + +Table 5: Statistics of the OGB datasets (Hu et al., 2020). + +
Dataset# nodes# edgesAvg. node degreeDiameter
ogbn-arxiv169,3431,166,24313.723
ogbn-mag1,939,74321,111,00721.76
ogbn-products2,449,02961,859,14050.527
ogbn-proteins132,53439,561,252597.09
ogbl-collab235,8681,285,4658.222
ogbl-citation22,927,96330,561,18720.721
+ +# E.2 QM9 + +Statistics of the QM9 dataset, as used in Brockschmidt (2020) are shown in Table 6. + +
TrainingValidationTest
# examples110,46210,00010,000
# nodes -average18.0318.0618.09
# edges - average18.6518.6718.72
Diameter - average6.356.356.35
+ +Table 6: Statistics of the QM9 chemical dataset (Ramakrishnan et al., 2014) as used by Brockschmidt (2020). + +# E.3 VARMISUSE + +Statistics of the VARMISUSE dataset, as used in Allamanis et al. (2018) and Brockschmidt (2020), are shown in Table 7. + +
TrainingValidationUnseenProject TestSeenProject Test
# graphs2543604265411703659974
# nodes - average2377174219593986
# edges - average72987851588212925
Diameter-average7.887.887.787.82
+ +Table 7: Statistics of the VARMISUSE dataset (Allamanis et al., 2018) as used by Brockschmidt (2020). + +# F ADDITIONAL RESULTS + +# F.1 DICTIONARYLOOKUP + +Figure 5 shows additional comparison between GATv2 and GIN (Xu et al., 2019) in the DICTIONARYLOOKUP problem. GATv2 easily achieves $100 \%$ train and test accuracy even for $k { = } 1 0 0$ and using only a single head. GIN, although considered as more expressive than other GNNs, cannot perfectly fit the training data (with a model size of $d = 1 2 8$ ) starting from $k { = } 2 0$ . + +# F.2 LINK-PREDICTION + +Results for link-prediction are showen in Table 8. + +![](images/4233ec4a02df2921f13b438873a45c7450236e836cc22f098f64cf61a279e12d.jpg) +Figure 5: Train and test accuracy across graph sizes in the DICTIONARYLOOKUP problem. GATv2 easily achieves $100 \%$ train and test accuracy even for $k { = } 1 0 0$ and using only a single head. GIN $\mathrm { { X u } }$ et al., 2019), although considered as more expressive than other GNNs, cannot perfectly fit the training data (with a model size of $d = 1 2 8$ ) starting from $k { = } 2 0$ . + +Table 8: Average Hits $@ 5 0$ (Table 8a) and mean reciprocal rank (MRR) (Table 8b) in link-prediction benchmarks from OGB (10 runs±std). The best result among GAT and GATv2 is marked in bold; the best result among all GNNs is marked in bold and underline. $\dagger$ was reported by Hu et al. (2020). + +
ogbl-citation2
80.04±0.25
80.44±0.10
79.84±0.19
75.95±1.3180.33±0.13
80.14±0.71
+ +(b) + +# F.3 QM9 + +Standard deviation for the QM9 results of Section 4.5 are presented in Table 9. + +# F.4 PUBMED CITATION NETWORK + +We tuned the following parameters for both GAT and GATv2: number of layers $\in \{ 0 , 1 , 2 \}$ , hidden size $\in \{ 8 , 1 6 , 3 2 \}$ , number of heads $\in \{ 1 , 4 , 8 \}$ , dropout $\in \{ 0 . 4 , 0 . 6 , 0 . 8 \}$ , bias $\in \{ T r u e , F a l s e \}$ , share weights $\in \{ T r u e , F a l s e \}$ , use residual $\dot { \in } \{ T r u e , F a l \dot { s } e \}$ . Table 10 shows the test accuracy (100 runs±stdev) using the best hyperparameters found for each model. + +It is important to note that PubMed has only 60 training nodes, which hinders expressive models such as GATv2 from exploiting their approximation and generalization advantages. Still, GATv2 is more accurate than GAT even in this small dataset. In Table 15, we show that this difference is statistically significant (p-value $< 0 . 0 0 0 1$ ). + +# G ADDITIONAL COMPARISON WITH TRANSFORMER-STYLE ATTENTION(DPGAT) + +The main goal of our paper is to highlight a severe theoretical limitation of the highly popular GAT architecture, and propose a minimal fix. + +
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.25 18.71±23.369.05±1.21 5.62±0.817.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h10.80±2.183.11±0.14
GAT8h28.10±16.45 7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.365.37±0.26 4.11±0.271.48±0.87+134.1% +0%
GATv21h22.94±17.34
10.19±0.6322.56±17.4615.04±4.585.23±0.362.46±0.65+91.6%
GATv28h6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
+ +Table 9: Average error rates (lower is better), 5 runs $\pm$ standard deviation for each property, on the QM9 dataset. The best result among GAT and GATv2 is marked in bold; the globally best result among all GNNs is marked in bold and underline. $\dagger$ was previously tuned and reported by Brockschmidt (2020). + +Table 10: Accuracy (100 runs±stdev) on Pubmed. GATv2 is more accurate than GAT. + +
ModelAccuracy
GAT78.1±0.59
GATv278.5±0.38
+ +We perform additional empirical comparison to DPGAT, which follows Luong et al. (2015) and the dot-product attention of the Transformer (Vaswani et al., 2017). We define DPGAT as: + +$$ +e \left( \pmb { h } _ { i } , \pmb { h } _ { j } \right) = \left( \left( \pmb { h } _ { i } ^ { \top } \pmb { Q } \right) \cdot \left( \pmb { h } _ { j } ^ { \top } \pmb { K } \right) ^ { \top } \right) / \sqrt { d _ { k } } +$$ + +Variants of DPGAT were used in prior work (Gao and Ji, 2019; Dwivedi and Bresson, 2020; Rong et al., 2020a; Velickovi ˇ c et al., 2020; Kim and Oh, 2021), and we consider it here for the conceptual ´ and empirical comparison with GAT. + +Despite its popularity, DPGAT is strictly weaker than GATv2. DPGAT provably performs dynamic attention for any set of node representations only if they are linearly independent (see Theorem 3 and its proof in Appendix G.1). Otherwise, there are examples of node representations that are linearly dependent and mappings $\varphi$ , for which dynamic attention does not hold (Appendix G.2). This constraint is not harmful when violated in practice, because every node has only a small set of neighbors, rather than all possible nodes in the graph; further, some nodes possibly never need to be “selected” in practice. + +G.1 PROOF THAT DPGAT PERFORMS DYNAMIC ATTENTION FOR LINEARLY INDEPENDENT NODE REPRESENTATIONS + +Theorem 3. A DPGAT layer computes dynamic attention for any set of node representations $\mathbb { K } = \mathbb { Q } = \{ h _ { 1 } , . . . , h _ { n } \}$ that are linearly independent. + +Proof. Let $\mathcal { G } = ( \nu , \mathcal { E } )$ be a graph modeled by a DPGAT layer, having linearly independent node representations $\{ h _ { 1 } , . . . , h _ { n } \}$ . Let $\varphi : [ n ] [ n ]$ be any node mapping $[ { \bar { n } } ] [ n ]$ . + +We denote the $\mathrm { i } ^ { t h }$ row of a matrix $M$ as $M _ { i }$ + +We define a matrix $_ { P }$ as: + +$$ +P _ { i , j } = { \left\{ \begin{array} { l l } { 1 } & { j = \varphi ( i ) } \\ { 0 } & { { \mathrm { o t h e r w i s e } } } \end{array} \right. } +$$ + +Let $\ b { X } \in \mathbb { R } ^ { n } \times \mathbb { R } ^ { d }$ be the matrix holding the graph’s node representations as its rows: + +$$ +\begin{array} { r } { \mathbf { \boldsymbol { X } } = \left[ \begin{array} { l l l } { - } & { h _ { 1 } } & { - } \\ { - } & { h _ { 2 } } & { - } \\ & { \vdots } \\ { - } & { h _ { n } } & { - } \end{array} \right] } \end{array} +$$ + +Since the rows of $\boldsymbol { X }$ are linearly independent, it necessarily holds that $d \geq n$ + +Next, we find weight matrices $Q \in \mathbb { R } ^ { d } \times \mathbb { R } ^ { d }$ and $K \in \mathbb { R } ^ { d } \times \mathbb { R } ^ { d }$ such that: + +$$ +( X Q ) \cdot ( X K ) ^ { \top } = P +$$ + +To satisfy Equation (16), we choose $Q$ and $\kappa$ such that $\mathbf { \Delta } X Q = U$ and $X K = P ^ { \top } U$ where $U$ is an orthonormal matrix $( \pmb { U } \cdot \pmb { U } ^ { \top } = \pmb { U } ^ { \top } \cdot \pmb { U } = I )$ . + +We can obtain $U$ using the singular value decomposition (SVD) of $\boldsymbol { X }$ : + +$$ +\pmb { X } = \pmb { U } \pmb { \Sigma } \pmb { V } ^ { \top } +$$ + +Since $\ b { \Sigma } \in \mathbb { R } ^ { n } \times \mathbb { R } ^ { n }$ and $\boldsymbol { X }$ has a full rank, $\pmb { \Sigma }$ is invertible, and thus: + +$$ +\pmb { X } \pmb { V } \pmb { \Sigma } ^ { - 1 } = \pmb { U } +$$ + +Now, we define $Q$ as follows: + +$$ +Q = V \Sigma ^ { - 1 } +$$ + +Note that $\mathbf { \Delta } X Q = U$ , as desired. + +To find $\kappa$ that satisfies $X K = P ^ { \top } U$ , we use Equation (17) and require: + +$$ +U \Sigma V ^ { \top } K = P ^ { \top } U +$$ + +and thus: + +$$ +\pmb { K } = \pmb { V } \pmb { \Sigma } ^ { - 1 } \pmb { U } ^ { T } \pmb { P } ^ { \top } \pmb { U } +$$ + +We define: + +$$ +z \left( h _ { i } , h _ { j } \right) = e \left( h _ { i } , h _ { j } \right) \cdot \sqrt { d _ { k } } +$$ + +Where $e$ is the attention score function of DPGAT (Equation (13)). + +Now, for a query $i$ and a key $j$ , and the corresponding representations $h _ { i } , h _ { j }$ + +$$ +\begin{array} { r } { z \left( \boldsymbol { h } _ { i } , \boldsymbol { h } _ { j } \right) = \left( \boldsymbol { h } _ { i } ^ { \top } \boldsymbol { Q } \right) \cdot \left( \boldsymbol { h } _ { j } ^ { \top } \boldsymbol { K } \right) ^ { \top } } \\ { = \left( \boldsymbol { X } _ { i } \boldsymbol { Q } \right) \cdot \left( \boldsymbol { X } _ { j } \boldsymbol { K } \right) ^ { \top } } \end{array} +$$ + +Since $X _ { i } Q = ( X Q ) _ { i }$ and $X _ { j } K = \left( X K \right) _ { j }$ , we get + +$$ +z \left( \pmb { h } _ { i } , \pmb { h } _ { j } \right) = \left( \pmb { X } \pmb { Q } \right) _ { i } \cdot \left( \left( \pmb { X } \pmb { K } \right) _ { j } \right) ^ { \top } = P _ { i , j } +$$ + +Therefore: + +$$ +z \left( \pmb { { h } } _ { i } , \pmb { { h } } _ { j } \right) = \left. \begin{array} { l l } { 1 } & { j = \varphi ( i ) } \\ { 0 } & { \mathrm { o t h e r w i s e } } \end{array} \right. +$$ + +And thus: + +$$ +e ( \pmb { h } _ { i } , \pmb { h } _ { j } ) = \left\{ \begin{array} { l l } { 1 / \sqrt { d _ { k } } } & { j = \varphi ( i ) } \\ { 0 } & { o t h e r w i s e } \end{array} \right. +$$ + +To conclude, for every selected query $i$ and any key $\mathscr { j } _ { \neq \varphi ( i ) }$ + +$$ +e \left( \pmb { h } _ { i } , \pmb { h } _ { \varphi ( i ) } \right) > e \left( \pmb { h } _ { i } , \pmb { h } _ { j } \right) +$$ + +and due to the increasing monotonicity of softmax: + +$$ +\alpha _ { i , \varphi ( i ) } > \alpha _ { i , j } +$$ + +Hence, a DPGAT layer computes dynamic attention. + +In the case that $d > n$ , we apply SVD to the full-rank matrix $X X ^ { \top } \in \mathbb { R } ^ { n \times n }$ , and follow the same steps to construct $Q$ and $\kappa$ . + +In the case that $Q \in \mathbb { R } ^ { d } \times \mathbb { R } ^ { d _ { k } }$ and $K \in \mathbb { R } ^ { d } \times \mathbb { R } ^ { d _ { k } }$ and $d _ { k } > d$ , we can use the same $Q$ and $\kappa$ (Equations (19) and (21)) padded with zeros. We define the $Q ^ { \prime } \in \mathbb { R } ^ { d } \times \mathbb { R } ^ { d _ { k e y } }$ and $K ^ { \prime } \in \mathbb { R } ^ { d } \times \mathbb { R } ^ { d _ { k e y } }$ as follows: + +$$ +\begin{array} { r l } & { Q _ { i , j } ^ { \prime } = \left\{ \begin{array} { l l } { Q _ { i , j } } & { j \leq d } \\ { 0 } & { \mathrm { o t h e r w i s e } } \end{array} \right. } \\ & { K _ { i , j } ^ { \prime } = \left\{ \begin{array} { l l } { K _ { i , j } } & { j \leq d } \\ { 0 } & { \mathrm { o t h e r w i s e } } \end{array} \right. } \end{array} +$$ + +# G.2 DPGAT IS STRICTLY WEAKER THAN GATV2 + +There are examples of node representations that are linearly dependent and mappings $\varphi$ , for which dynamic attention does not hold. First, we show a simple 2-dimensional example, and then we show the general case of such examples. + +$$ +\prod _ { \substack { h _ { 0 } = \hat { x } } } ^ { y } h _ { 2 } = \hat { x } + 2 \hat { y } +$$ + +Figure 6: An example for node representations that are linearly dependent, for which DPGAT cannot compute dynamic attention, because no query vector $q \in \mathbb { R } ^ { 2 }$ can “select” $\pmb { h } _ { 1 }$ . + +Consider the following linearly dependent set of vectors $\mathbb { K } = \mathbb { Q }$ (Figure 6): + +$$ +\begin{array} { l } { { \pmb { h } } _ { 0 } = { \pmb { \hat { x } } } } \\ { { \pmb { h } } _ { 1 } = { \pmb { \hat { x } } } + { \pmb { \hat { y } } } } \\ { { \pmb { h } } _ { 2 } = { \pmb { \hat { x } } } + 2 { \pmb { \hat { y } } } } \end{array} +$$ + +where $\hat { \pmb x }$ and $\hat { \pmb { y } }$ are the cartesian unit vectors. We define $\beta \in \{ 0 , 1 , 2 \}$ to express $\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \}$ using the same expression: + +$$ +\begin{array} { r } { \pmb { h } _ { \beta } = \hat { \pmb { x } } + \beta \hat { \pmb y } } \end{array} +$$ + +Let $\pmb q \in \mathbb { Q }$ be any query vector. For brevity, we define the unscaled dot-product attention score as $s$ + +$$ +s \left( \mathbf { q } , \boldsymbol { h } _ { \beta } \right) = e \left( \mathbf { q } , \boldsymbol { h } _ { \beta } \right) \cdot \sqrt { d _ { k } } +$$ + +Where $e$ is the attention score function of DPGAT (Equation (13)). The (unscaled) attention score between $\pmb q$ and $\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \}$ is: + +$$ +\begin{array} { r l } & { s \left( \pmb { q } , \pmb { h } _ { \beta } \right) = \left( \pmb { q } ^ { \top } \pmb { Q } \right) \left( \pmb { h } _ { \beta } ^ { \top } \pmb { K } \right) ^ { \top } } \\ & { \quad \quad \quad = \left( \pmb { q } ^ { \top } \pmb { Q } \right) \left( \left( \hat { \pmb { x } } + \beta \pmb { \hat { y } } \right) ^ { \top } \pmb { K } \right) ^ { \top } } \\ & { \quad \quad \quad = \left( \pmb { q } ^ { \top } \pmb { Q } \right) \left( \hat { \pmb { x } } ^ { \top } \pmb { K } + \beta \pmb { \hat { y } } ^ { \top } \pmb { K } \right) ^ { \top } } \\ & { \quad \quad \quad = \left( \pmb { q } ^ { \top } \pmb { Q } \right) \left( \hat { \pmb { x } } ^ { \top } \pmb { K } \right) ^ { \top } + \beta \left( \pmb { q } ^ { \top } \pmb { Q } \right) \left( \pmb { \hat { y } } ^ { \top } \pmb { K } \right) ^ { \top } } \end{array} +$$ + +The first term $\left( { { \pmb q } ^ { \top } } { \pmb Q } \right) \left( \hat { { \pmb x } } ^ { \top } { \pmb K } \right) ^ { \top }$ is unconditioned on $\beta$ , and thus shared for every $h _ { \beta }$ . Let us focus on the second term $\beta \left( \pmb { q } ^ { \top } \pmb { Q } \right) \left( \hat { \pmb { y } } ^ { \top } \pmb { K } \right) ^ { \top }$ . I f $\left( { { \pmb q } ^ { \top } } { \pmb Q } \right) \left( \hat { \pmb y } ^ { \top } { \pmb K } \right) ^ { \top } > 0$ , then: + +$$ +e \left( \pmb { q } , \pmb { h } _ { 2 } \right) > e \left( \pmb { q } , \pmb { h } _ { 1 } \right) +$$ + +Otherwise, if $\left( { { { q } ^ { \top } } { { Q } } } \right) \left( { \hat { { { y } } } ^ { \top } } K \right) ^ { \top } \le 0$ + +$$ +e \left( \pmb { q } , \pmb { h } _ { 0 } \right) \geq e \left( \pmb { q } , \pmb { h } _ { 1 } \right) +$$ + +Thus, for any query $\pmb q$ , the key $\pmb { h } _ { 1 }$ can never get the highest score, and thus cannot be “selected”. That is, the key $\pmb { h } _ { 1 }$ cannot satisfy that $e \left( \pmb q , \pmb h _ { 1 } \right)$ is strictly greater than any other key. + +In the general case, let $\boldsymbol { h } _ { 0 } , \boldsymbol { h } _ { 1 } \in \mathbb { R } ^ { d }$ be some non-zero vectors , and $\lambda$ is some scalar such that $0 < \lambda < 1$ . + +Consider the following linearly dependent set of vectors: + +$$ +\mathbb { K } = \mathbb { Q } = \{ \beta \pmb { h } _ { 1 } + \left( 1 - \beta \right) \pmb { h } _ { 0 } \ \lvert \ \beta \in \{ 0 , \lambda , 1 \} \} +$$ + +For any query $\pmb q \in \mathbb { Q }$ and $\beta \in \{ 0 , \lambda , 1 \}$ we define: + +$$ +s \left( \mathbf { q } , \beta \right) = e \left( \mathbf { q } , \left( \beta \pmb { h } _ { 1 } + \left( 1 - \beta \right) \pmb { h } _ { 0 } \right) \right) \cdot \sqrt { d _ { k } } +$$ + +Where $e$ is the attention score function of DPGAT (Equation (13)). + +Therefore: + +$$ +\begin{array} { r l } & { s \left( \boldsymbol { q } , \beta \right) = \left( \boldsymbol { q } ^ { \top } \boldsymbol { Q } \right) \left( \left( \beta h _ { 1 } + \left( 1 - \beta \right) h _ { 0 } \right) ^ { \top } \boldsymbol { K } \right) ^ { \top } } \\ & { \quad \quad = \left( \boldsymbol { q } ^ { \top } \boldsymbol { Q } \right) \left( \beta h _ { 1 } ^ { \top } \boldsymbol { K } + \left( 1 - \beta \right) h _ { 0 } ^ { \top } \boldsymbol { K } \right) ^ { \top } } \\ & { \quad \quad = \left( \boldsymbol { q } ^ { \top } \boldsymbol { Q } \right) \left( \beta h _ { 1 } ^ { \top } \boldsymbol { K } + h _ { 0 } ^ { \top } \boldsymbol { K } - \beta h _ { 0 } ^ { \top } \boldsymbol { K } \right) ^ { \top } } \\ & { \quad \quad = \left( \boldsymbol { q } ^ { \top } \boldsymbol { Q } \right) \left( \beta \left( h _ { 1 } ^ { \top } \boldsymbol { K } - h _ { 0 } ^ { \top } \boldsymbol { K } \right) + h _ { 0 } ^ { \top } \boldsymbol { K } \right) ^ { \top } } \\ & { \quad \quad = \beta \left( \boldsymbol { q } ^ { \top } \boldsymbol { Q } \right) \left( h _ { 1 } ^ { \top } \boldsymbol { K } - h _ { 0 } ^ { \top } \boldsymbol { K } \right) ^ { \top } + \left( \boldsymbol { q } ^ { \top } \boldsymbol { Q } \right) \left( h _ { 0 } ^ { \top } \boldsymbol { K } \right) ^ { \top } } \end{array} +$$ + +If $\left( { { q } ^ { \top } } Q \right) \left( h _ { 1 } ^ { \top } K - h _ { 0 } ^ { \top } K \right) ^ { \top } > 0 .$ + +$$ +e \left( \pmb q , \pmb h _ { 1 } \right) > e \left( \pmb q , \pmb h _ { \lambda } \right) +$$ + +Otherwise, if $\left( { \pmb q } ^ { \top } { \pmb Q } \right) \left( { \pmb h } _ { 1 } ^ { \top } { \pmb K } - { \pmb h } _ { 0 } ^ { \top } { \pmb K } \right) ^ { \top } \le 0 ,$ + +$$ +e \left( \pmb { q } , \pmb { h } _ { 0 } \right) \geq e \left( \pmb { q } , \pmb { h } _ { \lambda } \right) +$$ + +Thus, for any query $\pmb q$ , the key $h _ { \lambda }$ cannot be selected. That is, the key $\mathbf { \delta } _ { h _ { \lambda } }$ cannot satisfy that $\mathbf { \nabla } _ { \boldsymbol { \mathcal { C } } } \left( \mathbf { \nabla } _ { \mathbf { \boldsymbol { q } } } , \boldsymbol { h } _ { \lambda } \right)$ is strictly greater than any other key. Therefore, there are mappings $\varphi$ , for which dynamic attention does not hold. + +While we prove that GATv2 computes dynamic attention (Appendix B) for any set of node representations $\mathbb { K } = \mathbb { Q }$ , there are sets of node representations and mappings $\varphi$ for which dynamic attention does not hold for DPGAT. Thus, DPGAT is strictly weaker than GATv2. + +# G.3 EMPIRICAL EVALUATION + +Here we repeat the experiments of Section 4 with DPGAT. We remind that DPGAT is strictly weaker than our proposed GATv2 (see a proof in Appendix G.1). + +# H STATISTICAL SIGNIFICANCE + +Here we report the statistical significance of the strongest GATv2 and GAT models of the experiments reported in Section 4. + +![](images/0f010beab366169ea6709265d4f0cb66efdab3acea2d3044a4c540073bc64969.jpg) +Figure 7: Test accuracy compared to the noise ratio: GATv2 and DPGAT are more robust to structural noise compared to GAT. Each point is an average of 10 runs, error bars show standard deviation. + +Table 11: Accuracy (5 runs±stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both test sets, using GAT’s hyperparameters. $\dagger$ – previously reported by Brockschmidt (2020). + +
ModelSeenProjUnseenProj
No- AttentionGCN+87.2±1.581.4±2.3
GIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
DPGAT88.0±0.881.5±1.2
GATv288.0±1.182.8±1.7
+ +(a) + +Table 12: Average accuracy (Table 12a) and ROC-AUC (Table 12b) in node-prediction datasets (10 runs±std). In all datasets, GATv2 outperforms GAT. $\dagger -$ previously reported by Hu et al. (2020). + +
ModelAttn. Headsogbn-arxivogbn-productsogbn-mag
GCN+071.74±0.2978.97±0.3330.43±0.25
GraphSAGE†071.49±0.2778.70±0.3631.53±0.15
GAT171.59±0.3879.04±1.5432.20±1.46
871.54±0.3077.23±2.3731.75±1.60
DPGAT171.52±0.1776.49±0.7832.77±0.80
871.48±0.2673.53±0.4727.74±9.97
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.44
871.87±0.2578.46±2.4532.52±0.39
+ +(b) + +
ogbn-proteins
72.51±0.3577.68±0.20
70.77±5.7978.63±1.62
63.47±2.7972.88±0.59
77.23±3.3279.52±0.55
+ +Table 13: Average error rates (lower is better), 5 runs $\pm$ standard deviation for each property, on the QM9 dataset. The best result among GAT, GATv2 and DPGAT is marked in bold; the globally best result among all GNNs is marked in bold and underline. $\dagger$ was previously tuned and reported by Brockschmidt (2020). + +
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
DPGAT8h2.63±0.094.37±0.131.44±0.071.40±0.032.10±0.0732.59±34.7711.66±1.00
DPGAT1h3.20±0.178.35±0.781.71±0.032.17±0.142.88±0.1225.21±2.8665.79±39.84
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13 ±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.259.05±1.217.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h28.10±16.4518.71±23.365.62±0.8110.80±2.185.37±0.26
GAT8ht7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.364.11±0.273.11±0.14 1.48±0.87+134.1% +0%
DPGAT1h12.93±1.7013.83±2.556.37±0.28
DPGAT8h6.95±0.3213.32±2.39 7.09±0.5914.42±1.95 7.30±0.666.52±0.613.76±0.213.28±1.16 1.18±0.33+77.9% -9.7%
GATv21h5.23±0.36
GATv28h10.19±0.6322.56±17.4615.04±4.5822.94±17.342.46±0.65+91.6%
6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
+ +![](images/a9109830d2a6f78861878e365b1f89ef70a5d28fb8341f7508b1337e0e7390a2.jpg) +Figure 8: Test accuracy and statistical significance compared to the noise ratio: GATv2 is more robust to structural noise compared to GAT. Each point is an average of 10 runs, error bars show standard deviation. + +Table 14: Accuracy (5 runs $\pm$ stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both test sets, using GAT’s hyperparameters. $\dagger$ – previously reported by Brockschmidt (2020). + +
ModelSeenProjUnseenProj
GAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
p-value0.0480.049
+ +Table 15: Accuracy (100 runs±stdev) on Pubmed. GATv2 is more accurate than GAT. + +
Model Accuracy
GAT 78.1±0.59
GATv2 78.5±0.38
p-value <0.0001
+ +Table 16: Average accuracy (Table 16a) and ROC-AUC (Table 16b) in node-prediction datasets (30 runs±std). We report on the best GAT / GATv2 from Table 2. + +
(a)
Modelogbn-arxivogbn-productsogbn-mag
GAT71.65±0.3879.04±1.5432.36±1.10
GATv271.93±0.3580.63±0.7033.01±0.41
p-value0.0022<0.00010.0018
+ +(b) + +
ogbn-proteins
78.29 ±1.59
78.96±1.19 0.0349
+ +Table 17: Average Hits $\ @ 5 0$ (Table 17a) and mean reciprocal rank (MRR) (Table 17b) in linkprediction benchmarks from OGB (30 runs±std). We report on the best GAT / GATv2 from Table 8. + +
(a)
Modelogbl-collab w/o val edges w/ val edges
GAT42.24±2.2646.02±4.09
GATv243.82±2.2449.06±2.50
p-value0.00430.0005
+ +
(b)
ogbl-citation2
79.91±0.13
80.20±0.62
0.0075
+ +Table 18: Average error rates (lower is better), 20 runs $\pm$ standard deviation for each property, on the QM9 dataset. We report on GAT and GATv2 with 8 attention heads. + +
Predicted Property
Model1234567
GAT2.74±0.084.73±0.401.47±0.061.53±0.062.44±0.6055.21±42.3325.36±31.42
GATv22.67±0.084.28±0.231.43±0.051.51±0.072.21±0.0816.64± 1.1713.61±1.68
p-value0.0043<0.00010.01380.16910.04870.00010.0516
Predicted Property
Model8910111213
GAT GATv27.36±0.87 6.13±0.596.79±0.86 6.33±0.827.36±0.936.69±0.864.10±0.291.51±0.84
6.37±0.865.95±0.623.66±0.291.09±0.85
p-value<0.00010.04580.00060.0017<0.00010.0621
+ +# I COMPLEXITY ANALYSIS + +We repeat the definitions of GAT, GATv2 and DPGAT: + +GAT (Velickovi ˇ c et al., 2018): ´ + +GATv2 (our fixed version): + +DPGAT (Vaswani et al., 2017): + +$$ +\begin{array} { r l } & { e ( \boldsymbol { h } _ { i } , \boldsymbol { h } _ { j } ) = \mathrm { L e a k y R e L U } ( \boldsymbol { a } ^ { \top } \cdot [ \boldsymbol { W } \boldsymbol { h } _ { i } | | \boldsymbol { W } \boldsymbol { h } _ { j } | ) } \\ & { e ( \boldsymbol { h } _ { i } , \boldsymbol { h } _ { j } ) = \boldsymbol { a } ^ { \top } \mathrm { L e a k y R e L U } ( \boldsymbol { W } \cdot [ \boldsymbol { h } _ { i } | | \boldsymbol { h } _ { j } ] ) } \\ & { e ( \boldsymbol { h } _ { i } , \boldsymbol { h } _ { j } ) = ( ( \boldsymbol { h } _ { i } ^ { \top } \boldsymbol { Q } ) \cdot ( \boldsymbol { h } _ { j } ^ { \top } \boldsymbol { K } ) ^ { \top } ) / \sqrt { d ^ { \prime } } } \end{array} +$$ + +# I.1 TIME COMPLEXITY + +GAT As noted by Velickovi ˇ c et al. (2018), the time complexity of a single GAT head may be ´ expressed as $\mathcal { O } \left( | \mathcal { V } | d d ^ { \prime } + | \mathcal { E } | d ^ { \prime } \right)$ . Because of GAT’s static attention, this computation can be further optimized, by merging the linear layer $\mathbf { \delta } _ { \mathbf { a } _ { 1 } }$ with $W$ , merging $\mathbf { a } _ { 2 }$ with $W$ , and only then compute $a _ { \{ 1 , 2 \} } ^ { \top } W h _ { i }$ for every $i \in \mathcal V$ . + +GATv2 require the same computational cost as GAT’s declared complexity: $\mathcal { O } \left( | \mathcal { V } | d d ^ { \prime } + | \mathcal { E } | d ^ { \prime } \right)$ : we denote $W = [ W _ { 1 } \Vert W _ { 2 } ]$ , where $W _ { 1 } \in \mathbb { R } ^ { d ^ { \prime } \times d }$ and $W _ { 2 } ^ { d ^ { \prime } \times d }$ contain the left half and right half of the columns of $W$ , respectively. We can first compute $W _ { 1 } h _ { i }$ and $W _ { 2 } h _ { j }$ for every $i , j \in \mathcal { V }$ . This takes $\mathcal { O } \left( | \mathcal { V } | d d ^ { \prime } \right)$ . + +Then, for every edge $( j , i )$ , we compute LeakyReLU $\left( W \cdot \left[ \boldsymbol { h } _ { i } \right| \left| \boldsymbol { h } _ { j } \right] \right)$ using the precomputed $W _ { 1 } h _ { i }$ and $W _ { 2 } h _ { j }$ , since $W \cdot [ h _ { i } \| h _ { j } ] = W _ { 1 } h _ { i } + W _ { 2 } h _ { j }$ . This takes $\mathcal { O } \left( \vert \mathcal { E } \vert d ^ { \prime } \right)$ . + +Finally, computing the results of the linear layer $^ { a }$ takes additional $\mathcal { O } \left( | \mathcal { E } | d ^ { \prime } \right)$ time, and overall $\mathcal { O } \left( | \mathcal { V } | d d ^ { \prime } + \bar { | } \mathcal { E } | d ^ { \prime } \right)$ . + +DPGAT also takes the same time. We can first compute $\pmb { h } _ { i } ^ { \top } \pmb { Q }$ and $\pmb { h } _ { j } ^ { \top } \pmb { K }$ for every $i , j \in \mathcal { V }$ . This takes $\mathcal { O } \left( | \mathcal { V } | d d ^ { \prime } \right)$ . Computing the dot-product $\left( \pmb { h } _ { i } ^ { \top } \pmb { Q } \right) \left( \pmb { h } _ { j } ^ { \top } \pmb { K } \right) ^ { \top }$ for every edge $( j , i )$ takes additional $\mathcal { O } \left( | \mathcal { E } | d ^ { \prime } \right)$ time, and overall $\mathcal { O } \left( | \mathcal { V } | d d ^ { \prime } + | \mathcal { E } | d ^ { \prime } \right)$ . + +# I.2 PARAMETRIC COMPLEXITY + +
GATGATv2DPGAT
Official2d'+dd'd'+ 2dd'2ddk +dd'
In our experiments2d'+dd'd'+dd'2dd'
+ +Table 19: Number of parameters for each GNN type, in a single layer and a single attention head. + +All parametric costs are summarized in Table 19. All following calculations refer to a single layer having a single attention head, omitting bias vectors. + +GAT has learned vector and a matrix: $\pmb { a } \in \mathbb { R } ^ { 2 d ^ { \prime } }$ and $W \in \mathbb { R } ^ { d ^ { \prime } \times d }$ , thus overall $2 d ^ { \prime } + d d ^ { \prime }$ learned parameters. + +GATv2 has a matrix that is twice larger: $W \in \mathbb { R } ^ { d ^ { \prime } \times 2 d }$ , because it is applied on the concatenation $[ h _ { i } | | h _ { j } ]$ . Thus, the overall number of learned parameters is $d ^ { \prime } + 2 d d ^ { \prime }$ . However in our experiments, to rule out the increased number of parameters over GAT as the source of empirical difference, we constrained $W = [ W ^ { \prime } | | W ^ { \prime } ]$ , and thus the number of parameters were $d ^ { \prime } + d d ^ { \prime }$ . + +DPGAT has $Q$ and $\kappa$ matrices of sizes $d d _ { k }$ each, and additional $d d ^ { \prime }$ parameters in the value matrix $V$ , thus $2 d d _ { k } + d d ^ { \prime }$ parameters overall. However in our experiments, we constrained $Q = K$ and set $d _ { k } = d ^ { \prime }$ , and thus the number of parameters is only $2 d d ^ { \prime }$ . + +# REFERENCES + +Miltiadis Allamanis, Marc Brockschmidt, and Mahmoud Khademi. Learning to represent programs with graphs. In International Conference on Learning Representations, 2018. URL https://openreview.net/ forum?id $\equiv$ BJOFETxR-. + +Uri Alon and Eran Yahav. On the bottleneck of graph neural networks and its practical implications. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id $=$ i80OPhOCVH2. + +James Atwood and Don Towsley. Diffusion-convolutional neural networks. In Advances in neural information processing systems, pages 1993–2001, 2016. + +Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. CoRR, abs/1409.0473, 2014. URL http://arxiv.org/abs/1409.0473. + +Peter Battaglia, Razvan Pascanu, Matthew Lai, Danilo Jimenez Rezende, and Koray kavukcuoglu. Interaction networks for learning about objects, relations and physics. In Proceedings of the 30th International Conference on Neural Information Processing Systems, pages 4509–4517, 2016. + +Marc Brockschmidt. Gnn-film: Graph neural networks with feature-wise linear modulation. Proceedings of the 36th International Conference on Machine Learning, ICML, 2020. URL https://github.com/ microsoft/tf-gnn-samples. + +Michael M Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst. Geometric deep learning: going beyond euclidean data. IEEE Signal Processing Magazine, 34(4):18–42, 2017. + +Michael M. Bronstein, Joan Bruna, Taco Cohen, and Petar Velickovi ˇ c. Geometric deep learning: Grids, groups, ´ graphs, geodesics, and gauges, 2021. + +Dan Busbridge, Dane Sherburn, Pietro Cavallo, and Nils Y Hammerla. Relational graph attention networks. arXiv preprint arXiv:1904.05811, 2019. + +George Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of control, signals and systems, 2(4):303–314, 1989. + +Misha Denil, Sergio Gómez Colmenarejo, Serkan Cabi, David Saxton, and Nando de Freitas. Programmable agents. arXiv preprint arXiv:1706.06383, 2017. + +Yan Duan, Marcin Andrychowicz, Bradly Stadie, Jonathan Ho, Jonas Schneider, Ilya Sutskever, Pieter Abbeel, and Wojciech Zaremba. One-shot imitation learning. In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 1087–1098, 2017. + +David K Duvenaud, Dougal Maclaurin, Jorge Iparraguirre, Rafael Bombarell, Timothy Hirzel, Alán AspuruGuzik, and Ryan P Adams. Convolutional networks on graphs for learning molecular fingerprints. In Advances in neural information processing systems, pages 2224–2232, 2015. + +Vijay Prakash Dwivedi and Xavier Bresson. A generalization of transformer networks to graphs. arXiv preprint arXiv:2012.09699, 2020. + +Vijay Prakash Dwivedi, Chaitanya K Joshi, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. Benchmarking graph neural networks. arXiv preprint arXiv:2003.00982, 2020. + +Matthias Fey and Jan E. Lenssen. Fast graph representation learning with PyTorch Geometric. In ICLR Workshop on Representation Learning on Graphs and Manifolds, 2019. + +Ken-Ichi Funahashi. On the approximate realization of continuous mappings by neural networks. Neural networks, 2(3):183–192, 1989. + +Hongyang Gao and Shuiwang Ji. Graph representation learning via hard and channel-wise attention networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 741–749, 2019. + +Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural message passing for quantum chemistry. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 1263–1272. JMLR. org, 2017. + +Aleksa Gordic. pytorch-gat. ´ https://github.com/gordicaleksa/pytorch-GAT, 2020. + +Marco Gori, Gabriele Monfardini, and Franco Scarselli. A new model for learning in graph domains. In Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., volume 2, pages 729– 734. IEEE, 2005. + +Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. In Advances in neural information processing systems, pages 1024–1034, 2017. + +Kurt Hornik. Approximation capabilities of multilayer feedforward networks. Neural networks, 4(2):251–257, 1991. + +Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators. Neural networks, 2(5):359–366, 1989. + +Yedid Hoshen. Vain: attentional multi-agent predictive modeling. In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 2698–2708, 2017. + +Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open graph benchmark: Datasets for machine learning on graphs. arXiv preprint arXiv:2005.00687, 2020. + +Binxuan Huang and Kathleen M Carley. Syntax-aware aspect level sentiment classification with graph attention networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5472–5480, 2019. + +Qian Huang, Horace He, Abhay Singh, Ser-Nam Lim, and Austin Benson. Combining label propagation and simple models out-performs graph neural networks. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=8E1-f3VhX1o. + +Chaitanya Joshi. Transformers are graph neural networks. The Gradient, 2020. + +Dongkwan Kim and Alice Oh. How to find your friendly neighborhood: Graph attention design with self-supervision. In International Conference on Learning Representations, 2021. URL https:// openreview.net/forum?id $=$ Wi5KUNlqWty. + +Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In ICLR, 2017. + +Vineet Kosaraju, Amir Sadeghian, Roberto Martín-Martín, Ian Reid, Hamid Rezatofighi, and Silvio Savarese. Social-bigat: Multimodal trajectory forecasting using bicycle-gan and graph attention networks. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. URL https://proceedings. neurips.cc/paper/2019/file/d09bf41544a3365a46c9077ebb5e35c3-Paper.pdf. + +John Boaz Lee, Ryan A Rossi, Sungchul Kim, Nesreen K Ahmed, and Eunyee Koh. Attention models in graphs: A survey. arXiv preprint arXiv:1807.07984, 2018. + +Moshe Leshno, Vladimir Ya Lin, Allan Pinkus, and Shimon Schocken. Multilayer feedforward networks with a nonpolynomial activation function can approximate any function. Neural networks, 6(6):861–867, 1993. + +Guohao Li, Matthias Muller, Ali Thabet, and Bernard Ghanem. Deepgcns: Can gcns go as deep as cnns? In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9267–9276, 2019. + +Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for semisupervised learning. In Thirty-Second AAAI Conference on Artificial Intelligence, 2018. + +Yujia Li, Daniel Tarlow, Marc Brockschmidt, and Richard Zemel. Gated graph sequence neural networks. In International Conference on Learning Representations, 2016. + +Denis Lukovnikov and Asja Fischer. Gated relational graph attention networks, 2021. URL https:// openreview.net/forum?id=v-9E8egy_i. + +Thang Luong, Hieu Pham, and Christopher D. Manning. Effective approaches to attention-based neural machine translation. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, EMNLP 2015, Lisbon, Portugal, September 17-21, 2015, pages 1412–1421, 2015. URL http: //aclweb.org/anthology/D/D15/D15-1166.pdf. + +Nianzu Ma, Sahisnu Mazumder, Hao Wang, and Bing Liu. Entity-aware dependency-based deep graph attention network for comparative preference classification. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5782–5788, 2020. + +Federico Monti, Davide Boscaini, Jonathan Masci, Emanuele Rodola, Jan Svoboda, and Michael M Bronstein. Geometric deep learning on graphs and manifolds using mixture model cnns. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5115–5124, 2017. + +Deepak Nathani, Jatin Chauhan, Charu Sharma, and Manohar Kaul. Learning attention-based embeddings for relation prediction in knowledge graphs. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4710–4723, 2019. + +Sejun Park, Chulhee Yun, Jaeho Lee, and Jinwoo Shin. Minimum width for universal approximation. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id $\equiv$ O-XJwyoIF-k. + +Allan Pinkus. Approximation theory of the mlp model. Acta Numerica 1999: Volume 8, 8:143–195, 1999. + +Jiezhong Qiu, Jian Tang, Hao Ma, Yuxiao Dong, Kuansan Wang, and Jie Tang. Deepinf: Social influence prediction with deep learning. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’18), 2018. + +Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum chemistry structures and properties of 134 kilo molecules. Scientific data, 1:140022, 2014. + +Yu Rong, Yatao Bian, Tingyang Xu, Weiyang Xie, Ying Wei, Wenbing Huang, and Junzhou Huang. Selfsupervised graph transformer on large-scale molecular data. Advances in Neural Information Processing Systems, 33, 2020a. + +Yu Rong, Wenbing Huang, Tingyang Xu, and Junzhou Huang. Dropedge: Towards deep graph convolutional networks on node classification. In International Conference on Learning Representations, 2020b. URL https://openreview.net/forum?id $=$ Hkx1qkrKPr. + +Adam Santoro, David Raposo, David GT Barrett, Mateusz Malinowski, Razvan Pascanu, Peter Battaglia, and Timothy Lillicrap. A simple neural network module for relational reasoning. In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 4974–4983, 2017. + +Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. The graph neural network model. IEEE Transactions on Neural Networks, 20(1):61–80, 2008. + +Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Galligher, and Tina Eliassi-Rad. Collective classification in network data. AI magazine, 29(3):93–93, 2008. + +Yunsheng Shi, Zhengjie Huang, Shikun Feng, and Yu Sun. Masked label prediction: Unified massage passing model for semi-supervised classification. arXiv preprint arXiv:2009.03509, 2020. + +Kiran K Thekumparampil, Chong Wang, Sewoong Oh, and Li-Jia Li. Attention-based graph neural network for semi-supervised learning. arXiv preprint arXiv:1803.03735, 2018. + +Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, pages 6000–6010, 2017. + +Petar Velickovi ˇ c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph ´ attention networks. In International Conference on Learning Representations, 2018. + +Petar Velickovi ˇ c, Lars Buesing, Matthew Overlan, Razvan Pascanu, Oriol Vinyals, and Charles Blundell. Pointer ´ graph networks. Advances in Neural Information Processing Systems, 33, 2020. + +Petar et al. Velickovi ˇ c. Graph attention networks. 2018. ´ + +Guangtao Wang, Rex Ying, Jing Huang, and Jure Leskovec. Improving graph attention networks with large margin-based constraints. arXiv preprint arXiv:1910.11945, 2019a. + +Minjie Wang, Da Zheng, Zihao Ye, Quan Gan, Mufei Li, Xiang Song, Jinjing Zhou, Chao Ma, Lingfan Yu, Yu Gai, Tianjun Xiao, Tong He, George Karypis, Jinyang Li, and Zheng Zhang. Deep graph library: A graph-centric, highly-performant package for graph neural networks. arXiv preprint arXiv:1909.01315, 2019b. + +Xiao Wang, Houye Ji, Chuan Shi, Bai Wang, Yanfang Ye, Peng Cui, and Philip S Yu. Heterogeneous graph attention network. In The World Wide Web Conference, pages 2022–2032, 2019c. + +Yangkun Wang. Bag of tricks of semi-supervised classification with graph neural networks. arXiv preprint arXiv:2103.13355, 2021. + +Gail Weiss, Yoav Goldberg, and Eran Yahav. On the practical computational power of finite precision rnns for language recognition. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 740–745, 2018. +Felix Wu, Amauri Souza, Tianyi Zhang, Christopher Fifty, Tao Yu, and Kilian Weinberger. Simplifying graph convolutional networks. In International conference on machine learning, pages 6861–6871. PMLR, 2019. +Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and S Yu Philip. A comprehensive survey on graph neural networks. IEEE Transactions on Neural Networks and Learning Systems, 2020. +Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? In International Conference on Learning Representations, 2019. URL https://openreview.net/ forum?id $\equiv$ ryGs6iA5Km. +Yiding Yang, Jiayan Qiu, Mingli Song, Dacheng Tao, and Xinchao Wang. Distilling knowledge from graph convolutional networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020. +Seongjun Yun, Minbyul Jeong, Raehyun Kim, Jaewoo Kang, and Hyunwoo J Kim. Graph transformer networks. Advances in Neural Information Processing Systems, 32:11983–11993, 2019. +Hanqing Zeng, Hongkuan Zhou, Ajitesh Srivastava, Rajgopal Kannan, and Viktor Prasanna. Graphsaint: Graph sampling based inductive learning method. arXiv preprint arXiv:1907.04931, 2019. +Jiani Zhang, Xingjian Shi, Junyuan Xie, Hao Ma, Irwin King, and Dit-Yan Yeung. Gaan: Gated attention networks for learning on large and spatiotemporal graphs. In Proceedings of the Thirty-Fourth Conference on Uncertainty in Artificial Intelligence, pages 339–349, 2018. +Kai Zhang, Yaokang Zhu, Jun Wang, and Jie Zhang. Adaptive structural fingerprints for graph attention networks. In International Conference on Learning Representations, 2020. URL https://openreview.net/ forum?id $\equiv$ BJxWx0NYPr. +Lingxiao Zhao and Leman Akoglu. Pairnorm: Tackling oversmoothing in gnns. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id $\underline { { \underline { { \mathbf { \Pi } } } } } =$ rkecl1rtwB. \ No newline at end of file diff --git a/parse/dev/F72ximsx7C1/F72ximsx7C1_content_list.json b/parse/dev/F72ximsx7C1/F72ximsx7C1_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..95282ad0a12c0c0e6ea3f5bc40727b694fc9f402 --- /dev/null +++ b/parse/dev/F72ximsx7C1/F72ximsx7C1_content_list.json @@ -0,0 +1,4121 @@ +[ + { + "type": "text", + "text": "HOW ATTENTIVE ARE GRAPH ATTENTION NETWORKS? ", + "text_level": 1, + "bbox": [ + 174, + 99, + 681, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Shaked Brody \nTechnion \nshakedbr@cs.technion.ac.il ", + "bbox": [ + 184, + 170, + 439, + 212 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Uri Alon Language Technologies Institute Carnegie Mellon University ualon@cs.cmu.edu ", + "bbox": [ + 598, + 170, + 813, + 224 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Eran Yahav ", + "text_level": 1, + "bbox": [ + 184, + 247, + 267, + 261 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "yahave@cs.technion.ac.il ", + "bbox": [ + 186, + 276, + 419, + 289 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 325, + 544, + 340 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Graph Attention Networks (GATs) are one of the most popular GNN architectures and are considered as the state-of-the-art architecture for representation learning with graphs. In GAT, every node attends to its neighbors given its own representation as the query. However, in this paper we show that GAT computes a very limited kind of attention: the ranking of the attention scores is unconditioned on the query node. We formally define this restricted kind of attention as static attention and distinguish it from a strictly more expressive dynamic attention. Because GATs use a static attention mechanism, there are simple graph problems that GAT cannot express: in a controlled problem, we show that static attention hinders GAT from even fitting the training data. To remove this limitation, we introduce a simple fix by modifying the order of operations and propose GATv2: a dynamic graph attention variant that is strictly more expressive than GAT. We perform an extensive evaluation and show that GATv2 outperforms GAT across 12 OGB and other benchmarks while we match their parametric costs. Our code is available at https://github.com/tech-srl/how_attentive_are_ gats.1 GATv2 is available as part of the PyTorch Geometric library,2 the Deep Graph Library,3 and the TensorFlow GNN library.4 ", + "bbox": [ + 233, + 358, + 766, + 593 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 622, + 336, + 638 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Graph neural networks (GNNs; Gori et al., 2005; Scarselli et al., 2008) have seen increasing popularity over the past few years (Duvenaud et al., 2015; Atwood and Towsley, 2016; Bronstein et al., 2017; Monti et al., 2017). GNNs provide a general and efficient framework to learn from graph-structured data. Thus, GNNs are easily applicable in domains where the data can be represented as a set of nodes and the prediction depends on the relationships (edges) between the nodes. Such domains include molecules, social networks, product recommendation, computer programs and more. ", + "bbox": [ + 174, + 654, + 825, + 738 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In a GNN, each node iteratively updates its state by interacting with its neighbors. GNN variants (Wu et al., 2019; Xu et al., 2019; Li et al., 2016) mostly differ in how each node aggregates and combines the representations of its neighbors with its own. Velickovi ˇ c et al. (2018) pioneered the use ´ of attention-based neighborhood aggregation, in one of the most common GNN variants – Graph Attention Network (GAT). In GAT, every node updates its representation by attending to its neighbors using its own representation as the query. This generalizes the standard averaging or max-pooling of neighbors (Kipf and Welling, 2017; Hamilton et al., 2017), by allowing every node to compute a weighted average of its neighbors, and (softly) select its most relevant neighbors. The work of ", + "bbox": [ + 174, + 746, + 825, + 857 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/3fd2d6f1a250485c1844b504947dfd994af31b1f76fbbbea17a0ae62e23a556a.jpg", + "image_caption": [ + "Figure 1: In a complete bipartite graph of “query nodes” $\\{ q 0 , . . . , q 9 \\}$ and “key nodes” $\\{ k 0 , . . . , k 9 \\}$ : standard GAT (Figure 1a) computes static attention – the ranking of attention coefficients is global for all nodes in the graph, and is unconditioned on the query node. For example, all queries $_ { g 0 }$ to $q 9 ,$ ) attend mostly to the 8th key $( k 8 )$ . In contrast, GATv2 (Figure 1b) can actually compute dynamic attention, where every query has a different ranking of attention coefficients of the keys. " + ], + "image_footnote": [], + "bbox": [ + 187, + 97, + 802, + 516 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Velickovi ˇ c et al. also generalizes the Transformer’s (Vaswani et al., 2017) self-attention mechanism, ´ from sequences to graphs (Joshi, 2020). ", + "bbox": [ + 174, + 631, + 825, + 660 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Nowadays, GAT is one of the most popular GNN architectures (Bronstein et al., 2021) and is considered as the state-of-the-art neural architecture for learning with graphs (Wang et al., 2019a). Nevertheless, in this paper we show that GAT does not actually compute the expressive, well known, type of attention (Bahdanau et al., 2014), which we call dynamic attention. Instead, we show that GAT computes only a restricted “static” form of attention: for any query node, the attention function is monotonic with respect to the neighbor (key) scores. That is, the ranking (the argsort) of attention coefficients is identical for all nodes in the graph, and is unconditioned on the query node. This fact severely hurts the expressiveness of GAT, and is demonstrated in Figure 1a. ", + "bbox": [ + 174, + 666, + 825, + 777 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Supposedly, the conceptual idea of attention as the form of interaction between GNN nodes is orthogonal to the specific choice of attention function. However, Velickovi ˇ c et al.’s original design of ´ GAT has spread to a variety of domains (Yang et al., 2020; Wang et al., 2019c; Huang and Carley, 2019; Ma et al., 2020; Kosaraju et al., 2019; Nathani et al., 2019; Wu et al., 2020; Zhang et al., 2020) and has become the default implementation of “graph attention network” in all popular GNN libraries such as PyTorch Geometric (Fey and Lenssen, 2019), DGL (Wang et al., 2019b), and others (Dwivedi et al., 2020; Gordic, 2020; Brockschmidt, 2020). Many other works employed GNNs with attention ´ mechanisms other than the standard GAT’s (Yun et al. (2019), see also Appendix A). However, none of these works identified the monotonicity of GAT’s attention mechanism, the theoretical differences between attention types, nor empirically compared their performance. ", + "bbox": [ + 174, + 785, + 825, + 924 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "To overcome the limitation we identified in GAT, we introduce a simple fix to its attention function by only modifying the order of internal operations. The result is GATv2 – a graph attention variant that has a universal approximator attention function, and is thus strictly more expressive than GAT. The effect of fixing the attention function in GATv2 is demonstrated in Figure 1b. ", + "bbox": [ + 174, + 103, + 825, + 159 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In summary, our main contribution is identifying that one of the most popular GNN types, the graph attention network, does not compute dynamic attention, the kind of attention that it seems to compute. We introduce formal definitions for analyzing the expressive power of graph attention mechanisms (Definitions 3.1 and 3.2), and derive our claims theoretically (Theorem 1) from the equations of Velickovi ˇ c et al. (2018). Empirically, we use a synthetic problem to show that standard GAT ´ cannot express problems that require dynamic attention (Section 4.1). We introduce a simple fix by switching the order of internal operations in GAT, and propose GATv2, which does compute dynamic attention (Theorem 2). We further conduct a thorough empirical comparison of GAT and GATv2 and find that GATv2 outperforms GAT across 12 benchmarks of node-, link-, and graph-prediction. For example, GATv2 outperforms extensively tuned GNNs by over $1 . 4 \\%$ in the difficult “UnseenProj Test” set of the VarMisuse task (Allamanis et al., 2018), without any hyperparameter tuning; and GATv2 improves over an extensively-tuned GAT by $1 1 . 5 \\%$ in 13 prediction objectives in QM9. In node-prediction benchmarks from OGB (Hu et al., 2020), not only that GATv2 outperforms GAT with respect to accuracy – we find that dynamic attention provided a much better robustness to noise. ", + "bbox": [ + 173, + 166, + 825, + 361 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2 PRELIMINARIES ", + "text_level": 1, + "bbox": [ + 176, + 381, + 339, + 397 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "A directed graph $\\mathcal { G } = ( \\nu , \\mathcal { E } )$ contains nodes $\\mathcal { V } = \\{ 1 , . . . , n \\}$ and edges $\\mathcal { E } \\subseteq \\mathcal { V } \\times \\mathcal { V }$ , where $( j , i ) \\in \\mathcal { E }$ denotes an edge from a node $j$ to a node $i$ . We assume that every node $\\textit { i } \\in \\mathcal { V }$ has an initial representation $\\bar { \\boldsymbol { h } } _ { i } ^ { ( 0 ) } \\in \\mathbb { R } ^ { d _ { 0 } }$ . An undirected graph can be represented with bidirectional edges. ", + "bbox": [ + 174, + 411, + 825, + 458 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.1 GRAPH NEURAL NETWORKS ", + "text_level": 1, + "bbox": [ + 176, + 473, + 415, + 488 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "A graph neural network (GNN) layer updates every node representation by aggregating its neighbors’ representations. A layer’s input is a set of node representations $\\{ h _ { i } \\in \\bar { \\mathbb { R } } ^ { d } \\bar { | } \\ i \\stackrel { - } { \\in } \\mathcal { V } \\}$ and the set of edges $\\mathcal { E }$ . A layer outputs a new set of node representations $\\{ h _ { i } ^ { \\prime } \\in \\mathbb { R } ^ { d ^ { \\prime } } \\mid i \\in \\mathcal { V } \\}$ , where the same parametric function is applied to every node given its neighbors $\\mathcal { N } _ { i } = \\{ j \\in \\mathcal { V } \\mid ( j , i ) \\in \\mathcal { E } \\}$ : ", + "bbox": [ + 173, + 500, + 825, + 559 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/5c0f7e69ad5b558b5aa89dba0f3b2dc5e793350b81205f5ed8aaa6ea2bd60471.jpg", + "text": "$$\n\\begin{array} { r } { \\pmb { h } _ { i } ^ { \\prime } = f _ { \\theta } \\left( \\pmb { h } _ { i } , \\mathrm { A G G R E G A T E } \\left( \\{ \\pmb { h } _ { j } \\ | \\ j \\in \\mathcal { N } _ { i } \\} \\right) \\right) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 354, + 563, + 643, + 580 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The design of $f$ and AGGREGATE is what mostly distinguishes one type of GNN from the other. For example, a common variant of GraphSAGE (Hamilton et al., 2017) performs an element-wise mean as AGGREGATE, followed by concatenation with $\\boldsymbol { h } _ { i }$ , a linear layer and a ReLU as $f$ . ", + "bbox": [ + 174, + 585, + 825, + 627 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.2 GRAPH ATTENTION NETWORKS ", + "text_level": 1, + "bbox": [ + 176, + 643, + 436, + 657 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "GraphSAGE and many other popular GNN architectures $\\mathrm { { X u } }$ et al., 2019; Duvenaud et al., 2015) weigh all neighbors $j \\in \\mathcal N _ { i }$ with equal importance (e.g., mean or max-pooling as AGGREGATE). To address this limitation, GAT (Velickovi ˇ c et al., 2018) instantiates Equation (1) by computing a learned ´ weighted average of the representations of ${ \\mathcal { N } } _ { i }$ . A scoring function $e : \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\mathbb { R }$ computes a score for every edge $( j , i )$ , which indicates the importance of the features of the neighbor $j$ to the node $i$ : ", + "bbox": [ + 174, + 669, + 825, + 739 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/fdf1aac9f4eb59b6d40db1e9082b54ae55306d8a0e6db961c7777c502a2c36f3.jpg", + "text": "$$\ne ( \\pmb { h } _ { i } , \\pmb { h } _ { j } ) = \\mathrm { L e a k y R e L U } ( \\pmb { a } ^ { \\top } \\cdot [ \\pmb { W } \\pmb { h } _ { i } | | \\pmb { W } \\pmb { h } _ { j } | ) \n$$", + "text_format": "latex", + "bbox": [ + 339, + 744, + 656, + 765 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\pmb { a } \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }$ , $W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }$ are learned, and $\\parallel$ denotes vector concatenation. These attention scores are normalized across all neighbors $j \\in \\mathcal N _ { i }$ using softmax, and the attention function is defined as: ", + "bbox": [ + 174, + 770, + 828, + 800 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/b4309c7d70cbdbb892256ebb05106e903830abef53b0557adaa3453de5e3f136.jpg", + "text": "$$\n\\alpha _ { i j } = \\mathrm { s o f t m a x } _ { j } \\left( e \\left( h _ { i } , h _ { j } \\right) \\right) = \\frac { \\exp { \\left( e \\left( h _ { i } , h _ { j } \\right) \\right) } } { \\sum _ { j ^ { \\prime } \\in { \\mathcal { N } } _ { i } } \\exp { \\left( e \\left( h _ { i } , h _ { j ^ { \\prime } } \\right) \\right) } }\n$$", + "text_format": "latex", + "bbox": [ + 308, + 804, + 689, + 840 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Then, GAT computes a weighted average of the transformed features of the neighbor nodes (followed by a nonlinearity $\\sigma$ ) as the new representation of $i$ , using the normalized attention coefficients: ", + "bbox": [ + 173, + 844, + 828, + 873 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/e616a2cc35af6e2807af1d8d3102e0a51231e3b5e6c203fb765b2da392727b92.jpg", + "text": "$$\n\\begin{array} { r } { \\pmb { h } _ { i } ^ { \\prime } = \\sigma \\left( \\sum _ { j \\in \\mathcal { N } _ { i } } \\pmb { \\alpha } _ { i j } \\cdot \\pmb { W } \\pmb { h } _ { j } \\right) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 397, + 877, + 599, + 905 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "From now on, we will refer to Equations (2) to (4) as the definition of GAT. ", + "bbox": [ + 173, + 909, + 669, + 924 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 THE EXPRESSIVE POWER OF GRAPH ATTENTION MECHANISMS ", + "text_level": 1, + "bbox": [ + 174, + 102, + 735, + 118 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In this section, we explain why attention is limited when it is not dynamic (Section 3.1). We then show that GAT is severely constrained, because it can only compute static attention (Section 3.2). Next, we show how GAT can be fixed (Section 3.3), by simply modifying the order of operations. ", + "bbox": [ + 174, + 132, + 825, + 176 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We refer to a neural architecture (e.g., the scoring or the attention function of GAT) as a family of functions, parameterized by the learned parameters. An element in the family is a concrete function with specific trained weights. In the following, we use $[ n ]$ to denote the set $[ n ] = \\{ 1 , 2 , . . . , n \\} \\subset \\mathbb { N }$ . ", + "bbox": [ + 173, + 181, + 826, + 224 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.1 THE IMPORTANCE OF DYNAMIC WEIGHTING", + "text_level": 1, + "bbox": [ + 174, + 239, + 526, + 255 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Attention is a mechanism for computing a distribution over a set of input key vectors, given an additional query vector. If the attention function always weighs one key at least as much as any other key, unconditioned on the query, we say that this attention function is static: ", + "bbox": [ + 174, + 266, + 825, + 309 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Definition 3.1 (Static attention). A (possibly infinite) family of scoring functions $\\mathcal { F } \\subseteq$ $\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)$ computes static scoring for a given set of key vectors $\\mathbb { K } = \\{ k _ { 1 } , . . . , k _ { n } \\} \\subset \\mathbb { R } ^ { d }$ and query vectors $\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }$ , if for every $f \\in { \\mathcal { F } }$ there exists a “highest scoring” key $j _ { f } \\in [ n ]$ such that for every query $i \\in [ m ]$ and key $j \\in [ n ]$ it holds that $f \\left( \\boldsymbol { q } _ { i } , \\boldsymbol { k } _ { j _ { f } } \\right) \\geq f \\left( \\boldsymbol { q } _ { i } , \\boldsymbol { k } _ { j } \\right)$ . We say that a family of attention functions computes static attention given $\\Bbbk$ and $\\mathbb { Q }$ , if its scoring function computes static scoring, possibly followed by monotonic normalization such as softmax. ", + "bbox": [ + 173, + 311, + 825, + 400 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Static attention is very limited because every function $f \\in { \\mathcal { F } }$ has a key that is always selected, regardless of the query. Such functions cannot model situations where different keys have different relevance to different queries. Static attention is demonstrated in Figure 1a. ", + "bbox": [ + 174, + 411, + 825, + 454 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The general and powerful form of attention is dynamic attention: ", + "bbox": [ + 174, + 460, + 598, + 474 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Definition 3.2 (Dynamic attention). A (possibly infinite) family of scoring functions $\\mathcal { F } \\subseteq$ $\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)$ computes dynamic scoring for a given set of key vectors $\\mathbb { K } = \\{ \\boldsymbol { k } _ { 1 } , . . . , \\boldsymbol { k } _ { n } \\} \\subset \\mathbb { R } ^ { d }$ and query vectors $\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }$ , if for any mapping $\\varphi$ : $[ m ] [ n ]$ there exists $f \\in { \\mathcal { F } }$ such that for any query $i \\in [ m ]$ and any key $j _ { \\neq \\varphi ( i ) } \\in [ n ] \\colon f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right) > f \\left( q _ { i } , k _ { j } \\right) .$ . We say that a family of attention functions computes dynamic attention for $\\mathbb { K }$ and $\\mathbb { Q }$ , if its scoring function computes dynamic scoring, possibly followed by monotonic normalization such as softmax. ", + "bbox": [ + 174, + 478, + 825, + 565 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "That is, dynamic attention can select every key $\\varphi \\left( i \\right)$ using the query $i$ , by making $f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right)$ the maximal in $\\left\\{ f \\left( \\pmb { q } _ { i } , \\pmb { k } _ { j } \\right) \\mid j \\in [ n ] \\right\\}$ . Note that dynamic and static attention are exclusive properties, but they are not complementary. Further, every dynamic attention family has strict subsets of static attention families with respect to the same $\\mathbb { K }$ and $\\mathbb { Q }$ . Dynamic attention is demonstrated in Figure 1b. ", + "bbox": [ + 174, + 577, + 825, + 635 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Attending by decaying Another way to think about attention is the ability to “focus” on the most relevant inputs, given a query. Focusing is only possible by decaying other inputs, i.e., giving these decayed inputs lower scores than others. If one key is always given an equal or greater attention score than other keys (as in static attention), no query can ignore this key or decay this key’s score. ", + "bbox": [ + 174, + 643, + 825, + 700 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2 THE LIMITED EXPRESSIVITY OF GAT ", + "text_level": 1, + "bbox": [ + 176, + 717, + 477, + 731 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Although the scoring function $e$ can be defined in various ways, the original definition of Velickovi ˇ c´ et al. (2018) (Equation (2)) has become the de facto practice: it has spread to a variety of domains and is now the standard implementation of “graph attention network” in all popular GNN libraries (Fey and Lenssen, 2019; Wang et al., 2019b; Dwivedi et al., 2020; Gordic, 2020; Brockschmidt, 2020). ´ ", + "bbox": [ + 173, + 743, + 825, + 799 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The motivation of GAT is to compute a representation for every node as a weighted average of its neighbors. Statedly, GAT is inspired by the attention mechanism of Bahdanau et al. (2014) and the self-attention mechanism of the Transformer (Vaswani et al., 2017). Nonetheless: ", + "bbox": [ + 174, + 806, + 823, + 848 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Theorem 1. A GAT layer computes only static attention, for any set of node representations $\\mathbb { K } =$ $\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}$ . In particular, for $n > 1$ , a GAT layer does not compute dynamic attention. ", + "bbox": [ + 171, + 852, + 821, + 881 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Proof. Let $\\mathcal { G } = ( \\nu , \\mathcal { E } )$ be a graph modeled by a GAT layer with some $\\textbf { \\em a }$ and $W$ values (Equations (2) and (3)), and having node representations $\\{ h _ { 1 } , . . . , h _ { n } \\}$ . The learned parameter $\\textbf { \\em a }$ can be written as a concatenation $\\pmb { a } = [ \\pmb { a } _ { 1 } | | \\pmb { a } _ { 2 } ] \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }$ such that $\\mathbf { \\boldsymbol { a } } _ { 1 } , \\mathbf { \\boldsymbol { a } } _ { 2 } \\in \\mathbb { R } ^ { d ^ { \\prime } }$ , and Equation (2) can be re-written as: ", + "bbox": [ + 173, + 895, + 825, + 924 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 101, + 810, + 119 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/9a37260e84a76d8e8ef9574908ecbecd7391412b40e967772e2bb526c983de31.jpg", + "text": "$$\ne \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { a } _ { 1 } ^ { \\top } \\boldsymbol { W } \\boldsymbol { h } _ { i } + \\boldsymbol { a } _ { 2 } ^ { \\top } \\boldsymbol { W } \\boldsymbol { h } _ { j } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 333, + 125, + 663, + 145 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Since $\\nu$ is finite, there exists a node $j _ { m a x } \\in \\mathcal { V }$ such that ${ \\pmb a } _ { 2 } ^ { \\top } { \\pmb W } { \\pmb h } _ { j _ { m a x } }$ is maximal among all nodes $j \\in \\mathcal { V } \\left( j _ { m a x } \\right.$ is the $j _ { f }$ required by Definition 3.1). Due to the monotonicity of LeakyReLU and softmax, for every query node $i \\in \\mathcal V$ , the node $j _ { m a x }$ also leads to the maximal value of its attention distribution $\\{ \\alpha _ { i j } \\mid j \\in \\mathcal { V } \\}$ . Thus, from Definition 3.1 directly, $\\alpha$ computes only static attention. This also implies that $\\alpha$ does not compute dynamic attention, because in GAT, Definition 3.2 holds only for constant mappings $\\varphi$ that map all inputs to the same output. □ ", + "bbox": [ + 173, + 152, + 825, + 238 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The consequence of Theorem 1 is that for any set of nodes $\\nu$ and a trained GAT layer, the attention function $\\alpha$ defines a constant ranking (argsort) of the nodes, unconditioned on the query nodes $i$ . That is, we can denote $s _ { j } = { a } _ { 2 } ^ { \\top } \\bar { W } h _ { j }$ and get that for any choice of $\\boldsymbol { h } _ { i }$ , $\\alpha$ is monotonic with respect to the per-node scores $\\{ s _ { j } \\mid j \\in \\mathcal { V } \\}$ . This global ranking induces the local ranking of every neighborhood ${ \\mathcal { N } } _ { i }$ . The only effect of $\\boldsymbol { h } _ { i }$ is in the “sharpness” of the produced attention distribution. This is demonstrated in Figure 1a (bottom), where different curves denote different queries $( h _ { i } )$ . A discussion regarding the generalization to multi-head attention can be found in Appendix C. ", + "bbox": [ + 173, + 252, + 825, + 352 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.3 BUILDING DYNAMIC GRAPH ATTENTION NETWORKS ", + "text_level": 1, + "bbox": [ + 174, + 367, + 584, + 382 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "To create a dynamic graph attention network, we modify the order of internal operations in GAT and introduce GATv2 – a simple fix of GAT that has a strictly more expressive attention mechanism. ", + "bbox": [ + 174, + 393, + 825, + 422 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "GATv2 The main problem in the standard GAT scoring function (Equation (2)) is that the learned layers $W$ and $^ { a }$ are applied consecutively, and thus can be collapsed into a single linear layer. To fix this limitation, we simply apply the $\\textbf { \\em a }$ layer after the nonlinearity (LeakyReLU), and the $W$ layer after the concatenation, 5 effectively applying an MLP to compute the score for each query-key pair: ", + "bbox": [ + 176, + 433, + 825, + 489 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/761e7cc2081eda77a832d747f69378b8f8b3188a35f388893ea08877645e1885.jpg", + "text": "$$\n\\begin{array} { r l } & { e \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { a } ^ { \\top } \\cdot \\left[ \\boldsymbol { W } \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { W } \\boldsymbol { h } _ { j } \\right| \\right) \\right. } \\\\ & { e \\left. \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\boldsymbol { a } ^ { \\top } \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { W } \\cdot \\left[ \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { h } _ { j } \\right] \\right) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 464, + 496, + 777, + 537 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The simple modification makes a significant difference in the expressiveness of the attention function: ", + "bbox": [ + 178, + 542, + 825, + 558 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Theorem 2. A GATv2 layer computes dynamic attention for any set of node representations $\\mathbb { K } =$ $\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}$ . ", + "bbox": [ + 173, + 561, + 823, + 590 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We prove Theorem 2 in Appendix B. The main idea is that we can define an appropriate function that GATv2 will be a universal approximator (Cybenko, 1989; Hornik, 1991) of. In contrast, GAT (Equation (52)) cannot approximate any such desired function (Theorem 1). ", + "bbox": [ + 173, + 602, + 825, + 645 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Complexity GATv2 has the same time-complexity as GAT’s declared complexity: $\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)$ . However, by merging its linear layers, GAT can be computed faster than stated by Velickovi ˇ c et al. ´ (2018). For a detailed time- and parametric-complexity analysis, see Appendix I. ", + "bbox": [ + 174, + 654, + 826, + 696 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4 EVALUATION ", + "text_level": 1, + "bbox": [ + 176, + 717, + 313, + 733 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "First, we demonstrate the weakness of GAT using a simple synthetic problem that GAT cannot even fit (cannot even achieve high training accuracy), but is easily solvable by GATv2 (Section 4.1). Second, we show that GATv2 is much more robust to edge noise, because its dynamic attention mechanisms allow it to decay noisy (false) edges, while GAT’s performance severely decreases as noise increases (Section 4.2). Finally, we compare GAT and GATv2 across 12 benchmarks overall. (Sections 4.3 to 4.6 and appendix F.4). We find that GAT is inferior to GATv2 across all examined benchmarks. ", + "bbox": [ + 173, + 748, + 825, + 833 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Setup When previous results exist, we take hyperparameters that were tuned for GAT and use them in GATv2, without any additional tuning. Self-supervision (Kim and Oh, 2021; Rong et al., 2020a), graph regularization (Zhao and Akoglu, 2020; Rong et al., 2020b), and other tricks (Wang, 2021; Huang et al., 2021) are orthogonal to the contribution of the GNN layer itself, and may further improve all GNNs. In all experiments of GATv2, we constrain the learned matrix by setting $W = [ W ^ { \\prime } | | W ^ { \\prime } ]$ , to rule out the increased number of parameters over GAT as the source of empirical difference (see Appendix I.2). Training details, statistics, and code are provided in Appendix D. ", + "bbox": [ + 174, + 843, + 825, + 900 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/79aab347d93a4c3bb0bb39665869459e72b57008bc185548856fefb238200d5b.jpg", + "image_caption": [ + "Figure 2: The DICTIONARYLOOKUP problem of size $k { = } 4$ : every node in the bottom row has an alphabetic attribute $( \\{ { \\mathrm { A } } , { \\mathrm { B } } , { \\mathrm { C } } , \\ldots \\} )$ and a numeric value $( \\{ 1 , 2 , 3 , \\ldots \\} )$ ; every node in the upper row has only an attribute; the goal is to predict the value for each node in the upper row, using its attribute. " + ], + "image_footnote": [], + "bbox": [ + 212, + 99, + 364, + 193 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/eee3bdfe02443e30b1eb1a486a618dd6fcf3d8c448dc16f59b4318aa5c305de0.jpg", + "image_caption": [ + "Figure 3: The DICTIONARYLOOKUP problem: GATv2 easily achieves $100 \\%$ train and test accuracies even for $k { = } 1 0 0$ and using only a single head. " + ], + "image_footnote": [], + "bbox": [ + 421, + 99, + 816, + 268 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 347, + 825, + 388 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Our main goal is to compare dynamic and static graph attention mechanisms. However, for reference, we also include non-attentive baselines such as GCN (Kipf and Welling, 2017), GIN (Xu et al., 2019) and GraphSAGE (Hamilton et al., 2017). These non-attentive GNNs can be thought of as a special case of attention, where every node gives all its neighbors the same attention score. Additional comparison to a Transformer-style scaled dot-product attention (“DPGAT”), which is strictly weaker than our proposed GATv2 (see a proof in Appendix G.1), is shown in Appendix G. ", + "bbox": [ + 174, + 396, + 825, + 481 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.1 SYNTHETIC BENCHMARK: DICTIONARYLOOKUP ", + "text_level": 1, + "bbox": [ + 173, + 503, + 553, + 518 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The DICTIONARYLOOKUP problem is a contrived problem that we designed to test the ability of a GNN architecture to perform dynamic attention. Here, we demonstrate that GAT cannot learn this simple problem. Figure 2 shows a complete bipartite graph of $2 k$ nodes. Each “key node” in the bottom row has an attribute $( \\{ \\mathrm { A } , \\mathrm { B } , \\mathrm { C } , \\bar { \\ldots } \\} )$ and a value $( \\{ 1 , 2 , 3 , \\ldots \\} )$ . Each “query node” in the upper row has only an attribute $( \\{ \\mathrm { A } , \\mathrm { B } , \\mathrm { C } , \\ldots \\} )$ . The goal is to predict the value of every query node (upper row), according to its attribute. Each graph in the dataset has a different mapping from attributes to values. We created a separate dataset for each $k = \\{ 1 , 2 , 3 , \\ldots \\}$ , for which we trained a different model, and measured per-node accuracy. ", + "bbox": [ + 174, + 532, + 825, + 645 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Although this is a contrived problem, it is relevant to any subgraph with keys that share more than one query, and each query needs to attend to the keys differently. Such subgraphs are very common in a variety of real-world domains. This problem tests the layer itself because it can be solved using a single GNN layer, without suffering from multi-layer side-effects such as over-smoothing (Li et al., 2018), over-squashing (Alon and Yahav, 2021), or vanishing gradients (Li et al., 2019). Our code will be made publicly available, to serve as a testbed for future graph attention mechanisms. ", + "bbox": [ + 174, + 651, + 825, + 734 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Results Figure 3 shows the following surprising results: GAT with a single head $( \\mathrm { G A T _ { 1 } } _ { h } )$ ) failed to fit the training set for any value of $k$ , no matter for how many iterations it was trained, and after trying various training methods. Thus, it expectedly fails to generalize (resulting in low test accuracy). Using 8 heads, $\\mathrm { G A T } _ { 8 h }$ successfully fits the training set, but generalizes poorly to the test set. In contrast, GATv2 easily achieves $100 \\%$ training and $100 \\%$ test accuracies for any value of $k$ , and even for $k { = } 1 0 0$ (not shown) and using a single head, thanks to its ability to perform dynamic attention. These results clearly show the limitations of GAT, which are easily solved by GATv2. An additional comparison to GIN, which could not fit this dataset, is provided in Figure 5 in Appendix F.1. ", + "bbox": [ + 173, + 746, + 825, + 858 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Visualization Figure 1a (top) shows a heatmap of GAT’s attention scores in this DICTIONARYLOOKUP problem. As shown, all query nodes $q 0$ to $q 9$ attend mostly to the eighth key $( k 8 )$ , and have the same ranking of attention coefficients (Figure 1a (bottom)). In contrast, Figure 1b shows how GATv2 can select a different key node for every query node, because it computes dynamic attention. ", + "bbox": [ + 174, + 867, + 825, + 924 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The role of multi-head attention Velickovi ˇ c et al. (2018) found the role of multi-head attention to ´ be stabilizing the learning process. Nevertheless, Figure 3 shows that increasing the number of heads strictly increases training accuracy, and thus, the expressivity. Thus, GAT depends on having multiple attention heads. In contrast, even a single GATv2 head generalizes better than a multi-head GAT. ", + "bbox": [ + 174, + 103, + 823, + 160 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.2 ROBUSTNESS TO NOISE ", + "text_level": 1, + "bbox": [ + 176, + 180, + 379, + 194 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We examine the robustness of dynamic and static attention to noise. In particular, we focus on structural noise: given an input graph $\\mathcal { G } = ( \\nu , \\mathcal { E } )$ and a noise ratio $0 \\leq p \\leq 1$ , we randomly sample $| { \\mathcal { E } } | \\times p$ non-existing edges $\\mathcal { E } ^ { \\prime }$ from $\\mathcal { V } \\times \\mathcal { V } \\backslash \\mathcal { E }$ . We then train the GNN on the noisy graph $\\mathcal { G } ^ { \\prime } { = } ( \\dot { \\mathcal { V } } , \\mathcal { E } \\cup \\bar { \\mathcal { E } } ^ { \\prime } )$ . ", + "bbox": [ + 174, + 207, + 825, + 250 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/77bbac2b2bb1e1b15f9fdfb4dc28a2c29279f545cc08fbe57d963cffd3d7f569.jpg", + "image_caption": [ + "Figure 4: Test accuracy compared to the noise ratio: GATv2 is more robust to structural noise compared to GAT. Each point is an average of 10 runs, error bars show standard deviation. " + ], + "image_footnote": [], + "bbox": [ + 176, + 267, + 810, + 487 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Results Figure 8 shows the accuracy on two node-prediction datasets from the Open Graph Benchmark (OGB; Hu et al., 2020) as a function of the noise ratio $p$ . As $p$ increases, all models show a natural decline in test accuracy in both datasets. Yet, thanks to their ability to compute dynamic attention, GATv2 shows a milder degradation in accuracy compared to GAT, which shows a steeper descent. We hypothesize that the ability to perform dynamic attention helps the models distinguishing between given data edges $( \\mathcal { E } )$ and noise edges $( \\mathcal { E } ^ { \\prime } )$ ; in contrast, GAT cannot distinguish between edges, because it scores the source and target nodes separately. These results clearly demonstrate the robustness of dynamic attention over static attention in noisy settings, which are common in reality. ", + "bbox": [ + 174, + 547, + 825, + 660 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.3 PROGRAMS: VARMISUSE ", + "text_level": 1, + "bbox": [ + 174, + 679, + 390, + 694 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Setup VARMISUSE (Allamanis et al., 2018) is an inductive node-pointing problem that depends on 11 types of syntactic and semantic interactions between elements in computer programs. ", + "bbox": [ + 174, + 710, + 823, + 738 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We used the framework of Brockschmidt (2020), who performed an extensive hyperparameter tuning by searching over 30 configurations for every GNN type. We took their best GAT hyperparameters and used them to train GATv2, without further tuning. ", + "bbox": [ + 174, + 746, + 509, + 815 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Results As shown in Table 1, GATv2 is more accurate than GAT and other GNNs in the SeenProj test sets. Furthermore, GATv2 achieves an even higher improvement in the UnseenProj test set. Overall, these results demonstrate the power of GATv2 in modeling complex relational problems, especially ", + "bbox": [ + 176, + 827, + 511, + 919 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Table 1: Accuracy (5 runs $\\pm$ stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both test sets, using GAT’s hyperparameters. $\\dagger$ previously reported by Brockschmidt (2020). ", + "bbox": [ + 524, + 746, + 825, + 803 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/788a3fe581e9afc6eb6c6f6d21f417466d0cf87ea670d21ae7fcc0b84f5ff63a.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
ModelSeenProjUnseenProj
No-GCN+87.2±1.581.4±2.3
AttentionGIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
", + "bbox": [ + 529, + 814, + 810, + 905 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "since it outperforms extensively tuned models, without any further tuning by us. ", + "bbox": [ + 173, + 910, + 697, + 924 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.4 NODE-PREDICTION ", + "text_level": 1, + "bbox": [ + 176, + 103, + 351, + 117 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We further compare GATv2, GAT, and other GNNs on four node-prediction datasets from OGB. ", + "bbox": [ + 173, + 128, + 803, + 145 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/1ecb583dd83a8d1bd3d786a8ff4fc02573dabb8531d0aab9c0194499d5f90f53.jpg", + "table_caption": [ + "Table 2: Average accuracy (Table 2a) and ROC-AUC (Table 2b) in node-prediction datasets (10 runs±std). In all datasets, GATv2 outperforms GAT. † – previously reported by Hu et al. (2020). " + ], + "table_footnote": [], + "table_body": "
(a)(b)
ModelAttn. Headsogbn-arxivogbn-productsogbn-magogbn-proteins
GCN+071.74±0.2978.97±0.3330.43±0.2572.51±0.35
GraphSAGE+071.49±0.2778.70±0.3631.53±0.1577.68±0.20
GAT171.59±0.3879.04±1.5432.20±1.4670.77±5.79
871.54±0.3077.23±2.3731.75±1.6078.63±1.62
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.4477.23±3.32
871.87±0.2578.46±2.4532.52±0.3979.52±0.55
", + "bbox": [ + 179, + 195, + 815, + 338 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Results Results are shown in Table 2. In all settings and all datasets, GATv2 is more accurate than GAT and the non-attentive GNNs. Interestingly, in the datasets of Table 2a, even a single head of $G A T \\nu 2$ outperforms GAT with 8 heads. In Table 2b (ogbn-proteins), increasing the number of heads results in a major improvement for GAT (from 70.77 to 78.63), while GATv2 already gets most of the benefit using a single attention head. These results demonstrate the superiority of GATv2 over GAT in node prediction (and even with a single head), thanks to GATv2’s dynamic attention. ", + "bbox": [ + 174, + 354, + 826, + 439 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.5 GRAPH-PREDICTION: QM9 ", + "text_level": 1, + "bbox": [ + 176, + 455, + 405, + 469 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Setup In the QM9 dataset (Ramakrishnan et al., 2014; Gilmer et al., 2017), each graph is a molecule and the goal is to regress each graph to 13 real-valued quantum chemical properties. We used the implementation of Brockschmidt (2020) who performed an extensive hyperparameter search over 500 configurations; we took their best-found configuration of GAT to implement GATv2. ", + "bbox": [ + 174, + 484, + 825, + 541 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/00723c5a064bad5b41815a1955c05911652d470749e6c04360aa9dcf150d9e6d.jpg", + "table_caption": [ + "Table 3: Average error rates (lower is better), 5 runs for each property, on the QM9 dataset. The best result among GAT and GATv2 is marked in bold; the globally best result among all GNNs is marked in bold and underline. $\\dagger$ was previously tuned and reported by Brockschmidt (2020). " + ], + "table_footnote": [], + "table_body": "
Predicted Property
Model12345678910111213Rel. to GAT
GCNt3.214.221.451.622.4216.38 17.40 7.828.249.057.003.931.02-1.5%
GINt2.644.671.421.502.2715.63 12.93 5.8818.715.625.383.531.05-2.3%
GATt2.684.651.481.532.3152.39 14.87 7.6116.867.646.544.111.48+0%
GATv22.654.281.411.472.2916.37 14.03 6.076.286.605.973.571.59-11.5%
", + "bbox": [ + 186, + 607, + 812, + 713 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Results Table 3 shows the main results: GATv2 achieves a lower (better) average error than GAT, by $1 1 . 5 \\%$ relatively. GAT achieves the overall highest average error. In some properties, the non-attentive GNNs, GCN and GIN, perform best. We hypothesize that attention is not needed in modeling these properties. Generally, GATv2 achieves the lowest overall average relative error (rightmost column). ", + "bbox": [ + 174, + 729, + 825, + 786 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.6 LINK-PREDICTION", + "text_level": 1, + "bbox": [ + 176, + 803, + 346, + 816 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We compare GATv2, GAT, and other GNNs in link-prediction datasets from OGB. ", + "bbox": [ + 174, + 828, + 712, + 843 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Results Table 8 shows that in all datasets, GATv2 achieves a higher MRR than GAT, which achieves the lowest MRR. However, the non-attentive GraphSAGE performs better than all attentive GNNs. We hypothesize that attention might not be needed in these datasets. Another possibility is that dynamic attention is especially useful in graphs that have high node degrees: in ogbn-products and ogbn-proteins (Table 2) the average node degrees are 50.5 and 597, respectively (see Table 5 in ", + "bbox": [ + 174, + 853, + 825, + 924 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Appendix E). ogbl-collab and ogbl-citation2 (Table 8), however, have much lower average node degrees – of 8.2 and 20.7. We hypothesize that a dynamic attention mechanism is especially useful to select the most relevant neighbors when the total number of neighbors is high. We leave the study of the effect of the datasets’s average node degrees on the optimal GNN architecture for future work. ", + "bbox": [ + 174, + 103, + 825, + 160 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4.7 DISCUSSION ", + "text_level": 1, + "bbox": [ + 174, + 184, + 302, + 198 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In all examined benchmarks, we found that $G A T \\nu 2$ is more accurate than GAT. Further, we found that GATv2 is significantly more robust to noise than GAT. In the synthetic DICTIONARYLOOKUP benchmark (Section 4.1), GAT fails to express the data, and thus achieves even poor training accuracy. ", + "bbox": [ + 174, + 213, + 825, + 255 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In few of the benchmarks (Table 8 and some of the properties in Table 3) – a non-attentive model such as GCN or GIN achieved a higher accuracy than all GNNs that do use attention. We hypothesize that attention is not needed in these datasets. ", + "bbox": [ + 174, + 262, + 825, + 304 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Which graph attention mechanism should I use? It is usually impossible to determine in advance which architecture would perform best. A theoretically weaker model may perform better in practice, because a stronger model might overfit the training data if the task is “too simple” and does not require such expressiveness. Intuitively, we believe that the more complex the interactions between nodes are – the more benefit a GNN can take from theoretically stronger graph attention mechanisms such as GATv2. The main question is whether the problem has a global ranking of “influential” nodes (GAT is sufficient), or do different nodes have different rankings of neighbors (use GATv2). ", + "bbox": [ + 174, + 315, + 825, + 412 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Velickovi ˇ c, the author of GAT, has confirmed on Twitter ´ 6 that GAT was designed to work in the “easy-to-overfit” datasets of the time (2017), such as Cora, Citeseer and Pubmed (Sen et al., 2008), where the data might had an underlying static ranking of “globally important” nodes. Velickovi ˇ c´ agreed that newer and more challenging benchmarks may demand stronger attention mechanisms such as GATv2. that depends on the representation of the query node as well. In this paper, we revisit the traditional assumptions and show that many modern graph benchmarks and datasets contain more complex interactions, and thus require dynamic attention. ", + "bbox": [ + 173, + 419, + 825, + 517 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "5 CONCLUSION ", + "text_level": 1, + "bbox": [ + 176, + 544, + 318, + 560 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In this paper, we identify that the popular and widespread Graph Attention Network does not compute dynamic attention. Instead, the attention mechanism in the standard definition and implementations of GAT is only static: for any query, its neighbor-scoring is monotonic with respect to per-node scores. Further, the ranking (the argsort) of attention coefficients is identical for all nodes in the graph, and is unconditioned on the query node. As a result, GAT cannot even express simple alignment problems. To address this limitation, we introduce a simple fix and propose GATv2: by modifying the order of operations in GAT, GATv2 achieves a universal approximator attention function and is thus strictly more powerful than GAT. ", + "bbox": [ + 174, + 580, + 825, + 691 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We demonstrate the empirical advantage of GATv2 over GAT in a synthetic problem that requires dynamic selection of nodes, and in 12 benchmarks from OGB and other public datasets. Our experiments show that GATv2 outperforms GAT in all benchmarks while having the same parametric cost. ", + "bbox": [ + 174, + 699, + 826, + 741 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We encourage the community to use GATv2 instead or in addition to GAT whenever comparing new GNN architectures to the common strong baselines. In complex tasks and domains and in challenging datasets, a model that uses GAT as an internal component can replace it with GATv2 to benefit from a strictly more powerful model. To this end, we make our code publicly available at https: //github.com/tech-srl/how_attentive_are_gats ,7 and GATv2 is available as part of the PyTorch Geometric library,8 the Deep Graph Library,9 and the TensorFlow GNN library.10 ", + "bbox": [ + 174, + 747, + 825, + 832 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "ACKNOWLEDGMENTS ", + "text_level": 1, + "bbox": [ + 176, + 103, + 356, + 117 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We thank Gail Weiss for the helpful discussions, thorough feedback, and inspirational paper (Weiss et al., 2018). We also thank Petar Velickovi ˇ c for the useful initial discussion about the complexity ´ and implementation of GAT, Lucio Dery, and the Graph Representation Learning Reading Group at Mila11 for their useful comments and questions. ", + "bbox": [ + 174, + 133, + 825, + 190 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "A RELATED WORK ", + "text_level": 1, + "bbox": [ + 176, + 212, + 348, + 228 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Attention in GNNs Modeling pairwise interactions between elements in graph-structured data goes back to interaction networks (Battaglia et al., 2016; Hoshen, 2017) and relational networks (Santoro et al., 2017). The GAT formulation of Velickovi ˇ c et al. (2018) rose as the most popular framework ´ for attentional GNNs, thanks to its simplicity, generality, and applicability beyond reinforcement learning (Denil et al., 2017; Duan et al., 2017). Nevertheless, in this work, we show that the popular and widespread definition of GAT is severely constrained to static attention only. ", + "bbox": [ + 174, + 250, + 825, + 333 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Other graph attention mechanisms Many works employed GNNs with attention mechanisms other than the standard GAT’s (Zhang et al., 2018; Thekumparampil et al., 2018; Gao and Ji, 2019; Lukovnikov and Fischer, 2021; Shi et al., 2020; Dwivedi and Bresson, 2020; Busbridge et al., 2019; Rong et al., 2020a; Velickovi ˇ c et al., 2020), and Lee et al. (2018) conducted an extensive survey ´ of attention types in GNNs. However, none of these works identified the monotonicity of GAT’s attention mechanism, the theoretical differences between attention types, nor empirically compared their performance. Kim and Oh (2021) compared two graph attention mechanisms empirically, but in a specific self-supervised scenario, without observing the theoretical difference in their expressiveness. ", + "bbox": [ + 174, + 343, + 825, + 455 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "The static attention of GAT Qiu et al. (2018) recognized the order-preserving property of GAT, but did not identify the severe theoretical constraint that this property implies: the inability to perform dynamic attention (Theorem 1). Furthermore, they presented GAT’s monotonicity as a desired trait (!) To the best of our knowledge, our work is the first work to recognize the inability of GAT to perform dynamic attention and its practical harmful consequences. ", + "bbox": [ + 174, + 465, + 825, + 536 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "B PROOF FOR THEOREM 2 ", + "text_level": 1, + "bbox": [ + 176, + 558, + 410, + 574 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "For brevity, we repeat our definition of dynamic attention (Definition 3.2): ", + "bbox": [ + 174, + 590, + 658, + 604 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Definition 3.2 (Dynamic attention). A (possibly infinite) family of scoring functions $\\mathcal { F } \\subseteq$ $\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)$ computes dynamic scoring for a given set of key vectors $\\mathbb { K } = \\{ \\boldsymbol { k } _ { 1 } , . . . , \\boldsymbol { k } _ { n } \\} \\subset \\mathbb { R } ^ { d }$ and query vectors $\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }$ , if for any mapping $\\varphi$ : $[ m ] [ n ]$ there exists $f \\in { \\mathcal { F } }$ such that for any query $i \\in [ m ]$ and any key $j _ { \\neq \\varphi ( i ) } \\in [ n ] \\colon f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right) > f \\left( q _ { i } , k _ { j } \\right) .$ . We say that a family of attention functions computes dynamic attention for $\\mathbb { K }$ and $\\mathbb { Q }$ , if its scoring function computes dynamic scoring, possibly followed by monotonic normalization such as softmax. ", + "bbox": [ + 173, + 621, + 825, + 710 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Theorem 2. A GATv2 layer computes dynamic attention for any set of node representations $\\mathbb { K } =$ $\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}$ . ", + "bbox": [ + 169, + 715, + 823, + 744 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Proof. Let $\\mathcal { G } ~ = ~ ( \\nu , \\mathcal { E } )$ be a graph modeled by a GATv2 layer, having node representations $\\{ h _ { 1 } , . . . , h _ { n } \\}$ , and let $\\varphi : [ n ] \\to [ n ]$ be any node mapping $[ n ] [ n ]$ . We define $\\dot { g } : \\mathbb { R } ^ { 2 d } \\mathbb { R }$ as follows: ", + "bbox": [ + 173, + 763, + 825, + 806 + ], + "page_idx": 9 + }, + { + "type": "equation", + "img_path": "images/da7983c6b9a359f4db8ff7bd72c050147ef4fb4f9f548a408e173d5dcc722ea8.jpg", + "text": "$$\ng \\left( \\pmb { x } \\right) = \\left. { 1 } \\pmb { \\mathrm { \\rbrace } } \\ i : \\pmb { x } = \\left[ \\pmb { h } _ { i } \\right| \\left| \\pmb { h } _ { \\varphi ( i ) } \\right] \\right.\n$$", + "text_format": "latex", + "bbox": [ + 382, + 805, + 612, + 842 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Next, we define a continues function $\\widetilde { g } : { \\mathbb { R } } ^ { 2 d } \\to { \\mathbb { R } }$ that equals to $g$ in only specific $n ^ { 2 }$ inputs: ", + "bbox": [ + 169, + 856, + 779, + 872 + ], + "page_idx": 9 + }, + { + "type": "equation", + "img_path": "images/04d87a1789c2278580b6f5bdadc283de2be469ddb47bf2adf129bccb50b5cbec.jpg", + "text": "$$\n\\widetilde { g } ( [ h _ { i } \\| h _ { j } ] ) = g ( [ h _ { i } \\| h _ { j } ] ) , \\forall i , j \\in [ n ]\n$$", + "text_format": "latex", + "bbox": [ + 375, + 878, + 622, + 897 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "For all other inputs $\\boldsymbol { x } \\in \\mathbb { R } ^ { 2 d }$ , $\\widetilde g ( x )$ realizes to any values that maintain the continuity of $\\widetilde g$ (this is epossible because we fixed the values of $\\widetilde g$ for only a finite set of points). 12 ", + "bbox": [ + 169, + 103, + 823, + 132 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Thus, for every node $i \\in \\nu$ and $j _ { \\neq \\varphi ( i ) } \\in \\mathcal { V }$ : ", + "bbox": [ + 176, + 138, + 460, + 154 + ], + "page_idx": 10 + }, + { + "type": "equation", + "img_path": "images/e64247a44bd0da371c015f9ee20aa6ecc9ef48dfad9589c1616a16565ea446da.jpg", + "text": "$$\n1 = \\widetilde { g } \\left( \\left[ h _ { i } \\Vert h _ { \\varphi ( i ) } \\right] \\right) > \\widetilde { g } \\left( \\left[ h _ { i } \\Vert h _ { j } \\right] \\right) = 0\n$$", + "text_format": "latex", + "bbox": [ + 367, + 162, + 629, + 183 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "If we concatenate the two input vectors, and define the scoring function $e$ of GATv2 (Equation (7)) as a function of the concatenated vector $[ h _ { i } | | h _ { j } ]$ , from the universal approximation theorem (Hornik et al., 1989; Cybenko, 1989; Funahashi, 1989; Hornik, 1991), $e$ can approximate $\\widetilde g$ for any compact subset of R2d. ", + "bbox": [ + 173, + 195, + 825, + 252 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Thus, for any sufficiently small $\\epsilon$ (any $0 < \\epsilon < 1 / 2$ ) there exist parameters $W$ and $\\textbf { \\em a }$ such that for every node $i \\in \\nu$ and every $\\mathscr { j } _ { \\neq \\varphi ( i ) }$ : ", + "bbox": [ + 173, + 258, + 821, + 290 + ], + "page_idx": 10 + }, + { + "type": "equation", + "img_path": "images/87359cb3bdeba2a1c0aa11b6a1906074b8180a2e57b0bdb4e622f9fa259bc474.jpg", + "text": "$$\n{ e _ { W , a } } \\left( { h _ { i } } , { h _ { \\varphi ( i ) } } \\right) > 1 - \\epsilon > 0 + \\epsilon > { e _ { W , a } } \\left( { h _ { i } } , { h _ { j } } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 326, + 296, + 671, + 316 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "and due to the increasing monotonicity of softmax: ", + "bbox": [ + 173, + 323, + 513, + 338 + ], + "page_idx": 10 + }, + { + "type": "equation", + "img_path": "images/ae6999271be462896680164b08faadbcb45f1db0885e6c9aada7884447c43d1a.jpg", + "text": "$$\n\\alpha _ { i , \\varphi ( i ) } > \\alpha _ { i , j }\n$$", + "text_format": "latex", + "bbox": [ + 452, + 348, + 545, + 364 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "The choice of nonlinearity In general, these results hold if GATv2 had used any common nonpolynomial activation function (such as ReLU, sigmoid, or the hyperbolic tangent function). The LeakyReLU activation function of GATv2 does not change its universal approximation ability (Leshno et al., 1993; Pinkus, 1999; Park et al., 2021), and it was chosen only for consistency with the original definition of GAT. ", + "bbox": [ + 173, + 404, + 825, + 473 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "C GENERALIZATION TO MULTI-HEAD ATTENTION ", + "text_level": 1, + "bbox": [ + 174, + 496, + 601, + 511 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Velickovi ˇ c et al. (2018) found it beneficial to employ ´ $H$ separate attention heads and concatenate their outputs, similarly to Transformers. In this case, Theorem 1 holds for each head separately: every head $h \\in [ H ]$ has a (possibly different) node that maximizes $\\{ s _ { j } ^ { ( h ) } \\mid j \\in \\mathcal { V } \\}$ , and the output is the concatenation of $H$ static attention heads. ", + "bbox": [ + 174, + 526, + 825, + 587 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "D TRAINING DETAILS ", + "text_level": 1, + "bbox": [ + 176, + 608, + 370, + 625 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "In this section we elaborate on the training details of all of our experiments. All models use residual connections as in Velickovi ˇ c et al. (2018). All used code and data are publicly available under the ´ MIT license. ", + "bbox": [ + 176, + 640, + 823, + 683 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "D.1 NODE- AND LINK-PREDICTION ", + "text_level": 1, + "bbox": [ + 176, + 700, + 434, + 715 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "We used the provided splits of OGB (Hu et al., 2020) and the Adam optimizer. We tuned the following hyperparameters: number of layers $\\in \\{ 2 , 3 , 6 \\}$ , hidden size $\\in \\{ 6 4 , 1 2 8 , 2 5 6 \\}$ , learning rate $\\in \\{ 0 . 0 0 0 5 , 0 . 0 0 1 , 0 . 0 0 5 , 0 . 0 1 \\}$ and sampling method – full batch, GraphSAINT (Zeng et al., 2019) and NeighborSampling (Hamilton et al., 2017). We tuned hyperparameters according to validation score and early stopping. The final hyperparameters are detailed in Table 4. ", + "bbox": [ + 173, + 727, + 825, + 797 + ], + "page_idx": 10 + }, + { + "type": "table", + "img_path": "images/99ab02b3e8b0afbf67c8ce47dd5a7df3c52f36e24045d253fded2e4adebe327b.jpg", + "table_caption": [ + "Table 4: Training details of node- and link-prediction datasets. " + ], + "table_footnote": [], + "table_body": "
Dataset# layersHidden sizeLearning rateSampling method
ogbn-arxiv32560.01GraphSAINT
ogbn-products31280.001NeighborSampling
ogbn-mag22560.01NeighborSampling
ogbn-proteins6640.01NeighborSampling
ogbl-collab3640.001Full Batch
ogbl-citation232560.0005NeighborSampling
", + "bbox": [ + 248, + 101, + 748, + 213 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "D.2 ROBUSTNESS TO NOISE ", + "text_level": 1, + "bbox": [ + 174, + 265, + 383, + 280 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "In these experiments, we used the same best-found hyperparameters in node-prediction, with 8 attention heads in ogbn-arxiv and 1 head in ogbn-mag. Each point is an average of 10 runs. ", + "bbox": [ + 173, + 291, + 823, + 320 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "D.3 SYNTHETIC BENCHMARK: DICTIONARYLOOKUP ", + "text_level": 1, + "bbox": [ + 173, + 338, + 557, + 352 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "In all experiments, we used a learning rate decay of 0.5, a hidden size of $d = 1 2 8$ , a batch size of 1024, and the Adam optimizer. ", + "bbox": [ + 171, + 364, + 823, + 393 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "We created a separate dataset for every graph size $( k )$ , and we split each such dataset to train and test with a ratio of 80:20. Since this is a contrived problem, we did not use a validation set, and the reported test results can be thought of as validation results. Every model was trained on a fixed value of $k$ . Every key node (bottom row in Figure 2) was encoded as a sum of learned attribute embedding and a value embedding, followed by ReLU. ", + "bbox": [ + 174, + 400, + 825, + 469 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "We experimented with layer normalization, batch normalization, dropout, various activation functions and various learning rates. None of these changed the general trend, so the experiments in Figure 3 were conducted without any normalization, without dropout and a learning rate of 0.001. ", + "bbox": [ + 174, + 477, + 823, + 518 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "D.4 PROGRAMS: VARMISUSE ", + "text_level": 1, + "bbox": [ + 174, + 536, + 393, + 551 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "We used the code, splits, and the same best-found configurations as Brockschmidt (2020), who performed an extensive hyperparameter tuning by searching over 30 configurations for each GNN type. We trained each model five times. ", + "bbox": [ + 176, + 564, + 823, + 606 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "We took the best-found hyperparameters of Brockschmidt (2020) for GAT and used them to train GATv2, without any further tuning. ", + "bbox": [ + 174, + 613, + 823, + 641 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "D.5 GRAPH-PREDICTION: QM9 ", + "text_level": 1, + "bbox": [ + 176, + 659, + 410, + 674 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "We used the code and splits of Brockschmidt (2020) who performed an extensive hyperparameter search over 500 configurations. We took the best-found hyperparameters of Brockschmidt (2020) for GAT and used them to train GATv2. The only minor change from GAT is placing a residual connection after every layer, rather than after every other layer, which is within the experimented hyperparameter search that was reported by Brockschmidt (2020). ", + "bbox": [ + 174, + 685, + 825, + 756 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "D.6 COMPUTE AND RESOURCES ", + "text_level": 1, + "bbox": [ + 176, + 773, + 411, + 787 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Our experiments consumed approximately 100 days of GPU in total. We used cloud GPUs of type V100, and we used RTX 3080 and 3090 in local GPU machines. ", + "bbox": [ + 176, + 800, + 823, + 829 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "E DATA STATISTICS ", + "text_level": 1, + "bbox": [ + 176, + 851, + 354, + 867 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "E.1 NODE- AND LINK-PREDICTION DATASETS ", + "text_level": 1, + "bbox": [ + 174, + 882, + 509, + 897 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Statistics of the OGB datasets we used for node- and link-prediction are shown in Table 5. ", + "bbox": [ + 169, + 909, + 763, + 924 + ], + "page_idx": 11 + }, + { + "type": "table", + "img_path": "images/46e2ac78f7d9510797e6edc20553d155d54b390922301684be1a7d01d91f157e.jpg", + "table_caption": [ + "Table 5: Statistics of the OGB datasets (Hu et al., 2020). " + ], + "table_footnote": [], + "table_body": "
Dataset# nodes# edgesAvg. node degreeDiameter
ogbn-arxiv169,3431,166,24313.723
ogbn-mag1,939,74321,111,00721.76
ogbn-products2,449,02961,859,14050.527
ogbn-proteins132,53439,561,252597.09
ogbl-collab235,8681,285,4658.222
ogbl-citation22,927,96330,561,18720.721
", + "bbox": [ + 240, + 101, + 758, + 222 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "E.2 QM9 ", + "text_level": 1, + "bbox": [ + 174, + 276, + 254, + 291 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/e1a267d5274a5a15b6c2699809130678ba87975f4194a50a10aca2b3bd0a31f4.jpg", + "table_caption": [ + "Statistics of the QM9 dataset, as used in Brockschmidt (2020) are shown in Table 6. " + ], + "table_footnote": [ + "Table 6: Statistics of the QM9 chemical dataset (Ramakrishnan et al., 2014) as used by Brockschmidt (2020). " + ], + "table_body": "
TrainingValidationTest
# examples110,46210,00010,000
# nodes -average18.0318.0618.09
# edges - average18.6518.6718.72
Diameter - average6.356.356.35
", + "bbox": [ + 312, + 334, + 686, + 421 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "E.3 VARMISUSE ", + "text_level": 1, + "bbox": [ + 174, + 501, + 303, + 515 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/142de408d3dcc17abb31e482a2cedb6ea40aceb0d14995aec3ec00da71985a3a.jpg", + "table_caption": [ + "Statistics of the VARMISUSE dataset, as used in Allamanis et al. (2018) and Brockschmidt (2020), are shown in Table 7. " + ], + "table_footnote": [ + "Table 7: Statistics of the VARMISUSE dataset (Allamanis et al., 2018) as used by Brockschmidt (2020). " + ], + "table_body": "
TrainingValidationUnseenProject TestSeenProject Test
# graphs2543604265411703659974
# nodes - average2377174219593986
# edges - average72987851588212925
Diameter-average7.887.887.787.82
", + "bbox": [ + 207, + 570, + 790, + 659 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "F ADDITIONAL RESULTS ", + "text_level": 1, + "bbox": [ + 174, + 742, + 397, + 758 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "F.1 DICTIONARYLOOKUP ", + "text_level": 1, + "bbox": [ + 174, + 776, + 364, + 791 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Figure 5 shows additional comparison between GATv2 and GIN (Xu et al., 2019) in the DICTIONARYLOOKUP problem. GATv2 easily achieves $100 \\%$ train and test accuracy even for $k { = } 1 0 0$ and using only a single head. GIN, although considered as more expressive than other GNNs, cannot perfectly fit the training data (with a model size of $d = 1 2 8$ ) starting from $k { = } 2 0$ . ", + "bbox": [ + 173, + 804, + 826, + 861 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "F.2 LINK-PREDICTION", + "text_level": 1, + "bbox": [ + 174, + 881, + 344, + 896 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Results for link-prediction are showen in Table 8. ", + "bbox": [ + 173, + 909, + 498, + 924 + ], + "page_idx": 12 + }, + { + "type": "image", + "img_path": "images/4233ec4a02df2921f13b438873a45c7450236e836cc22f098f64cf61a279e12d.jpg", + "image_caption": [ + "Figure 5: Train and test accuracy across graph sizes in the DICTIONARYLOOKUP problem. GATv2 easily achieves $100 \\%$ train and test accuracy even for $k { = } 1 0 0$ and using only a single head. GIN $\\mathrm { { X u } }$ et al., 2019), although considered as more expressive than other GNNs, cannot perfectly fit the training data (with a model size of $d = 1 2 8$ ) starting from $k { = } 2 0$ . " + ], + "image_footnote": [], + "bbox": [ + 200, + 101, + 803, + 271 + ], + "page_idx": 13 + }, + { + "type": "table", + "img_path": "images/019284bd53268ef6e683724418d88ea9e9b95e3d7fc9ef3b6e444930e460100d.jpg", + "table_caption": [ + "Table 8: Average Hits $@ 5 0$ (Table 8a) and mean reciprocal rank (MRR) (Table 8b) in link-prediction benchmarks from OGB (10 runs±std). The best result among GAT and GATv2 is marked in bold; the best result among all GNNs is marked in bold and underline. $\\dagger$ was reported by Hu et al. (2020). " + ], + "table_footnote": [], + "table_body": "
ogbl-citation2
80.04±0.25
80.44±0.10
79.84±0.19
75.95±1.3180.33±0.13
80.14±0.71
", + "bbox": [ + 645, + 436, + 754, + 564 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "(b) ", + "bbox": [ + 684, + 417, + 705, + 430 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "F.3 QM9 ", + "text_level": 1, + "bbox": [ + 174, + 597, + 251, + 611 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Standard deviation for the QM9 results of Section 4.5 are presented in Table 9. ", + "bbox": [ + 173, + 626, + 689, + 641 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "F.4 PUBMED CITATION NETWORK ", + "text_level": 1, + "bbox": [ + 176, + 665, + 423, + 680 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "We tuned the following parameters for both GAT and GATv2: number of layers $\\in \\{ 0 , 1 , 2 \\}$ , hidden size $\\in \\{ 8 , 1 6 , 3 2 \\}$ , number of heads $\\in \\{ 1 , 4 , 8 \\}$ , dropout $\\in \\{ 0 . 4 , 0 . 6 , 0 . 8 \\}$ , bias $\\in \\{ T r u e , F a l s e \\}$ , share weights $\\in \\{ T r u e , F a l s e \\}$ , use residual $\\dot { \\in } \\{ T r u e , F a l \\dot { s } e \\}$ . Table 10 shows the test accuracy (100 runs±stdev) using the best hyperparameters found for each model. ", + "bbox": [ + 174, + 694, + 825, + 751 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "It is important to note that PubMed has only 60 training nodes, which hinders expressive models such as GATv2 from exploiting their approximation and generalization advantages. Still, GATv2 is more accurate than GAT even in this small dataset. In Table 15, we show that this difference is statistically significant (p-value $< 0 . 0 0 0 1$ ). ", + "bbox": [ + 174, + 757, + 825, + 814 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "G ADDITIONAL COMPARISON WITH TRANSFORMER-STYLE ATTENTION(DPGAT)", + "text_level": 1, + "bbox": [ + 173, + 840, + 784, + 876 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "The main goal of our paper is to highlight a severe theoretical limitation of the highly popular GAT architecture, and propose a minimal fix. ", + "bbox": [ + 173, + 895, + 823, + 924 + ], + "page_idx": 13 + }, + { + "type": "table", + "img_path": "images/7fd6def9d43f294d7cc625ec3140917897cf6fb89794df258cd7baf6db6c4453.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.25 18.71±23.369.05±1.21 5.62±0.817.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h10.80±2.183.11±0.14
GAT8h28.10±16.45 7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.365.37±0.26 4.11±0.271.48±0.87+134.1% +0%
GATv21h22.94±17.34
10.19±0.6322.56±17.4615.04±4.585.23±0.362.46±0.65+91.6%
GATv28h6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
", + "bbox": [ + 191, + 101, + 805, + 385 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Table 9: Average error rates (lower is better), 5 runs $\\pm$ standard deviation for each property, on the QM9 dataset. The best result among GAT and GATv2 is marked in bold; the globally best result among all GNNs is marked in bold and underline. $\\dagger$ was previously tuned and reported by Brockschmidt (2020). ", + "bbox": [ + 174, + 395, + 825, + 450 + ], + "page_idx": 14 + }, + { + "type": "table", + "img_path": "images/0b68db02c3f4936969ed4e3c1f6cc14ec2a34b6bc4e101b45bc1a8e2cd4f07cb.jpg", + "table_caption": [ + "Table 10: Accuracy (100 runs±stdev) on Pubmed. GATv2 is more accurate than GAT. " + ], + "table_footnote": [], + "table_body": "
ModelAccuracy
GAT78.1±0.59
GATv278.5±0.38
", + "bbox": [ + 437, + 492, + 560, + 547 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "We perform additional empirical comparison to DPGAT, which follows Luong et al. (2015) and the dot-product attention of the Transformer (Vaswani et al., 2017). We define DPGAT as: ", + "bbox": [ + 174, + 574, + 826, + 602 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/5b7116695b9f012e48806294910738c39d8fe1b477acc79bc36458ea30a52df0.jpg", + "text": "$$\ne \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right) = \\left( \\left( \\pmb { h } _ { i } ^ { \\top } \\pmb { Q } \\right) \\cdot \\left( \\pmb { h } _ { j } ^ { \\top } \\pmb { K } \\right) ^ { \\top } \\right) / \\sqrt { d _ { k } }\n$$", + "text_format": "latex", + "bbox": [ + 470, + 609, + 754, + 637 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Variants of DPGAT were used in prior work (Gao and Ji, 2019; Dwivedi and Bresson, 2020; Rong et al., 2020a; Velickovi ˇ c et al., 2020; Kim and Oh, 2021), and we consider it here for the conceptual ´ and empirical comparison with GAT. ", + "bbox": [ + 173, + 643, + 825, + 685 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Despite its popularity, DPGAT is strictly weaker than GATv2. DPGAT provably performs dynamic attention for any set of node representations only if they are linearly independent (see Theorem 3 and its proof in Appendix G.1). Otherwise, there are examples of node representations that are linearly dependent and mappings $\\varphi$ , for which dynamic attention does not hold (Appendix G.2). This constraint is not harmful when violated in practice, because every node has only a small set of neighbors, rather than all possible nodes in the graph; further, some nodes possibly never need to be “selected” in practice. ", + "bbox": [ + 173, + 693, + 825, + 791 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "G.1 PROOF THAT DPGAT PERFORMS DYNAMIC ATTENTION FOR LINEARLY INDEPENDENT NODE REPRESENTATIONS ", + "bbox": [ + 174, + 808, + 818, + 837 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Theorem 3. A DPGAT layer computes dynamic attention for any set of node representations $\\mathbb { K } = \\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}$ that are linearly independent. ", + "bbox": [ + 173, + 848, + 823, + 877 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Proof. Let $\\mathcal { G } = ( \\nu , \\mathcal { E } )$ be a graph modeled by a DPGAT layer, having linearly independent node representations $\\{ h _ { 1 } , . . . , h _ { n } \\}$ . Let $\\varphi : [ n ] [ n ]$ be any node mapping $[ { \\bar { n } } ] [ n ]$ . ", + "bbox": [ + 173, + 895, + 823, + 925 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "We denote the $\\mathrm { i } ^ { t h }$ row of a matrix $M$ as $M _ { i }$ ", + "bbox": [ + 174, + 102, + 468, + 118 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We define a matrix $_ { P }$ as: ", + "bbox": [ + 174, + 125, + 339, + 140 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/477f035d5e3de0a9ac649d2e748d973286069bc2d3286257c4f0cb84d9aab2f1.jpg", + "text": "$$\nP _ { i , j } = { \\left\\{ \\begin{array} { l l } { 1 } & { j = \\varphi ( i ) } \\\\ { 0 } & { { \\mathrm { o t h e r w i s e } } } \\end{array} \\right. }\n$$", + "text_format": "latex", + "bbox": [ + 421, + 136, + 573, + 171 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Let $\\ b { X } \\in \\mathbb { R } ^ { n } \\times \\mathbb { R } ^ { d }$ be the matrix holding the graph’s node representations as its rows: ", + "bbox": [ + 171, + 183, + 735, + 199 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/a6beda280a46d1888de1e211f54408173573f54ec526a68249ae83b483c0ab7c.jpg", + "text": "$$\n\\begin{array} { r } { \\mathbf { \\boldsymbol { X } } = \\left[ \\begin{array} { l l l } { - } & { h _ { 1 } } & { - } \\\\ { - } & { h _ { 2 } } & { - } \\\\ & { \\vdots } \\\\ { - } & { h _ { n } } & { - } \\end{array} \\right] } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 424, + 223, + 575, + 291 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Since the rows of $\\boldsymbol { X }$ are linearly independent, it necessarily holds that $d \\geq n$ ", + "bbox": [ + 173, + 304, + 678, + 320 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Next, we find weight matrices $Q \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d }$ and $K \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d }$ such that: ", + "bbox": [ + 173, + 324, + 665, + 340 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/9f1c806c98f690ce205e28a0f3dfd5c85857fcf8d4b7e2221d4f0574d3860dab.jpg", + "text": "$$\n( X Q ) \\cdot ( X K ) ^ { \\top } = P\n$$", + "text_format": "latex", + "bbox": [ + 423, + 345, + 575, + 366 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "To satisfy Equation (16), we choose $Q$ and $\\kappa$ such that $\\mathbf { \\Delta } X Q = U$ and $X K = P ^ { \\top } U$ where $U$ is an orthonormal matrix $( \\pmb { U } \\cdot \\pmb { U } ^ { \\top } = \\pmb { U } ^ { \\top } \\cdot \\pmb { U } = I )$ . ", + "bbox": [ + 174, + 369, + 826, + 400 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We can obtain $U$ using the singular value decomposition (SVD) of $\\boldsymbol { X }$ : ", + "bbox": [ + 173, + 406, + 637, + 421 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/f9cfa19086ea4d26d15fce64dd9d96515f34aa8bedfd1629f9cc5f660ba70dbc.jpg", + "text": "$$\n\\pmb { X } = \\pmb { U } \\pmb { \\Sigma } \\pmb { V } ^ { \\top }\n$$", + "text_format": "latex", + "bbox": [ + 450, + 426, + 549, + 444 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Since $\\ b { \\Sigma } \\in \\mathbb { R } ^ { n } \\times \\mathbb { R } ^ { n }$ and $\\boldsymbol { X }$ has a full rank, $\\pmb { \\Sigma }$ is invertible, and thus: ", + "bbox": [ + 173, + 450, + 625, + 465 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/70cdf4dfb1b4165f9d26012d5585e8c78028e46d0acb3e279bc127f8c2c78752.jpg", + "text": "$$\n\\pmb { X } \\pmb { V } \\pmb { \\Sigma } ^ { - 1 } = \\pmb { U }\n$$", + "text_format": "latex", + "bbox": [ + 446, + 470, + 552, + 488 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Now, we define $Q$ as follows: ", + "bbox": [ + 174, + 494, + 370, + 510 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/9bd0f61d5434954eb336f482c42efde0670901bd7906857e427cbf9ce9767b7d.jpg", + "text": "$$\nQ = V \\Sigma ^ { - 1 }\n$$", + "text_format": "latex", + "bbox": [ + 455, + 506, + 542, + 525 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Note that $\\mathbf { \\Delta } X Q = U$ , as desired. ", + "bbox": [ + 174, + 526, + 385, + 541 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "To find $\\kappa$ that satisfies $X K = P ^ { \\top } U$ , we use Equation (17) and require: ", + "bbox": [ + 174, + 547, + 656, + 564 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/3d72df2cc1041d92eefa2700e6a1e70c07ec67272321cf0aaab7d73fd139d2cf.jpg", + "text": "$$\nU \\Sigma V ^ { \\top } K = P ^ { \\top } U\n$$", + "text_format": "latex", + "bbox": [ + 429, + 568, + 568, + 587 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "and thus: ", + "bbox": [ + 173, + 594, + 235, + 608 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/772c658c6095e0d94634dd8eea267e0a482285f0cadc80317629464ad3ce6a70.jpg", + "text": "$$\n\\pmb { K } = \\pmb { V } \\pmb { \\Sigma } ^ { - 1 } \\pmb { U } ^ { T } \\pmb { P } ^ { \\top } \\pmb { U }\n$$", + "text_format": "latex", + "bbox": [ + 421, + 604, + 576, + 622 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We define: ", + "bbox": [ + 173, + 633, + 245, + 648 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/886128582d8bc69a605da787e1153820675d5a97d580f8b3aff8ea9ea4f27130.jpg", + "text": "$$\nz \\left( h _ { i } , h _ { j } \\right) = e \\left( h _ { i } , h _ { j } \\right) \\cdot \\sqrt { d _ { k } }\n$$", + "text_format": "latex", + "bbox": [ + 398, + 643, + 601, + 664 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Where $e$ is the attention score function of DPGAT (Equation (13)). ", + "bbox": [ + 174, + 665, + 612, + 681 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Now, for a query $i$ and a key $j$ , and the corresponding representations $h _ { i } , h _ { j }$ ", + "bbox": [ + 173, + 686, + 678, + 703 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/5bb78d4b38de8bdebc579d7a07c51cb15f221a216eada0f8ecec66a7a518907a.jpg", + "text": "$$\n\\begin{array} { r } { z \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\left( \\boldsymbol { h } _ { i } ^ { \\top } \\boldsymbol { Q } \\right) \\cdot \\left( \\boldsymbol { h } _ { j } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ { = \\left( \\boldsymbol { X } _ { i } \\boldsymbol { Q } \\right) \\cdot \\left( \\boldsymbol { X } _ { j } \\boldsymbol { K } \\right) ^ { \\top } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 390, + 708, + 607, + 756 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Since $X _ { i } Q = ( X Q ) _ { i }$ and $X _ { j } K = \\left( X K \\right) _ { j }$ , we get ", + "bbox": [ + 173, + 765, + 521, + 782 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/59a2f091ef7d23c498396a61c5a74d577cfec96df400b10cebc2cfcd7f00898b.jpg", + "text": "$$\nz \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right) = \\left( \\pmb { X } \\pmb { Q } \\right) _ { i } \\cdot \\left( \\left( \\pmb { X } \\pmb { K } \\right) _ { j } \\right) ^ { \\top } = P _ { i , j }\n$$", + "text_format": "latex", + "bbox": [ + 356, + 789, + 640, + 820 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Therefore: ", + "bbox": [ + 173, + 824, + 245, + 839 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/b0ddfd09ff9341e61d95b0c272b09198832f8e47a4984960beb2f63af1e6a774.jpg", + "text": "$$\nz \\left( \\pmb { { h } } _ { i } , \\pmb { { h } } _ { j } \\right) = \\left. \\begin{array} { l l } { 1 } & { j = \\varphi ( i ) } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right.\n$$", + "text_format": "latex", + "bbox": [ + 401, + 835, + 593, + 871 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "And thus: ", + "bbox": [ + 173, + 881, + 238, + 895 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/e74cf091b15eff9a87338446ecefff2fa8214729431ee33b2a327fe96985b788.jpg", + "text": "$$\ne ( \\pmb { h } _ { i } , \\pmb { h } _ { j } ) = \\left\\{ \\begin{array} { l l } { 1 / \\sqrt { d _ { k } } } & { j = \\varphi ( i ) } \\\\ { 0 } & { o t h e r w i s e } \\end{array} \\right.\n$$", + "text_format": "latex", + "bbox": [ + 380, + 891, + 617, + 928 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "To conclude, for every selected query $i$ and any key $\\mathscr { j } _ { \\neq \\varphi ( i ) }$ ", + "bbox": [ + 173, + 103, + 562, + 119 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/c78a2203b489ca5ecc4ed1f70e51b1ca20995f7e10ee304b4742a735b57e3b13.jpg", + "text": "$$\ne \\left( \\pmb { h } _ { i } , \\pmb { h } _ { \\varphi ( i ) } \\right) > e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 411, + 123, + 584, + 143 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "and due to the increasing monotonicity of softmax: ", + "bbox": [ + 173, + 146, + 513, + 161 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/4288e5c3d9b440fbda0b46235a3143151b8af70fc233595a40a6250b2fda7132.jpg", + "text": "$$\n\\alpha _ { i , \\varphi ( i ) } > \\alpha _ { i , j }\n$$", + "text_format": "latex", + "bbox": [ + 452, + 166, + 545, + 183 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Hence, a DPGAT layer computes dynamic attention. ", + "bbox": [ + 173, + 185, + 517, + 200 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "In the case that $d > n$ , we apply SVD to the full-rank matrix $X X ^ { \\top } \\in \\mathbb { R } ^ { n \\times n }$ , and follow the same steps to construct $Q$ and $\\kappa$ . ", + "bbox": [ + 173, + 205, + 823, + 236 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "In the case that $Q \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k } }$ and $K \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k } }$ and $d _ { k } > d$ , we can use the same $Q$ and $\\kappa$ (Equations (19) and (21)) padded with zeros. We define the $Q ^ { \\prime } \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k e y } }$ and $K ^ { \\prime } \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k e y } }$ as follows: ", + "bbox": [ + 173, + 239, + 825, + 284 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/31ad72d88f5cd93de23b90d47d003e95807d303624927413ae713c14bb4c7445.jpg", + "text": "$$\n\\begin{array} { r l } & { Q _ { i , j } ^ { \\prime } = \\left\\{ \\begin{array} { l l } { Q _ { i , j } } & { j \\leq d } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. } \\\\ & { K _ { i , j } ^ { \\prime } = \\left\\{ \\begin{array} { l l } { K _ { i , j } } & { j \\leq d } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 408, + 286, + 588, + 361 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "G.2 DPGAT IS STRICTLY WEAKER THAN GATV2 ", + "text_level": 1, + "bbox": [ + 173, + 393, + 531, + 410 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "There are examples of node representations that are linearly dependent and mappings $\\varphi$ , for which dynamic attention does not hold. First, we show a simple 2-dimensional example, and then we show the general case of such examples. ", + "bbox": [ + 173, + 420, + 825, + 463 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/b0f441cd12d6c4942d10f158b5b3baab44b607c6ac98f4f4863b6c4b09544de8.jpg", + "text": "$$\n\\prod _ { \\substack { h _ { 0 } = \\hat { x } } } ^ { y } h _ { 2 } = \\hat { x } + 2 \\hat { y }\n$$", + "text_format": "latex", + "bbox": [ + 416, + 478, + 581, + 570 + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "", + "image_caption": [ + "Figure 6: An example for node representations that are linearly dependent, for which DPGAT cannot compute dynamic attention, because no query vector $q \\in \\mathbb { R } ^ { 2 }$ can “select” $\\pmb { h } _ { 1 }$ . " + ], + "image_footnote": [], + "page_idx": 16 + }, + { + "type": "text", + "text": "Consider the following linearly dependent set of vectors $\\mathbb { K } = \\mathbb { Q }$ (Figure 6): ", + "bbox": [ + 173, + 626, + 668, + 641 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/a79e708698104ffef2ca0bc52c70500f9ea1b7a34cbabe95cef0ae1a5d18972f.jpg", + "text": "$$\n\\begin{array} { l } { { \\pmb { h } } _ { 0 } = { \\pmb { \\hat { x } } } } \\\\ { { \\pmb { h } } _ { 1 } = { \\pmb { \\hat { x } } } + { \\pmb { \\hat { y } } } } \\\\ { { \\pmb { h } } _ { 2 } = { \\pmb { \\hat { x } } } + 2 { \\pmb { \\hat { y } } } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 450, + 643, + 547, + 696 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "where $\\hat { \\pmb x }$ and $\\hat { \\pmb { y } }$ are the cartesian unit vectors. We define $\\beta \\in \\{ 0 , 1 , 2 \\}$ to express $\\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \\}$ using the same expression: ", + "bbox": [ + 173, + 700, + 825, + 728 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/575a2fa2b0ec9c91e579d6ce367d482b1b99ee7811ede8beb83e8868904546f6.jpg", + "text": "$$\n\\begin{array} { r } { \\pmb { h } _ { \\beta } = \\hat { \\pmb { x } } + \\beta \\hat { \\pmb y } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 450, + 727, + 549, + 744 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Let $\\pmb q \\in \\mathbb { Q }$ be any query vector. For brevity, we define the unscaled dot-product attention score as $s$ ", + "bbox": [ + 174, + 746, + 818, + 761 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/821407b7531cafa4a14e43215bbe169255bfb4789c144f517180083a952b491d.jpg", + "text": "$$\ns \\left( \\mathbf { q } , \\boldsymbol { h } _ { \\beta } \\right) = e \\left( \\mathbf { q } , \\boldsymbol { h } _ { \\beta } \\right) \\cdot \\sqrt { d _ { k } }\n$$", + "text_format": "latex", + "bbox": [ + 405, + 763, + 594, + 785 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Where $e$ is the attention score function of DPGAT (Equation (13)). The (unscaled) attention score between $\\pmb q$ and $\\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \\}$ is: ", + "bbox": [ + 174, + 787, + 825, + 818 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/2f076942dab3212541d4295ee604f1a329f957ef523d751087c6b0cd4b687f9c.jpg", + "text": "$$\n\\begin{array} { r l } & { s \\left( \\pmb { q } , \\pmb { h } _ { \\beta } \\right) = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { h } _ { \\beta } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\left( \\hat { \\pmb { x } } + \\beta \\pmb { \\hat { y } } \\right) ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { x } } ^ { \\top } \\pmb { K } + \\beta \\pmb { \\hat { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { x } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } + \\beta \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { \\hat { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 321, + 819, + 674, + 926 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "The first term $\\left( { { \\pmb q } ^ { \\top } } { \\pmb Q } \\right) \\left( \\hat { { \\pmb x } } ^ { \\top } { \\pmb K } \\right) ^ { \\top }$ is unconditioned on $\\beta$ , and thus shared for every $h _ { \\beta }$ . Let us focus on the second term $\\beta \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top }$ . I f $\\left( { { \\pmb q } ^ { \\top } } { \\pmb Q } \\right) \\left( \\hat { \\pmb y } ^ { \\top } { \\pmb K } \\right) ^ { \\top } > 0$ , then: ", + "bbox": [ + 174, + 101, + 825, + 140 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/6f622f464bc3ef858200573754232286c63e8529d0ef86327050512c98b3b6bb.jpg", + "text": "$$\ne \\left( \\pmb { q } , \\pmb { h } _ { 2 } \\right) > e \\left( \\pmb { q } , \\pmb { h } _ { 1 } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 428, + 146, + 570, + 164 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Otherwise, if $\\left( { { { q } ^ { \\top } } { { Q } } } \\right) \\left( { \\hat { { { y } } } ^ { \\top } } K \\right) ^ { \\top } \\le 0$ ", + "bbox": [ + 174, + 178, + 416, + 198 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/6d653bd6bf5e32345687239b6ae3b5ad8201768e8ab8d25482a8e6250c8af878.jpg", + "text": "$$\ne \\left( \\pmb { q } , \\pmb { h } _ { 0 } \\right) \\geq e \\left( \\pmb { q } , \\pmb { h } _ { 1 } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 428, + 203, + 570, + 220 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Thus, for any query $\\pmb q$ , the key $\\pmb { h } _ { 1 }$ can never get the highest score, and thus cannot be “selected”. That is, the key $\\pmb { h } _ { 1 }$ cannot satisfy that $e \\left( \\pmb q , \\pmb h _ { 1 } \\right)$ is strictly greater than any other key. ", + "bbox": [ + 171, + 227, + 823, + 256 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "In the general case, let $\\boldsymbol { h } _ { 0 } , \\boldsymbol { h } _ { 1 } \\in \\mathbb { R } ^ { d }$ be some non-zero vectors , and $\\lambda$ is some scalar such that $0 < \\lambda < 1$ . ", + "bbox": [ + 174, + 261, + 825, + 290 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Consider the following linearly dependent set of vectors: ", + "bbox": [ + 173, + 296, + 547, + 311 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/e8623e4e53f6695803304e77a9cbc56900b6b95c8cdcc5a075a4efdbe5d2ad8f.jpg", + "text": "$$\n\\mathbb { K } = \\mathbb { Q } = \\{ \\beta \\pmb { h } _ { 1 } + \\left( 1 - \\beta \\right) \\pmb { h } _ { 0 } \\ \\lvert \\ \\beta \\in \\{ 0 , \\lambda , 1 \\} \\}\n$$", + "text_format": "latex", + "bbox": [ + 343, + 318, + 653, + 335 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "For any query $\\pmb q \\in \\mathbb { Q }$ and $\\beta \\in \\{ 0 , \\lambda , 1 \\}$ we define: ", + "bbox": [ + 171, + 348, + 503, + 364 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/d12de243a09f1d853bf35794af7dc29a626a2511b35a198be04de03bb86289bc.jpg", + "text": "$$\ns \\left( \\mathbf { q } , \\beta \\right) = e \\left( \\mathbf { q } , \\left( \\beta \\pmb { h } _ { 1 } + \\left( 1 - \\beta \\right) \\pmb { h } _ { 0 } \\right) \\right) \\cdot \\sqrt { d _ { k } }\n$$", + "text_format": "latex", + "bbox": [ + 352, + 369, + 647, + 390 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Where $e$ is the attention score function of DPGAT (Equation (13)). ", + "bbox": [ + 173, + 396, + 611, + 411 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Therefore: ", + "bbox": [ + 173, + 417, + 245, + 431 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/7c0be19659c61b65b84ce7167e42f5a668f14cc33005c85e5794f45a13743e70.jpg", + "text": "$$\n\\begin{array} { r l } & { s \\left( \\boldsymbol { q } , \\beta \\right) = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\left( \\beta h _ { 1 } + \\left( 1 - \\beta \\right) h _ { 0 } \\right) ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta h _ { 1 } ^ { \\top } \\boldsymbol { K } + \\left( 1 - \\beta \\right) h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta h _ { 1 } ^ { \\top } \\boldsymbol { K } + h _ { 0 } ^ { \\top } \\boldsymbol { K } - \\beta h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta \\left( h _ { 1 } ^ { \\top } \\boldsymbol { K } - h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) + h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\beta \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( h _ { 1 } ^ { \\top } \\boldsymbol { K } - h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } + \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 297, + 435, + 699, + 564 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "If $\\left( { { q } ^ { \\top } } Q \\right) \\left( h _ { 1 } ^ { \\top } K - h _ { 0 } ^ { \\top } K \\right) ^ { \\top } > 0 .$ ", + "bbox": [ + 173, + 574, + 401, + 594 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/deb4a1feed49bffb7979a98bb27d95c6c0feddf5d85bf23d47b3859a4361738c.jpg", + "text": "$$\ne \\left( \\pmb q , \\pmb h _ { 1 } \\right) > e \\left( \\pmb q , \\pmb h _ { \\lambda } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 428, + 601, + 570, + 618 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Otherwise, if $\\left( { \\pmb q } ^ { \\top } { \\pmb Q } \\right) \\left( { \\pmb h } _ { 1 } ^ { \\top } { \\pmb K } - { \\pmb h } _ { 0 } ^ { \\top } { \\pmb K } \\right) ^ { \\top } \\le 0 ,$ ", + "bbox": [ + 176, + 632, + 475, + 652 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/3a6c3dc477df43b80ebcf249caa56c66f7e2f99c20283154da574924e7bb9913.jpg", + "text": "$$\ne \\left( \\pmb { q } , \\pmb { h } _ { 0 } \\right) \\geq e \\left( \\pmb { q } , \\pmb { h } _ { \\lambda } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 428, + 657, + 570, + 676 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Thus, for any query $\\pmb q$ , the key $h _ { \\lambda }$ cannot be selected. That is, the key $\\mathbf { \\delta } _ { h _ { \\lambda } }$ cannot satisfy that $\\mathbf { \\nabla } _ { \\boldsymbol { \\mathcal { C } } } \\left( \\mathbf { \\nabla } _ { \\mathbf { \\boldsymbol { q } } } , \\boldsymbol { h } _ { \\lambda } \\right)$ is strictly greater than any other key. Therefore, there are mappings $\\varphi$ , for which dynamic attention does not hold. ", + "bbox": [ + 174, + 681, + 825, + 723 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "While we prove that GATv2 computes dynamic attention (Appendix B) for any set of node representations $\\mathbb { K } = \\mathbb { Q }$ , there are sets of node representations and mappings $\\varphi$ for which dynamic attention does not hold for DPGAT. Thus, DPGAT is strictly weaker than GATv2. ", + "bbox": [ + 173, + 729, + 825, + 773 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "G.3 EMPIRICAL EVALUATION ", + "text_level": 1, + "bbox": [ + 174, + 789, + 393, + 804 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Here we repeat the experiments of Section 4 with DPGAT. We remind that DPGAT is strictly weaker than our proposed GATv2 (see a proof in Appendix G.1). ", + "bbox": [ + 173, + 814, + 825, + 844 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "H STATISTICAL SIGNIFICANCE ", + "text_level": 1, + "bbox": [ + 176, + 863, + 444, + 881 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Here we report the statistical significance of the strongest GATv2 and GAT models of the experiments reported in Section 4. ", + "bbox": [ + 173, + 895, + 823, + 924 + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/0f010beab366169ea6709265d4f0cb66efdab3acea2d3044a4c540073bc64969.jpg", + "image_caption": [ + "Figure 7: Test accuracy compared to the noise ratio: GATv2 and DPGAT are more robust to structural noise compared to GAT. Each point is an average of 10 runs, error bars show standard deviation. " + ], + "image_footnote": [], + "bbox": [ + 173, + 131, + 820, + 361 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/3729df644b0459c1ee338b4a1a91e066025f669f0d9f19b18d44d5b1a4c61729.jpg", + "table_caption": [ + "Table 11: Accuracy (5 runs±stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both test sets, using GAT’s hyperparameters. $\\dagger$ – previously reported by Brockschmidt (2020). " + ], + "table_footnote": [], + "table_body": "
ModelSeenProjUnseenProj
No- AttentionGCN+87.2±1.581.4±2.3
GIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
DPGAT88.0±0.881.5±1.2
GATv288.0±1.182.8±1.7
", + "bbox": [ + 356, + 508, + 642, + 613 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "(a) ", + "bbox": [ + 428, + 720, + 447, + 733 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/d2839ce14e5ce1e13e771cee93c7a8c132d130d89277bbce60c1fdfd0f5ae22c.jpg", + "table_caption": [ + "Table 12: Average accuracy (Table 12a) and ROC-AUC (Table 12b) in node-prediction datasets (10 runs±std). In all datasets, GATv2 outperforms GAT. $\\dagger -$ previously reported by Hu et al. (2020). " + ], + "table_footnote": [], + "table_body": "
ModelAttn. Headsogbn-arxivogbn-productsogbn-mag
GCN+071.74±0.2978.97±0.3330.43±0.25
GraphSAGE†071.49±0.2778.70±0.3631.53±0.15
GAT171.59±0.3879.04±1.5432.20±1.46
871.54±0.3077.23±2.3731.75±1.60
DPGAT171.52±0.1776.49±0.7832.77±0.80
871.48±0.2673.53±0.4727.74±9.97
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.44
871.87±0.2578.46±2.4532.52±0.39
", + "bbox": [ + 183, + 738, + 691, + 892 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "(b) ", + "bbox": [ + 750, + 720, + 769, + 733 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/4f095a99f6be29e043b71844162f7fb398e89952546ed4f514e85aa387699ba7.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
ogbn-proteins
72.51±0.3577.68±0.20
70.77±5.7978.63±1.62
63.47±2.7972.88±0.59
77.23±3.3279.52±0.55
", + "bbox": [ + 705, + 739, + 813, + 892 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/424d7314241e36f659f72bb1b181d1a8a5dd7c63cd357d190033bd0de94283ab.jpg", + "table_caption": [ + "Table 13: Average error rates (lower is better), 5 runs $\\pm$ standard deviation for each property, on the QM9 dataset. The best result among GAT, GATv2 and DPGAT is marked in bold; the globally best result among all GNNs is marked in bold and underline. $\\dagger$ was previously tuned and reported by Brockschmidt (2020). " + ], + "table_footnote": [], + "table_body": "
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
DPGAT8h2.63±0.094.37±0.131.44±0.071.40±0.032.10±0.0732.59±34.7711.66±1.00
DPGAT1h3.20±0.178.35±0.781.71±0.032.17±0.142.88±0.1225.21±2.8665.79±39.84
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13 ±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.259.05±1.217.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h28.10±16.4518.71±23.365.62±0.8110.80±2.185.37±0.26
GAT8ht7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.364.11±0.273.11±0.14 1.48±0.87+134.1% +0%
DPGAT1h12.93±1.7013.83±2.556.37±0.28
DPGAT8h6.95±0.3213.32±2.39 7.09±0.5914.42±1.95 7.30±0.666.52±0.613.76±0.213.28±1.16 1.18±0.33+77.9% -9.7%
GATv21h5.23±0.36
GATv28h10.19±0.6322.56±17.4615.04±4.5822.94±17.342.46±0.65+91.6%
6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
", + "bbox": [ + 187, + 169, + 808, + 517 + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/a9109830d2a6f78861878e365b1f89ef70a5d28fb8341f7508b1337e0e7390a2.jpg", + "image_caption": [ + "Figure 8: Test accuracy and statistical significance compared to the noise ratio: GATv2 is more robust to structural noise compared to GAT. Each point is an average of 10 runs, error bars show standard deviation. " + ], + "image_footnote": [], + "bbox": [ + 176, + 526, + 810, + 742 + ], + "page_idx": 19 + }, + { + "type": "table", + "img_path": "images/f92804deb5a5a72316e07d2be17398049bcfdc6e7f42478ed3d24944fc5f89a8.jpg", + "table_caption": [ + "Table 14: Accuracy (5 runs $\\pm$ stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both test sets, using GAT’s hyperparameters. $\\dagger$ – previously reported by Brockschmidt (2020). " + ], + "table_footnote": [], + "table_body": "
ModelSeenProjUnseenProj
GAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
p-value0.0480.049
", + "bbox": [ + 392, + 845, + 602, + 921 + ], + "page_idx": 19 + }, + { + "type": "table", + "img_path": "images/ae6edb6e2129580eec5eabb7cf79e1349f07ea69aed3861db50a0c1f92647b30.jpg", + "table_caption": [ + "Table 15: Accuracy (100 runs±stdev) on Pubmed. GATv2 is more accurate than GAT. " + ], + "table_footnote": [], + "table_body": "
Model Accuracy
GAT 78.1±0.59
GATv2 78.5±0.38
p-value <0.0001
", + "bbox": [ + 437, + 150, + 560, + 226 + ], + "page_idx": 20 + }, + { + "type": "table", + "img_path": "images/e60aeca2e5c137c25a39bb0dc4a695e603324776802578000663be86004f00e4.jpg", + "table_caption": [ + "Table 16: Average accuracy (Table 16a) and ROC-AUC (Table 16b) in node-prediction datasets (30 runs±std). We report on the best GAT / GATv2 from Table 2. " + ], + "table_footnote": [], + "table_body": "
(a)
Modelogbn-arxivogbn-productsogbn-mag
GAT71.65±0.3879.04±1.5432.36±1.10
GATv271.93±0.3580.63±0.7033.01±0.41
p-value0.0022<0.00010.0018
", + "bbox": [ + 261, + 320, + 612, + 415 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "(b) ", + "bbox": [ + 750, + 321, + 769, + 333 + ], + "page_idx": 20 + }, + { + "type": "table", + "img_path": "images/d8f3b42ff0eb2de1cdfe97747979be35ef59ee5710e92ab69714a713d6fb8ef3.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
ogbn-proteins
78.29 ±1.59
78.96±1.19 0.0349
", + "bbox": [ + 705, + 340, + 815, + 415 + ], + "page_idx": 20 + }, + { + "type": "table", + "img_path": "images/5e6e77c7ba89b1b9e1e81278e67fdd659e0c880740ef27e8eaf40978318fb84c.jpg", + "table_caption": [ + "Table 17: Average Hits $\\ @ 5 0$ (Table 17a) and mean reciprocal rank (MRR) (Table 17b) in linkprediction benchmarks from OGB (30 runs±std). We report on the best GAT / GATv2 from Table 8. " + ], + "table_footnote": [], + "table_body": "
(a)
Modelogbl-collab w/o val edges w/ val edges
GAT42.24±2.2646.02±4.09
GATv243.82±2.2449.06±2.50
p-value0.00430.0005
", + "bbox": [ + 315, + 511, + 576, + 617 + ], + "page_idx": 20 + }, + { + "type": "table", + "img_path": "images/ffa94db2bcee9b7ee4a6ee106d5e4ed602398ac903c08866ca032fe1aa1c88ab.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
(b)
ogbl-citation2
79.91±0.13
80.20±0.62
0.0075
", + "bbox": [ + 643, + 511, + 754, + 617 + ], + "page_idx": 20 + }, + { + "type": "table", + "img_path": "images/30a6b0c5270661bcc3a376e72370e2090929f0518cf0b3e006ae13065159783b.jpg", + "table_caption": [ + "Table 18: Average error rates (lower is better), 20 runs $\\pm$ standard deviation for each property, on the QM9 dataset. We report on GAT and GATv2 with 8 attention heads. " + ], + "table_footnote": [], + "table_body": "
Predicted Property
Model1234567
GAT2.74±0.084.73±0.401.47±0.061.53±0.062.44±0.6055.21±42.3325.36±31.42
GATv22.67±0.084.28±0.231.43±0.051.51±0.072.21±0.0816.64± 1.1713.61±1.68
p-value0.0043<0.00010.01380.16910.04870.00010.0516
Predicted Property
Model8910111213
GAT GATv27.36±0.87 6.13±0.596.79±0.86 6.33±0.827.36±0.936.69±0.864.10±0.291.51±0.84
6.37±0.865.95±0.623.66±0.291.09±0.85
p-value<0.00010.04580.00060.0017<0.00010.0621
", + "bbox": [ + 200, + 712, + 797, + 897 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "I COMPLEXITY ANALYSIS ", + "text_level": 1, + "bbox": [ + 174, + 102, + 406, + 118 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "We repeat the definitions of GAT, GATv2 and DPGAT: ", + "bbox": [ + 173, + 137, + 534, + 154 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "GAT (Velickovi ˇ c et al., 2018): ´ ", + "bbox": [ + 207, + 171, + 411, + 186 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "GATv2 (our fixed version): ", + "bbox": [ + 207, + 191, + 390, + 207 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "DPGAT (Vaswani et al., 2017): ", + "bbox": [ + 207, + 215, + 421, + 231 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/8a5e123ec02bb5ae8d73a9dbea17509e9c23670171b8eadac52309e582e12309.jpg", + "text": "$$\n\\begin{array} { r l } & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = \\mathrm { L e a k y R e L U } ( \\boldsymbol { a } ^ { \\top } \\cdot [ \\boldsymbol { W } \\boldsymbol { h } _ { i } | | \\boldsymbol { W } \\boldsymbol { h } _ { j } | ) } \\\\ & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = \\boldsymbol { a } ^ { \\top } \\mathrm { L e a k y R e L U } ( \\boldsymbol { W } \\cdot [ \\boldsymbol { h } _ { i } | | \\boldsymbol { h } _ { j } ] ) } \\\\ & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = ( ( \\boldsymbol { h } _ { i } ^ { \\top } \\boldsymbol { Q } ) \\cdot ( \\boldsymbol { h } _ { j } ^ { \\top } \\boldsymbol { K } ) ^ { \\top } ) / \\sqrt { d ^ { \\prime } } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 449, + 167, + 763, + 238 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "I.1 TIME COMPLEXITY ", + "text_level": 1, + "bbox": [ + 174, + 261, + 346, + 275 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "GAT As noted by Velickovi ˇ c et al. (2018), the time complexity of a single GAT head may be ´ expressed as $\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)$ . Because of GAT’s static attention, this computation can be further optimized, by merging the linear layer $\\mathbf { \\delta } _ { \\mathbf { a } _ { 1 } }$ with $W$ , merging $\\mathbf { a } _ { 2 }$ with $W$ , and only then compute $a _ { \\{ 1 , 2 \\} } ^ { \\top } W h _ { i }$ for every $i \\in \\mathcal V$ . ", + "bbox": [ + 173, + 294, + 825, + 352 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "GATv2 require the same computational cost as GAT’s declared complexity: $\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)$ : we denote $W = [ W _ { 1 } \\Vert W _ { 2 } ]$ , where $W _ { 1 } \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }$ and $W _ { 2 } ^ { d ^ { \\prime } \\times d }$ contain the left half and right half of the columns of $W$ , respectively. We can first compute $W _ { 1 } h _ { i }$ and $W _ { 2 } h _ { j }$ for every $i , j \\in \\mathcal { V }$ . This takes $\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } \\right)$ . ", + "bbox": [ + 173, + 363, + 825, + 424 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Then, for every edge $( j , i )$ , we compute LeakyReLU $\\left( W \\cdot \\left[ \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { h } _ { j } \\right] \\right)$ using the precomputed $W _ { 1 } h _ { i }$ and $W _ { 2 } h _ { j }$ , since $W \\cdot [ h _ { i } \\| h _ { j } ] = W _ { 1 } h _ { i } + W _ { 2 } h _ { j }$ . This takes $\\mathcal { O } \\left( \\vert \\mathcal { E } \\vert d ^ { \\prime } \\right)$ . ", + "bbox": [ + 173, + 429, + 823, + 460 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Finally, computing the results of the linear layer $^ { a }$ takes additional $\\mathcal { O } \\left( | \\mathcal { E } | d ^ { \\prime } \\right)$ time, and overall $\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + \\bar { | } \\mathcal { E } | d ^ { \\prime } \\right)$ . ", + "bbox": [ + 171, + 465, + 823, + 494 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "DPGAT also takes the same time. We can first compute $\\pmb { h } _ { i } ^ { \\top } \\pmb { Q }$ and $\\pmb { h } _ { j } ^ { \\top } \\pmb { K }$ for every $i , j \\in \\mathcal { V }$ . This takes $\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } \\right)$ . Computing the dot-product $\\left( \\pmb { h } _ { i } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { h } _ { j } ^ { \\top } \\pmb { K } \\right) ^ { \\top }$ for every edge $( j , i )$ takes additional $\\mathcal { O } \\left( | \\mathcal { E } | d ^ { \\prime } \\right)$ time, and overall $\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)$ . ", + "bbox": [ + 173, + 503, + 825, + 556 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "I.2 PARAMETRIC COMPLEXITY ", + "text_level": 1, + "bbox": [ + 174, + 580, + 401, + 595 + ], + "page_idx": 21 + }, + { + "type": "table", + "img_path": "images/7f473a828617dbd5ae061ebb1adaf2898962cae8e0636392f224b1bdad27d710.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
GATGATv2DPGAT
Official2d'+dd'd'+ 2dd'2ddk +dd'
In our experiments2d'+dd'd'+dd'2dd'
", + "bbox": [ + 295, + 622, + 702, + 683 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Table 19: Number of parameters for each GNN type, in a single layer and a single attention head. ", + "bbox": [ + 179, + 690, + 816, + 707 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "All parametric costs are summarized in Table 19. All following calculations refer to a single layer having a single attention head, omitting bias vectors. ", + "bbox": [ + 174, + 732, + 820, + 761 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "GAT has learned vector and a matrix: $\\pmb { a } \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }$ and $W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }$ , thus overall $2 d ^ { \\prime } + d d ^ { \\prime }$ learned parameters. ", + "bbox": [ + 174, + 772, + 823, + 803 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "GATv2 has a matrix that is twice larger: $W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times 2 d }$ , because it is applied on the concatenation $[ h _ { i } | | h _ { j } ]$ . Thus, the overall number of learned parameters is $d ^ { \\prime } + 2 d d ^ { \\prime }$ . However in our experiments, to rule out the increased number of parameters over GAT as the source of empirical difference, we constrained $W = [ W ^ { \\prime } | | W ^ { \\prime } ]$ , and thus the number of parameters were $d ^ { \\prime } + d d ^ { \\prime }$ . ", + "bbox": [ + 174, + 813, + 825, + 872 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "DPGAT has $Q$ and $\\kappa$ matrices of sizes $d d _ { k }$ each, and additional $d d ^ { \\prime }$ parameters in the value matrix $V$ , thus $2 d d _ { k } + d d ^ { \\prime }$ parameters overall. However in our experiments, we constrained $Q = K$ and set $d _ { k } = d ^ { \\prime }$ , and thus the number of parameters is only $2 d d ^ { \\prime }$ . ", + "bbox": [ + 174, + 881, + 823, + 924 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 174, + 103, + 287, + 117 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Miltiadis Allamanis, Marc Brockschmidt, and Mahmoud Khademi. Learning to represent programs with graphs. In International Conference on Learning Representations, 2018. URL https://openreview.net/ forum?id $\\equiv$ BJOFETxR-. ", + "bbox": [ + 173, + 125, + 825, + 164 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Uri Alon and Eran Yahav. On the bottleneck of graph neural networks and its practical implications. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id $=$ i80OPhOCVH2. ", + "bbox": [ + 171, + 172, + 823, + 212 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "James Atwood and Don Towsley. Diffusion-convolutional neural networks. In Advances in neural information processing systems, pages 1993–2001, 2016. ", + "bbox": [ + 173, + 222, + 821, + 250 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. CoRR, abs/1409.0473, 2014. URL http://arxiv.org/abs/1409.0473. ", + "bbox": [ + 171, + 257, + 823, + 285 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Peter Battaglia, Razvan Pascanu, Matthew Lai, Danilo Jimenez Rezende, and Koray kavukcuoglu. Interaction networks for learning about objects, relations and physics. In Proceedings of the 30th International Conference on Neural Information Processing Systems, pages 4509–4517, 2016. ", + "bbox": [ + 173, + 292, + 823, + 334 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Marc Brockschmidt. Gnn-film: Graph neural networks with feature-wise linear modulation. Proceedings of the 36th International Conference on Machine Learning, ICML, 2020. URL https://github.com/ microsoft/tf-gnn-samples. ", + "bbox": [ + 174, + 342, + 825, + 382 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Michael M Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst. Geometric deep learning: going beyond euclidean data. IEEE Signal Processing Magazine, 34(4):18–42, 2017. ", + "bbox": [ + 171, + 390, + 823, + 419 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Michael M. Bronstein, Joan Bruna, Taco Cohen, and Petar Velickovi ˇ c. Geometric deep learning: Grids, groups, ´ graphs, geodesics, and gauges, 2021. ", + "bbox": [ + 173, + 426, + 823, + 454 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Dan Busbridge, Dane Sherburn, Pietro Cavallo, and Nils Y Hammerla. Relational graph attention networks. arXiv preprint arXiv:1904.05811, 2019. ", + "bbox": [ + 173, + 463, + 823, + 491 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "George Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of control, signals and systems, 2(4):303–314, 1989. ", + "bbox": [ + 173, + 498, + 825, + 526 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Misha Denil, Sergio Gómez Colmenarejo, Serkan Cabi, David Saxton, and Nando de Freitas. Programmable agents. arXiv preprint arXiv:1706.06383, 2017. ", + "bbox": [ + 173, + 535, + 823, + 563 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Yan Duan, Marcin Andrychowicz, Bradly Stadie, Jonathan Ho, Jonas Schneider, Ilya Sutskever, Pieter Abbeel, and Wojciech Zaremba. One-shot imitation learning. In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 1087–1098, 2017. ", + "bbox": [ + 174, + 570, + 823, + 611 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "David K Duvenaud, Dougal Maclaurin, Jorge Iparraguirre, Rafael Bombarell, Timothy Hirzel, Alán AspuruGuzik, and Ryan P Adams. Convolutional networks on graphs for learning molecular fingerprints. In Advances in neural information processing systems, pages 2224–2232, 2015. ", + "bbox": [ + 174, + 619, + 823, + 660 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Vijay Prakash Dwivedi and Xavier Bresson. A generalization of transformer networks to graphs. arXiv preprint arXiv:2012.09699, 2020. ", + "bbox": [ + 173, + 667, + 823, + 695 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Vijay Prakash Dwivedi, Chaitanya K Joshi, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. Benchmarking graph neural networks. arXiv preprint arXiv:2003.00982, 2020. ", + "bbox": [ + 173, + 704, + 823, + 731 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Matthias Fey and Jan E. Lenssen. Fast graph representation learning with PyTorch Geometric. In ICLR Workshop on Representation Learning on Graphs and Manifolds, 2019. ", + "bbox": [ + 173, + 741, + 825, + 767 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Ken-Ichi Funahashi. On the approximate realization of continuous mappings by neural networks. Neural networks, 2(3):183–192, 1989. ", + "bbox": [ + 173, + 776, + 825, + 804 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Hongyang Gao and Shuiwang Ji. Graph representation learning via hard and channel-wise attention networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 741–749, 2019. ", + "bbox": [ + 173, + 813, + 825, + 852 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural message passing for quantum chemistry. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 1263–1272. JMLR. org, 2017. ", + "bbox": [ + 171, + 861, + 825, + 901 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Aleksa Gordic. pytorch-gat. ´ https://github.com/gordicaleksa/pytorch-GAT, 2020. ", + "bbox": [ + 168, + 910, + 766, + 924 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Marco Gori, Gabriele Monfardini, and Franco Scarselli. A new model for learning in graph domains. In Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., volume 2, pages 729– 734. IEEE, 2005. ", + "bbox": [ + 174, + 104, + 825, + 143 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. In Advances in neural information processing systems, pages 1024–1034, 2017. ", + "bbox": [ + 171, + 151, + 823, + 178 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Kurt Hornik. Approximation capabilities of multilayer feedforward networks. Neural networks, 4(2):251–257, 1991. ", + "bbox": [ + 174, + 185, + 823, + 212 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators. Neural networks, 2(5):359–366, 1989. ", + "bbox": [ + 173, + 219, + 823, + 247 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Yedid Hoshen. Vain: attentional multi-agent predictive modeling. In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 2698–2708, 2017. ", + "bbox": [ + 173, + 255, + 823, + 281 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open graph benchmark: Datasets for machine learning on graphs. arXiv preprint arXiv:2005.00687, 2020. ", + "bbox": [ + 176, + 289, + 825, + 327 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Binxuan Huang and Kathleen M Carley. Syntax-aware aspect level sentiment classification with graph attention networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5472–5480, 2019. ", + "bbox": [ + 174, + 335, + 826, + 387 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Qian Huang, Horace He, Abhay Singh, Ser-Nam Lim, and Austin Benson. Combining label propagation and simple models out-performs graph neural networks. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=8E1-f3VhX1o. ", + "bbox": [ + 176, + 395, + 823, + 435 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Chaitanya Joshi. Transformers are graph neural networks. The Gradient, 2020. ", + "bbox": [ + 173, + 443, + 640, + 457 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Dongkwan Kim and Alice Oh. How to find your friendly neighborhood: Graph attention design with self-supervision. In International Conference on Learning Representations, 2021. URL https:// openreview.net/forum?id $=$ Wi5KUNlqWty. ", + "bbox": [ + 174, + 464, + 820, + 503 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In ICLR, 2017. ", + "bbox": [ + 174, + 511, + 823, + 537 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Vineet Kosaraju, Amir Sadeghian, Roberto Martín-Martín, Ian Reid, Hamid Rezatofighi, and Silvio Savarese. Social-bigat: Multimodal trajectory forecasting using bicycle-gan and graph attention networks. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. URL https://proceedings. neurips.cc/paper/2019/file/d09bf41544a3365a46c9077ebb5e35c3-Paper.pdf. ", + "bbox": [ + 174, + 546, + 826, + 611 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "John Boaz Lee, Ryan A Rossi, Sungchul Kim, Nesreen K Ahmed, and Eunyee Koh. Attention models in graphs: A survey. arXiv preprint arXiv:1807.07984, 2018. ", + "bbox": [ + 173, + 617, + 825, + 645 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Moshe Leshno, Vladimir Ya Lin, Allan Pinkus, and Shimon Schocken. Multilayer feedforward networks with a nonpolynomial activation function can approximate any function. Neural networks, 6(6):861–867, 1993. ", + "bbox": [ + 173, + 652, + 823, + 680 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Guohao Li, Matthias Muller, Ali Thabet, and Bernard Ghanem. Deepgcns: Can gcns go as deep as cnns? In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9267–9276, 2019. ", + "bbox": [ + 174, + 686, + 825, + 714 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for semisupervised learning. In Thirty-Second AAAI Conference on Artificial Intelligence, 2018. ", + "bbox": [ + 174, + 720, + 821, + 748 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Yujia Li, Daniel Tarlow, Marc Brockschmidt, and Richard Zemel. Gated graph sequence neural networks. In International Conference on Learning Representations, 2016. ", + "bbox": [ + 174, + 756, + 823, + 784 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Denis Lukovnikov and Asja Fischer. Gated relational graph attention networks, 2021. URL https:// openreview.net/forum?id=v-9E8egy_i. ", + "bbox": [ + 173, + 790, + 825, + 818 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Thang Luong, Hieu Pham, and Christopher D. Manning. Effective approaches to attention-based neural machine translation. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, EMNLP 2015, Lisbon, Portugal, September 17-21, 2015, pages 1412–1421, 2015. URL http: //aclweb.org/anthology/D/D15/D15-1166.pdf. ", + "bbox": [ + 174, + 825, + 825, + 877 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Nianzu Ma, Sahisnu Mazumder, Hao Wang, and Bing Liu. Entity-aware dependency-based deep graph attention network for comparative preference classification. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5782–5788, 2020. ", + "bbox": [ + 174, + 885, + 825, + 924 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Federico Monti, Davide Boscaini, Jonathan Masci, Emanuele Rodola, Jan Svoboda, and Michael M Bronstein. Geometric deep learning on graphs and manifolds using mixture model cnns. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5115–5124, 2017. ", + "bbox": [ + 176, + 104, + 825, + 143 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Deepak Nathani, Jatin Chauhan, Charu Sharma, and Manohar Kaul. Learning attention-based embeddings for relation prediction in knowledge graphs. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4710–4723, 2019. ", + "bbox": [ + 176, + 151, + 821, + 190 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Sejun Park, Chulhee Yun, Jaeho Lee, and Jinwoo Shin. Minimum width for universal approximation. In International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id $\\equiv$ O-XJwyoIF-k. ", + "bbox": [ + 173, + 199, + 823, + 238 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Allan Pinkus. Approximation theory of the mlp model. Acta Numerica 1999: Volume 8, 8:143–195, 1999. ", + "bbox": [ + 169, + 246, + 802, + 261 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Jiezhong Qiu, Jian Tang, Hao Ma, Yuxiao Dong, Kuansan Wang, and Jie Tang. Deepinf: Social influence prediction with deep learning. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD’18), 2018. ", + "bbox": [ + 174, + 267, + 821, + 308 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum chemistry structures and properties of 134 kilo molecules. Scientific data, 1:140022, 2014. ", + "bbox": [ + 171, + 314, + 823, + 342 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Yu Rong, Yatao Bian, Tingyang Xu, Weiyang Xie, Ying Wei, Wenbing Huang, and Junzhou Huang. Selfsupervised graph transformer on large-scale molecular data. Advances in Neural Information Processing Systems, 33, 2020a. ", + "bbox": [ + 173, + 349, + 825, + 388 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Yu Rong, Wenbing Huang, Tingyang Xu, and Junzhou Huang. Dropedge: Towards deep graph convolutional networks on node classification. In International Conference on Learning Representations, 2020b. URL https://openreview.net/forum?id $=$ Hkx1qkrKPr. ", + "bbox": [ + 174, + 397, + 823, + 435 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Adam Santoro, David Raposo, David GT Barrett, Mateusz Malinowski, Razvan Pascanu, Peter Battaglia, and Timothy Lillicrap. A simple neural network module for relational reasoning. In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 4974–4983, 2017. ", + "bbox": [ + 174, + 444, + 823, + 483 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. The graph neural network model. IEEE Transactions on Neural Networks, 20(1):61–80, 2008. ", + "bbox": [ + 171, + 491, + 821, + 518 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Galligher, and Tina Eliassi-Rad. Collective classification in network data. AI magazine, 29(3):93–93, 2008. ", + "bbox": [ + 173, + 525, + 823, + 553 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Yunsheng Shi, Zhengjie Huang, Shikun Feng, and Yu Sun. Masked label prediction: Unified massage passing model for semi-supervised classification. arXiv preprint arXiv:2009.03509, 2020. ", + "bbox": [ + 176, + 560, + 823, + 588 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Kiran K Thekumparampil, Chong Wang, Sewoong Oh, and Li-Jia Li. Attention-based graph neural network for semi-supervised learning. arXiv preprint arXiv:1803.03735, 2018. ", + "bbox": [ + 171, + 594, + 823, + 622 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, pages 6000–6010, 2017. ", + "bbox": [ + 173, + 628, + 823, + 669 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Petar Velickovi ˇ c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph ´ attention networks. In International Conference on Learning Representations, 2018. ", + "bbox": [ + 171, + 676, + 821, + 704 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Petar Velickovi ˇ c, Lars Buesing, Matthew Overlan, Razvan Pascanu, Oriol Vinyals, and Charles Blundell. Pointer ´ graph networks. Advances in Neural Information Processing Systems, 33, 2020. ", + "bbox": [ + 176, + 712, + 823, + 738 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Petar et al. Velickovi ˇ c. Graph attention networks. 2018. ´ ", + "bbox": [ + 173, + 746, + 504, + 760 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Guangtao Wang, Rex Ying, Jing Huang, and Jure Leskovec. Improving graph attention networks with large margin-based constraints. arXiv preprint arXiv:1910.11945, 2019a. ", + "bbox": [ + 169, + 768, + 823, + 795 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Minjie Wang, Da Zheng, Zihao Ye, Quan Gan, Mufei Li, Xiang Song, Jinjing Zhou, Chao Ma, Lingfan Yu, Yu Gai, Tianjun Xiao, Tong He, George Karypis, Jinyang Li, and Zheng Zhang. Deep graph library: A graph-centric, highly-performant package for graph neural networks. arXiv preprint arXiv:1909.01315, 2019b. ", + "bbox": [ + 174, + 803, + 825, + 854 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Xiao Wang, Houye Ji, Chuan Shi, Bai Wang, Yanfang Ye, Peng Cui, and Philip S Yu. Heterogeneous graph attention network. In The World Wide Web Conference, pages 2022–2032, 2019c. ", + "bbox": [ + 173, + 863, + 821, + 888 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Yangkun Wang. Bag of tricks of semi-supervised classification with graph neural networks. arXiv preprint arXiv:2103.13355, 2021. ", + "bbox": [ + 173, + 897, + 825, + 922 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Gail Weiss, Yoav Goldberg, and Eran Yahav. On the practical computational power of finite precision rnns for language recognition. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 740–745, 2018. \nFelix Wu, Amauri Souza, Tianyi Zhang, Christopher Fifty, Tao Yu, and Kilian Weinberger. Simplifying graph convolutional networks. In International conference on machine learning, pages 6861–6871. PMLR, 2019. \nZonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and S Yu Philip. A comprehensive survey on graph neural networks. IEEE Transactions on Neural Networks and Learning Systems, 2020. \nKeyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? In International Conference on Learning Representations, 2019. URL https://openreview.net/ forum?id $\\equiv$ ryGs6iA5Km. \nYiding Yang, Jiayan Qiu, Mingli Song, Dacheng Tao, and Xinchao Wang. Distilling knowledge from graph convolutional networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020. \nSeongjun Yun, Minbyul Jeong, Raehyun Kim, Jaewoo Kang, and Hyunwoo J Kim. Graph transformer networks. Advances in Neural Information Processing Systems, 32:11983–11993, 2019. \nHanqing Zeng, Hongkuan Zhou, Ajitesh Srivastava, Rajgopal Kannan, and Viktor Prasanna. Graphsaint: Graph sampling based inductive learning method. arXiv preprint arXiv:1907.04931, 2019. \nJiani Zhang, Xingjian Shi, Junyuan Xie, Hao Ma, Irwin King, and Dit-Yan Yeung. Gaan: Gated attention networks for learning on large and spatiotemporal graphs. In Proceedings of the Thirty-Fourth Conference on Uncertainty in Artificial Intelligence, pages 339–349, 2018. \nKai Zhang, Yaokang Zhu, Jun Wang, and Jie Zhang. Adaptive structural fingerprints for graph attention networks. In International Conference on Learning Representations, 2020. URL https://openreview.net/ forum?id $\\equiv$ BJxWx0NYPr. \nLingxiao Zhao and Leman Akoglu. Pairnorm: Tackling oversmoothing in gnns. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ rkecl1rtwB. ", + "bbox": [ + 171, + 104, + 828, + 515 + ], + "page_idx": 25 + } +] \ No newline at end of file diff --git a/parse/dev/F72ximsx7C1/F72ximsx7C1_middle.json b/parse/dev/F72ximsx7C1/F72ximsx7C1_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..d9dd56a8811c8b727364301d149a68e327bf4a13 --- /dev/null +++ b/parse/dev/F72ximsx7C1/F72ximsx7C1_middle.json @@ -0,0 +1,71629 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 417, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 77, + 419, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 77, + 419, + 97 + ], + "score": 1.0, + "content": "HOW ATTENTIVE ARE GRAPH ATTENTION", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 204, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 204, + 117 + ], + "score": 1.0, + "content": "NETWORKS?", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 113, + 135, + 269, + 168 + ], + "lines": [ + { + "bbox": [ + 112, + 133, + 176, + 147 + ], + "spans": [ + { + "bbox": [ + 112, + 133, + 176, + 147 + ], + "score": 1.0, + "content": "Shaked Brody", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 145, + 153, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 153, + 158 + ], + "score": 1.0, + "content": "Technion", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 112, + 158, + 270, + 169 + ], + "spans": [ + { + "bbox": [ + 112, + 158, + 270, + 169 + ], + "score": 1.0, + "content": "shakedbr@cs.technion.ac.il", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 366, + 135, + 498, + 178 + ], + "lines": [ + { + "bbox": [ + 366, + 134, + 407, + 147 + ], + "spans": [ + { + "bbox": [ + 366, + 134, + 407, + 147 + ], + "score": 1.0, + "content": "Uri Alon", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 366, + 147, + 498, + 158 + ], + "spans": [ + { + "bbox": [ + 366, + 147, + 498, + 158 + ], + "score": 1.0, + "content": "Language Technologies Institute", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 366, + 156, + 480, + 169 + ], + "spans": [ + { + "bbox": [ + 366, + 156, + 480, + 169 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 366, + 168, + 465, + 180 + ], + "spans": [ + { + "bbox": [ + 366, + 168, + 465, + 180 + ], + "score": 1.0, + "content": "ualon@cs.cmu.edu", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.0 + }, + { + "type": "title", + "bbox": [ + 113, + 196, + 164, + 207 + ], + "lines": [ + { + "bbox": [ + 111, + 195, + 166, + 209 + ], + "spans": [ + { + "bbox": [ + 111, + 195, + 166, + 209 + ], + "score": 1.0, + "content": "Eran Yahav", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 114, + 219, + 257, + 229 + ], + "lines": [ + { + "bbox": [ + 111, + 218, + 258, + 230 + ], + "spans": [ + { + "bbox": [ + 111, + 218, + 258, + 230 + ], + "score": 1.0, + "content": "yahave@cs.technion.ac.il", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 278, + 258, + 333, + 270 + ], + "lines": [ + { + "bbox": [ + 276, + 257, + 336, + 272 + ], + "spans": [ + { + "bbox": [ + 276, + 257, + 336, + 272 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 143, + 284, + 469, + 470 + ], + "lines": [ + { + "bbox": [ + 142, + 284, + 469, + 297 + ], + "spans": [ + { + "bbox": [ + 142, + 284, + 469, + 297 + ], + "score": 1.0, + "content": "Graph Attention Networks (GATs) are one of the most popular GNN architectures", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 294, + 469, + 308 + ], + "spans": [ + { + "bbox": [ + 141, + 294, + 469, + 308 + ], + "score": 1.0, + "content": "and are considered as the state-of-the-art architecture for representation learning", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 306, + 470, + 319 + ], + "spans": [ + { + "bbox": [ + 141, + 306, + 470, + 319 + ], + "score": 1.0, + "content": "with graphs. In GAT, every node attends to its neighbors given its own represen-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 316, + 470, + 330 + ], + "spans": [ + { + "bbox": [ + 141, + 316, + 470, + 330 + ], + "score": 1.0, + "content": "tation as the query. However, in this paper we show that GAT computes a very", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 327, + 470, + 340 + ], + "spans": [ + { + "bbox": [ + 141, + 327, + 470, + 340 + ], + "score": 1.0, + "content": "limited kind of attention: the ranking of the attention scores is unconditioned", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 339, + 470, + 350 + ], + "spans": [ + { + "bbox": [ + 141, + 339, + 470, + 350 + ], + "score": 1.0, + "content": "on the query node. We formally define this restricted kind of attention as static", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 351, + 470, + 362 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 470, + 362 + ], + "score": 1.0, + "content": "attention and distinguish it from a strictly more expressive dynamic attention. Be-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 361, + 469, + 373 + ], + "spans": [ + { + "bbox": [ + 141, + 361, + 469, + 373 + ], + "score": 1.0, + "content": "cause GATs use a static attention mechanism, there are simple graph problems", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 371, + 469, + 384 + ], + "spans": [ + { + "bbox": [ + 141, + 371, + 469, + 384 + ], + "score": 1.0, + "content": "that GAT cannot express: in a controlled problem, we show that static attention", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 382, + 469, + 395 + ], + "spans": [ + { + "bbox": [ + 141, + 382, + 469, + 395 + ], + "score": 1.0, + "content": "hinders GAT from even fitting the training data. To remove this limitation, we", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 393, + 471, + 406 + ], + "spans": [ + { + "bbox": [ + 141, + 393, + 471, + 406 + ], + "score": 1.0, + "content": "introduce a simple fix by modifying the order of operations and propose GATv2:", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 405, + 470, + 417 + ], + "spans": [ + { + "bbox": [ + 141, + 405, + 470, + 417 + ], + "score": 1.0, + "content": "a dynamic graph attention variant that is strictly more expressive than GAT. We", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 416, + 470, + 427 + ], + "spans": [ + { + "bbox": [ + 141, + 416, + 470, + 427 + ], + "score": 1.0, + "content": "perform an extensive evaluation and show that GATv2 outperforms GAT across 12", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 426, + 470, + 439 + ], + "spans": [ + { + "bbox": [ + 141, + 426, + 470, + 439 + ], + "score": 1.0, + "content": "OGB and other benchmarks while we match their parametric costs. Our code is", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "spans": [ + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "score": 1.0, + "content": "available at https://github.com/tech-srl/how_attentive_are_", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 448, + 470, + 462 + ], + "spans": [ + { + "bbox": [ + 141, + 448, + 470, + 462 + ], + "score": 1.0, + "content": "gats.1 GATv2 is available as part of the PyTorch Geometric library,2 the Deep", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 459, + 348, + 472 + ], + "spans": [ + { + "bbox": [ + 141, + 459, + 348, + 472 + ], + "score": 1.0, + "content": "Graph Library,3 and the TensorFlow GNN library.4", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 108, + 493, + 206, + 506 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 208, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 208, + 509 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 505, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 532 + ], + "score": 1.0, + "content": "Graph neural networks (GNNs; Gori et al., 2005; Scarselli et al., 2008) have seen increasing popularity", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 530, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 506, + 542 + ], + "score": 1.0, + "content": "over the past few years (Duvenaud et al., 2015; Atwood and Towsley, 2016; Bronstein et al., 2017;", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 540, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 506, + 553 + ], + "score": 1.0, + "content": "Monti et al., 2017). GNNs provide a general and efficient framework to learn from graph-structured", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "score": 1.0, + "content": "data. Thus, GNNs are easily applicable in domains where the data can be represented as a set of", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 563, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 505, + 575 + ], + "score": 1.0, + "content": "nodes and the prediction depends on the relationships (edges) between the nodes. Such domains", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 573, + 478, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 478, + 587 + ], + "score": 1.0, + "content": "include molecules, social networks, product recommendation, computer programs and more.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 505, + 679 + ], + "lines": [ + { + "bbox": [ + 105, + 591, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 505, + 603 + ], + "score": 1.0, + "content": "In a GNN, each node iteratively updates its state by interacting with its neighbors. GNN variants", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "score": 1.0, + "content": "(Wu et al., 2019; Xu et al., 2019; Li et al., 2016) mostly differ in how each node aggregates and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 613, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 625 + ], + "score": 1.0, + "content": "combines the representations of its neighbors with its own. Velickovi ˇ c et al. (2018) pioneered the use ´", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "score": 1.0, + "content": "of attention-based neighborhood aggregation, in one of the most common GNN variants – Graph", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "score": 1.0, + "content": "Attention Network (GAT). In GAT, every node updates its representation by attending to its neighbors", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 644, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 659 + ], + "score": 1.0, + "content": "using its own representation as the query. This generalizes the standard averaging or max-pooling", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 656, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 506, + 669 + ], + "score": 1.0, + "content": "of neighbors (Kipf and Welling, 2017; Hamilton et al., 2017), by allowing every node to compute", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "a weighted average of its neighbors, and (softly) select its most relevant neighbors. The work of", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 39.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 119, + 689, + 499, + 732 + ], + "lines": [ + { + "bbox": [ + 118, + 687, + 500, + 701 + ], + "spans": [ + { + "bbox": [ + 118, + 687, + 500, + 701 + ], + "score": 1.0, + "content": "1An annotated implementation of GATv2 is available at https://nn.labml.ai/graphs/gatv2/", + "type": "text" + } + ] + }, + { + "bbox": [ + 117, + 696, + 428, + 713 + ], + "spans": [ + { + "bbox": [ + 117, + 696, + 428, + 713 + ], + "score": 1.0, + "content": "2from torch_geometric.nn.conv.gatv2_conv import GATv2Conv", + "type": "text" + } + ] + }, + { + "bbox": [ + 116, + 707, + 320, + 725 + ], + "spans": [ + { + "bbox": [ + 116, + 707, + 320, + 725 + ], + "score": 1.0, + "content": "3from dgl.nn.pytorch import GATv2Conv", + "type": "text" + } + ] + }, + { + "bbox": [ + 117, + 718, + 497, + 735 + ], + "spans": [ + { + "bbox": [ + 117, + 718, + 497, + 735 + ], + "score": 1.0, + "content": "4from tensorflow_gnn.graph.keras.layers.gat_v2 import GATv2Convolution", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 417, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 77, + 419, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 77, + 419, + 97 + ], + "score": 1.0, + "content": "HOW ATTENTIVE ARE GRAPH ATTENTION", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 204, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 204, + 117 + ], + "score": 1.0, + "content": "NETWORKS?", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "list", + "bbox": [ + 113, + 135, + 269, + 168 + ], + "lines": [ + { + "bbox": [ + 112, + 133, + 176, + 147 + ], + "spans": [ + { + "bbox": [ + 112, + 133, + 176, + 147 + ], + "score": 1.0, + "content": "Shaked Brody", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 145, + 153, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 153, + 158 + ], + "score": 1.0, + "content": "Technion", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 112, + 158, + 270, + 169 + ], + "spans": [ + { + "bbox": [ + 112, + 158, + 270, + 169 + ], + "score": 1.0, + "content": "shakedbr@cs.technion.ac.il", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + } + ], + "index": 4, + "bbox_fs": [ + 111, + 133, + 270, + 169 + ] + }, + { + "type": "text", + "bbox": [ + 366, + 135, + 498, + 178 + ], + "lines": [ + { + "bbox": [ + 366, + 134, + 407, + 147 + ], + "spans": [ + { + "bbox": [ + 366, + 134, + 407, + 147 + ], + "score": 1.0, + "content": "Uri Alon", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 366, + 147, + 498, + 158 + ], + "spans": [ + { + "bbox": [ + 366, + 147, + 498, + 158 + ], + "score": 1.0, + "content": "Language Technologies Institute", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 366, + 156, + 480, + 169 + ], + "spans": [ + { + "bbox": [ + 366, + 156, + 480, + 169 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 366, + 168, + 465, + 180 + ], + "spans": [ + { + "bbox": [ + 366, + 168, + 465, + 180 + ], + "score": 1.0, + "content": "ualon@cs.cmu.edu", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.0, + "bbox_fs": [ + 366, + 134, + 498, + 180 + ] + }, + { + "type": "title", + "bbox": [ + 113, + 196, + 164, + 207 + ], + "lines": [ + { + "bbox": [ + 111, + 195, + 166, + 209 + ], + "spans": [ + { + "bbox": [ + 111, + 195, + 166, + 209 + ], + "score": 1.0, + "content": "Eran Yahav", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 114, + 219, + 257, + 229 + ], + "lines": [ + { + "bbox": [ + 111, + 218, + 258, + 230 + ], + "spans": [ + { + "bbox": [ + 111, + 218, + 258, + 230 + ], + "score": 1.0, + "content": "yahave@cs.technion.ac.il", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 111, + 218, + 258, + 230 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 258, + 333, + 270 + ], + "lines": [ + { + "bbox": [ + 276, + 257, + 336, + 272 + ], + "spans": [ + { + "bbox": [ + 276, + 257, + 336, + 272 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 143, + 284, + 469, + 470 + ], + "lines": [ + { + "bbox": [ + 142, + 284, + 469, + 297 + ], + "spans": [ + { + "bbox": [ + 142, + 284, + 469, + 297 + ], + "score": 1.0, + "content": "Graph Attention Networks (GATs) are one of the most popular GNN architectures", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 294, + 469, + 308 + ], + "spans": [ + { + "bbox": [ + 141, + 294, + 469, + 308 + ], + "score": 1.0, + "content": "and are considered as the state-of-the-art architecture for representation learning", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 306, + 470, + 319 + ], + "spans": [ + { + "bbox": [ + 141, + 306, + 470, + 319 + ], + "score": 1.0, + "content": "with graphs. In GAT, every node attends to its neighbors given its own represen-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 316, + 470, + 330 + ], + "spans": [ + { + "bbox": [ + 141, + 316, + 470, + 330 + ], + "score": 1.0, + "content": "tation as the query. However, in this paper we show that GAT computes a very", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 327, + 470, + 340 + ], + "spans": [ + { + "bbox": [ + 141, + 327, + 470, + 340 + ], + "score": 1.0, + "content": "limited kind of attention: the ranking of the attention scores is unconditioned", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 339, + 470, + 350 + ], + "spans": [ + { + "bbox": [ + 141, + 339, + 470, + 350 + ], + "score": 1.0, + "content": "on the query node. We formally define this restricted kind of attention as static", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 351, + 470, + 362 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 470, + 362 + ], + "score": 1.0, + "content": "attention and distinguish it from a strictly more expressive dynamic attention. Be-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 361, + 469, + 373 + ], + "spans": [ + { + "bbox": [ + 141, + 361, + 469, + 373 + ], + "score": 1.0, + "content": "cause GATs use a static attention mechanism, there are simple graph problems", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 371, + 469, + 384 + ], + "spans": [ + { + "bbox": [ + 141, + 371, + 469, + 384 + ], + "score": 1.0, + "content": "that GAT cannot express: in a controlled problem, we show that static attention", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 382, + 469, + 395 + ], + "spans": [ + { + "bbox": [ + 141, + 382, + 469, + 395 + ], + "score": 1.0, + "content": "hinders GAT from even fitting the training data. To remove this limitation, we", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 393, + 471, + 406 + ], + "spans": [ + { + "bbox": [ + 141, + 393, + 471, + 406 + ], + "score": 1.0, + "content": "introduce a simple fix by modifying the order of operations and propose GATv2:", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 405, + 470, + 417 + ], + "spans": [ + { + "bbox": [ + 141, + 405, + 470, + 417 + ], + "score": 1.0, + "content": "a dynamic graph attention variant that is strictly more expressive than GAT. We", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 416, + 470, + 427 + ], + "spans": [ + { + "bbox": [ + 141, + 416, + 470, + 427 + ], + "score": 1.0, + "content": "perform an extensive evaluation and show that GATv2 outperforms GAT across 12", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 426, + 470, + 439 + ], + "spans": [ + { + "bbox": [ + 141, + 426, + 470, + 439 + ], + "score": 1.0, + "content": "OGB and other benchmarks while we match their parametric costs. Our code is", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "spans": [ + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "score": 1.0, + "content": "available at https://github.com/tech-srl/how_attentive_are_", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 448, + 470, + 462 + ], + "spans": [ + { + "bbox": [ + 141, + 448, + 470, + 462 + ], + "score": 1.0, + "content": "gats.1 GATv2 is available as part of the PyTorch Geometric library,2 the Deep", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 459, + 348, + 472 + ], + "spans": [ + { + "bbox": [ + 141, + 459, + 348, + 472 + ], + "score": 1.0, + "content": "Graph Library,3 and the TensorFlow GNN library.4", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 20, + "bbox_fs": [ + 141, + 284, + 471, + 472 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 493, + 206, + 506 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 208, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 208, + 509 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 505, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 532 + ], + "score": 1.0, + "content": "Graph neural networks (GNNs; Gori et al., 2005; Scarselli et al., 2008) have seen increasing popularity", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 530, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 506, + 542 + ], + "score": 1.0, + "content": "over the past few years (Duvenaud et al., 2015; Atwood and Towsley, 2016; Bronstein et al., 2017;", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 540, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 506, + 553 + ], + "score": 1.0, + "content": "Monti et al., 2017). GNNs provide a general and efficient framework to learn from graph-structured", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "score": 1.0, + "content": "data. Thus, GNNs are easily applicable in domains where the data can be represented as a set of", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 563, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 505, + 575 + ], + "score": 1.0, + "content": "nodes and the prediction depends on the relationships (edges) between the nodes. Such domains", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 573, + 478, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 478, + 587 + ], + "score": 1.0, + "content": "include molecules, social networks, product recommendation, computer programs and more.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 518, + 506, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 505, + 679 + ], + "lines": [ + { + "bbox": [ + 105, + 591, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 505, + 603 + ], + "score": 1.0, + "content": "In a GNN, each node iteratively updates its state by interacting with its neighbors. GNN variants", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "score": 1.0, + "content": "(Wu et al., 2019; Xu et al., 2019; Li et al., 2016) mostly differ in how each node aggregates and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 613, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 625 + ], + "score": 1.0, + "content": "combines the representations of its neighbors with its own. Velickovi ˇ c et al. (2018) pioneered the use ´", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "score": 1.0, + "content": "of attention-based neighborhood aggregation, in one of the most common GNN variants – Graph", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "score": 1.0, + "content": "Attention Network (GAT). In GAT, every node updates its representation by attending to its neighbors", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 644, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 659 + ], + "score": 1.0, + "content": "using its own representation as the query. This generalizes the standard averaging or max-pooling", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 656, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 506, + 669 + ], + "score": 1.0, + "content": "of neighbors (Kipf and Welling, 2017; Hamilton et al., 2017), by allowing every node to compute", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "a weighted average of its neighbors, and (softly) select its most relevant neighbors. The work of", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 591, + 506, + 680 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 77, + 491, + 409 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 77, + 491, + 409 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 77, + 491, + 409 + ], + "spans": [ + { + "bbox": [ + 115, + 77, + 491, + 409 + ], + "score": 0.957, + "type": "image", + "image_path": "3fd2d6f1a250485c1844b504947dfd994af31b1f76fbbbea17a0ae62e23a556a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 77, + 491, + 187.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 187.66666666666669, + 491, + 298.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 298.33333333333337, + 491, + 409.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 417, + 505, + 474 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 417, + 507, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 336, + 431 + ], + "score": 1.0, + "content": "Figure 1: In a complete bipartite graph of “query nodes”", + "type": "text" + }, + { + "bbox": [ + 336, + 418, + 383, + 430 + ], + "score": 0.93, + "content": "\\{ q 0 , . . . , q 9 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 417, + 453, + 431 + ], + "score": 1.0, + "content": "and “key nodes”", + "type": "text" + }, + { + "bbox": [ + 453, + 418, + 502, + 430 + ], + "score": 0.93, + "content": "\\{ k 0 , . . . , k 9 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 417, + 507, + 431 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "standard GAT (Figure 1a) computes static attention – the ranking of attention coefficients is global", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 439, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 482, + 453 + ], + "score": 1.0, + "content": "for all nodes in the graph, and is unconditioned on the query node. For example, all queries", + "type": "text" + }, + { + "bbox": [ + 482, + 441, + 494, + 451 + ], + "score": 0.64, + "content": "_ { g 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 439, + 505, + 453 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 450, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 119, + 462 + ], + "score": 0.62, + "content": "q 9 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 450, + 236, + 465 + ], + "score": 1.0, + "content": ") attend mostly to the 8th key", + "type": "text" + }, + { + "bbox": [ + 236, + 451, + 252, + 461 + ], + "score": 0.38, + "content": "( k 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 450, + 506, + 465 + ], + "score": 1.0, + "content": ". In contrast, GATv2 (Figure 1b) can actually compute dynamic", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 462, + 459, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 459, + 475 + ], + "score": 1.0, + "content": "attention, where every query has a different ranking of attention coefficients of the keys.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 498, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 506, + 514 + ], + "score": 1.0, + "content": "Velickovi ˇ c et al. also generalizes the Transformer’s (Vaswani et al., 2017) self-attention mechanism, ´", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 511, + 267, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 267, + 524 + ], + "score": 1.0, + "content": "from sequences to graphs (Joshi, 2020).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 528, + 505, + 616 + ], + "lines": [ + { + "bbox": [ + 106, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 505, + 540 + ], + "score": 1.0, + "content": "Nowadays, GAT is one of the most popular GNN architectures (Bronstein et al., 2021) and is", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "considered as the state-of-the-art neural architecture for learning with graphs (Wang et al., 2019a).", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "Nevertheless, in this paper we show that GAT does not actually compute the expressive, well known,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "type of attention (Bahdanau et al., 2014), which we call dynamic attention. Instead, we show that", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "GAT computes only a restricted “static” form of attention: for any query node, the attention function", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "is monotonic with respect to the neighbor (key) scores. That is, the ranking (the argsort) of attention", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 594, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "coefficients is identical for all nodes in the graph, and is unconditioned on the query node. This fact", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 605, + 410, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 410, + 617 + ], + "score": 1.0, + "content": "severely hurts the expressiveness of GAT, and is demonstrated in Figure 1a.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 107, + 622, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 622, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 506, + 634 + ], + "score": 1.0, + "content": "Supposedly, the conceptual idea of attention as the form of interaction between GNN nodes is", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 633, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 506, + 646 + ], + "score": 1.0, + "content": "orthogonal to the specific choice of attention function. However, Velickovi ˇ c et al.’s original design of ´", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 642, + 507, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 507, + 658 + ], + "score": 1.0, + "content": "GAT has spread to a variety of domains (Yang et al., 2020; Wang et al., 2019c; Huang and Carley,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "score": 1.0, + "content": "2019; Ma et al., 2020; Kosaraju et al., 2019; Nathani et al., 2019; Wu et al., 2020; Zhang et al., 2020)", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "and has become the default implementation of “graph attention network” in all popular GNN libraries", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "such as PyTorch Geometric (Fey and Lenssen, 2019), DGL (Wang et al., 2019b), and others (Dwivedi", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "et al., 2020; Gordic, 2020; Brockschmidt, 2020). Many other works employed GNNs with attention ´", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 699, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 504, + 711 + ], + "score": 1.0, + "content": "mechanisms other than the standard GAT’s (Yun et al. (2019), see also Appendix A). However, none", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "of these works identified the monotonicity of GAT’s attention mechanism, the theoretical differences", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 721, + 387, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 387, + 733 + ], + "score": 1.0, + "content": "between attention types, nor empirically compared their performance.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 22.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 77, + 491, + 409 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 77, + 491, + 409 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 77, + 491, + 409 + ], + "spans": [ + { + "bbox": [ + 115, + 77, + 491, + 409 + ], + "score": 0.957, + "type": "image", + "image_path": "3fd2d6f1a250485c1844b504947dfd994af31b1f76fbbbea17a0ae62e23a556a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 77, + 491, + 187.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 187.66666666666669, + 491, + 298.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 298.33333333333337, + 491, + 409.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 417, + 505, + 474 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 417, + 507, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 336, + 431 + ], + "score": 1.0, + "content": "Figure 1: In a complete bipartite graph of “query nodes”", + "type": "text" + }, + { + "bbox": [ + 336, + 418, + 383, + 430 + ], + "score": 0.93, + "content": "\\{ q 0 , . . . , q 9 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 417, + 453, + 431 + ], + "score": 1.0, + "content": "and “key nodes”", + "type": "text" + }, + { + "bbox": [ + 453, + 418, + 502, + 430 + ], + "score": 0.93, + "content": "\\{ k 0 , . . . , k 9 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 417, + 507, + 431 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "standard GAT (Figure 1a) computes static attention – the ranking of attention coefficients is global", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 439, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 482, + 453 + ], + "score": 1.0, + "content": "for all nodes in the graph, and is unconditioned on the query node. For example, all queries", + "type": "text" + }, + { + "bbox": [ + 482, + 441, + 494, + 451 + ], + "score": 0.64, + "content": "_ { g 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 439, + 505, + 453 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 450, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 119, + 462 + ], + "score": 0.62, + "content": "q 9 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 450, + 236, + 465 + ], + "score": 1.0, + "content": ") attend mostly to the 8th key", + "type": "text" + }, + { + "bbox": [ + 236, + 451, + 252, + 461 + ], + "score": 0.38, + "content": "( k 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 450, + 506, + 465 + ], + "score": 1.0, + "content": ". In contrast, GATv2 (Figure 1b) can actually compute dynamic", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 462, + 459, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 459, + 475 + ], + "score": 1.0, + "content": "attention, where every query has a different ranking of attention coefficients of the keys.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 498, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 506, + 514 + ], + "score": 1.0, + "content": "Velickovi ˇ c et al. also generalizes the Transformer’s (Vaswani et al., 2017) self-attention mechanism, ´", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 511, + 267, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 267, + 524 + ], + "score": 1.0, + "content": "from sequences to graphs (Joshi, 2020).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 498, + 506, + 524 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 528, + 505, + 616 + ], + "lines": [ + { + "bbox": [ + 106, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 505, + 540 + ], + "score": 1.0, + "content": "Nowadays, GAT is one of the most popular GNN architectures (Bronstein et al., 2021) and is", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "considered as the state-of-the-art neural architecture for learning with graphs (Wang et al., 2019a).", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "Nevertheless, in this paper we show that GAT does not actually compute the expressive, well known,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "type of attention (Bahdanau et al., 2014), which we call dynamic attention. Instead, we show that", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "GAT computes only a restricted “static” form of attention: for any query node, the attention function", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "is monotonic with respect to the neighbor (key) scores. That is, the ranking (the argsort) of attention", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 594, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "coefficients is identical for all nodes in the graph, and is unconditioned on the query node. This fact", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 605, + 410, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 410, + 617 + ], + "score": 1.0, + "content": "severely hurts the expressiveness of GAT, and is demonstrated in Figure 1a.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 528, + 506, + 617 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 622, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 622, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 506, + 634 + ], + "score": 1.0, + "content": "Supposedly, the conceptual idea of attention as the form of interaction between GNN nodes is", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 633, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 506, + 646 + ], + "score": 1.0, + "content": "orthogonal to the specific choice of attention function. However, Velickovi ˇ c et al.’s original design of ´", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 642, + 507, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 507, + 658 + ], + "score": 1.0, + "content": "GAT has spread to a variety of domains (Yang et al., 2020; Wang et al., 2019c; Huang and Carley,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "score": 1.0, + "content": "2019; Ma et al., 2020; Kosaraju et al., 2019; Nathani et al., 2019; Wu et al., 2020; Zhang et al., 2020)", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "and has become the default implementation of “graph attention network” in all popular GNN libraries", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "such as PyTorch Geometric (Fey and Lenssen, 2019), DGL (Wang et al., 2019b), and others (Dwivedi", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "et al., 2020; Gordic, 2020; Brockschmidt, 2020). Many other works employed GNNs with attention ´", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 699, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 504, + 711 + ], + "score": 1.0, + "content": "mechanisms other than the standard GAT’s (Yun et al. (2019), see also Appendix A). However, none", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "of these works identified the monotonicity of GAT’s attention mechanism, the theoretical differences", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 721, + 387, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 387, + 733 + ], + "score": 1.0, + "content": "between attention types, nor empirically compared their performance.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 622, + 507, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 126 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "To overcome the limitation we identified in GAT, we introduce a simple fix to its attention function", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "by only modifying the order of internal operations. The result is GATv2 – a graph attention variant", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 507, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 507, + 117 + ], + "score": 1.0, + "content": "that has a universal approximator attention function, and is thus strictly more expressive than GAT.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 434, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 434, + 127 + ], + "score": 1.0, + "content": "The effect of fixing the attention function in GATv2 is demonstrated in Figure 1b.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 505, + 286 + ], + "lines": [ + { + "bbox": [ + 105, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "In summary, our main contribution is identifying that one of the most popular GNN types, the graph", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 506, + 156 + ], + "score": 1.0, + "content": "attention network, does not compute dynamic attention, the kind of attention that it seems to compute.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "We introduce formal definitions for analyzing the expressive power of graph attention mechanisms", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 166, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 506, + 177 + ], + "score": 1.0, + "content": "(Definitions 3.1 and 3.2), and derive our claims theoretically (Theorem 1) from the equations of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "Velickovi ˇ c et al. (2018). Empirically, we use a synthetic problem to show that standard GAT ´ cannot", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 506, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 506, + 201 + ], + "score": 1.0, + "content": "express problems that require dynamic attention (Section 4.1). We introduce a simple fix by switching", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "the order of internal operations in GAT, and propose GATv2, which does compute dynamic attention", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 208, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 104, + 208, + 506, + 222 + ], + "score": 1.0, + "content": "(Theorem 2). We further conduct a thorough empirical comparison of GAT and GATv2 and find", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 220, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 220, + 506, + 233 + ], + "score": 1.0, + "content": "that GATv2 outperforms GAT across 12 benchmarks of node-, link-, and graph-prediction. For", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 230, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 366, + 245 + ], + "score": 1.0, + "content": "example, GATv2 outperforms extensively tuned GNNs by over", + "type": "text" + }, + { + "bbox": [ + 367, + 231, + 389, + 241 + ], + "score": 0.86, + "content": "1 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 230, + 506, + 245 + ], + "score": 1.0, + "content": "in the difficult “UnseenProj", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 242, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 506, + 255 + ], + "score": 1.0, + "content": "Test” set of the VarMisuse task (Allamanis et al., 2018), without any hyperparameter tuning; and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 252, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 252, + 319, + 265 + ], + "score": 1.0, + "content": "GATv2 improves over an extensively-tuned GAT by", + "type": "text" + }, + { + "bbox": [ + 320, + 253, + 346, + 263 + ], + "score": 0.88, + "content": "1 1 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 252, + 505, + 265 + ], + "score": 1.0, + "content": "in 13 prediction objectives in QM9. In", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 277 + ], + "score": 1.0, + "content": "node-prediction benchmarks from OGB (Hu et al., 2020), not only that GATv2 outperforms GAT", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 275, + 507, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 507, + 288 + ], + "score": 1.0, + "content": "with respect to accuracy – we find that dynamic attention provided a much better robustness to noise.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 10.5 + }, + { + "type": "title", + "bbox": [ + 108, + 302, + 208, + 315 + ], + "lines": [ + { + "bbox": [ + 105, + 301, + 209, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 209, + 317 + ], + "score": 1.0, + "content": "2 PRELIMINARIES", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 326, + 505, + 363 + ], + "lines": [ + { + "bbox": [ + 105, + 326, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 175, + 340 + ], + "score": 1.0, + "content": "A directed graph", + "type": "text" + }, + { + "bbox": [ + 176, + 326, + 222, + 339 + ], + "score": 0.93, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 326, + 284, + 340 + ], + "score": 1.0, + "content": "contains nodes", + "type": "text" + }, + { + "bbox": [ + 285, + 326, + 344, + 339 + ], + "score": 0.95, + "content": "\\mathcal { V } = \\{ 1 , . . . , n \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 326, + 387, + 340 + ], + "score": 1.0, + "content": "and edges", + "type": "text" + }, + { + "bbox": [ + 387, + 326, + 434, + 338 + ], + "score": 0.9, + "content": "\\mathcal { E } \\subseteq \\mathcal { V } \\times \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 326, + 465, + 340 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 465, + 326, + 504, + 339 + ], + "score": 0.92, + "content": "( j , i ) \\in \\mathcal { E }", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 338, + 505, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 232, + 350 + ], + "score": 1.0, + "content": "denotes an edge from a node", + "type": "text" + }, + { + "bbox": [ + 233, + 339, + 239, + 349 + ], + "score": 0.83, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 338, + 284, + 350 + ], + "score": 1.0, + "content": "to a node", + "type": "text" + }, + { + "bbox": [ + 285, + 339, + 290, + 348 + ], + "score": 0.64, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 338, + 416, + 350 + ], + "score": 1.0, + "content": ". We assume that every node", + "type": "text" + }, + { + "bbox": [ + 417, + 338, + 446, + 348 + ], + "score": 0.89, + "content": "\\textit { i } \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 338, + 505, + 350 + ], + "score": 1.0, + "content": "has an initial", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 103, + 347, + 480, + 366 + ], + "spans": [ + { + "bbox": [ + 103, + 347, + 165, + 366 + ], + "score": 1.0, + "content": "representation", + "type": "text" + }, + { + "bbox": [ + 165, + 348, + 211, + 363 + ], + "score": 0.93, + "content": "\\bar { \\boldsymbol { h } } _ { i } ^ { ( 0 ) } \\in \\mathbb { R } ^ { d _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 347, + 480, + 366 + ], + "score": 1.0, + "content": ". An undirected graph can be represented with bidirectional edges.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 108, + 375, + 254, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 375, + 255, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 255, + 389 + ], + "score": 1.0, + "content": "2.1 GRAPH NEURAL NETWORKS", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 396, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 104, + 396, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 104, + 396, + 506, + 408 + ], + "score": 1.0, + "content": "A graph neural network (GNN) layer updates every node representation by aggregating its neighbors’", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 407, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 367, + 420 + ], + "score": 1.0, + "content": "representations. A layer’s input is a set of node representations", + "type": "text" + }, + { + "bbox": [ + 367, + 407, + 446, + 419 + ], + "score": 0.91, + "content": "\\{ h _ { i } \\in \\bar { \\mathbb { R } } ^ { d } \\bar { | } \\ i \\stackrel { - } { \\in } \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 407, + 506, + 420 + ], + "score": 1.0, + "content": "and the set of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 419, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 132, + 433 + ], + "score": 1.0, + "content": "edges", + "type": "text" + }, + { + "bbox": [ + 132, + 420, + 140, + 430 + ], + "score": 0.71, + "content": "\\mathcal { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 419, + 351, + 433 + ], + "score": 1.0, + "content": ". A layer outputs a new set of node representations", + "type": "text" + }, + { + "bbox": [ + 352, + 419, + 434, + 432 + ], + "score": 0.89, + "content": "\\{ h _ { i } ^ { \\prime } \\in \\mathbb { R } ^ { d ^ { \\prime } } \\mid i \\in \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 419, + 506, + 433 + ], + "score": 1.0, + "content": ", where the same", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 431, + 473, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 362, + 444 + ], + "score": 1.0, + "content": "parametric function is applied to every node given its neighbors", + "type": "text" + }, + { + "bbox": [ + 363, + 432, + 469, + 444 + ], + "score": 0.84, + "content": "\\mathcal { N } _ { i } = \\{ j \\in \\mathcal { V } \\mid ( j , i ) \\in \\mathcal { E } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 431, + 473, + 444 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "interline_equation", + "bbox": [ + 217, + 446, + 394, + 460 + ], + "lines": [ + { + "bbox": [ + 217, + 446, + 394, + 460 + ], + "spans": [ + { + "bbox": [ + 217, + 446, + 394, + 460 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\pmb { h } _ { i } ^ { \\prime } = f _ { \\theta } \\left( \\pmb { h } _ { i } , \\mathrm { A G G R E G A T E } \\left( \\{ \\pmb { h } _ { j } \\ | \\ j \\in \\mathcal { N } _ { i } \\} \\right) \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "5c0f7e69ad5b558b5aa89dba0f3b2dc5e793350b81205f5ed8aaa6ea2bd60471.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 217, + 446, + 394, + 460 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 464, + 505, + 497 + ], + "lines": [ + { + "bbox": [ + 105, + 463, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 163, + 477 + ], + "score": 1.0, + "content": "The design of", + "type": "text" + }, + { + "bbox": [ + 163, + 464, + 170, + 475 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 463, + 506, + 477 + ], + "score": 1.0, + "content": "and AGGREGATE is what mostly distinguishes one type of GNN from the other. For", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 475, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 505, + 487 + ], + "score": 1.0, + "content": "example, a common variant of GraphSAGE (Hamilton et al., 2017) performs an element-wise mean", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 486, + 444, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 302, + 498 + ], + "score": 1.0, + "content": "as AGGREGATE, followed by concatenation with", + "type": "text" + }, + { + "bbox": [ + 302, + 486, + 313, + 497 + ], + "score": 0.87, + "content": "\\boldsymbol { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 486, + 433, + 498 + ], + "score": 1.0, + "content": ", a linear layer and a ReLU as", + "type": "text" + }, + { + "bbox": [ + 433, + 486, + 440, + 497 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 486, + 444, + 498 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "title", + "bbox": [ + 108, + 510, + 267, + 521 + ], + "lines": [ + { + "bbox": [ + 105, + 509, + 268, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 268, + 523 + ], + "score": 1.0, + "content": "2.2 GRAPH ATTENTION NETWORKS", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 530, + 505, + 586 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 343, + 543 + ], + "score": 1.0, + "content": "GraphSAGE and many other popular GNN architectures", + "type": "text" + }, + { + "bbox": [ + 344, + 532, + 357, + 541 + ], + "score": 0.27, + "content": "\\mathrm { { X u } }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 531, + 506, + 543 + ], + "score": 1.0, + "content": "et al., 2019; Duvenaud et al., 2015)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 187, + 555 + ], + "score": 1.0, + "content": "weigh all neighbors", + "type": "text" + }, + { + "bbox": [ + 187, + 542, + 216, + 553 + ], + "score": 0.91, + "content": "j \\in \\mathcal N _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 542, + 506, + 555 + ], + "score": 1.0, + "content": "with equal importance (e.g., mean or max-pooling as AGGREGATE). To", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 552, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 566 + ], + "score": 1.0, + "content": "address this limitation, GAT (Velickovi ˇ c et al., 2018) instantiates Equation (1) by computing a learned ´", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 561, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 273, + 578 + ], + "score": 1.0, + "content": "weighted average of the representations of", + "type": "text" + }, + { + "bbox": [ + 273, + 564, + 285, + 575 + ], + "score": 0.89, + "content": "{ \\mathcal { N } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 561, + 365, + 578 + ], + "score": 1.0, + "content": ". A scoring function", + "type": "text" + }, + { + "bbox": [ + 366, + 564, + 434, + 574 + ], + "score": 0.91, + "content": "e : \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 561, + 506, + 578 + ], + "score": 1.0, + "content": "computes a score", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 574, + 502, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 166, + 586 + ], + "score": 1.0, + "content": "for every edge", + "type": "text" + }, + { + "bbox": [ + 167, + 575, + 186, + 587 + ], + "score": 0.92, + "content": "( j , i )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 574, + 440, + 586 + ], + "score": 1.0, + "content": ", which indicates the importance of the features of the neighbor", + "type": "text" + }, + { + "bbox": [ + 440, + 577, + 445, + 586 + ], + "score": 0.86, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 574, + 493, + 586 + ], + "score": 1.0, + "content": "to the node", + "type": "text" + }, + { + "bbox": [ + 494, + 577, + 497, + 585 + ], + "score": 0.85, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 574, + 502, + 586 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34 + }, + { + "type": "interline_equation", + "bbox": [ + 208, + 590, + 402, + 606 + ], + "lines": [ + { + "bbox": [ + 208, + 590, + 402, + 606 + ], + "spans": [ + { + "bbox": [ + 208, + 590, + 402, + 606 + ], + "score": 0.91, + "content": "e ( \\pmb { h } _ { i } , \\pmb { h } _ { j } ) = \\mathrm { L e a k y R e L U } ( \\pmb { a } ^ { \\top } \\cdot [ \\pmb { W } \\pmb { h } _ { i } | | \\pmb { W } \\pmb { h } _ { j } | ) ", + "type": "interline_equation", + "image_path": "fdf1aac9f4eb59b6d40db1e9082b54ae55306d8a0e6db961c7777c502a2c36f3.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 208, + 590, + 402, + 606 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 610, + 507, + 634 + ], + "lines": [ + { + "bbox": [ + 104, + 607, + 507, + 626 + ], + "spans": [ + { + "bbox": [ + 104, + 607, + 132, + 626 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 610, + 170, + 622 + ], + "score": 0.88, + "content": "\\pmb { a } \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 607, + 174, + 626 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 174, + 610, + 225, + 622 + ], + "score": 0.89, + "content": "W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 607, + 290, + 626 + ], + "score": 1.0, + "content": "are learned, and", + "type": "text" + }, + { + "bbox": [ + 290, + 612, + 297, + 624 + ], + "score": 0.85, + "content": "\\parallel", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 607, + 507, + 626 + ], + "score": 1.0, + "content": "denotes vector concatenation. These attention scores", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 622, + 501, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 250, + 635 + ], + "score": 1.0, + "content": "are normalized across all neighbors", + "type": "text" + }, + { + "bbox": [ + 250, + 623, + 279, + 634 + ], + "score": 0.92, + "content": "j \\in \\mathcal N _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 622, + 501, + 635 + ], + "score": 1.0, + "content": "using softmax, and the attention function is defined as:", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "interline_equation", + "bbox": [ + 189, + 637, + 422, + 666 + ], + "lines": [ + { + "bbox": [ + 189, + 637, + 422, + 666 + ], + "spans": [ + { + "bbox": [ + 189, + 637, + 422, + 666 + ], + "score": 0.94, + "content": "\\alpha _ { i j } = \\mathrm { s o f t m a x } _ { j } \\left( e \\left( h _ { i } , h _ { j } \\right) \\right) = \\frac { \\exp { \\left( e \\left( h _ { i } , h _ { j } \\right) \\right) } } { \\sum _ { j ^ { \\prime } \\in { \\mathcal { N } } _ { i } } \\exp { \\left( e \\left( h _ { i } , h _ { j ^ { \\prime } } \\right) \\right) } }", + "type": "interline_equation", + "image_path": "b4309c7d70cbdbb892256ebb05106e903830abef53b0557adaa3453de5e3f136.jpg" + } + ] + } + ], + "index": 40.5, + "virtual_lines": [ + { + "bbox": [ + 189, + 637, + 422, + 651.5 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 189, + 651.5, + 422, + 666.0 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 669, + 507, + 692 + ], + "lines": [ + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "score": 1.0, + "content": "Then, GAT computes a weighted average of the transformed features of the neighbor nodes (followed", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 681, + 485, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 176, + 693 + ], + "score": 1.0, + "content": "by a nonlinearity", + "type": "text" + }, + { + "bbox": [ + 177, + 682, + 184, + 690 + ], + "score": 0.71, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 681, + 303, + 693 + ], + "score": 1.0, + "content": ") as the new representation of", + "type": "text" + }, + { + "bbox": [ + 303, + 683, + 307, + 690 + ], + "score": 0.81, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 681, + 485, + 693 + ], + "score": 1.0, + "content": ", using the normalized attention coefficients:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 695, + 367, + 717 + ], + "lines": [ + { + "bbox": [ + 243, + 695, + 367, + 717 + ], + "spans": [ + { + "bbox": [ + 243, + 695, + 367, + 717 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\pmb { h } _ { i } ^ { \\prime } = \\sigma \\left( \\sum _ { j \\in \\mathcal { N } _ { i } } \\pmb { \\alpha } _ { i j } \\cdot \\pmb { W } \\pmb { h } _ { j } \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "e616a2cc35af6e2807af1d8d3102e0a51231e3b5e6c203fb765b2da392727b92.jpg" + } + ] + } + ], + "index": 44, + "virtual_lines": [ + { + "bbox": [ + 243, + 695, + 367, + 717 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 410, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 409, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 409, + 732 + ], + "score": 1.0, + "content": "From now on, we will refer to Equations (2) to (4) as the definition of GAT.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 126 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "To overcome the limitation we identified in GAT, we introduce a simple fix to its attention function", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "by only modifying the order of internal operations. The result is GATv2 – a graph attention variant", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 507, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 507, + 117 + ], + "score": 1.0, + "content": "that has a universal approximator attention function, and is thus strictly more expressive than GAT.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 434, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 434, + 127 + ], + "score": 1.0, + "content": "The effect of fixing the attention function in GATv2 is demonstrated in Figure 1b.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 82, + 507, + 127 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 505, + 286 + ], + "lines": [ + { + "bbox": [ + 105, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "In summary, our main contribution is identifying that one of the most popular GNN types, the graph", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 506, + 156 + ], + "score": 1.0, + "content": "attention network, does not compute dynamic attention, the kind of attention that it seems to compute.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "We introduce formal definitions for analyzing the expressive power of graph attention mechanisms", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 166, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 506, + 177 + ], + "score": 1.0, + "content": "(Definitions 3.1 and 3.2), and derive our claims theoretically (Theorem 1) from the equations of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "Velickovi ˇ c et al. (2018). Empirically, we use a synthetic problem to show that standard GAT ´ cannot", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 506, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 506, + 201 + ], + "score": 1.0, + "content": "express problems that require dynamic attention (Section 4.1). We introduce a simple fix by switching", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "the order of internal operations in GAT, and propose GATv2, which does compute dynamic attention", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 208, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 104, + 208, + 506, + 222 + ], + "score": 1.0, + "content": "(Theorem 2). We further conduct a thorough empirical comparison of GAT and GATv2 and find", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 220, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 220, + 506, + 233 + ], + "score": 1.0, + "content": "that GATv2 outperforms GAT across 12 benchmarks of node-, link-, and graph-prediction. For", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 230, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 366, + 245 + ], + "score": 1.0, + "content": "example, GATv2 outperforms extensively tuned GNNs by over", + "type": "text" + }, + { + "bbox": [ + 367, + 231, + 389, + 241 + ], + "score": 0.86, + "content": "1 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 230, + 506, + 245 + ], + "score": 1.0, + "content": "in the difficult “UnseenProj", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 242, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 506, + 255 + ], + "score": 1.0, + "content": "Test” set of the VarMisuse task (Allamanis et al., 2018), without any hyperparameter tuning; and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 252, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 252, + 319, + 265 + ], + "score": 1.0, + "content": "GATv2 improves over an extensively-tuned GAT by", + "type": "text" + }, + { + "bbox": [ + 320, + 253, + 346, + 263 + ], + "score": 0.88, + "content": "1 1 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 252, + 505, + 265 + ], + "score": 1.0, + "content": "in 13 prediction objectives in QM9. In", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 277 + ], + "score": 1.0, + "content": "node-prediction benchmarks from OGB (Hu et al., 2020), not only that GATv2 outperforms GAT", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 275, + 507, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 507, + 288 + ], + "score": 1.0, + "content": "with respect to accuracy – we find that dynamic attention provided a much better robustness to noise.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 10.5, + "bbox_fs": [ + 104, + 132, + 507, + 288 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 302, + 208, + 315 + ], + "lines": [ + { + "bbox": [ + 105, + 301, + 209, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 209, + 317 + ], + "score": 1.0, + "content": "2 PRELIMINARIES", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 326, + 505, + 363 + ], + "lines": [ + { + "bbox": [ + 105, + 326, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 175, + 340 + ], + "score": 1.0, + "content": "A directed graph", + "type": "text" + }, + { + "bbox": [ + 176, + 326, + 222, + 339 + ], + "score": 0.93, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 326, + 284, + 340 + ], + "score": 1.0, + "content": "contains nodes", + "type": "text" + }, + { + "bbox": [ + 285, + 326, + 344, + 339 + ], + "score": 0.95, + "content": "\\mathcal { V } = \\{ 1 , . . . , n \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 326, + 387, + 340 + ], + "score": 1.0, + "content": "and edges", + "type": "text" + }, + { + "bbox": [ + 387, + 326, + 434, + 338 + ], + "score": 0.9, + "content": "\\mathcal { E } \\subseteq \\mathcal { V } \\times \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 326, + 465, + 340 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 465, + 326, + 504, + 339 + ], + "score": 0.92, + "content": "( j , i ) \\in \\mathcal { E }", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 338, + 505, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 232, + 350 + ], + "score": 1.0, + "content": "denotes an edge from a node", + "type": "text" + }, + { + "bbox": [ + 233, + 339, + 239, + 349 + ], + "score": 0.83, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 338, + 284, + 350 + ], + "score": 1.0, + "content": "to a node", + "type": "text" + }, + { + "bbox": [ + 285, + 339, + 290, + 348 + ], + "score": 0.64, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 338, + 416, + 350 + ], + "score": 1.0, + "content": ". We assume that every node", + "type": "text" + }, + { + "bbox": [ + 417, + 338, + 446, + 348 + ], + "score": 0.89, + "content": "\\textit { i } \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 338, + 505, + 350 + ], + "score": 1.0, + "content": "has an initial", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 103, + 347, + 480, + 366 + ], + "spans": [ + { + "bbox": [ + 103, + 347, + 165, + 366 + ], + "score": 1.0, + "content": "representation", + "type": "text" + }, + { + "bbox": [ + 165, + 348, + 211, + 363 + ], + "score": 0.93, + "content": "\\bar { \\boldsymbol { h } } _ { i } ^ { ( 0 ) } \\in \\mathbb { R } ^ { d _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 347, + 480, + 366 + ], + "score": 1.0, + "content": ". An undirected graph can be represented with bidirectional edges.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 103, + 326, + 505, + 366 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 375, + 254, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 375, + 255, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 255, + 389 + ], + "score": 1.0, + "content": "2.1 GRAPH NEURAL NETWORKS", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 396, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 104, + 396, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 104, + 396, + 506, + 408 + ], + "score": 1.0, + "content": "A graph neural network (GNN) layer updates every node representation by aggregating its neighbors’", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 407, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 367, + 420 + ], + "score": 1.0, + "content": "representations. A layer’s input is a set of node representations", + "type": "text" + }, + { + "bbox": [ + 367, + 407, + 446, + 419 + ], + "score": 0.91, + "content": "\\{ h _ { i } \\in \\bar { \\mathbb { R } } ^ { d } \\bar { | } \\ i \\stackrel { - } { \\in } \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 407, + 506, + 420 + ], + "score": 1.0, + "content": "and the set of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 419, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 132, + 433 + ], + "score": 1.0, + "content": "edges", + "type": "text" + }, + { + "bbox": [ + 132, + 420, + 140, + 430 + ], + "score": 0.71, + "content": "\\mathcal { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 419, + 351, + 433 + ], + "score": 1.0, + "content": ". A layer outputs a new set of node representations", + "type": "text" + }, + { + "bbox": [ + 352, + 419, + 434, + 432 + ], + "score": 0.89, + "content": "\\{ h _ { i } ^ { \\prime } \\in \\mathbb { R } ^ { d ^ { \\prime } } \\mid i \\in \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 419, + 506, + 433 + ], + "score": 1.0, + "content": ", where the same", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 431, + 473, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 362, + 444 + ], + "score": 1.0, + "content": "parametric function is applied to every node given its neighbors", + "type": "text" + }, + { + "bbox": [ + 363, + 432, + 469, + 444 + ], + "score": 0.84, + "content": "\\mathcal { N } _ { i } = \\{ j \\in \\mathcal { V } \\mid ( j , i ) \\in \\mathcal { E } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 431, + 473, + 444 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 104, + 396, + 506, + 444 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 217, + 446, + 394, + 460 + ], + "lines": [ + { + "bbox": [ + 217, + 446, + 394, + 460 + ], + "spans": [ + { + "bbox": [ + 217, + 446, + 394, + 460 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\pmb { h } _ { i } ^ { \\prime } = f _ { \\theta } \\left( \\pmb { h } _ { i } , \\mathrm { A G G R E G A T E } \\left( \\{ \\pmb { h } _ { j } \\ | \\ j \\in \\mathcal { N } _ { i } \\} \\right) \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "5c0f7e69ad5b558b5aa89dba0f3b2dc5e793350b81205f5ed8aaa6ea2bd60471.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 217, + 446, + 394, + 460 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 464, + 505, + 497 + ], + "lines": [ + { + "bbox": [ + 105, + 463, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 163, + 477 + ], + "score": 1.0, + "content": "The design of", + "type": "text" + }, + { + "bbox": [ + 163, + 464, + 170, + 475 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 463, + 506, + 477 + ], + "score": 1.0, + "content": "and AGGREGATE is what mostly distinguishes one type of GNN from the other. For", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 475, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 505, + 487 + ], + "score": 1.0, + "content": "example, a common variant of GraphSAGE (Hamilton et al., 2017) performs an element-wise mean", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 486, + 444, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 302, + 498 + ], + "score": 1.0, + "content": "as AGGREGATE, followed by concatenation with", + "type": "text" + }, + { + "bbox": [ + 302, + 486, + 313, + 497 + ], + "score": 0.87, + "content": "\\boldsymbol { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 486, + 433, + 498 + ], + "score": 1.0, + "content": ", a linear layer and a ReLU as", + "type": "text" + }, + { + "bbox": [ + 433, + 486, + 440, + 497 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 486, + 444, + 498 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 463, + 506, + 498 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 510, + 267, + 521 + ], + "lines": [ + { + "bbox": [ + 105, + 509, + 268, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 268, + 523 + ], + "score": 1.0, + "content": "2.2 GRAPH ATTENTION NETWORKS", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 530, + 505, + 586 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 343, + 543 + ], + "score": 1.0, + "content": "GraphSAGE and many other popular GNN architectures", + "type": "text" + }, + { + "bbox": [ + 344, + 532, + 357, + 541 + ], + "score": 0.27, + "content": "\\mathrm { { X u } }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 531, + 506, + 543 + ], + "score": 1.0, + "content": "et al., 2019; Duvenaud et al., 2015)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 187, + 555 + ], + "score": 1.0, + "content": "weigh all neighbors", + "type": "text" + }, + { + "bbox": [ + 187, + 542, + 216, + 553 + ], + "score": 0.91, + "content": "j \\in \\mathcal N _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 542, + 506, + 555 + ], + "score": 1.0, + "content": "with equal importance (e.g., mean or max-pooling as AGGREGATE). To", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 552, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 566 + ], + "score": 1.0, + "content": "address this limitation, GAT (Velickovi ˇ c et al., 2018) instantiates Equation (1) by computing a learned ´", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 561, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 273, + 578 + ], + "score": 1.0, + "content": "weighted average of the representations of", + "type": "text" + }, + { + "bbox": [ + 273, + 564, + 285, + 575 + ], + "score": 0.89, + "content": "{ \\mathcal { N } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 561, + 365, + 578 + ], + "score": 1.0, + "content": ". A scoring function", + "type": "text" + }, + { + "bbox": [ + 366, + 564, + 434, + 574 + ], + "score": 0.91, + "content": "e : \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 561, + 506, + 578 + ], + "score": 1.0, + "content": "computes a score", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 574, + 502, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 166, + 586 + ], + "score": 1.0, + "content": "for every edge", + "type": "text" + }, + { + "bbox": [ + 167, + 575, + 186, + 587 + ], + "score": 0.92, + "content": "( j , i )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 574, + 440, + 586 + ], + "score": 1.0, + "content": ", which indicates the importance of the features of the neighbor", + "type": "text" + }, + { + "bbox": [ + 440, + 577, + 445, + 586 + ], + "score": 0.86, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 574, + 493, + 586 + ], + "score": 1.0, + "content": "to the node", + "type": "text" + }, + { + "bbox": [ + 494, + 577, + 497, + 585 + ], + "score": 0.85, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 574, + 502, + 586 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 531, + 506, + 587 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 208, + 590, + 402, + 606 + ], + "lines": [ + { + "bbox": [ + 208, + 590, + 402, + 606 + ], + "spans": [ + { + "bbox": [ + 208, + 590, + 402, + 606 + ], + "score": 0.91, + "content": "e ( \\pmb { h } _ { i } , \\pmb { h } _ { j } ) = \\mathrm { L e a k y R e L U } ( \\pmb { a } ^ { \\top } \\cdot [ \\pmb { W } \\pmb { h } _ { i } | | \\pmb { W } \\pmb { h } _ { j } | ) ", + "type": "interline_equation", + "image_path": "fdf1aac9f4eb59b6d40db1e9082b54ae55306d8a0e6db961c7777c502a2c36f3.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 208, + 590, + 402, + 606 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 610, + 507, + 634 + ], + "lines": [ + { + "bbox": [ + 104, + 607, + 507, + 626 + ], + "spans": [ + { + "bbox": [ + 104, + 607, + 132, + 626 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 610, + 170, + 622 + ], + "score": 0.88, + "content": "\\pmb { a } \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 607, + 174, + 626 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 174, + 610, + 225, + 622 + ], + "score": 0.89, + "content": "W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 607, + 290, + 626 + ], + "score": 1.0, + "content": "are learned, and", + "type": "text" + }, + { + "bbox": [ + 290, + 612, + 297, + 624 + ], + "score": 0.85, + "content": "\\parallel", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 607, + 507, + 626 + ], + "score": 1.0, + "content": "denotes vector concatenation. These attention scores", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 622, + 501, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 250, + 635 + ], + "score": 1.0, + "content": "are normalized across all neighbors", + "type": "text" + }, + { + "bbox": [ + 250, + 623, + 279, + 634 + ], + "score": 0.92, + "content": "j \\in \\mathcal N _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 622, + 501, + 635 + ], + "score": 1.0, + "content": "using softmax, and the attention function is defined as:", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 104, + 607, + 507, + 635 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 189, + 637, + 422, + 666 + ], + "lines": [ + { + "bbox": [ + 189, + 637, + 422, + 666 + ], + "spans": [ + { + "bbox": [ + 189, + 637, + 422, + 666 + ], + "score": 0.94, + "content": "\\alpha _ { i j } = \\mathrm { s o f t m a x } _ { j } \\left( e \\left( h _ { i } , h _ { j } \\right) \\right) = \\frac { \\exp { \\left( e \\left( h _ { i } , h _ { j } \\right) \\right) } } { \\sum _ { j ^ { \\prime } \\in { \\mathcal { N } } _ { i } } \\exp { \\left( e \\left( h _ { i } , h _ { j ^ { \\prime } } \\right) \\right) } }", + "type": "interline_equation", + "image_path": "b4309c7d70cbdbb892256ebb05106e903830abef53b0557adaa3453de5e3f136.jpg" + } + ] + } + ], + "index": 40.5, + "virtual_lines": [ + { + "bbox": [ + 189, + 637, + 422, + 651.5 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 189, + 651.5, + 422, + 666.0 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 669, + 507, + 692 + ], + "lines": [ + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "score": 1.0, + "content": "Then, GAT computes a weighted average of the transformed features of the neighbor nodes (followed", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 681, + 485, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 176, + 693 + ], + "score": 1.0, + "content": "by a nonlinearity", + "type": "text" + }, + { + "bbox": [ + 177, + 682, + 184, + 690 + ], + "score": 0.71, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 681, + 303, + 693 + ], + "score": 1.0, + "content": ") as the new representation of", + "type": "text" + }, + { + "bbox": [ + 303, + 683, + 307, + 690 + ], + "score": 0.81, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 681, + 485, + 693 + ], + "score": 1.0, + "content": ", using the normalized attention coefficients:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 668, + 506, + 693 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 695, + 367, + 717 + ], + "lines": [ + { + "bbox": [ + 243, + 695, + 367, + 717 + ], + "spans": [ + { + "bbox": [ + 243, + 695, + 367, + 717 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\pmb { h } _ { i } ^ { \\prime } = \\sigma \\left( \\sum _ { j \\in \\mathcal { N } _ { i } } \\pmb { \\alpha } _ { i j } \\cdot \\pmb { W } \\pmb { h } _ { j } \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "e616a2cc35af6e2807af1d8d3102e0a51231e3b5e6c203fb765b2da392727b92.jpg" + } + ] + } + ], + "index": 44, + "virtual_lines": [ + { + "bbox": [ + 243, + 695, + 367, + 717 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 410, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 409, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 409, + 732 + ], + "score": 1.0, + "content": "From now on, we will refer to Equations (2) to (4) as the definition of GAT.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45, + "bbox_fs": [ + 106, + 720, + 409, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 450, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 452, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 452, + 96 + ], + "score": 1.0, + "content": "3 THE EXPRESSIVE POWER OF GRAPH ATTENTION MECHANISMS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 105, + 505, + 140 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "In this section, we explain why attention is limited when it is not dynamic (Section 3.1). We then", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 506, + 129 + ], + "score": 1.0, + "content": "show that GAT is severely constrained, because it can only compute static attention (Section 3.2).", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 497, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 497, + 140 + ], + "score": 1.0, + "content": "Next, we show how GAT can be fixed (Section 3.3), by simply modifying the order of operations.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 144, + 506, + 178 + ], + "lines": [ + { + "bbox": [ + 105, + 144, + 507, + 158 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 507, + 158 + ], + "score": 1.0, + "content": "We refer to a neural architecture (e.g., the scoring or the attention function of GAT) as a family of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 156, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 506, + 168 + ], + "score": 1.0, + "content": "functions, parameterized by the learned parameters. An element in the family is a concrete function", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 166, + 505, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 166, + 325, + 179 + ], + "score": 1.0, + "content": "with specific trained weights. In the following, we use", + "type": "text" + }, + { + "bbox": [ + 325, + 167, + 338, + 179 + ], + "score": 0.87, + "content": "[ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 166, + 406, + 179 + ], + "score": 1.0, + "content": "to denote the set", + "type": "text" + }, + { + "bbox": [ + 407, + 167, + 500, + 179 + ], + "score": 0.93, + "content": "[ n ] = \\{ 1 , 2 , . . . , n \\} \\subset \\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 166, + 505, + 179 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "title", + "bbox": [ + 107, + 190, + 322, + 202 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 322, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 322, + 203 + ], + "score": 1.0, + "content": "3.1 THE IMPORTANCE OF DYNAMIC WEIGHTING", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 211, + 505, + 245 + ], + "lines": [ + { + "bbox": [ + 105, + 210, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 505, + 225 + ], + "score": 1.0, + "content": "Attention is a mechanism for computing a distribution over a set of input key vectors, given an", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "score": 1.0, + "content": "additional query vector. If the attention function always weighs one key at least as much as any other", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 234, + 412, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 412, + 245 + ], + "score": 1.0, + "content": "key, unconditioned on the query, we say that this attention function is static:", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 247, + 505, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 484, + 261 + ], + "score": 1.0, + "content": "Definition 3.1 (Static attention). A (possibly infinite) family of scoring functions", + "type": "text" + }, + { + "bbox": [ + 485, + 248, + 505, + 259 + ], + "score": 0.85, + "content": "\\mathcal { F } \\subseteq", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 110, + 257, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 110, + 258, + 174, + 271 + ], + "score": 0.87, + "content": "\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 257, + 394, + 273 + ], + "score": 1.0, + "content": "computes static scoring for a given set of key vectors", + "type": "text" + }, + { + "bbox": [ + 395, + 258, + 487, + 271 + ], + "score": 0.92, + "content": "\\mathbb { K } = \\{ k _ { 1 } , . . . , k _ { n } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 257, + 506, + 273 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 270, + 504, + 286 + ], + "spans": [ + { + "bbox": [ + 104, + 270, + 161, + 286 + ], + "score": 1.0, + "content": "query vectors", + "type": "text" + }, + { + "bbox": [ + 162, + 271, + 254, + 284 + ], + "score": 0.92, + "content": "\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 270, + 303, + 286 + ], + "score": 1.0, + "content": ", if for every", + "type": "text" + }, + { + "bbox": [ + 303, + 272, + 330, + 283 + ], + "score": 0.91, + "content": "f \\in { \\mathcal { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 270, + 469, + 286 + ], + "score": 1.0, + "content": "there exists a “highest scoring” key", + "type": "text" + }, + { + "bbox": [ + 470, + 272, + 504, + 284 + ], + "score": 0.91, + "content": "j _ { f } \\in [ n ]", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 281, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 212, + 299 + ], + "score": 1.0, + "content": "such that for every query", + "type": "text" + }, + { + "bbox": [ + 213, + 284, + 245, + 296 + ], + "score": 0.92, + "content": "i \\in [ m ]", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 281, + 282, + 299 + ], + "score": 1.0, + "content": "and key", + "type": "text" + }, + { + "bbox": [ + 282, + 284, + 314, + 296 + ], + "score": 0.92, + "content": "j \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 281, + 368, + 299 + ], + "score": 1.0, + "content": "it holds that", + "type": "text" + }, + { + "bbox": [ + 368, + 283, + 466, + 297 + ], + "score": 0.9, + "content": "f \\left( \\boldsymbol { q } _ { i } , \\boldsymbol { k } _ { j _ { f } } \\right) \\geq f \\left( \\boldsymbol { q } _ { i } , \\boldsymbol { k } _ { j } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 281, + 505, + 299 + ], + "score": 1.0, + "content": ". We say", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 294, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 376, + 308 + ], + "score": 1.0, + "content": "that a family of attention functions computes static attention given", + "type": "text" + }, + { + "bbox": [ + 376, + 296, + 385, + 305 + ], + "score": 0.74, + "content": "\\Bbbk", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 294, + 403, + 308 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 403, + 296, + 412, + 306 + ], + "score": 0.85, + "content": "\\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 294, + 506, + 308 + ], + "score": 1.0, + "content": ", if its scoring function", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 306, + 463, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 463, + 319 + ], + "score": 1.0, + "content": "computes static scoring, possibly followed by monotonic normalization such as softmax.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 107, + 326, + 505, + 360 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 333, + 338 + ], + "score": 1.0, + "content": "Static attention is very limited because every function", + "type": "text" + }, + { + "bbox": [ + 334, + 326, + 364, + 338 + ], + "score": 0.92, + "content": "f \\in { \\mathcal { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 326, + 506, + 338 + ], + "score": 1.0, + "content": "has a key that is always selected,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "regardless of the query. Such functions cannot model situations where different keys have different", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 348, + 409, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 409, + 361 + ], + "score": 1.0, + "content": "relevance to different queries. Static attention is demonstrated in Figure 1a.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 365, + 366, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 365, + 367, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 367, + 379 + ], + "score": 1.0, + "content": "The general and powerful form of attention is dynamic attention:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 379, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 378, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 484, + 392 + ], + "score": 1.0, + "content": "Definition 3.2 (Dynamic attention). A (possibly infinite) family of scoring functions", + "type": "text" + }, + { + "bbox": [ + 485, + 379, + 505, + 390 + ], + "score": 0.84, + "content": "\\mathcal { F } \\subseteq", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 109, + 388, + 504, + 403 + ], + "spans": [ + { + "bbox": [ + 109, + 389, + 173, + 403 + ], + "score": 0.88, + "content": "\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 388, + 412, + 403 + ], + "score": 1.0, + "content": "computes dynamic scoring for a given set of key vectors", + "type": "text" + }, + { + "bbox": [ + 412, + 390, + 504, + 402 + ], + "score": 0.9, + "content": "\\mathbb { K } = \\{ \\boldsymbol { k } _ { 1 } , . . . , \\boldsymbol { k } _ { n } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 402, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 179, + 416 + ], + "score": 1.0, + "content": "and query vectors", + "type": "text" + }, + { + "bbox": [ + 180, + 402, + 272, + 415 + ], + "score": 0.91, + "content": "\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 402, + 352, + 416 + ], + "score": 1.0, + "content": ", if for any mapping", + "type": "text" + }, + { + "bbox": [ + 353, + 404, + 360, + 415 + ], + "score": 0.58, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 402, + 365, + 416 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 365, + 403, + 407, + 415 + ], + "score": 0.92, + "content": "[ m ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 402, + 456, + 416 + ], + "score": 1.0, + "content": "there exists", + "type": "text" + }, + { + "bbox": [ + 456, + 403, + 483, + 415 + ], + "score": 0.92, + "content": "f \\in { \\mathcal { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 402, + 506, + 416 + ], + "score": 1.0, + "content": "such", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 413, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 104, + 413, + 178, + 430 + ], + "score": 1.0, + "content": "that for any query", + "type": "text" + }, + { + "bbox": [ + 179, + 416, + 209, + 427 + ], + "score": 0.92, + "content": "i \\in [ m ]", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 413, + 259, + 430 + ], + "score": 1.0, + "content": "and any key", + "type": "text" + }, + { + "bbox": [ + 260, + 415, + 420, + 428 + ], + "score": 0.71, + "content": "j _ { \\neq \\varphi ( i ) } \\in [ n ] \\colon f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right) > f \\left( q _ { i } , k _ { j } \\right) .", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 413, + 506, + 430 + ], + "score": 1.0, + "content": ". We say that a family", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 425, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 330, + 439 + ], + "score": 1.0, + "content": "of attention functions computes dynamic attention for", + "type": "text" + }, + { + "bbox": [ + 330, + 427, + 340, + 436 + ], + "score": 0.75, + "content": "\\mathbb { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 425, + 358, + 439 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 359, + 427, + 368, + 437 + ], + "score": 0.84, + "content": "\\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 425, + 505, + 439 + ], + "score": 1.0, + "content": ", if its scoring function computes", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 436, + 434, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 434, + 450 + ], + "score": 1.0, + "content": "dynamic scoring, possibly followed by monotonic normalization such as softmax.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 457, + 505, + 503 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 297, + 471 + ], + "score": 1.0, + "content": "That is, dynamic attention can select every key", + "type": "text" + }, + { + "bbox": [ + 297, + 457, + 317, + 469 + ], + "score": 0.91, + "content": "\\varphi \\left( i \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 456, + 383, + 471 + ], + "score": 1.0, + "content": "using the query", + "type": "text" + }, + { + "bbox": [ + 383, + 458, + 388, + 468 + ], + "score": 0.63, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 456, + 437, + 471 + ], + "score": 1.0, + "content": ", by making", + "type": "text" + }, + { + "bbox": [ + 437, + 456, + 488, + 470 + ], + "score": 0.93, + "content": "f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 456, + 506, + 471 + ], + "score": 1.0, + "content": "the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 469, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 155, + 482 + ], + "score": 1.0, + "content": "maximal in", + "type": "text" + }, + { + "bbox": [ + 156, + 469, + 243, + 482 + ], + "score": 0.91, + "content": "\\left\\{ f \\left( \\pmb { q } _ { i } , \\pmb { k } _ { j } \\right) \\mid j \\in [ n ] \\right\\}", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 469, + 506, + 482 + ], + "score": 1.0, + "content": ". Note that dynamic and static attention are exclusive properties,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "but they are not complementary. Further, every dynamic attention family has strict subsets of static", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 491, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 274, + 503 + ], + "score": 1.0, + "content": "attention families with respect to the same", + "type": "text" + }, + { + "bbox": [ + 274, + 491, + 284, + 501 + ], + "score": 0.77, + "content": "\\mathbb { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 491, + 301, + 503 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 302, + 492, + 310, + 502 + ], + "score": 0.84, + "content": "\\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 491, + 506, + 503 + ], + "score": 1.0, + "content": ". Dynamic attention is demonstrated in Figure 1b.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 105, + 510, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 524 + ], + "score": 1.0, + "content": "Attending by decaying Another way to think about attention is the ability to “focus” on the most", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "relevant inputs, given a query. Focusing is only possible by decaying other inputs, i.e., giving these", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "decayed inputs lower scores than others. If one key is always given an equal or greater attention score", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 543, + 479, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 479, + 557 + ], + "score": 1.0, + "content": "than other keys (as in static attention), no query can ignore this key or decay this key’s score.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + }, + { + "type": "title", + "bbox": [ + 108, + 568, + 292, + 579 + ], + "lines": [ + { + "bbox": [ + 105, + 568, + 294, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 294, + 581 + ], + "score": 1.0, + "content": "3.2 THE LIMITED EXPRESSIVITY OF GAT", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 589, + 505, + 633 + ], + "lines": [ + { + "bbox": [ + 106, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 227, + 601 + ], + "score": 1.0, + "content": "Although the scoring function", + "type": "text" + }, + { + "bbox": [ + 228, + 591, + 234, + 599 + ], + "score": 0.49, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "can be defined in various ways, the original definition of Velickovi ˇ c´", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "et al. (2018) (Equation (2)) has become the de facto practice: it has spread to a variety of domains and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 610, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 104, + 610, + 505, + 624 + ], + "score": 1.0, + "content": "is now the standard implementation of “graph attention network” in all popular GNN libraries (Fey", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 622, + 500, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 500, + 634 + ], + "score": 1.0, + "content": "and Lenssen, 2019; Wang et al., 2019b; Dwivedi et al., 2020; Gordic, 2020; Brockschmidt, 2020). ´", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 107, + 639, + 504, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 638, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 505, + 652 + ], + "score": 1.0, + "content": "The motivation of GAT is to compute a representation for every node as a weighted average of its", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 650, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 650, + 505, + 662 + ], + "score": 1.0, + "content": "neighbors. Statedly, GAT is inspired by the attention mechanism of Bahdanau et al. (2014) and the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 661, + 433, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 661, + 433, + 673 + ], + "score": 1.0, + "content": "self-attention mechanism of the Transformer (Vaswani et al., 2017). Nonetheless:", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 105, + 675, + 503, + 698 + ], + "lines": [ + { + "bbox": [ + 105, + 673, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 484, + 689 + ], + "score": 1.0, + "content": "Theorem 1. A GAT layer computes only static attention, for any set of node representations", + "type": "text" + }, + { + "bbox": [ + 484, + 676, + 505, + 686 + ], + "score": 0.83, + "content": "\\mathbb { K } =", + "type": "inline_equation" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 685, + 479, + 698 + ], + "spans": [ + { + "bbox": [ + 106, + 686, + 179, + 698 + ], + "score": 0.92, + "content": "\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 685, + 252, + 698 + ], + "score": 1.0, + "content": ". In particular, for", + "type": "text" + }, + { + "bbox": [ + 253, + 687, + 275, + 696 + ], + "score": 0.87, + "content": "n > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 685, + 479, + 698 + ], + "score": 1.0, + "content": ", a GAT layer does not compute dynamic attention.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 151, + 723 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 151, + 710, + 198, + 722 + ], + "score": 0.92, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 708, + 381, + 723 + ], + "score": 1.0, + "content": "be a graph modeled by a GAT layer with some", + "type": "text" + }, + { + "bbox": [ + 381, + 712, + 388, + 720 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 708, + 406, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 406, + 710, + 419, + 720 + ], + "score": 0.51, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "values (Equations (2)", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 507, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 273, + 734 + ], + "score": 1.0, + "content": "and (3)), and having node representations", + "type": "text" + }, + { + "bbox": [ + 273, + 721, + 324, + 732 + ], + "score": 0.93, + "content": "\\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 720, + 420, + 734 + ], + "score": 1.0, + "content": ". The learned parameter", + "type": "text" + }, + { + "bbox": [ + 420, + 722, + 428, + 730 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 720, + 507, + 734 + ], + "score": 1.0, + "content": "can be written as a", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 450, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 452, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 452, + 96 + ], + "score": 1.0, + "content": "3 THE EXPRESSIVE POWER OF GRAPH ATTENTION MECHANISMS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 105, + 505, + 140 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "In this section, we explain why attention is limited when it is not dynamic (Section 3.1). We then", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 117, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 506, + 129 + ], + "score": 1.0, + "content": "show that GAT is severely constrained, because it can only compute static attention (Section 3.2).", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 497, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 497, + 140 + ], + "score": 1.0, + "content": "Next, we show how GAT can be fixed (Section 3.3), by simply modifying the order of operations.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 105, + 506, + 140 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 144, + 506, + 178 + ], + "lines": [ + { + "bbox": [ + 105, + 144, + 507, + 158 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 507, + 158 + ], + "score": 1.0, + "content": "We refer to a neural architecture (e.g., the scoring or the attention function of GAT) as a family of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 156, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 506, + 168 + ], + "score": 1.0, + "content": "functions, parameterized by the learned parameters. An element in the family is a concrete function", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 166, + 505, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 166, + 325, + 179 + ], + "score": 1.0, + "content": "with specific trained weights. In the following, we use", + "type": "text" + }, + { + "bbox": [ + 325, + 167, + 338, + 179 + ], + "score": 0.87, + "content": "[ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 166, + 406, + 179 + ], + "score": 1.0, + "content": "to denote the set", + "type": "text" + }, + { + "bbox": [ + 407, + 167, + 500, + 179 + ], + "score": 0.93, + "content": "[ n ] = \\{ 1 , 2 , . . . , n \\} \\subset \\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 166, + 505, + 179 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 144, + 507, + 179 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 190, + 322, + 202 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 322, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 322, + 203 + ], + "score": 1.0, + "content": "3.1 THE IMPORTANCE OF DYNAMIC WEIGHTING", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 211, + 505, + 245 + ], + "lines": [ + { + "bbox": [ + 105, + 210, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 505, + 225 + ], + "score": 1.0, + "content": "Attention is a mechanism for computing a distribution over a set of input key vectors, given an", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "score": 1.0, + "content": "additional query vector. If the attention function always weighs one key at least as much as any other", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 234, + 412, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 412, + 245 + ], + "score": 1.0, + "content": "key, unconditioned on the query, we say that this attention function is static:", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 210, + 505, + 245 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 247, + 505, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 484, + 261 + ], + "score": 1.0, + "content": "Definition 3.1 (Static attention). A (possibly infinite) family of scoring functions", + "type": "text" + }, + { + "bbox": [ + 485, + 248, + 505, + 259 + ], + "score": 0.85, + "content": "\\mathcal { F } \\subseteq", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 110, + 257, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 110, + 258, + 174, + 271 + ], + "score": 0.87, + "content": "\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 257, + 394, + 273 + ], + "score": 1.0, + "content": "computes static scoring for a given set of key vectors", + "type": "text" + }, + { + "bbox": [ + 395, + 258, + 487, + 271 + ], + "score": 0.92, + "content": "\\mathbb { K } = \\{ k _ { 1 } , . . . , k _ { n } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 257, + 506, + 273 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 270, + 504, + 286 + ], + "spans": [ + { + "bbox": [ + 104, + 270, + 161, + 286 + ], + "score": 1.0, + "content": "query vectors", + "type": "text" + }, + { + "bbox": [ + 162, + 271, + 254, + 284 + ], + "score": 0.92, + "content": "\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 270, + 303, + 286 + ], + "score": 1.0, + "content": ", if for every", + "type": "text" + }, + { + "bbox": [ + 303, + 272, + 330, + 283 + ], + "score": 0.91, + "content": "f \\in { \\mathcal { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 270, + 469, + 286 + ], + "score": 1.0, + "content": "there exists a “highest scoring” key", + "type": "text" + }, + { + "bbox": [ + 470, + 272, + 504, + 284 + ], + "score": 0.91, + "content": "j _ { f } \\in [ n ]", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 281, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 212, + 299 + ], + "score": 1.0, + "content": "such that for every query", + "type": "text" + }, + { + "bbox": [ + 213, + 284, + 245, + 296 + ], + "score": 0.92, + "content": "i \\in [ m ]", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 281, + 282, + 299 + ], + "score": 1.0, + "content": "and key", + "type": "text" + }, + { + "bbox": [ + 282, + 284, + 314, + 296 + ], + "score": 0.92, + "content": "j \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 281, + 368, + 299 + ], + "score": 1.0, + "content": "it holds that", + "type": "text" + }, + { + "bbox": [ + 368, + 283, + 466, + 297 + ], + "score": 0.9, + "content": "f \\left( \\boldsymbol { q } _ { i } , \\boldsymbol { k } _ { j _ { f } } \\right) \\geq f \\left( \\boldsymbol { q } _ { i } , \\boldsymbol { k } _ { j } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 281, + 505, + 299 + ], + "score": 1.0, + "content": ". We say", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 294, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 376, + 308 + ], + "score": 1.0, + "content": "that a family of attention functions computes static attention given", + "type": "text" + }, + { + "bbox": [ + 376, + 296, + 385, + 305 + ], + "score": 0.74, + "content": "\\Bbbk", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 294, + 403, + 308 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 403, + 296, + 412, + 306 + ], + "score": 0.85, + "content": "\\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 294, + 506, + 308 + ], + "score": 1.0, + "content": ", if its scoring function", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 306, + 463, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 463, + 319 + ], + "score": 1.0, + "content": "computes static scoring, possibly followed by monotonic normalization such as softmax.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5, + "bbox_fs": [ + 104, + 245, + 506, + 319 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 326, + 505, + 360 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 333, + 338 + ], + "score": 1.0, + "content": "Static attention is very limited because every function", + "type": "text" + }, + { + "bbox": [ + 334, + 326, + 364, + 338 + ], + "score": 0.92, + "content": "f \\in { \\mathcal { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 326, + 506, + 338 + ], + "score": 1.0, + "content": "has a key that is always selected,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "regardless of the query. Such functions cannot model situations where different keys have different", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 348, + 409, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 409, + 361 + ], + "score": 1.0, + "content": "relevance to different queries. Static attention is demonstrated in Figure 1a.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 106, + 326, + 506, + 361 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 365, + 366, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 365, + 367, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 367, + 379 + ], + "score": 1.0, + "content": "The general and powerful form of attention is dynamic attention:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 365, + 367, + 379 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 379, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 378, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 484, + 392 + ], + "score": 1.0, + "content": "Definition 3.2 (Dynamic attention). A (possibly infinite) family of scoring functions", + "type": "text" + }, + { + "bbox": [ + 485, + 379, + 505, + 390 + ], + "score": 0.84, + "content": "\\mathcal { F } \\subseteq", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 109, + 388, + 504, + 403 + ], + "spans": [ + { + "bbox": [ + 109, + 389, + 173, + 403 + ], + "score": 0.88, + "content": "\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 388, + 412, + 403 + ], + "score": 1.0, + "content": "computes dynamic scoring for a given set of key vectors", + "type": "text" + }, + { + "bbox": [ + 412, + 390, + 504, + 402 + ], + "score": 0.9, + "content": "\\mathbb { K } = \\{ \\boldsymbol { k } _ { 1 } , . . . , \\boldsymbol { k } _ { n } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 402, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 179, + 416 + ], + "score": 1.0, + "content": "and query vectors", + "type": "text" + }, + { + "bbox": [ + 180, + 402, + 272, + 415 + ], + "score": 0.91, + "content": "\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 402, + 352, + 416 + ], + "score": 1.0, + "content": ", if for any mapping", + "type": "text" + }, + { + "bbox": [ + 353, + 404, + 360, + 415 + ], + "score": 0.58, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 402, + 365, + 416 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 365, + 403, + 407, + 415 + ], + "score": 0.92, + "content": "[ m ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 402, + 456, + 416 + ], + "score": 1.0, + "content": "there exists", + "type": "text" + }, + { + "bbox": [ + 456, + 403, + 483, + 415 + ], + "score": 0.92, + "content": "f \\in { \\mathcal { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 402, + 506, + 416 + ], + "score": 1.0, + "content": "such", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 413, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 104, + 413, + 178, + 430 + ], + "score": 1.0, + "content": "that for any query", + "type": "text" + }, + { + "bbox": [ + 179, + 416, + 209, + 427 + ], + "score": 0.92, + "content": "i \\in [ m ]", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 413, + 259, + 430 + ], + "score": 1.0, + "content": "and any key", + "type": "text" + }, + { + "bbox": [ + 260, + 415, + 420, + 428 + ], + "score": 0.71, + "content": "j _ { \\neq \\varphi ( i ) } \\in [ n ] \\colon f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right) > f \\left( q _ { i } , k _ { j } \\right) .", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 413, + 506, + 430 + ], + "score": 1.0, + "content": ". We say that a family", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 425, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 330, + 439 + ], + "score": 1.0, + "content": "of attention functions computes dynamic attention for", + "type": "text" + }, + { + "bbox": [ + 330, + 427, + 340, + 436 + ], + "score": 0.75, + "content": "\\mathbb { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 425, + 358, + 439 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 359, + 427, + 368, + 437 + ], + "score": 0.84, + "content": "\\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 425, + 505, + 439 + ], + "score": 1.0, + "content": ", if its scoring function computes", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 436, + 434, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 434, + 450 + ], + "score": 1.0, + "content": "dynamic scoring, possibly followed by monotonic normalization such as softmax.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23.5, + "bbox_fs": [ + 104, + 378, + 506, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 457, + 505, + 503 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 297, + 471 + ], + "score": 1.0, + "content": "That is, dynamic attention can select every key", + "type": "text" + }, + { + "bbox": [ + 297, + 457, + 317, + 469 + ], + "score": 0.91, + "content": "\\varphi \\left( i \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 456, + 383, + 471 + ], + "score": 1.0, + "content": "using the query", + "type": "text" + }, + { + "bbox": [ + 383, + 458, + 388, + 468 + ], + "score": 0.63, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 456, + 437, + 471 + ], + "score": 1.0, + "content": ", by making", + "type": "text" + }, + { + "bbox": [ + 437, + 456, + 488, + 470 + ], + "score": 0.93, + "content": "f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 456, + 506, + 471 + ], + "score": 1.0, + "content": "the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 469, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 155, + 482 + ], + "score": 1.0, + "content": "maximal in", + "type": "text" + }, + { + "bbox": [ + 156, + 469, + 243, + 482 + ], + "score": 0.91, + "content": "\\left\\{ f \\left( \\pmb { q } _ { i } , \\pmb { k } _ { j } \\right) \\mid j \\in [ n ] \\right\\}", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 469, + 506, + 482 + ], + "score": 1.0, + "content": ". Note that dynamic and static attention are exclusive properties,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "but they are not complementary. Further, every dynamic attention family has strict subsets of static", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 491, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 274, + 503 + ], + "score": 1.0, + "content": "attention families with respect to the same", + "type": "text" + }, + { + "bbox": [ + 274, + 491, + 284, + 501 + ], + "score": 0.77, + "content": "\\mathbb { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 491, + 301, + 503 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 302, + 492, + 310, + 502 + ], + "score": 0.84, + "content": "\\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 491, + 506, + 503 + ], + "score": 1.0, + "content": ". Dynamic attention is demonstrated in Figure 1b.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 456, + 506, + 503 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 105, + 510, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 524 + ], + "score": 1.0, + "content": "Attending by decaying Another way to think about attention is the ability to “focus” on the most", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "relevant inputs, given a query. Focusing is only possible by decaying other inputs, i.e., giving these", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "decayed inputs lower scores than others. If one key is always given an equal or greater attention score", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 543, + 479, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 479, + 557 + ], + "score": 1.0, + "content": "than other keys (as in static attention), no query can ignore this key or decay this key’s score.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 510, + 506, + 557 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 568, + 292, + 579 + ], + "lines": [ + { + "bbox": [ + 105, + 568, + 294, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 294, + 581 + ], + "score": 1.0, + "content": "3.2 THE LIMITED EXPRESSIVITY OF GAT", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 589, + 505, + 633 + ], + "lines": [ + { + "bbox": [ + 106, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 227, + 601 + ], + "score": 1.0, + "content": "Although the scoring function", + "type": "text" + }, + { + "bbox": [ + 228, + 591, + 234, + 599 + ], + "score": 0.49, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "can be defined in various ways, the original definition of Velickovi ˇ c´", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "et al. (2018) (Equation (2)) has become the de facto practice: it has spread to a variety of domains and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 610, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 104, + 610, + 505, + 624 + ], + "score": 1.0, + "content": "is now the standard implementation of “graph attention network” in all popular GNN libraries (Fey", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 622, + 500, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 500, + 634 + ], + "score": 1.0, + "content": "and Lenssen, 2019; Wang et al., 2019b; Dwivedi et al., 2020; Gordic, 2020; Brockschmidt, 2020). ´", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5, + "bbox_fs": [ + 104, + 589, + 506, + 634 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 639, + 504, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 638, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 505, + 652 + ], + "score": 1.0, + "content": "The motivation of GAT is to compute a representation for every node as a weighted average of its", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 650, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 650, + 505, + 662 + ], + "score": 1.0, + "content": "neighbors. Statedly, GAT is inspired by the attention mechanism of Bahdanau et al. (2014) and the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 661, + 433, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 661, + 433, + 673 + ], + "score": 1.0, + "content": "self-attention mechanism of the Transformer (Vaswani et al., 2017). Nonetheless:", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 638, + 505, + 673 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 675, + 503, + 698 + ], + "lines": [ + { + "bbox": [ + 105, + 673, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 484, + 689 + ], + "score": 1.0, + "content": "Theorem 1. A GAT layer computes only static attention, for any set of node representations", + "type": "text" + }, + { + "bbox": [ + 484, + 676, + 505, + 686 + ], + "score": 0.83, + "content": "\\mathbb { K } =", + "type": "inline_equation" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 685, + 479, + 698 + ], + "spans": [ + { + "bbox": [ + 106, + 686, + 179, + 698 + ], + "score": 0.92, + "content": "\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 685, + 252, + 698 + ], + "score": 1.0, + "content": ". In particular, for", + "type": "text" + }, + { + "bbox": [ + 253, + 687, + 275, + 696 + ], + "score": 0.87, + "content": "n > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 685, + 479, + 698 + ], + "score": 1.0, + "content": ", a GAT layer does not compute dynamic attention.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 673, + 505, + 698 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 151, + 723 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 151, + 710, + 198, + 722 + ], + "score": 0.92, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 708, + 381, + 723 + ], + "score": 1.0, + "content": "be a graph modeled by a GAT layer with some", + "type": "text" + }, + { + "bbox": [ + 381, + 712, + 388, + 720 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 708, + 406, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 406, + 710, + 419, + 720 + ], + "score": 0.51, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "values (Equations (2)", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 507, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 273, + 734 + ], + "score": 1.0, + "content": "and (3)), and having node representations", + "type": "text" + }, + { + "bbox": [ + 273, + 721, + 324, + 732 + ], + "score": 0.93, + "content": "\\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 720, + 420, + 734 + ], + "score": 1.0, + "content": ". The learned parameter", + "type": "text" + }, + { + "bbox": [ + 420, + 722, + 428, + 730 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 720, + 507, + 734 + ], + "score": 1.0, + "content": "can be written as a", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 104, + 79, + 500, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 164, + 96 + ], + "score": 1.0, + "content": "concatenation", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 165, + 81, + 247, + 95 + ], + "score": 0.92, + "content": "\\pmb { a } = [ \\pmb { a } _ { 1 } | | \\pmb { a } _ { 2 } ] \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 248, + 79, + 287, + 96 + ], + "score": 1.0, + "content": "such that", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 288, + 81, + 341, + 94 + ], + "score": 0.92, + "content": "\\mathbf { \\boldsymbol { a } } _ { 1 } , \\mathbf { \\boldsymbol { a } } _ { 2 } \\in \\mathbb { R } ^ { d ^ { \\prime } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 341, + 79, + 500, + 96 + ], + "score": 1.0, + "content": ", and Equation (2) can be re-written as:", + "type": "text", + "cross_page": true + } + ], + "index": 0 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 708, + 507, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 80, + 496, + 95 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 500, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 164, + 96 + ], + "score": 1.0, + "content": "concatenation", + "type": "text" + }, + { + "bbox": [ + 165, + 81, + 247, + 95 + ], + "score": 0.92, + "content": "\\pmb { a } = [ \\pmb { a } _ { 1 } | | \\pmb { a } _ { 2 } ] \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 79, + 287, + 96 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 288, + 81, + 341, + 94 + ], + "score": 0.92, + "content": "\\mathbf { \\boldsymbol { a } } _ { 1 } , \\mathbf { \\boldsymbol { a } } _ { 2 } \\in \\mathbb { R } ^ { d ^ { \\prime } }", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 79, + 500, + 96 + ], + "score": 1.0, + "content": ", and Equation (2) can be re-written as:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 204, + 99, + 406, + 115 + ], + "lines": [ + { + "bbox": [ + 204, + 99, + 406, + 115 + ], + "spans": [ + { + "bbox": [ + 204, + 99, + 406, + 115 + ], + "score": 0.91, + "content": "e \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { a } _ { 1 } ^ { \\top } \\boldsymbol { W } \\boldsymbol { h } _ { i } + \\boldsymbol { a } _ { 2 } ^ { \\top } \\boldsymbol { W } \\boldsymbol { h } _ { j } \\right)", + "type": "interline_equation", + "image_path": "9a37260e84a76d8e8ef9574908ecbecd7391412b40e967772e2bb526c983de31.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 204, + 99, + 406, + 115 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 121, + 505, + 189 + ], + "lines": [ + { + "bbox": [ + 104, + 120, + 507, + 137 + ], + "spans": [ + { + "bbox": [ + 104, + 120, + 131, + 137 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 123, + 140, + 132 + ], + "score": 0.8, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 120, + 254, + 137 + ], + "score": 1.0, + "content": "is finite, there exists a node", + "type": "text" + }, + { + "bbox": [ + 254, + 122, + 295, + 133 + ], + "score": 0.92, + "content": "j _ { m a x } \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 120, + 335, + 137 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 336, + 121, + 387, + 134 + ], + "score": 0.92, + "content": "{ \\pmb a } _ { 2 } ^ { \\top } { \\pmb W } { \\pmb h } _ { j _ { m a x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 120, + 507, + 137 + ], + "score": 1.0, + "content": "is maximal among all nodes", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 160, + 145 + ], + "score": 0.63, + "content": "j \\in \\mathcal { V } \\left( j _ { m a x } \\right.", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 132, + 187, + 145 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 187, + 133, + 198, + 145 + ], + "score": 0.88, + "content": "j _ { f }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "required by Definition 3.1). Due to the monotonicity of LeakyReLU and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 144, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 231, + 156 + ], + "score": 1.0, + "content": "softmax, for every query node", + "type": "text" + }, + { + "bbox": [ + 231, + 145, + 255, + 154 + ], + "score": 0.91, + "content": "i \\in \\mathcal V", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 144, + 295, + 156 + ], + "score": 1.0, + "content": ", the node", + "type": "text" + }, + { + "bbox": [ + 295, + 145, + 316, + 156 + ], + "score": 0.91, + "content": "j _ { m a x }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 144, + 505, + 156 + ], + "score": 1.0, + "content": "also leads to the maximal value of its attention", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 153, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 154, + 168 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 155, + 155, + 210, + 167 + ], + "score": 0.92, + "content": "\\{ \\alpha _ { i j } \\mid j \\in \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 153, + 352, + 168 + ], + "score": 1.0, + "content": ". Thus, from Definition 3.1 directly,", + "type": "text" + }, + { + "bbox": [ + 352, + 157, + 360, + 165 + ], + "score": 0.76, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 153, + 506, + 168 + ], + "score": 1.0, + "content": "computes only static attention. This", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 166, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 174, + 179 + ], + "score": 1.0, + "content": "also implies that", + "type": "text" + }, + { + "bbox": [ + 175, + 169, + 182, + 176 + ], + "score": 0.8, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 166, + 506, + 179 + ], + "score": 1.0, + "content": "does not compute dynamic attention, because in GAT, Definition 3.2 holds only", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 176, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 198, + 190 + ], + "score": 1.0, + "content": "for constant mappings", + "type": "text" + }, + { + "bbox": [ + 198, + 178, + 206, + 189 + ], + "score": 0.83, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 176, + 362, + 190 + ], + "score": 1.0, + "content": "that map all inputs to the same output.", + "type": "text" + }, + { + "bbox": [ + 494, + 177, + 505, + 188 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 200, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 342, + 214 + ], + "score": 1.0, + "content": "The consequence of Theorem 1 is that for any set of nodes", + "type": "text" + }, + { + "bbox": [ + 343, + 202, + 351, + 212 + ], + "score": 0.72, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 200, + 505, + 214 + ], + "score": 1.0, + "content": "and a trained GAT layer, the attention", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 212, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 143, + 225 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 144, + 214, + 151, + 222 + ], + "score": 0.76, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 212, + 505, + 225 + ], + "score": 1.0, + "content": "defines a constant ranking (argsort) of the nodes, unconditioned on the query nodes", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 223, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 224, + 111, + 233 + ], + "score": 0.48, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 111, + 223, + 214, + 236 + ], + "score": 1.0, + "content": ". That is, we can denote", + "type": "text" + }, + { + "bbox": [ + 214, + 223, + 275, + 235 + ], + "score": 0.91, + "content": "s _ { j } = { a } _ { 2 } ^ { \\top } \\bar { W } h _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 223, + 402, + 236 + ], + "score": 1.0, + "content": "and get that for any choice of", + "type": "text" + }, + { + "bbox": [ + 403, + 223, + 414, + 234 + ], + "score": 0.84, + "content": "\\boldsymbol { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 223, + 418, + 236 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 419, + 225, + 426, + 233 + ], + "score": 0.66, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 223, + 505, + 236 + ], + "score": 1.0, + "content": "is monotonic with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 234, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 229, + 248 + ], + "score": 1.0, + "content": "respect to the per-node scores", + "type": "text" + }, + { + "bbox": [ + 229, + 235, + 280, + 246 + ], + "score": 0.89, + "content": "\\{ s _ { j } \\mid j \\in \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 234, + 506, + 248 + ], + "score": 1.0, + "content": ". This global ranking induces the local ranking of every", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 245, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 164, + 258 + ], + "score": 1.0, + "content": "neighborhood", + "type": "text" + }, + { + "bbox": [ + 165, + 245, + 177, + 256 + ], + "score": 0.88, + "content": "{ \\mathcal { N } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 245, + 256, + 258 + ], + "score": 1.0, + "content": ". The only effect of", + "type": "text" + }, + { + "bbox": [ + 256, + 246, + 267, + 256 + ], + "score": 0.88, + "content": "\\boldsymbol { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 245, + 506, + 258 + ], + "score": 1.0, + "content": "is in the “sharpness” of the produced attention distribution.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 255, + 506, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 473, + 269 + ], + "score": 1.0, + "content": "This is demonstrated in Figure 1a (bottom), where different curves denote different queries", + "type": "text" + }, + { + "bbox": [ + 474, + 256, + 491, + 267 + ], + "score": 0.85, + "content": "( h _ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 255, + 506, + 269 + ], + "score": 1.0, + "content": ". A", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 267, + 475, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 475, + 280 + ], + "score": 1.0, + "content": "discussion regarding the generalization to multi-head attention can be found in Appendix C.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 107, + 291, + 358, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 291, + 359, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 359, + 304 + ], + "score": 1.0, + "content": "3.3 BUILDING DYNAMIC GRAPH ATTENTION NETWORKS", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 312, + 505, + 335 + ], + "lines": [ + { + "bbox": [ + 106, + 313, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 325 + ], + "score": 1.0, + "content": "To create a dynamic graph attention network, we modify the order of internal operations in GAT and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 323, + 492, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 492, + 336 + ], + "score": 1.0, + "content": "introduce GATv2 – a simple fix of GAT that has a strictly more expressive attention mechanism.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 108, + 343, + 505, + 388 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 355 + ], + "score": 1.0, + "content": "GATv2 The main problem in the standard GAT scoring function (Equation (2)) is that the learned", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 355, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 132, + 367 + ], + "score": 1.0, + "content": "layers", + "type": "text" + }, + { + "bbox": [ + 133, + 355, + 146, + 365 + ], + "score": 0.44, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 355, + 164, + 367 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 164, + 356, + 172, + 365 + ], + "score": 0.71, + "content": "^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 355, + 505, + 367 + ], + "score": 1.0, + "content": "are applied consecutively, and thus can be collapsed into a single linear layer. To fix", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 252, + 378 + ], + "score": 1.0, + "content": "this limitation, we simply apply the", + "type": "text" + }, + { + "bbox": [ + 252, + 367, + 260, + 375 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 365, + 468, + 378 + ], + "score": 1.0, + "content": "layer after the nonlinearity (LeakyReLU), and the", + "type": "text" + }, + { + "bbox": [ + 468, + 366, + 481, + 376 + ], + "score": 0.46, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 365, + 506, + 378 + ], + "score": 1.0, + "content": "layer", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "score": 1.0, + "content": "after the concatenation, 5 effectively applying an MLP to compute the score for each query-key pair:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "interline_equation", + "bbox": [ + 284, + 393, + 476, + 426 + ], + "lines": [ + { + "bbox": [ + 284, + 393, + 476, + 426 + ], + "spans": [ + { + "bbox": [ + 284, + 393, + 476, + 426 + ], + "score": 0.87, + "content": "\\begin{array} { r l } & { e \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { a } ^ { \\top } \\cdot \\left[ \\boldsymbol { W } \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { W } \\boldsymbol { h } _ { j } \\right| \\right) \\right. } \\\\ & { e \\left. \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\boldsymbol { a } ^ { \\top } \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { W } \\cdot \\left[ \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { h } _ { j } \\right] \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "761e7cc2081eda77a832d747f69378b8f8b3188a35f388893ea08877645e1885.jpg" + } + ] + } + ], + "index": 22.5, + "virtual_lines": [ + { + "bbox": [ + 284, + 393, + 476, + 409.5 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 284, + 409.5, + 476, + 426.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 109, + 430, + 505, + 442 + ], + "lines": [ + { + "bbox": [ + 107, + 429, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 107, + 429, + 506, + 443 + ], + "score": 1.0, + "content": "The simple modification makes a significant difference in the expressiveness of the attention function:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 445, + 504, + 468 + ], + "lines": [ + { + "bbox": [ + 105, + 443, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 484, + 459 + ], + "score": 1.0, + "content": "Theorem 2. A GATv2 layer computes dynamic attention for any set of node representations", + "type": "text" + }, + { + "bbox": [ + 484, + 446, + 506, + 456 + ], + "score": 0.83, + "content": "\\mathbb { K } =", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 455, + 185, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 180, + 469 + ], + "score": 0.92, + "content": "\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 455, + 185, + 470 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 477, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 106, + 477, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 505, + 489 + ], + "score": 1.0, + "content": "We prove Theorem 2 in Appendix B. The main idea is that we can define an appropriate function", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 488, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 488, + 505, + 501 + ], + "score": 1.0, + "content": "that GATv2 will be a universal approximator (Cybenko, 1989; Hornik, 1991) of. In contrast, GAT", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 499, + 411, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 411, + 511 + ], + "score": 1.0, + "content": "(Equation (52)) cannot approximate any such desired function (Theorem 1).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 506, + 552 + ], + "lines": [ + { + "bbox": [ + 105, + 517, + 507, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 426, + 533 + ], + "score": 1.0, + "content": "Complexity GATv2 has the same time-complexity as GAT’s declared complexity:", + "type": "text" + }, + { + "bbox": [ + 427, + 519, + 503, + 531 + ], + "score": 0.9, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 517, + 507, + 533 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 529, + 507, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 507, + 542 + ], + "score": 1.0, + "content": "However, by merging its linear layers, GAT can be computed faster than stated by Velickovi ˇ c et al. ´", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 540, + 432, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 432, + 553 + ], + "score": 1.0, + "content": "(2018). For a detailed time- and parametric-complexity analysis, see Appendix I.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "title", + "bbox": [ + 108, + 568, + 192, + 581 + ], + "lines": [ + { + "bbox": [ + 104, + 567, + 195, + 583 + ], + "spans": [ + { + "bbox": [ + 104, + 567, + 195, + 583 + ], + "score": 1.0, + "content": "4 EVALUATION", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 593, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 605 + ], + "score": 1.0, + "content": "First, we demonstrate the weakness of GAT using a simple synthetic problem that GAT cannot even fit", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 604, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 506, + 617 + ], + "score": 1.0, + "content": "(cannot even achieve high training accuracy), but is easily solvable by GATv2 (Section 4.1). Second,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "score": 1.0, + "content": "we show that GATv2 is much more robust to edge noise, because its dynamic attention mechanisms", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "allow it to decay noisy (false) edges, while GAT’s performance severely decreases as noise increases", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "(Section 4.2). Finally, we compare GAT and GATv2 across 12 benchmarks overall. (Sections 4.3", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 648, + 499, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 499, + 660 + ], + "score": 1.0, + "content": "to 4.6 and appendix F.4). We find that GAT is inferior to GATv2 across all examined benchmarks.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 107, + 668, + 505, + 713 + ], + "lines": [ + { + "bbox": [ + 105, + 668, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 505, + 681 + ], + "score": 1.0, + "content": "Setup When previous results exist, we take hyperparameters that were tuned for GAT and use them", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 678, + 507, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 507, + 692 + ], + "score": 1.0, + "content": "in GATv2, without any additional tuning. Self-supervision (Kim and Oh, 2021; Rong et al., 2020a),", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 506, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 506, + 703 + ], + "score": 1.0, + "content": "graph regularization (Zhao and Akoglu, 2020; Rong et al., 2020b), and other tricks (Wang, 2021;", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 701, + 505, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 505, + 715 + ], + "score": 1.0, + "content": "Huang et al., 2021) are orthogonal to the contribution of the GNN layer itself, and may further improve", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 114, + 721, + 486, + 732 + ], + "lines": [ + { + "bbox": [ + 118, + 718, + 487, + 735 + ], + "spans": [ + { + "bbox": [ + 118, + 718, + 216, + 735 + ], + "score": 1.0, + "content": "5We also add a bias vector", + "type": "text" + }, + { + "bbox": [ + 216, + 722, + 222, + 730 + ], + "score": 0.25, + "content": "^ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 718, + 487, + 735 + ], + "score": 1.0, + "content": "before applying the nonlinearity, we omit this in Equation (7) for brevity.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 80, + 496, + 95 + ], + "lines": [], + "index": 0, + "bbox_fs": [ + 104, + 79, + 500, + 96 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 204, + 99, + 406, + 115 + ], + "lines": [ + { + "bbox": [ + 204, + 99, + 406, + 115 + ], + "spans": [ + { + "bbox": [ + 204, + 99, + 406, + 115 + ], + "score": 0.91, + "content": "e \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { a } _ { 1 } ^ { \\top } \\boldsymbol { W } \\boldsymbol { h } _ { i } + \\boldsymbol { a } _ { 2 } ^ { \\top } \\boldsymbol { W } \\boldsymbol { h } _ { j } \\right)", + "type": "interline_equation", + "image_path": "9a37260e84a76d8e8ef9574908ecbecd7391412b40e967772e2bb526c983de31.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 204, + 99, + 406, + 115 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 121, + 505, + 189 + ], + "lines": [ + { + "bbox": [ + 104, + 120, + 507, + 137 + ], + "spans": [ + { + "bbox": [ + 104, + 120, + 131, + 137 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 123, + 140, + 132 + ], + "score": 0.8, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 120, + 254, + 137 + ], + "score": 1.0, + "content": "is finite, there exists a node", + "type": "text" + }, + { + "bbox": [ + 254, + 122, + 295, + 133 + ], + "score": 0.92, + "content": "j _ { m a x } \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 120, + 335, + 137 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 336, + 121, + 387, + 134 + ], + "score": 0.92, + "content": "{ \\pmb a } _ { 2 } ^ { \\top } { \\pmb W } { \\pmb h } _ { j _ { m a x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 120, + 507, + 137 + ], + "score": 1.0, + "content": "is maximal among all nodes", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 160, + 145 + ], + "score": 0.63, + "content": "j \\in \\mathcal { V } \\left( j _ { m a x } \\right.", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 132, + 187, + 145 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 187, + 133, + 198, + 145 + ], + "score": 0.88, + "content": "j _ { f }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "required by Definition 3.1). Due to the monotonicity of LeakyReLU and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 144, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 231, + 156 + ], + "score": 1.0, + "content": "softmax, for every query node", + "type": "text" + }, + { + "bbox": [ + 231, + 145, + 255, + 154 + ], + "score": 0.91, + "content": "i \\in \\mathcal V", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 144, + 295, + 156 + ], + "score": 1.0, + "content": ", the node", + "type": "text" + }, + { + "bbox": [ + 295, + 145, + 316, + 156 + ], + "score": 0.91, + "content": "j _ { m a x }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 144, + 505, + 156 + ], + "score": 1.0, + "content": "also leads to the maximal value of its attention", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 153, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 154, + 168 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 155, + 155, + 210, + 167 + ], + "score": 0.92, + "content": "\\{ \\alpha _ { i j } \\mid j \\in \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 153, + 352, + 168 + ], + "score": 1.0, + "content": ". Thus, from Definition 3.1 directly,", + "type": "text" + }, + { + "bbox": [ + 352, + 157, + 360, + 165 + ], + "score": 0.76, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 153, + 506, + 168 + ], + "score": 1.0, + "content": "computes only static attention. This", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 166, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 174, + 179 + ], + "score": 1.0, + "content": "also implies that", + "type": "text" + }, + { + "bbox": [ + 175, + 169, + 182, + 176 + ], + "score": 0.8, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 166, + 506, + 179 + ], + "score": 1.0, + "content": "does not compute dynamic attention, because in GAT, Definition 3.2 holds only", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 176, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 198, + 190 + ], + "score": 1.0, + "content": "for constant mappings", + "type": "text" + }, + { + "bbox": [ + 198, + 178, + 206, + 189 + ], + "score": 0.83, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 176, + 362, + 190 + ], + "score": 1.0, + "content": "that map all inputs to the same output.", + "type": "text" + }, + { + "bbox": [ + 494, + 177, + 505, + 188 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5, + "bbox_fs": [ + 104, + 120, + 507, + 190 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 200, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 342, + 214 + ], + "score": 1.0, + "content": "The consequence of Theorem 1 is that for any set of nodes", + "type": "text" + }, + { + "bbox": [ + 343, + 202, + 351, + 212 + ], + "score": 0.72, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 200, + 505, + 214 + ], + "score": 1.0, + "content": "and a trained GAT layer, the attention", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 212, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 143, + 225 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 144, + 214, + 151, + 222 + ], + "score": 0.76, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 212, + 505, + 225 + ], + "score": 1.0, + "content": "defines a constant ranking (argsort) of the nodes, unconditioned on the query nodes", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 223, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 224, + 111, + 233 + ], + "score": 0.48, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 111, + 223, + 214, + 236 + ], + "score": 1.0, + "content": ". That is, we can denote", + "type": "text" + }, + { + "bbox": [ + 214, + 223, + 275, + 235 + ], + "score": 0.91, + "content": "s _ { j } = { a } _ { 2 } ^ { \\top } \\bar { W } h _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 223, + 402, + 236 + ], + "score": 1.0, + "content": "and get that for any choice of", + "type": "text" + }, + { + "bbox": [ + 403, + 223, + 414, + 234 + ], + "score": 0.84, + "content": "\\boldsymbol { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 223, + 418, + 236 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 419, + 225, + 426, + 233 + ], + "score": 0.66, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 223, + 505, + 236 + ], + "score": 1.0, + "content": "is monotonic with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 234, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 229, + 248 + ], + "score": 1.0, + "content": "respect to the per-node scores", + "type": "text" + }, + { + "bbox": [ + 229, + 235, + 280, + 246 + ], + "score": 0.89, + "content": "\\{ s _ { j } \\mid j \\in \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 234, + 506, + 248 + ], + "score": 1.0, + "content": ". This global ranking induces the local ranking of every", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 245, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 164, + 258 + ], + "score": 1.0, + "content": "neighborhood", + "type": "text" + }, + { + "bbox": [ + 165, + 245, + 177, + 256 + ], + "score": 0.88, + "content": "{ \\mathcal { N } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 245, + 256, + 258 + ], + "score": 1.0, + "content": ". The only effect of", + "type": "text" + }, + { + "bbox": [ + 256, + 246, + 267, + 256 + ], + "score": 0.88, + "content": "\\boldsymbol { h } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 245, + 506, + 258 + ], + "score": 1.0, + "content": "is in the “sharpness” of the produced attention distribution.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 255, + 506, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 473, + 269 + ], + "score": 1.0, + "content": "This is demonstrated in Figure 1a (bottom), where different curves denote different queries", + "type": "text" + }, + { + "bbox": [ + 474, + 256, + 491, + 267 + ], + "score": 0.85, + "content": "( h _ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 255, + 506, + 269 + ], + "score": 1.0, + "content": ". A", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 267, + 475, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 475, + 280 + ], + "score": 1.0, + "content": "discussion regarding the generalization to multi-head attention can be found in Appendix C.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 200, + 506, + 280 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 291, + 358, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 291, + 359, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 359, + 304 + ], + "score": 1.0, + "content": "3.3 BUILDING DYNAMIC GRAPH ATTENTION NETWORKS", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 312, + 505, + 335 + ], + "lines": [ + { + "bbox": [ + 106, + 313, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 325 + ], + "score": 1.0, + "content": "To create a dynamic graph attention network, we modify the order of internal operations in GAT and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 323, + 492, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 492, + 336 + ], + "score": 1.0, + "content": "introduce GATv2 – a simple fix of GAT that has a strictly more expressive attention mechanism.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 106, + 313, + 505, + 336 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 343, + 505, + 388 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 355 + ], + "score": 1.0, + "content": "GATv2 The main problem in the standard GAT scoring function (Equation (2)) is that the learned", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 355, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 132, + 367 + ], + "score": 1.0, + "content": "layers", + "type": "text" + }, + { + "bbox": [ + 133, + 355, + 146, + 365 + ], + "score": 0.44, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 355, + 164, + 367 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 164, + 356, + 172, + 365 + ], + "score": 0.71, + "content": "^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 355, + 505, + 367 + ], + "score": 1.0, + "content": "are applied consecutively, and thus can be collapsed into a single linear layer. To fix", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 365, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 252, + 378 + ], + "score": 1.0, + "content": "this limitation, we simply apply the", + "type": "text" + }, + { + "bbox": [ + 252, + 367, + 260, + 375 + ], + "score": 0.72, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 365, + 468, + 378 + ], + "score": 1.0, + "content": "layer after the nonlinearity (LeakyReLU), and the", + "type": "text" + }, + { + "bbox": [ + 468, + 366, + 481, + 376 + ], + "score": 0.46, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 365, + 506, + 378 + ], + "score": 1.0, + "content": "layer", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "score": 1.0, + "content": "after the concatenation, 5 effectively applying an MLP to compute the score for each query-key pair:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 344, + 506, + 390 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 284, + 393, + 476, + 426 + ], + "lines": [ + { + "bbox": [ + 284, + 393, + 476, + 426 + ], + "spans": [ + { + "bbox": [ + 284, + 393, + 476, + 426 + ], + "score": 0.87, + "content": "\\begin{array} { r l } & { e \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { a } ^ { \\top } \\cdot \\left[ \\boldsymbol { W } \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { W } \\boldsymbol { h } _ { j } \\right| \\right) \\right. } \\\\ & { e \\left. \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\boldsymbol { a } ^ { \\top } \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { W } \\cdot \\left[ \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { h } _ { j } \\right] \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "761e7cc2081eda77a832d747f69378b8f8b3188a35f388893ea08877645e1885.jpg" + } + ] + } + ], + "index": 22.5, + "virtual_lines": [ + { + "bbox": [ + 284, + 393, + 476, + 409.5 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 284, + 409.5, + 476, + 426.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 109, + 430, + 505, + 442 + ], + "lines": [ + { + "bbox": [ + 107, + 429, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 107, + 429, + 506, + 443 + ], + "score": 1.0, + "content": "The simple modification makes a significant difference in the expressiveness of the attention function:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 107, + 429, + 506, + 443 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 445, + 504, + 468 + ], + "lines": [ + { + "bbox": [ + 105, + 443, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 484, + 459 + ], + "score": 1.0, + "content": "Theorem 2. A GATv2 layer computes dynamic attention for any set of node representations", + "type": "text" + }, + { + "bbox": [ + 484, + 446, + 506, + 456 + ], + "score": 0.83, + "content": "\\mathbb { K } =", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 455, + 185, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 180, + 469 + ], + "score": 0.92, + "content": "\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 455, + 185, + 470 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 443, + 506, + 470 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 477, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 106, + 477, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 505, + 489 + ], + "score": 1.0, + "content": "We prove Theorem 2 in Appendix B. The main idea is that we can define an appropriate function", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 488, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 488, + 505, + 501 + ], + "score": 1.0, + "content": "that GATv2 will be a universal approximator (Cybenko, 1989; Hornik, 1991) of. In contrast, GAT", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 499, + 411, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 411, + 511 + ], + "score": 1.0, + "content": "(Equation (52)) cannot approximate any such desired function (Theorem 1).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 106, + 477, + 505, + 511 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 506, + 552 + ], + "lines": [ + { + "bbox": [ + 105, + 517, + 507, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 426, + 533 + ], + "score": 1.0, + "content": "Complexity GATv2 has the same time-complexity as GAT’s declared complexity:", + "type": "text" + }, + { + "bbox": [ + 427, + 519, + 503, + 531 + ], + "score": 0.9, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 517, + 507, + 533 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 529, + 507, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 507, + 542 + ], + "score": 1.0, + "content": "However, by merging its linear layers, GAT can be computed faster than stated by Velickovi ˇ c et al. ´", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 540, + 432, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 432, + 553 + ], + "score": 1.0, + "content": "(2018). For a detailed time- and parametric-complexity analysis, see Appendix I.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 517, + 507, + 553 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 568, + 192, + 581 + ], + "lines": [ + { + "bbox": [ + 104, + 567, + 195, + 583 + ], + "spans": [ + { + "bbox": [ + 104, + 567, + 195, + 583 + ], + "score": 1.0, + "content": "4 EVALUATION", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 593, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 605 + ], + "score": 1.0, + "content": "First, we demonstrate the weakness of GAT using a simple synthetic problem that GAT cannot even fit", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 604, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 506, + 617 + ], + "score": 1.0, + "content": "(cannot even achieve high training accuracy), but is easily solvable by GATv2 (Section 4.1). Second,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "score": 1.0, + "content": "we show that GATv2 is much more robust to edge noise, because its dynamic attention mechanisms", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "allow it to decay noisy (false) edges, while GAT’s performance severely decreases as noise increases", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "(Section 4.2). Finally, we compare GAT and GATv2 across 12 benchmarks overall. (Sections 4.3", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 648, + 499, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 499, + 660 + ], + "score": 1.0, + "content": "to 4.6 and appendix F.4). We find that GAT is inferior to GATv2 across all examined benchmarks.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 594, + 506, + 660 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 668, + 505, + 713 + ], + "lines": [ + { + "bbox": [ + 105, + 668, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 505, + 681 + ], + "score": 1.0, + "content": "Setup When previous results exist, we take hyperparameters that were tuned for GAT and use them", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 678, + 507, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 507, + 692 + ], + "score": 1.0, + "content": "in GATv2, without any additional tuning. Self-supervision (Kim and Oh, 2021; Rong et al., 2020a),", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 506, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 506, + 703 + ], + "score": 1.0, + "content": "graph regularization (Zhao and Akoglu, 2020; Rong et al., 2020b), and other tricks (Wang, 2021;", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 701, + 505, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 505, + 715 + ], + "score": 1.0, + "content": "Huang et al., 2021) are orthogonal to the contribution of the GNN layer itself, and may further improve", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 274, + 506, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 434, + 289 + ], + "score": 1.0, + "content": "all GNNs. In all experiments of GATv2, we constrain the learned matrix by setting", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 435, + 275, + 502, + 288 + ], + "score": 0.92, + "content": "W = [ W ^ { \\prime } | | W ^ { \\prime } ]", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 503, + 274, + 506, + 289 + ], + "score": 1.0, + "content": ",", + "type": "text", + "cross_page": true + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "to rule out the increased number of parameters over GAT as the source of empirical difference (see", + "type": "text", + "cross_page": true + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 298, + 429, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 429, + 309 + ], + "score": 1.0, + "content": "Appendix I.2). Training details, statistics, and code are provided in Appendix D.", + "type": "text", + "cross_page": true + } + ], + "index": 26 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 668, + 507, + 715 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 130, + 79, + 223, + 153 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 130, + 79, + 223, + 153 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 130, + 79, + 223, + 153 + ], + "spans": [ + { + "bbox": [ + 130, + 79, + 223, + 153 + ], + "score": 0.943, + "type": "image", + "image_path": "79aab347d93a4c3bb0bb39665869459e72b57008bc185548856fefb238200d5b.jpg" + } + ] + } + ], + "index": 0.5, + "virtual_lines": [ + { + "bbox": [ + 130, + 79, + 223, + 116.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 130, + 116.0, + 223, + 153.0 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 161, + 248, + 259 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 160, + 250, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 160, + 250, + 172 + ], + "score": 1.0, + "content": "Figure 2: The DICTIONARY-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 172, + 249, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 210, + 183 + ], + "score": 1.0, + "content": "LOOKUP problem of size", + "type": "text" + }, + { + "bbox": [ + 210, + 172, + 230, + 182 + ], + "score": 0.85, + "content": "k { = } 4", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 172, + 249, + 183 + ], + "score": 1.0, + "content": ": ev-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 183, + 248, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 248, + 194 + ], + "score": 1.0, + "content": "ery node in the bottom row has an", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 194, + 246, + 206 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 187, + 206 + ], + "score": 1.0, + "content": "alphabetic attribute", + "type": "text" + }, + { + "bbox": [ + 188, + 194, + 246, + 205 + ], + "score": 0.86, + "content": "( \\{ { \\mathrm { A } } , { \\mathrm { B } } , { \\mathrm { C } } , \\ldots \\} )", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 205, + 248, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 190, + 217 + ], + "score": 1.0, + "content": "and a numeric value", + "type": "text" + }, + { + "bbox": [ + 190, + 206, + 244, + 217 + ], + "score": 0.8, + "content": "( \\{ 1 , 2 , 3 , \\ldots \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 205, + 248, + 217 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 216, + 248, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 216, + 248, + 227 + ], + "score": 1.0, + "content": "every node in the upper row has", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 227, + 249, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 249, + 239 + ], + "score": 1.0, + "content": "only an attribute; the goal is to pre-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 238, + 248, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 248, + 249 + ], + "score": 1.0, + "content": "dict the value for each node in the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 249, + 225, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 225, + 260 + ], + "score": 1.0, + "content": "upper row, using its attribute.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6 + } + ], + "index": 3.25 + }, + { + "type": "image", + "bbox": [ + 258, + 79, + 500, + 213 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 258, + 79, + 500, + 213 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 258, + 79, + 500, + 213 + ], + "spans": [ + { + "bbox": [ + 258, + 79, + 500, + 213 + ], + "score": 0.967, + "type": "image", + "image_path": "eee3bdfe02443e30b1eb1a486a618dd6fcf3d8c448dc16f59b4318aa5c305de0.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 258, + 79, + 500, + 92.4 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 258, + 92.4, + 500, + 105.80000000000001 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 258, + 105.80000000000001, + 500, + 119.20000000000002 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 258, + 119.20000000000002, + 500, + 132.60000000000002 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 258, + 132.60000000000002, + 500, + 146.00000000000003 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 258, + 146.00000000000003, + 500, + 159.40000000000003 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 258, + 159.40000000000003, + 500, + 172.80000000000004 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 258, + 172.80000000000004, + 500, + 186.20000000000005 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 258, + 186.20000000000005, + 500, + 199.60000000000005 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 258, + 199.60000000000005, + 500, + 213.00000000000006 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 254, + 224, + 502, + 258 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 253, + 223, + 502, + 237 + ], + "spans": [ + { + "bbox": [ + 253, + 223, + 502, + 237 + ], + "score": 1.0, + "content": "Figure 3: The DICTIONARYLOOKUP problem: GATv2 easily", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 254, + 236, + 502, + 247 + ], + "spans": [ + { + "bbox": [ + 254, + 236, + 292, + 247 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + }, + { + "bbox": [ + 292, + 236, + 317, + 246 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 236, + 454, + 247 + ], + "score": 1.0, + "content": "train and test accuracies even for", + "type": "text" + }, + { + "bbox": [ + 455, + 236, + 484, + 246 + ], + "score": 0.88, + "content": "k { = } 1 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 236, + 502, + 247 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 254, + 246, + 356, + 258 + ], + "spans": [ + { + "bbox": [ + 254, + 246, + 356, + 258 + ], + "score": 1.0, + "content": "using only a single head.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + } + ], + "index": 18.75 + }, + { + "type": "text", + "bbox": [ + 107, + 275, + 505, + 308 + ], + "lines": [ + { + "bbox": [ + 105, + 274, + 506, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 434, + 289 + ], + "score": 1.0, + "content": "all GNNs. In all experiments of GATv2, we constrain the learned matrix by setting", + "type": "text" + }, + { + "bbox": [ + 435, + 275, + 502, + 288 + ], + "score": 0.92, + "content": "W = [ W ^ { \\prime } | | W ^ { \\prime } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 274, + 506, + 289 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "to rule out the increased number of parameters over GAT as the source of empirical difference (see", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 298, + 429, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 429, + 309 + ], + "score": 1.0, + "content": "Appendix I.2). Training details, statistics, and code are provided in Appendix D.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 314, + 505, + 381 + ], + "lines": [ + { + "bbox": [ + 106, + 315, + 507, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 507, + 327 + ], + "score": 1.0, + "content": "Our main goal is to compare dynamic and static graph attention mechanisms. However, for reference,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "we also include non-attentive baselines such as GCN (Kipf and Welling, 2017), GIN (Xu et al., 2019)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "and GraphSAGE (Hamilton et al., 2017). These non-attentive GNNs can be thought of as a special", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 347, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 506, + 360 + ], + "score": 1.0, + "content": "case of attention, where every node gives all its neighbors the same attention score. Additional", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 359, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 505, + 370 + ], + "score": 1.0, + "content": "comparison to a Transformer-style scaled dot-product attention (“DPGAT”), which is strictly weaker", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 369, + 439, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 439, + 382 + ], + "score": 1.0, + "content": "than our proposed GATv2 (see a proof in Appendix G.1), is shown in Appendix G.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5 + }, + { + "type": "title", + "bbox": [ + 106, + 399, + 339, + 411 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 340, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 340, + 412 + ], + "score": 1.0, + "content": "4.1 SYNTHETIC BENCHMARK: DICTIONARYLOOKUP", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 422, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 435 + ], + "score": 1.0, + "content": "The DICTIONARYLOOKUP problem is a contrived problem that we designed to test the ability of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 433, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 433, + 505, + 446 + ], + "score": 1.0, + "content": "a GNN architecture to perform dynamic attention. Here, we demonstrate that GAT cannot learn", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 380, + 457 + ], + "score": 1.0, + "content": "this simple problem. Figure 2 shows a complete bipartite graph of", + "type": "text" + }, + { + "bbox": [ + 380, + 445, + 392, + 455 + ], + "score": 0.85, + "content": "2 k", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "nodes. Each “key node” in", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 455, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 241, + 469 + ], + "score": 1.0, + "content": "the bottom row has an attribute", + "type": "text" + }, + { + "bbox": [ + 241, + 455, + 300, + 467 + ], + "score": 0.9, + "content": "( \\{ \\mathrm { A } , \\mathrm { B } , \\mathrm { C } , \\bar { \\ldots } \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 455, + 354, + 469 + ], + "score": 1.0, + "content": "and a value", + "type": "text" + }, + { + "bbox": [ + 355, + 456, + 407, + 467 + ], + "score": 0.88, + "content": "( \\{ 1 , 2 , 3 , \\ldots \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 455, + 506, + 469 + ], + "score": 1.0, + "content": ". Each “query node” in", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 465, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 104, + 465, + 252, + 481 + ], + "score": 1.0, + "content": "the upper row has only an attribute", + "type": "text" + }, + { + "bbox": [ + 252, + 467, + 311, + 479 + ], + "score": 0.83, + "content": "( \\{ \\mathrm { A } , \\mathrm { B } , \\mathrm { C } , \\ldots \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 465, + 505, + 481 + ], + "score": 1.0, + "content": ". The goal is to predict the value of every query", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 477, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 506, + 491 + ], + "score": 1.0, + "content": "node (upper row), according to its attribute. Each graph in the dataset has a different mapping from", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 342, + 501 + ], + "score": 1.0, + "content": "attributes to values. We created a separate dataset for each", + "type": "text" + }, + { + "bbox": [ + 342, + 488, + 408, + 501 + ], + "score": 0.93, + "content": "k = \\{ 1 , 2 , 3 , \\ldots \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 489, + 506, + 501 + ], + "score": 1.0, + "content": ", for which we trained a", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 498, + 308, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 308, + 513 + ], + "score": 1.0, + "content": "different model, and measured per-node accuracy.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 107, + 516, + 505, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "Although this is a contrived problem, it is relevant to any subgraph with keys that share more than", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "score": 1.0, + "content": "one query, and each query needs to attend to the keys differently. Such subgraphs are very common", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 537, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 104, + 537, + 506, + 552 + ], + "score": 1.0, + "content": "in a variety of real-world domains. This problem tests the layer itself because it can be solved using a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 549, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 506, + 562 + ], + "score": 1.0, + "content": "single GNN layer, without suffering from multi-layer side-effects such as over-smoothing (Li et al.,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "score": 1.0, + "content": "2018), over-squashing (Alon and Yahav, 2021), or vanishing gradients (Li et al., 2019). Our code", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 571, + 475, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 475, + 584 + ], + "score": 1.0, + "content": "will be made publicly available, to serve as a testbed for future graph attention mechanisms.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 44.5 + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 505, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 442, + 604 + ], + "score": 1.0, + "content": "Results Figure 3 shows the following surprising results: GAT with a single head", + "type": "text" + }, + { + "bbox": [ + 442, + 591, + 475, + 603 + ], + "score": 0.87, + "content": "( \\mathrm { G A T _ { 1 } } _ { h } )", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 591, + 505, + 604 + ], + "score": 1.0, + "content": ") failed", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 602, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 257, + 614 + ], + "score": 1.0, + "content": "to fit the training set for any value of", + "type": "text" + }, + { + "bbox": [ + 258, + 603, + 264, + 612 + ], + "score": 0.78, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 602, + 506, + 614 + ], + "score": 1.0, + "content": ", no matter for how many iterations it was trained, and after", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 613, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 506, + 626 + ], + "score": 1.0, + "content": "trying various training methods. Thus, it expectedly fails to generalize (resulting in low test accuracy).", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 170, + 636 + ], + "score": 1.0, + "content": "Using 8 heads,", + "type": "text" + }, + { + "bbox": [ + 171, + 624, + 201, + 635 + ], + "score": 0.9, + "content": "\\mathrm { G A T } _ { 8 h }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "successfully fits the training set, but generalizes poorly to the test set. In", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 235, + 647 + ], + "score": 1.0, + "content": "contrast, GATv2 easily achieves", + "type": "text" + }, + { + "bbox": [ + 235, + 635, + 258, + 645 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 635, + 309, + 647 + ], + "score": 1.0, + "content": "training and", + "type": "text" + }, + { + "bbox": [ + 309, + 635, + 333, + 645 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 635, + 457, + 647 + ], + "score": 1.0, + "content": "test accuracies for any value of", + "type": "text" + }, + { + "bbox": [ + 458, + 635, + 464, + 645 + ], + "score": 0.78, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 635, + 505, + 647 + ], + "score": 1.0, + "content": ", and even", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 120, + 658 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 121, + 646, + 151, + 657 + ], + "score": 0.88, + "content": "k { = } 1 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 645, + 506, + 658 + ], + "score": 1.0, + "content": "(not shown) and using a single head, thanks to its ability to perform dynamic attention.", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 657, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 669 + ], + "score": 1.0, + "content": "These results clearly show the limitations of GAT, which are easily solved by GATv2. An additional", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 668, + 478, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 478, + 681 + ], + "score": 1.0, + "content": "comparison to GIN, which could not fit this dataset, is provided in Figure 5 in Appendix F.1.", + "type": "text" + } + ], + "index": 55 + } + ], + "index": 51.5 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 507, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 507, + 700 + ], + "score": 1.0, + "content": "Visualization Figure 1a (top) shows a heatmap of GAT’s attention scores in this DICTIONARY-", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 288, + 711 + ], + "score": 1.0, + "content": "LOOKUP problem. As shown, all query nodes", + "type": "text" + }, + { + "bbox": [ + 289, + 700, + 300, + 710 + ], + "score": 0.77, + "content": "q 0", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 699, + 311, + 711 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 311, + 699, + 322, + 710 + ], + "score": 0.85, + "content": "q 9", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 699, + 446, + 711 + ], + "score": 1.0, + "content": "attend mostly to the eighth key", + "type": "text" + }, + { + "bbox": [ + 446, + 699, + 464, + 710 + ], + "score": 0.38, + "content": "( k 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 699, + 505, + 711 + ], + "score": 1.0, + "content": ", and have", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "the same ranking of attention coefficients (Figure 1a (bottom)). In contrast, Figure 1b shows how", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "GATv2 can select a different key node for every query node, because it computes dynamic attention.", + "type": "text" + } + ], + "index": 59 + } + ], + "index": 57.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 130, + 79, + 223, + 153 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 130, + 79, + 223, + 153 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 130, + 79, + 223, + 153 + ], + "spans": [ + { + "bbox": [ + 130, + 79, + 223, + 153 + ], + "score": 0.943, + "type": "image", + "image_path": "79aab347d93a4c3bb0bb39665869459e72b57008bc185548856fefb238200d5b.jpg" + } + ] + } + ], + "index": 0.5, + "virtual_lines": [ + { + "bbox": [ + 130, + 79, + 223, + 116.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 130, + 116.0, + 223, + 153.0 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 161, + 248, + 259 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 160, + 250, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 160, + 250, + 172 + ], + "score": 1.0, + "content": "Figure 2: The DICTIONARY-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 172, + 249, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 210, + 183 + ], + "score": 1.0, + "content": "LOOKUP problem of size", + "type": "text" + }, + { + "bbox": [ + 210, + 172, + 230, + 182 + ], + "score": 0.85, + "content": "k { = } 4", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 172, + 249, + 183 + ], + "score": 1.0, + "content": ": ev-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 183, + 248, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 248, + 194 + ], + "score": 1.0, + "content": "ery node in the bottom row has an", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 194, + 246, + 206 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 187, + 206 + ], + "score": 1.0, + "content": "alphabetic attribute", + "type": "text" + }, + { + "bbox": [ + 188, + 194, + 246, + 205 + ], + "score": 0.86, + "content": "( \\{ { \\mathrm { A } } , { \\mathrm { B } } , { \\mathrm { C } } , \\ldots \\} )", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 205, + 248, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 190, + 217 + ], + "score": 1.0, + "content": "and a numeric value", + "type": "text" + }, + { + "bbox": [ + 190, + 206, + 244, + 217 + ], + "score": 0.8, + "content": "( \\{ 1 , 2 , 3 , \\ldots \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 205, + 248, + 217 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 216, + 248, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 216, + 248, + 227 + ], + "score": 1.0, + "content": "every node in the upper row has", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 227, + 249, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 249, + 239 + ], + "score": 1.0, + "content": "only an attribute; the goal is to pre-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 238, + 248, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 248, + 249 + ], + "score": 1.0, + "content": "dict the value for each node in the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 249, + 225, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 225, + 260 + ], + "score": 1.0, + "content": "upper row, using its attribute.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6 + } + ], + "index": 3.25 + }, + { + "type": "image", + "bbox": [ + 258, + 79, + 500, + 213 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 258, + 79, + 500, + 213 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 258, + 79, + 500, + 213 + ], + "spans": [ + { + "bbox": [ + 258, + 79, + 500, + 213 + ], + "score": 0.967, + "type": "image", + "image_path": "eee3bdfe02443e30b1eb1a486a618dd6fcf3d8c448dc16f59b4318aa5c305de0.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 258, + 79, + 500, + 92.4 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 258, + 92.4, + 500, + 105.80000000000001 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 258, + 105.80000000000001, + 500, + 119.20000000000002 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 258, + 119.20000000000002, + 500, + 132.60000000000002 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 258, + 132.60000000000002, + 500, + 146.00000000000003 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 258, + 146.00000000000003, + 500, + 159.40000000000003 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 258, + 159.40000000000003, + 500, + 172.80000000000004 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 258, + 172.80000000000004, + 500, + 186.20000000000005 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 258, + 186.20000000000005, + 500, + 199.60000000000005 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 258, + 199.60000000000005, + 500, + 213.00000000000006 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 254, + 224, + 502, + 258 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 253, + 223, + 502, + 237 + ], + "spans": [ + { + "bbox": [ + 253, + 223, + 502, + 237 + ], + "score": 1.0, + "content": "Figure 3: The DICTIONARYLOOKUP problem: GATv2 easily", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 254, + 236, + 502, + 247 + ], + "spans": [ + { + "bbox": [ + 254, + 236, + 292, + 247 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + }, + { + "bbox": [ + 292, + 236, + 317, + 246 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 236, + 454, + 247 + ], + "score": 1.0, + "content": "train and test accuracies even for", + "type": "text" + }, + { + "bbox": [ + 455, + 236, + 484, + 246 + ], + "score": 0.88, + "content": "k { = } 1 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 236, + 502, + 247 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 254, + 246, + 356, + 258 + ], + "spans": [ + { + "bbox": [ + 254, + 246, + 356, + 258 + ], + "score": 1.0, + "content": "using only a single head.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + } + ], + "index": 18.75 + }, + { + "type": "text", + "bbox": [ + 107, + 275, + 505, + 308 + ], + "lines": [], + "index": 25, + "bbox_fs": [ + 105, + 274, + 506, + 309 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 314, + 505, + 381 + ], + "lines": [ + { + "bbox": [ + 106, + 315, + 507, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 507, + 327 + ], + "score": 1.0, + "content": "Our main goal is to compare dynamic and static graph attention mechanisms. However, for reference,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "we also include non-attentive baselines such as GCN (Kipf and Welling, 2017), GIN (Xu et al., 2019)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "and GraphSAGE (Hamilton et al., 2017). These non-attentive GNNs can be thought of as a special", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 347, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 506, + 360 + ], + "score": 1.0, + "content": "case of attention, where every node gives all its neighbors the same attention score. Additional", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 359, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 505, + 370 + ], + "score": 1.0, + "content": "comparison to a Transformer-style scaled dot-product attention (“DPGAT”), which is strictly weaker", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 369, + 439, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 439, + 382 + ], + "score": 1.0, + "content": "than our proposed GATv2 (see a proof in Appendix G.1), is shown in Appendix G.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 315, + 507, + 382 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 399, + 339, + 411 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 340, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 340, + 412 + ], + "score": 1.0, + "content": "4.1 SYNTHETIC BENCHMARK: DICTIONARYLOOKUP", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 422, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 435 + ], + "score": 1.0, + "content": "The DICTIONARYLOOKUP problem is a contrived problem that we designed to test the ability of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 433, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 433, + 505, + 446 + ], + "score": 1.0, + "content": "a GNN architecture to perform dynamic attention. Here, we demonstrate that GAT cannot learn", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 380, + 457 + ], + "score": 1.0, + "content": "this simple problem. Figure 2 shows a complete bipartite graph of", + "type": "text" + }, + { + "bbox": [ + 380, + 445, + 392, + 455 + ], + "score": 0.85, + "content": "2 k", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "nodes. Each “key node” in", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 455, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 241, + 469 + ], + "score": 1.0, + "content": "the bottom row has an attribute", + "type": "text" + }, + { + "bbox": [ + 241, + 455, + 300, + 467 + ], + "score": 0.9, + "content": "( \\{ \\mathrm { A } , \\mathrm { B } , \\mathrm { C } , \\bar { \\ldots } \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 455, + 354, + 469 + ], + "score": 1.0, + "content": "and a value", + "type": "text" + }, + { + "bbox": [ + 355, + 456, + 407, + 467 + ], + "score": 0.88, + "content": "( \\{ 1 , 2 , 3 , \\ldots \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 455, + 506, + 469 + ], + "score": 1.0, + "content": ". Each “query node” in", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 465, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 104, + 465, + 252, + 481 + ], + "score": 1.0, + "content": "the upper row has only an attribute", + "type": "text" + }, + { + "bbox": [ + 252, + 467, + 311, + 479 + ], + "score": 0.83, + "content": "( \\{ \\mathrm { A } , \\mathrm { B } , \\mathrm { C } , \\ldots \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 465, + 505, + 481 + ], + "score": 1.0, + "content": ". The goal is to predict the value of every query", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 477, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 506, + 491 + ], + "score": 1.0, + "content": "node (upper row), according to its attribute. Each graph in the dataset has a different mapping from", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 342, + 501 + ], + "score": 1.0, + "content": "attributes to values. We created a separate dataset for each", + "type": "text" + }, + { + "bbox": [ + 342, + 488, + 408, + 501 + ], + "score": 0.93, + "content": "k = \\{ 1 , 2 , 3 , \\ldots \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 489, + 506, + 501 + ], + "score": 1.0, + "content": ", for which we trained a", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 498, + 308, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 308, + 513 + ], + "score": 1.0, + "content": "different model, and measured per-node accuracy.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37.5, + "bbox_fs": [ + 104, + 423, + 506, + 513 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 516, + 505, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "Although this is a contrived problem, it is relevant to any subgraph with keys that share more than", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "score": 1.0, + "content": "one query, and each query needs to attend to the keys differently. Such subgraphs are very common", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 537, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 104, + 537, + 506, + 552 + ], + "score": 1.0, + "content": "in a variety of real-world domains. This problem tests the layer itself because it can be solved using a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 549, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 506, + 562 + ], + "score": 1.0, + "content": "single GNN layer, without suffering from multi-layer side-effects such as over-smoothing (Li et al.,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "score": 1.0, + "content": "2018), over-squashing (Alon and Yahav, 2021), or vanishing gradients (Li et al., 2019). Our code", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 571, + 475, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 475, + 584 + ], + "score": 1.0, + "content": "will be made publicly available, to serve as a testbed for future graph attention mechanisms.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 44.5, + "bbox_fs": [ + 104, + 516, + 506, + 584 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 505, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 442, + 604 + ], + "score": 1.0, + "content": "Results Figure 3 shows the following surprising results: GAT with a single head", + "type": "text" + }, + { + "bbox": [ + 442, + 591, + 475, + 603 + ], + "score": 0.87, + "content": "( \\mathrm { G A T _ { 1 } } _ { h } )", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 591, + 505, + 604 + ], + "score": 1.0, + "content": ") failed", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 602, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 257, + 614 + ], + "score": 1.0, + "content": "to fit the training set for any value of", + "type": "text" + }, + { + "bbox": [ + 258, + 603, + 264, + 612 + ], + "score": 0.78, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 602, + 506, + 614 + ], + "score": 1.0, + "content": ", no matter for how many iterations it was trained, and after", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 613, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 506, + 626 + ], + "score": 1.0, + "content": "trying various training methods. Thus, it expectedly fails to generalize (resulting in low test accuracy).", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 170, + 636 + ], + "score": 1.0, + "content": "Using 8 heads,", + "type": "text" + }, + { + "bbox": [ + 171, + 624, + 201, + 635 + ], + "score": 0.9, + "content": "\\mathrm { G A T } _ { 8 h }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "successfully fits the training set, but generalizes poorly to the test set. In", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 235, + 647 + ], + "score": 1.0, + "content": "contrast, GATv2 easily achieves", + "type": "text" + }, + { + "bbox": [ + 235, + 635, + 258, + 645 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 635, + 309, + 647 + ], + "score": 1.0, + "content": "training and", + "type": "text" + }, + { + "bbox": [ + 309, + 635, + 333, + 645 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 635, + 457, + 647 + ], + "score": 1.0, + "content": "test accuracies for any value of", + "type": "text" + }, + { + "bbox": [ + 458, + 635, + 464, + 645 + ], + "score": 0.78, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 635, + 505, + 647 + ], + "score": 1.0, + "content": ", and even", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 120, + 658 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 121, + 646, + 151, + 657 + ], + "score": 0.88, + "content": "k { = } 1 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 645, + 506, + 658 + ], + "score": 1.0, + "content": "(not shown) and using a single head, thanks to its ability to perform dynamic attention.", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 657, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 669 + ], + "score": 1.0, + "content": "These results clearly show the limitations of GAT, which are easily solved by GATv2. An additional", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 668, + 478, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 478, + 681 + ], + "score": 1.0, + "content": "comparison to GIN, which could not fit this dataset, is provided in Figure 5 in Appendix F.1.", + "type": "text" + } + ], + "index": 55 + } + ], + "index": 51.5, + "bbox_fs": [ + 105, + 591, + 506, + 681 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 507, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 507, + 700 + ], + "score": 1.0, + "content": "Visualization Figure 1a (top) shows a heatmap of GAT’s attention scores in this DICTIONARY-", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 288, + 711 + ], + "score": 1.0, + "content": "LOOKUP problem. As shown, all query nodes", + "type": "text" + }, + { + "bbox": [ + 289, + 700, + 300, + 710 + ], + "score": 0.77, + "content": "q 0", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 699, + 311, + 711 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 311, + 699, + 322, + 710 + ], + "score": 0.85, + "content": "q 9", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 699, + 446, + 711 + ], + "score": 1.0, + "content": "attend mostly to the eighth key", + "type": "text" + }, + { + "bbox": [ + 446, + 699, + 464, + 710 + ], + "score": 0.38, + "content": "( k 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 699, + 505, + 711 + ], + "score": 1.0, + "content": ", and have", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "the same ranking of attention coefficients (Figure 1a (bottom)). In contrast, Figure 1b shows how", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "GATv2 can select a different key node for every query node, because it computes dynamic attention.", + "type": "text" + } + ], + "index": 59 + } + ], + "index": 57.5, + "bbox_fs": [ + 105, + 687, + 507, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 94 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 94 + ], + "score": 1.0, + "content": "The role of multi-head attention Velickovi ˇ c et al. (2018) found the role of multi-head attention to ´", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "be stabilizing the learning process. Nevertheless, Figure 3 shows that increasing the number of heads", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "strictly increases training accuracy, and thus, the expressivity. Thus, GAT depends on having multiple", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 495, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 495, + 127 + ], + "score": 1.0, + "content": "attention heads. In contrast, even a single GATv2 head generalizes better than a multi-head GAT.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "title", + "bbox": [ + 108, + 143, + 232, + 154 + ], + "lines": [ + { + "bbox": [ + 105, + 142, + 234, + 155 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 234, + 155 + ], + "score": 1.0, + "content": "4.2 ROBUSTNESS TO NOISE", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 164, + 505, + 198 + ], + "lines": [ + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "We examine the robustness of dynamic and static attention to noise. In particular, we focus on", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 175, + 261, + 187 + ], + "score": 1.0, + "content": "structural noise: given an input graph", + "type": "text" + }, + { + "bbox": [ + 261, + 175, + 305, + 187 + ], + "score": 0.93, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 175, + 375, + 187 + ], + "score": 1.0, + "content": "and a noise ratio", + "type": "text" + }, + { + "bbox": [ + 376, + 176, + 414, + 187 + ], + "score": 0.92, + "content": "0 \\leq p \\leq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 175, + 505, + 187 + ], + "score": 1.0, + "content": ", we randomly sample", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 107, + 185, + 506, + 199 + ], + "spans": [ + { + "bbox": [ + 107, + 186, + 132, + 198 + ], + "score": 0.91, + "content": "| { \\mathcal { E } } | \\times p", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 185, + 208, + 199 + ], + "score": 1.0, + "content": "non-existing edges", + "type": "text" + }, + { + "bbox": [ + 208, + 186, + 218, + 196 + ], + "score": 0.86, + "content": "\\mathcal { E } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 185, + 240, + 199 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 240, + 187, + 274, + 198 + ], + "score": 0.92, + "content": "\\mathcal { V } \\times \\mathcal { V } \\backslash \\mathcal { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 185, + 442, + 199 + ], + "score": 1.0, + "content": ". We then train the GNN on the noisy graph", + "type": "text" + }, + { + "bbox": [ + 443, + 186, + 502, + 198 + ], + "score": 0.93, + "content": "\\mathcal { G } ^ { \\prime } { = } ( \\dot { \\mathcal { V } } , \\mathcal { E } \\cup \\bar { \\mathcal { E } } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 185, + 506, + 199 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "image", + "bbox": [ + 108, + 212, + 496, + 386 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 212, + 496, + 386 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 212, + 496, + 386 + ], + "spans": [ + { + "bbox": [ + 108, + 212, + 496, + 386 + ], + "score": 0.971, + "type": "image", + "image_path": "77bbac2b2bb1e1b15f9fdfb4dc28a2c29279f545cc08fbe57d963cffd3d7f569.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 108, + 212, + 496, + 270.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 108, + 270.0, + 496, + 328.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 108, + 328.0, + 496, + 386.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 394, + 505, + 417 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 395, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 406 + ], + "score": 1.0, + "content": "Figure 4: Test accuracy compared to the noise ratio: GATv2 is more robust to structural noise", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 405, + 469, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 469, + 419 + ], + "score": 1.0, + "content": "compared to GAT. Each point is an average of 10 runs, error bars show standard deviation.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + } + ], + "index": 10.25 + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 106, + 435, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 506, + 447 + ], + "score": 1.0, + "content": "Results Figure 8 shows the accuracy on two node-prediction datasets from the Open Graph Bench-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 446, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 359, + 457 + ], + "score": 1.0, + "content": "mark (OGB; Hu et al., 2020) as a function of the noise ratio", + "type": "text" + }, + { + "bbox": [ + 360, + 447, + 366, + 457 + ], + "score": 0.74, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 446, + 385, + 457 + ], + "score": 1.0, + "content": ". As", + "type": "text" + }, + { + "bbox": [ + 385, + 447, + 392, + 457 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 446, + 505, + 457 + ], + "score": 1.0, + "content": "increases, all models show", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "a natural decline in test accuracy in both datasets. Yet, thanks to their ability to compute dynamic", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "attention, GATv2 shows a milder degradation in accuracy compared to GAT, which shows a steeper", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 477, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 492 + ], + "score": 1.0, + "content": "descent. We hypothesize that the ability to perform dynamic attention helps the models distinguishing", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 489, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 214, + 502 + ], + "score": 1.0, + "content": "between given data edges", + "type": "text" + }, + { + "bbox": [ + 215, + 489, + 228, + 500 + ], + "score": 0.69, + "content": "( \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 489, + 298, + 502 + ], + "score": 1.0, + "content": "and noise edges", + "type": "text" + }, + { + "bbox": [ + 298, + 489, + 315, + 501 + ], + "score": 0.83, + "content": "( \\mathcal { E } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "; in contrast, GAT cannot distinguish between", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "edges, because it scores the source and target nodes separately. These results clearly demonstrate the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 512, + 502, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 502, + 524 + ], + "score": 1.0, + "content": "robustness of dynamic attention over static attention in noisy settings, which are common in reality.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 107, + 538, + 239, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 537, + 241, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 241, + 551 + ], + "score": 1.0, + "content": "4.3 PROGRAMS: VARMISUSE", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 504, + 585 + ], + "lines": [ + { + "bbox": [ + 106, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "Setup VARMISUSE (Allamanis et al., 2018) is an inductive node-pointing problem that depends on", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 574, + 460, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 460, + 587 + ], + "score": 1.0, + "content": "11 types of syntactic and semantic interactions between elements in computer programs.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 312, + 646 + ], + "lines": [ + { + "bbox": [ + 106, + 590, + 314, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 314, + 604 + ], + "score": 1.0, + "content": "We used the framework of Brockschmidt (2020),", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 602, + 313, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 313, + 615 + ], + "score": 1.0, + "content": "who performed an extensive hyperparameter tuning", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 613, + 313, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 313, + 625 + ], + "score": 1.0, + "content": "by searching over 30 configurations for every GNN", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 623, + 313, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 313, + 636 + ], + "score": 1.0, + "content": "type. We took their best GAT hyperparameters and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 632, + 308, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 308, + 649 + ], + "score": 1.0, + "content": "used them to train GATv2, without further tuning.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 108, + 655, + 313, + 728 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 315, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 315, + 667 + ], + "score": 1.0, + "content": "Results As shown in Table 1, GATv2 is more accu-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 666, + 313, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 313, + 677 + ], + "score": 1.0, + "content": "rate than GAT and other GNNs in the SeenProj test", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 677, + 313, + 688 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 313, + 688 + ], + "score": 1.0, + "content": "sets. Furthermore, GATv2 achieves an even higher", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 688, + 314, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 314, + 699 + ], + "score": 1.0, + "content": "improvement in the UnseenProj test set. Overall,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 699, + 313, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 313, + 711 + ], + "score": 1.0, + "content": "these results demonstrate the power of GATv2 in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 710, + 313, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 313, + 722 + ], + "score": 1.0, + "content": "modeling complex relational problems, especially", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 321, + 591, + 505, + 636 + ], + "lines": [ + { + "bbox": [ + 319, + 591, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 319, + 591, + 420, + 603 + ], + "score": 1.0, + "content": "Table 1: Accuracy (5 runs", + "type": "text" + }, + { + "bbox": [ + 421, + 592, + 429, + 601 + ], + "score": 0.35, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 591, + 506, + 603 + ], + "score": 1.0, + "content": "stdev) on VARMIS-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 320, + 602, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 320, + 602, + 505, + 613 + ], + "score": 1.0, + "content": "USE. GATv2 is more accurate than all GNNs", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 320, + 613, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 320, + 613, + 506, + 625 + ], + "score": 1.0, + "content": "in both test sets, using GAT’s hyperparameters.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 320, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 320, + 624, + 326, + 636 + ], + "score": 0.25, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 625, + 505, + 636 + ], + "score": 1.0, + "content": "previously reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "table", + "bbox": [ + 324, + 645, + 496, + 717 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 324, + 645, + 496, + 717 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 324, + 645, + 496, + 717 + ], + "spans": [ + { + "bbox": [ + 324, + 645, + 496, + 717 + ], + "score": 0.978, + "html": "
ModelSeenProjUnseenProj
No-GCN+87.2±1.581.4±2.3
AttentionGIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
", + "type": "table", + "image_path": "788a3fe581e9afc6eb6c6f6d21f417466d0cf87ea670d21ae7fcc0b84f5ff63a.jpg" + } + ] + } + ], + "index": 41.5, + "virtual_lines": [ + { + "bbox": [ + 324, + 645, + 496, + 657.0 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 324, + 657.0, + 496, + 669.0 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 324, + 669.0, + 496, + 681.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 324, + 681.0, + 496, + 693.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 324, + 693.0, + 496, + 705.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 324, + 705.0, + 496, + 717.0 + ], + "spans": [], + "index": 44 + } + ] + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 106, + 721, + 427, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 718, + 429, + 735 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 429, + 735 + ], + "score": 1.0, + "content": "since it outperforms extensively tuned models, without any further tuning by us.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 94 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 94 + ], + "score": 1.0, + "content": "The role of multi-head attention Velickovi ˇ c et al. (2018) found the role of multi-head attention to ´", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "be stabilizing the learning process. Nevertheless, Figure 3 shows that increasing the number of heads", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "strictly increases training accuracy, and thus, the expressivity. Thus, GAT depends on having multiple", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 495, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 495, + 127 + ], + "score": 1.0, + "content": "attention heads. In contrast, even a single GATv2 head generalizes better than a multi-head GAT.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 82, + 506, + 127 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 143, + 232, + 154 + ], + "lines": [ + { + "bbox": [ + 105, + 142, + 234, + 155 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 234, + 155 + ], + "score": 1.0, + "content": "4.2 ROBUSTNESS TO NOISE", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 164, + 505, + 198 + ], + "lines": [ + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "We examine the robustness of dynamic and static attention to noise. In particular, we focus on", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 175, + 261, + 187 + ], + "score": 1.0, + "content": "structural noise: given an input graph", + "type": "text" + }, + { + "bbox": [ + 261, + 175, + 305, + 187 + ], + "score": 0.93, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 175, + 375, + 187 + ], + "score": 1.0, + "content": "and a noise ratio", + "type": "text" + }, + { + "bbox": [ + 376, + 176, + 414, + 187 + ], + "score": 0.92, + "content": "0 \\leq p \\leq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 175, + 505, + 187 + ], + "score": 1.0, + "content": ", we randomly sample", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 107, + 185, + 506, + 199 + ], + "spans": [ + { + "bbox": [ + 107, + 186, + 132, + 198 + ], + "score": 0.91, + "content": "| { \\mathcal { E } } | \\times p", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 185, + 208, + 199 + ], + "score": 1.0, + "content": "non-existing edges", + "type": "text" + }, + { + "bbox": [ + 208, + 186, + 218, + 196 + ], + "score": 0.86, + "content": "\\mathcal { E } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 185, + 240, + 199 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 240, + 187, + 274, + 198 + ], + "score": 0.92, + "content": "\\mathcal { V } \\times \\mathcal { V } \\backslash \\mathcal { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 185, + 442, + 199 + ], + "score": 1.0, + "content": ". We then train the GNN on the noisy graph", + "type": "text" + }, + { + "bbox": [ + 443, + 186, + 502, + 198 + ], + "score": 0.93, + "content": "\\mathcal { G } ^ { \\prime } { = } ( \\dot { \\mathcal { V } } , \\mathcal { E } \\cup \\bar { \\mathcal { E } } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 185, + 506, + 199 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 163, + 506, + 199 + ] + }, + { + "type": "image", + "bbox": [ + 108, + 212, + 496, + 386 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 212, + 496, + 386 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 212, + 496, + 386 + ], + "spans": [ + { + "bbox": [ + 108, + 212, + 496, + 386 + ], + "score": 0.971, + "type": "image", + "image_path": "77bbac2b2bb1e1b15f9fdfb4dc28a2c29279f545cc08fbe57d963cffd3d7f569.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 108, + 212, + 496, + 270.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 108, + 270.0, + 496, + 328.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 108, + 328.0, + 496, + 386.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 394, + 505, + 417 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 395, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 406 + ], + "score": 1.0, + "content": "Figure 4: Test accuracy compared to the noise ratio: GATv2 is more robust to structural noise", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 405, + 469, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 469, + 419 + ], + "score": 1.0, + "content": "compared to GAT. Each point is an average of 10 runs, error bars show standard deviation.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + } + ], + "index": 10.25 + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 106, + 435, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 506, + 447 + ], + "score": 1.0, + "content": "Results Figure 8 shows the accuracy on two node-prediction datasets from the Open Graph Bench-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 446, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 359, + 457 + ], + "score": 1.0, + "content": "mark (OGB; Hu et al., 2020) as a function of the noise ratio", + "type": "text" + }, + { + "bbox": [ + 360, + 447, + 366, + 457 + ], + "score": 0.74, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 446, + 385, + 457 + ], + "score": 1.0, + "content": ". As", + "type": "text" + }, + { + "bbox": [ + 385, + 447, + 392, + 457 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 446, + 505, + 457 + ], + "score": 1.0, + "content": "increases, all models show", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "a natural decline in test accuracy in both datasets. Yet, thanks to their ability to compute dynamic", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "attention, GATv2 shows a milder degradation in accuracy compared to GAT, which shows a steeper", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 477, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 492 + ], + "score": 1.0, + "content": "descent. We hypothesize that the ability to perform dynamic attention helps the models distinguishing", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 489, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 214, + 502 + ], + "score": 1.0, + "content": "between given data edges", + "type": "text" + }, + { + "bbox": [ + 215, + 489, + 228, + 500 + ], + "score": 0.69, + "content": "( \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 489, + 298, + 502 + ], + "score": 1.0, + "content": "and noise edges", + "type": "text" + }, + { + "bbox": [ + 298, + 489, + 315, + 501 + ], + "score": 0.83, + "content": "( \\mathcal { E } ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "; in contrast, GAT cannot distinguish between", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "edges, because it scores the source and target nodes separately. These results clearly demonstrate the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 512, + 502, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 502, + 524 + ], + "score": 1.0, + "content": "robustness of dynamic attention over static attention in noisy settings, which are common in reality.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 435, + 506, + 524 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 538, + 239, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 537, + 241, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 241, + 551 + ], + "score": 1.0, + "content": "4.3 PROGRAMS: VARMISUSE", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 504, + 585 + ], + "lines": [ + { + "bbox": [ + 106, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "Setup VARMISUSE (Allamanis et al., 2018) is an inductive node-pointing problem that depends on", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 574, + 460, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 460, + 587 + ], + "score": 1.0, + "content": "11 types of syntactic and semantic interactions between elements in computer programs.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 106, + 563, + 505, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 312, + 646 + ], + "lines": [ + { + "bbox": [ + 106, + 590, + 314, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 314, + 604 + ], + "score": 1.0, + "content": "We used the framework of Brockschmidt (2020),", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 602, + 313, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 313, + 615 + ], + "score": 1.0, + "content": "who performed an extensive hyperparameter tuning", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 613, + 313, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 313, + 625 + ], + "score": 1.0, + "content": "by searching over 30 configurations for every GNN", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 623, + 313, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 313, + 636 + ], + "score": 1.0, + "content": "type. We took their best GAT hyperparameters and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 632, + 308, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 308, + 649 + ], + "score": 1.0, + "content": "used them to train GATv2, without further tuning.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 590, + 314, + 649 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 655, + 313, + 728 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 315, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 315, + 667 + ], + "score": 1.0, + "content": "Results As shown in Table 1, GATv2 is more accu-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 666, + 313, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 313, + 677 + ], + "score": 1.0, + "content": "rate than GAT and other GNNs in the SeenProj test", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 677, + 313, + 688 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 313, + 688 + ], + "score": 1.0, + "content": "sets. Furthermore, GATv2 achieves an even higher", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 688, + 314, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 314, + 699 + ], + "score": 1.0, + "content": "improvement in the UnseenProj test set. Overall,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 699, + 313, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 313, + 711 + ], + "score": 1.0, + "content": "these results demonstrate the power of GATv2 in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 710, + 313, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 313, + 722 + ], + "score": 1.0, + "content": "modeling complex relational problems, especially", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 654, + 315, + 722 + ] + }, + { + "type": "text", + "bbox": [ + 321, + 591, + 505, + 636 + ], + "lines": [ + { + "bbox": [ + 319, + 591, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 319, + 591, + 420, + 603 + ], + "score": 1.0, + "content": "Table 1: Accuracy (5 runs", + "type": "text" + }, + { + "bbox": [ + 421, + 592, + 429, + 601 + ], + "score": 0.35, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 591, + 506, + 603 + ], + "score": 1.0, + "content": "stdev) on VARMIS-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 320, + 602, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 320, + 602, + 505, + 613 + ], + "score": 1.0, + "content": "USE. GATv2 is more accurate than all GNNs", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 320, + 613, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 320, + 613, + 506, + 625 + ], + "score": 1.0, + "content": "in both test sets, using GAT’s hyperparameters.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 320, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 320, + 624, + 326, + 636 + ], + "score": 0.25, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 625, + 505, + 636 + ], + "score": 1.0, + "content": "previously reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5, + "bbox_fs": [ + 319, + 591, + 506, + 636 + ] + }, + { + "type": "table", + "bbox": [ + 324, + 645, + 496, + 717 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 324, + 645, + 496, + 717 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 324, + 645, + 496, + 717 + ], + "spans": [ + { + "bbox": [ + 324, + 645, + 496, + 717 + ], + "score": 0.978, + "html": "
ModelSeenProjUnseenProj
No-GCN+87.2±1.581.4±2.3
AttentionGIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
", + "type": "table", + "image_path": "788a3fe581e9afc6eb6c6f6d21f417466d0cf87ea670d21ae7fcc0b84f5ff63a.jpg" + } + ] + } + ], + "index": 41.5, + "virtual_lines": [ + { + "bbox": [ + 324, + 645, + 496, + 657.0 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 324, + 657.0, + 496, + 669.0 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 324, + 669.0, + 496, + 681.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 324, + 681.0, + 496, + 693.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 324, + 693.0, + 496, + 705.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 324, + 705.0, + 496, + 717.0 + ], + "spans": [], + "index": 44 + } + ] + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 106, + 721, + 427, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 718, + 429, + 735 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 429, + 735 + ], + "score": 1.0, + "content": "since it outperforms extensively tuned models, without any further tuning by us.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 718, + 429, + 735 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 215, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 216, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 216, + 95 + ], + "score": 1.0, + "content": "4.4 NODE-PREDICTION", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 492, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 101, + 492, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 101, + 492, + 116 + ], + "score": 1.0, + "content": "We further compare GATv2, GAT, and other GNNs on four node-prediction datasets from OGB.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "table", + "bbox": [ + 110, + 155, + 499, + 268 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 124, + 503, + 147 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "score": 1.0, + "content": "Table 2: Average accuracy (Table 2a) and ROC-AUC (Table 2b) in node-prediction datasets (10", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 135, + 493, + 148 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 493, + 148 + ], + "score": 1.0, + "content": "runs±std). In all datasets, GATv2 outperforms GAT. † – previously reported by Hu et al. (2020).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "table_body", + "bbox": [ + 110, + 155, + 499, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 155, + 499, + 268 + ], + "spans": [ + { + "bbox": [ + 110, + 155, + 499, + 268 + ], + "score": 0.638, + "html": "
(a)(b)
ModelAttn. Headsogbn-arxivogbn-productsogbn-magogbn-proteins
GCN+071.74±0.2978.97±0.3330.43±0.2572.51±0.35
GraphSAGE+071.49±0.2778.70±0.3631.53±0.1577.68±0.20
GAT171.59±0.3879.04±1.5432.20±1.4670.77±5.79
871.54±0.3077.23±2.3731.75±1.6078.63±1.62
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.4477.23±3.32
871.87±0.2578.46±2.4532.52±0.3979.52±0.55
", + "type": "table", + "image_path": "1ecb583dd83a8d1bd3d786a8ff4fc02573dabb8531d0aab9c0194499d5f90f53.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 110, + 155, + 499, + 192.66666666666666 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 110, + 192.66666666666666, + 499, + 230.33333333333331 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 110, + 230.33333333333331, + 499, + 268.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 107, + 281, + 506, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "Results Results are shown in Table 2. In all settings and all datasets, GATv2 is more accurate than", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "score": 1.0, + "content": "GAT and the non-attentive GNNs. Interestingly, in the datasets of Table 2a, even a single head of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 135, + 315 + ], + "score": 0.28, + "content": "G A T \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 304, + 505, + 316 + ], + "score": 1.0, + "content": "outperforms GAT with 8 heads. In Table 2b (ogbn-proteins), increasing the number of heads", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "results in a major improvement for GAT (from 70.77 to 78.63), while GATv2 already gets most of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "the benefit using a single attention head. These results demonstrate the superiority of GATv2 over", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 336, + 479, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 479, + 349 + ], + "score": 1.0, + "content": "GAT in node prediction (and even with a single head), thanks to GATv2’s dynamic attention.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9.5 + }, + { + "type": "title", + "bbox": [ + 108, + 361, + 248, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 250, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 250, + 374 + ], + "score": 1.0, + "content": "4.5 GRAPH-PREDICTION: QM9", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 384, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "Setup In the QM9 dataset (Ramakrishnan et al., 2014; Gilmer et al., 2017), each graph is a molecule", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "and the goal is to regress each graph to 13 real-valued quantum chemical properties. We used the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 407, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 506, + 419 + ], + "score": 1.0, + "content": "implementation of Brockschmidt (2020) who performed an extensive hyperparameter search over", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 417, + 465, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 465, + 430 + ], + "score": 1.0, + "content": "500 configurations; we took their best-found configuration of GAT to implement GATv2.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "table", + "bbox": [ + 114, + 481, + 497, + 565 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 438, + 503, + 473 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "Table 3: Average error rates (lower is better), 5 runs for each property, on the QM9 dataset. The best", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "result among GAT and GATv2 is marked in bold; the globally best result among all GNNs is marked", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 460, + 451, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 203, + 473 + ], + "score": 1.0, + "content": "in bold and underline.", + "type": "text" + }, + { + "bbox": [ + 203, + 461, + 209, + 473 + ], + "score": 0.75, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 460, + 451, + 473 + ], + "score": 1.0, + "content": "was previously tuned and reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "table_body", + "bbox": [ + 114, + 481, + 497, + 565 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 114, + 481, + 497, + 565 + ], + "spans": [ + { + "bbox": [ + 114, + 481, + 497, + 565 + ], + "score": 0.98, + "html": "
Predicted Property
Model12345678910111213Rel. to GAT
GCNt3.214.221.451.622.4216.38 17.40 7.828.249.057.003.931.02-1.5%
GINt2.644.671.421.502.2715.63 12.93 5.8818.715.625.383.531.05-2.3%
GATt2.684.651.481.532.3152.39 14.87 7.6116.867.646.544.111.48+0%
GATv22.654.281.411.472.2916.37 14.03 6.076.286.605.973.571.59-11.5%
", + "type": "table", + "image_path": "00723c5a064bad5b41815a1955c05911652d470749e6c04360aa9dcf150d9e6d.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 114, + 481, + 497, + 509.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 114, + 509.0, + 497, + 537.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 114, + 537.0, + 497, + 565.0 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 107, + 578, + 505, + 623 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "Results Table 3 shows the main results: GATv2 achieves a lower (better) average error than GAT, by", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 589, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 132, + 600 + ], + "score": 0.82, + "content": "1 1 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 589, + 506, + 603 + ], + "score": 1.0, + "content": "relatively. GAT achieves the overall highest average error. In some properties, the non-attentive", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "score": 1.0, + "content": "GNNs, GCN and GIN, perform best. We hypothesize that attention is not needed in modeling these", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "score": 1.0, + "content": "properties. Generally, GATv2 achieves the lowest overall average relative error (rightmost column).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "title", + "bbox": [ + 108, + 636, + 212, + 647 + ], + "lines": [ + { + "bbox": [ + 105, + 635, + 212, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 212, + 649 + ], + "score": 1.0, + "content": "4.6 LINK-PREDICTION", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 436, + 668 + ], + "lines": [ + { + "bbox": [ + 105, + 656, + 438, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 438, + 669 + ], + "score": 1.0, + "content": "We compare GATv2, GAT, and other GNNs in link-prediction datasets from OGB.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Results Table 8 shows that in all datasets, GATv2 achieves a higher MRR than GAT, which achieves", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 688, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 506, + 700 + ], + "score": 1.0, + "content": "the lowest MRR. However, the non-attentive GraphSAGE performs better than all attentive GNNs.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "We hypothesize that attention might not be needed in these datasets. Another possibility is that", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "dynamic attention is especially useful in graphs that have high node degrees: in ogbn-products and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "ogbn-proteins (Table 2) the average node degrees are 50.5 and 597, respectively (see Table 5 in", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 215, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 216, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 216, + 95 + ], + "score": 1.0, + "content": "4.4 NODE-PREDICTION", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 492, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 101, + 492, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 101, + 492, + 116 + ], + "score": 1.0, + "content": "We further compare GATv2, GAT, and other GNNs on four node-prediction datasets from OGB.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 101, + 492, + 116 + ] + }, + { + "type": "table", + "bbox": [ + 110, + 155, + 499, + 268 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 124, + 503, + 147 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "score": 1.0, + "content": "Table 2: Average accuracy (Table 2a) and ROC-AUC (Table 2b) in node-prediction datasets (10", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 135, + 493, + 148 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 493, + 148 + ], + "score": 1.0, + "content": "runs±std). In all datasets, GATv2 outperforms GAT. † – previously reported by Hu et al. (2020).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "table_body", + "bbox": [ + 110, + 155, + 499, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 155, + 499, + 268 + ], + "spans": [ + { + "bbox": [ + 110, + 155, + 499, + 268 + ], + "score": 0.638, + "html": "
(a)(b)
ModelAttn. Headsogbn-arxivogbn-productsogbn-magogbn-proteins
GCN+071.74±0.2978.97±0.3330.43±0.2572.51±0.35
GraphSAGE+071.49±0.2778.70±0.3631.53±0.1577.68±0.20
GAT171.59±0.3879.04±1.5432.20±1.4670.77±5.79
871.54±0.3077.23±2.3731.75±1.6078.63±1.62
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.4477.23±3.32
871.87±0.2578.46±2.4532.52±0.3979.52±0.55
", + "type": "table", + "image_path": "1ecb583dd83a8d1bd3d786a8ff4fc02573dabb8531d0aab9c0194499d5f90f53.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 110, + 155, + 499, + 192.66666666666666 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 110, + 192.66666666666666, + 499, + 230.33333333333331 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 110, + 230.33333333333331, + 499, + 268.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 107, + 281, + 506, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "Results Results are shown in Table 2. In all settings and all datasets, GATv2 is more accurate than", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "score": 1.0, + "content": "GAT and the non-attentive GNNs. Interestingly, in the datasets of Table 2a, even a single head of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 135, + 315 + ], + "score": 0.28, + "content": "G A T \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 304, + 505, + 316 + ], + "score": 1.0, + "content": "outperforms GAT with 8 heads. In Table 2b (ogbn-proteins), increasing the number of heads", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "results in a major improvement for GAT (from 70.77 to 78.63), while GATv2 already gets most of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "the benefit using a single attention head. These results demonstrate the superiority of GATv2 over", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 336, + 479, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 479, + 349 + ], + "score": 1.0, + "content": "GAT in node prediction (and even with a single head), thanks to GATv2’s dynamic attention.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 281, + 506, + 349 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 361, + 248, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 250, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 250, + 374 + ], + "score": 1.0, + "content": "4.5 GRAPH-PREDICTION: QM9", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 384, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "Setup In the QM9 dataset (Ramakrishnan et al., 2014; Gilmer et al., 2017), each graph is a molecule", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "and the goal is to regress each graph to 13 real-valued quantum chemical properties. We used the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 407, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 506, + 419 + ], + "score": 1.0, + "content": "implementation of Brockschmidt (2020) who performed an extensive hyperparameter search over", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 417, + 465, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 465, + 430 + ], + "score": 1.0, + "content": "500 configurations; we took their best-found configuration of GAT to implement GATv2.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 384, + 506, + 430 + ] + }, + { + "type": "table", + "bbox": [ + 114, + 481, + 497, + 565 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 438, + 503, + 473 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "Table 3: Average error rates (lower is better), 5 runs for each property, on the QM9 dataset. The best", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "result among GAT and GATv2 is marked in bold; the globally best result among all GNNs is marked", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 460, + 451, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 203, + 473 + ], + "score": 1.0, + "content": "in bold and underline.", + "type": "text" + }, + { + "bbox": [ + 203, + 461, + 209, + 473 + ], + "score": 0.75, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 460, + 451, + 473 + ], + "score": 1.0, + "content": "was previously tuned and reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "table_body", + "bbox": [ + 114, + 481, + 497, + 565 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 114, + 481, + 497, + 565 + ], + "spans": [ + { + "bbox": [ + 114, + 481, + 497, + 565 + ], + "score": 0.98, + "html": "
Predicted Property
Model12345678910111213Rel. to GAT
GCNt3.214.221.451.622.4216.38 17.40 7.828.249.057.003.931.02-1.5%
GINt2.644.671.421.502.2715.63 12.93 5.8818.715.625.383.531.05-2.3%
GATt2.684.651.481.532.3152.39 14.87 7.6116.867.646.544.111.48+0%
GATv22.654.281.411.472.2916.37 14.03 6.076.286.605.973.571.59-11.5%
", + "type": "table", + "image_path": "00723c5a064bad5b41815a1955c05911652d470749e6c04360aa9dcf150d9e6d.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 114, + 481, + 497, + 509.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 114, + 509.0, + 497, + 537.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 114, + 537.0, + 497, + 565.0 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 107, + 578, + 505, + 623 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "Results Table 3 shows the main results: GATv2 achieves a lower (better) average error than GAT, by", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 589, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 132, + 600 + ], + "score": 0.82, + "content": "1 1 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 589, + 506, + 603 + ], + "score": 1.0, + "content": "relatively. GAT achieves the overall highest average error. In some properties, the non-attentive", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "score": 1.0, + "content": "GNNs, GCN and GIN, perform best. We hypothesize that attention is not needed in modeling these", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "score": 1.0, + "content": "properties. Generally, GATv2 achieves the lowest overall average relative error (rightmost column).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 578, + 506, + 624 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 636, + 212, + 647 + ], + "lines": [ + { + "bbox": [ + 105, + 635, + 212, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 212, + 649 + ], + "score": 1.0, + "content": "4.6 LINK-PREDICTION", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 436, + 668 + ], + "lines": [ + { + "bbox": [ + 105, + 656, + 438, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 438, + 669 + ], + "score": 1.0, + "content": "We compare GATv2, GAT, and other GNNs in link-prediction datasets from OGB.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 656, + 438, + 669 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Results Table 8 shows that in all datasets, GATv2 achieves a higher MRR than GAT, which achieves", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 688, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 506, + 700 + ], + "score": 1.0, + "content": "the lowest MRR. However, the non-attentive GraphSAGE performs better than all attentive GNNs.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "We hypothesize that attention might not be needed in these datasets. Another possibility is that", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "dynamic attention is especially useful in graphs that have high node degrees: in ogbn-products and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "ogbn-proteins (Table 2) the average node degrees are 50.5 and 597, respectively (see Table 5 in", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 677, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Appendix E). ogbl-collab and ogbl-citation2 (Table 8), however, have much lower average node", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 105 + ], + "score": 1.0, + "content": "degrees – of 8.2 and 20.7. We hypothesize that a dynamic attention mechanism is especially useful to", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "select the most relevant neighbors when the total number of neighbors is high. We leave the study of", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 500, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 500, + 128 + ], + "score": 1.0, + "content": "the effect of the datasets’s average node degrees on the optimal GNN architecture for future work.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "title", + "bbox": [ + 107, + 146, + 185, + 157 + ], + "lines": [ + { + "bbox": [ + 105, + 144, + 187, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 187, + 159 + ], + "score": 1.0, + "content": "4.7 DISCUSSION", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 169, + 505, + 202 + ], + "lines": [ + { + "bbox": [ + 105, + 169, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 285, + 181 + ], + "score": 1.0, + "content": "In all examined benchmarks, we found that", + "type": "text" + }, + { + "bbox": [ + 285, + 169, + 315, + 180 + ], + "score": 0.26, + "content": "G A T \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 169, + 505, + 181 + ], + "score": 1.0, + "content": "is more accurate than GAT. Further, we found", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 181, + 504, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 504, + 191 + ], + "score": 1.0, + "content": "that GATv2 is significantly more robust to noise than GAT. In the synthetic DICTIONARYLOOKUP", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 190, + 507, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 507, + 205 + ], + "score": 1.0, + "content": "benchmark (Section 4.1), GAT fails to express the data, and thus achieves even poor training accuracy.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 208, + 505, + 241 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 505, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 220 + ], + "score": 1.0, + "content": "In few of the benchmarks (Table 8 and some of the properties in Table 3) – a non-attentive model", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "such as GCN or GIN achieved a higher accuracy than all GNNs that do use attention. We hypothesize", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 230, + 286, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 286, + 241 + ], + "score": 1.0, + "content": "that attention is not needed in these datasets.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 250, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "Which graph attention mechanism should I use? It is usually impossible to determine in advance", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 261, + 506, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 506, + 274 + ], + "score": 1.0, + "content": "which architecture would perform best. A theoretically weaker model may perform better in practice,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 272, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 506, + 284 + ], + "score": 1.0, + "content": "because a stronger model might overfit the training data if the task is “too simple” and does not", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 283, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 505, + 295 + ], + "score": 1.0, + "content": "require such expressiveness. Intuitively, we believe that the more complex the interactions between", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 294, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 306 + ], + "score": 1.0, + "content": "nodes are – the more benefit a GNN can take from theoretically stronger graph attention mechanisms", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "score": 1.0, + "content": "such as GATv2. The main question is whether the problem has a global ranking of “influential” nodes", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 316, + 474, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 474, + 329 + ], + "score": 1.0, + "content": "(GAT is sufficient), or do different nodes have different rankings of neighbors (use GATv2).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 332, + 505, + 410 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 345 + ], + "score": 1.0, + "content": "Velickovi ˇ c, the author of GAT, has confirmed on Twitter ´ 6 that GAT was designed to work in the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 343, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 506, + 356 + ], + "score": 1.0, + "content": "“easy-to-overfit” datasets of the time (2017), such as Cora, Citeseer and Pubmed (Sen et al., 2008),", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 354, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 505, + 367 + ], + "score": 1.0, + "content": "where the data might had an underlying static ranking of “globally important” nodes. Velickovi ˇ c´", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "agreed that newer and more challenging benchmarks may demand stronger attention mechanisms", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "such as GATv2. that depends on the representation of the query node as well. In this paper, we revisit", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 388, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 399 + ], + "score": 1.0, + "content": "the traditional assumptions and show that many modern graph benchmarks and datasets contain more", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 399, + 338, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 338, + 410 + ], + "score": 1.0, + "content": "complex interactions, and thus require dynamic attention.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21 + }, + { + "type": "title", + "bbox": [ + 108, + 431, + 195, + 444 + ], + "lines": [ + { + "bbox": [ + 104, + 430, + 197, + 447 + ], + "spans": [ + { + "bbox": [ + 104, + 430, + 197, + 447 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 460, + 505, + 548 + ], + "lines": [ + { + "bbox": [ + 106, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "In this paper, we identify that the popular and widespread Graph Attention Network does not compute", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 471, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 506, + 483 + ], + "score": 1.0, + "content": "dynamic attention. Instead, the attention mechanism in the standard definition and implementations of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 482, + 507, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 507, + 496 + ], + "score": 1.0, + "content": "GAT is only static: for any query, its neighbor-scoring is monotonic with respect to per-node scores.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "score": 1.0, + "content": "Further, the ranking (the argsort) of attention coefficients is identical for all nodes in the graph,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 505, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 505, + 517 + ], + "score": 1.0, + "content": "and is unconditioned on the query node. As a result, GAT cannot even express simple alignment", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 514, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 104, + 514, + 506, + 529 + ], + "score": 1.0, + "content": "problems. To address this limitation, we introduce a simple fix and propose GATv2: by modifying", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 527, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 505, + 538 + ], + "score": 1.0, + "content": "the order of operations in GAT, GATv2 achieves a universal approximator attention function and is", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 537, + 261, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 261, + 549 + ], + "score": 1.0, + "content": "thus strictly more powerful than GAT.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 554, + 506, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 553, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 568 + ], + "score": 1.0, + "content": "We demonstrate the empirical advantage of GATv2 over GAT in a synthetic problem that requires dy-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "score": 1.0, + "content": "namic selection of nodes, and in 12 benchmarks from OGB and other public datasets. Our experiments", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 575, + 483, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 483, + 590 + ], + "score": 1.0, + "content": "show that GATv2 outperforms GAT in all benchmarks while having the same parametric cost.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 592, + 505, + 659 + ], + "lines": [ + { + "bbox": [ + 105, + 592, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 506, + 606 + ], + "score": 1.0, + "content": "We encourage the community to use GATv2 instead or in addition to GAT whenever comparing new", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 602, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 506, + 618 + ], + "score": 1.0, + "content": "GNN architectures to the common strong baselines. In complex tasks and domains and in challenging", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "datasets, a model that uses GAT as an internal component can replace it with GATv2 to benefit", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 625, + 507, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 507, + 639 + ], + "score": 1.0, + "content": "from a strictly more powerful model. To this end, we make our code publicly available at https:", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "//github.com/tech-srl/how_attentive_are_gats ,7 and GATv2 is available as part", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 646, + 496, + 661 + ], + "spans": [ + { + "bbox": [ + 104, + 646, + 496, + 661 + ], + "score": 1.0, + "content": "of the PyTorch Geometric library,8 the Deep Graph Library,9 and the TensorFlow GNN library.10", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 678, + 499, + 732 + ], + "lines": [ + { + "bbox": [ + 117, + 675, + 428, + 691 + ], + "spans": [ + { + "bbox": [ + 117, + 675, + 428, + 691 + ], + "score": 1.0, + "content": "6https://twitter.com/PetarV_93/status/1399685979506675714", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 686, + 500, + 701 + ], + "spans": [ + { + "bbox": [ + 118, + 686, + 500, + 701 + ], + "score": 1.0, + "content": "7An annotated implementation of GATv2 is available at https://nn.labml.ai/graphs/gatv2/", + "type": "text" + } + ] + }, + { + "bbox": [ + 116, + 695, + 428, + 714 + ], + "spans": [ + { + "bbox": [ + 116, + 695, + 428, + 714 + ], + "score": 1.0, + "content": "8from torch_geometric.nn.conv.gatv2_conv import GATv2Conv", + "type": "text" + } + ] + }, + { + "bbox": [ + 116, + 708, + 320, + 723 + ], + "spans": [ + { + "bbox": [ + 116, + 708, + 320, + 723 + ], + "score": 1.0, + "content": "9from dgl.nn.pytorch import GATv2Conv", + "type": "text" + } + ] + }, + { + "bbox": [ + 114, + 717, + 498, + 736 + ], + "spans": [ + { + "bbox": [ + 114, + 717, + 498, + 736 + ], + "score": 1.0, + "content": "10from tensorflow_gnn.graph.keras.layers.gat_v2 import GATv2Convolution", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Appendix E). ogbl-collab and ogbl-citation2 (Table 8), however, have much lower average node", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 105 + ], + "score": 1.0, + "content": "degrees – of 8.2 and 20.7. We hypothesize that a dynamic attention mechanism is especially useful to", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "select the most relevant neighbors when the total number of neighbors is high. We leave the study of", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 500, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 500, + 128 + ], + "score": 1.0, + "content": "the effect of the datasets’s average node degrees on the optimal GNN architecture for future work.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 82, + 505, + 128 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 146, + 185, + 157 + ], + "lines": [ + { + "bbox": [ + 105, + 144, + 187, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 187, + 159 + ], + "score": 1.0, + "content": "4.7 DISCUSSION", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 169, + 505, + 202 + ], + "lines": [ + { + "bbox": [ + 105, + 169, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 285, + 181 + ], + "score": 1.0, + "content": "In all examined benchmarks, we found that", + "type": "text" + }, + { + "bbox": [ + 285, + 169, + 315, + 180 + ], + "score": 0.26, + "content": "G A T \\nu 2", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 169, + 505, + 181 + ], + "score": 1.0, + "content": "is more accurate than GAT. Further, we found", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 181, + 504, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 504, + 191 + ], + "score": 1.0, + "content": "that GATv2 is significantly more robust to noise than GAT. In the synthetic DICTIONARYLOOKUP", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 190, + 507, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 507, + 205 + ], + "score": 1.0, + "content": "benchmark (Section 4.1), GAT fails to express the data, and thus achieves even poor training accuracy.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 169, + 507, + 205 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 208, + 505, + 241 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 505, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 220 + ], + "score": 1.0, + "content": "In few of the benchmarks (Table 8 and some of the properties in Table 3) – a non-attentive model", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "such as GCN or GIN achieved a higher accuracy than all GNNs that do use attention. We hypothesize", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 230, + 286, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 286, + 241 + ], + "score": 1.0, + "content": "that attention is not needed in these datasets.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 207, + 505, + 241 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 250, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "Which graph attention mechanism should I use? It is usually impossible to determine in advance", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 261, + 506, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 506, + 274 + ], + "score": 1.0, + "content": "which architecture would perform best. A theoretically weaker model may perform better in practice,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 272, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 506, + 284 + ], + "score": 1.0, + "content": "because a stronger model might overfit the training data if the task is “too simple” and does not", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 283, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 505, + 295 + ], + "score": 1.0, + "content": "require such expressiveness. Intuitively, we believe that the more complex the interactions between", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 294, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 306 + ], + "score": 1.0, + "content": "nodes are – the more benefit a GNN can take from theoretically stronger graph attention mechanisms", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "score": 1.0, + "content": "such as GATv2. The main question is whether the problem has a global ranking of “influential” nodes", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 316, + 474, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 474, + 329 + ], + "score": 1.0, + "content": "(GAT is sufficient), or do different nodes have different rankings of neighbors (use GATv2).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 250, + 506, + 329 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 332, + 505, + 410 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 345 + ], + "score": 1.0, + "content": "Velickovi ˇ c, the author of GAT, has confirmed on Twitter ´ 6 that GAT was designed to work in the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 343, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 506, + 356 + ], + "score": 1.0, + "content": "“easy-to-overfit” datasets of the time (2017), such as Cora, Citeseer and Pubmed (Sen et al., 2008),", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 354, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 505, + 367 + ], + "score": 1.0, + "content": "where the data might had an underlying static ranking of “globally important” nodes. Velickovi ˇ c´", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "agreed that newer and more challenging benchmarks may demand stronger attention mechanisms", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "such as GATv2. that depends on the representation of the query node as well. In this paper, we revisit", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 388, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 399 + ], + "score": 1.0, + "content": "the traditional assumptions and show that many modern graph benchmarks and datasets contain more", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 399, + 338, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 338, + 410 + ], + "score": 1.0, + "content": "complex interactions, and thus require dynamic attention.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21, + "bbox_fs": [ + 104, + 332, + 506, + 410 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 431, + 195, + 444 + ], + "lines": [ + { + "bbox": [ + 104, + 430, + 197, + 447 + ], + "spans": [ + { + "bbox": [ + 104, + 430, + 197, + 447 + ], + "score": 1.0, + "content": "5 CONCLUSION", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 460, + 505, + 548 + ], + "lines": [ + { + "bbox": [ + 106, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "In this paper, we identify that the popular and widespread Graph Attention Network does not compute", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 471, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 506, + 483 + ], + "score": 1.0, + "content": "dynamic attention. Instead, the attention mechanism in the standard definition and implementations of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 482, + 507, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 507, + 496 + ], + "score": 1.0, + "content": "GAT is only static: for any query, its neighbor-scoring is monotonic with respect to per-node scores.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "score": 1.0, + "content": "Further, the ranking (the argsort) of attention coefficients is identical for all nodes in the graph,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 505, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 505, + 517 + ], + "score": 1.0, + "content": "and is unconditioned on the query node. As a result, GAT cannot even express simple alignment", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 514, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 104, + 514, + 506, + 529 + ], + "score": 1.0, + "content": "problems. To address this limitation, we introduce a simple fix and propose GATv2: by modifying", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 527, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 505, + 538 + ], + "score": 1.0, + "content": "the order of operations in GAT, GATv2 achieves a universal approximator attention function and is", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 537, + 261, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 261, + 549 + ], + "score": 1.0, + "content": "thus strictly more powerful than GAT.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5, + "bbox_fs": [ + 104, + 461, + 507, + 549 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 554, + 506, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 553, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 568 + ], + "score": 1.0, + "content": "We demonstrate the empirical advantage of GATv2 over GAT in a synthetic problem that requires dy-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 577 + ], + "score": 1.0, + "content": "namic selection of nodes, and in 12 benchmarks from OGB and other public datasets. Our experiments", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 575, + 483, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 483, + 590 + ], + "score": 1.0, + "content": "show that GATv2 outperforms GAT in all benchmarks while having the same parametric cost.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 553, + 506, + 590 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 592, + 505, + 659 + ], + "lines": [ + { + "bbox": [ + 105, + 592, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 506, + 606 + ], + "score": 1.0, + "content": "We encourage the community to use GATv2 instead or in addition to GAT whenever comparing new", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 602, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 506, + 618 + ], + "score": 1.0, + "content": "GNN architectures to the common strong baselines. In complex tasks and domains and in challenging", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "datasets, a model that uses GAT as an internal component can replace it with GATv2 to benefit", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 625, + 507, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 507, + 639 + ], + "score": 1.0, + "content": "from a strictly more powerful model. To this end, we make our code publicly available at https:", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "//github.com/tech-srl/how_attentive_are_gats ,7 and GATv2 is available as part", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 646, + 496, + 661 + ], + "spans": [ + { + "bbox": [ + 104, + 646, + 496, + 661 + ], + "score": 1.0, + "content": "of the PyTorch Geometric library,8 the Deep Graph Library,9 and the TensorFlow GNN library.10", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39.5, + "bbox_fs": [ + 104, + 592, + 507, + 661 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 218, + 93 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 219, + 96 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 219, + 96 + ], + "score": 1.0, + "content": "ACKNOWLEDGMENTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 151 + ], + "lines": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "We thank Gail Weiss for the helpful discussions, thorough feedback, and inspirational paper (Weiss", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 117, + 505, + 131 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 505, + 131 + ], + "score": 1.0, + "content": "et al., 2018). We also thank Petar Velickovi ˇ c for the useful initial discussion about the complexity ´", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 506, + 142 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 506, + 142 + ], + "score": 1.0, + "content": "and implementation of GAT, Lucio Dery, and the Graph Representation Learning Reading Group at", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 300, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 300, + 154 + ], + "score": 1.0, + "content": "Mila11 for their useful comments and questions.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 168, + 213, + 181 + ], + "lines": [ + { + "bbox": [ + 105, + 168, + 216, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 216, + 183 + ], + "score": 1.0, + "content": "A RELATED WORK", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 198, + 505, + 264 + ], + "lines": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "Attention in GNNs Modeling pairwise interactions between elements in graph-structured data goes", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 209, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 221 + ], + "score": 1.0, + "content": "back to interaction networks (Battaglia et al., 2016; Hoshen, 2017) and relational networks (Santoro", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "score": 1.0, + "content": "et al., 2017). The GAT formulation of Velickovi ˇ c et al. (2018) rose as the most popular framework ´", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 231, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 244 + ], + "score": 1.0, + "content": "for attentional GNNs, thanks to its simplicity, generality, and applicability beyond reinforcement", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "score": 1.0, + "content": "learning (Denil et al., 2017; Duan et al., 2017). Nevertheless, in this work, we show that the popular", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 252, + 430, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 430, + 266 + ], + "score": 1.0, + "content": "and widespread definition of GAT is severely constrained to static attention only.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 272, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 506, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 506, + 285 + ], + "score": 1.0, + "content": "Other graph attention mechanisms Many works employed GNNs with attention mechanisms", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 284, + 506, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 506, + 296 + ], + "score": 1.0, + "content": "other than the standard GAT’s (Zhang et al., 2018; Thekumparampil et al., 2018; Gao and Ji, 2019;", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 295, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 506, + 307 + ], + "score": 1.0, + "content": "Lukovnikov and Fischer, 2021; Shi et al., 2020; Dwivedi and Bresson, 2020; Busbridge et al., 2019;", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 304, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 104, + 304, + 506, + 320 + ], + "score": 1.0, + "content": "Rong et al., 2020a; Velickovi ˇ c et al., 2020), and Lee et al. (2018) conducted an extensive survey ´", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "of attention types in GNNs. However, none of these works identified the monotonicity of GAT’s", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 327, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 506, + 341 + ], + "score": 1.0, + "content": "attention mechanism, the theoretical differences between attention types, nor empirically compared", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 338, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 506, + 352 + ], + "score": 1.0, + "content": "their performance. Kim and Oh (2021) compared two graph attention mechanisms empirically, but in a", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 349, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 506, + 363 + ], + "score": 1.0, + "content": "specific self-supervised scenario, without observing the theoretical difference in their expressiveness.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 369, + 505, + 425 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "The static attention of GAT Qiu et al. (2018) recognized the order-preserving property of GAT, but", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "did not identify the severe theoretical constraint that this property implies: the inability to perform", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "score": 1.0, + "content": "dynamic attention (Theorem 1). Furthermore, they presented GAT’s monotonicity as a desired trait (!)", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "score": 1.0, + "content": "To the best of our knowledge, our work is the first work to recognize the inability of GAT to perform", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 414, + 339, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 414, + 339, + 426 + ], + "score": 1.0, + "content": "dynamic attention and its practical harmful consequences.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 108, + 442, + 251, + 455 + ], + "lines": [ + { + "bbox": [ + 105, + 441, + 252, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 252, + 456 + ], + "score": 1.0, + "content": "B PROOF FOR THEOREM 2", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 468, + 403, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 467, + 404, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 404, + 481 + ], + "score": 1.0, + "content": "For brevity, we repeat our definition of dynamic attention (Definition 3.2):", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 505, + 563 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 485, + 506 + ], + "score": 1.0, + "content": "Definition 3.2 (Dynamic attention). A (possibly infinite) family of scoring functions", + "type": "text" + }, + { + "bbox": [ + 485, + 493, + 505, + 504 + ], + "score": 0.84, + "content": "\\mathcal { F } \\subseteq", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 109, + 502, + 504, + 518 + ], + "spans": [ + { + "bbox": [ + 109, + 504, + 173, + 516 + ], + "score": 0.87, + "content": "\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 502, + 412, + 518 + ], + "score": 1.0, + "content": "computes dynamic scoring for a given set of key vectors", + "type": "text" + }, + { + "bbox": [ + 412, + 504, + 504, + 516 + ], + "score": 0.91, + "content": "\\mathbb { K } = \\{ \\boldsymbol { k } _ { 1 } , . . . , \\boldsymbol { k } _ { n } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 516, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 179, + 531 + ], + "score": 1.0, + "content": "and query vectors", + "type": "text" + }, + { + "bbox": [ + 180, + 516, + 272, + 529 + ], + "score": 0.91, + "content": "\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 516, + 352, + 531 + ], + "score": 1.0, + "content": ", if for any mapping", + "type": "text" + }, + { + "bbox": [ + 353, + 518, + 360, + 529 + ], + "score": 0.59, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 516, + 365, + 531 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 365, + 517, + 407, + 529 + ], + "score": 0.93, + "content": "[ m ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 516, + 456, + 531 + ], + "score": 1.0, + "content": "there exists", + "type": "text" + }, + { + "bbox": [ + 456, + 518, + 483, + 529 + ], + "score": 0.92, + "content": "f \\in { \\mathcal { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 516, + 505, + 531 + ], + "score": 1.0, + "content": "such", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 527, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 104, + 527, + 178, + 545 + ], + "score": 1.0, + "content": "that for any query", + "type": "text" + }, + { + "bbox": [ + 179, + 530, + 209, + 541 + ], + "score": 0.92, + "content": "i \\in [ m ]", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 527, + 259, + 545 + ], + "score": 1.0, + "content": "and any key", + "type": "text" + }, + { + "bbox": [ + 259, + 529, + 420, + 542 + ], + "score": 0.73, + "content": "j _ { \\neq \\varphi ( i ) } \\in [ n ] \\colon f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right) > f \\left( q _ { i } , k _ { j } \\right) .", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 527, + 506, + 545 + ], + "score": 1.0, + "content": ". We say that a family", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 540, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 330, + 553 + ], + "score": 1.0, + "content": "of attention functions computes dynamic attention for", + "type": "text" + }, + { + "bbox": [ + 330, + 541, + 340, + 550 + ], + "score": 0.76, + "content": "\\mathbb { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 540, + 358, + 553 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 359, + 541, + 368, + 551 + ], + "score": 0.85, + "content": "\\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 540, + 505, + 553 + ], + "score": 1.0, + "content": ", if its scoring function computes", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 551, + 435, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 435, + 564 + ], + "score": 1.0, + "content": "dynamic scoring, possibly followed by monotonic normalization such as softmax.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 104, + 567, + 504, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 484, + 581 + ], + "score": 1.0, + "content": "Theorem 2. A GATv2 layer computes dynamic attention for any set of node representations", + "type": "text" + }, + { + "bbox": [ + 484, + 567, + 505, + 578 + ], + "score": 0.84, + "content": "\\mathbb { K } =", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 576, + 185, + 592 + ], + "spans": [ + { + "bbox": [ + 107, + 578, + 180, + 590 + ], + "score": 0.91, + "content": "\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 576, + 185, + 592 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 106, + 605, + 505, + 639 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 154, + 618 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 154, + 606, + 207, + 618 + ], + "score": 0.92, + "content": "\\mathcal { G } ~ = ~ ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "be a graph modeled by a GATv2 layer, having node representations", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 107, + 616, + 504, + 631 + ], + "spans": [ + { + "bbox": [ + 107, + 617, + 158, + 629 + ], + "score": 0.92, + "content": "\\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 616, + 194, + 631 + ], + "score": 1.0, + "content": ", and let", + "type": "text" + }, + { + "bbox": [ + 195, + 617, + 255, + 629 + ], + "score": 0.92, + "content": "\\varphi : [ n ] \\to [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 616, + 350, + 631 + ], + "score": 1.0, + "content": "be any node mapping", + "type": "text" + }, + { + "bbox": [ + 350, + 617, + 393, + 629 + ], + "score": 0.93, + "content": "[ n ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 616, + 444, + 631 + ], + "score": 1.0, + "content": ". We define", + "type": "text" + }, + { + "bbox": [ + 445, + 616, + 504, + 629 + ], + "score": 0.92, + "content": "\\dot { g } : \\mathbb { R } ^ { 2 d } \\mathbb { R }", + "type": "inline_equation" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 627, + 153, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 153, + 640 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 638, + 375, + 667 + ], + "lines": [ + { + "bbox": [ + 234, + 638, + 375, + 667 + ], + "spans": [ + { + "bbox": [ + 234, + 638, + 375, + 667 + ], + "score": 0.93, + "content": "g \\left( \\pmb { x } \\right) = \\left. { 1 } \\pmb { \\mathrm { \\rbrace } } \\ i : \\pmb { x } = \\left[ \\pmb { h } _ { i } \\right| \\left| \\pmb { h } _ { \\varphi ( i ) } \\right] \\right.", + "type": "interline_equation", + "image_path": "da7983c6b9a359f4db8ff7bd72c050147ef4fb4f9f548a408e173d5dcc722ea8.jpg" + } + ] + } + ], + "index": 38.5, + "virtual_lines": [ + { + "bbox": [ + 234, + 638, + 375, + 652.5 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 234, + 652.5, + 375, + 667.0 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 678, + 477, + 691 + ], + "lines": [ + { + "bbox": [ + 104, + 676, + 479, + 693 + ], + "spans": [ + { + "bbox": [ + 104, + 676, + 255, + 693 + ], + "score": 1.0, + "content": "Next, we define a continues function", + "type": "text" + }, + { + "bbox": [ + 255, + 678, + 308, + 691 + ], + "score": 0.92, + "content": "\\widetilde { g } : { \\mathbb { R } } ^ { 2 d } \\to { \\mathbb { R } }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 676, + 365, + 693 + ], + "score": 1.0, + "content": "that equals to", + "type": "text" + }, + { + "bbox": [ + 365, + 681, + 371, + 691 + ], + "score": 0.82, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 676, + 436, + 693 + ], + "score": 1.0, + "content": "in only specific", + "type": "text" + }, + { + "bbox": [ + 436, + 678, + 447, + 689 + ], + "score": 0.87, + "content": "n ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 676, + 479, + 693 + ], + "score": 1.0, + "content": "inputs:", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "interline_equation", + "bbox": [ + 230, + 696, + 381, + 711 + ], + "lines": [ + { + "bbox": [ + 230, + 696, + 381, + 711 + ], + "spans": [ + { + "bbox": [ + 230, + 696, + 381, + 711 + ], + "score": 0.9, + "content": "\\widetilde { g } ( [ h _ { i } \\| h _ { j } ] ) = g ( [ h _ { i } \\| h _ { j } ] ) , \\forall i , j \\in [ n ]", + "type": "interline_equation", + "image_path": "04d87a1789c2278580b6f5bdadc283de2be469ddb47bf2adf129bccb50b5cbec.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 230, + 696, + 381, + 711 + ], + "spans": [], + "index": 41 + } + ] + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 117, + 721, + 212, + 732 + ], + "lines": [ + { + "bbox": [ + 115, + 718, + 213, + 735 + ], + "spans": [ + { + "bbox": [ + 115, + 718, + 213, + 735 + ], + "score": 1.0, + "content": "11https://grlmila.github.io/", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 218, + 93 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 219, + 96 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 219, + 96 + ], + "score": 1.0, + "content": "ACKNOWLEDGMENTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 151 + ], + "lines": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "We thank Gail Weiss for the helpful discussions, thorough feedback, and inspirational paper (Weiss", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 117, + 505, + 131 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 505, + 131 + ], + "score": 1.0, + "content": "et al., 2018). We also thank Petar Velickovi ˇ c for the useful initial discussion about the complexity ´", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 506, + 142 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 506, + 142 + ], + "score": 1.0, + "content": "and implementation of GAT, Lucio Dery, and the Graph Representation Learning Reading Group at", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 300, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 300, + 154 + ], + "score": 1.0, + "content": "Mila11 for their useful comments and questions.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 106, + 506, + 154 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 168, + 213, + 181 + ], + "lines": [ + { + "bbox": [ + 105, + 168, + 216, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 216, + 183 + ], + "score": 1.0, + "content": "A RELATED WORK", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 198, + 505, + 264 + ], + "lines": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "Attention in GNNs Modeling pairwise interactions between elements in graph-structured data goes", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 209, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 221 + ], + "score": 1.0, + "content": "back to interaction networks (Battaglia et al., 2016; Hoshen, 2017) and relational networks (Santoro", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "score": 1.0, + "content": "et al., 2017). The GAT formulation of Velickovi ˇ c et al. (2018) rose as the most popular framework ´", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 231, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 244 + ], + "score": 1.0, + "content": "for attentional GNNs, thanks to its simplicity, generality, and applicability beyond reinforcement", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "score": 1.0, + "content": "learning (Denil et al., 2017; Duan et al., 2017). Nevertheless, in this work, we show that the popular", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 252, + 430, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 430, + 266 + ], + "score": 1.0, + "content": "and widespread definition of GAT is severely constrained to static attention only.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 198, + 506, + 266 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 272, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 506, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 506, + 285 + ], + "score": 1.0, + "content": "Other graph attention mechanisms Many works employed GNNs with attention mechanisms", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 284, + 506, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 506, + 296 + ], + "score": 1.0, + "content": "other than the standard GAT’s (Zhang et al., 2018; Thekumparampil et al., 2018; Gao and Ji, 2019;", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 295, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 506, + 307 + ], + "score": 1.0, + "content": "Lukovnikov and Fischer, 2021; Shi et al., 2020; Dwivedi and Bresson, 2020; Busbridge et al., 2019;", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 304, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 104, + 304, + 506, + 320 + ], + "score": 1.0, + "content": "Rong et al., 2020a; Velickovi ˇ c et al., 2020), and Lee et al. (2018) conducted an extensive survey ´", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "of attention types in GNNs. However, none of these works identified the monotonicity of GAT’s", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 327, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 506, + 341 + ], + "score": 1.0, + "content": "attention mechanism, the theoretical differences between attention types, nor empirically compared", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 338, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 506, + 352 + ], + "score": 1.0, + "content": "their performance. Kim and Oh (2021) compared two graph attention mechanisms empirically, but in a", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 349, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 506, + 363 + ], + "score": 1.0, + "content": "specific self-supervised scenario, without observing the theoretical difference in their expressiveness.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5, + "bbox_fs": [ + 104, + 273, + 506, + 363 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 369, + 505, + 425 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "The static attention of GAT Qiu et al. (2018) recognized the order-preserving property of GAT, but", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "did not identify the severe theoretical constraint that this property implies: the inability to perform", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "score": 1.0, + "content": "dynamic attention (Theorem 1). Furthermore, they presented GAT’s monotonicity as a desired trait (!)", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "score": 1.0, + "content": "To the best of our knowledge, our work is the first work to recognize the inability of GAT to perform", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 414, + 339, + 426 + ], + "spans": [ + { + "bbox": [ + 106, + 414, + 339, + 426 + ], + "score": 1.0, + "content": "dynamic attention and its practical harmful consequences.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 369, + 506, + 426 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 442, + 251, + 455 + ], + "lines": [ + { + "bbox": [ + 105, + 441, + 252, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 252, + 456 + ], + "score": 1.0, + "content": "B PROOF FOR THEOREM 2", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 468, + 403, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 467, + 404, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 404, + 481 + ], + "score": 1.0, + "content": "For brevity, we repeat our definition of dynamic attention (Definition 3.2):", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 467, + 404, + 481 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 505, + 563 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 485, + 506 + ], + "score": 1.0, + "content": "Definition 3.2 (Dynamic attention). A (possibly infinite) family of scoring functions", + "type": "text" + }, + { + "bbox": [ + 485, + 493, + 505, + 504 + ], + "score": 0.84, + "content": "\\mathcal { F } \\subseteq", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 109, + 502, + 504, + 518 + ], + "spans": [ + { + "bbox": [ + 109, + 504, + 173, + 516 + ], + "score": 0.87, + "content": "\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 502, + 412, + 518 + ], + "score": 1.0, + "content": "computes dynamic scoring for a given set of key vectors", + "type": "text" + }, + { + "bbox": [ + 412, + 504, + 504, + 516 + ], + "score": 0.91, + "content": "\\mathbb { K } = \\{ \\boldsymbol { k } _ { 1 } , . . . , \\boldsymbol { k } _ { n } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 516, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 179, + 531 + ], + "score": 1.0, + "content": "and query vectors", + "type": "text" + }, + { + "bbox": [ + 180, + 516, + 272, + 529 + ], + "score": 0.91, + "content": "\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 516, + 352, + 531 + ], + "score": 1.0, + "content": ", if for any mapping", + "type": "text" + }, + { + "bbox": [ + 353, + 518, + 360, + 529 + ], + "score": 0.59, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 516, + 365, + 531 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 365, + 517, + 407, + 529 + ], + "score": 0.93, + "content": "[ m ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 516, + 456, + 531 + ], + "score": 1.0, + "content": "there exists", + "type": "text" + }, + { + "bbox": [ + 456, + 518, + 483, + 529 + ], + "score": 0.92, + "content": "f \\in { \\mathcal { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 516, + 505, + 531 + ], + "score": 1.0, + "content": "such", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 527, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 104, + 527, + 178, + 545 + ], + "score": 1.0, + "content": "that for any query", + "type": "text" + }, + { + "bbox": [ + 179, + 530, + 209, + 541 + ], + "score": 0.92, + "content": "i \\in [ m ]", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 527, + 259, + 545 + ], + "score": 1.0, + "content": "and any key", + "type": "text" + }, + { + "bbox": [ + 259, + 529, + 420, + 542 + ], + "score": 0.73, + "content": "j _ { \\neq \\varphi ( i ) } \\in [ n ] \\colon f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right) > f \\left( q _ { i } , k _ { j } \\right) .", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 527, + 506, + 545 + ], + "score": 1.0, + "content": ". We say that a family", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 540, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 330, + 553 + ], + "score": 1.0, + "content": "of attention functions computes dynamic attention for", + "type": "text" + }, + { + "bbox": [ + 330, + 541, + 340, + 550 + ], + "score": 0.76, + "content": "\\mathbb { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 540, + 358, + 553 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 359, + 541, + 368, + 551 + ], + "score": 0.85, + "content": "\\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 540, + 505, + 553 + ], + "score": 1.0, + "content": ", if its scoring function computes", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 551, + 435, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 435, + 564 + ], + "score": 1.0, + "content": "dynamic scoring, possibly followed by monotonic normalization such as softmax.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5, + "bbox_fs": [ + 104, + 492, + 506, + 564 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 567, + 504, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 484, + 581 + ], + "score": 1.0, + "content": "Theorem 2. A GATv2 layer computes dynamic attention for any set of node representations", + "type": "text" + }, + { + "bbox": [ + 484, + 567, + 505, + 578 + ], + "score": 0.84, + "content": "\\mathbb { K } =", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 576, + 185, + 592 + ], + "spans": [ + { + "bbox": [ + 107, + 578, + 180, + 590 + ], + "score": 0.91, + "content": "\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 576, + 185, + 592 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 564, + 505, + 592 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 605, + 505, + 639 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 154, + 618 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 154, + 606, + 207, + 618 + ], + "score": 0.92, + "content": "\\mathcal { G } ~ = ~ ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "be a graph modeled by a GATv2 layer, having node representations", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 107, + 616, + 504, + 631 + ], + "spans": [ + { + "bbox": [ + 107, + 617, + 158, + 629 + ], + "score": 0.92, + "content": "\\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 616, + 194, + 631 + ], + "score": 1.0, + "content": ", and let", + "type": "text" + }, + { + "bbox": [ + 195, + 617, + 255, + 629 + ], + "score": 0.92, + "content": "\\varphi : [ n ] \\to [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 616, + 350, + 631 + ], + "score": 1.0, + "content": "be any node mapping", + "type": "text" + }, + { + "bbox": [ + 350, + 617, + 393, + 629 + ], + "score": 0.93, + "content": "[ n ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 616, + 444, + 631 + ], + "score": 1.0, + "content": ". We define", + "type": "text" + }, + { + "bbox": [ + 445, + 616, + 504, + 629 + ], + "score": 0.92, + "content": "\\dot { g } : \\mathbb { R } ^ { 2 d } \\mathbb { R }", + "type": "inline_equation" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 627, + 153, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 153, + 640 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 605, + 505, + 640 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 638, + 375, + 667 + ], + "lines": [ + { + "bbox": [ + 234, + 638, + 375, + 667 + ], + "spans": [ + { + "bbox": [ + 234, + 638, + 375, + 667 + ], + "score": 0.93, + "content": "g \\left( \\pmb { x } \\right) = \\left. { 1 } \\pmb { \\mathrm { \\rbrace } } \\ i : \\pmb { x } = \\left[ \\pmb { h } _ { i } \\right| \\left| \\pmb { h } _ { \\varphi ( i ) } \\right] \\right.", + "type": "interline_equation", + "image_path": "da7983c6b9a359f4db8ff7bd72c050147ef4fb4f9f548a408e173d5dcc722ea8.jpg" + } + ] + } + ], + "index": 38.5, + "virtual_lines": [ + { + "bbox": [ + 234, + 638, + 375, + 652.5 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 234, + 652.5, + 375, + 667.0 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 678, + 477, + 691 + ], + "lines": [ + { + "bbox": [ + 104, + 676, + 479, + 693 + ], + "spans": [ + { + "bbox": [ + 104, + 676, + 255, + 693 + ], + "score": 1.0, + "content": "Next, we define a continues function", + "type": "text" + }, + { + "bbox": [ + 255, + 678, + 308, + 691 + ], + "score": 0.92, + "content": "\\widetilde { g } : { \\mathbb { R } } ^ { 2 d } \\to { \\mathbb { R } }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 676, + 365, + 693 + ], + "score": 1.0, + "content": "that equals to", + "type": "text" + }, + { + "bbox": [ + 365, + 681, + 371, + 691 + ], + "score": 0.82, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 676, + 436, + 693 + ], + "score": 1.0, + "content": "in only specific", + "type": "text" + }, + { + "bbox": [ + 436, + 678, + 447, + 689 + ], + "score": 0.87, + "content": "n ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 676, + 479, + 693 + ], + "score": 1.0, + "content": "inputs:", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40, + "bbox_fs": [ + 104, + 676, + 479, + 693 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 230, + 696, + 381, + 711 + ], + "lines": [ + { + "bbox": [ + 230, + 696, + 381, + 711 + ], + "spans": [ + { + "bbox": [ + 230, + 696, + 381, + 711 + ], + "score": 0.9, + "content": "\\widetilde { g } ( [ h _ { i } \\| h _ { j } ] ) = g ( [ h _ { i } \\| h _ { j } ] ) , \\forall i , j \\in [ n ]", + "type": "interline_equation", + "image_path": "04d87a1789c2278580b6f5bdadc283de2be469ddb47bf2adf129bccb50b5cbec.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 230, + 696, + 381, + 711 + ], + "spans": [], + "index": 41 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 187, + 95 + ], + "score": 1.0, + "content": "For all other inputs", + "type": "text" + }, + { + "bbox": [ + 188, + 82, + 223, + 93 + ], + "score": 0.89, + "content": "\\boldsymbol { x } \\in \\mathbb { R } ^ { 2 d }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 81, + 227, + 95 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 228, + 82, + 248, + 95 + ], + "score": 0.8, + "content": "\\widetilde g ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 81, + 466, + 95 + ], + "score": 1.0, + "content": "realizes to any values that maintain the continuity of", + "type": "text" + }, + { + "bbox": [ + 467, + 83, + 474, + 94 + ], + "score": 0.83, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 81, + 506, + 95 + ], + "score": 1.0, + "content": "(this is", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 92, + 407, + 106 + ], + "spans": [ + { + "bbox": [ + 104, + 92, + 265, + 106 + ], + "score": 1.0, + "content": "epossible because we fixed the values of", + "type": "text" + }, + { + "bbox": [ + 265, + 94, + 272, + 105 + ], + "score": 0.85, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 92, + 407, + 106 + ], + "score": 1.0, + "content": "for only a finite set of points). 12", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 108, + 110, + 282, + 122 + ], + "lines": [ + { + "bbox": [ + 106, + 109, + 283, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 109, + 191, + 124 + ], + "score": 1.0, + "content": "Thus, for every node", + "type": "text" + }, + { + "bbox": [ + 191, + 111, + 215, + 121 + ], + "score": 0.9, + "content": "i \\in \\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 109, + 233, + 124 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 234, + 111, + 279, + 124 + ], + "score": 0.92, + "content": "j _ { \\neq \\varphi ( i ) } \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 109, + 283, + 124 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 129, + 385, + 145 + ], + "lines": [ + { + "bbox": [ + 225, + 129, + 385, + 145 + ], + "spans": [ + { + "bbox": [ + 225, + 129, + 385, + 145 + ], + "score": 0.93, + "content": "1 = \\widetilde { g } \\left( \\left[ h _ { i } \\Vert h _ { \\varphi ( i ) } \\right] \\right) > \\widetilde { g } \\left( \\left[ h _ { i } \\Vert h _ { j } \\right] \\right) = 0", + "type": "interline_equation", + "image_path": "e64247a44bd0da371c015f9ee20aa6ecc9ef48dfad9589c1616a16565ea446da.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 225, + 129, + 385, + 145 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 155, + 505, + 200 + ], + "lines": [ + { + "bbox": [ + 104, + 155, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 104, + 155, + 388, + 168 + ], + "score": 1.0, + "content": "If we concatenate the two input vectors, and define the scoring function", + "type": "text" + }, + { + "bbox": [ + 388, + 158, + 394, + 166 + ], + "score": 0.71, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 155, + 506, + 168 + ], + "score": 1.0, + "content": "of GATv2 (Equation (7)) as", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 166, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 166, + 260, + 179 + ], + "score": 1.0, + "content": "a function of the concatenated vector", + "type": "text" + }, + { + "bbox": [ + 260, + 167, + 292, + 179 + ], + "score": 0.92, + "content": "[ h _ { i } | | h _ { j } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 166, + 506, + 179 + ], + "score": 1.0, + "content": ", from the universal approximation theorem (Hornik", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 177, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 354, + 191 + ], + "score": 1.0, + "content": "et al., 1989; Cybenko, 1989; Funahashi, 1989; Hornik, 1991),", + "type": "text" + }, + { + "bbox": [ + 354, + 180, + 361, + 188 + ], + "score": 0.67, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 177, + 430, + 191 + ], + "score": 1.0, + "content": "can approximate", + "type": "text" + }, + { + "bbox": [ + 430, + 178, + 437, + 190 + ], + "score": 0.85, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 177, + 505, + 191 + ], + "score": 1.0, + "content": "for any compact", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 187, + 166, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 187, + 166, + 200 + ], + "score": 1.0, + "content": "subset of R2d.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 106, + 205, + 503, + 230 + ], + "lines": [ + { + "bbox": [ + 105, + 204, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 236, + 218 + ], + "score": 1.0, + "content": "Thus, for any sufficiently small", + "type": "text" + }, + { + "bbox": [ + 236, + 208, + 242, + 215 + ], + "score": 0.72, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 204, + 264, + 218 + ], + "score": 1.0, + "content": "(any", + "type": "text" + }, + { + "bbox": [ + 264, + 206, + 313, + 218 + ], + "score": 0.91, + "content": "0 < \\epsilon < 1 / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 204, + 410, + 218 + ], + "score": 1.0, + "content": ") there exist parameters", + "type": "text" + }, + { + "bbox": [ + 411, + 206, + 424, + 216 + ], + "score": 0.8, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 204, + 442, + 218 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 443, + 208, + 450, + 216 + ], + "score": 0.77, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 204, + 505, + 218 + ], + "score": 1.0, + "content": "such that for", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 214, + 249, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 152, + 232 + ], + "score": 1.0, + "content": "every node", + "type": "text" + }, + { + "bbox": [ + 153, + 217, + 177, + 227 + ], + "score": 0.9, + "content": "i \\in \\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 214, + 219, + 232 + ], + "score": 1.0, + "content": "and every", + "type": "text" + }, + { + "bbox": [ + 219, + 218, + 244, + 230 + ], + "score": 0.92, + "content": "\\mathscr { j } _ { \\neq \\varphi ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 214, + 249, + 232 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "interline_equation", + "bbox": [ + 200, + 235, + 411, + 251 + ], + "lines": [ + { + "bbox": [ + 200, + 235, + 411, + 251 + ], + "spans": [ + { + "bbox": [ + 200, + 235, + 411, + 251 + ], + "score": 0.9, + "content": "{ e _ { W , a } } \\left( { h _ { i } } , { h _ { \\varphi ( i ) } } \\right) > 1 - \\epsilon > 0 + \\epsilon > { e _ { W , a } } \\left( { h _ { i } } , { h _ { j } } \\right)", + "type": "interline_equation", + "image_path": "87359cb3bdeba2a1c0aa11b6a1906074b8180a2e57b0bdb4e622f9fa259bc474.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 200, + 235, + 411, + 251 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 256, + 314, + 268 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 315, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 315, + 270 + ], + "score": 1.0, + "content": "and due to the increasing monotonicity of softmax:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "interline_equation", + "bbox": [ + 277, + 276, + 334, + 289 + ], + "lines": [ + { + "bbox": [ + 277, + 276, + 334, + 289 + ], + "spans": [ + { + "bbox": [ + 277, + 276, + 334, + 289 + ], + "score": 0.91, + "content": "\\alpha _ { i , \\varphi ( i ) } > \\alpha _ { i , j }", + "type": "interline_equation", + "image_path": "ae6999271be462896680164b08faadbcb45f1db0885e6c9aada7884447c43d1a.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 277, + 276, + 334, + 289 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 320, + 505, + 375 + ], + "lines": [ + { + "bbox": [ + 106, + 319, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 319, + 506, + 333 + ], + "score": 1.0, + "content": "The choice of nonlinearity In general, these results hold if GATv2 had used any common non-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "score": 1.0, + "content": "polynomial activation function (such as ReLU, sigmoid, or the hyperbolic tangent function). The", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 342, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 505, + 355 + ], + "score": 1.0, + "content": "LeakyReLU activation function of GATv2 does not change its universal approximation ability (Leshno", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "score": 1.0, + "content": "et al., 1993; Pinkus, 1999; Park et al., 2021), and it was chosen only for consistency with the original", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 364, + 181, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 181, + 376 + ], + "score": 1.0, + "content": "definition of GAT.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 393, + 368, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 370, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 370, + 408 + ], + "score": 1.0, + "content": "C GENERALIZATION TO MULTI-HEAD ATTENTION", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 417, + 505, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 325, + 430 + ], + "score": 1.0, + "content": "Velickovi ˇ c et al. (2018) found it beneficial to employ ´", + "type": "text" + }, + { + "bbox": [ + 325, + 419, + 336, + 428 + ], + "score": 0.79, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 418, + 505, + 430 + ], + "score": 1.0, + "content": "separate attention heads and concatenate", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "their outputs, similarly to Transformers. In this case, Theorem 1 holds for each head separately: every", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 438, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 128, + 459 + ], + "score": 1.0, + "content": "head", + "type": "text" + }, + { + "bbox": [ + 128, + 442, + 162, + 455 + ], + "score": 0.91, + "content": "h \\in [ H ]", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 438, + 354, + 459 + ], + "score": 1.0, + "content": "has a (possibly different) node that maximizes", + "type": "text" + }, + { + "bbox": [ + 354, + 440, + 414, + 456 + ], + "score": 0.93, + "content": "\\{ s _ { j } ^ { ( h ) } \\mid j \\in \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 438, + 505, + 459 + ], + "score": 1.0, + "content": ", and the output is the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 454, + 275, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 175, + 465 + ], + "score": 1.0, + "content": "concatenation of", + "type": "text" + }, + { + "bbox": [ + 175, + 454, + 185, + 464 + ], + "score": 0.84, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 454, + 275, + 465 + ], + "score": 1.0, + "content": "static attention heads.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20.5 + }, + { + "type": "title", + "bbox": [ + 108, + 482, + 227, + 495 + ], + "lines": [ + { + "bbox": [ + 105, + 480, + 229, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 229, + 497 + ], + "score": 1.0, + "content": "D TRAINING DETAILS", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 108, + 507, + 504, + 541 + ], + "lines": [ + { + "bbox": [ + 105, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "In this section we elaborate on the training details of all of our experiments. All models use residual", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "score": 1.0, + "content": "connections as in Velickovi ˇ c et al. (2018). All used code and data are publicly available under the ´", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 529, + 160, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 160, + 541 + ], + "score": 1.0, + "content": "MIT license.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + }, + { + "type": "title", + "bbox": [ + 108, + 555, + 266, + 567 + ], + "lines": [ + { + "bbox": [ + 106, + 555, + 267, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 267, + 568 + ], + "score": 1.0, + "content": "D.1 NODE- AND LINK-PREDICTION", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "score": 1.0, + "content": "We used the provided splits of OGB (Hu et al., 2020) and the Adam optimizer. We tuned the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 286, + 600 + ], + "score": 1.0, + "content": "following hyperparameters: number of layers", + "type": "text" + }, + { + "bbox": [ + 286, + 587, + 330, + 599 + ], + "score": 0.92, + "content": "\\in \\{ 2 , 3 , 6 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 587, + 380, + 600 + ], + "score": 1.0, + "content": ", hidden size", + "type": "text" + }, + { + "bbox": [ + 381, + 587, + 450, + 599 + ], + "score": 0.87, + "content": "\\in \\{ 6 4 , 1 2 8 , 2 5 6 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 587, + 505, + 600 + ], + "score": 1.0, + "content": ", learning rate", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 597, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 231, + 610 + ], + "score": 0.86, + "content": "\\in \\{ 0 . 0 0 0 5 , 0 . 0 0 1 , 0 . 0 0 5 , 0 . 0 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 597, + 505, + 611 + ], + "score": 1.0, + "content": "and sampling method – full batch, GraphSAINT (Zeng et al., 2019)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 608, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 506, + 622 + ], + "score": 1.0, + "content": "and NeighborSampling (Hamilton et al., 2017). We tuned hyperparameters according to validation", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 620, + 410, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 620, + 410, + 633 + ], + "score": 1.0, + "content": "score and early stopping. The final hyperparameters are detailed in Table 4.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 641, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 115, + 639, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 115, + 639, + 173, + 655 + ], + "score": 1.0, + "content": "12The function", + "type": "text" + }, + { + "bbox": [ + 173, + 642, + 180, + 652 + ], + "score": 0.8, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 639, + 506, + 655 + ], + "score": 1.0, + "content": "is a function that we define for the ease of proof, because the universal approximation", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 651, + 506, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 442, + 664 + ], + "score": 1.0, + "content": "etheorem is defined for continuous functions, and we only need the scoring function of GATv2", + "type": "text" + }, + { + "bbox": [ + 442, + 653, + 448, + 661 + ], + "score": 0.59, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 651, + 506, + 664 + ], + "score": 1.0, + "content": "to approximate", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 661, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 153, + 674 + ], + "score": 1.0, + "content": "the mapping", + "type": "text" + }, + { + "bbox": [ + 153, + 663, + 160, + 672 + ], + "score": 0.82, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 661, + 370, + 674 + ], + "score": 1.0, + "content": "in a finite set of points. So, we need the attention function", + "type": "text" + }, + { + "bbox": [ + 370, + 664, + 375, + 670 + ], + "score": 0.73, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 661, + 432, + 674 + ], + "score": 1.0, + "content": "to approximate", + "type": "text" + }, + { + "bbox": [ + 432, + 663, + 438, + 672 + ], + "score": 0.82, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 661, + 506, + 674 + ], + "score": 1.0, + "content": "(from Equation 8)", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 671, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 230, + 684 + ], + "score": 1.0, + "content": "in some specific points. But, since", + "type": "text" + }, + { + "bbox": [ + 230, + 674, + 235, + 682 + ], + "score": 0.77, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 671, + 337, + 684 + ], + "score": 1.0, + "content": "is not continuous, we define", + "type": "text" + }, + { + "bbox": [ + 337, + 672, + 343, + 682 + ], + "score": 0.84, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 671, + 506, + 684 + ], + "score": 1.0, + "content": "and use the universal approximation theorem", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 681, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 119, + 694 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 120, + 682, + 125, + 692 + ], + "score": 0.8, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 681, + 151, + 694 + ], + "score": 1.0, + "content": ". Since", + "type": "text" + }, + { + "bbox": [ + 151, + 684, + 157, + 691 + ], + "score": 0.62, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 681, + 209, + 694 + ], + "score": 1.0, + "content": "approximates", + "type": "text" + }, + { + "bbox": [ + 209, + 682, + 215, + 692 + ], + "score": 0.8, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 681, + 218, + 694 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 219, + 684, + 224, + 691 + ], + "score": 0.43, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 681, + 293, + 694 + ], + "score": 1.0, + "content": "also approximates", + "type": "text" + }, + { + "bbox": [ + 294, + 686, + 299, + 692 + ], + "score": 0.82, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 681, + 506, + 694 + ], + "score": 1.0, + "content": "ein our specific points, as a special case. We only require", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 691, + 506, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 122, + 703 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 122, + 693, + 128, + 702 + ], + "score": 0.86, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 691, + 197, + 703 + ], + "score": 1.0, + "content": "will be identical to", + "type": "text" + }, + { + "bbox": [ + 198, + 693, + 204, + 702 + ], + "score": 0.83, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 691, + 243, + 703 + ], + "score": 1.0, + "content": "ein specific", + "type": "text" + }, + { + "bbox": [ + 244, + 693, + 254, + 700 + ], + "score": 0.87, + "content": "n ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 691, + 279, + 703 + ], + "score": 1.0, + "content": "points", + "type": "text" + }, + { + "bbox": [ + 280, + 693, + 358, + 703 + ], + "score": 0.91, + "content": "\\{ [ h _ { i } | | h _ { j } ] \\mid i , j \\in [ n ] \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 691, + 506, + 703 + ], + "score": 1.0, + "content": ". For the rest of the input space, we don’t", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 701, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 242, + 713 + ], + "score": 1.0, + "content": "ehave any requirement on the value of", + "type": "text" + }, + { + "bbox": [ + 243, + 703, + 247, + 712 + ], + "score": 0.88, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 701, + 396, + 713 + ], + "score": 1.0, + "content": ", except for maintaining the continuity of", + "type": "text" + }, + { + "bbox": [ + 397, + 703, + 402, + 712 + ], + "score": 0.85, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 701, + 506, + 713 + ], + "score": 1.0, + "content": ". There exist infinitely many", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 711, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 156, + 725 + ], + "score": 1.0, + "content": "such possible", + "type": "text" + }, + { + "bbox": [ + 157, + 712, + 163, + 722 + ], + "score": 0.86, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 711, + 345, + 725 + ], + "score": 1.0, + "content": "e for every given set of keys, queries and a mapping", + "type": "text" + }, + { + "bbox": [ + 346, + 716, + 352, + 722 + ], + "score": 0.88, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 711, + 506, + 725 + ], + "score": 1.0, + "content": "e, but the concrete functions are not needed", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 721, + 157, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 157, + 733 + ], + "score": 1.0, + "content": "for the proof.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 495, + 294, + 504, + 304 + ], + "lines": [ + { + "bbox": [ + 496, + 296, + 504, + 304 + ], + "spans": [ + { + "bbox": [ + 496, + 296, + 504, + 304 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 187, + 95 + ], + "score": 1.0, + "content": "For all other inputs", + "type": "text" + }, + { + "bbox": [ + 188, + 82, + 223, + 93 + ], + "score": 0.89, + "content": "\\boldsymbol { x } \\in \\mathbb { R } ^ { 2 d }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 81, + 227, + 95 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 228, + 82, + 248, + 95 + ], + "score": 0.8, + "content": "\\widetilde g ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 81, + 466, + 95 + ], + "score": 1.0, + "content": "realizes to any values that maintain the continuity of", + "type": "text" + }, + { + "bbox": [ + 467, + 83, + 474, + 94 + ], + "score": 0.83, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 81, + 506, + 95 + ], + "score": 1.0, + "content": "(this is", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 92, + 407, + 106 + ], + "spans": [ + { + "bbox": [ + 104, + 92, + 265, + 106 + ], + "score": 1.0, + "content": "epossible because we fixed the values of", + "type": "text" + }, + { + "bbox": [ + 265, + 94, + 272, + 105 + ], + "score": 0.85, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 92, + 407, + 106 + ], + "score": 1.0, + "content": "for only a finite set of points). 12", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 104, + 81, + 506, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 110, + 282, + 122 + ], + "lines": [ + { + "bbox": [ + 106, + 109, + 283, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 109, + 191, + 124 + ], + "score": 1.0, + "content": "Thus, for every node", + "type": "text" + }, + { + "bbox": [ + 191, + 111, + 215, + 121 + ], + "score": 0.9, + "content": "i \\in \\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 109, + 233, + 124 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 234, + 111, + 279, + 124 + ], + "score": 0.92, + "content": "j _ { \\neq \\varphi ( i ) } \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 109, + 283, + 124 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 106, + 109, + 283, + 124 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 129, + 385, + 145 + ], + "lines": [ + { + "bbox": [ + 225, + 129, + 385, + 145 + ], + "spans": [ + { + "bbox": [ + 225, + 129, + 385, + 145 + ], + "score": 0.93, + "content": "1 = \\widetilde { g } \\left( \\left[ h _ { i } \\Vert h _ { \\varphi ( i ) } \\right] \\right) > \\widetilde { g } \\left( \\left[ h _ { i } \\Vert h _ { j } \\right] \\right) = 0", + "type": "interline_equation", + "image_path": "e64247a44bd0da371c015f9ee20aa6ecc9ef48dfad9589c1616a16565ea446da.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 225, + 129, + 385, + 145 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 155, + 505, + 200 + ], + "lines": [ + { + "bbox": [ + 104, + 155, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 104, + 155, + 388, + 168 + ], + "score": 1.0, + "content": "If we concatenate the two input vectors, and define the scoring function", + "type": "text" + }, + { + "bbox": [ + 388, + 158, + 394, + 166 + ], + "score": 0.71, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 155, + 506, + 168 + ], + "score": 1.0, + "content": "of GATv2 (Equation (7)) as", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 166, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 166, + 260, + 179 + ], + "score": 1.0, + "content": "a function of the concatenated vector", + "type": "text" + }, + { + "bbox": [ + 260, + 167, + 292, + 179 + ], + "score": 0.92, + "content": "[ h _ { i } | | h _ { j } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 166, + 506, + 179 + ], + "score": 1.0, + "content": ", from the universal approximation theorem (Hornik", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 177, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 354, + 191 + ], + "score": 1.0, + "content": "et al., 1989; Cybenko, 1989; Funahashi, 1989; Hornik, 1991),", + "type": "text" + }, + { + "bbox": [ + 354, + 180, + 361, + 188 + ], + "score": 0.67, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 177, + 430, + 191 + ], + "score": 1.0, + "content": "can approximate", + "type": "text" + }, + { + "bbox": [ + 430, + 178, + 437, + 190 + ], + "score": 0.85, + "content": "\\widetilde g", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 177, + 505, + 191 + ], + "score": 1.0, + "content": "for any compact", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 187, + 166, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 187, + 166, + 200 + ], + "score": 1.0, + "content": "subset of R2d.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5.5, + "bbox_fs": [ + 104, + 155, + 506, + 200 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 205, + 503, + 230 + ], + "lines": [ + { + "bbox": [ + 105, + 204, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 236, + 218 + ], + "score": 1.0, + "content": "Thus, for any sufficiently small", + "type": "text" + }, + { + "bbox": [ + 236, + 208, + 242, + 215 + ], + "score": 0.72, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 204, + 264, + 218 + ], + "score": 1.0, + "content": "(any", + "type": "text" + }, + { + "bbox": [ + 264, + 206, + 313, + 218 + ], + "score": 0.91, + "content": "0 < \\epsilon < 1 / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 204, + 410, + 218 + ], + "score": 1.0, + "content": ") there exist parameters", + "type": "text" + }, + { + "bbox": [ + 411, + 206, + 424, + 216 + ], + "score": 0.8, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 204, + 442, + 218 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 443, + 208, + 450, + 216 + ], + "score": 0.77, + "content": "\\textbf { \\em a }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 204, + 505, + 218 + ], + "score": 1.0, + "content": "such that for", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 214, + 249, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 152, + 232 + ], + "score": 1.0, + "content": "every node", + "type": "text" + }, + { + "bbox": [ + 153, + 217, + 177, + 227 + ], + "score": 0.9, + "content": "i \\in \\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 214, + 219, + 232 + ], + "score": 1.0, + "content": "and every", + "type": "text" + }, + { + "bbox": [ + 219, + 218, + 244, + 230 + ], + "score": 0.92, + "content": "\\mathscr { j } _ { \\neq \\varphi ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 214, + 249, + 232 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 204, + 505, + 232 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 200, + 235, + 411, + 251 + ], + "lines": [ + { + "bbox": [ + 200, + 235, + 411, + 251 + ], + "spans": [ + { + "bbox": [ + 200, + 235, + 411, + 251 + ], + "score": 0.9, + "content": "{ e _ { W , a } } \\left( { h _ { i } } , { h _ { \\varphi ( i ) } } \\right) > 1 - \\epsilon > 0 + \\epsilon > { e _ { W , a } } \\left( { h _ { i } } , { h _ { j } } \\right)", + "type": "interline_equation", + "image_path": "87359cb3bdeba2a1c0aa11b6a1906074b8180a2e57b0bdb4e622f9fa259bc474.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 200, + 235, + 411, + 251 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 256, + 314, + 268 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 315, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 315, + 270 + ], + "score": 1.0, + "content": "and due to the increasing monotonicity of softmax:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 254, + 315, + 270 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 277, + 276, + 334, + 289 + ], + "lines": [ + { + "bbox": [ + 277, + 276, + 334, + 289 + ], + "spans": [ + { + "bbox": [ + 277, + 276, + 334, + 289 + ], + "score": 0.91, + "content": "\\alpha _ { i , \\varphi ( i ) } > \\alpha _ { i , j }", + "type": "interline_equation", + "image_path": "ae6999271be462896680164b08faadbcb45f1db0885e6c9aada7884447c43d1a.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 277, + 276, + 334, + 289 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 320, + 505, + 375 + ], + "lines": [ + { + "bbox": [ + 106, + 319, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 319, + 506, + 333 + ], + "score": 1.0, + "content": "The choice of nonlinearity In general, these results hold if GATv2 had used any common non-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "score": 1.0, + "content": "polynomial activation function (such as ReLU, sigmoid, or the hyperbolic tangent function). The", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 342, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 505, + 355 + ], + "score": 1.0, + "content": "LeakyReLU activation function of GATv2 does not change its universal approximation ability (Leshno", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 505, + 367 + ], + "score": 1.0, + "content": "et al., 1993; Pinkus, 1999; Park et al., 2021), and it was chosen only for consistency with the original", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 364, + 181, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 181, + 376 + ], + "score": 1.0, + "content": "definition of GAT.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 319, + 506, + 376 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 393, + 368, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 370, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 370, + 408 + ], + "score": 1.0, + "content": "C GENERALIZATION TO MULTI-HEAD ATTENTION", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 417, + 505, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 325, + 430 + ], + "score": 1.0, + "content": "Velickovi ˇ c et al. (2018) found it beneficial to employ ´", + "type": "text" + }, + { + "bbox": [ + 325, + 419, + 336, + 428 + ], + "score": 0.79, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 418, + 505, + 430 + ], + "score": 1.0, + "content": "separate attention heads and concatenate", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "their outputs, similarly to Transformers. In this case, Theorem 1 holds for each head separately: every", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 438, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 128, + 459 + ], + "score": 1.0, + "content": "head", + "type": "text" + }, + { + "bbox": [ + 128, + 442, + 162, + 455 + ], + "score": 0.91, + "content": "h \\in [ H ]", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 438, + 354, + 459 + ], + "score": 1.0, + "content": "has a (possibly different) node that maximizes", + "type": "text" + }, + { + "bbox": [ + 354, + 440, + 414, + 456 + ], + "score": 0.93, + "content": "\\{ s _ { j } ^ { ( h ) } \\mid j \\in \\mathcal { V } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 438, + 505, + 459 + ], + "score": 1.0, + "content": ", and the output is the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 454, + 275, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 175, + 465 + ], + "score": 1.0, + "content": "concatenation of", + "type": "text" + }, + { + "bbox": [ + 175, + 454, + 185, + 464 + ], + "score": 0.84, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 454, + 275, + 465 + ], + "score": 1.0, + "content": "static attention heads.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 418, + 505, + 465 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 482, + 227, + 495 + ], + "lines": [ + { + "bbox": [ + 105, + 480, + 229, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 229, + 497 + ], + "score": 1.0, + "content": "D TRAINING DETAILS", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 108, + 507, + 504, + 541 + ], + "lines": [ + { + "bbox": [ + 105, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "In this section we elaborate on the training details of all of our experiments. All models use residual", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "score": 1.0, + "content": "connections as in Velickovi ˇ c et al. (2018). All used code and data are publicly available under the ´", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 529, + 160, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 160, + 541 + ], + "score": 1.0, + "content": "MIT license.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 507, + 505, + 541 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 555, + 266, + 567 + ], + "lines": [ + { + "bbox": [ + 106, + 555, + 267, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 267, + 568 + ], + "score": 1.0, + "content": "D.1 NODE- AND LINK-PREDICTION", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "score": 1.0, + "content": "We used the provided splits of OGB (Hu et al., 2020) and the Adam optimizer. We tuned the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 286, + 600 + ], + "score": 1.0, + "content": "following hyperparameters: number of layers", + "type": "text" + }, + { + "bbox": [ + 286, + 587, + 330, + 599 + ], + "score": 0.92, + "content": "\\in \\{ 2 , 3 , 6 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 587, + 380, + 600 + ], + "score": 1.0, + "content": ", hidden size", + "type": "text" + }, + { + "bbox": [ + 381, + 587, + 450, + 599 + ], + "score": 0.87, + "content": "\\in \\{ 6 4 , 1 2 8 , 2 5 6 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 587, + 505, + 600 + ], + "score": 1.0, + "content": ", learning rate", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 597, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 231, + 610 + ], + "score": 0.86, + "content": "\\in \\{ 0 . 0 0 0 5 , 0 . 0 0 1 , 0 . 0 0 5 , 0 . 0 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 597, + 505, + 611 + ], + "score": 1.0, + "content": "and sampling method – full batch, GraphSAINT (Zeng et al., 2019)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 608, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 506, + 622 + ], + "score": 1.0, + "content": "and NeighborSampling (Hamilton et al., 2017). We tuned hyperparameters according to validation", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 620, + 410, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 620, + 410, + 633 + ], + "score": 1.0, + "content": "score and early stopping. The final hyperparameters are detailed in Table 4.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 575, + 506, + 633 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 152, + 80, + 458, + 169 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 152, + 80, + 458, + 169 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 152, + 80, + 458, + 169 + ], + "spans": [ + { + "bbox": [ + 152, + 80, + 458, + 169 + ], + "score": 0.982, + "html": "
Dataset# layersHidden sizeLearning rateSampling method
ogbn-arxiv32560.01GraphSAINT
ogbn-products31280.001NeighborSampling
ogbn-mag22560.01NeighborSampling
ogbn-proteins6640.01NeighborSampling
ogbl-collab3640.001Full Batch
ogbl-citation232560.0005NeighborSampling
", + "type": "table", + "image_path": "99ab02b3e8b0afbf67c8ce47dd5a7df3c52f36e24045d253fded2e4adebe327b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 152, + 80, + 458, + 109.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 152, + 109.66666666666667, + 458, + 139.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 152, + 139.33333333333334, + 458, + 169.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 180, + 177, + 430, + 189 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 179, + 175, + 431, + 190 + ], + "spans": [ + { + "bbox": [ + 179, + 175, + 431, + 190 + ], + "score": 1.0, + "content": "Table 4: Training details of node- and link-prediction datasets.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "title", + "bbox": [ + 107, + 210, + 235, + 222 + ], + "lines": [ + { + "bbox": [ + 106, + 210, + 236, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 236, + 223 + ], + "score": 1.0, + "content": "D.2 ROBUSTNESS TO NOISE", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 231, + 504, + 254 + ], + "lines": [ + { + "bbox": [ + 106, + 231, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 506, + 244 + ], + "score": 1.0, + "content": "In these experiments, we used the same best-found hyperparameters in node-prediction, with 8", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 243, + 477, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 477, + 255 + ], + "score": 1.0, + "content": "attention heads in ogbn-arxiv and 1 head in ogbn-mag. Each point is an average of 10 runs.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "title", + "bbox": [ + 106, + 268, + 341, + 279 + ], + "lines": [ + { + "bbox": [ + 106, + 268, + 342, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 342, + 281 + ], + "score": 1.0, + "content": "D.3 SYNTHETIC BENCHMARK: DICTIONARYLOOKUP", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 105, + 289, + 504, + 312 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 405, + 303 + ], + "score": 1.0, + "content": "In all experiments, we used a learning rate decay of 0.5, a hidden size of", + "type": "text" + }, + { + "bbox": [ + 405, + 290, + 440, + 300 + ], + "score": 0.89, + "content": "d = 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 288, + 506, + 303 + ], + "score": 1.0, + "content": ", a batch size of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 300, + 231, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 231, + 312 + ], + "score": 1.0, + "content": "1024, and the Adam optimizer.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 317, + 505, + 372 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 312, + 330 + ], + "score": 1.0, + "content": "We created a separate dataset for every graph size", + "type": "text" + }, + { + "bbox": [ + 313, + 318, + 325, + 329 + ], + "score": 0.63, + "content": "( k )", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 317, + 505, + 330 + ], + "score": 1.0, + "content": ", and we split each such dataset to train and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 329, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 340 + ], + "score": 1.0, + "content": "test with a ratio of 80:20. Since this is a contrived problem, we did not use a validation set, and the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "reported test results can be thought of as validation results. Every model was trained on a fixed value", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 349, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 117, + 363 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 117, + 351, + 124, + 360 + ], + "score": 0.79, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 349, + 505, + 363 + ], + "score": 1.0, + "content": ". Every key node (bottom row in Figure 2) was encoded as a sum of learned attribute embedding", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 361, + 282, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 282, + 374 + ], + "score": 1.0, + "content": "and a value embedding, followed by ReLU.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 378, + 504, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "We experimented with layer normalization, batch normalization, dropout, various activation functions", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 389, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 505, + 402 + ], + "score": 1.0, + "content": "and various learning rates. None of these changed the general trend, so the experiments in Figure 3", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 400, + 463, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 463, + 412 + ], + "score": 1.0, + "content": "were conducted without any normalization, without dropout and a learning rate of 0.001.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 107, + 425, + 241, + 437 + ], + "lines": [ + { + "bbox": [ + 105, + 424, + 243, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 243, + 438 + ], + "score": 1.0, + "content": "D.4 PROGRAMS: VARMISUSE", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 108, + 447, + 504, + 480 + ], + "lines": [ + { + "bbox": [ + 106, + 447, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 505, + 459 + ], + "score": 1.0, + "content": "We used the code, splits, and the same best-found configurations as Brockschmidt (2020), who", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 457, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 506, + 471 + ], + "score": 1.0, + "content": "performed an extensive hyperparameter tuning by searching over 30 configurations for each GNN", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 469, + 267, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 267, + 480 + ], + "score": 1.0, + "content": "type. We trained each model five times.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 504, + 508 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "score": 1.0, + "content": "We took the best-found hyperparameters of Brockschmidt (2020) for GAT and used them to train", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 496, + 250, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 250, + 510 + ], + "score": 1.0, + "content": "GATv2, without any further tuning.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "title", + "bbox": [ + 108, + 522, + 251, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 520, + 252, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 252, + 537 + ], + "score": 1.0, + "content": "D.5 GRAPH-PREDICTION: QM9", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 543, + 505, + 599 + ], + "lines": [ + { + "bbox": [ + 106, + 543, + 506, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 506, + 557 + ], + "score": 1.0, + "content": "We used the code and splits of Brockschmidt (2020) who performed an extensive hyperparameter", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 554, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 568 + ], + "score": 1.0, + "content": "search over 500 configurations. We took the best-found hyperparameters of Brockschmidt (2020)", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "score": 1.0, + "content": "for GAT and used them to train GATv2. The only minor change from GAT is placing a residual", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "score": 1.0, + "content": "connection after every layer, rather than after every other layer, which is within the experimented", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 587, + 372, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 372, + 600 + ], + "score": 1.0, + "content": "hyperparameter search that was reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 108, + 613, + 252, + 624 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 253, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 253, + 626 + ], + "score": 1.0, + "content": "D.6 COMPUTE AND RESOURCES", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 108, + 634, + 504, + 657 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 648 + ], + "score": 1.0, + "content": "Our experiments consumed approximately 100 days of GPU in total. We used cloud GPUs of type", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 646, + 365, + 657 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 365, + 657 + ], + "score": 1.0, + "content": "V100, and we used RTX 3080 and 3090 in local GPU machines.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "title", + "bbox": [ + 108, + 674, + 217, + 687 + ], + "lines": [ + { + "bbox": [ + 105, + 672, + 218, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 218, + 689 + ], + "score": 1.0, + "content": "E DATA STATISTICS", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "title", + "bbox": [ + 107, + 699, + 312, + 711 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 313, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 313, + 712 + ], + "score": 1.0, + "content": "E.1 NODE- AND LINK-PREDICTION DATASETS", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 104, + 720, + 467, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 468, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 468, + 732 + ], + "score": 1.0, + "content": "Statistics of the OGB datasets we used for node- and link-prediction are shown in Table 5.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 152, + 80, + 458, + 169 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 152, + 80, + 458, + 169 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 152, + 80, + 458, + 169 + ], + "spans": [ + { + "bbox": [ + 152, + 80, + 458, + 169 + ], + "score": 0.982, + "html": "
Dataset# layersHidden sizeLearning rateSampling method
ogbn-arxiv32560.01GraphSAINT
ogbn-products31280.001NeighborSampling
ogbn-mag22560.01NeighborSampling
ogbn-proteins6640.01NeighborSampling
ogbl-collab3640.001Full Batch
ogbl-citation232560.0005NeighborSampling
", + "type": "table", + "image_path": "99ab02b3e8b0afbf67c8ce47dd5a7df3c52f36e24045d253fded2e4adebe327b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 152, + 80, + 458, + 109.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 152, + 109.66666666666667, + 458, + 139.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 152, + 139.33333333333334, + 458, + 169.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 180, + 177, + 430, + 189 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 179, + 175, + 431, + 190 + ], + "spans": [ + { + "bbox": [ + 179, + 175, + 431, + 190 + ], + "score": 1.0, + "content": "Table 4: Training details of node- and link-prediction datasets.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "title", + "bbox": [ + 107, + 210, + 235, + 222 + ], + "lines": [ + { + "bbox": [ + 106, + 210, + 236, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 236, + 223 + ], + "score": 1.0, + "content": "D.2 ROBUSTNESS TO NOISE", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 231, + 504, + 254 + ], + "lines": [ + { + "bbox": [ + 106, + 231, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 506, + 244 + ], + "score": 1.0, + "content": "In these experiments, we used the same best-found hyperparameters in node-prediction, with 8", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 243, + 477, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 477, + 255 + ], + "score": 1.0, + "content": "attention heads in ogbn-arxiv and 1 head in ogbn-mag. Each point is an average of 10 runs.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 106, + 231, + 506, + 255 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 268, + 341, + 279 + ], + "lines": [ + { + "bbox": [ + 106, + 268, + 342, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 342, + 281 + ], + "score": 1.0, + "content": "D.3 SYNTHETIC BENCHMARK: DICTIONARYLOOKUP", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 105, + 289, + 504, + 312 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 405, + 303 + ], + "score": 1.0, + "content": "In all experiments, we used a learning rate decay of 0.5, a hidden size of", + "type": "text" + }, + { + "bbox": [ + 405, + 290, + 440, + 300 + ], + "score": 0.89, + "content": "d = 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 288, + 506, + 303 + ], + "score": 1.0, + "content": ", a batch size of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 300, + 231, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 231, + 312 + ], + "score": 1.0, + "content": "1024, and the Adam optimizer.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 288, + 506, + 312 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 317, + 505, + 372 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 312, + 330 + ], + "score": 1.0, + "content": "We created a separate dataset for every graph size", + "type": "text" + }, + { + "bbox": [ + 313, + 318, + 325, + 329 + ], + "score": 0.63, + "content": "( k )", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 317, + 505, + 330 + ], + "score": 1.0, + "content": ", and we split each such dataset to train and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 329, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 340 + ], + "score": 1.0, + "content": "test with a ratio of 80:20. Since this is a contrived problem, we did not use a validation set, and the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "reported test results can be thought of as validation results. Every model was trained on a fixed value", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 349, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 117, + 363 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 117, + 351, + 124, + 360 + ], + "score": 0.79, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 349, + 505, + 363 + ], + "score": 1.0, + "content": ". Every key node (bottom row in Figure 2) was encoded as a sum of learned attribute embedding", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 361, + 282, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 282, + 374 + ], + "score": 1.0, + "content": "and a value embedding, followed by ReLU.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 317, + 505, + 374 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 378, + 504, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "We experimented with layer normalization, batch normalization, dropout, various activation functions", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 389, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 505, + 402 + ], + "score": 1.0, + "content": "and various learning rates. None of these changed the general trend, so the experiments in Figure 3", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 400, + 463, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 463, + 412 + ], + "score": 1.0, + "content": "were conducted without any normalization, without dropout and a learning rate of 0.001.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 106, + 378, + 505, + 412 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 425, + 241, + 437 + ], + "lines": [ + { + "bbox": [ + 105, + 424, + 243, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 243, + 438 + ], + "score": 1.0, + "content": "D.4 PROGRAMS: VARMISUSE", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 108, + 447, + 504, + 480 + ], + "lines": [ + { + "bbox": [ + 106, + 447, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 505, + 459 + ], + "score": 1.0, + "content": "We used the code, splits, and the same best-found configurations as Brockschmidt (2020), who", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 457, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 506, + 471 + ], + "score": 1.0, + "content": "performed an extensive hyperparameter tuning by searching over 30 configurations for each GNN", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 469, + 267, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 267, + 480 + ], + "score": 1.0, + "content": "type. We trained each model five times.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 447, + 506, + 480 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 504, + 508 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 505, + 498 + ], + "score": 1.0, + "content": "We took the best-found hyperparameters of Brockschmidt (2020) for GAT and used them to train", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 496, + 250, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 250, + 510 + ], + "score": 1.0, + "content": "GATv2, without any further tuning.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 486, + 505, + 510 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 522, + 251, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 520, + 252, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 252, + 537 + ], + "score": 1.0, + "content": "D.5 GRAPH-PREDICTION: QM9", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 543, + 505, + 599 + ], + "lines": [ + { + "bbox": [ + 106, + 543, + 506, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 506, + 557 + ], + "score": 1.0, + "content": "We used the code and splits of Brockschmidt (2020) who performed an extensive hyperparameter", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 554, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 568 + ], + "score": 1.0, + "content": "search over 500 configurations. We took the best-found hyperparameters of Brockschmidt (2020)", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "score": 1.0, + "content": "for GAT and used them to train GATv2. The only minor change from GAT is placing a residual", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "score": 1.0, + "content": "connection after every layer, rather than after every other layer, which is within the experimented", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 587, + 372, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 372, + 600 + ], + "score": 1.0, + "content": "hyperparameter search that was reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 543, + 506, + 600 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 613, + 252, + 624 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 253, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 253, + 626 + ], + "score": 1.0, + "content": "D.6 COMPUTE AND RESOURCES", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 108, + 634, + 504, + 657 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 648 + ], + "score": 1.0, + "content": "Our experiments consumed approximately 100 days of GPU in total. We used cloud GPUs of type", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 646, + 365, + 657 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 365, + 657 + ], + "score": 1.0, + "content": "V100, and we used RTX 3080 and 3090 in local GPU machines.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 633, + 505, + 657 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 674, + 217, + 687 + ], + "lines": [ + { + "bbox": [ + 105, + 672, + 218, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 218, + 689 + ], + "score": 1.0, + "content": "E DATA STATISTICS", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "title", + "bbox": [ + 107, + 699, + 312, + 711 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 313, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 313, + 712 + ], + "score": 1.0, + "content": "E.1 NODE- AND LINK-PREDICTION DATASETS", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 104, + 720, + 467, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 468, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 468, + 732 + ], + "score": 1.0, + "content": "Statistics of the OGB datasets we used for node- and link-prediction are shown in Table 5.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 106, + 720, + 468, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 147, + 80, + 464, + 176 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 147, + 80, + 464, + 176 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 80, + 464, + 176 + ], + "spans": [ + { + "bbox": [ + 147, + 80, + 464, + 176 + ], + "score": 0.981, + "html": "
Dataset# nodes# edgesAvg. node degreeDiameter
ogbn-arxiv169,3431,166,24313.723
ogbn-mag1,939,74321,111,00721.76
ogbn-products2,449,02961,859,14050.527
ogbn-proteins132,53439,561,252597.09
ogbl-collab235,8681,285,4658.222
ogbl-citation22,927,96330,561,18720.721
", + "type": "table", + "image_path": "46e2ac78f7d9510797e6edc20553d155d54b390922301684be1a7d01d91f157e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 147, + 80, + 464, + 112.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 147, + 112.0, + 464, + 144.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 147, + 144.0, + 464, + 176.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 191, + 183, + 419, + 196 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 191, + 181, + 419, + 197 + ], + "spans": [ + { + "bbox": [ + 191, + 181, + 419, + 197 + ], + "score": 1.0, + "content": "Table 5: Statistics of the OGB datasets (Hu et al., 2020).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "title", + "bbox": [ + 107, + 219, + 156, + 231 + ], + "lines": [ + { + "bbox": [ + 104, + 218, + 158, + 234 + ], + "spans": [ + { + "bbox": [ + 104, + 218, + 158, + 234 + ], + "score": 1.0, + "content": "E.2 QM9", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "table", + "bbox": [ + 191, + 265, + 420, + 334 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 241, + 442, + 254 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 240, + 443, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 443, + 254 + ], + "score": 1.0, + "content": "Statistics of the QM9 dataset, as used in Brockschmidt (2020) are shown in Table 6.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "table_body", + "bbox": [ + 191, + 265, + 420, + 334 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 191, + 265, + 420, + 334 + ], + "spans": [ + { + "bbox": [ + 191, + 265, + 420, + 334 + ], + "score": 0.977, + "html": "
TrainingValidationTest
# examples110,46210,00010,000
# nodes -average18.0318.0618.09
# edges - average18.6518.6718.72
Diameter - average6.356.356.35
", + "type": "table", + "image_path": "e1a267d5274a5a15b6c2699809130678ba87975f4194a50a10aca2b3bd0a31f4.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 191, + 265, + 420, + 278.8 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 191, + 278.8, + 420, + 292.6 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 191, + 292.6, + 420, + 306.40000000000003 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 191, + 306.40000000000003, + 420, + 320.20000000000005 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 191, + 320.20000000000005, + 420, + 334.00000000000006 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 105, + 342, + 504, + 365 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "score": 1.0, + "content": "Table 6: Statistics of the QM9 chemical dataset (Ramakrishnan et al., 2014) as used by Brockschmidt", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 351, + 140, + 366 + ], + "spans": [ + { + "bbox": [ + 104, + 351, + 140, + 366 + ], + "score": 1.0, + "content": "(2020).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 107, + 397, + 186, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 187, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 187, + 410 + ], + "score": 1.0, + "content": "E.3 VARMISUSE", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "table", + "bbox": [ + 127, + 452, + 484, + 522 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 419, + 506, + 442 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 419, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 419, + 506, + 432 + ], + "score": 1.0, + "content": "Statistics of the VARMISUSE dataset, as used in Allamanis et al. (2018) and Brockschmidt (2020),", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 430, + 194, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 194, + 441 + ], + "score": 1.0, + "content": "are shown in Table 7.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "table_body", + "bbox": [ + 127, + 452, + 484, + 522 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 127, + 452, + 484, + 522 + ], + "spans": [ + { + "bbox": [ + 127, + 452, + 484, + 522 + ], + "score": 0.738, + "html": "
TrainingValidationUnseenProject TestSeenProject Test
# graphs2543604265411703659974
# nodes - average2377174219593986
# edges - average72987851588212925
Diameter-average7.887.887.787.82
", + "type": "table", + "image_path": "142de408d3dcc17abb31e482a2cedb6ea40aceb0d14995aec3ec00da71985a3a.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 127, + 452, + 484, + 475.3333333333333 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 127, + 475.3333333333333, + 484, + 498.66666666666663 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 127, + 498.66666666666663, + 484, + 522.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 529, + 504, + 552 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "score": 1.0, + "content": "Table 7: Statistics of the VARMISUSE dataset (Allamanis et al., 2018) as used by Brockschmidt", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 538, + 140, + 554 + ], + "spans": [ + { + "bbox": [ + 104, + 538, + 140, + 554 + ], + "score": 1.0, + "content": "(2020).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 107, + 588, + 243, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 245, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 245, + 603 + ], + "score": 1.0, + "content": "F ADDITIONAL RESULTS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "title", + "bbox": [ + 107, + 615, + 223, + 627 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 224, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 224, + 628 + ], + "score": 1.0, + "content": "F.1 DICTIONARYLOOKUP", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 506, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 638, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 506, + 649 + ], + "score": 1.0, + "content": "Figure 5 shows additional comparison between GATv2 and GIN (Xu et al., 2019) in the DICTIO-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 303, + 661 + ], + "score": 1.0, + "content": "NARYLOOKUP problem. GATv2 easily achieves", + "type": "text" + }, + { + "bbox": [ + 304, + 649, + 329, + 659 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 649, + 457, + 661 + ], + "score": 1.0, + "content": "train and test accuracy even for", + "type": "text" + }, + { + "bbox": [ + 457, + 649, + 487, + 659 + ], + "score": 0.88, + "content": "k { = } 1 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "using only a single head. GIN, although considered as more expressive than other GNNs, cannot", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 671, + 430, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 309, + 683 + ], + "score": 1.0, + "content": "perfectly fit the training data (with a model size of", + "type": "text" + }, + { + "bbox": [ + 309, + 671, + 344, + 681 + ], + "score": 0.89, + "content": "d = 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 671, + 402, + 683 + ], + "score": 1.0, + "content": ") starting from", + "type": "text" + }, + { + "bbox": [ + 402, + 671, + 427, + 681 + ], + "score": 0.87, + "content": "k { = } 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 671, + 430, + 683 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 107, + 698, + 211, + 710 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 212, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 212, + 711 + ], + "score": 1.0, + "content": "F.2 LINK-PREDICTION", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 305, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 306, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 306, + 732 + ], + "score": 1.0, + "content": "Results for link-prediction are showen in Table 8.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 147, + 80, + 464, + 176 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 147, + 80, + 464, + 176 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 80, + 464, + 176 + ], + "spans": [ + { + "bbox": [ + 147, + 80, + 464, + 176 + ], + "score": 0.981, + "html": "
Dataset# nodes# edgesAvg. node degreeDiameter
ogbn-arxiv169,3431,166,24313.723
ogbn-mag1,939,74321,111,00721.76
ogbn-products2,449,02961,859,14050.527
ogbn-proteins132,53439,561,252597.09
ogbl-collab235,8681,285,4658.222
ogbl-citation22,927,96330,561,18720.721
", + "type": "table", + "image_path": "46e2ac78f7d9510797e6edc20553d155d54b390922301684be1a7d01d91f157e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 147, + 80, + 464, + 112.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 147, + 112.0, + 464, + 144.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 147, + 144.0, + 464, + 176.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 191, + 183, + 419, + 196 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 191, + 181, + 419, + 197 + ], + "spans": [ + { + "bbox": [ + 191, + 181, + 419, + 197 + ], + "score": 1.0, + "content": "Table 5: Statistics of the OGB datasets (Hu et al., 2020).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "title", + "bbox": [ + 107, + 219, + 156, + 231 + ], + "lines": [ + { + "bbox": [ + 104, + 218, + 158, + 234 + ], + "spans": [ + { + "bbox": [ + 104, + 218, + 158, + 234 + ], + "score": 1.0, + "content": "E.2 QM9", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "table", + "bbox": [ + 191, + 265, + 420, + 334 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 241, + 442, + 254 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 240, + 443, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 443, + 254 + ], + "score": 1.0, + "content": "Statistics of the QM9 dataset, as used in Brockschmidt (2020) are shown in Table 6.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "table_body", + "bbox": [ + 191, + 265, + 420, + 334 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 191, + 265, + 420, + 334 + ], + "spans": [ + { + "bbox": [ + 191, + 265, + 420, + 334 + ], + "score": 0.977, + "html": "
TrainingValidationTest
# examples110,46210,00010,000
# nodes -average18.0318.0618.09
# edges - average18.6518.6718.72
Diameter - average6.356.356.35
", + "type": "table", + "image_path": "e1a267d5274a5a15b6c2699809130678ba87975f4194a50a10aca2b3bd0a31f4.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 191, + 265, + 420, + 278.8 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 191, + 278.8, + 420, + 292.6 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 191, + 292.6, + 420, + 306.40000000000003 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 191, + 306.40000000000003, + 420, + 320.20000000000005 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 191, + 320.20000000000005, + 420, + 334.00000000000006 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 105, + 342, + 504, + 365 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "score": 1.0, + "content": "Table 6: Statistics of the QM9 chemical dataset (Ramakrishnan et al., 2014) as used by Brockschmidt", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 351, + 140, + 366 + ], + "spans": [ + { + "bbox": [ + 104, + 351, + 140, + 366 + ], + "score": 1.0, + "content": "(2020).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 107, + 397, + 186, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 187, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 187, + 410 + ], + "score": 1.0, + "content": "E.3 VARMISUSE", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "table", + "bbox": [ + 127, + 452, + 484, + 522 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 419, + 506, + 442 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 419, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 419, + 506, + 432 + ], + "score": 1.0, + "content": "Statistics of the VARMISUSE dataset, as used in Allamanis et al. (2018) and Brockschmidt (2020),", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 430, + 194, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 194, + 441 + ], + "score": 1.0, + "content": "are shown in Table 7.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "table_body", + "bbox": [ + 127, + 452, + 484, + 522 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 127, + 452, + 484, + 522 + ], + "spans": [ + { + "bbox": [ + 127, + 452, + 484, + 522 + ], + "score": 0.738, + "html": "
TrainingValidationUnseenProject TestSeenProject Test
# graphs2543604265411703659974
# nodes - average2377174219593986
# edges - average72987851588212925
Diameter-average7.887.887.787.82
", + "type": "table", + "image_path": "142de408d3dcc17abb31e482a2cedb6ea40aceb0d14995aec3ec00da71985a3a.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 127, + 452, + 484, + 475.3333333333333 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 127, + 475.3333333333333, + 484, + 498.66666666666663 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 127, + 498.66666666666663, + 484, + 522.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 529, + 504, + 552 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "score": 1.0, + "content": "Table 7: Statistics of the VARMISUSE dataset (Allamanis et al., 2018) as used by Brockschmidt", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 538, + 140, + 554 + ], + "spans": [ + { + "bbox": [ + 104, + 538, + 140, + 554 + ], + "score": 1.0, + "content": "(2020).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 107, + 588, + 243, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 245, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 245, + 603 + ], + "score": 1.0, + "content": "F ADDITIONAL RESULTS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "title", + "bbox": [ + 107, + 615, + 223, + 627 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 224, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 224, + 628 + ], + "score": 1.0, + "content": "F.1 DICTIONARYLOOKUP", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 506, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 638, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 506, + 649 + ], + "score": 1.0, + "content": "Figure 5 shows additional comparison between GATv2 and GIN (Xu et al., 2019) in the DICTIO-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 303, + 661 + ], + "score": 1.0, + "content": "NARYLOOKUP problem. GATv2 easily achieves", + "type": "text" + }, + { + "bbox": [ + 304, + 649, + 329, + 659 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 649, + 457, + 661 + ], + "score": 1.0, + "content": "train and test accuracy even for", + "type": "text" + }, + { + "bbox": [ + 457, + 649, + 487, + 659 + ], + "score": 0.88, + "content": "k { = } 1 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "using only a single head. GIN, although considered as more expressive than other GNNs, cannot", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 671, + 430, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 309, + 683 + ], + "score": 1.0, + "content": "perfectly fit the training data (with a model size of", + "type": "text" + }, + { + "bbox": [ + 309, + 671, + 344, + 681 + ], + "score": 0.89, + "content": "d = 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 671, + 402, + 683 + ], + "score": 1.0, + "content": ") starting from", + "type": "text" + }, + { + "bbox": [ + 402, + 671, + 427, + 681 + ], + "score": 0.87, + "content": "k { = } 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 671, + 430, + 683 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 638, + 506, + 683 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 698, + 211, + 710 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 212, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 212, + 711 + ], + "score": 1.0, + "content": "F.2 LINK-PREDICTION", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 305, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 306, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 306, + 732 + ], + "score": 1.0, + "content": "Results for link-prediction are showen in Table 8.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 106, + 720, + 306, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 123, + 80, + 492, + 215 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 123, + 80, + 492, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 80, + 492, + 215 + ], + "spans": [ + { + "bbox": [ + 123, + 80, + 492, + 215 + ], + "score": 0.971, + "type": "image", + "image_path": "4233ec4a02df2921f13b438873a45c7450236e836cc22f098f64cf61a279e12d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 123, + 80, + 492, + 125.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 123, + 125.0, + 492, + 170.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 123, + 170.0, + 492, + 215.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 226, + 505, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 226, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 506, + 239 + ], + "score": 1.0, + "content": "Figure 5: Train and test accuracy across graph sizes in the DICTIONARYLOOKUP problem. GATv2", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 171, + 249 + ], + "score": 1.0, + "content": "easily achieves", + "type": "text" + }, + { + "bbox": [ + 171, + 237, + 196, + 248 + ], + "score": 0.88, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 237, + 329, + 249 + ], + "score": 1.0, + "content": "train and test accuracy even for", + "type": "text" + }, + { + "bbox": [ + 329, + 237, + 359, + 248 + ], + "score": 0.9, + "content": "k { = } 1 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "and using only a single head. GIN", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 249, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 109, + 260 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 109, + 249, + 124, + 259 + ], + "score": 0.38, + "content": "\\mathrm { { X u } }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 249, + 505, + 260 + ], + "score": 1.0, + "content": "et al., 2019), although considered as more expressive than other GNNs, cannot perfectly fit the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 259, + 367, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 245, + 272 + ], + "score": 1.0, + "content": "training data (with a model size of", + "type": "text" + }, + { + "bbox": [ + 246, + 259, + 281, + 270 + ], + "score": 0.88, + "content": "d = 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 259, + 339, + 272 + ], + "score": 1.0, + "content": ") starting from", + "type": "text" + }, + { + "bbox": [ + 339, + 259, + 363, + 270 + ], + "score": 0.88, + "content": "k { = } 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 259, + 367, + 272 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "table", + "bbox": [ + 395, + 346, + 462, + 447 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 287, + 505, + 321 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 194, + 300 + ], + "score": 1.0, + "content": "Table 8: Average Hits", + "type": "text" + }, + { + "bbox": [ + 194, + 288, + 213, + 298 + ], + "score": 0.66, + "content": "@ 5 0", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 287, + 505, + 300 + ], + "score": 1.0, + "content": "(Table 8a) and mean reciprocal rank (MRR) (Table 8b) in link-prediction", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 298, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 506, + 311 + ], + "score": 1.0, + "content": "benchmarks from OGB (10 runs±std). The best result among GAT and GATv2 is marked in bold;", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 309, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 367, + 322 + ], + "score": 1.0, + "content": "the best result among all GNNs is marked in bold and underline.", + "type": "text" + }, + { + "bbox": [ + 367, + 310, + 374, + 321 + ], + "score": 0.75, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 309, + 505, + 322 + ], + "score": 1.0, + "content": "was reported by Hu et al. (2020).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "table_body", + "bbox": [ + 395, + 346, + 462, + 447 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 395, + 346, + 462, + 447 + ], + "spans": [ + { + "bbox": [ + 395, + 346, + 462, + 447 + ], + "score": 0.722, + "html": "
ogbl-citation2
80.04±0.25
80.44±0.10
79.84±0.19
75.95±1.3180.33±0.13
80.14±0.71
", + "type": "table", + "image_path": "019284bd53268ef6e683724418d88ea9e9b95e3d7fc9ef3b6e444930e460100d.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 395, + 346, + 462, + 447 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 419, + 331, + 432, + 341 + ], + "lines": [ + { + "bbox": [ + 417, + 330, + 433, + 343 + ], + "spans": [ + { + "bbox": [ + 417, + 330, + 433, + 343 + ], + "score": 1.0, + "content": "(b)", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 107, + 473, + 154, + 484 + ], + "lines": [ + { + "bbox": [ + 104, + 470, + 157, + 488 + ], + "spans": [ + { + "bbox": [ + 104, + 470, + 157, + 488 + ], + "score": 1.0, + "content": "F.3 QM9", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 422, + 508 + ], + "lines": [ + { + "bbox": [ + 106, + 496, + 423, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 423, + 509 + ], + "score": 1.0, + "content": "Standard deviation for the QM9 results of Section 4.5 are presented in Table 9.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "title", + "bbox": [ + 108, + 527, + 259, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 527, + 261, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 261, + 540 + ], + "score": 1.0, + "content": "F.4 PUBMED CITATION NETWORK", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 550, + 505, + 595 + ], + "lines": [ + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 427, + 563 + ], + "score": 1.0, + "content": "We tuned the following parameters for both GAT and GATv2: number of layers", + "type": "text" + }, + { + "bbox": [ + 427, + 550, + 471, + 562 + ], + "score": 0.91, + "content": "\\in \\{ 0 , 1 , 2 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 549, + 505, + 563 + ], + "score": 1.0, + "content": ", hidden", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 125, + 574 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 126, + 561, + 180, + 573 + ], + "score": 0.83, + "content": "\\in \\{ 8 , 1 6 , 3 2 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 560, + 254, + 574 + ], + "score": 1.0, + "content": ", number of heads", + "type": "text" + }, + { + "bbox": [ + 255, + 561, + 299, + 573 + ], + "score": 0.92, + "content": "\\in \\{ 1 , 4 , 8 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 560, + 338, + 574 + ], + "score": 1.0, + "content": ", dropout", + "type": "text" + }, + { + "bbox": [ + 338, + 561, + 406, + 573 + ], + "score": 0.87, + "content": "\\in \\{ 0 . 4 , 0 . 6 , 0 . 8 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 560, + 428, + 574 + ], + "score": 1.0, + "content": ", bias", + "type": "text" + }, + { + "bbox": [ + 429, + 562, + 502, + 573 + ], + "score": 0.78, + "content": "\\in \\{ T r u e , F a l s e \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 560, + 506, + 574 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 572, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 165, + 586 + ], + "score": 1.0, + "content": "share weights", + "type": "text" + }, + { + "bbox": [ + 165, + 573, + 237, + 585 + ], + "score": 0.9, + "content": "\\in \\{ T r u e , F a l s e \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 572, + 292, + 586 + ], + "score": 1.0, + "content": ", use residual", + "type": "text" + }, + { + "bbox": [ + 292, + 573, + 365, + 585 + ], + "score": 0.88, + "content": "\\dot { \\in } \\{ T r u e , F a l \\dot { s } e \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 572, + 506, + 586 + ], + "score": 1.0, + "content": ". Table 10 shows the test accuracy", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 583, + 395, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 395, + 595 + ], + "score": 1.0, + "content": "(100 runs±stdev) using the best hyperparameters found for each model.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 105, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "It is important to note that PubMed has only 60 training nodes, which hinders expressive models", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 612, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 505, + 623 + ], + "score": 1.0, + "content": "such as GATv2 from exploiting their approximation and generalization advantages. Still, GATv2", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 623, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 505, + 633 + ], + "score": 1.0, + "content": "is more accurate than GAT even in this small dataset. In Table 15, we show that this difference is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 633, + 280, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 234, + 645 + ], + "score": 1.0, + "content": "statistically significant (p-value", + "type": "text" + }, + { + "bbox": [ + 234, + 633, + 274, + 644 + ], + "score": 0.81, + "content": "< 0 . 0 0 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 633, + 280, + 645 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20.5 + }, + { + "type": "title", + "bbox": [ + 106, + 666, + 480, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 482, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 482, + 681 + ], + "score": 1.0, + "content": "G ADDITIONAL COMPARISON WITH TRANSFORMER-STYLE ATTENTION", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 126, + 680, + 181, + 696 + ], + "spans": [ + { + "bbox": [ + 126, + 680, + 181, + 696 + ], + "score": 1.0, + "content": "(DPGAT)", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 505, + 721 + ], + "score": 1.0, + "content": "The main goal of our paper is to highlight a severe theoretical limitation of the highly popular GAT", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 720, + 268, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 268, + 733 + ], + "score": 1.0, + "content": "architecture, and propose a minimal fix.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 123, + 80, + 492, + 215 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 123, + 80, + 492, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 80, + 492, + 215 + ], + "spans": [ + { + "bbox": [ + 123, + 80, + 492, + 215 + ], + "score": 0.971, + "type": "image", + "image_path": "4233ec4a02df2921f13b438873a45c7450236e836cc22f098f64cf61a279e12d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 123, + 80, + 492, + 125.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 123, + 125.0, + 492, + 170.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 123, + 170.0, + 492, + 215.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 226, + 505, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 226, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 506, + 239 + ], + "score": 1.0, + "content": "Figure 5: Train and test accuracy across graph sizes in the DICTIONARYLOOKUP problem. GATv2", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 171, + 249 + ], + "score": 1.0, + "content": "easily achieves", + "type": "text" + }, + { + "bbox": [ + 171, + 237, + 196, + 248 + ], + "score": 0.88, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 237, + 329, + 249 + ], + "score": 1.0, + "content": "train and test accuracy even for", + "type": "text" + }, + { + "bbox": [ + 329, + 237, + 359, + 248 + ], + "score": 0.9, + "content": "k { = } 1 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "and using only a single head. GIN", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 249, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 109, + 260 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 109, + 249, + 124, + 259 + ], + "score": 0.38, + "content": "\\mathrm { { X u } }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 249, + 505, + 260 + ], + "score": 1.0, + "content": "et al., 2019), although considered as more expressive than other GNNs, cannot perfectly fit the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 259, + 367, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 245, + 272 + ], + "score": 1.0, + "content": "training data (with a model size of", + "type": "text" + }, + { + "bbox": [ + 246, + 259, + 281, + 270 + ], + "score": 0.88, + "content": "d = 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 259, + 339, + 272 + ], + "score": 1.0, + "content": ") starting from", + "type": "text" + }, + { + "bbox": [ + 339, + 259, + 363, + 270 + ], + "score": 0.88, + "content": "k { = } 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 259, + 367, + 272 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "table", + "bbox": [ + 395, + 346, + 462, + 447 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 287, + 505, + 321 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 194, + 300 + ], + "score": 1.0, + "content": "Table 8: Average Hits", + "type": "text" + }, + { + "bbox": [ + 194, + 288, + 213, + 298 + ], + "score": 0.66, + "content": "@ 5 0", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 287, + 505, + 300 + ], + "score": 1.0, + "content": "(Table 8a) and mean reciprocal rank (MRR) (Table 8b) in link-prediction", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 298, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 506, + 311 + ], + "score": 1.0, + "content": "benchmarks from OGB (10 runs±std). The best result among GAT and GATv2 is marked in bold;", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 309, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 367, + 322 + ], + "score": 1.0, + "content": "the best result among all GNNs is marked in bold and underline.", + "type": "text" + }, + { + "bbox": [ + 367, + 310, + 374, + 321 + ], + "score": 0.75, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 309, + 505, + 322 + ], + "score": 1.0, + "content": "was reported by Hu et al. (2020).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "table_body", + "bbox": [ + 395, + 346, + 462, + 447 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 395, + 346, + 462, + 447 + ], + "spans": [ + { + "bbox": [ + 395, + 346, + 462, + 447 + ], + "score": 0.722, + "html": "
ogbl-citation2
80.04±0.25
80.44±0.10
79.84±0.19
75.95±1.3180.33±0.13
80.14±0.71
", + "type": "table", + "image_path": "019284bd53268ef6e683724418d88ea9e9b95e3d7fc9ef3b6e444930e460100d.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 395, + 346, + 462, + 447 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 419, + 331, + 432, + 341 + ], + "lines": [ + { + "bbox": [ + 417, + 330, + 433, + 343 + ], + "spans": [ + { + "bbox": [ + 417, + 330, + 433, + 343 + ], + "score": 1.0, + "content": "(b)", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 417, + 330, + 433, + 343 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 473, + 154, + 484 + ], + "lines": [ + { + "bbox": [ + 104, + 470, + 157, + 488 + ], + "spans": [ + { + "bbox": [ + 104, + 470, + 157, + 488 + ], + "score": 1.0, + "content": "F.3 QM9", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 422, + 508 + ], + "lines": [ + { + "bbox": [ + 106, + 496, + 423, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 423, + 509 + ], + "score": 1.0, + "content": "Standard deviation for the QM9 results of Section 4.5 are presented in Table 9.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 106, + 496, + 423, + 509 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 527, + 259, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 527, + 261, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 261, + 540 + ], + "score": 1.0, + "content": "F.4 PUBMED CITATION NETWORK", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 550, + 505, + 595 + ], + "lines": [ + { + "bbox": [ + 105, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 427, + 563 + ], + "score": 1.0, + "content": "We tuned the following parameters for both GAT and GATv2: number of layers", + "type": "text" + }, + { + "bbox": [ + 427, + 550, + 471, + 562 + ], + "score": 0.91, + "content": "\\in \\{ 0 , 1 , 2 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 549, + 505, + 563 + ], + "score": 1.0, + "content": ", hidden", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 125, + 574 + ], + "score": 1.0, + "content": "size", + "type": "text" + }, + { + "bbox": [ + 126, + 561, + 180, + 573 + ], + "score": 0.83, + "content": "\\in \\{ 8 , 1 6 , 3 2 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 560, + 254, + 574 + ], + "score": 1.0, + "content": ", number of heads", + "type": "text" + }, + { + "bbox": [ + 255, + 561, + 299, + 573 + ], + "score": 0.92, + "content": "\\in \\{ 1 , 4 , 8 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 560, + 338, + 574 + ], + "score": 1.0, + "content": ", dropout", + "type": "text" + }, + { + "bbox": [ + 338, + 561, + 406, + 573 + ], + "score": 0.87, + "content": "\\in \\{ 0 . 4 , 0 . 6 , 0 . 8 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 560, + 428, + 574 + ], + "score": 1.0, + "content": ", bias", + "type": "text" + }, + { + "bbox": [ + 429, + 562, + 502, + 573 + ], + "score": 0.78, + "content": "\\in \\{ T r u e , F a l s e \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 560, + 506, + 574 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 572, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 165, + 586 + ], + "score": 1.0, + "content": "share weights", + "type": "text" + }, + { + "bbox": [ + 165, + 573, + 237, + 585 + ], + "score": 0.9, + "content": "\\in \\{ T r u e , F a l s e \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 572, + 292, + 586 + ], + "score": 1.0, + "content": ", use residual", + "type": "text" + }, + { + "bbox": [ + 292, + 573, + 365, + 585 + ], + "score": 0.88, + "content": "\\dot { \\in } \\{ T r u e , F a l \\dot { s } e \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 572, + 506, + 586 + ], + "score": 1.0, + "content": ". Table 10 shows the test accuracy", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 583, + 395, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 395, + 595 + ], + "score": 1.0, + "content": "(100 runs±stdev) using the best hyperparameters found for each model.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 549, + 506, + 595 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 105, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "It is important to note that PubMed has only 60 training nodes, which hinders expressive models", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 612, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 505, + 623 + ], + "score": 1.0, + "content": "such as GATv2 from exploiting their approximation and generalization advantages. Still, GATv2", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 623, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 505, + 633 + ], + "score": 1.0, + "content": "is more accurate than GAT even in this small dataset. In Table 15, we show that this difference is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 633, + 280, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 234, + 645 + ], + "score": 1.0, + "content": "statistically significant (p-value", + "type": "text" + }, + { + "bbox": [ + 234, + 633, + 274, + 644 + ], + "score": 0.81, + "content": "< 0 . 0 0 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 633, + 280, + 645 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 600, + 505, + 645 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 666, + 480, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 482, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 482, + 681 + ], + "score": 1.0, + "content": "G ADDITIONAL COMPARISON WITH TRANSFORMER-STYLE ATTENTION", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 126, + 680, + 181, + 696 + ], + "spans": [ + { + "bbox": [ + 126, + 680, + 181, + 696 + ], + "score": 1.0, + "content": "(DPGAT)", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 505, + 721 + ], + "score": 1.0, + "content": "The main goal of our paper is to highlight a severe theoretical limitation of the highly popular GAT", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 720, + 268, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 268, + 733 + ], + "score": 1.0, + "content": "architecture, and propose a minimal fix.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 708, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 117, + 80, + 493, + 305 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 117, + 80, + 493, + 305 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 117, + 80, + 493, + 305 + ], + "spans": [ + { + "bbox": [ + 117, + 80, + 493, + 305 + ], + "score": 0.939, + "html": "
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.25 18.71±23.369.05±1.21 5.62±0.817.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h10.80±2.183.11±0.14
GAT8h28.10±16.45 7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.365.37±0.26 4.11±0.271.48±0.87+134.1% +0%
GATv21h22.94±17.34
10.19±0.6322.56±17.4615.04±4.585.23±0.362.46±0.65+91.6%
GATv28h6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
", + "type": "table", + "image_path": "7fd6def9d43f294d7cc625ec3140917897cf6fb89794df258cd7baf6db6c4453.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 117, + 80, + 493, + 155.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 117, + 155.0, + 493, + 230.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 117, + 230.0, + 493, + 305.0 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 313, + 505, + 357 + ], + "lines": [ + { + "bbox": [ + 105, + 312, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 326, + 326 + ], + "score": 1.0, + "content": "Table 9: Average error rates (lower is better), 5 runs", + "type": "text" + }, + { + "bbox": [ + 326, + 314, + 337, + 324 + ], + "score": 0.74, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 312, + 505, + 326 + ], + "score": 1.0, + "content": "standard deviation for each property, on", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 324, + 506, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 506, + 337 + ], + "score": 1.0, + "content": "the QM9 dataset. The best result among GAT and GATv2 is marked in bold; the globally best", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 343, + 348 + ], + "score": 1.0, + "content": "result among all GNNs is marked in bold and underline.", + "type": "text" + }, + { + "bbox": [ + 343, + 336, + 349, + 347 + ], + "score": 0.74, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "was previously tuned and reported by", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 346, + 196, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 196, + 358 + ], + "score": 1.0, + "content": "Brockschmidt (2020).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "table", + "bbox": [ + 268, + 390, + 343, + 434 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 131, + 369, + 478, + 382 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 131, + 369, + 479, + 382 + ], + "spans": [ + { + "bbox": [ + 131, + 369, + 479, + 382 + ], + "score": 1.0, + "content": "Table 10: Accuracy (100 runs±stdev) on Pubmed. GATv2 is more accurate than GAT.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "table_body", + "bbox": [ + 268, + 390, + 343, + 434 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 268, + 390, + 343, + 434 + ], + "spans": [ + { + "bbox": [ + 268, + 390, + 343, + 434 + ], + "score": 0.959, + "html": "
ModelAccuracy
GAT78.1±0.59
GATv278.5±0.38
", + "type": "table", + "image_path": "0b68db02c3f4936969ed4e3c1f6cc14ec2a34b6bc4e101b45bc1a8e2cd4f07cb.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 268, + 390, + 343, + 412.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 268, + 412.0, + 343, + 434.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 7.75 + }, + { + "type": "text", + "bbox": [ + 107, + 455, + 506, + 477 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 505, + 468 + ], + "score": 1.0, + "content": "We perform additional empirical comparison to DPGAT, which follows Luong et al. (2015) and the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 465, + 453, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 453, + 478 + ], + "score": 1.0, + "content": "dot-product attention of the Transformer (Vaswani et al., 2017). We define DPGAT as:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "interline_equation", + "bbox": [ + 288, + 483, + 462, + 505 + ], + "lines": [ + { + "bbox": [ + 288, + 483, + 462, + 505 + ], + "spans": [ + { + "bbox": [ + 288, + 483, + 462, + 505 + ], + "score": 0.89, + "content": "e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right) = \\left( \\left( \\pmb { h } _ { i } ^ { \\top } \\pmb { Q } \\right) \\cdot \\left( \\pmb { h } _ { j } ^ { \\top } \\pmb { K } \\right) ^ { \\top } \\right) / \\sqrt { d _ { k } }", + "type": "interline_equation", + "image_path": "5b7116695b9f012e48806294910738c39d8fe1b477acc79bc36458ea30a52df0.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 288, + 483, + 462, + 505 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 510, + 505, + 543 + ], + "lines": [ + { + "bbox": [ + 105, + 508, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 505, + 523 + ], + "score": 1.0, + "content": "Variants of DPGAT were used in prior work (Gao and Ji, 2019; Dwivedi and Bresson, 2020; Rong", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "et al., 2020a; Velickovi ˇ c et al., 2020; Kim and Oh, 2021), and we consider it here for the conceptual ´", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 532, + 256, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 256, + 544 + ], + "score": 1.0, + "content": "and empirical comparison with GAT.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 549, + 505, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 505, + 562 + ], + "score": 1.0, + "content": "Despite its popularity, DPGAT is strictly weaker than GATv2. DPGAT provably performs dynamic", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "attention for any set of node representations only if they are linearly independent (see Theorem 3", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "score": 1.0, + "content": "and its proof in Appendix G.1). Otherwise, there are examples of node representations that are", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 245, + 595 + ], + "score": 1.0, + "content": "linearly dependent and mappings", + "type": "text" + }, + { + "bbox": [ + 246, + 584, + 254, + 594 + ], + "score": 0.76, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 581, + 506, + 595 + ], + "score": 1.0, + "content": ", for which dynamic attention does not hold (Appendix G.2).", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 592, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 506, + 605 + ], + "score": 1.0, + "content": "This constraint is not harmful when violated in practice, because every node has only a small set of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "neighbors, rather than all possible nodes in the graph; further, some nodes possibly never need to be", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 614, + 195, + 628 + ], + "spans": [ + { + "bbox": [ + 104, + 614, + 195, + 628 + ], + "score": 1.0, + "content": "“selected” in practice.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 640, + 501, + 663 + ], + "lines": [ + { + "bbox": [ + 106, + 640, + 502, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 502, + 653 + ], + "score": 1.0, + "content": "G.1 PROOF THAT DPGAT PERFORMS DYNAMIC ATTENTION FOR LINEARLY INDEPENDENT", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 131, + 651, + 244, + 664 + ], + "spans": [ + { + "bbox": [ + 131, + 651, + 244, + 664 + ], + "score": 1.0, + "content": "NODE REPRESENTATIONS", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 672, + 504, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 671, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 686 + ], + "score": 1.0, + "content": "Theorem 3. A DPGAT layer computes dynamic attention for any set of node representations", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 682, + 321, + 697 + ], + "spans": [ + { + "bbox": [ + 107, + 683, + 200, + 695 + ], + "score": 0.93, + "content": "\\mathbb { K } = \\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 682, + 321, + 697 + ], + "score": 1.0, + "content": "that are linearly independent.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 733 + ], + "lines": [ + { + "bbox": [ + 105, + 707, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 152, + 723 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 153, + 710, + 200, + 721 + ], + "score": 0.92, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 707, + 505, + 723 + ], + "score": 1.0, + "content": "be a graph modeled by a DPGAT layer, having linearly independent node", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 720, + 429, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 169, + 734 + ], + "score": 1.0, + "content": "representations", + "type": "text" + }, + { + "bbox": [ + 170, + 721, + 221, + 732 + ], + "score": 0.93, + "content": "\\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 720, + 241, + 734 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 241, + 720, + 295, + 732 + ], + "score": 0.93, + "content": "\\varphi : [ n ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 720, + 385, + 734 + ], + "score": 1.0, + "content": "be any node mapping", + "type": "text" + }, + { + "bbox": [ + 385, + 720, + 424, + 732 + ], + "score": 0.92, + "content": "[ { \\bar { n } } ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 720, + 429, + 734 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 117, + 80, + 493, + 305 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 117, + 80, + 493, + 305 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 117, + 80, + 493, + 305 + ], + "spans": [ + { + "bbox": [ + 117, + 80, + 493, + 305 + ], + "score": 0.939, + "html": "
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.25 18.71±23.369.05±1.21 5.62±0.817.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h10.80±2.183.11±0.14
GAT8h28.10±16.45 7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.365.37±0.26 4.11±0.271.48±0.87+134.1% +0%
GATv21h22.94±17.34
10.19±0.6322.56±17.4615.04±4.585.23±0.362.46±0.65+91.6%
GATv28h6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
", + "type": "table", + "image_path": "7fd6def9d43f294d7cc625ec3140917897cf6fb89794df258cd7baf6db6c4453.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 117, + 80, + 493, + 155.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 117, + 155.0, + 493, + 230.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 117, + 230.0, + 493, + 305.0 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 313, + 505, + 357 + ], + "lines": [ + { + "bbox": [ + 105, + 312, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 326, + 326 + ], + "score": 1.0, + "content": "Table 9: Average error rates (lower is better), 5 runs", + "type": "text" + }, + { + "bbox": [ + 326, + 314, + 337, + 324 + ], + "score": 0.74, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 312, + 505, + 326 + ], + "score": 1.0, + "content": "standard deviation for each property, on", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 324, + 506, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 506, + 337 + ], + "score": 1.0, + "content": "the QM9 dataset. The best result among GAT and GATv2 is marked in bold; the globally best", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 343, + 348 + ], + "score": 1.0, + "content": "result among all GNNs is marked in bold and underline.", + "type": "text" + }, + { + "bbox": [ + 343, + 336, + 349, + 347 + ], + "score": 0.74, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "was previously tuned and reported by", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 346, + 196, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 196, + 358 + ], + "score": 1.0, + "content": "Brockschmidt (2020).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 312, + 506, + 358 + ] + }, + { + "type": "table", + "bbox": [ + 268, + 390, + 343, + 434 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 131, + 369, + 478, + 382 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 131, + 369, + 479, + 382 + ], + "spans": [ + { + "bbox": [ + 131, + 369, + 479, + 382 + ], + "score": 1.0, + "content": "Table 10: Accuracy (100 runs±stdev) on Pubmed. GATv2 is more accurate than GAT.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "table_body", + "bbox": [ + 268, + 390, + 343, + 434 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 268, + 390, + 343, + 434 + ], + "spans": [ + { + "bbox": [ + 268, + 390, + 343, + 434 + ], + "score": 0.959, + "html": "
ModelAccuracy
GAT78.1±0.59
GATv278.5±0.38
", + "type": "table", + "image_path": "0b68db02c3f4936969ed4e3c1f6cc14ec2a34b6bc4e101b45bc1a8e2cd4f07cb.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 268, + 390, + 343, + 412.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 268, + 412.0, + 343, + 434.0 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 7.75 + }, + { + "type": "text", + "bbox": [ + 107, + 455, + 506, + 477 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 505, + 468 + ], + "score": 1.0, + "content": "We perform additional empirical comparison to DPGAT, which follows Luong et al. (2015) and the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 465, + 453, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 453, + 478 + ], + "score": 1.0, + "content": "dot-product attention of the Transformer (Vaswani et al., 2017). We define DPGAT as:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 106, + 455, + 505, + 478 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 288, + 483, + 462, + 505 + ], + "lines": [ + { + "bbox": [ + 288, + 483, + 462, + 505 + ], + "spans": [ + { + "bbox": [ + 288, + 483, + 462, + 505 + ], + "score": 0.89, + "content": "e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right) = \\left( \\left( \\pmb { h } _ { i } ^ { \\top } \\pmb { Q } \\right) \\cdot \\left( \\pmb { h } _ { j } ^ { \\top } \\pmb { K } \\right) ^ { \\top } \\right) / \\sqrt { d _ { k } }", + "type": "interline_equation", + "image_path": "5b7116695b9f012e48806294910738c39d8fe1b477acc79bc36458ea30a52df0.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 288, + 483, + 462, + 505 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 510, + 505, + 543 + ], + "lines": [ + { + "bbox": [ + 105, + 508, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 505, + 523 + ], + "score": 1.0, + "content": "Variants of DPGAT were used in prior work (Gao and Ji, 2019; Dwivedi and Bresson, 2020; Rong", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "et al., 2020a; Velickovi ˇ c et al., 2020; Kim and Oh, 2021), and we consider it here for the conceptual ´", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 532, + 256, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 256, + 544 + ], + "score": 1.0, + "content": "and empirical comparison with GAT.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 508, + 505, + 544 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 549, + 505, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 549, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 505, + 562 + ], + "score": 1.0, + "content": "Despite its popularity, DPGAT is strictly weaker than GATv2. DPGAT provably performs dynamic", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "attention for any set of node representations only if they are linearly independent (see Theorem 3", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 506, + 583 + ], + "score": 1.0, + "content": "and its proof in Appendix G.1). Otherwise, there are examples of node representations that are", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 245, + 595 + ], + "score": 1.0, + "content": "linearly dependent and mappings", + "type": "text" + }, + { + "bbox": [ + 246, + 584, + 254, + 594 + ], + "score": 0.76, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 581, + 506, + 595 + ], + "score": 1.0, + "content": ", for which dynamic attention does not hold (Appendix G.2).", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 592, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 506, + 605 + ], + "score": 1.0, + "content": "This constraint is not harmful when violated in practice, because every node has only a small set of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "neighbors, rather than all possible nodes in the graph; further, some nodes possibly never need to be", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 614, + 195, + 628 + ], + "spans": [ + { + "bbox": [ + 104, + 614, + 195, + 628 + ], + "score": 1.0, + "content": "“selected” in practice.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19, + "bbox_fs": [ + 104, + 549, + 506, + 628 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 640, + 501, + 663 + ], + "lines": [ + { + "bbox": [ + 106, + 640, + 502, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 502, + 653 + ], + "score": 1.0, + "content": "G.1 PROOF THAT DPGAT PERFORMS DYNAMIC ATTENTION FOR LINEARLY INDEPENDENT", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 131, + 651, + 244, + 664 + ], + "spans": [ + { + "bbox": [ + 131, + 651, + 244, + 664 + ], + "score": 1.0, + "content": "NODE REPRESENTATIONS", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 640, + 502, + 664 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 672, + 504, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 671, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 686 + ], + "score": 1.0, + "content": "Theorem 3. A DPGAT layer computes dynamic attention for any set of node representations", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 682, + 321, + 697 + ], + "spans": [ + { + "bbox": [ + 107, + 683, + 200, + 695 + ], + "score": 0.93, + "content": "\\mathbb { K } = \\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 682, + 321, + 697 + ], + "score": 1.0, + "content": "that are linearly independent.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 671, + 505, + 697 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 733 + ], + "lines": [ + { + "bbox": [ + 105, + 707, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 152, + 723 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 153, + 710, + 200, + 721 + ], + "score": 0.92, + "content": "\\mathcal { G } = ( \\nu , \\mathcal { E } )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 707, + 505, + 723 + ], + "score": 1.0, + "content": "be a graph modeled by a DPGAT layer, having linearly independent node", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 720, + 429, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 169, + 734 + ], + "score": 1.0, + "content": "representations", + "type": "text" + }, + { + "bbox": [ + 170, + 721, + 221, + 732 + ], + "score": 0.93, + "content": "\\{ h _ { 1 } , . . . , h _ { n } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 720, + 241, + 734 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 241, + 720, + 295, + 732 + ], + "score": 0.93, + "content": "\\varphi : [ n ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 720, + 385, + 734 + ], + "score": 1.0, + "content": "be any node mapping", + "type": "text" + }, + { + "bbox": [ + 385, + 720, + 424, + 732 + ], + "score": 0.92, + "content": "[ { \\bar { n } } ] [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 720, + 429, + 734 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 707, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 81, + 287, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 284, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 164, + 96 + ], + "score": 1.0, + "content": "We denote the", + "type": "text" + }, + { + "bbox": [ + 165, + 82, + 177, + 92 + ], + "score": 0.88, + "content": "\\mathrm { i } ^ { t h }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 79, + 243, + 96 + ], + "score": 1.0, + "content": "row of a matrix", + "type": "text" + }, + { + "bbox": [ + 243, + 83, + 256, + 93 + ], + "score": 0.8, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 79, + 268, + 96 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 269, + 83, + 284, + 93 + ], + "score": 0.88, + "content": "M _ { i }", + "type": "inline_equation" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 99, + 208, + 111 + ], + "lines": [ + { + "bbox": [ + 106, + 98, + 209, + 112 + ], + "spans": [ + { + "bbox": [ + 106, + 98, + 184, + 112 + ], + "score": 1.0, + "content": "We define a matrix", + "type": "text" + }, + { + "bbox": [ + 184, + 100, + 194, + 109 + ], + "score": 0.81, + "content": "_ { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 98, + 209, + 112 + ], + "score": 1.0, + "content": "as:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 108, + 351, + 136 + ], + "lines": [ + { + "bbox": [ + 258, + 108, + 351, + 136 + ], + "spans": [ + { + "bbox": [ + 258, + 108, + 351, + 136 + ], + "score": 0.93, + "content": "P _ { i , j } = { \\left\\{ \\begin{array} { l l } { 1 } & { j = \\varphi ( i ) } \\\\ { 0 } & { { \\mathrm { o t h e r w i s e } } } \\end{array} \\right. }", + "type": "interline_equation", + "image_path": "477f035d5e3de0a9ac649d2e748d973286069bc2d3286257c4f0cb84d9aab2f1.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 258, + 108, + 351, + 136 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 145, + 450, + 158 + ], + "lines": [ + { + "bbox": [ + 105, + 144, + 451, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 122, + 159 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 145, + 182, + 156 + ], + "score": 0.92, + "content": "\\ b { X } \\in \\mathbb { R } ^ { n } \\times \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 144, + 451, + 159 + ], + "score": 1.0, + "content": "be the matrix holding the graph’s node representations as its rows:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 260, + 177, + 352, + 231 + ], + "lines": [ + { + "bbox": [ + 260, + 177, + 352, + 231 + ], + "spans": [ + { + "bbox": [ + 260, + 177, + 352, + 231 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathbf { \\boldsymbol { X } } = \\left[ \\begin{array} { l l l } { - } & { h _ { 1 } } & { - } \\\\ { - } & { h _ { 2 } } & { - } \\\\ & { \\vdots } \\\\ { - } & { h _ { n } } & { - } \\end{array} \\right] } \\end{array}", + "type": "interline_equation", + "image_path": "a6beda280a46d1888de1e211f54408173573f54ec526a68249ae83b483c0ab7c.jpg" + } + ] + } + ], + "index": 4.5, + "virtual_lines": [ + { + "bbox": [ + 260, + 177, + 352, + 204.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 260, + 204.0, + 352, + 231.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 241, + 415, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 414, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 178, + 256 + ], + "score": 1.0, + "content": "Since the rows of", + "type": "text" + }, + { + "bbox": [ + 178, + 242, + 189, + 252 + ], + "score": 0.84, + "content": "\\boldsymbol { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 239, + 388, + 256 + ], + "score": 1.0, + "content": "are linearly independent, it necessarily holds that", + "type": "text" + }, + { + "bbox": [ + 389, + 242, + 414, + 253 + ], + "score": 0.91, + "content": "d \\geq n", + "type": "inline_equation" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 407, + 270 + ], + "lines": [ + { + "bbox": [ + 104, + 255, + 407, + 272 + ], + "spans": [ + { + "bbox": [ + 104, + 255, + 228, + 272 + ], + "score": 1.0, + "content": "Next, we find weight matrices", + "type": "text" + }, + { + "bbox": [ + 229, + 257, + 286, + 270 + ], + "score": 0.92, + "content": "Q \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 255, + 305, + 272 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 305, + 257, + 364, + 269 + ], + "score": 0.92, + "content": "K \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 255, + 407, + 272 + ], + "score": 1.0, + "content": "such that:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 259, + 274, + 352, + 290 + ], + "lines": [ + { + "bbox": [ + 259, + 274, + 352, + 290 + ], + "spans": [ + { + "bbox": [ + 259, + 274, + 352, + 290 + ], + "score": 0.91, + "content": "( X Q ) \\cdot ( X K ) ^ { \\top } = P", + "type": "interline_equation", + "image_path": "9f1c806c98f690ce205e28a0f3dfd5c85857fcf8d4b7e2221d4f0574d3860dab.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 259, + 274, + 352, + 290 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 293, + 506, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 293, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 254, + 306 + ], + "score": 1.0, + "content": "To satisfy Equation (16), we choose", + "type": "text" + }, + { + "bbox": [ + 255, + 295, + 264, + 305 + ], + "score": 0.82, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 293, + 283, + 306 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 283, + 294, + 294, + 304 + ], + "score": 0.84, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 293, + 335, + 306 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 335, + 294, + 377, + 306 + ], + "score": 0.93, + "content": "\\mathbf { \\Delta } X Q = U", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 293, + 396, + 306 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 396, + 293, + 456, + 304 + ], + "score": 0.92, + "content": "X K = P ^ { \\top } U", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 293, + 484, + 306 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 484, + 295, + 494, + 304 + ], + "score": 0.83, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 293, + 506, + 306 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 303, + 308, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 200, + 319 + ], + "score": 1.0, + "content": "an orthonormal matrix", + "type": "text" + }, + { + "bbox": [ + 200, + 305, + 302, + 316 + ], + "score": 0.87, + "content": "( \\pmb { U } \\cdot \\pmb { U } ^ { \\top } = \\pmb { U } ^ { \\top } \\cdot \\pmb { U } = I )", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 303, + 308, + 319 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 322, + 390, + 334 + ], + "lines": [ + { + "bbox": [ + 106, + 321, + 390, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 165, + 335 + ], + "score": 1.0, + "content": "We can obtain", + "type": "text" + }, + { + "bbox": [ + 165, + 322, + 176, + 332 + ], + "score": 0.81, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 321, + 375, + 335 + ], + "score": 1.0, + "content": "using the singular value decomposition (SVD) of", + "type": "text" + }, + { + "bbox": [ + 375, + 322, + 386, + 332 + ], + "score": 0.83, + "content": "\\boldsymbol { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 321, + 390, + 335 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "interline_equation", + "bbox": [ + 276, + 338, + 336, + 352 + ], + "lines": [ + { + "bbox": [ + 276, + 338, + 336, + 352 + ], + "spans": [ + { + "bbox": [ + 276, + 338, + 336, + 352 + ], + "score": 0.9, + "content": "\\pmb { X } = \\pmb { U } \\pmb { \\Sigma } \\pmb { V } ^ { \\top }", + "type": "interline_equation", + "image_path": "f9cfa19086ea4d26d15fce64dd9d96515f34aa8bedfd1629f9cc5f660ba70dbc.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 276, + 338, + 336, + 352 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 357, + 383, + 369 + ], + "lines": [ + { + "bbox": [ + 105, + 356, + 384, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 131, + 371 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 358, + 190, + 368 + ], + "score": 0.92, + "content": "\\ b { \\Sigma } \\in \\mathbb { R } ^ { n } \\times \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 356, + 208, + 371 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 209, + 358, + 219, + 368 + ], + "score": 0.81, + "content": "\\boldsymbol { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 356, + 282, + 371 + ], + "score": 1.0, + "content": "has a full rank,", + "type": "text" + }, + { + "bbox": [ + 283, + 358, + 292, + 368 + ], + "score": 0.83, + "content": "\\pmb { \\Sigma }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 356, + 384, + 371 + ], + "score": 1.0, + "content": "is invertible, and thus:", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 273, + 373, + 338, + 387 + ], + "lines": [ + { + "bbox": [ + 273, + 373, + 338, + 387 + ], + "spans": [ + { + "bbox": [ + 273, + 373, + 338, + 387 + ], + "score": 0.9, + "content": "\\pmb { X } \\pmb { V } \\pmb { \\Sigma } ^ { - 1 } = \\pmb { U }", + "type": "interline_equation", + "image_path": "70cdf4dfb1b4165f9d26012d5585e8c78028e46d0acb3e279bc127f8c2c78752.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 273, + 373, + 338, + 387 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 227, + 404 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 228, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 171, + 406 + ], + "score": 1.0, + "content": "Now, we define", + "type": "text" + }, + { + "bbox": [ + 171, + 393, + 181, + 404 + ], + "score": 0.84, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 391, + 228, + 406 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "interline_equation", + "bbox": [ + 279, + 401, + 332, + 416 + ], + "lines": [ + { + "bbox": [ + 279, + 401, + 332, + 416 + ], + "spans": [ + { + "bbox": [ + 279, + 401, + 332, + 416 + ], + "score": 0.9, + "content": "Q = V \\Sigma ^ { - 1 }", + "type": "interline_equation", + "image_path": "9bd0f61d5434954eb336f482c42efde0670901bd7906857e427cbf9ce9767b7d.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 279, + 401, + 332, + 416 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 417, + 236, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 417, + 236, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 146, + 431 + ], + "score": 1.0, + "content": "Note that", + "type": "text" + }, + { + "bbox": [ + 146, + 418, + 188, + 430 + ], + "score": 0.92, + "content": "\\mathbf { \\Delta } X Q = U", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 417, + 236, + 431 + ], + "score": 1.0, + "content": ", as desired.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 402, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 432, + 403, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 137, + 449 + ], + "score": 1.0, + "content": "To find", + "type": "text" + }, + { + "bbox": [ + 137, + 435, + 149, + 445 + ], + "score": 0.8, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 432, + 201, + 449 + ], + "score": 1.0, + "content": "that satisfies", + "type": "text" + }, + { + "bbox": [ + 201, + 434, + 261, + 445 + ], + "score": 0.92, + "content": "X K = P ^ { \\top } U", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 432, + 403, + 449 + ], + "score": 1.0, + "content": ", we use Equation (17) and require:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 450, + 348, + 465 + ], + "lines": [ + { + "bbox": [ + 263, + 450, + 348, + 465 + ], + "spans": [ + { + "bbox": [ + 263, + 450, + 348, + 465 + ], + "score": 0.9, + "content": "U \\Sigma V ^ { \\top } K = P ^ { \\top } U", + "type": "interline_equation", + "image_path": "3d72df2cc1041d92eefa2700e6a1e70c07ec67272321cf0aaab7d73fd139d2cf.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 263, + 450, + 348, + 465 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 471, + 144, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 146, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 146, + 483 + ], + "score": 1.0, + "content": "and thus:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 479, + 353, + 493 + ], + "lines": [ + { + "bbox": [ + 258, + 479, + 353, + 493 + ], + "spans": [ + { + "bbox": [ + 258, + 479, + 353, + 493 + ], + "score": 0.9, + "content": "\\pmb { K } = \\pmb { V } \\pmb { \\Sigma } ^ { - 1 } \\pmb { U } ^ { T } \\pmb { P } ^ { \\top } \\pmb { U }", + "type": "interline_equation", + "image_path": "772c658c6095e0d94634dd8eea267e0a482285f0cadc80317629464ad3ce6a70.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 258, + 479, + 353, + 493 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 502, + 150, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 501, + 152, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 152, + 515 + ], + "score": 1.0, + "content": "We define:", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 244, + 510, + 368, + 526 + ], + "lines": [ + { + "bbox": [ + 244, + 510, + 368, + 526 + ], + "spans": [ + { + "bbox": [ + 244, + 510, + 368, + 526 + ], + "score": 0.92, + "content": "z \\left( h _ { i } , h _ { j } \\right) = e \\left( h _ { i } , h _ { j } \\right) \\cdot \\sqrt { d _ { k } }", + "type": "interline_equation", + "image_path": "886128582d8bc69a605da787e1153820675d5a97d580f8b3aff8ea9ea4f27130.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 244, + 510, + 368, + 526 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 527, + 375, + 540 + ], + "lines": [ + { + "bbox": [ + 105, + 526, + 374, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 135, + 541 + ], + "score": 1.0, + "content": "Where", + "type": "text" + }, + { + "bbox": [ + 135, + 530, + 141, + 537 + ], + "score": 0.75, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 526, + 374, + 541 + ], + "score": 1.0, + "content": "is the attention score function of DPGAT (Equation (13)).", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 544, + 415, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 542, + 412, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 176, + 560 + ], + "score": 1.0, + "content": "Now, for a query", + "type": "text" + }, + { + "bbox": [ + 176, + 546, + 181, + 555 + ], + "score": 0.76, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 542, + 222, + 560 + ], + "score": 1.0, + "content": "and a key", + "type": "text" + }, + { + "bbox": [ + 223, + 546, + 228, + 556 + ], + "score": 0.81, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 542, + 386, + 560 + ], + "score": 1.0, + "content": ", and the corresponding representations", + "type": "text" + }, + { + "bbox": [ + 386, + 545, + 412, + 557 + ], + "score": 0.92, + "content": "h _ { i } , h _ { j }", + "type": "inline_equation" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 561, + 372, + 599 + ], + "lines": [ + { + "bbox": [ + 239, + 561, + 372, + 599 + ], + "spans": [ + { + "bbox": [ + 239, + 561, + 372, + 599 + ], + "score": 0.93, + "content": "\\begin{array} { r } { z \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\left( \\boldsymbol { h } _ { i } ^ { \\top } \\boldsymbol { Q } \\right) \\cdot \\left( \\boldsymbol { h } _ { j } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ { = \\left( \\boldsymbol { X } _ { i } \\boldsymbol { Q } \\right) \\cdot \\left( \\boldsymbol { X } _ { j } \\boldsymbol { K } \\right) ^ { \\top } } \\end{array}", + "type": "interline_equation", + "image_path": "5bb78d4b38de8bdebc579d7a07c51cb15f221a216eada0f8ecec66a7a518907a.jpg" + } + ] + } + ], + "index": 26.5, + "virtual_lines": [ + { + "bbox": [ + 239, + 561, + 372, + 580.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 239, + 580.0, + 372, + 599.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 319, + 620 + ], + "lines": [ + { + "bbox": [ + 105, + 604, + 320, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 131, + 623 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 607, + 197, + 620 + ], + "score": 0.93, + "content": "X _ { i } Q = ( X Q ) _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 604, + 215, + 623 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 216, + 607, + 286, + 621 + ], + "score": 0.93, + "content": "X _ { j } K = \\left( X K \\right) _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 604, + 320, + 623 + ], + "score": 1.0, + "content": ", we get", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 625, + 392, + 650 + ], + "lines": [ + { + "bbox": [ + 218, + 625, + 392, + 650 + ], + "spans": [ + { + "bbox": [ + 218, + 625, + 392, + 650 + ], + "score": 0.93, + "content": "z \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right) = \\left( \\pmb { X } \\pmb { Q } \\right) _ { i } \\cdot \\left( \\left( \\pmb { X } \\pmb { K } \\right) _ { j } \\right) ^ { \\top } = P _ { i , j }", + "type": "interline_equation", + "image_path": "59a2f091ef7d23c498396a61c5a74d577cfec96df400b10cebc2cfcd7f00898b.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 218, + 625, + 392, + 650 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 653, + 150, + 665 + ], + "lines": [ + { + "bbox": [ + 105, + 652, + 151, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 652, + 151, + 666 + ], + "score": 1.0, + "content": "Therefore:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "interline_equation", + "bbox": [ + 246, + 662, + 363, + 690 + ], + "lines": [ + { + "bbox": [ + 246, + 662, + 363, + 690 + ], + "spans": [ + { + "bbox": [ + 246, + 662, + 363, + 690 + ], + "score": 0.93, + "content": "z \\left( \\pmb { { h } } _ { i } , \\pmb { { h } } _ { j } \\right) = \\left. \\begin{array} { l l } { 1 } & { j = \\varphi ( i ) } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "b0ddfd09ff9341e61d95b0c272b09198832f8e47a4984960beb2f63af1e6a774.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 246, + 662, + 363, + 690 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 698, + 146, + 709 + ], + "lines": [ + { + "bbox": [ + 106, + 697, + 148, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 697, + 148, + 711 + ], + "score": 1.0, + "content": "And thus:", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 706, + 378, + 735 + ], + "lines": [ + { + "bbox": [ + 233, + 706, + 378, + 735 + ], + "spans": [ + { + "bbox": [ + 233, + 706, + 378, + 735 + ], + "score": 0.94, + "content": "e ( \\pmb { h } _ { i } , \\pmb { h } _ { j } ) = \\left\\{ \\begin{array} { l l } { 1 / \\sqrt { d _ { k } } } & { j = \\varphi ( i ) } \\\\ { 0 } & { o t h e r w i s e } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "e74cf091b15eff9a87338446ecefff2fa8214729431ee33b2a327fe96985b788.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 233, + 706, + 378, + 735 + ], + "spans": [], + "index": 33 + } + ] + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 81, + 287, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 284, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 164, + 96 + ], + "score": 1.0, + "content": "We denote the", + "type": "text" + }, + { + "bbox": [ + 165, + 82, + 177, + 92 + ], + "score": 0.88, + "content": "\\mathrm { i } ^ { t h }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 79, + 243, + 96 + ], + "score": 1.0, + "content": "row of a matrix", + "type": "text" + }, + { + "bbox": [ + 243, + 83, + 256, + 93 + ], + "score": 0.8, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 79, + 268, + 96 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 269, + 83, + 284, + 93 + ], + "score": 0.88, + "content": "M _ { i }", + "type": "inline_equation" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 105, + 79, + 284, + 96 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 99, + 208, + 111 + ], + "lines": [ + { + "bbox": [ + 106, + 98, + 209, + 112 + ], + "spans": [ + { + "bbox": [ + 106, + 98, + 184, + 112 + ], + "score": 1.0, + "content": "We define a matrix", + "type": "text" + }, + { + "bbox": [ + 184, + 100, + 194, + 109 + ], + "score": 0.81, + "content": "_ { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 98, + 209, + 112 + ], + "score": 1.0, + "content": "as:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 98, + 209, + 112 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 108, + 351, + 136 + ], + "lines": [ + { + "bbox": [ + 258, + 108, + 351, + 136 + ], + "spans": [ + { + "bbox": [ + 258, + 108, + 351, + 136 + ], + "score": 0.93, + "content": "P _ { i , j } = { \\left\\{ \\begin{array} { l l } { 1 } & { j = \\varphi ( i ) } \\\\ { 0 } & { { \\mathrm { o t h e r w i s e } } } \\end{array} \\right. }", + "type": "interline_equation", + "image_path": "477f035d5e3de0a9ac649d2e748d973286069bc2d3286257c4f0cb84d9aab2f1.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 258, + 108, + 351, + 136 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 145, + 450, + 158 + ], + "lines": [ + { + "bbox": [ + 105, + 144, + 451, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 122, + 159 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 145, + 182, + 156 + ], + "score": 0.92, + "content": "\\ b { X } \\in \\mathbb { R } ^ { n } \\times \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 144, + 451, + 159 + ], + "score": 1.0, + "content": "be the matrix holding the graph’s node representations as its rows:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 144, + 451, + 159 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 260, + 177, + 352, + 231 + ], + "lines": [ + { + "bbox": [ + 260, + 177, + 352, + 231 + ], + "spans": [ + { + "bbox": [ + 260, + 177, + 352, + 231 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathbf { \\boldsymbol { X } } = \\left[ \\begin{array} { l l l } { - } & { h _ { 1 } } & { - } \\\\ { - } & { h _ { 2 } } & { - } \\\\ & { \\vdots } \\\\ { - } & { h _ { n } } & { - } \\end{array} \\right] } \\end{array}", + "type": "interline_equation", + "image_path": "a6beda280a46d1888de1e211f54408173573f54ec526a68249ae83b483c0ab7c.jpg" + } + ] + } + ], + "index": 4.5, + "virtual_lines": [ + { + "bbox": [ + 260, + 177, + 352, + 204.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 260, + 204.0, + 352, + 231.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 241, + 415, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 414, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 178, + 256 + ], + "score": 1.0, + "content": "Since the rows of", + "type": "text" + }, + { + "bbox": [ + 178, + 242, + 189, + 252 + ], + "score": 0.84, + "content": "\\boldsymbol { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 239, + 388, + 256 + ], + "score": 1.0, + "content": "are linearly independent, it necessarily holds that", + "type": "text" + }, + { + "bbox": [ + 389, + 242, + 414, + 253 + ], + "score": 0.91, + "content": "d \\geq n", + "type": "inline_equation" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 239, + 414, + 256 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 407, + 270 + ], + "lines": [ + { + "bbox": [ + 104, + 255, + 407, + 272 + ], + "spans": [ + { + "bbox": [ + 104, + 255, + 228, + 272 + ], + "score": 1.0, + "content": "Next, we find weight matrices", + "type": "text" + }, + { + "bbox": [ + 229, + 257, + 286, + 270 + ], + "score": 0.92, + "content": "Q \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 255, + 305, + 272 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 305, + 257, + 364, + 269 + ], + "score": 0.92, + "content": "K \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 255, + 407, + 272 + ], + "score": 1.0, + "content": "such that:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 104, + 255, + 407, + 272 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 259, + 274, + 352, + 290 + ], + "lines": [ + { + "bbox": [ + 259, + 274, + 352, + 290 + ], + "spans": [ + { + "bbox": [ + 259, + 274, + 352, + 290 + ], + "score": 0.91, + "content": "( X Q ) \\cdot ( X K ) ^ { \\top } = P", + "type": "interline_equation", + "image_path": "9f1c806c98f690ce205e28a0f3dfd5c85857fcf8d4b7e2221d4f0574d3860dab.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 259, + 274, + 352, + 290 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 293, + 506, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 293, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 254, + 306 + ], + "score": 1.0, + "content": "To satisfy Equation (16), we choose", + "type": "text" + }, + { + "bbox": [ + 255, + 295, + 264, + 305 + ], + "score": 0.82, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 293, + 283, + 306 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 283, + 294, + 294, + 304 + ], + "score": 0.84, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 293, + 335, + 306 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 335, + 294, + 377, + 306 + ], + "score": 0.93, + "content": "\\mathbf { \\Delta } X Q = U", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 293, + 396, + 306 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 396, + 293, + 456, + 304 + ], + "score": 0.92, + "content": "X K = P ^ { \\top } U", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 293, + 484, + 306 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 484, + 295, + 494, + 304 + ], + "score": 0.83, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 293, + 506, + 306 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 303, + 308, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 200, + 319 + ], + "score": 1.0, + "content": "an orthonormal matrix", + "type": "text" + }, + { + "bbox": [ + 200, + 305, + 302, + 316 + ], + "score": 0.87, + "content": "( \\pmb { U } \\cdot \\pmb { U } ^ { \\top } = \\pmb { U } ^ { \\top } \\cdot \\pmb { U } = I )", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 303, + 308, + 319 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 293, + 506, + 319 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 322, + 390, + 334 + ], + "lines": [ + { + "bbox": [ + 106, + 321, + 390, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 165, + 335 + ], + "score": 1.0, + "content": "We can obtain", + "type": "text" + }, + { + "bbox": [ + 165, + 322, + 176, + 332 + ], + "score": 0.81, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 321, + 375, + 335 + ], + "score": 1.0, + "content": "using the singular value decomposition (SVD) of", + "type": "text" + }, + { + "bbox": [ + 375, + 322, + 386, + 332 + ], + "score": 0.83, + "content": "\\boldsymbol { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 321, + 390, + 335 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 321, + 390, + 335 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 276, + 338, + 336, + 352 + ], + "lines": [ + { + "bbox": [ + 276, + 338, + 336, + 352 + ], + "spans": [ + { + "bbox": [ + 276, + 338, + 336, + 352 + ], + "score": 0.9, + "content": "\\pmb { X } = \\pmb { U } \\pmb { \\Sigma } \\pmb { V } ^ { \\top }", + "type": "interline_equation", + "image_path": "f9cfa19086ea4d26d15fce64dd9d96515f34aa8bedfd1629f9cc5f660ba70dbc.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 276, + 338, + 336, + 352 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 357, + 383, + 369 + ], + "lines": [ + { + "bbox": [ + 105, + 356, + 384, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 131, + 371 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 358, + 190, + 368 + ], + "score": 0.92, + "content": "\\ b { \\Sigma } \\in \\mathbb { R } ^ { n } \\times \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 356, + 208, + 371 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 209, + 358, + 219, + 368 + ], + "score": 0.81, + "content": "\\boldsymbol { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 356, + 282, + 371 + ], + "score": 1.0, + "content": "has a full rank,", + "type": "text" + }, + { + "bbox": [ + 283, + 358, + 292, + 368 + ], + "score": 0.83, + "content": "\\pmb { \\Sigma }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 356, + 384, + 371 + ], + "score": 1.0, + "content": "is invertible, and thus:", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 356, + 384, + 371 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 273, + 373, + 338, + 387 + ], + "lines": [ + { + "bbox": [ + 273, + 373, + 338, + 387 + ], + "spans": [ + { + "bbox": [ + 273, + 373, + 338, + 387 + ], + "score": 0.9, + "content": "\\pmb { X } \\pmb { V } \\pmb { \\Sigma } ^ { - 1 } = \\pmb { U }", + "type": "interline_equation", + "image_path": "70cdf4dfb1b4165f9d26012d5585e8c78028e46d0acb3e279bc127f8c2c78752.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 273, + 373, + 338, + 387 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 392, + 227, + 404 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 228, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 171, + 406 + ], + "score": 1.0, + "content": "Now, we define", + "type": "text" + }, + { + "bbox": [ + 171, + 393, + 181, + 404 + ], + "score": 0.84, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 391, + 228, + 406 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 391, + 228, + 406 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 279, + 401, + 332, + 416 + ], + "lines": [ + { + "bbox": [ + 279, + 401, + 332, + 416 + ], + "spans": [ + { + "bbox": [ + 279, + 401, + 332, + 416 + ], + "score": 0.9, + "content": "Q = V \\Sigma ^ { - 1 }", + "type": "interline_equation", + "image_path": "9bd0f61d5434954eb336f482c42efde0670901bd7906857e427cbf9ce9767b7d.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 279, + 401, + 332, + 416 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 417, + 236, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 417, + 236, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 146, + 431 + ], + "score": 1.0, + "content": "Note that", + "type": "text" + }, + { + "bbox": [ + 146, + 418, + 188, + 430 + ], + "score": 0.92, + "content": "\\mathbf { \\Delta } X Q = U", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 417, + 236, + 431 + ], + "score": 1.0, + "content": ", as desired.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 417, + 236, + 431 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 402, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 432, + 403, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 137, + 449 + ], + "score": 1.0, + "content": "To find", + "type": "text" + }, + { + "bbox": [ + 137, + 435, + 149, + 445 + ], + "score": 0.8, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 432, + 201, + 449 + ], + "score": 1.0, + "content": "that satisfies", + "type": "text" + }, + { + "bbox": [ + 201, + 434, + 261, + 445 + ], + "score": 0.92, + "content": "X K = P ^ { \\top } U", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 432, + 403, + 449 + ], + "score": 1.0, + "content": ", we use Equation (17) and require:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 432, + 403, + 449 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 450, + 348, + 465 + ], + "lines": [ + { + "bbox": [ + 263, + 450, + 348, + 465 + ], + "spans": [ + { + "bbox": [ + 263, + 450, + 348, + 465 + ], + "score": 0.9, + "content": "U \\Sigma V ^ { \\top } K = P ^ { \\top } U", + "type": "interline_equation", + "image_path": "3d72df2cc1041d92eefa2700e6a1e70c07ec67272321cf0aaab7d73fd139d2cf.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 263, + 450, + 348, + 465 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 471, + 144, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 146, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 146, + 483 + ], + "score": 1.0, + "content": "and thus:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 470, + 146, + 483 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 479, + 353, + 493 + ], + "lines": [ + { + "bbox": [ + 258, + 479, + 353, + 493 + ], + "spans": [ + { + "bbox": [ + 258, + 479, + 353, + 493 + ], + "score": 0.9, + "content": "\\pmb { K } = \\pmb { V } \\pmb { \\Sigma } ^ { - 1 } \\pmb { U } ^ { T } \\pmb { P } ^ { \\top } \\pmb { U }", + "type": "interline_equation", + "image_path": "772c658c6095e0d94634dd8eea267e0a482285f0cadc80317629464ad3ce6a70.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 258, + 479, + 353, + 493 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 502, + 150, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 501, + 152, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 152, + 515 + ], + "score": 1.0, + "content": "We define:", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 501, + 152, + 515 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 244, + 510, + 368, + 526 + ], + "lines": [ + { + "bbox": [ + 244, + 510, + 368, + 526 + ], + "spans": [ + { + "bbox": [ + 244, + 510, + 368, + 526 + ], + "score": 0.92, + "content": "z \\left( h _ { i } , h _ { j } \\right) = e \\left( h _ { i } , h _ { j } \\right) \\cdot \\sqrt { d _ { k } }", + "type": "interline_equation", + "image_path": "886128582d8bc69a605da787e1153820675d5a97d580f8b3aff8ea9ea4f27130.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 244, + 510, + 368, + 526 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 527, + 375, + 540 + ], + "lines": [ + { + "bbox": [ + 105, + 526, + 374, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 135, + 541 + ], + "score": 1.0, + "content": "Where", + "type": "text" + }, + { + "bbox": [ + 135, + 530, + 141, + 537 + ], + "score": 0.75, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 526, + 374, + 541 + ], + "score": 1.0, + "content": "is the attention score function of DPGAT (Equation (13)).", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 526, + 374, + 541 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 544, + 415, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 542, + 412, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 176, + 560 + ], + "score": 1.0, + "content": "Now, for a query", + "type": "text" + }, + { + "bbox": [ + 176, + 546, + 181, + 555 + ], + "score": 0.76, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 542, + 222, + 560 + ], + "score": 1.0, + "content": "and a key", + "type": "text" + }, + { + "bbox": [ + 223, + 546, + 228, + 556 + ], + "score": 0.81, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 542, + 386, + 560 + ], + "score": 1.0, + "content": ", and the corresponding representations", + "type": "text" + }, + { + "bbox": [ + 386, + 545, + 412, + 557 + ], + "score": 0.92, + "content": "h _ { i } , h _ { j }", + "type": "inline_equation" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 542, + 412, + 560 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 561, + 372, + 599 + ], + "lines": [ + { + "bbox": [ + 239, + 561, + 372, + 599 + ], + "spans": [ + { + "bbox": [ + 239, + 561, + 372, + 599 + ], + "score": 0.93, + "content": "\\begin{array} { r } { z \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\left( \\boldsymbol { h } _ { i } ^ { \\top } \\boldsymbol { Q } \\right) \\cdot \\left( \\boldsymbol { h } _ { j } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ { = \\left( \\boldsymbol { X } _ { i } \\boldsymbol { Q } \\right) \\cdot \\left( \\boldsymbol { X } _ { j } \\boldsymbol { K } \\right) ^ { \\top } } \\end{array}", + "type": "interline_equation", + "image_path": "5bb78d4b38de8bdebc579d7a07c51cb15f221a216eada0f8ecec66a7a518907a.jpg" + } + ] + } + ], + "index": 26.5, + "virtual_lines": [ + { + "bbox": [ + 239, + 561, + 372, + 580.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 239, + 580.0, + 372, + 599.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 319, + 620 + ], + "lines": [ + { + "bbox": [ + 105, + 604, + 320, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 131, + 623 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 607, + 197, + 620 + ], + "score": 0.93, + "content": "X _ { i } Q = ( X Q ) _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 604, + 215, + 623 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 216, + 607, + 286, + 621 + ], + "score": 0.93, + "content": "X _ { j } K = \\left( X K \\right) _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 604, + 320, + 623 + ], + "score": 1.0, + "content": ", we get", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 604, + 320, + 623 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 625, + 392, + 650 + ], + "lines": [ + { + "bbox": [ + 218, + 625, + 392, + 650 + ], + "spans": [ + { + "bbox": [ + 218, + 625, + 392, + 650 + ], + "score": 0.93, + "content": "z \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right) = \\left( \\pmb { X } \\pmb { Q } \\right) _ { i } \\cdot \\left( \\left( \\pmb { X } \\pmb { K } \\right) _ { j } \\right) ^ { \\top } = P _ { i , j }", + "type": "interline_equation", + "image_path": "59a2f091ef7d23c498396a61c5a74d577cfec96df400b10cebc2cfcd7f00898b.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 218, + 625, + 392, + 650 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 653, + 150, + 665 + ], + "lines": [ + { + "bbox": [ + 105, + 652, + 151, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 652, + 151, + 666 + ], + "score": 1.0, + "content": "Therefore:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 652, + 151, + 666 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 246, + 662, + 363, + 690 + ], + "lines": [ + { + "bbox": [ + 246, + 662, + 363, + 690 + ], + "spans": [ + { + "bbox": [ + 246, + 662, + 363, + 690 + ], + "score": 0.93, + "content": "z \\left( \\pmb { { h } } _ { i } , \\pmb { { h } } _ { j } \\right) = \\left. \\begin{array} { l l } { 1 } & { j = \\varphi ( i ) } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "b0ddfd09ff9341e61d95b0c272b09198832f8e47a4984960beb2f63af1e6a774.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 246, + 662, + 363, + 690 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 698, + 146, + 709 + ], + "lines": [ + { + "bbox": [ + 106, + 697, + 148, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 697, + 148, + 711 + ], + "score": 1.0, + "content": "And thus:", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32, + "bbox_fs": [ + 106, + 697, + 148, + 711 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 706, + 378, + 735 + ], + "lines": [ + { + "bbox": [ + 233, + 706, + 378, + 735 + ], + "spans": [ + { + "bbox": [ + 233, + 706, + 378, + 735 + ], + "score": 0.94, + "content": "e ( \\pmb { h } _ { i } , \\pmb { h } _ { j } ) = \\left\\{ \\begin{array} { l l } { 1 / \\sqrt { d _ { k } } } & { j = \\varphi ( i ) } \\\\ { 0 } & { o t h e r w i s e } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "e74cf091b15eff9a87338446ecefff2fa8214729431ee33b2a327fe96985b788.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 233, + 706, + 378, + 735 + ], + "spans": [], + "index": 33 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 344, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 340, + 99 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 258, + 99 + ], + "score": 1.0, + "content": "To conclude, for every selected query", + "type": "text" + }, + { + "bbox": [ + 258, + 84, + 263, + 92 + ], + "score": 0.78, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 78, + 314, + 99 + ], + "score": 1.0, + "content": "and any key", + "type": "text" + }, + { + "bbox": [ + 314, + 83, + 340, + 96 + ], + "score": 0.92, + "content": "\\mathscr { j } _ { \\neq \\varphi ( i ) }", + "type": "inline_equation" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 252, + 98, + 358, + 114 + ], + "lines": [ + { + "bbox": [ + 252, + 98, + 358, + 114 + ], + "spans": [ + { + "bbox": [ + 252, + 98, + 358, + 114 + ], + "score": 0.92, + "content": "e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { \\varphi ( i ) } \\right) > e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right)", + "type": "interline_equation", + "image_path": "c78a2203b489ca5ecc4ed1f70e51b1ca20995f7e10ee304b4742a735b57e3b13.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 252, + 98, + 358, + 114 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 116, + 314, + 128 + ], + "lines": [ + { + "bbox": [ + 105, + 115, + 315, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 315, + 129 + ], + "score": 1.0, + "content": "and due to the increasing monotonicity of softmax:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "interline_equation", + "bbox": [ + 277, + 132, + 334, + 145 + ], + "lines": [ + { + "bbox": [ + 277, + 132, + 334, + 145 + ], + "spans": [ + { + "bbox": [ + 277, + 132, + 334, + 145 + ], + "score": 0.92, + "content": "\\alpha _ { i , \\varphi ( i ) } > \\alpha _ { i , j }", + "type": "interline_equation", + "image_path": "4288e5c3d9b440fbda0b46235a3143151b8af70fc233595a40a6250b2fda7132.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 277, + 132, + 334, + 145 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 147, + 317, + 159 + ], + "lines": [ + { + "bbox": [ + 105, + 146, + 318, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 318, + 162 + ], + "score": 1.0, + "content": "Hence, a DPGAT layer computes dynamic attention.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 163, + 504, + 187 + ], + "lines": [ + { + "bbox": [ + 104, + 161, + 507, + 178 + ], + "spans": [ + { + "bbox": [ + 104, + 161, + 170, + 178 + ], + "score": 1.0, + "content": "In the case that", + "type": "text" + }, + { + "bbox": [ + 170, + 165, + 195, + 174 + ], + "score": 0.9, + "content": "d > n", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 161, + 353, + 178 + ], + "score": 1.0, + "content": ", we apply SVD to the full-rank matrix", + "type": "text" + }, + { + "bbox": [ + 353, + 163, + 417, + 175 + ], + "score": 0.92, + "content": "X X ^ { \\top } \\in \\mathbb { R } ^ { n \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 161, + 507, + 178 + ], + "score": 1.0, + "content": ", and follow the same", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 175, + 221, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 175, + 178, + 186 + ], + "score": 1.0, + "content": "steps to construct", + "type": "text" + }, + { + "bbox": [ + 178, + 176, + 188, + 187 + ], + "score": 0.86, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 175, + 206, + 186 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 207, + 176, + 218, + 185 + ], + "score": 0.85, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 175, + 221, + 186 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 106, + 190, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 189, + 504, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 172, + 204 + ], + "score": 1.0, + "content": "In the case that", + "type": "text" + }, + { + "bbox": [ + 172, + 191, + 235, + 203 + ], + "score": 0.93, + "content": "Q \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 189, + 255, + 204 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 255, + 191, + 320, + 203 + ], + "score": 0.91, + "content": "K \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 189, + 340, + 204 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 340, + 192, + 372, + 203 + ], + "score": 0.9, + "content": "d _ { k } > d", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 189, + 463, + 204 + ], + "score": 1.0, + "content": ", we can use the same", + "type": "text" + }, + { + "bbox": [ + 463, + 192, + 473, + 203 + ], + "score": 0.83, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 189, + 492, + 204 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 493, + 192, + 504, + 202 + ], + "score": 0.79, + "content": "\\kappa", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 202, + 503, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 341, + 214 + ], + "score": 1.0, + "content": "(Equations (19) and (21)) padded with zeros. We define the", + "type": "text" + }, + { + "bbox": [ + 341, + 203, + 411, + 214 + ], + "score": 0.92, + "content": "Q ^ { \\prime } \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k e y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 202, + 430, + 214 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 431, + 203, + 503, + 213 + ], + "score": 0.91, + "content": "K ^ { \\prime } \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k e y } }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 213, + 153, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 153, + 226 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 250, + 227, + 360, + 286 + ], + "lines": [ + { + "bbox": [ + 250, + 227, + 360, + 286 + ], + "spans": [ + { + "bbox": [ + 250, + 227, + 360, + 286 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { Q _ { i , j } ^ { \\prime } = \\left\\{ \\begin{array} { l l } { Q _ { i , j } } & { j \\leq d } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. } \\\\ & { K _ { i , j } ^ { \\prime } = \\left\\{ \\begin{array} { l l } { K _ { i , j } } & { j \\leq d } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. } \\end{array}", + "type": "interline_equation", + "image_path": "31ad72d88f5cd93de23b90d47d003e95807d303624927413ae713c14bb4c7445.jpg" + } + ] + } + ], + "index": 10.5, + "virtual_lines": [ + { + "bbox": [ + 250, + 227, + 360, + 256.5 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 250, + 256.5, + 360, + 286.0 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 312, + 325, + 325 + ], + "lines": [ + { + "bbox": [ + 106, + 312, + 325, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 325, + 326 + ], + "score": 1.0, + "content": "G.2 DPGAT IS STRICTLY WEAKER THAN GATV2", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 367 + ], + "lines": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 452, + 346 + ], + "score": 1.0, + "content": "There are examples of node representations that are linearly dependent and mappings", + "type": "text" + }, + { + "bbox": [ + 452, + 336, + 460, + 345 + ], + "score": 0.8, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 334, + 505, + 346 + ], + "score": 1.0, + "content": ", for which", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 345, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 505, + 357 + ], + "score": 1.0, + "content": "dynamic attention does not hold. First, we show a simple 2-dimensional example, and then we show", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 356, + 246, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 246, + 368 + ], + "score": 1.0, + "content": "the general case of such examples.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 255, + 379, + 356, + 452 + ], + "lines": [ + { + "bbox": [ + 257, + 380, + 356, + 452 + ], + "spans": [ + { + "bbox": [ + 257, + 380, + 356, + 452 + ], + "score": 0.59, + "content": "\\prod _ { \\substack { h _ { 0 } = \\hat { x } } } ^ { y } h _ { 2 } = \\hat { x } + 2 \\hat { y }", + "type": "interline_equation", + "image_path": "b0f441cd12d6c4942d10f158b5b3baab44b607c6ac98f4f4863b6c4b09544de8.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 255, + 379, + 356, + 415.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 255, + 415.5, + 356, + 452.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "image", + "bbox": [], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 105, + 461, + 505, + 485 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "Figure 6: An example for node representations that are linearly dependent, for which DPGAT cannot", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 472, + 417, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 319, + 485 + ], + "score": 1.0, + "content": "compute dynamic attention, because no query vector", + "type": "text" + }, + { + "bbox": [ + 319, + 472, + 349, + 484 + ], + "score": 0.92, + "content": "q \\in \\mathbb { R } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 473, + 401, + 485 + ], + "score": 1.0, + "content": "can “select”", + "type": "text" + }, + { + "bbox": [ + 401, + 473, + 413, + 484 + ], + "score": 0.89, + "content": "\\pmb { h } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 473, + 417, + 485 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 409, + 508 + ], + "lines": [ + { + "bbox": [ + 106, + 494, + 409, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 332, + 510 + ], + "score": 1.0, + "content": "Consider the following linearly dependent set of vectors", + "type": "text" + }, + { + "bbox": [ + 333, + 496, + 363, + 507 + ], + "score": 0.89, + "content": "\\mathbb { K } = \\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 494, + 409, + 510 + ], + "score": 1.0, + "content": "(Figure 6):", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 276, + 510, + 335, + 552 + ], + "lines": [ + { + "bbox": [ + 276, + 510, + 335, + 552 + ], + "spans": [ + { + "bbox": [ + 276, + 510, + 335, + 552 + ], + "score": 0.88, + "content": "\\begin{array} { l } { { \\pmb { h } } _ { 0 } = { \\pmb { \\hat { x } } } } \\\\ { { \\pmb { h } } _ { 1 } = { \\pmb { \\hat { x } } } + { \\pmb { \\hat { y } } } } \\\\ { { \\pmb { h } } _ { 2 } = { \\pmb { \\hat { x } } } + 2 { \\pmb { \\hat { y } } } } \\end{array}", + "type": "interline_equation", + "image_path": "a79e708698104ffef2ca0bc52c70500f9ea1b7a34cbabe95cef0ae1a5d18972f.jpg" + } + ] + } + ], + "index": 21.5, + "virtual_lines": [ + { + "bbox": [ + 276, + 510, + 335, + 531.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 276, + 531.0, + 335, + 552.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 555, + 505, + 577 + ], + "lines": [ + { + "bbox": [ + 104, + 552, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 104, + 552, + 133, + 569 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 556, + 141, + 565 + ], + "score": 0.83, + "content": "\\hat { \\pmb x }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 552, + 159, + 569 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 159, + 556, + 167, + 567 + ], + "score": 0.85, + "content": "\\hat { \\pmb { y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 552, + 329, + 569 + ], + "score": 1.0, + "content": "are the cartesian unit vectors. We define", + "type": "text" + }, + { + "bbox": [ + 330, + 555, + 383, + 567 + ], + "score": 0.93, + "content": "\\beta \\in \\{ 0 , 1 , 2 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 552, + 426, + 569 + ], + "score": 1.0, + "content": "to express", + "type": "text" + }, + { + "bbox": [ + 427, + 555, + 480, + 567 + ], + "score": 0.93, + "content": "\\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 552, + 506, + 569 + ], + "score": 1.0, + "content": "using", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 566, + 191, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 191, + 578 + ], + "score": 1.0, + "content": "the same expression:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "interline_equation", + "bbox": [ + 276, + 576, + 336, + 590 + ], + "lines": [ + { + "bbox": [ + 276, + 576, + 336, + 590 + ], + "spans": [ + { + "bbox": [ + 276, + 576, + 336, + 590 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\pmb { h } _ { \\beta } = \\hat { \\pmb { x } } + \\beta \\hat { \\pmb y } } \\end{array}", + "type": "interline_equation", + "image_path": "575a2fa2b0ec9c91e579d6ce367d482b1b99ee7811ede8beb83e8868904546f6.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 276, + 576, + 336, + 590 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 501, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 589, + 498, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 122, + 605 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 122, + 591, + 149, + 603 + ], + "score": 0.92, + "content": "\\pmb q \\in \\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 589, + 492, + 605 + ], + "score": 1.0, + "content": "be any query vector. For brevity, we define the unscaled dot-product attention score as", + "type": "text" + }, + { + "bbox": [ + 492, + 594, + 498, + 601 + ], + "score": 0.75, + "content": "s", + "type": "inline_equation" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "interline_equation", + "bbox": [ + 248, + 605, + 364, + 622 + ], + "lines": [ + { + "bbox": [ + 248, + 605, + 364, + 622 + ], + "spans": [ + { + "bbox": [ + 248, + 605, + 364, + 622 + ], + "score": 0.93, + "content": "s \\left( \\mathbf { q } , \\boldsymbol { h } _ { \\beta } \\right) = e \\left( \\mathbf { q } , \\boldsymbol { h } _ { \\beta } \\right) \\cdot \\sqrt { d _ { k } }", + "type": "interline_equation", + "image_path": "821407b7531cafa4a14e43215bbe169255bfb4789c144f517180083a952b491d.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 248, + 605, + 364, + 622 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 624, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 135, + 637 + ], + "score": 1.0, + "content": "Where", + "type": "text" + }, + { + "bbox": [ + 136, + 626, + 141, + 634 + ], + "score": 0.79, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 623, + 505, + 637 + ], + "score": 1.0, + "content": "is the attention score function of DPGAT (Equation (13)). The (unscaled) attention score", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 635, + 234, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 142, + 648 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 142, + 637, + 149, + 647 + ], + "score": 0.8, + "content": "\\pmb q", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 635, + 167, + 648 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 168, + 635, + 221, + 648 + ], + "score": 0.93, + "content": "\\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 635, + 234, + 648 + ], + "score": 1.0, + "content": "is:", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "interline_equation", + "bbox": [ + 197, + 649, + 413, + 734 + ], + "lines": [ + { + "bbox": [ + 197, + 649, + 413, + 734 + ], + "spans": [ + { + "bbox": [ + 197, + 649, + 413, + 734 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { s \\left( \\pmb { q } , \\pmb { h } _ { \\beta } \\right) = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { h } _ { \\beta } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\left( \\hat { \\pmb { x } } + \\beta \\pmb { \\hat { y } } \\right) ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { x } } ^ { \\top } \\pmb { K } + \\beta \\pmb { \\hat { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { x } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } + \\beta \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { \\hat { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\end{array}", + "type": "interline_equation", + "image_path": "2f076942dab3212541d4295ee604f1a329f957ef523d751087c6b0cd4b687f9c.jpg" + } + ] + } + ], + "index": 32.5, + "virtual_lines": [ + { + "bbox": [ + 197, + 649, + 413, + 663.1666666666666 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 197, + 663.1666666666666, + 413, + 677.3333333333333 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 197, + 677.3333333333333, + 413, + 691.4999999999999 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 197, + 691.4999999999999, + 413, + 705.6666666666665 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 197, + 705.6666666666665, + 413, + 719.8333333333331 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 197, + 719.8333333333331, + 413, + 733.9999999999998 + ], + "spans": [], + "index": 35 + } + ] + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 495, + 289, + 504, + 299 + ], + "lines": [ + { + "bbox": [ + 496, + 291, + 504, + 299 + ], + "spans": [ + { + "bbox": [ + 496, + 291, + 504, + 299 + ], + "score": 0.997, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 344, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 340, + 99 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 258, + 99 + ], + "score": 1.0, + "content": "To conclude, for every selected query", + "type": "text" + }, + { + "bbox": [ + 258, + 84, + 263, + 92 + ], + "score": 0.78, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 78, + 314, + 99 + ], + "score": 1.0, + "content": "and any key", + "type": "text" + }, + { + "bbox": [ + 314, + 83, + 340, + 96 + ], + "score": 0.92, + "content": "\\mathscr { j } _ { \\neq \\varphi ( i ) }", + "type": "inline_equation" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 105, + 78, + 340, + 99 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 252, + 98, + 358, + 114 + ], + "lines": [ + { + "bbox": [ + 252, + 98, + 358, + 114 + ], + "spans": [ + { + "bbox": [ + 252, + 98, + 358, + 114 + ], + "score": 0.92, + "content": "e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { \\varphi ( i ) } \\right) > e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right)", + "type": "interline_equation", + "image_path": "c78a2203b489ca5ecc4ed1f70e51b1ca20995f7e10ee304b4742a735b57e3b13.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 252, + 98, + 358, + 114 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 116, + 314, + 128 + ], + "lines": [ + { + "bbox": [ + 105, + 115, + 315, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 315, + 129 + ], + "score": 1.0, + "content": "and due to the increasing monotonicity of softmax:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 115, + 315, + 129 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 277, + 132, + 334, + 145 + ], + "lines": [ + { + "bbox": [ + 277, + 132, + 334, + 145 + ], + "spans": [ + { + "bbox": [ + 277, + 132, + 334, + 145 + ], + "score": 0.92, + "content": "\\alpha _ { i , \\varphi ( i ) } > \\alpha _ { i , j }", + "type": "interline_equation", + "image_path": "4288e5c3d9b440fbda0b46235a3143151b8af70fc233595a40a6250b2fda7132.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 277, + 132, + 334, + 145 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 147, + 317, + 159 + ], + "lines": [ + { + "bbox": [ + 105, + 146, + 318, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 318, + 162 + ], + "score": 1.0, + "content": "Hence, a DPGAT layer computes dynamic attention.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 146, + 318, + 162 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 163, + 504, + 187 + ], + "lines": [ + { + "bbox": [ + 104, + 161, + 507, + 178 + ], + "spans": [ + { + "bbox": [ + 104, + 161, + 170, + 178 + ], + "score": 1.0, + "content": "In the case that", + "type": "text" + }, + { + "bbox": [ + 170, + 165, + 195, + 174 + ], + "score": 0.9, + "content": "d > n", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 161, + 353, + 178 + ], + "score": 1.0, + "content": ", we apply SVD to the full-rank matrix", + "type": "text" + }, + { + "bbox": [ + 353, + 163, + 417, + 175 + ], + "score": 0.92, + "content": "X X ^ { \\top } \\in \\mathbb { R } ^ { n \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 161, + 507, + 178 + ], + "score": 1.0, + "content": ", and follow the same", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 175, + 221, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 175, + 178, + 186 + ], + "score": 1.0, + "content": "steps to construct", + "type": "text" + }, + { + "bbox": [ + 178, + 176, + 188, + 187 + ], + "score": 0.86, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 175, + 206, + 186 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 207, + 176, + 218, + 185 + ], + "score": 0.85, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 175, + 221, + 186 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 104, + 161, + 507, + 187 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 190, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 189, + 504, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 172, + 204 + ], + "score": 1.0, + "content": "In the case that", + "type": "text" + }, + { + "bbox": [ + 172, + 191, + 235, + 203 + ], + "score": 0.93, + "content": "Q \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 189, + 255, + 204 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 255, + 191, + 320, + 203 + ], + "score": 0.91, + "content": "K \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 189, + 340, + 204 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 340, + 192, + 372, + 203 + ], + "score": 0.9, + "content": "d _ { k } > d", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 189, + 463, + 204 + ], + "score": 1.0, + "content": ", we can use the same", + "type": "text" + }, + { + "bbox": [ + 463, + 192, + 473, + 203 + ], + "score": 0.83, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 189, + 492, + 204 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 493, + 192, + 504, + 202 + ], + "score": 0.79, + "content": "\\kappa", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 202, + 503, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 341, + 214 + ], + "score": 1.0, + "content": "(Equations (19) and (21)) padded with zeros. We define the", + "type": "text" + }, + { + "bbox": [ + 341, + 203, + 411, + 214 + ], + "score": 0.92, + "content": "Q ^ { \\prime } \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k e y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 202, + 430, + 214 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 431, + 203, + 503, + 213 + ], + "score": 0.91, + "content": "K ^ { \\prime } \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k e y } }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 213, + 153, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 153, + 226 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 189, + 504, + 226 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 250, + 227, + 360, + 286 + ], + "lines": [ + { + "bbox": [ + 250, + 227, + 360, + 286 + ], + "spans": [ + { + "bbox": [ + 250, + 227, + 360, + 286 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { Q _ { i , j } ^ { \\prime } = \\left\\{ \\begin{array} { l l } { Q _ { i , j } } & { j \\leq d } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. } \\\\ & { K _ { i , j } ^ { \\prime } = \\left\\{ \\begin{array} { l l } { K _ { i , j } } & { j \\leq d } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. } \\end{array}", + "type": "interline_equation", + "image_path": "31ad72d88f5cd93de23b90d47d003e95807d303624927413ae713c14bb4c7445.jpg" + } + ] + } + ], + "index": 10.5, + "virtual_lines": [ + { + "bbox": [ + 250, + 227, + 360, + 256.5 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 250, + 256.5, + 360, + 286.0 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 312, + 325, + 325 + ], + "lines": [ + { + "bbox": [ + 106, + 312, + 325, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 325, + 326 + ], + "score": 1.0, + "content": "G.2 DPGAT IS STRICTLY WEAKER THAN GATV2", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 367 + ], + "lines": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 452, + 346 + ], + "score": 1.0, + "content": "There are examples of node representations that are linearly dependent and mappings", + "type": "text" + }, + { + "bbox": [ + 452, + 336, + 460, + 345 + ], + "score": 0.8, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 334, + 505, + 346 + ], + "score": 1.0, + "content": ", for which", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 345, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 505, + 357 + ], + "score": 1.0, + "content": "dynamic attention does not hold. First, we show a simple 2-dimensional example, and then we show", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 356, + 246, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 246, + 368 + ], + "score": 1.0, + "content": "the general case of such examples.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 106, + 334, + 505, + 368 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 255, + 379, + 356, + 452 + ], + "lines": [ + { + "bbox": [ + 257, + 380, + 356, + 452 + ], + "spans": [ + { + "bbox": [ + 257, + 380, + 356, + 452 + ], + "score": 0.59, + "content": "\\prod _ { \\substack { h _ { 0 } = \\hat { x } } } ^ { y } h _ { 2 } = \\hat { x } + 2 \\hat { y }", + "type": "interline_equation", + "image_path": "b0f441cd12d6c4942d10f158b5b3baab44b607c6ac98f4f4863b6c4b09544de8.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 255, + 379, + 356, + 415.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 255, + 415.5, + 356, + 452.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "image", + "bbox": [], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 105, + 461, + 505, + 485 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 474 + ], + "score": 1.0, + "content": "Figure 6: An example for node representations that are linearly dependent, for which DPGAT cannot", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 472, + 417, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 319, + 485 + ], + "score": 1.0, + "content": "compute dynamic attention, because no query vector", + "type": "text" + }, + { + "bbox": [ + 319, + 472, + 349, + 484 + ], + "score": 0.92, + "content": "q \\in \\mathbb { R } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 473, + 401, + 485 + ], + "score": 1.0, + "content": "can “select”", + "type": "text" + }, + { + "bbox": [ + 401, + 473, + 413, + 484 + ], + "score": 0.89, + "content": "\\pmb { h } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 473, + 417, + 485 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 409, + 508 + ], + "lines": [ + { + "bbox": [ + 106, + 494, + 409, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 332, + 510 + ], + "score": 1.0, + "content": "Consider the following linearly dependent set of vectors", + "type": "text" + }, + { + "bbox": [ + 333, + 496, + 363, + 507 + ], + "score": 0.89, + "content": "\\mathbb { K } = \\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 494, + 409, + 510 + ], + "score": 1.0, + "content": "(Figure 6):", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 106, + 494, + 409, + 510 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 276, + 510, + 335, + 552 + ], + "lines": [ + { + "bbox": [ + 276, + 510, + 335, + 552 + ], + "spans": [ + { + "bbox": [ + 276, + 510, + 335, + 552 + ], + "score": 0.88, + "content": "\\begin{array} { l } { { \\pmb { h } } _ { 0 } = { \\pmb { \\hat { x } } } } \\\\ { { \\pmb { h } } _ { 1 } = { \\pmb { \\hat { x } } } + { \\pmb { \\hat { y } } } } \\\\ { { \\pmb { h } } _ { 2 } = { \\pmb { \\hat { x } } } + 2 { \\pmb { \\hat { y } } } } \\end{array}", + "type": "interline_equation", + "image_path": "a79e708698104ffef2ca0bc52c70500f9ea1b7a34cbabe95cef0ae1a5d18972f.jpg" + } + ] + } + ], + "index": 21.5, + "virtual_lines": [ + { + "bbox": [ + 276, + 510, + 335, + 531.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 276, + 531.0, + 335, + 552.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 555, + 505, + 577 + ], + "lines": [ + { + "bbox": [ + 104, + 552, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 104, + 552, + 133, + 569 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 556, + 141, + 565 + ], + "score": 0.83, + "content": "\\hat { \\pmb x }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 552, + 159, + 569 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 159, + 556, + 167, + 567 + ], + "score": 0.85, + "content": "\\hat { \\pmb { y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 552, + 329, + 569 + ], + "score": 1.0, + "content": "are the cartesian unit vectors. We define", + "type": "text" + }, + { + "bbox": [ + 330, + 555, + 383, + 567 + ], + "score": 0.93, + "content": "\\beta \\in \\{ 0 , 1 , 2 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 552, + 426, + 569 + ], + "score": 1.0, + "content": "to express", + "type": "text" + }, + { + "bbox": [ + 427, + 555, + 480, + 567 + ], + "score": 0.93, + "content": "\\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 552, + 506, + 569 + ], + "score": 1.0, + "content": "using", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 566, + 191, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 191, + 578 + ], + "score": 1.0, + "content": "the same expression:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 104, + 552, + 506, + 578 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 276, + 576, + 336, + 590 + ], + "lines": [ + { + "bbox": [ + 276, + 576, + 336, + 590 + ], + "spans": [ + { + "bbox": [ + 276, + 576, + 336, + 590 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\pmb { h } _ { \\beta } = \\hat { \\pmb { x } } + \\beta \\hat { \\pmb y } } \\end{array}", + "type": "interline_equation", + "image_path": "575a2fa2b0ec9c91e579d6ce367d482b1b99ee7811ede8beb83e8868904546f6.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 276, + 576, + 336, + 590 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 501, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 589, + 498, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 122, + 605 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 122, + 591, + 149, + 603 + ], + "score": 0.92, + "content": "\\pmb q \\in \\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 589, + 492, + 605 + ], + "score": 1.0, + "content": "be any query vector. For brevity, we define the unscaled dot-product attention score as", + "type": "text" + }, + { + "bbox": [ + 492, + 594, + 498, + 601 + ], + "score": 0.75, + "content": "s", + "type": "inline_equation" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 589, + 498, + 605 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 248, + 605, + 364, + 622 + ], + "lines": [ + { + "bbox": [ + 248, + 605, + 364, + 622 + ], + "spans": [ + { + "bbox": [ + 248, + 605, + 364, + 622 + ], + "score": 0.93, + "content": "s \\left( \\mathbf { q } , \\boldsymbol { h } _ { \\beta } \\right) = e \\left( \\mathbf { q } , \\boldsymbol { h } _ { \\beta } \\right) \\cdot \\sqrt { d _ { k } }", + "type": "interline_equation", + "image_path": "821407b7531cafa4a14e43215bbe169255bfb4789c144f517180083a952b491d.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 248, + 605, + 364, + 622 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 624, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 135, + 637 + ], + "score": 1.0, + "content": "Where", + "type": "text" + }, + { + "bbox": [ + 136, + 626, + 141, + 634 + ], + "score": 0.79, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 623, + 505, + 637 + ], + "score": 1.0, + "content": "is the attention score function of DPGAT (Equation (13)). The (unscaled) attention score", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 635, + 234, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 142, + 648 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 142, + 637, + 149, + 647 + ], + "score": 0.8, + "content": "\\pmb q", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 635, + 167, + 648 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 168, + 635, + 221, + 648 + ], + "score": 0.93, + "content": "\\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 635, + 234, + 648 + ], + "score": 1.0, + "content": "is:", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 106, + 623, + 505, + 648 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 197, + 649, + 413, + 734 + ], + "lines": [ + { + "bbox": [ + 197, + 649, + 413, + 734 + ], + "spans": [ + { + "bbox": [ + 197, + 649, + 413, + 734 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { s \\left( \\pmb { q } , \\pmb { h } _ { \\beta } \\right) = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { h } _ { \\beta } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\left( \\hat { \\pmb { x } } + \\beta \\pmb { \\hat { y } } \\right) ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { x } } ^ { \\top } \\pmb { K } + \\beta \\pmb { \\hat { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { x } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } + \\beta \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { \\hat { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\end{array}", + "type": "interline_equation", + "image_path": "2f076942dab3212541d4295ee604f1a329f957ef523d751087c6b0cd4b687f9c.jpg" + } + ] + } + ], + "index": 32.5, + "virtual_lines": [ + { + "bbox": [ + 197, + 649, + 413, + 663.1666666666666 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 197, + 663.1666666666666, + 413, + 677.3333333333333 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 197, + 677.3333333333333, + 413, + 691.4999999999999 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 197, + 691.4999999999999, + 413, + 705.6666666666665 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 197, + 705.6666666666665, + 413, + 719.8333333333331 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 197, + 719.8333333333331, + 413, + 733.9999999999998 + ], + "spans": [], + "index": 35 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 80, + 505, + 111 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 507, + 98 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 164, + 98 + ], + "score": 1.0, + "content": "The first term", + "type": "text" + }, + { + "bbox": [ + 165, + 80, + 236, + 96 + ], + "score": 0.9, + "content": "\\left( { { \\pmb q } ^ { \\top } } { \\pmb Q } \\right) \\left( \\hat { { \\pmb x } } ^ { \\top } { \\pmb K } \\right) ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 79, + 320, + 98 + ], + "score": 1.0, + "content": "is unconditioned on", + "type": "text" + }, + { + "bbox": [ + 320, + 84, + 327, + 95 + ], + "score": 0.82, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 79, + 435, + 98 + ], + "score": 1.0, + "content": ", and thus shared for every", + "type": "text" + }, + { + "bbox": [ + 435, + 84, + 448, + 96 + ], + "score": 0.88, + "content": "h _ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 79, + 507, + 98 + ], + "score": 1.0, + "content": ". Let us focus", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 103, + 95, + 397, + 112 + ], + "spans": [ + { + "bbox": [ + 103, + 95, + 184, + 112 + ], + "score": 1.0, + "content": "on the second term", + "type": "text" + }, + { + "bbox": [ + 185, + 96, + 264, + 112 + ], + "score": 0.88, + "content": "\\beta \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 95, + 279, + 112 + ], + "score": 1.0, + "content": ". I f", + "type": "text" + }, + { + "bbox": [ + 279, + 95, + 369, + 112 + ], + "score": 0.9, + "content": "\\left( { { \\pmb q } ^ { \\top } } { \\pmb Q } \\right) \\left( \\hat { \\pmb y } ^ { \\top } { \\pmb K } \\right) ^ { \\top } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 95, + 397, + 112 + ], + "score": 1.0, + "content": ", then:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 116, + 349, + 130 + ], + "lines": [ + { + "bbox": [ + 262, + 116, + 349, + 130 + ], + "spans": [ + { + "bbox": [ + 262, + 116, + 349, + 130 + ], + "score": 0.9, + "content": "e \\left( \\pmb { q } , \\pmb { h } _ { 2 } \\right) > e \\left( \\pmb { q } , \\pmb { h } _ { 1 } \\right)", + "type": "interline_equation", + "image_path": "6f622f464bc3ef858200573754232286c63e8529d0ef86327050512c98b3b6bb.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 262, + 116, + 349, + 130 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 141, + 255, + 157 + ], + "lines": [ + { + "bbox": [ + 103, + 140, + 252, + 158 + ], + "spans": [ + { + "bbox": [ + 103, + 140, + 162, + 158 + ], + "score": 1.0, + "content": "Otherwise, if", + "type": "text" + }, + { + "bbox": [ + 162, + 141, + 252, + 157 + ], + "score": 0.9, + "content": "\\left( { { { q } ^ { \\top } } { { Q } } } \\right) \\left( { \\hat { { { y } } } ^ { \\top } } K \\right) ^ { \\top } \\le 0", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 161, + 349, + 175 + ], + "lines": [ + { + "bbox": [ + 262, + 161, + 349, + 175 + ], + "spans": [ + { + "bbox": [ + 262, + 161, + 349, + 175 + ], + "score": 0.91, + "content": "e \\left( \\pmb { q } , \\pmb { h } _ { 0 } \\right) \\geq e \\left( \\pmb { q } , \\pmb { h } _ { 1 } \\right)", + "type": "interline_equation", + "image_path": "6d653bd6bf5e32345687239b6ae3b5ad8201768e8ab8d25482a8e6250c8af878.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 262, + 161, + 349, + 175 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 180, + 504, + 203 + ], + "lines": [ + { + "bbox": [ + 106, + 180, + 506, + 192 + ], + "spans": [ + { + "bbox": [ + 106, + 180, + 185, + 192 + ], + "score": 1.0, + "content": "Thus, for any query", + "type": "text" + }, + { + "bbox": [ + 185, + 182, + 192, + 191 + ], + "score": 0.75, + "content": "\\pmb q", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 180, + 226, + 192 + ], + "score": 1.0, + "content": ", the key", + "type": "text" + }, + { + "bbox": [ + 226, + 181, + 239, + 191 + ], + "score": 0.87, + "content": "\\pmb { h } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 180, + 506, + 192 + ], + "score": 1.0, + "content": "can never get the highest score, and thus cannot be “selected”. That", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 191, + 423, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 191, + 149, + 204 + ], + "score": 1.0, + "content": "is, the key", + "type": "text" + }, + { + "bbox": [ + 149, + 191, + 162, + 202 + ], + "score": 0.88, + "content": "\\pmb { h } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 191, + 238, + 204 + ], + "score": 1.0, + "content": "cannot satisfy that", + "type": "text" + }, + { + "bbox": [ + 238, + 191, + 274, + 203 + ], + "score": 0.92, + "content": "e \\left( \\pmb q , \\pmb h _ { 1 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 191, + 423, + 204 + ], + "score": 1.0, + "content": "is strictly greater than any other key.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 107, + 207, + 505, + 230 + ], + "lines": [ + { + "bbox": [ + 105, + 206, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 204, + 221 + ], + "score": 1.0, + "content": "In the general case, let", + "type": "text" + }, + { + "bbox": [ + 204, + 207, + 259, + 219 + ], + "score": 0.93, + "content": "\\boldsymbol { h } _ { 0 } , \\boldsymbol { h } _ { 1 } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 206, + 394, + 221 + ], + "score": 1.0, + "content": "be some non-zero vectors , and", + "type": "text" + }, + { + "bbox": [ + 394, + 208, + 402, + 218 + ], + "score": 0.81, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 206, + 506, + 221 + ], + "score": 1.0, + "content": "is some scalar such that", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 218, + 155, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 150, + 230 + ], + "score": 0.9, + "content": "0 < \\lambda < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 218, + 155, + 231 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 235, + 335, + 247 + ], + "lines": [ + { + "bbox": [ + 107, + 235, + 335, + 248 + ], + "spans": [ + { + "bbox": [ + 107, + 235, + 335, + 248 + ], + "score": 1.0, + "content": "Consider the following linearly dependent set of vectors:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 252, + 400, + 266 + ], + "lines": [ + { + "bbox": [ + 210, + 252, + 400, + 266 + ], + "spans": [ + { + "bbox": [ + 210, + 252, + 400, + 266 + ], + "score": 0.92, + "content": "\\mathbb { K } = \\mathbb { Q } = \\{ \\beta \\pmb { h } _ { 1 } + \\left( 1 - \\beta \\right) \\pmb { h } _ { 0 } \\ \\lvert \\ \\beta \\in \\{ 0 , \\lambda , 1 \\} \\}", + "type": "interline_equation", + "image_path": "e8623e4e53f6695803304e77a9cbc56900b6b95c8cdcc5a075a4efdbe5d2ad8f.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 210, + 252, + 400, + 266 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 276, + 308, + 289 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 309, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 165, + 290 + ], + "score": 1.0, + "content": "For any query", + "type": "text" + }, + { + "bbox": [ + 165, + 277, + 192, + 289 + ], + "score": 0.92, + "content": "\\pmb q \\in \\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 276, + 209, + 290 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 210, + 277, + 264, + 289 + ], + "score": 0.94, + "content": "\\beta \\in \\{ 0 , \\lambda , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 276, + 309, + 290 + ], + "score": 1.0, + "content": "we define:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 293, + 396, + 309 + ], + "lines": [ + { + "bbox": [ + 216, + 293, + 396, + 309 + ], + "spans": [ + { + "bbox": [ + 216, + 293, + 396, + 309 + ], + "score": 0.9, + "content": "s \\left( \\mathbf { q } , \\beta \\right) = e \\left( \\mathbf { q } , \\left( \\beta \\pmb { h } _ { 1 } + \\left( 1 - \\beta \\right) \\pmb { h } _ { 0 } \\right) \\right) \\cdot \\sqrt { d _ { k } }", + "type": "interline_equation", + "image_path": "d12de243a09f1d853bf35794af7dc29a626a2511b35a198be04de03bb86289bc.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 216, + 293, + 396, + 309 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 314, + 374, + 326 + ], + "lines": [ + { + "bbox": [ + 105, + 312, + 374, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 135, + 328 + ], + "score": 1.0, + "content": "Where", + "type": "text" + }, + { + "bbox": [ + 135, + 317, + 141, + 324 + ], + "score": 0.77, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 312, + 374, + 328 + ], + "score": 1.0, + "content": "is the attention score function of DPGAT (Equation (13)).", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 150, + 342 + ], + "lines": [ + { + "bbox": [ + 106, + 330, + 151, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 151, + 344 + ], + "score": 1.0, + "content": "Therefore:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 182, + 345, + 428, + 447 + ], + "lines": [ + { + "bbox": [ + 182, + 345, + 428, + 447 + ], + "spans": [ + { + "bbox": [ + 182, + 345, + 428, + 447 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { s \\left( \\boldsymbol { q } , \\beta \\right) = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\left( \\beta h _ { 1 } + \\left( 1 - \\beta \\right) h _ { 0 } \\right) ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta h _ { 1 } ^ { \\top } \\boldsymbol { K } + \\left( 1 - \\beta \\right) h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta h _ { 1 } ^ { \\top } \\boldsymbol { K } + h _ { 0 } ^ { \\top } \\boldsymbol { K } - \\beta h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta \\left( h _ { 1 } ^ { \\top } \\boldsymbol { K } - h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) + h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\beta \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( h _ { 1 } ^ { \\top } \\boldsymbol { K } - h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } + \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\end{array}", + "type": "interline_equation", + "image_path": "7c0be19659c61b65b84ce7167e42f5a668f14cc33005c85e5794f45a13743e70.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 182, + 345, + 428, + 379.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 182, + 379.0, + 428, + 413.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 182, + 413.0, + 428, + 447.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 455, + 246, + 471 + ], + "lines": [ + { + "bbox": [ + 103, + 455, + 244, + 472 + ], + "spans": [ + { + "bbox": [ + 103, + 455, + 115, + 472 + ], + "score": 1.0, + "content": "If", + "type": "text" + }, + { + "bbox": [ + 116, + 455, + 244, + 472 + ], + "score": 0.79, + "content": "\\left( { { q } ^ { \\top } } Q \\right) \\left( h _ { 1 } ^ { \\top } K - h _ { 0 } ^ { \\top } K \\right) ^ { \\top } > 0 .", + "type": "inline_equation" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 476, + 349, + 490 + ], + "lines": [ + { + "bbox": [ + 262, + 476, + 349, + 490 + ], + "spans": [ + { + "bbox": [ + 262, + 476, + 349, + 490 + ], + "score": 0.87, + "content": "e \\left( \\pmb q , \\pmb h _ { 1 } \\right) > e \\left( \\pmb q , \\pmb h _ { \\lambda } \\right)", + "type": "interline_equation", + "image_path": "deb4a1feed49bffb7979a98bb27d95c6c0feddf5d85bf23d47b3859a4361738c.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 262, + 476, + 349, + 490 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 501, + 291, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 289, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 162, + 518 + ], + "score": 1.0, + "content": "Otherwise, if", + "type": "text" + }, + { + "bbox": [ + 162, + 500, + 289, + 518 + ], + "score": 0.87, + "content": "\\left( { \\pmb q } ^ { \\top } { \\pmb Q } \\right) \\left( { \\pmb h } _ { 1 } ^ { \\top } { \\pmb K } - { \\pmb h } _ { 0 } ^ { \\top } { \\pmb K } \\right) ^ { \\top } \\le 0 ,", + "type": "inline_equation" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 521, + 349, + 536 + ], + "lines": [ + { + "bbox": [ + 262, + 521, + 349, + 536 + ], + "spans": [ + { + "bbox": [ + 262, + 521, + 349, + 536 + ], + "score": 0.9, + "content": "e \\left( \\pmb { q } , \\pmb { h } _ { 0 } \\right) \\geq e \\left( \\pmb { q } , \\pmb { h } _ { \\lambda } \\right)", + "type": "interline_equation", + "image_path": "3a6c3dc477df43b80ebcf249caa56c66f7e2f99c20283154da574924e7bb9913.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 262, + 521, + 349, + 536 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 185, + 553 + ], + "score": 1.0, + "content": "Thus, for any query", + "type": "text" + }, + { + "bbox": [ + 185, + 542, + 192, + 552 + ], + "score": 0.75, + "content": "\\pmb q", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 539, + 226, + 553 + ], + "score": 1.0, + "content": ", the key", + "type": "text" + }, + { + "bbox": [ + 226, + 541, + 240, + 551 + ], + "score": 0.89, + "content": "h _ { \\lambda }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 539, + 380, + 553 + ], + "score": 1.0, + "content": "cannot be selected. That is, the key", + "type": "text" + }, + { + "bbox": [ + 380, + 541, + 393, + 551 + ], + "score": 0.88, + "content": "\\mathbf { \\delta } _ { h _ { \\lambda } }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 539, + 467, + 553 + ], + "score": 1.0, + "content": "cannot satisfy that", + "type": "text" + }, + { + "bbox": [ + 468, + 540, + 505, + 552 + ], + "score": 0.92, + "content": "\\mathbf { \\nabla } _ { \\boldsymbol { \\mathcal { C } } } \\left( \\mathbf { \\nabla } _ { \\mathbf { \\boldsymbol { q } } } , \\boldsymbol { h } _ { \\lambda } \\right)", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 378, + 563 + ], + "score": 1.0, + "content": "is strictly greater than any other key. Therefore, there are mappings", + "type": "text" + }, + { + "bbox": [ + 378, + 553, + 385, + 563 + ], + "score": 0.84, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 550, + 505, + 563 + ], + "score": 1.0, + "content": ", for which dynamic attention", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 562, + 165, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 165, + 573 + ], + "score": 1.0, + "content": "does not hold.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 578, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 506, + 592 + ], + "score": 1.0, + "content": "While we prove that GATv2 computes dynamic attention (Appendix B) for any set of node represen-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 589, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 136, + 603 + ], + "score": 1.0, + "content": "tations", + "type": "text" + }, + { + "bbox": [ + 136, + 590, + 166, + 601 + ], + "score": 0.92, + "content": "\\mathbb { K } = \\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 589, + 379, + 603 + ], + "score": 1.0, + "content": ", there are sets of node representations and mappings", + "type": "text" + }, + { + "bbox": [ + 380, + 592, + 388, + 601 + ], + "score": 0.83, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 589, + 505, + 603 + ], + "score": 1.0, + "content": "for which dynamic attention", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 600, + 397, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 397, + 613 + ], + "score": 1.0, + "content": "does not hold for DPGAT. Thus, DPGAT is strictly weaker than GATv2.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "title", + "bbox": [ + 107, + 625, + 241, + 637 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 243, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 243, + 639 + ], + "score": 1.0, + "content": "G.3 EMPIRICAL EVALUATION", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 645, + 505, + 669 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "Here we repeat the experiments of Section 4 with DPGAT. We remind that DPGAT is strictly weaker", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 657, + 337, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 337, + 670 + ], + "score": 1.0, + "content": "than our proposed GATv2 (see a proof in Appendix G.1).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "title", + "bbox": [ + 108, + 684, + 272, + 698 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 275, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 275, + 699 + ], + "score": 1.0, + "content": "H STATISTICAL SIGNIFICANCE", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "Here we report the statistical significance of the strongest GATv2 and GAT models of the experiments", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 720, + 195, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 195, + 732 + ], + "score": 1.0, + "content": "reported in Section 4.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 80, + 505, + 111 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 507, + 98 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 164, + 98 + ], + "score": 1.0, + "content": "The first term", + "type": "text" + }, + { + "bbox": [ + 165, + 80, + 236, + 96 + ], + "score": 0.9, + "content": "\\left( { { \\pmb q } ^ { \\top } } { \\pmb Q } \\right) \\left( \\hat { { \\pmb x } } ^ { \\top } { \\pmb K } \\right) ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 79, + 320, + 98 + ], + "score": 1.0, + "content": "is unconditioned on", + "type": "text" + }, + { + "bbox": [ + 320, + 84, + 327, + 95 + ], + "score": 0.82, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 79, + 435, + 98 + ], + "score": 1.0, + "content": ", and thus shared for every", + "type": "text" + }, + { + "bbox": [ + 435, + 84, + 448, + 96 + ], + "score": 0.88, + "content": "h _ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 79, + 507, + 98 + ], + "score": 1.0, + "content": ". Let us focus", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 103, + 95, + 397, + 112 + ], + "spans": [ + { + "bbox": [ + 103, + 95, + 184, + 112 + ], + "score": 1.0, + "content": "on the second term", + "type": "text" + }, + { + "bbox": [ + 185, + 96, + 264, + 112 + ], + "score": 0.88, + "content": "\\beta \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 95, + 279, + 112 + ], + "score": 1.0, + "content": ". I f", + "type": "text" + }, + { + "bbox": [ + 279, + 95, + 369, + 112 + ], + "score": 0.9, + "content": "\\left( { { \\pmb q } ^ { \\top } } { \\pmb Q } \\right) \\left( \\hat { \\pmb y } ^ { \\top } { \\pmb K } \\right) ^ { \\top } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 95, + 397, + 112 + ], + "score": 1.0, + "content": ", then:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 103, + 79, + 507, + 112 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 116, + 349, + 130 + ], + "lines": [ + { + "bbox": [ + 262, + 116, + 349, + 130 + ], + "spans": [ + { + "bbox": [ + 262, + 116, + 349, + 130 + ], + "score": 0.9, + "content": "e \\left( \\pmb { q } , \\pmb { h } _ { 2 } \\right) > e \\left( \\pmb { q } , \\pmb { h } _ { 1 } \\right)", + "type": "interline_equation", + "image_path": "6f622f464bc3ef858200573754232286c63e8529d0ef86327050512c98b3b6bb.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 262, + 116, + 349, + 130 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 141, + 255, + 157 + ], + "lines": [ + { + "bbox": [ + 103, + 140, + 252, + 158 + ], + "spans": [ + { + "bbox": [ + 103, + 140, + 162, + 158 + ], + "score": 1.0, + "content": "Otherwise, if", + "type": "text" + }, + { + "bbox": [ + 162, + 141, + 252, + 157 + ], + "score": 0.9, + "content": "\\left( { { { q } ^ { \\top } } { { Q } } } \\right) \\left( { \\hat { { { y } } } ^ { \\top } } K \\right) ^ { \\top } \\le 0", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 103, + 140, + 252, + 158 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 161, + 349, + 175 + ], + "lines": [ + { + "bbox": [ + 262, + 161, + 349, + 175 + ], + "spans": [ + { + "bbox": [ + 262, + 161, + 349, + 175 + ], + "score": 0.91, + "content": "e \\left( \\pmb { q } , \\pmb { h } _ { 0 } \\right) \\geq e \\left( \\pmb { q } , \\pmb { h } _ { 1 } \\right)", + "type": "interline_equation", + "image_path": "6d653bd6bf5e32345687239b6ae3b5ad8201768e8ab8d25482a8e6250c8af878.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 262, + 161, + 349, + 175 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 180, + 504, + 203 + ], + "lines": [ + { + "bbox": [ + 106, + 180, + 506, + 192 + ], + "spans": [ + { + "bbox": [ + 106, + 180, + 185, + 192 + ], + "score": 1.0, + "content": "Thus, for any query", + "type": "text" + }, + { + "bbox": [ + 185, + 182, + 192, + 191 + ], + "score": 0.75, + "content": "\\pmb q", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 180, + 226, + 192 + ], + "score": 1.0, + "content": ", the key", + "type": "text" + }, + { + "bbox": [ + 226, + 181, + 239, + 191 + ], + "score": 0.87, + "content": "\\pmb { h } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 180, + 506, + 192 + ], + "score": 1.0, + "content": "can never get the highest score, and thus cannot be “selected”. That", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 191, + 423, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 191, + 149, + 204 + ], + "score": 1.0, + "content": "is, the key", + "type": "text" + }, + { + "bbox": [ + 149, + 191, + 162, + 202 + ], + "score": 0.88, + "content": "\\pmb { h } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 191, + 238, + 204 + ], + "score": 1.0, + "content": "cannot satisfy that", + "type": "text" + }, + { + "bbox": [ + 238, + 191, + 274, + 203 + ], + "score": 0.92, + "content": "e \\left( \\pmb q , \\pmb h _ { 1 } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 191, + 423, + 204 + ], + "score": 1.0, + "content": "is strictly greater than any other key.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 106, + 180, + 506, + 204 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 207, + 505, + 230 + ], + "lines": [ + { + "bbox": [ + 105, + 206, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 204, + 221 + ], + "score": 1.0, + "content": "In the general case, let", + "type": "text" + }, + { + "bbox": [ + 204, + 207, + 259, + 219 + ], + "score": 0.93, + "content": "\\boldsymbol { h } _ { 0 } , \\boldsymbol { h } _ { 1 } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 206, + 394, + 221 + ], + "score": 1.0, + "content": "be some non-zero vectors , and", + "type": "text" + }, + { + "bbox": [ + 394, + 208, + 402, + 218 + ], + "score": 0.81, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 206, + 506, + 221 + ], + "score": 1.0, + "content": "is some scalar such that", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 218, + 155, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 150, + 230 + ], + "score": 0.9, + "content": "0 < \\lambda < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 218, + 155, + 231 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 206, + 506, + 231 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 235, + 335, + 247 + ], + "lines": [ + { + "bbox": [ + 107, + 235, + 335, + 248 + ], + "spans": [ + { + "bbox": [ + 107, + 235, + 335, + 248 + ], + "score": 1.0, + "content": "Consider the following linearly dependent set of vectors:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 107, + 235, + 335, + 248 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 252, + 400, + 266 + ], + "lines": [ + { + "bbox": [ + 210, + 252, + 400, + 266 + ], + "spans": [ + { + "bbox": [ + 210, + 252, + 400, + 266 + ], + "score": 0.92, + "content": "\\mathbb { K } = \\mathbb { Q } = \\{ \\beta \\pmb { h } _ { 1 } + \\left( 1 - \\beta \\right) \\pmb { h } _ { 0 } \\ \\lvert \\ \\beta \\in \\{ 0 , \\lambda , 1 \\} \\}", + "type": "interline_equation", + "image_path": "e8623e4e53f6695803304e77a9cbc56900b6b95c8cdcc5a075a4efdbe5d2ad8f.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 210, + 252, + 400, + 266 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 276, + 308, + 289 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 309, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 165, + 290 + ], + "score": 1.0, + "content": "For any query", + "type": "text" + }, + { + "bbox": [ + 165, + 277, + 192, + 289 + ], + "score": 0.92, + "content": "\\pmb q \\in \\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 276, + 209, + 290 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 210, + 277, + 264, + 289 + ], + "score": 0.94, + "content": "\\beta \\in \\{ 0 , \\lambda , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 276, + 309, + 290 + ], + "score": 1.0, + "content": "we define:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 276, + 309, + 290 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 293, + 396, + 309 + ], + "lines": [ + { + "bbox": [ + 216, + 293, + 396, + 309 + ], + "spans": [ + { + "bbox": [ + 216, + 293, + 396, + 309 + ], + "score": 0.9, + "content": "s \\left( \\mathbf { q } , \\beta \\right) = e \\left( \\mathbf { q } , \\left( \\beta \\pmb { h } _ { 1 } + \\left( 1 - \\beta \\right) \\pmb { h } _ { 0 } \\right) \\right) \\cdot \\sqrt { d _ { k } }", + "type": "interline_equation", + "image_path": "d12de243a09f1d853bf35794af7dc29a626a2511b35a198be04de03bb86289bc.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 216, + 293, + 396, + 309 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 314, + 374, + 326 + ], + "lines": [ + { + "bbox": [ + 105, + 312, + 374, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 135, + 328 + ], + "score": 1.0, + "content": "Where", + "type": "text" + }, + { + "bbox": [ + 135, + 317, + 141, + 324 + ], + "score": 0.77, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 312, + 374, + 328 + ], + "score": 1.0, + "content": "is the attention score function of DPGAT (Equation (13)).", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 312, + 374, + 328 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 150, + 342 + ], + "lines": [ + { + "bbox": [ + 106, + 330, + 151, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 151, + 344 + ], + "score": 1.0, + "content": "Therefore:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 106, + 330, + 151, + 344 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 182, + 345, + 428, + 447 + ], + "lines": [ + { + "bbox": [ + 182, + 345, + 428, + 447 + ], + "spans": [ + { + "bbox": [ + 182, + 345, + 428, + 447 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { s \\left( \\boldsymbol { q } , \\beta \\right) = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\left( \\beta h _ { 1 } + \\left( 1 - \\beta \\right) h _ { 0 } \\right) ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta h _ { 1 } ^ { \\top } \\boldsymbol { K } + \\left( 1 - \\beta \\right) h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta h _ { 1 } ^ { \\top } \\boldsymbol { K } + h _ { 0 } ^ { \\top } \\boldsymbol { K } - \\beta h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta \\left( h _ { 1 } ^ { \\top } \\boldsymbol { K } - h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) + h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\beta \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( h _ { 1 } ^ { \\top } \\boldsymbol { K } - h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } + \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\end{array}", + "type": "interline_equation", + "image_path": "7c0be19659c61b65b84ce7167e42f5a668f14cc33005c85e5794f45a13743e70.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 182, + 345, + 428, + 379.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 182, + 379.0, + 428, + 413.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 182, + 413.0, + 428, + 447.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 455, + 246, + 471 + ], + "lines": [ + { + "bbox": [ + 103, + 455, + 244, + 472 + ], + "spans": [ + { + "bbox": [ + 103, + 455, + 115, + 472 + ], + "score": 1.0, + "content": "If", + "type": "text" + }, + { + "bbox": [ + 116, + 455, + 244, + 472 + ], + "score": 0.79, + "content": "\\left( { { q } ^ { \\top } } Q \\right) \\left( h _ { 1 } ^ { \\top } K - h _ { 0 } ^ { \\top } K \\right) ^ { \\top } > 0 .", + "type": "inline_equation" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 103, + 455, + 244, + 472 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 476, + 349, + 490 + ], + "lines": [ + { + "bbox": [ + 262, + 476, + 349, + 490 + ], + "spans": [ + { + "bbox": [ + 262, + 476, + 349, + 490 + ], + "score": 0.87, + "content": "e \\left( \\pmb q , \\pmb h _ { 1 } \\right) > e \\left( \\pmb q , \\pmb h _ { \\lambda } \\right)", + "type": "interline_equation", + "image_path": "deb4a1feed49bffb7979a98bb27d95c6c0feddf5d85bf23d47b3859a4361738c.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 262, + 476, + 349, + 490 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 501, + 291, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 289, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 162, + 518 + ], + "score": 1.0, + "content": "Otherwise, if", + "type": "text" + }, + { + "bbox": [ + 162, + 500, + 289, + 518 + ], + "score": 0.87, + "content": "\\left( { \\pmb q } ^ { \\top } { \\pmb Q } \\right) \\left( { \\pmb h } _ { 1 } ^ { \\top } { \\pmb K } - { \\pmb h } _ { 0 } ^ { \\top } { \\pmb K } \\right) ^ { \\top } \\le 0 ,", + "type": "inline_equation" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 500, + 289, + 518 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 521, + 349, + 536 + ], + "lines": [ + { + "bbox": [ + 262, + 521, + 349, + 536 + ], + "spans": [ + { + "bbox": [ + 262, + 521, + 349, + 536 + ], + "score": 0.9, + "content": "e \\left( \\pmb { q } , \\pmb { h } _ { 0 } \\right) \\geq e \\left( \\pmb { q } , \\pmb { h } _ { \\lambda } \\right)", + "type": "interline_equation", + "image_path": "3a6c3dc477df43b80ebcf249caa56c66f7e2f99c20283154da574924e7bb9913.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 262, + 521, + 349, + 536 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 185, + 553 + ], + "score": 1.0, + "content": "Thus, for any query", + "type": "text" + }, + { + "bbox": [ + 185, + 542, + 192, + 552 + ], + "score": 0.75, + "content": "\\pmb q", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 539, + 226, + 553 + ], + "score": 1.0, + "content": ", the key", + "type": "text" + }, + { + "bbox": [ + 226, + 541, + 240, + 551 + ], + "score": 0.89, + "content": "h _ { \\lambda }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 539, + 380, + 553 + ], + "score": 1.0, + "content": "cannot be selected. That is, the key", + "type": "text" + }, + { + "bbox": [ + 380, + 541, + 393, + 551 + ], + "score": 0.88, + "content": "\\mathbf { \\delta } _ { h _ { \\lambda } }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 539, + 467, + 553 + ], + "score": 1.0, + "content": "cannot satisfy that", + "type": "text" + }, + { + "bbox": [ + 468, + 540, + 505, + 552 + ], + "score": 0.92, + "content": "\\mathbf { \\nabla } _ { \\boldsymbol { \\mathcal { C } } } \\left( \\mathbf { \\nabla } _ { \\mathbf { \\boldsymbol { q } } } , \\boldsymbol { h } _ { \\lambda } \\right)", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 378, + 563 + ], + "score": 1.0, + "content": "is strictly greater than any other key. Therefore, there are mappings", + "type": "text" + }, + { + "bbox": [ + 378, + 553, + 385, + 563 + ], + "score": 0.84, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 550, + 505, + 563 + ], + "score": 1.0, + "content": ", for which dynamic attention", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 562, + 165, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 165, + 573 + ], + "score": 1.0, + "content": "does not hold.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 539, + 505, + 573 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 578, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 506, + 592 + ], + "score": 1.0, + "content": "While we prove that GATv2 computes dynamic attention (Appendix B) for any set of node represen-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 589, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 136, + 603 + ], + "score": 1.0, + "content": "tations", + "type": "text" + }, + { + "bbox": [ + 136, + 590, + 166, + 601 + ], + "score": 0.92, + "content": "\\mathbb { K } = \\mathbb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 589, + 379, + 603 + ], + "score": 1.0, + "content": ", there are sets of node representations and mappings", + "type": "text" + }, + { + "bbox": [ + 380, + 592, + 388, + 601 + ], + "score": 0.83, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 589, + 505, + 603 + ], + "score": 1.0, + "content": "for which dynamic attention", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 600, + 397, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 397, + 613 + ], + "score": 1.0, + "content": "does not hold for DPGAT. Thus, DPGAT is strictly weaker than GATv2.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 578, + 506, + 613 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 625, + 241, + 637 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 243, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 243, + 639 + ], + "score": 1.0, + "content": "G.3 EMPIRICAL EVALUATION", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 645, + 505, + 669 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "Here we repeat the experiments of Section 4 with DPGAT. We remind that DPGAT is strictly weaker", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 657, + 337, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 337, + 670 + ], + "score": 1.0, + "content": "than our proposed GATv2 (see a proof in Appendix G.1).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 106, + 646, + 505, + 670 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 684, + 272, + 698 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 275, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 275, + 699 + ], + "score": 1.0, + "content": "H STATISTICAL SIGNIFICANCE", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "Here we report the statistical significance of the strongest GATv2 and GAT models of the experiments", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 720, + 195, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 195, + 732 + ], + "score": 1.0, + "content": "reported in Section 4.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 708, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 104, + 502, + 286 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 104, + 502, + 286 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 104, + 502, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 502, + 286 + ], + "score": 0.971, + "type": "image", + "image_path": "0f010beab366169ea6709265d4f0cb66efdab3acea2d3044a4c540073bc64969.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 104, + 502, + 164.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 164.66666666666666, + 502, + 225.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 225.33333333333331, + 502, + 286.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 295, + 506, + 318 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "score": 1.0, + "content": "Figure 7: Test accuracy compared to the noise ratio: GATv2 and DPGAT are more robust to structural", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 306, + 492, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 492, + 319 + ], + "score": 1.0, + "content": "noise compared to GAT. Each point is an average of 10 runs, error bars show standard deviation.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "table", + "bbox": [ + 218, + 403, + 393, + 486 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 371, + 505, + 395 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 372, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 505, + 384 + ], + "score": 1.0, + "content": "Table 11: Accuracy (5 runs±stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 383, + 463, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 267, + 396 + ], + "score": 1.0, + "content": "test sets, using GAT’s hyperparameters.", + "type": "text" + }, + { + "bbox": [ + 267, + 384, + 273, + 395 + ], + "score": 0.4, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 383, + 463, + 396 + ], + "score": 1.0, + "content": "– previously reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 218, + 403, + 393, + 486 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 218, + 403, + 393, + 486 + ], + "spans": [ + { + "bbox": [ + 218, + 403, + 393, + 486 + ], + "score": 0.975, + "html": "
ModelSeenProjUnseenProj
No- AttentionGCN+87.2±1.581.4±2.3
GIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
DPGAT88.0±0.881.5±1.2
GATv288.0±1.182.8±1.7
", + "type": "table", + "image_path": "3729df644b0459c1ee338b4a1a91e066025f669f0d9f19b18d44d5b1a4c61729.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 218, + 403, + 393, + 416.8333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 218, + 416.8333333333333, + 393, + 430.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 218, + 430.66666666666663, + 393, + 444.49999999999994 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 218, + 444.49999999999994, + 393, + 458.33333333333326 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 218, + 458.33333333333326, + 393, + 472.1666666666666 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 218, + 472.1666666666666, + 393, + 485.9999999999999 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 262, + 571, + 274, + 581 + ], + "lines": [ + { + "bbox": [ + 261, + 570, + 276, + 583 + ], + "spans": [ + { + "bbox": [ + 261, + 570, + 276, + 583 + ], + "score": 1.0, + "content": "(a)", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "table", + "bbox": [ + 112, + 585, + 423, + 707 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 537, + 506, + 561 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 537, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 505, + 551 + ], + "score": 1.0, + "content": "Table 12: Average accuracy (Table 12a) and ROC-AUC (Table 12b) in node-prediction datasets (10", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 549, + 493, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 318, + 562 + ], + "score": 1.0, + "content": "runs±std). In all datasets, GATv2 outperforms GAT.", + "type": "text" + }, + { + "bbox": [ + 318, + 550, + 333, + 561 + ], + "score": 0.4, + "content": "\\dagger -", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 549, + 493, + 562 + ], + "score": 1.0, + "content": "previously reported by Hu et al. (2020).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "table_body", + "bbox": [ + 112, + 585, + 423, + 707 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 112, + 585, + 423, + 707 + ], + "spans": [ + { + "bbox": [ + 112, + 585, + 423, + 707 + ], + "score": 0.971, + "html": "
ModelAttn. Headsogbn-arxivogbn-productsogbn-mag
GCN+071.74±0.2978.97±0.3330.43±0.25
GraphSAGE†071.49±0.2778.70±0.3631.53±0.15
GAT171.59±0.3879.04±1.5432.20±1.46
871.54±0.3077.23±2.3731.75±1.60
DPGAT171.52±0.1776.49±0.7832.77±0.80
871.48±0.2673.53±0.4727.74±9.97
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.44
871.87±0.2578.46±2.4532.52±0.39
", + "type": "table", + "image_path": "d2839ce14e5ce1e13e771cee93c7a8c132d130d89277bbce60c1fdfd0f5ae22c.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 112, + 585, + 423, + 625.6666666666666 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 112, + 625.6666666666666, + 423, + 666.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 112, + 666.3333333333333, + 423, + 706.9999999999999 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 15.75 + }, + { + "type": "text", + "bbox": [ + 459, + 571, + 471, + 581 + ], + "lines": [ + { + "bbox": [ + 457, + 570, + 473, + 583 + ], + "spans": [ + { + "bbox": [ + 457, + 570, + 473, + 583 + ], + "score": 1.0, + "content": "(b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "table", + "bbox": [ + 432, + 586, + 498, + 707 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 432, + 586, + 498, + 707 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 432, + 586, + 498, + 707 + ], + "spans": [ + { + "bbox": [ + 432, + 586, + 498, + 707 + ], + "score": 0.955, + "html": "
ogbn-proteins
72.51±0.3577.68±0.20
70.77±5.7978.63±1.62
63.47±2.7972.88±0.59
77.23±3.3279.52±0.55
", + "type": "table", + "image_path": "4f095a99f6be29e043b71844162f7fb398e89952546ed4f514e85aa387699ba7.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 432, + 586, + 498, + 707 + ], + "spans": [], + "index": 19 + } + ] + } + ], + "index": 19 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "19", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 104, + 502, + 286 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 104, + 502, + 286 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 104, + 502, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 502, + 286 + ], + "score": 0.971, + "type": "image", + "image_path": "0f010beab366169ea6709265d4f0cb66efdab3acea2d3044a4c540073bc64969.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 104, + 502, + 164.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 164.66666666666666, + 502, + 225.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 225.33333333333331, + 502, + 286.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 295, + 506, + 318 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 505, + 308 + ], + "score": 1.0, + "content": "Figure 7: Test accuracy compared to the noise ratio: GATv2 and DPGAT are more robust to structural", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 306, + 492, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 492, + 319 + ], + "score": 1.0, + "content": "noise compared to GAT. Each point is an average of 10 runs, error bars show standard deviation.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "table", + "bbox": [ + 218, + 403, + 393, + 486 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 371, + 505, + 395 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 372, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 505, + 384 + ], + "score": 1.0, + "content": "Table 11: Accuracy (5 runs±stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 383, + 463, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 267, + 396 + ], + "score": 1.0, + "content": "test sets, using GAT’s hyperparameters.", + "type": "text" + }, + { + "bbox": [ + 267, + 384, + 273, + 395 + ], + "score": 0.4, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 383, + 463, + 396 + ], + "score": 1.0, + "content": "– previously reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 218, + 403, + 393, + 486 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 218, + 403, + 393, + 486 + ], + "spans": [ + { + "bbox": [ + 218, + 403, + 393, + 486 + ], + "score": 0.975, + "html": "
ModelSeenProjUnseenProj
No- AttentionGCN+87.2±1.581.4±2.3
GIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
DPGAT88.0±0.881.5±1.2
GATv288.0±1.182.8±1.7
", + "type": "table", + "image_path": "3729df644b0459c1ee338b4a1a91e066025f669f0d9f19b18d44d5b1a4c61729.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 218, + 403, + 393, + 416.8333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 218, + 416.8333333333333, + 393, + 430.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 218, + 430.66666666666663, + 393, + 444.49999999999994 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 218, + 444.49999999999994, + 393, + 458.33333333333326 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 218, + 458.33333333333326, + 393, + 472.1666666666666 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 218, + 472.1666666666666, + 393, + 485.9999999999999 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 262, + 571, + 274, + 581 + ], + "lines": [ + { + "bbox": [ + 261, + 570, + 276, + 583 + ], + "spans": [ + { + "bbox": [ + 261, + 570, + 276, + 583 + ], + "score": 1.0, + "content": "(a)", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 261, + 570, + 276, + 583 + ] + }, + { + "type": "table", + "bbox": [ + 112, + 585, + 423, + 707 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 537, + 506, + 561 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 537, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 505, + 551 + ], + "score": 1.0, + "content": "Table 12: Average accuracy (Table 12a) and ROC-AUC (Table 12b) in node-prediction datasets (10", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 549, + 493, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 318, + 562 + ], + "score": 1.0, + "content": "runs±std). In all datasets, GATv2 outperforms GAT.", + "type": "text" + }, + { + "bbox": [ + 318, + 550, + 333, + 561 + ], + "score": 0.4, + "content": "\\dagger -", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 549, + 493, + 562 + ], + "score": 1.0, + "content": "previously reported by Hu et al. (2020).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "table_body", + "bbox": [ + 112, + 585, + 423, + 707 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 112, + 585, + 423, + 707 + ], + "spans": [ + { + "bbox": [ + 112, + 585, + 423, + 707 + ], + "score": 0.971, + "html": "
ModelAttn. Headsogbn-arxivogbn-productsogbn-mag
GCN+071.74±0.2978.97±0.3330.43±0.25
GraphSAGE†071.49±0.2778.70±0.3631.53±0.15
GAT171.59±0.3879.04±1.5432.20±1.46
871.54±0.3077.23±2.3731.75±1.60
DPGAT171.52±0.1776.49±0.7832.77±0.80
871.48±0.2673.53±0.4727.74±9.97
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.44
871.87±0.2578.46±2.4532.52±0.39
", + "type": "table", + "image_path": "d2839ce14e5ce1e13e771cee93c7a8c132d130d89277bbce60c1fdfd0f5ae22c.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 112, + 585, + 423, + 625.6666666666666 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 112, + 625.6666666666666, + 423, + 666.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 112, + 666.3333333333333, + 423, + 706.9999999999999 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 15.75 + }, + { + "type": "text", + "bbox": [ + 459, + 571, + 471, + 581 + ], + "lines": [ + { + "bbox": [ + 457, + 570, + 473, + 583 + ], + "spans": [ + { + "bbox": [ + 457, + 570, + 473, + 583 + ], + "score": 1.0, + "content": "(b)", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 457, + 570, + 473, + 583 + ] + }, + { + "type": "table", + "bbox": [ + 432, + 586, + 498, + 707 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 432, + 586, + 498, + 707 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 432, + 586, + 498, + 707 + ], + "spans": [ + { + "bbox": [ + 432, + 586, + 498, + 707 + ], + "score": 0.955, + "html": "
ogbn-proteins
72.51±0.3577.68±0.20
70.77±5.7978.63±1.62
63.47±2.7972.88±0.59
77.23±3.3279.52±0.55
", + "type": "table", + "image_path": "4f095a99f6be29e043b71844162f7fb398e89952546ed4f514e85aa387699ba7.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 432, + 586, + 498, + 707 + ], + "spans": [], + "index": 19 + } + ] + } + ], + "index": 19 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 115, + 134, + 495, + 410 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 505, + 124 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 79, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 319, + 93 + ], + "score": 1.0, + "content": "Table 13: Average error rates (lower is better), 5 runs", + "type": "text" + }, + { + "bbox": [ + 319, + 81, + 329, + 91 + ], + "score": 0.76, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 79, + 505, + 93 + ], + "score": 1.0, + "content": "standard deviation for each property, on the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 90, + 506, + 104 + ], + "spans": [ + { + "bbox": [ + 105, + 90, + 506, + 104 + ], + "score": 1.0, + "content": "QM9 dataset. The best result among GAT, GATv2 and DPGAT is marked in bold; the globally best", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 102, + 505, + 115 + ], + "spans": [ + { + "bbox": [ + 104, + 102, + 343, + 115 + ], + "score": 1.0, + "content": "result among all GNNs is marked in bold and underline.", + "type": "text" + }, + { + "bbox": [ + 343, + 103, + 349, + 114 + ], + "score": 0.71, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 102, + 505, + 115 + ], + "score": 1.0, + "content": "was previously tuned and reported by", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 112, + 197, + 125 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 197, + 125 + ], + "score": 1.0, + "content": "Brockschmidt (2020).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 115, + 134, + 495, + 410 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 134, + 495, + 410 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 495, + 410 + ], + "score": 0.576, + "html": "
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
DPGAT8h2.63±0.094.37±0.131.44±0.071.40±0.032.10±0.0732.59±34.7711.66±1.00
DPGAT1h3.20±0.178.35±0.781.71±0.032.17±0.142.88±0.1225.21±2.8665.79±39.84
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13 ±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.259.05±1.217.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h28.10±16.4518.71±23.365.62±0.8110.80±2.185.37±0.26
GAT8ht7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.364.11±0.273.11±0.14 1.48±0.87+134.1% +0%
DPGAT1h12.93±1.7013.83±2.556.37±0.28
DPGAT8h6.95±0.3213.32±2.39 7.09±0.5914.42±1.95 7.30±0.666.52±0.613.76±0.213.28±1.16 1.18±0.33+77.9% -9.7%
GATv21h5.23±0.36
GATv28h10.19±0.6322.56±17.4615.04±4.5822.94±17.342.46±0.65+91.6%
6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
", + "type": "table", + "image_path": "424d7314241e36f659f72bb1b181d1a8a5dd7c63cd357d190033bd0de94283ab.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 115, + 134, + 495, + 226.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 115, + 226.0, + 495, + 318.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 115, + 318.0, + 495, + 410.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "image", + "bbox": [ + 108, + 417, + 496, + 588 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 417, + 496, + 588 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 417, + 496, + 588 + ], + "spans": [ + { + "bbox": [ + 108, + 417, + 496, + 588 + ], + "score": 0.719, + "type": "image", + "image_path": "a9109830d2a6f78861878e365b1f89ef70a5d28fb8341f7508b1337e0e7390a2.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 108, + 417, + 496, + 474.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 108, + 474.0, + 496, + 531.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 108, + 531.0, + 496, + 588.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 597, + 504, + 631 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "Figure 8: Test accuracy and statistical significance compared to the noise ratio: GATv2 is more robust", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 609, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 620 + ], + "score": 1.0, + "content": "to structural noise compared to GAT. Each point is an average of 10 runs, error bars show standard", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 618, + 149, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 149, + 632 + ], + "score": 1.0, + "content": "deviation.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "table", + "bbox": [ + 240, + 670, + 369, + 730 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 638, + 505, + 662 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 639, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 214, + 650 + ], + "score": 1.0, + "content": "Table 14: Accuracy (5 runs", + "type": "text" + }, + { + "bbox": [ + 214, + 640, + 222, + 649 + ], + "score": 0.37, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 639, + 505, + 650 + ], + "score": 1.0, + "content": "stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 650, + 464, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 267, + 663 + ], + "score": 1.0, + "content": "test sets, using GAT’s hyperparameters.", + "type": "text" + }, + { + "bbox": [ + 267, + 650, + 273, + 662 + ], + "score": 0.26, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 650, + 464, + 663 + ], + "score": 1.0, + "content": "– previously reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "table_body", + "bbox": [ + 240, + 670, + 369, + 730 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 240, + 670, + 369, + 730 + ], + "spans": [ + { + "bbox": [ + 240, + 670, + 369, + 730 + ], + "score": 0.969, + "html": "
ModelSeenProjUnseenProj
GAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
p-value0.0480.049
", + "type": "table", + "image_path": "f92804deb5a5a72316e07d2be17398049bcfdc6e7f42478ed3d24944fc5f89a8.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 240, + 670, + 369, + 700.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 240, + 700.0, + 369, + 730.0 + ], + "spans": [], + "index": 16 + } + ] + } + ], + "index": 14.5 + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "20", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 115, + 134, + 495, + 410 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 505, + 124 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 79, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 319, + 93 + ], + "score": 1.0, + "content": "Table 13: Average error rates (lower is better), 5 runs", + "type": "text" + }, + { + "bbox": [ + 319, + 81, + 329, + 91 + ], + "score": 0.76, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 79, + 505, + 93 + ], + "score": 1.0, + "content": "standard deviation for each property, on the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 90, + 506, + 104 + ], + "spans": [ + { + "bbox": [ + 105, + 90, + 506, + 104 + ], + "score": 1.0, + "content": "QM9 dataset. The best result among GAT, GATv2 and DPGAT is marked in bold; the globally best", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 102, + 505, + 115 + ], + "spans": [ + { + "bbox": [ + 104, + 102, + 343, + 115 + ], + "score": 1.0, + "content": "result among all GNNs is marked in bold and underline.", + "type": "text" + }, + { + "bbox": [ + 343, + 103, + 349, + 114 + ], + "score": 0.71, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 102, + 505, + 115 + ], + "score": 1.0, + "content": "was previously tuned and reported by", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 112, + 197, + 125 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 197, + 125 + ], + "score": 1.0, + "content": "Brockschmidt (2020).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 115, + 134, + 495, + 410 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 134, + 495, + 410 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 495, + 410 + ], + "score": 0.576, + "html": "
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
DPGAT8h2.63±0.094.37±0.131.44±0.071.40±0.032.10±0.0732.59±34.7711.66±1.00
DPGAT1h3.20±0.178.35±0.781.71±0.032.17±0.142.88±0.1225.21±2.8665.79±39.84
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13 ±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.259.05±1.217.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h28.10±16.4518.71±23.365.62±0.8110.80±2.185.37±0.26
GAT8ht7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.364.11±0.273.11±0.14 1.48±0.87+134.1% +0%
DPGAT1h12.93±1.7013.83±2.556.37±0.28
DPGAT8h6.95±0.3213.32±2.39 7.09±0.5914.42±1.95 7.30±0.666.52±0.613.76±0.213.28±1.16 1.18±0.33+77.9% -9.7%
GATv21h5.23±0.36
GATv28h10.19±0.6322.56±17.4615.04±4.5822.94±17.342.46±0.65+91.6%
6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
", + "type": "table", + "image_path": "424d7314241e36f659f72bb1b181d1a8a5dd7c63cd357d190033bd0de94283ab.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 115, + 134, + 495, + 226.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 115, + 226.0, + 495, + 318.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 115, + 318.0, + 495, + 410.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "image", + "bbox": [ + 108, + 417, + 496, + 588 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 417, + 496, + 588 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 417, + 496, + 588 + ], + "spans": [ + { + "bbox": [ + 108, + 417, + 496, + 588 + ], + "score": 0.719, + "type": "image", + "image_path": "a9109830d2a6f78861878e365b1f89ef70a5d28fb8341f7508b1337e0e7390a2.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 108, + 417, + 496, + 474.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 108, + 474.0, + 496, + 531.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 108, + 531.0, + 496, + 588.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 597, + 504, + 631 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "Figure 8: Test accuracy and statistical significance compared to the noise ratio: GATv2 is more robust", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 609, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 620 + ], + "score": 1.0, + "content": "to structural noise compared to GAT. Each point is an average of 10 runs, error bars show standard", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 618, + 149, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 149, + 632 + ], + "score": 1.0, + "content": "deviation.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "table", + "bbox": [ + 240, + 670, + 369, + 730 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 638, + 505, + 662 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 639, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 214, + 650 + ], + "score": 1.0, + "content": "Table 14: Accuracy (5 runs", + "type": "text" + }, + { + "bbox": [ + 214, + 640, + 222, + 649 + ], + "score": 0.37, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 639, + 505, + 650 + ], + "score": 1.0, + "content": "stdev) on VARMISUSE. GATv2 is more accurate than all GNNs in both", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 650, + 464, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 267, + 663 + ], + "score": 1.0, + "content": "test sets, using GAT’s hyperparameters.", + "type": "text" + }, + { + "bbox": [ + 267, + 650, + 273, + 662 + ], + "score": 0.26, + "content": "\\dagger", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 650, + 464, + 663 + ], + "score": 1.0, + "content": "– previously reported by Brockschmidt (2020).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "table_body", + "bbox": [ + 240, + 670, + 369, + 730 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 240, + 670, + 369, + 730 + ], + "spans": [ + { + "bbox": [ + 240, + 670, + 369, + 730 + ], + "score": 0.969, + "html": "
ModelSeenProjUnseenProj
GAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
p-value0.0480.049
", + "type": "table", + "image_path": "f92804deb5a5a72316e07d2be17398049bcfdc6e7f42478ed3d24944fc5f89a8.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 240, + 670, + 369, + 700.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 240, + 700.0, + 369, + 730.0 + ], + "spans": [], + "index": 16 + } + ] + } + ], + "index": 14.5 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 268, + 119, + 343, + 179 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 131, + 99, + 478, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 131, + 98, + 479, + 111 + ], + "spans": [ + { + "bbox": [ + 131, + 98, + 479, + 111 + ], + "score": 1.0, + "content": "Table 15: Accuracy (100 runs±stdev) on Pubmed. GATv2 is more accurate than GAT.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 268, + 119, + 343, + 179 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 268, + 119, + 343, + 179 + ], + "spans": [ + { + "bbox": [ + 268, + 119, + 343, + 179 + ], + "score": 0.966, + "html": "
Model Accuracy
GAT 78.1±0.59
GATv2 78.5±0.38
p-value <0.0001
", + "type": "table", + "image_path": "ae6edb6e2129580eec5eabb7cf79e1349f07ea69aed3861db50a0c1f92647b30.jpg" + } + ] + } + ], + "index": 1.5, + "virtual_lines": [ + { + "bbox": [ + 268, + 119, + 343, + 149.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 268, + 149.0, + 343, + 179.0 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 0.75 + }, + { + "type": "table", + "bbox": [ + 160, + 254, + 375, + 329 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 222, + 504, + 245 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "score": 1.0, + "content": "Table 16: Average accuracy (Table 16a) and ROC-AUC (Table 16b) in node-prediction datasets (30", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 234, + 355, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 355, + 245 + ], + "score": 1.0, + "content": "runs±std). We report on the best GAT / GATv2 from Table 2.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "table_body", + "bbox": [ + 160, + 254, + 375, + 329 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 160, + 254, + 375, + 329 + ], + "spans": [ + { + "bbox": [ + 160, + 254, + 375, + 329 + ], + "score": 0.887, + "html": "
(a)
Modelogbn-arxivogbn-productsogbn-mag
GAT71.65±0.3879.04±1.5432.36±1.10
GATv271.93±0.3580.63±0.7033.01±0.41
p-value0.0022<0.00010.0018
", + "type": "table", + "image_path": "e60aeca2e5c137c25a39bb0dc4a695e603324776802578000663be86004f00e4.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 160, + 254, + 375, + 269.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 160, + 269.0, + 375, + 284.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 160, + 284.0, + 375, + 299.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 160, + 299.0, + 375, + 314.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 160, + 314.0, + 375, + 329.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "text", + "bbox": [ + 459, + 255, + 471, + 264 + ], + "lines": [ + { + "bbox": [ + 457, + 254, + 473, + 267 + ], + "spans": [ + { + "bbox": [ + 457, + 254, + 473, + 267 + ], + "score": 1.0, + "content": "(b)", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "table", + "bbox": [ + 432, + 270, + 499, + 329 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 432, + 270, + 499, + 329 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 432, + 270, + 499, + 329 + ], + "spans": [ + { + "bbox": [ + 432, + 270, + 499, + 329 + ], + "score": 0.936, + "html": "
ogbn-proteins
78.29 ±1.59
78.96±1.19 0.0349
", + "type": "table", + "image_path": "d8f3b42ff0eb2de1cdfe97747979be35ef59ee5710e92ab69714a713d6fb8ef3.jpg" + } + ] + } + ], + "index": 10.5, + "virtual_lines": [ + { + "bbox": [ + 432, + 270, + 499, + 299.5 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 432, + 299.5, + 499, + 329.0 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 10.5 + }, + { + "type": "table", + "bbox": [ + 193, + 405, + 353, + 489 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 371, + 505, + 395 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 372, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 205, + 384 + ], + "score": 1.0, + "content": "Table 17: Average Hits", + "type": "text" + }, + { + "bbox": [ + 206, + 372, + 226, + 383 + ], + "score": 0.61, + "content": "\\ @ 5 0", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 372, + 506, + 384 + ], + "score": 1.0, + "content": "(Table 17a) and mean reciprocal rank (MRR) (Table 17b) in link-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "prediction benchmarks from OGB (30 runs±std). We report on the best GAT / GATv2 from Table 8.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "table_body", + "bbox": [ + 193, + 405, + 353, + 489 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 193, + 405, + 353, + 489 + ], + "spans": [ + { + "bbox": [ + 193, + 405, + 353, + 489 + ], + "score": 0.903, + "html": "
(a)
Modelogbl-collab w/o val edges w/ val edges
GAT42.24±2.2646.02±4.09
GATv243.82±2.2449.06±2.50
p-value0.00430.0005
", + "type": "table", + "image_path": "5e6e77c7ba89b1b9e1e81278e67fdd659e0c880740ef27e8eaf40978318fb84c.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 193, + 405, + 353, + 419.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 193, + 419.0, + 353, + 433.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 193, + 433.0, + 353, + 447.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 193, + 447.0, + 353, + 461.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 193, + 461.0, + 353, + 475.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 193, + 475.0, + 353, + 489.0 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 15.5 + }, + { + "type": "table", + "bbox": [ + 394, + 405, + 462, + 489 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 394, + 405, + 462, + 489 + ], + "group_id": 5, + "lines": [ + { + "bbox": [ + 394, + 405, + 462, + 489 + ], + "spans": [ + { + "bbox": [ + 394, + 405, + 462, + 489 + ], + "score": 0.9, + "html": "
(b)
ogbl-citation2
79.91±0.13
80.20±0.62
0.0075
", + "type": "table", + "image_path": "ffa94db2bcee9b7ee4a6ee106d5e4ed602398ac903c08866ca032fe1aa1c88ab.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 394, + 405, + 462, + 489 + ], + "spans": [], + "index": 21 + } + ] + } + ], + "index": 21 + }, + { + "type": "table", + "bbox": [ + 123, + 564, + 488, + 711 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 533, + 503, + 556 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 105, + 532, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 321, + 547 + ], + "score": 1.0, + "content": "Table 18: Average error rates (lower is better), 20 runs", + "type": "text" + }, + { + "bbox": [ + 321, + 534, + 332, + 544 + ], + "score": 0.61, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 532, + 505, + 547 + ], + "score": 1.0, + "content": "standard deviation for each property, on the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 544, + 381, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 381, + 557 + ], + "score": 1.0, + "content": "QM9 dataset. We report on GAT and GATv2 with 8 attention heads.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "table_body", + "bbox": [ + 123, + 564, + 488, + 711 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 123, + 564, + 488, + 711 + ], + "spans": [ + { + "bbox": [ + 123, + 564, + 488, + 711 + ], + "score": 0.742, + "html": "
Predicted Property
Model1234567
GAT2.74±0.084.73±0.401.47±0.061.53±0.062.44±0.6055.21±42.3325.36±31.42
GATv22.67±0.084.28±0.231.43±0.051.51±0.072.21±0.0816.64± 1.1713.61±1.68
p-value0.0043<0.00010.01380.16910.04870.00010.0516
Predicted Property
Model8910111213
GAT GATv27.36±0.87 6.13±0.596.79±0.86 6.33±0.827.36±0.936.69±0.864.10±0.291.51±0.84
6.37±0.865.95±0.623.66±0.291.09±0.85
p-value<0.00010.04580.00060.0017<0.00010.0621
", + "type": "table", + "image_path": "30a6b0c5270661bcc3a376e72370e2090929f0518cf0b3e006ae13065159783b.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 123, + 564, + 488, + 613.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 123, + 613.0, + 488, + 662.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 123, + 662.0, + 488, + 711.0 + ], + "spans": [], + "index": 26 + } + ] + } + ], + "index": 23.75 + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 268, + 119, + 343, + 179 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 131, + 99, + 478, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 131, + 98, + 479, + 111 + ], + "spans": [ + { + "bbox": [ + 131, + 98, + 479, + 111 + ], + "score": 1.0, + "content": "Table 15: Accuracy (100 runs±stdev) on Pubmed. GATv2 is more accurate than GAT.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 268, + 119, + 343, + 179 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 268, + 119, + 343, + 179 + ], + "spans": [ + { + "bbox": [ + 268, + 119, + 343, + 179 + ], + "score": 0.966, + "html": "
Model Accuracy
GAT 78.1±0.59
GATv2 78.5±0.38
p-value <0.0001
", + "type": "table", + "image_path": "ae6edb6e2129580eec5eabb7cf79e1349f07ea69aed3861db50a0c1f92647b30.jpg" + } + ] + } + ], + "index": 1.5, + "virtual_lines": [ + { + "bbox": [ + 268, + 119, + 343, + 149.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 268, + 149.0, + 343, + 179.0 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 0.75 + }, + { + "type": "table", + "bbox": [ + 160, + 254, + 375, + 329 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 222, + 504, + 245 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "score": 1.0, + "content": "Table 16: Average accuracy (Table 16a) and ROC-AUC (Table 16b) in node-prediction datasets (30", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 234, + 355, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 355, + 245 + ], + "score": 1.0, + "content": "runs±std). We report on the best GAT / GATv2 from Table 2.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "table_body", + "bbox": [ + 160, + 254, + 375, + 329 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 160, + 254, + 375, + 329 + ], + "spans": [ + { + "bbox": [ + 160, + 254, + 375, + 329 + ], + "score": 0.887, + "html": "
(a)
Modelogbn-arxivogbn-productsogbn-mag
GAT71.65±0.3879.04±1.5432.36±1.10
GATv271.93±0.3580.63±0.7033.01±0.41
p-value0.0022<0.00010.0018
", + "type": "table", + "image_path": "e60aeca2e5c137c25a39bb0dc4a695e603324776802578000663be86004f00e4.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 160, + 254, + 375, + 269.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 160, + 269.0, + 375, + 284.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 160, + 284.0, + 375, + 299.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 160, + 299.0, + 375, + 314.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 160, + 314.0, + 375, + 329.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "text", + "bbox": [ + 459, + 255, + 471, + 264 + ], + "lines": [ + { + "bbox": [ + 457, + 254, + 473, + 267 + ], + "spans": [ + { + "bbox": [ + 457, + 254, + 473, + 267 + ], + "score": 1.0, + "content": "(b)", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 457, + 254, + 473, + 267 + ] + }, + { + "type": "table", + "bbox": [ + 432, + 270, + 499, + 329 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 432, + 270, + 499, + 329 + ], + "group_id": 4, + "lines": [ + { + "bbox": [ + 432, + 270, + 499, + 329 + ], + "spans": [ + { + "bbox": [ + 432, + 270, + 499, + 329 + ], + "score": 0.936, + "html": "
ogbn-proteins
78.29 ±1.59
78.96±1.19 0.0349
", + "type": "table", + "image_path": "d8f3b42ff0eb2de1cdfe97747979be35ef59ee5710e92ab69714a713d6fb8ef3.jpg" + } + ] + } + ], + "index": 10.5, + "virtual_lines": [ + { + "bbox": [ + 432, + 270, + 499, + 299.5 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 432, + 299.5, + 499, + 329.0 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 10.5 + }, + { + "type": "table", + "bbox": [ + 193, + 405, + 353, + 489 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 371, + 505, + 395 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 372, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 205, + 384 + ], + "score": 1.0, + "content": "Table 17: Average Hits", + "type": "text" + }, + { + "bbox": [ + 206, + 372, + 226, + 383 + ], + "score": 0.61, + "content": "\\ @ 5 0", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 372, + 506, + 384 + ], + "score": 1.0, + "content": "(Table 17a) and mean reciprocal rank (MRR) (Table 17b) in link-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "prediction benchmarks from OGB (30 runs±std). We report on the best GAT / GATv2 from Table 8.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "table_body", + "bbox": [ + 193, + 405, + 353, + 489 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 193, + 405, + 353, + 489 + ], + "spans": [ + { + "bbox": [ + 193, + 405, + 353, + 489 + ], + "score": 0.903, + "html": "
(a)
Modelogbl-collab w/o val edges w/ val edges
GAT42.24±2.2646.02±4.09
GATv243.82±2.2449.06±2.50
p-value0.00430.0005
", + "type": "table", + "image_path": "5e6e77c7ba89b1b9e1e81278e67fdd659e0c880740ef27e8eaf40978318fb84c.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 193, + 405, + 353, + 419.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 193, + 419.0, + 353, + 433.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 193, + 433.0, + 353, + 447.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 193, + 447.0, + 353, + 461.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 193, + 461.0, + 353, + 475.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 193, + 475.0, + 353, + 489.0 + ], + "spans": [], + "index": 20 + } + ] + } + ], + "index": 15.5 + }, + { + "type": "table", + "bbox": [ + 394, + 405, + 462, + 489 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 394, + 405, + 462, + 489 + ], + "group_id": 5, + "lines": [ + { + "bbox": [ + 394, + 405, + 462, + 489 + ], + "spans": [ + { + "bbox": [ + 394, + 405, + 462, + 489 + ], + "score": 0.9, + "html": "
(b)
ogbl-citation2
79.91±0.13
80.20±0.62
0.0075
", + "type": "table", + "image_path": "ffa94db2bcee9b7ee4a6ee106d5e4ed602398ac903c08866ca032fe1aa1c88ab.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 394, + 405, + 462, + 489 + ], + "spans": [], + "index": 21 + } + ] + } + ], + "index": 21 + }, + { + "type": "table", + "bbox": [ + 123, + 564, + 488, + 711 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 533, + 503, + 556 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 105, + 532, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 321, + 547 + ], + "score": 1.0, + "content": "Table 18: Average error rates (lower is better), 20 runs", + "type": "text" + }, + { + "bbox": [ + 321, + 534, + 332, + 544 + ], + "score": 0.61, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 532, + 505, + 547 + ], + "score": 1.0, + "content": "standard deviation for each property, on the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 544, + 381, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 381, + 557 + ], + "score": 1.0, + "content": "QM9 dataset. We report on GAT and GATv2 with 8 attention heads.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "table_body", + "bbox": [ + 123, + 564, + 488, + 711 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 123, + 564, + 488, + 711 + ], + "spans": [ + { + "bbox": [ + 123, + 564, + 488, + 711 + ], + "score": 0.742, + "html": "
Predicted Property
Model1234567
GAT2.74±0.084.73±0.401.47±0.061.53±0.062.44±0.6055.21±42.3325.36±31.42
GATv22.67±0.084.28±0.231.43±0.051.51±0.072.21±0.0816.64± 1.1713.61±1.68
p-value0.0043<0.00010.01380.16910.04870.00010.0516
Predicted Property
Model8910111213
GAT GATv27.36±0.87 6.13±0.596.79±0.86 6.33±0.827.36±0.936.69±0.864.10±0.291.51±0.84
6.37±0.865.95±0.623.66±0.291.09±0.85
p-value<0.00010.04580.00060.0017<0.00010.0621
", + "type": "table", + "image_path": "30a6b0c5270661bcc3a376e72370e2090929f0518cf0b3e006ae13065159783b.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 123, + 564, + 488, + 613.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 123, + 613.0, + 488, + 662.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 123, + 662.0, + 488, + 711.0 + ], + "spans": [], + "index": 26 + } + ] + } + ], + "index": 23.75 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 249, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 250, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 250, + 96 + ], + "score": 1.0, + "content": "I COMPLEXITY ANALYSIS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 109, + 327, + 122 + ], + "lines": [ + { + "bbox": [ + 105, + 109, + 328, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 109, + 328, + 123 + ], + "score": 1.0, + "content": "We repeat the definitions of GAT, GATv2 and DPGAT:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 127, + 136, + 252, + 148 + ], + "lines": [ + { + "bbox": [ + 127, + 135, + 252, + 150 + ], + "spans": [ + { + "bbox": [ + 127, + 135, + 252, + 150 + ], + "score": 1.0, + "content": "GAT (Velickovi ˇ c et al., 2018): ´", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 127, + 152, + 239, + 164 + ], + "lines": [ + { + "bbox": [ + 127, + 151, + 241, + 166 + ], + "spans": [ + { + "bbox": [ + 127, + 151, + 241, + 166 + ], + "score": 1.0, + "content": "GATv2 (our fixed version):", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 127, + 171, + 258, + 183 + ], + "lines": [ + { + "bbox": [ + 127, + 169, + 258, + 185 + ], + "spans": [ + { + "bbox": [ + 127, + 169, + 258, + 185 + ], + "score": 1.0, + "content": "DPGAT (Vaswani et al., 2017):", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 275, + 133, + 467, + 189 + ], + "lines": [ + { + "bbox": [ + 275, + 133, + 467, + 189 + ], + "spans": [ + { + "bbox": [ + 275, + 133, + 467, + 189 + ], + "score": 0.86, + "content": "\\begin{array} { r l } & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = \\mathrm { L e a k y R e L U } ( \\boldsymbol { a } ^ { \\top } \\cdot [ \\boldsymbol { W } \\boldsymbol { h } _ { i } | | \\boldsymbol { W } \\boldsymbol { h } _ { j } | ) } \\\\ & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = \\boldsymbol { a } ^ { \\top } \\mathrm { L e a k y R e L U } ( \\boldsymbol { W } \\cdot [ \\boldsymbol { h } _ { i } | | \\boldsymbol { h } _ { j } ] ) } \\\\ & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = ( ( \\boldsymbol { h } _ { i } ^ { \\top } \\boldsymbol { Q } ) \\cdot ( \\boldsymbol { h } _ { j } ^ { \\top } \\boldsymbol { K } ) ^ { \\top } ) / \\sqrt { d ^ { \\prime } } } \\end{array}", + "type": "interline_equation", + "image_path": "8a5e123ec02bb5ae8d73a9dbea17509e9c23670171b8eadac52309e582e12309.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 275, + 133, + 467, + 147.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 275, + 147.0, + 467, + 161.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 275, + 161.0, + 467, + 175.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 275, + 175.0, + 467, + 189.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 207, + 212, + 218 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 214, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 214, + 219 + ], + "score": 1.0, + "content": "I.1 TIME COMPLEXITY", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 233, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 232, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 246 + ], + "score": 1.0, + "content": "GAT As noted by Velickovi ˇ c et al. (2018), the time complexity of a single GAT head may be ´", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 245, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 159, + 257 + ], + "score": 1.0, + "content": "expressed as", + "type": "text" + }, + { + "bbox": [ + 159, + 245, + 235, + 257 + ], + "score": 0.91, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 245, + 506, + 257 + ], + "score": 1.0, + "content": ". Because of GAT’s static attention, this computation can be further", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 255, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 266, + 268 + ], + "score": 1.0, + "content": "optimized, by merging the linear layer", + "type": "text" + }, + { + "bbox": [ + 267, + 257, + 279, + 267 + ], + "score": 0.85, + "content": "\\mathbf { \\delta } _ { \\mathbf { a } _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 255, + 301, + 268 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 302, + 256, + 315, + 266 + ], + "score": 0.73, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 255, + 356, + 268 + ], + "score": 1.0, + "content": ", merging", + "type": "text" + }, + { + "bbox": [ + 356, + 257, + 368, + 267 + ], + "score": 0.85, + "content": "\\mathbf { a } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 255, + 390, + 268 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 391, + 256, + 404, + 266 + ], + "score": 0.75, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 255, + 505, + 268 + ], + "score": 1.0, + "content": ", and only then compute", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 266, + 223, + 282 + ], + "spans": [ + { + "bbox": [ + 107, + 266, + 155, + 281 + ], + "score": 0.93, + "content": "a _ { \\{ 1 , 2 \\} } ^ { \\top } W h _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 266, + 195, + 282 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 195, + 267, + 219, + 277 + ], + "score": 0.89, + "content": "i \\in \\mathcal V", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 266, + 223, + 282 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 288, + 505, + 336 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 410, + 302 + ], + "score": 1.0, + "content": "GATv2 require the same computational cost as GAT’s declared complexity:", + "type": "text" + }, + { + "bbox": [ + 410, + 289, + 487, + 301 + ], + "score": 0.91, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 288, + 505, + 302 + ], + "score": 1.0, + "content": ": we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 101, + 295, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 101, + 295, + 137, + 320 + ], + "score": 1.0, + "content": "denote", + "type": "text" + }, + { + "bbox": [ + 137, + 302, + 204, + 315 + ], + "score": 0.92, + "content": "W = [ W _ { 1 } \\Vert W _ { 2 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 295, + 236, + 320 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 236, + 300, + 289, + 313 + ], + "score": 0.93, + "content": "W _ { 1 } \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 295, + 308, + 320 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 308, + 300, + 339, + 314 + ], + "score": 0.9, + "content": "W _ { 2 } ^ { d ^ { \\prime } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 295, + 506, + 320 + ], + "score": 1.0, + "content": "contain the left half and right half of the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 314, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 154, + 326 + ], + "score": 1.0, + "content": "columns of", + "type": "text" + }, + { + "bbox": [ + 155, + 315, + 167, + 324 + ], + "score": 0.79, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 314, + 312, + 326 + ], + "score": 1.0, + "content": ", respectively. We can first compute", + "type": "text" + }, + { + "bbox": [ + 312, + 314, + 339, + 325 + ], + "score": 0.88, + "content": "W _ { 1 } h _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 314, + 357, + 326 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 357, + 314, + 384, + 326 + ], + "score": 0.92, + "content": "W _ { 2 } h _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 314, + 424, + 326 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 424, + 314, + 457, + 325 + ], + "score": 0.91, + "content": "i , j \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 314, + 505, + 326 + ], + "score": 1.0, + "content": ". This takes", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 322, + 156, + 338 + ], + "spans": [ + { + "bbox": [ + 107, + 324, + 151, + 337 + ], + "score": 0.92, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 322, + 156, + 338 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 504, + 365 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 504, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 192, + 356 + ], + "score": 1.0, + "content": "Then, for every edge", + "type": "text" + }, + { + "bbox": [ + 192, + 342, + 213, + 353 + ], + "score": 0.9, + "content": "( j , i )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 339, + 323, + 356 + ], + "score": 1.0, + "content": ", we compute LeakyReLU", + "type": "text" + }, + { + "bbox": [ + 323, + 341, + 381, + 353 + ], + "score": 0.9, + "content": "\\left( W \\cdot \\left[ \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { h } _ { j } \\right] \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 339, + 477, + 356 + ], + "score": 1.0, + "content": "using the precomputed", + "type": "text" + }, + { + "bbox": [ + 478, + 341, + 504, + 353 + ], + "score": 0.89, + "content": "W _ { 1 } h _ { i }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 351, + 397, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 123, + 366 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 353, + 151, + 365 + ], + "score": 0.92, + "content": "W _ { 2 } h _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 351, + 177, + 366 + ], + "score": 1.0, + "content": ", since", + "type": "text" + }, + { + "bbox": [ + 178, + 352, + 307, + 365 + ], + "score": 0.93, + "content": "W \\cdot [ h _ { i } \\| h _ { j } ] = W _ { 1 } h _ { i } + W _ { 2 } h _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 351, + 354, + 366 + ], + "score": 1.0, + "content": ". This takes", + "type": "text" + }, + { + "bbox": [ + 354, + 353, + 392, + 365 + ], + "score": 0.92, + "content": "\\mathcal { O } \\left( \\vert \\mathcal { E } \\vert d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 351, + 397, + 366 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 105, + 369, + 504, + 392 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 312, + 382 + ], + "score": 1.0, + "content": "Finally, computing the results of the linear layer", + "type": "text" + }, + { + "bbox": [ + 313, + 372, + 320, + 379 + ], + "score": 0.74, + "content": "^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 368, + 391, + 382 + ], + "score": 1.0, + "content": "takes additional", + "type": "text" + }, + { + "bbox": [ + 392, + 369, + 429, + 381 + ], + "score": 0.93, + "content": "\\mathcal { O } \\left( | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 368, + 505, + 382 + ], + "score": 1.0, + "content": "time, and overall", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 379, + 187, + 393 + ], + "spans": [ + { + "bbox": [ + 107, + 380, + 182, + 393 + ], + "score": 0.9, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + \\bar { | } \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 379, + 187, + 393 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 106, + 399, + 505, + 441 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 338, + 414 + ], + "score": 1.0, + "content": "DPGAT also takes the same time. We can first compute", + "type": "text" + }, + { + "bbox": [ + 338, + 399, + 362, + 412 + ], + "score": 0.92, + "content": "\\pmb { h } _ { i } ^ { \\top } \\pmb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 398, + 380, + 414 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 380, + 399, + 405, + 414 + ], + "score": 0.95, + "content": "\\pmb { h } _ { j } ^ { \\top } \\pmb { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 398, + 446, + 414 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 446, + 400, + 479, + 412 + ], + "score": 0.91, + "content": "i , j \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 398, + 506, + 414 + ], + "score": 1.0, + "content": ". This", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 414, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 104, + 414, + 129, + 430 + ], + "score": 1.0, + "content": "takes", + "type": "text" + }, + { + "bbox": [ + 129, + 415, + 173, + 428 + ], + "score": 0.93, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 414, + 287, + 430 + ], + "score": 1.0, + "content": ". Computing the dot-product", + "type": "text" + }, + { + "bbox": [ + 287, + 414, + 359, + 430 + ], + "score": 0.86, + "content": "\\left( \\pmb { h } _ { i } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { h } _ { j } ^ { \\top } \\pmb { K } \\right) ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 414, + 419, + 430 + ], + "score": 1.0, + "content": "for every edge", + "type": "text" + }, + { + "bbox": [ + 420, + 416, + 440, + 428 + ], + "score": 0.91, + "content": "( j , i )", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 414, + 506, + 430 + ], + "score": 1.0, + "content": "takes additional", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 428, + 296, + 441 + ], + "spans": [ + { + "bbox": [ + 107, + 428, + 145, + 441 + ], + "score": 0.93, + "content": "\\mathcal { O } \\left( | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 428, + 216, + 441 + ], + "score": 1.0, + "content": "time, and overall", + "type": "text" + }, + { + "bbox": [ + 216, + 428, + 292, + 441 + ], + "score": 0.92, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 428, + 296, + 441 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 107, + 460, + 246, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 461, + 247, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 247, + 473 + ], + "score": 1.0, + "content": "I.2 PARAMETRIC COMPLEXITY", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "table", + "bbox": [ + 181, + 493, + 430, + 541 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 181, + 493, + 430, + 541 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 181, + 493, + 430, + 541 + ], + "spans": [ + { + "bbox": [ + 181, + 493, + 430, + 541 + ], + "score": 0.975, + "html": "
GATGATv2DPGAT
Official2d'+dd'd'+ 2dd'2ddk +dd'
In our experiments2d'+dd'd'+dd'2dd'
", + "type": "table", + "image_path": "7f473a828617dbd5ae061ebb1adaf2898962cae8e0636392f224b1bdad27d710.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 181, + 493, + 430, + 509.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 181, + 509.0, + 430, + 525.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 181, + 525.0, + 430, + 541.0 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 110, + 547, + 500, + 560 + ], + "lines": [ + { + "bbox": [ + 109, + 546, + 501, + 562 + ], + "spans": [ + { + "bbox": [ + 109, + 546, + 501, + 562 + ], + "score": 1.0, + "content": "Table 19: Number of parameters for each GNN type, in a single layer and a single attention head.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 580, + 502, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 579, + 504, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 504, + 594 + ], + "score": 1.0, + "content": "All parametric costs are summarized in Table 19. All following calculations refer to a single layer", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 592, + 318, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 318, + 604 + ], + "score": 1.0, + "content": "having a single attention head, omitting bias vectors.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 107, + 612, + 504, + 636 + ], + "lines": [ + { + "bbox": [ + 104, + 609, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 104, + 609, + 265, + 627 + ], + "score": 1.0, + "content": "GAT has learned vector and a matrix:", + "type": "text" + }, + { + "bbox": [ + 266, + 611, + 304, + 624 + ], + "score": 0.92, + "content": "\\pmb { a } \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 609, + 324, + 627 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 325, + 611, + 376, + 624 + ], + "score": 0.92, + "content": "W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 609, + 430, + 627 + ], + "score": 1.0, + "content": ", thus overall", + "type": "text" + }, + { + "bbox": [ + 431, + 613, + 471, + 624 + ], + "score": 0.9, + "content": "2 d ^ { \\prime } + d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 609, + 506, + 627 + ], + "score": 1.0, + "content": "learned", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 625, + 155, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 155, + 636 + ], + "score": 1.0, + "content": "parameters.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 107, + 644, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 104, + 642, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 642, + 274, + 659 + ], + "score": 1.0, + "content": "GATv2 has a matrix that is twice larger:", + "type": "text" + }, + { + "bbox": [ + 275, + 644, + 329, + 656 + ], + "score": 0.92, + "content": "W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times 2 d }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 642, + 506, + 659 + ], + "score": 1.0, + "content": ", because it is applied on the concatenation", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 107, + 656, + 507, + 669 + ], + "spans": [ + { + "bbox": [ + 107, + 657, + 139, + 669 + ], + "score": 0.92, + "content": "[ h _ { i } | | h _ { j } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 656, + 344, + 669 + ], + "score": 1.0, + "content": ". Thus, the overall number of learned parameters is", + "type": "text" + }, + { + "bbox": [ + 344, + 657, + 383, + 667 + ], + "score": 0.92, + "content": "d ^ { \\prime } + 2 d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 656, + 507, + 669 + ], + "score": 1.0, + "content": ". However in our experiments,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 668, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 505, + 680 + ], + "score": 1.0, + "content": "to rule out the increased number of parameters over GAT as the source of empirical difference, we", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 678, + 428, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 155, + 691 + ], + "score": 1.0, + "content": "constrained", + "type": "text" + }, + { + "bbox": [ + 155, + 678, + 223, + 691 + ], + "score": 0.94, + "content": "W = [ W ^ { \\prime } | | W ^ { \\prime } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 678, + 389, + 691 + ], + "score": 1.0, + "content": ", and thus the number of parameters were", + "type": "text" + }, + { + "bbox": [ + 389, + 679, + 423, + 689 + ], + "score": 0.92, + "content": "d ^ { \\prime } + d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 678, + 428, + 691 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 158, + 711 + ], + "score": 1.0, + "content": "DPGAT has", + "type": "text" + }, + { + "bbox": [ + 158, + 699, + 168, + 710 + ], + "score": 0.82, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 698, + 186, + 711 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 187, + 699, + 198, + 709 + ], + "score": 0.79, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 698, + 267, + 711 + ], + "score": 1.0, + "content": "matrices of sizes", + "type": "text" + }, + { + "bbox": [ + 268, + 699, + 284, + 710 + ], + "score": 0.89, + "content": "d d _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 698, + 367, + 711 + ], + "score": 1.0, + "content": "each, and additional", + "type": "text" + }, + { + "bbox": [ + 367, + 699, + 381, + 709 + ], + "score": 0.87, + "content": "d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "parameters in the value matrix", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 107, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 107, + 710, + 117, + 720 + ], + "score": 0.77, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 709, + 140, + 722 + ], + "score": 1.0, + "content": ", thus", + "type": "text" + }, + { + "bbox": [ + 141, + 710, + 187, + 720 + ], + "score": 0.91, + "content": "2 d d _ { k } + d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 709, + 453, + 722 + ], + "score": 1.0, + "content": "parameters overall. However in our experiments, we constrained", + "type": "text" + }, + { + "bbox": [ + 453, + 710, + 487, + 721 + ], + "score": 0.91, + "content": "Q = K", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 720, + 350, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 120, + 732 + ], + "score": 1.0, + "content": "set", + "type": "text" + }, + { + "bbox": [ + 120, + 721, + 152, + 732 + ], + "score": 0.92, + "content": "d _ { k } = d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 720, + 326, + 732 + ], + "score": 1.0, + "content": ", and thus the number of parameters is only", + "type": "text" + }, + { + "bbox": [ + 327, + 721, + 345, + 730 + ], + "score": 0.88, + "content": "2 d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 720, + 350, + 732 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + } + ], + "page_idx": 21, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 249, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 250, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 250, + 96 + ], + "score": 1.0, + "content": "I COMPLEXITY ANALYSIS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 109, + 327, + 122 + ], + "lines": [ + { + "bbox": [ + 105, + 109, + 328, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 109, + 328, + 123 + ], + "score": 1.0, + "content": "We repeat the definitions of GAT, GATv2 and DPGAT:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 109, + 328, + 123 + ] + }, + { + "type": "text", + "bbox": [ + 127, + 136, + 252, + 148 + ], + "lines": [ + { + "bbox": [ + 127, + 135, + 252, + 150 + ], + "spans": [ + { + "bbox": [ + 127, + 135, + 252, + 150 + ], + "score": 1.0, + "content": "GAT (Velickovi ˇ c et al., 2018): ´", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 127, + 135, + 252, + 150 + ] + }, + { + "type": "text", + "bbox": [ + 127, + 152, + 239, + 164 + ], + "lines": [ + { + "bbox": [ + 127, + 151, + 241, + 166 + ], + "spans": [ + { + "bbox": [ + 127, + 151, + 241, + 166 + ], + "score": 1.0, + "content": "GATv2 (our fixed version):", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 127, + 151, + 241, + 166 + ] + }, + { + "type": "text", + "bbox": [ + 127, + 171, + 258, + 183 + ], + "lines": [ + { + "bbox": [ + 127, + 169, + 258, + 185 + ], + "spans": [ + { + "bbox": [ + 127, + 169, + 258, + 185 + ], + "score": 1.0, + "content": "DPGAT (Vaswani et al., 2017):", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 127, + 169, + 258, + 185 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 275, + 133, + 467, + 189 + ], + "lines": [ + { + "bbox": [ + 275, + 133, + 467, + 189 + ], + "spans": [ + { + "bbox": [ + 275, + 133, + 467, + 189 + ], + "score": 0.86, + "content": "\\begin{array} { r l } & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = \\mathrm { L e a k y R e L U } ( \\boldsymbol { a } ^ { \\top } \\cdot [ \\boldsymbol { W } \\boldsymbol { h } _ { i } | | \\boldsymbol { W } \\boldsymbol { h } _ { j } | ) } \\\\ & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = \\boldsymbol { a } ^ { \\top } \\mathrm { L e a k y R e L U } ( \\boldsymbol { W } \\cdot [ \\boldsymbol { h } _ { i } | | \\boldsymbol { h } _ { j } ] ) } \\\\ & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = ( ( \\boldsymbol { h } _ { i } ^ { \\top } \\boldsymbol { Q } ) \\cdot ( \\boldsymbol { h } _ { j } ^ { \\top } \\boldsymbol { K } ) ^ { \\top } ) / \\sqrt { d ^ { \\prime } } } \\end{array}", + "type": "interline_equation", + "image_path": "8a5e123ec02bb5ae8d73a9dbea17509e9c23670171b8eadac52309e582e12309.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 275, + 133, + 467, + 147.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 275, + 147.0, + 467, + 161.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 275, + 161.0, + 467, + 175.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 275, + 175.0, + 467, + 189.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 207, + 212, + 218 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 214, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 214, + 219 + ], + "score": 1.0, + "content": "I.1 TIME COMPLEXITY", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 233, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 232, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 246 + ], + "score": 1.0, + "content": "GAT As noted by Velickovi ˇ c et al. (2018), the time complexity of a single GAT head may be ´", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 245, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 159, + 257 + ], + "score": 1.0, + "content": "expressed as", + "type": "text" + }, + { + "bbox": [ + 159, + 245, + 235, + 257 + ], + "score": 0.91, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 245, + 506, + 257 + ], + "score": 1.0, + "content": ". Because of GAT’s static attention, this computation can be further", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 255, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 266, + 268 + ], + "score": 1.0, + "content": "optimized, by merging the linear layer", + "type": "text" + }, + { + "bbox": [ + 267, + 257, + 279, + 267 + ], + "score": 0.85, + "content": "\\mathbf { \\delta } _ { \\mathbf { a } _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 255, + 301, + 268 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 302, + 256, + 315, + 266 + ], + "score": 0.73, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 255, + 356, + 268 + ], + "score": 1.0, + "content": ", merging", + "type": "text" + }, + { + "bbox": [ + 356, + 257, + 368, + 267 + ], + "score": 0.85, + "content": "\\mathbf { a } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 255, + 390, + 268 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 391, + 256, + 404, + 266 + ], + "score": 0.75, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 255, + 505, + 268 + ], + "score": 1.0, + "content": ", and only then compute", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 266, + 223, + 282 + ], + "spans": [ + { + "bbox": [ + 107, + 266, + 155, + 281 + ], + "score": 0.93, + "content": "a _ { \\{ 1 , 2 \\} } ^ { \\top } W h _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 266, + 195, + 282 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 195, + 267, + 219, + 277 + ], + "score": 0.89, + "content": "i \\in \\mathcal V", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 266, + 223, + 282 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 232, + 506, + 282 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 288, + 505, + 336 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 410, + 302 + ], + "score": 1.0, + "content": "GATv2 require the same computational cost as GAT’s declared complexity:", + "type": "text" + }, + { + "bbox": [ + 410, + 289, + 487, + 301 + ], + "score": 0.91, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 288, + 505, + 302 + ], + "score": 1.0, + "content": ": we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 101, + 295, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 101, + 295, + 137, + 320 + ], + "score": 1.0, + "content": "denote", + "type": "text" + }, + { + "bbox": [ + 137, + 302, + 204, + 315 + ], + "score": 0.92, + "content": "W = [ W _ { 1 } \\Vert W _ { 2 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 295, + 236, + 320 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 236, + 300, + 289, + 313 + ], + "score": 0.93, + "content": "W _ { 1 } \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 295, + 308, + 320 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 308, + 300, + 339, + 314 + ], + "score": 0.9, + "content": "W _ { 2 } ^ { d ^ { \\prime } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 295, + 506, + 320 + ], + "score": 1.0, + "content": "contain the left half and right half of the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 314, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 154, + 326 + ], + "score": 1.0, + "content": "columns of", + "type": "text" + }, + { + "bbox": [ + 155, + 315, + 167, + 324 + ], + "score": 0.79, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 314, + 312, + 326 + ], + "score": 1.0, + "content": ", respectively. We can first compute", + "type": "text" + }, + { + "bbox": [ + 312, + 314, + 339, + 325 + ], + "score": 0.88, + "content": "W _ { 1 } h _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 314, + 357, + 326 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 357, + 314, + 384, + 326 + ], + "score": 0.92, + "content": "W _ { 2 } h _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 314, + 424, + 326 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 424, + 314, + 457, + 325 + ], + "score": 0.91, + "content": "i , j \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 314, + 505, + 326 + ], + "score": 1.0, + "content": ". This takes", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 322, + 156, + 338 + ], + "spans": [ + { + "bbox": [ + 107, + 324, + 151, + 337 + ], + "score": 0.92, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 322, + 156, + 338 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 101, + 288, + 506, + 338 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 504, + 365 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 504, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 192, + 356 + ], + "score": 1.0, + "content": "Then, for every edge", + "type": "text" + }, + { + "bbox": [ + 192, + 342, + 213, + 353 + ], + "score": 0.9, + "content": "( j , i )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 339, + 323, + 356 + ], + "score": 1.0, + "content": ", we compute LeakyReLU", + "type": "text" + }, + { + "bbox": [ + 323, + 341, + 381, + 353 + ], + "score": 0.9, + "content": "\\left( W \\cdot \\left[ \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { h } _ { j } \\right] \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 339, + 477, + 356 + ], + "score": 1.0, + "content": "using the precomputed", + "type": "text" + }, + { + "bbox": [ + 478, + 341, + 504, + 353 + ], + "score": 0.89, + "content": "W _ { 1 } h _ { i }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 351, + 397, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 123, + 366 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 353, + 151, + 365 + ], + "score": 0.92, + "content": "W _ { 2 } h _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 351, + 177, + 366 + ], + "score": 1.0, + "content": ", since", + "type": "text" + }, + { + "bbox": [ + 178, + 352, + 307, + 365 + ], + "score": 0.93, + "content": "W \\cdot [ h _ { i } \\| h _ { j } ] = W _ { 1 } h _ { i } + W _ { 2 } h _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 351, + 354, + 366 + ], + "score": 1.0, + "content": ". This takes", + "type": "text" + }, + { + "bbox": [ + 354, + 353, + 392, + 365 + ], + "score": 0.92, + "content": "\\mathcal { O } \\left( \\vert \\mathcal { E } \\vert d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 351, + 397, + 366 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 339, + 504, + 366 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 369, + 504, + 392 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 312, + 382 + ], + "score": 1.0, + "content": "Finally, computing the results of the linear layer", + "type": "text" + }, + { + "bbox": [ + 313, + 372, + 320, + 379 + ], + "score": 0.74, + "content": "^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 368, + 391, + 382 + ], + "score": 1.0, + "content": "takes additional", + "type": "text" + }, + { + "bbox": [ + 392, + 369, + 429, + 381 + ], + "score": 0.93, + "content": "\\mathcal { O } \\left( | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 368, + 505, + 382 + ], + "score": 1.0, + "content": "time, and overall", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 379, + 187, + 393 + ], + "spans": [ + { + "bbox": [ + 107, + 380, + 182, + 393 + ], + "score": 0.9, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + \\bar { | } \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 379, + 187, + 393 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 368, + 505, + 393 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 399, + 505, + 441 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 338, + 414 + ], + "score": 1.0, + "content": "DPGAT also takes the same time. We can first compute", + "type": "text" + }, + { + "bbox": [ + 338, + 399, + 362, + 412 + ], + "score": 0.92, + "content": "\\pmb { h } _ { i } ^ { \\top } \\pmb { Q }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 398, + 380, + 414 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 380, + 399, + 405, + 414 + ], + "score": 0.95, + "content": "\\pmb { h } _ { j } ^ { \\top } \\pmb { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 398, + 446, + 414 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 446, + 400, + 479, + 412 + ], + "score": 0.91, + "content": "i , j \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 398, + 506, + 414 + ], + "score": 1.0, + "content": ". This", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 414, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 104, + 414, + 129, + 430 + ], + "score": 1.0, + "content": "takes", + "type": "text" + }, + { + "bbox": [ + 129, + 415, + 173, + 428 + ], + "score": 0.93, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 414, + 287, + 430 + ], + "score": 1.0, + "content": ". Computing the dot-product", + "type": "text" + }, + { + "bbox": [ + 287, + 414, + 359, + 430 + ], + "score": 0.86, + "content": "\\left( \\pmb { h } _ { i } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { h } _ { j } ^ { \\top } \\pmb { K } \\right) ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 414, + 419, + 430 + ], + "score": 1.0, + "content": "for every edge", + "type": "text" + }, + { + "bbox": [ + 420, + 416, + 440, + 428 + ], + "score": 0.91, + "content": "( j , i )", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 414, + 506, + 430 + ], + "score": 1.0, + "content": "takes additional", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 428, + 296, + 441 + ], + "spans": [ + { + "bbox": [ + 107, + 428, + 145, + 441 + ], + "score": 0.93, + "content": "\\mathcal { O } \\left( | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 428, + 216, + 441 + ], + "score": 1.0, + "content": "time, and overall", + "type": "text" + }, + { + "bbox": [ + 216, + 428, + 292, + 441 + ], + "score": 0.92, + "content": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 428, + 296, + 441 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 104, + 398, + 506, + 441 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 460, + 246, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 461, + 247, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 247, + 473 + ], + "score": 1.0, + "content": "I.2 PARAMETRIC COMPLEXITY", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "table", + "bbox": [ + 181, + 493, + 430, + 541 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 181, + 493, + 430, + 541 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 181, + 493, + 430, + 541 + ], + "spans": [ + { + "bbox": [ + 181, + 493, + 430, + 541 + ], + "score": 0.975, + "html": "
GATGATv2DPGAT
Official2d'+dd'd'+ 2dd'2ddk +dd'
In our experiments2d'+dd'd'+dd'2dd'
", + "type": "table", + "image_path": "7f473a828617dbd5ae061ebb1adaf2898962cae8e0636392f224b1bdad27d710.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 181, + 493, + 430, + 509.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 181, + 509.0, + 430, + 525.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 181, + 525.0, + 430, + 541.0 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 110, + 547, + 500, + 560 + ], + "lines": [ + { + "bbox": [ + 109, + 546, + 501, + 562 + ], + "spans": [ + { + "bbox": [ + 109, + 546, + 501, + 562 + ], + "score": 1.0, + "content": "Table 19: Number of parameters for each GNN type, in a single layer and a single attention head.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 109, + 546, + 501, + 562 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 580, + 502, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 579, + 504, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 504, + 594 + ], + "score": 1.0, + "content": "All parametric costs are summarized in Table 19. All following calculations refer to a single layer", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 592, + 318, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 318, + 604 + ], + "score": 1.0, + "content": "having a single attention head, omitting bias vectors.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 579, + 504, + 604 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 612, + 504, + 636 + ], + "lines": [ + { + "bbox": [ + 104, + 609, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 104, + 609, + 265, + 627 + ], + "score": 1.0, + "content": "GAT has learned vector and a matrix:", + "type": "text" + }, + { + "bbox": [ + 266, + 611, + 304, + 624 + ], + "score": 0.92, + "content": "\\pmb { a } \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 609, + 324, + 627 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 325, + 611, + 376, + 624 + ], + "score": 0.92, + "content": "W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 609, + 430, + 627 + ], + "score": 1.0, + "content": ", thus overall", + "type": "text" + }, + { + "bbox": [ + 431, + 613, + 471, + 624 + ], + "score": 0.9, + "content": "2 d ^ { \\prime } + d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 609, + 506, + 627 + ], + "score": 1.0, + "content": "learned", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 625, + 155, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 155, + 636 + ], + "score": 1.0, + "content": "parameters.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 104, + 609, + 506, + 636 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 644, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 104, + 642, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 642, + 274, + 659 + ], + "score": 1.0, + "content": "GATv2 has a matrix that is twice larger:", + "type": "text" + }, + { + "bbox": [ + 275, + 644, + 329, + 656 + ], + "score": 0.92, + "content": "W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times 2 d }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 642, + 506, + 659 + ], + "score": 1.0, + "content": ", because it is applied on the concatenation", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 107, + 656, + 507, + 669 + ], + "spans": [ + { + "bbox": [ + 107, + 657, + 139, + 669 + ], + "score": 0.92, + "content": "[ h _ { i } | | h _ { j } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 656, + 344, + 669 + ], + "score": 1.0, + "content": ". Thus, the overall number of learned parameters is", + "type": "text" + }, + { + "bbox": [ + 344, + 657, + 383, + 667 + ], + "score": 0.92, + "content": "d ^ { \\prime } + 2 d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 656, + 507, + 669 + ], + "score": 1.0, + "content": ". However in our experiments,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 668, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 505, + 680 + ], + "score": 1.0, + "content": "to rule out the increased number of parameters over GAT as the source of empirical difference, we", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 678, + 428, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 155, + 691 + ], + "score": 1.0, + "content": "constrained", + "type": "text" + }, + { + "bbox": [ + 155, + 678, + 223, + 691 + ], + "score": 0.94, + "content": "W = [ W ^ { \\prime } | | W ^ { \\prime } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 678, + 389, + 691 + ], + "score": 1.0, + "content": ", and thus the number of parameters were", + "type": "text" + }, + { + "bbox": [ + 389, + 679, + 423, + 689 + ], + "score": 0.92, + "content": "d ^ { \\prime } + d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 678, + 428, + 691 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5, + "bbox_fs": [ + 104, + 642, + 507, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 158, + 711 + ], + "score": 1.0, + "content": "DPGAT has", + "type": "text" + }, + { + "bbox": [ + 158, + 699, + 168, + 710 + ], + "score": 0.82, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 698, + 186, + 711 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 187, + 699, + 198, + 709 + ], + "score": 0.79, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 698, + 267, + 711 + ], + "score": 1.0, + "content": "matrices of sizes", + "type": "text" + }, + { + "bbox": [ + 268, + 699, + 284, + 710 + ], + "score": 0.89, + "content": "d d _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 698, + 367, + 711 + ], + "score": 1.0, + "content": "each, and additional", + "type": "text" + }, + { + "bbox": [ + 367, + 699, + 381, + 709 + ], + "score": 0.87, + "content": "d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "parameters in the value matrix", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 107, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 107, + 710, + 117, + 720 + ], + "score": 0.77, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 709, + 140, + 722 + ], + "score": 1.0, + "content": ", thus", + "type": "text" + }, + { + "bbox": [ + 141, + 710, + 187, + 720 + ], + "score": 0.91, + "content": "2 d d _ { k } + d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 709, + 453, + 722 + ], + "score": 1.0, + "content": "parameters overall. However in our experiments, we constrained", + "type": "text" + }, + { + "bbox": [ + 453, + 710, + 487, + 721 + ], + "score": 0.91, + "content": "Q = K", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 720, + 350, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 120, + 732 + ], + "score": 1.0, + "content": "set", + "type": "text" + }, + { + "bbox": [ + 120, + 721, + 152, + 732 + ], + "score": 0.92, + "content": "d _ { k } = d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 720, + 326, + 732 + ], + "score": 1.0, + "content": ", and thus the number of parameters is only", + "type": "text" + }, + { + "bbox": [ + 327, + 721, + 345, + 730 + ], + "score": 0.88, + "content": "2 d d ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 720, + 350, + 732 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 106, + 698, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 99, + 505, + 130 + ], + "lines": [ + { + "bbox": [ + 105, + 98, + 506, + 111 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 506, + 111 + ], + "score": 1.0, + "content": "Miltiadis Allamanis, Marc Brockschmidt, and Mahmoud Khademi. Learning to represent programs with graphs.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 110, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 116, + 110, + 505, + 120 + ], + "score": 1.0, + "content": "In International Conference on Learning Representations, 2018. URL https://openreview.net/", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 117, + 119, + 219, + 129 + ], + "spans": [ + { + "bbox": [ + 117, + 119, + 160, + 129 + ], + "score": 1.0, + "content": "forum?id", + "type": "text" + }, + { + "bbox": [ + 160, + 120, + 166, + 128 + ], + "score": 0.38, + "content": "\\equiv", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 119, + 219, + 129 + ], + "score": 1.0, + "content": "BJOFETxR-.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 105, + 137, + 504, + 168 + ], + "lines": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "Uri Alon and Eran Yahav. On the bottleneck of graph neural networks and its practical implications. In Interna-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 148, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 148, + 498, + 158 + ], + "score": 1.0, + "content": "tional Conference on Learning Representations, 2021. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 498, + 150, + 505, + 157 + ], + "score": 0.29, + "content": "=", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 158, + 180, + 168 + ], + "spans": [ + { + "bbox": [ + 116, + 158, + 180, + 168 + ], + "score": 1.0, + "content": "i80OPhOCVH2.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 176, + 503, + 198 + ], + "lines": [ + { + "bbox": [ + 105, + 175, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 505, + 189 + ], + "score": 1.0, + "content": "James Atwood and Don Towsley. Diffusion-convolutional neural networks. In Advances in neural information", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 187, + 279, + 197 + ], + "spans": [ + { + "bbox": [ + 115, + 187, + 279, + 197 + ], + "score": 1.0, + "content": "processing systems, pages 1993–2001, 2016.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 105, + 204, + 504, + 226 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 217 + ], + "score": 1.0, + "content": "Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 214, + 480, + 227 + ], + "spans": [ + { + "bbox": [ + 116, + 214, + 480, + 227 + ], + "score": 1.0, + "content": "align and translate. CoRR, abs/1409.0473, 2014. URL http://arxiv.org/abs/1409.0473.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 504, + 265 + ], + "lines": [ + { + "bbox": [ + 106, + 233, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 506, + 245 + ], + "score": 1.0, + "content": "Peter Battaglia, Razvan Pascanu, Matthew Lai, Danilo Jimenez Rezende, and Koray kavukcuoglu. Interaction", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 244, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 115, + 244, + 506, + 255 + ], + "score": 1.0, + "content": "networks for learning about objects, relations and physics. In Proceedings of the 30th International Conference", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 253, + 365, + 265 + ], + "spans": [ + { + "bbox": [ + 115, + 253, + 365, + 265 + ], + "score": 1.0, + "content": "on Neural Information Processing Systems, pages 4509–4517, 2016.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 271, + 505, + 303 + ], + "lines": [ + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "score": 1.0, + "content": "Marc Brockschmidt. Gnn-film: Graph neural networks with feature-wise linear modulation. Proceedings of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 116, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "the 36th International Conference on Machine Learning, ICML, 2020. URL https://github.com/", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 293, + 252, + 303 + ], + "spans": [ + { + "bbox": [ + 116, + 293, + 252, + 303 + ], + "score": 1.0, + "content": "microsoft/tf-gnn-samples.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 105, + 309, + 504, + 332 + ], + "lines": [ + { + "bbox": [ + 105, + 309, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 505, + 323 + ], + "score": 1.0, + "content": "Michael M Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst. Geometric deep", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 320, + 460, + 332 + ], + "spans": [ + { + "bbox": [ + 115, + 320, + 460, + 332 + ], + "score": 1.0, + "content": "learning: going beyond euclidean data. IEEE Signal Processing Magazine, 34(4):18–42, 2017.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 504, + 360 + ], + "lines": [ + { + "bbox": [ + 104, + 336, + 507, + 353 + ], + "spans": [ + { + "bbox": [ + 104, + 336, + 507, + 353 + ], + "score": 1.0, + "content": "Michael M. Bronstein, Joan Bruna, Taco Cohen, and Petar Velickovi ˇ c. Geometric deep learning: Grids, groups, ´", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 349, + 252, + 360 + ], + "spans": [ + { + "bbox": [ + 116, + 349, + 252, + 360 + ], + "score": 1.0, + "content": "graphs, geodesics, and gauges, 2021.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 504, + 389 + ], + "lines": [ + { + "bbox": [ + 106, + 367, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 506, + 379 + ], + "score": 1.0, + "content": "Dan Busbridge, Dane Sherburn, Pietro Cavallo, and Nils Y Hammerla. Relational graph attention networks.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 378, + 263, + 389 + ], + "spans": [ + { + "bbox": [ + 115, + 378, + 263, + 389 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1904.05811, 2019.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 395, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "George Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of control, signals", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 406, + 240, + 417 + ], + "spans": [ + { + "bbox": [ + 115, + 406, + 240, + 417 + ], + "score": 1.0, + "content": "and systems, 2(4):303–314, 1989.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 424, + 504, + 446 + ], + "lines": [ + { + "bbox": [ + 106, + 424, + 504, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 504, + 435 + ], + "score": 1.0, + "content": "Misha Denil, Sergio Gómez Colmenarejo, Serkan Cabi, David Saxton, and Nando de Freitas. Programmable", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 435, + 291, + 446 + ], + "spans": [ + { + "bbox": [ + 115, + 435, + 291, + 446 + ], + "score": 1.0, + "content": "agents. arXiv preprint arXiv:1706.06383, 2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 452, + 504, + 484 + ], + "lines": [ + { + "bbox": [ + 106, + 453, + 506, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 506, + 464 + ], + "score": 1.0, + "content": "Yan Duan, Marcin Andrychowicz, Bradly Stadie, Jonathan Ho, Jonas Schneider, Ilya Sutskever, Pieter Abbeel,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 462, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 115, + 462, + 506, + 475 + ], + "score": 1.0, + "content": "and Wojciech Zaremba. One-shot imitation learning. In Proceedings of the 31st International Conference on", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 473, + 353, + 485 + ], + "spans": [ + { + "bbox": [ + 115, + 473, + 353, + 485 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, pages 1087–1098, 2017.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 491, + 504, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "David K Duvenaud, Dougal Maclaurin, Jorge Iparraguirre, Rafael Bombarell, Timothy Hirzel, Alán Aspuru-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 117, + 502, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 117, + 502, + 505, + 513 + ], + "score": 1.0, + "content": "Guzik, and Ryan P Adams. Convolutional networks on graphs for learning molecular fingerprints. In", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 114, + 511, + 394, + 523 + ], + "spans": [ + { + "bbox": [ + 114, + 511, + 394, + 523 + ], + "score": 1.0, + "content": "Advances in neural information processing systems, pages 2224–2232, 2015.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 529, + 504, + 551 + ], + "lines": [ + { + "bbox": [ + 106, + 528, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 505, + 543 + ], + "score": 1.0, + "content": "Vijay Prakash Dwivedi and Xavier Bresson. A generalization of transformer networks to graphs. arXiv preprint", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 540, + 209, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 540, + 209, + 550 + ], + "score": 1.0, + "content": "arXiv:2012.09699, 2020.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 106, + 558, + 504, + 579 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "score": 1.0, + "content": "Vijay Prakash Dwivedi, Chaitanya K Joshi, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. Benchmarking", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 568, + 348, + 580 + ], + "spans": [ + { + "bbox": [ + 115, + 568, + 348, + 580 + ], + "score": 1.0, + "content": "graph neural networks. arXiv preprint arXiv:2003.00982, 2020.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 587, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 599 + ], + "score": 1.0, + "content": "Matthias Fey and Jan E. Lenssen. Fast graph representation learning with PyTorch Geometric. In ICLR Workshop", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 597, + 338, + 608 + ], + "spans": [ + { + "bbox": [ + 116, + 597, + 338, + 608 + ], + "score": 1.0, + "content": "on Representation Learning on Graphs and Manifolds, 2019.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 615, + 505, + 637 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "Ken-Ichi Funahashi. On the approximate realization of continuous mappings by neural networks. Neural", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 626, + 230, + 636 + ], + "spans": [ + { + "bbox": [ + 115, + 626, + 230, + 636 + ], + "score": 1.0, + "content": "networks, 2(3):183–192, 1989.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 106, + 644, + 505, + 675 + ], + "lines": [ + { + "bbox": [ + 106, + 644, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 506, + 656 + ], + "score": 1.0, + "content": "Hongyang Gao and Shuiwang Ji. Graph representation learning via hard and channel-wise attention networks.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 114, + 653, + 507, + 667 + ], + "spans": [ + { + "bbox": [ + 114, + 653, + 507, + 667 + ], + "score": 1.0, + "content": "In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 664, + 198, + 675 + ], + "spans": [ + { + "bbox": [ + 115, + 664, + 198, + 675 + ], + "score": 1.0, + "content": "pages 741–749, 2019.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 105, + 682, + 505, + 714 + ], + "lines": [ + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "score": 1.0, + "content": "Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural message passing", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 692, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 115, + 692, + 505, + 705 + ], + "score": 1.0, + "content": "for quantum chemistry. In Proceedings of the 34th International Conference on Machine Learning-Volume", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 703, + 264, + 714 + ], + "spans": [ + { + "bbox": [ + 116, + 703, + 264, + 714 + ], + "score": 1.0, + "content": "70, pages 1263–1272. JMLR. org, 2017.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45 + }, + { + "type": "text", + "bbox": [ + 103, + 721, + 469, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 719, + 470, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 470, + 734 + ], + "score": 1.0, + "content": "Aleksa Gordic. pytorch-gat. ´ https://github.com/gordicaleksa/pytorch-GAT, 2020.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47 + } + ], + "page_idx": 22, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 99, + 505, + 130 + ], + "lines": [ + { + "bbox": [ + 105, + 98, + 506, + 111 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 506, + 111 + ], + "score": 1.0, + "content": "Miltiadis Allamanis, Marc Brockschmidt, and Mahmoud Khademi. Learning to represent programs with graphs.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 110, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 116, + 110, + 505, + 120 + ], + "score": 1.0, + "content": "In International Conference on Learning Representations, 2018. URL https://openreview.net/", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 117, + 119, + 219, + 129 + ], + "spans": [ + { + "bbox": [ + 117, + 119, + 160, + 129 + ], + "score": 1.0, + "content": "forum?id", + "type": "text" + }, + { + "bbox": [ + 160, + 120, + 166, + 128 + ], + "score": 0.38, + "content": "\\equiv", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 119, + 219, + 129 + ], + "score": 1.0, + "content": "BJOFETxR-.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 98, + 506, + 129 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 137, + 504, + 168 + ], + "lines": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "Uri Alon and Eran Yahav. On the bottleneck of graph neural networks and its practical implications. In Interna-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 148, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 148, + 498, + 158 + ], + "score": 1.0, + "content": "tional Conference on Learning Representations, 2021. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 498, + 150, + 505, + 157 + ], + "score": 0.29, + "content": "=", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 158, + 180, + 168 + ], + "spans": [ + { + "bbox": [ + 116, + 158, + 180, + 168 + ], + "score": 1.0, + "content": "i80OPhOCVH2.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 106, + 137, + 505, + 168 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 176, + 503, + 198 + ], + "lines": [ + { + "bbox": [ + 105, + 175, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 505, + 189 + ], + "score": 1.0, + "content": "James Atwood and Don Towsley. Diffusion-convolutional neural networks. In Advances in neural information", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 187, + 279, + 197 + ], + "spans": [ + { + "bbox": [ + 115, + 187, + 279, + 197 + ], + "score": 1.0, + "content": "processing systems, pages 1993–2001, 2016.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 175, + 505, + 197 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 204, + 504, + 226 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 217 + ], + "score": 1.0, + "content": "Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 214, + 480, + 227 + ], + "spans": [ + { + "bbox": [ + 116, + 214, + 480, + 227 + ], + "score": 1.0, + "content": "align and translate. CoRR, abs/1409.0473, 2014. URL http://arxiv.org/abs/1409.0473.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 203, + 505, + 227 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 504, + 265 + ], + "lines": [ + { + "bbox": [ + 106, + 233, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 506, + 245 + ], + "score": 1.0, + "content": "Peter Battaglia, Razvan Pascanu, Matthew Lai, Danilo Jimenez Rezende, and Koray kavukcuoglu. Interaction", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 244, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 115, + 244, + 506, + 255 + ], + "score": 1.0, + "content": "networks for learning about objects, relations and physics. In Proceedings of the 30th International Conference", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 253, + 365, + 265 + ], + "spans": [ + { + "bbox": [ + 115, + 253, + 365, + 265 + ], + "score": 1.0, + "content": "on Neural Information Processing Systems, pages 4509–4517, 2016.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 106, + 233, + 506, + 265 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 271, + 505, + 303 + ], + "lines": [ + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "score": 1.0, + "content": "Marc Brockschmidt. Gnn-film: Graph neural networks with feature-wise linear modulation. Proceedings of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 116, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "the 36th International Conference on Machine Learning, ICML, 2020. URL https://github.com/", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 293, + 252, + 303 + ], + "spans": [ + { + "bbox": [ + 116, + 293, + 252, + 303 + ], + "score": 1.0, + "content": "microsoft/tf-gnn-samples.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 270, + 506, + 303 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 309, + 504, + 332 + ], + "lines": [ + { + "bbox": [ + 105, + 309, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 505, + 323 + ], + "score": 1.0, + "content": "Michael M Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst. Geometric deep", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 320, + 460, + 332 + ], + "spans": [ + { + "bbox": [ + 115, + 320, + 460, + 332 + ], + "score": 1.0, + "content": "learning: going beyond euclidean data. IEEE Signal Processing Magazine, 34(4):18–42, 2017.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 309, + 505, + 332 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 504, + 360 + ], + "lines": [ + { + "bbox": [ + 104, + 336, + 507, + 353 + ], + "spans": [ + { + "bbox": [ + 104, + 336, + 507, + 353 + ], + "score": 1.0, + "content": "Michael M. Bronstein, Joan Bruna, Taco Cohen, and Petar Velickovi ˇ c. Geometric deep learning: Grids, groups, ´", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 349, + 252, + 360 + ], + "spans": [ + { + "bbox": [ + 116, + 349, + 252, + 360 + ], + "score": 1.0, + "content": "graphs, geodesics, and gauges, 2021.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 104, + 336, + 507, + 360 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 504, + 389 + ], + "lines": [ + { + "bbox": [ + 106, + 367, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 506, + 379 + ], + "score": 1.0, + "content": "Dan Busbridge, Dane Sherburn, Pietro Cavallo, and Nils Y Hammerla. Relational graph attention networks.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 378, + 263, + 389 + ], + "spans": [ + { + "bbox": [ + 115, + 378, + 263, + 389 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1904.05811, 2019.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 106, + 367, + 506, + 389 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 395, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "George Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of control, signals", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 406, + 240, + 417 + ], + "spans": [ + { + "bbox": [ + 115, + 406, + 240, + 417 + ], + "score": 1.0, + "content": "and systems, 2(4):303–314, 1989.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 395, + 505, + 417 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 424, + 504, + 446 + ], + "lines": [ + { + "bbox": [ + 106, + 424, + 504, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 504, + 435 + ], + "score": 1.0, + "content": "Misha Denil, Sergio Gómez Colmenarejo, Serkan Cabi, David Saxton, and Nando de Freitas. Programmable", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 435, + 291, + 446 + ], + "spans": [ + { + "bbox": [ + 115, + 435, + 291, + 446 + ], + "score": 1.0, + "content": "agents. arXiv preprint arXiv:1706.06383, 2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 424, + 504, + 446 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 452, + 504, + 484 + ], + "lines": [ + { + "bbox": [ + 106, + 453, + 506, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 506, + 464 + ], + "score": 1.0, + "content": "Yan Duan, Marcin Andrychowicz, Bradly Stadie, Jonathan Ho, Jonas Schneider, Ilya Sutskever, Pieter Abbeel,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 462, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 115, + 462, + 506, + 475 + ], + "score": 1.0, + "content": "and Wojciech Zaremba. One-shot imitation learning. In Proceedings of the 31st International Conference on", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 473, + 353, + 485 + ], + "spans": [ + { + "bbox": [ + 115, + 473, + 353, + 485 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, pages 1087–1098, 2017.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 106, + 453, + 506, + 485 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 491, + 504, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "David K Duvenaud, Dougal Maclaurin, Jorge Iparraguirre, Rafael Bombarell, Timothy Hirzel, Alán Aspuru-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 117, + 502, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 117, + 502, + 505, + 513 + ], + "score": 1.0, + "content": "Guzik, and Ryan P Adams. Convolutional networks on graphs for learning molecular fingerprints. In", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 114, + 511, + 394, + 523 + ], + "spans": [ + { + "bbox": [ + 114, + 511, + 394, + 523 + ], + "score": 1.0, + "content": "Advances in neural information processing systems, pages 2224–2232, 2015.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 490, + 506, + 523 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 529, + 504, + 551 + ], + "lines": [ + { + "bbox": [ + 106, + 528, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 505, + 543 + ], + "score": 1.0, + "content": "Vijay Prakash Dwivedi and Xavier Bresson. A generalization of transformer networks to graphs. arXiv preprint", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 540, + 209, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 540, + 209, + 550 + ], + "score": 1.0, + "content": "arXiv:2012.09699, 2020.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 106, + 528, + 505, + 550 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 558, + 504, + 579 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "score": 1.0, + "content": "Vijay Prakash Dwivedi, Chaitanya K Joshi, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. Benchmarking", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 568, + 348, + 580 + ], + "spans": [ + { + "bbox": [ + 115, + 568, + 348, + 580 + ], + "score": 1.0, + "content": "graph neural networks. arXiv preprint arXiv:2003.00982, 2020.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 106, + 558, + 505, + 580 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 587, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 599 + ], + "score": 1.0, + "content": "Matthias Fey and Jan E. Lenssen. Fast graph representation learning with PyTorch Geometric. In ICLR Workshop", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 597, + 338, + 608 + ], + "spans": [ + { + "bbox": [ + 116, + 597, + 338, + 608 + ], + "score": 1.0, + "content": "on Representation Learning on Graphs and Manifolds, 2019.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 106, + 587, + 505, + 608 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 615, + 505, + 637 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "Ken-Ichi Funahashi. On the approximate realization of continuous mappings by neural networks. Neural", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 626, + 230, + 636 + ], + "spans": [ + { + "bbox": [ + 115, + 626, + 230, + 636 + ], + "score": 1.0, + "content": "networks, 2(3):183–192, 1989.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 106, + 615, + 505, + 636 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 644, + 505, + 675 + ], + "lines": [ + { + "bbox": [ + 106, + 644, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 506, + 656 + ], + "score": 1.0, + "content": "Hongyang Gao and Shuiwang Ji. Graph representation learning via hard and channel-wise attention networks.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 114, + 653, + 507, + 667 + ], + "spans": [ + { + "bbox": [ + 114, + 653, + 507, + 667 + ], + "score": 1.0, + "content": "In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 664, + 198, + 675 + ], + "spans": [ + { + "bbox": [ + 115, + 664, + 198, + 675 + ], + "score": 1.0, + "content": "pages 741–749, 2019.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 106, + 644, + 507, + 675 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 682, + 505, + 714 + ], + "lines": [ + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "score": 1.0, + "content": "Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural message passing", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 692, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 115, + 692, + 505, + 705 + ], + "score": 1.0, + "content": "for quantum chemistry. In Proceedings of the 34th International Conference on Machine Learning-Volume", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 703, + 264, + 714 + ], + "spans": [ + { + "bbox": [ + 116, + 703, + 264, + 714 + ], + "score": 1.0, + "content": "70, pages 1263–1272. JMLR. org, 2017.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 681, + 505, + 714 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 721, + 469, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 719, + 470, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 470, + 734 + ], + "score": 1.0, + "content": "Aleksa Gordic. pytorch-gat. ´ https://github.com/gordicaleksa/pytorch-GAT, 2020.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47, + "bbox_fs": [ + 106, + 719, + 470, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 83, + 505, + 114 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Marco Gori, Gabriele Monfardini, and Franco Scarselli. A new model for learning in graph domains. In", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 506, + 104 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 506, + 104 + ], + "score": 1.0, + "content": "Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., volume 2, pages 729–", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 103, + 181, + 114 + ], + "spans": [ + { + "bbox": [ + 115, + 103, + 181, + 114 + ], + "score": 1.0, + "content": "734. IEEE, 2005.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 105, + 120, + 504, + 141 + ], + "lines": [ + { + "bbox": [ + 106, + 119, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 119, + 505, + 133 + ], + "score": 1.0, + "content": "Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. In Advances", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 131, + 359, + 142 + ], + "spans": [ + { + "bbox": [ + 115, + 131, + 359, + 142 + ], + "score": 1.0, + "content": "in neural information processing systems, pages 1024–1034, 2017.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 147, + 504, + 168 + ], + "lines": [ + { + "bbox": [ + 106, + 147, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 505, + 158 + ], + "score": 1.0, + "content": "Kurt Hornik. Approximation capabilities of multilayer feedforward networks. Neural networks, 4(2):251–257,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 157, + 140, + 169 + ], + "spans": [ + { + "bbox": [ + 115, + 157, + 140, + 169 + ], + "score": 1.0, + "content": "1991.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 504, + 196 + ], + "lines": [ + { + "bbox": [ + 106, + 175, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 175, + 505, + 186 + ], + "score": 1.0, + "content": "Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 186, + 314, + 196 + ], + "spans": [ + { + "bbox": [ + 116, + 186, + 314, + 196 + ], + "score": 1.0, + "content": "approximators. Neural networks, 2(5):359–366, 1989.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 202, + 504, + 223 + ], + "lines": [ + { + "bbox": [ + 107, + 202, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 107, + 202, + 505, + 213 + ], + "score": 1.0, + "content": "Yedid Hoshen. Vain: attentional multi-agent predictive modeling. In Proceedings of the 31st International", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 212, + 408, + 224 + ], + "spans": [ + { + "bbox": [ + 116, + 212, + 408, + 224 + ], + "score": 1.0, + "content": "Conference on Neural Information Processing Systems, pages 2698–2708, 2017.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 108, + 229, + 505, + 259 + ], + "lines": [ + { + "bbox": [ + 106, + 228, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 505, + 241 + ], + "score": 1.0, + "content": "Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 239, + 507, + 251 + ], + "spans": [ + { + "bbox": [ + 115, + 239, + 507, + 251 + ], + "score": 1.0, + "content": "Leskovec. Open graph benchmark: Datasets for machine learning on graphs. arXiv preprint arXiv:2005.00687,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 114, + 248, + 141, + 262 + ], + "spans": [ + { + "bbox": [ + 114, + 248, + 141, + 262 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 266, + 506, + 307 + ], + "lines": [ + { + "bbox": [ + 105, + 265, + 506, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 506, + 279 + ], + "score": 1.0, + "content": "Binxuan Huang and Kathleen M Carley. Syntax-aware aspect level sentiment classification with graph attention", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 277, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 116, + 277, + 505, + 288 + ], + "score": 1.0, + "content": "networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 285, + 507, + 299 + ], + "spans": [ + { + "bbox": [ + 115, + 285, + 507, + 299 + ], + "score": 1.0, + "content": "the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5472–5480,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 296, + 140, + 308 + ], + "spans": [ + { + "bbox": [ + 115, + 296, + 140, + 308 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 108, + 313, + 504, + 345 + ], + "lines": [ + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "score": 1.0, + "content": "Qian Huang, Horace He, Abhay Singh, Ser-Nam Lim, and Austin Benson. Combining label propagation and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 324, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 115, + 324, + 506, + 336 + ], + "score": 1.0, + "content": "simple models out-performs graph neural networks. In International Conference on Learning Representations,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 335, + 397, + 345 + ], + "spans": [ + { + "bbox": [ + 116, + 335, + 397, + 345 + ], + "score": 1.0, + "content": "2021. URL https://openreview.net/forum?id=8E1-f3VhX1o.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 351, + 392, + 362 + ], + "lines": [ + { + "bbox": [ + 106, + 351, + 392, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 392, + 363 + ], + "score": 1.0, + "content": "Chaitanya Joshi. Transformers are graph neural networks. The Gradient, 2020.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 368, + 502, + 399 + ], + "lines": [ + { + "bbox": [ + 106, + 369, + 504, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 504, + 379 + ], + "score": 1.0, + "content": "Dongkwan Kim and Alice Oh. How to find your friendly neighborhood: Graph attention design with", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 379, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 505, + 390 + ], + "score": 1.0, + "content": "self-supervision. In International Conference on Learning Representations, 2021. URL https://", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 388, + 310, + 401 + ], + "spans": [ + { + "bbox": [ + 116, + 388, + 241, + 401 + ], + "score": 1.0, + "content": "openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 241, + 390, + 246, + 397 + ], + "score": 0.28, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 388, + 310, + 401 + ], + "score": 1.0, + "content": "Wi5KUNlqWty.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 405, + 504, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "score": 1.0, + "content": "Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In ICLR,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 415, + 141, + 427 + ], + "spans": [ + { + "bbox": [ + 115, + 415, + 141, + 427 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 433, + 506, + 484 + ], + "lines": [ + { + "bbox": [ + 107, + 433, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 107, + 433, + 506, + 444 + ], + "score": 1.0, + "content": "Vineet Kosaraju, Amir Sadeghian, Roberto Martín-Martín, Ian Reid, Hamid Rezatofighi, and Silvio Savarese.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 443, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 116, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "Social-bigat: Multimodal trajectory forecasting using bicycle-gan and graph attention networks. In H. Wallach,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 453, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 116, + 453, + 506, + 465 + ], + "score": 1.0, + "content": "H. Larochelle, A. Beygelzimer, F. d Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Infor-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 461, + 507, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 461, + 507, + 476 + ], + "score": 1.0, + "content": "mation Processing Systems, volume 32. Curran Associates, Inc., 2019. URL https://proceedings.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 473, + 493, + 485 + ], + "spans": [ + { + "bbox": [ + 115, + 473, + 493, + 485 + ], + "score": 1.0, + "content": "neurips.cc/paper/2019/file/d09bf41544a3365a46c9077ebb5e35c3-Paper.pdf.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 489, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 488, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 502 + ], + "score": 1.0, + "content": "John Boaz Lee, Ryan A Rossi, Sungchul Kim, Nesreen K Ahmed, and Eunyee Koh. Attention models in graphs:", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 501, + 299, + 511 + ], + "spans": [ + { + "bbox": [ + 115, + 501, + 299, + 511 + ], + "score": 1.0, + "content": "A survey. arXiv preprint arXiv:1807.07984, 2018.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 106, + 517, + 504, + 539 + ], + "lines": [ + { + "bbox": [ + 106, + 518, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 506, + 529 + ], + "score": 1.0, + "content": "Moshe Leshno, Vladimir Ya Lin, Allan Pinkus, and Shimon Schocken. Multilayer feedforward networks with a", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 528, + 494, + 539 + ], + "spans": [ + { + "bbox": [ + 115, + 528, + 494, + 539 + ], + "score": 1.0, + "content": "nonpolynomial activation function can approximate any function. Neural networks, 6(6):861–867, 1993.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 107, + 544, + 505, + 566 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "score": 1.0, + "content": "Guohao Li, Matthias Muller, Ali Thabet, and Bernard Ghanem. Deepgcns: Can gcns go as deep as cnns? In", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 554, + 486, + 566 + ], + "spans": [ + { + "bbox": [ + 115, + 554, + 486, + 566 + ], + "score": 1.0, + "content": "Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9267–9276, 2019.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 107, + 571, + 503, + 593 + ], + "lines": [ + { + "bbox": [ + 107, + 572, + 504, + 583 + ], + "spans": [ + { + "bbox": [ + 107, + 572, + 504, + 583 + ], + "score": 1.0, + "content": "Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for semi-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 582, + 435, + 594 + ], + "spans": [ + { + "bbox": [ + 116, + 582, + 435, + 594 + ], + "score": 1.0, + "content": "supervised learning. In Thirty-Second AAAI Conference on Artificial Intelligence, 2018.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 504, + 621 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "score": 1.0, + "content": "Yujia Li, Daniel Tarlow, Marc Brockschmidt, and Richard Zemel. Gated graph sequence neural networks. In", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 609, + 339, + 621 + ], + "spans": [ + { + "bbox": [ + 116, + 609, + 339, + 621 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2016.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "Denis Lukovnikov and Asja Fischer. Gated relational graph attention networks, 2021. URL https://", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 637, + 305, + 649 + ], + "spans": [ + { + "bbox": [ + 116, + 637, + 305, + 649 + ], + "score": 1.0, + "content": "openreview.net/forum?id=v-9E8egy_i.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 505, + 666 + ], + "score": 1.0, + "content": "Thang Luong, Hieu Pham, and Christopher D. Manning. Effective approaches to attention-based neural", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 663, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 115, + 663, + 506, + 677 + ], + "score": 1.0, + "content": "machine translation. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 673, + 507, + 686 + ], + "spans": [ + { + "bbox": [ + 115, + 673, + 507, + 686 + ], + "score": 1.0, + "content": "Processing, EMNLP 2015, Lisbon, Portugal, September 17-21, 2015, pages 1412–1421, 2015. URL http:", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 685, + 341, + 695 + ], + "spans": [ + { + "bbox": [ + 116, + 685, + 341, + 695 + ], + "score": 1.0, + "content": "//aclweb.org/anthology/D/D15/D15-1166.pdf.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45.5 + }, + { + "type": "text", + "bbox": [ + 107, + 701, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 701, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 701, + 505, + 713 + ], + "score": 1.0, + "content": "Nianzu Ma, Sahisnu Mazumder, Hao Wang, and Bing Liu. Entity-aware dependency-based deep graph attention", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 114, + 711, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 114, + 711, + 506, + 723 + ], + "score": 1.0, + "content": "network for comparative preference classification. In Proceedings of the 58th Annual Meeting of the", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 114, + 721, + 363, + 732 + ], + "spans": [ + { + "bbox": [ + 114, + 721, + 363, + 732 + ], + "score": 1.0, + "content": "Association for Computational Linguistics, pages 5782–5788, 2020.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 49 + } + ], + "page_idx": 23, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 83, + 505, + 114 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Marco Gori, Gabriele Monfardini, and Franco Scarselli. A new model for learning in graph domains. In", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 506, + 104 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 506, + 104 + ], + "score": 1.0, + "content": "Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., volume 2, pages 729–", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 103, + 181, + 114 + ], + "spans": [ + { + "bbox": [ + 115, + 103, + 181, + 114 + ], + "score": 1.0, + "content": "734. IEEE, 2005.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 82, + 506, + 114 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 120, + 504, + 141 + ], + "lines": [ + { + "bbox": [ + 106, + 119, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 119, + 505, + 133 + ], + "score": 1.0, + "content": "Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. In Advances", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 131, + 359, + 142 + ], + "spans": [ + { + "bbox": [ + 115, + 131, + 359, + 142 + ], + "score": 1.0, + "content": "in neural information processing systems, pages 1024–1034, 2017.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 106, + 119, + 505, + 142 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 147, + 504, + 168 + ], + "lines": [ + { + "bbox": [ + 106, + 147, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 505, + 158 + ], + "score": 1.0, + "content": "Kurt Hornik. Approximation capabilities of multilayer feedforward networks. Neural networks, 4(2):251–257,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 157, + 140, + 169 + ], + "spans": [ + { + "bbox": [ + 115, + 157, + 140, + 169 + ], + "score": 1.0, + "content": "1991.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 106, + 147, + 505, + 169 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 504, + 196 + ], + "lines": [ + { + "bbox": [ + 106, + 175, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 175, + 505, + 186 + ], + "score": 1.0, + "content": "Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 186, + 314, + 196 + ], + "spans": [ + { + "bbox": [ + 116, + 186, + 314, + 196 + ], + "score": 1.0, + "content": "approximators. Neural networks, 2(5):359–366, 1989.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 106, + 175, + 505, + 196 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 202, + 504, + 223 + ], + "lines": [ + { + "bbox": [ + 107, + 202, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 107, + 202, + 505, + 213 + ], + "score": 1.0, + "content": "Yedid Hoshen. Vain: attentional multi-agent predictive modeling. In Proceedings of the 31st International", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 212, + 408, + 224 + ], + "spans": [ + { + "bbox": [ + 116, + 212, + 408, + 224 + ], + "score": 1.0, + "content": "Conference on Neural Information Processing Systems, pages 2698–2708, 2017.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 107, + 202, + 505, + 224 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 229, + 505, + 259 + ], + "lines": [ + { + "bbox": [ + 106, + 228, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 505, + 241 + ], + "score": 1.0, + "content": "Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 239, + 507, + 251 + ], + "spans": [ + { + "bbox": [ + 115, + 239, + 507, + 251 + ], + "score": 1.0, + "content": "Leskovec. Open graph benchmark: Datasets for machine learning on graphs. arXiv preprint arXiv:2005.00687,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 114, + 248, + 141, + 262 + ], + "spans": [ + { + "bbox": [ + 114, + 248, + 141, + 262 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 106, + 228, + 507, + 262 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 266, + 506, + 307 + ], + "lines": [ + { + "bbox": [ + 105, + 265, + 506, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 506, + 279 + ], + "score": 1.0, + "content": "Binxuan Huang and Kathleen M Carley. Syntax-aware aspect level sentiment classification with graph attention", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 277, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 116, + 277, + 505, + 288 + ], + "score": 1.0, + "content": "networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 285, + 507, + 299 + ], + "spans": [ + { + "bbox": [ + 115, + 285, + 507, + 299 + ], + "score": 1.0, + "content": "the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5472–5480,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 296, + 140, + 308 + ], + "spans": [ + { + "bbox": [ + 115, + 296, + 140, + 308 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 265, + 507, + 308 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 313, + 504, + 345 + ], + "lines": [ + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "score": 1.0, + "content": "Qian Huang, Horace He, Abhay Singh, Ser-Nam Lim, and Austin Benson. Combining label propagation and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 324, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 115, + 324, + 506, + 336 + ], + "score": 1.0, + "content": "simple models out-performs graph neural networks. In International Conference on Learning Representations,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 335, + 397, + 345 + ], + "spans": [ + { + "bbox": [ + 116, + 335, + 397, + 345 + ], + "score": 1.0, + "content": "2021. URL https://openreview.net/forum?id=8E1-f3VhX1o.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 313, + 506, + 345 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 351, + 392, + 362 + ], + "lines": [ + { + "bbox": [ + 106, + 351, + 392, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 392, + 363 + ], + "score": 1.0, + "content": "Chaitanya Joshi. Transformers are graph neural networks. The Gradient, 2020.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 351, + 392, + 363 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 368, + 502, + 399 + ], + "lines": [ + { + "bbox": [ + 106, + 369, + 504, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 504, + 379 + ], + "score": 1.0, + "content": "Dongkwan Kim and Alice Oh. How to find your friendly neighborhood: Graph attention design with", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 379, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 505, + 390 + ], + "score": 1.0, + "content": "self-supervision. In International Conference on Learning Representations, 2021. URL https://", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 388, + 310, + 401 + ], + "spans": [ + { + "bbox": [ + 116, + 388, + 241, + 401 + ], + "score": 1.0, + "content": "openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 241, + 390, + 246, + 397 + ], + "score": 0.28, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 388, + 310, + 401 + ], + "score": 1.0, + "content": "Wi5KUNlqWty.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 369, + 505, + 401 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 405, + 504, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "score": 1.0, + "content": "Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In ICLR,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 415, + 141, + 427 + ], + "spans": [ + { + "bbox": [ + 115, + 415, + 141, + 427 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 404, + 506, + 427 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 433, + 506, + 484 + ], + "lines": [ + { + "bbox": [ + 107, + 433, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 107, + 433, + 506, + 444 + ], + "score": 1.0, + "content": "Vineet Kosaraju, Amir Sadeghian, Roberto Martín-Martín, Ian Reid, Hamid Rezatofighi, and Silvio Savarese.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 443, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 116, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "Social-bigat: Multimodal trajectory forecasting using bicycle-gan and graph attention networks. In H. Wallach,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 453, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 116, + 453, + 506, + 465 + ], + "score": 1.0, + "content": "H. Larochelle, A. Beygelzimer, F. d Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Infor-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 461, + 507, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 461, + 507, + 476 + ], + "score": 1.0, + "content": "mation Processing Systems, volume 32. Curran Associates, Inc., 2019. URL https://proceedings.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 473, + 493, + 485 + ], + "spans": [ + { + "bbox": [ + 115, + 473, + 493, + 485 + ], + "score": 1.0, + "content": "neurips.cc/paper/2019/file/d09bf41544a3365a46c9077ebb5e35c3-Paper.pdf.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29, + "bbox_fs": [ + 107, + 433, + 507, + 485 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 489, + 505, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 488, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 502 + ], + "score": 1.0, + "content": "John Boaz Lee, Ryan A Rossi, Sungchul Kim, Nesreen K Ahmed, and Eunyee Koh. Attention models in graphs:", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 501, + 299, + 511 + ], + "spans": [ + { + "bbox": [ + 115, + 501, + 299, + 511 + ], + "score": 1.0, + "content": "A survey. arXiv preprint arXiv:1807.07984, 2018.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 488, + 506, + 511 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 517, + 504, + 539 + ], + "lines": [ + { + "bbox": [ + 106, + 518, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 506, + 529 + ], + "score": 1.0, + "content": "Moshe Leshno, Vladimir Ya Lin, Allan Pinkus, and Shimon Schocken. Multilayer feedforward networks with a", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 528, + 494, + 539 + ], + "spans": [ + { + "bbox": [ + 115, + 528, + 494, + 539 + ], + "score": 1.0, + "content": "nonpolynomial activation function can approximate any function. Neural networks, 6(6):861–867, 1993.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 106, + 518, + 506, + 539 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 544, + 505, + 566 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "score": 1.0, + "content": "Guohao Li, Matthias Muller, Ali Thabet, and Bernard Ghanem. Deepgcns: Can gcns go as deep as cnns? In", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 554, + 486, + 566 + ], + "spans": [ + { + "bbox": [ + 115, + 554, + 486, + 566 + ], + "score": 1.0, + "content": "Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9267–9276, 2019.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 545, + 505, + 566 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 571, + 503, + 593 + ], + "lines": [ + { + "bbox": [ + 107, + 572, + 504, + 583 + ], + "spans": [ + { + "bbox": [ + 107, + 572, + 504, + 583 + ], + "score": 1.0, + "content": "Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for semi-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 582, + 435, + 594 + ], + "spans": [ + { + "bbox": [ + 116, + 582, + 435, + 594 + ], + "score": 1.0, + "content": "supervised learning. In Thirty-Second AAAI Conference on Artificial Intelligence, 2018.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 107, + 572, + 504, + 594 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 504, + 621 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "score": 1.0, + "content": "Yujia Li, Daniel Tarlow, Marc Brockschmidt, and Richard Zemel. Gated graph sequence neural networks. In", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 609, + 339, + 621 + ], + "spans": [ + { + "bbox": [ + 116, + 609, + 339, + 621 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2016.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 106, + 599, + 506, + 621 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "Denis Lukovnikov and Asja Fischer. Gated relational graph attention networks, 2021. URL https://", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 637, + 305, + 649 + ], + "spans": [ + { + "bbox": [ + 116, + 637, + 305, + 649 + ], + "score": 1.0, + "content": "openreview.net/forum?id=v-9E8egy_i.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 106, + 626, + 505, + 649 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 505, + 666 + ], + "score": 1.0, + "content": "Thang Luong, Hieu Pham, and Christopher D. Manning. Effective approaches to attention-based neural", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 663, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 115, + 663, + 506, + 677 + ], + "score": 1.0, + "content": "machine translation. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 673, + 507, + 686 + ], + "spans": [ + { + "bbox": [ + 115, + 673, + 507, + 686 + ], + "score": 1.0, + "content": "Processing, EMNLP 2015, Lisbon, Portugal, September 17-21, 2015, pages 1412–1421, 2015. URL http:", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 685, + 341, + 695 + ], + "spans": [ + { + "bbox": [ + 116, + 685, + 341, + 695 + ], + "score": 1.0, + "content": "//aclweb.org/anthology/D/D15/D15-1166.pdf.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45.5, + "bbox_fs": [ + 106, + 654, + 507, + 695 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 701, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 701, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 701, + 505, + 713 + ], + "score": 1.0, + "content": "Nianzu Ma, Sahisnu Mazumder, Hao Wang, and Bing Liu. Entity-aware dependency-based deep graph attention", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 114, + 711, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 114, + 711, + 506, + 723 + ], + "score": 1.0, + "content": "network for comparative preference classification. In Proceedings of the 58th Annual Meeting of the", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 114, + 721, + 363, + 732 + ], + "spans": [ + { + "bbox": [ + 114, + 721, + 363, + 732 + ], + "score": 1.0, + "content": "Association for Computational Linguistics, pages 5782–5788, 2020.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 49, + "bbox_fs": [ + 106, + 701, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 83, + 505, + 114 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "Federico Monti, Davide Boscaini, Jonathan Masci, Emanuele Rodola, Jan Svoboda, and Michael M Bronstein.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "Geometric deep learning on graphs and manifolds using mixture model cnns. In Proceedings of the IEEE", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 103, + 408, + 115 + ], + "spans": [ + { + "bbox": [ + 115, + 103, + 408, + 115 + ], + "score": 1.0, + "content": "conference on computer vision and pattern recognition, pages 5115–5124, 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 108, + 120, + 503, + 151 + ], + "lines": [ + { + "bbox": [ + 106, + 120, + 505, + 132 + ], + "spans": [ + { + "bbox": [ + 106, + 120, + 505, + 132 + ], + "score": 1.0, + "content": "Deepak Nathani, Jatin Chauhan, Charu Sharma, and Manohar Kaul. Learning attention-based embeddings for", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 131, + 504, + 142 + ], + "spans": [ + { + "bbox": [ + 115, + 131, + 504, + 142 + ], + "score": 1.0, + "content": "relation prediction in knowledge graphs. In Proceedings of the 57th Annual Meeting of the Association for", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 140, + 306, + 152 + ], + "spans": [ + { + "bbox": [ + 116, + 140, + 306, + 152 + ], + "score": 1.0, + "content": "Computational Linguistics, pages 4710–4723, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 158, + 504, + 189 + ], + "lines": [ + { + "bbox": [ + 106, + 157, + 506, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 157, + 506, + 170 + ], + "score": 1.0, + "content": "Sejun Park, Chulhee Yun, Jaeho Lee, and Jinwoo Shin. Minimum width for universal approximation. In Interna-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 167, + 504, + 179 + ], + "spans": [ + { + "bbox": [ + 116, + 167, + 497, + 179 + ], + "score": 1.0, + "content": "tional Conference on Learning Representations, 2021. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 498, + 169, + 504, + 177 + ], + "score": 0.33, + "content": "\\equiv", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 178, + 182, + 189 + ], + "spans": [ + { + "bbox": [ + 116, + 178, + 182, + 189 + ], + "score": 1.0, + "content": "O-XJwyoIF-k.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 104, + 195, + 491, + 207 + ], + "lines": [ + { + "bbox": [ + 106, + 195, + 490, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 490, + 207 + ], + "score": 1.0, + "content": "Allan Pinkus. Approximation theory of the mlp model. Acta Numerica 1999: Volume 8, 8:143–195, 1999.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 212, + 503, + 244 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 505, + 225 + ], + "score": 1.0, + "content": "Jiezhong Qiu, Jian Tang, Hao Ma, Yuxiao Dong, Kuansan Wang, and Jie Tang. Deepinf: Social influence", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 222, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 116, + 222, + 505, + 235 + ], + "score": 1.0, + "content": "prediction with deep learning. In Proceedings of the 24th ACM SIGKDD International Conference on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 232, + 325, + 244 + ], + "spans": [ + { + "bbox": [ + 116, + 232, + 325, + 244 + ], + "score": 1.0, + "content": "Knowledge Discovery and Data Mining (KDD’18), 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 105, + 249, + 504, + 271 + ], + "lines": [ + { + "bbox": [ + 105, + 249, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 505, + 263 + ], + "score": 1.0, + "content": "Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum chemistry", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 260, + 407, + 272 + ], + "spans": [ + { + "bbox": [ + 115, + 260, + 407, + 272 + ], + "score": 1.0, + "content": "structures and properties of 134 kilo molecules. Scientific data, 1:140022, 2014.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 277, + 505, + 308 + ], + "lines": [ + { + "bbox": [ + 106, + 277, + 506, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 506, + 289 + ], + "score": 1.0, + "content": "Yu Rong, Yatao Bian, Tingyang Xu, Weiyang Xie, Ying Wei, Wenbing Huang, and Junzhou Huang. Self-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 286, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 115, + 286, + 505, + 300 + ], + "score": 1.0, + "content": "supervised graph transformer on large-scale molecular data. Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 298, + 189, + 308 + ], + "spans": [ + { + "bbox": [ + 116, + 298, + 189, + 308 + ], + "score": 1.0, + "content": "Systems, 33, 2020a.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 315, + 504, + 345 + ], + "lines": [ + { + "bbox": [ + 106, + 315, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 506, + 327 + ], + "score": 1.0, + "content": "Yu Rong, Wenbing Huang, Tingyang Xu, and Junzhou Huang. Dropedge: Towards deep graph convolutional", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 326, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 116, + 326, + 505, + 336 + ], + "score": 1.0, + "content": "networks on node classification. In International Conference on Learning Representations, 2020b. URL", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 335, + 348, + 347 + ], + "spans": [ + { + "bbox": [ + 115, + 335, + 284, + 347 + ], + "score": 1.0, + "content": "https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 284, + 336, + 290, + 343 + ], + "score": 0.48, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 335, + 348, + 347 + ], + "score": 1.0, + "content": "Hkx1qkrKPr.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 352, + 504, + 383 + ], + "lines": [ + { + "bbox": [ + 106, + 353, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 505, + 363 + ], + "score": 1.0, + "content": "Adam Santoro, David Raposo, David GT Barrett, Mateusz Malinowski, Razvan Pascanu, Peter Battaglia, and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 363, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 116, + 363, + 505, + 373 + ], + "score": 1.0, + "content": "Timothy Lillicrap. A simple neural network module for relational reasoning. In Proceedings of the 31st", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 114, + 372, + 459, + 384 + ], + "spans": [ + { + "bbox": [ + 114, + 372, + 459, + 384 + ], + "score": 1.0, + "content": "International Conference on Neural Information Processing Systems, pages 4974–4983, 2017.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 105, + 389, + 503, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 390, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 505, + 401 + ], + "score": 1.0, + "content": "Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. The graph", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 399, + 419, + 411 + ], + "spans": [ + { + "bbox": [ + 115, + 399, + 419, + 411 + ], + "score": 1.0, + "content": "neural network model. IEEE Transactions on Neural Networks, 20(1):61–80, 2008.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 106, + 416, + 504, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "score": 1.0, + "content": "Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Galligher, and Tina Eliassi-Rad. Collective", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 427, + 347, + 438 + ], + "spans": [ + { + "bbox": [ + 116, + 427, + 347, + 438 + ], + "score": 1.0, + "content": "classification in network data. AI magazine, 29(3):93–93, 2008.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 108, + 444, + 504, + 466 + ], + "lines": [ + { + "bbox": [ + 106, + 442, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 506, + 458 + ], + "score": 1.0, + "content": "Yunsheng Shi, Zhengjie Huang, Shikun Feng, and Yu Sun. Masked label prediction: Unified massage passing", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 454, + 413, + 466 + ], + "spans": [ + { + "bbox": [ + 115, + 454, + 413, + 466 + ], + "score": 1.0, + "content": "model for semi-supervised classification. arXiv preprint arXiv:2009.03509, 2020.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 105, + 471, + 504, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "Kiran K Thekumparampil, Chong Wang, Sewoong Oh, and Li-Jia Li. Attention-based graph neural network for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 482, + 358, + 493 + ], + "spans": [ + { + "bbox": [ + 116, + 482, + 358, + 493 + ], + "score": 1.0, + "content": "semi-supervised learning. arXiv preprint arXiv:1803.03735, 2018.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 498, + 504, + 530 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 506, + 511 + ], + "score": 1.0, + "content": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 508, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 115, + 508, + 505, + 522 + ], + "score": 1.0, + "content": "and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, pages", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 520, + 185, + 530 + ], + "spans": [ + { + "bbox": [ + 116, + 520, + 185, + 530 + ], + "score": 1.0, + "content": "6000–6010, 2017.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 105, + 536, + 503, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "score": 1.0, + "content": "Petar Velickovi ˇ c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph ´", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 117, + 547, + 422, + 558 + ], + "spans": [ + { + "bbox": [ + 117, + 547, + 422, + 558 + ], + "score": 1.0, + "content": "attention networks. In International Conference on Learning Representations, 2018.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 108, + 564, + 504, + 585 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 575 + ], + "score": 1.0, + "content": "Petar Velickovi ˇ c, Lars Buesing, Matthew Overlan, Razvan Pascanu, Oriol Vinyals, and Charles Blundell. Pointer ´", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 574, + 406, + 586 + ], + "spans": [ + { + "bbox": [ + 115, + 574, + 406, + 586 + ], + "score": 1.0, + "content": "graph networks. Advances in Neural Information Processing Systems, 33, 2020.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 309, + 602 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 309, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 309, + 603 + ], + "score": 1.0, + "content": "Petar et al. Velickovi ˇ c. Graph attention networks. 2018. ´", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 104, + 609, + 504, + 630 + ], + "lines": [ + { + "bbox": [ + 106, + 607, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 607, + 505, + 622 + ], + "score": 1.0, + "content": "Guangtao Wang, Rex Ying, Jing Huang, and Jure Leskovec. Improving graph attention networks with large", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 619, + 361, + 630 + ], + "spans": [ + { + "bbox": [ + 115, + 619, + 361, + 630 + ], + "score": 1.0, + "content": "margin-based constraints. arXiv preprint arXiv:1910.11945, 2019a.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 636, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 506, + 648 + ], + "score": 1.0, + "content": "Minjie Wang, Da Zheng, Zihao Ye, Quan Gan, Mufei Li, Xiang Song, Jinjing Zhou, Chao Ma, Lingfan Yu,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 117, + 646, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 117, + 646, + 506, + 658 + ], + "score": 1.0, + "content": "Yu Gai, Tianjun Xiao, Tong He, George Karypis, Jinyang Li, and Zheng Zhang. Deep graph library: A", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 115, + 655, + 506, + 669 + ], + "score": 1.0, + "content": "graph-centric, highly-performant package for graph neural networks. arXiv preprint arXiv:1909.01315,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 664, + 145, + 678 + ], + "spans": [ + { + "bbox": [ + 115, + 664, + 145, + 678 + ], + "score": 1.0, + "content": "2019b.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 106, + 684, + 503, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 684, + 505, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 505, + 696 + ], + "score": 1.0, + "content": "Xiao Wang, Houye Ji, Chuan Shi, Bai Wang, Yanfang Ye, Peng Cui, and Philip S Yu. Heterogeneous graph", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 693, + 411, + 705 + ], + "spans": [ + { + "bbox": [ + 116, + 693, + 411, + 705 + ], + "score": 1.0, + "content": "attention network. In The World Wide Web Conference, pages 2022–2032, 2019c.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46.5 + }, + { + "type": "text", + "bbox": [ + 106, + 711, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 723 + ], + "score": 1.0, + "content": "Yangkun Wang. Bag of tricks of semi-supervised classification with graph neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 115, + 721, + 209, + 731 + ], + "spans": [ + { + "bbox": [ + 115, + 721, + 209, + 731 + ], + "score": 1.0, + "content": "arXiv:2103.13355, 2021.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48.5 + } + ], + "page_idx": 24, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 83, + 505, + 114 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "Federico Monti, Davide Boscaini, Jonathan Masci, Emanuele Rodola, Jan Svoboda, and Michael M Bronstein.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "Geometric deep learning on graphs and manifolds using mixture model cnns. In Proceedings of the IEEE", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 103, + 408, + 115 + ], + "spans": [ + { + "bbox": [ + 115, + 103, + 408, + 115 + ], + "score": 1.0, + "content": "conference on computer vision and pattern recognition, pages 5115–5124, 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 82, + 506, + 115 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 120, + 503, + 151 + ], + "lines": [ + { + "bbox": [ + 106, + 120, + 505, + 132 + ], + "spans": [ + { + "bbox": [ + 106, + 120, + 505, + 132 + ], + "score": 1.0, + "content": "Deepak Nathani, Jatin Chauhan, Charu Sharma, and Manohar Kaul. Learning attention-based embeddings for", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 131, + 504, + 142 + ], + "spans": [ + { + "bbox": [ + 115, + 131, + 504, + 142 + ], + "score": 1.0, + "content": "relation prediction in knowledge graphs. In Proceedings of the 57th Annual Meeting of the Association for", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 140, + 306, + 152 + ], + "spans": [ + { + "bbox": [ + 116, + 140, + 306, + 152 + ], + "score": 1.0, + "content": "Computational Linguistics, pages 4710–4723, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 120, + 505, + 152 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 158, + 504, + 189 + ], + "lines": [ + { + "bbox": [ + 106, + 157, + 506, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 157, + 506, + 170 + ], + "score": 1.0, + "content": "Sejun Park, Chulhee Yun, Jaeho Lee, and Jinwoo Shin. Minimum width for universal approximation. In Interna-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 167, + 504, + 179 + ], + "spans": [ + { + "bbox": [ + 116, + 167, + 497, + 179 + ], + "score": 1.0, + "content": "tional Conference on Learning Representations, 2021. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 498, + 169, + 504, + 177 + ], + "score": 0.33, + "content": "\\equiv", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 178, + 182, + 189 + ], + "spans": [ + { + "bbox": [ + 116, + 178, + 182, + 189 + ], + "score": 1.0, + "content": "O-XJwyoIF-k.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 157, + 506, + 189 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 195, + 491, + 207 + ], + "lines": [ + { + "bbox": [ + 106, + 195, + 490, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 490, + 207 + ], + "score": 1.0, + "content": "Allan Pinkus. Approximation theory of the mlp model. Acta Numerica 1999: Volume 8, 8:143–195, 1999.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 106, + 195, + 490, + 207 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 212, + 503, + 244 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 505, + 225 + ], + "score": 1.0, + "content": "Jiezhong Qiu, Jian Tang, Hao Ma, Yuxiao Dong, Kuansan Wang, and Jie Tang. Deepinf: Social influence", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 222, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 116, + 222, + 505, + 235 + ], + "score": 1.0, + "content": "prediction with deep learning. In Proceedings of the 24th ACM SIGKDD International Conference on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 232, + 325, + 244 + ], + "spans": [ + { + "bbox": [ + 116, + 232, + 325, + 244 + ], + "score": 1.0, + "content": "Knowledge Discovery and Data Mining (KDD’18), 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 212, + 505, + 244 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 249, + 504, + 271 + ], + "lines": [ + { + "bbox": [ + 105, + 249, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 505, + 263 + ], + "score": 1.0, + "content": "Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum chemistry", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 260, + 407, + 272 + ], + "spans": [ + { + "bbox": [ + 115, + 260, + 407, + 272 + ], + "score": 1.0, + "content": "structures and properties of 134 kilo molecules. Scientific data, 1:140022, 2014.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 249, + 505, + 272 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 277, + 505, + 308 + ], + "lines": [ + { + "bbox": [ + 106, + 277, + 506, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 506, + 289 + ], + "score": 1.0, + "content": "Yu Rong, Yatao Bian, Tingyang Xu, Weiyang Xie, Ying Wei, Wenbing Huang, and Junzhou Huang. Self-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 286, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 115, + 286, + 505, + 300 + ], + "score": 1.0, + "content": "supervised graph transformer on large-scale molecular data. Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 298, + 189, + 308 + ], + "spans": [ + { + "bbox": [ + 116, + 298, + 189, + 308 + ], + "score": 1.0, + "content": "Systems, 33, 2020a.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 106, + 277, + 506, + 308 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 315, + 504, + 345 + ], + "lines": [ + { + "bbox": [ + 106, + 315, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 506, + 327 + ], + "score": 1.0, + "content": "Yu Rong, Wenbing Huang, Tingyang Xu, and Junzhou Huang. Dropedge: Towards deep graph convolutional", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 326, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 116, + 326, + 505, + 336 + ], + "score": 1.0, + "content": "networks on node classification. In International Conference on Learning Representations, 2020b. URL", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 335, + 348, + 347 + ], + "spans": [ + { + "bbox": [ + 115, + 335, + 284, + 347 + ], + "score": 1.0, + "content": "https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 284, + 336, + 290, + 343 + ], + "score": 0.48, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 335, + 348, + 347 + ], + "score": 1.0, + "content": "Hkx1qkrKPr.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 106, + 315, + 506, + 347 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 352, + 504, + 383 + ], + "lines": [ + { + "bbox": [ + 106, + 353, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 505, + 363 + ], + "score": 1.0, + "content": "Adam Santoro, David Raposo, David GT Barrett, Mateusz Malinowski, Razvan Pascanu, Peter Battaglia, and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 363, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 116, + 363, + 505, + 373 + ], + "score": 1.0, + "content": "Timothy Lillicrap. A simple neural network module for relational reasoning. In Proceedings of the 31st", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 114, + 372, + 459, + 384 + ], + "spans": [ + { + "bbox": [ + 114, + 372, + 459, + 384 + ], + "score": 1.0, + "content": "International Conference on Neural Information Processing Systems, pages 4974–4983, 2017.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 353, + 505, + 384 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 389, + 503, + 411 + ], + "lines": [ + { + "bbox": [ + 106, + 390, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 505, + 401 + ], + "score": 1.0, + "content": "Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. The graph", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 399, + 419, + 411 + ], + "spans": [ + { + "bbox": [ + 115, + 399, + 419, + 411 + ], + "score": 1.0, + "content": "neural network model. IEEE Transactions on Neural Networks, 20(1):61–80, 2008.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 106, + 390, + 505, + 411 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 416, + 504, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "score": 1.0, + "content": "Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Galligher, and Tina Eliassi-Rad. Collective", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 427, + 347, + 438 + ], + "spans": [ + { + "bbox": [ + 116, + 427, + 347, + 438 + ], + "score": 1.0, + "content": "classification in network data. AI magazine, 29(3):93–93, 2008.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 106, + 417, + 505, + 438 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 444, + 504, + 466 + ], + "lines": [ + { + "bbox": [ + 106, + 442, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 506, + 458 + ], + "score": 1.0, + "content": "Yunsheng Shi, Zhengjie Huang, Shikun Feng, and Yu Sun. Masked label prediction: Unified massage passing", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 454, + 413, + 466 + ], + "spans": [ + { + "bbox": [ + 115, + 454, + 413, + 466 + ], + "score": 1.0, + "content": "model for semi-supervised classification. arXiv preprint arXiv:2009.03509, 2020.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 106, + 442, + 506, + 466 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 471, + 504, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "Kiran K Thekumparampil, Chong Wang, Sewoong Oh, and Li-Jia Li. Attention-based graph neural network for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 482, + 358, + 493 + ], + "spans": [ + { + "bbox": [ + 116, + 482, + 358, + 493 + ], + "score": 1.0, + "content": "semi-supervised learning. arXiv preprint arXiv:1803.03735, 2018.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 106, + 471, + 506, + 493 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 498, + 504, + 530 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 506, + 511 + ], + "score": 1.0, + "content": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 508, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 115, + 508, + 505, + 522 + ], + "score": 1.0, + "content": "and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, pages", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 520, + 185, + 530 + ], + "spans": [ + { + "bbox": [ + 116, + 520, + 185, + 530 + ], + "score": 1.0, + "content": "6000–6010, 2017.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 498, + 506, + 530 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 536, + 503, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "score": 1.0, + "content": "Petar Velickovi ˇ c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph ´", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 117, + 547, + 422, + 558 + ], + "spans": [ + { + "bbox": [ + 117, + 547, + 422, + 558 + ], + "score": 1.0, + "content": "attention networks. In International Conference on Learning Representations, 2018.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 535, + 505, + 558 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 564, + 504, + 585 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 575 + ], + "score": 1.0, + "content": "Petar Velickovi ˇ c, Lars Buesing, Matthew Overlan, Razvan Pascanu, Oriol Vinyals, and Charles Blundell. Pointer ´", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 574, + 406, + 586 + ], + "spans": [ + { + "bbox": [ + 115, + 574, + 406, + 586 + ], + "score": 1.0, + "content": "graph networks. Advances in Neural Information Processing Systems, 33, 2020.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 106, + 564, + 505, + 586 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 309, + 602 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 309, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 309, + 603 + ], + "score": 1.0, + "content": "Petar et al. Velickovi ˇ c. Graph attention networks. 2018. ´", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39, + "bbox_fs": [ + 106, + 591, + 309, + 603 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 609, + 504, + 630 + ], + "lines": [ + { + "bbox": [ + 106, + 607, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 607, + 505, + 622 + ], + "score": 1.0, + "content": "Guangtao Wang, Rex Ying, Jing Huang, and Jure Leskovec. Improving graph attention networks with large", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 619, + 361, + 630 + ], + "spans": [ + { + "bbox": [ + 115, + 619, + 361, + 630 + ], + "score": 1.0, + "content": "margin-based constraints. arXiv preprint arXiv:1910.11945, 2019a.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 106, + 607, + 505, + 630 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 636, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 506, + 648 + ], + "score": 1.0, + "content": "Minjie Wang, Da Zheng, Zihao Ye, Quan Gan, Mufei Li, Xiang Song, Jinjing Zhou, Chao Ma, Lingfan Yu,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 117, + 646, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 117, + 646, + 506, + 658 + ], + "score": 1.0, + "content": "Yu Gai, Tianjun Xiao, Tong He, George Karypis, Jinyang Li, and Zheng Zhang. Deep graph library: A", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 115, + 655, + 506, + 669 + ], + "score": 1.0, + "content": "graph-centric, highly-performant package for graph neural networks. arXiv preprint arXiv:1909.01315,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 664, + 145, + 678 + ], + "spans": [ + { + "bbox": [ + 115, + 664, + 145, + 678 + ], + "score": 1.0, + "content": "2019b.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43.5, + "bbox_fs": [ + 106, + 636, + 506, + 678 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 684, + 503, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 684, + 505, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 505, + 696 + ], + "score": 1.0, + "content": "Xiao Wang, Houye Ji, Chuan Shi, Bai Wang, Yanfang Ye, Peng Cui, and Philip S Yu. Heterogeneous graph", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 693, + 411, + 705 + ], + "spans": [ + { + "bbox": [ + 116, + 693, + 411, + 705 + ], + "score": 1.0, + "content": "attention network. In The World Wide Web Conference, pages 2022–2032, 2019c.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46.5, + "bbox_fs": [ + 106, + 684, + 505, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 711, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 723 + ], + "score": 1.0, + "content": "Yangkun Wang. Bag of tricks of semi-supervised classification with graph neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 115, + 721, + 209, + 731 + ], + "spans": [ + { + "bbox": [ + 115, + 721, + 209, + 731 + ], + "score": 1.0, + "content": "arXiv:2103.13355, 2021.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48.5, + "bbox_fs": [ + 106, + 710, + 505, + 731 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 83, + 507, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 83, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 506, + 97 + ], + "score": 1.0, + "content": "Gail Weiss, Yoav Goldberg, and Eran Yahav. On the practical computational power of finite precision rnns", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 92, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 115, + 92, + 506, + 107 + ], + "score": 1.0, + "content": "for language recognition. In Proceedings of the 56th Annual Meeting of the Association for Computational", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 103, + 335, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 103, + 335, + 116 + ], + "score": 1.0, + "content": "Linguistics (Volume 2: Short Papers), pages 740–745, 2018.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 120, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 104, + 120, + 506, + 135 + ], + "score": 1.0, + "content": "Felix Wu, Amauri Souza, Tianyi Zhang, Christopher Fifty, Tao Yu, and Kilian Weinberger. Simplifying graph", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 131, + 504, + 144 + ], + "spans": [ + { + "bbox": [ + 115, + 131, + 504, + 144 + ], + "score": 1.0, + "content": "convolutional networks. In International conference on machine learning, pages 6861–6871. PMLR, 2019.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 148, + 506, + 163 + ], + "spans": [ + { + "bbox": [ + 104, + 148, + 506, + 163 + ], + "score": 1.0, + "content": "Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and S Yu Philip. A comprehensive", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 114, + 159, + 490, + 172 + ], + "spans": [ + { + "bbox": [ + 114, + 159, + 490, + 172 + ], + "score": 1.0, + "content": "survey on graph neural networks. IEEE Transactions on Neural Networks and Learning Systems, 2020.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 176, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 190 + ], + "score": 1.0, + "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks?", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 115, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "In International Conference on Learning Representations, 2019. URL https://openreview.net/", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 198, + 224, + 210 + ], + "spans": [ + { + "bbox": [ + 116, + 198, + 160, + 210 + ], + "score": 1.0, + "content": "forum?id", + "type": "text" + }, + { + "bbox": [ + 160, + 198, + 166, + 206 + ], + "score": 0.54, + "content": "\\equiv", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 198, + 224, + 210 + ], + "score": 1.0, + "content": "ryGs6iA5Km.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "Yiding Yang, Jiayan Qiu, Mingli Song, Dacheng Tao, and Xinchao Wang. Distilling knowledge from graph", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 116, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "convolutional networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 114, + 235, + 236, + 247 + ], + "spans": [ + { + "bbox": [ + 114, + 235, + 236, + 247 + ], + "score": 1.0, + "content": "Recognition (CVPR), June 2020.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 253, + 506, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 506, + 266 + ], + "score": 1.0, + "content": "Seongjun Yun, Minbyul Jeong, Raehyun Kim, Jaewoo Kang, and Hyunwoo J Kim. Graph transformer networks.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 263, + 396, + 275 + ], + "spans": [ + { + "bbox": [ + 115, + 263, + 396, + 275 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 32:11983–11993, 2019.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "score": 1.0, + "content": "Hanqing Zeng, Hongkuan Zhou, Ajitesh Srivastava, Rajgopal Kannan, and Viktor Prasanna. Graphsaint: Graph", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 291, + 420, + 303 + ], + "spans": [ + { + "bbox": [ + 115, + 291, + 420, + 303 + ], + "score": 1.0, + "content": "sampling based inductive learning method. arXiv preprint arXiv:1907.04931, 2019.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 308, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 505, + 321 + ], + "score": 1.0, + "content": "Jiani Zhang, Xingjian Shi, Junyuan Xie, Hao Ma, Irwin King, and Dit-Yan Yeung. Gaan: Gated attention", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 114, + 318, + 506, + 332 + ], + "spans": [ + { + "bbox": [ + 114, + 318, + 506, + 332 + ], + "score": 1.0, + "content": "networks for learning on large and spatiotemporal graphs. In Proceedings of the Thirty-Fourth Conference on", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 328, + 333, + 341 + ], + "spans": [ + { + "bbox": [ + 115, + 328, + 333, + 341 + ], + "score": 1.0, + "content": "Uncertainty in Artificial Intelligence, pages 339–349, 2018.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 345, + 507, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 507, + 360 + ], + "score": 1.0, + "content": "Kai Zhang, Yaokang Zhu, Jun Wang, and Jie Zhang. Adaptive structural fingerprints for graph attention networks.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 357, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 115, + 357, + 506, + 369 + ], + "score": 1.0, + "content": "In International Conference on Learning Representations, 2020. URL https://openreview.net/", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 366, + 224, + 379 + ], + "spans": [ + { + "bbox": [ + 116, + 366, + 160, + 379 + ], + "score": 1.0, + "content": "forum?id", + "type": "text" + }, + { + "bbox": [ + 160, + 367, + 166, + 375 + ], + "score": 0.37, + "content": "\\equiv", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 366, + 224, + 379 + ], + "score": 1.0, + "content": "BJxWx0NYPr.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 383, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 506, + 398 + ], + "score": 1.0, + "content": "Lingxiao Zhao and Leman Akoglu. Pairnorm: Tackling oversmoothing in gnns. In International Conference on", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 394, + 487, + 407 + ], + "spans": [ + { + "bbox": [ + 115, + 394, + 424, + 407 + ], + "score": 1.0, + "content": "Learning Representations, 2020. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 424, + 396, + 430, + 402 + ], + "score": 0.38, + "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 394, + 487, + 407 + ], + "score": 1.0, + "content": "rkecl1rtwB.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 12 + } + ], + "page_idx": 25, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "26", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 83, + 507, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 83, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 506, + 97 + ], + "score": 1.0, + "content": "Gail Weiss, Yoav Goldberg, and Eran Yahav. On the practical computational power of finite precision rnns", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 92, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 115, + 92, + 506, + 107 + ], + "score": 1.0, + "content": "for language recognition. In Proceedings of the 56th Annual Meeting of the Association for Computational", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 103, + 335, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 103, + 335, + 116 + ], + "score": 1.0, + "content": "Linguistics (Volume 2: Short Papers), pages 740–745, 2018.", + "type": "text" + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 120, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 104, + 120, + 506, + 135 + ], + "score": 1.0, + "content": "Felix Wu, Amauri Souza, Tianyi Zhang, Christopher Fifty, Tao Yu, and Kilian Weinberger. Simplifying graph", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 131, + 504, + 144 + ], + "spans": [ + { + "bbox": [ + 115, + 131, + 504, + 144 + ], + "score": 1.0, + "content": "convolutional networks. In International conference on machine learning, pages 6861–6871. PMLR, 2019.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 148, + 506, + 163 + ], + "spans": [ + { + "bbox": [ + 104, + 148, + 506, + 163 + ], + "score": 1.0, + "content": "Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and S Yu Philip. A comprehensive", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 114, + 159, + 490, + 172 + ], + "spans": [ + { + "bbox": [ + 114, + 159, + 490, + 172 + ], + "score": 1.0, + "content": "survey on graph neural networks. IEEE Transactions on Neural Networks and Learning Systems, 2020.", + "type": "text" + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 176, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 190 + ], + "score": 1.0, + "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks?", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 115, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "In International Conference on Learning Representations, 2019. URL https://openreview.net/", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 198, + 224, + 210 + ], + "spans": [ + { + "bbox": [ + 116, + 198, + 160, + 210 + ], + "score": 1.0, + "content": "forum?id", + "type": "text" + }, + { + "bbox": [ + 160, + 198, + 166, + 206 + ], + "score": 0.54, + "content": "\\equiv", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 198, + 224, + 210 + ], + "score": 1.0, + "content": "ryGs6iA5Km.", + "type": "text" + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "Yiding Yang, Jiayan Qiu, Mingli Song, Dacheng Tao, and Xinchao Wang. Distilling knowledge from graph", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 116, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "convolutional networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 114, + 235, + 236, + 247 + ], + "spans": [ + { + "bbox": [ + 114, + 235, + 236, + 247 + ], + "score": 1.0, + "content": "Recognition (CVPR), June 2020.", + "type": "text" + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 253, + 506, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 506, + 266 + ], + "score": 1.0, + "content": "Seongjun Yun, Minbyul Jeong, Raehyun Kim, Jaewoo Kang, and Hyunwoo J Kim. Graph transformer networks.", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 263, + 396, + 275 + ], + "spans": [ + { + "bbox": [ + 115, + 263, + 396, + 275 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 32:11983–11993, 2019.", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "score": 1.0, + "content": "Hanqing Zeng, Hongkuan Zhou, Ajitesh Srivastava, Rajgopal Kannan, and Viktor Prasanna. Graphsaint: Graph", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 291, + 420, + 303 + ], + "spans": [ + { + "bbox": [ + 115, + 291, + 420, + 303 + ], + "score": 1.0, + "content": "sampling based inductive learning method. arXiv preprint arXiv:1907.04931, 2019.", + "type": "text" + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 308, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 505, + 321 + ], + "score": 1.0, + "content": "Jiani Zhang, Xingjian Shi, Junyuan Xie, Hao Ma, Irwin King, and Dit-Yan Yeung. Gaan: Gated attention", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 114, + 318, + 506, + 332 + ], + "spans": [ + { + "bbox": [ + 114, + 318, + 506, + 332 + ], + "score": 1.0, + "content": "networks for learning on large and spatiotemporal graphs. In Proceedings of the Thirty-Fourth Conference on", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 328, + 333, + 341 + ], + "spans": [ + { + "bbox": [ + 115, + 328, + 333, + 341 + ], + "score": 1.0, + "content": "Uncertainty in Artificial Intelligence, pages 339–349, 2018.", + "type": "text" + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 345, + 507, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 507, + 360 + ], + "score": 1.0, + "content": "Kai Zhang, Yaokang Zhu, Jun Wang, and Jie Zhang. Adaptive structural fingerprints for graph attention networks.", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 357, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 115, + 357, + 506, + 369 + ], + "score": 1.0, + "content": "In International Conference on Learning Representations, 2020. URL https://openreview.net/", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 366, + 224, + 379 + ], + "spans": [ + { + "bbox": [ + 116, + 366, + 160, + 379 + ], + "score": 1.0, + "content": "forum?id", + "type": "text" + }, + { + "bbox": [ + 160, + 367, + 166, + 375 + ], + "score": 0.37, + "content": "\\equiv", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 366, + 224, + 379 + ], + "score": 1.0, + "content": "BJxWx0NYPr.", + "type": "text" + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 383, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 506, + 398 + ], + "score": 1.0, + "content": "Lingxiao Zhao and Leman Akoglu. Pairnorm: Tackling oversmoothing in gnns. In International Conference on", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 394, + 487, + 407 + ], + "spans": [ + { + "bbox": [ + 115, + 394, + 424, + 407 + ], + "score": 1.0, + "content": "Learning Representations, 2020. URL https://openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 424, + 396, + 430, + 402 + ], + "score": 0.38, + "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 394, + 487, + 407 + ], + "score": 1.0, + "content": "rkecl1rtwB.", + "type": "text" + } + ], + "index": 24, + "is_list_end_line": true + } + ], + "index": 12, + "bbox_fs": [ + 104, + 83, + 507, + 407 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/dev/F72ximsx7C1/F72ximsx7C1_model.json b/parse/dev/F72ximsx7C1/F72ximsx7C1_model.json new file mode 100644 index 0000000000000000000000000000000000000000..cb1682084ca61b39ddfe888a773cdb1207ec120a --- /dev/null +++ b/parse/dev/F72ximsx7C1/F72ximsx7C1_model.json @@ -0,0 +1,38804 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 791, + 1304, + 791, + 1304, + 1308, + 398, + 1308 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 299, + 1441, + 1403, + 1441, + 1403, + 1627, + 299, + 1627 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1642, + 1404, + 1642, + 1404, + 1888, + 298, + 1888 + ], + "score": 0.973 + }, + { + "category_id": 2, + "poly": [ + 331, + 1914, + 1387, + 1914, + 1387, + 2034, + 331, + 2034 + ], + "score": 0.961 + }, + { + "category_id": 0, + "poly": [ + 298, + 221, + 1160, + 221, + 1160, + 323, + 298, + 323 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 315, + 375, + 748, + 375, + 748, + 468, + 315, + 468 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 302, + 1371, + 573, + 1371, + 573, + 1406, + 302, + 1406 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 816, + 75, + 816, + 104, + 300, + 104 + ], + "score": 0.858 + }, + { + "category_id": 0, + "poly": [ + 773, + 719, + 927, + 719, + 927, + 752, + 773, + 752 + ], + "score": 0.853 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 856, + 2089, + 856, + 2112, + 841, + 2112 + ], + "score": 0.735 + }, + { + "category_id": 1, + "poly": [ + 1019, + 377, + 1386, + 377, + 1386, + 497, + 1019, + 497 + ], + "score": 0.577 + }, + { + "category_id": 1, + "poly": [ + 318, + 610, + 714, + 610, + 714, + 638, + 318, + 638 + ], + "score": 0.572 + }, + { + "category_id": 0, + "poly": [ + 314, + 547, + 457, + 547, + 457, + 576, + 314, + 576 + ], + "score": 0.434 + }, + { + "category_id": 0, + "poly": [ + 1021, + 377, + 1128, + 377, + 1128, + 406, + 1021, + 406 + ], + "score": 0.277 + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1909.0, + 1390.0, + 1909.0, + 1390.0, + 1949.0, + 328.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1936.0, + 1189.0, + 1936.0, + 1189.0, + 1982.0, + 325.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1966.0, + 891.0, + 1966.0, + 891.0, + 2014.0, + 324.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1996.0, + 1383.0, + 1996.0, + 1383.0, + 2043.0, + 326.0, + 2043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 216.0, + 1165.0, + 216.0, + 1165.0, + 272.0, + 294.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 271.0, + 568.0, + 271.0, + 568.0, + 327.0, + 294.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1368.0, + 579.0, + 1368.0, + 579.0, + 1415.0, + 294.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 715.0, + 934.0, + 715.0, + 934.0, + 758.0, + 768.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2088.0, + 859.0, + 2088.0, + 859.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 542.0, + 462.0, + 542.0, + 462.0, + 581.0, + 311.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 374.0, + 1132.0, + 374.0, + 1132.0, + 411.0, + 1017.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 791.0, + 1304.0, + 791.0, + 1304.0, + 825.0, + 396.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 817.0, + 1305.0, + 817.0, + 1305.0, + 858.0, + 393.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 851.0, + 1308.0, + 851.0, + 1308.0, + 887.0, + 394.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 880.0, + 1306.0, + 880.0, + 1306.0, + 918.0, + 393.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 911.0, + 1307.0, + 911.0, + 1307.0, + 947.0, + 393.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 943.0, + 1306.0, + 943.0, + 1306.0, + 974.0, + 393.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 975.0, + 1307.0, + 975.0, + 1307.0, + 1008.0, + 394.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1005.0, + 1305.0, + 1005.0, + 1305.0, + 1038.0, + 394.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1033.0, + 1305.0, + 1033.0, + 1305.0, + 1068.0, + 393.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1063.0, + 1305.0, + 1063.0, + 1305.0, + 1099.0, + 393.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1093.0, + 1309.0, + 1093.0, + 1309.0, + 1130.0, + 392.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1126.0, + 1306.0, + 1126.0, + 1306.0, + 1159.0, + 393.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1157.0, + 1307.0, + 1157.0, + 1307.0, + 1188.0, + 393.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1186.0, + 1306.0, + 1186.0, + 1306.0, + 1222.0, + 393.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1217.0, + 1305.0, + 1217.0, + 1305.0, + 1252.0, + 393.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1246.0, + 1306.0, + 1246.0, + 1306.0, + 1285.0, + 393.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1277.0, + 968.0, + 1277.0, + 968.0, + 1312.0, + 394.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1441.0, + 1405.0, + 1441.0, + 1405.0, + 1478.0, + 294.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1473.0, + 1408.0, + 1473.0, + 1408.0, + 1507.0, + 293.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1501.0, + 1407.0, + 1501.0, + 1407.0, + 1538.0, + 293.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1534.0, + 1407.0, + 1534.0, + 1407.0, + 1570.0, + 294.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1566.0, + 1405.0, + 1566.0, + 1405.0, + 1598.0, + 295.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1594.0, + 1330.0, + 1594.0, + 1330.0, + 1631.0, + 294.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1642.0, + 1404.0, + 1642.0, + 1404.0, + 1676.0, + 294.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1671.0, + 1405.0, + 1671.0, + 1405.0, + 1707.0, + 293.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1704.0, + 1404.0, + 1704.0, + 1404.0, + 1737.0, + 294.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1733.0, + 1405.0, + 1733.0, + 1405.0, + 1770.0, + 293.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1765.0, + 1404.0, + 1765.0, + 1404.0, + 1799.0, + 294.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1791.0, + 1405.0, + 1791.0, + 1405.0, + 1832.0, + 292.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1823.0, + 1406.0, + 1823.0, + 1406.0, + 1861.0, + 292.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1855.0, + 1407.0, + 1855.0, + 1407.0, + 1890.0, + 292.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 371.0, + 491.0, + 371.0, + 491.0, + 411.0, + 313.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 404.0, + 425.0, + 404.0, + 425.0, + 439.0, + 311.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 439.0, + 750.0, + 439.0, + 750.0, + 470.0, + 313.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 375.0, + 1131.0, + 375.0, + 1131.0, + 408.0, + 1018.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 409.0, + 1386.0, + 409.0, + 1386.0, + 440.0, + 1017.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 435.0, + 1335.0, + 435.0, + 1335.0, + 472.0, + 1017.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 468.0, + 1292.0, + 468.0, + 1292.0, + 500.0, + 1019.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 607.0, + 718.0, + 607.0, + 718.0, + 641.0, + 310.0, + 641.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1728, + 1405, + 1728, + 1405, + 2034, + 298, + 2034 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1467, + 1405, + 1467, + 1405, + 1713, + 298, + 1713 + ], + "score": 0.979 + }, + { + "category_id": 4, + "poly": [ + 296, + 1161, + 1405, + 1161, + 1405, + 1317, + 296, + 1317 + ], + "score": 0.967 + }, + { + "category_id": 3, + "poly": [ + 320, + 221, + 1366, + 221, + 1366, + 1137, + 320, + 1137 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 299, + 1390, + 1403, + 1390, + 1403, + 1453, + 299, + 1453 + ], + "score": 0.938 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.717 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.152 + }, + { + "category_id": 3, + "poly": [ + 894, + 215, + 1363, + 215, + 1363, + 1131, + 894, + 1131 + ], + "score": 0.139 + }, + { + "category_id": 13, + "poly": [ + 935, + 1162, + 1066, + 1162, + 1066, + 1196, + 935, + 1196 + ], + "score": 0.93, + "latex": "\\{ q 0 , . . . , q 9 \\}" + }, + { + "category_id": 13, + "poly": [ + 1261, + 1162, + 1396, + 1162, + 1396, + 1196, + 1261, + 1196 + ], + "score": 0.93, + "latex": "\\{ k 0 , . . . , k 9 \\}" + }, + { + "category_id": 13, + "poly": [ + 1341, + 1225, + 1373, + 1225, + 1373, + 1255, + 1341, + 1255 + ], + "score": 0.64, + "latex": "_ { g 0 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1255, + 331, + 1255, + 331, + 1286, + 297, + 1286 + ], + "score": 0.62, + "latex": "q 9 ," + }, + { + "category_id": 13, + "poly": [ + 658, + 1255, + 701, + 1255, + 701, + 1283, + 658, + 1283 + ], + "score": 0.38, + "latex": "( k 8 )" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1159.0, + 934.0, + 1159.0, + 934.0, + 1199.0, + 293.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1159.0, + 1260.0, + 1159.0, + 1260.0, + 1199.0, + 1067.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1159.0, + 1410.0, + 1159.0, + 1410.0, + 1199.0, + 1397.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1191.0, + 1405.0, + 1191.0, + 1405.0, + 1228.0, + 294.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1220.0, + 1340.0, + 1220.0, + 1340.0, + 1260.0, + 293.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1220.0, + 1405.0, + 1220.0, + 1405.0, + 1260.0, + 1374.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1251.0, + 657.0, + 1251.0, + 657.0, + 1292.0, + 332.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1251.0, + 1406.0, + 1251.0, + 1406.0, + 1292.0, + 702.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1284.0, + 1277.0, + 1284.0, + 1277.0, + 1322.0, + 294.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 222.0, + 786.0, + 222.0, + 786.0, + 251.0, + 375.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 220.0, + 1355.0, + 220.0, + 1355.0, + 253.0, + 940.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 262.0, + 794.0, + 262.0, + 794.0, + 303.0, + 326.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 270.0, + 931.0, + 270.0, + 931.0, + 297.0, + 892.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 265.0, + 1360.0, + 265.0, + 1360.0, + 291.0, + 933.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 303.0, + 794.0, + 303.0, + 794.0, + 340.0, + 326.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 302.0, + 1360.0, + 302.0, + 1360.0, + 332.0, + 936.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 343.0, + 365.0, + 343.0, + 365.0, + 374.0, + 321.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 342.0, + 794.0, + 342.0, + 794.0, + 371.0, + 368.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 340.0, + 1360.0, + 340.0, + 1360.0, + 378.0, + 893.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 379.0, + 368.0, + 379.0, + 368.0, + 418.0, + 325.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 380.0, + 794.0, + 380.0, + 794.0, + 409.0, + 369.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 379.0, + 935.0, + 379.0, + 935.0, + 417.0, + 893.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 380.0, + 1360.0, + 380.0, + 1360.0, + 409.0, + 936.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 418.0, + 794.0, + 418.0, + 794.0, + 447.0, + 370.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 420.0, + 934.0, + 420.0, + 934.0, + 454.0, + 896.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 418.0, + 1361.0, + 418.0, + 1361.0, + 447.0, + 935.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 456.0, + 368.0, + 456.0, + 368.0, + 494.0, + 325.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 457.0, + 794.0, + 457.0, + 794.0, + 485.0, + 370.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 457.0, + 935.0, + 457.0, + 935.0, + 493.0, + 893.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 457.0, + 1360.0, + 457.0, + 1360.0, + 485.0, + 938.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 497.0, + 367.0, + 497.0, + 367.0, + 528.0, + 321.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 493.0, + 793.0, + 493.0, + 793.0, + 524.0, + 369.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 497.0, + 934.0, + 497.0, + 934.0, + 530.0, + 894.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 496.0, + 1360.0, + 496.0, + 1360.0, + 521.0, + 936.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 533.0, + 369.0, + 533.0, + 369.0, + 571.0, + 325.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 534.0, + 793.0, + 534.0, + 793.0, + 560.0, + 370.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 536.0, + 932.0, + 536.0, + 932.0, + 566.0, + 889.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 532.0, + 1361.0, + 532.0, + 1361.0, + 561.0, + 937.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 570.0, + 793.0, + 570.0, + 793.0, + 608.0, + 325.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 571.0, + 935.0, + 571.0, + 935.0, + 607.0, + 893.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 572.0, + 1362.0, + 572.0, + 1362.0, + 600.0, + 938.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 608.0, + 368.0, + 608.0, + 368.0, + 646.0, + 325.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 611.0, + 793.0, + 611.0, + 793.0, + 637.0, + 370.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 612.0, + 932.0, + 612.0, + 932.0, + 641.0, + 888.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 611.0, + 1361.0, + 611.0, + 1361.0, + 637.0, + 937.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 658.0, + 940.0, + 658.0, + 940.0, + 697.0, + 885.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 677.0, + 487.0, + 677.0, + 487.0, + 952.0, + 438.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 681.0, + 371.0, + 681.0, + 371.0, + 717.0, + 319.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 693.0, + 1052.0, + 693.0, + 1052.0, + 731.0, + 1036.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 693.0, + 1306.0, + 693.0, + 1306.0, + 738.0, + 1285.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 704.0, + 1095.0, + 704.0, + 1095.0, + 731.0, + 1077.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 733.0, + 937.0, + 733.0, + 937.0, + 768.0, + 886.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 726.0, + 1092.0, + 726.0, + 1092.0, + 734.0, + 1079.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 728.0, + 1175.0, + 728.0, + 1175.0, + 743.0, + 1165.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 723.0, + 1261.0, + 723.0, + 1261.0, + 746.0, + 1251.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 723.0, + 1342.0, + 723.0, + 1342.0, + 735.0, + 1332.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 745.0, + 1359.0, + 745.0, + 1359.0, + 1023.0, + 1305.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 771.0, + 371.0, + 771.0, + 371.0, + 810.0, + 318.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 780.0, + 739.0, + 780.0, + 739.0, + 861.0, + 713.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 797.0, + 747.0, + 797.0, + 747.0, + 837.0, + 736.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 804.0, + 940.0, + 804.0, + 940.0, + 842.0, + 886.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 840.0, + 724.0, + 840.0, + 724.0, + 861.0, + 716.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 846.0, + 736.0, + 846.0, + 736.0, + 860.0, + 726.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 860.0, + 371.0, + 860.0, + 371.0, + 900.0, + 318.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 879.0, + 938.0, + 879.0, + 938.0, + 913.0, + 886.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 941.0, + 702.0, + 941.0, + 702.0, + 954.0, + 694.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 953.0, + 370.0, + 953.0, + 370.0, + 988.0, + 319.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 951.0, + 937.0, + 951.0, + 937.0, + 986.0, + 886.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 978.0, + 419.0, + 978.0, + 419.0, + 1000.0, + 386.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 970.0, + 776.0, + 970.0, + 776.0, + 1028.0, + 760.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 992.0, + 547.0, + 992.0, + 547.0, + 1007.0, + 512.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 1011.0, + 399.0, + 1011.0, + 399.0, + 1024.0, + 389.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1009.0, + 1051.0, + 1009.0, + 1051.0, + 1025.0, + 1036.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 1009.0, + 1091.0, + 1009.0, + 1091.0, + 1031.0, + 1079.0, + 1031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1024.0, + 937.0, + 1024.0, + 937.0, + 1059.0, + 886.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1031.0, + 987.0, + 1031.0, + 987.0, + 1042.0, + 964.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 1032.0, + 1022.0, + 1032.0, + 1022.0, + 1040.0, + 1013.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1064.0, + 788.0, + 1064.0, + 788.0, + 1089.0, + 374.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1060.0, + 1356.0, + 1060.0, + 1356.0, + 1089.0, + 940.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1105.0, + 856.0, + 1105.0, + 856.0, + 1139.0, + 313.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1106.0, + 1372.0, + 1106.0, + 1372.0, + 1139.0, + 881.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.25, + 301.5, + 938.25, + 301.5, + 938.25, + 339.5, + 888.25, + 339.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 418.0, + 372.0, + 418.0, + 372.0, + 454.0, + 321.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 1022.0, + 1342.0, + 1022.0, + 1342.0, + 1040.5, + 1287.0, + 1040.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 222.0, + 1355.0, + 222.0, + 1355.0, + 252.0, + 942.0, + 252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 266.0, + 935.0, + 266.0, + 935.0, + 305.0, + 895.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 266.0, + 1361.0, + 266.0, + 1361.0, + 292.0, + 937.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 296.0, + 1361.0, + 296.0, + 1361.0, + 343.0, + 894.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 341.0, + 933.0, + 341.0, + 933.0, + 377.0, + 896.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 340.0, + 1361.0, + 340.0, + 1361.0, + 369.0, + 937.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 380.0, + 933.0, + 380.0, + 933.0, + 416.0, + 896.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 378.0, + 1361.0, + 378.0, + 1361.0, + 408.0, + 935.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 418.0, + 935.0, + 418.0, + 935.0, + 457.0, + 895.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 418.0, + 1361.0, + 418.0, + 1361.0, + 445.0, + 938.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 457.0, + 933.0, + 457.0, + 933.0, + 493.0, + 896.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 455.0, + 1361.0, + 455.0, + 1361.0, + 485.0, + 937.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 496.0, + 934.0, + 496.0, + 934.0, + 529.0, + 896.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 494.0, + 1362.0, + 494.0, + 1362.0, + 524.0, + 936.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 532.0, + 1364.0, + 532.0, + 1364.0, + 563.0, + 936.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 574.0, + 934.0, + 574.0, + 934.0, + 604.0, + 890.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 572.0, + 1362.0, + 572.0, + 1362.0, + 599.0, + 937.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 609.0, + 1361.0, + 609.0, + 1361.0, + 647.0, + 892.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 658.0, + 942.0, + 658.0, + 942.0, + 697.0, + 885.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 731.0, + 943.0, + 731.0, + 943.0, + 768.0, + 885.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 695.0, + 1049.0, + 695.0, + 1049.0, + 727.0, + 1039.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 698.0, + 1093.0, + 698.0, + 1093.0, + 723.0, + 1078.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 719.0, + 1177.0, + 719.0, + 1177.0, + 743.0, + 1165.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 734.0, + 1217.0, + 734.0, + 1217.0, + 746.0, + 1210.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 725.0, + 1260.0, + 725.0, + 1260.0, + 738.0, + 1252.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 697.0, + 1304.0, + 697.0, + 1304.0, + 743.0, + 1289.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1331.0, + 699.0, + 1348.0, + 699.0, + 1348.0, + 749.0, + 1331.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 742.0, + 1358.0, + 742.0, + 1358.0, + 1024.0, + 1304.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 804.0, + 943.0, + 804.0, + 943.0, + 841.0, + 885.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 878.0, + 942.0, + 878.0, + 942.0, + 913.0, + 887.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 949.0, + 940.0, + 949.0, + 940.0, + 988.0, + 886.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1022.0, + 943.0, + 1022.0, + 943.0, + 1060.0, + 886.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1059.0, + 1358.0, + 1059.0, + 1358.0, + 1092.0, + 940.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1105.0, + 1369.0, + 1105.0, + 1369.0, + 1137.0, + 889.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 534.5, + 937.0, + 534.5, + 937.0, + 567.0, + 891.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1728.0, + 1406.0, + 1728.0, + 1406.0, + 1762.0, + 294.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1759.0, + 1406.0, + 1759.0, + 1406.0, + 1795.0, + 295.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1785.0, + 1410.0, + 1785.0, + 1410.0, + 1828.0, + 292.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1818.0, + 1407.0, + 1818.0, + 1407.0, + 1854.0, + 293.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1849.0, + 1406.0, + 1849.0, + 1406.0, + 1885.0, + 293.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1881.0, + 1405.0, + 1881.0, + 1405.0, + 1917.0, + 295.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1910.0, + 1405.0, + 1910.0, + 1405.0, + 1946.0, + 293.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1944.0, + 1402.0, + 1944.0, + 1402.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1974.0, + 1405.0, + 1974.0, + 1405.0, + 2005.0, + 296.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2003.0, + 1077.0, + 2003.0, + 1077.0, + 2038.0, + 295.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1469.0, + 1405.0, + 1469.0, + 1405.0, + 1502.0, + 295.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1497.0, + 1407.0, + 1497.0, + 1407.0, + 1534.0, + 293.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1529.0, + 1408.0, + 1529.0, + 1408.0, + 1565.0, + 293.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1560.0, + 1407.0, + 1560.0, + 1407.0, + 1595.0, + 293.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1591.0, + 1405.0, + 1591.0, + 1405.0, + 1625.0, + 295.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1621.0, + 1405.0, + 1621.0, + 1405.0, + 1655.0, + 293.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1652.0, + 1406.0, + 1652.0, + 1406.0, + 1686.0, + 295.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1683.0, + 1139.0, + 1683.0, + 1139.0, + 1716.0, + 296.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1386.0, + 1408.0, + 1386.0, + 1408.0, + 1428.0, + 294.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1420.0, + 744.0, + 1420.0, + 744.0, + 1456.0, + 295.0, + 1456.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 367, + 1405, + 367, + 1405, + 796, + 297, + 796 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1473, + 1404, + 1473, + 1404, + 1630, + 298, + 1630 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1101, + 1405, + 1101, + 1405, + 1232, + 297, + 1232 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 906, + 1403, + 906, + 1403, + 1010, + 298, + 1010 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 228, + 1403, + 228, + 1403, + 352, + 299, + 352 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 1289, + 1404, + 1289, + 1404, + 1383, + 298, + 1383 + ], + "score": 0.975 + }, + { + "category_id": 8, + "poly": [ + 525, + 1774, + 1173, + 1774, + 1173, + 1852, + 525, + 1852 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 297, + 1860, + 1410, + 1860, + 1410, + 1923, + 297, + 1923 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 676, + 1934, + 1020, + 1934, + 1020, + 1994, + 676, + 1994 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 298, + 1697, + 1409, + 1697, + 1409, + 1763, + 298, + 1763 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 579, + 1641, + 1119, + 1641, + 1119, + 1683, + 579, + 1683 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 606, + 1241, + 1093, + 1241, + 1093, + 1280, + 606, + 1280 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 297, + 2002, + 1139, + 2002, + 1139, + 2036, + 297, + 2036 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 301, + 1417, + 743, + 1417, + 743, + 1449, + 301, + 1449 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 104, + 300, + 104 + ], + "score": 0.901 + }, + { + "category_id": 0, + "poly": [ + 301, + 1044, + 707, + 1044, + 707, + 1077, + 301, + 1077 + ], + "score": 0.9 + }, + { + "category_id": 0, + "poly": [ + 300, + 840, + 578, + 840, + 578, + 875, + 300, + 875 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1648, + 1399, + 1648, + 1399, + 1677, + 1366, + 1677 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1245, + 1399, + 1245, + 1399, + 1275, + 1366, + 1275 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1948, + 1400, + 1948, + 1400, + 1977, + 1366, + 1977 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1795, + 1400, + 1795, + 1400, + 1824, + 1366, + 1824 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.685 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.457 + }, + { + "category_id": 13, + "poly": [ + 792, + 908, + 957, + 908, + 957, + 942, + 792, + 942 + ], + "score": 0.95, + "latex": "\\mathcal { V } = \\{ 1 , . . . , n \\}" + }, + { + "category_id": 14, + "poly": [ + 677, + 1933, + 1022, + 1933, + 1022, + 1994, + 677, + 1994 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\pmb { h } _ { i } ^ { \\prime } = \\sigma \\left( \\sum _ { j \\in \\mathcal { N } _ { i } } \\pmb { \\alpha } _ { i j } \\cdot \\pmb { W } \\pmb { h } _ { j } \\right) } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 526, + 1771, + 1173, + 1771, + 1173, + 1852, + 526, + 1852 + ], + "score": 0.94, + "latex": "\\alpha _ { i j } = \\mathrm { s o f t m a x } _ { j } \\left( e \\left( h _ { i } , h _ { j } \\right) \\right) = \\frac { \\exp { \\left( e \\left( h _ { i } , h _ { j } \\right) \\right) } } { \\sum _ { j ^ { \\prime } \\in { \\mathcal { N } } _ { i } } \\exp { \\left( e \\left( h _ { i } , h _ { j ^ { \\prime } } \\right) \\right) } }" + }, + { + "category_id": 13, + "poly": [ + 461, + 969, + 588, + 969, + 588, + 1011, + 461, + 1011 + ], + "score": 0.93, + "latex": "\\bar { \\boldsymbol { h } } _ { i } ^ { ( 0 ) } \\in \\mathbb { R } ^ { d _ { 0 } }" + }, + { + "category_id": 13, + "poly": [ + 489, + 908, + 617, + 908, + 617, + 942, + 489, + 942 + ], + "score": 0.93, + "latex": "\\mathcal { G } = ( \\nu , \\mathcal { E } )" + }, + { + "category_id": 13, + "poly": [ + 1293, + 908, + 1401, + 908, + 1401, + 942, + 1293, + 942 + ], + "score": 0.92, + "latex": "( j , i ) \\in \\mathcal { E }" + }, + { + "category_id": 13, + "poly": [ + 697, + 1732, + 777, + 1732, + 777, + 1762, + 697, + 1762 + ], + "score": 0.92, + "latex": "j \\in \\mathcal N _ { i }" + }, + { + "category_id": 13, + "poly": [ + 464, + 1598, + 519, + 1598, + 519, + 1631, + 464, + 1631 + ], + "score": 0.92, + "latex": "( j , i )" + }, + { + "category_id": 14, + "poly": [ + 580, + 1640, + 1118, + 1640, + 1118, + 1684, + 580, + 1684 + ], + "score": 0.91, + "latex": "e ( \\pmb { h } _ { i } , \\pmb { h } _ { j } ) = \\mathrm { L e a k y R e L U } ( \\pmb { a } ^ { \\top } \\cdot [ \\pmb { W } \\pmb { h } _ { i } | | \\pmb { W } \\pmb { h } _ { j } | ) " + }, + { + "category_id": 13, + "poly": [ + 522, + 1507, + 602, + 1507, + 602, + 1538, + 522, + 1538 + ], + "score": 0.91, + "latex": "j \\in \\mathcal N _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1022, + 1132, + 1239, + 1132, + 1239, + 1166, + 1022, + 1166 + ], + "score": 0.91, + "latex": "\\{ h _ { i } \\in \\bar { \\mathbb { R } } ^ { d } \\bar { | } \\ i \\stackrel { - } { \\in } \\mathcal { V } \\}" + }, + { + "category_id": 14, + "poly": [ + 603, + 1240, + 1096, + 1240, + 1096, + 1280, + 603, + 1280 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\pmb { h } _ { i } ^ { \\prime } = f _ { \\theta } \\left( \\pmb { h } _ { i } , \\mathrm { A G G R E G A T E } \\left( \\{ \\pmb { h } _ { j } \\ | \\ j \\in \\mathcal { N } _ { i } \\} \\right) \\right) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1017, + 1569, + 1208, + 1569, + 1208, + 1596, + 1017, + 1596 + ], + "score": 0.91, + "latex": "e : \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 1077, + 908, + 1207, + 908, + 1207, + 939, + 1077, + 939 + ], + "score": 0.9, + "latex": "\\mathcal { E } \\subseteq \\mathcal { V } \\times \\mathcal { V }" + }, + { + "category_id": 13, + "poly": [ + 1159, + 941, + 1241, + 941, + 1241, + 969, + 1159, + 969 + ], + "score": 0.89, + "latex": "\\textit { i } \\in \\mathcal { V }" + }, + { + "category_id": 13, + "poly": [ + 978, + 1166, + 1207, + 1166, + 1207, + 1201, + 978, + 1201 + ], + "score": 0.89, + "latex": "\\{ h _ { i } ^ { \\prime } \\in \\mathbb { R } ^ { d ^ { \\prime } } \\mid i \\in \\mathcal { V } \\}" + }, + { + "category_id": 13, + "poly": [ + 486, + 1695, + 626, + 1695, + 626, + 1729, + 486, + 1729 + ], + "score": 0.89, + "latex": "W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }" + }, + { + "category_id": 13, + "poly": [ + 761, + 1568, + 794, + 1568, + 794, + 1598, + 761, + 1598 + ], + "score": 0.89, + "latex": "{ \\mathcal { N } } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 369, + 1695, + 473, + 1695, + 473, + 1729, + 369, + 1729 + ], + "score": 0.88, + "latex": "\\pmb { a } \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }" + }, + { + "category_id": 13, + "poly": [ + 889, + 704, + 963, + 704, + 963, + 733, + 889, + 733 + ], + "score": 0.88, + "latex": "1 1 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 840, + 1352, + 871, + 1352, + 871, + 1381, + 840, + 1381 + ], + "score": 0.87, + "latex": "\\boldsymbol { h } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1224, + 1604, + 1238, + 1604, + 1238, + 1630, + 1224, + 1630 + ], + "score": 0.86, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 455, + 1291, + 474, + 1291, + 474, + 1322, + 455, + 1322 + ], + "score": 0.86, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1020, + 643, + 1082, + 643, + 1082, + 672, + 1020, + 672 + ], + "score": 0.86, + "latex": "1 . 4 \\%" + }, + { + "category_id": 13, + "poly": [ + 1373, + 1605, + 1383, + 1605, + 1383, + 1625, + 1373, + 1625 + ], + "score": 0.85, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 808, + 1701, + 825, + 1701, + 825, + 1734, + 808, + 1734 + ], + "score": 0.85, + "latex": "\\parallel" + }, + { + "category_id": 13, + "poly": [ + 1204, + 1352, + 1223, + 1352, + 1223, + 1383, + 1204, + 1383 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1009, + 1200, + 1303, + 1200, + 1303, + 1234, + 1009, + 1234 + ], + "score": 0.84, + "latex": "\\mathcal { N } _ { i } = \\{ j \\in \\mathcal { V } \\mid ( j , i ) \\in \\mathcal { E } \\}" + }, + { + "category_id": 13, + "poly": [ + 648, + 942, + 666, + 942, + 666, + 972, + 648, + 972 + ], + "score": 0.83, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 843, + 1898, + 853, + 1898, + 853, + 1919, + 843, + 1919 + ], + "score": 0.81, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 369, + 1169, + 390, + 1169, + 390, + 1196, + 369, + 1196 + ], + "score": 0.71, + "latex": "\\mathcal { E }" + }, + { + "category_id": 13, + "poly": [ + 492, + 1897, + 512, + 1897, + 512, + 1919, + 492, + 1919 + ], + "score": 0.71, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 792, + 943, + 806, + 943, + 806, + 967, + 792, + 967 + ], + "score": 0.64, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 956, + 1478, + 994, + 1478, + 994, + 1504, + 956, + 1504 + ], + "score": 0.27, + "latex": "\\mathrm { { X u } }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1414.0, + 746.0, + 1414.0, + 746.0, + 1454.0, + 294.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1042.0, + 711.0, + 1042.0, + 711.0, + 1081.0, + 293.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 837.0, + 582.0, + 837.0, + 582.0, + 881.0, + 293.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 368.0, + 1403.0, + 368.0, + 1403.0, + 403.0, + 294.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 398.0, + 1407.0, + 398.0, + 1407.0, + 435.0, + 292.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 428.0, + 1406.0, + 428.0, + 1406.0, + 465.0, + 292.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 462.0, + 1406.0, + 462.0, + 1406.0, + 494.0, + 295.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 489.0, + 1406.0, + 489.0, + 1406.0, + 526.0, + 294.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 520.0, + 1406.0, + 520.0, + 1406.0, + 560.0, + 294.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 551.0, + 1405.0, + 551.0, + 1405.0, + 587.0, + 295.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 580.0, + 1407.0, + 580.0, + 1407.0, + 618.0, + 291.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 612.0, + 1407.0, + 612.0, + 1407.0, + 648.0, + 295.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 640.0, + 1019.0, + 640.0, + 1019.0, + 681.0, + 292.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 640.0, + 1407.0, + 640.0, + 1407.0, + 681.0, + 1083.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 673.0, + 1406.0, + 673.0, + 1406.0, + 709.0, + 295.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 702.0, + 888.0, + 702.0, + 888.0, + 738.0, + 295.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 702.0, + 1403.0, + 702.0, + 1403.0, + 738.0, + 964.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 734.0, + 1406.0, + 734.0, + 1406.0, + 770.0, + 294.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 765.0, + 1410.0, + 765.0, + 1410.0, + 800.0, + 294.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1475.0, + 955.0, + 1475.0, + 955.0, + 1509.0, + 297.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 1475.0, + 1406.0, + 1475.0, + 1406.0, + 1509.0, + 995.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1506.0, + 521.0, + 1506.0, + 521.0, + 1543.0, + 294.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 1506.0, + 1406.0, + 1506.0, + 1406.0, + 1543.0, + 603.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1535.0, + 1405.0, + 1535.0, + 1405.0, + 1573.0, + 294.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1561.0, + 760.0, + 1561.0, + 760.0, + 1606.0, + 292.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1561.0, + 1016.0, + 1561.0, + 1016.0, + 1606.0, + 795.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 1561.0, + 1406.0, + 1561.0, + 1406.0, + 1606.0, + 1209.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1597.0, + 463.0, + 1597.0, + 463.0, + 1630.0, + 296.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 1597.0, + 1223.0, + 1597.0, + 1223.0, + 1630.0, + 520.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1597.0, + 1372.0, + 1597.0, + 1372.0, + 1630.0, + 1239.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1384.0, + 1597.0, + 1397.0, + 1597.0, + 1397.0, + 1630.0, + 1384.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1100.0, + 1406.0, + 1100.0, + 1406.0, + 1136.0, + 291.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1132.0, + 1021.0, + 1132.0, + 1021.0, + 1169.0, + 294.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1132.0, + 1406.0, + 1132.0, + 1406.0, + 1169.0, + 1240.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1165.0, + 368.0, + 1165.0, + 368.0, + 1204.0, + 292.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1165.0, + 977.0, + 1165.0, + 977.0, + 1204.0, + 391.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 1165.0, + 1406.0, + 1165.0, + 1406.0, + 1204.0, + 1208.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1198.0, + 1008.0, + 1198.0, + 1008.0, + 1235.0, + 292.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 1198.0, + 1314.0, + 1198.0, + 1314.0, + 1235.0, + 1304.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 906.0, + 488.0, + 906.0, + 488.0, + 945.0, + 293.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 906.0, + 791.0, + 906.0, + 791.0, + 945.0, + 618.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 906.0, + 1076.0, + 906.0, + 1076.0, + 945.0, + 958.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 906.0, + 1292.0, + 906.0, + 1292.0, + 945.0, + 1208.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 906.0, + 1407.0, + 906.0, + 1407.0, + 945.0, + 1402.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 939.0, + 647.0, + 939.0, + 647.0, + 973.0, + 294.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 939.0, + 791.0, + 939.0, + 791.0, + 973.0, + 667.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 939.0, + 1158.0, + 939.0, + 1158.0, + 973.0, + 807.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1242.0, + 939.0, + 1405.0, + 939.0, + 1405.0, + 973.0, + 1242.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 964.0, + 460.0, + 964.0, + 460.0, + 1019.0, + 288.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 964.0, + 1334.0, + 964.0, + 1334.0, + 1019.0, + 589.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 230.0, + 1403.0, + 230.0, + 1403.0, + 263.0, + 294.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 261.0, + 1408.0, + 261.0, + 1408.0, + 297.0, + 294.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 291.0, + 1410.0, + 291.0, + 1410.0, + 327.0, + 293.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 322.0, + 1206.0, + 322.0, + 1206.0, + 354.0, + 295.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1287.0, + 454.0, + 1287.0, + 454.0, + 1325.0, + 294.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1287.0, + 1408.0, + 1287.0, + 1408.0, + 1325.0, + 475.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1320.0, + 1405.0, + 1320.0, + 1405.0, + 1354.0, + 294.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1352.0, + 839.0, + 1352.0, + 839.0, + 1386.0, + 294.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1352.0, + 1203.0, + 1352.0, + 1203.0, + 1386.0, + 872.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1352.0, + 1235.0, + 1352.0, + 1235.0, + 1386.0, + 1224.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1856.0, + 1406.0, + 1856.0, + 1406.0, + 1897.0, + 293.0, + 1897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1892.0, + 491.0, + 1892.0, + 491.0, + 1926.0, + 294.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1892.0, + 842.0, + 1892.0, + 842.0, + 1926.0, + 513.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1892.0, + 1348.0, + 1892.0, + 1348.0, + 1926.0, + 854.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1688.0, + 368.0, + 1688.0, + 368.0, + 1739.0, + 291.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 1688.0, + 485.0, + 1688.0, + 485.0, + 1739.0, + 474.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 1688.0, + 807.0, + 1688.0, + 807.0, + 1739.0, + 627.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 1688.0, + 1410.0, + 1688.0, + 1410.0, + 1739.0, + 826.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1729.0, + 696.0, + 1729.0, + 696.0, + 1765.0, + 296.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 1729.0, + 1393.0, + 1729.0, + 1393.0, + 1765.0, + 778.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2000.0, + 1138.0, + 2000.0, + 1138.0, + 2036.0, + 295.0, + 2036.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 688, + 1404, + 688, + 1404, + 883, + 297, + 883 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1053, + 1403, + 1053, + 1403, + 1247, + 298, + 1247 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1270, + 1405, + 1270, + 1405, + 1398, + 298, + 1398 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1637, + 1404, + 1637, + 1404, + 1760, + 297, + 1760 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1419, + 1404, + 1419, + 1404, + 1544, + 298, + 1544 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 299, + 906, + 1405, + 906, + 1405, + 1000, + 299, + 1000 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 294, + 1405, + 294, + 1405, + 389, + 298, + 389 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 297, + 402, + 1406, + 402, + 1406, + 497, + 297, + 497 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 299, + 1775, + 1400, + 1775, + 1400, + 1868, + 299, + 1868 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 299, + 588, + 1403, + 588, + 1403, + 682, + 299, + 682 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 295, + 1971, + 1404, + 1971, + 1404, + 2035, + 295, + 2035 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 292, + 1876, + 1399, + 1876, + 1399, + 1939, + 292, + 1939 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 298, + 1014, + 1019, + 1014, + 1019, + 1046, + 298, + 1046 + ], + "score": 0.914 + }, + { + "category_id": 0, + "poly": [ + 298, + 225, + 1251, + 225, + 1251, + 262, + 298, + 262 + ], + "score": 0.901 + }, + { + "category_id": 0, + "poly": [ + 298, + 530, + 896, + 530, + 896, + 563, + 298, + 563 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.896 + }, + { + "category_id": 0, + "poly": [ + 300, + 1579, + 813, + 1579, + 813, + 1611, + 300, + 1611 + ], + "score": 0.867 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 857, + 2089, + 857, + 2111, + 840, + 2111 + ], + "score": 0.773 + }, + { + "category_id": 13, + "poly": [ + 761, + 2003, + 902, + 2003, + 902, + 2036, + 761, + 2036 + ], + "score": 0.93, + "latex": "\\{ h _ { 1 } , . . . , h _ { n } \\}" + }, + { + "category_id": 13, + "poly": [ + 1216, + 1268, + 1358, + 1268, + 1358, + 1308, + 1216, + 1308 + ], + "score": 0.93, + "latex": "f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right)" + }, + { + "category_id": 13, + "poly": [ + 1131, + 464, + 1391, + 464, + 1391, + 498, + 1131, + 498 + ], + "score": 0.93, + "latex": "[ n ] = \\{ 1 , 2 , . . . , n \\} \\subset \\mathbb { N }" + }, + { + "category_id": 13, + "poly": [ + 450, + 754, + 707, + 754, + 707, + 790, + 450, + 790 + ], + "score": 0.92, + "latex": "\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1906, + 498, + 1906, + 498, + 1940, + 297, + 1940 + ], + "score": 0.92, + "latex": "\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}" + }, + { + "category_id": 13, + "poly": [ + 422, + 1973, + 550, + 1973, + 550, + 2006, + 422, + 2006 + ], + "score": 0.92, + "latex": "\\mathcal { G } = ( \\nu , \\mathcal { E } )" + }, + { + "category_id": 13, + "poly": [ + 1015, + 1120, + 1132, + 1120, + 1132, + 1154, + 1015, + 1154 + ], + "score": 0.92, + "latex": "[ m ] [ n ]" + }, + { + "category_id": 13, + "poly": [ + 928, + 908, + 1013, + 908, + 1013, + 939, + 928, + 939 + ], + "score": 0.92, + "latex": "f \\in { \\mathcal { F } }" + }, + { + "category_id": 13, + "poly": [ + 1098, + 719, + 1353, + 719, + 1353, + 754, + 1098, + 754 + ], + "score": 0.92, + "latex": "\\mathbb { K } = \\{ k _ { 1 } , . . . , k _ { n } \\} \\subset \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 592, + 791, + 683, + 791, + 683, + 824, + 592, + 824 + ], + "score": 0.92, + "latex": "i \\in [ m ]" + }, + { + "category_id": 13, + "poly": [ + 498, + 1156, + 583, + 1156, + 583, + 1188, + 498, + 1188 + ], + "score": 0.92, + "latex": "i \\in [ m ]" + }, + { + "category_id": 13, + "poly": [ + 786, + 790, + 873, + 790, + 873, + 824, + 786, + 824 + ], + "score": 0.92, + "latex": "j \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 1268, + 1122, + 1344, + 1122, + 1344, + 1154, + 1268, + 1154 + ], + "score": 0.92, + "latex": "f \\in { \\mathcal { F } }" + }, + { + "category_id": 13, + "poly": [ + 500, + 1118, + 758, + 1118, + 758, + 1155, + 500, + 1155 + ], + "score": 0.91, + "latex": "\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1306, + 756, + 1400, + 756, + 1400, + 790, + 1306, + 790 + ], + "score": 0.91, + "latex": "j _ { f } \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 843, + 756, + 919, + 756, + 919, + 788, + 843, + 788 + ], + "score": 0.91, + "latex": "f \\in { \\mathcal { F } }" + }, + { + "category_id": 13, + "poly": [ + 434, + 1303, + 676, + 1303, + 676, + 1339, + 434, + 1339 + ], + "score": 0.91, + "latex": "\\left\\{ f \\left( \\pmb { q } _ { i } , \\pmb { k } _ { j } \\right) \\mid j \\in [ n ] \\right\\}" + }, + { + "category_id": 13, + "poly": [ + 826, + 1271, + 883, + 1271, + 883, + 1305, + 826, + 1305 + ], + "score": 0.91, + "latex": "\\varphi \\left( i \\right)" + }, + { + "category_id": 13, + "poly": [ + 1146, + 1085, + 1401, + 1085, + 1401, + 1119, + 1146, + 1119 + ], + "score": 0.9, + "latex": "\\mathbb { K } = \\{ \\boldsymbol { k } _ { 1 } , . . . , \\boldsymbol { k } _ { n } \\} \\subset \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1024, + 788, + 1297, + 788, + 1297, + 825, + 1024, + 825 + ], + "score": 0.9, + "latex": "f \\left( \\boldsymbol { q } _ { i } , \\boldsymbol { k } _ { j _ { f } } \\right) \\geq f \\left( \\boldsymbol { q } _ { i } , \\boldsymbol { k } _ { j } \\right)" + }, + { + "category_id": 13, + "poly": [ + 305, + 1083, + 483, + 1083, + 483, + 1120, + 305, + 1120 + ], + "score": 0.88, + "latex": "\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)" + }, + { + "category_id": 13, + "poly": [ + 905, + 464, + 939, + 464, + 939, + 498, + 905, + 498 + ], + "score": 0.87, + "latex": "[ n ]" + }, + { + "category_id": 13, + "poly": [ + 306, + 718, + 485, + 718, + 485, + 754, + 306, + 754 + ], + "score": 0.87, + "latex": "\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)" + }, + { + "category_id": 13, + "poly": [ + 703, + 1909, + 766, + 1909, + 766, + 1935, + 703, + 1935 + ], + "score": 0.87, + "latex": "n > 1" + }, + { + "category_id": 13, + "poly": [ + 1348, + 690, + 1404, + 690, + 1404, + 720, + 1348, + 720 + ], + "score": 0.85, + "latex": "\\mathcal { F } \\subseteq" + }, + { + "category_id": 13, + "poly": [ + 1122, + 824, + 1147, + 824, + 1147, + 852, + 1122, + 852 + ], + "score": 0.85, + "latex": "\\mathbb { Q }" + }, + { + "category_id": 13, + "poly": [ + 839, + 1367, + 863, + 1367, + 863, + 1396, + 839, + 1396 + ], + "score": 0.84, + "latex": "\\mathbb { Q }" + }, + { + "category_id": 13, + "poly": [ + 998, + 1188, + 1023, + 1188, + 1023, + 1216, + 998, + 1216 + ], + "score": 0.84, + "latex": "\\mathbb { Q }" + }, + { + "category_id": 13, + "poly": [ + 1348, + 1055, + 1404, + 1055, + 1404, + 1085, + 1348, + 1085 + ], + "score": 0.84, + "latex": "\\mathcal { F } \\subseteq" + }, + { + "category_id": 13, + "poly": [ + 1347, + 1878, + 1405, + 1878, + 1405, + 1907, + 1347, + 1907 + ], + "score": 0.83, + "latex": "\\mathbb { K } =" + }, + { + "category_id": 13, + "poly": [ + 763, + 1366, + 789, + 1366, + 789, + 1393, + 763, + 1393 + ], + "score": 0.77, + "latex": "\\mathbb { K }" + }, + { + "category_id": 13, + "poly": [ + 919, + 1188, + 945, + 1188, + 945, + 1213, + 919, + 1213 + ], + "score": 0.75, + "latex": "\\mathbb { K }" + }, + { + "category_id": 13, + "poly": [ + 1046, + 823, + 1072, + 823, + 1072, + 848, + 1046, + 848 + ], + "score": 0.74, + "latex": "\\Bbbk" + }, + { + "category_id": 13, + "poly": [ + 1169, + 2008, + 1190, + 2008, + 1190, + 2030, + 1169, + 2030 + ], + "score": 0.72, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 1060, + 1979, + 1080, + 1979, + 1080, + 2000, + 1060, + 2000 + ], + "score": 0.72, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 723, + 1153, + 1167, + 1153, + 1167, + 1190, + 723, + 1190 + ], + "score": 0.71, + "latex": "j _ { \\neq \\varphi ( i ) } \\in [ n ] \\colon f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right) > f \\left( q _ { i } , k _ { j } \\right) ." + }, + { + "category_id": 13, + "poly": [ + 725, + 1154, + 863, + 1154, + 863, + 1190, + 725, + 1190 + ], + "score": 0.64, + "latex": "j _ { \\neq \\varphi ( i ) } \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 1066, + 1274, + 1079, + 1274, + 1079, + 1300, + 1066, + 1300 + ], + "score": 0.63, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 981, + 1124, + 1002, + 1124, + 1002, + 1153, + 981, + 1153 + ], + "score": 0.58, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 1129, + 1973, + 1166, + 1973, + 1166, + 2001, + 1129, + 2001 + ], + "score": 0.51, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 634, + 1644, + 651, + 1644, + 651, + 1665, + 634, + 1665 + ], + "score": 0.49, + "latex": "e" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 222.0, + 1256.0, + 222.0, + 1256.0, + 267.0, + 290.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 530.0, + 897.0, + 530.0, + 897.0, + 566.0, + 295.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1579.0, + 817.0, + 1579.0, + 817.0, + 1615.0, + 294.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 683.0, + 1347.0, + 683.0, + 1347.0, + 726.0, + 292.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 714.0, + 305.0, + 714.0, + 305.0, + 759.0, + 297.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 714.0, + 1097.0, + 714.0, + 1097.0, + 759.0, + 486.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 714.0, + 1406.0, + 714.0, + 1406.0, + 759.0, + 1354.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 751.0, + 449.0, + 751.0, + 449.0, + 796.0, + 291.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 751.0, + 842.0, + 751.0, + 842.0, + 796.0, + 708.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 751.0, + 1305.0, + 751.0, + 1305.0, + 796.0, + 920.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 751.0, + 1406.0, + 751.0, + 1406.0, + 796.0, + 1401.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 783.0, + 591.0, + 783.0, + 591.0, + 832.0, + 292.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 783.0, + 785.0, + 783.0, + 785.0, + 832.0, + 684.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 783.0, + 1023.0, + 783.0, + 1023.0, + 832.0, + 874.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 783.0, + 1405.0, + 783.0, + 1405.0, + 832.0, + 1298.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 817.0, + 1045.0, + 817.0, + 1045.0, + 858.0, + 294.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 817.0, + 1121.0, + 817.0, + 1121.0, + 858.0, + 1073.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 817.0, + 1406.0, + 817.0, + 1406.0, + 858.0, + 1148.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 851.0, + 1287.0, + 851.0, + 1287.0, + 888.0, + 295.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1050.0, + 1347.0, + 1050.0, + 1347.0, + 1091.0, + 293.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1078.0, + 304.0, + 1078.0, + 304.0, + 1122.0, + 296.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1078.0, + 1145.0, + 1078.0, + 1145.0, + 1122.0, + 484.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1078.0, + 1407.0, + 1078.0, + 1407.0, + 1122.0, + 1402.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1119.0, + 499.0, + 1119.0, + 499.0, + 1158.0, + 293.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1119.0, + 980.0, + 1119.0, + 980.0, + 1158.0, + 759.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1119.0, + 1014.0, + 1119.0, + 1014.0, + 1158.0, + 1003.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1119.0, + 1267.0, + 1119.0, + 1267.0, + 1158.0, + 1133.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1119.0, + 1406.0, + 1119.0, + 1406.0, + 1158.0, + 1345.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1149.0, + 497.0, + 1149.0, + 497.0, + 1197.0, + 291.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1149.0, + 722.0, + 1149.0, + 722.0, + 1197.0, + 584.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 1149.0, + 1407.0, + 1149.0, + 1407.0, + 1197.0, + 1168.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1181.0, + 918.0, + 1181.0, + 918.0, + 1221.0, + 293.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 1181.0, + 997.0, + 1181.0, + 997.0, + 1221.0, + 946.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1181.0, + 1405.0, + 1181.0, + 1405.0, + 1221.0, + 1024.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1213.0, + 1208.0, + 1213.0, + 1208.0, + 1250.0, + 296.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1269.0, + 825.0, + 1269.0, + 825.0, + 1309.0, + 292.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 884.0, + 1269.0, + 1065.0, + 1269.0, + 1065.0, + 1309.0, + 884.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1269.0, + 1215.0, + 1269.0, + 1215.0, + 1309.0, + 1080.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 1269.0, + 1406.0, + 1269.0, + 1406.0, + 1309.0, + 1359.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1303.0, + 433.0, + 1303.0, + 433.0, + 1340.0, + 293.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1303.0, + 1408.0, + 1303.0, + 1408.0, + 1340.0, + 677.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1335.0, + 1403.0, + 1335.0, + 1403.0, + 1368.0, + 296.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1366.0, + 762.0, + 1366.0, + 762.0, + 1399.0, + 296.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1366.0, + 838.0, + 1366.0, + 838.0, + 1399.0, + 790.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 1366.0, + 1407.0, + 1366.0, + 1407.0, + 1399.0, + 864.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1638.0, + 633.0, + 1638.0, + 633.0, + 1671.0, + 295.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1638.0, + 1405.0, + 1638.0, + 1405.0, + 1671.0, + 652.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1667.0, + 1406.0, + 1667.0, + 1406.0, + 1703.0, + 294.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1697.0, + 1405.0, + 1697.0, + 1405.0, + 1735.0, + 291.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1728.0, + 1390.0, + 1728.0, + 1390.0, + 1762.0, + 292.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1419.0, + 1406.0, + 1419.0, + 1406.0, + 1456.0, + 294.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1450.0, + 1404.0, + 1450.0, + 1404.0, + 1486.0, + 293.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1481.0, + 1404.0, + 1481.0, + 1404.0, + 1517.0, + 295.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1511.0, + 1332.0, + 1511.0, + 1332.0, + 1548.0, + 293.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 907.0, + 927.0, + 907.0, + 927.0, + 941.0, + 295.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 907.0, + 1406.0, + 907.0, + 1406.0, + 941.0, + 1014.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 937.0, + 1403.0, + 937.0, + 1403.0, + 971.0, + 295.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 969.0, + 1137.0, + 969.0, + 1137.0, + 1003.0, + 295.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 294.0, + 1406.0, + 294.0, + 1406.0, + 332.0, + 293.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 326.0, + 1407.0, + 326.0, + 1407.0, + 360.0, + 295.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 355.0, + 1383.0, + 355.0, + 1383.0, + 391.0, + 293.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 402.0, + 1409.0, + 402.0, + 1409.0, + 440.0, + 294.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 434.0, + 1406.0, + 434.0, + 1406.0, + 468.0, + 294.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 463.0, + 904.0, + 463.0, + 904.0, + 499.0, + 292.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 463.0, + 1130.0, + 463.0, + 1130.0, + 499.0, + 940.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1392.0, + 463.0, + 1403.0, + 463.0, + 1403.0, + 499.0, + 1392.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1774.0, + 1405.0, + 1774.0, + 1405.0, + 1812.0, + 294.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1807.0, + 1404.0, + 1807.0, + 1404.0, + 1841.0, + 295.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1837.0, + 1205.0, + 1837.0, + 1205.0, + 1871.0, + 295.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 586.0, + 1405.0, + 586.0, + 1405.0, + 627.0, + 293.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 619.0, + 1405.0, + 619.0, + 1405.0, + 654.0, + 293.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 652.0, + 1147.0, + 652.0, + 1147.0, + 683.0, + 294.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1967.0, + 421.0, + 1967.0, + 421.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1967.0, + 1059.0, + 1967.0, + 1059.0, + 2009.0, + 551.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 1967.0, + 1128.0, + 1967.0, + 1128.0, + 2009.0, + 1081.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 1967.0, + 1407.0, + 1967.0, + 1407.0, + 2009.0, + 1167.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 760.0, + 2000.0, + 760.0, + 2040.0, + 294.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 2000.0, + 1168.0, + 2000.0, + 1168.0, + 2040.0, + 903.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 2000.0, + 1409.0, + 2000.0, + 1409.0, + 2040.0, + 1191.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1872.0, + 1346.0, + 1872.0, + 1346.0, + 1914.0, + 292.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1905.0, + 702.0, + 1905.0, + 702.0, + 1941.0, + 499.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1905.0, + 1331.0, + 1905.0, + 1331.0, + 1941.0, + 767.0, + 1941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1014.0, + 1022.0, + 1014.0, + 1022.0, + 1053.0, + 294.0, + 1053.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 558, + 1405, + 558, + 1405, + 776, + 297, + 776 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1649, + 1404, + 1649, + 1404, + 1835, + 297, + 1835 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 337, + 1403, + 337, + 1403, + 525, + 297, + 525 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 300, + 954, + 1403, + 954, + 1403, + 1079, + 300, + 1079 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 299, + 1858, + 1404, + 1858, + 1404, + 1982, + 299, + 1982 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 297, + 1325, + 1405, + 1325, + 1405, + 1420, + 297, + 1420 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 564, + 278, + 1130, + 278, + 1130, + 321, + 564, + 321 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 298, + 869, + 1403, + 869, + 1403, + 932, + 298, + 932 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 299, + 224, + 1380, + 224, + 1380, + 264, + 299, + 264 + ], + "score": 0.922 + }, + { + "category_id": 8, + "poly": [ + 793, + 1142, + 1282, + 1142, + 1282, + 1181, + 793, + 1181 + ], + "score": 0.917 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 816, + 74, + 816, + 105, + 299, + 105 + ], + "score": 0.909 + }, + { + "category_id": 1, + "poly": [ + 297, + 1238, + 1402, + 1238, + 1402, + 1301, + 297, + 1301 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 300, + 1580, + 536, + 1580, + 536, + 1615, + 300, + 1615 + ], + "score": 0.898 + }, + { + "category_id": 9, + "poly": [ + 1366, + 286, + 1400, + 286, + 1400, + 316, + 1366, + 316 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 317, + 2004, + 1352, + 2004, + 1352, + 2035, + 317, + 2035 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 299, + 811, + 996, + 811, + 996, + 843, + 299, + 843 + ], + "score": 0.877 + }, + { + "category_id": 1, + "poly": [ + 304, + 1195, + 1404, + 1195, + 1404, + 1229, + 304, + 1229 + ], + "score": 0.876 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1102, + 1400, + 1102, + 1400, + 1131, + 1366, + 1131 + ], + "score": 0.848 + }, + { + "category_id": 8, + "poly": [ + 798, + 1096, + 1317, + 1096, + 1317, + 1135, + 798, + 1135 + ], + "score": 0.838 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1148, + 1400, + 1148, + 1400, + 1177, + 1366, + 1177 + ], + "score": 0.83 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.751 + }, + { + "category_id": 8, + "poly": [ + 374, + 1095, + 1318, + 1095, + 1318, + 1135, + 374, + 1135 + ], + "score": 0.743 + }, + { + "category_id": 1, + "poly": [ + 298, + 1441, + 1406, + 1441, + 1406, + 1536, + 298, + 1536 + ], + "score": 0.74 + }, + { + "category_id": 13, + "poly": [ + 459, + 226, + 688, + 226, + 688, + 264, + 459, + 264 + ], + "score": 0.92, + "latex": "\\pmb { a } = [ \\pmb { a } _ { 1 } | | \\pmb { a } _ { 2 } ] \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }" + }, + { + "category_id": 13, + "poly": [ + 431, + 432, + 586, + 432, + 586, + 466, + 431, + 466 + ], + "score": 0.92, + "latex": "\\{ \\alpha _ { i j } \\mid j \\in \\mathcal { V } \\}" + }, + { + "category_id": 13, + "poly": [ + 934, + 337, + 1075, + 337, + 1075, + 374, + 934, + 374 + ], + "score": 0.92, + "latex": "{ \\pmb a } _ { 2 } ^ { \\top } { \\pmb W } { \\pmb h } _ { j _ { m a x } }" + }, + { + "category_id": 13, + "poly": [ + 800, + 226, + 948, + 226, + 948, + 263, + 800, + 263 + ], + "score": 0.92, + "latex": "\\mathbf { \\boldsymbol { a } } _ { 1 } , \\mathbf { \\boldsymbol { a } } _ { 2 } \\in \\mathbb { R } ^ { d ^ { \\prime } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1268, + 501, + 1268, + 501, + 1303, + 297, + 1303 + ], + "score": 0.92, + "latex": "\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}" + }, + { + "category_id": 13, + "poly": [ + 708, + 341, + 822, + 341, + 822, + 372, + 708, + 372 + ], + "score": 0.92, + "latex": "j _ { m a x } \\in \\mathcal { V }" + }, + { + "category_id": 14, + "poly": [ + 569, + 277, + 1130, + 277, + 1130, + 322, + 569, + 322 + ], + "score": 0.91, + "latex": "e \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { a } _ { 1 } ^ { \\top } \\boldsymbol { W } \\boldsymbol { h } _ { i } + \\boldsymbol { a } _ { 2 } ^ { \\top } \\boldsymbol { W } \\boldsymbol { h } _ { j } \\right)" + }, + { + "category_id": 13, + "poly": [ + 596, + 620, + 766, + 620, + 766, + 654, + 596, + 654 + ], + "score": 0.91, + "latex": "s _ { j } = { a } _ { 2 } ^ { \\top } \\bar { W } h _ { j }" + }, + { + "category_id": 13, + "poly": [ + 821, + 403, + 880, + 403, + 880, + 434, + 821, + 434 + ], + "score": 0.91, + "latex": "j _ { m a x }" + }, + { + "category_id": 13, + "poly": [ + 643, + 403, + 709, + 403, + 709, + 430, + 643, + 430 + ], + "score": 0.91, + "latex": "i \\in \\mathcal V" + }, + { + "category_id": 13, + "poly": [ + 1187, + 1442, + 1398, + 1442, + 1398, + 1476, + 1187, + 1476 + ], + "score": 0.9, + "latex": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 638, + 654, + 780, + 654, + 780, + 685, + 638, + 685 + ], + "score": 0.89, + "latex": "\\{ s _ { j } \\mid j \\in \\mathcal { V } \\}" + }, + { + "category_id": 13, + "poly": [ + 459, + 683, + 493, + 683, + 493, + 713, + 459, + 713 + ], + "score": 0.88, + "latex": "{ \\mathcal { N } } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 713, + 686, + 743, + 686, + 743, + 713, + 713, + 713 + ], + "score": 0.88, + "latex": "\\boldsymbol { h } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 522, + 372, + 552, + 372, + 552, + 405, + 522, + 405 + ], + "score": 0.88, + "latex": "j _ { f }" + }, + { + "category_id": 14, + "poly": [ + 791, + 1092, + 1324, + 1092, + 1324, + 1185, + 791, + 1185 + ], + "score": 0.87, + "latex": "\\begin{array} { r l } & { e \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { a } ^ { \\top } \\cdot \\left[ \\boldsymbol { W } \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { W } \\boldsymbol { h } _ { j } \\right| \\right) \\right. } \\\\ & { e \\left. \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\boldsymbol { a } ^ { \\top } \\mathrm { L e a k y R e L U } \\left( \\boldsymbol { W } \\cdot \\left[ \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { h } _ { j } \\right] \\right) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1317, + 713, + 1365, + 713, + 1365, + 744, + 1317, + 744 + ], + "score": 0.85, + "latex": "( h _ { i } )" + }, + { + "category_id": 13, + "poly": [ + 1120, + 622, + 1150, + 622, + 1150, + 652, + 1120, + 652 + ], + "score": 0.84, + "latex": "\\boldsymbol { h } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 552, + 497, + 574, + 497, + 574, + 525, + 552, + 525 + ], + "score": 0.83, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 1347, + 1239, + 1406, + 1239, + 1406, + 1268, + 1347, + 1268 + ], + "score": 0.83, + "latex": "\\mathbb { K } =" + }, + { + "category_id": 13, + "poly": [ + 487, + 470, + 508, + 470, + 508, + 490, + 487, + 490 + ], + "score": 0.8, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 367, + 342, + 390, + 342, + 390, + 368, + 367, + 368 + ], + "score": 0.8, + "latex": "\\nu" + }, + { + "category_id": 13, + "poly": [ + 400, + 595, + 422, + 595, + 422, + 619, + 400, + 619 + ], + "score": 0.76, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 980, + 437, + 1002, + 437, + 1002, + 460, + 980, + 460 + ], + "score": 0.76, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 953, + 562, + 977, + 562, + 977, + 589, + 953, + 589 + ], + "score": 0.72, + "latex": "\\nu" + }, + { + "category_id": 13, + "poly": [ + 702, + 1021, + 724, + 1021, + 724, + 1044, + 702, + 1044 + ], + "score": 0.72, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 457, + 991, + 478, + 991, + 478, + 1014, + 457, + 1014 + ], + "score": 0.71, + "latex": "^ { a }" + }, + { + "category_id": 13, + "poly": [ + 1164, + 626, + 1186, + 626, + 1186, + 649, + 1164, + 649 + ], + "score": 0.66, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 297, + 371, + 447, + 371, + 447, + 403, + 297, + 403 + ], + "score": 0.63, + "latex": "j \\in \\mathcal { V } \\left( j _ { m a x } \\right." + }, + { + "category_id": 13, + "poly": [ + 297, + 624, + 310, + 624, + 310, + 649, + 297, + 649 + ], + "score": 0.48, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 1301, + 1017, + 1338, + 1017, + 1338, + 1045, + 1301, + 1045 + ], + "score": 0.46, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 370, + 987, + 407, + 987, + 407, + 1015, + 370, + 1015 + ], + "score": 0.44, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 602, + 2008, + 618, + 2008, + 618, + 2030, + 602, + 2030 + ], + "score": 0.25, + "latex": "^ { b }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1576.0, + 543.0, + 1576.0, + 543.0, + 1622.0, + 291.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1995.0, + 601.0, + 1995.0, + 601.0, + 2043.0, + 329.0, + 2043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 1995.0, + 1355.0, + 1995.0, + 1355.0, + 2043.0, + 619.0, + 2043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 811.0, + 999.0, + 811.0, + 999.0, + 847.0, + 295.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 558.0, + 952.0, + 558.0, + 952.0, + 595.0, + 294.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 558.0, + 1405.0, + 558.0, + 1405.0, + 595.0, + 978.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 589.0, + 399.0, + 589.0, + 399.0, + 626.0, + 292.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 589.0, + 1405.0, + 589.0, + 1405.0, + 626.0, + 423.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 621.0, + 595.0, + 621.0, + 595.0, + 656.0, + 311.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 621.0, + 1119.0, + 621.0, + 1119.0, + 656.0, + 767.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 621.0, + 1163.0, + 621.0, + 1163.0, + 656.0, + 1151.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 621.0, + 1405.0, + 621.0, + 1405.0, + 656.0, + 1187.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 651.0, + 637.0, + 651.0, + 637.0, + 689.0, + 292.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 651.0, + 1406.0, + 651.0, + 1406.0, + 689.0, + 781.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 682.0, + 458.0, + 682.0, + 458.0, + 717.0, + 295.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 682.0, + 712.0, + 682.0, + 712.0, + 717.0, + 494.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 682.0, + 1408.0, + 682.0, + 1408.0, + 717.0, + 744.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 710.0, + 1316.0, + 710.0, + 1316.0, + 748.0, + 294.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 710.0, + 1407.0, + 710.0, + 1407.0, + 748.0, + 1366.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 744.0, + 1320.0, + 744.0, + 1320.0, + 779.0, + 295.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1651.0, + 1405.0, + 1651.0, + 1405.0, + 1683.0, + 296.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1679.0, + 1407.0, + 1679.0, + 1407.0, + 1715.0, + 294.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1711.0, + 1406.0, + 1711.0, + 1406.0, + 1747.0, + 294.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1742.0, + 1406.0, + 1742.0, + 1406.0, + 1778.0, + 294.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1771.0, + 1405.0, + 1771.0, + 1405.0, + 1807.0, + 294.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1802.0, + 1387.0, + 1802.0, + 1387.0, + 1836.0, + 292.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 335.0, + 366.0, + 335.0, + 366.0, + 382.0, + 291.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 335.0, + 707.0, + 335.0, + 707.0, + 382.0, + 391.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 335.0, + 933.0, + 335.0, + 933.0, + 382.0, + 823.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 335.0, + 1409.0, + 335.0, + 1409.0, + 382.0, + 1076.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 369.0, + 296.0, + 369.0, + 296.0, + 405.0, + 292.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 369.0, + 521.0, + 369.0, + 521.0, + 405.0, + 448.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 369.0, + 1405.0, + 369.0, + 1405.0, + 405.0, + 553.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 400.0, + 642.0, + 400.0, + 642.0, + 436.0, + 292.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 400.0, + 820.0, + 400.0, + 820.0, + 436.0, + 710.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 400.0, + 1405.0, + 400.0, + 1405.0, + 436.0, + 881.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 427.0, + 430.0, + 427.0, + 430.0, + 468.0, + 292.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 427.0, + 979.0, + 427.0, + 979.0, + 468.0, + 587.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 427.0, + 1407.0, + 427.0, + 1407.0, + 468.0, + 1003.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 462.0, + 486.0, + 462.0, + 486.0, + 498.0, + 295.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 462.0, + 1407.0, + 462.0, + 1407.0, + 498.0, + 509.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 491.0, + 551.0, + 491.0, + 551.0, + 529.0, + 294.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 491.0, + 1006.0, + 491.0, + 1006.0, + 529.0, + 575.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 493.0, + 1405.0, + 493.0, + 1405.0, + 524.0, + 1374.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 956.0, + 1403.0, + 956.0, + 1403.0, + 988.0, + 296.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 987.0, + 369.0, + 987.0, + 369.0, + 1020.0, + 295.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 987.0, + 456.0, + 987.0, + 456.0, + 1020.0, + 408.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 987.0, + 1405.0, + 987.0, + 1405.0, + 1020.0, + 479.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1014.0, + 701.0, + 1014.0, + 701.0, + 1052.0, + 293.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 1014.0, + 1300.0, + 1014.0, + 1300.0, + 1052.0, + 725.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1014.0, + 1407.0, + 1014.0, + 1407.0, + 1052.0, + 1339.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1045.0, + 1408.0, + 1045.0, + 1408.0, + 1084.0, + 294.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1858.0, + 1405.0, + 1858.0, + 1405.0, + 1894.0, + 294.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1886.0, + 1409.0, + 1886.0, + 1409.0, + 1924.0, + 292.0, + 1924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1919.0, + 1407.0, + 1919.0, + 1407.0, + 1954.0, + 293.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1948.0, + 1405.0, + 1948.0, + 1405.0, + 1987.0, + 293.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1326.0, + 1403.0, + 1326.0, + 1403.0, + 1360.0, + 295.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1358.0, + 1405.0, + 1358.0, + 1405.0, + 1392.0, + 295.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1388.0, + 1143.0, + 1388.0, + 1143.0, + 1422.0, + 296.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 871.0, + 1403.0, + 871.0, + 1403.0, + 903.0, + 297.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 899.0, + 1369.0, + 899.0, + 1369.0, + 935.0, + 296.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 221.0, + 458.0, + 221.0, + 458.0, + 269.0, + 291.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 221.0, + 799.0, + 221.0, + 799.0, + 269.0, + 689.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 221.0, + 1391.0, + 221.0, + 1391.0, + 269.0, + 949.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1233.0, + 1346.0, + 1233.0, + 1346.0, + 1276.0, + 292.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 1266.0, + 516.0, + 1266.0, + 516.0, + 1306.0, + 502.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1194.0, + 1408.0, + 1194.0, + 1408.0, + 1233.0, + 299.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1437.0, + 1186.0, + 1437.0, + 1186.0, + 1481.0, + 293.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1437.0, + 1409.0, + 1437.0, + 1409.0, + 1481.0, + 1399.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1471.0, + 1411.0, + 1471.0, + 1411.0, + 1508.0, + 292.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1502.0, + 1202.0, + 1502.0, + 1202.0, + 1538.0, + 293.0, + 1538.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1174, + 1403, + 1174, + 1403, + 1420, + 298, + 1420 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 1642, + 1405, + 1642, + 1405, + 1889, + 297, + 1889 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1435, + 1404, + 1435, + 1404, + 1619, + 298, + 1619 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 874, + 1404, + 874, + 1404, + 1059, + 298, + 1059 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1911, + 1404, + 1911, + 1404, + 2034, + 299, + 2034 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 766, + 1405, + 766, + 1405, + 858, + 298, + 858 + ], + "score": 0.97 + }, + { + "category_id": 3, + "poly": [ + 718, + 221, + 1390, + 221, + 1390, + 594, + 718, + 594 + ], + "score": 0.967 + }, + { + "category_id": 4, + "poly": [ + 296, + 448, + 690, + 448, + 690, + 722, + 296, + 722 + ], + "score": 0.967 + }, + { + "category_id": 4, + "poly": [ + 706, + 624, + 1396, + 624, + 1396, + 718, + 706, + 718 + ], + "score": 0.957 + }, + { + "category_id": 3, + "poly": [ + 362, + 221, + 622, + 221, + 622, + 427, + 362, + 427 + ], + "score": 0.943 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 297, + 1111, + 943, + 1111, + 943, + 1144, + 297, + 1144 + ], + "score": 0.831 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 858, + 2089, + 858, + 2112, + 840, + 2112 + ], + "score": 0.774 + }, + { + "category_id": 13, + "poly": [ + 952, + 1358, + 1136, + 1358, + 1136, + 1392, + 952, + 1392 + ], + "score": 0.93, + "latex": "k = \\{ 1 , 2 , 3 , \\ldots \\}" + }, + { + "category_id": 13, + "poly": [ + 1209, + 765, + 1397, + 765, + 1397, + 800, + 1209, + 800 + ], + "score": 0.92, + "latex": "W = [ W ^ { \\prime } | | W ^ { \\prime } ]" + }, + { + "category_id": 13, + "poly": [ + 671, + 1266, + 835, + 1266, + 835, + 1299, + 671, + 1299 + ], + "score": 0.9, + "latex": "( \\{ \\mathrm { A } , \\mathrm { B } , \\mathrm { C } , \\bar { \\ldots } \\} )" + }, + { + "category_id": 13, + "poly": [ + 475, + 1736, + 559, + 1736, + 559, + 1766, + 475, + 1766 + ], + "score": 0.9, + "latex": "\\mathrm { G A T } _ { 8 h }" + }, + { + "category_id": 13, + "poly": [ + 1264, + 656, + 1347, + 656, + 1347, + 684, + 1264, + 684 + ], + "score": 0.88, + "latex": "k { = } 1 0 0" + }, + { + "category_id": 13, + "poly": [ + 987, + 1267, + 1131, + 1267, + 1131, + 1299, + 987, + 1299 + ], + "score": 0.88, + "latex": "( \\{ 1 , 2 , 3 , \\ldots \\} )" + }, + { + "category_id": 13, + "poly": [ + 337, + 1796, + 420, + 1796, + 420, + 1825, + 337, + 1825 + ], + "score": 0.88, + "latex": "k { = } 1 0 0" + }, + { + "category_id": 13, + "poly": [ + 813, + 656, + 881, + 656, + 881, + 685, + 813, + 685 + ], + "score": 0.87, + "latex": "100 \\%" + }, + { + "category_id": 13, + "poly": [ + 1229, + 1644, + 1322, + 1644, + 1322, + 1675, + 1229, + 1675 + ], + "score": 0.87, + "latex": "( \\mathrm { G A T _ { 1 } } _ { h } )" + }, + { + "category_id": 13, + "poly": [ + 654, + 1765, + 719, + 1765, + 719, + 1794, + 654, + 1794 + ], + "score": 0.87, + "latex": "100 \\%" + }, + { + "category_id": 13, + "poly": [ + 861, + 1765, + 927, + 1765, + 927, + 1794, + 861, + 1794 + ], + "score": 0.87, + "latex": "100 \\%" + }, + { + "category_id": 13, + "poly": [ + 523, + 540, + 686, + 540, + 686, + 572, + 523, + 572 + ], + "score": 0.86, + "latex": "( \\{ { \\mathrm { A } } , { \\mathrm { B } } , { \\mathrm { C } } , \\ldots \\} )" + }, + { + "category_id": 13, + "poly": [ + 1057, + 1237, + 1089, + 1237, + 1089, + 1264, + 1057, + 1264 + ], + "score": 0.85, + "latex": "2 k" + }, + { + "category_id": 13, + "poly": [ + 865, + 1944, + 896, + 1944, + 896, + 1974, + 865, + 1974 + ], + "score": 0.85, + "latex": "q 9" + }, + { + "category_id": 13, + "poly": [ + 585, + 480, + 640, + 480, + 640, + 508, + 585, + 508 + ], + "score": 0.85, + "latex": "k { = } 4" + }, + { + "category_id": 13, + "poly": [ + 702, + 1299, + 865, + 1299, + 865, + 1331, + 702, + 1331 + ], + "score": 0.83, + "latex": "( \\{ \\mathrm { A } , \\mathrm { B } , \\mathrm { C } , \\ldots \\} )" + }, + { + "category_id": 13, + "poly": [ + 529, + 573, + 679, + 573, + 679, + 604, + 529, + 604 + ], + "score": 0.8, + "latex": "( \\{ 1 , 2 , 3 , \\ldots \\} )" + }, + { + "category_id": 13, + "poly": [ + 1273, + 1766, + 1291, + 1766, + 1291, + 1792, + 1273, + 1792 + ], + "score": 0.78, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 717, + 1676, + 735, + 1676, + 735, + 1701, + 717, + 1701 + ], + "score": 0.78, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 803, + 1945, + 834, + 1945, + 834, + 1974, + 803, + 1974 + ], + "score": 0.77, + "latex": "q 0" + }, + { + "category_id": 13, + "poly": [ + 780, + 565, + 798, + 565, + 798, + 589, + 780, + 589 + ], + "score": 0.51, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1241, + 1943, + 1290, + 1943, + 1290, + 1973, + 1241, + 1973 + ], + "score": 0.38, + "latex": "( k 8 )" + }, + { + "category_id": 13, + "poly": [ + 1253, + 328, + 1327, + 328, + 1327, + 356, + 1253, + 356 + ], + "score": 0.38, + "latex": "\\mathrm { G A T _ { 1 h } }" + }, + { + "category_id": 13, + "poly": [ + 1212, + 296, + 1328, + 296, + 1328, + 324, + 1212, + 324 + ], + "score": 0.38, + "latex": "O \\mathrm { - } \\mathbf { G } \\mathbf { A } \\mathrm { T } _ { 1 h }" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 232.0, + 787.0, + 232.0, + 787.0, + 261.0, + 741.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 227.0, + 1377.0, + 227.0, + 1377.0, + 263.0, + 1213.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 258.0, + 787.0, + 258.0, + 787.0, + 290.0, + 752.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 271.0, + 1238.0, + 271.0, + 1238.0, + 286.0, + 1217.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 262.0, + 1376.0, + 262.0, + 1376.0, + 294.0, + 1247.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 282.0, + 791.0, + 282.0, + 791.0, + 426.0, + 717.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 281.0, + 810.0, + 281.0, + 810.0, + 292.0, + 797.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 303.0, + 1239.0, + 303.0, + 1239.0, + 351.0, + 1217.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.0, + 291.0, + 1385.0, + 291.0, + 1385.0, + 360.0, + 1329.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 419.0, + 787.0, + 419.0, + 787.0, + 451.0, + 752.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 446.0, + 787.0, + 446.0, + 787.0, + 478.0, + 752.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 475.0, + 786.0, + 475.0, + 786.0, + 503.0, + 754.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 505.0, + 785.0, + 505.0, + 785.0, + 527.0, + 766.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 520.0, + 1207.0, + 520.0, + 1207.0, + 544.0, + 792.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 559.0, + 1210.0, + 559.0, + 1210.0, + 597.0, + 799.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 446.0, + 696.0, + 446.0, + 696.0, + 480.0, + 293.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 480.0, + 584.0, + 480.0, + 584.0, + 511.0, + 295.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 480.0, + 694.0, + 480.0, + 694.0, + 511.0, + 641.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 509.0, + 691.0, + 509.0, + 691.0, + 540.0, + 293.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 540.0, + 522.0, + 540.0, + 522.0, + 574.0, + 295.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 540.0, + 691.0, + 540.0, + 691.0, + 574.0, + 687.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 570.0, + 528.0, + 570.0, + 528.0, + 604.0, + 294.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 570.0, + 691.0, + 570.0, + 691.0, + 604.0, + 680.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 602.0, + 691.0, + 602.0, + 691.0, + 633.0, + 295.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 633.0, + 693.0, + 633.0, + 693.0, + 664.0, + 295.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 662.0, + 690.0, + 662.0, + 690.0, + 692.0, + 296.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 694.0, + 626.0, + 694.0, + 626.0, + 724.0, + 294.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 622.0, + 1397.0, + 622.0, + 1397.0, + 659.0, + 705.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 656.0, + 812.0, + 656.0, + 812.0, + 687.0, + 707.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 656.0, + 1263.0, + 656.0, + 1263.0, + 687.0, + 882.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 656.0, + 1397.0, + 656.0, + 1397.0, + 687.0, + 1348.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 686.0, + 990.0, + 686.0, + 990.0, + 719.0, + 706.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 220.0, + 626.0, + 220.0, + 626.0, + 285.0, + 362.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 375.0, + 426.0, + 375.0, + 426.0, + 425.0, + 367.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 370.0, + 618.0, + 370.0, + 618.0, + 429.0, + 428.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1110.0, + 945.0, + 1110.0, + 945.0, + 1147.0, + 294.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1176.0, + 1407.0, + 1176.0, + 1407.0, + 1209.0, + 294.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1204.0, + 1405.0, + 1204.0, + 1405.0, + 1239.0, + 291.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1237.0, + 1056.0, + 1237.0, + 1056.0, + 1271.0, + 294.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1237.0, + 1405.0, + 1237.0, + 1405.0, + 1271.0, + 1090.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1266.0, + 670.0, + 1266.0, + 670.0, + 1303.0, + 293.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1266.0, + 986.0, + 1266.0, + 986.0, + 1303.0, + 836.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1266.0, + 1407.0, + 1266.0, + 1407.0, + 1303.0, + 1132.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1293.0, + 701.0, + 1293.0, + 701.0, + 1338.0, + 291.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1293.0, + 1405.0, + 1293.0, + 1405.0, + 1338.0, + 866.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1326.0, + 1407.0, + 1326.0, + 1407.0, + 1364.0, + 292.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1359.0, + 951.0, + 1359.0, + 951.0, + 1393.0, + 294.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1359.0, + 1407.0, + 1359.0, + 1407.0, + 1393.0, + 1137.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1386.0, + 856.0, + 1386.0, + 856.0, + 1426.0, + 293.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1644.0, + 1228.0, + 1644.0, + 1228.0, + 1678.0, + 295.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 1644.0, + 1405.0, + 1644.0, + 1405.0, + 1678.0, + 1323.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1674.0, + 716.0, + 1674.0, + 716.0, + 1706.0, + 292.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1674.0, + 1407.0, + 1674.0, + 1407.0, + 1706.0, + 736.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1704.0, + 1408.0, + 1704.0, + 1408.0, + 1740.0, + 292.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1735.0, + 474.0, + 1735.0, + 474.0, + 1769.0, + 295.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1735.0, + 1405.0, + 1735.0, + 1405.0, + 1769.0, + 560.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1766.0, + 653.0, + 1766.0, + 653.0, + 1799.0, + 294.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1766.0, + 860.0, + 1766.0, + 860.0, + 1799.0, + 720.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1766.0, + 1272.0, + 1766.0, + 1272.0, + 1799.0, + 928.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 1766.0, + 1405.0, + 1766.0, + 1405.0, + 1799.0, + 1292.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1794.0, + 336.0, + 1794.0, + 336.0, + 1829.0, + 292.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 1794.0, + 1407.0, + 1794.0, + 1407.0, + 1829.0, + 421.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1826.0, + 1405.0, + 1826.0, + 1405.0, + 1859.0, + 295.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1856.0, + 1328.0, + 1856.0, + 1328.0, + 1892.0, + 292.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1434.0, + 1404.0, + 1434.0, + 1404.0, + 1470.0, + 294.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1466.0, + 1405.0, + 1466.0, + 1405.0, + 1502.0, + 292.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1493.0, + 1408.0, + 1493.0, + 1408.0, + 1535.0, + 291.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1527.0, + 1408.0, + 1527.0, + 1408.0, + 1563.0, + 293.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1555.0, + 1406.0, + 1555.0, + 1406.0, + 1592.0, + 293.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1588.0, + 1321.0, + 1588.0, + 1321.0, + 1623.0, + 293.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 875.0, + 1409.0, + 875.0, + 1409.0, + 911.0, + 296.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 905.0, + 1405.0, + 905.0, + 1405.0, + 937.0, + 294.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 933.0, + 1405.0, + 933.0, + 1405.0, + 969.0, + 294.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 965.0, + 1406.0, + 965.0, + 1406.0, + 1001.0, + 293.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 998.0, + 1405.0, + 998.0, + 1405.0, + 1030.0, + 294.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1025.0, + 1220.0, + 1025.0, + 1220.0, + 1063.0, + 292.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 1409.0, + 1911.0, + 1409.0, + 1947.0, + 294.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1943.0, + 802.0, + 1943.0, + 802.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 1943.0, + 864.0, + 1943.0, + 864.0, + 1976.0, + 835.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1943.0, + 1240.0, + 1943.0, + 1240.0, + 1976.0, + 897.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 1943.0, + 1404.0, + 1943.0, + 1404.0, + 1976.0, + 1291.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2007.0, + 293.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2000.0, + 1408.0, + 2000.0, + 1408.0, + 2039.0, + 292.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 763.0, + 1208.0, + 763.0, + 1208.0, + 804.0, + 292.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 763.0, + 1407.0, + 763.0, + 1407.0, + 804.0, + 1398.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 798.0, + 1405.0, + 798.0, + 1405.0, + 831.0, + 295.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 828.0, + 1192.0, + 828.0, + 1192.0, + 861.0, + 295.0, + 861.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1207, + 1405, + 1207, + 1405, + 1453, + 298, + 1453 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 900, + 1793, + 1380, + 1793, + 1380, + 1993, + 900, + 1993 + ], + "score": 0.978, + "html": "
ModelSeenProjUnseenProj
No-GCN+87.2±1.581.4±2.3
AttentionGIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
" + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1402, + 229, + 1402, + 353, + 298, + 353 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 457, + 1405, + 457, + 1405, + 551, + 299, + 551 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1643, + 867, + 1643, + 867, + 1795, + 298, + 1795 + ], + "score": 0.972 + }, + { + "category_id": 3, + "poly": [ + 302, + 589, + 1379, + 589, + 1379, + 1073, + 302, + 1073 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 300, + 1820, + 871, + 1820, + 871, + 2024, + 300, + 2024 + ], + "score": 0.964 + }, + { + "category_id": 4, + "poly": [ + 297, + 1097, + 1403, + 1097, + 1403, + 1161, + 297, + 1161 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 298, + 1564, + 1401, + 1564, + 1401, + 1627, + 298, + 1627 + ], + "score": 0.949 + }, + { + "category_id": 0, + "poly": [ + 299, + 1496, + 665, + 1496, + 665, + 1528, + 299, + 1528 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 300, + 398, + 647, + 398, + 647, + 428, + 300, + 428 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.897 + }, + { + "category_id": 1, + "poly": [ + 297, + 2004, + 1187, + 2004, + 1187, + 2034, + 297, + 2034 + ], + "score": 0.859 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.772 + }, + { + "category_id": 1, + "poly": [ + 892, + 1644, + 1405, + 1644, + 1405, + 1768, + 892, + 1768 + ], + "score": 0.601 + }, + { + "category_id": 6, + "poly": [ + 892, + 1644, + 1405, + 1644, + 1405, + 1768, + 892, + 1768 + ], + "score": 0.465 + }, + { + "category_id": 13, + "poly": [ + 1231, + 518, + 1397, + 518, + 1397, + 552, + 1231, + 552 + ], + "score": 0.93, + "latex": "\\mathcal { G } ^ { \\prime } { = } ( \\dot { \\mathcal { V } } , \\mathcal { E } \\cup \\bar { \\mathcal { E } } ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 726, + 488, + 848, + 488, + 848, + 521, + 726, + 521 + ], + "score": 0.93, + "latex": "\\mathcal { G } = ( \\nu , \\mathcal { E } )" + }, + { + "category_id": 13, + "poly": [ + 1045, + 490, + 1151, + 490, + 1151, + 520, + 1045, + 520 + ], + "score": 0.92, + "latex": "0 \\leq p \\leq 1" + }, + { + "category_id": 13, + "poly": [ + 669, + 520, + 762, + 520, + 762, + 551, + 669, + 551 + ], + "score": 0.92, + "latex": "\\mathcal { V } \\times \\mathcal { V } \\backslash \\mathcal { E }" + }, + { + "category_id": 13, + "poly": [ + 299, + 518, + 369, + 518, + 369, + 552, + 299, + 552 + ], + "score": 0.91, + "latex": "| { \\mathcal { E } } | \\times p" + }, + { + "category_id": 13, + "poly": [ + 579, + 519, + 607, + 519, + 607, + 547, + 579, + 547 + ], + "score": 0.86, + "latex": "\\mathcal { E } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 830, + 1361, + 875, + 1361, + 875, + 1392, + 830, + 1392 + ], + "score": 0.83, + "latex": "( \\mathcal { E } ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 1072, + 1244, + 1091, + 1244, + 1091, + 1270, + 1072, + 1270 + ], + "score": 0.79, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1000, + 1244, + 1017, + 1244, + 1017, + 1270, + 1000, + 1270 + ], + "score": 0.74, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 598, + 1361, + 635, + 1361, + 635, + 1391, + 598, + 1391 + ], + "score": 0.69, + "latex": "( \\mathcal { E } )" + }, + { + "category_id": 13, + "poly": [ + 1170, + 1646, + 1193, + 1646, + 1193, + 1672, + 1170, + 1672 + ], + "score": 0.35, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 890, + 1736, + 908, + 1736, + 908, + 1769, + 890, + 1769 + ], + "score": 0.25, + "latex": "\\dagger" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 584.0, + 390.0, + 584.0, + 390.0, + 629.0, + 345.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 603.0, + 790.0, + 603.0, + 790.0, + 649.0, + 529.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 607.0, + 1341.0, + 607.0, + 1341.0, + 646.0, + 1072.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 669.0, + 437.0, + 669.0, + 437.0, + 690.0, + 424.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 652.0, + 557.0, + 652.0, + 557.0, + 675.0, + 530.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 640.0, + 638.0, + 640.0, + 638.0, + 683.0, + 562.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 635.0, + 950.0, + 635.0, + 950.0, + 694.0, + 892.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 651.0, + 1109.0, + 651.0, + 1109.0, + 675.0, + 1081.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 640.0, + 1189.0, + 640.0, + 1189.0, + 683.0, + 1113.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 684.0, + 388.0, + 684.0, + 388.0, + 728.0, + 342.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 706.0, + 342.0, + 706.0, + 342.0, + 829.0, + 300.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 708.0, + 892.0, + 708.0, + 892.0, + 829.0, + 852.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 720.0, + 1240.0, + 720.0, + 1240.0, + 732.0, + 1230.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 728.0, + 482.0, + 728.0, + 482.0, + 747.0, + 463.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 731.0, + 519.0, + 731.0, + 519.0, + 743.0, + 509.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 735.0, + 990.0, + 735.0, + 990.0, + 749.0, + 974.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 745.0, + 566.0, + 745.0, + 566.0, + 765.0, + 546.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 771.0, + 524.0, + 771.0, + 524.0, + 787.0, + 507.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 768.0, + 605.0, + 768.0, + 605.0, + 779.0, + 593.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 760.0, + 939.0, + 760.0, + 939.0, + 801.0, + 893.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 785.0, + 386.0, + 785.0, + 386.0, + 825.0, + 340.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 824.0, + 603.0, + 824.0, + 603.0, + 832.0, + 593.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 825.0, + 730.0, + 825.0, + 730.0, + 835.0, + 720.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 815.0, + 1075.0, + 815.0, + 1075.0, + 831.0, + 1058.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 835.0, + 647.0, + 835.0, + 647.0, + 843.0, + 637.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 847.0, + 1115.0, + 847.0, + 1115.0, + 863.0, + 1102.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 852.0, + 1157.0, + 852.0, + 1157.0, + 863.0, + 1145.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 852.0, + 1196.0, + 852.0, + 1196.0, + 861.0, + 1186.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 857.0, + 693.0, + 857.0, + 693.0, + 877.0, + 673.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 869.0, + 733.0, + 869.0, + 733.0, + 880.0, + 720.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 869.0, + 815.0, + 869.0, + 815.0, + 883.0, + 800.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 867.0, + 940.0, + 867.0, + 940.0, + 910.0, + 891.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 872.0, + 1240.0, + 872.0, + 1240.0, + 882.0, + 1230.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 880.0, + 391.0, + 880.0, + 391.0, + 930.0, + 338.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 895.0, + 1286.0, + 895.0, + 1286.0, + 909.0, + 1270.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 917.0, + 816.0, + 917.0, + 816.0, + 928.0, + 804.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 945.0, + 403.0, + 945.0, + 403.0, + 979.0, + 373.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 943.0, + 500.0, + 943.0, + 500.0, + 983.0, + 443.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 943.0, + 593.0, + 943.0, + 593.0, + 983.0, + 528.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 943.0, + 671.0, + 943.0, + 671.0, + 983.0, + 610.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 943.0, + 757.0, + 943.0, + 757.0, + 983.0, + 696.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 943.0, + 838.0, + 943.0, + 838.0, + 983.0, + 779.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 947.0, + 955.0, + 947.0, + 955.0, + 979.0, + 925.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 943.0, + 1053.0, + 943.0, + 1053.0, + 983.0, + 995.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 943.0, + 1138.0, + 943.0, + 1138.0, + 983.0, + 1078.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 943.0, + 1222.0, + 943.0, + 1222.0, + 983.0, + 1162.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 943.0, + 1308.0, + 943.0, + 1308.0, + 983.0, + 1248.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 944.0, + 1387.0, + 944.0, + 1387.0, + 981.0, + 1332.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 988.0, + 685.0, + 988.0, + 685.0, + 1022.0, + 510.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 988.0, + 1237.0, + 988.0, + 1237.0, + 1023.0, + 1061.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1037.0, + 680.0, + 1037.0, + 680.0, + 1075.0, + 515.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 1034.0, + 1232.0, + 1034.0, + 1232.0, + 1082.0, + 1071.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 631.5, + 410.0, + 631.5, + 410.0, + 649.5, + 389.0, + 649.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.25, + 884.0, + 778.25, + 884.0, + 778.25, + 907.0, + 755.25, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1098.0, + 1403.0, + 1098.0, + 1403.0, + 1130.0, + 297.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1126.0, + 1305.0, + 1126.0, + 1305.0, + 1164.0, + 293.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1493.0, + 670.0, + 1493.0, + 670.0, + 1533.0, + 294.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 395.0, + 650.0, + 395.0, + 650.0, + 433.0, + 294.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 1643.0, + 1169.0, + 1643.0, + 1169.0, + 1676.0, + 888.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1643.0, + 1407.0, + 1643.0, + 1407.0, + 1676.0, + 1194.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1674.0, + 1404.0, + 1674.0, + 1404.0, + 1705.0, + 890.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1704.0, + 1408.0, + 1704.0, + 1408.0, + 1738.0, + 889.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 1737.0, + 1404.0, + 1737.0, + 1404.0, + 1769.0, + 909.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1209.0, + 1407.0, + 1209.0, + 1407.0, + 1242.0, + 296.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1239.0, + 999.0, + 1239.0, + 999.0, + 1272.0, + 295.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 1239.0, + 1071.0, + 1239.0, + 1071.0, + 1272.0, + 1018.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1239.0, + 1405.0, + 1239.0, + 1405.0, + 1272.0, + 1092.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1269.0, + 1405.0, + 1269.0, + 1405.0, + 1304.0, + 292.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1298.0, + 1406.0, + 1298.0, + 1406.0, + 1335.0, + 293.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1326.0, + 1405.0, + 1326.0, + 1405.0, + 1368.0, + 292.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1359.0, + 597.0, + 1359.0, + 597.0, + 1397.0, + 292.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1359.0, + 829.0, + 1359.0, + 829.0, + 1397.0, + 636.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1359.0, + 1406.0, + 1359.0, + 1406.0, + 1397.0, + 876.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1392.0, + 1403.0, + 1392.0, + 1403.0, + 1426.0, + 295.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1423.0, + 1396.0, + 1423.0, + 1396.0, + 1456.0, + 295.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 229.0, + 1406.0, + 229.0, + 1406.0, + 263.0, + 293.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 261.0, + 1407.0, + 261.0, + 1407.0, + 297.0, + 293.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 292.0, + 1404.0, + 292.0, + 1404.0, + 325.0, + 294.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 321.0, + 1377.0, + 321.0, + 1377.0, + 354.0, + 294.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 455.0, + 1406.0, + 455.0, + 1406.0, + 493.0, + 294.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 487.0, + 725.0, + 487.0, + 725.0, + 521.0, + 295.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 487.0, + 1044.0, + 487.0, + 1044.0, + 521.0, + 849.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 487.0, + 1403.0, + 487.0, + 1403.0, + 521.0, + 1152.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 516.0, + 298.0, + 516.0, + 298.0, + 554.0, + 295.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 516.0, + 578.0, + 516.0, + 578.0, + 554.0, + 370.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 516.0, + 668.0, + 516.0, + 668.0, + 554.0, + 608.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 516.0, + 1230.0, + 516.0, + 1230.0, + 554.0, + 763.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 516.0, + 1408.0, + 516.0, + 1408.0, + 554.0, + 1398.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1640.0, + 873.0, + 1640.0, + 873.0, + 1678.0, + 296.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1673.0, + 871.0, + 1673.0, + 871.0, + 1709.0, + 294.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1704.0, + 870.0, + 1704.0, + 870.0, + 1738.0, + 295.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1733.0, + 871.0, + 1733.0, + 871.0, + 1769.0, + 293.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1758.0, + 858.0, + 1758.0, + 858.0, + 1803.0, + 293.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1818.0, + 875.0, + 1818.0, + 875.0, + 1853.0, + 295.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1850.0, + 871.0, + 1850.0, + 871.0, + 1883.0, + 295.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1883.0, + 872.0, + 1883.0, + 872.0, + 1912.0, + 295.0, + 1912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1912.0, + 873.0, + 1912.0, + 873.0, + 1944.0, + 294.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1943.0, + 872.0, + 1943.0, + 872.0, + 1975.0, + 295.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 871.0, + 1973.0, + 871.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 875.0, + 2002.0, + 875.0, + 2031.0, + 295.0, + 2031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1565.0, + 1405.0, + 1565.0, + 1405.0, + 1601.0, + 296.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1596.0, + 1279.0, + 1596.0, + 1279.0, + 1631.0, + 296.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1997.0, + 1192.0, + 1997.0, + 1192.0, + 2043.0, + 292.0, + 2043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 1643.0, + 1169.0, + 1643.0, + 1169.0, + 1676.0, + 888.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1643.0, + 1407.0, + 1643.0, + 1407.0, + 1676.0, + 1194.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1674.0, + 1404.0, + 1674.0, + 1404.0, + 1705.0, + 890.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1704.0, + 1408.0, + 1704.0, + 1408.0, + 1738.0, + 889.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 1737.0, + 1404.0, + 1737.0, + 1404.0, + 1769.0, + 909.0, + 1769.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 317, + 1337, + 1382, + 1337, + 1382, + 1570, + 317, + 1570 + ], + "score": 0.98, + "html": "
Predicted Property
Model12345678910111213Rel. to GAT
GCNt3.214.221.451.622.4216.38 17.40 7.828.249.057.003.931.02-1.5%
GINt2.644.671.421.502.2715.63 12.93 5.8818.715.625.383.531.05-2.3%
GATt2.684.651.481.532.3152.39 14.87 7.6116.867.646.544.111.48+0%
GATv22.654.281.411.472.2916.37 14.03 6.076.286.605.973.571.59-11.5%
" + }, + { + "category_id": 1, + "poly": [ + 299, + 1069, + 1404, + 1069, + 1404, + 1193, + 299, + 1193 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 1880, + 1404, + 1880, + 1404, + 2034, + 299, + 2034 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 1608, + 1403, + 1608, + 1403, + 1731, + 298, + 1731 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 298, + 782, + 1406, + 782, + 1406, + 968, + 298, + 968 + ], + "score": 0.953 + }, + { + "category_id": 6, + "poly": [ + 299, + 1219, + 1399, + 1219, + 1399, + 1315, + 299, + 1315 + ], + "score": 0.939 + }, + { + "category_id": 6, + "poly": [ + 297, + 345, + 1398, + 345, + 1398, + 411, + 297, + 411 + ], + "score": 0.913 + }, + { + "category_id": 0, + "poly": [ + 300, + 1767, + 589, + 1767, + 589, + 1799, + 300, + 1799 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 300, + 229, + 598, + 229, + 598, + 261, + 300, + 261 + ], + "score": 0.907 + }, + { + "category_id": 1, + "poly": [ + 297, + 286, + 1367, + 286, + 1367, + 320, + 297, + 320 + ], + "score": 0.899 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.897 + }, + { + "category_id": 1, + "poly": [ + 299, + 1823, + 1212, + 1823, + 1212, + 1857, + 299, + 1857 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.793 + }, + { + "category_id": 0, + "poly": [ + 301, + 1003, + 691, + 1003, + 691, + 1036, + 301, + 1036 + ], + "score": 0.689 + }, + { + "category_id": 1, + "poly": [ + 301, + 1003, + 691, + 1003, + 691, + 1036, + 301, + 1036 + ], + "score": 0.231 + }, + { + "category_id": 1, + "poly": [ + 1275, + 437, + 1310, + 437, + 1310, + 465, + 1275, + 465 + ], + "score": 0.191 + }, + { + "category_id": 9, + "poly": [ + 1275, + 437, + 1310, + 437, + 1310, + 465, + 1275, + 465 + ], + "score": 0.126 + }, + { + "category_id": 13, + "poly": [ + 296, + 1639, + 369, + 1639, + 369, + 1668, + 296, + 1668 + ], + "score": 0.82, + "latex": "1 1 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 565, + 1283, + 583, + 1283, + 583, + 1314, + 565, + 1314 + ], + "score": 0.75, + "latex": "\\dagger" + }, + { + "category_id": 13, + "poly": [ + 296, + 845, + 377, + 845, + 377, + 875, + 296, + 875 + ], + "score": 0.28, + "latex": "G A T \\nu 2" + }, + { + "category_id": 5, + "poly": [ + 306, + 433, + 1388, + 433, + 1388, + 746, + 306, + 746 + ], + "score": 0.638, + "html": "
(a)(b)
ModelAttn. Headsogbn-arxivogbn-productsogbn-magogbn-proteins
GCN+071.74±0.2978.97±0.3330.43±0.2572.51±0.35
GraphSAGE+071.49±0.2778.70±0.3631.53±0.1577.68±0.20
GAT171.59±0.3879.04±1.5432.20±1.4670.77±5.79
871.54±0.3077.23±2.3731.75±1.6078.63±1.62
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.4477.23±3.32
871.87±0.2578.46±2.4532.52±0.3979.52±0.55
" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1217.0, + 1407.0, + 1217.0, + 1407.0, + 1256.0, + 292.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1251.0, + 1404.0, + 1251.0, + 1404.0, + 1285.0, + 294.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1280.0, + 564.0, + 1280.0, + 564.0, + 1316.0, + 294.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1280.0, + 1255.0, + 1280.0, + 1255.0, + 1316.0, + 584.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 348.0, + 1403.0, + 348.0, + 1403.0, + 381.0, + 296.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 377.0, + 1372.0, + 377.0, + 1372.0, + 413.0, + 295.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1764.0, + 591.0, + 1764.0, + 591.0, + 1803.0, + 294.0, + 1803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 226.0, + 600.0, + 226.0, + 600.0, + 265.0, + 294.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 2085.0, + 859.0, + 2085.0, + 859.0, + 2116.0, + 836.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1001.0, + 696.0, + 1001.0, + 696.0, + 1040.0, + 294.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1069.0, + 1405.0, + 1069.0, + 1405.0, + 1104.0, + 293.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1102.0, + 1404.0, + 1102.0, + 1404.0, + 1135.0, + 294.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1133.0, + 1406.0, + 1133.0, + 1406.0, + 1166.0, + 295.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1161.0, + 1292.0, + 1161.0, + 1292.0, + 1197.0, + 294.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1882.0, + 1405.0, + 1882.0, + 1405.0, + 1915.0, + 295.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1912.0, + 1408.0, + 1912.0, + 1408.0, + 1945.0, + 295.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1941.0, + 1406.0, + 1941.0, + 1406.0, + 1978.0, + 294.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2009.0, + 295.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 1406.0, + 2002.0, + 1406.0, + 2039.0, + 295.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1606.0, + 1405.0, + 1606.0, + 1405.0, + 1644.0, + 292.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1637.0, + 295.0, + 1637.0, + 295.0, + 1675.0, + 292.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1637.0, + 1407.0, + 1637.0, + 1407.0, + 1675.0, + 370.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1670.0, + 1404.0, + 1670.0, + 1404.0, + 1706.0, + 293.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1699.0, + 1406.0, + 1699.0, + 1406.0, + 1736.0, + 292.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 781.0, + 1406.0, + 781.0, + 1406.0, + 818.0, + 293.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 812.0, + 1408.0, + 812.0, + 1408.0, + 849.0, + 293.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 846.0, + 1404.0, + 846.0, + 1404.0, + 878.0, + 378.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 874.0, + 1406.0, + 874.0, + 1406.0, + 909.0, + 293.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 903.0, + 1407.0, + 903.0, + 1407.0, + 941.0, + 293.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 934.0, + 1331.0, + 934.0, + 1331.0, + 971.0, + 293.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 282.0, + 1369.0, + 282.0, + 1369.0, + 324.0, + 293.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1823.0, + 1219.0, + 1823.0, + 1219.0, + 1861.0, + 294.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1001.0, + 696.0, + 1001.0, + 696.0, + 1040.0, + 294.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 435.0, + 1315.0, + 435.0, + 1315.0, + 472.0, + 1271.0, + 472.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1279, + 1405, + 1279, + 1405, + 1524, + 298, + 1524 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 695, + 1403, + 695, + 1403, + 909, + 298, + 909 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 924, + 1404, + 924, + 1404, + 1139, + 297, + 1139 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 229, + 1403, + 229, + 1403, + 353, + 299, + 353 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 1647, + 1405, + 1647, + 1405, + 1833, + 299, + 1833 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 299, + 1540, + 1406, + 1540, + 1406, + 1632, + 299, + 1632 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 299, + 578, + 1403, + 578, + 1403, + 670, + 299, + 670 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 299, + 471, + 1405, + 471, + 1405, + 563, + 299, + 563 + ], + "score": 0.969 + }, + { + "category_id": 2, + "poly": [ + 328, + 1884, + 1388, + 1884, + 1388, + 2035, + 328, + 2035 + ], + "score": 0.959 + }, + { + "category_id": 0, + "poly": [ + 300, + 1199, + 543, + 1199, + 543, + 1235, + 300, + 1235 + ], + "score": 0.895 + }, + { + "category_id": 0, + "poly": [ + 299, + 406, + 515, + 406, + 515, + 438, + 299, + 438 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 104, + 300, + 104 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2111, + 840, + 2111 + ], + "score": 0.801 + }, + { + "category_id": 13, + "poly": [ + 793, + 472, + 876, + 472, + 876, + 501, + 793, + 501 + ], + "score": 0.26, + "latex": "G A T \\nu 2" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1876.0, + 1189.0, + 1876.0, + 1189.0, + 1920.0, + 325.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1908.0, + 1391.0, + 1908.0, + 1391.0, + 1949.0, + 328.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1933.0, + 1189.0, + 1933.0, + 1189.0, + 1986.0, + 324.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1993.0, + 1384.0, + 1993.0, + 1384.0, + 2046.0, + 317.0, + 2046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.25, + 1969.0, + 889.25, + 1969.0, + 889.25, + 2010.5, + 324.25, + 2010.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1196.0, + 549.0, + 1196.0, + 549.0, + 1244.0, + 291.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 402.0, + 520.0, + 402.0, + 520.0, + 444.0, + 293.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1281.0, + 1405.0, + 1281.0, + 1405.0, + 1314.0, + 295.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1310.0, + 1406.0, + 1310.0, + 1406.0, + 1344.0, + 295.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1339.0, + 1410.0, + 1339.0, + 1410.0, + 1379.0, + 292.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1370.0, + 1407.0, + 1370.0, + 1407.0, + 1406.0, + 293.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1404.0, + 1405.0, + 1404.0, + 1405.0, + 1437.0, + 295.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1428.0, + 1407.0, + 1428.0, + 1407.0, + 1472.0, + 291.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1464.0, + 1405.0, + 1464.0, + 1405.0, + 1497.0, + 295.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1494.0, + 725.0, + 1494.0, + 725.0, + 1527.0, + 296.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 695.0, + 1405.0, + 695.0, + 1405.0, + 730.0, + 294.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 725.0, + 1407.0, + 725.0, + 1407.0, + 762.0, + 293.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 757.0, + 1407.0, + 757.0, + 1407.0, + 791.0, + 294.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 787.0, + 1405.0, + 787.0, + 1405.0, + 822.0, + 293.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 818.0, + 1405.0, + 818.0, + 1405.0, + 852.0, + 294.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 847.0, + 1405.0, + 847.0, + 1405.0, + 882.0, + 294.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 879.0, + 1317.0, + 879.0, + 1317.0, + 914.0, + 294.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 924.0, + 1405.0, + 924.0, + 1405.0, + 959.0, + 295.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 954.0, + 1406.0, + 954.0, + 1406.0, + 990.0, + 291.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 985.0, + 1405.0, + 985.0, + 1405.0, + 1020.0, + 294.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1018.0, + 1404.0, + 1018.0, + 1404.0, + 1051.0, + 294.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1047.0, + 1405.0, + 1047.0, + 1405.0, + 1082.0, + 295.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1079.0, + 1405.0, + 1079.0, + 1405.0, + 1110.0, + 296.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1110.0, + 940.0, + 1110.0, + 940.0, + 1141.0, + 296.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 264.0, + 293.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 262.0, + 1405.0, + 262.0, + 1405.0, + 294.0, + 295.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 292.0, + 1405.0, + 292.0, + 1405.0, + 325.0, + 293.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 320.0, + 1390.0, + 320.0, + 1390.0, + 356.0, + 294.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1647.0, + 1406.0, + 1647.0, + 1406.0, + 1684.0, + 294.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1674.0, + 1406.0, + 1674.0, + 1406.0, + 1717.0, + 292.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1707.0, + 1403.0, + 1707.0, + 1403.0, + 1743.0, + 294.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1737.0, + 1409.0, + 1737.0, + 1409.0, + 1776.0, + 293.0, + 1776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1768.0, + 1408.0, + 1768.0, + 1408.0, + 1806.0, + 292.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1796.0, + 1380.0, + 1796.0, + 1380.0, + 1837.0, + 290.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1537.0, + 1406.0, + 1537.0, + 1406.0, + 1578.0, + 293.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1572.0, + 1404.0, + 1572.0, + 1404.0, + 1605.0, + 294.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1599.0, + 1344.0, + 1599.0, + 1344.0, + 1639.0, + 293.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 577.0, + 1404.0, + 577.0, + 1404.0, + 612.0, + 293.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 610.0, + 1404.0, + 610.0, + 1404.0, + 643.0, + 294.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 640.0, + 795.0, + 640.0, + 795.0, + 672.0, + 295.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 471.0, + 792.0, + 471.0, + 792.0, + 505.0, + 294.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 471.0, + 1405.0, + 471.0, + 1405.0, + 505.0, + 877.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 503.0, + 1402.0, + 503.0, + 1402.0, + 533.0, + 297.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 529.0, + 1410.0, + 529.0, + 1410.0, + 571.0, + 293.0, + 571.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 757, + 1405, + 757, + 1405, + 1003, + 298, + 1003 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1026, + 1404, + 1026, + 1404, + 1181, + 299, + 1181 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 550, + 1403, + 550, + 1403, + 735, + 298, + 735 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1369, + 1404, + 1369, + 1404, + 1565, + 297, + 1565 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 297, + 1403, + 297, + 1403, + 421, + 299, + 421 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1682, + 1404, + 1682, + 1404, + 1775, + 297, + 1775 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 290, + 1575, + 1402, + 1575, + 1402, + 1640, + 290, + 1640 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 650, + 1776, + 1042, + 1776, + 1042, + 1851, + 650, + 1851 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 640, + 1936, + 1058, + 1936, + 1058, + 1978, + 640, + 1978 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 289, + 1884, + 1327, + 1884, + 1327, + 1921, + 289, + 1921 + ], + "score": 0.926 + }, + { + "category_id": 2, + "poly": [ + 326, + 2004, + 589, + 2004, + 589, + 2034, + 326, + 2034 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 298, + 1300, + 1121, + 1300, + 1121, + 1333, + 298, + 1333 + ], + "score": 0.913 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.895 + }, + { + "category_id": 0, + "poly": [ + 302, + 1228, + 698, + 1228, + 698, + 1264, + 302, + 1264 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1942, + 1400, + 1942, + 1400, + 1971, + 1366, + 1971 + ], + "score": 0.88 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1799, + 1400, + 1799, + 1400, + 1828, + 1366, + 1828 + ], + "score": 0.877 + }, + { + "category_id": 0, + "poly": [ + 302, + 469, + 594, + 469, + 594, + 505, + 302, + 505 + ], + "score": 0.876 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.872 + }, + { + "category_id": 0, + "poly": [ + 302, + 228, + 607, + 228, + 607, + 261, + 302, + 261 + ], + "score": 0.834 + }, + { + "category_id": 14, + "poly": [ + 650, + 1773, + 1044, + 1773, + 1044, + 1853, + 650, + 1853 + ], + "score": 0.93, + "latex": "g \\left( \\pmb { x } \\right) = \\left. { 1 } \\pmb { \\mathrm { \\rbrace } } \\ i : \\pmb { x } = \\left[ \\pmb { h } _ { i } \\right| \\left| \\pmb { h } _ { \\varphi ( i ) } \\right] \\right." + }, + { + "category_id": 13, + "poly": [ + 1015, + 1437, + 1132, + 1437, + 1132, + 1471, + 1015, + 1471 + ], + "score": 0.93, + "latex": "[ m ] [ n ]" + }, + { + "category_id": 13, + "poly": [ + 974, + 1715, + 1092, + 1715, + 1092, + 1749, + 974, + 1749 + ], + "score": 0.93, + "latex": "[ n ] [ n ]" + }, + { + "category_id": 13, + "poly": [ + 430, + 1684, + 577, + 1684, + 577, + 1717, + 430, + 1717 + ], + "score": 0.92, + "latex": "\\mathcal { G } ~ = ~ ( \\nu , \\mathcal { E } )" + }, + { + "category_id": 13, + "poly": [ + 299, + 1715, + 439, + 1715, + 439, + 1749, + 299, + 1749 + ], + "score": 0.92, + "latex": "\\{ h _ { 1 } , . . . , h _ { n } \\}" + }, + { + "category_id": 13, + "poly": [ + 542, + 1715, + 709, + 1715, + 709, + 1749, + 542, + 1749 + ], + "score": 0.92, + "latex": "\\varphi : [ n ] \\to [ n ]" + }, + { + "category_id": 13, + "poly": [ + 498, + 1473, + 583, + 1473, + 583, + 1505, + 498, + 1505 + ], + "score": 0.92, + "latex": "i \\in [ m ]" + }, + { + "category_id": 13, + "poly": [ + 710, + 1885, + 857, + 1885, + 857, + 1920, + 710, + 1920 + ], + "score": 0.92, + "latex": "\\widetilde { g } : { \\mathbb { R } } ^ { 2 d } \\to { \\mathbb { R } }" + }, + { + "category_id": 13, + "poly": [ + 1268, + 1439, + 1344, + 1439, + 1344, + 1471, + 1268, + 1471 + ], + "score": 0.92, + "latex": "f \\in { \\mathcal { F } }" + }, + { + "category_id": 13, + "poly": [ + 1237, + 1713, + 1401, + 1713, + 1401, + 1748, + 1237, + 1748 + ], + "score": 0.92, + "latex": "\\dot { g } : \\mathbb { R } ^ { 2 d } \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 500, + 1436, + 758, + 1436, + 758, + 1472, + 500, + 1472 + ], + "score": 0.91, + "latex": "\\mathbb { Q } = \\{ \\ b { q } _ { 1 } , . . . , \\ b { q } _ { m } \\} \\subset \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1606, + 502, + 1606, + 502, + 1641, + 298, + 1641 + ], + "score": 0.91, + "latex": "\\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}" + }, + { + "category_id": 13, + "poly": [ + 1146, + 1402, + 1400, + 1402, + 1400, + 1436, + 1146, + 1436 + ], + "score": 0.91, + "latex": "\\mathbb { K } = \\{ \\boldsymbol { k } _ { 1 } , . . . , \\boldsymbol { k } _ { n } \\} \\subset \\mathbb { R } ^ { d }" + }, + { + "category_id": 14, + "poly": [ + 639, + 1936, + 1059, + 1936, + 1059, + 1976, + 639, + 1976 + ], + "score": 0.9, + "latex": "\\widetilde { g } ( [ h _ { i } \\| h _ { j } ] ) = g ( [ h _ { i } \\| h _ { j } ] ) , \\forall i , j \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 1213, + 1886, + 1244, + 1886, + 1244, + 1915, + 1213, + 1915 + ], + "score": 0.87, + "latex": "n ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 305, + 1400, + 483, + 1400, + 483, + 1436, + 305, + 1436 + ], + "score": 0.87, + "latex": "\\left( \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d } \\to \\mathbb { R } \\right)" + }, + { + "category_id": 13, + "poly": [ + 998, + 1505, + 1023, + 1505, + 1023, + 1533, + 998, + 1533 + ], + "score": 0.85, + "latex": "\\mathbb { Q }" + }, + { + "category_id": 13, + "poly": [ + 1349, + 1372, + 1404, + 1372, + 1404, + 1402, + 1349, + 1402 + ], + "score": 0.84, + "latex": "\\mathcal { F } \\subseteq" + }, + { + "category_id": 13, + "poly": [ + 1347, + 1577, + 1405, + 1577, + 1405, + 1606, + 1347, + 1606 + ], + "score": 0.84, + "latex": "\\mathbb { K } =" + }, + { + "category_id": 13, + "poly": [ + 1015, + 1894, + 1033, + 1894, + 1033, + 1920, + 1015, + 1920 + ], + "score": 0.82, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 919, + 1505, + 945, + 1505, + 945, + 1530, + 919, + 1530 + ], + "score": 0.76, + "latex": "\\mathbb { K }" + }, + { + "category_id": 13, + "poly": [ + 722, + 1470, + 1167, + 1470, + 1167, + 1506, + 722, + 1506 + ], + "score": 0.73, + "latex": "j _ { \\neq \\varphi ( i ) } \\in [ n ] \\colon f \\left( q _ { i } , k _ { \\varphi ( i ) } \\right) > f \\left( q _ { i } , k _ { j } \\right) ." + }, + { + "category_id": 13, + "poly": [ + 725, + 1471, + 863, + 1471, + 863, + 1507, + 725, + 1507 + ], + "score": 0.63, + "latex": "j _ { \\neq \\varphi ( i ) } \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 981, + 1441, + 1002, + 1441, + 1002, + 1470, + 981, + 1470 + ], + "score": 0.59, + "latex": "\\varphi" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1996.0, + 594.0, + 1996.0, + 594.0, + 2043.0, + 321.0, + 2043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1226.0, + 702.0, + 1226.0, + 702.0, + 1269.0, + 294.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 467.0, + 600.0, + 467.0, + 600.0, + 510.0, + 294.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2123.0, + 830.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 228.0, + 611.0, + 228.0, + 611.0, + 267.0, + 298.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 759.0, + 1406.0, + 759.0, + 1406.0, + 792.0, + 296.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 790.0, + 1408.0, + 790.0, + 1408.0, + 824.0, + 295.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 820.0, + 1407.0, + 820.0, + 1407.0, + 854.0, + 295.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 845.0, + 1407.0, + 845.0, + 1407.0, + 891.0, + 291.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 881.0, + 1405.0, + 881.0, + 1405.0, + 915.0, + 295.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 911.0, + 1406.0, + 911.0, + 1406.0, + 948.0, + 293.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 941.0, + 1407.0, + 941.0, + 1407.0, + 979.0, + 293.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 971.0, + 1408.0, + 971.0, + 1408.0, + 1009.0, + 293.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1026.0, + 1406.0, + 1026.0, + 1406.0, + 1064.0, + 294.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1056.0, + 1404.0, + 1056.0, + 1404.0, + 1092.0, + 294.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1086.0, + 1405.0, + 1086.0, + 1405.0, + 1123.0, + 294.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1115.0, + 1404.0, + 1115.0, + 1404.0, + 1154.0, + 292.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1151.0, + 944.0, + 1151.0, + 944.0, + 1185.0, + 297.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 551.0, + 1404.0, + 551.0, + 1404.0, + 585.0, + 294.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 583.0, + 1405.0, + 583.0, + 1405.0, + 615.0, + 294.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 611.0, + 1407.0, + 611.0, + 1407.0, + 648.0, + 292.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 642.0, + 1405.0, + 642.0, + 1405.0, + 678.0, + 293.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 672.0, + 1405.0, + 672.0, + 1405.0, + 709.0, + 293.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 702.0, + 1197.0, + 702.0, + 1197.0, + 739.0, + 293.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1369.0, + 1348.0, + 1369.0, + 1348.0, + 1406.0, + 295.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1396.0, + 304.0, + 1396.0, + 304.0, + 1441.0, + 297.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1396.0, + 1145.0, + 1396.0, + 1145.0, + 1441.0, + 484.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1396.0, + 1406.0, + 1396.0, + 1406.0, + 1441.0, + 1401.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1435.0, + 499.0, + 1435.0, + 499.0, + 1475.0, + 293.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1435.0, + 980.0, + 1435.0, + 980.0, + 1475.0, + 759.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1435.0, + 1014.0, + 1435.0, + 1014.0, + 1475.0, + 1003.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1435.0, + 1267.0, + 1435.0, + 1267.0, + 1475.0, + 1133.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1435.0, + 1405.0, + 1435.0, + 1405.0, + 1475.0, + 1345.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1466.0, + 497.0, + 1466.0, + 497.0, + 1515.0, + 291.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1466.0, + 721.0, + 1466.0, + 721.0, + 1515.0, + 584.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 1466.0, + 1407.0, + 1466.0, + 1407.0, + 1515.0, + 1168.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1500.0, + 918.0, + 1500.0, + 918.0, + 1537.0, + 295.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 1500.0, + 997.0, + 1500.0, + 997.0, + 1537.0, + 946.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1500.0, + 1405.0, + 1500.0, + 1405.0, + 1537.0, + 1024.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1532.0, + 1210.0, + 1532.0, + 1210.0, + 1569.0, + 296.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 296.0, + 1404.0, + 296.0, + 1404.0, + 332.0, + 294.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 327.0, + 1404.0, + 327.0, + 1404.0, + 365.0, + 292.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 358.0, + 1407.0, + 358.0, + 1407.0, + 396.0, + 293.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 386.0, + 836.0, + 386.0, + 836.0, + 428.0, + 293.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1681.0, + 429.0, + 1681.0, + 429.0, + 1719.0, + 295.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1681.0, + 1404.0, + 1681.0, + 1404.0, + 1719.0, + 578.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1712.0, + 298.0, + 1712.0, + 298.0, + 1753.0, + 292.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1712.0, + 541.0, + 1712.0, + 541.0, + 1753.0, + 440.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1712.0, + 973.0, + 1712.0, + 973.0, + 1753.0, + 710.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1712.0, + 1236.0, + 1712.0, + 1236.0, + 1753.0, + 1093.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1712.0, + 1409.0, + 1712.0, + 1409.0, + 1753.0, + 1402.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1744.0, + 427.0, + 1744.0, + 427.0, + 1779.0, + 294.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1568.0, + 1346.0, + 1568.0, + 1346.0, + 1615.0, + 292.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1601.0, + 297.0, + 1601.0, + 297.0, + 1646.0, + 293.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1601.0, + 514.0, + 1601.0, + 514.0, + 1646.0, + 503.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1880.0, + 709.0, + 1880.0, + 709.0, + 1925.0, + 290.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 1880.0, + 1014.0, + 1880.0, + 1014.0, + 1925.0, + 858.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1880.0, + 1212.0, + 1880.0, + 1212.0, + 1925.0, + 1034.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 1880.0, + 1332.0, + 1880.0, + 1332.0, + 1925.0, + 1245.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1299.0, + 1123.0, + 1299.0, + 1123.0, + 1338.0, + 295.0, + 1338.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1161, + 1405, + 1161, + 1405, + 1292, + 298, + 1292 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1600, + 1404, + 1600, + 1404, + 1756, + 297, + 1756 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 889, + 1405, + 889, + 1405, + 1044, + 297, + 1044 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 433, + 1404, + 433, + 1404, + 556, + 297, + 556 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 300, + 1410, + 1402, + 1410, + 1402, + 1503, + 300, + 1503 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 291, + 228, + 1402, + 228, + 1402, + 294, + 291, + 294 + ], + "score": 0.956 + }, + { + "category_id": 2, + "poly": [ + 296, + 1782, + 1404, + 1782, + 1404, + 2035, + 296, + 2035 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 297, + 571, + 1399, + 571, + 1399, + 639, + 297, + 639 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 631, + 360, + 1071, + 360, + 1071, + 406, + 631, + 406 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 556, + 657, + 1143, + 657, + 1143, + 699, + 556, + 699 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 297, + 712, + 874, + 712, + 874, + 746, + 297, + 746 + ], + "score": 0.93 + }, + { + "category_id": 8, + "poly": [ + 769, + 769, + 929, + 769, + 929, + 803, + 769, + 803 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 301, + 307, + 784, + 307, + 784, + 341, + 301, + 341 + ], + "score": 0.925 + }, + { + "category_id": 0, + "poly": [ + 298, + 1093, + 1024, + 1093, + 1024, + 1127, + 298, + 1127 + ], + "score": 0.912 + }, + { + "category_id": 9, + "poly": [ + 1352, + 365, + 1399, + 365, + 1399, + 396, + 1352, + 396 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 816, + 74, + 816, + 105, + 299, + 105 + ], + "score": 0.91 + }, + { + "category_id": 9, + "poly": [ + 1352, + 766, + 1399, + 766, + 1399, + 797, + 1352, + 797 + ], + "score": 0.905 + }, + { + "category_id": 0, + "poly": [ + 300, + 1340, + 633, + 1340, + 633, + 1376, + 300, + 1376 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1352, + 662, + 1400, + 662, + 1400, + 692, + 1352, + 692 + ], + "score": 0.892 + }, + { + "category_id": 0, + "poly": [ + 301, + 1544, + 740, + 1544, + 740, + 1575, + 301, + 1575 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 861, + 2088, + 861, + 2113, + 835, + 2113 + ], + "score": 0.847 + }, + { + "category_id": 2, + "poly": [ + 1375, + 819, + 1401, + 819, + 1401, + 846, + 1375, + 846 + ], + "score": 0.797 + }, + { + "category_id": 14, + "poly": [ + 626, + 359, + 1071, + 359, + 1071, + 404, + 626, + 404 + ], + "score": 0.93, + "latex": "1 = \\widetilde { g } \\left( \\left[ h _ { i } \\Vert h _ { \\varphi ( i ) } \\right] \\right) > \\widetilde { g } \\left( \\left[ h _ { i } \\Vert h _ { j } \\right] \\right) = 0" + }, + { + "category_id": 13, + "poly": [ + 985, + 1223, + 1151, + 1223, + 1151, + 1269, + 985, + 1269 + ], + "score": 0.93, + "latex": "\\{ s _ { j } ^ { ( h ) } \\mid j \\in \\mathcal { V } \\}" + }, + { + "category_id": 13, + "poly": [ + 650, + 309, + 775, + 309, + 775, + 345, + 650, + 345 + ], + "score": 0.92, + "latex": "j _ { \\neq \\varphi ( i ) } \\in \\mathcal { V }" + }, + { + "category_id": 13, + "poly": [ + 796, + 1633, + 917, + 1633, + 917, + 1666, + 796, + 1666 + ], + "score": 0.92, + "latex": "\\in \\{ 2 , 3 , 6 \\}" + }, + { + "category_id": 13, + "poly": [ + 610, + 606, + 680, + 606, + 680, + 640, + 610, + 640 + ], + "score": 0.92, + "latex": "\\mathscr { j } _ { \\neq \\varphi ( i ) }" + }, + { + "category_id": 13, + "poly": [ + 724, + 465, + 812, + 465, + 812, + 499, + 724, + 499 + ], + "score": 0.92, + "latex": "[ h _ { i } | | h _ { j } ]" + }, + { + "category_id": 13, + "poly": [ + 735, + 573, + 872, + 573, + 872, + 606, + 735, + 606 + ], + "score": 0.91, + "latex": "0 < \\epsilon < 1 / 2" + }, + { + "category_id": 13, + "poly": [ + 778, + 1926, + 996, + 1926, + 996, + 1953, + 778, + 1953 + ], + "score": 0.91, + "latex": "\\{ [ h _ { i } | | h _ { j } ] \\mid i , j \\in [ n ] \\}" + }, + { + "category_id": 13, + "poly": [ + 358, + 1230, + 451, + 1230, + 451, + 1264, + 358, + 1264 + ], + "score": 0.91, + "latex": "h \\in [ H ]" + }, + { + "category_id": 14, + "poly": [ + 770, + 767, + 930, + 767, + 930, + 803, + 770, + 803 + ], + "score": 0.91, + "latex": "\\alpha _ { i , \\varphi ( i ) } > \\alpha _ { i , j }" + }, + { + "category_id": 14, + "poly": [ + 556, + 654, + 1143, + 654, + 1143, + 699, + 556, + 699 + ], + "score": 0.9, + "latex": "{ e _ { W , a } } \\left( { h _ { i } } , { h _ { \\varphi ( i ) } } \\right) > 1 - \\epsilon > 0 + \\epsilon > { e _ { W , a } } \\left( { h _ { i } } , { h _ { j } } \\right)" + }, + { + "category_id": 13, + "poly": [ + 425, + 605, + 492, + 605, + 492, + 632, + 425, + 632 + ], + "score": 0.9, + "latex": "i \\in \\nu" + }, + { + "category_id": 13, + "poly": [ + 533, + 309, + 599, + 309, + 599, + 337, + 533, + 337 + ], + "score": 0.9, + "latex": "i \\in \\nu" + }, + { + "category_id": 13, + "poly": [ + 404, + 525, + 449, + 525, + 449, + 553, + 404, + 553 + ], + "score": 0.89, + "latex": "\\mathbb { R } ^ { 2 d }" + }, + { + "category_id": 13, + "poly": [ + 523, + 228, + 622, + 228, + 622, + 260, + 523, + 260 + ], + "score": 0.89, + "latex": "\\boldsymbol { x } \\in \\mathbb { R } ^ { 2 d }" + }, + { + "category_id": 13, + "poly": [ + 962, + 1991, + 979, + 1991, + 979, + 2007, + 962, + 2007 + ], + "score": 0.88, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 675, + 1955, + 688, + 1955, + 688, + 1979, + 675, + 1979 + ], + "score": 0.88, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 678, + 1925, + 706, + 1925, + 706, + 1946, + 678, + 1946 + ], + "score": 0.87, + "latex": "n ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1059, + 1633, + 1251, + 1633, + 1251, + 1665, + 1059, + 1665 + ], + "score": 0.87, + "latex": "\\in \\{ 6 4 , 1 2 8 , 2 5 6 \\}" + }, + { + "category_id": 13, + "poly": [ + 293, + 1663, + 643, + 1663, + 643, + 1697, + 293, + 1697 + ], + "score": 0.86, + "latex": "\\in \\{ 0 . 0 0 0 5 , 0 . 0 0 1 , 0 . 0 0 5 , 0 . 0 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 341, + 1925, + 357, + 1925, + 357, + 1951, + 341, + 1951 + ], + "score": 0.86, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 437, + 1979, + 453, + 1979, + 453, + 2007, + 437, + 2007 + ], + "score": 0.86, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 1103, + 1955, + 1118, + 1955, + 1118, + 1979, + 1103, + 1979 + ], + "score": 0.85, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 738, + 262, + 757, + 262, + 757, + 294, + 738, + 294 + ], + "score": 0.85, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 1197, + 496, + 1215, + 496, + 1215, + 528, + 1197, + 528 + ], + "score": 0.85, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 938, + 1868, + 955, + 1868, + 955, + 1896, + 938, + 1896 + ], + "score": 0.84, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 488, + 1263, + 516, + 1263, + 516, + 1289, + 488, + 1289 + ], + "score": 0.84, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 1298, + 231, + 1317, + 231, + 1317, + 263, + 1298, + 263 + ], + "score": 0.83, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 550, + 1927, + 567, + 1927, + 567, + 1951, + 550, + 1951 + ], + "score": 0.83, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 426, + 1843, + 447, + 1843, + 447, + 1868, + 426, + 1868 + ], + "score": 0.82, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 1201, + 1843, + 1218, + 1843, + 1218, + 1868, + 1201, + 1868 + ], + "score": 0.82, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 817, + 1907, + 831, + 1907, + 831, + 1924, + 817, + 1924 + ], + "score": 0.82, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 583, + 1897, + 598, + 1897, + 598, + 1924, + 583, + 1924 + ], + "score": 0.8, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 634, + 229, + 689, + 229, + 689, + 264, + 634, + 264 + ], + "score": 0.8, + "latex": "\\widetilde g ( x )" + }, + { + "category_id": 13, + "poly": [ + 334, + 1896, + 349, + 1896, + 349, + 1924, + 334, + 1924 + ], + "score": 0.8, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 1142, + 574, + 1178, + 574, + 1178, + 600, + 1142, + 600 + ], + "score": 0.8, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 482, + 1784, + 500, + 1784, + 500, + 1812, + 482, + 1812 + ], + "score": 0.8, + "latex": "\\widetilde g" + }, + { + "category_id": 13, + "poly": [ + 905, + 1164, + 934, + 1164, + 934, + 1190, + 905, + 1190 + ], + "score": 0.79, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 1231, + 578, + 1252, + 578, + 1252, + 600, + 1231, + 600 + ], + "score": 0.77, + "latex": "\\textbf { \\em a }" + }, + { + "category_id": 13, + "poly": [ + 640, + 1873, + 655, + 1873, + 655, + 1896, + 640, + 1896 + ], + "score": 0.77, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 1029, + 1845, + 1043, + 1845, + 1043, + 1863, + 1029, + 1863 + ], + "score": 0.73, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 658, + 579, + 674, + 579, + 674, + 599, + 658, + 599 + ], + "score": 0.72, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 1079, + 441, + 1095, + 441, + 1095, + 462, + 1079, + 462 + ], + "score": 0.71, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 986, + 501, + 1003, + 501, + 1003, + 523, + 986, + 523 + ], + "score": 0.67, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 422, + 1900, + 438, + 1900, + 438, + 1920, + 422, + 1920 + ], + "score": 0.62, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 1230, + 1816, + 1246, + 1816, + 1246, + 1837, + 1230, + 1837 + ], + "score": 0.59, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 609, + 1902, + 623, + 1902, + 623, + 1920, + 609, + 1920 + ], + "score": 0.43, + "latex": "e" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1776.0, + 481.0, + 1776.0, + 481.0, + 1820.0, + 320.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1776.0, + 1408.0, + 1776.0, + 1408.0, + 1820.0, + 501.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1809.0, + 1229.0, + 1809.0, + 1229.0, + 1845.0, + 293.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 1809.0, + 1406.0, + 1809.0, + 1406.0, + 1845.0, + 1247.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1837.0, + 425.0, + 1837.0, + 425.0, + 1873.0, + 293.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1837.0, + 1028.0, + 1837.0, + 1028.0, + 1873.0, + 448.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1837.0, + 1200.0, + 1837.0, + 1200.0, + 1873.0, + 1044.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 1837.0, + 1406.0, + 1837.0, + 1406.0, + 1873.0, + 1219.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1866.0, + 639.0, + 1866.0, + 639.0, + 1900.0, + 293.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1866.0, + 937.0, + 1866.0, + 937.0, + 1900.0, + 656.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1866.0, + 1406.0, + 1866.0, + 1406.0, + 1900.0, + 956.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1894.0, + 333.0, + 1894.0, + 333.0, + 1928.0, + 294.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1894.0, + 421.0, + 1894.0, + 421.0, + 1928.0, + 350.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1894.0, + 582.0, + 1894.0, + 582.0, + 1928.0, + 439.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 1894.0, + 608.0, + 1894.0, + 608.0, + 1928.0, + 599.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 1894.0, + 816.0, + 1894.0, + 816.0, + 1928.0, + 624.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1894.0, + 1406.0, + 1894.0, + 1406.0, + 1928.0, + 832.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1921.0, + 340.0, + 1921.0, + 340.0, + 1955.0, + 294.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1921.0, + 549.0, + 1921.0, + 549.0, + 1955.0, + 358.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1921.0, + 677.0, + 1921.0, + 677.0, + 1955.0, + 568.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1921.0, + 777.0, + 1921.0, + 777.0, + 1955.0, + 707.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 1921.0, + 1407.0, + 1921.0, + 1407.0, + 1955.0, + 997.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1948.0, + 674.0, + 1948.0, + 674.0, + 1983.0, + 293.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 1948.0, + 1102.0, + 1948.0, + 1102.0, + 1983.0, + 689.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1948.0, + 1406.0, + 1948.0, + 1406.0, + 1983.0, + 1119.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1976.0, + 436.0, + 1976.0, + 436.0, + 2014.0, + 293.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 1976.0, + 961.0, + 1976.0, + 961.0, + 2014.0, + 454.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1976.0, + 1406.0, + 1976.0, + 1406.0, + 2014.0, + 980.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 437.0, + 2003.0, + 437.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1086.0, + 1028.0, + 1086.0, + 1028.0, + 1134.0, + 293.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1336.0, + 637.0, + 1336.0, + 637.0, + 1383.0, + 294.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1542.0, + 744.0, + 1542.0, + 744.0, + 1579.0, + 295.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 824.0, + 1401.0, + 824.0, + 1401.0, + 845.0, + 1379.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1162.0, + 904.0, + 1162.0, + 904.0, + 1195.0, + 296.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 1162.0, + 1405.0, + 1162.0, + 1405.0, + 1195.0, + 935.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1189.0, + 1405.0, + 1189.0, + 1405.0, + 1230.0, + 292.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1217.0, + 357.0, + 1217.0, + 357.0, + 1277.0, + 293.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1217.0, + 984.0, + 1217.0, + 984.0, + 1277.0, + 452.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 1217.0, + 1405.0, + 1217.0, + 1405.0, + 1277.0, + 1152.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1263.0, + 487.0, + 1263.0, + 487.0, + 1293.0, + 294.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1263.0, + 765.0, + 1263.0, + 765.0, + 1293.0, + 517.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1599.0, + 1406.0, + 1599.0, + 1406.0, + 1638.0, + 294.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1631.0, + 795.0, + 1631.0, + 795.0, + 1669.0, + 295.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1631.0, + 1058.0, + 1631.0, + 1058.0, + 1669.0, + 918.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 1631.0, + 1405.0, + 1631.0, + 1405.0, + 1669.0, + 1252.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 1661.0, + 1405.0, + 1661.0, + 1405.0, + 1698.0, + 644.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1691.0, + 1406.0, + 1691.0, + 1406.0, + 1730.0, + 294.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1723.0, + 1141.0, + 1723.0, + 1141.0, + 1761.0, + 292.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 888.0, + 1408.0, + 888.0, + 1408.0, + 927.0, + 295.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 923.0, + 1403.0, + 923.0, + 1403.0, + 956.0, + 296.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 951.0, + 1405.0, + 951.0, + 1405.0, + 988.0, + 294.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 978.0, + 1405.0, + 978.0, + 1405.0, + 1020.0, + 292.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1012.0, + 505.0, + 1012.0, + 505.0, + 1045.0, + 295.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 432.0, + 1078.0, + 432.0, + 1078.0, + 469.0, + 291.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 432.0, + 1406.0, + 432.0, + 1406.0, + 469.0, + 1096.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 463.0, + 723.0, + 463.0, + 723.0, + 498.0, + 292.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 463.0, + 1407.0, + 463.0, + 1407.0, + 498.0, + 813.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 494.0, + 985.0, + 494.0, + 985.0, + 531.0, + 292.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 494.0, + 1196.0, + 494.0, + 1196.0, + 531.0, + 1004.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 494.0, + 1405.0, + 494.0, + 1405.0, + 531.0, + 1216.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 520.5, + 462.0, + 520.5, + 462.0, + 558.0, + 290.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1411.0, + 1404.0, + 1411.0, + 1404.0, + 1444.0, + 294.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1441.0, + 1404.0, + 1441.0, + 1404.0, + 1475.0, + 294.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1471.0, + 447.0, + 1471.0, + 447.0, + 1505.0, + 294.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 226.0, + 522.0, + 226.0, + 522.0, + 266.0, + 294.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 226.0, + 633.0, + 226.0, + 633.0, + 266.0, + 623.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 226.0, + 1297.0, + 226.0, + 1297.0, + 266.0, + 690.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 226.0, + 1407.0, + 226.0, + 1407.0, + 266.0, + 1318.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 257.0, + 737.0, + 257.0, + 737.0, + 297.0, + 289.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 257.0, + 1131.0, + 257.0, + 1131.0, + 297.0, + 758.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 568.0, + 657.0, + 568.0, + 657.0, + 608.0, + 293.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 568.0, + 734.0, + 568.0, + 734.0, + 608.0, + 675.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 568.0, + 1141.0, + 568.0, + 1141.0, + 608.0, + 873.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 568.0, + 1230.0, + 568.0, + 1230.0, + 608.0, + 1179.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 568.0, + 1405.0, + 568.0, + 1405.0, + 608.0, + 1253.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 597.0, + 424.0, + 597.0, + 424.0, + 646.0, + 293.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 597.0, + 609.0, + 597.0, + 609.0, + 646.0, + 493.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 597.0, + 694.0, + 597.0, + 694.0, + 646.0, + 681.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 708.0, + 877.0, + 708.0, + 877.0, + 752.0, + 292.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 303.0, + 532.0, + 303.0, + 532.0, + 347.0, + 295.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 303.0, + 649.0, + 303.0, + 649.0, + 347.0, + 600.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 303.0, + 788.0, + 303.0, + 788.0, + 347.0, + 776.0, + 347.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 424, + 223, + 1274, + 223, + 1274, + 470, + 424, + 470 + ], + "score": 0.982, + "html": "
Dataset# layersHidden sizeLearning rateSampling method
ogbn-arxiv32560.01GraphSAINT
ogbn-products31280.001NeighborSampling
ogbn-mag22560.01NeighborSampling
ogbn-proteins6640.01NeighborSampling
ogbl-collab3640.001Full Batch
ogbl-citation232560.0005NeighborSampling
" + }, + { + "category_id": 1, + "poly": [ + 299, + 1510, + 1404, + 1510, + 1404, + 1665, + 299, + 1665 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 882, + 1403, + 882, + 1403, + 1035, + 298, + 1035 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 1051, + 1401, + 1051, + 1401, + 1144, + 299, + 1144 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 301, + 1243, + 1402, + 1243, + 1402, + 1334, + 301, + 1334 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 295, + 644, + 1401, + 644, + 1401, + 707, + 295, + 707 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 300, + 1762, + 1400, + 1762, + 1400, + 1825, + 300, + 1825 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 299, + 1350, + 1401, + 1350, + 1401, + 1413, + 299, + 1413 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 294, + 804, + 1400, + 804, + 1400, + 867, + 294, + 867 + ], + "score": 0.941 + }, + { + "category_id": 0, + "poly": [ + 298, + 585, + 654, + 585, + 654, + 618, + 298, + 618 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 291, + 2001, + 1298, + 2001, + 1298, + 2034, + 291, + 2034 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 299, + 1183, + 672, + 1183, + 672, + 1215, + 299, + 1215 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 815, + 75, + 815, + 105, + 298, + 105 + ], + "score": 0.893 + }, + { + "category_id": 0, + "poly": [ + 301, + 1704, + 701, + 1704, + 701, + 1736, + 301, + 1736 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 300, + 1873, + 605, + 1873, + 605, + 1909, + 300, + 1909 + ], + "score": 0.889 + }, + { + "category_id": 0, + "poly": [ + 298, + 1944, + 869, + 1944, + 869, + 1976, + 298, + 1976 + ], + "score": 0.876 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2112, + 836, + 2112 + ], + "score": 0.824 + }, + { + "category_id": 6, + "poly": [ + 500, + 492, + 1197, + 492, + 1197, + 525, + 500, + 525 + ], + "score": 0.796 + }, + { + "category_id": 0, + "poly": [ + 301, + 1452, + 699, + 1452, + 699, + 1484, + 301, + 1484 + ], + "score": 0.751 + }, + { + "category_id": 0, + "poly": [ + 297, + 746, + 949, + 746, + 949, + 777, + 297, + 777 + ], + "score": 0.634 + }, + { + "category_id": 1, + "poly": [ + 297, + 746, + 949, + 746, + 949, + 777, + 297, + 777 + ], + "score": 0.344 + }, + { + "category_id": 1, + "poly": [ + 301, + 1452, + 699, + 1452, + 699, + 1484, + 301, + 1484 + ], + "score": 0.15 + }, + { + "category_id": 13, + "poly": [ + 1126, + 806, + 1223, + 806, + 1223, + 834, + 1126, + 834 + ], + "score": 0.89, + "latex": "d = 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 327, + 976, + 345, + 976, + 345, + 1002, + 327, + 1002 + ], + "score": 0.79, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 870, + 885, + 905, + 885, + 905, + 914, + 870, + 914 + ], + "score": 0.63, + "latex": "( k )" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 584.0, + 656.0, + 584.0, + 656.0, + 620.0, + 295.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1180.0, + 676.0, + 1180.0, + 676.0, + 1219.0, + 294.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 70.0, + 818.0, + 70.0, + 818.0, + 110.0, + 294.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1701.0, + 705.0, + 1701.0, + 705.0, + 1740.0, + 294.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1868.0, + 608.0, + 1868.0, + 608.0, + 1914.0, + 293.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1941.0, + 871.0, + 1941.0, + 871.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 488.0, + 1198.0, + 488.0, + 1198.0, + 529.0, + 499.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1447.0, + 702.0, + 1447.0, + 702.0, + 1492.0, + 293.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 745.0, + 951.0, + 745.0, + 951.0, + 781.0, + 295.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1510.0, + 1408.0, + 1510.0, + 1408.0, + 1548.0, + 295.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1541.0, + 1406.0, + 1541.0, + 1406.0, + 1578.0, + 294.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1570.0, + 1405.0, + 1570.0, + 1405.0, + 1609.0, + 293.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1605.0, + 1403.0, + 1605.0, + 1403.0, + 1638.0, + 295.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1633.0, + 1035.0, + 1633.0, + 1035.0, + 1668.0, + 293.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 882.0, + 869.0, + 882.0, + 869.0, + 919.0, + 296.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 882.0, + 1405.0, + 882.0, + 1405.0, + 919.0, + 906.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 914.0, + 1404.0, + 914.0, + 1404.0, + 947.0, + 294.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 943.0, + 1405.0, + 943.0, + 1405.0, + 977.0, + 292.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 972.0, + 326.0, + 972.0, + 326.0, + 1011.0, + 293.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 972.0, + 1404.0, + 972.0, + 1404.0, + 1011.0, + 346.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1003.0, + 785.0, + 1003.0, + 785.0, + 1039.0, + 294.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1052.0, + 1405.0, + 1052.0, + 1405.0, + 1085.0, + 295.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1083.0, + 1405.0, + 1083.0, + 1405.0, + 1117.0, + 295.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1113.0, + 1287.0, + 1113.0, + 1287.0, + 1147.0, + 295.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1243.0, + 1404.0, + 1243.0, + 1404.0, + 1277.0, + 296.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1272.0, + 1406.0, + 1272.0, + 1406.0, + 1309.0, + 294.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1304.0, + 744.0, + 1304.0, + 744.0, + 1336.0, + 296.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 644.0, + 1406.0, + 644.0, + 1406.0, + 679.0, + 295.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 676.0, + 1327.0, + 676.0, + 1327.0, + 711.0, + 295.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1759.0, + 1404.0, + 1759.0, + 1404.0, + 1801.0, + 294.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1795.0, + 1016.0, + 1795.0, + 1016.0, + 1827.0, + 296.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1350.0, + 1405.0, + 1350.0, + 1405.0, + 1385.0, + 295.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1378.0, + 696.0, + 1378.0, + 696.0, + 1419.0, + 294.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 802.0, + 1125.0, + 802.0, + 1125.0, + 842.0, + 293.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 802.0, + 1407.0, + 802.0, + 1407.0, + 842.0, + 1224.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 834.0, + 644.0, + 834.0, + 644.0, + 869.0, + 296.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2001.0, + 1301.0, + 2001.0, + 1301.0, + 2036.0, + 295.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 745.0, + 951.0, + 745.0, + 951.0, + 781.0, + 295.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1447.0, + 702.0, + 1447.0, + 702.0, + 1492.0, + 293.0, + 1492.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 410, + 224, + 1289, + 224, + 1289, + 489, + 410, + 489 + ], + "score": 0.981, + "html": "
Dataset# nodes# edgesAvg. node degreeDiameter
ogbn-arxiv169,3431,166,24313.723
ogbn-mag1,939,74321,111,00721.76
ogbn-products2,449,02961,859,14050.527
ogbn-proteins132,53439,561,252597.09
ogbl-collab235,8681,285,4658.222
ogbl-citation22,927,96330,561,18720.721
" + }, + { + "category_id": 5, + "poly": [ + 532, + 738, + 1167, + 738, + 1167, + 929, + 532, + 929 + ], + "score": 0.977, + "html": "
TrainingValidationTest
# examples110,46210,00010,000
# nodes -average18.0318.0618.09
# edges - average18.6518.6718.72
Diameter - average6.356.356.35
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1771, + 1407, + 1771, + 1407, + 1896, + 297, + 1896 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 297, + 2001, + 849, + 2001, + 849, + 2035, + 297, + 2035 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 299, + 1104, + 518, + 1104, + 518, + 1136, + 299, + 1136 + ], + "score": 0.907 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.9 + }, + { + "category_id": 0, + "poly": [ + 299, + 1710, + 622, + 1710, + 622, + 1743, + 299, + 1743 + ], + "score": 0.9 + }, + { + "category_id": 0, + "poly": [ + 298, + 610, + 435, + 610, + 435, + 643, + 298, + 643 + ], + "score": 0.896 + }, + { + "category_id": 0, + "poly": [ + 299, + 1941, + 588, + 1941, + 588, + 1973, + 299, + 1973 + ], + "score": 0.884 + }, + { + "category_id": 0, + "poly": [ + 298, + 1636, + 677, + 1636, + 677, + 1672, + 298, + 1672 + ], + "score": 0.878 + }, + { + "category_id": 6, + "poly": [ + 532, + 510, + 1164, + 510, + 1164, + 545, + 532, + 545 + ], + "score": 0.856 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2113, + 836, + 2113 + ], + "score": 0.844 + }, + { + "category_id": 6, + "poly": [ + 301, + 1164, + 1407, + 1164, + 1407, + 1230, + 301, + 1230 + ], + "score": 0.833 + }, + { + "category_id": 7, + "poly": [ + 294, + 951, + 1401, + 951, + 1401, + 1014, + 294, + 1014 + ], + "score": 0.683 + }, + { + "category_id": 7, + "poly": [ + 297, + 1471, + 1402, + 1471, + 1402, + 1535, + 297, + 1535 + ], + "score": 0.429 + }, + { + "category_id": 6, + "poly": [ + 302, + 670, + 1229, + 670, + 1229, + 706, + 302, + 706 + ], + "score": 0.332 + }, + { + "category_id": 1, + "poly": [ + 302, + 670, + 1229, + 670, + 1229, + 706, + 302, + 706 + ], + "score": 0.135 + }, + { + "category_id": 6, + "poly": [ + 294, + 951, + 1401, + 951, + 1401, + 1014, + 294, + 1014 + ], + "score": 0.122 + }, + { + "category_id": 13, + "poly": [ + 860, + 1864, + 957, + 1864, + 957, + 1892, + 860, + 1892 + ], + "score": 0.89, + "latex": "d = 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 1271, + 1804, + 1354, + 1804, + 1354, + 1833, + 1271, + 1833 + ], + "score": 0.88, + "latex": "k { = } 1 0 0" + }, + { + "category_id": 13, + "poly": [ + 1119, + 1864, + 1187, + 1864, + 1187, + 1892, + 1119, + 1892 + ], + "score": 0.87, + "latex": "k { = } 2 0" + }, + { + "category_id": 13, + "poly": [ + 845, + 1804, + 914, + 1804, + 914, + 1833, + 845, + 1833 + ], + "score": 0.87, + "latex": "100 \\%" + }, + { + "category_id": 5, + "poly": [ + 353, + 1257, + 1347, + 1257, + 1347, + 1450, + 353, + 1450 + ], + "score": 0.738, + "html": "
TrainingValidationUnseenProject TestSeenProject Test
# graphs2543604265411703659974
# nodes - average2377174219593986
# edges - average72987851588212925
Diameter-average7.887.887.787.82
" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1099.0, + 521.0, + 1099.0, + 521.0, + 1141.0, + 292.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1708.0, + 624.0, + 1708.0, + 624.0, + 1746.0, + 293.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 606.0, + 440.0, + 606.0, + 440.0, + 650.0, + 291.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1941.0, + 590.0, + 1941.0, + 590.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1634.0, + 681.0, + 1634.0, + 681.0, + 1677.0, + 292.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 505.0, + 1165.0, + 505.0, + 1165.0, + 549.0, + 532.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1165.0, + 1407.0, + 1165.0, + 1407.0, + 1201.0, + 295.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1196.0, + 541.0, + 1196.0, + 541.0, + 1227.0, + 294.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 949.0, + 1404.0, + 949.0, + 1404.0, + 985.0, + 293.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 977.0, + 389.0, + 977.0, + 389.0, + 1019.0, + 291.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1467.0, + 1404.0, + 1467.0, + 1404.0, + 1506.0, + 292.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1497.0, + 389.0, + 1497.0, + 389.0, + 1539.0, + 291.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 669.0, + 1233.0, + 669.0, + 1233.0, + 708.0, + 296.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 949.0, + 1404.0, + 949.0, + 1404.0, + 985.0, + 293.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 977.0, + 389.0, + 977.0, + 389.0, + 1019.0, + 291.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1773.0, + 1408.0, + 1773.0, + 1408.0, + 1805.0, + 295.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1804.0, + 844.0, + 1804.0, + 844.0, + 1837.0, + 294.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 1804.0, + 1270.0, + 1804.0, + 1270.0, + 1837.0, + 915.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1804.0, + 1405.0, + 1804.0, + 1405.0, + 1837.0, + 1355.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1836.0, + 1404.0, + 1836.0, + 1404.0, + 1869.0, + 296.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1865.0, + 859.0, + 1865.0, + 859.0, + 1898.0, + 294.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1865.0, + 1118.0, + 1865.0, + 1118.0, + 1898.0, + 958.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 1865.0, + 1197.0, + 1865.0, + 1197.0, + 1898.0, + 1188.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2000.0, + 851.0, + 2000.0, + 851.0, + 2036.0, + 295.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 669.0, + 1233.0, + 669.0, + 1233.0, + 708.0, + 296.0, + 708.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 1667, + 1404, + 1667, + 1404, + 1792, + 299, + 1792 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1529, + 1405, + 1529, + 1405, + 1653, + 298, + 1653 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 342, + 224, + 1369, + 224, + 1369, + 598, + 342, + 598 + ], + "score": 0.971 + }, + { + "category_id": 4, + "poly": [ + 297, + 628, + 1405, + 628, + 1405, + 753, + 297, + 753 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 297, + 1971, + 1401, + 1971, + 1401, + 2034, + 297, + 2034 + ], + "score": 0.942 + }, + { + "category_id": 0, + "poly": [ + 297, + 1852, + 1335, + 1852, + 1335, + 1929, + 297, + 1929 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 297, + 1379, + 1173, + 1379, + 1173, + 1412, + 297, + 1412 + ], + "score": 0.918 + }, + { + "category_id": 0, + "poly": [ + 298, + 1315, + 430, + 1315, + 430, + 1347, + 298, + 1347 + ], + "score": 0.909 + }, + { + "category_id": 0, + "poly": [ + 301, + 1465, + 722, + 1465, + 722, + 1498, + 301, + 1498 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.895 + }, + { + "category_id": 6, + "poly": [ + 300, + 798, + 1405, + 798, + 1405, + 894, + 300, + 894 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2112, + 835, + 2112 + ], + "score": 0.842 + }, + { + "category_id": 5, + "poly": [ + 1099, + 963, + 1284, + 963, + 1284, + 1244, + 1099, + 1244 + ], + "score": 0.722, + "html": "
ogbl-citation2
80.04±0.25
80.44±0.10
79.84±0.19
75.95±1.3180.33±0.13
80.14±0.71
" + }, + { + "category_id": 1, + "poly": [ + 1165, + 920, + 1200, + 920, + 1200, + 948, + 1165, + 948 + ], + "score": 0.171 + }, + { + "category_id": 13, + "poly": [ + 709, + 1561, + 833, + 1561, + 833, + 1594, + 709, + 1594 + ], + "score": 0.92, + "latex": "\\in \\{ 1 , 4 , 8 \\}" + }, + { + "category_id": 13, + "poly": [ + 1188, + 1530, + 1310, + 1530, + 1310, + 1563, + 1188, + 1563 + ], + "score": 0.91, + "latex": "\\in \\{ 0 , 1 , 2 \\}" + }, + { + "category_id": 13, + "poly": [ + 915, + 661, + 998, + 661, + 998, + 689, + 915, + 689 + ], + "score": 0.9, + "latex": "k { = } 1 0 0" + }, + { + "category_id": 13, + "poly": [ + 460, + 1592, + 661, + 1592, + 661, + 1625, + 460, + 1625 + ], + "score": 0.9, + "latex": "\\in \\{ T r u e , F a l s e \\}" + }, + { + "category_id": 13, + "poly": [ + 813, + 1592, + 1016, + 1592, + 1016, + 1625, + 813, + 1625 + ], + "score": 0.88, + "latex": "\\dot { \\in } \\{ T r u e , F a l \\dot { s } e \\}" + }, + { + "category_id": 13, + "poly": [ + 684, + 722, + 781, + 722, + 781, + 750, + 684, + 750 + ], + "score": 0.88, + "latex": "d = 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 476, + 661, + 545, + 661, + 545, + 690, + 476, + 690 + ], + "score": 0.88, + "latex": "100 \\%" + }, + { + "category_id": 13, + "poly": [ + 943, + 722, + 1011, + 722, + 1011, + 750, + 943, + 750 + ], + "score": 0.88, + "latex": "k { = } 2 0" + }, + { + "category_id": 13, + "poly": [ + 940, + 1561, + 1128, + 1561, + 1128, + 1594, + 940, + 1594 + ], + "score": 0.87, + "latex": "\\in \\{ 0 . 4 , 0 . 6 , 0 . 8 \\}" + }, + { + "category_id": 13, + "poly": [ + 350, + 1561, + 500, + 1561, + 500, + 1594, + 350, + 1594 + ], + "score": 0.83, + "latex": "\\in \\{ 8 , 1 6 , 3 2 \\}" + }, + { + "category_id": 13, + "poly": [ + 651, + 1761, + 762, + 1761, + 762, + 1791, + 651, + 1791 + ], + "score": 0.81, + "latex": "< 0 . 0 0 0 1" + }, + { + "category_id": 13, + "poly": [ + 1192, + 1562, + 1397, + 1562, + 1397, + 1594, + 1192, + 1594 + ], + "score": 0.78, + "latex": "\\in \\{ T r u e , F a l s e \\}" + }, + { + "category_id": 13, + "poly": [ + 1021, + 862, + 1039, + 862, + 1039, + 894, + 1021, + 894 + ], + "score": 0.75, + "latex": "\\dagger" + }, + { + "category_id": 13, + "poly": [ + 540, + 800, + 594, + 800, + 594, + 829, + 540, + 829 + ], + "score": 0.66, + "latex": "@ 5 0" + }, + { + "category_id": 13, + "poly": [ + 568, + 569, + 586, + 569, + 586, + 594, + 568, + 594 + ], + "score": 0.46, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 305, + 692, + 345, + 692, + 345, + 720, + 305, + 720 + ], + "score": 0.38, + "latex": "\\mathrm { { X u } }" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 229.0, + 414.0, + 229.0, + 414.0, + 262.0, + 365.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 257.0, + 413.0, + 257.0, + 413.0, + 289.0, + 375.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 266.0, + 476.0, + 266.0, + 476.0, + 275.0, + 463.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 286.0, + 413.0, + 286.0, + 413.0, + 316.0, + 375.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 313.0, + 413.0, + 313.0, + 413.0, + 345.0, + 377.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 320.0, + 376.0, + 320.0, + 376.0, + 432.0, + 342.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 340.0, + 587.0, + 340.0, + 587.0, + 355.0, + 574.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 343.0, + 412.0, + 343.0, + 412.0, + 370.0, + 378.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 354.0, + 1361.0, + 354.0, + 1361.0, + 387.0, + 1158.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 371.0, + 412.0, + 371.0, + 412.0, + 398.0, + 378.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 397.0, + 412.0, + 397.0, + 412.0, + 425.0, + 378.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 382.0, + 1353.0, + 382.0, + 1353.0, + 423.0, + 1192.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 424.0, + 413.0, + 424.0, + 413.0, + 455.0, + 375.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 429.0, + 1188.0, + 429.0, + 1188.0, + 443.0, + 1170.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 415.0, + 1310.0, + 415.0, + 1310.0, + 452.0, + 1195.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 451.0, + 413.0, + 451.0, + 413.0, + 483.0, + 375.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 461.0, + 1187.0, + 461.0, + 1187.0, + 475.0, + 1171.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 447.0, + 1301.0, + 447.0, + 1301.0, + 484.0, + 1195.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 478.0, + 414.0, + 478.0, + 414.0, + 510.0, + 377.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 478.0, + 810.0, + 478.0, + 810.0, + 503.0, + 787.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 508.0, + 414.0, + 508.0, + 414.0, + 534.0, + 389.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 506.0, + 1027.0, + 506.0, + 1027.0, + 520.0, + 1010.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 506.0, + 1135.0, + 506.0, + 1135.0, + 520.0, + 1120.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 522.0, + 488.0, + 522.0, + 488.0, + 553.0, + 453.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 522.0, + 598.0, + 522.0, + 598.0, + 553.0, + 562.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 522.0, + 709.0, + 522.0, + 709.0, + 553.0, + 671.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 522.0, + 818.0, + 522.0, + 818.0, + 552.0, + 780.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 522.0, + 927.0, + 522.0, + 927.0, + 553.0, + 890.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 522.0, + 1036.0, + 522.0, + 1036.0, + 552.0, + 999.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 523.0, + 1144.0, + 523.0, + 1144.0, + 551.0, + 1111.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 564.0, + 567.0, + 564.0, + 567.0, + 602.0, + 563.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 564.0, + 998.0, + 564.0, + 998.0, + 602.0, + 587.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 628.0, + 1406.0, + 628.0, + 1406.0, + 664.0, + 292.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 661.0, + 475.0, + 661.0, + 475.0, + 694.0, + 295.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 661.0, + 914.0, + 661.0, + 914.0, + 694.0, + 546.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 661.0, + 1405.0, + 661.0, + 1405.0, + 694.0, + 999.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 692.0, + 304.0, + 692.0, + 304.0, + 724.0, + 295.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 692.0, + 1403.0, + 692.0, + 1403.0, + 724.0, + 346.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 721.0, + 683.0, + 721.0, + 683.0, + 756.0, + 292.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 721.0, + 942.0, + 721.0, + 942.0, + 756.0, + 782.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 721.0, + 1021.0, + 721.0, + 1021.0, + 756.0, + 1012.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1850.0, + 1340.0, + 1850.0, + 1340.0, + 1894.0, + 292.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1890.0, + 504.0, + 1890.0, + 504.0, + 1935.0, + 351.0, + 1935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1308.0, + 437.0, + 1308.0, + 437.0, + 1358.0, + 289.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1464.0, + 727.0, + 1464.0, + 727.0, + 1501.0, + 293.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 799.0, + 539.0, + 799.0, + 539.0, + 834.0, + 295.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 799.0, + 1405.0, + 799.0, + 1405.0, + 834.0, + 595.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 830.0, + 1407.0, + 830.0, + 1407.0, + 864.0, + 295.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 859.0, + 1020.0, + 859.0, + 1020.0, + 896.0, + 294.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 859.0, + 1404.0, + 859.0, + 1404.0, + 896.0, + 1040.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1669.0, + 1403.0, + 1669.0, + 1403.0, + 1701.0, + 293.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1700.0, + 1405.0, + 1700.0, + 1405.0, + 1733.0, + 295.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1732.0, + 1405.0, + 1732.0, + 1405.0, + 1761.0, + 295.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1761.0, + 650.0, + 1761.0, + 650.0, + 1794.0, + 294.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1761.0, + 779.0, + 1761.0, + 779.0, + 1794.0, + 763.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1527.0, + 1187.0, + 1527.0, + 1187.0, + 1565.0, + 293.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1311.0, + 1527.0, + 1405.0, + 1527.0, + 1405.0, + 1565.0, + 1311.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1558.0, + 349.0, + 1558.0, + 349.0, + 1597.0, + 293.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1558.0, + 708.0, + 1558.0, + 708.0, + 1597.0, + 501.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 1558.0, + 939.0, + 1558.0, + 939.0, + 1597.0, + 834.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1558.0, + 1191.0, + 1558.0, + 1191.0, + 1597.0, + 1129.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1558.0, + 1407.0, + 1558.0, + 1407.0, + 1597.0, + 1398.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1589.0, + 459.0, + 1589.0, + 459.0, + 1628.0, + 292.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 1589.0, + 812.0, + 1589.0, + 812.0, + 1628.0, + 662.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1589.0, + 1406.0, + 1589.0, + 1406.0, + 1628.0, + 1017.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1621.0, + 1099.0, + 1621.0, + 1099.0, + 1654.0, + 293.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1969.0, + 1403.0, + 1969.0, + 1403.0, + 2005.0, + 295.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2001.0, + 745.0, + 2001.0, + 745.0, + 2037.0, + 295.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1378.0, + 1175.0, + 1378.0, + 1175.0, + 1416.0, + 295.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 918.0, + 1205.0, + 918.0, + 1205.0, + 955.0, + 1161.0, + 955.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1525, + 1405, + 1525, + 1405, + 1742, + 297, + 1742 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1417, + 1404, + 1417, + 1404, + 1510, + 297, + 1510 + ], + "score": 0.974 + }, + { + "category_id": 5, + "poly": [ + 747, + 1085, + 953, + 1085, + 953, + 1207, + 747, + 1207 + ], + "score": 0.959, + "html": "
ModelAccuracy
GAT78.1±0.59
GATv278.5±0.38
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1264, + 1406, + 1264, + 1406, + 1327, + 298, + 1327 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 295, + 1971, + 1400, + 1971, + 1400, + 2037, + 295, + 2037 + ], + "score": 0.941 + }, + { + "category_id": 5, + "poly": [ + 327, + 223, + 1372, + 223, + 1372, + 849, + 327, + 849 + ], + "score": 0.939, + "html": "
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.25 18.71±23.369.05±1.21 5.62±0.817.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h10.80±2.183.11±0.14
GAT8h28.10±16.45 7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.365.37±0.26 4.11±0.271.48±0.87+134.1% +0%
GATv21h22.94±17.34
10.19±0.6322.56±17.4615.04±4.585.23±0.362.46±0.65+91.6%
GATv28h6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
" + }, + { + "category_id": 6, + "poly": [ + 365, + 1027, + 1328, + 1027, + 1328, + 1062, + 365, + 1062 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.898 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1358, + 1399, + 1358, + 1399, + 1389, + 1353, + 1389 + ], + "score": 0.895 + }, + { + "category_id": 1, + "poly": [ + 297, + 1869, + 1400, + 1869, + 1400, + 1932, + 297, + 1932 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2113, + 835, + 2113 + ], + "score": 0.846 + }, + { + "category_id": 8, + "poly": [ + 370, + 1344, + 1286, + 1344, + 1286, + 1398, + 370, + 1398 + ], + "score": 0.831 + }, + { + "category_id": 1, + "poly": [ + 298, + 871, + 1405, + 871, + 1405, + 994, + 298, + 994 + ], + "score": 0.768 + }, + { + "category_id": 1, + "poly": [ + 299, + 1778, + 1392, + 1778, + 1392, + 1842, + 299, + 1842 + ], + "score": 0.767 + }, + { + "category_id": 6, + "poly": [ + 298, + 871, + 1405, + 871, + 1405, + 994, + 298, + 994 + ], + "score": 0.255 + }, + { + "category_id": 0, + "poly": [ + 299, + 1778, + 1392, + 1778, + 1392, + 1842, + 299, + 1842 + ], + "score": 0.191 + }, + { + "category_id": 1, + "poly": [ + 370, + 1344, + 1286, + 1344, + 1286, + 1398, + 370, + 1398 + ], + "score": 0.116 + }, + { + "category_id": 13, + "poly": [ + 298, + 1899, + 557, + 1899, + 557, + 1933, + 298, + 1933 + ], + "score": 0.93, + "latex": "\\mathbb { K } = \\mathbb { Q } = \\{ h _ { 1 } , . . . , h _ { n } \\}" + }, + { + "category_id": 13, + "poly": [ + 473, + 2003, + 614, + 2003, + 614, + 2036, + 473, + 2036 + ], + "score": 0.93, + "latex": "\\{ h _ { 1 } , . . . , h _ { n } \\}" + }, + { + "category_id": 13, + "poly": [ + 671, + 2002, + 822, + 2002, + 822, + 2036, + 671, + 2036 + ], + "score": 0.93, + "latex": "\\varphi : [ n ] [ n ]" + }, + { + "category_id": 13, + "poly": [ + 1071, + 2002, + 1179, + 2002, + 1179, + 2036, + 1071, + 2036 + ], + "score": 0.92, + "latex": "[ { \\bar { n } } ] [ n ]" + }, + { + "category_id": 13, + "poly": [ + 425, + 1973, + 556, + 1973, + 556, + 2004, + 425, + 2004 + ], + "score": 0.92, + "latex": "\\mathcal { G } = ( \\nu , \\mathcal { E } )" + }, + { + "category_id": 14, + "poly": [ + 802, + 1342, + 1284, + 1342, + 1284, + 1403, + 802, + 1403 + ], + "score": 0.89, + "latex": "e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right) = \\left( \\left( \\pmb { h } _ { i } ^ { \\top } \\pmb { Q } \\right) \\cdot \\left( \\pmb { h } _ { j } ^ { \\top } \\pmb { K } \\right) ^ { \\top } \\right) / \\sqrt { d _ { k } }" + }, + { + "category_id": 13, + "poly": [ + 684, + 1623, + 706, + 1623, + 706, + 1650, + 684, + 1650 + ], + "score": 0.76, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 954, + 934, + 972, + 934, + 972, + 966, + 954, + 966 + ], + "score": 0.74, + "latex": "\\dagger" + }, + { + "category_id": 13, + "poly": [ + 907, + 874, + 937, + 874, + 937, + 901, + 907, + 901 + ], + "score": 0.74, + "latex": "\\pm" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 1027.0, + 1331.0, + 1027.0, + 1331.0, + 1062.0, + 366.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 867.0, + 906.0, + 867.0, + 906.0, + 907.0, + 292.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 867.0, + 1405.0, + 867.0, + 1405.0, + 907.0, + 938.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 901.0, + 1407.0, + 901.0, + 1407.0, + 937.0, + 293.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 932.0, + 953.0, + 932.0, + 953.0, + 967.0, + 293.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 932.0, + 1403.0, + 932.0, + 1403.0, + 967.0, + 973.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 963.0, + 547.0, + 963.0, + 547.0, + 996.0, + 295.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1778.0, + 1397.0, + 1778.0, + 1397.0, + 1814.0, + 297.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 1810.0, + 679.0, + 1810.0, + 679.0, + 1845.0, + 366.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1527.0, + 1403.0, + 1527.0, + 1403.0, + 1562.0, + 296.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1557.0, + 1405.0, + 1557.0, + 1405.0, + 1590.0, + 294.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1588.0, + 1406.0, + 1588.0, + 1406.0, + 1622.0, + 295.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1615.0, + 683.0, + 1615.0, + 683.0, + 1655.0, + 292.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1615.0, + 1408.0, + 1615.0, + 1408.0, + 1655.0, + 707.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1646.0, + 1407.0, + 1646.0, + 1407.0, + 1683.0, + 292.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1679.0, + 1405.0, + 1679.0, + 1405.0, + 1714.0, + 294.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1708.0, + 543.0, + 1708.0, + 543.0, + 1745.0, + 291.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1413.0, + 1405.0, + 1413.0, + 1405.0, + 1455.0, + 292.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1448.0, + 1404.0, + 1448.0, + 1404.0, + 1484.0, + 294.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1480.0, + 713.0, + 1480.0, + 713.0, + 1513.0, + 293.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1265.0, + 1404.0, + 1265.0, + 1404.0, + 1301.0, + 296.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1294.0, + 1261.0, + 1294.0, + 1261.0, + 1330.0, + 295.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1966.0, + 424.0, + 1966.0, + 424.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1966.0, + 1405.0, + 1966.0, + 1405.0, + 2009.0, + 557.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 472.0, + 2000.0, + 472.0, + 2040.0, + 294.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 2000.0, + 670.0, + 2000.0, + 670.0, + 2040.0, + 615.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 2000.0, + 1070.0, + 2000.0, + 1070.0, + 2040.0, + 823.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 2000.0, + 1194.0, + 2000.0, + 1194.0, + 2040.0, + 1180.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1865.0, + 1405.0, + 1865.0, + 1405.0, + 1908.0, + 293.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1895.0, + 297.0, + 1895.0, + 297.0, + 1937.0, + 293.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1895.0, + 894.0, + 1895.0, + 894.0, + 1937.0, + 558.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 867.0, + 906.0, + 867.0, + 906.0, + 907.0, + 292.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 867.0, + 1405.0, + 867.0, + 1405.0, + 907.0, + 938.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 901.0, + 1407.0, + 901.0, + 1407.0, + 937.0, + 293.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 932.0, + 953.0, + 932.0, + 953.0, + 967.0, + 293.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 932.0, + 1403.0, + 932.0, + 1403.0, + 967.0, + 973.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 963.0, + 547.0, + 963.0, + 547.0, + 996.0, + 295.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1778.0, + 1397.0, + 1778.0, + 1397.0, + 1814.0, + 297.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 1810.0, + 679.0, + 1810.0, + 679.0, + 1845.0, + 366.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1354.0, + 738.0, + 1354.0, + 738.0, + 1392.0, + 370.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 796.0, + 1342.0, + 801.0, + 1342.0, + 801.0, + 1401.0, + 796.0, + 1401.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 687, + 1841, + 1006, + 1841, + 1006, + 1916, + 687, + 1916 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 646, + 1963, + 1051, + 1963, + 1051, + 2041, + 646, + 2041 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 719, + 303, + 978, + 303, + 978, + 378, + 719, + 378 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 723, + 493, + 972, + 493, + 972, + 645, + 723, + 645 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 298, + 816, + 1406, + 816, + 1406, + 882, + 298, + 882 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 720, + 764, + 980, + 764, + 980, + 806, + 720, + 806 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 608, + 1742, + 1090, + 1742, + 1090, + 1805, + 608, + 1805 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 682, + 1420, + 1021, + 1420, + 1021, + 1463, + 682, + 1463 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 769, + 942, + 928, + 942, + 928, + 980, + 769, + 980 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 297, + 993, + 1066, + 993, + 1066, + 1027, + 297, + 1027 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 297, + 895, + 1086, + 895, + 1086, + 929, + 297, + 929 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 296, + 1684, + 888, + 1684, + 888, + 1724, + 296, + 1724 + ], + "score": 0.931 + }, + { + "category_id": 8, + "poly": [ + 719, + 1333, + 980, + 1333, + 980, + 1372, + 719, + 1372 + ], + "score": 0.931 + }, + { + "category_id": 8, + "poly": [ + 776, + 1118, + 920, + 1118, + 920, + 1155, + 776, + 1155 + ], + "score": 0.93 + }, + { + "category_id": 8, + "poly": [ + 731, + 1255, + 966, + 1255, + 966, + 1293, + 731, + 1293 + ], + "score": 0.93 + }, + { + "category_id": 8, + "poly": [ + 764, + 1039, + 937, + 1039, + 937, + 1074, + 764, + 1074 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 293, + 405, + 1250, + 405, + 1250, + 441, + 293, + 441 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 297, + 1940, + 408, + 1940, + 408, + 1972, + 297, + 1972 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 298, + 1090, + 631, + 1090, + 631, + 1123, + 298, + 1123 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 296, + 1816, + 417, + 1816, + 417, + 1848, + 296, + 1848 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 297, + 1309, + 402, + 1309, + 402, + 1340, + 297, + 1340 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 296, + 1513, + 1155, + 1513, + 1155, + 1548, + 296, + 1548 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 297, + 671, + 1153, + 671, + 1153, + 706, + 297, + 706 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 298, + 1160, + 657, + 1160, + 657, + 1194, + 298, + 1194 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 299, + 227, + 799, + 227, + 799, + 262, + 299, + 262 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 297, + 1395, + 419, + 1395, + 419, + 1428, + 297, + 1428 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 298, + 1206, + 1117, + 1206, + 1117, + 1242, + 298, + 1242 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 298, + 276, + 579, + 276, + 579, + 309, + 298, + 309 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 297, + 715, + 1133, + 715, + 1133, + 751, + 297, + 751 + ], + "score": 0.916 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 816, + 73, + 816, + 106, + 297, + 106 + ], + "score": 0.915 + }, + { + "category_id": 9, + "poly": [ + 1351, + 325, + 1402, + 325, + 1402, + 356, + 1351, + 356 + ], + "score": 0.901 + }, + { + "category_id": 1, + "poly": [ + 298, + 1466, + 1043, + 1466, + 1043, + 1500, + 298, + 1500 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1351, + 770, + 1400, + 770, + 1400, + 801, + 1351, + 801 + ], + "score": 0.898 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1123, + 1401, + 1123, + 1401, + 1154, + 1351, + 1154 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1987, + 1401, + 1987, + 1401, + 2019, + 1351, + 2019 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1428, + 1401, + 1428, + 1401, + 1459, + 1351, + 1459 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1760, + 1401, + 1760, + 1401, + 1791, + 1352, + 1791 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1574, + 1400, + 1574, + 1400, + 1606, + 1352, + 1606 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1352, + 947, + 1400, + 947, + 1400, + 979, + 1352, + 979 + ], + "score": 0.892 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1262, + 1400, + 1262, + 1400, + 1293, + 1352, + 1293 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1351, + 552, + 1401, + 552, + 1401, + 584, + 1351, + 584 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1863, + 1401, + 1863, + 1401, + 1894, + 1352, + 1894 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1624, + 1401, + 1624, + 1401, + 1654, + 1351, + 1654 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1044, + 1401, + 1044, + 1401, + 1075, + 1352, + 1075 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1340, + 1400, + 1340, + 1400, + 1371, + 1352, + 1371 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2113, + 835, + 2113 + ], + "score": 0.86 + }, + { + "category_id": 8, + "poly": [ + 664, + 1562, + 1030, + 1562, + 1030, + 1660, + 664, + 1660 + ], + "score": 0.639 + }, + { + "category_id": 8, + "poly": [ + 783, + 1619, + 1023, + 1619, + 1023, + 1660, + 783, + 1660 + ], + "score": 0.606 + }, + { + "category_id": 8, + "poly": [ + 666, + 1563, + 1029, + 1563, + 1029, + 1612, + 666, + 1612 + ], + "score": 0.596 + }, + { + "category_id": 14, + "poly": [ + 649, + 1963, + 1051, + 1963, + 1051, + 2042, + 649, + 2042 + ], + "score": 0.94, + "latex": "e ( \\pmb { h } _ { i } , \\pmb { h } _ { j } ) = \\left\\{ \\begin{array} { l l } { 1 / \\sqrt { d _ { k } } } & { j = \\varphi ( i ) } \\\\ { 0 } & { o t h e r w i s e } \\end{array} \\right." + }, + { + "category_id": 14, + "poly": [ + 686, + 1840, + 1010, + 1840, + 1010, + 1917, + 686, + 1917 + ], + "score": 0.93, + "latex": "z \\left( \\pmb { { h } } _ { i } , \\pmb { { h } } _ { j } \\right) = \\left. \\begin{array} { l l } { 1 } & { j = \\varphi ( i ) } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right." + }, + { + "category_id": 14, + "poly": [ + 717, + 301, + 976, + 301, + 976, + 379, + 717, + 379 + ], + "score": 0.93, + "latex": "P _ { i , j } = { \\left\\{ \\begin{array} { l l } { 1 } & { j = \\varphi ( i ) } \\\\ { 0 } & { { \\mathrm { o t h e r w i s e } } } \\end{array} \\right. }" + }, + { + "category_id": 13, + "poly": [ + 932, + 818, + 1049, + 818, + 1049, + 850, + 932, + 850 + ], + "score": 0.93, + "latex": "\\mathbf { \\Delta } X Q = U" + }, + { + "category_id": 14, + "poly": [ + 607, + 1737, + 1091, + 1737, + 1091, + 1806, + 607, + 1806 + ], + "score": 0.93, + "latex": "z \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right) = \\left( \\pmb { X } \\pmb { Q } \\right) _ { i } \\cdot \\left( \\left( \\pmb { X } \\pmb { K } \\right) _ { j } \\right) ^ { \\top } = P _ { i , j }" + }, + { + "category_id": 14, + "poly": [ + 723, + 492, + 978, + 492, + 978, + 643, + 723, + 643 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\mathbf { \\boldsymbol { X } } = \\left[ \\begin{array} { l l l } { - } & { h _ { 1 } } & { - } \\\\ { - } & { h _ { 2 } } & { - } \\\\ & { \\vdots } \\\\ { - } & { h _ { n } } & { - } \\end{array} \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 367, + 1688, + 548, + 1688, + 548, + 1723, + 367, + 1723 + ], + "score": 0.93, + "latex": "X _ { i } Q = ( X Q ) _ { i }" + }, + { + "category_id": 14, + "poly": [ + 664, + 1559, + 1034, + 1559, + 1034, + 1664, + 664, + 1664 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { z \\left( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } \\right) = \\left( \\boldsymbol { h } _ { i } ^ { \\top } \\boldsymbol { Q } \\right) \\cdot \\left( \\boldsymbol { h } _ { j } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ { = \\left( \\boldsymbol { X } _ { i } \\boldsymbol { Q } \\right) \\cdot \\left( \\boldsymbol { X } _ { j } \\boldsymbol { K } \\right) ^ { \\top } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 600, + 1688, + 796, + 1688, + 796, + 1727, + 600, + 1727 + ], + "score": 0.93, + "latex": "X _ { j } K = \\left( X K \\right) _ { j }" + }, + { + "category_id": 13, + "poly": [ + 342, + 404, + 508, + 404, + 508, + 436, + 342, + 436 + ], + "score": 0.92, + "latex": "\\ b { X } \\in \\mathbb { R } ^ { n } \\times \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 561, + 1206, + 726, + 1206, + 726, + 1238, + 561, + 1238 + ], + "score": 0.92, + "latex": "X K = P ^ { \\top } U" + }, + { + "category_id": 14, + "poly": [ + 679, + 1418, + 1023, + 1418, + 1023, + 1463, + 679, + 1463 + ], + "score": 0.92, + "latex": "z \\left( h _ { i } , h _ { j } \\right) = e \\left( h _ { i } , h _ { j } \\right) \\cdot \\sqrt { d _ { k } }" + }, + { + "category_id": 13, + "poly": [ + 637, + 716, + 797, + 716, + 797, + 751, + 637, + 751 + ], + "score": 0.92, + "latex": "Q \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 849, + 716, + 1013, + 716, + 1013, + 749, + 849, + 749 + ], + "score": 0.92, + "latex": "K \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 407, + 1163, + 524, + 1163, + 524, + 1195, + 407, + 1195 + ], + "score": 0.92, + "latex": "\\mathbf { \\Delta } X Q = U" + }, + { + "category_id": 13, + "poly": [ + 1101, + 816, + 1267, + 816, + 1267, + 847, + 1101, + 847 + ], + "score": 0.92, + "latex": "X K = P ^ { \\top } U" + }, + { + "category_id": 13, + "poly": [ + 1074, + 1515, + 1146, + 1515, + 1146, + 1549, + 1074, + 1549 + ], + "score": 0.92, + "latex": "h _ { i } , h _ { j }" + }, + { + "category_id": 13, + "poly": [ + 367, + 996, + 529, + 996, + 529, + 1024, + 367, + 1024 + ], + "score": 0.92, + "latex": "\\ b { \\Sigma } \\in \\mathbb { R } ^ { n } \\times \\mathbb { R } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 720, + 763, + 980, + 763, + 980, + 806, + 720, + 806 + ], + "score": 0.91, + "latex": "( X Q ) \\cdot ( X K ) ^ { \\top } = P" + }, + { + "category_id": 13, + "poly": [ + 1081, + 674, + 1151, + 674, + 1151, + 703, + 1081, + 703 + ], + "score": 0.91, + "latex": "d \\geq n" + }, + { + "category_id": 14, + "poly": [ + 760, + 1037, + 939, + 1037, + 939, + 1075, + 760, + 1075 + ], + "score": 0.9, + "latex": "\\pmb { X } \\pmb { V } \\pmb { \\Sigma } ^ { - 1 } = \\pmb { U }" + }, + { + "category_id": 14, + "poly": [ + 767, + 940, + 934, + 940, + 934, + 978, + 767, + 978 + ], + "score": 0.9, + "latex": "\\pmb { X } = \\pmb { U } \\pmb { \\Sigma } \\pmb { V } ^ { \\top }" + }, + { + "category_id": 14, + "poly": [ + 731, + 1252, + 969, + 1252, + 969, + 1293, + 731, + 1293 + ], + "score": 0.9, + "latex": "U \\Sigma V ^ { \\top } K = P ^ { \\top } U" + }, + { + "category_id": 14, + "poly": [ + 775, + 1116, + 923, + 1116, + 923, + 1156, + 775, + 1156 + ], + "score": 0.9, + "latex": "Q = V \\Sigma ^ { - 1 }" + }, + { + "category_id": 14, + "poly": [ + 717, + 1331, + 983, + 1331, + 983, + 1371, + 717, + 1371 + ], + "score": 0.9, + "latex": "\\pmb { K } = \\pmb { V } \\pmb { \\Sigma } ^ { - 1 } \\pmb { U } ^ { T } \\pmb { P } ^ { \\top } \\pmb { U }" + }, + { + "category_id": 13, + "poly": [ + 748, + 231, + 790, + 231, + 790, + 261, + 748, + 261 + ], + "score": 0.88, + "latex": "M _ { i }" + }, + { + "category_id": 13, + "poly": [ + 459, + 228, + 494, + 228, + 494, + 258, + 459, + 258 + ], + "score": 0.88, + "latex": "\\mathrm { i } ^ { t h }" + }, + { + "category_id": 13, + "poly": [ + 558, + 848, + 839, + 848, + 839, + 879, + 558, + 879 + ], + "score": 0.87, + "latex": "( \\pmb { U } \\cdot \\pmb { U } ^ { \\top } = \\pmb { U } ^ { \\top } \\cdot \\pmb { U } = I )" + }, + { + "category_id": 13, + "poly": [ + 476, + 1093, + 504, + 1093, + 504, + 1123, + 476, + 1123 + ], + "score": 0.84, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 497, + 674, + 527, + 674, + 527, + 700, + 497, + 700 + ], + "score": 0.84, + "latex": "\\boldsymbol { X }" + }, + { + "category_id": 13, + "poly": [ + 788, + 819, + 819, + 819, + 819, + 846, + 788, + 846 + ], + "score": 0.84, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 787, + 996, + 813, + 996, + 813, + 1023, + 787, + 1023 + ], + "score": 0.83, + "latex": "\\pmb { \\Sigma }" + }, + { + "category_id": 13, + "poly": [ + 1044, + 897, + 1074, + 897, + 1074, + 924, + 1044, + 924 + ], + "score": 0.83, + "latex": "\\boldsymbol { X }" + }, + { + "category_id": 13, + "poly": [ + 1346, + 820, + 1374, + 820, + 1374, + 846, + 1346, + 846 + ], + "score": 0.83, + "latex": "U" + }, + { + "category_id": 13, + "poly": [ + 709, + 820, + 736, + 820, + 736, + 849, + 709, + 849 + ], + "score": 0.82, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 620, + 1517, + 635, + 1517, + 635, + 1547, + 620, + 1547 + ], + "score": 0.81, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 461, + 897, + 489, + 897, + 489, + 924, + 461, + 924 + ], + "score": 0.81, + "latex": "U" + }, + { + "category_id": 13, + "poly": [ + 581, + 997, + 611, + 997, + 611, + 1023, + 581, + 1023 + ], + "score": 0.81, + "latex": "\\boldsymbol { X }" + }, + { + "category_id": 13, + "poly": [ + 513, + 279, + 540, + 279, + 540, + 305, + 513, + 305 + ], + "score": 0.81, + "latex": "_ { P }" + }, + { + "category_id": 13, + "poly": [ + 383, + 1210, + 414, + 1210, + 414, + 1237, + 383, + 1237 + ], + "score": 0.8, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 676, + 231, + 713, + 231, + 713, + 259, + 676, + 259 + ], + "score": 0.8, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 491, + 1518, + 504, + 1518, + 504, + 1542, + 491, + 1542 + ], + "score": 0.76, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 377, + 1474, + 393, + 1474, + 393, + 1494, + 377, + 1494 + ], + "score": 0.75, + "latex": "e" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 815.0, + 708.0, + 815.0, + 708.0, + 852.0, + 293.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 815.0, + 787.0, + 815.0, + 787.0, + 852.0, + 737.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 815.0, + 931.0, + 815.0, + 931.0, + 852.0, + 820.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 815.0, + 1100.0, + 815.0, + 1100.0, + 852.0, + 1050.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 815.0, + 1345.0, + 815.0, + 1345.0, + 852.0, + 1268.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 815.0, + 1406.0, + 815.0, + 1406.0, + 852.0, + 1375.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 843.0, + 557.0, + 843.0, + 557.0, + 887.0, + 292.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 843.0, + 856.0, + 843.0, + 856.0, + 887.0, + 840.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 991.0, + 366.0, + 991.0, + 366.0, + 1032.0, + 294.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 991.0, + 580.0, + 991.0, + 580.0, + 1032.0, + 530.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 991.0, + 786.0, + 991.0, + 786.0, + 1032.0, + 612.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 991.0, + 1069.0, + 991.0, + 1069.0, + 1032.0, + 814.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 893.0, + 460.0, + 893.0, + 460.0, + 933.0, + 296.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 893.0, + 1043.0, + 893.0, + 1043.0, + 933.0, + 490.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 893.0, + 1085.0, + 893.0, + 1085.0, + 933.0, + 1075.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1680.0, + 366.0, + 1680.0, + 366.0, + 1733.0, + 293.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1680.0, + 599.0, + 1680.0, + 599.0, + 1733.0, + 549.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 1680.0, + 889.0, + 1680.0, + 889.0, + 1733.0, + 797.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 402.0, + 341.0, + 402.0, + 341.0, + 443.0, + 293.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 402.0, + 1254.0, + 402.0, + 1254.0, + 443.0, + 509.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1937.0, + 413.0, + 1937.0, + 413.0, + 1975.0, + 295.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1087.0, + 475.0, + 1087.0, + 475.0, + 1128.0, + 294.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1087.0, + 635.0, + 1087.0, + 635.0, + 1128.0, + 505.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1812.0, + 421.0, + 1812.0, + 421.0, + 1851.0, + 294.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1306.0, + 408.0, + 1306.0, + 408.0, + 1342.0, + 292.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1506.0, + 490.0, + 1506.0, + 490.0, + 1557.0, + 292.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1506.0, + 619.0, + 1506.0, + 619.0, + 1557.0, + 505.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1506.0, + 1073.0, + 1506.0, + 1073.0, + 1557.0, + 636.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1506.0, + 1159.0, + 1506.0, + 1159.0, + 1557.0, + 1147.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 665.0, + 496.0, + 665.0, + 496.0, + 712.0, + 294.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 665.0, + 1080.0, + 665.0, + 1080.0, + 712.0, + 528.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 665.0, + 1158.0, + 665.0, + 1158.0, + 712.0, + 1152.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1159.0, + 406.0, + 1159.0, + 406.0, + 1199.0, + 294.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1159.0, + 657.0, + 1159.0, + 657.0, + 1199.0, + 525.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 222.0, + 458.0, + 222.0, + 458.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 222.0, + 675.0, + 222.0, + 675.0, + 268.0, + 495.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 222.0, + 747.0, + 222.0, + 747.0, + 268.0, + 714.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 222.0, + 804.0, + 222.0, + 804.0, + 268.0, + 791.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1392.0, + 423.0, + 1392.0, + 423.0, + 1431.0, + 293.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1201.0, + 382.0, + 1201.0, + 382.0, + 1248.0, + 292.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1201.0, + 560.0, + 1201.0, + 560.0, + 1248.0, + 415.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 1201.0, + 1121.0, + 1201.0, + 1121.0, + 1248.0, + 727.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 273.0, + 512.0, + 273.0, + 512.0, + 312.0, + 295.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 273.0, + 581.0, + 273.0, + 581.0, + 312.0, + 541.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 711.0, + 636.0, + 711.0, + 636.0, + 756.0, + 291.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 711.0, + 848.0, + 711.0, + 848.0, + 756.0, + 798.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 711.0, + 1133.0, + 711.0, + 1133.0, + 756.0, + 1014.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1462.0, + 376.0, + 1462.0, + 376.0, + 1505.0, + 294.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1462.0, + 1040.0, + 1462.0, + 1040.0, + 1505.0, + 394.0, + 1505.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 927, + 1405, + 927, + 1405, + 1021, + 297, + 1021 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 530, + 1403, + 530, + 1403, + 627, + 297, + 627 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 296, + 454, + 1402, + 454, + 1402, + 520, + 296, + 520 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 690, + 1686, + 1009, + 1686, + 1009, + 1728, + 690, + 1728 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 296, + 1542, + 1404, + 1542, + 1404, + 1604, + 296, + 1604 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 298, + 1734, + 1404, + 1734, + 1404, + 1800, + 298, + 1800 + ], + "score": 0.938 + }, + { + "category_id": 8, + "poly": [ + 701, + 276, + 997, + 276, + 997, + 316, + 701, + 316 + ], + "score": 0.933 + }, + { + "category_id": 8, + "poly": [ + 769, + 370, + 928, + 370, + 928, + 403, + 769, + 403 + ], + "score": 0.933 + }, + { + "category_id": 8, + "poly": [ + 696, + 718, + 999, + 718, + 999, + 793, + 696, + 793 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 663, + 1864, + 993, + 1864, + 993, + 1924, + 663, + 1924 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 297, + 228, + 957, + 228, + 957, + 264, + 297, + 264 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 297, + 323, + 873, + 323, + 873, + 356, + 297, + 356 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 409, + 882, + 409, + 882, + 443, + 297, + 443 + ], + "score": 0.926 + }, + { + "category_id": 8, + "poly": [ + 767, + 1605, + 931, + 1605, + 931, + 1637, + 767, + 1637 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 296, + 1378, + 1137, + 1378, + 1137, + 1412, + 296, + 1412 + ], + "score": 0.923 + }, + { + "category_id": 8, + "poly": [ + 697, + 634, + 995, + 634, + 995, + 708, + 697, + 708 + ], + "score": 0.922 + }, + { + "category_id": 8, + "poly": [ + 661, + 1931, + 1007, + 1931, + 1007, + 1977, + 661, + 1977 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 297, + 869, + 903, + 869, + 903, + 903, + 297, + 903 + ], + "score": 0.916 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 816, + 73, + 816, + 106, + 297, + 106 + ], + "score": 0.914 + }, + { + "category_id": 8, + "poly": [ + 660, + 1983, + 1141, + 1983, + 1141, + 2031, + 660, + 2031 + ], + "score": 0.907 + }, + { + "category_id": 1, + "poly": [ + 298, + 1642, + 1392, + 1642, + 1392, + 1675, + 298, + 1675 + ], + "score": 0.907 + }, + { + "category_id": 8, + "poly": [ + 554, + 1810, + 889, + 1810, + 889, + 1857, + 554, + 1857 + ], + "score": 0.902 + }, + { + "category_id": 9, + "poly": [ + 1352, + 367, + 1400, + 367, + 1400, + 398, + 1352, + 398 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1351, + 740, + 1400, + 740, + 1400, + 771, + 1351, + 771 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1351, + 656, + 1400, + 656, + 1400, + 686, + 1351, + 686 + ], + "score": 0.888 + }, + { + "category_id": 8, + "poly": [ + 770, + 1500, + 929, + 1500, + 929, + 1533, + 770, + 1533 + ], + "score": 0.886 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1605, + 1399, + 1605, + 1399, + 1635, + 1352, + 1635 + ], + "score": 0.885 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1423, + 1400, + 1423, + 1400, + 1453, + 1352, + 1453 + ], + "score": 0.885 + }, + { + "category_id": 9, + "poly": [ + 1352, + 280, + 1400, + 280, + 1400, + 310, + 1352, + 310 + ], + "score": 0.884 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1693, + 1400, + 1693, + 1400, + 1723, + 1352, + 1723 + ], + "score": 0.884 + }, + { + "category_id": 8, + "poly": [ + 772, + 1422, + 862, + 1422, + 862, + 1454, + 772, + 1454 + ], + "score": 0.881 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1821, + 1400, + 1821, + 1400, + 1851, + 1352, + 1851 + ], + "score": 0.88 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1882, + 1400, + 1882, + 1400, + 1912, + 1352, + 1912 + ], + "score": 0.88 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1994, + 1401, + 1994, + 1401, + 2024, + 1352, + 2024 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1941, + 1400, + 1941, + 1400, + 1972, + 1352, + 1972 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1502, + 1400, + 1502, + 1400, + 1531, + 1352, + 1531 + ], + "score": 0.872 + }, + { + "category_id": 8, + "poly": [ + 771, + 1462, + 915, + 1462, + 915, + 1494, + 771, + 1494 + ], + "score": 0.869 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 863, + 2087, + 863, + 2112, + 835, + 2112 + ], + "score": 0.866 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1463, + 1400, + 1463, + 1400, + 1492, + 1352, + 1492 + ], + "score": 0.857 + }, + { + "category_id": 2, + "poly": [ + 1375, + 804, + 1402, + 804, + 1402, + 833, + 1375, + 833 + ], + "score": 0.803 + }, + { + "category_id": 4, + "poly": [ + 294, + 1283, + 1403, + 1283, + 1403, + 1348, + 294, + 1348 + ], + "score": 0.802 + }, + { + "category_id": 3, + "poly": [ + 711, + 1054, + 984, + 1054, + 984, + 1255, + 711, + 1255 + ], + "score": 0.598 + }, + { + "category_id": 8, + "poly": [ + 711, + 1054, + 984, + 1054, + 984, + 1255, + 711, + 1255 + ], + "score": 0.57 + }, + { + "category_id": 1, + "poly": [ + 294, + 1283, + 1403, + 1283, + 1403, + 1348, + 294, + 1348 + ], + "score": 0.1 + }, + { + "category_id": 14, + "poly": [ + 549, + 1804, + 1148, + 1804, + 1148, + 2039, + 549, + 2039 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { s \\left( \\pmb { q } , \\pmb { h } _ { \\beta } \\right) = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { h } _ { \\beta } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\left( \\hat { \\pmb { x } } + \\beta \\pmb { \\hat { y } } \\right) ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { x } } ^ { \\top } \\pmb { K } + \\beta \\pmb { \\hat { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad \\quad = \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { x } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } + \\beta \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { \\hat { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 467, + 1766, + 614, + 1766, + 614, + 1800, + 467, + 1800 + ], + "score": 0.93, + "latex": "\\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \\}" + }, + { + "category_id": 13, + "poly": [ + 917, + 1543, + 1064, + 1543, + 1064, + 1577, + 917, + 1577 + ], + "score": 0.93, + "latex": "\\beta \\in \\{ 0 , 1 , 2 \\}" + }, + { + "category_id": 13, + "poly": [ + 1187, + 1543, + 1334, + 1543, + 1334, + 1577, + 1187, + 1577 + ], + "score": 0.93, + "latex": "\\{ h _ { 0 } , h _ { 1 } , h _ { 2 } \\}" + }, + { + "category_id": 13, + "poly": [ + 479, + 531, + 655, + 531, + 655, + 565, + 479, + 565 + ], + "score": 0.93, + "latex": "Q \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k } }" + }, + { + "category_id": 14, + "poly": [ + 689, + 1682, + 1013, + 1682, + 1013, + 1728, + 689, + 1728 + ], + "score": 0.93, + "latex": "s \\left( \\mathbf { q } , \\boldsymbol { h } _ { \\beta } \\right) = e \\left( \\mathbf { q } , \\boldsymbol { h } _ { \\beta } \\right) \\cdot \\sqrt { d _ { k } }" + }, + { + "category_id": 13, + "poly": [ + 983, + 454, + 1160, + 454, + 1160, + 487, + 983, + 487 + ], + "score": 0.92, + "latex": "X X ^ { \\top } \\in \\mathbb { R } ^ { n \\times n }" + }, + { + "category_id": 13, + "poly": [ + 341, + 1644, + 416, + 1644, + 416, + 1675, + 341, + 1675 + ], + "score": 0.92, + "latex": "\\pmb q \\in \\mathbb { Q }" + }, + { + "category_id": 14, + "poly": [ + 767, + 1602, + 934, + 1602, + 934, + 1640, + 767, + 1640 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\pmb { h } _ { \\beta } = \\hat { \\pmb { x } } + \\beta \\hat { \\pmb y } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 702, + 274, + 997, + 274, + 997, + 317, + 702, + 317 + ], + "score": 0.92, + "latex": "e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { \\varphi ( i ) } \\right) > e \\left( \\pmb { h } _ { i } , \\pmb { h } _ { j } \\right)" + }, + { + "category_id": 13, + "poly": [ + 949, + 564, + 1144, + 564, + 1144, + 597, + 949, + 597 + ], + "score": 0.92, + "latex": "Q ^ { \\prime } \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k e y } }" + }, + { + "category_id": 14, + "poly": [ + 695, + 631, + 1001, + 631, + 1001, + 797, + 695, + 797 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { Q _ { i , j } ^ { \\prime } = \\left\\{ \\begin{array} { l l } { Q _ { i , j } } & { j \\leq d } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. } \\\\ & { K _ { i , j } ^ { \\prime } = \\left\\{ \\begin{array} { l l } { K _ { i , j } } & { j \\leq d } \\\\ { 0 } & { \\mathrm { o t h e r w i s e } } \\end{array} \\right. } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 874, + 233, + 946, + 233, + 946, + 267, + 874, + 267 + ], + "score": 0.92, + "latex": "\\mathscr { j } _ { \\neq \\varphi ( i ) }" + }, + { + "category_id": 14, + "poly": [ + 770, + 369, + 930, + 369, + 930, + 404, + 770, + 404 + ], + "score": 0.92, + "latex": "\\alpha _ { i , \\varphi ( i ) } > \\alpha _ { i , j }" + }, + { + "category_id": 13, + "poly": [ + 888, + 1313, + 972, + 1313, + 972, + 1346, + 888, + 1346 + ], + "score": 0.92, + "latex": "q \\in \\mathbb { R } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 711, + 531, + 891, + 531, + 891, + 564, + 711, + 564 + ], + "score": 0.91, + "latex": "K \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k } }" + }, + { + "category_id": 13, + "poly": [ + 1198, + 564, + 1399, + 564, + 1399, + 594, + 1198, + 594 + ], + "score": 0.91, + "latex": "K ^ { \\prime } \\in \\mathbb { R } ^ { d } \\times \\mathbb { R } ^ { d _ { k e y } }" + }, + { + "category_id": 13, + "poly": [ + 947, + 535, + 1034, + 535, + 1034, + 565, + 947, + 565 + ], + "score": 0.9, + "latex": "d _ { k } > d" + }, + { + "category_id": 13, + "poly": [ + 474, + 459, + 544, + 459, + 544, + 486, + 474, + 486 + ], + "score": 0.9, + "latex": "d > n" + }, + { + "category_id": 13, + "poly": [ + 925, + 1380, + 1009, + 1380, + 1009, + 1411, + 925, + 1411 + ], + "score": 0.89, + "latex": "\\mathbb { K } = \\mathbb { Q }" + }, + { + "category_id": 13, + "poly": [ + 1115, + 1316, + 1149, + 1316, + 1149, + 1346, + 1115, + 1346 + ], + "score": 0.89, + "latex": "\\pmb { h } _ { 1 }" + }, + { + "category_id": 14, + "poly": [ + 767, + 1418, + 932, + 1418, + 932, + 1536, + 767, + 1536 + ], + "score": 0.88, + "latex": "\\begin{array} { l } { { \\pmb { h } } _ { 0 } = { \\pmb { \\hat { x } } } } \\\\ { { \\pmb { h } } _ { 1 } = { \\pmb { \\hat { x } } } + { \\pmb { \\hat { y } } } } \\\\ { { \\pmb { h } } _ { 2 } = { \\pmb { \\hat { x } } } + 2 { \\pmb { \\hat { y } } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 497, + 489, + 524, + 489, + 524, + 520, + 497, + 520 + ], + "score": 0.86, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 443, + 1545, + 464, + 1545, + 464, + 1576, + 443, + 1576 + ], + "score": 0.85, + "latex": "\\hat { \\pmb { y } }" + }, + { + "category_id": 13, + "poly": [ + 575, + 489, + 606, + 489, + 606, + 515, + 575, + 515 + ], + "score": 0.85, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 1288, + 536, + 1316, + 536, + 1316, + 565, + 1288, + 565 + ], + "score": 0.83, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 371, + 1545, + 393, + 1545, + 393, + 1571, + 371, + 1571 + ], + "score": 0.83, + "latex": "\\hat { \\pmb x }" + }, + { + "category_id": 13, + "poly": [ + 1257, + 934, + 1279, + 934, + 1279, + 961, + 1257, + 961 + ], + "score": 0.8, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 397, + 1772, + 416, + 1772, + 416, + 1799, + 397, + 1799 + ], + "score": 0.8, + "latex": "\\pmb q" + }, + { + "category_id": 13, + "poly": [ + 378, + 1741, + 394, + 1741, + 394, + 1763, + 378, + 1763 + ], + "score": 0.79, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 1370, + 535, + 1401, + 535, + 1401, + 562, + 1370, + 562 + ], + "score": 0.79, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 718, + 234, + 731, + 234, + 731, + 258, + 718, + 258 + ], + "score": 0.78, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 1369, + 1650, + 1384, + 1650, + 1384, + 1670, + 1369, + 1670 + ], + "score": 0.75, + "latex": "s" + }, + { + "category_id": 14, + "poly": [ + 714, + 1056, + 990, + 1056, + 990, + 1257, + 714, + 1257 + ], + "score": 0.59, + "latex": "\\prod _ { \\substack { h _ { 0 } = \\hat { x } } } ^ { y } h _ { 2 } = \\hat { x } + 2 \\hat { y }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 869.0, + 904.0, + 869.0, + 904.0, + 906.0, + 296.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 809.0, + 1401.0, + 809.0, + 1401.0, + 831.0, + 1380.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1283.0, + 1405.0, + 1283.0, + 1405.0, + 1318.0, + 294.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1314.0, + 887.0, + 1314.0, + 887.0, + 1349.0, + 293.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1314.0, + 1114.0, + 1314.0, + 1114.0, + 1349.0, + 973.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 1314.0, + 1159.0, + 1314.0, + 1159.0, + 1349.0, + 1150.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1052.0, + 713.0, + 1052.0, + 713.0, + 1086.0, + 710.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 928.0, + 1256.0, + 928.0, + 1256.0, + 962.0, + 295.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1280.0, + 928.0, + 1405.0, + 928.0, + 1405.0, + 962.0, + 1280.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 959.0, + 1405.0, + 959.0, + 1405.0, + 993.0, + 296.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 990.0, + 685.0, + 990.0, + 685.0, + 1024.0, + 296.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 527.0, + 478.0, + 527.0, + 478.0, + 569.0, + 292.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 527.0, + 710.0, + 527.0, + 710.0, + 569.0, + 656.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 527.0, + 946.0, + 527.0, + 946.0, + 569.0, + 892.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 527.0, + 1287.0, + 527.0, + 1287.0, + 569.0, + 1035.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 527.0, + 1369.0, + 527.0, + 1369.0, + 569.0, + 1317.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 527.0, + 1405.0, + 527.0, + 1405.0, + 569.0, + 1402.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 563.0, + 948.0, + 563.0, + 948.0, + 597.0, + 292.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 563.0, + 1197.0, + 563.0, + 1197.0, + 597.0, + 1145.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 563.0, + 1404.0, + 563.0, + 1404.0, + 597.0, + 1400.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 593.0, + 427.0, + 593.0, + 427.0, + 629.0, + 293.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 448.0, + 473.0, + 448.0, + 473.0, + 496.0, + 290.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 448.0, + 982.0, + 448.0, + 982.0, + 496.0, + 545.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 448.0, + 1409.0, + 448.0, + 1409.0, + 496.0, + 1161.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 488.0, + 496.0, + 488.0, + 496.0, + 519.0, + 296.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 488.0, + 574.0, + 488.0, + 574.0, + 519.0, + 525.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 488.0, + 616.0, + 488.0, + 616.0, + 519.0, + 607.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1535.0, + 370.0, + 1535.0, + 370.0, + 1583.0, + 291.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1535.0, + 442.0, + 1535.0, + 442.0, + 1583.0, + 394.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 1535.0, + 916.0, + 1535.0, + 916.0, + 1583.0, + 465.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1535.0, + 1186.0, + 1535.0, + 1186.0, + 1583.0, + 1065.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 1535.0, + 1406.0, + 1535.0, + 1406.0, + 1583.0, + 1335.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1573.0, + 533.0, + 1573.0, + 533.0, + 1608.0, + 295.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1732.0, + 377.0, + 1732.0, + 377.0, + 1771.0, + 295.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1732.0, + 1404.0, + 1732.0, + 1404.0, + 1771.0, + 395.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1766.0, + 396.0, + 1766.0, + 396.0, + 1802.0, + 296.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1766.0, + 466.0, + 1766.0, + 466.0, + 1802.0, + 417.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 1766.0, + 651.0, + 1766.0, + 651.0, + 1802.0, + 615.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 219.0, + 717.0, + 219.0, + 717.0, + 275.0, + 292.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 219.0, + 873.0, + 219.0, + 873.0, + 275.0, + 732.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 219.0, + 962.0, + 219.0, + 962.0, + 275.0, + 947.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 320.0, + 876.0, + 320.0, + 876.0, + 360.0, + 293.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 406.0, + 886.0, + 406.0, + 886.0, + 450.0, + 293.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1373.0, + 924.0, + 1373.0, + 924.0, + 1419.0, + 295.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1373.0, + 1138.0, + 1373.0, + 1138.0, + 1419.0, + 1010.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1638.0, + 340.0, + 1638.0, + 340.0, + 1681.0, + 294.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1638.0, + 1368.0, + 1638.0, + 1368.0, + 1681.0, + 417.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1385.0, + 1638.0, + 1398.0, + 1638.0, + 1398.0, + 1681.0, + 1385.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1283.0, + 1405.0, + 1283.0, + 1405.0, + 1318.0, + 294.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1314.0, + 887.0, + 1314.0, + 887.0, + 1349.0, + 293.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1314.0, + 1114.0, + 1314.0, + 1114.0, + 1349.0, + 973.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 1314.0, + 1159.0, + 1314.0, + 1159.0, + 1349.0, + 1150.0, + 1349.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1608, + 1405, + 1608, + 1405, + 1703, + 297, + 1703 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1501, + 1405, + 1501, + 1405, + 1593, + 298, + 1593 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 224, + 1404, + 224, + 1404, + 311, + 298, + 311 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 293, + 501, + 1400, + 501, + 1400, + 565, + 293, + 565 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 297, + 1794, + 1403, + 1794, + 1403, + 1860, + 297, + 1860 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 600, + 819, + 1097, + 819, + 1097, + 863, + 600, + 863 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 729, + 450, + 970, + 450, + 970, + 489, + 729, + 489 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 297, + 1970, + 1402, + 1970, + 1402, + 2034, + 297, + 2034 + ], + "score": 0.937 + }, + { + "category_id": 1, + "poly": [ + 298, + 577, + 1403, + 577, + 1403, + 640, + 298, + 640 + ], + "score": 0.935 + }, + { + "category_id": 8, + "poly": [ + 730, + 324, + 970, + 324, + 970, + 363, + 730, + 363 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 294, + 768, + 858, + 768, + 858, + 804, + 294, + 804 + ], + "score": 0.931 + }, + { + "category_id": 8, + "poly": [ + 598, + 1138, + 1078, + 1138, + 1078, + 1183, + 598, + 1183 + ], + "score": 0.926 + }, + { + "category_id": 8, + "poly": [ + 728, + 1324, + 971, + 1324, + 971, + 1363, + 728, + 1363 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 298, + 392, + 710, + 392, + 710, + 438, + 298, + 438 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 300, + 1393, + 810, + 1393, + 810, + 1437, + 300, + 1437 + ], + "score": 0.925 + }, + { + "category_id": 8, + "poly": [ + 588, + 701, + 1109, + 701, + 1109, + 741, + 588, + 741 + ], + "score": 0.923 + }, + { + "category_id": 8, + "poly": [ + 601, + 1085, + 1070, + 1085, + 1070, + 1130, + 601, + 1130 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 296, + 873, + 1039, + 873, + 1039, + 907, + 296, + 907 + ], + "score": 0.921 + }, + { + "category_id": 8, + "poly": [ + 600, + 1188, + 1185, + 1188, + 1185, + 1237, + 600, + 1237 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 297, + 655, + 932, + 655, + 932, + 688, + 297, + 688 + ], + "score": 0.919 + }, + { + "category_id": 8, + "poly": [ + 509, + 966, + 1052, + 966, + 1052, + 1026, + 509, + 1026 + ], + "score": 0.919 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 816, + 73, + 816, + 106, + 297, + 106 + ], + "score": 0.915 + }, + { + "category_id": 0, + "poly": [ + 299, + 1738, + 671, + 1738, + 671, + 1771, + 299, + 1771 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 297, + 921, + 417, + 921, + 417, + 952, + 297, + 952 + ], + "score": 0.908 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1452, + 1400, + 1452, + 1400, + 1484, + 1352, + 1484 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 300, + 1901, + 758, + 1901, + 758, + 1939, + 300, + 1939 + ], + "score": 0.905 + }, + { + "category_id": 8, + "poly": [ + 602, + 1033, + 1041, + 1033, + 1041, + 1078, + 602, + 1078 + ], + "score": 0.905 + }, + { + "category_id": 8, + "poly": [ + 728, + 1450, + 971, + 1450, + 971, + 1489, + 728, + 1489 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1352, + 452, + 1400, + 452, + 1400, + 483, + 1352, + 483 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1352, + 327, + 1400, + 327, + 1400, + 358, + 1352, + 358 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1326, + 1400, + 1326, + 1400, + 1358, + 1351, + 1358 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1351, + 826, + 1401, + 826, + 1401, + 857, + 1351, + 857 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1351, + 705, + 1400, + 705, + 1400, + 735, + 1351, + 735 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1043, + 1400, + 1043, + 1400, + 1073, + 1352, + 1073 + ], + "score": 0.884 + }, + { + "category_id": 9, + "poly": [ + 1352, + 982, + 1400, + 982, + 1400, + 1013, + 1352, + 1013 + ], + "score": 0.884 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1199, + 1401, + 1199, + 1401, + 1229, + 1351, + 1229 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 297, + 1266, + 685, + 1266, + 685, + 1311, + 297, + 1311 + ], + "score": 0.878 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1148, + 1400, + 1148, + 1400, + 1177, + 1352, + 1177 + ], + "score": 0.873 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1094, + 1400, + 1094, + 1400, + 1125, + 1352, + 1125 + ], + "score": 0.872 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2113, + 836, + 2113 + ], + "score": 0.855 + }, + { + "category_id": 14, + "poly": [ + 507, + 960, + 1189, + 960, + 1189, + 1244, + 507, + 1244 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { s \\left( \\boldsymbol { q } , \\beta \\right) = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\left( \\beta h _ { 1 } + \\left( 1 - \\beta \\right) h _ { 0 } \\right) ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta h _ { 1 } ^ { \\top } \\boldsymbol { K } + \\left( 1 - \\beta \\right) h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta h _ { 1 } ^ { \\top } \\boldsymbol { K } + h _ { 0 } ^ { \\top } \\boldsymbol { K } - \\beta h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( \\beta \\left( h _ { 1 } ^ { \\top } \\boldsymbol { K } - h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) + h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\\\ & { \\quad \\quad = \\beta \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( h _ { 1 } ^ { \\top } \\boldsymbol { K } - h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } + \\left( \\boldsymbol { q } ^ { \\top } \\boldsymbol { Q } \\right) \\left( h _ { 0 } ^ { \\top } \\boldsymbol { K } \\right) ^ { \\top } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 584, + 770, + 734, + 770, + 734, + 804, + 584, + 804 + ], + "score": 0.94, + "latex": "\\beta \\in \\{ 0 , \\lambda , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 569, + 576, + 720, + 576, + 720, + 611, + 569, + 611 + ], + "score": 0.93, + "latex": "\\boldsymbol { h } _ { 0 } , \\boldsymbol { h } _ { 1 } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 663, + 532, + 763, + 532, + 763, + 565, + 663, + 565 + ], + "score": 0.92, + "latex": "e \\left( \\pmb q , \\pmb h _ { 1 } \\right)" + }, + { + "category_id": 13, + "poly": [ + 379, + 1640, + 462, + 1640, + 462, + 1671, + 379, + 1671 + ], + "score": 0.92, + "latex": "\\mathbb { K } = \\mathbb { Q }" + }, + { + "category_id": 13, + "poly": [ + 460, + 772, + 534, + 772, + 534, + 803, + 460, + 803 + ], + "score": 0.92, + "latex": "\\pmb q \\in \\mathbb { Q }" + }, + { + "category_id": 13, + "poly": [ + 1300, + 1501, + 1403, + 1501, + 1403, + 1535, + 1300, + 1535 + ], + "score": 0.92, + "latex": "\\mathbf { \\nabla } _ { \\boldsymbol { \\mathcal { C } } } \\left( \\mathbf { \\nabla } _ { \\mathbf { \\boldsymbol { q } } } , \\boldsymbol { h } _ { \\lambda } \\right)" + }, + { + "category_id": 14, + "poly": [ + 585, + 700, + 1113, + 700, + 1113, + 740, + 585, + 740 + ], + "score": 0.92, + "latex": "\\mathbb { K } = \\mathbb { Q } = \\{ \\beta \\pmb { h } _ { 1 } + \\left( 1 - \\beta \\right) \\pmb { h } _ { 0 } \\ \\lvert \\ \\beta \\in \\{ 0 , \\lambda , 1 \\} \\}" + }, + { + "category_id": 14, + "poly": [ + 729, + 449, + 971, + 449, + 971, + 488, + 729, + 488 + ], + "score": 0.91, + "latex": "e \\left( \\pmb { q } , \\pmb { h } _ { 0 } \\right) \\geq e \\left( \\pmb { q } , \\pmb { h } _ { 1 } \\right)" + }, + { + "category_id": 13, + "poly": [ + 451, + 392, + 701, + 392, + 701, + 438, + 451, + 438 + ], + "score": 0.9, + "latex": "\\left( { { { q } ^ { \\top } } { { Q } } } \\right) \\left( { \\hat { { { y } } } ^ { \\top } } K \\right) ^ { \\top } \\le 0" + }, + { + "category_id": 13, + "poly": [ + 459, + 223, + 656, + 223, + 656, + 269, + 459, + 269 + ], + "score": 0.9, + "latex": "\\left( { { \\pmb q } ^ { \\top } } { \\pmb Q } \\right) \\left( \\hat { { \\pmb x } } ^ { \\top } { \\pmb K } \\right) ^ { \\top }" + }, + { + "category_id": 14, + "poly": [ + 729, + 323, + 971, + 323, + 971, + 362, + 729, + 362 + ], + "score": 0.9, + "latex": "e \\left( \\pmb { q } , \\pmb { h } _ { 2 } \\right) > e \\left( \\pmb { q } , \\pmb { h } _ { 1 } \\right)" + }, + { + "category_id": 14, + "poly": [ + 601, + 816, + 1102, + 816, + 1102, + 861, + 601, + 861 + ], + "score": 0.9, + "latex": "s \\left( \\mathbf { q } , \\beta \\right) = e \\left( \\mathbf { q } , \\left( \\beta \\pmb { h } _ { 1 } + \\left( 1 - \\beta \\right) \\pmb { h } _ { 0 } \\right) \\right) \\cdot \\sqrt { d _ { k } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 610, + 418, + 610, + 418, + 639, + 297, + 639 + ], + "score": 0.9, + "latex": "0 < \\lambda < 1" + }, + { + "category_id": 13, + "poly": [ + 776, + 266, + 1026, + 266, + 1026, + 313, + 776, + 313 + ], + "score": 0.9, + "latex": "\\left( { { \\pmb q } ^ { \\top } } { \\pmb Q } \\right) \\left( \\hat { \\pmb y } ^ { \\top } { \\pmb K } \\right) ^ { \\top } > 0" + }, + { + "category_id": 14, + "poly": [ + 728, + 1448, + 971, + 1448, + 971, + 1489, + 728, + 1489 + ], + "score": 0.9, + "latex": "e \\left( \\pmb { q } , \\pmb { h } _ { 0 } \\right) \\geq e \\left( \\pmb { q } , \\pmb { h } _ { \\lambda } \\right)" + }, + { + "category_id": 13, + "poly": [ + 630, + 1503, + 667, + 1503, + 667, + 1532, + 630, + 1532 + ], + "score": 0.89, + "latex": "h _ { \\lambda }" + }, + { + "category_id": 13, + "poly": [ + 416, + 533, + 451, + 533, + 451, + 563, + 416, + 563 + ], + "score": 0.88, + "latex": "\\pmb { h } _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1057, + 1503, + 1093, + 1503, + 1093, + 1532, + 1057, + 1532 + ], + "score": 0.88, + "latex": "\\mathbf { \\delta } _ { h _ { \\lambda } }" + }, + { + "category_id": 13, + "poly": [ + 514, + 269, + 735, + 269, + 735, + 313, + 514, + 313 + ], + "score": 0.88, + "latex": "\\beta \\left( \\pmb { q } ^ { \\top } \\pmb { Q } \\right) \\left( \\hat { \\pmb { y } } ^ { \\top } \\pmb { K } \\right) ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 1211, + 234, + 1247, + 234, + 1247, + 268, + 1211, + 268 + ], + "score": 0.88, + "latex": "h _ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 630, + 503, + 664, + 503, + 664, + 532, + 630, + 532 + ], + "score": 0.87, + "latex": "\\pmb { h } _ { 1 }" + }, + { + "category_id": 14, + "poly": [ + 728, + 1323, + 972, + 1323, + 972, + 1362, + 728, + 1362 + ], + "score": 0.87, + "latex": "e \\left( \\pmb q , \\pmb h _ { 1 } \\right) > e \\left( \\pmb q , \\pmb h _ { \\lambda } \\right)" + }, + { + "category_id": 13, + "poly": [ + 451, + 1391, + 803, + 1391, + 803, + 1439, + 451, + 1439 + ], + "score": 0.87, + "latex": "\\left( { \\pmb q } ^ { \\top } { \\pmb Q } \\right) \\left( { \\pmb h } _ { 1 } ^ { \\top } { \\pmb K } - { \\pmb h } _ { 0 } ^ { \\top } { \\pmb K } \\right) ^ { \\top } \\le 0 ," + }, + { + "category_id": 13, + "poly": [ + 1051, + 1537, + 1072, + 1537, + 1072, + 1564, + 1051, + 1564 + ], + "score": 0.84, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 1056, + 1645, + 1078, + 1645, + 1078, + 1671, + 1056, + 1671 + ], + "score": 0.83, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 890, + 234, + 910, + 234, + 910, + 265, + 890, + 265 + ], + "score": 0.82, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 1097, + 580, + 1117, + 580, + 1117, + 606, + 1097, + 606 + ], + "score": 0.81, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 323, + 1265, + 678, + 1265, + 678, + 1313, + 323, + 1313 + ], + "score": 0.79, + "latex": "\\left( { { q } ^ { \\top } } Q \\right) \\left( h _ { 1 } ^ { \\top } K - h _ { 0 } ^ { \\top } K \\right) ^ { \\top } > 0 ." + }, + { + "category_id": 13, + "poly": [ + 377, + 881, + 393, + 881, + 393, + 902, + 377, + 902 + ], + "score": 0.77, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 516, + 507, + 535, + 507, + 535, + 533, + 516, + 533 + ], + "score": 0.75, + "latex": "\\pmb q" + }, + { + "category_id": 13, + "poly": [ + 516, + 1507, + 535, + 1507, + 535, + 1534, + 516, + 1534 + ], + "score": 0.75, + "latex": "\\pmb q" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1734.0, + 675.0, + 1734.0, + 675.0, + 1775.0, + 294.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1897.0, + 764.0, + 1897.0, + 764.0, + 1944.0, + 292.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2122.0, + 832.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1608.0, + 1406.0, + 1608.0, + 1406.0, + 1646.0, + 294.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1638.0, + 378.0, + 1638.0, + 378.0, + 1675.0, + 294.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1638.0, + 1055.0, + 1638.0, + 1055.0, + 1675.0, + 463.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 1638.0, + 1405.0, + 1638.0, + 1405.0, + 1675.0, + 1079.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1669.0, + 1105.0, + 1669.0, + 1105.0, + 1704.0, + 294.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1498.0, + 515.0, + 1498.0, + 515.0, + 1537.0, + 293.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1498.0, + 629.0, + 1498.0, + 629.0, + 1537.0, + 536.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 1498.0, + 1056.0, + 1498.0, + 1056.0, + 1537.0, + 668.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 1498.0, + 1299.0, + 1498.0, + 1299.0, + 1537.0, + 1094.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1530.0, + 1050.0, + 1530.0, + 1050.0, + 1566.0, + 293.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1530.0, + 1405.0, + 1530.0, + 1405.0, + 1566.0, + 1073.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1563.0, + 459.0, + 1563.0, + 459.0, + 1593.0, + 296.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 220.0, + 458.0, + 220.0, + 458.0, + 273.0, + 291.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 220.0, + 889.0, + 220.0, + 889.0, + 273.0, + 657.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 220.0, + 1210.0, + 220.0, + 1210.0, + 273.0, + 911.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 220.0, + 1409.0, + 220.0, + 1409.0, + 273.0, + 1248.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 265.0, + 513.0, + 265.0, + 513.0, + 313.0, + 288.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 265.0, + 775.0, + 265.0, + 775.0, + 313.0, + 736.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 265.0, + 1103.0, + 265.0, + 1103.0, + 313.0, + 1027.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 500.0, + 515.0, + 500.0, + 515.0, + 536.0, + 295.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 500.0, + 629.0, + 500.0, + 629.0, + 536.0, + 536.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 500.0, + 1406.0, + 500.0, + 1406.0, + 536.0, + 665.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 532.0, + 415.0, + 532.0, + 415.0, + 568.0, + 295.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 532.0, + 662.0, + 532.0, + 662.0, + 568.0, + 452.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 532.0, + 1176.0, + 532.0, + 1176.0, + 568.0, + 764.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1795.0, + 1405.0, + 1795.0, + 1405.0, + 1831.0, + 296.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1826.0, + 938.0, + 1826.0, + 938.0, + 1862.0, + 295.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1967.0, + 1406.0, + 1967.0, + 1406.0, + 2008.0, + 292.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 543.0, + 2002.0, + 543.0, + 2035.0, + 293.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 574.0, + 568.0, + 574.0, + 568.0, + 614.0, + 292.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 574.0, + 1096.0, + 574.0, + 1096.0, + 614.0, + 721.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 574.0, + 1408.0, + 574.0, + 1408.0, + 614.0, + 1118.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 606.0, + 296.0, + 606.0, + 296.0, + 643.0, + 292.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 606.0, + 431.0, + 606.0, + 431.0, + 643.0, + 419.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 768.0, + 459.0, + 768.0, + 459.0, + 808.0, + 295.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 768.0, + 583.0, + 768.0, + 583.0, + 808.0, + 535.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 768.0, + 859.0, + 768.0, + 859.0, + 808.0, + 735.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 391.0, + 450.0, + 391.0, + 450.0, + 440.0, + 288.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 391.0, + 717.0, + 391.0, + 717.0, + 440.0, + 702.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1389.0, + 450.0, + 1389.0, + 450.0, + 1440.0, + 293.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1389.0, + 815.0, + 1389.0, + 815.0, + 1440.0, + 804.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 869.0, + 376.0, + 869.0, + 376.0, + 912.0, + 294.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 869.0, + 1040.0, + 869.0, + 1040.0, + 912.0, + 394.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 655.0, + 933.0, + 655.0, + 933.0, + 691.0, + 298.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 917.0, + 420.0, + 917.0, + 420.0, + 956.0, + 295.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1264.0, + 322.0, + 1264.0, + 322.0, + 1313.0, + 288.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1264.0, + 692.0, + 1264.0, + 692.0, + 1313.0, + 679.0, + 1313.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 606, + 1122, + 1093, + 1122, + 1093, + 1350, + 606, + 1350 + ], + "score": 0.975, + "html": "
ModelSeenProjUnseenProj
No- AttentionGCN+87.2±1.581.4±2.3
GIN+87.1±0.181.1±0.9
AttentionGAT+86.9±0.781.2±0.9
DPGAT88.0±0.881.5±1.2
GATv288.0±1.182.8±1.7
" + }, + { + "category_id": 3, + "poly": [ + 295, + 289, + 1397, + 289, + 1397, + 797, + 295, + 797 + ], + "score": 0.971 + }, + { + "category_id": 5, + "poly": [ + 313, + 1626, + 1175, + 1626, + 1175, + 1964, + 313, + 1964 + ], + "score": 0.971, + "html": "
ModelAttn. Headsogbn-arxivogbn-productsogbn-mag
GCN+071.74±0.2978.97±0.3330.43±0.25
GraphSAGE†071.49±0.2778.70±0.3631.53±0.15
GAT171.59±0.3879.04±1.5432.20±1.46
871.54±0.3077.23±2.3731.75±1.60
DPGAT171.52±0.1776.49±0.7832.77±0.80
871.48±0.2673.53±0.4727.74±9.97
GATv2 (this work)171.78±0.1880.63±0.7032.61±0.44
871.87±0.2578.46±2.4532.52±0.39
" + }, + { + "category_id": 5, + "poly": [ + 1200, + 1628, + 1386, + 1628, + 1386, + 1964, + 1200, + 1964 + ], + "score": 0.955, + "html": "
ogbn-proteins
72.51±0.3577.68±0.20
70.77±5.7978.63±1.62
63.47±2.7972.88±0.59
77.23±3.3279.52±0.55
" + }, + { + "category_id": 4, + "poly": [ + 294, + 820, + 1406, + 820, + 1406, + 886, + 294, + 886 + ], + "score": 0.946 + }, + { + "category_id": 6, + "poly": [ + 295, + 1032, + 1404, + 1032, + 1404, + 1099, + 295, + 1099 + ], + "score": 0.937 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 105, + 299, + 105 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 864, + 2087, + 864, + 2113, + 835, + 2113 + ], + "score": 0.864 + }, + { + "category_id": 6, + "poly": [ + 293, + 1494, + 1406, + 1494, + 1406, + 1561, + 293, + 1561 + ], + "score": 0.855 + }, + { + "category_id": 1, + "poly": [ + 1275, + 1587, + 1310, + 1587, + 1310, + 1615, + 1275, + 1615 + ], + "score": 0.281 + }, + { + "category_id": 1, + "poly": [ + 730, + 1588, + 763, + 1588, + 763, + 1615, + 730, + 1615 + ], + "score": 0.13 + }, + { + "category_id": 13, + "poly": [ + 743, + 1067, + 761, + 1067, + 761, + 1099, + 743, + 1099 + ], + "score": 0.4, + "latex": "\\dagger" + }, + { + "category_id": 13, + "poly": [ + 885, + 1528, + 925, + 1528, + 925, + 1559, + 885, + 1559 + ], + "score": 0.4, + "latex": "\\dagger -" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 292.0, + 368.0, + 292.0, + 368.0, + 323.0, + 333.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 306.0, + 810.0, + 306.0, + 810.0, + 340.0, + 684.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 306.0, + 1362.0, + 306.0, + 1362.0, + 340.0, + 1239.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 344.0, + 716.0, + 344.0, + 716.0, + 364.0, + 691.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 334.0, + 805.0, + 334.0, + 805.0, + 372.0, + 717.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 326.0, + 934.0, + 326.0, + 934.0, + 363.0, + 912.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 345.0, + 1262.0, + 345.0, + 1262.0, + 362.0, + 1244.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 335.0, + 1356.0, + 335.0, + 1356.0, + 369.0, + 1270.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 364.0, + 781.0, + 364.0, + 781.0, + 400.0, + 686.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 358.0, + 931.0, + 358.0, + 931.0, + 399.0, + 881.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 364.0, + 1333.0, + 364.0, + 1333.0, + 400.0, + 1239.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 397.0, + 370.0, + 397.0, + 370.0, + 538.0, + 294.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 445.0, + 468.0, + 445.0, + 468.0, + 454.0, + 455.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 431.0, + 884.0, + 431.0, + 884.0, + 539.0, + 848.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 404.0, + 977.0, + 404.0, + 977.0, + 466.0, + 959.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 444.0, + 1020.0, + 444.0, + 1020.0, + 462.0, + 1006.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 413.0, + 1071.0, + 413.0, + 1071.0, + 463.0, + 1048.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 424.0, + 1116.0, + 424.0, + 1116.0, + 444.0, + 1094.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 432.0, + 1162.0, + 432.0, + 1162.0, + 451.0, + 1140.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 410.0, + 1207.0, + 410.0, + 1207.0, + 453.0, + 1185.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 413.0, + 1253.0, + 413.0, + 1253.0, + 447.0, + 1232.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 418.0, + 1301.0, + 418.0, + 1301.0, + 459.0, + 1278.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 440.0, + 1343.0, + 440.0, + 1343.0, + 467.0, + 1324.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 415.0, + 1387.0, + 415.0, + 1387.0, + 432.0, + 1372.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 464.0, + 563.0, + 464.0, + 563.0, + 481.0, + 547.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 461.0, + 1113.0, + 461.0, + 1113.0, + 476.0, + 1098.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 489.0, + 514.0, + 489.0, + 514.0, + 505.0, + 500.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 485.0, + 608.0, + 485.0, + 608.0, + 499.0, + 592.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 481.0, + 921.0, + 481.0, + 921.0, + 516.0, + 881.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 506.0, + 369.0, + 506.0, + 369.0, + 540.0, + 330.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 508.0, + 1019.0, + 508.0, + 1019.0, + 519.0, + 1009.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 515.0, + 558.0, + 515.0, + 558.0, + 525.0, + 548.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 542.0, + 608.0, + 542.0, + 608.0, + 557.0, + 591.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 537.0, + 1068.0, + 537.0, + 1068.0, + 553.0, + 1050.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 553.0, + 656.0, + 553.0, + 656.0, + 572.0, + 634.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 558.0, + 788.0, + 558.0, + 788.0, + 570.0, + 778.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 571.0, + 788.0, + 571.0, + 788.0, + 582.0, + 778.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 570.0, + 1113.0, + 570.0, + 1113.0, + 586.0, + 1098.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 570.0, + 1162.0, + 570.0, + 1162.0, + 590.0, + 1140.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 573.0, + 1205.0, + 573.0, + 1205.0, + 589.0, + 1189.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 581.0, + 701.0, + 581.0, + 701.0, + 604.0, + 681.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 591.0, + 745.0, + 591.0, + 745.0, + 608.0, + 730.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 580.0, + 840.0, + 580.0, + 840.0, + 613.0, + 817.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 591.0, + 1384.0, + 591.0, + 1384.0, + 601.0, + 1374.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 611.0, + 371.0, + 611.0, + 371.0, + 652.0, + 329.0, + 652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 614.0, + 795.0, + 614.0, + 795.0, + 635.0, + 771.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 599.0, + 922.0, + 599.0, + 922.0, + 633.0, + 881.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 595.0, + 1250.0, + 595.0, + 1250.0, + 610.0, + 1233.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1280.0, + 622.0, + 1295.0, + 622.0, + 1295.0, + 637.0, + 1280.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 647.0, + 836.0, + 647.0, + 836.0, + 662.0, + 820.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 677.0, + 383.0, + 677.0, + 383.0, + 706.0, + 358.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 675.0, + 484.0, + 675.0, + 484.0, + 709.0, + 438.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 673.0, + 578.0, + 673.0, + 578.0, + 710.0, + 528.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 673.0, + 671.0, + 673.0, + 671.0, + 710.0, + 619.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 676.0, + 760.0, + 676.0, + 760.0, + 709.0, + 712.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 672.0, + 852.0, + 672.0, + 852.0, + 710.0, + 803.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 677.0, + 932.0, + 677.0, + 932.0, + 705.0, + 909.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 676.0, + 1035.0, + 676.0, + 1035.0, + 709.0, + 990.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 673.0, + 1130.0, + 673.0, + 1130.0, + 710.0, + 1080.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 673.0, + 1222.0, + 673.0, + 1222.0, + 710.0, + 1170.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 673.0, + 1313.0, + 673.0, + 1313.0, + 710.0, + 1262.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 672.0, + 1404.0, + 672.0, + 1404.0, + 710.0, + 1354.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 717.0, + 1237.0, + 717.0, + 1237.0, + 748.0, + 1061.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 764.0, + 680.0, + 764.0, + 680.0, + 797.0, + 515.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 759.0, + 1230.0, + 759.0, + 1230.0, + 804.0, + 1071.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 347.5, + 419.0, + 347.5, + 419.0, + 369.0, + 371.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.25, + 468.0, + 1160.25, + 468.0, + 1160.25, + 477.0, + 1142.25, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.25, + 713.5, + 687.25, + 713.5, + 687.25, + 750.0, + 507.25, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 821.0, + 1405.0, + 821.0, + 1405.0, + 857.0, + 296.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 852.0, + 1369.0, + 852.0, + 1369.0, + 887.0, + 293.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1034.0, + 1404.0, + 1034.0, + 1404.0, + 1067.0, + 297.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1066.0, + 742.0, + 1066.0, + 742.0, + 1102.0, + 295.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1066.0, + 1288.0, + 1066.0, + 1288.0, + 1102.0, + 762.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 70.0, + 817.0, + 70.0, + 817.0, + 110.0, + 294.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1494.0, + 1405.0, + 1494.0, + 1405.0, + 1531.0, + 292.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1526.0, + 884.0, + 1526.0, + 884.0, + 1563.0, + 295.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1526.0, + 1370.0, + 1526.0, + 1370.0, + 1563.0, + 926.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1585.0, + 1315.0, + 1585.0, + 1315.0, + 1622.0, + 1271.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 1585.0, + 767.0, + 1585.0, + 767.0, + 1621.0, + 727.0, + 1621.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 669, + 1863, + 1026, + 1863, + 1026, + 2029, + 669, + 2029 + ], + "score": 0.969, + "html": "
ModelSeenProjUnseenProj
GAT+86.9±0.781.2±0.9
GATv288.0±1.182.8±1.7
p-value0.0480.049
" + }, + { + "category_id": 6, + "poly": [ + 297, + 223, + 1405, + 223, + 1405, + 347, + 297, + 347 + ], + "score": 0.956 + }, + { + "category_id": 4, + "poly": [ + 296, + 1661, + 1401, + 1661, + 1401, + 1754, + 296, + 1754 + ], + "score": 0.949 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.9 + }, + { + "category_id": 6, + "poly": [ + 295, + 1774, + 1403, + 1774, + 1403, + 1840, + 295, + 1840 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 866, + 2088, + 866, + 2114, + 834, + 2114 + ], + "score": 0.844 + }, + { + "category_id": 3, + "poly": [ + 301, + 1161, + 1379, + 1161, + 1379, + 1636, + 301, + 1636 + ], + "score": 0.719 + }, + { + "category_id": 5, + "poly": [ + 322, + 373, + 1376, + 373, + 1376, + 1140, + 322, + 1140 + ], + "score": 0.576, + "html": "
Predicted Property
Model1234567
GCNt3.21±0.064.22±0.451.45±0.011.62±0.042.42±0.1416.38±0.4917.40±3.56
GINt2.64±0.114.67±0.521.42±0.011.50±0.092.27±0.0915.63±1.4012.93±1.81
GAT1h3.08±0.087.82±1.421.79±0.103.96±1.513.58±1.0335.43±29.9116.5±10.65
GAT8h2.68±0.064.65±0.441.48±0.031.53±0.072.31±0.0652.39±42.5814.87±2.88
DPGAT8h2.63±0.094.37±0.131.44±0.071.40±0.032.10±0.0732.59±34.7711.66±1.00
DPGAT1h3.20±0.178.35±0.781.71±0.032.17±0.142.88±0.1225.21±2.8665.79±39.84
GATv21h3.04±0.066.38±0.621.68±0.042.18±0.612.82±0.2520.56±0.7077.13 ±37.93
GATv28h2.65±0.054.28±0.271.41±0.041.47±0.032.29±0.1516.37±0.9714.03±1.39
Predicted Property
Model GCNt8910111213GAT8h
GINt7.82±0.80 5.88±1.018.24±1.259.05±1.217.00±1.51 5.38±0.753.93±0.48 3.53±0.371.02±0.05 1.05±0.11-1.5% -2.3%
GAT1h28.10±16.4518.71±23.365.62±0.8110.80±2.185.37±0.26
GAT8ht7.61±0.4620.80±13.40 6.86±0.5315.80±5.87 7.64±0.926.54±0.364.11±0.273.11±0.14 1.48±0.87+134.1% +0%
DPGAT1h12.93±1.7013.83±2.556.37±0.28
DPGAT8h6.95±0.3213.32±2.39 7.09±0.5914.42±1.95 7.30±0.666.52±0.613.76±0.213.28±1.16 1.18±0.33+77.9% -9.7%
GATv21h5.23±0.36
GATv28h10.19±0.6322.56±17.4615.04±4.5822.94±17.342.46±0.65+91.6%
6.07±0.776.28±0.836.60±0.795.97±0.943.57±0.361.59±0.96-11.5%
" + }, + { + "category_id": 13, + "poly": [ + 888, + 226, + 916, + 226, + 916, + 254, + 888, + 254 + ], + "score": 0.76, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 954, + 287, + 972, + 287, + 972, + 318, + 954, + 318 + ], + "score": 0.71, + "latex": "\\dagger" + }, + { + "category_id": 13, + "poly": [ + 596, + 1778, + 619, + 1778, + 619, + 1804, + 596, + 1804 + ], + "score": 0.37, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 743, + 1808, + 761, + 1808, + 761, + 1839, + 743, + 1839 + ], + "score": 0.26, + "latex": "\\dagger" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 221.0, + 887.0, + 221.0, + 887.0, + 259.0, + 291.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 221.0, + 1405.0, + 221.0, + 1405.0, + 259.0, + 917.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 252.0, + 1407.0, + 252.0, + 1407.0, + 290.0, + 294.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 284.0, + 953.0, + 284.0, + 953.0, + 321.0, + 291.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 284.0, + 1405.0, + 284.0, + 1405.0, + 321.0, + 973.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 313.0, + 548.0, + 313.0, + 548.0, + 349.0, + 294.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1660.0, + 1407.0, + 1660.0, + 1407.0, + 1698.0, + 294.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1692.0, + 1405.0, + 1692.0, + 1405.0, + 1724.0, + 292.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1719.0, + 415.0, + 1719.0, + 415.0, + 1757.0, + 294.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1776.0, + 595.0, + 1776.0, + 595.0, + 1808.0, + 297.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 1776.0, + 1404.0, + 1776.0, + 1404.0, + 1808.0, + 620.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1806.0, + 742.0, + 1806.0, + 742.0, + 1842.0, + 294.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1806.0, + 1291.0, + 1806.0, + 1291.0, + 1842.0, + 762.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2122.0, + 829.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1158.0, + 387.0, + 1158.0, + 387.0, + 1196.0, + 344.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 1174.0, + 790.0, + 1174.0, + 790.0, + 1219.0, + 550.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1174.0, + 1342.0, + 1174.0, + 1342.0, + 1219.0, + 1100.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1222.0, + 576.0, + 1222.0, + 576.0, + 1243.0, + 553.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 1210.0, + 658.0, + 1210.0, + 658.0, + 1252.0, + 582.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1200.0, + 957.0, + 1200.0, + 957.0, + 1266.0, + 890.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1210.0, + 1211.0, + 1210.0, + 1211.0, + 1252.0, + 1099.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1254.0, + 390.0, + 1254.0, + 390.0, + 1399.0, + 301.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1297.0, + 482.0, + 1297.0, + 482.0, + 1316.0, + 462.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1298.0, + 525.0, + 1298.0, + 525.0, + 1317.0, + 505.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1277.0, + 893.0, + 1277.0, + 893.0, + 1400.0, + 853.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1275.0, + 992.0, + 1275.0, + 992.0, + 1322.0, + 973.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1315.0, + 565.0, + 1315.0, + 565.0, + 1335.0, + 547.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1323.0, + 1157.0, + 1323.0, + 1157.0, + 1331.0, + 1147.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1357.0, + 1325.0, + 1368.0, + 1325.0, + 1368.0, + 1334.0, + 1357.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1339.0, + 526.0, + 1339.0, + 526.0, + 1359.0, + 506.0, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1331.0, + 938.0, + 1331.0, + 938.0, + 1371.0, + 893.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1353.0, + 388.0, + 1353.0, + 388.0, + 1395.0, + 339.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1361.0, + 564.0, + 1361.0, + 564.0, + 1378.0, + 548.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 1359.0, + 1029.0, + 1359.0, + 1029.0, + 1371.0, + 1018.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1387.0, + 609.0, + 1387.0, + 609.0, + 1408.0, + 588.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1383.0, + 1076.0, + 1383.0, + 1076.0, + 1402.0, + 1055.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 1399.0, + 653.0, + 1399.0, + 653.0, + 1418.0, + 633.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1403.0, + 774.0, + 1403.0, + 774.0, + 1415.0, + 764.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1413.0, + 1117.0, + 1413.0, + 1117.0, + 1433.0, + 1099.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 1418.0, + 1158.0, + 1418.0, + 1158.0, + 1433.0, + 1141.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 1416.0, + 1203.0, + 1416.0, + 1203.0, + 1437.0, + 1184.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1450.0, + 391.0, + 1450.0, + 391.0, + 1498.0, + 338.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1436.0, + 736.0, + 1436.0, + 736.0, + 1451.0, + 719.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 1437.0, + 815.0, + 1437.0, + 815.0, + 1451.0, + 800.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1437.0, + 939.0, + 1437.0, + 939.0, + 1480.0, + 892.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1438.0, + 1245.0, + 1438.0, + 1245.0, + 1455.0, + 1229.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1463.0, + 1285.0, + 1463.0, + 1285.0, + 1478.0, + 1269.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 1444.0, + 1331.0, + 1444.0, + 1331.0, + 1463.0, + 1310.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 1485.0, + 818.0, + 1485.0, + 818.0, + 1500.0, + 802.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1513.0, + 402.0, + 1513.0, + 402.0, + 1548.0, + 374.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 1512.0, + 499.0, + 1512.0, + 499.0, + 1552.0, + 445.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 1511.0, + 586.0, + 1511.0, + 586.0, + 1552.0, + 528.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 1511.0, + 669.0, + 1511.0, + 669.0, + 1552.0, + 612.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1512.0, + 757.0, + 1512.0, + 757.0, + 1551.0, + 696.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1513.0, + 836.0, + 1513.0, + 836.0, + 1549.0, + 784.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1515.0, + 954.0, + 1515.0, + 954.0, + 1548.0, + 926.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 1515.0, + 1051.0, + 1515.0, + 1051.0, + 1551.0, + 997.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 1511.0, + 1137.0, + 1511.0, + 1137.0, + 1552.0, + 1079.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 1511.0, + 1222.0, + 1511.0, + 1222.0, + 1552.0, + 1164.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1513.0, + 1306.0, + 1513.0, + 1306.0, + 1549.0, + 1251.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 1511.0, + 1389.0, + 1511.0, + 1389.0, + 1551.0, + 1333.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1558.0, + 665.0, + 1558.0, + 665.0, + 1593.0, + 533.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1558.0, + 1215.0, + 1558.0, + 1215.0, + 1593.0, + 1084.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1601.0, + 680.0, + 1601.0, + 680.0, + 1639.0, + 514.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1598.0, + 1231.0, + 1598.0, + 1231.0, + 1646.0, + 1072.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 1202.5, + 414.0, + 1202.5, + 414.0, + 1221.0, + 388.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 1219.5, + 480.0, + 1219.5, + 480.0, + 1273.0, + 414.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1262.0, + 1108.0, + 1262.0, + 1108.0, + 1290.5, + 1017.0, + 1290.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 1263.5, + 1020.0, + 1263.5, + 1020.0, + 1288.0, + 991.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 1273.5, + 1274.0, + 1273.5, + 1274.0, + 1297.5, + 1234.0, + 1297.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1278.5, + 552.0, + 1278.5, + 552.0, + 1304.0, + 513.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.75, + 1325.5, + 756.75, + 1325.5, + 756.75, + 1372.0, + 638.75, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.75, + 1311.5, + 639.75, + 1311.5, + 639.75, + 1338.5, + 594.75, + 1338.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1322.0, + 1105.0, + 1322.0, + 1105.0, + 1346.0, + 1066.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1318.0, + 1239.0, + 1318.0, + 1239.0, + 1343.5, + 1206.0, + 1343.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 1313.0, + 1281.0, + 1313.0, + 1281.0, + 1336.5, + 1245.0, + 1336.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.25, + 1325.0, + 560.25, + 1325.0, + 560.25, + 1350.0, + 528.25, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1347.5, + 638.0, + 1347.5, + 638.0, + 1374.5, + 600.0, + 1374.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1163.75, + 1344.0, + 1197.75, + 1344.0, + 1197.75, + 1369.0, + 1163.75, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1345.5, + 1353.0, + 1345.5, + 1353.0, + 1366.5, + 1325.0, + 1366.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.25, + 1385.5, + 728.25, + 1385.5, + 728.25, + 1419.5, + 706.25, + 1419.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 1367.0, + 689.0, + 1367.0, + 689.0, + 1378.5, + 675.0, + 1378.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 1423.0, + 684.0, + 1423.0, + 684.0, + 1441.0, + 673.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1453.5, + 791.0, + 1453.5, + 791.0, + 1477.5, + 749.0, + 1477.5 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 747, + 333, + 953, + 333, + 953, + 498, + 747, + 498 + ], + "score": 0.966, + "html": "
Model Accuracy
GAT 78.1±0.59
GATv2 78.5±0.38
p-value <0.0001
" + }, + { + "category_id": 6, + "poly": [ + 297, + 1482, + 1399, + 1482, + 1399, + 1546, + 297, + 1546 + ], + "score": 0.942 + }, + { + "category_id": 5, + "poly": [ + 1200, + 751, + 1387, + 751, + 1387, + 914, + 1200, + 914 + ], + "score": 0.936, + "html": "
ogbn-proteins
78.29 ±1.59
78.96±1.19 0.0349
" + }, + { + "category_id": 6, + "poly": [ + 296, + 1033, + 1405, + 1033, + 1405, + 1099, + 296, + 1099 + ], + "score": 0.908 + }, + { + "category_id": 6, + "poly": [ + 364, + 275, + 1330, + 275, + 1330, + 309, + 364, + 309 + ], + "score": 0.906 + }, + { + "category_id": 5, + "poly": [ + 538, + 1125, + 983, + 1125, + 983, + 1360, + 538, + 1360 + ], + "score": 0.903, + "html": "
(a)
Modelogbl-collab w/o val edges w/ val edges
GAT42.24±2.2646.02±4.09
GATv243.82±2.2449.06±2.50
p-value0.00430.0005
" + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.902 + }, + { + "category_id": 5, + "poly": [ + 1097, + 1125, + 1285, + 1125, + 1285, + 1359, + 1097, + 1359 + ], + "score": 0.9, + "html": "
(b)
ogbl-citation2
79.91±0.13
80.20±0.62
0.0075
" + }, + { + "category_id": 6, + "poly": [ + 294, + 617, + 1402, + 617, + 1402, + 682, + 294, + 682 + ], + "score": 0.846 + }, + { + "category_id": 5, + "poly": [ + 344, + 1567, + 1358, + 1567, + 1358, + 1977, + 344, + 1977 + ], + "score": 0.742, + "html": "
Predicted Property
Model1234567
GAT2.74±0.084.73±0.401.47±0.061.53±0.062.44±0.6055.21±42.3325.36±31.42
GATv22.67±0.084.28±0.231.43±0.051.51±0.072.21±0.0816.64± 1.1713.61±1.68
p-value0.0043<0.00010.01380.16910.04870.00010.0516
Predicted Property
Model8910111213
GAT GATv27.36±0.87 6.13±0.596.79±0.86 6.33±0.827.36±0.936.69±0.864.10±0.291.51±0.84
6.37±0.865.95±0.623.66±0.291.09±0.85
p-value<0.00010.04580.00060.0017<0.00010.0621
" + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 863, + 2088, + 863, + 2113, + 834, + 2113 + ], + "score": 0.703 + }, + { + "category_id": 1, + "poly": [ + 1275, + 709, + 1310, + 709, + 1310, + 736, + 1275, + 736 + ], + "score": 0.496 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 863, + 2088, + 863, + 2113, + 834, + 2113 + ], + "score": 0.13 + }, + { + "category_id": 1, + "poly": [ + 294, + 617, + 1402, + 617, + 1402, + 682, + 294, + 682 + ], + "score": 0.094 + }, + { + "category_id": 1, + "poly": [ + 1165, + 1125, + 1201, + 1125, + 1201, + 1154, + 1165, + 1154 + ], + "score": 0.092 + }, + { + "category_id": 13, + "poly": [ + 894, + 1485, + 923, + 1485, + 923, + 1513, + 894, + 1513 + ], + "score": 0.61, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 573, + 1035, + 629, + 1035, + 629, + 1065, + 573, + 1065 + ], + "score": 0.61, + "latex": "\\ @ 5 0" + }, + { + "category_id": 5, + "poly": [ + 446, + 707, + 1042, + 707, + 1042, + 915, + 446, + 915 + ], + "score": 0.887, + "html": "
(a)
Modelogbn-arxivogbn-productsogbn-mag
GAT71.65±0.3879.04±1.5432.36±1.10
GATv271.93±0.3580.63±0.7033.01±0.41
p-value0.0022<0.00010.0018
" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1479.0, + 893.0, + 1479.0, + 893.0, + 1520.0, + 292.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1479.0, + 1404.0, + 1479.0, + 1404.0, + 1520.0, + 924.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1513.0, + 1061.0, + 1513.0, + 1061.0, + 1549.0, + 297.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1036.0, + 572.0, + 1036.0, + 572.0, + 1069.0, + 295.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 1036.0, + 1407.0, + 1036.0, + 1407.0, + 1069.0, + 630.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1066.0, + 1405.0, + 1066.0, + 1405.0, + 1100.0, + 294.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 274.0, + 1331.0, + 274.0, + 1331.0, + 310.0, + 366.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 617.0, + 1404.0, + 617.0, + 1404.0, + 653.0, + 294.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 650.0, + 988.0, + 650.0, + 988.0, + 683.0, + 294.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 707.0, + 1314.0, + 707.0, + 1314.0, + 743.0, + 1271.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 617.0, + 1404.0, + 617.0, + 1404.0, + 653.0, + 294.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 650.0, + 988.0, + 650.0, + 988.0, + 683.0, + 294.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 1123.0, + 1204.0, + 1123.0, + 1204.0, + 1162.0, + 1162.0, + 1162.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1790, + 1405, + 1790, + 1405, + 1920, + 298, + 1920 + ], + "score": 0.976 + }, + { + "category_id": 5, + "poly": [ + 504, + 1372, + 1195, + 1372, + 1195, + 1503, + 504, + 1503 + ], + "score": 0.975, + "html": "
GATGATv2DPGAT
Official2d'+dd'd'+ 2dd'2ddk +dd'
In our experiments2d'+dd'd'+dd'2dd'
" + }, + { + "category_id": 1, + "poly": [ + 297, + 802, + 1405, + 802, + 1405, + 936, + 297, + 936 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1109, + 1404, + 1109, + 1404, + 1226, + 297, + 1226 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 649, + 1404, + 649, + 1404, + 776, + 297, + 776 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 299, + 1940, + 1401, + 1940, + 1401, + 2035, + 299, + 2035 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 296, + 947, + 1401, + 947, + 1401, + 1015, + 296, + 1015 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 299, + 1612, + 1397, + 1612, + 1397, + 1677, + 299, + 1677 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 298, + 1700, + 1401, + 1700, + 1401, + 1767, + 298, + 1767 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 294, + 1025, + 1401, + 1025, + 1401, + 1091, + 294, + 1091 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 297, + 305, + 910, + 305, + 910, + 341, + 297, + 341 + ], + "score": 0.918 + }, + { + "category_id": 8, + "poly": [ + 767, + 422, + 1260, + 422, + 1260, + 460, + 767, + 460 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 816, + 74, + 816, + 106, + 298, + 106 + ], + "score": 0.894 + }, + { + "category_id": 0, + "poly": [ + 299, + 575, + 591, + 575, + 591, + 608, + 299, + 608 + ], + "score": 0.882 + }, + { + "category_id": 0, + "poly": [ + 298, + 1280, + 686, + 1280, + 686, + 1313, + 298, + 1313 + ], + "score": 0.876 + }, + { + "category_id": 8, + "poly": [ + 768, + 375, + 1299, + 375, + 1299, + 416, + 768, + 416 + ], + "score": 0.876 + }, + { + "category_id": 9, + "poly": [ + 1351, + 478, + 1401, + 478, + 1401, + 509, + 1351, + 509 + ], + "score": 0.873 + }, + { + "category_id": 9, + "poly": [ + 1352, + 381, + 1400, + 381, + 1400, + 412, + 1352, + 412 + ], + "score": 0.87 + }, + { + "category_id": 9, + "poly": [ + 1352, + 427, + 1401, + 427, + 1401, + 457, + 1352, + 457 + ], + "score": 0.864 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2113, + 835, + 2113 + ], + "score": 0.855 + }, + { + "category_id": 0, + "poly": [ + 299, + 226, + 694, + 226, + 694, + 262, + 299, + 262 + ], + "score": 0.835 + }, + { + "category_id": 1, + "poly": [ + 306, + 1522, + 1390, + 1522, + 1390, + 1557, + 306, + 1557 + ], + "score": 0.808 + }, + { + "category_id": 8, + "poly": [ + 349, + 468, + 1235, + 468, + 1235, + 521, + 349, + 521 + ], + "score": 0.616 + }, + { + "category_id": 8, + "poly": [ + 772, + 467, + 1233, + 467, + 1233, + 522, + 772, + 522 + ], + "score": 0.609 + }, + { + "category_id": 8, + "poly": [ + 350, + 422, + 1259, + 422, + 1259, + 460, + 350, + 460 + ], + "score": 0.437 + }, + { + "category_id": 1, + "poly": [ + 354, + 475, + 717, + 475, + 717, + 511, + 354, + 511 + ], + "score": 0.162 + }, + { + "category_id": 8, + "poly": [ + 386, + 374, + 1296, + 374, + 1296, + 416, + 386, + 416 + ], + "score": 0.144 + }, + { + "category_id": 1, + "poly": [ + 353, + 424, + 664, + 424, + 664, + 458, + 353, + 458 + ], + "score": 0.136 + }, + { + "category_id": 1, + "poly": [ + 355, + 378, + 702, + 378, + 702, + 413, + 355, + 413 + ], + "score": 0.113 + }, + { + "category_id": 8, + "poly": [ + 349, + 468, + 1234, + 468, + 1234, + 521, + 349, + 521 + ], + "score": 0.092 + }, + { + "category_id": 13, + "poly": [ + 1057, + 1109, + 1127, + 1109, + 1127, + 1151, + 1057, + 1151 + ], + "score": 0.95, + "latex": "\\pmb { h } _ { j } ^ { \\top } \\pmb { K }" + }, + { + "category_id": 13, + "poly": [ + 433, + 1886, + 620, + 1886, + 620, + 1920, + 433, + 1920 + ], + "score": 0.94, + "latex": "W = [ W ^ { \\prime } | | W ^ { \\prime } ]" + }, + { + "category_id": 13, + "poly": [ + 298, + 1191, + 404, + 1191, + 404, + 1226, + 298, + 1226 + ], + "score": 0.93, + "latex": "\\mathcal { O } \\left( | \\mathcal { E } | d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 657, + 836, + 805, + 836, + 805, + 872, + 657, + 872 + ], + "score": 0.93, + "latex": "W _ { 1 } \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }" + }, + { + "category_id": 13, + "poly": [ + 495, + 980, + 854, + 980, + 854, + 1015, + 495, + 1015 + ], + "score": 0.93, + "latex": "W \\cdot [ h _ { i } \\| h _ { j } ] = W _ { 1 } h _ { i } + W _ { 2 } h _ { j }" + }, + { + "category_id": 13, + "poly": [ + 1089, + 1026, + 1194, + 1026, + 1194, + 1061, + 1089, + 1061 + ], + "score": 0.93, + "latex": "\\mathcal { O } \\left( | \\mathcal { E } | d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 298, + 740, + 432, + 740, + 432, + 782, + 298, + 782 + ], + "score": 0.93, + "latex": "a _ { \\{ 1 , 2 \\} } ^ { \\top } W h _ { i }" + }, + { + "category_id": 13, + "poly": [ + 360, + 1155, + 481, + 1155, + 481, + 1190, + 360, + 1190 + ], + "score": 0.93, + "latex": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 335, + 2003, + 424, + 2003, + 424, + 2034, + 335, + 2034 + ], + "score": 0.92, + "latex": "d _ { k } = d ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 941, + 1110, + 1006, + 1110, + 1006, + 1147, + 941, + 1147 + ], + "score": 0.92, + "latex": "\\pmb { h } _ { i } ^ { \\top } \\pmb { Q }" + }, + { + "category_id": 13, + "poly": [ + 764, + 1790, + 916, + 1790, + 916, + 1824, + 764, + 1824 + ], + "score": 0.92, + "latex": "W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times 2 d }" + }, + { + "category_id": 13, + "poly": [ + 382, + 840, + 569, + 840, + 569, + 875, + 382, + 875 + ], + "score": 0.92, + "latex": "W = [ W _ { 1 } \\Vert W _ { 2 } ]" + }, + { + "category_id": 13, + "poly": [ + 1083, + 1887, + 1177, + 1887, + 1177, + 1916, + 1083, + 1916 + ], + "score": 0.92, + "latex": "d ^ { \\prime } + d d ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 298, + 902, + 420, + 902, + 420, + 937, + 298, + 937 + ], + "score": 0.92, + "latex": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 986, + 981, + 1091, + 981, + 1091, + 1014, + 986, + 1014 + ], + "score": 0.92, + "latex": "\\mathcal { O } \\left( \\vert \\mathcal { E } \\vert d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 739, + 1699, + 847, + 1699, + 847, + 1734, + 739, + 1734 + ], + "score": 0.92, + "latex": "\\pmb { a } \\in \\mathbb { R } ^ { 2 d ^ { \\prime } }" + }, + { + "category_id": 13, + "poly": [ + 994, + 873, + 1068, + 873, + 1068, + 906, + 994, + 906 + ], + "score": 0.92, + "latex": "W _ { 2 } h _ { j }" + }, + { + "category_id": 13, + "poly": [ + 602, + 1190, + 812, + 1190, + 812, + 1225, + 602, + 1225 + ], + "score": 0.92, + "latex": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 903, + 1699, + 1046, + 1699, + 1046, + 1734, + 903, + 1734 + ], + "score": 0.92, + "latex": "W \\in \\mathbb { R } ^ { d ^ { \\prime } \\times d }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1825, + 387, + 1825, + 387, + 1860, + 298, + 1860 + ], + "score": 0.92, + "latex": "[ h _ { i } | | h _ { j } ]" + }, + { + "category_id": 13, + "poly": [ + 345, + 981, + 420, + 981, + 420, + 1015, + 345, + 1015 + ], + "score": 0.92, + "latex": "W _ { 2 } h _ { j }" + }, + { + "category_id": 13, + "poly": [ + 958, + 1825, + 1066, + 1825, + 1066, + 1855, + 958, + 1855 + ], + "score": 0.92, + "latex": "d ^ { \\prime } + 2 d d ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1260, + 1973, + 1353, + 1973, + 1353, + 2004, + 1260, + 2004 + ], + "score": 0.91, + "latex": "Q = K" + }, + { + "category_id": 13, + "poly": [ + 1141, + 803, + 1353, + 803, + 1353, + 838, + 1141, + 838 + ], + "score": 0.91, + "latex": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 1180, + 873, + 1271, + 873, + 1271, + 904, + 1180, + 904 + ], + "score": 0.91, + "latex": "i , j \\in \\mathcal { V }" + }, + { + "category_id": 13, + "poly": [ + 1241, + 1113, + 1333, + 1113, + 1333, + 1145, + 1241, + 1145 + ], + "score": 0.91, + "latex": "i , j \\in \\mathcal { V }" + }, + { + "category_id": 13, + "poly": [ + 444, + 681, + 654, + 681, + 654, + 715, + 444, + 715 + ], + "score": 0.91, + "latex": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + | \\mathcal { E } | d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 392, + 1973, + 521, + 1973, + 521, + 2002, + 392, + 2002 + ], + "score": 0.91, + "latex": "2 d d _ { k } + d d ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1167, + 1156, + 1223, + 1156, + 1223, + 1190, + 1167, + 1190 + ], + "score": 0.91, + "latex": "( j , i )" + }, + { + "category_id": 13, + "poly": [ + 1198, + 1704, + 1309, + 1704, + 1309, + 1735, + 1198, + 1735 + ], + "score": 0.9, + "latex": "2 d ^ { \\prime } + d d ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 857, + 835, + 943, + 835, + 943, + 874, + 857, + 874 + ], + "score": 0.9, + "latex": "W _ { 2 } ^ { d ^ { \\prime } \\times d }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1057, + 508, + 1057, + 508, + 1092, + 298, + 1092 + ], + "score": 0.9, + "latex": "\\mathcal { O } \\left( | \\mathcal { V } | d d ^ { \\prime } + \\bar { | } \\mathcal { E } | d ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 899, + 949, + 1059, + 949, + 1059, + 982, + 899, + 982 + ], + "score": 0.9, + "latex": "\\left( W \\cdot \\left[ \\boldsymbol { h } _ { i } \\right| \\left| \\boldsymbol { h } _ { j } \\right] \\right)" + }, + { + "category_id": 13, + "poly": [ + 536, + 950, + 592, + 950, + 592, + 981, + 536, + 981 + ], + "score": 0.9, + "latex": "( j , i )" + }, + { + "category_id": 13, + "poly": [ + 1328, + 949, + 1400, + 949, + 1400, + 981, + 1328, + 981 + ], + "score": 0.89, + "latex": "W _ { 1 } h _ { i }" + }, + { + "category_id": 13, + "poly": [ + 745, + 1944, + 789, + 1944, + 789, + 1973, + 745, + 1973 + ], + "score": 0.89, + "latex": "d d _ { k }" + }, + { + "category_id": 13, + "poly": [ + 543, + 743, + 609, + 743, + 609, + 770, + 543, + 770 + ], + "score": 0.89, + "latex": "i \\in \\mathcal V" + }, + { + "category_id": 13, + "poly": [ + 869, + 874, + 942, + 874, + 942, + 903, + 869, + 903 + ], + "score": 0.88, + "latex": "W _ { 1 } h _ { i }" + }, + { + "category_id": 13, + "poly": [ + 909, + 2003, + 961, + 2003, + 961, + 2030, + 909, + 2030 + ], + "score": 0.88, + "latex": "2 d d ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1022, + 1942, + 1060, + 1942, + 1060, + 1971, + 1022, + 1971 + ], + "score": 0.87, + "latex": "d d ^ { \\prime }" + }, + { + "category_id": 14, + "poly": [ + 765, + 370, + 1299, + 370, + 1299, + 527, + 765, + 527 + ], + "score": 0.86, + "latex": "\\begin{array} { r l } & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = \\mathrm { L e a k y R e L U } ( \\boldsymbol { a } ^ { \\top } \\cdot [ \\boldsymbol { W } \\boldsymbol { h } _ { i } | | \\boldsymbol { W } \\boldsymbol { h } _ { j } | ) } \\\\ & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = \\boldsymbol { a } ^ { \\top } \\mathrm { L e a k y R e L U } ( \\boldsymbol { W } \\cdot [ \\boldsymbol { h } _ { i } | | \\boldsymbol { h } _ { j } ] ) } \\\\ & { e ( \\boldsymbol { h } _ { i } , \\boldsymbol { h } _ { j } ) = ( ( \\boldsymbol { h } _ { i } ^ { \\top } \\boldsymbol { Q } ) \\cdot ( \\boldsymbol { h } _ { j } ^ { \\top } \\boldsymbol { K } ) ^ { \\top } ) / \\sqrt { d ^ { \\prime } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 799, + 1150, + 998, + 1150, + 998, + 1195, + 799, + 1195 + ], + "score": 0.86, + "latex": "\\left( \\pmb { h } _ { i } ^ { \\top } \\pmb { Q } \\right) \\left( \\pmb { h } _ { j } ^ { \\top } \\pmb { K } \\right) ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 990, + 715, + 1024, + 715, + 1024, + 742, + 990, + 742 + ], + "score": 0.85, + "latex": "\\mathbf { a } _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 742, + 716, + 776, + 716, + 776, + 742, + 742, + 742 + ], + "score": 0.85, + "latex": "\\mathbf { \\delta } _ { \\mathbf { a } _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 441, + 1944, + 469, + 1944, + 469, + 1973, + 441, + 1973 + ], + "score": 0.82, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 431, + 875, + 466, + 875, + 466, + 900, + 431, + 900 + ], + "score": 0.79, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 520, + 1944, + 551, + 1944, + 551, + 1970, + 520, + 1970 + ], + "score": 0.79, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 298, + 1973, + 326, + 1973, + 326, + 2001, + 298, + 2001 + ], + "score": 0.77, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 1087, + 712, + 1124, + 712, + 1124, + 740, + 1087, + 740 + ], + "score": 0.75, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 870, + 1034, + 891, + 1034, + 891, + 1055, + 870, + 1055 + ], + "score": 0.74, + "latex": "^ { a }" + }, + { + "category_id": 13, + "poly": [ + 839, + 712, + 876, + 712, + 876, + 740, + 839, + 740 + ], + "score": 0.73, + "latex": "W" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 109.0, + 296.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 575.0, + 596.0, + 575.0, + 596.0, + 611.0, + 294.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1281.0, + 688.0, + 1281.0, + 688.0, + 1314.0, + 295.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 222.0, + 697.0, + 222.0, + 697.0, + 267.0, + 291.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1786.0, + 763.0, + 1786.0, + 763.0, + 1832.0, + 291.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1786.0, + 1408.0, + 1786.0, + 1408.0, + 1832.0, + 917.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 1824.0, + 957.0, + 1824.0, + 957.0, + 1861.0, + 388.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1824.0, + 1410.0, + 1824.0, + 1410.0, + 1861.0, + 1067.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1856.0, + 1405.0, + 1856.0, + 1405.0, + 1891.0, + 293.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1885.0, + 432.0, + 1885.0, + 432.0, + 1920.0, + 292.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1885.0, + 1082.0, + 1885.0, + 1082.0, + 1920.0, + 621.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 1885.0, + 1191.0, + 1885.0, + 1191.0, + 1920.0, + 1178.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 801.0, + 1140.0, + 801.0, + 1140.0, + 840.0, + 294.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 801.0, + 1405.0, + 801.0, + 1405.0, + 840.0, + 1354.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 821.0, + 381.0, + 821.0, + 381.0, + 891.0, + 283.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 821.0, + 656.0, + 821.0, + 656.0, + 891.0, + 570.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 821.0, + 856.0, + 821.0, + 856.0, + 891.0, + 806.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 821.0, + 1406.0, + 821.0, + 1406.0, + 891.0, + 944.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 873.0, + 430.0, + 873.0, + 430.0, + 907.0, + 295.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 873.0, + 868.0, + 873.0, + 868.0, + 907.0, + 467.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 873.0, + 993.0, + 873.0, + 993.0, + 907.0, + 943.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 873.0, + 1179.0, + 873.0, + 1179.0, + 907.0, + 1069.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 873.0, + 1405.0, + 873.0, + 1405.0, + 907.0, + 1272.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 897.0, + 297.0, + 897.0, + 297.0, + 941.0, + 293.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 897.0, + 435.0, + 897.0, + 435.0, + 941.0, + 421.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1108.0, + 940.0, + 1108.0, + 940.0, + 1150.0, + 294.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 1108.0, + 1056.0, + 1108.0, + 1056.0, + 1150.0, + 1007.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1108.0, + 1240.0, + 1108.0, + 1240.0, + 1150.0, + 1128.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1108.0, + 1407.0, + 1108.0, + 1407.0, + 1150.0, + 1334.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1150.0, + 359.0, + 1150.0, + 359.0, + 1195.0, + 291.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1150.0, + 798.0, + 1150.0, + 798.0, + 1195.0, + 482.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1150.0, + 1166.0, + 1150.0, + 1166.0, + 1195.0, + 999.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1150.0, + 1406.0, + 1150.0, + 1406.0, + 1195.0, + 1224.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1189.0, + 601.0, + 1189.0, + 601.0, + 1227.0, + 405.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1189.0, + 824.0, + 1189.0, + 824.0, + 1227.0, + 813.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 646.0, + 1405.0, + 646.0, + 1405.0, + 685.0, + 294.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 681.0, + 443.0, + 681.0, + 443.0, + 715.0, + 292.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 681.0, + 1407.0, + 681.0, + 1407.0, + 715.0, + 655.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 711.0, + 741.0, + 711.0, + 741.0, + 747.0, + 292.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 711.0, + 838.0, + 711.0, + 838.0, + 747.0, + 777.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 711.0, + 989.0, + 711.0, + 989.0, + 747.0, + 877.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 711.0, + 1086.0, + 711.0, + 1086.0, + 747.0, + 1025.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 711.0, + 1405.0, + 711.0, + 1405.0, + 747.0, + 1125.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 741.0, + 297.0, + 741.0, + 297.0, + 786.0, + 294.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 741.0, + 542.0, + 741.0, + 542.0, + 786.0, + 433.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 741.0, + 620.0, + 741.0, + 620.0, + 786.0, + 610.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1941.0, + 440.0, + 1941.0, + 440.0, + 1975.0, + 295.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 1941.0, + 519.0, + 1941.0, + 519.0, + 1975.0, + 470.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1941.0, + 744.0, + 1941.0, + 744.0, + 1975.0, + 552.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1941.0, + 1021.0, + 1941.0, + 1021.0, + 1975.0, + 790.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 1941.0, + 1405.0, + 1941.0, + 1405.0, + 1975.0, + 1061.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1972.0, + 391.0, + 1972.0, + 391.0, + 2006.0, + 327.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1972.0, + 1259.0, + 1972.0, + 1259.0, + 2006.0, + 522.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2006.0, + 1354.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 334.0, + 2002.0, + 334.0, + 2036.0, + 295.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 2002.0, + 908.0, + 2002.0, + 908.0, + 2036.0, + 425.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 2002.0, + 974.0, + 2002.0, + 974.0, + 2036.0, + 962.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 942.0, + 535.0, + 942.0, + 535.0, + 989.0, + 293.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 942.0, + 898.0, + 942.0, + 898.0, + 989.0, + 593.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 942.0, + 1327.0, + 942.0, + 1327.0, + 989.0, + 1060.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 942.0, + 1406.0, + 942.0, + 1406.0, + 989.0, + 1401.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 977.0, + 344.0, + 977.0, + 344.0, + 1018.0, + 295.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 977.0, + 494.0, + 977.0, + 494.0, + 1018.0, + 421.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 977.0, + 985.0, + 977.0, + 985.0, + 1018.0, + 855.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 977.0, + 1105.0, + 977.0, + 1105.0, + 1018.0, + 1092.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1610.0, + 1402.0, + 1610.0, + 1402.0, + 1652.0, + 294.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1647.0, + 885.0, + 1647.0, + 885.0, + 1679.0, + 296.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1693.0, + 738.0, + 1693.0, + 738.0, + 1744.0, + 291.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1693.0, + 902.0, + 1693.0, + 902.0, + 1744.0, + 848.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 1693.0, + 1197.0, + 1693.0, + 1197.0, + 1744.0, + 1047.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 1693.0, + 1407.0, + 1693.0, + 1407.0, + 1744.0, + 1310.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1738.0, + 433.0, + 1738.0, + 433.0, + 1767.0, + 292.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1023.0, + 869.0, + 1023.0, + 869.0, + 1063.0, + 294.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1023.0, + 1088.0, + 1023.0, + 1088.0, + 1063.0, + 892.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1023.0, + 1404.0, + 1023.0, + 1404.0, + 1063.0, + 1195.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 1055.0, + 521.0, + 1055.0, + 521.0, + 1093.0, + 509.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 305.0, + 912.0, + 305.0, + 912.0, + 344.0, + 294.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1518.0, + 1392.0, + 1518.0, + 1392.0, + 1562.0, + 305.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 472.0, + 717.0, + 472.0, + 717.0, + 514.0, + 354.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 420.0, + 671.0, + 420.0, + 671.0, + 462.0, + 354.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 375.0, + 702.0, + 375.0, + 702.0, + 418.0, + 353.0, + 418.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 21, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.885 + }, + { + "category_id": 0, + "poly": [ + 299, + 228, + 489, + 228, + 489, + 261, + 299, + 261 + ], + "score": 0.854 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2113, + 835, + 2113 + ], + "score": 0.839 + }, + { + "category_id": 1, + "poly": [ + 297, + 1099, + 1404, + 1099, + 1404, + 1160, + 297, + 1160 + ], + "score": 0.658 + }, + { + "category_id": 1, + "poly": [ + 296, + 1178, + 1400, + 1178, + 1400, + 1239, + 296, + 1239 + ], + "score": 0.629 + }, + { + "category_id": 1, + "poly": [ + 297, + 1020, + 1402, + 1020, + 1402, + 1081, + 297, + 1081 + ], + "score": 0.603 + }, + { + "category_id": 1, + "poly": [ + 295, + 1551, + 1402, + 1551, + 1402, + 1611, + 295, + 1611 + ], + "score": 0.577 + }, + { + "category_id": 1, + "poly": [ + 299, + 1258, + 1402, + 1258, + 1402, + 1346, + 299, + 1346 + ], + "score": 0.576 + }, + { + "category_id": 1, + "poly": [ + 297, + 941, + 1402, + 941, + 1402, + 1002, + 297, + 1002 + ], + "score": 0.575 + }, + { + "category_id": 1, + "poly": [ + 296, + 1631, + 1403, + 1631, + 1403, + 1691, + 296, + 1691 + ], + "score": 0.573 + }, + { + "category_id": 1, + "poly": [ + 296, + 277, + 1405, + 277, + 1405, + 362, + 296, + 362 + ], + "score": 0.568 + }, + { + "category_id": 1, + "poly": [ + 298, + 1366, + 1402, + 1366, + 1402, + 1454, + 298, + 1454 + ], + "score": 0.538 + }, + { + "category_id": 1, + "poly": [ + 296, + 1471, + 1402, + 1471, + 1402, + 1531, + 296, + 1531 + ], + "score": 0.531 + }, + { + "category_id": 1, + "poly": [ + 295, + 1710, + 1403, + 1710, + 1403, + 1770, + 295, + 1770 + ], + "score": 0.531 + }, + { + "category_id": 1, + "poly": [ + 293, + 383, + 1400, + 383, + 1400, + 469, + 293, + 469 + ], + "score": 0.511 + }, + { + "category_id": 1, + "poly": [ + 294, + 861, + 1402, + 861, + 1402, + 923, + 294, + 923 + ], + "score": 0.51 + }, + { + "category_id": 1, + "poly": [ + 295, + 489, + 1398, + 489, + 1398, + 550, + 295, + 550 + ], + "score": 0.503 + }, + { + "category_id": 1, + "poly": [ + 299, + 754, + 1404, + 754, + 1404, + 844, + 299, + 844 + ], + "score": 0.502 + }, + { + "category_id": 1, + "poly": [ + 297, + 647, + 1402, + 647, + 1402, + 737, + 297, + 737 + ], + "score": 0.482 + }, + { + "category_id": 1, + "poly": [ + 288, + 2005, + 1305, + 2005, + 1305, + 2035, + 288, + 2035 + ], + "score": 0.481 + }, + { + "category_id": 1, + "poly": [ + 294, + 569, + 1402, + 569, + 1402, + 629, + 294, + 629 + ], + "score": 0.479 + }, + { + "category_id": 1, + "poly": [ + 295, + 1790, + 1403, + 1790, + 1403, + 1876, + 295, + 1876 + ], + "score": 0.468 + }, + { + "category_id": 1, + "poly": [ + 294, + 1896, + 1404, + 1896, + 1404, + 1984, + 294, + 1984 + ], + "score": 0.468 + }, + { + "category_id": 13, + "poly": [ + 446, + 336, + 462, + 336, + 462, + 357, + 446, + 357 + ], + "score": 0.38, + "latex": "\\equiv" + }, + { + "category_id": 13, + "poly": [ + 1385, + 417, + 1403, + 417, + 1403, + 437, + 1385, + 437 + ], + "score": 0.29, + "latex": "=" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 228.0, + 490.0, + 228.0, + 490.0, + 265.0, + 297.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2124.0, + 828.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1099.0, + 1404.0, + 1099.0, + 1404.0, + 1135.0, + 296.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1130.0, + 668.0, + 1130.0, + 668.0, + 1160.0, + 322.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1180.0, + 1402.0, + 1180.0, + 1402.0, + 1211.0, + 296.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1210.0, + 810.0, + 1210.0, + 810.0, + 1239.0, + 322.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1022.0, + 1406.0, + 1022.0, + 1406.0, + 1053.0, + 296.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1051.0, + 731.0, + 1051.0, + 731.0, + 1081.0, + 322.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1551.0, + 1405.0, + 1551.0, + 1405.0, + 1587.0, + 295.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1580.0, + 968.0, + 1580.0, + 968.0, + 1612.0, + 322.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1259.0, + 1407.0, + 1259.0, + 1407.0, + 1291.0, + 297.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1286.0, + 1406.0, + 1286.0, + 1406.0, + 1320.0, + 322.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1315.0, + 983.0, + 1315.0, + 983.0, + 1348.0, + 322.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 934.0, + 1409.0, + 934.0, + 1409.0, + 983.0, + 291.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 972.0, + 700.0, + 972.0, + 700.0, + 1000.0, + 323.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1631.0, + 1405.0, + 1631.0, + 1405.0, + 1666.0, + 295.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1660.0, + 941.0, + 1660.0, + 941.0, + 1691.0, + 323.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 274.0, + 1407.0, + 274.0, + 1407.0, + 310.0, + 294.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 307.0, + 1403.0, + 307.0, + 1403.0, + 336.0, + 324.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 333.0, + 445.0, + 333.0, + 445.0, + 361.0, + 325.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 333.0, + 609.0, + 333.0, + 609.0, + 361.0, + 463.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1363.0, + 1407.0, + 1363.0, + 1407.0, + 1402.0, + 292.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1396.0, + 1404.0, + 1396.0, + 1404.0, + 1425.0, + 325.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1422.0, + 1097.0, + 1422.0, + 1097.0, + 1455.0, + 318.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1468.0, + 1404.0, + 1468.0, + 1404.0, + 1509.0, + 296.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1501.0, + 581.0, + 1501.0, + 581.0, + 1530.0, + 320.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1711.0, + 1405.0, + 1711.0, + 1405.0, + 1742.0, + 297.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1740.0, + 639.0, + 1740.0, + 639.0, + 1767.0, + 322.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 382.0, + 1405.0, + 382.0, + 1405.0, + 417.0, + 295.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 412.0, + 1384.0, + 412.0, + 1384.0, + 441.0, + 323.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 440.0, + 502.0, + 440.0, + 502.0, + 469.0, + 323.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 859.0, + 1405.0, + 859.0, + 1405.0, + 898.0, + 293.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 891.0, + 1278.0, + 891.0, + 1278.0, + 924.0, + 321.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 488.0, + 1403.0, + 488.0, + 1403.0, + 525.0, + 294.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 522.0, + 777.0, + 522.0, + 777.0, + 548.0, + 321.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 751.0, + 1408.0, + 751.0, + 1408.0, + 790.0, + 294.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 784.0, + 1404.0, + 784.0, + 1404.0, + 817.0, + 323.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 814.0, + 700.0, + 814.0, + 700.0, + 844.0, + 323.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 648.0, + 1406.0, + 648.0, + 1406.0, + 681.0, + 295.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 678.0, + 1406.0, + 678.0, + 1406.0, + 710.0, + 321.0, + 710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 705.0, + 1014.0, + 705.0, + 1014.0, + 737.0, + 322.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1998.0, + 1306.0, + 1998.0, + 1306.0, + 2040.0, + 295.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 566.0, + 1405.0, + 566.0, + 1405.0, + 605.0, + 294.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 597.0, + 1336.0, + 597.0, + 1336.0, + 631.0, + 323.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1790.0, + 1406.0, + 1790.0, + 1406.0, + 1823.0, + 295.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1814.0, + 1410.0, + 1814.0, + 1410.0, + 1854.0, + 319.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1847.0, + 551.0, + 1847.0, + 551.0, + 1876.0, + 320.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1892.0, + 1405.0, + 1892.0, + 1405.0, + 1933.0, + 292.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1924.0, + 1405.0, + 1924.0, + 1405.0, + 1959.0, + 321.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1953.0, + 734.0, + 1953.0, + 734.0, + 1986.0, + 323.0, + 1986.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 22, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 816, + 75, + 816, + 105, + 300, + 105 + ], + "score": 0.873 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2112, + 835, + 2112 + ], + "score": 0.831 + }, + { + "category_id": 1, + "poly": [ + 299, + 1513, + 1403, + 1513, + 1403, + 1573, + 299, + 1573 + ], + "score": 0.821 + }, + { + "category_id": 1, + "poly": [ + 295, + 977, + 1090, + 977, + 1090, + 1008, + 295, + 1008 + ], + "score": 0.817 + }, + { + "category_id": 1, + "poly": [ + 295, + 1361, + 1404, + 1361, + 1404, + 1421, + 295, + 1421 + ], + "score": 0.813 + }, + { + "category_id": 1, + "poly": [ + 299, + 1665, + 1402, + 1665, + 1402, + 1725, + 299, + 1725 + ], + "score": 0.81 + }, + { + "category_id": 1, + "poly": [ + 297, + 1438, + 1401, + 1438, + 1401, + 1498, + 297, + 1498 + ], + "score": 0.81 + }, + { + "category_id": 1, + "poly": [ + 299, + 1024, + 1396, + 1024, + 1396, + 1110, + 299, + 1110 + ], + "score": 0.808 + }, + { + "category_id": 1, + "poly": [ + 296, + 1741, + 1404, + 1741, + 1404, + 1801, + 296, + 1801 + ], + "score": 0.807 + }, + { + "category_id": 1, + "poly": [ + 300, + 872, + 1401, + 872, + 1401, + 959, + 300, + 959 + ], + "score": 0.804 + }, + { + "category_id": 1, + "poly": [ + 298, + 1127, + 1402, + 1127, + 1402, + 1186, + 298, + 1186 + ], + "score": 0.8 + }, + { + "category_id": 1, + "poly": [ + 298, + 1588, + 1399, + 1588, + 1399, + 1649, + 298, + 1649 + ], + "score": 0.8 + }, + { + "category_id": 1, + "poly": [ + 299, + 231, + 1404, + 231, + 1404, + 317, + 299, + 317 + ], + "score": 0.783 + }, + { + "category_id": 1, + "poly": [ + 292, + 335, + 1402, + 335, + 1402, + 393, + 292, + 393 + ], + "score": 0.78 + }, + { + "category_id": 1, + "poly": [ + 295, + 562, + 1402, + 562, + 1402, + 622, + 295, + 622 + ], + "score": 0.765 + }, + { + "category_id": 1, + "poly": [ + 298, + 409, + 1400, + 409, + 1400, + 469, + 298, + 469 + ], + "score": 0.763 + }, + { + "category_id": 1, + "poly": [ + 298, + 740, + 1407, + 740, + 1407, + 854, + 298, + 854 + ], + "score": 0.76 + }, + { + "category_id": 1, + "poly": [ + 301, + 638, + 1405, + 638, + 1405, + 722, + 301, + 722 + ], + "score": 0.76 + }, + { + "category_id": 1, + "poly": [ + 296, + 486, + 1402, + 486, + 1402, + 545, + 296, + 545 + ], + "score": 0.758 + }, + { + "category_id": 1, + "poly": [ + 299, + 1949, + 1403, + 1949, + 1403, + 2034, + 299, + 2034 + ], + "score": 0.743 + }, + { + "category_id": 1, + "poly": [ + 298, + 1203, + 1407, + 1203, + 1407, + 1346, + 298, + 1346 + ], + "score": 0.718 + }, + { + "category_id": 1, + "poly": [ + 299, + 1818, + 1405, + 1818, + 1405, + 1932, + 299, + 1932 + ], + "score": 0.64 + }, + { + "category_id": 13, + "poly": [ + 671, + 1084, + 686, + 1084, + 686, + 1104, + 671, + 1104 + ], + "score": 0.28, + "latex": "=" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 2084.0, + 872.0, + 2084.0, + 872.0, + 2124.0, + 828.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1515.0, + 1404.0, + 1515.0, + 1404.0, + 1547.0, + 295.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1541.0, + 1350.0, + 1541.0, + 1350.0, + 1574.0, + 320.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 975.0, + 1091.0, + 975.0, + 1091.0, + 1010.0, + 297.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1358.0, + 1406.0, + 1358.0, + 1406.0, + 1397.0, + 293.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1392.0, + 833.0, + 1392.0, + 833.0, + 1421.0, + 322.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1664.0, + 1406.0, + 1664.0, + 1406.0, + 1700.0, + 295.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1694.0, + 944.0, + 1694.0, + 944.0, + 1725.0, + 323.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1439.0, + 1406.0, + 1439.0, + 1406.0, + 1470.0, + 296.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1468.0, + 1374.0, + 1468.0, + 1374.0, + 1499.0, + 322.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1026.0, + 1400.0, + 1026.0, + 1400.0, + 1055.0, + 296.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1053.0, + 1404.0, + 1053.0, + 1404.0, + 1086.0, + 323.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1080.0, + 670.0, + 1080.0, + 670.0, + 1115.0, + 323.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 1080.0, + 862.0, + 1080.0, + 862.0, + 1115.0, + 687.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1740.0, + 1404.0, + 1740.0, + 1404.0, + 1775.0, + 295.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1770.0, + 849.0, + 1770.0, + 849.0, + 1805.0, + 324.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 871.0, + 1406.0, + 871.0, + 1406.0, + 906.0, + 294.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 900.0, + 1408.0, + 900.0, + 1408.0, + 935.0, + 321.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 931.0, + 1103.0, + 931.0, + 1103.0, + 960.0, + 324.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1124.0, + 1407.0, + 1124.0, + 1407.0, + 1160.0, + 294.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1153.0, + 392.0, + 1153.0, + 392.0, + 1188.0, + 321.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1591.0, + 1401.0, + 1591.0, + 1401.0, + 1622.0, + 298.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1617.0, + 1209.0, + 1617.0, + 1209.0, + 1652.0, + 323.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 230.0, + 1405.0, + 230.0, + 1405.0, + 264.0, + 294.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 259.0, + 1406.0, + 259.0, + 1406.0, + 291.0, + 322.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 287.0, + 504.0, + 287.0, + 504.0, + 317.0, + 322.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 333.0, + 1405.0, + 333.0, + 1405.0, + 370.0, + 295.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 364.0, + 998.0, + 364.0, + 998.0, + 395.0, + 320.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 563.0, + 1404.0, + 563.0, + 1404.0, + 594.0, + 298.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 590.0, + 1135.0, + 590.0, + 1135.0, + 623.0, + 323.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 410.0, + 1405.0, + 410.0, + 1405.0, + 441.0, + 296.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 437.0, + 389.0, + 437.0, + 389.0, + 471.0, + 321.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 738.0, + 1407.0, + 738.0, + 1407.0, + 777.0, + 292.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 772.0, + 1405.0, + 772.0, + 1405.0, + 801.0, + 324.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 793.0, + 1409.0, + 793.0, + 1409.0, + 833.0, + 320.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 824.0, + 389.0, + 824.0, + 389.0, + 856.0, + 322.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 635.0, + 1405.0, + 635.0, + 1405.0, + 672.0, + 295.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 666.0, + 1409.0, + 666.0, + 1409.0, + 698.0, + 321.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 689.0, + 392.0, + 689.0, + 392.0, + 728.0, + 319.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 487.0, + 1403.0, + 487.0, + 1403.0, + 518.0, + 296.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 517.0, + 873.0, + 517.0, + 873.0, + 545.0, + 323.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1949.0, + 1404.0, + 1949.0, + 1404.0, + 1981.0, + 295.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1976.0, + 1407.0, + 1976.0, + 1407.0, + 2011.0, + 319.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 2004.0, + 1009.0, + 2004.0, + 1009.0, + 2035.0, + 319.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1204.0, + 1408.0, + 1204.0, + 1408.0, + 1235.0, + 298.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1233.0, + 1408.0, + 1233.0, + 1408.0, + 1264.0, + 324.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1261.0, + 1408.0, + 1261.0, + 1408.0, + 1292.0, + 324.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1283.0, + 1410.0, + 1283.0, + 1410.0, + 1324.0, + 320.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1314.0, + 1371.0, + 1314.0, + 1371.0, + 1349.0, + 322.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1819.0, + 1405.0, + 1819.0, + 1405.0, + 1852.0, + 297.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1843.0, + 1407.0, + 1843.0, + 1407.0, + 1883.0, + 321.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1871.0, + 1410.0, + 1871.0, + 1410.0, + 1908.0, + 321.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1903.0, + 949.0, + 1903.0, + 949.0, + 1933.0, + 324.0, + 1933.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 23, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.879 + }, + { + "category_id": 1, + "poly": [ + 296, + 1643, + 859, + 1643, + 859, + 1674, + 296, + 1674 + ], + "score": 0.84 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2112, + 835, + 2112 + ], + "score": 0.834 + }, + { + "category_id": 1, + "poly": [ + 291, + 1692, + 1402, + 1692, + 1402, + 1752, + 291, + 1752 + ], + "score": 0.787 + }, + { + "category_id": 1, + "poly": [ + 295, + 1977, + 1405, + 1977, + 1405, + 2033, + 295, + 2033 + ], + "score": 0.778 + }, + { + "category_id": 1, + "poly": [ + 300, + 1567, + 1402, + 1567, + 1402, + 1627, + 300, + 1627 + ], + "score": 0.775 + }, + { + "category_id": 1, + "poly": [ + 293, + 1490, + 1399, + 1490, + 1399, + 1551, + 293, + 1551 + ], + "score": 0.742 + }, + { + "category_id": 1, + "poly": [ + 297, + 1158, + 1401, + 1158, + 1401, + 1217, + 297, + 1217 + ], + "score": 0.719 + }, + { + "category_id": 1, + "poly": [ + 301, + 231, + 1403, + 231, + 1403, + 318, + 301, + 318 + ], + "score": 0.71 + }, + { + "category_id": 1, + "poly": [ + 298, + 875, + 1401, + 875, + 1401, + 961, + 298, + 961 + ], + "score": 0.708 + }, + { + "category_id": 1, + "poly": [ + 297, + 1900, + 1399, + 1900, + 1399, + 1958, + 297, + 1958 + ], + "score": 0.704 + }, + { + "category_id": 1, + "poly": [ + 298, + 978, + 1400, + 978, + 1400, + 1066, + 298, + 1066 + ], + "score": 0.703 + }, + { + "category_id": 1, + "poly": [ + 300, + 1234, + 1401, + 1234, + 1401, + 1295, + 300, + 1295 + ], + "score": 0.69 + }, + { + "category_id": 1, + "poly": [ + 294, + 1082, + 1399, + 1082, + 1399, + 1142, + 294, + 1142 + ], + "score": 0.679 + }, + { + "category_id": 1, + "poly": [ + 292, + 1311, + 1401, + 1311, + 1401, + 1370, + 292, + 1370 + ], + "score": 0.668 + }, + { + "category_id": 1, + "poly": [ + 297, + 770, + 1404, + 770, + 1404, + 857, + 297, + 857 + ], + "score": 0.667 + }, + { + "category_id": 1, + "poly": [ + 300, + 335, + 1399, + 335, + 1399, + 422, + 300, + 422 + ], + "score": 0.649 + }, + { + "category_id": 1, + "poly": [ + 296, + 439, + 1401, + 439, + 1401, + 525, + 296, + 525 + ], + "score": 0.643 + }, + { + "category_id": 1, + "poly": [ + 298, + 1768, + 1405, + 1768, + 1405, + 1881, + 298, + 1881 + ], + "score": 0.638 + }, + { + "category_id": 1, + "poly": [ + 294, + 694, + 1400, + 694, + 1400, + 755, + 294, + 755 + ], + "score": 0.627 + }, + { + "category_id": 1, + "poly": [ + 298, + 590, + 1399, + 590, + 1399, + 678, + 298, + 678 + ], + "score": 0.566 + }, + { + "category_id": 1, + "poly": [ + 291, + 542, + 1364, + 542, + 1364, + 575, + 291, + 575 + ], + "score": 0.563 + }, + { + "category_id": 1, + "poly": [ + 297, + 1386, + 1401, + 1386, + 1401, + 1473, + 297, + 1473 + ], + "score": 0.479 + }, + { + "category_id": 13, + "poly": [ + 790, + 936, + 806, + 936, + 806, + 955, + 790, + 955 + ], + "score": 0.48, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1384, + 472, + 1402, + 472, + 1402, + 492, + 1384, + 492 + ], + "score": 0.33, + "latex": "\\equiv" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1643.0, + 861.0, + 1643.0, + 861.0, + 1675.0, + 295.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1688.0, + 1404.0, + 1688.0, + 1404.0, + 1730.0, + 295.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1721.0, + 1005.0, + 1721.0, + 1005.0, + 1752.0, + 322.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 1403.0, + 1973.0, + 1403.0, + 2011.0, + 295.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2004.0, + 581.0, + 2004.0, + 581.0, + 2033.0, + 322.0, + 2033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1568.0, + 1404.0, + 1568.0, + 1404.0, + 1599.0, + 296.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1596.0, + 1129.0, + 1596.0, + 1129.0, + 1628.0, + 321.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1488.0, + 1405.0, + 1488.0, + 1405.0, + 1526.0, + 293.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1520.0, + 1173.0, + 1520.0, + 1173.0, + 1552.0, + 325.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1160.0, + 1403.0, + 1160.0, + 1403.0, + 1191.0, + 296.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1188.0, + 966.0, + 1188.0, + 966.0, + 1219.0, + 324.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 230.0, + 1407.0, + 230.0, + 1407.0, + 264.0, + 294.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 259.0, + 1404.0, + 259.0, + 1404.0, + 292.0, + 324.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 287.0, + 1136.0, + 287.0, + 1136.0, + 320.0, + 321.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 875.0, + 1406.0, + 875.0, + 1406.0, + 911.0, + 296.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 906.0, + 1405.0, + 906.0, + 1405.0, + 935.0, + 323.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 931.0, + 789.0, + 931.0, + 789.0, + 966.0, + 322.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 931.0, + 967.0, + 931.0, + 967.0, + 966.0, + 807.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1901.0, + 1403.0, + 1901.0, + 1403.0, + 1935.0, + 296.0, + 1935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1927.0, + 1143.0, + 1927.0, + 1143.0, + 1961.0, + 323.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 981.0, + 1404.0, + 981.0, + 1404.0, + 1010.0, + 297.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1009.0, + 1405.0, + 1009.0, + 1405.0, + 1038.0, + 324.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1034.0, + 1275.0, + 1034.0, + 1275.0, + 1069.0, + 319.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1230.0, + 1406.0, + 1230.0, + 1406.0, + 1273.0, + 295.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1263.0, + 1148.0, + 1263.0, + 1148.0, + 1296.0, + 321.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1084.0, + 1403.0, + 1084.0, + 1403.0, + 1116.0, + 297.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1110.0, + 1165.0, + 1110.0, + 1165.0, + 1143.0, + 321.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1311.0, + 1406.0, + 1311.0, + 1406.0, + 1345.0, + 296.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1341.0, + 995.0, + 1341.0, + 995.0, + 1370.0, + 324.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 771.0, + 1407.0, + 771.0, + 1407.0, + 804.0, + 297.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 797.0, + 1405.0, + 797.0, + 1405.0, + 836.0, + 320.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 828.0, + 525.0, + 828.0, + 525.0, + 858.0, + 324.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 335.0, + 1404.0, + 335.0, + 1404.0, + 368.0, + 296.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 366.0, + 1402.0, + 366.0, + 1402.0, + 395.0, + 322.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 391.0, + 852.0, + 391.0, + 852.0, + 424.0, + 324.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 438.0, + 1406.0, + 438.0, + 1406.0, + 473.0, + 295.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 466.0, + 1383.0, + 466.0, + 1383.0, + 498.0, + 323.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 497.0, + 506.0, + 497.0, + 506.0, + 526.0, + 324.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1769.0, + 1407.0, + 1769.0, + 1407.0, + 1802.0, + 296.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1797.0, + 1406.0, + 1797.0, + 1406.0, + 1830.0, + 325.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1822.0, + 1408.0, + 1822.0, + 1408.0, + 1859.0, + 321.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1847.0, + 403.0, + 1847.0, + 403.0, + 1885.0, + 321.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 693.0, + 1404.0, + 693.0, + 1404.0, + 731.0, + 294.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 723.0, + 1132.0, + 723.0, + 1132.0, + 756.0, + 321.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 589.0, + 1405.0, + 589.0, + 1405.0, + 627.0, + 292.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 619.0, + 1405.0, + 619.0, + 1405.0, + 655.0, + 323.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 647.0, + 905.0, + 647.0, + 905.0, + 680.0, + 324.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 542.0, + 1363.0, + 542.0, + 1363.0, + 577.0, + 297.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1386.0, + 1406.0, + 1386.0, + 1406.0, + 1420.0, + 295.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1413.0, + 1405.0, + 1413.0, + 1405.0, + 1451.0, + 321.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1445.0, + 514.0, + 1445.0, + 514.0, + 1474.0, + 324.0, + 1474.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 24, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 816, + 75, + 816, + 105, + 300, + 105 + ], + "score": 0.876 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2113, + 834, + 2113 + ], + "score": 0.836 + }, + { + "category_id": 1, + "poly": [ + 293, + 231, + 1409, + 231, + 1409, + 1134, + 293, + 1134 + ], + "score": 0.656 + }, + { + "category_id": 13, + "poly": [ + 446, + 552, + 463, + 552, + 463, + 573, + 446, + 573 + ], + "score": 0.54, + "latex": "\\equiv" + }, + { + "category_id": 13, + "poly": [ + 1179, + 1100, + 1196, + 1100, + 1196, + 1119, + 1179, + 1119 + ], + "score": 0.38, + "latex": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =" + }, + { + "category_id": 13, + "poly": [ + 446, + 1022, + 462, + 1022, + 462, + 1043, + 446, + 1043 + ], + "score": 0.37, + "latex": "\\equiv" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2121.0, + 830.0, + 2121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 231.0, + 1407.0, + 231.0, + 1407.0, + 270.0, + 294.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 257.0, + 1407.0, + 257.0, + 1407.0, + 299.0, + 320.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 288.0, + 931.0, + 288.0, + 931.0, + 323.0, + 322.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 334.0, + 1406.0, + 334.0, + 1406.0, + 375.0, + 291.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 365.0, + 1402.0, + 365.0, + 1402.0, + 400.0, + 322.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 413.0, + 1407.0, + 413.0, + 1407.0, + 453.0, + 291.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 442.0, + 1363.0, + 442.0, + 1363.0, + 479.0, + 319.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 491.0, + 1407.0, + 491.0, + 1407.0, + 528.0, + 294.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 521.0, + 1406.0, + 521.0, + 1406.0, + 556.0, + 322.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 550.0, + 445.0, + 550.0, + 445.0, + 585.0, + 324.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 550.0, + 623.0, + 550.0, + 623.0, + 585.0, + 464.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 595.0, + 1406.0, + 595.0, + 1406.0, + 636.0, + 294.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 627.0, + 1403.0, + 627.0, + 1403.0, + 659.0, + 323.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 654.0, + 656.0, + 654.0, + 656.0, + 687.0, + 319.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 704.0, + 1408.0, + 704.0, + 1408.0, + 739.0, + 295.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 733.0, + 1102.0, + 733.0, + 1102.0, + 764.0, + 320.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 779.0, + 1405.0, + 779.0, + 1405.0, + 817.0, + 292.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 809.0, + 1168.0, + 809.0, + 1168.0, + 844.0, + 322.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 857.0, + 1405.0, + 857.0, + 1405.0, + 894.0, + 292.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 884.0, + 1406.0, + 884.0, + 1406.0, + 924.0, + 319.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 913.0, + 926.0, + 913.0, + 926.0, + 948.0, + 322.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 961.0, + 1410.0, + 961.0, + 1410.0, + 1001.0, + 292.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 992.0, + 1407.0, + 992.0, + 1407.0, + 1027.0, + 322.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1018.0, + 445.0, + 1018.0, + 445.0, + 1054.0, + 324.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1018.0, + 623.0, + 1018.0, + 623.0, + 1054.0, + 463.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1066.0, + 1406.0, + 1066.0, + 1406.0, + 1106.0, + 292.0, + 1106.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1096.0, + 1178.0, + 1096.0, + 1178.0, + 1131.0, + 320.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1096.0, + 1355.0, + 1096.0, + 1355.0, + 1131.0, + 1197.0, + 1131.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 25, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/dev/HH3GHN_Q1Ba/images/0060862d868ac5fe982d975b2c04d47a9ec5eb182298c2566f40587c87e97c1a.jpg b/parse/dev/HH3GHN_Q1Ba/images/0060862d868ac5fe982d975b2c04d47a9ec5eb182298c2566f40587c87e97c1a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3d06b7ddb0648827b5f153ff3b77d85a74441c0d --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/0060862d868ac5fe982d975b2c04d47a9ec5eb182298c2566f40587c87e97c1a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bca3ebed68471dcc109f9a71179f14a2725f5b7b9f3ae458caaf7ee1ada1151b +size 71036 diff --git a/parse/dev/HH3GHN_Q1Ba/images/0b5a60f19f06e4f1669620d951fa8a3907482a9ffadc74383ab61e566377c557.jpg b/parse/dev/HH3GHN_Q1Ba/images/0b5a60f19f06e4f1669620d951fa8a3907482a9ffadc74383ab61e566377c557.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7655a58862160d55f91ec84a3023c4b6ebaddc0d --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/0b5a60f19f06e4f1669620d951fa8a3907482a9ffadc74383ab61e566377c557.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a4961744b9be148b6f73b8586a231589f0d4bfbc87bfddbb146c2ee798d69a4 +size 3129 diff --git a/parse/dev/HH3GHN_Q1Ba/images/239ec8dcdfb85587594ebab6155ba059af73935b65fca6fbcec9ef605ea8f5bc.jpg b/parse/dev/HH3GHN_Q1Ba/images/239ec8dcdfb85587594ebab6155ba059af73935b65fca6fbcec9ef605ea8f5bc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..87cfb36f6440579f522ef8621b5b4d46a154d987 --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/239ec8dcdfb85587594ebab6155ba059af73935b65fca6fbcec9ef605ea8f5bc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c1560661f632f31997576fca3f26f9aaa406dfe44a7f06c5f7702c3cfe529b0 +size 4853 diff --git a/parse/dev/HH3GHN_Q1Ba/images/50108642be8a4c2c7027cfb5147d23af23c60ecccb1e9ee9b41c7fcc9847795e.jpg b/parse/dev/HH3GHN_Q1Ba/images/50108642be8a4c2c7027cfb5147d23af23c60ecccb1e9ee9b41c7fcc9847795e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ce09f350e3ad26f9a2460182d4c3b415bdb2ddea --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/50108642be8a4c2c7027cfb5147d23af23c60ecccb1e9ee9b41c7fcc9847795e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3bd6bbcc2e5f3bbabfba6238821dbcefeb8a88125378fc3801680ee026c43bcb +size 37943 diff --git a/parse/dev/HH3GHN_Q1Ba/images/96de090c059271dae874bda0fd03cd1e0d69236b6e868c2364af673b44477a31.jpg b/parse/dev/HH3GHN_Q1Ba/images/96de090c059271dae874bda0fd03cd1e0d69236b6e868c2364af673b44477a31.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a37bde4d75e2822002587c5eb490ca1579af7b87 --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/96de090c059271dae874bda0fd03cd1e0d69236b6e868c2364af673b44477a31.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a263d3a7827018658e777abd165cdb9efc235cc29584d775a1e301af694f005a +size 29683 diff --git a/parse/dev/HH3GHN_Q1Ba/images/a7cfb413d91d48729a1156fa423cdb687d726f0529babdbf272581c26ffdee54.jpg b/parse/dev/HH3GHN_Q1Ba/images/a7cfb413d91d48729a1156fa423cdb687d726f0529babdbf272581c26ffdee54.jpg new file mode 100644 index 0000000000000000000000000000000000000000..77323009644551662d097abef36a4b8c231d9b3e --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/a7cfb413d91d48729a1156fa423cdb687d726f0529babdbf272581c26ffdee54.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c550d5950ed6e2083603da52f3fd63e664ae21d165b7450b22078330184e357b +size 3512 diff --git a/parse/dev/HH3GHN_Q1Ba/images/b13a33960517ca91aaf32900110e25e78e7402d3ebe4398b08fa7ab4ef1e6f59.jpg b/parse/dev/HH3GHN_Q1Ba/images/b13a33960517ca91aaf32900110e25e78e7402d3ebe4398b08fa7ab4ef1e6f59.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0c6ba54460c0b9a1497b9878ca990f6c7686aa20 --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/b13a33960517ca91aaf32900110e25e78e7402d3ebe4398b08fa7ab4ef1e6f59.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd64878239a47ccdc2facfe49a474a12c3e75f72415dee6536035da599cef737 +size 7689 diff --git a/parse/dev/HH3GHN_Q1Ba/images/c1eddb6fc702095e2d788d16d4e7ef83f4cdbef34f777f2984677508d830e0ca.jpg b/parse/dev/HH3GHN_Q1Ba/images/c1eddb6fc702095e2d788d16d4e7ef83f4cdbef34f777f2984677508d830e0ca.jpg new file mode 100644 index 0000000000000000000000000000000000000000..938893b1954ea576fa8ed4f2df9add4c6330e463 --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/c1eddb6fc702095e2d788d16d4e7ef83f4cdbef34f777f2984677508d830e0ca.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7cb16e3c96e4390c9f823f955e7ed4bcf88ab82e9d55ab5a7a688d86f32bcbc0 +size 6387 diff --git a/parse/dev/HH3GHN_Q1Ba/images/d1b58f5a19eb70b0b23b55cc1d29a3d1719d9aa847405fa988ca77e31641401a.jpg b/parse/dev/HH3GHN_Q1Ba/images/d1b58f5a19eb70b0b23b55cc1d29a3d1719d9aa847405fa988ca77e31641401a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ec6d94c88e5ae32e5b44b7807bd910b6854848f2 --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/d1b58f5a19eb70b0b23b55cc1d29a3d1719d9aa847405fa988ca77e31641401a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:580138f1286b58212bfa2e390cd7b8a0bef0a67037e22aef243e35f7277b2a49 +size 9483 diff --git a/parse/dev/HH3GHN_Q1Ba/images/e7d09a7b0dd300ef33c8470b1d1b81d3e6fb58a44fa4b4d9a7ad43dd56d4c278.jpg b/parse/dev/HH3GHN_Q1Ba/images/e7d09a7b0dd300ef33c8470b1d1b81d3e6fb58a44fa4b4d9a7ad43dd56d4c278.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3b1d8f729cfbec0046b467aa3ffc7b39707bd3cd --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/e7d09a7b0dd300ef33c8470b1d1b81d3e6fb58a44fa4b4d9a7ad43dd56d4c278.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f017252884fe737f1801c99a95bc1d58737b91c232667d5fbf8f246caa55a42b +size 20026 diff --git a/parse/dev/HH3GHN_Q1Ba/images/ec47a08c6d75085626dd7a5877acc5569093e4599fcb40f0df28f7779060214e.jpg b/parse/dev/HH3GHN_Q1Ba/images/ec47a08c6d75085626dd7a5877acc5569093e4599fcb40f0df28f7779060214e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f484b126564cbadeaf690d9e83f06d894c0e4412 --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/ec47a08c6d75085626dd7a5877acc5569093e4599fcb40f0df28f7779060214e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3da3d5a925521eea9953b328a183bed7691f7d2b9b89b29837e4bac9111a615 +size 23692 diff --git a/parse/dev/HH3GHN_Q1Ba/images/f2f463e93ac956465641d4f807d61427f85855a556a4602832e7ea39ee382cf7.jpg b/parse/dev/HH3GHN_Q1Ba/images/f2f463e93ac956465641d4f807d61427f85855a556a4602832e7ea39ee382cf7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..270fa46902d57982abadc9b33d1cfe4c76740240 --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/f2f463e93ac956465641d4f807d61427f85855a556a4602832e7ea39ee382cf7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6f2d5b843299bbd6cd62cdc4bee63d370af2b3b81a98477e11c5bbefdd7068e +size 3060 diff --git a/parse/dev/HH3GHN_Q1Ba/images/f387ff3283156b753832cc3bbab074ab1a2fdae0e4a14e26a9803f932b7be04d.jpg b/parse/dev/HH3GHN_Q1Ba/images/f387ff3283156b753832cc3bbab074ab1a2fdae0e4a14e26a9803f932b7be04d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6b0950be17dc7e5ef18f88f5940ca621687f7717 --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/f387ff3283156b753832cc3bbab074ab1a2fdae0e4a14e26a9803f932b7be04d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b543b1fa4e1cf73bce0b30c6841f82e8133bc97bc20801a6858eb2734713d68 +size 9813 diff --git a/parse/dev/HH3GHN_Q1Ba/images/fe8f2ca377d41e351308ab27f3827ff91e061b6cf5b831419a9de48ff4932750.jpg b/parse/dev/HH3GHN_Q1Ba/images/fe8f2ca377d41e351308ab27f3827ff91e061b6cf5b831419a9de48ff4932750.jpg new file mode 100644 index 0000000000000000000000000000000000000000..637e767d8e172e406600e298c9b58c4e42a6e9bc --- /dev/null +++ b/parse/dev/HH3GHN_Q1Ba/images/fe8f2ca377d41e351308ab27f3827ff91e061b6cf5b831419a9de48ff4932750.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b291be83c97658faeb7bb77b2dbd990b33dee07c7f05d81bbad4f8d1feace061 +size 11582 diff --git a/parse/dev/ITw9edRDlD/images/1ee9c99e7812e40c6978800126e5e620c636f1b53224d74a08f05c34d216602c.jpg b/parse/dev/ITw9edRDlD/images/1ee9c99e7812e40c6978800126e5e620c636f1b53224d74a08f05c34d216602c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d915c0ef04b2b9de1594ec4b08423a908fef56e8 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/1ee9c99e7812e40c6978800126e5e620c636f1b53224d74a08f05c34d216602c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58779c924da0fd00e36d460e4e23b9f6ae4eeb1a461c0cb1c85e4b3b68a605db +size 3360 diff --git a/parse/dev/ITw9edRDlD/images/2151594b0144a294047300f651dcce9bfc628b1e2f77cedebc288ed61472039a.jpg b/parse/dev/ITw9edRDlD/images/2151594b0144a294047300f651dcce9bfc628b1e2f77cedebc288ed61472039a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b93a10fda4e9b32eea8d3dde57f3016fc341b431 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/2151594b0144a294047300f651dcce9bfc628b1e2f77cedebc288ed61472039a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9eaf79fe1f6f968d8d2a5b711163e9672e73d02b01aab7e8376c47ad614edee4 +size 84151 diff --git a/parse/dev/ITw9edRDlD/images/25889826914510fd5999570fda3860a6fdf394b61d83134a7bd26d56c641e5c6.jpg b/parse/dev/ITw9edRDlD/images/25889826914510fd5999570fda3860a6fdf394b61d83134a7bd26d56c641e5c6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4134f13f1fa2c6cbe794f97ce24bb0374248e83e --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/25889826914510fd5999570fda3860a6fdf394b61d83134a7bd26d56c641e5c6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4f46cb97343252344161fc1cc9830305ca938b7fe65d175cfc378f0145b3094 +size 21982 diff --git a/parse/dev/ITw9edRDlD/images/2a0b37bcf22a17705eb78cf5e057a2c33e380e7c33abe875a1f08da0f76248e3.jpg b/parse/dev/ITw9edRDlD/images/2a0b37bcf22a17705eb78cf5e057a2c33e380e7c33abe875a1f08da0f76248e3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db7fadf59c6dc39e966d9a15621389c0a0d61235 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/2a0b37bcf22a17705eb78cf5e057a2c33e380e7c33abe875a1f08da0f76248e3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:15dbe148fbd17d5960bb6d37761e0ddb19d3441350d07f8a1f1bcf662258babf +size 6202 diff --git a/parse/dev/ITw9edRDlD/images/2e488cd5d0123faca69d3ad78b6b5874fe2c1df04992be3b932ab2f92c00f572.jpg b/parse/dev/ITw9edRDlD/images/2e488cd5d0123faca69d3ad78b6b5874fe2c1df04992be3b932ab2f92c00f572.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1b586c56f7aeafaf6faf32168d704038e3183277 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/2e488cd5d0123faca69d3ad78b6b5874fe2c1df04992be3b932ab2f92c00f572.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:91b2b7ffbb8e23c148420b411ea056ed16725133a21d742125d68cbaf85f9ca9 +size 6773 diff --git a/parse/dev/ITw9edRDlD/images/3489b3534ad2521f4b7d830b0635c6120f9030d86dee59d62ef68e143d62e50e.jpg b/parse/dev/ITw9edRDlD/images/3489b3534ad2521f4b7d830b0635c6120f9030d86dee59d62ef68e143d62e50e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d5ede44e6751b68340e9d11f0a198c906c91965 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/3489b3534ad2521f4b7d830b0635c6120f9030d86dee59d62ef68e143d62e50e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:363edde54f1b4cdb05ecf6e91fcc79581a4c4abd72aed816509a160c86ec16f2 +size 33098 diff --git a/parse/dev/ITw9edRDlD/images/38be85b71e7e7bada1a8b5a5180dbe5be8df06d1287fef6f83fcabce79b76a66.jpg b/parse/dev/ITw9edRDlD/images/38be85b71e7e7bada1a8b5a5180dbe5be8df06d1287fef6f83fcabce79b76a66.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2020b782810f1edb16cf866fa9afff09daae9d47 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/38be85b71e7e7bada1a8b5a5180dbe5be8df06d1287fef6f83fcabce79b76a66.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bb851540119197745217171b88b789a6089374234ba4b73ee1730005e247b1c2 +size 45624 diff --git a/parse/dev/ITw9edRDlD/images/38d1b3571deb00098b83f0272d6e8d14ac421eb06ebecf23348f75954da88ad8.jpg b/parse/dev/ITw9edRDlD/images/38d1b3571deb00098b83f0272d6e8d14ac421eb06ebecf23348f75954da88ad8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..58507bb4c5502022b8c9f6aa8ea4cfea03c0d7a8 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/38d1b3571deb00098b83f0272d6e8d14ac421eb06ebecf23348f75954da88ad8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c6ddbe5e092f147a23e556d2f04c7d40da7f8829c248f3c15f527a90679f60c1 +size 40804 diff --git a/parse/dev/ITw9edRDlD/images/3b5832ac7d413a729a7b0b439ac6e558fb55d2924cc8f9f534d9caea6166a616.jpg b/parse/dev/ITw9edRDlD/images/3b5832ac7d413a729a7b0b439ac6e558fb55d2924cc8f9f534d9caea6166a616.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a0efab2b17819a2ae78d76cebaec04c08df01e2d --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/3b5832ac7d413a729a7b0b439ac6e558fb55d2924cc8f9f534d9caea6166a616.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d15113c074b54a038a55a62aa3da52b5fa1d62fe42224114196ef1b16839d040 +size 8351 diff --git a/parse/dev/ITw9edRDlD/images/58619c355bf48fd739a8be2b0fcd8ad7e47c401ff272f138604393721c9fb37d.jpg b/parse/dev/ITw9edRDlD/images/58619c355bf48fd739a8be2b0fcd8ad7e47c401ff272f138604393721c9fb37d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3e989c87537fadd9f10dcf4c4f5928ed3d496f80 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/58619c355bf48fd739a8be2b0fcd8ad7e47c401ff272f138604393721c9fb37d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7dc225a85fcfc043ca9ad075e98a7e5de5f479c7f392b0c76578c17dac551fc0 +size 86782 diff --git a/parse/dev/ITw9edRDlD/images/6df8e9b5949e4aab27210961a1263e3d0c8f4e152a055e4a5272939a20d94f46.jpg b/parse/dev/ITw9edRDlD/images/6df8e9b5949e4aab27210961a1263e3d0c8f4e152a055e4a5272939a20d94f46.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c2c904208c25c9f63c7929b23ad974fc881ce340 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/6df8e9b5949e4aab27210961a1263e3d0c8f4e152a055e4a5272939a20d94f46.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddf6b7aa7ec991433a2bc1485e4100b0dd8b3768664e091a353f92d5c6b984cc +size 3917 diff --git a/parse/dev/ITw9edRDlD/images/8af0b76128070114c8b41becd05c431422903df72aa8755799c99fb78638ba59.jpg b/parse/dev/ITw9edRDlD/images/8af0b76128070114c8b41becd05c431422903df72aa8755799c99fb78638ba59.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8384ff5f6065a19c18eec46cfd53ad24378255b9 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/8af0b76128070114c8b41becd05c431422903df72aa8755799c99fb78638ba59.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:876e40711805cd6d394824f252c833d86975f3d2a7013d489640b3e4b25daaae +size 3241 diff --git a/parse/dev/ITw9edRDlD/images/8b509ff42cbe3fe7c78e89bc130afcf7ca2207b38d561841d48ee943b4ca6f56.jpg b/parse/dev/ITw9edRDlD/images/8b509ff42cbe3fe7c78e89bc130afcf7ca2207b38d561841d48ee943b4ca6f56.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ac08b056622f553e957223cfc3416caf33777828 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/8b509ff42cbe3fe7c78e89bc130afcf7ca2207b38d561841d48ee943b4ca6f56.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b0879e568eb51a9ee18bdab466a7e5e969717f2005997b54032079260218604 +size 9098 diff --git a/parse/dev/ITw9edRDlD/images/97ad61b112eae0c20313f648501431aa4dc61eed1e7e8b1c4819d7f3e8289df0.jpg b/parse/dev/ITw9edRDlD/images/97ad61b112eae0c20313f648501431aa4dc61eed1e7e8b1c4819d7f3e8289df0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3b6d2ddcd53098ce18f812c28a65a044220dfb71 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/97ad61b112eae0c20313f648501431aa4dc61eed1e7e8b1c4819d7f3e8289df0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a697ed74b49332a8650fbfd548dab1f012ed1e2916baff1bed7a3a05eeb00b7 +size 11322 diff --git a/parse/dev/ITw9edRDlD/images/9f821e06113838668f620fece0360ab9f0a920b262835347b401956be4719242.jpg b/parse/dev/ITw9edRDlD/images/9f821e06113838668f620fece0360ab9f0a920b262835347b401956be4719242.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db113cc69cf2ab63f3a2aef6090ec81a93e99c60 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/9f821e06113838668f620fece0360ab9f0a920b262835347b401956be4719242.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4be2cbb6aea87a8d07dee60e1bdca933b496c819c5caeda1c03f6c62a69c058c +size 74452 diff --git a/parse/dev/ITw9edRDlD/images/a17689acc3ce9e552ab8b7a81583f8a3ebe58973b339b735aaa2567308af5ee4.jpg b/parse/dev/ITw9edRDlD/images/a17689acc3ce9e552ab8b7a81583f8a3ebe58973b339b735aaa2567308af5ee4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1b2c57ae5290c94b651675e6cd4ff7b3732f41d7 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/a17689acc3ce9e552ab8b7a81583f8a3ebe58973b339b735aaa2567308af5ee4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13cb6eeee5fa77db2d3525aa95cafe9697d312e358c0196eb3c64cb45395f26a +size 13222 diff --git a/parse/dev/ITw9edRDlD/images/aeaa81388b37adea8775476623a02dd5dc3201b4f674b5d84db994bf94e56554.jpg b/parse/dev/ITw9edRDlD/images/aeaa81388b37adea8775476623a02dd5dc3201b4f674b5d84db994bf94e56554.jpg new file mode 100644 index 0000000000000000000000000000000000000000..65c3c64156f2fd726c1e1ad0ab7a014066cd31fd --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/aeaa81388b37adea8775476623a02dd5dc3201b4f674b5d84db994bf94e56554.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c097edca4908558ff75582236ba0f94dae1cd33846388d3d35e1a1431a4f485f +size 35392 diff --git a/parse/dev/ITw9edRDlD/images/b232dac27063b9946e0f425e9311d8f3af48091cb9b2c8555faa06ef2edfb1e7.jpg b/parse/dev/ITw9edRDlD/images/b232dac27063b9946e0f425e9311d8f3af48091cb9b2c8555faa06ef2edfb1e7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e7f1d4c129066d5fa077c864435b51573e35ed8c --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/b232dac27063b9946e0f425e9311d8f3af48091cb9b2c8555faa06ef2edfb1e7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e73c03bb5ca1e3cfc556c58eadc7771a5cbb5ef1461bdc8e8c72adf36e682019 +size 43358 diff --git a/parse/dev/ITw9edRDlD/images/bbb64d43dfd754116f39b15b55fbaf9e5565772cd7c477a77fef3cabb9c1d7e0.jpg b/parse/dev/ITw9edRDlD/images/bbb64d43dfd754116f39b15b55fbaf9e5565772cd7c477a77fef3cabb9c1d7e0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..96687ff9ca8d986b954a4efd5080ff7afc1882d7 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/bbb64d43dfd754116f39b15b55fbaf9e5565772cd7c477a77fef3cabb9c1d7e0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b587c5b3e87db91f7a2c0a46c1029170fb0c11902fe99ac237b9c35bce640ad6 +size 89735 diff --git a/parse/dev/ITw9edRDlD/images/d10024623233c36dfd850f5da05c6889fbd52b7b00b95c5a9183a6c3f28e8126.jpg b/parse/dev/ITw9edRDlD/images/d10024623233c36dfd850f5da05c6889fbd52b7b00b95c5a9183a6c3f28e8126.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1cae33e9352573794a6cd3f8092959f7ca7bb4ea --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/d10024623233c36dfd850f5da05c6889fbd52b7b00b95c5a9183a6c3f28e8126.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9e7f7df515036392dd8707472da965637a7ff38810432f46ec2a4185a38912b +size 8295 diff --git a/parse/dev/ITw9edRDlD/images/d12d7e0942eee875ba02e247fdac2ce86743247347c9c86b6be84199e00a15af.jpg b/parse/dev/ITw9edRDlD/images/d12d7e0942eee875ba02e247fdac2ce86743247347c9c86b6be84199e00a15af.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0bc35051ead5aac6711efa4daae01b1ece6f60b4 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/d12d7e0942eee875ba02e247fdac2ce86743247347c9c86b6be84199e00a15af.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:508c10a8755bb931d12fe0ad5b34a19c5fed33df378145e1db010b92584dd001 +size 5337 diff --git a/parse/dev/ITw9edRDlD/images/d16bd2849e775991762208e8b49a2f7a38216d53692d7d4f7b27279d9ad989af.jpg b/parse/dev/ITw9edRDlD/images/d16bd2849e775991762208e8b49a2f7a38216d53692d7d4f7b27279d9ad989af.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d4a17c5c5662c19de626bbbfb42e029c2e3773f6 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/d16bd2849e775991762208e8b49a2f7a38216d53692d7d4f7b27279d9ad989af.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58f23227c7ffe661b83ea98a8cec855d00565f1a469433b3b1637fa6b3e50841 +size 10860 diff --git a/parse/dev/ITw9edRDlD/images/e0243250ca9c28b579b35ffb406027eb4dd45b7f519cc9e46cf77b8cc90342d9.jpg b/parse/dev/ITw9edRDlD/images/e0243250ca9c28b579b35ffb406027eb4dd45b7f519cc9e46cf77b8cc90342d9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ced2421929cad605de35c54f5932f16e6b657491 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/e0243250ca9c28b579b35ffb406027eb4dd45b7f519cc9e46cf77b8cc90342d9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ceccefcea4b3c7862146b166325643693f725ff20a7a15ad708ba02414fcbf4 +size 12780 diff --git a/parse/dev/ITw9edRDlD/images/e3f92f76fe9724d253565131e8cb279499fb11c062052e48ec6575ba7683ccb7.jpg b/parse/dev/ITw9edRDlD/images/e3f92f76fe9724d253565131e8cb279499fb11c062052e48ec6575ba7683ccb7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..70beb8b75065de41b6ec53275cafb6160193a938 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/e3f92f76fe9724d253565131e8cb279499fb11c062052e48ec6575ba7683ccb7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c12f3079fe232e0279ed92bb59865612c4437cb1f10c7ab825f39106e4c8f019 +size 75404 diff --git a/parse/dev/ITw9edRDlD/images/e40144d99c5950df6256a55196ba929f579447f60232c87be728623c6f571c7d.jpg b/parse/dev/ITw9edRDlD/images/e40144d99c5950df6256a55196ba929f579447f60232c87be728623c6f571c7d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4301f5394da3be0eb3dd653d1bb4d205f7e8f475 --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/e40144d99c5950df6256a55196ba929f579447f60232c87be728623c6f571c7d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bb835632c6747a40d4f0dc8f01be011ea90efb32d3ffc4f42dcc6c24ecb91de +size 21532 diff --git a/parse/dev/ITw9edRDlD/images/f69b19a9a3642de63e2e396a4a87ed592deb194782e67f78c0e5072ad4656e8b.jpg b/parse/dev/ITw9edRDlD/images/f69b19a9a3642de63e2e396a4a87ed592deb194782e67f78c0e5072ad4656e8b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8dcd31c8b46d8f368d2d8ea7dc5a5176b3feb38e --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/f69b19a9a3642de63e2e396a4a87ed592deb194782e67f78c0e5072ad4656e8b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:476eda5da08b2455473223aa2537919be65f2854c36eea985b3f1976d2008b5d +size 51401 diff --git a/parse/dev/ITw9edRDlD/images/fd58c559a7dbe1dd65712e151478cf16dcab40d1f576cbcec179385137efbd9a.jpg b/parse/dev/ITw9edRDlD/images/fd58c559a7dbe1dd65712e151478cf16dcab40d1f576cbcec179385137efbd9a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5b85e7a9596d9da2a3dc32291c8ebd4e3137ab7a --- /dev/null +++ b/parse/dev/ITw9edRDlD/images/fd58c559a7dbe1dd65712e151478cf16dcab40d1f576cbcec179385137efbd9a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7d8ed97bc7877c138132c5def997fcd63eae16e1ed4ee1a845b0da0932e1e5e +size 15136 diff --git a/parse/dev/LKEYuYNOqx/LKEYuYNOqx_model.json b/parse/dev/LKEYuYNOqx/LKEYuYNOqx_model.json new file mode 100644 index 0000000000000000000000000000000000000000..0b6297e459326666c73e7d14be1bf7eb9ff34b50 --- /dev/null +++ b/parse/dev/LKEYuYNOqx/LKEYuYNOqx_model.json @@ -0,0 +1,23470 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 721, + 1304, + 721, + 1304, + 1269, + 398, + 1269 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1396, + 1406, + 1396, + 1406, + 1761, + 298, + 1761 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 299, + 1775, + 1403, + 1775, + 1403, + 1928, + 299, + 1928 + ], + "score": 0.968 + }, + { + "category_id": 2, + "poly": [ + 333, + 1951, + 757, + 1951, + 757, + 1979, + 333, + 1979 + ], + "score": 0.914 + }, + { + "category_id": 0, + "poly": [ + 299, + 1325, + 530, + 1325, + 530, + 1364, + 299, + 1364 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 323, + 270, + 1379, + 270, + 1379, + 328, + 323, + 328 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 298, + 2033, + 1071, + 2033, + 1071, + 2062, + 298, + 2062 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 787, + 650, + 912, + 650, + 912, + 686, + 787, + 686 + ], + "score": 0.885 + }, + { + "category_id": 1, + "poly": [ + 439, + 439, + 1256, + 439, + 1256, + 477, + 439, + 477 + ], + "score": 0.481 + }, + { + "category_id": 1, + "poly": [ + 561, + 547, + 1142, + 547, + 1142, + 570, + 561, + 570 + ], + "score": 0.409 + }, + { + "category_id": 1, + "poly": [ + 542, + 504, + 1160, + 504, + 1160, + 537, + 542, + 537 + ], + "score": 0.374 + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1946.0, + 760.0, + 1946.0, + 760.0, + 1986.0, + 329.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1322.0, + 536.0, + 1322.0, + 536.0, + 1370.0, + 292.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 271.0, + 1378.0, + 271.0, + 1378.0, + 329.0, + 322.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2028.0, + 1073.0, + 2028.0, + 1073.0, + 2066.0, + 295.0, + 2066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 647.0, + 920.0, + 647.0, + 920.0, + 691.0, + 782.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 722.0, + 1306.0, + 722.0, + 1306.0, + 755.0, + 395.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 750.0, + 1308.0, + 750.0, + 1308.0, + 786.0, + 391.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 782.0, + 1305.0, + 782.0, + 1305.0, + 815.0, + 394.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 813.0, + 1305.0, + 813.0, + 1305.0, + 845.0, + 394.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 843.0, + 1309.0, + 843.0, + 1309.0, + 879.0, + 394.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 874.0, + 1307.0, + 874.0, + 1307.0, + 905.0, + 394.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 903.0, + 1305.0, + 903.0, + 1305.0, + 936.0, + 394.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 934.0, + 1306.0, + 934.0, + 1306.0, + 967.0, + 394.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 966.0, + 1304.0, + 966.0, + 1304.0, + 996.0, + 395.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 995.0, + 1308.0, + 995.0, + 1308.0, + 1026.0, + 393.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1024.0, + 1306.0, + 1024.0, + 1306.0, + 1058.0, + 393.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1053.0, + 1305.0, + 1053.0, + 1305.0, + 1091.0, + 394.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1086.0, + 1307.0, + 1086.0, + 1307.0, + 1116.0, + 393.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1115.0, + 1306.0, + 1115.0, + 1306.0, + 1148.0, + 392.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1145.0, + 1305.0, + 1145.0, + 1305.0, + 1178.0, + 393.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1175.0, + 1304.0, + 1175.0, + 1304.0, + 1208.0, + 393.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1205.0, + 1305.0, + 1205.0, + 1305.0, + 1241.0, + 394.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1237.0, + 1253.0, + 1237.0, + 1253.0, + 1271.0, + 392.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1396.0, + 1403.0, + 1396.0, + 1403.0, + 1430.0, + 296.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1426.0, + 1403.0, + 1426.0, + 1403.0, + 1460.0, + 295.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1456.0, + 1406.0, + 1456.0, + 1406.0, + 1493.0, + 293.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1485.0, + 1408.0, + 1485.0, + 1408.0, + 1522.0, + 295.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1516.0, + 1406.0, + 1516.0, + 1406.0, + 1552.0, + 293.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1546.0, + 1406.0, + 1546.0, + 1406.0, + 1583.0, + 292.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1576.0, + 1404.0, + 1576.0, + 1404.0, + 1612.0, + 292.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1608.0, + 1406.0, + 1608.0, + 1406.0, + 1643.0, + 292.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1639.0, + 1406.0, + 1639.0, + 1406.0, + 1673.0, + 293.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1667.0, + 1408.0, + 1667.0, + 1408.0, + 1706.0, + 293.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1700.0, + 1409.0, + 1700.0, + 1409.0, + 1733.0, + 295.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1730.0, + 1224.0, + 1730.0, + 1224.0, + 1764.0, + 295.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1774.0, + 1407.0, + 1774.0, + 1407.0, + 1809.0, + 293.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1803.0, + 1407.0, + 1803.0, + 1407.0, + 1841.0, + 293.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1836.0, + 1407.0, + 1836.0, + 1407.0, + 1873.0, + 294.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1867.0, + 1404.0, + 1867.0, + 1404.0, + 1900.0, + 297.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1896.0, + 1405.0, + 1896.0, + 1405.0, + 1932.0, + 294.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 440.0, + 681.0, + 440.0, + 681.0, + 477.0, + 441.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 439.0, + 961.0, + 439.0, + 961.0, + 479.0, + 777.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 440.0, + 1255.0, + 440.0, + 1255.0, + 479.0, + 1070.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 545.0, + 1145.0, + 545.0, + 1145.0, + 575.0, + 556.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 499.0, + 1165.0, + 499.0, + 1165.0, + 544.0, + 538.0, + 544.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 561, + 1406, + 561, + 1406, + 958, + 298, + 958 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1165, + 1405, + 1165, + 1405, + 1382, + 297, + 1382 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1536, + 1404, + 1536, + 1404, + 1742, + 297, + 1742 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 969, + 1404, + 969, + 1404, + 1155, + 297, + 1155 + ], + "score": 0.981 + }, + { + "category_id": 3, + "poly": [ + 346, + 126, + 1337, + 126, + 1337, + 419, + 346, + 419 + ], + "score": 0.973 + }, + { + "category_id": 4, + "poly": [ + 296, + 438, + 1407, + 438, + 1407, + 551, + 296, + 551 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 295, + 1803, + 1404, + 1803, + 1404, + 1868, + 295, + 1868 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 667, + 1493, + 1032, + 1493, + 1032, + 1531, + 667, + 1531 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 674, + 1751, + 1024, + 1751, + 1024, + 1793, + 674, + 1793 + ], + "score": 0.937 + }, + { + "category_id": 8, + "poly": [ + 636, + 1963, + 1064, + 1963, + 1064, + 2003, + 636, + 2003 + ], + "score": 0.931 + }, + { + "category_id": 0, + "poly": [ + 298, + 1403, + 527, + 1403, + 527, + 1439, + 298, + 1439 + ], + "score": 0.926 + }, + { + "category_id": 8, + "poly": [ + 560, + 1878, + 1142, + 1878, + 1142, + 1919, + 560, + 1919 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 293, + 1452, + 1315, + 1452, + 1315, + 1486, + 293, + 1486 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 304, + 1923, + 1335, + 1923, + 1335, + 1957, + 304, + 1957 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1498, + 1399, + 1498, + 1399, + 1527, + 1366, + 1527 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1758, + 1400, + 1758, + 1400, + 1787, + 1366, + 1787 + ], + "score": 0.874 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1968, + 1401, + 1968, + 1401, + 1996, + 1365, + 1996 + ], + "score": 0.867 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1884, + 1400, + 1884, + 1400, + 1913, + 1366, + 1913 + ], + "score": 0.861 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.755 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 859, + 2061, + 859, + 2085, + 840, + 2085 + ], + "score": 0.111 + }, + { + "category_id": 14, + "poly": [ + 674, + 1747, + 1024, + 1747, + 1024, + 1794, + 674, + 1794 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { d \\mathbf { x } _ { t } = - \\frac { 1 } { 2 } \\beta _ { t } \\mathbf { x } _ { t } d t + \\sqrt { \\beta _ { t } } d \\mathbf { w } _ { t } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 531, + 1677, + 678, + 1677, + 678, + 1712, + 531, + 1712 + ], + "score": 0.93, + "latex": "\\sigma _ { t } ^ { 2 } = 1 - \\alpha _ { t } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 374, + 1537, + 528, + 1537, + 528, + 1570, + 374, + 1570 + ], + "score": 0.93, + "latex": "\\mathbf { x } _ { 0 } \\sim p _ { 0 } ( \\mathbf { x } _ { 0 } )" + }, + { + "category_id": 14, + "poly": [ + 634, + 1961, + 1063, + 1961, + 1063, + 2003, + 634, + 2003 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { d \\mathbf { x } _ { t } = - \\frac { 1 } { 2 } \\beta _ { t } \\left[ \\mathbf { x } _ { t } + \\nabla _ { \\mathbf { x } _ { t } } \\log p _ { t } ( \\mathbf { x } _ { t } ) \\right] d t , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 477, + 1570, + 578, + 1570, + 578, + 1602, + 477, + 1602 + ], + "score": 0.92, + "latex": "t \\in [ 0 , 1 ]" + }, + { + "category_id": 14, + "poly": [ + 668, + 1491, + 1028, + 1491, + 1028, + 1531, + 668, + 1531 + ], + "score": 0.91, + "latex": "p _ { t } ( \\mathbf { x } _ { t } | \\mathbf { x } _ { 0 } ) = \\mathcal { N } ( \\mathbf { x } _ { t } ; \\alpha _ { t } \\mathbf { x } _ { 0 } , \\sigma _ { t } ^ { 2 } I ) ," + }, + { + "category_id": 14, + "poly": [ + 558, + 1874, + 1139, + 1874, + 1139, + 1919, + 558, + 1919 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { d \\mathbf { x } _ { t } = - \\frac { 1 } { 2 } \\beta _ { t } \\left[ \\mathbf { x } _ { t } + 2 \\nabla _ { \\mathbf { x } _ { t } } \\log p _ { t } ( \\mathbf { x } _ { t } ) \\right] d t + \\sqrt { \\beta _ { t } } d \\mathbf { w } _ { t } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 686, + 1644, + 927, + 1644, + 927, + 1682, + 686, + 1682 + ], + "score": 0.9, + "latex": "p _ { 1 } ( \\mathbf { x } _ { 1 } ) \\approx \\mathcal { N } ( \\mathbf { x } _ { 1 } ; \\mathbf { 0 } , \\bar { I } )" + }, + { + "category_id": 13, + "poly": [ + 570, + 1804, + 714, + 1804, + 714, + 1838, + 570, + 1838 + ], + "score": 0.89, + "latex": "\\mathbf { x } _ { 0 } \\sim p _ { 0 } ( \\mathbf { x } _ { 0 } )" + }, + { + "category_id": 13, + "poly": [ + 765, + 1808, + 801, + 1808, + 801, + 1835, + 765, + 1835 + ], + "score": 0.88, + "latex": "\\mathbf { w } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1056, + 1573, + 1087, + 1573, + 1087, + 1599, + 1056, + 1599 + ], + "score": 0.87, + "latex": "\\alpha _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1109, + 1542, + 1139, + 1542, + 1139, + 1568, + 1109, + 1568 + ], + "score": 0.86, + "latex": "\\mathbf { x } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1137, + 1573, + 1166, + 1573, + 1166, + 1599, + 1137, + 1599 + ], + "score": 0.85, + "latex": "\\sigma _ { t }" + }, + { + "category_id": 13, + "poly": [ + 371, + 1801, + 557, + 1801, + 557, + 1841, + 371, + 1841 + ], + "score": 0.85, + "latex": "\\begin{array} { r } { \\beta _ { t } = - \\frac { d } { d t } \\log \\alpha _ { t } ^ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1252, + 1613, + 1266, + 1613, + 1266, + 1637, + 1252, + 1637 + ], + "score": 0.75, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 891, + 1599, + 923, + 1599, + 923, + 1650, + 891, + 1650 + ], + "score": 0.6, + "latex": "\\frac { \\alpha _ { t } ^ { 2 } } { \\sigma _ { t } ^ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 848, + 1598, + 923, + 1598, + 923, + 1650, + 848, + 1650 + ], + "score": 0.53, + "latex": "\\begin{array} { r } { \\log { \\frac { \\alpha _ { t } ^ { 2 } } { \\sigma _ { t } ^ { 2 } } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 853, + 394, + 915, + 394, + 915, + 418, + 853, + 418 + ], + "score": 0.34, + "latex": "p _ { 1 } ( \\mathbf { x } _ { 1 } )" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 327.0, + 1311.0, + 327.0, + 1311.0, + 355.0, + 1079.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 347.0, + 1244.0, + 347.0, + 1244.0, + 379.0, + 1079.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 372.0, + 1327.0, + 372.0, + 1327.0, + 400.0, + 1079.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 389.0, + 437.0, + 389.0, + 437.0, + 422.0, + 363.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 391.0, + 805.0, + 391.0, + 805.0, + 419.0, + 545.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 387.0, + 852.0, + 387.0, + 852.0, + 421.0, + 847.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 387.0, + 921.0, + 387.0, + 921.0, + 421.0, + 916.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 395.0, + 1269.0, + 395.0, + 1269.0, + 423.0, + 1080.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 435.0, + 1407.0, + 435.0, + 1407.0, + 473.0, + 294.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 465.0, + 1408.0, + 465.0, + 1408.0, + 498.0, + 295.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 493.0, + 1405.0, + 493.0, + 1405.0, + 526.0, + 295.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 520.0, + 1404.0, + 520.0, + 1404.0, + 553.0, + 294.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1397.0, + 530.0, + 1397.0, + 530.0, + 1449.0, + 289.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 863.0, + 2058.0, + 863.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 563.0, + 1404.0, + 563.0, + 1404.0, + 595.0, + 296.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 593.0, + 1403.0, + 593.0, + 1403.0, + 625.0, + 295.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 621.0, + 1406.0, + 621.0, + 1406.0, + 659.0, + 293.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 648.0, + 1409.0, + 648.0, + 1409.0, + 691.0, + 292.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 682.0, + 1409.0, + 682.0, + 1409.0, + 718.0, + 293.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 713.0, + 1403.0, + 713.0, + 1403.0, + 745.0, + 296.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 744.0, + 1404.0, + 744.0, + 1404.0, + 779.0, + 293.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 771.0, + 1404.0, + 771.0, + 1404.0, + 809.0, + 293.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 804.0, + 1406.0, + 804.0, + 1406.0, + 840.0, + 295.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 833.0, + 1407.0, + 833.0, + 1407.0, + 872.0, + 292.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 858.0, + 1407.0, + 858.0, + 1407.0, + 905.0, + 291.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 895.0, + 1403.0, + 895.0, + 1403.0, + 928.0, + 296.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 924.0, + 1399.0, + 924.0, + 1399.0, + 961.0, + 292.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1167.0, + 1406.0, + 1167.0, + 1406.0, + 1202.0, + 295.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1198.0, + 1406.0, + 1198.0, + 1406.0, + 1233.0, + 294.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1229.0, + 1405.0, + 1229.0, + 1405.0, + 1264.0, + 294.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1259.0, + 1406.0, + 1259.0, + 1406.0, + 1293.0, + 294.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1288.0, + 1405.0, + 1288.0, + 1405.0, + 1323.0, + 295.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1321.0, + 1405.0, + 1321.0, + 1405.0, + 1353.0, + 292.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1350.0, + 1363.0, + 1350.0, + 1363.0, + 1384.0, + 294.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1537.0, + 373.0, + 1537.0, + 373.0, + 1571.0, + 296.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 1537.0, + 1108.0, + 1537.0, + 1108.0, + 1571.0, + 529.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1537.0, + 1402.0, + 1537.0, + 1402.0, + 1571.0, + 1140.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1567.0, + 476.0, + 1567.0, + 476.0, + 1603.0, + 294.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 1567.0, + 1055.0, + 1567.0, + 1055.0, + 1603.0, + 579.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1567.0, + 1136.0, + 1567.0, + 1136.0, + 1603.0, + 1088.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 1567.0, + 1404.0, + 1567.0, + 1404.0, + 1603.0, + 1167.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1602.0, + 847.0, + 1602.0, + 847.0, + 1648.0, + 286.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1602.0, + 1251.0, + 1602.0, + 1251.0, + 1648.0, + 924.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 1602.0, + 1411.0, + 1602.0, + 1411.0, + 1648.0, + 1267.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1649.0, + 685.0, + 1649.0, + 685.0, + 1682.0, + 296.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1649.0, + 1404.0, + 1649.0, + 1404.0, + 1682.0, + 928.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1676.0, + 530.0, + 1676.0, + 530.0, + 1713.0, + 294.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1676.0, + 1405.0, + 1676.0, + 1405.0, + 1713.0, + 679.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1708.0, + 1405.0, + 1708.0, + 1405.0, + 1745.0, + 294.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 1601.5, + 922.0, + 1601.5, + 922.0, + 1609.5, + 902.0, + 1609.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 971.0, + 1406.0, + 971.0, + 1406.0, + 1003.0, + 297.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 999.0, + 1406.0, + 999.0, + 1406.0, + 1035.0, + 295.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1031.0, + 1406.0, + 1031.0, + 1406.0, + 1067.0, + 294.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1059.0, + 1405.0, + 1059.0, + 1405.0, + 1096.0, + 294.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1089.0, + 1406.0, + 1089.0, + 1406.0, + 1127.0, + 294.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1121.0, + 1324.0, + 1121.0, + 1324.0, + 1156.0, + 295.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1787.0, + 370.0, + 1787.0, + 370.0, + 1872.0, + 284.0, + 1872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1787.0, + 569.0, + 1787.0, + 569.0, + 1872.0, + 558.0, + 1872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1787.0, + 764.0, + 1787.0, + 764.0, + 1872.0, + 715.0, + 1872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 1787.0, + 1415.0, + 1787.0, + 1415.0, + 1872.0, + 802.0, + 1872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1447.0, + 1321.0, + 1447.0, + 1321.0, + 1493.0, + 292.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1922.0, + 1338.0, + 1922.0, + 1338.0, + 1961.0, + 298.0, + 1961.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1755, + 1406, + 1755, + 1406, + 2004, + 296, + 2004 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1006, + 1405, + 1006, + 1405, + 1138, + 297, + 1138 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1230, + 1405, + 1230, + 1405, + 1324, + 298, + 1324 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1420, + 1404, + 1420, + 1404, + 1514, + 298, + 1514 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 200, + 1405, + 200, + 1405, + 328, + 298, + 328 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 338, + 1405, + 338, + 1405, + 456, + 297, + 456 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 514, + 1668, + 1181, + 1668, + 1181, + 1747, + 514, + 1747 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 571, + 1527, + 1127, + 1527, + 1127, + 1602, + 571, + 1602 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 521, + 1150, + 1175, + 1150, + 1175, + 1220, + 521, + 1220 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 561, + 1338, + 1136, + 1338, + 1136, + 1406, + 561, + 1406 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 300, + 558, + 1405, + 558, + 1405, + 635, + 300, + 635 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 449, + 466, + 1250, + 466, + 1250, + 546, + 449, + 546 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 301, + 857, + 1416, + 857, + 1416, + 935, + 301, + 935 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 730, + 647, + 968, + 647, + 968, + 684, + 730, + 684 + ], + "score": 0.938 + }, + { + "category_id": 0, + "poly": [ + 297, + 953, + 936, + 953, + 936, + 994, + 297, + 994 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 296, + 1613, + 894, + 1613, + 894, + 1656, + 296, + 1656 + ], + "score": 0.922 + }, + { + "category_id": 8, + "poly": [ + 381, + 794, + 1308, + 794, + 1308, + 840, + 381, + 840 + ], + "score": 0.916 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1688, + 1400, + 1688, + 1400, + 1720, + 1351, + 1720 + ], + "score": 0.907 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1548, + 1400, + 1548, + 1400, + 1580, + 1351, + 1580 + ], + "score": 0.9 + }, + { + "category_id": 1, + "poly": [ + 299, + 698, + 1409, + 698, + 1409, + 783, + 299, + 783 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1170, + 1400, + 1170, + 1400, + 1200, + 1365, + 1200 + ], + "score": 0.879 + }, + { + "category_id": 9, + "poly": [ + 1365, + 650, + 1400, + 650, + 1400, + 679, + 1365, + 679 + ], + "score": 0.873 + }, + { + "category_id": 9, + "poly": [ + 1365, + 487, + 1401, + 487, + 1401, + 517, + 1365, + 517 + ], + "score": 0.867 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1356, + 1400, + 1356, + 1400, + 1387, + 1365, + 1387 + ], + "score": 0.862 + }, + { + "category_id": 9, + "poly": [ + 1365, + 800, + 1401, + 800, + 1401, + 831, + 1365, + 831 + ], + "score": 0.857 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.665 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 860, + 2061, + 860, + 2085, + 841, + 2085 + ], + "score": 0.496 + }, + { + "category_id": 14, + "poly": [ + 451, + 465, + 1248, + 465, + 1248, + 545, + 451, + 545 + ], + "score": 0.94, + "latex": "\\frac { d \\bar { \\mathbf { x } } _ { t } } { d \\gamma _ { t } } = \\sqrt { 1 + \\gamma _ { t } ^ { 2 } } \\frac { d \\mathbf { x } _ { t } } { d t } \\frac { d t } { d \\gamma _ { t } } + \\mathbf { x } _ { t } \\frac { \\gamma _ { t } } { \\sqrt { 1 + \\gamma _ { t } ^ { 2 } } } = - \\frac { \\gamma _ { t } } { \\sqrt { 1 + \\gamma _ { t } ^ { 2 } } } \\nabla _ { \\mathbf { x } _ { t } } \\log p _ { t } ( \\mathbf { x } _ { t } ) ," + }, + { + "category_id": 13, + "poly": [ + 740, + 1420, + 898, + 1420, + 898, + 1476, + 740, + 1476 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\gamma _ { t } = \\sqrt { \\frac { 1 - \\alpha _ { t } ^ { 2 } } { \\alpha _ { t } ^ { 2 } } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 803, + 403, + 954, + 403, + 954, + 456, + 803, + 456 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\gamma _ { t } = \\sqrt { \\frac { 1 - \\alpha _ { t } ^ { 2 } } { \\alpha _ { t } ^ { 2 } } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 465, + 697, + 754, + 697, + 754, + 753, + 465, + 753 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\sigma _ { t } = \\sqrt { 1 - \\alpha _ { t } ^ { 2 } } = \\frac { \\gamma _ { t } } { \\sqrt { \\gamma _ { t } ^ { 2 } + 1 } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 671, + 890, + 798, + 890, + 798, + 937, + 671, + 937 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { g ( t ) = \\frac { \\beta _ { t } } { 2 \\sigma _ { t } ^ { 2 } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 742, + 1619, + 841, + 1619, + 841, + 1653, + 742, + 1653 + ], + "score": 0.94, + "latex": "\\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t )" + }, + { + "category_id": 13, + "poly": [ + 783, + 559, + 968, + 559, + 968, + 606, + 783, + 606 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\beta ( t ) \\frac { d t } { d \\gamma _ { t } } = \\frac { 2 \\gamma _ { t } } { \\gamma _ { t } ^ { 2 } + 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 373, + 1613, + 470, + 1613, + 470, + 1657, + 373, + 1657 + ], + "score": 0.93, + "latex": "\\frac { \\partial \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\partial \\mathbf { x } _ { t } }" + }, + { + "category_id": 14, + "poly": [ + 516, + 1666, + 1184, + 1666, + 1184, + 1748, + 516, + 1748 + ], + "score": 0.93, + "latex": "\\frac { d \\mathbf { x } _ { t } } { d \\gamma _ { t } } = \\frac { \\partial \\mathbf { x } _ { t } } { \\partial \\bar { \\mathbf { x } } _ { t } } \\frac { d \\bar { \\mathbf { x } } _ { t } } { d \\gamma _ { t } } + \\frac { \\partial \\mathbf { x } _ { t } } { \\partial \\gamma _ { t } } = \\frac { 1 } { \\sqrt { \\gamma _ { t } ^ { 2 } + 1 } } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) - \\frac { \\gamma _ { t } } { 1 + \\gamma _ { t } ^ { 2 } } \\mathbf { x } _ { t } ." + }, + { + "category_id": 13, + "poly": [ + 849, + 1473, + 998, + 1473, + 998, + 1517, + 849, + 1517 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { d _ { \\gamma _ { t } } \\epsilon _ { \\theta } : = \\frac { d \\epsilon _ { \\theta } } { d \\gamma _ { t } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 841, + 1099, + 989, + 1099, + 989, + 1140, + 841, + 1140 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\frac { d \\mathbf { y } } { d t } = \\pmb { f } ( \\mathbf { y } , t ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 372, + 203, + 538, + 203, + 538, + 237, + 372, + 237 + ], + "score": 0.93, + "latex": "\\nabla _ { \\mathbf { x } _ { t } } \\log p _ { t } ( \\mathbf { x } _ { t } )" + }, + { + "category_id": 14, + "poly": [ + 523, + 1149, + 1173, + 1149, + 1173, + 1222, + 523, + 1222 + ], + "score": 0.93, + "latex": "\\mathbf { y } _ { t _ { n + 1 } } = \\mathbf { y } _ { t _ { n } } + h _ { n } { \\frac { d \\mathbf { y } } { d t } } | _ { ( \\mathbf { y } _ { t _ { n } } , t _ { n } ) } + \\cdot \\cdot \\cdot + { \\frac { 1 } { p ! } } h _ { n } ^ { p } { \\frac { d ^ { p } \\mathbf { y } } { d t ^ { p } } } | _ { ( \\mathbf { y } _ { t _ { n } } , t _ { n } ) } ," + }, + { + "category_id": 13, + "poly": [ + 374, + 749, + 474, + 749, + 474, + 783, + 374, + 783 + ], + "score": 0.93, + "latex": "\\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t )" + }, + { + "category_id": 14, + "poly": [ + 572, + 1526, + 1125, + 1526, + 1125, + 1601, + 572, + 1601 + ], + "score": 0.93, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) = \\frac { \\partial \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\partial \\mathbf { x } _ { t } } \\frac { d \\mathbf { x } _ { t } } { d \\gamma _ { t } } + \\frac { \\partial \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\partial t } \\frac { d t } { d \\gamma _ { t } } ," + }, + { + "category_id": 13, + "poly": [ + 1006, + 406, + 1203, + 406, + 1203, + 447, + 1006, + 447 + ], + "score": 0.93, + "latex": "\\bar { \\mathbf { x } } _ { t } = \\mathbf { x } _ { t } \\sqrt { 1 + \\gamma _ { t } ^ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 380, + 264, + 582, + 264, + 582, + 297, + 380, + 297 + ], + "score": 0.93, + "latex": "\\mathbf { x } _ { 1 } \\sim \\mathcal { N } ( \\mathbf { x } _ { 1 } ; \\mathbf { 0 } , I )" + }, + { + "category_id": 13, + "poly": [ + 374, + 1434, + 586, + 1434, + 586, + 1469, + 374, + 1469 + ], + "score": 0.92, + "latex": "h _ { n } = \\gamma _ { t _ { n + 1 } } - \\gamma _ { t _ { n } }" + }, + { + "category_id": 13, + "poly": [ + 942, + 858, + 1041, + 858, + 1041, + 893, + 942, + 893 + ], + "score": 0.92, + "latex": "g ( t ) = 1" + }, + { + "category_id": 14, + "poly": [ + 562, + 1336, + 1134, + 1336, + 1134, + 1408, + 562, + 1408 + ], + "score": 0.92, + "latex": "\\bar { \\mathbf { x } } _ { t _ { n + 1 } } = \\bar { \\mathbf { x } } _ { t _ { n } } + h _ { n } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t _ { n } } , t _ { n } ) + \\frac { 1 } { 2 } h _ { n } ^ { 2 } \\frac { d \\epsilon _ { \\theta } } { d \\gamma _ { t } } | _ { ( \\mathbf { x } _ { t _ { n } } , t _ { n } ) } ," + }, + { + "category_id": 13, + "poly": [ + 1070, + 558, + 1318, + 558, + 1318, + 603, + 1070, + 603 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { s _ { \\theta } ( \\mathbf { x } _ { t } , t ) : = - \\frac { \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\sigma _ { t } } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 730, + 646, + 967, + 646, + 967, + 684, + 730, + 684 + ], + "score": 0.92, + "latex": "d \\bar { \\mathbf { x } } _ { t } = \\epsilon _ { \\theta } \\left( \\mathbf { x } _ { t } , t \\right) d \\gamma _ { t } ," + }, + { + "category_id": 13, + "poly": [ + 1073, + 294, + 1375, + 294, + 1375, + 329, + 1073, + 329 + ], + "score": 0.91, + "latex": "s _ { \\theta } ( \\mathbf { x } _ { t } , t ) \\approx \\nabla _ { \\mathbf { x } _ { t } } \\log p _ { t } ( \\mathbf { x } _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 372, + 1233, + 555, + 1233, + 555, + 1265, + 372, + 1265 + ], + "score": 0.91, + "latex": "h _ { n } = t _ { n + 1 } - t _ { n }" + }, + { + "category_id": 13, + "poly": [ + 632, + 560, + 673, + 560, + 673, + 600, + 632, + 600 + ], + "score": 0.91, + "latex": "\\frac { d { \\bf x } _ { t } } { d t }" + }, + { + "category_id": 13, + "poly": [ + 1356, + 862, + 1403, + 862, + 1403, + 893, + 1356, + 893 + ], + "score": 0.9, + "latex": "g ( t )" + }, + { + "category_id": 13, + "poly": [ + 405, + 861, + 583, + 861, + 583, + 891, + 405, + 891 + ], + "score": 0.89, + "latex": "0 < t _ { \\mathrm { c u t o f f } } \\ll 1" + }, + { + "category_id": 14, + "poly": [ + 383, + 792, + 1309, + 792, + 1309, + 838, + 383, + 838 + ], + "score": 0.88, + "latex": "\\operatorname* { m i n } _ { \\pmb { \\theta } } \\mathbb { E } _ { t \\sim \\mathcal { U } [ t _ { \\mathrm { c u t o f f } } , 1 ] , \\mathbf { x } _ { 0 } \\sim p ( \\mathbf { x } _ { 0 } ) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } \\left[ g ( t ) \\lVert \\epsilon - \\epsilon _ { \\boldsymbol { \\theta } } ( \\mathbf { x } _ { t } , t ) \\rVert _ { 2 } ^ { 2 } \\right] , \\quad \\mathbf { x } _ { t } = \\alpha _ { t } \\mathbf { x } _ { 0 } + \\sigma _ { t } \\epsilon ," + }, + { + "category_id": 13, + "poly": [ + 1136, + 1438, + 1167, + 1438, + 1167, + 1463, + 1136, + 1463 + ], + "score": 0.84, + "latex": "\\alpha _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1041, + 1107, + 1059, + 1107, + 1059, + 1136, + 1041, + 1136 + ], + "score": 0.82, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1078, + 1073, + 1095, + 1073, + 1095, + 1101, + 1078, + 1101 + ], + "score": 0.81, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 813, + 1072, + 830, + 1072, + 830, + 1100, + 813, + 1100 + ], + "score": 0.8, + "latex": "p" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 950.0, + 941.0, + 950.0, + 941.0, + 1000.0, + 288.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 837.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1757.0, + 1403.0, + 1757.0, + 1403.0, + 1791.0, + 294.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1787.0, + 1406.0, + 1787.0, + 1406.0, + 1821.0, + 294.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1818.0, + 1406.0, + 1818.0, + 1406.0, + 1852.0, + 294.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1847.0, + 1406.0, + 1847.0, + 1406.0, + 1883.0, + 294.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1877.0, + 1407.0, + 1877.0, + 1407.0, + 1915.0, + 293.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1909.0, + 1406.0, + 1909.0, + 1406.0, + 1943.0, + 294.0, + 1943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1940.0, + 1404.0, + 1940.0, + 1404.0, + 1974.0, + 295.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1967.0, + 1261.0, + 1967.0, + 1261.0, + 2007.0, + 291.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1008.0, + 1405.0, + 1008.0, + 1405.0, + 1042.0, + 295.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1036.0, + 1405.0, + 1036.0, + 1405.0, + 1073.0, + 294.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1067.0, + 812.0, + 1067.0, + 812.0, + 1105.0, + 294.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1067.0, + 1077.0, + 1067.0, + 1077.0, + 1105.0, + 831.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1067.0, + 1406.0, + 1067.0, + 1406.0, + 1105.0, + 1096.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1097.0, + 840.0, + 1097.0, + 840.0, + 1143.0, + 292.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1097.0, + 1040.0, + 1097.0, + 1040.0, + 1143.0, + 990.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 1097.0, + 1259.0, + 1097.0, + 1259.0, + 1143.0, + 1060.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1230.0, + 371.0, + 1230.0, + 371.0, + 1267.0, + 293.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1230.0, + 1406.0, + 1230.0, + 1406.0, + 1267.0, + 556.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1260.0, + 1406.0, + 1260.0, + 1406.0, + 1298.0, + 293.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1289.0, + 721.0, + 1289.0, + 721.0, + 1327.0, + 295.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1417.0, + 373.0, + 1417.0, + 373.0, + 1474.0, + 292.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1417.0, + 739.0, + 1417.0, + 739.0, + 1474.0, + 587.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1417.0, + 904.0, + 1417.0, + 904.0, + 1474.0, + 899.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1431.0, + 1135.0, + 1431.0, + 1135.0, + 1465.0, + 908.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 1431.0, + 1404.0, + 1431.0, + 1404.0, + 1465.0, + 1168.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1471.0, + 848.0, + 1471.0, + 848.0, + 1516.0, + 292.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1475.0, + 1347.0, + 1475.0, + 1347.0, + 1519.0, + 999.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.25, + 1472.5, + 998.25, + 1472.5, + 998.25, + 1498.0, + 956.25, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 202.0, + 371.0, + 202.0, + 371.0, + 239.0, + 293.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 202.0, + 1408.0, + 202.0, + 1408.0, + 239.0, + 539.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 231.0, + 1407.0, + 231.0, + 1407.0, + 269.0, + 292.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 264.0, + 379.0, + 264.0, + 379.0, + 297.0, + 296.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 264.0, + 1405.0, + 264.0, + 1405.0, + 297.0, + 583.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 294.0, + 1072.0, + 294.0, + 1072.0, + 329.0, + 295.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 294.0, + 1386.0, + 294.0, + 1386.0, + 329.0, + 1376.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 337.0, + 1406.0, + 337.0, + 1406.0, + 374.0, + 295.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 366.0, + 1406.0, + 366.0, + 1406.0, + 405.0, + 292.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 406.0, + 802.0, + 406.0, + 802.0, + 452.0, + 292.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 397.0, + 1005.0, + 397.0, + 1005.0, + 458.0, + 955.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 397.0, + 1313.0, + 397.0, + 1313.0, + 458.0, + 1204.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 544.0, + 631.0, + 544.0, + 631.0, + 619.0, + 284.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 544.0, + 782.0, + 544.0, + 782.0, + 619.0, + 674.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 533.0, + 1069.0, + 533.0, + 1069.0, + 631.0, + 969.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 533.0, + 1406.0, + 533.0, + 1406.0, + 631.0, + 1319.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 596.0, + 1363.0, + 596.0, + 1363.0, + 639.0, + 291.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.25, + 576.0, + 873.25, + 576.0, + 873.25, + 605.0, + 831.25, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 854.0, + 404.0, + 854.0, + 404.0, + 896.0, + 294.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 854.0, + 941.0, + 854.0, + 941.0, + 896.0, + 584.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 854.0, + 1355.0, + 854.0, + 1355.0, + 896.0, + 1042.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 889.0, + 670.0, + 889.0, + 670.0, + 939.0, + 294.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 889.0, + 1337.0, + 889.0, + 1337.0, + 939.0, + 799.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1615.0, + 372.0, + 1615.0, + 372.0, + 1650.0, + 293.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1631.0, + 446.0, + 1631.0, + 446.0, + 1665.0, + 398.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1608.0, + 895.0, + 1608.0, + 895.0, + 1657.0, + 365.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 693.0, + 464.0, + 693.0, + 464.0, + 750.0, + 282.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 699.0, + 1405.0, + 699.0, + 1405.0, + 738.0, + 755.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 715.0, + 762.0, + 715.0, + 762.0, + 758.0, + 755.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 749.0, + 373.0, + 749.0, + 373.0, + 784.0, + 297.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 749.0, + 1218.0, + 749.0, + 1218.0, + 784.0, + 475.0, + 784.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 565, + 1047, + 565, + 1047, + 1021, + 297, + 1021 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1440, + 1403, + 1440, + 1403, + 1563, + 299, + 1563 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 340, + 151, + 1355, + 151, + 1355, + 482, + 340, + 482 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 1193, + 1408, + 1193, + 1408, + 1383, + 297, + 1383 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 298, + 1883, + 1404, + 1883, + 1404, + 2009, + 298, + 2009 + ], + "score": 0.97 + }, + { + "category_id": 4, + "poly": [ + 1065, + 1006, + 1404, + 1006, + 1404, + 1173, + 1065, + 1173 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 297, + 1034, + 1046, + 1034, + 1046, + 1193, + 297, + 1193 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 292, + 1681, + 1403, + 1681, + 1403, + 1751, + 292, + 1751 + ], + "score": 0.948 + }, + { + "category_id": 4, + "poly": [ + 295, + 493, + 1404, + 493, + 1404, + 551, + 295, + 551 + ], + "score": 0.941 + }, + { + "category_id": 8, + "poly": [ + 532, + 1758, + 1166, + 1758, + 1166, + 1828, + 532, + 1828 + ], + "score": 0.939 + }, + { + "category_id": 3, + "poly": [ + 1079, + 579, + 1393, + 579, + 1393, + 997, + 1079, + 997 + ], + "score": 0.937 + }, + { + "category_id": 0, + "poly": [ + 298, + 1400, + 785, + 1400, + 785, + 1434, + 298, + 1434 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 296, + 1837, + 1397, + 1837, + 1397, + 1874, + 296, + 1874 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1777, + 1400, + 1777, + 1400, + 1808, + 1352, + 1808 + ], + "score": 0.886 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1585, + 1400, + 1585, + 1400, + 1615, + 1352, + 1615 + ], + "score": 0.838 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2084, + 841, + 2084 + ], + "score": 0.776 + }, + { + "category_id": 8, + "poly": [ + 339, + 1566, + 1317, + 1566, + 1317, + 1673, + 339, + 1673 + ], + "score": 0.706 + }, + { + "category_id": 8, + "poly": [ + 344, + 1566, + 1315, + 1566, + 1315, + 1673, + 344, + 1673 + ], + "score": 0.213 + }, + { + "category_id": 13, + "poly": [ + 406, + 1712, + 503, + 1712, + 503, + 1757, + 406, + 1757 + ], + "score": 0.94, + "latex": "\\frac { \\partial \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\partial \\mathbf { x } _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 351, + 1976, + 451, + 1976, + 451, + 2009, + 351, + 2009 + ], + "score": 0.93, + "latex": "\\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t )" + }, + { + "category_id": 13, + "poly": [ + 414, + 1291, + 482, + 1291, + 482, + 1324, + 414, + 1324 + ], + "score": 0.93, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 14, + "poly": [ + 339, + 1563, + 1314, + 1563, + 1314, + 1676, + 339, + 1676 + ], + "score": 0.93, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) = \\frac { 1 } { \\sqrt { \\gamma _ { t } ^ { 2 } + 1 } } \\underbrace { \\frac { \\partial \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\partial \\mathbf { x } _ { t } } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } _ { \\mathrm { J V P _ { 1 } } } - \\frac { \\gamma _ { t } } { 1 + \\gamma _ { t } ^ { 2 } } \\underbrace { \\frac { \\partial \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\partial \\mathbf { x } _ { t } } \\mathbf { x } _ { t } } _ { \\mathrm { J V P _ { 2 } } } + \\frac { \\partial \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\partial t } \\frac { d t } { d \\gamma _ { t } } ." + }, + { + "category_id": 13, + "poly": [ + 643, + 777, + 728, + 777, + 728, + 810, + 643, + 810 + ], + "score": 0.92, + "latex": "\\hat { \\mathbf { x } } _ { t } ( \\Delta t )" + }, + { + "category_id": 13, + "poly": [ + 709, + 1155, + 753, + 1155, + 753, + 1197, + 709, + 1197 + ], + "score": 0.92, + "latex": "\\frac { d ^ { p } \\mathbf { y } } { d t ^ { p } }" + }, + { + "category_id": 13, + "poly": [ + 1014, + 1836, + 1226, + 1836, + 1226, + 1875, + 1014, + 1875 + ], + "score": 0.92, + "latex": "\\nabla _ { \\mathbf { x } _ { t } } ^ { \\top } \\nabla _ { \\mathbf { x } _ { t } } \\log p _ { t } ( \\mathbf { x } _ { t } )" + }, + { + "category_id": 14, + "poly": [ + 532, + 1755, + 1165, + 1755, + 1165, + 1828, + 532, + 1828 + ], + "score": 0.92, + "latex": "\\frac { \\partial \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\partial \\mathbf { x } _ { t } } = - \\sigma _ { t } \\frac { \\partial \\mathbf { s } _ { \\theta } ( \\mathbf { x } _ { t } , t ) } { \\partial \\mathbf { x } _ { t } } \\approx - \\sigma _ { t } \\nabla _ { \\mathbf { x } _ { t } } ^ { \\top } \\nabla _ { \\mathbf { x } _ { t } } \\log p _ { t } ( \\mathbf { x } _ { t } ) ," + }, + { + "category_id": 13, + "poly": [ + 1061, + 1885, + 1128, + 1885, + 1128, + 1918, + 1061, + 1918 + ], + "score": 0.92, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1067, + 1090, + 1141, + 1090, + 1141, + 1120, + 1067, + 1120 + ], + "score": 0.91, + "latex": "\\xi _ { t } ( \\Delta t )" + }, + { + "category_id": 13, + "poly": [ + 299, + 1220, + 343, + 1220, + 343, + 1266, + 299, + 1266 + ], + "score": 0.9, + "latex": "\\textstyle { \\frac { d ^ { 2 } \\mathbf { y } } { d t ^ { 2 } } }" + }, + { + "category_id": 13, + "poly": [ + 1063, + 1262, + 1142, + 1262, + 1142, + 1292, + 1063, + 1292 + ], + "score": 0.9, + "latex": "\\xi _ { t } ( \\Delta t )" + }, + { + "category_id": 13, + "poly": [ + 298, + 838, + 503, + 838, + 503, + 871, + 298, + 871 + ], + "score": 0.9, + "latex": "t \\in \\{ 0 . 1 , 0 . 2 , 0 . 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 984, + 1227, + 1399, + 1227, + 1399, + 1264, + 984, + 1264 + ], + "score": 0.9, + "latex": "( f ( { \\bf y } _ { t _ { n } } , t _ { n } ) - f ( { \\bf y } _ { t _ { n - 1 } } , t _ { n - 1 } ) ) / h _ { n - 1 }" + }, + { + "category_id": 13, + "poly": [ + 765, + 1261, + 799, + 1261, + 799, + 1291, + 765, + 1291 + ], + "score": 0.89, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 778, + 331, + 778, + 331, + 808, + 298, + 808 + ], + "score": 0.87, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 645, + 691, + 676, + 691, + 676, + 717, + 645, + 717 + ], + "score": 0.86, + "latex": "\\mathbf { x } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1189, + 1949, + 1220, + 1949, + 1220, + 1976, + 1189, + 1976 + ], + "score": 0.85, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1325, + 1292, + 1362, + 1292, + 1362, + 1319, + 1325, + 1319 + ], + "score": 0.84, + "latex": "\\Delta t" + }, + { + "category_id": 13, + "poly": [ + 810, + 1475, + 841, + 1475, + 841, + 1502, + 810, + 1502 + ], + "score": 0.83, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 368, + 812, + 398, + 812, + 398, + 838, + 368, + 838 + ], + "score": 0.83, + "latex": "\\mathbf { x } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 634, + 1890, + 664, + 1890, + 664, + 1915, + 634, + 1915 + ], + "score": 0.82, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 686, + 747, + 722, + 747, + 722, + 774, + 686, + 774 + ], + "score": 0.78, + "latex": "\\Delta t" + }, + { + "category_id": 13, + "poly": [ + 1070, + 1323, + 1085, + 1323, + 1085, + 1348, + 1070, + 1348 + ], + "score": 0.75, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 371, + 901, + 385, + 901, + 385, + 926, + 371, + 926 + ], + "score": 0.6, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 926, + 495, + 958, + 495, + 958, + 522, + 926, + 522 + ], + "score": 0.4, + "latex": "( b )" + }, + { + "category_id": 13, + "poly": [ + 884, + 687, + 930, + 687, + 930, + 716, + 884, + 716 + ], + "score": 0.3, + "latex": "1 0 \\mathrm { k }" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 160.0, + 496.0, + 160.0, + 496.0, + 183.0, + 462.0, + 183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 166.0, + 512.0, + 166.0, + 512.0, + 181.0, + 497.0, + 181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 159.0, + 1232.0, + 159.0, + 1232.0, + 186.0, + 1179.0, + 186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 183.0, + 471.0, + 183.0, + 471.0, + 198.0, + 458.0, + 198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 183.0, + 518.0, + 183.0, + 518.0, + 198.0, + 503.0, + 198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 183.0, + 1189.0, + 183.0, + 1189.0, + 198.0, + 1176.0, + 198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 182.0, + 1235.0, + 182.0, + 1235.0, + 199.0, + 1221.0, + 199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 203.0, + 395.0, + 203.0, + 395.0, + 213.0, + 385.0, + 213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 195.0, + 411.0, + 195.0, + 411.0, + 205.0, + 399.0, + 205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 203.0, + 434.0, + 203.0, + 434.0, + 229.0, + 418.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 199.0, + 479.0, + 199.0, + 479.0, + 213.0, + 464.0, + 213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 208.0, + 492.0, + 208.0, + 492.0, + 219.0, + 482.0, + 219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 199.0, + 512.0, + 199.0, + 512.0, + 213.0, + 496.0, + 213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 202.0, + 558.0, + 202.0, + 558.0, + 229.0, + 543.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 198.0, + 573.0, + 198.0, + 573.0, + 203.0, + 568.0, + 203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 220.0, + 598.0, + 220.0, + 598.0, + 229.0, + 590.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 207.0, + 796.0, + 207.0, + 796.0, + 235.0, + 775.0, + 235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 221.0, + 911.0, + 221.0, + 911.0, + 233.0, + 902.0, + 233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 201.0, + 1113.0, + 201.0, + 1113.0, + 232.0, + 1093.0, + 232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 198.0, + 1129.0, + 198.0, + 1129.0, + 207.0, + 1118.0, + 207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 202.0, + 1154.0, + 202.0, + 1154.0, + 239.0, + 1136.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 198.0, + 1232.0, + 198.0, + 1232.0, + 220.0, + 1180.0, + 220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 212.0, + 1271.0, + 212.0, + 1271.0, + 239.0, + 1254.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 199.0, + 1318.0, + 199.0, + 1318.0, + 232.0, + 1298.0, + 232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 238.0, + 392.0, + 238.0, + 392.0, + 244.0, + 386.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 244.0, + 410.0, + 244.0, + 410.0, + 251.0, + 399.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 236.0, + 427.0, + 236.0, + 427.0, + 247.0, + 417.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 233.0, + 561.0, + 233.0, + 561.0, + 247.0, + 547.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 242.0, + 576.0, + 242.0, + 576.0, + 251.0, + 565.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 233.0, + 593.0, + 233.0, + 593.0, + 247.0, + 579.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 230.0, + 787.0, + 230.0, + 787.0, + 246.0, + 774.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 235.0, + 1131.0, + 235.0, + 1131.0, + 255.0, + 1100.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 233.0, + 1295.0, + 233.0, + 1295.0, + 255.0, + 1264.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 233.0, + 1314.0, + 233.0, + 1314.0, + 251.0, + 1296.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 268.0, + 525.0, + 268.0, + 525.0, + 278.0, + 513.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 269.0, + 821.0, + 269.0, + 821.0, + 284.0, + 806.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 267.0, + 1180.0, + 267.0, + 1180.0, + 279.0, + 1165.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 277.0, + 1096.0, + 277.0, + 1096.0, + 290.0, + 1081.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 277.0, + 1325.0, + 277.0, + 1325.0, + 285.0, + 1314.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 286.0, + 359.0, + 286.0, + 359.0, + 294.0, + 349.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 284.0, + 811.0, + 284.0, + 811.0, + 296.0, + 802.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 282.0, + 1079.0, + 282.0, + 1079.0, + 296.0, + 1065.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 282.0, + 1113.0, + 282.0, + 1113.0, + 296.0, + 1100.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 284.0, + 1310.0, + 284.0, + 1310.0, + 295.0, + 1300.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 285.0, + 1342.0, + 285.0, + 1342.0, + 294.0, + 1334.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 301.0, + 353.0, + 301.0, + 353.0, + 312.0, + 342.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 301.0, + 399.0, + 301.0, + 399.0, + 312.0, + 390.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 301.0, + 585.0, + 301.0, + 585.0, + 311.0, + 576.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 301.0, + 631.0, + 301.0, + 631.0, + 311.0, + 622.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 299.0, + 1073.0, + 299.0, + 1073.0, + 314.0, + 1059.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 297.0, + 1119.0, + 297.0, + 1119.0, + 315.0, + 1105.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 296.0, + 1305.0, + 296.0, + 1305.0, + 315.0, + 1291.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1341.0, + 301.0, + 1352.0, + 301.0, + 1352.0, + 312.0, + 1341.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 318.0, + 360.0, + 318.0, + 360.0, + 328.0, + 348.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 320.0, + 391.0, + 320.0, + 391.0, + 325.0, + 385.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 318.0, + 593.0, + 318.0, + 593.0, + 328.0, + 583.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 317.0, + 1080.0, + 317.0, + 1080.0, + 330.0, + 1065.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 319.0, + 1312.0, + 319.0, + 1312.0, + 328.0, + 1300.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 318.0, + 1343.0, + 318.0, + 1343.0, + 328.0, + 1333.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 323.0, + 1100.0, + 323.0, + 1100.0, + 337.0, + 1081.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 324.0, + 1330.0, + 324.0, + 1330.0, + 337.0, + 1313.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 336.0, + 1175.0, + 336.0, + 1175.0, + 345.0, + 1166.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 336.0, + 1244.0, + 336.0, + 1244.0, + 345.0, + 1233.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 349.0, + 853.0, + 349.0, + 853.0, + 358.0, + 841.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 368.0, + 395.0, + 368.0, + 395.0, + 376.0, + 385.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 360.0, + 410.0, + 360.0, + 410.0, + 368.0, + 400.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 367.0, + 427.0, + 367.0, + 427.0, + 377.0, + 417.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 366.0, + 561.0, + 366.0, + 561.0, + 379.0, + 547.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 360.0, + 576.0, + 360.0, + 576.0, + 370.0, + 564.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 359.0, + 794.0, + 359.0, + 794.0, + 410.0, + 759.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 366.0, + 921.0, + 366.0, + 921.0, + 380.0, + 908.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 359.0, + 1130.0, + 359.0, + 1130.0, + 371.0, + 1115.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 364.0, + 1146.0, + 364.0, + 1146.0, + 378.0, + 1132.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 363.0, + 1278.0, + 363.0, + 1278.0, + 378.0, + 1263.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 358.0, + 1296.0, + 358.0, + 1296.0, + 373.0, + 1281.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1297.0, + 364.0, + 1313.0, + 364.0, + 1313.0, + 378.0, + 1297.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 384.0, + 385.0, + 384.0, + 385.0, + 394.0, + 376.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 380.0, + 435.0, + 380.0, + 435.0, + 409.0, + 418.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 394.0, + 492.0, + 394.0, + 492.0, + 404.0, + 483.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 401.0, + 510.0, + 401.0, + 510.0, + 410.0, + 500.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 380.0, + 558.0, + 380.0, + 558.0, + 410.0, + 540.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 384.0, + 599.0, + 384.0, + 599.0, + 394.0, + 590.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 383.0, + 911.0, + 383.0, + 911.0, + 394.0, + 901.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 377.0, + 1155.0, + 377.0, + 1155.0, + 418.0, + 1091.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 387.0, + 1204.0, + 387.0, + 1204.0, + 439.0, + 1181.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 374.0, + 1281.0, + 374.0, + 1281.0, + 415.0, + 1252.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 377.0, + 1320.0, + 377.0, + 1320.0, + 398.0, + 1303.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 416.0, + 479.0, + 416.0, + 479.0, + 446.0, + 457.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 418.0, + 516.0, + 418.0, + 516.0, + 446.0, + 497.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 432.0, + 836.0, + 432.0, + 836.0, + 441.0, + 827.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 410.0, + 1239.0, + 410.0, + 1239.0, + 461.0, + 1172.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 437.0, + 496.0, + 437.0, + 496.0, + 452.0, + 480.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 454.0, + 573.0, + 454.0, + 573.0, + 484.0, + 399.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 452.0, + 903.0, + 452.0, + 903.0, + 483.0, + 786.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 454.0, + 1265.0, + 454.0, + 1265.0, + 483.0, + 1144.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 192.5, + 1304.0, + 192.5, + 1304.0, + 213.5, + 1253.0, + 213.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.75, + 235.0, + 917.75, + 235.0, + 917.75, + 244.0, + 908.75, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.75, + 234.5, + 1153.75, + 234.5, + 1153.75, + 247.5, + 1125.75, + 247.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 314.0, + 1108.0, + 314.0, + 1108.0, + 333.0, + 1101.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 364.0, + 1120.0, + 364.0, + 1120.0, + 376.5, + 1093.0, + 376.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 390.0, + 1231.0, + 390.0, + 1231.0, + 412.5, + 1193.0, + 412.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.75, + 400.5, + 473.75, + 400.5, + 473.75, + 409.5, + 468.75, + 409.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 396.0, + 1320.0, + 396.0, + 1320.0, + 420.0, + 1273.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1003.0, + 1405.0, + 1003.0, + 1405.0, + 1040.0, + 1065.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1034.0, + 1406.0, + 1034.0, + 1406.0, + 1063.0, + 1065.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1062.0, + 1157.0, + 1062.0, + 1157.0, + 1090.0, + 1065.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1059.0, + 1405.0, + 1059.0, + 1405.0, + 1092.0, + 1173.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1088.0, + 1405.0, + 1088.0, + 1405.0, + 1122.0, + 1142.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1119.0, + 1405.0, + 1119.0, + 1405.0, + 1147.0, + 1064.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1147.0, + 1354.0, + 1147.0, + 1354.0, + 1173.0, + 1067.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 490.0, + 925.0, + 490.0, + 925.0, + 526.0, + 294.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 490.0, + 1407.0, + 490.0, + 1407.0, + 526.0, + 959.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 518.0, + 1300.0, + 518.0, + 1300.0, + 554.0, + 294.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 583.0, + 1130.0, + 583.0, + 1130.0, + 601.0, + 1091.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 588.0, + 1258.0, + 588.0, + 1258.0, + 610.0, + 1168.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 606.0, + 1258.0, + 606.0, + 1258.0, + 625.0, + 1169.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 617.0, + 1134.0, + 617.0, + 1134.0, + 634.0, + 1094.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 624.0, + 1167.0, + 624.0, + 1167.0, + 637.0, + 1140.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 622.0, + 1258.0, + 622.0, + 1258.0, + 641.0, + 1169.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 636.0, + 1100.0, + 636.0, + 1100.0, + 715.0, + 1077.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 638.0, + 1253.0, + 638.0, + 1253.0, + 656.0, + 1168.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 648.0, + 1132.0, + 648.0, + 1132.0, + 669.0, + 1093.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 655.0, + 1253.0, + 655.0, + 1253.0, + 672.0, + 1168.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 676.0, + 1165.0, + 676.0, + 1165.0, + 683.0, + 1143.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 670.0, + 1253.0, + 670.0, + 1253.0, + 690.0, + 1169.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 682.0, + 1133.0, + 682.0, + 1133.0, + 702.0, + 1093.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 716.0, + 1131.0, + 716.0, + 1131.0, + 733.0, + 1094.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 748.0, + 1132.0, + 748.0, + 1132.0, + 768.0, + 1093.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 746.0, + 1372.0, + 746.0, + 1372.0, + 764.0, + 1172.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 778.0, + 1139.0, + 778.0, + 1139.0, + 799.0, + 1104.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 789.0, + 1214.0, + 789.0, + 1214.0, + 809.0, + 1165.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 804.0, + 1214.0, + 804.0, + 1214.0, + 824.0, + 1167.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 807.0, + 1169.0, + 807.0, + 1169.0, + 820.0, + 1141.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 812.0, + 1138.0, + 812.0, + 1138.0, + 834.0, + 1104.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 822.0, + 1213.0, + 822.0, + 1213.0, + 840.0, + 1168.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 823.0, + 1169.0, + 823.0, + 1169.0, + 837.0, + 1138.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 846.0, + 1115.0, + 846.0, + 1115.0, + 901.0, + 1087.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 847.0, + 1131.0, + 847.0, + 1131.0, + 865.0, + 1103.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 880.0, + 1131.0, + 880.0, + 1131.0, + 898.0, + 1104.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 913.0, + 1130.0, + 913.0, + 1130.0, + 933.0, + 1105.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 948.0, + 1132.0, + 948.0, + 1132.0, + 966.0, + 1105.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 950.0, + 1371.0, + 950.0, + 1371.0, + 960.0, + 1330.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 967.0, + 1158.0, + 967.0, + 1158.0, + 985.0, + 1129.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 966.0, + 1205.0, + 966.0, + 1205.0, + 987.0, + 1173.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 966.0, + 1253.0, + 966.0, + 1253.0, + 987.0, + 1220.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 966.0, + 1299.0, + 966.0, + 1299.0, + 987.0, + 1266.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 967.0, + 1344.0, + 967.0, + 1344.0, + 985.0, + 1314.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 968.0, + 1390.0, + 968.0, + 1390.0, + 985.0, + 1359.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 981.0, + 1296.0, + 981.0, + 1296.0, + 1001.0, + 1224.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 739.0, + 1375.0, + 739.0, + 1375.0, + 752.0, + 1338.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 929.0, + 1235.0, + 929.0, + 1235.0, + 939.5, + 1198.0, + 939.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1399.0, + 787.0, + 1399.0, + 787.0, + 1439.0, + 295.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 565.0, + 1046.0, + 565.0, + 1046.0, + 597.0, + 298.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 597.0, + 1045.0, + 597.0, + 1045.0, + 628.0, + 297.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 625.0, + 1047.0, + 625.0, + 1047.0, + 659.0, + 293.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 656.0, + 1050.0, + 656.0, + 1050.0, + 690.0, + 293.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 687.0, + 644.0, + 687.0, + 644.0, + 720.0, + 294.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 687.0, + 883.0, + 687.0, + 883.0, + 720.0, + 677.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 687.0, + 1046.0, + 687.0, + 1046.0, + 720.0, + 931.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 717.0, + 1047.0, + 717.0, + 1047.0, + 750.0, + 294.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 745.0, + 685.0, + 745.0, + 685.0, + 782.0, + 294.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 745.0, + 1047.0, + 745.0, + 1047.0, + 782.0, + 723.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 778.0, + 642.0, + 778.0, + 642.0, + 812.0, + 332.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 778.0, + 1050.0, + 778.0, + 1050.0, + 812.0, + 729.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 808.0, + 367.0, + 808.0, + 367.0, + 843.0, + 294.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 808.0, + 1047.0, + 808.0, + 1047.0, + 843.0, + 399.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 837.0, + 297.0, + 837.0, + 297.0, + 873.0, + 293.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 837.0, + 1048.0, + 837.0, + 1048.0, + 873.0, + 504.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 868.0, + 1047.0, + 868.0, + 1047.0, + 902.0, + 294.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 898.0, + 370.0, + 898.0, + 370.0, + 933.0, + 293.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 898.0, + 1046.0, + 898.0, + 1046.0, + 933.0, + 386.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 926.0, + 1048.0, + 926.0, + 1048.0, + 966.0, + 294.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 959.0, + 1048.0, + 959.0, + 1048.0, + 992.0, + 293.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 989.0, + 1001.0, + 989.0, + 1001.0, + 1025.0, + 294.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1442.0, + 1405.0, + 1442.0, + 1405.0, + 1475.0, + 295.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1472.0, + 809.0, + 1472.0, + 809.0, + 1505.0, + 295.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 1472.0, + 1405.0, + 1472.0, + 1405.0, + 1505.0, + 842.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1504.0, + 1404.0, + 1504.0, + 1404.0, + 1535.0, + 294.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1533.0, + 1408.0, + 1533.0, + 1408.0, + 1569.0, + 295.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1187.0, + 1406.0, + 1187.0, + 1406.0, + 1227.0, + 292.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1219.0, + 983.0, + 1219.0, + 983.0, + 1269.0, + 344.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1219.0, + 1409.0, + 1219.0, + 1409.0, + 1269.0, + 1400.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1258.0, + 764.0, + 1258.0, + 764.0, + 1294.0, + 296.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 1258.0, + 1062.0, + 1258.0, + 1062.0, + 1294.0, + 800.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 1258.0, + 1406.0, + 1258.0, + 1406.0, + 1294.0, + 1143.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1288.0, + 413.0, + 1288.0, + 413.0, + 1328.0, + 294.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 1288.0, + 1324.0, + 1288.0, + 1324.0, + 1328.0, + 483.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 1288.0, + 1408.0, + 1288.0, + 1408.0, + 1328.0, + 1363.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1318.0, + 1069.0, + 1318.0, + 1069.0, + 1355.0, + 292.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1318.0, + 1406.0, + 1318.0, + 1406.0, + 1355.0, + 1086.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1351.0, + 984.0, + 1351.0, + 984.0, + 1384.0, + 296.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1884.0, + 633.0, + 1884.0, + 633.0, + 1920.0, + 294.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 1884.0, + 1060.0, + 1884.0, + 1060.0, + 1920.0, + 665.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1884.0, + 1405.0, + 1884.0, + 1405.0, + 1920.0, + 1129.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1915.0, + 1406.0, + 1915.0, + 1406.0, + 1948.0, + 294.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1945.0, + 1188.0, + 1945.0, + 1188.0, + 1980.0, + 292.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 1945.0, + 1408.0, + 1945.0, + 1408.0, + 1980.0, + 1221.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1975.0, + 350.0, + 1975.0, + 350.0, + 2011.0, + 292.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1975.0, + 1405.0, + 1975.0, + 1405.0, + 2011.0, + 452.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1033.0, + 1048.0, + 1033.0, + 1048.0, + 1070.0, + 295.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1063.0, + 1045.0, + 1063.0, + 1045.0, + 1099.0, + 294.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1094.0, + 1048.0, + 1094.0, + 1048.0, + 1130.0, + 293.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1121.0, + 1051.0, + 1121.0, + 1051.0, + 1165.0, + 291.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1154.0, + 708.0, + 1154.0, + 708.0, + 1198.0, + 294.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 1154.0, + 1045.0, + 1154.0, + 1045.0, + 1198.0, + 754.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1679.0, + 1406.0, + 1679.0, + 1406.0, + 1717.0, + 294.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1705.0, + 405.0, + 1705.0, + 405.0, + 1755.0, + 295.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1705.0, + 1109.0, + 1705.0, + 1109.0, + 1755.0, + 504.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1834.0, + 1013.0, + 1834.0, + 1013.0, + 1878.0, + 294.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 1834.0, + 1403.0, + 1834.0, + 1403.0, + 1878.0, + 1227.0, + 1878.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1631, + 1405, + 1631, + 1405, + 1819, + 297, + 1819 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1823, + 1405, + 1823, + 1405, + 2009, + 298, + 2009 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1005, + 1406, + 1005, + 1406, + 1249, + 297, + 1249 + ], + "score": 0.968 + }, + { + "category_id": 4, + "poly": [ + 879, + 542, + 1404, + 542, + 1404, + 683, + 879, + 683 + ], + "score": 0.964 + }, + { + "category_id": 3, + "poly": [ + 878, + 292, + 1413, + 292, + 1413, + 515, + 878, + 515 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 297, + 695, + 1405, + 695, + 1405, + 1001, + 297, + 1001 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 296, + 272, + 858, + 272, + 858, + 695, + 296, + 695 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 295, + 202, + 1405, + 202, + 1405, + 266, + 295, + 266 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 298, + 1348, + 1405, + 1348, + 1405, + 1448, + 298, + 1448 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 299, + 1454, + 1404, + 1454, + 1404, + 1547, + 299, + 1547 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 518, + 1258, + 1179, + 1258, + 1179, + 1330, + 518, + 1330 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 305, + 1560, + 1325, + 1560, + 1325, + 1604, + 305, + 1604 + ], + "score": 0.908 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1276, + 1400, + 1276, + 1400, + 1306, + 1352, + 1306 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1354, + 1565, + 1400, + 1565, + 1400, + 1597, + 1354, + 1597 + ], + "score": 0.861 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.705 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.124 + }, + { + "category_id": 13, + "poly": [ + 766, + 1633, + 899, + 1633, + 899, + 1668, + 766, + 1668 + ], + "score": 0.94, + "latex": "g _ { \\mathrm { d } } ( t ) = \\gamma _ { t } ^ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 518, + 1256, + 1178, + 1256, + 1178, + 1329, + 518, + 1329 + ], + "score": 0.94, + "latex": "k _ { \\psi } = - \\frac { 1 } { \\gamma _ { t } } k _ { \\psi } ^ { ( 1 ) } + \\frac { \\gamma _ { t } } { 1 + \\gamma _ { t } ^ { 2 } } k _ { \\psi } ^ { ( 2 ) } + \\frac { 1 } { \\gamma _ { t } ( 1 + \\gamma _ { t } ^ { 2 } ) } k _ { \\psi } ^ { ( 3 ) } \\approx d _ { \\gamma _ { t } } \\epsilon _ { \\theta } ," + }, + { + "category_id": 13, + "poly": [ + 487, + 605, + 588, + 605, + 588, + 637, + 487, + 637 + ], + "score": 0.93, + "latex": "\\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t )" + }, + { + "category_id": 13, + "poly": [ + 1163, + 1036, + 1397, + 1036, + 1397, + 1070, + 1163, + 1070 + ], + "score": 0.93, + "latex": "p _ { 0 } ( \\mathbf { x } _ { 0 } ) = \\delta ( \\mathbf { x } _ { 0 } = \\mathbf { 0 } )" + }, + { + "category_id": 13, + "poly": [ + 746, + 1913, + 892, + 1913, + 892, + 1951, + 746, + 1951 + ], + "score": 0.92, + "latex": "\\nabla _ { \\mathbf { x } _ { t } } ^ { \\top } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , \\dot { t } )" + }, + { + "category_id": 13, + "poly": [ + 1172, + 970, + 1239, + 970, + 1239, + 1002, + 1172, + 1002 + ], + "score": 0.92, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 363, + 635, + 472, + 635, + 472, + 668, + 363, + 668 + ], + "score": 0.92, + "latex": "k _ { \\psi } ( \\mathbf { x } _ { t } , t )" + }, + { + "category_id": 13, + "poly": [ + 997, + 1006, + 1065, + 1006, + 1065, + 1039, + 997, + 1039 + ], + "score": 0.92, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 744, + 878, + 811, + 878, + 811, + 911, + 744, + 911 + ], + "score": 0.92, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 574, + 1638, + 712, + 1638, + 712, + 1666, + 574, + 1666 + ], + "score": 0.92, + "latex": "\\alpha _ { t } \\mathbf { x } _ { 0 } + \\sigma _ { t } \\mathbf { \\epsilon } \\mathbf { \\epsilon }" + }, + { + "category_id": 13, + "poly": [ + 543, + 302, + 610, + 302, + 610, + 335, + 543, + 335 + ], + "score": 0.92, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1203, + 1757, + 1270, + 1757, + 1270, + 1790, + 1203, + 1790 + ], + "score": 0.92, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 737, + 363, + 803, + 363, + 803, + 396, + 737, + 396 + ], + "score": 0.92, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1292, + 1453, + 1400, + 1453, + 1400, + 1488, + 1292, + 1488 + ], + "score": 0.92, + "latex": "t \\in [ 0 , T ]" + }, + { + "category_id": 13, + "poly": [ + 1106, + 1634, + 1161, + 1634, + 1161, + 1668, + 1106, + 1668 + ], + "score": 0.91, + "latex": "1 / \\gamma _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1057, + 1455, + 1123, + 1455, + 1123, + 1488, + 1057, + 1488 + ], + "score": 0.91, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 968, + 571, + 1031, + 571, + 1031, + 600, + 968, + 600 + ], + "score": 0.91, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1009, + 1096, + 1111, + 1096, + 1111, + 1129, + 1009, + 1129 + ], + "score": 0.91, + "latex": "t \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 446, + 1729, + 513, + 1729, + 513, + 1759, + 446, + 1759 + ], + "score": 0.9, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 442, + 1697, + 510, + 1697, + 510, + 1728, + 442, + 1728 + ], + "score": 0.9, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 911, + 1455, + 950, + 1455, + 950, + 1488, + 911, + 1488 + ], + "score": 0.9, + "latex": "k _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 1174, + 655, + 1209, + 655, + 1209, + 684, + 1174, + 684 + ], + "score": 0.89, + "latex": "k _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 1066, + 1758, + 1102, + 1758, + 1102, + 1790, + 1066, + 1790 + ], + "score": 0.89, + "latex": "k _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 1296, + 627, + 1331, + 627, + 1331, + 656, + 1296, + 656 + ], + "score": 0.89, + "latex": "k _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 972, + 204, + 1033, + 204, + 1033, + 234, + 972, + 234 + ], + "score": 0.89, + "latex": "\\mathrm { J V P _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 1331, + 1697, + 1370, + 1697, + 1370, + 1729, + 1331, + 1729 + ], + "score": 0.89, + "latex": "k _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 434, + 1127, + 496, + 1127, + 496, + 1158, + 434, + 1158 + ], + "score": 0.89, + "latex": "\\mathrm { J V P _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 1195, + 544, + 1231, + 544, + 1231, + 573, + 1195, + 573 + ], + "score": 0.89, + "latex": "k _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 524, + 1516, + 558, + 1516, + 558, + 1546, + 524, + 1546 + ], + "score": 0.89, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 696, + 336, + 696, + 336, + 729, + 297, + 729 + ], + "score": 0.89, + "latex": "k _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 809, + 1128, + 879, + 1128, + 879, + 1155, + 809, + 1155 + ], + "score": 0.88, + "latex": "t 0" + }, + { + "category_id": 13, + "poly": [ + 958, + 1128, + 1019, + 1128, + 1019, + 1158, + 958, + 1158 + ], + "score": 0.88, + "latex": "\\mathrm { J V P _ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 1361, + 908, + 1400, + 908, + 1400, + 941, + 1361, + 941 + ], + "score": 0.88, + "latex": "k _ { \\psi }" + }, + { + "category_id": 14, + "poly": [ + 315, + 1559, + 1338, + 1559, + 1338, + 1619, + 315, + 1619 + ], + "score": 0.88, + "latex": "\\operatorname* { m i n } _ { \\psi } \\mathbb { E } _ { t \\sim \\mathcal { U } [ t _ { \\mathrm { c u t o f f } } , 1 ] , \\mathbf { x } _ { 0 } \\sim p ( \\mathbf { x } _ { 0 } ) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } \\left[ g _ { \\mathrm { d } } ( t ) \\| k _ { \\psi } ( \\alpha _ { t } \\mathbf { x } _ { 0 } + \\sigma _ { t } \\epsilon , t ) - d _ { \\gamma _ { t } } \\epsilon _ { \\theta } ( \\alpha _ { t } \\mathbf { x } _ { 0 } + \\sigma _ { t } \\epsilon , t ) \\| _ { 2 } ^ { 2 } \\right]" + }, + { + "category_id": 13, + "poly": [ + 298, + 1758, + 335, + 1758, + 335, + 1791, + 298, + 1791 + ], + "score": 0.88, + "latex": "k _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 678, + 696, + 720, + 696, + 720, + 725, + 678, + 725 + ], + "score": 0.86, + "latex": "2 \\%" + }, + { + "category_id": 13, + "poly": [ + 672, + 850, + 704, + 850, + 704, + 878, + 672, + 878 + ], + "score": 0.86, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 343, + 1488, + 373, + 1488, + 373, + 1515, + 343, + 1515 + ], + "score": 0.85, + "latex": "\\mathbf { x } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1246, + 760, + 1277, + 760, + 1277, + 787, + 1246, + 787 + ], + "score": 0.85, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 727, + 547, + 758, + 547, + 758, + 575, + 727, + 575 + ], + "score": 0.85, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1100, + 657, + 1129, + 657, + 1129, + 682, + 1100, + 682 + ], + "score": 0.84, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1222, + 630, + 1250, + 630, + 1250, + 653, + 1222, + 653 + ], + "score": 0.83, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 825, + 579, + 856, + 579, + 856, + 605, + 825, + 605 + ], + "score": 0.83, + "latex": "\\mathbf { x } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1256, + 208, + 1287, + 208, + 1287, + 234, + 1256, + 234 + ], + "score": 0.83, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 826, + 790, + 858, + 790, + 858, + 817, + 826, + 817 + ], + "score": 0.82, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 509, + 1391, + 540, + 1391, + 540, + 1417, + 509, + 1417 + ], + "score": 0.81, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 824, + 457, + 856, + 457, + 856, + 484, + 824, + 484 + ], + "score": 0.81, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 496, + 1350, + 635, + 1350, + 635, + 1386, + 496, + 1386 + ], + "score": 0.81, + "latex": "i \\in \\{ 1 , 2 , 3 \\}" + }, + { + "category_id": 13, + "poly": [ + 369, + 1345, + 485, + 1345, + 485, + 1390, + 369, + 1390 + ], + "score": 0.77, + "latex": "k _ { \\psi } ^ { ( i ) } ( \\mathbf { x } _ { t } , t )" + }, + { + "category_id": 13, + "poly": [ + 1200, + 1669, + 1214, + 1669, + 1214, + 1693, + 1200, + 1693 + ], + "score": 0.7, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 880, + 631, + 908, + 631, + 908, + 655, + 880, + 655 + ], + "score": 0.7, + "latex": "\\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 881, + 480, + 897, + 480, + 897, + 505, + 881, + 505 + ], + "score": 0.43, + "latex": "t -" + }, + { + "category_id": 13, + "poly": [ + 369, + 1345, + 636, + 1345, + 636, + 1390, + 369, + 1390 + ], + "score": 0.42, + "latex": "k _ { \\psi } ^ { ( i ) } ( \\mathbf { x } _ { t } , t ) , i \\in \\{ 1 , 2 , 3 \\}" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 541.0, + 1194.0, + 541.0, + 1194.0, + 573.0, + 879.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 541.0, + 1403.0, + 541.0, + 1403.0, + 573.0, + 1232.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 570.0, + 967.0, + 570.0, + 967.0, + 599.0, + 877.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 570.0, + 1405.0, + 570.0, + 1405.0, + 599.0, + 1032.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 598.0, + 1404.0, + 598.0, + 1404.0, + 628.0, + 878.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 626.0, + 1221.0, + 626.0, + 1221.0, + 658.0, + 909.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 626.0, + 1295.0, + 626.0, + 1295.0, + 658.0, + 1251.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 626.0, + 1404.0, + 626.0, + 1404.0, + 658.0, + 1332.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 652.0, + 1099.0, + 652.0, + 1099.0, + 687.0, + 877.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 652.0, + 1173.0, + 652.0, + 1173.0, + 687.0, + 1130.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 652.0, + 1221.0, + 652.0, + 1221.0, + 687.0, + 1210.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 356.0, + 1052.0, + 356.0, + 1052.0, + 372.0, + 1010.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 380.0, + 913.0, + 380.0, + 913.0, + 406.0, + 881.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 376.0, + 1272.0, + 376.0, + 1272.0, + 410.0, + 1154.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 378.0, + 1415.0, + 378.0, + 1415.0, + 411.0, + 1305.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 474.0, + 953.0, + 474.0, + 953.0, + 515.0, + 906.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 481.0, + 910.0, + 481.0, + 910.0, + 505.0, + 898.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.75, + 334.0, + 1274.75, + 334.0, + 1274.75, + 362.0, + 1219.75, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2092.0, + 838.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2092.0, + 838.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1634.0, + 573.0, + 1634.0, + 573.0, + 1670.0, + 295.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 1634.0, + 765.0, + 1634.0, + 765.0, + 1670.0, + 713.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 1634.0, + 1105.0, + 1634.0, + 1105.0, + 1670.0, + 900.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 1634.0, + 1406.0, + 1634.0, + 1406.0, + 1670.0, + 1162.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1661.0, + 1199.0, + 1661.0, + 1199.0, + 1704.0, + 292.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1661.0, + 1406.0, + 1661.0, + 1406.0, + 1704.0, + 1215.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1693.0, + 441.0, + 1693.0, + 441.0, + 1733.0, + 292.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 1693.0, + 1330.0, + 1693.0, + 1330.0, + 1733.0, + 511.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1371.0, + 1693.0, + 1406.0, + 1693.0, + 1406.0, + 1733.0, + 1371.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1725.0, + 445.0, + 1725.0, + 445.0, + 1760.0, + 294.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1725.0, + 1405.0, + 1725.0, + 1405.0, + 1760.0, + 514.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1756.0, + 1065.0, + 1756.0, + 1065.0, + 1792.0, + 336.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1756.0, + 1202.0, + 1756.0, + 1202.0, + 1792.0, + 1103.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1756.0, + 1405.0, + 1756.0, + 1405.0, + 1792.0, + 1271.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1787.0, + 1369.0, + 1787.0, + 1369.0, + 1821.0, + 292.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1823.0, + 1405.0, + 1823.0, + 1405.0, + 1859.0, + 295.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1852.0, + 1406.0, + 1852.0, + 1406.0, + 1890.0, + 293.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1884.0, + 1408.0, + 1884.0, + 1408.0, + 1920.0, + 295.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1913.0, + 745.0, + 1913.0, + 745.0, + 1953.0, + 293.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1913.0, + 1407.0, + 1913.0, + 1407.0, + 1953.0, + 893.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1945.0, + 1405.0, + 1945.0, + 1405.0, + 1980.0, + 293.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1976.0, + 1405.0, + 1976.0, + 1405.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1002.0, + 996.0, + 1002.0, + 996.0, + 1041.0, + 292.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1002.0, + 1407.0, + 1002.0, + 1407.0, + 1041.0, + 1066.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1037.0, + 1162.0, + 1037.0, + 1162.0, + 1071.0, + 294.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1037.0, + 1407.0, + 1037.0, + 1407.0, + 1071.0, + 1398.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1065.0, + 1406.0, + 1065.0, + 1406.0, + 1100.0, + 294.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1096.0, + 1008.0, + 1096.0, + 1008.0, + 1131.0, + 294.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1096.0, + 1406.0, + 1096.0, + 1406.0, + 1131.0, + 1112.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1126.0, + 433.0, + 1126.0, + 433.0, + 1160.0, + 292.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 1126.0, + 808.0, + 1126.0, + 808.0, + 1160.0, + 497.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 1126.0, + 957.0, + 1126.0, + 957.0, + 1160.0, + 880.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1126.0, + 1406.0, + 1126.0, + 1406.0, + 1160.0, + 1020.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1157.0, + 1406.0, + 1157.0, + 1406.0, + 1190.0, + 294.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1188.0, + 1404.0, + 1188.0, + 1404.0, + 1221.0, + 291.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1218.0, + 849.0, + 1218.0, + 849.0, + 1251.0, + 295.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 693.0, + 296.0, + 693.0, + 296.0, + 730.0, + 292.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 693.0, + 677.0, + 693.0, + 677.0, + 730.0, + 337.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 693.0, + 1407.0, + 693.0, + 1407.0, + 730.0, + 721.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 724.0, + 1405.0, + 724.0, + 1405.0, + 760.0, + 292.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 755.0, + 1245.0, + 755.0, + 1245.0, + 787.0, + 295.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 755.0, + 1403.0, + 755.0, + 1403.0, + 787.0, + 1278.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 786.0, + 825.0, + 786.0, + 825.0, + 821.0, + 294.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 786.0, + 1405.0, + 786.0, + 1405.0, + 821.0, + 859.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 814.0, + 1405.0, + 814.0, + 1405.0, + 853.0, + 292.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 844.0, + 671.0, + 844.0, + 671.0, + 882.0, + 292.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 844.0, + 1407.0, + 844.0, + 1407.0, + 882.0, + 705.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 877.0, + 743.0, + 877.0, + 743.0, + 913.0, + 294.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 877.0, + 1406.0, + 877.0, + 1406.0, + 913.0, + 812.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 905.0, + 1360.0, + 905.0, + 1360.0, + 946.0, + 291.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 905.0, + 1406.0, + 905.0, + 1406.0, + 946.0, + 1401.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 937.0, + 1406.0, + 937.0, + 1406.0, + 971.0, + 292.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 969.0, + 1171.0, + 969.0, + 1171.0, + 1004.0, + 295.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 969.0, + 1408.0, + 969.0, + 1408.0, + 1004.0, + 1240.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 268.0, + 860.0, + 268.0, + 860.0, + 306.0, + 295.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 301.0, + 542.0, + 301.0, + 542.0, + 335.0, + 293.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 301.0, + 861.0, + 301.0, + 861.0, + 335.0, + 611.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 331.0, + 860.0, + 331.0, + 860.0, + 365.0, + 294.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 361.0, + 736.0, + 361.0, + 736.0, + 396.0, + 295.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 361.0, + 861.0, + 361.0, + 861.0, + 396.0, + 804.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 392.0, + 864.0, + 392.0, + 864.0, + 426.0, + 295.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 422.0, + 861.0, + 422.0, + 861.0, + 454.0, + 295.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 451.0, + 823.0, + 451.0, + 823.0, + 489.0, + 292.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 451.0, + 861.0, + 451.0, + 861.0, + 489.0, + 857.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 484.0, + 861.0, + 484.0, + 861.0, + 515.0, + 295.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 512.0, + 861.0, + 512.0, + 861.0, + 547.0, + 294.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 545.0, + 726.0, + 545.0, + 726.0, + 576.0, + 295.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 545.0, + 860.0, + 545.0, + 860.0, + 576.0, + 759.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 572.0, + 824.0, + 572.0, + 824.0, + 610.0, + 294.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 572.0, + 861.0, + 572.0, + 861.0, + 610.0, + 857.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 604.0, + 486.0, + 604.0, + 486.0, + 638.0, + 294.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 604.0, + 860.0, + 604.0, + 860.0, + 638.0, + 589.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 634.0, + 362.0, + 634.0, + 362.0, + 668.0, + 295.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 634.0, + 862.0, + 634.0, + 862.0, + 668.0, + 473.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 663.0, + 860.0, + 663.0, + 860.0, + 701.0, + 294.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 202.0, + 971.0, + 202.0, + 971.0, + 238.0, + 295.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 202.0, + 1255.0, + 202.0, + 1255.0, + 238.0, + 1034.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1288.0, + 202.0, + 1405.0, + 202.0, + 1405.0, + 238.0, + 1288.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 235.0, + 1407.0, + 235.0, + 1407.0, + 267.0, + 297.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1342.0, + 368.0, + 1342.0, + 368.0, + 1397.0, + 289.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1342.0, + 1407.0, + 1342.0, + 1407.0, + 1397.0, + 637.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1384.0, + 508.0, + 1384.0, + 508.0, + 1422.0, + 293.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1384.0, + 1405.0, + 1384.0, + 1405.0, + 1422.0, + 541.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1414.0, + 1410.0, + 1414.0, + 1410.0, + 1453.0, + 293.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1452.0, + 910.0, + 1452.0, + 910.0, + 1490.0, + 294.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 1452.0, + 1056.0, + 1452.0, + 1056.0, + 1490.0, + 951.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1452.0, + 1291.0, + 1452.0, + 1291.0, + 1490.0, + 1124.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1452.0, + 1404.0, + 1452.0, + 1404.0, + 1490.0, + 1401.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1485.0, + 342.0, + 1485.0, + 342.0, + 1519.0, + 295.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1485.0, + 1405.0, + 1485.0, + 1405.0, + 1519.0, + 374.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1515.0, + 523.0, + 1515.0, + 523.0, + 1549.0, + 293.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 1515.0, + 1276.0, + 1515.0, + 1276.0, + 1549.0, + 559.0, + 1549.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 538, + 1405, + 538, + 1405, + 1025, + 298, + 1025 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1659, + 1404, + 1659, + 1404, + 2024, + 298, + 2024 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1038, + 1405, + 1038, + 1405, + 1402, + 298, + 1402 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 202, + 1405, + 202, + 1405, + 386, + 298, + 386 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1492, + 1405, + 1492, + 1405, + 1645, + 298, + 1645 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 1417, + 1401, + 1417, + 1401, + 1479, + 299, + 1479 + ], + "score": 0.946 + }, + { + "category_id": 0, + "poly": [ + 299, + 458, + 547, + 458, + 547, + 496, + 299, + 496 + ], + "score": 0.909 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2085, + 840, + 2085 + ], + "score": 0.779 + }, + { + "category_id": 15, + "poly": [ + 293.0, + 457.0, + 552.0, + 457.0, + 552.0, + 501.0, + 293.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2091.0, + 840.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 536.0, + 1408.0, + 536.0, + 1408.0, + 572.0, + 293.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 567.0, + 1407.0, + 567.0, + 1407.0, + 602.0, + 295.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 597.0, + 1405.0, + 597.0, + 1405.0, + 635.0, + 293.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 629.0, + 1405.0, + 629.0, + 1405.0, + 664.0, + 293.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 660.0, + 1402.0, + 660.0, + 1402.0, + 691.0, + 296.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 687.0, + 1405.0, + 687.0, + 1405.0, + 723.0, + 293.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 720.0, + 1405.0, + 720.0, + 1405.0, + 755.0, + 295.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 751.0, + 1406.0, + 751.0, + 1406.0, + 785.0, + 295.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 780.0, + 1405.0, + 780.0, + 1405.0, + 815.0, + 295.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 811.0, + 1403.0, + 811.0, + 1403.0, + 844.0, + 296.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 841.0, + 1403.0, + 841.0, + 1403.0, + 874.0, + 293.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 868.0, + 1403.0, + 868.0, + 1403.0, + 907.0, + 293.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 900.0, + 1408.0, + 900.0, + 1408.0, + 937.0, + 292.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 928.0, + 1406.0, + 928.0, + 1406.0, + 970.0, + 292.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 960.0, + 1405.0, + 960.0, + 1405.0, + 996.0, + 293.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 992.0, + 1407.0, + 992.0, + 1407.0, + 1026.0, + 296.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1656.0, + 1405.0, + 1656.0, + 1405.0, + 1695.0, + 296.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1690.0, + 1405.0, + 1690.0, + 1405.0, + 1725.0, + 292.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1720.0, + 1406.0, + 1720.0, + 1406.0, + 1754.0, + 294.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1751.0, + 1404.0, + 1751.0, + 1404.0, + 1785.0, + 294.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1780.0, + 1405.0, + 1780.0, + 1405.0, + 1816.0, + 293.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1811.0, + 1405.0, + 1811.0, + 1405.0, + 1845.0, + 293.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1841.0, + 1406.0, + 1841.0, + 1406.0, + 1875.0, + 294.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1869.0, + 1405.0, + 1869.0, + 1405.0, + 1906.0, + 293.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1902.0, + 1406.0, + 1902.0, + 1406.0, + 1937.0, + 292.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1929.0, + 1405.0, + 1929.0, + 1405.0, + 1970.0, + 292.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1963.0, + 1405.0, + 1963.0, + 1405.0, + 1996.0, + 294.0, + 1996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1992.0, + 1142.0, + 1992.0, + 1142.0, + 2028.0, + 293.0, + 2028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1040.0, + 1407.0, + 1040.0, + 1407.0, + 1070.0, + 297.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1068.0, + 1407.0, + 1068.0, + 1407.0, + 1103.0, + 295.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1098.0, + 1407.0, + 1098.0, + 1407.0, + 1132.0, + 295.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1129.0, + 1405.0, + 1129.0, + 1405.0, + 1163.0, + 295.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1160.0, + 1405.0, + 1160.0, + 1405.0, + 1193.0, + 295.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1190.0, + 1405.0, + 1190.0, + 1405.0, + 1224.0, + 295.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1221.0, + 1405.0, + 1221.0, + 1405.0, + 1255.0, + 293.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1249.0, + 1407.0, + 1249.0, + 1407.0, + 1285.0, + 293.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1280.0, + 1405.0, + 1280.0, + 1405.0, + 1314.0, + 295.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1309.0, + 1405.0, + 1309.0, + 1405.0, + 1346.0, + 293.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1342.0, + 1405.0, + 1342.0, + 1405.0, + 1376.0, + 295.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1370.0, + 1408.0, + 1370.0, + 1408.0, + 1406.0, + 293.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 200.0, + 1405.0, + 200.0, + 1405.0, + 240.0, + 293.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 231.0, + 1409.0, + 231.0, + 1409.0, + 270.0, + 293.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 260.0, + 1410.0, + 260.0, + 1410.0, + 300.0, + 292.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 294.0, + 1407.0, + 294.0, + 1407.0, + 330.0, + 295.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 324.0, + 1405.0, + 324.0, + 1405.0, + 360.0, + 295.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 354.0, + 1408.0, + 354.0, + 1408.0, + 390.0, + 293.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1493.0, + 1407.0, + 1493.0, + 1407.0, + 1526.0, + 296.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1522.0, + 1402.0, + 1522.0, + 1402.0, + 1555.0, + 296.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1551.0, + 1406.0, + 1551.0, + 1406.0, + 1587.0, + 293.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1582.0, + 1408.0, + 1582.0, + 1408.0, + 1619.0, + 296.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1612.0, + 1138.0, + 1612.0, + 1138.0, + 1650.0, + 293.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1416.0, + 1405.0, + 1416.0, + 1405.0, + 1452.0, + 295.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1445.0, + 1385.0, + 1445.0, + 1385.0, + 1482.0, + 294.0, + 1482.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1519, + 1405, + 1519, + 1405, + 2006, + 297, + 2006 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1233, + 1405, + 1233, + 1405, + 1507, + 298, + 1507 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 956, + 1405, + 956, + 1405, + 1141, + 298, + 1141 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 551, + 1404, + 551, + 1404, + 747, + 298, + 747 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 760, + 1404, + 760, + 1404, + 943, + 298, + 943 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 370, + 1404, + 370, + 1404, + 494, + 298, + 494 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 298, + 265, + 1404, + 265, + 1404, + 356, + 298, + 356 + ], + "score": 0.97 + }, + { + "category_id": 0, + "poly": [ + 299, + 1183, + 578, + 1183, + 578, + 1216, + 299, + 1216 + ], + "score": 0.918 + }, + { + "category_id": 0, + "poly": [ + 298, + 198, + 532, + 198, + 532, + 237, + 298, + 237 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 295, + 506, + 1394, + 506, + 1394, + 540, + 295, + 540 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2084, + 841, + 2084 + ], + "score": 0.704 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2084, + 841, + 2084 + ], + "score": 0.129 + }, + { + "category_id": 13, + "poly": [ + 1075, + 615, + 1289, + 615, + 1289, + 659, + 1075, + 659 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\mathbf { x } _ { 0 } = \\frac { \\mathbf { x } _ { t } - \\sigma _ { t } \\epsilon _ { \\theta } \\left( \\mathbf { x } _ { t } , t \\right) } { \\alpha _ { t } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 395, + 851, + 626, + 851, + 626, + 885, + 395, + 885 + ], + "score": 0.92, + "latex": "\\mathcal { N } ( \\mathbf { 0 } , I ) \\approx p _ { t = 1 } ( \\mathbf { x } _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 925, + 618, + 1016, + 618, + 1016, + 650, + 925, + 650 + ], + "score": 0.91, + "latex": "t { = } 1 0 ^ { - 3 }" + }, + { + "category_id": 13, + "poly": [ + 1049, + 551, + 1141, + 551, + 1141, + 582, + 1049, + 582 + ], + "score": 0.91, + "latex": "t { = } 1 0 ^ { - 3 }" + }, + { + "category_id": 13, + "poly": [ + 889, + 988, + 927, + 988, + 927, + 1021, + 889, + 1021 + ], + "score": 0.89, + "latex": "k _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 1033, + 1975, + 1093, + 1975, + 1093, + 2004, + 1033, + 2004 + ], + "score": 0.88, + "latex": "1 0 ^ { - 2 }" + }, + { + "category_id": 13, + "poly": [ + 342, + 1519, + 615, + 1519, + 615, + 1554, + 342, + 1554 + ], + "score": 0.87, + "latex": "\\mathrm { N F E s } \\in \\{ 1 0 , 1 5 , 2 0 , 2 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 630, + 1019, + 702, + 1019, + 702, + 1048, + 630, + 1048 + ], + "score": 0.87, + "latex": "1 4 . 4 \\%" + }, + { + "category_id": 13, + "poly": [ + 933, + 555, + 982, + 555, + 982, + 582, + 933, + 582 + ], + "score": 0.86, + "latex": "t { = } 1" + }, + { + "category_id": 13, + "poly": [ + 500, + 1019, + 571, + 1019, + 571, + 1048, + 500, + 1048 + ], + "score": 0.83, + "latex": "1 4 . 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 332, + 1019, + 404, + 1019, + 404, + 1049, + 332, + 1049 + ], + "score": 0.81, + "latex": "1 . 4 7 \\%" + }, + { + "category_id": 13, + "poly": [ + 414, + 1019, + 488, + 1019, + 488, + 1049, + 414, + 1049 + ], + "score": 0.8, + "latex": "2 . 8 3 \\%" + }, + { + "category_id": 13, + "poly": [ + 1178, + 1583, + 1214, + 1583, + 1214, + 1610, + 1178, + 1610 + ], + "score": 0.53, + "latex": "\\mathord { \\mathrm { : } } = 5" + }, + { + "category_id": 13, + "poly": [ + 352, + 1704, + 404, + 1704, + 404, + 1731, + 352, + 1731 + ], + "score": 0.38, + "latex": "\\scriptstyle { \\mathsf { s } } = 2 0" + }, + { + "category_id": 13, + "poly": [ + 297, + 1703, + 405, + 1703, + 405, + 1732, + 297, + 1732 + ], + "score": 0.32, + "latex": "\\mathrm { N F E s } { = } 2 0" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1181.0, + 582.0, + 1181.0, + 582.0, + 1223.0, + 293.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 191.0, + 538.0, + 191.0, + 538.0, + 247.0, + 288.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2092.0, + 840.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1518.0, + 341.0, + 1518.0, + 341.0, + 1556.0, + 294.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1518.0, + 1406.0, + 1518.0, + 1406.0, + 1556.0, + 616.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1548.0, + 1405.0, + 1548.0, + 1405.0, + 1586.0, + 294.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1582.0, + 1177.0, + 1582.0, + 1177.0, + 1616.0, + 295.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1582.0, + 1405.0, + 1582.0, + 1405.0, + 1616.0, + 1215.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1612.0, + 1405.0, + 1612.0, + 1405.0, + 1646.0, + 295.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1644.0, + 1403.0, + 1644.0, + 1403.0, + 1674.0, + 296.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1673.0, + 1406.0, + 1673.0, + 1406.0, + 1707.0, + 295.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1699.0, + 296.0, + 1699.0, + 296.0, + 1739.0, + 292.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1699.0, + 1406.0, + 1699.0, + 1406.0, + 1739.0, + 406.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1733.0, + 1407.0, + 1733.0, + 1407.0, + 1767.0, + 294.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1764.0, + 1405.0, + 1764.0, + 1405.0, + 1798.0, + 295.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1793.0, + 1406.0, + 1793.0, + 1406.0, + 1827.0, + 294.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1824.0, + 1406.0, + 1824.0, + 1406.0, + 1857.0, + 294.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1854.0, + 1405.0, + 1854.0, + 1405.0, + 1888.0, + 294.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1885.0, + 1405.0, + 1885.0, + 1405.0, + 1919.0, + 295.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1914.0, + 1405.0, + 1914.0, + 1405.0, + 1948.0, + 295.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1946.0, + 1405.0, + 1946.0, + 1405.0, + 1980.0, + 295.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1032.0, + 1972.0, + 1032.0, + 2010.0, + 294.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 1972.0, + 1113.0, + 1972.0, + 1113.0, + 2010.0, + 1094.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1232.0, + 1406.0, + 1232.0, + 1406.0, + 1269.0, + 295.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1264.0, + 1406.0, + 1264.0, + 1406.0, + 1296.0, + 296.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1295.0, + 1407.0, + 1295.0, + 1407.0, + 1327.0, + 296.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1322.0, + 1405.0, + 1322.0, + 1405.0, + 1358.0, + 295.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1352.0, + 1408.0, + 1352.0, + 1408.0, + 1391.0, + 293.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1382.0, + 1406.0, + 1382.0, + 1406.0, + 1419.0, + 293.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1414.0, + 1406.0, + 1414.0, + 1406.0, + 1448.0, + 293.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1444.0, + 1405.0, + 1444.0, + 1405.0, + 1480.0, + 295.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1475.0, + 1209.0, + 1475.0, + 1209.0, + 1508.0, + 295.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 958.0, + 1405.0, + 958.0, + 1405.0, + 990.0, + 296.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 987.0, + 888.0, + 987.0, + 888.0, + 1023.0, + 293.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 987.0, + 1407.0, + 987.0, + 1407.0, + 1023.0, + 928.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1018.0, + 331.0, + 1018.0, + 331.0, + 1054.0, + 295.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1018.0, + 413.0, + 1018.0, + 413.0, + 1054.0, + 405.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1018.0, + 499.0, + 1018.0, + 499.0, + 1054.0, + 489.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1018.0, + 629.0, + 1018.0, + 629.0, + 1054.0, + 572.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1018.0, + 1407.0, + 1018.0, + 1407.0, + 1054.0, + 703.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1045.0, + 1408.0, + 1045.0, + 1408.0, + 1085.0, + 292.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1079.0, + 1405.0, + 1079.0, + 1405.0, + 1115.0, + 295.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1108.0, + 1397.0, + 1108.0, + 1397.0, + 1145.0, + 293.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 551.0, + 932.0, + 551.0, + 932.0, + 588.0, + 296.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 551.0, + 1048.0, + 551.0, + 1048.0, + 588.0, + 983.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 551.0, + 1405.0, + 551.0, + 1405.0, + 588.0, + 1142.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 584.0, + 1402.0, + 584.0, + 1402.0, + 617.0, + 296.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 604.0, + 924.0, + 604.0, + 924.0, + 662.0, + 288.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 604.0, + 1074.0, + 604.0, + 1074.0, + 662.0, + 1017.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 604.0, + 1413.0, + 604.0, + 1413.0, + 662.0, + 1290.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 653.0, + 1406.0, + 653.0, + 1406.0, + 690.0, + 294.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 683.0, + 1409.0, + 683.0, + 1409.0, + 719.0, + 293.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 712.0, + 1318.0, + 712.0, + 1318.0, + 752.0, + 293.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 758.0, + 1406.0, + 758.0, + 1406.0, + 795.0, + 293.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 790.0, + 1405.0, + 790.0, + 1405.0, + 824.0, + 293.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 821.0, + 1406.0, + 821.0, + 1406.0, + 857.0, + 294.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 852.0, + 394.0, + 852.0, + 394.0, + 887.0, + 294.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 852.0, + 1405.0, + 852.0, + 1405.0, + 887.0, + 627.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 882.0, + 1406.0, + 882.0, + 1406.0, + 917.0, + 294.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 910.0, + 1352.0, + 910.0, + 1352.0, + 948.0, + 293.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 369.0, + 1405.0, + 369.0, + 1405.0, + 405.0, + 293.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 402.0, + 1405.0, + 402.0, + 1405.0, + 434.0, + 294.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 433.0, + 1408.0, + 433.0, + 1408.0, + 466.0, + 294.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 464.0, + 1234.0, + 464.0, + 1234.0, + 496.0, + 294.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 264.0, + 1405.0, + 264.0, + 1405.0, + 301.0, + 293.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 294.0, + 1409.0, + 294.0, + 1409.0, + 331.0, + 293.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 325.0, + 1407.0, + 325.0, + 1407.0, + 357.0, + 294.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 501.0, + 1397.0, + 501.0, + 1397.0, + 545.0, + 292.0, + 545.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1085, + 1405, + 1085, + 1405, + 1571, + 297, + 1571 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 299, + 1635, + 1012, + 1635, + 1012, + 1849, + 299, + 1849 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1854, + 1013, + 1854, + 1013, + 2007, + 299, + 2007 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 1037, + 1579, + 1414, + 1579, + 1414, + 1895, + 1037, + 1895 + ], + "score": 0.961 + }, + { + "category_id": 4, + "poly": [ + 1045, + 419, + 1383, + 419, + 1383, + 643, + 1045, + 643 + ], + "score": 0.959 + }, + { + "category_id": 5, + "poly": [ + 299, + 721, + 1024, + 721, + 1024, + 1052, + 299, + 1052 + ], + "score": 0.947, + "html": "
MethodNFEs=5NFEs=10NFEs=15NFEs=20NFEs=25
GENIE(ours) (*)11.25.284.493.943.64
GENIE (ours)13.95.974.493.943.67
DDIM[58] (*)27.611.27.355.875.16
DDIM[58]29.711.27.355.875.16
S-PNDM[63]35.910.36.615.204.51
F-PNDM[63]N/AN/A10.35.964.73
Euler-Maruyama32523016411280.3
FastDDIM [64] (t)19.90-5.05-
Learned Sampler 66] (t/*)12.47.865.904.724.25
Learned Sampler [66] (t)14.38.155.944.894.47
Analytic DDIM [65] (t)-14.0--5.71
CLD-SGM[60]334306236162106
VESDE-PC[57]461461461461462
" + }, + { + "category_id": 4, + "poly": [ + 1055, + 725, + 1404, + 725, + 1404, + 1058, + 1055, + 1058 + ], + "score": 0.938 + }, + { + "category_id": 4, + "poly": [ + 1034, + 1899, + 1406, + 1899, + 1406, + 1956, + 1034, + 1956 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 299, + 1594, + 650, + 1594, + 650, + 1627, + 299, + 1627 + ], + "score": 0.912 + }, + { + "category_id": 3, + "poly": [ + 323, + 134, + 1016, + 134, + 1016, + 687, + 323, + 687 + ], + "score": 0.847 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2085, + 840, + 2085 + ], + "score": 0.799 + }, + { + "category_id": 3, + "poly": [ + 1053, + 140, + 1368, + 140, + 1368, + 366, + 1053, + 366 + ], + "score": 0.4 + }, + { + "category_id": 13, + "poly": [ + 298, + 1916, + 419, + 1916, + 419, + 1947, + 298, + 1947 + ], + "score": 0.93, + "latex": "w \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t )" + }, + { + "category_id": 13, + "poly": [ + 619, + 1947, + 729, + 1947, + 729, + 1975, + 619, + 1975 + ], + "score": 0.92, + "latex": "w > 1 . 0" + }, + { + "category_id": 13, + "poly": [ + 379, + 1885, + 479, + 1885, + 479, + 1917, + 379, + 1917 + ], + "score": 0.92, + "latex": "\\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t )" + }, + { + "category_id": 13, + "poly": [ + 550, + 1884, + 1015, + 1884, + 1015, + 1917, + 550, + 1917 + ], + "score": 0.9, + "latex": "\\hat { \\epsilon } _ { \\theta } ( \\mathbf { x } _ { t } , t , c , w ) = ( 1 + w ) \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t , c ) -" + }, + { + "category_id": 13, + "poly": [ + 887, + 1917, + 1012, + 1917, + 1012, + 1948, + 887, + 1948 + ], + "score": 0.89, + "latex": "\\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t , c )" + }, + { + "category_id": 13, + "poly": [ + 977, + 1207, + 1045, + 1207, + 1045, + 1236, + 977, + 1236 + ], + "score": 0.88, + "latex": "100 \\%" + }, + { + "category_id": 13, + "poly": [ + 1052, + 209, + 1108, + 209, + 1108, + 233, + 1052, + 233 + ], + "score": 0.42, + "latex": "\\twoheadleftarrow" + }, + { + "category_id": 13, + "poly": [ + 1051, + 142, + 1108, + 142, + 1108, + 169, + 1051, + 169 + ], + "score": 0.37, + "latex": "\\twoheadleftarrow" + }, + { + "category_id": 13, + "poly": [ + 1051, + 177, + 1108, + 177, + 1108, + 201, + 1051, + 201 + ], + "score": 0.31, + "latex": "\\twoheadleftarrow" + }, + { + "category_id": 13, + "poly": [ + 1051, + 241, + 1108, + 241, + 1108, + 265, + 1051, + 265 + ], + "score": 0.26, + "latex": "\\twoheadleftarrow" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 1580.0, + 1087.0, + 1580.0, + 1087.0, + 1604.0, + 1060.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 1590.0, + 1397.0, + 1590.0, + 1397.0, + 1613.0, + 1273.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1609.0, + 1401.0, + 1609.0, + 1401.0, + 1634.0, + 1268.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1623.0, + 1088.0, + 1623.0, + 1088.0, + 1652.0, + 1058.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1629.0, + 1401.0, + 1629.0, + 1401.0, + 1655.0, + 1268.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 1651.0, + 1395.0, + 1651.0, + 1395.0, + 1674.0, + 1275.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1668.0, + 1088.0, + 1668.0, + 1088.0, + 1696.0, + 1058.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1669.0, + 1398.0, + 1669.0, + 1398.0, + 1694.0, + 1269.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1697.0, + 1088.0, + 1697.0, + 1088.0, + 1740.0, + 1034.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1689.0, + 1398.0, + 1689.0, + 1398.0, + 1714.0, + 1269.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1758.0, + 1089.0, + 1758.0, + 1089.0, + 1785.0, + 1059.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1767.0, + 1095.0, + 1767.0, + 1095.0, + 1775.0, + 1087.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1782.0, + 1110.0, + 1782.0, + 1110.0, + 1787.0, + 1105.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 1805.0, + 1086.0, + 1805.0, + 1086.0, + 1828.0, + 1067.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1823.0, + 1159.0, + 1823.0, + 1159.0, + 1843.0, + 1109.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1853.0, + 1098.0, + 1853.0, + 1098.0, + 1876.0, + 1077.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1850.0, + 1183.0, + 1850.0, + 1183.0, + 1877.0, + 1147.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 1851.0, + 1260.0, + 1851.0, + 1260.0, + 1877.0, + 1225.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1851.0, + 1336.0, + 1851.0, + 1336.0, + 1877.0, + 1302.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 1851.0, + 1414.0, + 1851.0, + 1414.0, + 1877.0, + 1379.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1876.0, + 1305.0, + 1876.0, + 1305.0, + 1898.0, + 1195.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1772.0, + 1360.0, + 1772.0, + 1360.0, + 1796.5, + 1213.0, + 1796.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1815.0, + 1154.0, + 1815.0, + 1154.0, + 1829.0, + 1120.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.75, + 1813.5, + 1208.75, + 1813.5, + 1208.75, + 1830.5, + 1161.75, + 1830.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 418.0, + 1385.0, + 418.0, + 1385.0, + 452.0, + 1043.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 448.0, + 1386.0, + 448.0, + 1386.0, + 478.0, + 1042.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 474.0, + 1386.0, + 474.0, + 1386.0, + 506.0, + 1043.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 505.0, + 1382.0, + 505.0, + 1382.0, + 533.0, + 1044.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 530.0, + 1384.0, + 530.0, + 1384.0, + 561.0, + 1044.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 558.0, + 1382.0, + 558.0, + 1382.0, + 589.0, + 1044.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 587.0, + 1382.0, + 587.0, + 1382.0, + 617.0, + 1044.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 612.0, + 1263.0, + 612.0, + 1263.0, + 645.0, + 1043.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 725.0, + 1404.0, + 725.0, + 1404.0, + 755.0, + 1054.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 753.0, + 1406.0, + 753.0, + 1406.0, + 783.0, + 1052.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 781.0, + 1407.0, + 781.0, + 1407.0, + 811.0, + 1054.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 808.0, + 1406.0, + 808.0, + 1406.0, + 838.0, + 1054.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 836.0, + 1406.0, + 836.0, + 1406.0, + 866.0, + 1055.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 864.0, + 1406.0, + 864.0, + 1406.0, + 893.0, + 1055.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 893.0, + 1407.0, + 893.0, + 1407.0, + 920.0, + 1055.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 918.0, + 1406.0, + 918.0, + 1406.0, + 951.0, + 1053.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 948.0, + 1405.0, + 948.0, + 1405.0, + 978.0, + 1055.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 973.0, + 1406.0, + 973.0, + 1406.0, + 1009.0, + 1051.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1000.0, + 1405.0, + 1000.0, + 1405.0, + 1033.0, + 1053.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1029.0, + 1190.0, + 1029.0, + 1190.0, + 1063.0, + 1052.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1898.0, + 1406.0, + 1898.0, + 1406.0, + 1929.0, + 1033.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1926.0, + 1400.0, + 1926.0, + 1400.0, + 1958.0, + 1032.0, + 1958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1588.0, + 653.0, + 1588.0, + 653.0, + 1633.0, + 293.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 138.0, + 374.0, + 138.0, + 374.0, + 173.0, + 342.0, + 173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 137.0, + 670.0, + 137.0, + 670.0, + 162.0, + 589.0, + 162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 138.0, + 1014.0, + 138.0, + 1014.0, + 166.0, + 940.0, + 166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 167.0, + 372.0, + 167.0, + 372.0, + 197.0, + 344.0, + 197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 193.0, + 373.0, + 193.0, + 373.0, + 225.0, + 345.0, + 225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 208.0, + 453.0, + 208.0, + 453.0, + 218.0, + 445.0, + 218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 221.0, + 372.0, + 221.0, + 372.0, + 250.0, + 344.0, + 250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 246.0, + 345.0, + 246.0, + 345.0, + 284.0, + 324.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 246.0, + 392.0, + 246.0, + 392.0, + 283.0, + 373.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 247.0, + 370.0, + 247.0, + 370.0, + 275.0, + 345.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 264.0, + 459.0, + 264.0, + 459.0, + 297.0, + 443.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 272.0, + 375.0, + 272.0, + 375.0, + 396.0, + 342.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 279.0, + 520.0, + 279.0, + 520.0, + 289.0, + 511.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 312.0, + 453.0, + 312.0, + 453.0, + 322.0, + 445.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 325.0, + 520.0, + 325.0, + 520.0, + 338.0, + 509.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 314.0, + 995.0, + 314.0, + 995.0, + 334.0, + 985.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 335.0, + 521.0, + 335.0, + 521.0, + 348.0, + 509.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 341.0, + 586.0, + 341.0, + 586.0, + 367.0, + 572.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 353.0, + 649.0, + 353.0, + 649.0, + 366.0, + 641.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 367.0, + 653.0, + 367.0, + 653.0, + 376.0, + 644.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 401.0, + 373.0, + 401.0, + 373.0, + 432.0, + 343.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 400.0, + 669.0, + 400.0, + 669.0, + 424.0, + 551.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 400.0, + 1013.0, + 400.0, + 1013.0, + 423.0, + 858.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 431.0, + 372.0, + 431.0, + 372.0, + 463.0, + 344.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 427.0, + 521.0, + 427.0, + 521.0, + 442.0, + 509.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 462.0, + 373.0, + 462.0, + 373.0, + 492.0, + 344.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 488.0, + 372.0, + 488.0, + 372.0, + 520.0, + 344.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 506.0, + 372.0, + 506.0, + 372.0, + 548.0, + 324.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 520.0, + 867.0, + 520.0, + 867.0, + 531.0, + 854.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 544.0, + 377.0, + 544.0, + 377.0, + 653.0, + 341.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 544.0, + 453.0, + 544.0, + 453.0, + 555.0, + 445.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 608.0, + 649.0, + 608.0, + 649.0, + 617.0, + 641.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 590.0, + 994.0, + 590.0, + 994.0, + 630.0, + 980.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 619.0, + 651.0, + 619.0, + 651.0, + 636.0, + 641.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 633.0, + 585.0, + 633.0, + 585.0, + 646.0, + 573.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 649.0, + 392.0, + 649.0, + 392.0, + 670.0, + 376.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 647.0, + 463.0, + 647.0, + 463.0, + 672.0, + 437.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 646.0, + 528.0, + 646.0, + 528.0, + 672.0, + 502.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 645.0, + 595.0, + 645.0, + 595.0, + 673.0, + 564.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 646.0, + 658.0, + 646.0, + 658.0, + 672.0, + 631.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 649.0, + 737.0, + 649.0, + 737.0, + 669.0, + 721.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 646.0, + 809.0, + 646.0, + 809.0, + 671.0, + 782.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 646.0, + 874.0, + 646.0, + 874.0, + 672.0, + 847.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 645.0, + 940.0, + 645.0, + 940.0, + 673.0, + 909.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 646.0, + 1003.0, + 646.0, + 1003.0, + 672.0, + 976.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 668.0, + 543.0, + 668.0, + 543.0, + 692.0, + 496.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 668.0, + 889.0, + 668.0, + 889.0, + 692.0, + 841.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2090.0, + 838.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 139.0, + 1257.0, + 139.0, + 1257.0, + 174.0, + 1109.0, + 174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 173.0, + 1230.0, + 173.0, + 1230.0, + 205.0, + 1109.0, + 205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 202.0, + 1258.0, + 202.0, + 1258.0, + 237.0, + 1109.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 233.0, + 1257.0, + 233.0, + 1257.0, + 270.0, + 1109.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 278.0, + 1090.0, + 278.0, + 1090.0, + 295.0, + 1074.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 267.0, + 1369.0, + 267.0, + 1369.0, + 302.0, + 1103.0, + 302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 299.0, + 1306.0, + 299.0, + 1306.0, + 334.0, + 1104.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 332.0, + 1357.0, + 332.0, + 1357.0, + 367.0, + 1104.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1083.0, + 1406.0, + 1083.0, + 1406.0, + 1122.0, + 292.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1117.0, + 1406.0, + 1117.0, + 1406.0, + 1151.0, + 295.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1146.0, + 1406.0, + 1146.0, + 1406.0, + 1180.0, + 295.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1177.0, + 1405.0, + 1177.0, + 1405.0, + 1211.0, + 295.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1206.0, + 976.0, + 1206.0, + 976.0, + 1240.0, + 295.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1206.0, + 1406.0, + 1206.0, + 1406.0, + 1240.0, + 1046.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1236.0, + 1407.0, + 1236.0, + 1407.0, + 1271.0, + 295.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1267.0, + 1406.0, + 1267.0, + 1406.0, + 1305.0, + 295.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1297.0, + 1406.0, + 1297.0, + 1406.0, + 1332.0, + 296.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1326.0, + 1406.0, + 1326.0, + 1406.0, + 1365.0, + 292.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1357.0, + 1405.0, + 1357.0, + 1405.0, + 1391.0, + 295.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1389.0, + 1405.0, + 1389.0, + 1405.0, + 1423.0, + 295.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1417.0, + 1403.0, + 1417.0, + 1403.0, + 1451.0, + 295.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1449.0, + 1405.0, + 1449.0, + 1405.0, + 1483.0, + 295.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1479.0, + 1403.0, + 1479.0, + 1403.0, + 1514.0, + 294.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1510.0, + 1403.0, + 1510.0, + 1403.0, + 1544.0, + 295.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1538.0, + 771.0, + 1538.0, + 771.0, + 1575.0, + 294.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1634.0, + 1013.0, + 1634.0, + 1013.0, + 1667.0, + 297.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1664.0, + 1013.0, + 1664.0, + 1013.0, + 1698.0, + 294.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1693.0, + 1013.0, + 1693.0, + 1013.0, + 1728.0, + 293.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1724.0, + 1014.0, + 1724.0, + 1014.0, + 1760.0, + 294.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1757.0, + 1014.0, + 1757.0, + 1014.0, + 1790.0, + 295.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1786.0, + 1016.0, + 1786.0, + 1016.0, + 1820.0, + 294.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1817.0, + 771.0, + 1817.0, + 771.0, + 1850.0, + 295.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1851.0, + 1013.0, + 1851.0, + 1013.0, + 1887.0, + 296.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1883.0, + 378.0, + 1883.0, + 378.0, + 1917.0, + 294.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 1883.0, + 549.0, + 1883.0, + 549.0, + 1917.0, + 480.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 1914.0, + 886.0, + 1914.0, + 886.0, + 1948.0, + 420.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1944.0, + 618.0, + 1944.0, + 618.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1944.0, + 1018.0, + 1944.0, + 1018.0, + 1979.0, + 730.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1976.0, + 694.0, + 1976.0, + 694.0, + 2008.0, + 297.0, + 2008.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 874, + 1404, + 874, + 1404, + 967, + 298, + 967 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1642, + 936, + 1642, + 936, + 1824, + 299, + 1824 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 972, + 1403, + 972, + 1403, + 1156, + 297, + 1156 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1383, + 1403, + 1383, + 1403, + 1566, + 297, + 1566 + ], + "score": 0.972 + }, + { + "category_id": 5, + "poly": [ + 972, + 1660, + 1377, + 1660, + 1377, + 1795, + 972, + 1795 + ], + "score": 0.971, + "html": "
MethodNFEs=5NFEs=10NFEs=15
GENIE (ours)5.534.904.83
DDIM [58]9.476.645.85
S-PNDM[63]14.611.08.83
F-PNDM [[63]N/AN/A11.7
" + }, + { + "category_id": 5, + "poly": [ + 808, + 1216, + 1397, + 1216, + 1397, + 1353, + 808, + 1353 + ], + "score": 0.971, + "html": "
AblationNFEs=5NFEs=10NFEs=15NFEs=20NFEs=25
Standard13.96.044.493.943.67
No mixed14.76.324.824.314.10
No weighting14.87.455.895.174.80
Bigger model13.75.584.464.053.77
" + }, + { + "category_id": 1, + "poly": [ + 299, + 1825, + 1406, + 1825, + 1406, + 2008, + 299, + 2008 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 298, + 1232, + 783, + 1232, + 783, + 1383, + 298, + 1383 + ], + "score": 0.967 + }, + { + "category_id": 6, + "poly": [ + 955, + 1586, + 1404, + 1586, + 1404, + 1643, + 955, + 1643 + ], + "score": 0.931 + }, + { + "category_id": 3, + "poly": [ + 304, + 135, + 1367, + 135, + 1367, + 729, + 304, + 729 + ], + "score": 0.927 + }, + { + "category_id": 8, + "poly": [ + 507, + 812, + 1189, + 812, + 1189, + 851, + 507, + 851 + ], + "score": 0.927 + }, + { + "category_id": 0, + "poly": [ + 299, + 1187, + 560, + 1187, + 560, + 1218, + 299, + 1218 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 298, + 1597, + 508, + 1597, + 508, + 1630, + 298, + 1630 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1352, + 815, + 1400, + 815, + 1400, + 845, + 1352, + 845 + ], + "score": 0.882 + }, + { + "category_id": 6, + "poly": [ + 817, + 1185, + 1385, + 1185, + 1385, + 1212, + 817, + 1212 + ], + "score": 0.869 + }, + { + "category_id": 4, + "poly": [ + 1223, + 145, + 1400, + 145, + 1400, + 371, + 1223, + 371 + ], + "score": 0.835 + }, + { + "category_id": 4, + "poly": [ + 295, + 736, + 1402, + 736, + 1402, + 792, + 295, + 792 + ], + "score": 0.831 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 859, + 2061, + 859, + 2084, + 840, + 2084 + ], + "score": 0.798 + }, + { + "category_id": 1, + "poly": [ + 295, + 736, + 1402, + 736, + 1402, + 792, + 295, + 792 + ], + "score": 0.121 + }, + { + "category_id": 13, + "poly": [ + 720, + 874, + 881, + 874, + 881, + 908, + 720, + 908 + ], + "score": 0.93, + "latex": "d _ { \\gamma _ { t } } \\mathbf { \\epsilon } _ { \\theta } ( \\mathbf { x } _ { t } , t , c )" + }, + { + "category_id": 13, + "poly": [ + 930, + 874, + 1066, + 874, + 1066, + 908, + 930, + 908 + ], + "score": 0.93, + "latex": "d _ { \\gamma _ { t } } \\mathbf { \\epsilon } _ { \\theta } ( \\mathbf { x } _ { t } , t )" + }, + { + "category_id": 13, + "poly": [ + 868, + 1383, + 980, + 1383, + 980, + 1416, + 868, + 1416 + ], + "score": 0.93, + "latex": "g _ { \\mathrm { d } } ( t ) = 1" + }, + { + "category_id": 13, + "poly": [ + 954, + 1413, + 1080, + 1413, + 1080, + 1447, + 954, + 1447 + ], + "score": 0.93, + "latex": "g _ { \\mathrm { d } } ( t ) \\stackrel { \\cdot } { = } \\gamma _ { t } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 505, + 1323, + 573, + 1323, + 573, + 1356, + 505, + 1356 + ], + "score": 0.91, + "latex": "d _ { \\gamma _ { t } } \\epsilon _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 525, + 1764, + 649, + 1764, + 649, + 1793, + 525, + 1793 + ], + "score": 0.89, + "latex": "1 2 8 \\times 1 2 8" + }, + { + "category_id": 14, + "poly": [ + 510, + 811, + 1189, + 811, + 1189, + 851, + 510, + 851 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { d _ { \\gamma _ { t } } \\hat { \\epsilon } _ { \\theta } ( \\mathbf { x } _ { t } , t , c , w ) = ( 1 + w ) d _ { \\gamma _ { t } } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t , c ) - w d _ { \\gamma _ { t } } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 296, + 1794, + 589, + 1794, + 589, + 1823, + 296, + 1823 + ], + "score": 0.88, + "latex": "1 2 8 \\times 1 2 8 \\to 5 1 2 \\times 5 1 2" + }, + { + "category_id": 13, + "poly": [ + 298, + 764, + 329, + 764, + 329, + 790, + 298, + 790 + ], + "score": 0.84, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 760, + 1886, + 871, + 1886, + 871, + 1914, + 760, + 1914 + ], + "score": 0.51, + "latex": "\\mathrm { N F E s } { = } 1 5" + }, + { + "category_id": 13, + "poly": [ + 296, + 1915, + 395, + 1915, + 395, + 1944, + 296, + 1944 + ], + "score": 0.44, + "latex": "\\mathrm { N F E s } { = } 5" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1583.0, + 1408.0, + 1583.0, + 1408.0, + 1619.0, + 953.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1614.0, + 1224.0, + 1614.0, + 1224.0, + 1643.0, + 956.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 142.0, + 392.0, + 142.0, + 392.0, + 162.0, + 327.0, + 162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 140.0, + 496.0, + 140.0, + 496.0, + 163.0, + 419.0, + 163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 140.0, + 593.0, + 140.0, + 593.0, + 163.0, + 516.0, + 163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 142.0, + 693.0, + 142.0, + 693.0, + 162.0, + 626.0, + 162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 140.0, + 798.0, + 140.0, + 798.0, + 163.0, + 719.0, + 163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 140.0, + 895.0, + 140.0, + 895.0, + 163.0, + 815.0, + 163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 142.0, + 993.0, + 142.0, + 993.0, + 162.0, + 925.0, + 162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 140.0, + 1096.0, + 140.0, + 1096.0, + 163.0, + 1018.0, + 163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 140.0, + 1193.0, + 140.0, + 1193.0, + 163.0, + 1115.0, + 163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 145.0, + 1310.0, + 145.0, + 1310.0, + 181.0, + 1221.0, + 181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 171.0, + 935.0, + 171.0, + 935.0, + 242.0, + 693.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 171.0, + 1377.0, + 171.0, + 1377.0, + 209.0, + 1220.0, + 209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 203.0, + 1328.0, + 203.0, + 1328.0, + 235.0, + 1217.0, + 235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 212.0, + 1371.0, + 212.0, + 1371.0, + 222.0, + 1361.0, + 222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 227.0, + 1377.0, + 227.0, + 1377.0, + 266.0, + 1218.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 253.0, + 1310.0, + 253.0, + 1310.0, + 293.0, + 1218.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 257.0, + 1377.0, + 257.0, + 1377.0, + 290.0, + 1333.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 283.0, + 1296.0, + 283.0, + 1296.0, + 319.0, + 1218.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 284.0, + 1377.0, + 284.0, + 1377.0, + 319.0, + 1323.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 322.0, + 781.0, + 322.0, + 781.0, + 334.0, + 754.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 318.0, + 889.0, + 318.0, + 889.0, + 334.0, + 846.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 312.0, + 1105.0, + 312.0, + 1105.0, + 338.0, + 1046.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 310.0, + 1378.0, + 310.0, + 1378.0, + 349.0, + 1218.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 339.0, + 1375.0, + 339.0, + 1375.0, + 374.0, + 1221.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 377.0, + 528.0, + 377.0, + 528.0, + 403.0, + 386.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 377.0, + 827.0, + 377.0, + 827.0, + 403.0, + 686.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 377.0, + 1127.0, + 377.0, + 1127.0, + 403.0, + 987.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 422.0, + 441.0, + 422.0, + 441.0, + 449.0, + 358.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 424.0, + 728.0, + 424.0, + 728.0, + 445.0, + 643.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 424.0, + 867.0, + 424.0, + 867.0, + 448.0, + 780.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 424.0, + 1005.0, + 424.0, + 1005.0, + 446.0, + 911.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 457.0, + 1100.0, + 457.0, + 1100.0, + 478.0, + 1078.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 459.0, + 1373.0, + 459.0, + 1373.0, + 483.0, + 1320.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 477.0, + 1371.0, + 477.0, + 1371.0, + 501.0, + 1321.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 490.0, + 328.0, + 490.0, + 328.0, + 534.0, + 311.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 513.0, + 1079.0, + 513.0, + 1079.0, + 630.0, + 1052.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 527.0, + 1098.0, + 527.0, + 1098.0, + 548.0, + 1079.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 593.0, + 1036.0, + 593.0, + 1036.0, + 742.0, + 559.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 596.0, + 1100.0, + 596.0, + 1100.0, + 615.0, + 1079.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 626.0, + 1098.0, + 626.0, + 1098.0, + 643.0, + 1085.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 633.0, + 329.0, + 633.0, + 329.0, + 691.0, + 309.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 655.0, + 1098.0, + 655.0, + 1098.0, + 681.0, + 1085.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 691.0, + 1117.0, + 691.0, + 1117.0, + 704.0, + 1100.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 689.0, + 1146.0, + 689.0, + 1146.0, + 707.0, + 1121.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 689.0, + 1204.0, + 689.0, + 1204.0, + 707.0, + 1180.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 689.0, + 1262.0, + 689.0, + 1262.0, + 708.0, + 1238.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 689.0, + 1319.0, + 689.0, + 1319.0, + 708.0, + 1295.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 690.0, + 1370.0, + 690.0, + 1370.0, + 706.0, + 1353.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 706.0, + 1263.0, + 706.0, + 1263.0, + 731.0, + 1218.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.25, + 314.0, + 1202.25, + 314.0, + 1202.25, + 338.0, + 1142.25, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 425.5, + 568.0, + 425.5, + 568.0, + 442.0, + 509.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1184.0, + 564.0, + 1184.0, + 564.0, + 1223.0, + 294.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1590.0, + 512.0, + 1590.0, + 512.0, + 1641.0, + 291.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1181.0, + 1385.0, + 1181.0, + 1385.0, + 1217.0, + 818.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 144.0, + 1310.0, + 144.0, + 1310.0, + 185.0, + 1219.0, + 185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1369.0, + 145.0, + 1403.0, + 145.0, + 1403.0, + 180.0, + 1369.0, + 180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 175.0, + 1378.0, + 175.0, + 1378.0, + 205.0, + 1222.0, + 205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 203.0, + 1327.0, + 203.0, + 1327.0, + 237.0, + 1222.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1357.0, + 203.0, + 1400.0, + 203.0, + 1400.0, + 234.0, + 1357.0, + 234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 230.0, + 1401.0, + 230.0, + 1401.0, + 263.0, + 1221.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 258.0, + 1306.0, + 258.0, + 1306.0, + 289.0, + 1221.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 258.0, + 1402.0, + 258.0, + 1402.0, + 290.0, + 1335.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 286.0, + 1293.0, + 286.0, + 1293.0, + 316.0, + 1221.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 283.0, + 1399.0, + 283.0, + 1399.0, + 317.0, + 1324.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 314.0, + 1399.0, + 314.0, + 1399.0, + 347.0, + 1223.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 339.0, + 1390.0, + 339.0, + 1390.0, + 375.0, + 1222.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 734.0, + 1405.0, + 734.0, + 1405.0, + 766.0, + 295.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 759.0, + 1148.0, + 759.0, + 1148.0, + 797.0, + 330.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 871.0, + 719.0, + 871.0, + 719.0, + 912.0, + 293.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 871.0, + 929.0, + 871.0, + 929.0, + 912.0, + 882.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 871.0, + 1406.0, + 871.0, + 1406.0, + 912.0, + 1067.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 904.0, + 1405.0, + 904.0, + 1405.0, + 940.0, + 292.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 933.0, + 1377.0, + 933.0, + 1377.0, + 972.0, + 293.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1640.0, + 937.0, + 1640.0, + 937.0, + 1675.0, + 296.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1672.0, + 937.0, + 1672.0, + 937.0, + 1706.0, + 294.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1699.0, + 941.0, + 1699.0, + 941.0, + 1738.0, + 293.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1734.0, + 937.0, + 1734.0, + 937.0, + 1764.0, + 294.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1761.0, + 524.0, + 1761.0, + 524.0, + 1797.0, + 294.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1761.0, + 938.0, + 1761.0, + 938.0, + 1797.0, + 650.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1790.0, + 295.0, + 1790.0, + 295.0, + 1828.0, + 292.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1790.0, + 937.0, + 1790.0, + 937.0, + 1828.0, + 590.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 971.0, + 1405.0, + 971.0, + 1405.0, + 1007.0, + 295.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 999.0, + 1408.0, + 999.0, + 1408.0, + 1042.0, + 292.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1034.0, + 1405.0, + 1034.0, + 1405.0, + 1067.0, + 292.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1065.0, + 1403.0, + 1065.0, + 1403.0, + 1097.0, + 295.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1092.0, + 1408.0, + 1092.0, + 1408.0, + 1130.0, + 292.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1124.0, + 977.0, + 1124.0, + 977.0, + 1160.0, + 292.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1381.0, + 867.0, + 1381.0, + 867.0, + 1419.0, + 291.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1381.0, + 1406.0, + 1381.0, + 1406.0, + 1419.0, + 981.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1409.0, + 953.0, + 1409.0, + 953.0, + 1452.0, + 291.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 1409.0, + 1408.0, + 1409.0, + 1408.0, + 1452.0, + 1081.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1443.0, + 1407.0, + 1443.0, + 1407.0, + 1478.0, + 295.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1472.0, + 1407.0, + 1472.0, + 1407.0, + 1511.0, + 293.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1504.0, + 1405.0, + 1504.0, + 1405.0, + 1539.0, + 293.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1535.0, + 988.0, + 1535.0, + 988.0, + 1567.0, + 296.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1823.0, + 1406.0, + 1823.0, + 1406.0, + 1858.0, + 296.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1853.0, + 1406.0, + 1853.0, + 1406.0, + 1889.0, + 293.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1884.0, + 759.0, + 1884.0, + 759.0, + 1919.0, + 296.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1884.0, + 1406.0, + 1884.0, + 1406.0, + 1919.0, + 872.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1913.0, + 1408.0, + 1913.0, + 1408.0, + 1949.0, + 396.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1946.0, + 1409.0, + 1946.0, + 1409.0, + 1981.0, + 294.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1976.0, + 972.0, + 1976.0, + 972.0, + 2011.0, + 294.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1231.0, + 786.0, + 1231.0, + 786.0, + 1264.0, + 297.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1260.0, + 783.0, + 1260.0, + 783.0, + 1294.0, + 295.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1291.0, + 786.0, + 1291.0, + 786.0, + 1324.0, + 294.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1321.0, + 504.0, + 1321.0, + 504.0, + 1355.0, + 295.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1321.0, + 783.0, + 1321.0, + 783.0, + 1355.0, + 574.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1354.0, + 784.0, + 1354.0, + 784.0, + 1384.0, + 294.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 734.0, + 1405.0, + 734.0, + 1405.0, + 766.0, + 295.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 759.0, + 1148.0, + 759.0, + 1148.0, + 797.0, + 330.0, + 797.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1311, + 1405, + 1311, + 1405, + 1585, + 298, + 1585 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1591, + 1404, + 1591, + 1404, + 1835, + 298, + 1835 + ], + "score": 0.98 + }, + { + "category_id": 3, + "poly": [ + 406, + 178, + 1291, + 178, + 1291, + 1065, + 406, + 1065 + ], + "score": 0.978 + }, + { + "category_id": 4, + "poly": [ + 297, + 1084, + 1404, + 1084, + 1404, + 1237, + 297, + 1237 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 299, + 1945, + 1402, + 1945, + 1402, + 2007, + 299, + 2007 + ], + "score": 0.945 + }, + { + "category_id": 0, + "poly": [ + 301, + 1877, + 575, + 1877, + 575, + 1914, + 301, + 1914 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 298, + 1267, + 523, + 1267, + 523, + 1303, + 298, + 1303 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 837, + 2062, + 865, + 2062, + 865, + 2085, + 837, + 2085 + ], + "score": 0.851 + }, + { + "category_id": 13, + "poly": [ + 901, + 1084, + 1186, + 1084, + 1186, + 1115, + 901, + 1115 + ], + "score": 0.9, + "latex": "1 2 8 \\times 1 2 8 \\to 5 1 2 \\times 5 1 2" + }, + { + "category_id": 13, + "poly": [ + 499, + 1176, + 620, + 1176, + 620, + 1205, + 499, + 1205 + ], + "score": 0.9, + "latex": "1 2 8 \\times 1 2 8" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1085.0, + 900.0, + 1085.0, + 900.0, + 1118.0, + 296.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 1085.0, + 1405.0, + 1085.0, + 1405.0, + 1118.0, + 1187.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1114.0, + 1405.0, + 1114.0, + 1405.0, + 1151.0, + 295.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1145.0, + 1405.0, + 1145.0, + 1405.0, + 1182.0, + 295.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1172.0, + 498.0, + 1172.0, + 498.0, + 1213.0, + 294.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1172.0, + 1406.0, + 1172.0, + 1406.0, + 1213.0, + 621.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1207.0, + 1206.0, + 1207.0, + 1206.0, + 1240.0, + 295.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1872.0, + 579.0, + 1872.0, + 579.0, + 1922.0, + 295.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1260.0, + 528.0, + 1260.0, + 528.0, + 1311.0, + 290.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1310.0, + 1405.0, + 1310.0, + 1405.0, + 1347.0, + 295.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1339.0, + 1405.0, + 1339.0, + 1405.0, + 1375.0, + 295.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1371.0, + 1406.0, + 1371.0, + 1406.0, + 1408.0, + 293.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1401.0, + 1406.0, + 1401.0, + 1406.0, + 1438.0, + 295.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1430.0, + 1406.0, + 1430.0, + 1406.0, + 1469.0, + 293.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1464.0, + 1403.0, + 1464.0, + 1403.0, + 1496.0, + 296.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1493.0, + 1407.0, + 1493.0, + 1407.0, + 1530.0, + 295.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1521.0, + 1405.0, + 1521.0, + 1405.0, + 1560.0, + 293.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1550.0, + 608.0, + 1550.0, + 608.0, + 1590.0, + 293.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1590.0, + 1404.0, + 1590.0, + 1404.0, + 1626.0, + 295.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1625.0, + 1405.0, + 1625.0, + 1405.0, + 1654.0, + 296.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1647.0, + 1405.0, + 1647.0, + 1405.0, + 1689.0, + 292.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1683.0, + 1404.0, + 1683.0, + 1404.0, + 1717.0, + 296.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1712.0, + 1404.0, + 1712.0, + 1404.0, + 1746.0, + 293.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1739.0, + 1405.0, + 1739.0, + 1405.0, + 1779.0, + 293.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1774.0, + 1406.0, + 1774.0, + 1406.0, + 1807.0, + 294.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1804.0, + 973.0, + 1804.0, + 973.0, + 1837.0, + 296.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1981.0, + 294.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1394.0, + 1972.0, + 1394.0, + 2011.0, + 294.0, + 2011.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 302, + 244, + 1411, + 244, + 1411, + 2017, + 302, + 2017 + ], + "score": 0.912 + }, + { + "category_id": 0, + "poly": [ + 298, + 200, + 455, + 200, + 455, + 234, + 298, + 234 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 835, + 2061, + 864, + 2061, + 864, + 2087, + 835, + 2087 + ], + "score": 0.823 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 197.0, + 460.0, + 197.0, + 460.0, + 239.0, + 295.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2058.0, + 869.0, + 2058.0, + 869.0, + 2097.0, + 831.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 245.0, + 1407.0, + 245.0, + 1407.0, + 282.0, + 320.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 274.0, + 837.0, + 274.0, + 837.0, + 309.0, + 357.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 321.0, + 1409.0, + 321.0, + 1409.0, + 362.0, + 314.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 354.0, + 1274.0, + 354.0, + 1274.0, + 385.0, + 363.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 403.0, + 1405.0, + 403.0, + 1405.0, + 434.0, + 322.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 428.0, + 912.0, + 428.0, + 912.0, + 463.0, + 359.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 477.0, + 1407.0, + 477.0, + 1407.0, + 514.0, + 318.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 506.0, + 761.0, + 506.0, + 761.0, + 541.0, + 359.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 557.0, + 1405.0, + 557.0, + 1405.0, + 588.0, + 322.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 584.0, + 1103.0, + 584.0, + 1103.0, + 615.0, + 361.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 631.0, + 1405.0, + 631.0, + 1405.0, + 668.0, + 318.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 662.0, + 747.0, + 662.0, + 747.0, + 693.0, + 363.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 709.0, + 1409.0, + 709.0, + 1409.0, + 744.0, + 316.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 738.0, + 878.0, + 738.0, + 878.0, + 770.0, + 365.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 785.0, + 1407.0, + 785.0, + 1407.0, + 820.0, + 316.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 816.0, + 1125.0, + 816.0, + 1125.0, + 848.0, + 363.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 863.0, + 1407.0, + 863.0, + 1407.0, + 900.0, + 320.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 891.0, + 1246.0, + 891.0, + 1246.0, + 926.0, + 359.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 937.0, + 1405.0, + 937.0, + 1405.0, + 976.0, + 304.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 971.0, + 1113.0, + 971.0, + 1113.0, + 1002.0, + 363.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1017.0, + 1407.0, + 1017.0, + 1407.0, + 1052.0, + 304.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1047.0, + 622.0, + 1047.0, + 622.0, + 1078.0, + 361.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1093.0, + 1405.0, + 1093.0, + 1405.0, + 1130.0, + 306.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1125.0, + 1238.0, + 1125.0, + 1238.0, + 1156.0, + 363.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1170.0, + 1409.0, + 1170.0, + 1409.0, + 1207.0, + 304.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1199.0, + 1318.0, + 1199.0, + 1318.0, + 1236.0, + 359.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1250.0, + 1407.0, + 1250.0, + 1407.0, + 1281.0, + 308.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1277.0, + 1385.0, + 1277.0, + 1385.0, + 1308.0, + 363.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1328.0, + 1403.0, + 1328.0, + 1403.0, + 1359.0, + 308.0, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1353.0, + 1407.0, + 1353.0, + 1407.0, + 1390.0, + 361.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1382.0, + 890.0, + 1382.0, + 890.0, + 1413.0, + 363.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1429.0, + 1403.0, + 1429.0, + 1403.0, + 1466.0, + 306.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1460.0, + 707.0, + 1460.0, + 707.0, + 1491.0, + 363.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1505.0, + 1405.0, + 1505.0, + 1405.0, + 1542.0, + 306.0, + 1542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1536.0, + 1129.0, + 1536.0, + 1129.0, + 1568.0, + 365.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1583.0, + 1405.0, + 1583.0, + 1405.0, + 1620.0, + 306.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1614.0, + 1403.0, + 1614.0, + 1403.0, + 1646.0, + 365.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1642.0, + 841.0, + 1642.0, + 841.0, + 1673.0, + 363.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1690.0, + 1407.0, + 1690.0, + 1407.0, + 1722.0, + 308.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1718.0, + 1304.0, + 1718.0, + 1304.0, + 1749.0, + 363.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1763.0, + 1409.0, + 1763.0, + 1409.0, + 1804.0, + 302.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1796.0, + 1316.0, + 1796.0, + 1316.0, + 1827.0, + 363.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1845.0, + 1405.0, + 1845.0, + 1405.0, + 1876.0, + 310.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1872.0, + 1405.0, + 1872.0, + 1405.0, + 1903.0, + 361.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1893.0, + 735.0, + 1893.0, + 735.0, + 1934.0, + 358.0, + 1934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1942.0, + 1407.0, + 1942.0, + 1407.0, + 1987.0, + 302.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1977.0, + 1113.0, + 1977.0, + 1113.0, + 2008.0, + 363.0, + 2008.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 303, + 121, + 1410, + 121, + 1410, + 2025, + 303, + 2025 + ], + "score": 0.825 + }, + { + "category_id": 2, + "poly": [ + 835, + 2061, + 866, + 2061, + 866, + 2087, + 835, + 2087 + ], + "score": 0.82 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 870.0, + 2057.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 203.0, + 1406.0, + 203.0, + 1406.0, + 236.0, + 310.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 230.0, + 1104.0, + 230.0, + 1104.0, + 263.0, + 362.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 282.0, + 1406.0, + 282.0, + 1406.0, + 321.0, + 305.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 309.0, + 919.0, + 309.0, + 919.0, + 347.0, + 360.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 361.0, + 1408.0, + 361.0, + 1408.0, + 401.0, + 307.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 392.0, + 1397.0, + 392.0, + 1397.0, + 426.0, + 364.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 443.0, + 1408.0, + 443.0, + 1408.0, + 482.0, + 307.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 470.0, + 1349.0, + 470.0, + 1349.0, + 509.0, + 360.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 522.0, + 1408.0, + 522.0, + 1408.0, + 559.0, + 303.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 553.0, + 1102.0, + 553.0, + 1102.0, + 587.0, + 364.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 603.0, + 1403.0, + 603.0, + 1403.0, + 637.0, + 310.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 633.0, + 970.0, + 633.0, + 970.0, + 666.0, + 362.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 683.0, + 1410.0, + 683.0, + 1410.0, + 722.0, + 305.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 712.0, + 437.0, + 712.0, + 437.0, + 747.0, + 362.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 762.0, + 1408.0, + 762.0, + 1408.0, + 802.0, + 305.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 793.0, + 833.0, + 793.0, + 833.0, + 827.0, + 362.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 843.0, + 1408.0, + 843.0, + 1408.0, + 883.0, + 305.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 868.0, + 1408.0, + 868.0, + 1408.0, + 910.0, + 358.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 900.0, + 431.0, + 900.0, + 431.0, + 937.0, + 360.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 950.0, + 1410.0, + 950.0, + 1410.0, + 987.0, + 303.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 979.0, + 1408.0, + 979.0, + 1408.0, + 1017.0, + 360.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1003.0, + 432.0, + 1003.0, + 432.0, + 1047.0, + 362.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1058.0, + 1408.0, + 1058.0, + 1408.0, + 1096.0, + 303.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1090.0, + 1297.0, + 1090.0, + 1297.0, + 1123.0, + 362.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1140.0, + 1403.0, + 1140.0, + 1403.0, + 1173.0, + 310.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1171.0, + 1255.0, + 1171.0, + 1255.0, + 1205.0, + 362.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1219.0, + 1408.0, + 1219.0, + 1408.0, + 1259.0, + 305.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1250.0, + 1343.0, + 1250.0, + 1343.0, + 1284.0, + 362.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1301.0, + 1403.0, + 1301.0, + 1403.0, + 1334.0, + 307.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1326.0, + 1410.0, + 1326.0, + 1410.0, + 1369.0, + 360.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1357.0, + 431.0, + 1357.0, + 431.0, + 1392.0, + 360.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1407.0, + 1406.0, + 1407.0, + 1406.0, + 1447.0, + 305.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1436.0, + 911.0, + 1436.0, + 911.0, + 1474.0, + 360.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1486.0, + 1408.0, + 1486.0, + 1408.0, + 1528.0, + 303.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1520.0, + 1271.0, + 1520.0, + 1271.0, + 1553.0, + 362.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1572.0, + 1406.0, + 1572.0, + 1406.0, + 1605.0, + 310.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1595.0, + 1408.0, + 1595.0, + 1408.0, + 1637.0, + 360.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1624.0, + 846.0, + 1624.0, + 846.0, + 1662.0, + 358.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1678.0, + 1403.0, + 1678.0, + 1403.0, + 1712.0, + 307.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1708.0, + 1403.0, + 1708.0, + 1403.0, + 1741.0, + 364.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1756.0, + 1408.0, + 1756.0, + 1408.0, + 1793.0, + 303.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1789.0, + 1406.0, + 1789.0, + 1406.0, + 1822.0, + 362.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1816.0, + 544.0, + 1816.0, + 544.0, + 1850.0, + 362.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1864.0, + 1408.0, + 1864.0, + 1408.0, + 1904.0, + 305.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 1895.0, + 1309.0, + 1895.0, + 1309.0, + 1929.0, + 360.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1946.0, + 1408.0, + 1946.0, + 1408.0, + 1985.0, + 305.0, + 1985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1975.0, + 1062.0, + 1975.0, + 1062.0, + 2010.0, + 358.0, + 2010.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 306, + 200, + 1409, + 200, + 1409, + 2007, + 306, + 2007 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2087, + 836, + 2087 + ], + "score": 0.8 + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 831.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 200.0, + 1406.0, + 200.0, + 1406.0, + 237.0, + 305.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 225.0, + 1406.0, + 225.0, + 1406.0, + 269.0, + 361.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 261.0, + 598.0, + 261.0, + 598.0, + 289.0, + 363.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 309.0, + 1402.0, + 309.0, + 1402.0, + 341.0, + 307.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 335.0, + 1406.0, + 335.0, + 1406.0, + 374.0, + 363.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 365.0, + 598.0, + 365.0, + 598.0, + 402.0, + 361.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 414.0, + 1404.0, + 414.0, + 1404.0, + 452.0, + 305.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 446.0, + 994.0, + 446.0, + 994.0, + 478.0, + 363.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 492.0, + 1404.0, + 492.0, + 1404.0, + 529.0, + 305.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 519.0, + 1228.0, + 519.0, + 1228.0, + 559.0, + 359.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 573.0, + 1408.0, + 573.0, + 1408.0, + 611.0, + 305.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 603.0, + 960.0, + 603.0, + 960.0, + 635.0, + 361.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 655.0, + 1406.0, + 655.0, + 1406.0, + 686.0, + 309.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 682.0, + 945.0, + 682.0, + 945.0, + 714.0, + 363.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 732.0, + 1404.0, + 732.0, + 1404.0, + 770.0, + 305.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 754.0, + 1384.0, + 754.0, + 1384.0, + 800.0, + 359.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 810.0, + 1406.0, + 810.0, + 1406.0, + 845.0, + 305.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 835.0, + 1408.0, + 835.0, + 1408.0, + 875.0, + 359.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 865.0, + 699.0, + 865.0, + 699.0, + 903.0, + 361.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 917.0, + 1408.0, + 917.0, + 1408.0, + 955.0, + 307.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 945.0, + 1030.0, + 945.0, + 1030.0, + 982.0, + 361.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 994.0, + 1404.0, + 994.0, + 1404.0, + 1030.0, + 305.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1020.0, + 1408.0, + 1020.0, + 1408.0, + 1062.0, + 359.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1047.0, + 435.0, + 1047.0, + 435.0, + 1088.0, + 363.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1102.0, + 1406.0, + 1102.0, + 1406.0, + 1139.0, + 307.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1129.0, + 1194.0, + 1129.0, + 1194.0, + 1165.0, + 359.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1181.0, + 1408.0, + 1181.0, + 1408.0, + 1217.0, + 305.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1211.0, + 913.0, + 1211.0, + 913.0, + 1243.0, + 361.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1260.0, + 1406.0, + 1260.0, + 1406.0, + 1296.0, + 305.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1290.0, + 1390.0, + 1290.0, + 1390.0, + 1322.0, + 363.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1338.0, + 1408.0, + 1338.0, + 1408.0, + 1376.0, + 305.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1370.0, + 1406.0, + 1370.0, + 1406.0, + 1401.0, + 363.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1397.0, + 731.0, + 1397.0, + 731.0, + 1429.0, + 363.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1445.0, + 1408.0, + 1445.0, + 1408.0, + 1483.0, + 307.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1477.0, + 848.0, + 1477.0, + 848.0, + 1509.0, + 363.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1525.0, + 1408.0, + 1525.0, + 1408.0, + 1562.0, + 307.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1554.0, + 505.0, + 1554.0, + 505.0, + 1586.0, + 363.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1602.0, + 1406.0, + 1602.0, + 1406.0, + 1640.0, + 307.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1632.0, + 1315.0, + 1632.0, + 1315.0, + 1670.0, + 361.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1686.0, + 1406.0, + 1686.0, + 1406.0, + 1717.0, + 307.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1713.0, + 1406.0, + 1713.0, + 1406.0, + 1745.0, + 363.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1739.0, + 622.0, + 1739.0, + 622.0, + 1771.0, + 365.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1791.0, + 1406.0, + 1791.0, + 1406.0, + 1823.0, + 307.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1819.0, + 1214.0, + 1819.0, + 1214.0, + 1850.0, + 363.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 1870.0, + 1404.0, + 1870.0, + 1404.0, + 1902.0, + 309.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1898.0, + 1129.0, + 1898.0, + 1129.0, + 1930.0, + 363.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1946.0, + 1408.0, + 1946.0, + 1408.0, + 1981.0, + 303.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1972.0, + 436.0, + 1972.0, + 436.0, + 2011.0, + 359.0, + 2011.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 304, + 192, + 1410, + 192, + 1410, + 2020, + 304, + 2020 + ], + "score": 0.879 + }, + { + "category_id": 2, + "poly": [ + 835, + 2060, + 865, + 2060, + 865, + 2086, + 835, + 2086 + ], + "score": 0.641 + }, + { + "category_id": 2, + "poly": [ + 835, + 2060, + 865, + 2060, + 865, + 2086, + 835, + 2086 + ], + "score": 0.459 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 200.0, + 1404.0, + 200.0, + 1404.0, + 238.0, + 308.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 226.0, + 1406.0, + 226.0, + 1406.0, + 269.0, + 357.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 259.0, + 599.0, + 259.0, + 599.0, + 291.0, + 361.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 311.0, + 1404.0, + 311.0, + 1404.0, + 343.0, + 308.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 335.0, + 1404.0, + 335.0, + 1404.0, + 375.0, + 359.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 367.0, + 599.0, + 367.0, + 599.0, + 399.0, + 363.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 417.0, + 1404.0, + 417.0, + 1404.0, + 449.0, + 310.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 443.0, + 1301.0, + 443.0, + 1301.0, + 481.0, + 361.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 493.0, + 1408.0, + 493.0, + 1408.0, + 532.0, + 306.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 522.0, + 799.0, + 522.0, + 799.0, + 560.0, + 361.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 570.0, + 1408.0, + 570.0, + 1408.0, + 612.0, + 302.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 600.0, + 986.0, + 600.0, + 986.0, + 636.0, + 359.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 650.0, + 1406.0, + 650.0, + 1406.0, + 690.0, + 304.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 682.0, + 938.0, + 682.0, + 938.0, + 714.0, + 361.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 730.0, + 1408.0, + 730.0, + 1408.0, + 767.0, + 304.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 757.0, + 1408.0, + 757.0, + 1408.0, + 797.0, + 357.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 787.0, + 431.0, + 787.0, + 431.0, + 821.0, + 361.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 835.0, + 1408.0, + 835.0, + 1408.0, + 875.0, + 304.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 867.0, + 585.0, + 867.0, + 585.0, + 899.0, + 363.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 911.0, + 1410.0, + 911.0, + 1410.0, + 957.0, + 302.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 945.0, + 619.0, + 945.0, + 619.0, + 981.0, + 359.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 998.0, + 1404.0, + 998.0, + 1404.0, + 1030.0, + 308.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1026.0, + 1075.0, + 1026.0, + 1075.0, + 1058.0, + 363.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1076.0, + 1404.0, + 1076.0, + 1404.0, + 1108.0, + 308.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1104.0, + 1404.0, + 1104.0, + 1404.0, + 1136.0, + 363.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1130.0, + 732.0, + 1130.0, + 732.0, + 1166.0, + 357.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1180.0, + 1406.0, + 1180.0, + 1406.0, + 1218.0, + 306.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1206.0, + 671.0, + 1206.0, + 671.0, + 1247.0, + 359.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1263.0, + 1406.0, + 1263.0, + 1406.0, + 1295.0, + 308.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1284.0, + 433.0, + 1284.0, + 433.0, + 1326.0, + 359.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1335.0, + 1408.0, + 1335.0, + 1408.0, + 1377.0, + 304.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1367.0, + 431.0, + 1367.0, + 431.0, + 1401.0, + 361.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1417.0, + 1406.0, + 1417.0, + 1406.0, + 1455.0, + 306.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1443.0, + 1406.0, + 1443.0, + 1406.0, + 1484.0, + 359.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1474.0, + 823.0, + 1474.0, + 823.0, + 1512.0, + 361.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1524.0, + 1408.0, + 1524.0, + 1408.0, + 1562.0, + 306.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1552.0, + 1154.0, + 1552.0, + 1154.0, + 1588.0, + 359.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1602.0, + 1408.0, + 1602.0, + 1408.0, + 1638.0, + 304.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1628.0, + 1405.0, + 1628.0, + 1405.0, + 1670.0, + 361.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1660.0, + 601.0, + 1660.0, + 601.0, + 1692.0, + 363.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1708.0, + 1406.0, + 1708.0, + 1406.0, + 1747.0, + 308.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1735.0, + 1276.0, + 1735.0, + 1276.0, + 1775.0, + 359.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1791.0, + 1402.0, + 1791.0, + 1402.0, + 1823.0, + 308.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1819.0, + 1004.0, + 1819.0, + 1004.0, + 1851.0, + 363.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1869.0, + 1406.0, + 1869.0, + 1406.0, + 1901.0, + 310.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1897.0, + 817.0, + 1897.0, + 817.0, + 1929.0, + 363.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1945.0, + 1406.0, + 1945.0, + 1406.0, + 1984.0, + 306.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1974.0, + 621.0, + 1974.0, + 621.0, + 2010.0, + 359.0, + 2010.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 147, + 1409, + 147, + 1409, + 2021, + 295, + 2021 + ], + "score": 0.822 + }, + { + "category_id": 2, + "poly": [ + 835, + 2061, + 865, + 2061, + 865, + 2086, + 835, + 2086 + ], + "score": 0.811 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 869.0, + 2058.0, + 869.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 204.0, + 1404.0, + 204.0, + 1404.0, + 237.0, + 308.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 231.0, + 624.0, + 231.0, + 624.0, + 264.0, + 361.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 276.0, + 1406.0, + 276.0, + 1406.0, + 313.0, + 304.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 307.0, + 1406.0, + 307.0, + 1406.0, + 340.0, + 363.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 333.0, + 430.0, + 333.0, + 430.0, + 368.0, + 361.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 379.0, + 1406.0, + 379.0, + 1406.0, + 418.0, + 306.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 412.0, + 1406.0, + 412.0, + 1406.0, + 445.0, + 361.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 436.0, + 433.0, + 436.0, + 433.0, + 471.0, + 361.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 486.0, + 1404.0, + 486.0, + 1404.0, + 519.0, + 308.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 510.0, + 1408.0, + 510.0, + 1408.0, + 551.0, + 357.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 541.0, + 700.0, + 541.0, + 700.0, + 576.0, + 357.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 586.0, + 1408.0, + 586.0, + 1408.0, + 625.0, + 306.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 617.0, + 955.0, + 617.0, + 955.0, + 650.0, + 359.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 662.0, + 1406.0, + 662.0, + 1406.0, + 702.0, + 306.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 693.0, + 1071.0, + 693.0, + 1071.0, + 726.0, + 361.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 743.0, + 1404.0, + 743.0, + 1404.0, + 776.0, + 310.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 769.0, + 624.0, + 769.0, + 624.0, + 802.0, + 361.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 817.0, + 1404.0, + 817.0, + 1404.0, + 850.0, + 308.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 845.0, + 1339.0, + 845.0, + 1339.0, + 878.0, + 361.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 891.0, + 1406.0, + 891.0, + 1406.0, + 930.0, + 306.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 922.0, + 892.0, + 922.0, + 892.0, + 954.0, + 365.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 969.0, + 1404.0, + 969.0, + 1404.0, + 1002.0, + 308.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 998.0, + 711.0, + 998.0, + 711.0, + 1028.0, + 359.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1045.0, + 1404.0, + 1045.0, + 1404.0, + 1078.0, + 308.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1072.0, + 1134.0, + 1072.0, + 1134.0, + 1105.0, + 361.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1121.0, + 1404.0, + 1121.0, + 1404.0, + 1154.0, + 308.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1148.0, + 546.0, + 1148.0, + 546.0, + 1181.0, + 361.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1193.0, + 1410.0, + 1193.0, + 1410.0, + 1230.0, + 304.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1222.0, + 1408.0, + 1222.0, + 1408.0, + 1261.0, + 361.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1248.0, + 713.0, + 1248.0, + 713.0, + 1288.0, + 359.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1300.0, + 1404.0, + 1300.0, + 1404.0, + 1333.0, + 308.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1322.0, + 1406.0, + 1322.0, + 1406.0, + 1366.0, + 359.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1355.0, + 584.0, + 1355.0, + 584.0, + 1388.0, + 363.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1401.0, + 1402.0, + 1401.0, + 1402.0, + 1434.0, + 293.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1429.0, + 1406.0, + 1429.0, + 1406.0, + 1469.0, + 361.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1460.0, + 749.0, + 1460.0, + 749.0, + 1493.0, + 361.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1503.0, + 1406.0, + 1503.0, + 1406.0, + 1543.0, + 291.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1534.0, + 1191.0, + 1534.0, + 1191.0, + 1567.0, + 363.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1577.0, + 1404.0, + 1577.0, + 1404.0, + 1617.0, + 291.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1610.0, + 1406.0, + 1610.0, + 1406.0, + 1643.0, + 365.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1639.0, + 1406.0, + 1639.0, + 1406.0, + 1672.0, + 363.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1666.0, + 1122.0, + 1666.0, + 1122.0, + 1699.0, + 363.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1713.0, + 1406.0, + 1713.0, + 1406.0, + 1746.0, + 293.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1742.0, + 1406.0, + 1742.0, + 1406.0, + 1775.0, + 365.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1767.0, + 700.0, + 1767.0, + 700.0, + 1804.0, + 357.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1814.0, + 1406.0, + 1814.0, + 1406.0, + 1853.0, + 291.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1845.0, + 1404.0, + 1845.0, + 1404.0, + 1878.0, + 363.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1872.0, + 873.0, + 1872.0, + 873.0, + 1911.0, + 361.0, + 1911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1921.0, + 1406.0, + 1921.0, + 1406.0, + 1954.0, + 296.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1946.0, + 1406.0, + 1946.0, + 1406.0, + 1985.0, + 359.0, + 1985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1976.0, + 1324.0, + 1976.0, + 1324.0, + 2009.0, + 365.0, + 2009.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 292, + 201, + 1409, + 201, + 1409, + 870, + 292, + 870 + ], + "score": 0.807 + }, + { + "category_id": 2, + "poly": [ + 836, + 2062, + 865, + 2062, + 865, + 2085, + 836, + 2085 + ], + "score": 0.754 + }, + { + "category_id": 2, + "poly": [ + 836, + 2062, + 865, + 2062, + 865, + 2086, + 836, + 2086 + ], + "score": 0.193 + }, + { + "category_id": 13, + "poly": [ + 1243, + 236, + 1265, + 236, + 1265, + 258, + 1243, + 258 + ], + "score": 0.66, + "latex": "^ +" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 871.0, + 2058.0, + 871.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2058.0, + 871.0, + 2058.0, + 871.0, + 2095.0, + 831.0, + 2095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 203.0, + 1407.0, + 203.0, + 1407.0, + 238.0, + 294.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 227.0, + 1242.0, + 227.0, + 1242.0, + 267.0, + 360.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 227.0, + 1408.0, + 227.0, + 1408.0, + 267.0, + 1266.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 255.0, + 640.0, + 255.0, + 640.0, + 293.0, + 361.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 308.0, + 1406.0, + 308.0, + 1406.0, + 343.0, + 294.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 336.0, + 1407.0, + 336.0, + 1407.0, + 371.0, + 361.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 362.0, + 431.0, + 362.0, + 431.0, + 398.0, + 360.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 409.0, + 1408.0, + 409.0, + 1408.0, + 451.0, + 290.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 442.0, + 556.0, + 442.0, + 556.0, + 475.0, + 361.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 490.0, + 1407.0, + 490.0, + 1407.0, + 525.0, + 294.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 519.0, + 1407.0, + 519.0, + 1407.0, + 554.0, + 362.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 544.0, + 430.0, + 544.0, + 430.0, + 580.0, + 360.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 596.0, + 1407.0, + 596.0, + 1407.0, + 632.0, + 293.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 623.0, + 848.0, + 623.0, + 848.0, + 658.0, + 362.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 670.0, + 1408.0, + 670.0, + 1408.0, + 713.0, + 291.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 701.0, + 1406.0, + 701.0, + 1406.0, + 735.0, + 361.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 728.0, + 430.0, + 728.0, + 430.0, + 762.0, + 361.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 778.0, + 1406.0, + 778.0, + 1406.0, + 814.0, + 293.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 805.0, + 1408.0, + 805.0, + 1408.0, + 843.0, + 359.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 833.0, + 468.0, + 833.0, + 468.0, + 869.0, + 360.0, + 869.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 406, + 298, + 1405, + 298, + 1405, + 497, + 406, + 497 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 407, + 725, + 1405, + 725, + 1405, + 1103, + 407, + 1103 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 407, + 1801, + 1405, + 1801, + 1405, + 1996, + 407, + 1996 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 407, + 546, + 1404, + 546, + 1404, + 676, + 407, + 676 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 363, + 685, + 671, + 685, + 671, + 715, + 363, + 715 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 365, + 506, + 840, + 506, + 840, + 538, + 365, + 538 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 364, + 256, + 580, + 256, + 580, + 288, + 364, + 288 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 298, + 199, + 435, + 199, + 435, + 235, + 298, + 235 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 864, + 2061, + 864, + 2085, + 836, + 2085 + ], + "score": 0.849 + }, + { + "category_id": 1, + "poly": [ + 365, + 1760, + 1200, + 1760, + 1200, + 1793, + 365, + 1793 + ], + "score": 0.837 + }, + { + "category_id": 1, + "poly": [ + 368, + 1113, + 1400, + 1113, + 1400, + 1146, + 368, + 1146 + ], + "score": 0.828 + }, + { + "category_id": 1, + "poly": [ + 404, + 1154, + 1407, + 1154, + 1407, + 1750, + 404, + 1750 + ], + "score": 0.811 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 197.0, + 440.0, + 197.0, + 440.0, + 238.0, + 296.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 869.0, + 2058.0, + 869.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 294.0, + 1405.0, + 294.0, + 1405.0, + 333.0, + 405.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 327.0, + 801.0, + 327.0, + 801.0, + 364.0, + 449.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 362.0, + 1172.0, + 362.0, + 1172.0, + 398.0, + 404.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 399.0, + 1408.0, + 399.0, + 1408.0, + 434.0, + 405.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 431.0, + 1405.0, + 431.0, + 1405.0, + 470.0, + 404.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 463.0, + 596.0, + 463.0, + 596.0, + 500.0, + 449.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 723.0, + 1408.0, + 723.0, + 1408.0, + 758.0, + 405.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 755.0, + 1404.0, + 755.0, + 1404.0, + 788.0, + 450.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 786.0, + 1221.0, + 786.0, + 1221.0, + 819.0, + 451.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 822.0, + 446.0, + 822.0, + 446.0, + 852.0, + 406.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 820.0, + 1405.0, + 820.0, + 1405.0, + 855.0, + 449.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 848.0, + 1408.0, + 848.0, + 1408.0, + 886.0, + 450.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 879.0, + 796.0, + 879.0, + 796.0, + 918.0, + 450.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 919.0, + 446.0, + 919.0, + 446.0, + 948.0, + 407.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 916.0, + 1408.0, + 916.0, + 1408.0, + 952.0, + 449.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 946.0, + 1406.0, + 946.0, + 1406.0, + 981.0, + 447.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 978.0, + 1406.0, + 978.0, + 1406.0, + 1011.0, + 451.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1009.0, + 1017.0, + 1009.0, + 1017.0, + 1042.0, + 450.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1040.0, + 1405.0, + 1040.0, + 1405.0, + 1078.0, + 407.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1074.0, + 1256.0, + 1074.0, + 1256.0, + 1106.0, + 452.0, + 1106.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1799.0, + 1407.0, + 1799.0, + 1407.0, + 1836.0, + 404.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1829.0, + 660.0, + 1829.0, + 660.0, + 1866.0, + 449.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1866.0, + 1405.0, + 1866.0, + 1405.0, + 1902.0, + 405.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1897.0, + 951.0, + 1897.0, + 951.0, + 1933.0, + 449.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1932.0, + 1405.0, + 1932.0, + 1405.0, + 1968.0, + 408.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1964.0, + 926.0, + 1964.0, + 926.0, + 1999.0, + 448.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 544.0, + 1407.0, + 544.0, + 1407.0, + 583.0, + 403.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 576.0, + 1406.0, + 576.0, + 1406.0, + 613.0, + 450.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 607.0, + 1056.0, + 607.0, + 1056.0, + 640.0, + 450.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 642.0, + 1180.0, + 642.0, + 1180.0, + 679.0, + 404.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 679.0, + 675.0, + 679.0, + 675.0, + 722.0, + 357.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 503.0, + 846.0, + 503.0, + 846.0, + 543.0, + 359.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 252.0, + 586.0, + 252.0, + 586.0, + 291.0, + 361.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1756.0, + 1205.0, + 1756.0, + 1205.0, + 1799.0, + 357.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1111.0, + 1404.0, + 1111.0, + 1404.0, + 1151.0, + 358.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1154.0, + 1410.0, + 1154.0, + 1410.0, + 1188.0, + 405.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 1182.0, + 1409.0, + 1182.0, + 1409.0, + 1220.0, + 447.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1216.0, + 1405.0, + 1216.0, + 1405.0, + 1246.0, + 448.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1244.0, + 1408.0, + 1244.0, + 1408.0, + 1278.0, + 449.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1275.0, + 1406.0, + 1275.0, + 1406.0, + 1310.0, + 451.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1307.0, + 1408.0, + 1307.0, + 1408.0, + 1339.0, + 452.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1332.0, + 1409.0, + 1332.0, + 1409.0, + 1371.0, + 449.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1366.0, + 681.0, + 1366.0, + 681.0, + 1399.0, + 451.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1400.0, + 1405.0, + 1400.0, + 1405.0, + 1436.0, + 406.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1429.0, + 1409.0, + 1429.0, + 1409.0, + 1466.0, + 448.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1459.0, + 609.0, + 1459.0, + 609.0, + 1498.0, + 449.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1497.0, + 1398.0, + 1497.0, + 1398.0, + 1532.0, + 405.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1531.0, + 1404.0, + 1531.0, + 1404.0, + 1566.0, + 403.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1564.0, + 704.0, + 1564.0, + 704.0, + 1597.0, + 451.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1597.0, + 1405.0, + 1597.0, + 1405.0, + 1633.0, + 408.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1626.0, + 1409.0, + 1626.0, + 1409.0, + 1664.0, + 449.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1659.0, + 1409.0, + 1659.0, + 1409.0, + 1695.0, + 449.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1689.0, + 1406.0, + 1689.0, + 1406.0, + 1723.0, + 449.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1721.0, + 861.0, + 1721.0, + 861.0, + 1753.0, + 452.0, + 1753.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/dev/Loek7hfb46P/Loek7hfb46P_origin.pdf b/parse/dev/Loek7hfb46P/Loek7hfb46P_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..dd88bf57097b2368175632669bc2f460fa0499ea --- /dev/null +++ b/parse/dev/Loek7hfb46P/Loek7hfb46P_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa12013bd634cb0e39cff0beb4fce314d519bfbc6a2865c1ba2dd20c538abaf1 +size 35283899 diff --git a/parse/dev/MSBDFwGYwwt/images/0580b61767cedf17a0057ed6c2e16df757d7ae1e4f2aac057ec346f1a30d91c0.jpg b/parse/dev/MSBDFwGYwwt/images/0580b61767cedf17a0057ed6c2e16df757d7ae1e4f2aac057ec346f1a30d91c0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fca708cce9ef4d3441d9c98c95db611c84ec0442 --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/0580b61767cedf17a0057ed6c2e16df757d7ae1e4f2aac057ec346f1a30d91c0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48c5d123ddafd656ed164d7c6bbeda9a23b49faa717c7f4ca626b2e18afdf90a +size 7637 diff --git a/parse/dev/MSBDFwGYwwt/images/1e5c49815114f48a1860fda80030b01e5e0711a29a132c0141bf683ed5540bbd.jpg b/parse/dev/MSBDFwGYwwt/images/1e5c49815114f48a1860fda80030b01e5e0711a29a132c0141bf683ed5540bbd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..228aff8936fe78e4562298e89d194eba1dd3a656 --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/1e5c49815114f48a1860fda80030b01e5e0711a29a132c0141bf683ed5540bbd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fbced6cffc02b23ad06417bf1988f545ff8e1c37ee86550977bbfe8f11868684 +size 19215 diff --git a/parse/dev/MSBDFwGYwwt/images/21446138e7b58ef7fa06708aaa42a3e254895c4f524c58878edc2c4efac74221.jpg b/parse/dev/MSBDFwGYwwt/images/21446138e7b58ef7fa06708aaa42a3e254895c4f524c58878edc2c4efac74221.jpg new file mode 100644 index 0000000000000000000000000000000000000000..14524748de0714fcbac3a1f6a44447574042a95a --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/21446138e7b58ef7fa06708aaa42a3e254895c4f524c58878edc2c4efac74221.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8686d45281a3e7c040c4102f3bd1a357d1f7bed4d3026150471753c40b717e7f +size 12278 diff --git a/parse/dev/MSBDFwGYwwt/images/28a7f8bb724b01d8b8249243e97576a3f357fe5f8f1c8233e2bc6ae24fd887a0.jpg b/parse/dev/MSBDFwGYwwt/images/28a7f8bb724b01d8b8249243e97576a3f357fe5f8f1c8233e2bc6ae24fd887a0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..772bf948f5c8c0d29a902fe601b7c015bbb3bd11 --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/28a7f8bb724b01d8b8249243e97576a3f357fe5f8f1c8233e2bc6ae24fd887a0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ff285f986352bfa4c052db71db431e83993afd76d48eb85aa422ff781bdec79 +size 37092 diff --git a/parse/dev/MSBDFwGYwwt/images/28bb2249b676fce13ceb44cb9f5d7b9a687a29bec215dafdfc3182443ba799ed.jpg b/parse/dev/MSBDFwGYwwt/images/28bb2249b676fce13ceb44cb9f5d7b9a687a29bec215dafdfc3182443ba799ed.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cebdbd65a9da20e8d6d52ed8749d5155902ce10b --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/28bb2249b676fce13ceb44cb9f5d7b9a687a29bec215dafdfc3182443ba799ed.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f69ed5df239d8fe90cb3ff7491f2f193a1eb687bc15c2db3b935e786415fce3b +size 13915 diff --git a/parse/dev/MSBDFwGYwwt/images/3fe70a5fa5e568ea300a36e7abe2116f37ac51b7e74db9b44dd4570a6c06d456.jpg b/parse/dev/MSBDFwGYwwt/images/3fe70a5fa5e568ea300a36e7abe2116f37ac51b7e74db9b44dd4570a6c06d456.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e15bd19845eb470a6929580c26859a93224181dd --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/3fe70a5fa5e568ea300a36e7abe2116f37ac51b7e74db9b44dd4570a6c06d456.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b8b165b0af115696a9b08dff523196bb7d5db43e94c89b8199ae067670d0e1a +size 100882 diff --git a/parse/dev/MSBDFwGYwwt/images/433506758c93192d844b3ef47d777cc94fd12a2a093fa8cc235eab2d96c11c16.jpg b/parse/dev/MSBDFwGYwwt/images/433506758c93192d844b3ef47d777cc94fd12a2a093fa8cc235eab2d96c11c16.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ed4e157b25eb71554dbc2f9032685ab0f29e1695 --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/433506758c93192d844b3ef47d777cc94fd12a2a093fa8cc235eab2d96c11c16.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02adcb3233ea178698134375d6abe90dd04f9167f28609d9e7ded75c1411edb7 +size 54733 diff --git a/parse/dev/MSBDFwGYwwt/images/4e4ea3792074d9baa9783628607af7660a5456f6dba1dc48a72b2e24a80bc68a.jpg b/parse/dev/MSBDFwGYwwt/images/4e4ea3792074d9baa9783628607af7660a5456f6dba1dc48a72b2e24a80bc68a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f1e5fb658917eb63e691e574c4f99d9ba05e0b56 --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/4e4ea3792074d9baa9783628607af7660a5456f6dba1dc48a72b2e24a80bc68a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:207cd5a293eb32b5c268ddf93d0e14cd7af91b58b52894ff4e95be9ecba40ee6 +size 4725 diff --git a/parse/dev/MSBDFwGYwwt/images/6336bb11015c76070699a93df3fbf4911d80201fb970f31640e7199218d771f7.jpg b/parse/dev/MSBDFwGYwwt/images/6336bb11015c76070699a93df3fbf4911d80201fb970f31640e7199218d771f7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f7c62306e8e31d7cbbcf5611f0391377c37dc4a7 --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/6336bb11015c76070699a93df3fbf4911d80201fb970f31640e7199218d771f7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1cb9a0560a021e3c32bbe720611fa98ff53231de71fc2e558e2b1fd00bd9b2a +size 36151 diff --git a/parse/dev/MSBDFwGYwwt/images/6a2bf7b1897ac7721d08e30d6bdd23cf4b46304eaec2d710da3164f34c80d63f.jpg b/parse/dev/MSBDFwGYwwt/images/6a2bf7b1897ac7721d08e30d6bdd23cf4b46304eaec2d710da3164f34c80d63f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d9aa3414b1c5f7585ff441b6fa5711386ae04e92 --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/6a2bf7b1897ac7721d08e30d6bdd23cf4b46304eaec2d710da3164f34c80d63f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02e8a521166c3ed99c614dbcbb3b57705aeafabc36014c692d18b71839a4aed2 +size 104078 diff --git a/parse/dev/MSBDFwGYwwt/images/ad405327cc92ad77506750401708e4f007f58e16f633a762200fcbf18fde2c5d.jpg b/parse/dev/MSBDFwGYwwt/images/ad405327cc92ad77506750401708e4f007f58e16f633a762200fcbf18fde2c5d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8d8db7df182089bad231cc4002aa5086a560e1ec --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/ad405327cc92ad77506750401708e4f007f58e16f633a762200fcbf18fde2c5d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:840286f3f8baf4b1ff5a342067553988946502b23f8ae53b50aaf46ba117a4aa +size 9106 diff --git a/parse/dev/MSBDFwGYwwt/images/eff82a8928b58a7d62890bdc17d1250163e22dccc27770dfac1bde99f15855f3.jpg b/parse/dev/MSBDFwGYwwt/images/eff82a8928b58a7d62890bdc17d1250163e22dccc27770dfac1bde99f15855f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b5b83de47f32d15f5c7162b5c76e9911065829aa --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/eff82a8928b58a7d62890bdc17d1250163e22dccc27770dfac1bde99f15855f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:592b649b0727a4a2d83ecc71f2dccd504ec743930cb9d6e08754e904466c08de +size 73127 diff --git a/parse/dev/MSBDFwGYwwt/images/f16b1a41aa648df27ad37860bf43f1e981fa6c499b9aab1a8677e5a27f44dd2b.jpg b/parse/dev/MSBDFwGYwwt/images/f16b1a41aa648df27ad37860bf43f1e981fa6c499b9aab1a8677e5a27f44dd2b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d669b77d4b382f257d8906fac8a66d57640fd4c8 --- /dev/null +++ b/parse/dev/MSBDFwGYwwt/images/f16b1a41aa648df27ad37860bf43f1e981fa6c499b9aab1a8677e5a27f44dd2b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8380d8d616e91b13b6616e84811b2d83f2f76259322a898bed89ca2d399d5c7 +size 57390 diff --git a/parse/dev/NHeAUKlTO8/NHeAUKlTO8_layout.pdf b/parse/dev/NHeAUKlTO8/NHeAUKlTO8_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6996499bcd6ed4ada0c496c31986bffcaf3f90ad --- /dev/null +++ b/parse/dev/NHeAUKlTO8/NHeAUKlTO8_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0595efa85a5c318f131f3a34eb7352ece45ff1ca4b0750003a3fe2a25403fd2c +size 897189 diff --git a/parse/dev/NHeAUKlTO8/NHeAUKlTO8_span.pdf b/parse/dev/NHeAUKlTO8/NHeAUKlTO8_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..54ee84c253a9bb730a506b654fb5b439ea3d7a25 --- /dev/null +++ b/parse/dev/NHeAUKlTO8/NHeAUKlTO8_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a52d543098cd4bc72a76485b16ea7ed721486d7fd57af2e234f6a1811f8411ac +size 946681 diff --git a/parse/dev/OboQ71j1Bn/OboQ71j1Bn_middle.json b/parse/dev/OboQ71j1Bn/OboQ71j1Bn_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..96883e091eb3606ad613753585e6dd066196b4de --- /dev/null +++ b/parse/dev/OboQ71j1Bn/OboQ71j1Bn_middle.json @@ -0,0 +1,56275 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 78, + 506, + 136 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 509, + 99 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 509, + 99 + ], + "score": 1.0, + "content": "REDUCE, REUSE, RECYCLE: COMPOSITIONAL GENER-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 100, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 505, + 118 + ], + "score": 1.0, + "content": "ATION WITH ENERGY-BASED DIFFUSION MODELS AND", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 119, + 167, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 119, + 167, + 138 + ], + "score": 1.0, + "content": "MCMC", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 112, + 155, + 244, + 177 + ], + "lines": [ + { + "bbox": [ + 113, + 156, + 201, + 167 + ], + "spans": [ + { + "bbox": [ + 113, + 156, + 201, + 167 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 112, + 166, + 245, + 178 + ], + "spans": [ + { + "bbox": [ + 112, + 166, + 245, + 178 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 278, + 206, + 333, + 218 + ], + "lines": [ + { + "bbox": [ + 276, + 206, + 335, + 219 + ], + "spans": [ + { + "bbox": [ + 276, + 206, + 335, + 219 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 143, + 228, + 468, + 416 + ], + "lines": [ + { + "bbox": [ + 141, + 229, + 469, + 243 + ], + "spans": [ + { + "bbox": [ + 141, + 229, + 469, + 243 + ], + "score": 1.0, + "content": "Since their introduction, diffusion models have quickly become the prevailing", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 241, + 470, + 253 + ], + "spans": [ + { + "bbox": [ + 141, + 241, + 470, + 253 + ], + "score": 1.0, + "content": "approach to generative modeling in many domains. They can be interpreted", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 252, + 469, + 264 + ], + "spans": [ + { + "bbox": [ + 141, + 252, + 469, + 264 + ], + "score": 1.0, + "content": "as learning the gradients of a time-varying sequence of log-probability density", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 262, + 470, + 275 + ], + "spans": [ + { + "bbox": [ + 141, + 262, + 470, + 275 + ], + "score": 1.0, + "content": "functions. This interpretation has motivated classifier-based and classifier-free", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 274, + 470, + 285 + ], + "spans": [ + { + "bbox": [ + 141, + 274, + 470, + 285 + ], + "score": 1.0, + "content": "guidance as methods for post-hoc control of diffusion models. In this work, we", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 284, + 470, + 297 + ], + "spans": [ + { + "bbox": [ + 141, + 284, + 470, + 297 + ], + "score": 1.0, + "content": "build upon these ideas using the score-based interpretation of diffusion models, and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 296, + 470, + 307 + ], + "spans": [ + { + "bbox": [ + 141, + 296, + 470, + 307 + ], + "score": 1.0, + "content": "explore alternative ways to condition, modify, and reuse diffusion models for tasks", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 306, + 469, + 319 + ], + "spans": [ + { + "bbox": [ + 141, + 306, + 469, + 319 + ], + "score": 1.0, + "content": "involving compositional generation and guidance. In particular, we investigate why", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 318, + 470, + 329 + ], + "spans": [ + { + "bbox": [ + 142, + 318, + 470, + 329 + ], + "score": 1.0, + "content": "certain types of composition fail using current techniques and present a number", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 328, + 470, + 341 + ], + "spans": [ + { + "bbox": [ + 141, + 328, + 470, + 341 + ], + "score": 1.0, + "content": "of solutions. We conclude that the sampler (not the model) is responsible for this", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 339, + 470, + 352 + ], + "spans": [ + { + "bbox": [ + 141, + 339, + 470, + 352 + ], + "score": 1.0, + "content": "failure and propose new samplers, inspired by MCMC, which enable successful", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 351, + 470, + 363 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 470, + 363 + ], + "score": 1.0, + "content": "compositional generation. Further, we propose an energy-based parameterization", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 361, + 469, + 373 + ], + "spans": [ + { + "bbox": [ + 141, + 361, + 469, + 373 + ], + "score": 1.0, + "content": "of diffusion models which enables the use of new compositional operators and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 372, + 470, + 385 + ], + "spans": [ + { + "bbox": [ + 141, + 372, + 470, + 385 + ], + "score": 1.0, + "content": "more sophisticated, Metropolis-corrected samplers. Intriguingly we find these", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 384, + 469, + 396 + ], + "spans": [ + { + "bbox": [ + 142, + 384, + 469, + 396 + ], + "score": 1.0, + "content": "samplers lead to notable improvements in compositional generation across a wide", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 394, + 470, + 407 + ], + "spans": [ + { + "bbox": [ + 142, + 394, + 470, + 407 + ], + "score": 1.0, + "content": "variety of problems such as classifier-guided ImageNet modeling and compositional", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 406, + 245, + 417 + ], + "spans": [ + { + "bbox": [ + 142, + 406, + 245, + 417 + ], + "score": 1.0, + "content": "text-to-image generation.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 434, + 206, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 433, + 208, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 208, + 450 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 456, + 505, + 577 + ], + "lines": [ + { + "bbox": [ + 106, + 457, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 506, + 469 + ], + "score": 1.0, + "content": "In recent years, tremendous progress has been made in generative modeling across a variety of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "domains (Brown et al., 2020; Brock et al., 2018; Ho et al., 2020). These models now serve as", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 478, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 493 + ], + "score": 1.0, + "content": "powerful priors for downstream applications such as code generation (Li et al., 2022), text-to-image", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 507, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 507, + 502 + ], + "score": 1.0, + "content": "generation (Saharia et al., 2022), question-answering (Brown et al., 2020) and many more. However,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "to fit this complex data, generative models have grown inexorably larger (requiring 10’s or even 100’s", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 511, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 506, + 525 + ], + "score": 1.0, + "content": "of billions of parameters) (Kaplan et al., 2020) and require datasets containing non-negligible fractions", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 523, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 506, + 535 + ], + "score": 1.0, + "content": "of the entire internet, making it costly and difficult to train and or finetune such models. Despite this,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "score": 1.0, + "content": "some of the most compelling applications of large generative models do not rely on finetuning. For", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 544, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 506, + 558 + ], + "score": 1.0, + "content": "example, prompting (Brown et al., 2020) has been a successful strategy to selectively extract insights", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 556, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 505, + 568 + ], + "score": 1.0, + "content": "from large models. In this paper, we explore an alternative to finetuning and prompting, through", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 566, + 500, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 500, + 578 + ], + "score": 1.0, + "content": "which we may repurpose the underlying prior learned by generative models for downstream tasks.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "Diffusion Models (Sohl-Dickstein et al., 2015; Song & Ermon, 2019; Ho et al., 2020) are a recently", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "popular approach to generative modeling which have demonstrated a favorable combination of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "score": 1.0, + "content": "scalability, sample quality, and log-likelihood. A key feature of diffusion models is the ability for", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 617, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 505, + 628 + ], + "score": 1.0, + "content": "their sampling to be “guided” after training. This involves combining the pre-trained Diffusion Model", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 626, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 131, + 639 + ], + "score": 0.91, + "content": "p _ { \\theta } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 626, + 227, + 640 + ], + "score": 1.0, + "content": "with a predictive model", + "type": "text" + }, + { + "bbox": [ + 228, + 627, + 260, + 639 + ], + "score": 0.94, + "content": "p _ { \\theta } ( y | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 626, + 363, + 640 + ], + "score": 1.0, + "content": "to generate samples from", + "type": "text" + }, + { + "bbox": [ + 363, + 627, + 396, + 639 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 626, + 506, + 640 + ], + "score": 1.0, + "content": ". This predictive model can", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 637, + 507, + 651 + ], + "spans": [ + { + "bbox": [ + 104, + 637, + 507, + 651 + ], + "score": 1.0, + "content": "be either explicitly defined (such as a pre-trained classifier) (Sohl-Dickstein et al., 2015; Dhariwal &", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "Nichol, 2021) or an implicit predictive model defined through the combination of a conditional and", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "unconditional generative model (Ho & Salimans, 2022). These forms of conditioning are particularly", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 670, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 505, + 684 + ], + "score": 1.0, + "content": "appealing (especially the former) as they allow us to reuse pre-trained generative models for many", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 682, + 377, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 377, + 694 + ], + "score": 1.0, + "content": "downstream applications, beyond those considered at training time.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "These conditioning methods are a form of model composition, i.e. combining probabilistic models", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "together to create new models. Compositional models have a long history back to early work on", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 721, + 506, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 506, + 732 + ], + "score": 1.0, + "content": "Mixtures-Of-Experts (Jacobs et al., 1991) and Product-Of-Experts models (Hinton, 2002; Mayraz &", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 78, + 506, + 136 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 509, + 99 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 509, + 99 + ], + "score": 1.0, + "content": "REDUCE, REUSE, RECYCLE: COMPOSITIONAL GENER-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 100, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 505, + 118 + ], + "score": 1.0, + "content": "ATION WITH ENERGY-BASED DIFFUSION MODELS AND", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 119, + 167, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 119, + 167, + 138 + ], + "score": 1.0, + "content": "MCMC", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 112, + 155, + 244, + 177 + ], + "lines": [ + { + "bbox": [ + 113, + 156, + 201, + 167 + ], + "spans": [ + { + "bbox": [ + 113, + 156, + 201, + 167 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 112, + 166, + 245, + 178 + ], + "spans": [ + { + "bbox": [ + 112, + 166, + 245, + 178 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 112, + 156, + 245, + 178 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 206, + 333, + 218 + ], + "lines": [ + { + "bbox": [ + 276, + 206, + 335, + 219 + ], + "spans": [ + { + "bbox": [ + 276, + 206, + 335, + 219 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 143, + 228, + 468, + 416 + ], + "lines": [ + { + "bbox": [ + 141, + 229, + 469, + 243 + ], + "spans": [ + { + "bbox": [ + 141, + 229, + 469, + 243 + ], + "score": 1.0, + "content": "Since their introduction, diffusion models have quickly become the prevailing", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 241, + 470, + 253 + ], + "spans": [ + { + "bbox": [ + 141, + 241, + 470, + 253 + ], + "score": 1.0, + "content": "approach to generative modeling in many domains. They can be interpreted", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 252, + 469, + 264 + ], + "spans": [ + { + "bbox": [ + 141, + 252, + 469, + 264 + ], + "score": 1.0, + "content": "as learning the gradients of a time-varying sequence of log-probability density", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 262, + 470, + 275 + ], + "spans": [ + { + "bbox": [ + 141, + 262, + 470, + 275 + ], + "score": 1.0, + "content": "functions. This interpretation has motivated classifier-based and classifier-free", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 274, + 470, + 285 + ], + "spans": [ + { + "bbox": [ + 141, + 274, + 470, + 285 + ], + "score": 1.0, + "content": "guidance as methods for post-hoc control of diffusion models. In this work, we", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 284, + 470, + 297 + ], + "spans": [ + { + "bbox": [ + 141, + 284, + 470, + 297 + ], + "score": 1.0, + "content": "build upon these ideas using the score-based interpretation of diffusion models, and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 296, + 470, + 307 + ], + "spans": [ + { + "bbox": [ + 141, + 296, + 470, + 307 + ], + "score": 1.0, + "content": "explore alternative ways to condition, modify, and reuse diffusion models for tasks", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 306, + 469, + 319 + ], + "spans": [ + { + "bbox": [ + 141, + 306, + 469, + 319 + ], + "score": 1.0, + "content": "involving compositional generation and guidance. In particular, we investigate why", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 318, + 470, + 329 + ], + "spans": [ + { + "bbox": [ + 142, + 318, + 470, + 329 + ], + "score": 1.0, + "content": "certain types of composition fail using current techniques and present a number", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 328, + 470, + 341 + ], + "spans": [ + { + "bbox": [ + 141, + 328, + 470, + 341 + ], + "score": 1.0, + "content": "of solutions. We conclude that the sampler (not the model) is responsible for this", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 339, + 470, + 352 + ], + "spans": [ + { + "bbox": [ + 141, + 339, + 470, + 352 + ], + "score": 1.0, + "content": "failure and propose new samplers, inspired by MCMC, which enable successful", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 351, + 470, + 363 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 470, + 363 + ], + "score": 1.0, + "content": "compositional generation. Further, we propose an energy-based parameterization", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 361, + 469, + 373 + ], + "spans": [ + { + "bbox": [ + 141, + 361, + 469, + 373 + ], + "score": 1.0, + "content": "of diffusion models which enables the use of new compositional operators and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 372, + 470, + 385 + ], + "spans": [ + { + "bbox": [ + 141, + 372, + 470, + 385 + ], + "score": 1.0, + "content": "more sophisticated, Metropolis-corrected samplers. Intriguingly we find these", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 384, + 469, + 396 + ], + "spans": [ + { + "bbox": [ + 142, + 384, + 469, + 396 + ], + "score": 1.0, + "content": "samplers lead to notable improvements in compositional generation across a wide", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 394, + 470, + 407 + ], + "spans": [ + { + "bbox": [ + 142, + 394, + 470, + 407 + ], + "score": 1.0, + "content": "variety of problems such as classifier-guided ImageNet modeling and compositional", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 406, + 245, + 417 + ], + "spans": [ + { + "bbox": [ + 142, + 406, + 245, + 417 + ], + "score": 1.0, + "content": "text-to-image generation.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 14, + "bbox_fs": [ + 141, + 229, + 470, + 417 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 434, + 206, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 433, + 208, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 208, + 450 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 456, + 505, + 577 + ], + "lines": [ + { + "bbox": [ + 106, + 457, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 506, + 469 + ], + "score": 1.0, + "content": "In recent years, tremendous progress has been made in generative modeling across a variety of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "domains (Brown et al., 2020; Brock et al., 2018; Ho et al., 2020). These models now serve as", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 478, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 493 + ], + "score": 1.0, + "content": "powerful priors for downstream applications such as code generation (Li et al., 2022), text-to-image", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 507, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 507, + 502 + ], + "score": 1.0, + "content": "generation (Saharia et al., 2022), question-answering (Brown et al., 2020) and many more. However,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "to fit this complex data, generative models have grown inexorably larger (requiring 10’s or even 100’s", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 511, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 506, + 525 + ], + "score": 1.0, + "content": "of billions of parameters) (Kaplan et al., 2020) and require datasets containing non-negligible fractions", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 523, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 506, + 535 + ], + "score": 1.0, + "content": "of the entire internet, making it costly and difficult to train and or finetune such models. Despite this,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "score": 1.0, + "content": "some of the most compelling applications of large generative models do not rely on finetuning. For", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 544, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 506, + 558 + ], + "score": 1.0, + "content": "example, prompting (Brown et al., 2020) has been a successful strategy to selectively extract insights", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 556, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 505, + 568 + ], + "score": 1.0, + "content": "from large models. In this paper, we explore an alternative to finetuning and prompting, through", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 566, + 500, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 500, + 578 + ], + "score": 1.0, + "content": "which we may repurpose the underlying prior learned by generative models for downstream tasks.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 457, + 507, + 578 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "Diffusion Models (Sohl-Dickstein et al., 2015; Song & Ermon, 2019; Ho et al., 2020) are a recently", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "popular approach to generative modeling which have demonstrated a favorable combination of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 618 + ], + "score": 1.0, + "content": "scalability, sample quality, and log-likelihood. A key feature of diffusion models is the ability for", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 617, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 505, + 628 + ], + "score": 1.0, + "content": "their sampling to be “guided” after training. This involves combining the pre-trained Diffusion Model", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 626, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 131, + 639 + ], + "score": 0.91, + "content": "p _ { \\theta } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 626, + 227, + 640 + ], + "score": 1.0, + "content": "with a predictive model", + "type": "text" + }, + { + "bbox": [ + 228, + 627, + 260, + 639 + ], + "score": 0.94, + "content": "p _ { \\theta } ( y | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 626, + 363, + 640 + ], + "score": 1.0, + "content": "to generate samples from", + "type": "text" + }, + { + "bbox": [ + 363, + 627, + 396, + 639 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 626, + 506, + 640 + ], + "score": 1.0, + "content": ". This predictive model can", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 637, + 507, + 651 + ], + "spans": [ + { + "bbox": [ + 104, + 637, + 507, + 651 + ], + "score": 1.0, + "content": "be either explicitly defined (such as a pre-trained classifier) (Sohl-Dickstein et al., 2015; Dhariwal &", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "Nichol, 2021) or an implicit predictive model defined through the combination of a conditional and", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "unconditional generative model (Ho & Salimans, 2022). These forms of conditioning are particularly", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 670, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 505, + 684 + ], + "score": 1.0, + "content": "appealing (especially the former) as they allow us to reuse pre-trained generative models for many", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 682, + 377, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 377, + 694 + ], + "score": 1.0, + "content": "downstream applications, beyond those considered at training time.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 39.5, + "bbox_fs": [ + 104, + 583, + 507, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "These conditioning methods are a form of model composition, i.e. combining probabilistic models", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "together to create new models. Compositional models have a long history back to early work on", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 721, + 506, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 506, + 732 + ], + "score": 1.0, + "content": "Mixtures-Of-Experts (Jacobs et al., 1991) and Product-Of-Experts models (Hinton, 2002; Mayraz &", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46, + "bbox_fs": [ + 106, + 698, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 82, + 501, + 213 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 82, + 501, + 213 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 82, + 501, + 213 + ], + "spans": [ + { + "bbox": [ + 110, + 82, + 501, + 213 + ], + "score": 0.973, + "type": "image", + "image_path": "e319142d3909a1775051d2649b2726a3b96107e89f1adbd0fbb4ad3ca99ee902.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 82, + 501, + 125.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 125.66666666666666, + 501, + 169.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 169.33333333333331, + 501, + 212.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 222, + 504, + 253 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 222, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 222, + 505, + 234 + ], + "score": 1.0, + "content": "Figure 1: Creating new models through composition. Simple operators enable diffusion models to be", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 232, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 244 + ], + "score": 1.0, + "content": "composed without retraining in settings such (a) products, (b) classifier conditioning, (c) compositional text-to-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 243, + 475, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 475, + 254 + ], + "score": 1.0, + "content": "image generation with a product (left) and a mixture (right). All samples generated by trained models.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 258, + 506, + 302 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 506, + 271 + ], + "score": 1.0, + "content": "Hinton, 2000). Here, many simple models or predictors were combined to increase their capacity.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 269, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 269, + 506, + 281 + ], + "score": 1.0, + "content": "Much of this early work on model composition was done in the context of Energy-Based Mod-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "els (Hinton, 2002), an alternative class of generative model which bears many similarities to diffusion", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 291, + 141, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 141, + 302 + ], + "score": 1.0, + "content": "models.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 307, + 506, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 308, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 506, + 320 + ], + "score": 1.0, + "content": "In this work, we explore the ways that diffusion models can be reused and composed with one-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 319, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 506, + 331 + ], + "score": 1.0, + "content": "another. First, we introduce a set of methods which allow pre-trained diffusion models to be", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 330, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 506, + 343 + ], + "score": 1.0, + "content": "composed, with one-another and with other models, to create new models without retraining. Second,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 340, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 506, + 354 + ], + "score": 1.0, + "content": "we illustrate how existing methods for composing diffusion models are not fully correct, and propose", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "a remedy to these issues with MCMC-derived sampling. Next, we propose the use of an energy-based", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "score": 1.0, + "content": "parameterization for diffusion models, where the unnormalized density of each reverse diffusion", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 373, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 506, + 386 + ], + "score": 1.0, + "content": "distribution is explicitly modeled. We illustrate how this parameterization enables both additional", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 385, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 506, + 397 + ], + "score": 1.0, + "content": "ways to compose diffusion models, as well as the use of more powerful Metropolis-adjusted MCMC", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "samplers. Finally, we demonstrate the effectiveness of our approach in settings from 2D data to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 406, + 500, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 500, + 419 + ], + "score": 1.0, + "content": "high-resolution text-to-image generation. An illustration of our domains can be found in Figure 1.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 107, + 428, + 200, + 441 + ], + "lines": [ + { + "bbox": [ + 104, + 426, + 202, + 444 + ], + "spans": [ + { + "bbox": [ + 104, + 426, + 202, + 444 + ], + "score": 1.0, + "content": "2 BACKGROUND", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 107, + 449, + 219, + 461 + ], + "lines": [ + { + "bbox": [ + 105, + 448, + 220, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 220, + 462 + ], + "score": 1.0, + "content": "2.1 DIFFUSION MODELS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 467, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 306, + 479 + ], + "score": 1.0, + "content": "Diffusion models seek to model a data distribution", + "type": "text" + }, + { + "bbox": [ + 306, + 468, + 329, + 479 + ], + "score": 0.91, + "content": "q ( x _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 466, + 506, + 479 + ], + "score": 1.0, + "content": ". We augment this distribution with auxiliary", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 474, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 104, + 474, + 146, + 494 + ], + "score": 1.0, + "content": "variables", + "type": "text" + }, + { + "bbox": [ + 146, + 478, + 180, + 490 + ], + "score": 0.92, + "content": "\\{ x _ { t } \\} _ { t = 1 } ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 474, + 309, + 494 + ], + "score": 1.0, + "content": "defining a Gaussian diffusion", + "type": "text" + }, + { + "bbox": [ + 309, + 478, + 505, + 490 + ], + "score": 0.87, + "content": "q ( x _ { 0 } , \\dots , x _ { T } ) = q ( x _ { 0 } ) q ( x _ { 1 } | x _ { 0 } ) \\dots q ( x _ { T } | x _ { T - 1 } )", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 488, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 233, + 502 + ], + "score": 1.0, + "content": "where each transition is defined", + "type": "text" + }, + { + "bbox": [ + 234, + 489, + 395, + 501 + ], + "score": 0.89, + "content": "q ( x _ { t } | x _ { t - 1 } ) = \\mathcal { N } ( x _ { t } ; \\sqrt { 1 - \\beta _ { t } } x _ { t - 1 } , \\beta _ { t } \\bar { I } )", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 488, + 434, + 502 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 434, + 490, + 480, + 500 + ], + "score": 0.88, + "content": "0 < \\beta _ { t } \\le 1", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 488, + 506, + 502 + ], + "score": 1.0, + "content": ". This", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 498, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 215, + 514 + ], + "score": 1.0, + "content": "transition first scales down", + "type": "text" + }, + { + "bbox": [ + 215, + 502, + 235, + 511 + ], + "score": 0.85, + "content": "x _ { t - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 498, + 249, + 514 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 250, + 501, + 285, + 511 + ], + "score": 0.86, + "content": "\\sqrt { 1 - \\beta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 498, + 452, + 514 + ], + "score": 1.0, + "content": "and then adds Gaussian noise of variance", + "type": "text" + }, + { + "bbox": [ + 452, + 501, + 462, + 511 + ], + "score": 0.88, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 498, + 506, + 514 + ], + "score": 1.0, + "content": ". For large", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 510, + 278, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 138, + 524 + ], + "score": 1.0, + "content": "enough", + "type": "text" + }, + { + "bbox": [ + 138, + 512, + 147, + 521 + ], + "score": 0.82, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 510, + 204, + 524 + ], + "score": 1.0, + "content": ", we will have", + "type": "text" + }, + { + "bbox": [ + 204, + 511, + 274, + 523 + ], + "score": 0.92, + "content": "q ( x _ { T } ) \\approx \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 510, + 278, + 524 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 527, + 505, + 564 + ], + "lines": [ + { + "bbox": [ + 106, + 527, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 208, + 541 + ], + "score": 1.0, + "content": "Our model takes the form", + "type": "text" + }, + { + "bbox": [ + 209, + 527, + 258, + 540 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x _ { t - 1 } | x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 527, + 433, + 541 + ], + "score": 1.0, + "content": "and seeks to learn the reverse distribution of", + "type": "text" + }, + { + "bbox": [ + 433, + 528, + 477, + 540 + ], + "score": 0.92, + "content": "q ( x _ { t } | x _ { t - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 527, + 506, + 541 + ], + "score": 1.0, + "content": "which", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 173, + 551 + ], + "score": 1.0, + "content": "seeks to denoise", + "type": "text" + }, + { + "bbox": [ + 174, + 541, + 184, + 550 + ], + "score": 0.81, + "content": "x _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 539, + 195, + 551 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 196, + 541, + 216, + 551 + ], + "score": 0.85, + "content": "x _ { t - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 539, + 302, + 551 + ], + "score": 1.0, + "content": ". In the limit of small", + "type": "text" + }, + { + "bbox": [ + 302, + 540, + 312, + 551 + ], + "score": 0.86, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "this reversal becomes Gaussian (Sohl-Dickstein", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 550, + 462, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 279, + 564 + ], + "score": 1.0, + "content": "et al., 2015) so we parameterize our model", + "type": "text" + }, + { + "bbox": [ + 279, + 550, + 438, + 564 + ], + "score": 0.9, + "content": "p _ { \\theta } ( x _ { t - 1 } | x _ { t } ) = \\mathcal { N } ( x _ { t - 1 } ; \\mu _ { \\theta } ( x _ { t } , t ) , \\tilde { \\beta } _ { t } I )", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 550, + 462, + 564 + ], + "score": 1.0, + "content": "with:", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 567, + 386, + 587 + ], + "lines": [ + { + "bbox": [ + 225, + 567, + 386, + 587 + ], + "spans": [ + { + "bbox": [ + 225, + 567, + 386, + 587 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mu _ { \\theta } ( x _ { t } , t ) = \\frac { 1 } { \\sqrt { \\alpha _ { t } } } \\left( x _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - { { \\bar { \\alpha } } _ { t } } } } \\epsilon _ { \\theta } ( x _ { t } , t ) \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "940729b4adcc76d5314d7c0ae61a214fc9e4b28c0bae9237135fbe91acf48531.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 225, + 567, + 386, + 587 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 410, + 605 + ], + "lines": [ + { + "bbox": [ + 104, + 587, + 407, + 609 + ], + "spans": [ + { + "bbox": [ + 104, + 587, + 133, + 609 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 593, + 168, + 605 + ], + "score": 0.93, + "content": "\\epsilon _ { \\theta } ( x _ { t } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 587, + 267, + 609 + ], + "score": 1.0, + "content": "is a neural network, and", + "type": "text" + }, + { + "bbox": [ + 267, + 591, + 307, + 604 + ], + "score": 0.86, + "content": "\\alpha _ { t } , \\bar { \\alpha } _ { t } , \\tilde { \\beta } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 587, + 374, + 609 + ], + "score": 1.0, + "content": "are functions of", + "type": "text" + }, + { + "bbox": [ + 374, + 591, + 407, + 605 + ], + "score": 0.92, + "content": "\\{ \\beta _ { t } \\} _ { t = 1 } ^ { T }", + "type": "inline_equation" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 609, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 104, + 608, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 104, + 608, + 278, + 622 + ], + "score": 1.0, + "content": "A useful feature of the diffusion process", + "type": "text" + }, + { + "bbox": [ + 278, + 612, + 284, + 621 + ], + "score": 0.8, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 608, + 506, + 622 + ], + "score": 1.0, + "content": "is that we can analytically derive any time marginal", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 621, + 504, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 256, + 635 + ], + "score": 0.92, + "content": "q ( x _ { t } | x _ { 0 } ) = \\mathcal { N } ( x _ { t } ; \\sqrt { 1 - \\sigma _ { t } ^ { 2 } } x _ { 0 } , \\sigma _ { t } ^ { 2 } I )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 621, + 311, + 636 + ], + "score": 1.0, + "content": "where again", + "type": "text" + }, + { + "bbox": [ + 311, + 624, + 321, + 633 + ], + "score": 0.82, + "content": "\\sigma _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 621, + 389, + 636 + ], + "score": 1.0, + "content": "is a function of", + "type": "text" + }, + { + "bbox": [ + 389, + 621, + 422, + 635 + ], + "score": 0.91, + "content": "\\{ \\beta _ { t } \\} _ { t = 1 } ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 621, + 493, + 636 + ], + "score": 1.0, + "content": ". We can sample", + "type": "text" + }, + { + "bbox": [ + 493, + 624, + 504, + 633 + ], + "score": 0.77, + "content": "x _ { t }", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 634, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 304, + 649 + ], + "score": 1.0, + "content": "from this distribution using reparameterization, i.e", + "type": "text" + }, + { + "bbox": [ + 305, + 634, + 426, + 648 + ], + "score": 0.93, + "content": "x _ { t } ( x _ { 0 } , \\epsilon ) = \\sqrt { 1 - \\sigma _ { t } ^ { 2 } } x _ { 0 } + \\sigma _ { t } \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 635, + 452, + 649 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 453, + 635, + 503, + 648 + ], + "score": 0.92, + "content": "\\epsilon \\sim \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 635, + 506, + 649 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 646, + 375, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 375, + 659 + ], + "score": 1.0, + "content": "Exploiting this, diffusion models are typically trained with the loss", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "interline_equation", + "bbox": [ + 155, + 662, + 455, + 678 + ], + "lines": [ + { + "bbox": [ + 155, + 662, + 455, + 678 + ], + "spans": [ + { + "bbox": [ + 155, + 662, + 455, + 678 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathcal { L } ( \\theta ) = \\sum _ { t = 1 } ^ { T } \\mathcal { L } _ { t } ( \\theta ) , \\qquad \\mathcal { L } _ { t } ( \\theta ) = \\mathbb { E } _ { q ( x _ { 0 } ) , N ( \\epsilon ; 0 , I ) } \\left[ | | \\epsilon - \\epsilon _ { \\theta } ( x _ { t } ( x _ { 0 } , \\epsilon ) , t ) | | ^ { 2 } \\right] . } \\end{array}", + "type": "interline_equation", + "image_path": "6987c428394ab12452ed372b67a1ae940d352ea11c2d7c7ee4dc337dcfeaa70f.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 155, + 662, + 455, + 678 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 131, + 701 + ], + "score": 1.0, + "content": "Once", + "type": "text" + }, + { + "bbox": [ + 132, + 687, + 163, + 700 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 686, + 260, + 701 + ], + "score": 1.0, + "content": "is trained, we recover", + "type": "text" + }, + { + "bbox": [ + 261, + 688, + 293, + 700 + ], + "score": 0.94, + "content": "\\mu _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 686, + 435, + 701 + ], + "score": 1.0, + "content": "with Equation 1 to parameterize", + "type": "text" + }, + { + "bbox": [ + 436, + 687, + 485, + 700 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x _ { t - 1 } | x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 686, + 506, + 701 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "perform ancestral sampling (also known as the reverse process) to reverse the diffusion, i.e sample", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 107, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 107, + 709, + 165, + 721 + ], + "score": 0.93, + "content": "\\overline { { x _ { T } } } \\sim \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 709, + 203, + 722 + ], + "score": 1.0, + "content": ", then for", + "type": "text" + }, + { + "bbox": [ + 204, + 710, + 267, + 720 + ], + "score": 0.9, + "content": "t = T - 1 1", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 709, + 302, + 722 + ], + "score": 1.0, + "content": ", sample", + "type": "text" + }, + { + "bbox": [ + 302, + 710, + 385, + 722 + ], + "score": 0.92, + "content": "x _ { t - 1 } \\sim p _ { \\theta } ( x _ { t - 1 } | x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 709, + 506, + 722 + ], + "score": 1.0, + "content": ". A more detailed description", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 721, + 224, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 224, + 732 + ], + "score": 1.0, + "content": "can be found in Appendix B.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 82, + 501, + 213 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 82, + 501, + 213 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 82, + 501, + 213 + ], + "spans": [ + { + "bbox": [ + 110, + 82, + 501, + 213 + ], + "score": 0.973, + "type": "image", + "image_path": "e319142d3909a1775051d2649b2726a3b96107e89f1adbd0fbb4ad3ca99ee902.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 82, + 501, + 125.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 125.66666666666666, + 501, + 169.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 169.33333333333331, + 501, + 212.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 222, + 504, + 253 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 222, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 222, + 505, + 234 + ], + "score": 1.0, + "content": "Figure 1: Creating new models through composition. Simple operators enable diffusion models to be", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 232, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 244 + ], + "score": 1.0, + "content": "composed without retraining in settings such (a) products, (b) classifier conditioning, (c) compositional text-to-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 243, + 475, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 475, + 254 + ], + "score": 1.0, + "content": "image generation with a product (left) and a mixture (right). All samples generated by trained models.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 258, + 506, + 302 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 506, + 271 + ], + "score": 1.0, + "content": "Hinton, 2000). Here, many simple models or predictors were combined to increase their capacity.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 269, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 269, + 506, + 281 + ], + "score": 1.0, + "content": "Much of this early work on model composition was done in the context of Energy-Based Mod-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "els (Hinton, 2002), an alternative class of generative model which bears many similarities to diffusion", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 291, + 141, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 141, + 302 + ], + "score": 1.0, + "content": "models.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 257, + 506, + 302 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 307, + 506, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 308, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 506, + 320 + ], + "score": 1.0, + "content": "In this work, we explore the ways that diffusion models can be reused and composed with one-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 319, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 506, + 331 + ], + "score": 1.0, + "content": "another. First, we introduce a set of methods which allow pre-trained diffusion models to be", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 330, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 506, + 343 + ], + "score": 1.0, + "content": "composed, with one-another and with other models, to create new models without retraining. Second,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 340, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 506, + 354 + ], + "score": 1.0, + "content": "we illustrate how existing methods for composing diffusion models are not fully correct, and propose", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "a remedy to these issues with MCMC-derived sampling. Next, we propose the use of an energy-based", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "score": 1.0, + "content": "parameterization for diffusion models, where the unnormalized density of each reverse diffusion", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 373, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 506, + 386 + ], + "score": 1.0, + "content": "distribution is explicitly modeled. We illustrate how this parameterization enables both additional", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 385, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 506, + 397 + ], + "score": 1.0, + "content": "ways to compose diffusion models, as well as the use of more powerful Metropolis-adjusted MCMC", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "samplers. Finally, we demonstrate the effectiveness of our approach in settings from 2D data to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 406, + 500, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 500, + 419 + ], + "score": 1.0, + "content": "high-resolution text-to-image generation. An illustration of our domains can be found in Figure 1.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 308, + 506, + 419 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 428, + 200, + 441 + ], + "lines": [ + { + "bbox": [ + 104, + 426, + 202, + 444 + ], + "spans": [ + { + "bbox": [ + 104, + 426, + 202, + 444 + ], + "score": 1.0, + "content": "2 BACKGROUND", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 107, + 449, + 219, + 461 + ], + "lines": [ + { + "bbox": [ + 105, + 448, + 220, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 220, + 462 + ], + "score": 1.0, + "content": "2.1 DIFFUSION MODELS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 467, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 306, + 479 + ], + "score": 1.0, + "content": "Diffusion models seek to model a data distribution", + "type": "text" + }, + { + "bbox": [ + 306, + 468, + 329, + 479 + ], + "score": 0.91, + "content": "q ( x _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 466, + 506, + 479 + ], + "score": 1.0, + "content": ". We augment this distribution with auxiliary", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 474, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 104, + 474, + 146, + 494 + ], + "score": 1.0, + "content": "variables", + "type": "text" + }, + { + "bbox": [ + 146, + 478, + 180, + 490 + ], + "score": 0.92, + "content": "\\{ x _ { t } \\} _ { t = 1 } ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 474, + 309, + 494 + ], + "score": 1.0, + "content": "defining a Gaussian diffusion", + "type": "text" + }, + { + "bbox": [ + 309, + 478, + 505, + 490 + ], + "score": 0.87, + "content": "q ( x _ { 0 } , \\dots , x _ { T } ) = q ( x _ { 0 } ) q ( x _ { 1 } | x _ { 0 } ) \\dots q ( x _ { T } | x _ { T - 1 } )", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 488, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 233, + 502 + ], + "score": 1.0, + "content": "where each transition is defined", + "type": "text" + }, + { + "bbox": [ + 234, + 489, + 395, + 501 + ], + "score": 0.89, + "content": "q ( x _ { t } | x _ { t - 1 } ) = \\mathcal { N } ( x _ { t } ; \\sqrt { 1 - \\beta _ { t } } x _ { t - 1 } , \\beta _ { t } \\bar { I } )", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 488, + 434, + 502 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 434, + 490, + 480, + 500 + ], + "score": 0.88, + "content": "0 < \\beta _ { t } \\le 1", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 488, + 506, + 502 + ], + "score": 1.0, + "content": ". This", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 498, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 215, + 514 + ], + "score": 1.0, + "content": "transition first scales down", + "type": "text" + }, + { + "bbox": [ + 215, + 502, + 235, + 511 + ], + "score": 0.85, + "content": "x _ { t - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 498, + 249, + 514 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 250, + 501, + 285, + 511 + ], + "score": 0.86, + "content": "\\sqrt { 1 - \\beta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 498, + 452, + 514 + ], + "score": 1.0, + "content": "and then adds Gaussian noise of variance", + "type": "text" + }, + { + "bbox": [ + 452, + 501, + 462, + 511 + ], + "score": 0.88, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 498, + 506, + 514 + ], + "score": 1.0, + "content": ". For large", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 510, + 278, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 138, + 524 + ], + "score": 1.0, + "content": "enough", + "type": "text" + }, + { + "bbox": [ + 138, + 512, + 147, + 521 + ], + "score": 0.82, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 510, + 204, + 524 + ], + "score": 1.0, + "content": ", we will have", + "type": "text" + }, + { + "bbox": [ + 204, + 511, + 274, + 523 + ], + "score": 0.92, + "content": "q ( x _ { T } ) \\approx \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 510, + 278, + 524 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24, + "bbox_fs": [ + 104, + 466, + 506, + 524 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 527, + 505, + 564 + ], + "lines": [ + { + "bbox": [ + 106, + 527, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 208, + 541 + ], + "score": 1.0, + "content": "Our model takes the form", + "type": "text" + }, + { + "bbox": [ + 209, + 527, + 258, + 540 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x _ { t - 1 } | x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 527, + 433, + 541 + ], + "score": 1.0, + "content": "and seeks to learn the reverse distribution of", + "type": "text" + }, + { + "bbox": [ + 433, + 528, + 477, + 540 + ], + "score": 0.92, + "content": "q ( x _ { t } | x _ { t - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 527, + 506, + 541 + ], + "score": 1.0, + "content": "which", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 173, + 551 + ], + "score": 1.0, + "content": "seeks to denoise", + "type": "text" + }, + { + "bbox": [ + 174, + 541, + 184, + 550 + ], + "score": 0.81, + "content": "x _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 539, + 195, + 551 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 196, + 541, + 216, + 551 + ], + "score": 0.85, + "content": "x _ { t - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 539, + 302, + 551 + ], + "score": 1.0, + "content": ". In the limit of small", + "type": "text" + }, + { + "bbox": [ + 302, + 540, + 312, + 551 + ], + "score": 0.86, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "this reversal becomes Gaussian (Sohl-Dickstein", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 550, + 462, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 279, + 564 + ], + "score": 1.0, + "content": "et al., 2015) so we parameterize our model", + "type": "text" + }, + { + "bbox": [ + 279, + 550, + 438, + 564 + ], + "score": 0.9, + "content": "p _ { \\theta } ( x _ { t - 1 } | x _ { t } ) = \\mathcal { N } ( x _ { t - 1 } ; \\mu _ { \\theta } ( x _ { t } , t ) , \\tilde { \\beta } _ { t } I )", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 550, + 462, + 564 + ], + "score": 1.0, + "content": "with:", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 527, + 506, + 564 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 567, + 386, + 587 + ], + "lines": [ + { + "bbox": [ + 225, + 567, + 386, + 587 + ], + "spans": [ + { + "bbox": [ + 225, + 567, + 386, + 587 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mu _ { \\theta } ( x _ { t } , t ) = \\frac { 1 } { \\sqrt { \\alpha _ { t } } } \\left( x _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - { { \\bar { \\alpha } } _ { t } } } } \\epsilon _ { \\theta } ( x _ { t } , t ) \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "940729b4adcc76d5314d7c0ae61a214fc9e4b28c0bae9237135fbe91acf48531.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 225, + 567, + 386, + 587 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 410, + 605 + ], + "lines": [ + { + "bbox": [ + 104, + 587, + 407, + 609 + ], + "spans": [ + { + "bbox": [ + 104, + 587, + 133, + 609 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 593, + 168, + 605 + ], + "score": 0.93, + "content": "\\epsilon _ { \\theta } ( x _ { t } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 587, + 267, + 609 + ], + "score": 1.0, + "content": "is a neural network, and", + "type": "text" + }, + { + "bbox": [ + 267, + 591, + 307, + 604 + ], + "score": 0.86, + "content": "\\alpha _ { t } , \\bar { \\alpha } _ { t } , \\tilde { \\beta } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 587, + 374, + 609 + ], + "score": 1.0, + "content": "are functions of", + "type": "text" + }, + { + "bbox": [ + 374, + 591, + 407, + 605 + ], + "score": 0.92, + "content": "\\{ \\beta _ { t } \\} _ { t = 1 } ^ { T }", + "type": "inline_equation" + } + ], + "index": 31 + } + ], + "index": 31, + "bbox_fs": [ + 104, + 587, + 407, + 609 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 609, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 104, + 608, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 104, + 608, + 278, + 622 + ], + "score": 1.0, + "content": "A useful feature of the diffusion process", + "type": "text" + }, + { + "bbox": [ + 278, + 612, + 284, + 621 + ], + "score": 0.8, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 608, + 506, + 622 + ], + "score": 1.0, + "content": "is that we can analytically derive any time marginal", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 621, + 504, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 256, + 635 + ], + "score": 0.92, + "content": "q ( x _ { t } | x _ { 0 } ) = \\mathcal { N } ( x _ { t } ; \\sqrt { 1 - \\sigma _ { t } ^ { 2 } } x _ { 0 } , \\sigma _ { t } ^ { 2 } I )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 621, + 311, + 636 + ], + "score": 1.0, + "content": "where again", + "type": "text" + }, + { + "bbox": [ + 311, + 624, + 321, + 633 + ], + "score": 0.82, + "content": "\\sigma _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 621, + 389, + 636 + ], + "score": 1.0, + "content": "is a function of", + "type": "text" + }, + { + "bbox": [ + 389, + 621, + 422, + 635 + ], + "score": 0.91, + "content": "\\{ \\beta _ { t } \\} _ { t = 1 } ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 621, + 493, + 636 + ], + "score": 1.0, + "content": ". We can sample", + "type": "text" + }, + { + "bbox": [ + 493, + 624, + 504, + 633 + ], + "score": 0.77, + "content": "x _ { t }", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 634, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 304, + 649 + ], + "score": 1.0, + "content": "from this distribution using reparameterization, i.e", + "type": "text" + }, + { + "bbox": [ + 305, + 634, + 426, + 648 + ], + "score": 0.93, + "content": "x _ { t } ( x _ { 0 } , \\epsilon ) = \\sqrt { 1 - \\sigma _ { t } ^ { 2 } } x _ { 0 } + \\sigma _ { t } \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 635, + 452, + 649 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 453, + 635, + 503, + 648 + ], + "score": 0.92, + "content": "\\epsilon \\sim \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 635, + 506, + 649 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 646, + 375, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 375, + 659 + ], + "score": 1.0, + "content": "Exploiting this, diffusion models are typically trained with the loss", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5, + "bbox_fs": [ + 104, + 608, + 506, + 659 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 155, + 662, + 455, + 678 + ], + "lines": [ + { + "bbox": [ + 155, + 662, + 455, + 678 + ], + "spans": [ + { + "bbox": [ + 155, + 662, + 455, + 678 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathcal { L } ( \\theta ) = \\sum _ { t = 1 } ^ { T } \\mathcal { L } _ { t } ( \\theta ) , \\qquad \\mathcal { L } _ { t } ( \\theta ) = \\mathbb { E } _ { q ( x _ { 0 } ) , N ( \\epsilon ; 0 , I ) } \\left[ | | \\epsilon - \\epsilon _ { \\theta } ( x _ { t } ( x _ { 0 } , \\epsilon ) , t ) | | ^ { 2 } \\right] . } \\end{array}", + "type": "interline_equation", + "image_path": "6987c428394ab12452ed372b67a1ae940d352ea11c2d7c7ee4dc337dcfeaa70f.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 155, + 662, + 455, + 678 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 131, + 701 + ], + "score": 1.0, + "content": "Once", + "type": "text" + }, + { + "bbox": [ + 132, + 687, + 163, + 700 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 686, + 260, + 701 + ], + "score": 1.0, + "content": "is trained, we recover", + "type": "text" + }, + { + "bbox": [ + 261, + 688, + 293, + 700 + ], + "score": 0.94, + "content": "\\mu _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 686, + 435, + 701 + ], + "score": 1.0, + "content": "with Equation 1 to parameterize", + "type": "text" + }, + { + "bbox": [ + 436, + 687, + 485, + 700 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x _ { t - 1 } | x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 686, + 506, + 701 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "perform ancestral sampling (also known as the reverse process) to reverse the diffusion, i.e sample", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 107, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 107, + 709, + 165, + 721 + ], + "score": 0.93, + "content": "\\overline { { x _ { T } } } \\sim \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 709, + 203, + 722 + ], + "score": 1.0, + "content": ", then for", + "type": "text" + }, + { + "bbox": [ + 204, + 710, + 267, + 720 + ], + "score": 0.9, + "content": "t = T - 1 1", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 709, + 302, + 722 + ], + "score": 1.0, + "content": ", sample", + "type": "text" + }, + { + "bbox": [ + 302, + 710, + 385, + 722 + ], + "score": 0.92, + "content": "x _ { t - 1 } \\sim p _ { \\theta } ( x _ { t - 1 } | x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 709, + 506, + 722 + ], + "score": 1.0, + "content": ". A more detailed description", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 721, + 224, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 224, + 732 + ], + "score": 1.0, + "content": "can be found in Appendix B.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 686, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 81, + 346, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 347, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 347, + 95 + ], + "score": 1.0, + "content": "2.2 ENERGY-BASED MODELS AND MCMC SAMPLING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 99, + 506, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 506, + 114 + ], + "score": 1.0, + "content": "Energy-Based Models (EBMs) are a class of probabilistic model which parameterize a distribution as", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 111, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 111, + 167, + 128 + ], + "score": 0.94, + "content": "\\begin{array} { r } { p _ { \\theta } ( x ) = \\frac { e ^ { f _ { \\theta } ( x ) } } { Z ( \\theta ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 113, + 296, + 127 + ], + "score": 1.0, + "content": "where the normalizing constant", + "type": "text" + }, + { + "bbox": [ + 297, + 112, + 374, + 127 + ], + "score": 0.94, + "content": "\\begin{array} { r } { Z ( \\theta ) = \\int e ^ { f _ { \\theta } ( x ) } d x } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 113, + 506, + 127 + ], + "score": 1.0, + "content": "is not modeled. Choosing not to", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "model this quantity gives the model much more flexibility but comes with considerable limitations.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 506, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 506, + 151 + ], + "score": 1.0, + "content": "We can no longer efficiently compute likelihoods or draw samples from the model. This complicates", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 149, + 403, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 403, + 161 + ], + "score": 1.0, + "content": "training, as most generative models are trained by maximizing likelihood.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 505, + 200 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "score": 1.0, + "content": "One popular method for EBM training is denoising score matching. This approach minimizes the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 175, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 505, + 190 + ], + "score": 1.0, + "content": "Fisher Divergence1 between the model and a Gaussian-smoothed version of the data distribution", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 187, + 400, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 243, + 200 + ], + "score": 0.91, + "content": "\\begin{array} { r } { q _ { \\sigma } ( x ) = \\int q ( x ^ { \\prime } ) \\mathcal { N } ( x ; x ^ { \\prime } , \\sigma ^ { 2 } I ) d x ^ { \\prime } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 187, + 400, + 200 + ], + "score": 1.0, + "content": "by minimizing the following objective", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 205, + 406, + 226 + ], + "lines": [ + { + "bbox": [ + 205, + 205, + 406, + 226 + ], + "spans": [ + { + "bbox": [ + 205, + 205, + 406, + 226 + ], + "score": 0.92, + "content": "\\mathcal { I } _ { \\sigma } ( \\theta ) = \\mathbb { E } _ { q ( x ) N ( \\epsilon ; 0 , I ) } \\left[ \\big | \\big | \\frac { \\epsilon } { \\sigma } + \\nabla _ { x } f _ { \\theta } ( x + \\sigma \\epsilon ) \\big | \\big | ^ { 2 } \\right] .", + "type": "interline_equation", + "image_path": "e103012da360353b2d5aa138d91378a04471bff41436a1289314694db474dd4b.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 205, + 205, + 406, + 226 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 505, + 300 + ], + "lines": [ + { + "bbox": [ + 104, + 231, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 104, + 231, + 254, + 247 + ], + "score": 1.0, + "content": "When minimized, this ensures that", + "type": "text" + }, + { + "bbox": [ + 254, + 232, + 318, + 246 + ], + "score": 0.93, + "content": "e ^ { f _ { \\theta } ( x ) } \\propto q _ { \\sigma } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 231, + 379, + 247 + ], + "score": 1.0, + "content": "and therefore", + "type": "text" + }, + { + "bbox": [ + 379, + 233, + 485, + 246 + ], + "score": 0.92, + "content": "\\nabla _ { x } f _ { \\theta } ( x ) = \\nabla _ { x } \\log { q _ { \\sigma } ( x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 231, + 506, + 247 + ], + "score": 1.0, + "content": ". To", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 244, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 505, + 257 + ], + "score": 1.0, + "content": "estimate likelihoods or sample from our model, we must rely on approximate methods, such as", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 256, + 504, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 504, + 267 + ], + "score": 1.0, + "content": "MCMC sampling or numerical ODE integration. MCMC works by simulating a Markov chain", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 266, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 160, + 280 + ], + "score": 1.0, + "content": "beginning at", + "type": "text" + }, + { + "bbox": [ + 160, + 267, + 208, + 279 + ], + "score": 0.93, + "content": "x _ { 0 } \\sim p ( x _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 266, + 350, + 280 + ], + "score": 1.0, + "content": "and using a transition distribution", + "type": "text" + }, + { + "bbox": [ + 350, + 266, + 419, + 279 + ], + "score": 0.92, + "content": "x _ { t } \\sim k ( \\bar { x } _ { t } | x _ { t - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 266, + 434, + 280 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 435, + 267, + 457, + 279 + ], + "score": 0.92, + "content": "k ( \\cdot | \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 266, + 505, + 280 + ], + "score": 1.0, + "content": "has certain", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 277, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 104, + 277, + 393, + 291 + ], + "score": 1.0, + "content": "properties, namely invariance w.r.t. the target and ergodicity, then as", + "type": "text" + }, + { + "bbox": [ + 394, + 279, + 426, + 288 + ], + "score": 0.87, + "content": "t \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 277, + 431, + 291 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 431, + 280, + 441, + 289 + ], + "score": 0.82, + "content": "x _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 277, + 506, + 291 + ], + "score": 1.0, + "content": "converges to a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 288, + 251, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 251, + 302 + ], + "score": 1.0, + "content": "sample from our target distribution.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 305, + 504, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 305, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 505, + 318 + ], + "score": 1.0, + "content": "Perhaps the most popular MCMC sampling algorithm for EBMs is Unadjusted Langevin Dynamics", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 315, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 505, + 330 + ], + "score": 1.0, + "content": "(ULA) (Roberts & Tweedie, 1996; Du & Mordatch, 2019; Nijkamp et al., 2020) which is defined by", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 333, + 406, + 353 + ], + "lines": [ + { + "bbox": [ + 205, + 333, + 406, + 353 + ], + "spans": [ + { + "bbox": [ + 205, + 333, + 406, + 353 + ], + "score": 0.92, + "content": "\\begin{array} { r } { k ( x _ { t } | x _ { t - 1 } ) = N \\left( x _ { t } ; x _ { t - 1 } + \\frac { \\sigma ^ { 2 } } { 2 } \\nabla _ { x } f _ { \\theta } ( x _ { t - 1 } ) , \\sigma ^ { 2 } I \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "113e4260d357d7ff1b47335eea991e588075f78cdb8da7cb5b7d79d90a6dfd24.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 205, + 333, + 406, + 353 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 360, + 505, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 359, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 331, + 375 + ], + "score": 1.0, + "content": "This resembles a step of gradient ascent (with step-size", + "type": "text" + }, + { + "bbox": [ + 331, + 359, + 343, + 375 + ], + "score": 0.86, + "content": "\\frac { \\sigma ^ { 2 } } { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 360, + 506, + 375 + ], + "score": 1.0, + "content": ") with added Gaussian noise of variance", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 372, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 107, + 374, + 118, + 385 + ], + "score": 0.84, + "content": "\\sigma ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 372, + 505, + 388 + ], + "score": 1.0, + "content": ". This transition is based on a discretization of the Langevin SDE. In the limit of infinitesimally", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 384, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 104, + 384, + 131, + 399 + ], + "score": 1.0, + "content": "small", + "type": "text" + }, + { + "bbox": [ + 131, + 387, + 138, + 395 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 384, + 505, + 399 + ], + "score": 1.0, + "content": "this approach will draw exact samples. To handle the error accrued when using larger step", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 396, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 409 + ], + "score": 1.0, + "content": "sizes, a Metropolis correction can be added giving the Metropolis-Adjusted-Langevin-Algorithm", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 406, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 484, + 420 + ], + "score": 1.0, + "content": "(MALA) (Besag, 1994). With Metropolis correction, we first generate a proposed update", + "type": "text" + }, + { + "bbox": [ + 484, + 408, + 505, + 418 + ], + "score": 0.84, + "content": "\\hat { x } \\sim", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 418, + 502, + 438 + ], + "spans": [ + { + "bbox": [ + 107, + 421, + 148, + 434 + ], + "score": 0.91, + "content": "k ( x | x _ { t - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 421, + 239, + 435 + ], + "score": 1.0, + "content": ", then with probability", + "type": "text" + }, + { + "bbox": [ + 239, + 418, + 351, + 438 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\operatorname* { m i n } \\left( 1 , \\frac { e ^ { f _ { \\theta } ( \\hat { x } ) } } { e ^ { f _ { \\theta } ( x _ { t - 1 } ) } } \\frac { k \\left( x _ { t - 1 } | \\hat { x } \\right) } { k \\left( \\hat { x } | x _ { t - 1 } \\right) } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 421, + 380, + 435 + ], + "score": 1.0, + "content": "we set", + "type": "text" + }, + { + "bbox": [ + 380, + 422, + 410, + 433 + ], + "score": 0.89, + "content": "\\boldsymbol { x } _ { t } = \\boldsymbol { \\hat { x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 421, + 455, + 435 + ], + "score": 1.0, + "content": ", otherwise", + "type": "text" + }, + { + "bbox": [ + 455, + 424, + 497, + 434 + ], + "score": 0.86, + "content": "x _ { t } = x _ { t - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 421, + 502, + 435 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 441, + 506, + 498 + ], + "lines": [ + { + "bbox": [ + 106, + 442, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 505, + 454 + ], + "score": 1.0, + "content": "Hamiltonian Monte Carlo (HMC) (Duane et al., 1987; Neal, 1996) is a more advanced MCMC", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "score": 1.0, + "content": "sampling method which augments the state-space with auxiliary momentum variables and numerically", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "integrates energy-conserving Hamiltonian dynamics to advance the sampler. HMC is typically applied", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 475, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 506, + 487 + ], + "score": 1.0, + "content": "with a Metropolis correction, but an approximate variant can be used without it (U-HMC) (Geffner &", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 485, + 389, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 389, + 500 + ], + "score": 1.0, + "content": "Domke, 2021). See Appendix C.1 for details of HMC variants we use.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 107, + 507, + 383, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 507, + 383, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 383, + 520 + ], + "score": 1.0, + "content": "2.3 RELATIONSHIP BETWEEN DIFFUSION MODELS AND EBMS", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 505, + 559 + ], + "lines": [ + { + "bbox": [ + 105, + 525, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 537 + ], + "score": 1.0, + "content": "Diffusion models and EBMs are closely related. For instance, Song & Ermon (2019) uses an EBM", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 535, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 104, + 535, + 506, + 551 + ], + "score": 1.0, + "content": "perspective to propose a close cousin to diffusion models. We can see from inspection that the training", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 547, + 504, + 560 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 504, + 560 + ], + "score": 1.0, + "content": "objective of diffusion models is identical (up to a constant) to the denoising score matching objective", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "interline_equation", + "bbox": [ + 177, + 564, + 434, + 586 + ], + "lines": [ + { + "bbox": [ + 177, + 564, + 434, + 586 + ], + "spans": [ + { + "bbox": [ + 177, + 564, + 434, + 586 + ], + "score": 0.92, + "content": "\\sigma _ { t } ^ { 2 } \\mathcal { I } _ { \\sigma _ { t } } ( \\theta ) = \\mathbb { E } _ { q ( x ) \\mathcal { N } ( \\epsilon ; 0 , I ) } \\left[ | | \\epsilon + \\sigma _ { t } \\nabla _ { x } f _ { \\theta } ( x + \\sigma _ { t } \\epsilon ) | | ^ { 2 } \\right] = \\mathcal { L } _ { t } ( \\theta )", + "type": "interline_equation", + "image_path": "19cf0a67d18e3205da24d1fac2da27f1bd314de81b22e0a33e4f779af41543da.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 177, + 564, + 434, + 586 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 590, + 505, + 652 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 208, + 604 + ], + "score": 1.0, + "content": "where we have replaced", + "type": "text" + }, + { + "bbox": [ + 208, + 591, + 239, + 603 + ], + "score": 0.93, + "content": "\\epsilon _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 590, + 262, + 604 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 263, + 591, + 343, + 604 + ], + "score": 0.91, + "content": "- \\sigma _ { t } \\nabla _ { x } f _ { \\theta } ( x + \\sigma _ { t } \\epsilon )", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 590, + 420, + 604 + ], + "score": 1.0, + "content": ". Thus by training", + "type": "text" + }, + { + "bbox": [ + 421, + 591, + 451, + 603 + ], + "score": 0.91, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 590, + 505, + 604 + ], + "score": 1.0, + "content": "to minimize", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 602, + 504, + 619 + ], + "spans": [ + { + "bbox": [ + 104, + 602, + 481, + 619 + ], + "score": 1.0, + "content": "Equation 2, we can recover the diffused data distribution score with ∇x log qσ(x) ≈ − ϵθ(x,t)σ .", + "type": "text" + }, + { + "bbox": [ + 476, + 604, + 504, + 616 + ], + "score": 1.0, + "content": "From", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 617, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 184, + 631 + ], + "score": 1.0, + "content": "this, we can define", + "type": "text" + }, + { + "bbox": [ + 184, + 617, + 272, + 630 + ], + "score": 0.95, + "content": "\\epsilon _ { \\theta } ( x , t ) = \\nabla _ { x } f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 617, + 506, + 631 + ], + "score": 1.0, + "content": "(the derivative of an explicitly defined scalar function) to", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 280, + 641 + ], + "score": 1.0, + "content": "learn a noise-conditional potential function", + "type": "text" + }, + { + "bbox": [ + 280, + 629, + 311, + 641 + ], + "score": 0.93, + "content": "f _ { \\boldsymbol { \\theta } } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 629, + 505, + 641 + ], + "score": 1.0, + "content": ". We later demonstrate the benefits of this in two", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "score": 1.0, + "content": "ways; it enables the use of more sophisticated sampling algorithms and more forms of composition.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37 + }, + { + "type": "title", + "bbox": [ + 108, + 661, + 261, + 673 + ], + "lines": [ + { + "bbox": [ + 106, + 662, + 262, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 262, + 673 + ], + "score": 1.0, + "content": "2.4 CONTROLLABLE GENERATION", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 678, + 505, + 713 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 276, + 692 + ], + "score": 1.0, + "content": "It may be convenient to train a model of", + "type": "text" + }, + { + "bbox": [ + 276, + 679, + 296, + 691 + ], + "score": 0.92, + "content": "p ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 678, + 325, + 692 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 325, + 681, + 332, + 689 + ], + "score": 0.78, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 678, + 506, + 692 + ], + "score": 1.0, + "content": "is, say, the distribution of all images, but", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 325, + 703 + ], + "score": 1.0, + "content": "in practice we often want to generate samples from", + "type": "text" + }, + { + "bbox": [ + 326, + 690, + 353, + 702 + ], + "score": 0.92, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 690, + 384, + 703 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 384, + 692, + 391, + 702 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 690, + 505, + 703 + ], + "score": 1.0, + "content": "is some attribute, label, or", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "feature. This can be accomplished within the framework of diffusion models by introducing a learned", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 118, + 720, + 423, + 734 + ], + "lines": [ + { + "bbox": [ + 118, + 718, + 421, + 735 + ], + "spans": [ + { + "bbox": [ + 118, + 718, + 244, + 735 + ], + "score": 1.0, + "content": "1The Fisher divergence is defined:", + "type": "text" + }, + { + "bbox": [ + 244, + 720, + 421, + 733 + ], + "score": 0.74, + "content": "\\mathbf { F } ( p | | q ) = \\mathbb { E } _ { p } \\left[ | | \\nabla _ { x } \\log p ( x ) - \\nabla _ { x } \\log q ( x ) | | ^ { 2 } \\right] .", + "type": "inline_equation" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 81, + 346, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 347, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 347, + 95 + ], + "score": 1.0, + "content": "2.2 ENERGY-BASED MODELS AND MCMC SAMPLING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 99, + 506, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 506, + 114 + ], + "score": 1.0, + "content": "Energy-Based Models (EBMs) are a class of probabilistic model which parameterize a distribution as", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 111, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 111, + 167, + 128 + ], + "score": 0.94, + "content": "\\begin{array} { r } { p _ { \\theta } ( x ) = \\frac { e ^ { f _ { \\theta } ( x ) } } { Z ( \\theta ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 113, + 296, + 127 + ], + "score": 1.0, + "content": "where the normalizing constant", + "type": "text" + }, + { + "bbox": [ + 297, + 112, + 374, + 127 + ], + "score": 0.94, + "content": "\\begin{array} { r } { Z ( \\theta ) = \\int e ^ { f _ { \\theta } ( x ) } d x } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 113, + 506, + 127 + ], + "score": 1.0, + "content": "is not modeled. Choosing not to", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "model this quantity gives the model much more flexibility but comes with considerable limitations.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 138, + 506, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 506, + 151 + ], + "score": 1.0, + "content": "We can no longer efficiently compute likelihoods or draw samples from the model. This complicates", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 149, + 403, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 403, + 161 + ], + "score": 1.0, + "content": "training, as most generative models are trained by maximizing likelihood.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 99, + 506, + 161 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 505, + 200 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "score": 1.0, + "content": "One popular method for EBM training is denoising score matching. This approach minimizes the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 175, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 505, + 190 + ], + "score": 1.0, + "content": "Fisher Divergence1 between the model and a Gaussian-smoothed version of the data distribution", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 187, + 400, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 243, + 200 + ], + "score": 0.91, + "content": "\\begin{array} { r } { q _ { \\sigma } ( x ) = \\int q ( x ^ { \\prime } ) \\mathcal { N } ( x ; x ^ { \\prime } , \\sigma ^ { 2 } I ) d x ^ { \\prime } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 187, + 400, + 200 + ], + "score": 1.0, + "content": "by minimizing the following objective", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 164, + 505, + 200 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 205, + 406, + 226 + ], + "lines": [ + { + "bbox": [ + 205, + 205, + 406, + 226 + ], + "spans": [ + { + "bbox": [ + 205, + 205, + 406, + 226 + ], + "score": 0.92, + "content": "\\mathcal { I } _ { \\sigma } ( \\theta ) = \\mathbb { E } _ { q ( x ) N ( \\epsilon ; 0 , I ) } \\left[ \\big | \\big | \\frac { \\epsilon } { \\sigma } + \\nabla _ { x } f _ { \\theta } ( x + \\sigma \\epsilon ) \\big | \\big | ^ { 2 } \\right] .", + "type": "interline_equation", + "image_path": "e103012da360353b2d5aa138d91378a04471bff41436a1289314694db474dd4b.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 205, + 205, + 406, + 226 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 505, + 300 + ], + "lines": [ + { + "bbox": [ + 104, + 231, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 104, + 231, + 254, + 247 + ], + "score": 1.0, + "content": "When minimized, this ensures that", + "type": "text" + }, + { + "bbox": [ + 254, + 232, + 318, + 246 + ], + "score": 0.93, + "content": "e ^ { f _ { \\theta } ( x ) } \\propto q _ { \\sigma } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 231, + 379, + 247 + ], + "score": 1.0, + "content": "and therefore", + "type": "text" + }, + { + "bbox": [ + 379, + 233, + 485, + 246 + ], + "score": 0.92, + "content": "\\nabla _ { x } f _ { \\theta } ( x ) = \\nabla _ { x } \\log { q _ { \\sigma } ( x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 231, + 506, + 247 + ], + "score": 1.0, + "content": ". To", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 244, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 505, + 257 + ], + "score": 1.0, + "content": "estimate likelihoods or sample from our model, we must rely on approximate methods, such as", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 256, + 504, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 256, + 504, + 267 + ], + "score": 1.0, + "content": "MCMC sampling or numerical ODE integration. MCMC works by simulating a Markov chain", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 266, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 160, + 280 + ], + "score": 1.0, + "content": "beginning at", + "type": "text" + }, + { + "bbox": [ + 160, + 267, + 208, + 279 + ], + "score": 0.93, + "content": "x _ { 0 } \\sim p ( x _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 266, + 350, + 280 + ], + "score": 1.0, + "content": "and using a transition distribution", + "type": "text" + }, + { + "bbox": [ + 350, + 266, + 419, + 279 + ], + "score": 0.92, + "content": "x _ { t } \\sim k ( \\bar { x } _ { t } | x _ { t - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 266, + 434, + 280 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 435, + 267, + 457, + 279 + ], + "score": 0.92, + "content": "k ( \\cdot | \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 266, + 505, + 280 + ], + "score": 1.0, + "content": "has certain", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 277, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 104, + 277, + 393, + 291 + ], + "score": 1.0, + "content": "properties, namely invariance w.r.t. the target and ergodicity, then as", + "type": "text" + }, + { + "bbox": [ + 394, + 279, + 426, + 288 + ], + "score": 0.87, + "content": "t \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 277, + 431, + 291 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 431, + 280, + 441, + 289 + ], + "score": 0.82, + "content": "x _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 277, + 506, + 291 + ], + "score": 1.0, + "content": "converges to a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 288, + 251, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 251, + 302 + ], + "score": 1.0, + "content": "sample from our target distribution.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5, + "bbox_fs": [ + 104, + 231, + 506, + 302 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 305, + 504, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 305, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 505, + 318 + ], + "score": 1.0, + "content": "Perhaps the most popular MCMC sampling algorithm for EBMs is Unadjusted Langevin Dynamics", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 315, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 505, + 330 + ], + "score": 1.0, + "content": "(ULA) (Roberts & Tweedie, 1996; Du & Mordatch, 2019; Nijkamp et al., 2020) which is defined by", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 305, + 505, + 330 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 333, + 406, + 353 + ], + "lines": [ + { + "bbox": [ + 205, + 333, + 406, + 353 + ], + "spans": [ + { + "bbox": [ + 205, + 333, + 406, + 353 + ], + "score": 0.92, + "content": "\\begin{array} { r } { k ( x _ { t } | x _ { t - 1 } ) = N \\left( x _ { t } ; x _ { t - 1 } + \\frac { \\sigma ^ { 2 } } { 2 } \\nabla _ { x } f _ { \\theta } ( x _ { t - 1 } ) , \\sigma ^ { 2 } I \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "113e4260d357d7ff1b47335eea991e588075f78cdb8da7cb5b7d79d90a6dfd24.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 205, + 333, + 406, + 353 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 360, + 505, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 359, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 331, + 375 + ], + "score": 1.0, + "content": "This resembles a step of gradient ascent (with step-size", + "type": "text" + }, + { + "bbox": [ + 331, + 359, + 343, + 375 + ], + "score": 0.86, + "content": "\\frac { \\sigma ^ { 2 } } { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 360, + 506, + 375 + ], + "score": 1.0, + "content": ") with added Gaussian noise of variance", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 372, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 107, + 374, + 118, + 385 + ], + "score": 0.84, + "content": "\\sigma ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 372, + 505, + 388 + ], + "score": 1.0, + "content": ". This transition is based on a discretization of the Langevin SDE. In the limit of infinitesimally", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 384, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 104, + 384, + 131, + 399 + ], + "score": 1.0, + "content": "small", + "type": "text" + }, + { + "bbox": [ + 131, + 387, + 138, + 395 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 384, + 505, + 399 + ], + "score": 1.0, + "content": "this approach will draw exact samples. To handle the error accrued when using larger step", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 396, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 409 + ], + "score": 1.0, + "content": "sizes, a Metropolis correction can be added giving the Metropolis-Adjusted-Langevin-Algorithm", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 406, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 484, + 420 + ], + "score": 1.0, + "content": "(MALA) (Besag, 1994). With Metropolis correction, we first generate a proposed update", + "type": "text" + }, + { + "bbox": [ + 484, + 408, + 505, + 418 + ], + "score": 0.84, + "content": "\\hat { x } \\sim", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 418, + 502, + 438 + ], + "spans": [ + { + "bbox": [ + 107, + 421, + 148, + 434 + ], + "score": 0.91, + "content": "k ( x | x _ { t - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 421, + 239, + 435 + ], + "score": 1.0, + "content": ", then with probability", + "type": "text" + }, + { + "bbox": [ + 239, + 418, + 351, + 438 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\operatorname* { m i n } \\left( 1 , \\frac { e ^ { f _ { \\theta } ( \\hat { x } ) } } { e ^ { f _ { \\theta } ( x _ { t - 1 } ) } } \\frac { k \\left( x _ { t - 1 } | \\hat { x } \\right) } { k \\left( \\hat { x } | x _ { t - 1 } \\right) } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 421, + 380, + 435 + ], + "score": 1.0, + "content": "we set", + "type": "text" + }, + { + "bbox": [ + 380, + 422, + 410, + 433 + ], + "score": 0.89, + "content": "\\boldsymbol { x } _ { t } = \\boldsymbol { \\hat { x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 421, + 455, + 435 + ], + "score": 1.0, + "content": ", otherwise", + "type": "text" + }, + { + "bbox": [ + 455, + 424, + 497, + 434 + ], + "score": 0.86, + "content": "x _ { t } = x _ { t - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 421, + 502, + 435 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 104, + 359, + 506, + 438 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 441, + 506, + 498 + ], + "lines": [ + { + "bbox": [ + 106, + 442, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 505, + 454 + ], + "score": 1.0, + "content": "Hamiltonian Monte Carlo (HMC) (Duane et al., 1987; Neal, 1996) is a more advanced MCMC", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "score": 1.0, + "content": "sampling method which augments the state-space with auxiliary momentum variables and numerically", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "integrates energy-conserving Hamiltonian dynamics to advance the sampler. HMC is typically applied", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 475, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 506, + 487 + ], + "score": 1.0, + "content": "with a Metropolis correction, but an approximate variant can be used without it (U-HMC) (Geffner &", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 485, + 389, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 389, + 500 + ], + "score": 1.0, + "content": "Domke, 2021). See Appendix C.1 for details of HMC variants we use.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 442, + 506, + 500 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 507, + 383, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 507, + 383, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 383, + 520 + ], + "score": 1.0, + "content": "2.3 RELATIONSHIP BETWEEN DIFFUSION MODELS AND EBMS", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 505, + 559 + ], + "lines": [ + { + "bbox": [ + 105, + 525, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 537 + ], + "score": 1.0, + "content": "Diffusion models and EBMs are closely related. For instance, Song & Ermon (2019) uses an EBM", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 535, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 104, + 535, + 506, + 551 + ], + "score": 1.0, + "content": "perspective to propose a close cousin to diffusion models. We can see from inspection that the training", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 547, + 504, + 560 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 504, + 560 + ], + "score": 1.0, + "content": "objective of diffusion models is identical (up to a constant) to the denoising score matching objective", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32, + "bbox_fs": [ + 104, + 525, + 506, + 560 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 177, + 564, + 434, + 586 + ], + "lines": [ + { + "bbox": [ + 177, + 564, + 434, + 586 + ], + "spans": [ + { + "bbox": [ + 177, + 564, + 434, + 586 + ], + "score": 0.92, + "content": "\\sigma _ { t } ^ { 2 } \\mathcal { I } _ { \\sigma _ { t } } ( \\theta ) = \\mathbb { E } _ { q ( x ) \\mathcal { N } ( \\epsilon ; 0 , I ) } \\left[ | | \\epsilon + \\sigma _ { t } \\nabla _ { x } f _ { \\theta } ( x + \\sigma _ { t } \\epsilon ) | | ^ { 2 } \\right] = \\mathcal { L } _ { t } ( \\theta )", + "type": "interline_equation", + "image_path": "19cf0a67d18e3205da24d1fac2da27f1bd314de81b22e0a33e4f779af41543da.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 177, + 564, + 434, + 586 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 590, + 505, + 652 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 208, + 604 + ], + "score": 1.0, + "content": "where we have replaced", + "type": "text" + }, + { + "bbox": [ + 208, + 591, + 239, + 603 + ], + "score": 0.93, + "content": "\\epsilon _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 590, + 262, + 604 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 263, + 591, + 343, + 604 + ], + "score": 0.91, + "content": "- \\sigma _ { t } \\nabla _ { x } f _ { \\theta } ( x + \\sigma _ { t } \\epsilon )", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 590, + 420, + 604 + ], + "score": 1.0, + "content": ". Thus by training", + "type": "text" + }, + { + "bbox": [ + 421, + 591, + 451, + 603 + ], + "score": 0.91, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 590, + 505, + 604 + ], + "score": 1.0, + "content": "to minimize", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 602, + 504, + 619 + ], + "spans": [ + { + "bbox": [ + 104, + 602, + 481, + 619 + ], + "score": 1.0, + "content": "Equation 2, we can recover the diffused data distribution score with ∇x log qσ(x) ≈ − ϵθ(x,t)σ .", + "type": "text" + }, + { + "bbox": [ + 476, + 604, + 504, + 616 + ], + "score": 1.0, + "content": "From", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 617, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 184, + 631 + ], + "score": 1.0, + "content": "this, we can define", + "type": "text" + }, + { + "bbox": [ + 184, + 617, + 272, + 630 + ], + "score": 0.95, + "content": "\\epsilon _ { \\theta } ( x , t ) = \\nabla _ { x } f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 617, + 506, + 631 + ], + "score": 1.0, + "content": "(the derivative of an explicitly defined scalar function) to", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 280, + 641 + ], + "score": 1.0, + "content": "learn a noise-conditional potential function", + "type": "text" + }, + { + "bbox": [ + 280, + 629, + 311, + 641 + ], + "score": 0.93, + "content": "f _ { \\boldsymbol { \\theta } } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 629, + 505, + 641 + ], + "score": 1.0, + "content": ". We later demonstrate the benefits of this in two", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "score": 1.0, + "content": "ways; it enables the use of more sophisticated sampling algorithms and more forms of composition.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37, + "bbox_fs": [ + 104, + 590, + 506, + 652 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 661, + 261, + 673 + ], + "lines": [ + { + "bbox": [ + 106, + 662, + 262, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 262, + 673 + ], + "score": 1.0, + "content": "2.4 CONTROLLABLE GENERATION", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 678, + 505, + 713 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 276, + 692 + ], + "score": 1.0, + "content": "It may be convenient to train a model of", + "type": "text" + }, + { + "bbox": [ + 276, + 679, + 296, + 691 + ], + "score": 0.92, + "content": "p ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 678, + 325, + 692 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 325, + 681, + 332, + 689 + ], + "score": 0.78, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 678, + 506, + 692 + ], + "score": 1.0, + "content": "is, say, the distribution of all images, but", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 325, + 703 + ], + "score": 1.0, + "content": "in practice we often want to generate samples from", + "type": "text" + }, + { + "bbox": [ + 326, + 690, + 353, + 702 + ], + "score": 0.92, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 690, + 384, + 703 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 384, + 692, + 391, + 702 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 690, + 505, + 703 + ], + "score": 1.0, + "content": "is some attribute, label, or", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "feature. This can be accomplished within the framework of diffusion models by introducing a learned", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 678, + 506, + 713 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 176, + 96 + ], + "score": 1.0, + "content": "predictive model", + "type": "text" + }, + { + "bbox": [ + 176, + 82, + 216, + 95 + ], + "score": 0.92, + "content": "p _ { \\theta } ( y | x ; t )", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 81, + 473, + 96 + ], + "score": 1.0, + "content": ", i.e a time-conditional model of the distribution of some feature", + "type": "text" + }, + { + "bbox": [ + 473, + 85, + 480, + 94 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "given", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 93, + 352, + 106 + ], + "spans": [ + { + "bbox": [ + 107, + 98, + 113, + 104 + ], + "score": 0.68, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 93, + 319, + 106 + ], + "score": 1.0, + "content": ". We can then exploit Bayes’ rule to notice that (for", + "type": "text" + }, + { + "bbox": [ + 320, + 94, + 345, + 104 + ], + "score": 0.89, + "content": "\\lambda = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 93, + 352, + 106 + ], + "score": 1.0, + "content": "),", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 188, + 110, + 422, + 124 + ], + "lines": [ + { + "bbox": [ + 188, + 110, + 422, + 124 + ], + "spans": [ + { + "bbox": [ + 188, + 110, + 422, + 124 + ], + "score": 0.9, + "content": "\\nabla _ { x } \\log p _ { \\theta } ( x | y ; t ) = \\nabla _ { x } \\log p _ { \\theta } ( x ; t ) + \\lambda \\nabla _ { x } \\log p _ { \\theta } ( y | x ; t ) .", + "type": "interline_equation", + "image_path": "2d61799ec2213b965cea5aa84fb50d4d3ccdc2dc6badc72b0a58fe1ba4bf4cb7.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 188, + 110, + 422, + 124 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 129, + 504, + 163 + ], + "lines": [ + { + "bbox": [ + 106, + 130, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 106, + 130, + 505, + 142 + ], + "score": 1.0, + "content": "In practice, when using the right side of Equation 6 for sampling, it is beneficial to increase the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 141, + 505, + 153 + ], + "spans": [ + { + "bbox": [ + 107, + 141, + 172, + 153 + ], + "score": 1.0, + "content": "‘guidance scale’", + "type": "text" + }, + { + "bbox": [ + 172, + 142, + 179, + 151 + ], + "score": 0.82, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 141, + 202, + 153 + ], + "score": 1.0, + "content": "to be", + "type": "text" + }, + { + "bbox": [ + 202, + 142, + 219, + 151 + ], + "score": 0.77, + "content": "> 1", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 141, + 505, + 153 + ], + "score": 1.0, + "content": "(Dhariwal & Nichol, 2021). Thus, we can re-purpose the unconditional", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 152, + 318, + 163 + ], + "spans": [ + { + "bbox": [ + 106, + 152, + 318, + 163 + ], + "score": 1.0, + "content": "diffusion model and turn it into a conditional model.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 168, + 505, + 202 + ], + "lines": [ + { + "bbox": [ + 105, + 168, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 416, + 182 + ], + "score": 1.0, + "content": "If instead of a classifier, we have a both an unconditional diffusion model", + "type": "text" + }, + { + "bbox": [ + 416, + 169, + 478, + 181 + ], + "score": 0.93, + "content": "\\nabla _ { x } \\log p _ { \\theta } ( x ; t )", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 168, + 506, + 182 + ], + "score": 1.0, + "content": "and a", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 219, + 193 + ], + "score": 1.0, + "content": "conditional diffusion model", + "type": "text" + }, + { + "bbox": [ + 219, + 180, + 289, + 192 + ], + "score": 0.93, + "content": "\\nabla _ { x } \\log p _ { \\theta } ( x | y ; t )", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 180, + 506, + 193 + ], + "score": 1.0, + "content": ", we can again utilize Bayes’ rule to derive an implicit", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 191, + 222, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 222, + 203 + ], + "score": 1.0, + "content": "predictive model’s gradients", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 208, + 418, + 222 + ], + "lines": [ + { + "bbox": [ + 192, + 208, + 418, + 222 + ], + "spans": [ + { + "bbox": [ + 192, + 208, + 418, + 222 + ], + "score": 0.9, + "content": "\\nabla _ { x } \\log p _ { \\theta } ( y | x ; t ) = \\nabla _ { x } \\log p _ { \\theta } ( x | y ; t ) - \\nabla _ { x } \\log p _ { \\theta } ( x ; t )", + "type": "interline_equation", + "image_path": "c19db12076954632df487316e74642c08fdbe9f83c5c0a9daceaa1f17c518ab7.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 192, + 208, + 418, + 222 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 227, + 505, + 338 + ], + "lines": [ + { + "bbox": [ + 106, + 228, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 505, + 240 + ], + "score": 1.0, + "content": "which can be used to replace the explicit model in Equation 6, giving what is known as classifier-free", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 238, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 506, + 252 + ], + "score": 1.0, + "content": "guidance (Ho & Salimans, 2022). This method has led to incredible performance, but comes at a", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "cost to modularity. This contrasts with the classifier-guidance setting, where we only need to train", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 260, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 506, + 273 + ], + "score": 1.0, + "content": "a single (costly) generative model. We can then attach any predictive model we would like to for", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 271, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 506, + 284 + ], + "score": 1.0, + "content": "conditioning. This is beneficial as it is often much easier and cheaper to train predictive models than a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 439, + 295 + ], + "score": 1.0, + "content": "flexible generative model. In the classifier-free setting, we must know exactly which", + "type": "text" + }, + { + "bbox": [ + 439, + 284, + 446, + 294 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 281, + 506, + 295 + ], + "score": 1.0, + "content": "we would like", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "score": 1.0, + "content": "to condition on, and incorporate these labels into model training. In both guidance settings, we use", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 305, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 305, + 505, + 316 + ], + "score": 1.0, + "content": "our (possibly implicit) predictive model to modify the learned score of our model. We then perform", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 315, + 506, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 506, + 328 + ], + "score": 1.0, + "content": "diffusion model sampling as we would in the unconditional setting. We will see later that even in toy", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 327, + 301, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 301, + 338 + ], + "score": 1.0, + "content": "settings, this is often not the optimal thing to do.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 108, + 350, + 392, + 363 + ], + "lines": [ + { + "bbox": [ + 104, + 349, + 393, + 366 + ], + "spans": [ + { + "bbox": [ + 104, + 349, + 393, + 366 + ], + "score": 1.0, + "content": "3 COMPOSITIONAL GENERATION BEYOND GUIDANCE", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 373, + 506, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "score": 1.0, + "content": "Most work on conditional diffusion models has come in the form of classifier or classifier-free", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 385, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 506, + 397 + ], + "score": 1.0, + "content": "guidance, but these are far from the only ways we can compose distributions to obtain new models.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "These ideas have been studied primarily in the context of EBMs because most compositional operators", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 407, + 428, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 428, + 419 + ], + "score": 1.0, + "content": "leave the resulting distribution unnormalized. We outline various options below.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 105, + 423, + 503, + 446 + ], + "lines": [ + { + "bbox": [ + 106, + 423, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 261, + 435 + ], + "score": 1.0, + "content": "Products: We can take a product of", + "type": "text" + }, + { + "bbox": [ + 262, + 424, + 272, + 433 + ], + "score": 0.82, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 423, + 505, + 435 + ], + "score": 1.0, + "content": "distributions and re-normalize to create a new distribution,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 434, + 390, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 390, + 447 + ], + "score": 1.0, + "content": "roughly equivalent to the “intersection” of the composite distributions,", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 194, + 446, + 417, + 464 + ], + "lines": [ + { + "bbox": [ + 194, + 446, + 417, + 464 + ], + "spans": [ + { + "bbox": [ + 194, + 446, + 417, + 464 + ], + "score": 0.9, + "content": "\\begin{array} { r } { q ^ { \\mathrm { p r o d } } ( x ) = \\frac { 1 } { Z } \\prod _ { i = 1 } ^ { N } q ^ { i } ( x ) , \\qquad Z = \\int \\prod _ { i = 1 } ^ { N } q ^ { i } ( x ) d x . } \\end{array}", + "type": "interline_equation", + "image_path": "40a824ac23ec00bffb84e71b84eee7fda94557895d18f538f68809f327fd7c3a.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 194, + 446, + 417, + 464 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 469, + 505, + 525 + ], + "lines": [ + { + "bbox": [ + 105, + 468, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 247, + 483 + ], + "score": 1.0, + "content": "Regions of high probability under", + "type": "text" + }, + { + "bbox": [ + 248, + 469, + 280, + 482 + ], + "score": 0.93, + "content": "q ^ { \\mathrm { p r o d } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 468, + 468, + 483 + ], + "score": 1.0, + "content": "will typically have high probability under all", + "type": "text" + }, + { + "bbox": [ + 468, + 469, + 490, + 482 + ], + "score": 0.91, + "content": "q ^ { i } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 468, + 505, + 483 + ], + "score": 1.0, + "content": ". A", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 480, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 495 + ], + "score": 1.0, + "content": "simple product model can be seen in Figure 2. These ideas were initially proposed to increase", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 492, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 505 + ], + "score": 1.0, + "content": "the capacity of weaker models by allowing individual “experts” to model specific features in the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 502, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 518 + ], + "score": 1.0, + "content": "input (Hinton, 2002), and were recently demonstrated at scale in the image domain using Deep", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 514, + 272, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 272, + 527 + ], + "score": 1.0, + "content": "Energy-Based Models (Du et al., 2020a).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 531, + 505, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 530, + 507, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 507, + 544 + ], + "score": 1.0, + "content": "The approaches to guidance discussed in Section 2.4 define product models with only two experts.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "The first models the relative density of the input data and the second models the conditional probability", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 552, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 118, + 566 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 555, + 124, + 565 + ], + "score": 0.75, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 552, + 473, + 566 + ], + "score": 1.0, + "content": ". Combining these by a product models likely inputs which have the desired property", + "type": "text" + }, + { + "bbox": [ + 473, + 555, + 479, + 565 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 552, + 506, + 566 + ], + "score": 1.0, + "content": ". This", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 563, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 505, + 577 + ], + "score": 1.0, + "content": "form of composition has become popular for diffusion models since they do not directly model the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 574, + 503, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 503, + 589 + ], + "score": 1.0, + "content": "probability, but instead the gradient of the log-probability which can also be composed in this way.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 592, + 505, + 625 + ], + "lines": [ + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "score": 1.0, + "content": "Mixtures: Complementary to the product or intersection is the mixture or union of multiple", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 240, + 615 + ], + "score": 1.0, + "content": "distributions. We can combine", + "type": "text" + }, + { + "bbox": [ + 241, + 603, + 251, + 613 + ], + "score": 0.78, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "distributions through a mixture to create a new distribution", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 614, + 379, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 379, + 626 + ], + "score": 1.0, + "content": "equivalent to the union of the concepts captured in each distribution", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 626, + 356, + 642 + ], + "lines": [ + { + "bbox": [ + 254, + 626, + 356, + 642 + ], + "spans": [ + { + "bbox": [ + 254, + 626, + 356, + 642 + ], + "score": 0.92, + "content": "\\begin{array} { r } { q ^ { \\mathrm { m i x } } ( x ) = \\frac { 1 } { N } \\sum _ { i = 1 } ^ { N } q ^ { i } ( x ) } \\end{array}", + "type": "interline_equation", + "image_path": "70997b5cc7cb83498d865d3735905da06b0a1bf1894f6b1cdc7868a5ee834879.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 254, + 626, + 356, + 642 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 433, + 662 + ], + "score": 1.0, + "content": "where regions of high probability consist of regions of high probability under any", + "type": "text" + }, + { + "bbox": [ + 433, + 648, + 456, + 661 + ], + "score": 0.92, + "content": "q ^ { i } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 648, + 505, + 662 + ], + "score": 1.0, + "content": ". We cannot", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "compose score-functions to define mixtures (unlike products). Instead, we need a model which", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 670, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 506, + 684 + ], + "score": 1.0, + "content": "specifies probability. Generating from mixtures of energy based models requires knowing the ratio", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 682, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 505, + 694 + ], + "score": 1.0, + "content": "of normalizers between the models. In our experiments, we assume this ratio is 1. A simple", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 692, + 323, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 323, + 705 + ], + "score": 1.0, + "content": "compositional mixture model can be seen in Figure 2.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 105, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 285, + 723 + ], + "score": 1.0, + "content": "Negation: Finally, given two distributions", + "type": "text" + }, + { + "bbox": [ + 285, + 710, + 309, + 722 + ], + "score": 0.92, + "content": "p _ { 0 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 708, + 327, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 327, + 710, + 351, + 722 + ], + "score": 0.91, + "content": "p _ { 1 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 708, + 506, + 723 + ], + "score": 1.0, + "content": ", we can explicitly invert the density of", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 131, + 732 + ], + "score": 0.91, + "content": "p _ { 1 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 719, + 194, + 734 + ], + "score": 1.0, + "content": "with respect to", + "type": "text" + }, + { + "bbox": [ + 195, + 721, + 219, + 732 + ], + "score": 0.92, + "content": "p _ { 0 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 719, + 506, + 734 + ], + "score": 1.0, + "content": ", which constructs a new distribution which assigns high likelihood to", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 176, + 96 + ], + "score": 1.0, + "content": "predictive model", + "type": "text" + }, + { + "bbox": [ + 176, + 82, + 216, + 95 + ], + "score": 0.92, + "content": "p _ { \\theta } ( y | x ; t )", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 81, + 473, + 96 + ], + "score": 1.0, + "content": ", i.e a time-conditional model of the distribution of some feature", + "type": "text" + }, + { + "bbox": [ + 473, + 85, + 480, + 94 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "given", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 93, + 352, + 106 + ], + "spans": [ + { + "bbox": [ + 107, + 98, + 113, + 104 + ], + "score": 0.68, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 93, + 319, + 106 + ], + "score": 1.0, + "content": ". We can then exploit Bayes’ rule to notice that (for", + "type": "text" + }, + { + "bbox": [ + 320, + 94, + 345, + 104 + ], + "score": 0.89, + "content": "\\lambda = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 93, + 352, + 106 + ], + "score": 1.0, + "content": "),", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 81, + 506, + 106 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 188, + 110, + 422, + 124 + ], + "lines": [ + { + "bbox": [ + 188, + 110, + 422, + 124 + ], + "spans": [ + { + "bbox": [ + 188, + 110, + 422, + 124 + ], + "score": 0.9, + "content": "\\nabla _ { x } \\log p _ { \\theta } ( x | y ; t ) = \\nabla _ { x } \\log p _ { \\theta } ( x ; t ) + \\lambda \\nabla _ { x } \\log p _ { \\theta } ( y | x ; t ) .", + "type": "interline_equation", + "image_path": "2d61799ec2213b965cea5aa84fb50d4d3ccdc2dc6badc72b0a58fe1ba4bf4cb7.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 188, + 110, + 422, + 124 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 129, + 504, + 163 + ], + "lines": [ + { + "bbox": [ + 106, + 130, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 106, + 130, + 505, + 142 + ], + "score": 1.0, + "content": "In practice, when using the right side of Equation 6 for sampling, it is beneficial to increase the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 141, + 505, + 153 + ], + "spans": [ + { + "bbox": [ + 107, + 141, + 172, + 153 + ], + "score": 1.0, + "content": "‘guidance scale’", + "type": "text" + }, + { + "bbox": [ + 172, + 142, + 179, + 151 + ], + "score": 0.82, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 141, + 202, + 153 + ], + "score": 1.0, + "content": "to be", + "type": "text" + }, + { + "bbox": [ + 202, + 142, + 219, + 151 + ], + "score": 0.77, + "content": "> 1", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 141, + 505, + 153 + ], + "score": 1.0, + "content": "(Dhariwal & Nichol, 2021). Thus, we can re-purpose the unconditional", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 152, + 318, + 163 + ], + "spans": [ + { + "bbox": [ + 106, + 152, + 318, + 163 + ], + "score": 1.0, + "content": "diffusion model and turn it into a conditional model.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 130, + 505, + 163 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 168, + 505, + 202 + ], + "lines": [ + { + "bbox": [ + 105, + 168, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 416, + 182 + ], + "score": 1.0, + "content": "If instead of a classifier, we have a both an unconditional diffusion model", + "type": "text" + }, + { + "bbox": [ + 416, + 169, + 478, + 181 + ], + "score": 0.93, + "content": "\\nabla _ { x } \\log p _ { \\theta } ( x ; t )", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 168, + 506, + 182 + ], + "score": 1.0, + "content": "and a", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 219, + 193 + ], + "score": 1.0, + "content": "conditional diffusion model", + "type": "text" + }, + { + "bbox": [ + 219, + 180, + 289, + 192 + ], + "score": 0.93, + "content": "\\nabla _ { x } \\log p _ { \\theta } ( x | y ; t )", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 180, + 506, + 193 + ], + "score": 1.0, + "content": ", we can again utilize Bayes’ rule to derive an implicit", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 191, + 222, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 222, + 203 + ], + "score": 1.0, + "content": "predictive model’s gradients", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 168, + 506, + 203 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 208, + 418, + 222 + ], + "lines": [ + { + "bbox": [ + 192, + 208, + 418, + 222 + ], + "spans": [ + { + "bbox": [ + 192, + 208, + 418, + 222 + ], + "score": 0.9, + "content": "\\nabla _ { x } \\log p _ { \\theta } ( y | x ; t ) = \\nabla _ { x } \\log p _ { \\theta } ( x | y ; t ) - \\nabla _ { x } \\log p _ { \\theta } ( x ; t )", + "type": "interline_equation", + "image_path": "c19db12076954632df487316e74642c08fdbe9f83c5c0a9daceaa1f17c518ab7.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 192, + 208, + 418, + 222 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 227, + 505, + 338 + ], + "lines": [ + { + "bbox": [ + 106, + 228, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 505, + 240 + ], + "score": 1.0, + "content": "which can be used to replace the explicit model in Equation 6, giving what is known as classifier-free", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 238, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 506, + 252 + ], + "score": 1.0, + "content": "guidance (Ho & Salimans, 2022). This method has led to incredible performance, but comes at a", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "cost to modularity. This contrasts with the classifier-guidance setting, where we only need to train", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 260, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 506, + 273 + ], + "score": 1.0, + "content": "a single (costly) generative model. We can then attach any predictive model we would like to for", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 271, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 506, + 284 + ], + "score": 1.0, + "content": "conditioning. This is beneficial as it is often much easier and cheaper to train predictive models than a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 439, + 295 + ], + "score": 1.0, + "content": "flexible generative model. In the classifier-free setting, we must know exactly which", + "type": "text" + }, + { + "bbox": [ + 439, + 284, + 446, + 294 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 281, + 506, + 295 + ], + "score": 1.0, + "content": "we would like", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "score": 1.0, + "content": "to condition on, and incorporate these labels into model training. In both guidance settings, we use", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 305, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 305, + 505, + 316 + ], + "score": 1.0, + "content": "our (possibly implicit) predictive model to modify the learned score of our model. We then perform", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 315, + 506, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 506, + 328 + ], + "score": 1.0, + "content": "diffusion model sampling as we would in the unconditional setting. We will see later that even in toy", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 327, + 301, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 301, + 338 + ], + "score": 1.0, + "content": "settings, this is often not the optimal thing to do.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 228, + 506, + 338 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 350, + 392, + 363 + ], + "lines": [ + { + "bbox": [ + 104, + 349, + 393, + 366 + ], + "spans": [ + { + "bbox": [ + 104, + 349, + 393, + 366 + ], + "score": 1.0, + "content": "3 COMPOSITIONAL GENERATION BEYOND GUIDANCE", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 373, + 506, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "score": 1.0, + "content": "Most work on conditional diffusion models has come in the form of classifier or classifier-free", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 385, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 506, + 397 + ], + "score": 1.0, + "content": "guidance, but these are far from the only ways we can compose distributions to obtain new models.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "These ideas have been studied primarily in the context of EBMs because most compositional operators", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 407, + 428, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 428, + 419 + ], + "score": 1.0, + "content": "leave the resulting distribution unnormalized. We outline various options below.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 374, + 506, + 419 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 423, + 503, + 446 + ], + "lines": [ + { + "bbox": [ + 106, + 423, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 261, + 435 + ], + "score": 1.0, + "content": "Products: We can take a product of", + "type": "text" + }, + { + "bbox": [ + 262, + 424, + 272, + 433 + ], + "score": 0.82, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 423, + 505, + 435 + ], + "score": 1.0, + "content": "distributions and re-normalize to create a new distribution,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 434, + 390, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 390, + 447 + ], + "score": 1.0, + "content": "roughly equivalent to the “intersection” of the composite distributions,", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 423, + 505, + 447 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 194, + 446, + 417, + 464 + ], + "lines": [ + { + "bbox": [ + 194, + 446, + 417, + 464 + ], + "spans": [ + { + "bbox": [ + 194, + 446, + 417, + 464 + ], + "score": 0.9, + "content": "\\begin{array} { r } { q ^ { \\mathrm { p r o d } } ( x ) = \\frac { 1 } { Z } \\prod _ { i = 1 } ^ { N } q ^ { i } ( x ) , \\qquad Z = \\int \\prod _ { i = 1 } ^ { N } q ^ { i } ( x ) d x . } \\end{array}", + "type": "interline_equation", + "image_path": "40a824ac23ec00bffb84e71b84eee7fda94557895d18f538f68809f327fd7c3a.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 194, + 446, + 417, + 464 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 469, + 505, + 525 + ], + "lines": [ + { + "bbox": [ + 105, + 468, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 247, + 483 + ], + "score": 1.0, + "content": "Regions of high probability under", + "type": "text" + }, + { + "bbox": [ + 248, + 469, + 280, + 482 + ], + "score": 0.93, + "content": "q ^ { \\mathrm { p r o d } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 468, + 468, + 483 + ], + "score": 1.0, + "content": "will typically have high probability under all", + "type": "text" + }, + { + "bbox": [ + 468, + 469, + 490, + 482 + ], + "score": 0.91, + "content": "q ^ { i } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 468, + 505, + 483 + ], + "score": 1.0, + "content": ". A", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 480, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 495 + ], + "score": 1.0, + "content": "simple product model can be seen in Figure 2. These ideas were initially proposed to increase", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 492, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 505 + ], + "score": 1.0, + "content": "the capacity of weaker models by allowing individual “experts” to model specific features in the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 502, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 518 + ], + "score": 1.0, + "content": "input (Hinton, 2002), and were recently demonstrated at scale in the image domain using Deep", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 514, + 272, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 272, + 527 + ], + "score": 1.0, + "content": "Energy-Based Models (Du et al., 2020a).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 468, + 505, + 527 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 531, + 505, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 530, + 507, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 507, + 544 + ], + "score": 1.0, + "content": "The approaches to guidance discussed in Section 2.4 define product models with only two experts.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "The first models the relative density of the input data and the second models the conditional probability", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 552, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 118, + 566 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 555, + 124, + 565 + ], + "score": 0.75, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 552, + 473, + 566 + ], + "score": 1.0, + "content": ". Combining these by a product models likely inputs which have the desired property", + "type": "text" + }, + { + "bbox": [ + 473, + 555, + 479, + 565 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 552, + 506, + 566 + ], + "score": 1.0, + "content": ". This", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 563, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 505, + 577 + ], + "score": 1.0, + "content": "form of composition has become popular for diffusion models since they do not directly model the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 574, + 503, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 503, + 589 + ], + "score": 1.0, + "content": "probability, but instead the gradient of the log-probability which can also be composed in this way.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 530, + 507, + 589 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 592, + 505, + 625 + ], + "lines": [ + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "score": 1.0, + "content": "Mixtures: Complementary to the product or intersection is the mixture or union of multiple", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 240, + 615 + ], + "score": 1.0, + "content": "distributions. We can combine", + "type": "text" + }, + { + "bbox": [ + 241, + 603, + 251, + 613 + ], + "score": 0.78, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "distributions through a mixture to create a new distribution", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 614, + 379, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 379, + 626 + ], + "score": 1.0, + "content": "equivalent to the union of the concepts captured in each distribution", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 592, + 505, + 626 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 626, + 356, + 642 + ], + "lines": [ + { + "bbox": [ + 254, + 626, + 356, + 642 + ], + "spans": [ + { + "bbox": [ + 254, + 626, + 356, + 642 + ], + "score": 0.92, + "content": "\\begin{array} { r } { q ^ { \\mathrm { m i x } } ( x ) = \\frac { 1 } { N } \\sum _ { i = 1 } ^ { N } q ^ { i } ( x ) } \\end{array}", + "type": "interline_equation", + "image_path": "70997b5cc7cb83498d865d3735905da06b0a1bf1894f6b1cdc7868a5ee834879.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 254, + 626, + 356, + 642 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 433, + 662 + ], + "score": 1.0, + "content": "where regions of high probability consist of regions of high probability under any", + "type": "text" + }, + { + "bbox": [ + 433, + 648, + 456, + 661 + ], + "score": 0.92, + "content": "q ^ { i } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 648, + 505, + 662 + ], + "score": 1.0, + "content": ". We cannot", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "compose score-functions to define mixtures (unlike products). Instead, we need a model which", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 670, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 506, + 684 + ], + "score": 1.0, + "content": "specifies probability. Generating from mixtures of energy based models requires knowing the ratio", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 682, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 505, + 694 + ], + "score": 1.0, + "content": "of normalizers between the models. In our experiments, we assume this ratio is 1. A simple", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 692, + 323, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 323, + 705 + ], + "score": 1.0, + "content": "compositional mixture model can be seen in Figure 2.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 648, + 506, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 285, + 723 + ], + "score": 1.0, + "content": "Negation: Finally, given two distributions", + "type": "text" + }, + { + "bbox": [ + 285, + 710, + 309, + 722 + ], + "score": 0.92, + "content": "p _ { 0 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 708, + 327, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 327, + 710, + 351, + 722 + ], + "score": 0.91, + "content": "p _ { 1 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 708, + 506, + 723 + ], + "score": 1.0, + "content": ", we can explicitly invert the density of", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 131, + 732 + ], + "score": 0.91, + "content": "p _ { 1 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 719, + 194, + 734 + ], + "score": 1.0, + "content": "with respect to", + "type": "text" + }, + { + "bbox": [ + 195, + 721, + 219, + 732 + ], + "score": 0.92, + "content": "p _ { 0 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 719, + 506, + 734 + ], + "score": 1.0, + "content": ", which constructs a new distribution which assigns high likelihood to", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 200, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 143, + 216 + ], + "score": 1.0, + "content": "points in", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 144, + 202, + 168, + 214 + ], + "score": 0.91, + "content": "p _ { 0 } ( x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 168, + 200, + 226, + 216 + ], + "score": 1.0, + "content": "that are not in", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 227, + 202, + 251, + 214 + ], + "score": 0.92, + "content": "p _ { 1 } ( x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 251, + 200, + 354, + 216 + ], + "score": 1.0, + "content": "(Du et al., 2020a), where", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 354, + 204, + 362, + 212 + ], + "score": 0.76, + "content": "\\alpha", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 362, + 200, + 480, + 216 + ], + "score": 1.0, + "content": "controls the degree we invert", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 481, + 202, + 505, + 214 + ], + "score": 0.91, + "content": "p _ { 1 } ( x )", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 213, + 257, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 139, + 225 + ], + "score": 1.0, + "content": "(we use", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 139, + 214, + 173, + 223 + ], + "score": 0.88, + "content": "\\alpha = 0 . 5", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 173, + 213, + 257, + 225 + ], + "score": 1.0, + "content": "in our experiments).", + "type": "text", + "cross_page": true + } + ], + "index": 8 + } + ], + "index": 47.5, + "bbox_fs": [ + 105, + 708, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 165, + 73, + 447, + 152 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 165, + 73, + 447, + 152 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 165, + 73, + 447, + 152 + ], + "spans": [ + { + "bbox": [ + 165, + 73, + 447, + 152 + ], + "score": 0.968, + "type": "image", + "image_path": "3b058e38c0a52dd0c4de8b4c338a59265fea6f03565646084366161750c0614b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 165, + 73, + 447, + 99.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 165, + 99.33333333333333, + 447, + 125.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 165, + 125.66666666666666, + 447, + 152.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 160, + 506, + 200 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "score": 1.0, + "content": "Figure 2: An illustration of product and mixture compositional models, and the improved sampling", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 169, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 505, + 181 + ], + "score": 1.0, + "content": "performance of MCMC in both cases. Left to right: Component distributions, ground truth composed", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 180, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 180, + 505, + 191 + ], + "score": 1.0, + "content": "distribution, reverse diffusion samples, HMC samples. Top: product, bottom: mixture. Reverse diffusion fails to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 190, + 223, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 223, + 201 + ], + "score": 1.0, + "content": "sample from composed models.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 104, + 202, + 504, + 223 + ], + "lines": [ + { + "bbox": [ + 105, + 200, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 143, + 216 + ], + "score": 1.0, + "content": "points in", + "type": "text" + }, + { + "bbox": [ + 144, + 202, + 168, + 214 + ], + "score": 0.91, + "content": "p _ { 0 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 200, + 226, + 216 + ], + "score": 1.0, + "content": "that are not in", + "type": "text" + }, + { + "bbox": [ + 227, + 202, + 251, + 214 + ], + "score": 0.92, + "content": "p _ { 1 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 200, + 354, + 216 + ], + "score": 1.0, + "content": "(Du et al., 2020a), where", + "type": "text" + }, + { + "bbox": [ + 354, + 204, + 362, + 212 + ], + "score": 0.76, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 200, + 480, + 216 + ], + "score": 1.0, + "content": "controls the degree we invert", + "type": "text" + }, + { + "bbox": [ + 481, + 202, + 505, + 214 + ], + "score": 0.91, + "content": "p _ { 1 } ( x )", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 213, + 257, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 139, + 225 + ], + "score": 1.0, + "content": "(we use", + "type": "text" + }, + { + "bbox": [ + 139, + 214, + 173, + 223 + ], + "score": 0.88, + "content": "\\alpha = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 213, + 257, + 225 + ], + "score": 1.0, + "content": "in our experiments).", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "interline_equation", + "bbox": [ + 268, + 222, + 342, + 241 + ], + "lines": [ + { + "bbox": [ + 268, + 222, + 342, + 241 + ], + "spans": [ + { + "bbox": [ + 268, + 222, + 342, + 241 + ], + "score": 0.93, + "content": "\\begin{array} { r } { q ^ { \\mathrm { n e g } } ( x ) \\propto \\frac { q ^ { 0 } ( x ) } { q ^ { 1 } ( x ) ^ { \\alpha } } . } \\end{array}", + "type": "interline_equation", + "image_path": "5b9f022991ccae73fab62deacd939af281eb5e7fc60c8578f0ca55db5904e5fe.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 268, + 222, + 342, + 241 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 244, + 505, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "score": 1.0, + "content": "We can combine negation with our previous operators, in a nested manner to construct complex", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 254, + 273, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 273, + 267 + ], + "score": 1.0, + "content": "combinations of distributions (Figure 5).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 271, + 505, + 316 + ], + "lines": [ + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "In section 2.3, we showed how diffusion models can be interpreted as approximating the gradient", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 281, + 506, + 296 + ], + "spans": [ + { + "bbox": [ + 107, + 283, + 155, + 294 + ], + "score": 0.9, + "content": "\\nabla _ { x } \\log { q ( x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 281, + 382, + 296 + ], + "score": 1.0, + "content": ", but do not learn an explicit model of the log-likelihood", + "type": "text" + }, + { + "bbox": [ + 383, + 283, + 416, + 295 + ], + "score": 0.89, + "content": "\\log q ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 281, + 506, + 296 + ], + "score": 1.0, + "content": ". This means with the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 293, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 143, + 307 + ], + "score": 1.0, + "content": "standard", + "type": "text" + }, + { + "bbox": [ + 143, + 294, + 174, + 306 + ], + "score": 0.92, + "content": "\\epsilon _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 293, + 505, + 307 + ], + "score": 1.0, + "content": "-parameterization we can, in theory, utilize product and negation composition, but", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 305, + 209, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 209, + 317 + ], + "score": 1.0, + "content": "not mixture composition.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "title", + "bbox": [ + 105, + 326, + 470, + 339 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 472, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 472, + 340 + ], + "score": 1.0, + "content": "4 SCALING COMPOSITIONAL GENERATION WITH DIFFUSION MODELS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 348, + 505, + 426 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 505, + 361 + ], + "score": 1.0, + "content": "While highly compositional, EBMs present many challenges. The lack of a normalized probability", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "score": 1.0, + "content": "function makes training, evaluation, and sampling very difficult. Much progress has been made to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 371, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 506, + 383 + ], + "score": 1.0, + "content": "scale these models (Du & Mordatch, 2019; Nijkamp et al., 2020; Grathwohl et al., 2019; Du et al.,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "score": 1.0, + "content": "2020b; Grathwohl et al., 2021), but EBMs still lag behind other approaches in terms of efficiency and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "score": 1.0, + "content": "scalability. In contrast, diffusion models have demonstrated very impressive scalability. Fortuitously,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 404, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 505, + 415 + ], + "score": 1.0, + "content": "diffusion models have similarities to EBMs, such as their training objective and their score-based", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 415, + 409, + 427 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 409, + 427 + ], + "score": 1.0, + "content": "interpretation, which makes many forms of composition readily applicable.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 431, + 505, + 497 + ], + "lines": [ + { + "bbox": [ + 106, + 430, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 505, + 443 + ], + "score": 1.0, + "content": "Unfortunately, when two diffusion models are composed into, for example, a product model", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 439, + 507, + 457 + ], + "spans": [ + { + "bbox": [ + 107, + 442, + 202, + 454 + ], + "score": 0.93, + "content": "q ^ { \\mathrm { p r o d } } ( x ) \\propto \\dot { q ^ { 1 } } ( x ) q ^ { 2 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 439, + 507, + 457 + ], + "score": 1.0, + "content": ", issues arise if the model which reverses the diffusion uses a score esti-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 452, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 505, + 466 + ], + "score": 1.0, + "content": "mate obtained by adding the score estimates of the two models. We see in Figure 2 that composing", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 464, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 476 + ], + "score": 1.0, + "content": "two models in such a way leads indeed to sub-par samples. This is because to sample from this", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "score": 1.0, + "content": "product distribution using standard reverse diffusion (Song et al., 2021), one would need to compute", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 485, + 378, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 378, + 499 + ], + "score": 1.0, + "content": "instead the score of the diffused target product distribution given by", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5 + }, + { + "type": "interline_equation", + "bbox": [ + 199, + 497, + 412, + 511 + ], + "lines": [ + { + "bbox": [ + 199, + 497, + 412, + 511 + ], + "spans": [ + { + "bbox": [ + 199, + 497, + 412, + 511 + ], + "score": 0.87, + "content": "\\begin{array} { r } { \\nabla _ { x } \\log \\tilde { q } _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } ) = \\nabla _ { x } \\log \\left( \\int d x _ { 0 } q ^ { 1 } ( x _ { 0 } ) q ^ { 2 } ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "b65e152d5aeca559f6d6fd19735670d3459067a487312f225fd397ab87b0a019.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 199, + 497, + 412, + 511 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 492, + 525 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 493, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 122, + 528 + ], + "score": 1.0, + "content": "For", + "type": "text" + }, + { + "bbox": [ + 123, + 514, + 146, + 524 + ], + "score": 0.89, + "content": "t > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 512, + 493, + 528 + ], + "score": 1.0, + "content": ", this quantity is not equal to the sum of the scores of the two models which is given by", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 529, + 465, + 543 + ], + "lines": [ + { + "bbox": [ + 145, + 529, + 465, + 543 + ], + "spans": [ + { + "bbox": [ + 145, + 529, + 465, + 543 + ], + "score": 0.85, + "content": "\\begin{array} { r } { \\nabla _ { x } \\log q _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } ) = \\nabla _ { x } \\log \\left( \\int d x _ { 0 } q ^ { 1 } ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) \\right) + \\nabla _ { x } \\log \\left( \\int d x _ { 0 } q ^ { 2 } ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "24bf69b118ac9b51a6a200fc36441b770e26c58dd5d7bef98b58a45bee53ee7a.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 145, + 529, + 465, + 543 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 545, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 104, + 543, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 104, + 543, + 506, + 559 + ], + "score": 1.0, + "content": "Therefore, plugging the composed score function into the standard ancestral sampling procedure", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "score": 1.0, + "content": "discussed in Section 2.1, which we refer to as “reverse diffusion,” does not correspond to sampling", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "from the composed model, and thus reverse diffusion sampling will generate incorrect samples from", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 577, + 459, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 459, + 590 + ], + "score": 1.0, + "content": "composed distributions. This effect can be seen in Figure 2, with details in Appendix D.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 104, + 592, + 507, + 612 + ], + "spans": [ + { + "bbox": [ + 104, + 592, + 218, + 612 + ], + "score": 1.0, + "content": "The score of the distribution", + "type": "text" + }, + { + "bbox": [ + 218, + 596, + 254, + 609 + ], + "score": 0.92, + "content": "q _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 592, + 444, + 612 + ], + "score": 1.0, + "content": "in Equation 12 is easy to compute, unlike that of", + "type": "text" + }, + { + "bbox": [ + 445, + 596, + 482, + 609 + ], + "score": 0.92, + "content": "\\tilde { q } _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 592, + 507, + 612 + ], + "score": 1.0, + "content": "from", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 608, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 104, + 608, + 209, + 624 + ], + "score": 1.0, + "content": "Equation 11. In addition,", + "type": "text" + }, + { + "bbox": [ + 209, + 610, + 245, + 622 + ], + "score": 0.91, + "content": "q _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 608, + 506, + 624 + ], + "score": 1.0, + "content": "describes a sequence of distributions which smoothly interpolate", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 619, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 144, + 634 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 144, + 622, + 176, + 633 + ], + "score": 0.9, + "content": "q ^ { \\mathrm { p r o d } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 619, + 189, + 634 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 189, + 623, + 214, + 631 + ], + "score": 0.88, + "content": "t = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 619, + 234, + 634 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 235, + 622, + 267, + 633 + ], + "score": 0.92, + "content": "\\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 619, + 280, + 634 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 281, + 623, + 308, + 631 + ], + "score": 0.88, + "content": "t = T", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 619, + 506, + 634 + ], + "score": 1.0, + "content": ", though this sequence of distributions does not", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 631, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 505, + 644 + ], + "score": 1.0, + "content": "correspond to the distributions that result from the standard forward diffusion process described in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "score": 1.0, + "content": "Section 2.1, leading the reverse diffusion sampling to generate poor samples. We discuss how we", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 653, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 104, + 653, + 338, + 669 + ], + "score": 1.0, + "content": "may utilize MCMC samplers, which use our knowledge of", + "type": "text" + }, + { + "bbox": [ + 338, + 656, + 403, + 668 + ], + "score": 0.93, + "content": "\\nabla _ { x } \\log { q _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 653, + 506, + 669 + ], + "score": 1.0, + "content": ", to correctly sample from", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 104, + 664, + 453, + 684 + ], + "spans": [ + { + "bbox": [ + 104, + 664, + 211, + 684 + ], + "score": 1.0, + "content": "intermediate distributions", + "type": "text" + }, + { + "bbox": [ + 212, + 669, + 249, + 682 + ], + "score": 0.92, + "content": "\\tilde { q } _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 664, + 453, + 684 + ], + "score": 1.0, + "content": ", leading to accurate composed sample generation.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40 + }, + { + "type": "title", + "bbox": [ + 108, + 690, + 290, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 690, + 291, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 690, + 291, + 703 + ], + "score": 1.0, + "content": "4.1 IMPROVING SAMPLING WITH MCMC", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 205, + 723 + ], + "score": 1.0, + "content": "In order to sample from", + "type": "text" + }, + { + "bbox": [ + 206, + 711, + 237, + 722 + ], + "score": 0.92, + "content": "q ^ { \\mathrm { p r o d } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "using the combined score function from Equation 12, we can use", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "annealed MCMC sampling, described below in Algorithm 1. This method applies MCMC transition", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 165, + 73, + 447, + 152 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 165, + 73, + 447, + 152 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 165, + 73, + 447, + 152 + ], + "spans": [ + { + "bbox": [ + 165, + 73, + 447, + 152 + ], + "score": 0.968, + "type": "image", + "image_path": "3b058e38c0a52dd0c4de8b4c338a59265fea6f03565646084366161750c0614b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 165, + 73, + 447, + 99.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 165, + 99.33333333333333, + 447, + 125.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 165, + 125.66666666666666, + 447, + 152.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 160, + 506, + 200 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 505, + 173 + ], + "score": 1.0, + "content": "Figure 2: An illustration of product and mixture compositional models, and the improved sampling", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 169, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 505, + 181 + ], + "score": 1.0, + "content": "performance of MCMC in both cases. Left to right: Component distributions, ground truth composed", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 180, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 180, + 505, + 191 + ], + "score": 1.0, + "content": "distribution, reverse diffusion samples, HMC samples. Top: product, bottom: mixture. Reverse diffusion fails to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 190, + 223, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 223, + 201 + ], + "score": 1.0, + "content": "sample from composed models.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 104, + 202, + 504, + 223 + ], + "lines": [], + "index": 7.5, + "bbox_fs": [ + 105, + 200, + 505, + 225 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 268, + 222, + 342, + 241 + ], + "lines": [ + { + "bbox": [ + 268, + 222, + 342, + 241 + ], + "spans": [ + { + "bbox": [ + 268, + 222, + 342, + 241 + ], + "score": 0.93, + "content": "\\begin{array} { r } { q ^ { \\mathrm { n e g } } ( x ) \\propto \\frac { q ^ { 0 } ( x ) } { q ^ { 1 } ( x ) ^ { \\alpha } } . } \\end{array}", + "type": "interline_equation", + "image_path": "5b9f022991ccae73fab62deacd939af281eb5e7fc60c8578f0ca55db5904e5fe.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 268, + 222, + 342, + 241 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 244, + 505, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "score": 1.0, + "content": "We can combine negation with our previous operators, in a nested manner to construct complex", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 254, + 273, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 273, + 267 + ], + "score": 1.0, + "content": "combinations of distributions (Figure 5).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 106, + 244, + 505, + 267 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 271, + 505, + 316 + ], + "lines": [ + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "In section 2.3, we showed how diffusion models can be interpreted as approximating the gradient", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 281, + 506, + 296 + ], + "spans": [ + { + "bbox": [ + 107, + 283, + 155, + 294 + ], + "score": 0.9, + "content": "\\nabla _ { x } \\log { q ( x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 281, + 382, + 296 + ], + "score": 1.0, + "content": ", but do not learn an explicit model of the log-likelihood", + "type": "text" + }, + { + "bbox": [ + 383, + 283, + 416, + 295 + ], + "score": 0.89, + "content": "\\log q ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 281, + 506, + 296 + ], + "score": 1.0, + "content": ". This means with the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 293, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 143, + 307 + ], + "score": 1.0, + "content": "standard", + "type": "text" + }, + { + "bbox": [ + 143, + 294, + 174, + 306 + ], + "score": 0.92, + "content": "\\epsilon _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 293, + 505, + 307 + ], + "score": 1.0, + "content": "-parameterization we can, in theory, utilize product and negation composition, but", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 305, + 209, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 209, + 317 + ], + "score": 1.0, + "content": "not mixture composition.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 271, + 506, + 317 + ] + }, + { + "type": "title", + "bbox": [ + 105, + 326, + 470, + 339 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 472, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 472, + 340 + ], + "score": 1.0, + "content": "4 SCALING COMPOSITIONAL GENERATION WITH DIFFUSION MODELS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 348, + 505, + 426 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 505, + 361 + ], + "score": 1.0, + "content": "While highly compositional, EBMs present many challenges. The lack of a normalized probability", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "score": 1.0, + "content": "function makes training, evaluation, and sampling very difficult. Much progress has been made to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 371, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 506, + 383 + ], + "score": 1.0, + "content": "scale these models (Du & Mordatch, 2019; Nijkamp et al., 2020; Grathwohl et al., 2019; Du et al.,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "score": 1.0, + "content": "2020b; Grathwohl et al., 2021), but EBMs still lag behind other approaches in terms of efficiency and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "score": 1.0, + "content": "scalability. In contrast, diffusion models have demonstrated very impressive scalability. Fortuitously,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 404, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 505, + 415 + ], + "score": 1.0, + "content": "diffusion models have similarities to EBMs, such as their training objective and their score-based", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 415, + 409, + 427 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 409, + 427 + ], + "score": 1.0, + "content": "interpretation, which makes many forms of composition readily applicable.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 348, + 506, + 427 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 431, + 505, + 497 + ], + "lines": [ + { + "bbox": [ + 106, + 430, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 505, + 443 + ], + "score": 1.0, + "content": "Unfortunately, when two diffusion models are composed into, for example, a product model", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 439, + 507, + 457 + ], + "spans": [ + { + "bbox": [ + 107, + 442, + 202, + 454 + ], + "score": 0.93, + "content": "q ^ { \\mathrm { p r o d } } ( x ) \\propto \\dot { q ^ { 1 } } ( x ) q ^ { 2 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 439, + 507, + 457 + ], + "score": 1.0, + "content": ", issues arise if the model which reverses the diffusion uses a score esti-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 452, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 505, + 466 + ], + "score": 1.0, + "content": "mate obtained by adding the score estimates of the two models. We see in Figure 2 that composing", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 464, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 476 + ], + "score": 1.0, + "content": "two models in such a way leads indeed to sub-par samples. This is because to sample from this", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "score": 1.0, + "content": "product distribution using standard reverse diffusion (Song et al., 2021), one would need to compute", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 485, + 378, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 378, + 499 + ], + "score": 1.0, + "content": "instead the score of the diffused target product distribution given by", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 430, + 507, + 499 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 199, + 497, + 412, + 511 + ], + "lines": [ + { + "bbox": [ + 199, + 497, + 412, + 511 + ], + "spans": [ + { + "bbox": [ + 199, + 497, + 412, + 511 + ], + "score": 0.87, + "content": "\\begin{array} { r } { \\nabla _ { x } \\log \\tilde { q } _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } ) = \\nabla _ { x } \\log \\left( \\int d x _ { 0 } q ^ { 1 } ( x _ { 0 } ) q ^ { 2 } ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "b65e152d5aeca559f6d6fd19735670d3459067a487312f225fd397ab87b0a019.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 199, + 497, + 412, + 511 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 492, + 525 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 493, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 122, + 528 + ], + "score": 1.0, + "content": "For", + "type": "text" + }, + { + "bbox": [ + 123, + 514, + 146, + 524 + ], + "score": 0.89, + "content": "t > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 512, + 493, + 528 + ], + "score": 1.0, + "content": ", this quantity is not equal to the sum of the scores of the two models which is given by", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 512, + 493, + 528 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 529, + 465, + 543 + ], + "lines": [ + { + "bbox": [ + 145, + 529, + 465, + 543 + ], + "spans": [ + { + "bbox": [ + 145, + 529, + 465, + 543 + ], + "score": 0.85, + "content": "\\begin{array} { r } { \\nabla _ { x } \\log q _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } ) = \\nabla _ { x } \\log \\left( \\int d x _ { 0 } q ^ { 1 } ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) \\right) + \\nabla _ { x } \\log \\left( \\int d x _ { 0 } q ^ { 2 } ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "24bf69b118ac9b51a6a200fc36441b770e26c58dd5d7bef98b58a45bee53ee7a.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 145, + 529, + 465, + 543 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 545, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 104, + 543, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 104, + 543, + 506, + 559 + ], + "score": 1.0, + "content": "Therefore, plugging the composed score function into the standard ancestral sampling procedure", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "score": 1.0, + "content": "discussed in Section 2.1, which we refer to as “reverse diffusion,” does not correspond to sampling", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "from the composed model, and thus reverse diffusion sampling will generate incorrect samples from", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 577, + 459, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 459, + 590 + ], + "score": 1.0, + "content": "composed distributions. This effect can be seen in Figure 2, with details in Appendix D.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 104, + 543, + 506, + 590 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 104, + 592, + 507, + 612 + ], + "spans": [ + { + "bbox": [ + 104, + 592, + 218, + 612 + ], + "score": 1.0, + "content": "The score of the distribution", + "type": "text" + }, + { + "bbox": [ + 218, + 596, + 254, + 609 + ], + "score": 0.92, + "content": "q _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 592, + 444, + 612 + ], + "score": 1.0, + "content": "in Equation 12 is easy to compute, unlike that of", + "type": "text" + }, + { + "bbox": [ + 445, + 596, + 482, + 609 + ], + "score": 0.92, + "content": "\\tilde { q } _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 592, + 507, + 612 + ], + "score": 1.0, + "content": "from", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 608, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 104, + 608, + 209, + 624 + ], + "score": 1.0, + "content": "Equation 11. In addition,", + "type": "text" + }, + { + "bbox": [ + 209, + 610, + 245, + 622 + ], + "score": 0.91, + "content": "q _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 608, + 506, + 624 + ], + "score": 1.0, + "content": "describes a sequence of distributions which smoothly interpolate", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 619, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 144, + 634 + ], + "score": 1.0, + "content": "between", + "type": "text" + }, + { + "bbox": [ + 144, + 622, + 176, + 633 + ], + "score": 0.9, + "content": "q ^ { \\mathrm { p r o d } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 619, + 189, + 634 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 189, + 623, + 214, + 631 + ], + "score": 0.88, + "content": "t = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 619, + 234, + 634 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 235, + 622, + 267, + 633 + ], + "score": 0.92, + "content": "\\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 619, + 280, + 634 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 281, + 623, + 308, + 631 + ], + "score": 0.88, + "content": "t = T", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 619, + 506, + 634 + ], + "score": 1.0, + "content": ", though this sequence of distributions does not", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 631, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 505, + 644 + ], + "score": 1.0, + "content": "correspond to the distributions that result from the standard forward diffusion process described in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "score": 1.0, + "content": "Section 2.1, leading the reverse diffusion sampling to generate poor samples. We discuss how we", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 653, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 104, + 653, + 338, + 669 + ], + "score": 1.0, + "content": "may utilize MCMC samplers, which use our knowledge of", + "type": "text" + }, + { + "bbox": [ + 338, + 656, + 403, + 668 + ], + "score": 0.93, + "content": "\\nabla _ { x } \\log { q _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 653, + 506, + 669 + ], + "score": 1.0, + "content": ", to correctly sample from", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 104, + 664, + 453, + 684 + ], + "spans": [ + { + "bbox": [ + 104, + 664, + 211, + 684 + ], + "score": 1.0, + "content": "intermediate distributions", + "type": "text" + }, + { + "bbox": [ + 212, + 669, + 249, + 682 + ], + "score": 0.92, + "content": "\\tilde { q } _ { t } ^ { \\mathrm { p r o d } } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 664, + 453, + 684 + ], + "score": 1.0, + "content": ", leading to accurate composed sample generation.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40, + "bbox_fs": [ + 104, + 592, + 507, + 684 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 690, + 290, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 690, + 291, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 690, + 291, + 703 + ], + "score": 1.0, + "content": "4.1 IMPROVING SAMPLING WITH MCMC", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 205, + 723 + ], + "score": 1.0, + "content": "In order to sample from", + "type": "text" + }, + { + "bbox": [ + 206, + 711, + 237, + 722 + ], + "score": 0.92, + "content": "q ^ { \\mathrm { p r o d } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "using the combined score function from Equation 12, we can use", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "annealed MCMC sampling, described below in Algorithm 1. This method applies MCMC transition", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "kernels to a sequence of distributions which begins with a known, tractable distribution and concludes", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 235, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 249 + ], + "score": 1.0, + "content": "at our target distribution. Annealed MCMC has a long history enabling sampling from very complex", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 247, + 304, + 259 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 304, + 259 + ], + "score": 1.0, + "content": "distributions (Neal, 2001; Song & Ermon, 2019).", + "type": "text", + "cross_page": true + } + ], + "index": 7 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 708, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 175, + 80, + 434, + 189 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 175, + 80, + 434, + 189 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 175, + 80, + 434, + 189 + ], + "spans": [ + { + "bbox": [ + 175, + 80, + 434, + 189 + ], + "score": 0.976, + "html": "
Model SamplerProductMixture
RAISE↑LL↑Var↓ln(MMD)↓LL↑Var↓
Reverse1.55-6.470.063=
ScoreULA2.371.790.026=
U-HMC2.522.400.021
Reverse (equal steps)2.27-2.920.046==-
Reverse1.37-6.030.064-3.84-2.170.020
ULA2.361.840.027-4.210.570.013
EBMMALA2.642.730.013-4.381.290.008
U-HMC2.632.450.022-4.691.030.010
HMC2.712.720.009-4.481.300.007
", + "type": "table", + "image_path": "668b3f3e453ab1de0f64e31c641774d6f3f5e85ba4ed7ea487e4fedf36134640.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 175, + 80, + 434, + 116.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 175, + 116.33333333333334, + 434, + 152.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 175, + 152.66666666666669, + 434, + 189.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 194, + 505, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 193, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 505, + 205 + ], + "score": 1.0, + "content": "Table 1: Quantitative results on 2D composition. Energy based parameterization enables mixture", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 204, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 215 + ], + "score": 1.0, + "content": "compositional models, and MCMC sampling leads to better samples from compositional diffusion models.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 224, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "kernels to a sequence of distributions which begins with a known, tractable distribution and concludes", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 235, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 249 + ], + "score": 1.0, + "content": "at our target distribution. Annealed MCMC has a long history enabling sampling from very complex", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 247, + 304, + 259 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 304, + 259 + ], + "score": 1.0, + "content": "distributions (Neal, 2001; Song & Ermon, 2019).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 263, + 336, + 406 + ], + "lines": [ + { + "bbox": [ + 106, + 262, + 337, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 272, + 276 + ], + "score": 1.0, + "content": "We explore two types of transition kernels", + "type": "text" + }, + { + "bbox": [ + 273, + 263, + 299, + 275 + ], + "score": 0.91, + "content": "k _ { t } ( \\cdot | \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 262, + 337, + 276 + ], + "score": 1.0, + "content": "based on", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 274, + 336, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 336, + 287 + ], + "score": 1.0, + "content": "Langevin Dynamics (Equation 4) and HMC. When using", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 285, + 337, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 160, + 297 + ], + "score": 1.0, + "content": "the standard", + "type": "text" + }, + { + "bbox": [ + 161, + 286, + 192, + 297 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 285, + 337, + 297 + ], + "score": 1.0, + "content": "-parameterization, we do not have", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 297, + 336, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 336, + 309 + ], + "score": 1.0, + "content": "access to an explicitly defined energy-function meaning", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 307, + 336, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 336, + 319 + ], + "score": 1.0, + "content": "we cannot utilize any MCMC sampler with Metropolis", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 318, + 336, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 336, + 330 + ], + "score": 1.0, + "content": "corrections. Thus, we only utilize the ULA and U-HMC", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 329, + 337, + 341 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 337, + 341 + ], + "score": 1.0, + "content": "samplers described in Section 2.2. These samplers are not", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 340, + 337, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 337, + 352 + ], + "score": 1.0, + "content": "exact, but can in practice generate good results. In the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 352, + 336, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 336, + 363 + ], + "score": 1.0, + "content": "next section we detail how Metropolis corrections may be", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 363, + 336, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 336, + 373 + ], + "score": 1.0, + "content": "incorporated. Full details of our samplers can be found", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 373, + 336, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 336, + 385 + ], + "score": 1.0, + "content": "in Appendix C.1. While continuous time sampling in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 384, + 337, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 337, + 396 + ], + "score": 1.0, + "content": "diffusion models Song et al. (2021) is also referred to as", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 394, + 337, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 337, + 407 + ], + "score": 1.0, + "content": "ULA, the MCMC sampling procedure is run across time", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 344, + 271, + 471, + 282 + ], + "lines": [ + { + "bbox": [ + 344, + 270, + 472, + 284 + ], + "spans": [ + { + "bbox": [ + 344, + 270, + 472, + 284 + ], + "score": 1.0, + "content": "Algorithm 1 Annealed MCMC", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 346, + 286, + 504, + 389 + ], + "lines": [ + { + "bbox": [ + 353, + 286, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 353, + 286, + 452, + 298 + ], + "score": 1.0, + "content": "Input: Transition kernels", + "type": "text" + }, + { + "bbox": [ + 452, + 286, + 476, + 298 + ], + "score": 0.9, + "content": "k _ { t } ( \\cdot | \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 286, + 505, + 298 + ], + "score": 1.0, + "content": ", Initial", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 352, + 295, + 492, + 309 + ], + "spans": [ + { + "bbox": [ + 352, + 295, + 396, + 309 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 396, + 297, + 417, + 308 + ], + "score": 0.86, + "content": "p _ { T } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 295, + 483, + 309 + ], + "score": 1.0, + "content": ", Number of steps", + "type": "text" + }, + { + "bbox": [ + 483, + 297, + 492, + 306 + ], + "score": 0.75, + "content": "N", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 353, + 306, + 456, + 317 + ], + "spans": [ + { + "bbox": [ + 353, + 307, + 400, + 317 + ], + "score": 0.8, + "content": "x r \\sim p _ { T } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 306, + 456, + 317 + ], + "score": 1.0, + "content": "# Initialize.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 352, + 315, + 428, + 327 + ], + "spans": [ + { + "bbox": [ + 352, + 315, + 366, + 327 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 366, + 317, + 415, + 327 + ], + "score": 0.69, + "content": "t = T , \\dots , 0", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 315, + 428, + 327 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 362, + 326, + 441, + 338 + ], + "spans": [ + { + "bbox": [ + 362, + 326, + 376, + 338 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 376, + 327, + 426, + 337 + ], + "score": 0.71, + "content": "i = 1 , \\ldots , N", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 326, + 441, + 338 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 373, + 337, + 424, + 348 + ], + "spans": [ + { + "bbox": [ + 373, + 337, + 424, + 348 + ], + "score": 0.64, + "content": "x _ { t } \\sim k _ { t } ( \\cdot | x _ { t } )", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 362, + 345, + 394, + 357 + ], + "spans": [ + { + "bbox": [ + 362, + 345, + 394, + 357 + ], + "score": 1.0, + "content": "end for", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 363, + 358, + 403, + 367 + ], + "spans": [ + { + "bbox": [ + 363, + 358, + 403, + 367 + ], + "score": 0.29, + "content": "x _ { t - 1 } = x _ { t }", + "type": "inline_equation", + "image_path": "3284f9216971b69178ae154c5095b0ffaf65d1198916bf4cf72a7b8cd64827c4.jpg" + } + ], + "index": 29 + }, + { + "bbox": [ + 352, + 366, + 384, + 377 + ], + "spans": [ + { + "bbox": [ + 352, + 366, + 384, + 377 + ], + "score": 1.0, + "content": "end for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 352, + 375, + 395, + 389 + ], + "spans": [ + { + "bbox": [ + 352, + 375, + 395, + 389 + ], + "score": 1.0, + "content": "return x0", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 407, + 505, + 442 + ], + "lines": [ + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "score": 1.0, + "content": "(and corresponds to the same sampling procedure as discretized diffusion discussed in Section 2.1),", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 101, + 414, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 101, + 414, + 428, + 437 + ], + "score": 1.0, + "content": "as opposed to being used to sample from each intermediate distribution q˜prodt (", + "type": "text" + }, + { + "bbox": [ + 439, + 417, + 506, + 433 + ], + "score": 1.0, + "content": ". Thus applying", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 429, + 400, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 400, + 443 + ], + "score": 1.0, + "content": "continuous sampling gives the same issues as reverse diffusion sampling.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 447, + 505, + 491 + ], + "lines": [ + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "score": 1.0, + "content": "We can see again in Figure 2 that applying this MCMC sampling procedure allows samples from", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "score": 1.0, + "content": "the composed distribution to be faithfully generated with no modification to the underlying diffusion", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "score": 1.0, + "content": "models. Quantitative results can be found in Table 1 which further imply that the choice of sampler", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 481, + 460, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 460, + 492 + ], + "score": 1.0, + "content": "may be responsible for prior failures in compositional generation with diffusion models.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "title", + "bbox": [ + 108, + 503, + 291, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 503, + 293, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 293, + 515 + ], + "score": 1.0, + "content": "4.2 ENERGY-BASED PARAMETERIZATION", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 506, + 566 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "As noted in Section 3, we are unable to use mixture composition without an explicitly parameterized", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 381, + 546 + ], + "score": 1.0, + "content": "likelihood function. But, if we parameterize a potential function", + "type": "text" + }, + { + "bbox": [ + 382, + 533, + 414, + 545 + ], + "score": 0.93, + "content": "f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 533, + 506, + 546 + ], + "score": 1.0, + "content": "and implicitly define", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 107, + 544, + 507, + 557 + ], + "spans": [ + { + "bbox": [ + 107, + 544, + 197, + 556 + ], + "score": 0.91, + "content": "\\epsilon _ { \\theta } ( x , t ) = \\nabla _ { x } f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 544, + 507, + 557 + ], + "score": 1.0, + "content": "we can recover an explicit estimate of the (unnormalized) log-likelihood –", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 555, + 367, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 367, + 568 + ], + "score": 1.0, + "content": "enabling us to utilize all presented forms for model composition.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 106, + 572, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 571, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 505, + 585 + ], + "score": 1.0, + "content": "Additionally, an explicit estimate of log-likelihood enables the use of more accurate samplers. As", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 582, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 244, + 596 + ], + "score": 1.0, + "content": "explained above, with the standard", + "type": "text" + }, + { + "bbox": [ + 245, + 583, + 275, + 595 + ], + "score": 0.92, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 582, + 506, + 596 + ], + "score": 1.0, + "content": "-parameterization we can only utilize unadjusted samplers.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 594, + 506, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 506, + 607 + ], + "score": 1.0, + "content": "While they can perform well in practice, there exist many distributions from which they cannot", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 104, + 604, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 104, + 604, + 506, + 618 + ], + "score": 1.0, + "content": "generate decent samples (Roberts & Tweedie, 1996) such as targets with lighter-than-Gaussian tails", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "where the ULA chain is transient. Additionally, for an accurate approximation to the Langevin SDE,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "ULA will need increasingly small stepsizes as the curvature of the log-likelihood gradient increases", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 638, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 505, + 651 + ], + "score": 1.0, + "content": "which can lead to arbitrarily slow mixing (Durmus & Moulines, 2019). In these settings a Metropolis", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 648, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 663 + ], + "score": 1.0, + "content": "correction can greatly improve sample quality and convergence. Again this issue can be solved by", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 659, + 476, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 141, + 674 + ], + "score": 1.0, + "content": "defining", + "type": "text" + }, + { + "bbox": [ + 142, + 660, + 230, + 672 + ], + "score": 0.93, + "content": "\\epsilon _ { \\theta } ( x , t ) = \\nabla _ { x } f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 659, + 440, + 674 + ], + "score": 1.0, + "content": "for some explicitly defined scalar potential function", + "type": "text" + }, + { + "bbox": [ + 440, + 660, + 471, + 672 + ], + "score": 0.93, + "content": "f _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 659, + 476, + 674 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 48 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Energy-based parameterizations have been explored in the past (Salimans & Ho, 2021) and were", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "found to perform comparably to score-based models for unconditional generative modeling. In that", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "setting the score parameterization is then preferable as computing the gradient of the energy requires", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "more computation. In the compositional setting, however, the additional flexibility enabled by explicit", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "(unnormalized) log-probability estimation motivates a re-exploration of the energy-parameterization.", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 55 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 175, + 80, + 434, + 189 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 175, + 80, + 434, + 189 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 175, + 80, + 434, + 189 + ], + "spans": [ + { + "bbox": [ + 175, + 80, + 434, + 189 + ], + "score": 0.976, + "html": "
Model SamplerProductMixture
RAISE↑LL↑Var↓ln(MMD)↓LL↑Var↓
Reverse1.55-6.470.063=
ScoreULA2.371.790.026=
U-HMC2.522.400.021
Reverse (equal steps)2.27-2.920.046==-
Reverse1.37-6.030.064-3.84-2.170.020
ULA2.361.840.027-4.210.570.013
EBMMALA2.642.730.013-4.381.290.008
U-HMC2.632.450.022-4.691.030.010
HMC2.712.720.009-4.481.300.007
", + "type": "table", + "image_path": "668b3f3e453ab1de0f64e31c641774d6f3f5e85ba4ed7ea487e4fedf36134640.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 175, + 80, + 434, + 116.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 175, + 116.33333333333334, + 434, + 152.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 175, + 152.66666666666669, + 434, + 189.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 194, + 505, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 193, + 505, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 505, + 205 + ], + "score": 1.0, + "content": "Table 1: Quantitative results on 2D composition. Energy based parameterization enables mixture", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 204, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 215 + ], + "score": 1.0, + "content": "compositional models, and MCMC sampling leads to better samples from compositional diffusion models.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 224, + 505, + 258 + ], + "lines": [], + "index": 6, + "bbox_fs": [ + 105, + 225, + 505, + 259 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 263, + 336, + 406 + ], + "lines": [ + { + "bbox": [ + 106, + 262, + 337, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 272, + 276 + ], + "score": 1.0, + "content": "We explore two types of transition kernels", + "type": "text" + }, + { + "bbox": [ + 273, + 263, + 299, + 275 + ], + "score": 0.91, + "content": "k _ { t } ( \\cdot | \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 262, + 337, + 276 + ], + "score": 1.0, + "content": "based on", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 274, + 336, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 336, + 287 + ], + "score": 1.0, + "content": "Langevin Dynamics (Equation 4) and HMC. When using", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 285, + 337, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 160, + 297 + ], + "score": 1.0, + "content": "the standard", + "type": "text" + }, + { + "bbox": [ + 161, + 286, + 192, + 297 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 285, + 337, + 297 + ], + "score": 1.0, + "content": "-parameterization, we do not have", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 297, + 336, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 336, + 309 + ], + "score": 1.0, + "content": "access to an explicitly defined energy-function meaning", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 307, + 336, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 336, + 319 + ], + "score": 1.0, + "content": "we cannot utilize any MCMC sampler with Metropolis", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 318, + 336, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 336, + 330 + ], + "score": 1.0, + "content": "corrections. Thus, we only utilize the ULA and U-HMC", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 329, + 337, + 341 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 337, + 341 + ], + "score": 1.0, + "content": "samplers described in Section 2.2. These samplers are not", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 340, + 337, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 337, + 352 + ], + "score": 1.0, + "content": "exact, but can in practice generate good results. In the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 352, + 336, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 336, + 363 + ], + "score": 1.0, + "content": "next section we detail how Metropolis corrections may be", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 363, + 336, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 336, + 373 + ], + "score": 1.0, + "content": "incorporated. Full details of our samplers can be found", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 373, + 336, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 336, + 385 + ], + "score": 1.0, + "content": "in Appendix C.1. While continuous time sampling in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 384, + 337, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 337, + 396 + ], + "score": 1.0, + "content": "diffusion models Song et al. (2021) is also referred to as", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 394, + 337, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 337, + 407 + ], + "score": 1.0, + "content": "ULA, the MCMC sampling procedure is run across time", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 262, + 337, + 407 + ] + }, + { + "type": "title", + "bbox": [ + 344, + 271, + 471, + 282 + ], + "lines": [ + { + "bbox": [ + 344, + 270, + 472, + 284 + ], + "spans": [ + { + "bbox": [ + 344, + 270, + 472, + 284 + ], + "score": 1.0, + "content": "Algorithm 1 Annealed MCMC", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "list", + "bbox": [ + 346, + 286, + 504, + 389 + ], + "lines": [ + { + "bbox": [ + 353, + 286, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 353, + 286, + 452, + 298 + ], + "score": 1.0, + "content": "Input: Transition kernels", + "type": "text" + }, + { + "bbox": [ + 452, + 286, + 476, + 298 + ], + "score": 0.9, + "content": "k _ { t } ( \\cdot | \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 286, + 505, + 298 + ], + "score": 1.0, + "content": ", Initial", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 352, + 295, + 492, + 309 + ], + "spans": [ + { + "bbox": [ + 352, + 295, + 396, + 309 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 396, + 297, + 417, + 308 + ], + "score": 0.86, + "content": "p _ { T } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 295, + 483, + 309 + ], + "score": 1.0, + "content": ", Number of steps", + "type": "text" + }, + { + "bbox": [ + 483, + 297, + 492, + 306 + ], + "score": 0.75, + "content": "N", + "type": "inline_equation" + } + ], + "index": 23, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 353, + 306, + 456, + 317 + ], + "spans": [ + { + "bbox": [ + 353, + 307, + 400, + 317 + ], + "score": 0.8, + "content": "x r \\sim p _ { T } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 306, + 456, + 317 + ], + "score": 1.0, + "content": "# Initialize.", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 352, + 315, + 428, + 327 + ], + "spans": [ + { + "bbox": [ + 352, + 315, + 366, + 327 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 366, + 317, + 415, + 327 + ], + "score": 0.69, + "content": "t = T , \\dots , 0", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 315, + 428, + 327 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 362, + 326, + 441, + 338 + ], + "spans": [ + { + "bbox": [ + 362, + 326, + 376, + 338 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 376, + 327, + 426, + 337 + ], + "score": 0.71, + "content": "i = 1 , \\ldots , N", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 326, + 441, + 338 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 373, + 337, + 424, + 348 + ], + "spans": [ + { + "bbox": [ + 373, + 337, + 424, + 348 + ], + "score": 0.64, + "content": "x _ { t } \\sim k _ { t } ( \\cdot | x _ { t } )", + "type": "inline_equation" + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 362, + 345, + 394, + 357 + ], + "spans": [ + { + "bbox": [ + 362, + 345, + 394, + 357 + ], + "score": 1.0, + "content": "end for", + "type": "text" + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 363, + 358, + 403, + 367 + ], + "spans": [ + { + "bbox": [ + 363, + 358, + 403, + 367 + ], + "score": 0.29, + "content": "x _ { t - 1 } = x _ { t }", + "type": "inline_equation", + "image_path": "3284f9216971b69178ae154c5095b0ffaf65d1198916bf4cf72a7b8cd64827c4.jpg" + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 352, + 366, + 384, + 377 + ], + "spans": [ + { + "bbox": [ + 352, + 366, + 384, + 377 + ], + "score": 1.0, + "content": "end for", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 352, + 375, + 395, + 389 + ], + "spans": [ + { + "bbox": [ + 352, + 375, + 395, + 389 + ], + "score": 1.0, + "content": "return x0", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 26.5, + "bbox_fs": [ + 352, + 286, + 505, + 389 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 407, + 505, + 442 + ], + "lines": [ + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 506, + 418 + ], + "score": 1.0, + "content": "(and corresponds to the same sampling procedure as discretized diffusion discussed in Section 2.1),", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 101, + 414, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 101, + 414, + 428, + 437 + ], + "score": 1.0, + "content": "as opposed to being used to sample from each intermediate distribution q˜prodt (", + "type": "text" + }, + { + "bbox": [ + 439, + 417, + 506, + 433 + ], + "score": 1.0, + "content": ". Thus applying", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 429, + 400, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 400, + 443 + ], + "score": 1.0, + "content": "continuous sampling gives the same issues as reverse diffusion sampling.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 101, + 405, + 506, + 443 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 447, + 505, + 491 + ], + "lines": [ + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "score": 1.0, + "content": "We can see again in Figure 2 that applying this MCMC sampling procedure allows samples from", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "score": 1.0, + "content": "the composed distribution to be faithfully generated with no modification to the underlying diffusion", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "score": 1.0, + "content": "models. Quantitative results can be found in Table 1 which further imply that the choice of sampler", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 481, + 460, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 460, + 492 + ], + "score": 1.0, + "content": "may be responsible for prior failures in compositional generation with diffusion models.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 446, + 506, + 492 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 503, + 291, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 503, + 293, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 293, + 515 + ], + "score": 1.0, + "content": "4.2 ENERGY-BASED PARAMETERIZATION", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 506, + 566 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "As noted in Section 3, we are unable to use mixture composition without an explicitly parameterized", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 381, + 546 + ], + "score": 1.0, + "content": "likelihood function. But, if we parameterize a potential function", + "type": "text" + }, + { + "bbox": [ + 382, + 533, + 414, + 545 + ], + "score": 0.93, + "content": "f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 533, + 506, + 546 + ], + "score": 1.0, + "content": "and implicitly define", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 107, + 544, + 507, + 557 + ], + "spans": [ + { + "bbox": [ + 107, + 544, + 197, + 556 + ], + "score": 0.91, + "content": "\\epsilon _ { \\theta } ( x , t ) = \\nabla _ { x } f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 544, + 507, + 557 + ], + "score": 1.0, + "content": "we can recover an explicit estimate of the (unnormalized) log-likelihood –", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 555, + 367, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 367, + 568 + ], + "score": 1.0, + "content": "enabling us to utilize all presented forms for model composition.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 522, + 507, + 568 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 572, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 571, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 505, + 585 + ], + "score": 1.0, + "content": "Additionally, an explicit estimate of log-likelihood enables the use of more accurate samplers. As", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 582, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 244, + 596 + ], + "score": 1.0, + "content": "explained above, with the standard", + "type": "text" + }, + { + "bbox": [ + 245, + 583, + 275, + 595 + ], + "score": 0.92, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 582, + 506, + 596 + ], + "score": 1.0, + "content": "-parameterization we can only utilize unadjusted samplers.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 594, + 506, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 506, + 607 + ], + "score": 1.0, + "content": "While they can perform well in practice, there exist many distributions from which they cannot", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 104, + 604, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 104, + 604, + 506, + 618 + ], + "score": 1.0, + "content": "generate decent samples (Roberts & Tweedie, 1996) such as targets with lighter-than-Gaussian tails", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "where the ULA chain is transient. Additionally, for an accurate approximation to the Langevin SDE,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "ULA will need increasingly small stepsizes as the curvature of the log-likelihood gradient increases", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 638, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 505, + 651 + ], + "score": 1.0, + "content": "which can lead to arbitrarily slow mixing (Durmus & Moulines, 2019). In these settings a Metropolis", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 648, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 663 + ], + "score": 1.0, + "content": "correction can greatly improve sample quality and convergence. Again this issue can be solved by", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 659, + 476, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 141, + 674 + ], + "score": 1.0, + "content": "defining", + "type": "text" + }, + { + "bbox": [ + 142, + 660, + 230, + 672 + ], + "score": 0.93, + "content": "\\epsilon _ { \\theta } ( x , t ) = \\nabla _ { x } f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 659, + 440, + 674 + ], + "score": 1.0, + "content": "for some explicitly defined scalar potential function", + "type": "text" + }, + { + "bbox": [ + 440, + 660, + 471, + 672 + ], + "score": 0.93, + "content": "f _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 659, + 476, + 674 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 48, + "bbox_fs": [ + 104, + 571, + 506, + 674 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Energy-based parameterizations have been explored in the past (Salimans & Ho, 2021) and were", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "found to perform comparably to score-based models for unconditional generative modeling. In that", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "setting the score parameterization is then preferable as computing the gradient of the energy requires", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "more computation. In the compositional setting, however, the additional flexibility enabled by explicit", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "(unnormalized) log-probability estimation motivates a re-exploration of the energy-parameterization.", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 55, + "bbox_fs": [ + 105, + 677, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 82, + 502, + 205 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 82, + 502, + 205 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 82, + 502, + 205 + ], + "spans": [ + { + "bbox": [ + 111, + 82, + 502, + 205 + ], + "score": 0.845, + "type": "image", + "image_path": "17d85044c3b8849e7fab51be1678ab0ee00be24a391f698285ed722c9fdd8dc2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 82, + 502, + 123.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 123.0, + 502, + 164.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 164.0, + 502, + 205.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 206, + 505, + 256 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 206, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 206, + 505, + 217 + ], + "score": 1.0, + "content": "Figure 3: (a) Composition enables the positions of multiple shapes to be simultaneously controlled, while", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 215, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 228 + ], + "score": 1.0, + "content": "training only conditions on the location of one object per image. Reverse diffusion samples place shapes", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "score": 1.0, + "content": "in incorrect locations. MCMC generates samples that satisfy all constraints. (b) Metropolis adjustment", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 236, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 506, + 248 + ], + "score": 1.0, + "content": "significantly improves generation performance across sampling steps. As more MCMC steps are run (at", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 245, + 420, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 420, + 258 + ], + "score": 1.0, + "content": "each timestep), generation accuracy of combinations of 5 cubes improves significantly.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 262, + 505, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "score": 1.0, + "content": "We explored a number of energy-based parameterizations for diffusion models and ran a pi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 272, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 506, + 286 + ], + "score": 1.0, + "content": "lot study on ImageNet. In this study we found it best to parameterize the log probability as", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 282, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 107, + 283, + 211, + 296 + ], + "score": 0.91, + "content": "f _ { \\theta } ( x , t ) \\dot { { \\bf \\theta } } = - | | s _ { \\theta } \\overline { { ( x , t ) } } | | ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 282, + 245, + 297 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 246, + 284, + 278, + 296 + ], + "score": 0.94, + "content": "s _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 282, + 505, + 297 + ], + "score": 1.0, + "content": "is a vector-output neural network, like those used in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 294, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 107, + 296, + 137, + 307 + ], + "score": 0.91, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 294, + 506, + 308 + ], + "score": 1.0, + "content": "-parameterized diffusion models. Full details on our study can be found in Appendix E. From", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 306, + 362, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 362, + 318 + ], + "score": 1.0, + "content": "here on, all energy-based diffusion models take the above form.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 322, + 505, + 378 + ], + "lines": [ + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "Our energy-parameterized models enable us to use MALA and HMC samplers which produce our", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "score": 1.0, + "content": "best compositional generation results by a large margin. An additional benefit of these samplers", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 345, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 505, + 357 + ], + "score": 1.0, + "content": "is that, through monitoring their acceptance rates, we are able to derive an effective automated", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 356, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 506, + 369 + ], + "score": 1.0, + "content": "method for tuning their hyper-parameters (a notoriously difficult task prior) which is not available for", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 366, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 379 + ], + "score": 1.0, + "content": "unadjusted samplers. Details of our samplers and tuning procedures can be found in Appendix C.1.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 384, + 200, + 396 + ], + "lines": [ + { + "bbox": [ + 105, + 382, + 201, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 201, + 398 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 506, + 503 + ], + "lines": [ + { + "bbox": [ + 106, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "We experiment with various model parameterizations and sampling schemes for compositional", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "score": 1.0, + "content": "generation with diffusion models. We first investigate these ideas on some illustrative 2D datasets,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "then move to the image domain with an artificial dataset of shapes. Here, we compose a model", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "conditioned on the location of a single shape with itself to condition on the location of all of the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 448, + 507, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 507, + 461 + ], + "score": 1.0, + "content": "shapes in the image. After this we experiment with classifier guidance on the ImageNet dataset.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 458, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 506, + 472 + ], + "score": 1.0, + "content": "Finally, we self-compose text-to-image models to generate from compositions of various text prompts.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "Full details of all experiments can be found in Appendix G. Throughout we compare our proposed", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 481, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 481, + 506, + 494 + ], + "score": 1.0, + "content": "improvements with a score-parameterized model using standard reverse diffusion sampling. We note", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 492, + 350, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 350, + 505 + ], + "score": 1.0, + "content": "that this baseline is exactly the approach of Liu et al. (2022).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 107, + 508, + 192, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 193, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 193, + 522 + ], + "score": 1.0, + "content": "5.1 2D DENSITIES", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 524, + 505, + 634 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "We train diffusion models using both parameterizations and study the impact of various sampling", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 534, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 104, + 534, + 506, + 547 + ], + "score": 1.0, + "content": "approaches for compositional generation. Samples are evaluated using RAISE (Burda et al., 2015)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 307, + 558 + ], + "score": 1.0, + "content": "(which gives lower bounds on log-likelihood) and", + "type": "text" + }, + { + "bbox": [ + 307, + 545, + 337, + 556 + ], + "score": 0.71, + "content": "\\mathbf { M M D } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 545, + 505, + 558 + ], + "score": 1.0, + "content": ", LL (log-likelihood of generated samples", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "under composed distribution), and Var (L2 difference of variance of GMMs fit on generated samples", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "compared to GMMs of the composed distribution). Results can be found in Table 1 and visualizations", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "can be seen in Figure 2. All MCMC sampling methods improve sample quality and likelihood, with", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 589, + 506, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 506, + 602 + ], + "score": 1.0, + "content": "Metropolis adjusted methods performing the best. All MCMC experiments use the same number", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "of score function evaluations. We include a baseline, labeled “Reverse (equal steps)” which is a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "score": 1.0, + "content": "diffusion model trained with more steps such that reverse diffusion sampling has the same cost as our", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 621, + 507, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 507, + 637 + ], + "score": 1.0, + "content": "MCMC samplers. We see that simply adding more time-steps does not solve compositional sampling.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 108, + 642, + 216, + 653 + ], + "lines": [ + { + "bbox": [ + 105, + 640, + 218, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 218, + 655 + ], + "score": 1.0, + "content": "5.2 COMPOSING CUBES", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 659, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 672 + ], + "score": 1.0, + "content": "Next, we train models on a dataset of images containing between 1 and 5 examples of various shapes", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 670, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 383, + 683 + ], + "score": 1.0, + "content": "taken from CLEVR (Johnson et al., 2017). We train our models to fit", + "type": "text" + }, + { + "bbox": [ + 383, + 670, + 411, + 682 + ], + "score": 0.93, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 670, + 438, + 683 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 439, + 672, + 446, + 682 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 670, + 506, + 683 + ], + "score": 1.0, + "content": "is the location", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 682, + 506, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 693 + ], + "score": 1.0, + "content": "of one of the shapes in the image. We then compose this conditional model with itself to create a", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 692, + 438, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 390, + 705 + ], + "score": 1.0, + "content": "product model which defines the distribution of images conditioned on", + "type": "text" + }, + { + "bbox": [ + 391, + 695, + 397, + 702 + ], + "score": 0.72, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 692, + 438, + 705 + ], + "score": 1.0, + "content": "shapes as", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 711, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 117, + 709, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 117, + 709, + 505, + 724 + ], + "score": 1.0, + "content": "2For the mixture we use MMD to replace RAISE likelihood based evaluation as we encountered numerical", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 721, + 317, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 317, + 732 + ], + "score": 1.0, + "content": "stability issues with RAISE when applying to the mixture.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 82, + 502, + 205 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 82, + 502, + 205 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 82, + 502, + 205 + ], + "spans": [ + { + "bbox": [ + 111, + 82, + 502, + 205 + ], + "score": 0.845, + "type": "image", + "image_path": "17d85044c3b8849e7fab51be1678ab0ee00be24a391f698285ed722c9fdd8dc2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 82, + 502, + 123.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 123.0, + 502, + 164.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 164.0, + 502, + 205.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 206, + 505, + 256 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 206, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 206, + 505, + 217 + ], + "score": 1.0, + "content": "Figure 3: (a) Composition enables the positions of multiple shapes to be simultaneously controlled, while", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 215, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 228 + ], + "score": 1.0, + "content": "training only conditions on the location of one object per image. Reverse diffusion samples place shapes", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "score": 1.0, + "content": "in incorrect locations. MCMC generates samples that satisfy all constraints. (b) Metropolis adjustment", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 236, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 506, + 248 + ], + "score": 1.0, + "content": "significantly improves generation performance across sampling steps. As more MCMC steps are run (at", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 245, + 420, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 420, + 258 + ], + "score": 1.0, + "content": "each timestep), generation accuracy of combinations of 5 cubes improves significantly.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 262, + 505, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "score": 1.0, + "content": "We explored a number of energy-based parameterizations for diffusion models and ran a pi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 272, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 506, + 286 + ], + "score": 1.0, + "content": "lot study on ImageNet. In this study we found it best to parameterize the log probability as", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 282, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 107, + 283, + 211, + 296 + ], + "score": 0.91, + "content": "f _ { \\theta } ( x , t ) \\dot { { \\bf \\theta } } = - | | s _ { \\theta } \\overline { { ( x , t ) } } | | ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 282, + 245, + 297 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 246, + 284, + 278, + 296 + ], + "score": 0.94, + "content": "s _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 282, + 505, + 297 + ], + "score": 1.0, + "content": "is a vector-output neural network, like those used in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 294, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 107, + 296, + 137, + 307 + ], + "score": 0.91, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 294, + 506, + 308 + ], + "score": 1.0, + "content": "-parameterized diffusion models. Full details on our study can be found in Appendix E. From", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 306, + 362, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 362, + 318 + ], + "score": 1.0, + "content": "here on, all energy-based diffusion models take the above form.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 261, + 506, + 318 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 322, + 505, + 378 + ], + "lines": [ + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "Our energy-parameterized models enable us to use MALA and HMC samplers which produce our", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "score": 1.0, + "content": "best compositional generation results by a large margin. An additional benefit of these samplers", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 345, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 505, + 357 + ], + "score": 1.0, + "content": "is that, through monitoring their acceptance rates, we are able to derive an effective automated", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 356, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 506, + 369 + ], + "score": 1.0, + "content": "method for tuning their hyper-parameters (a notoriously difficult task prior) which is not available for", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 366, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 379 + ], + "score": 1.0, + "content": "unadjusted samplers. Details of our samplers and tuning procedures can be found in Appendix C.1.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 322, + 506, + 379 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 384, + 200, + 396 + ], + "lines": [ + { + "bbox": [ + 105, + 382, + 201, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 201, + 398 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 506, + 503 + ], + "lines": [ + { + "bbox": [ + 106, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "We experiment with various model parameterizations and sampling schemes for compositional", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "score": 1.0, + "content": "generation with diffusion models. We first investigate these ideas on some illustrative 2D datasets,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "then move to the image domain with an artificial dataset of shapes. Here, we compose a model", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "conditioned on the location of a single shape with itself to condition on the location of all of the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 448, + 507, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 507, + 461 + ], + "score": 1.0, + "content": "shapes in the image. After this we experiment with classifier guidance on the ImageNet dataset.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 458, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 506, + 472 + ], + "score": 1.0, + "content": "Finally, we self-compose text-to-image models to generate from compositions of various text prompts.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "Full details of all experiments can be found in Appendix G. Throughout we compare our proposed", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 481, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 481, + 506, + 494 + ], + "score": 1.0, + "content": "improvements with a score-parameterized model using standard reverse diffusion sampling. We note", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 492, + 350, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 350, + 505 + ], + "score": 1.0, + "content": "that this baseline is exactly the approach of Liu et al. (2022).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 404, + 507, + 505 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 508, + 192, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 193, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 193, + 522 + ], + "score": 1.0, + "content": "5.1 2D DENSITIES", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 524, + 505, + 634 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "We train diffusion models using both parameterizations and study the impact of various sampling", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 534, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 104, + 534, + 506, + 547 + ], + "score": 1.0, + "content": "approaches for compositional generation. Samples are evaluated using RAISE (Burda et al., 2015)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 307, + 558 + ], + "score": 1.0, + "content": "(which gives lower bounds on log-likelihood) and", + "type": "text" + }, + { + "bbox": [ + 307, + 545, + 337, + 556 + ], + "score": 0.71, + "content": "\\mathbf { M M D } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 545, + 505, + 558 + ], + "score": 1.0, + "content": ", LL (log-likelihood of generated samples", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "under composed distribution), and Var (L2 difference of variance of GMMs fit on generated samples", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "compared to GMMs of the composed distribution). Results can be found in Table 1 and visualizations", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "can be seen in Figure 2. All MCMC sampling methods improve sample quality and likelihood, with", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 589, + 506, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 506, + 602 + ], + "score": 1.0, + "content": "Metropolis adjusted methods performing the best. All MCMC experiments use the same number", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "of score function evaluations. We include a baseline, labeled “Reverse (equal steps)” which is a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 506, + 624 + ], + "score": 1.0, + "content": "diffusion model trained with more steps such that reverse diffusion sampling has the same cost as our", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 621, + 507, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 507, + 637 + ], + "score": 1.0, + "content": "MCMC samplers. We see that simply adding more time-steps does not solve compositional sampling.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5, + "bbox_fs": [ + 104, + 523, + 507, + 637 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 642, + 216, + 653 + ], + "lines": [ + { + "bbox": [ + 105, + 640, + 218, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 218, + 655 + ], + "score": 1.0, + "content": "5.2 COMPOSING CUBES", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 659, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 672 + ], + "score": 1.0, + "content": "Next, we train models on a dataset of images containing between 1 and 5 examples of various shapes", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 670, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 383, + 683 + ], + "score": 1.0, + "content": "taken from CLEVR (Johnson et al., 2017). We train our models to fit", + "type": "text" + }, + { + "bbox": [ + 383, + 670, + 411, + 682 + ], + "score": 0.93, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 670, + 438, + 683 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 439, + 672, + 446, + 682 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 670, + 506, + 683 + ], + "score": 1.0, + "content": "is the location", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 682, + 506, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 693 + ], + "score": 1.0, + "content": "of one of the shapes in the image. We then compose this conditional model with itself to create a", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 692, + 438, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 390, + 705 + ], + "score": 1.0, + "content": "product model which defines the distribution of images conditioned on", + "type": "text" + }, + { + "bbox": [ + 391, + 695, + 397, + 702 + ], + "score": 0.72, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 692, + 438, + 705 + ], + "score": 1.0, + "content": "shapes as", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 658, + 506, + 705 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 150, + 82, + 463, + 161 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 150, + 82, + 463, + 161 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 150, + 82, + 463, + 161 + ], + "spans": [ + { + "bbox": [ + 150, + 82, + 463, + 161 + ], + "score": 0.965, + "type": "image", + "image_path": "2cfc6ef18e34277e46354121c22d0afa7e309308989f6606b15d53e01dc2c92c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 150, + 82, + 463, + 108.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 150, + 108.33333333333333, + 463, + 134.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 150, + 134.66666666666666, + 463, + 161.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 170, + 505, + 191 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 169, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 505, + 182 + ], + "score": 1.0, + "content": "Figure 4: Classifier-guided generation on ImageNet. HMC leads to higher fidelity and more class-identified", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 179, + 260, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 260, + 192 + ], + "score": 1.0, + "content": "images than reverse diffusion sampling.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "table", + "bbox": [ + 297, + 196, + 487, + 279 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 297, + 196, + 487, + 279 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 297, + 196, + 487, + 279 + ], + "spans": [ + { + "bbox": [ + 297, + 196, + 487, + 279 + ], + "score": 0.958, + "html": "
ModelSamplerInception Score ↑FID↓Accuracy ↑
ScoreReverse29.1030.4618.64
LA29.3530.4965.81
U-HMC32.1926.8989.93
EBMReverse28.0533.5818.60
LA28.1233.4566.28
MALA30.4332.2283.65
U-HMC31.3932.0890.83
HMC33.4630.5294.61
", + "type": "table", + "image_path": "6c966eed623729080551929998da9a616b2e2daf0197f942ea9c5d641cb88931.jpg" + } + ] + } + ], + "index": 11.0, + "virtual_lines": [ + { + "bbox": [ + 297, + 196, + 487, + 209.83333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 297, + 209.83333333333334, + 487, + 223.66666666666669 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 297, + 223.66666666666669, + 487, + 237.50000000000003 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 297, + 237.50000000000003, + 487, + 251.33333333333337 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 297, + 251.33333333333337, + 487, + 265.1666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 297, + 265.1666666666667, + 487, + 279.0 + ], + "spans": [], + "index": 16 + } + ] + } + ], + "index": 11.0 + }, + { + "type": "table", + "bbox": [ + 117, + 196, + 273, + 280 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 117, + 196, + 273, + 280 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 196, + 273, + 280 + ], + "spans": [ + { + "bbox": [ + 117, + 196, + 273, + 280 + ], + "score": 0.967, + "html": "
ModelSamplerCombinations
12345
Reverse70.868.266.364.157.4
ScoreULA75.073.471.867.960.2
U-HMC79.176.073.671.162.3
Reverse71.067.162.558.151.0
ULA81.371.866.659.654.8
EBMMALA85.474.471.165.663.9
U-HMC84.581.379.274.268.1
HMC91.682.980.176.572.7
", + "type": "table", + "image_path": "3434619d2f0822114bb8ccdd5cc8723519895a9c96c0bd159ff61268db246fbf.jpg" + } + ] + } + ], + "index": 10.0, + "virtual_lines": [ + { + "bbox": [ + 117, + 196, + 273, + 210.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 117, + 210.0, + 273, + 224.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 117, + 224.0, + 273, + 238.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 117, + 238.0, + 273, + 252.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 117, + 252.0, + 273, + 266.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 117, + 266.0, + 273, + 280.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 284, + 287, + 304 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 283, + 288, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 288, + 295 + ], + "score": 1.0, + "content": "Table 2: MCMC Sampling enables more com-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 294, + 258, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 258, + 304 + ], + "score": 1.0, + "content": "positional cube generation on CLEVR.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + } + ], + "index": 13.75 + }, + { + "type": "title", + "bbox": [ + 296, + 283, + 492, + 303 + ], + "lines": [ + { + "bbox": [ + 295, + 282, + 493, + 294 + ], + "spans": [ + { + "bbox": [ + 295, + 282, + 493, + 294 + ], + "score": 1.0, + "content": "Table 3: MCMC Sampling enables better classifier", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 295, + 293, + 451, + 304 + ], + "spans": [ + { + "bbox": [ + 295, + 293, + 345, + 304 + ], + "score": 1.0, + "content": "guidance on", + "type": "text" + }, + { + "bbox": [ + 345, + 293, + 379, + 303 + ], + "score": 0.5, + "content": "{ \\bf 1 2 8 x 1 2 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 293, + 451, + 304 + ], + "score": 1.0, + "content": "ImageNet dataset.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "interline_equation", + "bbox": [ + 169, + 306, + 443, + 339 + ], + "lines": [ + { + "bbox": [ + 169, + 306, + 443, + 339 + ], + "spans": [ + { + "bbox": [ + 169, + 306, + 443, + 339 + ], + "score": 0.94, + "content": "\\log p _ { \\theta } ( x | y _ { 1 } , \\dots , y _ { c } ) = \\log p _ { \\theta } ( x ) + \\sum _ { i = 1 } ^ { c } \\left( \\log p _ { \\theta } ( x | y _ { i } ) - \\log p _ { \\theta } ( x ) \\right) .", + "type": "interline_equation", + "image_path": "8a4a6ce42887bd8ae6b78c9e1e720fdb36bb06692deacfcc58d44c8c4953d0c3.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 169, + 306, + 443, + 317.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 169, + 317.0, + 443, + 328.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 169, + 328.0, + 443, + 339.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 341, + 505, + 419 + ], + "lines": [ + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "score": 1.0, + "content": "We then sample using various methods, where for each number of combination of cubes, the same", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 352, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 505, + 365 + ], + "score": 1.0, + "content": "number of score function evaluations are used, and evaluate each by the fraction of samples which have", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "all objects placed in the correct location (as determined by a learned classifier). Results can be found", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 375, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 106, + 375, + 506, + 387 + ], + "score": 1.0, + "content": "in Table 2, where we see MCMC sampling leads to improvements and the Metropolis adjustment", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "enabled by the energy-based parameterization leads to further improvements. We qualitatively", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "illustrate results in Figure 3, and see more accurate generations with more steps of sampling, with", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 407, + 329, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 329, + 420 + ], + "score": 1.0, + "content": "more substantial increases with Metropolis adjustment.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 108, + 426, + 249, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 250, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 250, + 439 + ], + "score": 1.0, + "content": "5.3 CLASSIFIER CONDITIONING", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 109, + 443, + 504, + 466 + ], + "lines": [ + { + "bbox": [ + 107, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 107, + 443, + 505, + 456 + ], + "score": 1.0, + "content": "Next, we train unconditional diffusion models and a noise-conditioned classifier on ImageNet. We", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 107, + 455, + 210, + 467 + ], + "spans": [ + { + "bbox": [ + 107, + 455, + 210, + 467 + ], + "score": 1.0, + "content": "compose these models as", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 467, + 419, + 481 + ], + "lines": [ + { + "bbox": [ + 192, + 467, + 419, + 481 + ], + "spans": [ + { + "bbox": [ + 192, + 467, + 419, + 481 + ], + "score": 0.89, + "content": "\\nabla _ { \\boldsymbol { x } } \\log { p _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } | \\boldsymbol { y } , t ) } = \\nabla _ { \\boldsymbol { x } } \\log { p _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } | t ) } + \\nabla _ { \\boldsymbol { x } } \\log { p _ { \\boldsymbol { \\theta } } ( \\boldsymbol { y } | \\boldsymbol { x } , t ) } .", + "type": "interline_equation", + "image_path": "ecf67e67ac7e9d4648b86eca4f763be4b1596ff4e3c92b2932aef87f24373fcf.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 192, + 467, + 419, + 481 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 484, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "and sample using the corresponding score functions. We compare various samplers and model", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "score": 1.0, + "content": "parameterizations on classifier accuracy, FID (Heusel et al., 2017) and Inception Score. Quantitative", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 104, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "results can be seen in Table 3 and qualitative results seen in Figure 4. We find that MCMC improves", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 518, + 478, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 478, + 530 + ], + "score": 1.0, + "content": "performance over reverse sampling, with further improvements from Metropolis corrections.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "title", + "bbox": [ + 107, + 536, + 197, + 548 + ], + "lines": [ + { + "bbox": [ + 106, + 536, + 198, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 198, + 549 + ], + "score": 1.0, + "content": "5.4 TEXT-2-IMAGE", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 554, + 506, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 567 + ], + "score": 1.0, + "content": "Perhaps the most well-known results achieved with diffusion models are in text-to-image", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 564, + 507, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 442, + 578 + ], + "score": 1.0, + "content": "generation (Ramesh et al., 2022; Saharia et al., 2022). Here we model", + "type": "text" + }, + { + "bbox": [ + 443, + 565, + 503, + 577 + ], + "score": 0.91, + "content": "p _ { \\theta } \\mathrm { ( } x _ { \\mathrm { i m a g e } } | y _ { \\mathrm { t e x t } } \\mathrm { ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 564, + 507, + 578 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "While generated images generated are photo-realistic, they can fail to generate images", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 587, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 599 + ], + "score": 1.0, + "content": "from prompts which specify multiple concepts at a time (Liu et al., 2022) such as", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 148, + 610 + ], + "score": 0.9, + "content": "y _ { \\mathrm { t e x t } } = { } ^ { \\ast } \\mathrm { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "horse on a sandy beach or a grass plain on a not sunny day′′. To deal with these issues", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 312, + 622 + ], + "score": 1.0, + "content": "we can dissect the prompt into smaller components", + "type": "text" + }, + { + "bbox": [ + 312, + 611, + 354, + 621 + ], + "score": 0.9, + "content": "y _ { 1 } , \\ldots , y _ { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 608, + 505, + 622 + ], + "score": 1.0, + "content": ", parameterize models conditioned on", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 619, + 506, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 174, + 633 + ], + "score": 1.0, + "content": "each component", + "type": "text" + }, + { + "bbox": [ + 174, + 620, + 209, + 632 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x | y _ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 619, + 506, + 633 + ], + "score": 1.0, + "content": "and compose these models using our introduced operators. We can parse", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 630, + 201, + 643 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 201, + 643 + ], + "score": 1.0, + "content": "the above example into", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 156, + 646, + 454, + 659 + ], + "lines": [ + { + "bbox": [ + 157, + 646, + 454, + 660 + ], + "spans": [ + { + "bbox": [ + 157, + 646, + 454, + 660 + ], + "score": 1.0, + "content": "“A horse” AND (“A sandy beach” OR “Grass plains”) AND (NOT “Sunny”)", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 48 + }, + { + "type": "text", + "bbox": [ + 105, + 663, + 385, + 675 + ], + "lines": [ + { + "bbox": [ + 106, + 663, + 387, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 387, + 676 + ], + "score": 1.0, + "content": "which can be used to define the following (unnormalized) distribution", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 49 + }, + { + "type": "interline_equation", + "bbox": [ + 132, + 678, + 478, + 707 + ], + "lines": [ + { + "bbox": [ + 132, + 678, + 478, + 707 + ], + "spans": [ + { + "bbox": [ + 132, + 678, + 478, + 707 + ], + "score": 0.93, + "content": "p _ { \\theta } ^ { \\mathrm { c o m p } } ( x | y _ { \\mathrm { t e x t } } ) \\underset { \\sim } { \\sim } \\frac { p _ { \\theta } ( x | ^ { \\ast } \\mathrm { A } \\mathrm { h o r s e } ^ { \\cdot \\ast } ) \\left[ \\frac { 1 } { 2 } p _ { \\theta } ( x | ^ { \\ast } \\mathrm { A } \\mathrm { s a n d y } \\mathrm { b e a c h } ^ { \\cdot \\ast } ) + \\frac { 1 } { 2 } p _ { \\theta } ( x | ^ { \\ast } \\mathrm { G r a s s } \\mathrm { p l a i n s } ^ { \\cdot \\ast } ) \\right] } { p _ { \\theta } ( x | ^ { \\ast } \\mathrm { S u n n y } ^ { \\prime } ) ^ { \\alpha } }", + "type": "interline_equation", + "image_path": "8f5fb74146561a4c0217ce9088cf147a544ce42581031bdbdf42b24033ef2ed9.jpg" + } + ] + } + ], + "index": 51, + "virtual_lines": [ + { + "bbox": [ + 132, + 678, + 478, + 687.6666666666666 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 132, + 687.6666666666666, + 478, + 697.3333333333333 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 132, + 697.3333333333333, + 478, + 706.9999999999999 + ], + "spans": [], + "index": 52 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Liu et al. (2022) demonstrated that composing models this way can improve the efficacy of these", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 104, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 104, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "kinds of generations, but was restricted to composition using classifier-free guidance. We train a", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 53.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 150, + 82, + 463, + 161 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 150, + 82, + 463, + 161 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 150, + 82, + 463, + 161 + ], + "spans": [ + { + "bbox": [ + 150, + 82, + 463, + 161 + ], + "score": 0.965, + "type": "image", + "image_path": "2cfc6ef18e34277e46354121c22d0afa7e309308989f6606b15d53e01dc2c92c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 150, + 82, + 463, + 108.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 150, + 108.33333333333333, + 463, + 134.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 150, + 134.66666666666666, + 463, + 161.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 170, + 505, + 191 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 169, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 505, + 182 + ], + "score": 1.0, + "content": "Figure 4: Classifier-guided generation on ImageNet. HMC leads to higher fidelity and more class-identified", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 179, + 260, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 260, + 192 + ], + "score": 1.0, + "content": "images than reverse diffusion sampling.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "table", + "bbox": [ + 297, + 196, + 487, + 279 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 297, + 196, + 487, + 279 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 297, + 196, + 487, + 279 + ], + "spans": [ + { + "bbox": [ + 297, + 196, + 487, + 279 + ], + "score": 0.958, + "html": "
ModelSamplerInception Score ↑FID↓Accuracy ↑
ScoreReverse29.1030.4618.64
LA29.3530.4965.81
U-HMC32.1926.8989.93
EBMReverse28.0533.5818.60
LA28.1233.4566.28
MALA30.4332.2283.65
U-HMC31.3932.0890.83
HMC33.4630.5294.61
", + "type": "table", + "image_path": "6c966eed623729080551929998da9a616b2e2daf0197f942ea9c5d641cb88931.jpg" + } + ] + } + ], + "index": 11.0, + "virtual_lines": [ + { + "bbox": [ + 297, + 196, + 487, + 209.83333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 297, + 209.83333333333334, + 487, + 223.66666666666669 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 297, + 223.66666666666669, + 487, + 237.50000000000003 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 297, + 237.50000000000003, + 487, + 251.33333333333337 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 297, + 251.33333333333337, + 487, + 265.1666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 297, + 265.1666666666667, + 487, + 279.0 + ], + "spans": [], + "index": 16 + } + ] + } + ], + "index": 11.0 + }, + { + "type": "table", + "bbox": [ + 117, + 196, + 273, + 280 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 117, + 196, + 273, + 280 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 196, + 273, + 280 + ], + "spans": [ + { + "bbox": [ + 117, + 196, + 273, + 280 + ], + "score": 0.967, + "html": "
ModelSamplerCombinations
12345
Reverse70.868.266.364.157.4
ScoreULA75.073.471.867.960.2
U-HMC79.176.073.671.162.3
Reverse71.067.162.558.151.0
ULA81.371.866.659.654.8
EBMMALA85.474.471.165.663.9
U-HMC84.581.379.274.268.1
HMC91.682.980.176.572.7
", + "type": "table", + "image_path": "3434619d2f0822114bb8ccdd5cc8723519895a9c96c0bd159ff61268db246fbf.jpg" + } + ] + } + ], + "index": 10.0, + "virtual_lines": [ + { + "bbox": [ + 117, + 196, + 273, + 210.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 117, + 210.0, + 273, + 224.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 117, + 224.0, + 273, + 238.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 117, + 238.0, + 273, + 252.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 117, + 252.0, + 273, + 266.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 117, + 266.0, + 273, + 280.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 284, + 287, + 304 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 283, + 288, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 288, + 295 + ], + "score": 1.0, + "content": "Table 2: MCMC Sampling enables more com-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 294, + 258, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 258, + 304 + ], + "score": 1.0, + "content": "positional cube generation on CLEVR.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + } + ], + "index": 13.75 + }, + { + "type": "title", + "bbox": [ + 296, + 283, + 492, + 303 + ], + "lines": [ + { + "bbox": [ + 295, + 282, + 493, + 294 + ], + "spans": [ + { + "bbox": [ + 295, + 282, + 493, + 294 + ], + "score": 1.0, + "content": "Table 3: MCMC Sampling enables better classifier", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 295, + 293, + 451, + 304 + ], + "spans": [ + { + "bbox": [ + 295, + 293, + 345, + 304 + ], + "score": 1.0, + "content": "guidance on", + "type": "text" + }, + { + "bbox": [ + 345, + 293, + 379, + 303 + ], + "score": 0.5, + "content": "{ \\bf 1 2 8 x 1 2 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 293, + 451, + 304 + ], + "score": 1.0, + "content": "ImageNet dataset.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "interline_equation", + "bbox": [ + 169, + 306, + 443, + 339 + ], + "lines": [ + { + "bbox": [ + 169, + 306, + 443, + 339 + ], + "spans": [ + { + "bbox": [ + 169, + 306, + 443, + 339 + ], + "score": 0.94, + "content": "\\log p _ { \\theta } ( x | y _ { 1 } , \\dots , y _ { c } ) = \\log p _ { \\theta } ( x ) + \\sum _ { i = 1 } ^ { c } \\left( \\log p _ { \\theta } ( x | y _ { i } ) - \\log p _ { \\theta } ( x ) \\right) .", + "type": "interline_equation", + "image_path": "8a4a6ce42887bd8ae6b78c9e1e720fdb36bb06692deacfcc58d44c8c4953d0c3.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 169, + 306, + 443, + 317.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 169, + 317.0, + 443, + 328.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 169, + 328.0, + 443, + 339.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 341, + 505, + 419 + ], + "lines": [ + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "score": 1.0, + "content": "We then sample using various methods, where for each number of combination of cubes, the same", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 352, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 505, + 365 + ], + "score": 1.0, + "content": "number of score function evaluations are used, and evaluate each by the fraction of samples which have", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "all objects placed in the correct location (as determined by a learned classifier). Results can be found", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 375, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 106, + 375, + 506, + 387 + ], + "score": 1.0, + "content": "in Table 2, where we see MCMC sampling leads to improvements and the Metropolis adjustment", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "enabled by the energy-based parameterization leads to further improvements. We qualitatively", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "illustrate results in Figure 3, and see more accurate generations with more steps of sampling, with", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 407, + 329, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 329, + 420 + ], + "score": 1.0, + "content": "more substantial increases with Metropolis adjustment.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 341, + 506, + 420 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 426, + 249, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 250, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 250, + 439 + ], + "score": 1.0, + "content": "5.3 CLASSIFIER CONDITIONING", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 109, + 443, + 504, + 466 + ], + "lines": [ + { + "bbox": [ + 107, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 107, + 443, + 505, + 456 + ], + "score": 1.0, + "content": "Next, we train unconditional diffusion models and a noise-conditioned classifier on ImageNet. We", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 107, + 455, + 210, + 467 + ], + "spans": [ + { + "bbox": [ + 107, + 455, + 210, + 467 + ], + "score": 1.0, + "content": "compose these models as", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 107, + 443, + 505, + 467 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 467, + 419, + 481 + ], + "lines": [ + { + "bbox": [ + 192, + 467, + 419, + 481 + ], + "spans": [ + { + "bbox": [ + 192, + 467, + 419, + 481 + ], + "score": 0.89, + "content": "\\nabla _ { \\boldsymbol { x } } \\log { p _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } | \\boldsymbol { y } , t ) } = \\nabla _ { \\boldsymbol { x } } \\log { p _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } | t ) } + \\nabla _ { \\boldsymbol { x } } \\log { p _ { \\boldsymbol { \\theta } } ( \\boldsymbol { y } | \\boldsymbol { x } , t ) } .", + "type": "interline_equation", + "image_path": "ecf67e67ac7e9d4648b86eca4f763be4b1596ff4e3c92b2932aef87f24373fcf.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 192, + 467, + 419, + 481 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 484, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "and sample using the corresponding score functions. We compare various samplers and model", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "score": 1.0, + "content": "parameterizations on classifier accuracy, FID (Heusel et al., 2017) and Inception Score. Quantitative", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 104, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "results can be seen in Table 3 and qualitative results seen in Figure 4. We find that MCMC improves", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 518, + 478, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 478, + 530 + ], + "score": 1.0, + "content": "performance over reverse sampling, with further improvements from Metropolis corrections.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5, + "bbox_fs": [ + 104, + 484, + 506, + 530 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 536, + 197, + 548 + ], + "lines": [ + { + "bbox": [ + 106, + 536, + 198, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 198, + 549 + ], + "score": 1.0, + "content": "5.4 TEXT-2-IMAGE", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 554, + 506, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 567 + ], + "score": 1.0, + "content": "Perhaps the most well-known results achieved with diffusion models are in text-to-image", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 564, + 507, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 442, + 578 + ], + "score": 1.0, + "content": "generation (Ramesh et al., 2022; Saharia et al., 2022). Here we model", + "type": "text" + }, + { + "bbox": [ + 443, + 565, + 503, + 577 + ], + "score": 0.91, + "content": "p _ { \\theta } \\mathrm { ( } x _ { \\mathrm { i m a g e } } | y _ { \\mathrm { t e x t } } \\mathrm { ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 564, + 507, + 578 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "While generated images generated are photo-realistic, they can fail to generate images", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 587, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 599 + ], + "score": 1.0, + "content": "from prompts which specify multiple concepts at a time (Liu et al., 2022) such as", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 148, + 610 + ], + "score": 0.9, + "content": "y _ { \\mathrm { t e x t } } = { } ^ { \\ast } \\mathrm { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "horse on a sandy beach or a grass plain on a not sunny day′′. To deal with these issues", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 312, + 622 + ], + "score": 1.0, + "content": "we can dissect the prompt into smaller components", + "type": "text" + }, + { + "bbox": [ + 312, + 611, + 354, + 621 + ], + "score": 0.9, + "content": "y _ { 1 } , \\ldots , y _ { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 608, + 505, + 622 + ], + "score": 1.0, + "content": ", parameterize models conditioned on", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 619, + 506, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 174, + 633 + ], + "score": 1.0, + "content": "each component", + "type": "text" + }, + { + "bbox": [ + 174, + 620, + 209, + 632 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x | y _ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 619, + 506, + 633 + ], + "score": 1.0, + "content": "and compose these models using our introduced operators. We can parse", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 630, + 201, + 643 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 201, + 643 + ], + "score": 1.0, + "content": "the above example into", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 552, + 507, + 643 + ] + }, + { + "type": "text", + "bbox": [ + 156, + 646, + 454, + 659 + ], + "lines": [ + { + "bbox": [ + 157, + 646, + 454, + 660 + ], + "spans": [ + { + "bbox": [ + 157, + 646, + 454, + 660 + ], + "score": 1.0, + "content": "“A horse” AND (“A sandy beach” OR “Grass plains”) AND (NOT “Sunny”)", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 48, + "bbox_fs": [ + 157, + 646, + 454, + 660 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 663, + 385, + 675 + ], + "lines": [ + { + "bbox": [ + 106, + 663, + 387, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 387, + 676 + ], + "score": 1.0, + "content": "which can be used to define the following (unnormalized) distribution", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 49, + "bbox_fs": [ + 106, + 663, + 387, + 676 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 132, + 678, + 478, + 707 + ], + "lines": [ + { + "bbox": [ + 132, + 678, + 478, + 707 + ], + "spans": [ + { + "bbox": [ + 132, + 678, + 478, + 707 + ], + "score": 0.93, + "content": "p _ { \\theta } ^ { \\mathrm { c o m p } } ( x | y _ { \\mathrm { t e x t } } ) \\underset { \\sim } { \\sim } \\frac { p _ { \\theta } ( x | ^ { \\ast } \\mathrm { A } \\mathrm { h o r s e } ^ { \\cdot \\ast } ) \\left[ \\frac { 1 } { 2 } p _ { \\theta } ( x | ^ { \\ast } \\mathrm { A } \\mathrm { s a n d y } \\mathrm { b e a c h } ^ { \\cdot \\ast } ) + \\frac { 1 } { 2 } p _ { \\theta } ( x | ^ { \\ast } \\mathrm { G r a s s } \\mathrm { p l a i n s } ^ { \\cdot \\ast } ) \\right] } { p _ { \\theta } ( x | ^ { \\ast } \\mathrm { S u n n y } ^ { \\prime } ) ^ { \\alpha } }", + "type": "interline_equation", + "image_path": "8f5fb74146561a4c0217ce9088cf147a544ce42581031bdbdf42b24033ef2ed9.jpg" + } + ] + } + ], + "index": 51, + "virtual_lines": [ + { + "bbox": [ + 132, + 678, + 478, + 687.6666666666666 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 132, + 687.6666666666666, + 478, + 697.3333333333333 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 132, + 697.3333333333333, + 478, + 706.9999999999999 + ], + "spans": [], + "index": 52 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Liu et al. (2022) demonstrated that composing models this way can improve the efficacy of these", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 104, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 104, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "kinds of generations, but was restricted to composition using classifier-free guidance. We train a", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 53.5, + "bbox_fs": [ + 104, + 709, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 135, + 83, + 468, + 234 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 135, + 83, + 468, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 135, + 83, + 468, + 234 + ], + "spans": [ + { + "bbox": [ + 135, + 83, + 468, + 234 + ], + "score": 0.974, + "type": "image", + "image_path": "2a15afd56676a0d2412690f421933cf678a1b434bcb3b129ab0f2d857d5a134b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 135, + 83, + 468, + 133.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 135, + 133.33333333333334, + 468, + 183.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 135, + 183.66666666666669, + 468, + 234.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 108, + 236, + 502, + 247 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 235, + 504, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 235, + 504, + 248 + ], + "score": 1.0, + "content": "Figure 5: Energy based parameterization enables high-resolution compositional text-to-image synthesis.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 107, + 252, + 504, + 285 + ], + "lines": [ + { + "bbox": [ + 105, + 252, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 265 + ], + "score": 1.0, + "content": "energy-parameterized diffusion model for text conditional 64x64 image generation and illustrate", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 262, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 288, + 276 + ], + "score": 1.0, + "content": "composed results in Figure 5 (upsampled to", + "type": "text" + }, + { + "bbox": [ + 288, + 263, + 336, + 274 + ], + "score": 0.73, + "content": "1 0 2 4 \\mathrm { x } 1 0 2 4 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 262, + 505, + 276 + ], + "score": 1.0, + "content": "). We find that composition enables more", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 274, + 408, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 408, + 286 + ], + "score": 1.0, + "content": "faithful generations of scenes in Figure 6 with more results in Appendix A.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "image", + "bbox": [ + 140, + 292, + 473, + 425 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 140, + 292, + 473, + 425 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 140, + 292, + 473, + 425 + ], + "spans": [ + { + "bbox": [ + 140, + 292, + 473, + 425 + ], + "score": 0.972, + "type": "image", + "image_path": "a8563d2ff94e18860887b533bd40d910214c5ca2c2780232891039d74b2ddf8c.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 140, + 292, + 473, + 336.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 140, + 336.3333333333333, + 473, + 380.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 140, + 380.66666666666663, + 473, + 424.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 160, + 429, + 454, + 440 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 158, + 428, + 456, + 442 + ], + "spans": [ + { + "bbox": [ + 158, + 428, + 456, + 442 + ], + "score": 1.0, + "content": "Figure 6: Composing text descriptions enables more accurate scene generation.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "index": 9.0 + }, + { + "type": "title", + "bbox": [ + 107, + 451, + 191, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 449, + 192, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 192, + 466 + ], + "score": 1.0, + "content": "6 DISCUSSION", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 472, + 506, + 561 + ], + "lines": [ + { + "bbox": [ + 106, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "Limitations. Our work demonstrates that diffusion models, in combination with MCMC-based", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 484, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 506, + 496 + ], + "score": 1.0, + "content": "sampling procedures, can be composed in novel ways capable of generating high-quality samples.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 494, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 506 + ], + "score": 1.0, + "content": "However, our proposed solutions have a number of drawbacks. First, more sophisticated MCMC", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 506, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 506, + 518 + ], + "score": 1.0, + "content": "samplers come at a higher cost than the standard sampling approach and can take 5-times longer to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "score": 1.0, + "content": "generate samples than typical diffusion sampling. Second, we have shown that energy-parameterized", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 527, + 507, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 507, + 541 + ], + "score": 1.0, + "content": "models enable the use of more sophisticated sampling techniques, garnering further improvements.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "Unfortunately, this requires a second backward-pass through the model to compute the derivative", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 549, + 507, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 507, + 561 + ], + "score": 1.0, + "content": "implicitly, leading them to have double the memory and compute cost of score-parameterized models.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 105, + 566, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 577 + ], + "score": 1.0, + "content": "While these are considerable drawbacks, we note the focus of this work is to demonstrate that such", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "score": 1.0, + "content": "things are possible within the framework of diffusion models. We believe there is much that can be", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 588, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 600 + ], + "score": 1.0, + "content": "done to achieve the benefits of our sampling procedures at less cost such as distillation (Salimans &", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 599, + 431, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 431, + 611 + ], + "score": 1.0, + "content": "Ho, 2022) and easier-to-differentiate neural networks (Chen & Duvenaud, 2019).", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 616, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "Finally we note that not all models can be effectively composed together. For example if we wanted", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 205, + 639 + ], + "score": 1.0, + "content": "to model the product of", + "type": "text" + }, + { + "bbox": [ + 206, + 628, + 288, + 639 + ], + "score": 0.94, + "content": "\\mathcal { N } ( - 1 0 , 1 ) \\mathcal { N } ( 1 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 626, + 505, + 639 + ], + "score": 1.0, + "content": ", the resulting distribution’s support is far outside the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "support of the constituent models. To accurately model this, our constituent models would need to be", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "near-perfect far outside the training distribution. Thus it is unlikely for good results to be obtained.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 506, + 673 + ], + "score": 1.0, + "content": "The same care should be taken in the text-2-image setting with, for example, contradicting prompts.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Conclusion. In this work we have explored the ways that pretrained diffusion models can be", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "composed to model new distributions. We demonstrate ways that na¨ıve implementations fail, and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 698, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 713 + ], + "score": 1.0, + "content": "present two ways that performance can be improved: MCMC sampling and energy-parameterized", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "diffusion models. We find our proposed methods lead to notable improvement across a variety of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 720, + 328, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 328, + 733 + ], + "score": 1.0, + "content": "domains, scales, and different compositional operators.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 135, + 83, + 468, + 234 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 135, + 83, + 468, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 135, + 83, + 468, + 234 + ], + "spans": [ + { + "bbox": [ + 135, + 83, + 468, + 234 + ], + "score": 0.974, + "type": "image", + "image_path": "2a15afd56676a0d2412690f421933cf678a1b434bcb3b129ab0f2d857d5a134b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 135, + 83, + 468, + 133.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 135, + 133.33333333333334, + 468, + 183.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 135, + 183.66666666666669, + 468, + 234.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 108, + 236, + 502, + 247 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 235, + 504, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 235, + 504, + 248 + ], + "score": 1.0, + "content": "Figure 5: Energy based parameterization enables high-resolution compositional text-to-image synthesis.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 107, + 252, + 504, + 285 + ], + "lines": [ + { + "bbox": [ + 105, + 252, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 265 + ], + "score": 1.0, + "content": "energy-parameterized diffusion model for text conditional 64x64 image generation and illustrate", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 262, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 288, + 276 + ], + "score": 1.0, + "content": "composed results in Figure 5 (upsampled to", + "type": "text" + }, + { + "bbox": [ + 288, + 263, + 336, + 274 + ], + "score": 0.73, + "content": "1 0 2 4 \\mathrm { x } 1 0 2 4 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 262, + 505, + 276 + ], + "score": 1.0, + "content": "). We find that composition enables more", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 274, + 408, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 408, + 286 + ], + "score": 1.0, + "content": "faithful generations of scenes in Figure 6 with more results in Appendix A.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 252, + 506, + 286 + ] + }, + { + "type": "image", + "bbox": [ + 140, + 292, + 473, + 425 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 140, + 292, + 473, + 425 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 140, + 292, + 473, + 425 + ], + "spans": [ + { + "bbox": [ + 140, + 292, + 473, + 425 + ], + "score": 0.972, + "type": "image", + "image_path": "a8563d2ff94e18860887b533bd40d910214c5ca2c2780232891039d74b2ddf8c.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 140, + 292, + 473, + 336.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 140, + 336.3333333333333, + 473, + 380.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 140, + 380.66666666666663, + 473, + 424.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 160, + 429, + 454, + 440 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 158, + 428, + 456, + 442 + ], + "spans": [ + { + "bbox": [ + 158, + 428, + 456, + 442 + ], + "score": 1.0, + "content": "Figure 6: Composing text descriptions enables more accurate scene generation.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "index": 9.0 + }, + { + "type": "title", + "bbox": [ + 107, + 451, + 191, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 449, + 192, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 192, + 466 + ], + "score": 1.0, + "content": "6 DISCUSSION", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 472, + 506, + 561 + ], + "lines": [ + { + "bbox": [ + 106, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "Limitations. Our work demonstrates that diffusion models, in combination with MCMC-based", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 484, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 506, + 496 + ], + "score": 1.0, + "content": "sampling procedures, can be composed in novel ways capable of generating high-quality samples.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 494, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 506 + ], + "score": 1.0, + "content": "However, our proposed solutions have a number of drawbacks. First, more sophisticated MCMC", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 506, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 506, + 518 + ], + "score": 1.0, + "content": "samplers come at a higher cost than the standard sampling approach and can take 5-times longer to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 529 + ], + "score": 1.0, + "content": "generate samples than typical diffusion sampling. Second, we have shown that energy-parameterized", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 527, + 507, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 507, + 541 + ], + "score": 1.0, + "content": "models enable the use of more sophisticated sampling techniques, garnering further improvements.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "Unfortunately, this requires a second backward-pass through the model to compute the derivative", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 549, + 507, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 507, + 561 + ], + "score": 1.0, + "content": "implicitly, leading them to have double the memory and compute cost of score-parameterized models.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 471, + 507, + 561 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 105, + 566, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 577 + ], + "score": 1.0, + "content": "While these are considerable drawbacks, we note the focus of this work is to demonstrate that such", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "score": 1.0, + "content": "things are possible within the framework of diffusion models. We believe there is much that can be", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 588, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 600 + ], + "score": 1.0, + "content": "done to achieve the benefits of our sampling procedures at less cost such as distillation (Salimans &", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 599, + 431, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 431, + 611 + ], + "score": 1.0, + "content": "Ho, 2022) and easier-to-differentiate neural networks (Chen & Duvenaud, 2019).", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 566, + 506, + 611 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 616, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "Finally we note that not all models can be effectively composed together. For example if we wanted", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 205, + 639 + ], + "score": 1.0, + "content": "to model the product of", + "type": "text" + }, + { + "bbox": [ + 206, + 628, + 288, + 639 + ], + "score": 0.94, + "content": "\\mathcal { N } ( - 1 0 , 1 ) \\mathcal { N } ( 1 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 626, + 505, + 639 + ], + "score": 1.0, + "content": ", the resulting distribution’s support is far outside the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "support of the constituent models. To accurately model this, our constituent models would need to be", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "near-perfect far outside the training distribution. Thus it is unlikely for good results to be obtained.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 506, + 673 + ], + "score": 1.0, + "content": "The same care should be taken in the text-2-image setting with, for example, contradicting prompts.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 615, + 506, + 673 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Conclusion. In this work we have explored the ways that pretrained diffusion models can be", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "composed to model new distributions. We demonstrate ways that na¨ıve implementations fail, and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 698, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 713 + ], + "score": 1.0, + "content": "present two ways that performance can be improved: MCMC sampling and energy-parameterized", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "diffusion models. We find our proposed methods lead to notable improvement across a variety of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 720, + 328, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 328, + 733 + ], + "score": 1.0, + "content": "domains, scales, and different compositional operators.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 677, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 175, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 106, + 504, + 130 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "Julian Besag. Comments on “Representations of knowledge in complex systems” by U. Grenander", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 117, + 382, + 131 + ], + "spans": [ + { + "bbox": [ + 115, + 117, + 382, + 131 + ], + "score": 1.0, + "content": "and MI Miller. J. Roy. Statist. Soc. Ser. B, 56(591-592):4, 1994. 3", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 106, + 137, + 504, + 160 + ], + "lines": [ + { + "bbox": [ + 106, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 106, + 136, + 505, + 151 + ], + "score": 1.0, + "content": "Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale GAN training for high fidelity", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 148, + 387, + 160 + ], + "spans": [ + { + "bbox": [ + 115, + 148, + 387, + 160 + ], + "score": 1.0, + "content": "natural image synthesis. arXiv preprint arXiv:1809.11096, 2018. 1", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 108, + 167, + 504, + 202 + ], + "lines": [ + { + "bbox": [ + 106, + 167, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 506, + 180 + ], + "score": 1.0, + "content": "Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 179, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 116, + 179, + 505, + 191 + ], + "score": 1.0, + "content": "Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 189, + 502, + 202 + ], + "spans": [ + { + "bbox": [ + 115, + 189, + 502, + 202 + ], + "score": 1.0, + "content": "few-shot learners. Advances in Neural Information Processing Systems, 33:1877–1901, 2020. 1", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 208, + 505, + 243 + ], + "lines": [ + { + "bbox": [ + 107, + 209, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 107, + 209, + 506, + 221 + ], + "score": 1.0, + "content": "Yuri Burda, Roger Grosse, and Ruslan Salakhutdinov. Accurate and conservative estimates of mrf", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 220, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 116, + 220, + 506, + 233 + ], + "score": 1.0, + "content": "log-likelihood using reverse annealing. In Artificial Intelligence and Statistics, pp. 102–110. PMLR,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 230, + 151, + 243 + ], + "spans": [ + { + "bbox": [ + 115, + 230, + 151, + 243 + ], + "score": 1.0, + "content": "2015. 7", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 105, + 250, + 504, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 249, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 506, + 264 + ], + "score": 1.0, + "content": "Ricky TQ Chen and David K Duvenaud. Neural networks with cheap differential operators. Advances", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 261, + 339, + 274 + ], + "spans": [ + { + "bbox": [ + 115, + 261, + 339, + 274 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems, 32, 2019. 9", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 107, + 280, + 504, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "score": 1.0, + "content": "Prafulla Dhariwal and Alexander Quinn Nichol. Diffusion models beat GANs on image synthesis. In", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 291, + 375, + 304 + ], + "spans": [ + { + "bbox": [ + 114, + 291, + 375, + 304 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2021. 1, 4", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 107, + 311, + 504, + 334 + ], + "lines": [ + { + "bbox": [ + 106, + 311, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 506, + 324 + ], + "score": 1.0, + "content": "Yilun Du and Igor Mordatch. Implicit generation and generalization in energy-based models. arXiv", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 114, + 322, + 288, + 335 + ], + "spans": [ + { + "bbox": [ + 114, + 322, + 288, + 335 + ], + "score": 1.0, + "content": "preprint arXiv:1903.08689, 2019. 3, 5, 18", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 342, + 504, + 365 + ], + "lines": [ + { + "bbox": [ + 107, + 342, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 107, + 342, + 506, + 354 + ], + "score": 1.0, + "content": "Yilun Du, Shuang Li, and Igor Mordatch. Compositional visual generation with energy based models.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 114, + 353, + 441, + 365 + ], + "spans": [ + { + "bbox": [ + 114, + 353, + 441, + 365 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 33:6637–6647, 2020a. 4, 5", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 107, + 371, + 503, + 395 + ], + "lines": [ + { + "bbox": [ + 106, + 371, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 505, + 385 + ], + "score": 1.0, + "content": "Yilun Du, Shuang Li, Joshua Tenenbaum, and Igor Mordatch. Improved contrastive divergence", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 383, + 427, + 396 + ], + "spans": [ + { + "bbox": [ + 116, + 383, + 427, + 396 + ], + "score": 1.0, + "content": "training of energy based models. arXiv preprint arXiv:2012.01316, 2020b. 5", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 108, + 402, + 503, + 425 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "Simon Duane, Anthony D Kennedy, Brian J Pendleton, and Duncan Roweth. Hybrid Monte Carlo.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 413, + 291, + 426 + ], + "spans": [ + { + "bbox": [ + 115, + 413, + 291, + 426 + ], + "score": 1.0, + "content": "Physics letters B, 195(2):216–222, 1987. 3", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 432, + 504, + 456 + ], + "lines": [ + { + "bbox": [ + 105, + 432, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 505, + 446 + ], + "score": 1.0, + "content": "Alain Durmus and Eric Moulines. High-dimensional Bayesian inference via the unadjusted Langevin", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 444, + 321, + 456 + ], + "spans": [ + { + "bbox": [ + 116, + 444, + 321, + 456 + ], + "score": 1.0, + "content": "algorithm. Bernoulli, 25(4A):2854–2882, 2019. 6", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 462, + 504, + 487 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 475 + ], + "score": 1.0, + "content": "Tomas Geffner and Justin Domke. MCMC variational inference via uncorrected Hamiltonian", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 475, + 462, + 487 + ], + "spans": [ + { + "bbox": [ + 115, + 475, + 462, + 487 + ], + "score": 1.0, + "content": "annealing. Advances in Neural Information Processing Systems, 34:639–651, 2021. 3", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 505, + 528 + ], + "lines": [ + { + "bbox": [ + 105, + 493, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 506, + 507 + ], + "score": 1.0, + "content": "Will Grathwohl, Kuan-Chieh Wang, Jorn-Henrik Jacobsen, David Duvenaud, Mohammad Norouzi, ¨", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 115, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "and Kevin Swersky. Your classifier is secretly an energy based model and you should treat it like", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 516, + 309, + 528 + ], + "spans": [ + { + "bbox": [ + 115, + 516, + 309, + 528 + ], + "score": 1.0, + "content": "one. arXiv preprint arXiv:1912.03263, 2019. 5", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 108, + 534, + 505, + 569 + ], + "lines": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "Will Grathwohl, Jacob Jin Kelly, Milad Hashemi, Mohammad Norouzi, Kevin Swersky, and David", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 547, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 115, + 547, + 505, + 559 + ], + "score": 1.0, + "content": "Duvenaud. No MCMC for me: Amortized sampling for fast and stable training of energy-based", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 557, + 420, + 570 + ], + "spans": [ + { + "bbox": [ + 115, + 557, + 420, + 570 + ], + "score": 1.0, + "content": "models. In International Conference on Learning Representations, 2021. 5", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 576, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 588, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 116, + 588, + 505, + 600 + ], + "score": 1.0, + "content": "trained by a two time-scale update rule converge to a local Nash equilibrium. Advances in Neural", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 599, + 299, + 611 + ], + "spans": [ + { + "bbox": [ + 115, + 599, + 299, + 611 + ], + "score": 1.0, + "content": "Information Processing Systems, 30, 2017. 8", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 618, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 617, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 506, + 631 + ], + "score": 1.0, + "content": "Geoffrey E Hinton. Training products of experts by minimizing contrastive divergence. Neural", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 628, + 302, + 641 + ], + "spans": [ + { + "bbox": [ + 116, + 628, + 302, + 641 + ], + "score": 1.0, + "content": "Computation, 14(8):1771–1800, 2002. 1, 2, 4", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 104, + 648, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 114, + 658, + 160, + 670 + ], + "spans": [ + { + "bbox": [ + 114, + 658, + 160, + 670 + ], + "score": 1.0, + "content": "2022. 1, 4", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 106, + 678, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 690, + 405, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 405, + 703 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, 33:6840–6851, 2020. 1, 20, 21", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. Adaptive mixtures of", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 720, + 343, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 343, + 733 + ], + "score": 1.0, + "content": "local experts. Neural Computation, 3(1):79–87, 1991. 1", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 175, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 106, + 504, + 130 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "Julian Besag. Comments on “Representations of knowledge in complex systems” by U. Grenander", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 117, + 382, + 131 + ], + "spans": [ + { + "bbox": [ + 115, + 117, + 382, + 131 + ], + "score": 1.0, + "content": "and MI Miller. J. Roy. Statist. Soc. Ser. B, 56(591-592):4, 1994. 3", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 106, + 106, + 505, + 131 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 137, + 504, + 160 + ], + "lines": [ + { + "bbox": [ + 106, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 106, + 136, + 505, + 151 + ], + "score": 1.0, + "content": "Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale GAN training for high fidelity", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 148, + 387, + 160 + ], + "spans": [ + { + "bbox": [ + 115, + 148, + 387, + 160 + ], + "score": 1.0, + "content": "natural image synthesis. arXiv preprint arXiv:1809.11096, 2018. 1", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 106, + 136, + 505, + 160 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 167, + 504, + 202 + ], + "lines": [ + { + "bbox": [ + 106, + 167, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 506, + 180 + ], + "score": 1.0, + "content": "Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 179, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 116, + 179, + 505, + 191 + ], + "score": 1.0, + "content": "Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 189, + 502, + 202 + ], + "spans": [ + { + "bbox": [ + 115, + 189, + 502, + 202 + ], + "score": 1.0, + "content": "few-shot learners. Advances in Neural Information Processing Systems, 33:1877–1901, 2020. 1", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 106, + 167, + 506, + 202 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 208, + 505, + 243 + ], + "lines": [ + { + "bbox": [ + 107, + 209, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 107, + 209, + 506, + 221 + ], + "score": 1.0, + "content": "Yuri Burda, Roger Grosse, and Ruslan Salakhutdinov. Accurate and conservative estimates of mrf", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 220, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 116, + 220, + 506, + 233 + ], + "score": 1.0, + "content": "log-likelihood using reverse annealing. In Artificial Intelligence and Statistics, pp. 102–110. PMLR,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 230, + 151, + 243 + ], + "spans": [ + { + "bbox": [ + 115, + 230, + 151, + 243 + ], + "score": 1.0, + "content": "2015. 7", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 107, + 209, + 506, + 243 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 250, + 504, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 249, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 506, + 264 + ], + "score": 1.0, + "content": "Ricky TQ Chen and David K Duvenaud. Neural networks with cheap differential operators. Advances", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 261, + 339, + 274 + ], + "spans": [ + { + "bbox": [ + 115, + 261, + 339, + 274 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems, 32, 2019. 9", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 249, + 506, + 274 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 280, + 504, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "score": 1.0, + "content": "Prafulla Dhariwal and Alexander Quinn Nichol. Diffusion models beat GANs on image synthesis. In", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 291, + 375, + 304 + ], + "spans": [ + { + "bbox": [ + 114, + 291, + 375, + 304 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2021. 1, 4", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 280, + 505, + 304 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 311, + 504, + 334 + ], + "lines": [ + { + "bbox": [ + 106, + 311, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 506, + 324 + ], + "score": 1.0, + "content": "Yilun Du and Igor Mordatch. Implicit generation and generalization in energy-based models. arXiv", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 114, + 322, + 288, + 335 + ], + "spans": [ + { + "bbox": [ + 114, + 322, + 288, + 335 + ], + "score": 1.0, + "content": "preprint arXiv:1903.08689, 2019. 3, 5, 18", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 106, + 311, + 506, + 335 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 342, + 504, + 365 + ], + "lines": [ + { + "bbox": [ + 107, + 342, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 107, + 342, + 506, + 354 + ], + "score": 1.0, + "content": "Yilun Du, Shuang Li, and Igor Mordatch. Compositional visual generation with energy based models.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 114, + 353, + 441, + 365 + ], + "spans": [ + { + "bbox": [ + 114, + 353, + 441, + 365 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 33:6637–6647, 2020a. 4, 5", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 107, + 342, + 506, + 365 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 371, + 503, + 395 + ], + "lines": [ + { + "bbox": [ + 106, + 371, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 505, + 385 + ], + "score": 1.0, + "content": "Yilun Du, Shuang Li, Joshua Tenenbaum, and Igor Mordatch. Improved contrastive divergence", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 383, + 427, + 396 + ], + "spans": [ + { + "bbox": [ + 116, + 383, + 427, + 396 + ], + "score": 1.0, + "content": "training of energy based models. arXiv preprint arXiv:2012.01316, 2020b. 5", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 371, + 505, + 396 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 402, + 503, + 425 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "Simon Duane, Anthony D Kennedy, Brian J Pendleton, and Duncan Roweth. Hybrid Monte Carlo.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 413, + 291, + 426 + ], + "spans": [ + { + "bbox": [ + 115, + 413, + 291, + 426 + ], + "score": 1.0, + "content": "Physics letters B, 195(2):216–222, 1987. 3", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 106, + 403, + 505, + 426 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 432, + 504, + 456 + ], + "lines": [ + { + "bbox": [ + 105, + 432, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 505, + 446 + ], + "score": 1.0, + "content": "Alain Durmus and Eric Moulines. High-dimensional Bayesian inference via the unadjusted Langevin", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 444, + 321, + 456 + ], + "spans": [ + { + "bbox": [ + 116, + 444, + 321, + 456 + ], + "score": 1.0, + "content": "algorithm. Bernoulli, 25(4A):2854–2882, 2019. 6", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 432, + 505, + 456 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 462, + 504, + 487 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 475 + ], + "score": 1.0, + "content": "Tomas Geffner and Justin Domke. MCMC variational inference via uncorrected Hamiltonian", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 475, + 462, + 487 + ], + "spans": [ + { + "bbox": [ + 115, + 475, + 462, + 487 + ], + "score": 1.0, + "content": "annealing. Advances in Neural Information Processing Systems, 34:639–651, 2021. 3", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 464, + 505, + 487 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 505, + 528 + ], + "lines": [ + { + "bbox": [ + 105, + 493, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 506, + 507 + ], + "score": 1.0, + "content": "Will Grathwohl, Kuan-Chieh Wang, Jorn-Henrik Jacobsen, David Duvenaud, Mohammad Norouzi, ¨", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 115, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "and Kevin Swersky. Your classifier is secretly an energy based model and you should treat it like", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 516, + 309, + 528 + ], + "spans": [ + { + "bbox": [ + 115, + 516, + 309, + 528 + ], + "score": 1.0, + "content": "one. arXiv preprint arXiv:1912.03263, 2019. 5", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 493, + 506, + 528 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 534, + 505, + 569 + ], + "lines": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "Will Grathwohl, Jacob Jin Kelly, Milad Hashemi, Mohammad Norouzi, Kevin Swersky, and David", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 547, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 115, + 547, + 505, + 559 + ], + "score": 1.0, + "content": "Duvenaud. No MCMC for me: Amortized sampling for fast and stable training of energy-based", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 557, + 420, + 570 + ], + "spans": [ + { + "bbox": [ + 115, + 557, + 420, + 570 + ], + "score": 1.0, + "content": "models. In International Conference on Learning Representations, 2021. 5", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 106, + 535, + 505, + 570 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 576, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 588, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 116, + 588, + 505, + 600 + ], + "score": 1.0, + "content": "trained by a two time-scale update rule converge to a local Nash equilibrium. Advances in Neural", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 599, + 299, + 611 + ], + "spans": [ + { + "bbox": [ + 115, + 599, + 299, + 611 + ], + "score": 1.0, + "content": "Information Processing Systems, 30, 2017. 8", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 576, + 505, + 611 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 618, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 617, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 506, + 631 + ], + "score": 1.0, + "content": "Geoffrey E Hinton. Training products of experts by minimizing contrastive divergence. Neural", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 628, + 302, + 641 + ], + "spans": [ + { + "bbox": [ + 116, + 628, + 302, + 641 + ], + "score": 1.0, + "content": "Computation, 14(8):1771–1800, 2002. 1, 2, 4", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 617, + 506, + 641 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 648, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 114, + 658, + 160, + 670 + ], + "spans": [ + { + "bbox": [ + 114, + 658, + 160, + 670 + ], + "score": 1.0, + "content": "2022. 1, 4", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 648, + 506, + 670 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 678, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 690, + 405, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 405, + 703 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, 33:6840–6851, 2020. 1, 20, 21", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 678, + 506, + 703 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. Adaptive mixtures of", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 720, + 343, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 343, + 733 + ], + "score": 1.0, + "content": "local experts. Neural Computation, 3(1):79–87, 1991. 1", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 106, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 506, + 126 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Justin Johnson, Bharath Hariharan, Laurens Van Der Maaten, Li Fei-Fei, C Lawrence Zitnick, and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "Ross Girshick. Clevr: A diagnostic dataset for compositional language and elementary visual", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 105, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 116, + 105, + 506, + 117 + ], + "score": 1.0, + "content": "reasoning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 115, + 215, + 127 + ], + "spans": [ + { + "bbox": [ + 115, + 115, + 215, + 127 + ], + "score": 1.0, + "content": "pp. 2901–2910, 2017. 7", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 107, + 134, + 503, + 167 + ], + "lines": [ + { + "bbox": [ + 105, + 134, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 505, + 146 + ], + "score": 1.0, + "content": "Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 145, + 504, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 145, + 504, + 158 + ], + "score": 1.0, + "content": "Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 156, + 286, + 168 + ], + "spans": [ + { + "bbox": [ + 116, + 156, + 286, + 168 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2001.08361, 2020. 1", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 504, + 209 + ], + "lines": [ + { + "bbox": [ + 107, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 107, + 175, + 505, + 187 + ], + "score": 1.0, + "content": "Yujia Li, David Choi, Junyoung Chung, Nate Kushman, Julian Schrittwieser, Remi Leblond, Tom ´", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 116, + 186, + 505, + 199 + ], + "score": 1.0, + "content": "Eccles, James Keeling, Felix Gimeno, Agustin Dal Lago, et al. Competition-level code generation", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 197, + 353, + 209 + ], + "spans": [ + { + "bbox": [ + 116, + 197, + 353, + 209 + ], + "score": 1.0, + "content": "with alphacode. arXiv preprint arXiv:2203.07814, 2022. 1", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 215, + 503, + 239 + ], + "lines": [ + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "score": 1.0, + "content": "Nan Liu, Shuang Li, Yilun Du, Josh Tenenbaum, and Antonio Torralba. Learning to compose visual", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 226, + 425, + 239 + ], + "spans": [ + { + "bbox": [ + 115, + 226, + 425, + 239 + ], + "score": 1.0, + "content": "relations. Advances in Neural Information Processing Systems, 34, 2021. 20", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 505, + 268 + ], + "lines": [ + { + "bbox": [ + 106, + 246, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 505, + 258 + ], + "score": 1.0, + "content": "Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, and Joshua B Tenenbaum. Compositional visual", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 256, + 487, + 269 + ], + "spans": [ + { + "bbox": [ + 115, + 256, + 487, + 269 + ], + "score": 1.0, + "content": "generation with composable diffusion models. arXiv preprint arXiv:2206.01714, 2022. 7, 8", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 106, + 275, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 506, + 288 + ], + "score": 1.0, + "content": "Guy Mayraz and Geoffrey E Hinton. Recognizing hand-written digits using hierarchical products of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 287, + 412, + 299 + ], + "spans": [ + { + "bbox": [ + 115, + 287, + 412, + 299 + ], + "score": 1.0, + "content": "experts. Advances in Neural Information Processing Systems, 13, 2000. 1", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 305, + 505, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 506, + 320 + ], + "score": 1.0, + "content": "Radford M Neal. Monte Carlo implementation. In Bayesian Learning for Neural Networks, pp.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 316, + 264, + 329 + ], + "spans": [ + { + "bbox": [ + 116, + 316, + 264, + 329 + ], + "score": 1.0, + "content": "55–98. Springer, 1996. 3, 14, 15, 16", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 105, + 335, + 504, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 334, + 507, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 507, + 349 + ], + "score": 1.0, + "content": "Radford M Neal. Annealed importance sampling. Statistics and Computing, 11(2):125–139, 2001. 6", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 504, + 388 + ], + "lines": [ + { + "bbox": [ + 105, + 354, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 506, + 367 + ], + "score": 1.0, + "content": "Erik Nijkamp, Mitch Hill, Tian Han, Song-Chun Zhu, and Ying Nian Wu. On the anatomy of", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 116, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "mcmc-based maximum likelihood learning of energy-based models. In Proceedings of the AAAI", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 376, + 438, + 389 + ], + "spans": [ + { + "bbox": [ + 116, + 376, + 438, + 389 + ], + "score": 1.0, + "content": "Conference on Artificial Intelligence, volume 34, pp. 5272–5280, 2020. 3, 5, 18", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 395, + 505, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 395, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 506, + 408 + ], + "score": 1.0, + "content": "Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 406, + 477, + 418 + ], + "spans": [ + { + "bbox": [ + 115, + 406, + 477, + 418 + ], + "score": 1.0, + "content": "conditional image generation with clip latents. arXiv preprint arXiv:2204.06125, 2022. 8", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 105, + 424, + 504, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 425, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 505, + 438 + ], + "score": 1.0, + "content": "Gareth O Roberts and Richard L Tweedie. Exponential convergence of Langevin distributions and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 435, + 381, + 448 + ], + "spans": [ + { + "bbox": [ + 115, + 435, + 381, + 448 + ], + "score": 1.0, + "content": "their discrete approximations. Bernoulli, pp. 341–363, 1996. 3, 6", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 106, + 454, + 506, + 499 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 506, + 468 + ], + "score": 1.0, + "content": "Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 466, + 507, + 479 + ], + "spans": [ + { + "bbox": [ + 115, + 466, + 507, + 479 + ], + "score": 1.0, + "content": "Kamyar Seyed Ghasemipour, Burcu Karagol Ayan, S Sara Mahdavi, Rapha Gontijo Lopes, et al.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 476, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 115, + 476, + 506, + 490 + ], + "score": 1.0, + "content": "Photorealistic text-to-image diffusion models with deep language understanding. arXiv preprint", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 488, + 239, + 500 + ], + "spans": [ + { + "bbox": [ + 115, + 488, + 239, + 500 + ], + "score": 1.0, + "content": "arXiv:2205.11487, 2022. 1, 8", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 105, + 506, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "Tim Salimans and Jonathan Ho. Should ebms model the energy or the score? In Energy Based", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 517, + 289, + 530 + ], + "spans": [ + { + "bbox": [ + 115, + 517, + 289, + 530 + ], + "score": 1.0, + "content": "Models Workshop-ICLR 2021, 2021. 6, 18", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 105, + 536, + 505, + 559 + ], + "lines": [ + { + "bbox": [ + 105, + 536, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 506, + 550 + ], + "score": 1.0, + "content": "Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. arXiv", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 548, + 261, + 559 + ], + "spans": [ + { + "bbox": [ + 115, + 548, + 261, + 559 + ], + "score": 1.0, + "content": "preprint arXiv:2202.00512, 2022. 9", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 505, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 577, + 507, + 591 + ], + "spans": [ + { + "bbox": [ + 115, + 577, + 507, + 591 + ], + "score": 1.0, + "content": "learning using nonequilibrium thermodynamics. In International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 588, + 258, + 600 + ], + "spans": [ + { + "bbox": [ + 115, + 588, + 258, + 600 + ], + "score": 1.0, + "content": "pp. 2256–2265. PMLR, 2015. 1, 2", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 607, + 505, + 630 + ], + "lines": [ + { + "bbox": [ + 107, + 607, + 507, + 620 + ], + "spans": [ + { + "bbox": [ + 107, + 607, + 507, + 620 + ], + "score": 1.0, + "content": "Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 114, + 618, + 399, + 631 + ], + "spans": [ + { + "bbox": [ + 114, + 618, + 399, + 631 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 32, 2019. 1, 3, 6", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 109, + 637, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 107, + 637, + 504, + 650 + ], + "spans": [ + { + "bbox": [ + 107, + 637, + 504, + 650 + ], + "score": 1.0, + "content": "Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 648, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 648, + 506, + 660 + ], + "score": 1.0, + "content": "Poole. Score-based generative modeling through stochastic differential equations. In International", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 658, + 344, + 672 + ], + "spans": [ + { + "bbox": [ + 115, + 658, + 344, + 672 + ], + "score": 1.0, + "content": "Conference on Learning Representations, 2021. 5, 6, 17", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 104, + 678, + 504, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "Hongyi Zhang, Yann N Dauphin, and Tengyu Ma. Fixup initialization: Residual learning without", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 689, + 354, + 700 + ], + "spans": [ + { + "bbox": [ + 115, + 689, + 354, + 700 + ], + "score": 1.0, + "content": "normalization. arXiv preprint arXiv:1901.09321, 2019. 20", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 506, + 126 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Justin Johnson, Bharath Hariharan, Laurens Van Der Maaten, Li Fei-Fei, C Lawrence Zitnick, and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "Ross Girshick. Clevr: A diagnostic dataset for compositional language and elementary visual", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 105, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 116, + 105, + 506, + 117 + ], + "score": 1.0, + "content": "reasoning. In Proceedings of the IEEE conference on Computer Vision and Pattern Recognition,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 115, + 215, + 127 + ], + "spans": [ + { + "bbox": [ + 115, + 115, + 215, + 127 + ], + "score": 1.0, + "content": "pp. 2901–2910, 2017. 7", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 106, + 82, + 506, + 127 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 134, + 503, + 167 + ], + "lines": [ + { + "bbox": [ + 105, + 134, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 505, + 146 + ], + "score": 1.0, + "content": "Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 145, + 504, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 145, + 504, + 158 + ], + "score": 1.0, + "content": "Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 156, + 286, + 168 + ], + "spans": [ + { + "bbox": [ + 116, + 156, + 286, + 168 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2001.08361, 2020. 1", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 134, + 505, + 168 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 504, + 209 + ], + "lines": [ + { + "bbox": [ + 107, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 107, + 175, + 505, + 187 + ], + "score": 1.0, + "content": "Yujia Li, David Choi, Junyoung Chung, Nate Kushman, Julian Schrittwieser, Remi Leblond, Tom ´", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 116, + 186, + 505, + 199 + ], + "score": 1.0, + "content": "Eccles, James Keeling, Felix Gimeno, Agustin Dal Lago, et al. Competition-level code generation", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 197, + 353, + 209 + ], + "spans": [ + { + "bbox": [ + 116, + 197, + 353, + 209 + ], + "score": 1.0, + "content": "with alphacode. arXiv preprint arXiv:2203.07814, 2022. 1", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 107, + 175, + 505, + 209 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 215, + 503, + 239 + ], + "lines": [ + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 229 + ], + "score": 1.0, + "content": "Nan Liu, Shuang Li, Yilun Du, Josh Tenenbaum, and Antonio Torralba. Learning to compose visual", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 226, + 425, + 239 + ], + "spans": [ + { + "bbox": [ + 115, + 226, + 425, + 239 + ], + "score": 1.0, + "content": "relations. Advances in Neural Information Processing Systems, 34, 2021. 20", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 215, + 505, + 239 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 505, + 268 + ], + "lines": [ + { + "bbox": [ + 106, + 246, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 505, + 258 + ], + "score": 1.0, + "content": "Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, and Joshua B Tenenbaum. Compositional visual", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 256, + 487, + 269 + ], + "spans": [ + { + "bbox": [ + 115, + 256, + 487, + 269 + ], + "score": 1.0, + "content": "generation with composable diffusion models. arXiv preprint arXiv:2206.01714, 2022. 7, 8", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 106, + 246, + 505, + 269 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 106, + 275, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 506, + 288 + ], + "score": 1.0, + "content": "Guy Mayraz and Geoffrey E Hinton. Recognizing hand-written digits using hierarchical products of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 287, + 412, + 299 + ], + "spans": [ + { + "bbox": [ + 115, + 287, + 412, + 299 + ], + "score": 1.0, + "content": "experts. Advances in Neural Information Processing Systems, 13, 2000. 1", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 106, + 275, + 506, + 299 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 305, + 505, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 506, + 320 + ], + "score": 1.0, + "content": "Radford M Neal. Monte Carlo implementation. In Bayesian Learning for Neural Networks, pp.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 316, + 264, + 329 + ], + "spans": [ + { + "bbox": [ + 116, + 316, + 264, + 329 + ], + "score": 1.0, + "content": "55–98. Springer, 1996. 3, 14, 15, 16", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 303, + 506, + 329 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 335, + 504, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 334, + 507, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 507, + 349 + ], + "score": 1.0, + "content": "Radford M Neal. Annealed importance sampling. Statistics and Computing, 11(2):125–139, 2001. 6", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 334, + 507, + 349 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 504, + 388 + ], + "lines": [ + { + "bbox": [ + 105, + 354, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 506, + 367 + ], + "score": 1.0, + "content": "Erik Nijkamp, Mitch Hill, Tian Han, Song-Chun Zhu, and Ying Nian Wu. On the anatomy of", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 116, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "mcmc-based maximum likelihood learning of energy-based models. In Proceedings of the AAAI", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 376, + 438, + 389 + ], + "spans": [ + { + "bbox": [ + 116, + 376, + 438, + 389 + ], + "score": 1.0, + "content": "Conference on Artificial Intelligence, volume 34, pp. 5272–5280, 2020. 3, 5, 18", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 354, + 506, + 389 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 395, + 505, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 395, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 506, + 408 + ], + "score": 1.0, + "content": "Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 406, + 477, + 418 + ], + "spans": [ + { + "bbox": [ + 115, + 406, + 477, + 418 + ], + "score": 1.0, + "content": "conditional image generation with clip latents. arXiv preprint arXiv:2204.06125, 2022. 8", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 106, + 395, + 506, + 418 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 424, + 504, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 425, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 505, + 438 + ], + "score": 1.0, + "content": "Gareth O Roberts and Richard L Tweedie. Exponential convergence of Langevin distributions and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 435, + 381, + 448 + ], + "spans": [ + { + "bbox": [ + 115, + 435, + 381, + 448 + ], + "score": 1.0, + "content": "their discrete approximations. Bernoulli, pp. 341–363, 1996. 3, 6", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 106, + 425, + 505, + 448 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 454, + 506, + 499 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 506, + 468 + ], + "score": 1.0, + "content": "Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 466, + 507, + 479 + ], + "spans": [ + { + "bbox": [ + 115, + 466, + 507, + 479 + ], + "score": 1.0, + "content": "Kamyar Seyed Ghasemipour, Burcu Karagol Ayan, S Sara Mahdavi, Rapha Gontijo Lopes, et al.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 476, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 115, + 476, + 506, + 490 + ], + "score": 1.0, + "content": "Photorealistic text-to-image diffusion models with deep language understanding. arXiv preprint", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 488, + 239, + 500 + ], + "spans": [ + { + "bbox": [ + 115, + 488, + 239, + 500 + ], + "score": 1.0, + "content": "arXiv:2205.11487, 2022. 1, 8", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 455, + 507, + 500 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 506, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "Tim Salimans and Jonathan Ho. Should ebms model the energy or the score? In Energy Based", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 517, + 289, + 530 + ], + "spans": [ + { + "bbox": [ + 115, + 517, + 289, + 530 + ], + "score": 1.0, + "content": "Models Workshop-ICLR 2021, 2021. 6, 18", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 506, + 506, + 530 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 536, + 505, + 559 + ], + "lines": [ + { + "bbox": [ + 105, + 536, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 506, + 550 + ], + "score": 1.0, + "content": "Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. arXiv", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 548, + 261, + 559 + ], + "spans": [ + { + "bbox": [ + 115, + 548, + 261, + 559 + ], + "score": 1.0, + "content": "preprint arXiv:2202.00512, 2022. 9", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 536, + 506, + 559 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 505, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 577, + 507, + 591 + ], + "spans": [ + { + "bbox": [ + 115, + 577, + 507, + 591 + ], + "score": 1.0, + "content": "learning using nonequilibrium thermodynamics. In International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 588, + 258, + 600 + ], + "spans": [ + { + "bbox": [ + 115, + 588, + 258, + 600 + ], + "score": 1.0, + "content": "pp. 2256–2265. PMLR, 2015. 1, 2", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 565, + 507, + 600 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 607, + 505, + 630 + ], + "lines": [ + { + "bbox": [ + 107, + 607, + 507, + 620 + ], + "spans": [ + { + "bbox": [ + 107, + 607, + 507, + 620 + ], + "score": 1.0, + "content": "Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 114, + 618, + 399, + 631 + ], + "spans": [ + { + "bbox": [ + 114, + 618, + 399, + 631 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 32, 2019. 1, 3, 6", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 107, + 607, + 507, + 631 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 637, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 107, + 637, + 504, + 650 + ], + "spans": [ + { + "bbox": [ + 107, + 637, + 504, + 650 + ], + "score": 1.0, + "content": "Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 648, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 648, + 506, + 660 + ], + "score": 1.0, + "content": "Poole. Score-based generative modeling through stochastic differential equations. In International", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 658, + 344, + 672 + ], + "spans": [ + { + "bbox": [ + 115, + 658, + 344, + 672 + ], + "score": 1.0, + "content": "Conference on Learning Representations, 2021. 5, 6, 17", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40, + "bbox_fs": [ + 107, + 637, + 506, + 672 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 678, + 504, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "Hongyi Zhang, Yann N Dauphin, and Tengyu Ma. Fixup initialization: Residual learning without", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 689, + 354, + 700 + ], + "spans": [ + { + "bbox": [ + 115, + 689, + 354, + 700 + ], + "score": 1.0, + "content": "normalization. arXiv preprint arXiv:1901.09321, 2019. 20", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 106, + 678, + 505, + 700 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 167, + 95 + ], + "lines": [ + { + "bbox": [ + 104, + 77, + 170, + 97 + ], + "spans": [ + { + "bbox": [ + 104, + 77, + 170, + 97 + ], + "score": 1.0, + "content": "Appendix", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 104, + 506, + 170 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "score": 1.0, + "content": "In this appendix, we present additional text-to-image results in Section A. We present detailed", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 113, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 505, + 129 + ], + "score": 1.0, + "content": "derivations of diffusion models in Section B. We present additional information on MCMC sampling", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 126, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 506, + 138 + ], + "score": 1.0, + "content": "in Section C.1. We provide additional derivations on composing diffusion models in Section D.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 136, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 506, + 150 + ], + "score": 1.0, + "content": "We discuss different parameterizations of energy based diffusion models in Section E. We further", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 148, + 506, + 161 + ], + "spans": [ + { + "bbox": [ + 106, + 148, + 506, + 161 + ], + "score": 1.0, + "content": "provide additional example 2D compositions in Section F. Finally, we provide experimental details in", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 158, + 152, + 172 + ], + "spans": [ + { + "bbox": [ + 104, + 158, + 152, + 172 + ], + "score": 1.0, + "content": "Section G.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 108, + 197, + 263, + 210 + ], + "lines": [ + { + "bbox": [ + 105, + 196, + 264, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 264, + 212 + ], + "score": 1.0, + "content": "A TEXT-TO-IMAGE RESULTS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 229, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 505, + 242 + ], + "score": 1.0, + "content": "We present additional use cases of composing models in different text-to-image domains. First, in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "score": 1.0, + "content": "Figure A1, we illustrate how composing two separate energy parameterized diffusion models enables", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "score": 1.0, + "content": "us to more accurately generate images that have more detailed information in the caption. Next, in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 262, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 506, + 277 + ], + "score": 1.0, + "content": "Figure A2, we illustrate how composing two separate energy parameterized diffusion models further", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 273, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 506, + 286 + ], + "score": 1.0, + "content": "enable us to accurately generate images with the correct colors assigned to each object. We further", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 285, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 505, + 298 + ], + "score": 1.0, + "content": "show in Figure A3 how composing the negation of one energy parameterized diffusion model with", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 296, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 296, + 505, + 308 + ], + "score": 1.0, + "content": "another other enables us to generate images where one commonly occurring co-founding factor does", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 305, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 506, + 321 + ], + "score": 1.0, + "content": "occur (i.e. a sandy beach without coastal water). Finally, we illustrate in Figure A4, how composing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 317, + 470, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 470, + 331 + ], + "score": 1.0, + "content": "multiple diffusion models enables us to render the number of objects in a scene accurately.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12 + }, + { + "type": "image", + "bbox": [ + 115, + 351, + 504, + 673 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 351, + 504, + 673 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 351, + 504, + 673 + ], + "spans": [ + { + "bbox": [ + 115, + 351, + 504, + 673 + ], + "score": 0.978, + "type": "image", + "image_path": "97c0bf123e3c015a3bcd1f74d6a3ba0c74f732c744e61cbde1a5ba54b67dc81b.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 115, + 351, + 504, + 458.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 115, + 458.3333333333333, + 504, + 565.6666666666666 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 115, + 565.6666666666666, + 504, + 673.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 682, + 505, + 712 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 681, + 507, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 507, + 694 + ], + "score": 1.0, + "content": "Figure A1: By composing energy based diffusion models, we can render more detailed information in images.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 691, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 704 + ], + "score": 1.0, + "content": "In the above images, we can more accurately render details such as Central Park (top) or the effect of snowing", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 701, + 143, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 143, + 714 + ], + "score": 1.0, + "content": "(bottom).", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + } + ], + "index": 19.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 167, + 95 + ], + "lines": [ + { + "bbox": [ + 104, + 77, + 170, + 97 + ], + "spans": [ + { + "bbox": [ + 104, + 77, + 170, + 97 + ], + "score": 1.0, + "content": "Appendix", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 104, + 506, + 170 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "score": 1.0, + "content": "In this appendix, we present additional text-to-image results in Section A. We present detailed", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 113, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 505, + 129 + ], + "score": 1.0, + "content": "derivations of diffusion models in Section B. We present additional information on MCMC sampling", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 126, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 506, + 138 + ], + "score": 1.0, + "content": "in Section C.1. We provide additional derivations on composing diffusion models in Section D.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 136, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 506, + 150 + ], + "score": 1.0, + "content": "We discuss different parameterizations of energy based diffusion models in Section E. We further", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 148, + 506, + 161 + ], + "spans": [ + { + "bbox": [ + 106, + 148, + 506, + 161 + ], + "score": 1.0, + "content": "provide additional example 2D compositions in Section F. Finally, we provide experimental details in", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 158, + 152, + 172 + ], + "spans": [ + { + "bbox": [ + 104, + 158, + 152, + 172 + ], + "score": 1.0, + "content": "Section G.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5, + "bbox_fs": [ + 104, + 105, + 506, + 172 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 197, + 263, + 210 + ], + "lines": [ + { + "bbox": [ + 105, + 196, + 264, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 264, + 212 + ], + "score": 1.0, + "content": "A TEXT-TO-IMAGE RESULTS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 229, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 106, + 230, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 505, + 242 + ], + "score": 1.0, + "content": "We present additional use cases of composing models in different text-to-image domains. First, in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "score": 1.0, + "content": "Figure A1, we illustrate how composing two separate energy parameterized diffusion models enables", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "score": 1.0, + "content": "us to more accurately generate images that have more detailed information in the caption. Next, in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 262, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 506, + 277 + ], + "score": 1.0, + "content": "Figure A2, we illustrate how composing two separate energy parameterized diffusion models further", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 273, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 506, + 286 + ], + "score": 1.0, + "content": "enable us to accurately generate images with the correct colors assigned to each object. We further", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 285, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 505, + 298 + ], + "score": 1.0, + "content": "show in Figure A3 how composing the negation of one energy parameterized diffusion model with", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 296, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 296, + 505, + 308 + ], + "score": 1.0, + "content": "another other enables us to generate images where one commonly occurring co-founding factor does", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 305, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 506, + 321 + ], + "score": 1.0, + "content": "occur (i.e. a sandy beach without coastal water). Finally, we illustrate in Figure A4, how composing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 317, + 470, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 470, + 331 + ], + "score": 1.0, + "content": "multiple diffusion models enables us to render the number of objects in a scene accurately.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 230, + 506, + 331 + ] + }, + { + "type": "image", + "bbox": [ + 115, + 351, + 504, + 673 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 351, + 504, + 673 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 351, + 504, + 673 + ], + "spans": [ + { + "bbox": [ + 115, + 351, + 504, + 673 + ], + "score": 0.978, + "type": "image", + "image_path": "97c0bf123e3c015a3bcd1f74d6a3ba0c74f732c744e61cbde1a5ba54b67dc81b.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 115, + 351, + 504, + 458.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 115, + 458.3333333333333, + 504, + 565.6666666666666 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 115, + 565.6666666666666, + 504, + 673.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 682, + 505, + 712 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 681, + 507, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 507, + 694 + ], + "score": 1.0, + "content": "Figure A1: By composing energy based diffusion models, we can render more detailed information in images.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 691, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 704 + ], + "score": 1.0, + "content": "In the above images, we can more accurately render details such as Central Park (top) or the effect of snowing", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 701, + 143, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 143, + 714 + ], + "score": 1.0, + "content": "(bottom).", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + } + ], + "index": 19.5 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 80, + 504, + 406 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 80, + 504, + 406 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 80, + 504, + 406 + ], + "spans": [ + { + "bbox": [ + 111, + 80, + 504, + 406 + ], + "score": 0.976, + "type": "image", + "image_path": "f487ea0a145e2d5a6ebb1381cab3f50126e5df4e2a2c547ab9d39ae4719bcd22.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 80, + 504, + 188.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 188.66666666666669, + 504, + 297.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 297.33333333333337, + 504, + 406.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 414, + 505, + 435 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 414, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 506, + 426 + ], + "score": 1.0, + "content": "Figure A2: By composing energy based diffusion models, we can more accurately render different colors of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 424, + 174, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 174, + 436 + ], + "score": 1.0, + "content": "objects in a scene.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 107, + 448, + 373, + 462 + ], + "lines": [ + { + "bbox": [ + 105, + 447, + 373, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 373, + 463 + ], + "score": 1.0, + "content": "B DETAILED DERIVATION OF DIFFUSION MODELS", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 474, + 506, + 507 + ], + "lines": [ + { + "bbox": [ + 105, + 474, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 304, + 487 + ], + "score": 1.0, + "content": "Diffusion models seek to model a data distribution", + "type": "text" + }, + { + "bbox": [ + 304, + 475, + 328, + 487 + ], + "score": 0.92, + "content": "q ( x _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 474, + 506, + 487 + ], + "score": 1.0, + "content": "(written this way for notational convenience)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 484, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 255, + 499 + ], + "score": 1.0, + "content": "and define a series of latent variables", + "type": "text" + }, + { + "bbox": [ + 255, + 487, + 300, + 497 + ], + "score": 0.88, + "content": "x _ { 1 } , \\dots , x _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 484, + 437, + 499 + ], + "score": 1.0, + "content": "generated from a Markov process", + "type": "text" + }, + { + "bbox": [ + 437, + 486, + 505, + 498 + ], + "score": 0.93, + "content": "x _ { t } \\sim q ( x _ { t } | x _ { t - 1 } )", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 496, + 134, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 134, + 509 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 507, + 394, + 528 + ], + "lines": [ + { + "bbox": [ + 218, + 507, + 394, + 528 + ], + "spans": [ + { + "bbox": [ + 218, + 507, + 394, + 528 + ], + "score": 0.94, + "content": "q ( x _ { t } | x _ { t - 1 } ) = N \\left( x _ { t } ; \\sqrt { 1 - \\beta _ { t } } x _ { t - 1 } , \\beta _ { t } I \\right) .", + "type": "interline_equation", + "image_path": "86e6899980603373150d00f0eb114b88edfa14a7523b58e584433e599c76fc56.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 218, + 507, + 394, + 528 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 535, + 505, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 387, + 549 + ], + "score": 1.0, + "content": "A unique and useful property of this process is that all time marginals", + "type": "text" + }, + { + "bbox": [ + 388, + 536, + 423, + 548 + ], + "score": 0.93, + "content": "q ( x _ { t } | x _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 535, + 505, + 549 + ], + "score": 1.0, + "content": "can be computed in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 547, + 227, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 227, + 558 + ], + "score": 1.0, + "content": "closed form and are Gaussian", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 557, + 387, + 584 + ], + "lines": [ + { + "bbox": [ + 224, + 557, + 387, + 584 + ], + "spans": [ + { + "bbox": [ + 224, + 557, + 387, + 584 + ], + "score": 0.95, + "content": "q ( x _ { t } | x _ { 0 } ) = { \\mathcal { N } } \\left( x _ { t } ; { \\sqrt { 1 - \\sigma _ { t } ^ { 2 } } } x _ { 0 } , \\sigma _ { t } ^ { 2 } I \\right) .", + "type": "interline_equation", + "image_path": "fbc6493eb51569ea46822bd220904a95e546c47f4c45583948d03a3528cb28cf.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 224, + 557, + 387, + 584 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 586, + 505, + 611 + ], + "lines": [ + { + "bbox": [ + 104, + 585, + 504, + 602 + ], + "spans": [ + { + "bbox": [ + 104, + 585, + 134, + 602 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 587, + 192, + 599 + ], + "score": 0.9, + "content": "\\sigma _ { t } ^ { 2 } = 1 - \\bar { \\alpha } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 585, + 212, + 602 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 212, + 586, + 299, + 601 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\bar { \\alpha } _ { t } = \\prod _ { t = 1 } ^ { T } ( 1 - \\beta _ { t } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 585, + 399, + 602 + ], + "score": 1.0, + "content": ". We can see that if all", + "type": "text" + }, + { + "bbox": [ + 399, + 588, + 432, + 599 + ], + "score": 0.92, + "content": "\\beta _ { t } ~ > ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 585, + 469, + 602 + ], + "score": 1.0, + "content": ", then as", + "type": "text" + }, + { + "bbox": [ + 470, + 589, + 504, + 598 + ], + "score": 0.89, + "content": "t \\infty", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 107, + 597, + 208, + 612 + ], + "spans": [ + { + "bbox": [ + 107, + 599, + 203, + 611 + ], + "score": 0.89, + "content": "q ( x _ { t } | x _ { 0 } ) { \\overset { } { \\to } } { \\mathcal { N } } ( x _ { t } ; 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 597, + 208, + 612 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 104, + 615, + 506, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 507, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 206, + 630 + ], + "score": 1.0, + "content": "We seek to train a model", + "type": "text" + }, + { + "bbox": [ + 207, + 615, + 256, + 628 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x _ { t - 1 } | x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 614, + 319, + 630 + ], + "score": 1.0, + "content": "which reverses", + "type": "text" + }, + { + "bbox": [ + 319, + 615, + 363, + 627 + ], + "score": 0.92, + "content": "q ( x _ { t } | x _ { t - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 614, + 507, + 630 + ], + "score": 1.0, + "content": "step-wise with a parametric model.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 625, + 433, + 643 + ], + "spans": [ + { + "bbox": [ + 106, + 628, + 334, + 641 + ], + "score": 1.0, + "content": "We can analytically derive the variance of the reversal as", + "type": "text" + }, + { + "bbox": [ + 335, + 627, + 388, + 643 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tilde { \\beta } _ { t } = \\frac { 1 - \\bar { \\alpha } _ { t - 1 } } { 1 - \\bar { \\alpha } _ { t } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 625, + 433, + 641 + ], + "score": 1.0, + "content": "1−α¯t−1 and define", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 644, + 385, + 659 + ], + "lines": [ + { + "bbox": [ + 225, + 644, + 385, + 659 + ], + "spans": [ + { + "bbox": [ + 225, + 644, + 385, + 659 + ], + "score": 0.88, + "content": "p _ { \\theta } ( x _ { t } | x _ { t + 1 } ) = \\mathcal { N } ( x _ { t } ; \\mu _ { \\theta } ( x _ { t - 1 } , t ) , \\tilde { \\beta } _ { t } I )", + "type": "interline_equation", + "image_path": "e609eab2ce5becd81546000823444dd517ab414d7423003569361626baeee9aa.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 225, + 644, + 385, + 659 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 660, + 502, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 138, + 672 + ], + "score": 1.0, + "content": "and set", + "type": "text" + }, + { + "bbox": [ + 139, + 660, + 212, + 672 + ], + "score": 0.94, + "content": "p ( x _ { T } ) = \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 658, + 505, + 672 + ], + "score": 1.0, + "content": ". We train this model to maximize a variational bound on the marginal", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 669, + 151, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 151, + 683 + ], + "score": 1.0, + "content": "likelihood", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 681, + 468, + 732 + ], + "lines": [ + { + "bbox": [ + 124, + 681, + 468, + 732 + ], + "spans": [ + { + "bbox": [ + 124, + 681, + 468, + 732 + ], + "score": 0.93, + "content": "\\begin{array} { r l r } { { \\log p _ { \\theta } ( x _ { 0 } ) \\geq { \\mathbf E } _ { q ( x _ { 1 } , \\dots , x _ { T } | x _ { 0 } ) } [ \\log p _ { \\theta } ( x _ { 0 } | x _ { 1 } ) + \\sum _ { t = 1 } ^ { T } { D _ { K L } ( q ( x _ { t } | x _ { t - 1 } , x _ { 0 } ) | | p ( x _ { t } | | x _ { t - 1 } ) ) } } } \\\\ & { } & { + { D _ { K L } ( q ( x _ { T } | x _ { 0 } ) | | p ( x _ { T } ) ) } ] . } \\end{array}", + "type": "interline_equation", + "image_path": "730c9c8c662ad50b514c92f1b721f5ee671f186a6e19c787734a3bf5f9c1e012.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 124, + 681, + 468, + 698.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 124, + 698.0, + 468, + 715.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 124, + 715.0, + 468, + 732.0 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 80, + 504, + 406 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 80, + 504, + 406 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 80, + 504, + 406 + ], + "spans": [ + { + "bbox": [ + 111, + 80, + 504, + 406 + ], + "score": 0.976, + "type": "image", + "image_path": "f487ea0a145e2d5a6ebb1381cab3f50126e5df4e2a2c547ab9d39ae4719bcd22.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 80, + 504, + 188.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 188.66666666666669, + 504, + 297.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 297.33333333333337, + 504, + 406.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 414, + 505, + 435 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 414, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 506, + 426 + ], + "score": 1.0, + "content": "Figure A2: By composing energy based diffusion models, we can more accurately render different colors of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 424, + 174, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 174, + 436 + ], + "score": 1.0, + "content": "objects in a scene.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 107, + 448, + 373, + 462 + ], + "lines": [ + { + "bbox": [ + 105, + 447, + 373, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 373, + 463 + ], + "score": 1.0, + "content": "B DETAILED DERIVATION OF DIFFUSION MODELS", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 474, + 506, + 507 + ], + "lines": [ + { + "bbox": [ + 105, + 474, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 304, + 487 + ], + "score": 1.0, + "content": "Diffusion models seek to model a data distribution", + "type": "text" + }, + { + "bbox": [ + 304, + 475, + 328, + 487 + ], + "score": 0.92, + "content": "q ( x _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 474, + 506, + 487 + ], + "score": 1.0, + "content": "(written this way for notational convenience)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 484, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 255, + 499 + ], + "score": 1.0, + "content": "and define a series of latent variables", + "type": "text" + }, + { + "bbox": [ + 255, + 487, + 300, + 497 + ], + "score": 0.88, + "content": "x _ { 1 } , \\dots , x _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 484, + 437, + 499 + ], + "score": 1.0, + "content": "generated from a Markov process", + "type": "text" + }, + { + "bbox": [ + 437, + 486, + 505, + 498 + ], + "score": 0.93, + "content": "x _ { t } \\sim q ( x _ { t } | x _ { t - 1 } )", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 496, + 134, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 134, + 509 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 474, + 506, + 509 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 507, + 394, + 528 + ], + "lines": [ + { + "bbox": [ + 218, + 507, + 394, + 528 + ], + "spans": [ + { + "bbox": [ + 218, + 507, + 394, + 528 + ], + "score": 0.94, + "content": "q ( x _ { t } | x _ { t - 1 } ) = N \\left( x _ { t } ; \\sqrt { 1 - \\beta _ { t } } x _ { t - 1 } , \\beta _ { t } I \\right) .", + "type": "interline_equation", + "image_path": "86e6899980603373150d00f0eb114b88edfa14a7523b58e584433e599c76fc56.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 218, + 507, + 394, + 528 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 535, + 505, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 387, + 549 + ], + "score": 1.0, + "content": "A unique and useful property of this process is that all time marginals", + "type": "text" + }, + { + "bbox": [ + 388, + 536, + 423, + 548 + ], + "score": 0.93, + "content": "q ( x _ { t } | x _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 535, + 505, + 549 + ], + "score": 1.0, + "content": "can be computed in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 547, + 227, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 227, + 558 + ], + "score": 1.0, + "content": "closed form and are Gaussian", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 535, + 505, + 558 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 557, + 387, + 584 + ], + "lines": [ + { + "bbox": [ + 224, + 557, + 387, + 584 + ], + "spans": [ + { + "bbox": [ + 224, + 557, + 387, + 584 + ], + "score": 0.95, + "content": "q ( x _ { t } | x _ { 0 } ) = { \\mathcal { N } } \\left( x _ { t } ; { \\sqrt { 1 - \\sigma _ { t } ^ { 2 } } } x _ { 0 } , \\sigma _ { t } ^ { 2 } I \\right) .", + "type": "interline_equation", + "image_path": "fbc6493eb51569ea46822bd220904a95e546c47f4c45583948d03a3528cb28cf.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 224, + 557, + 387, + 584 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 586, + 505, + 611 + ], + "lines": [ + { + "bbox": [ + 104, + 585, + 504, + 602 + ], + "spans": [ + { + "bbox": [ + 104, + 585, + 134, + 602 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 587, + 192, + 599 + ], + "score": 0.9, + "content": "\\sigma _ { t } ^ { 2 } = 1 - \\bar { \\alpha } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 585, + 212, + 602 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 212, + 586, + 299, + 601 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\bar { \\alpha } _ { t } = \\prod _ { t = 1 } ^ { T } ( 1 - \\beta _ { t } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 585, + 399, + 602 + ], + "score": 1.0, + "content": ". We can see that if all", + "type": "text" + }, + { + "bbox": [ + 399, + 588, + 432, + 599 + ], + "score": 0.92, + "content": "\\beta _ { t } ~ > ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 585, + 469, + 602 + ], + "score": 1.0, + "content": ", then as", + "type": "text" + }, + { + "bbox": [ + 470, + 589, + 504, + 598 + ], + "score": 0.89, + "content": "t \\infty", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 107, + 597, + 208, + 612 + ], + "spans": [ + { + "bbox": [ + 107, + 599, + 203, + 611 + ], + "score": 0.89, + "content": "q ( x _ { t } | x _ { 0 } ) { \\overset { } { \\to } } { \\mathcal { N } } ( x _ { t } ; 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 597, + 208, + 612 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 104, + 585, + 504, + 612 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 615, + 506, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 507, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 206, + 630 + ], + "score": 1.0, + "content": "We seek to train a model", + "type": "text" + }, + { + "bbox": [ + 207, + 615, + 256, + 628 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x _ { t - 1 } | x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 614, + 319, + 630 + ], + "score": 1.0, + "content": "which reverses", + "type": "text" + }, + { + "bbox": [ + 319, + 615, + 363, + 627 + ], + "score": 0.92, + "content": "q ( x _ { t } | x _ { t - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 614, + 507, + 630 + ], + "score": 1.0, + "content": "step-wise with a parametric model.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 625, + 433, + 643 + ], + "spans": [ + { + "bbox": [ + 106, + 628, + 334, + 641 + ], + "score": 1.0, + "content": "We can analytically derive the variance of the reversal as", + "type": "text" + }, + { + "bbox": [ + 335, + 627, + 388, + 643 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tilde { \\beta } _ { t } = \\frac { 1 - \\bar { \\alpha } _ { t - 1 } } { 1 - \\bar { \\alpha } _ { t } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 625, + 433, + 641 + ], + "score": 1.0, + "content": "1−α¯t−1 and define", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 614, + 507, + 643 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 644, + 385, + 659 + ], + "lines": [ + { + "bbox": [ + 225, + 644, + 385, + 659 + ], + "spans": [ + { + "bbox": [ + 225, + 644, + 385, + 659 + ], + "score": 0.88, + "content": "p _ { \\theta } ( x _ { t } | x _ { t + 1 } ) = \\mathcal { N } ( x _ { t } ; \\mu _ { \\theta } ( x _ { t - 1 } , t ) , \\tilde { \\beta } _ { t } I )", + "type": "interline_equation", + "image_path": "e609eab2ce5becd81546000823444dd517ab414d7423003569361626baeee9aa.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 225, + 644, + 385, + 659 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 660, + 502, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 138, + 672 + ], + "score": 1.0, + "content": "and set", + "type": "text" + }, + { + "bbox": [ + 139, + 660, + 212, + 672 + ], + "score": 0.94, + "content": "p ( x _ { T } ) = \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 658, + 505, + 672 + ], + "score": 1.0, + "content": ". We train this model to maximize a variational bound on the marginal", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 669, + 151, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 151, + 683 + ], + "score": 1.0, + "content": "likelihood", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 658, + 505, + 683 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 681, + 468, + 732 + ], + "lines": [ + { + "bbox": [ + 124, + 681, + 468, + 732 + ], + "spans": [ + { + "bbox": [ + 124, + 681, + 468, + 732 + ], + "score": 0.93, + "content": "\\begin{array} { r l r } { { \\log p _ { \\theta } ( x _ { 0 } ) \\geq { \\mathbf E } _ { q ( x _ { 1 } , \\dots , x _ { T } | x _ { 0 } ) } [ \\log p _ { \\theta } ( x _ { 0 } | x _ { 1 } ) + \\sum _ { t = 1 } ^ { T } { D _ { K L } ( q ( x _ { t } | x _ { t - 1 } , x _ { 0 } ) | | p ( x _ { t } | | x _ { t - 1 } ) ) } } } \\\\ & { } & { + { D _ { K L } ( q ( x _ { T } | x _ { 0 } ) | | p ( x _ { T } ) ) } ] . } \\end{array}", + "type": "interline_equation", + "image_path": "730c9c8c662ad50b514c92f1b721f5ee671f186a6e19c787734a3bf5f9c1e012.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 124, + 681, + 468, + 698.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 124, + 698.0, + 468, + 715.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 124, + 715.0, + 468, + 732.0 + ], + "spans": [], + "index": 22 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 80, + 503, + 405 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 80, + 503, + 405 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 80, + 503, + 405 + ], + "spans": [ + { + "bbox": [ + 112, + 80, + 503, + 405 + ], + "score": 0.978, + "type": "image", + "image_path": "97483bdb0cce6ccf6854f297ede1fbe5b048eaf3ebdac470949060c514476bad.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 80, + 503, + 188.33333333333331 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 188.33333333333331, + 503, + 296.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 296.66666666666663, + 503, + 404.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 413, + 505, + 435 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 412, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 505, + 427 + ], + "score": 1.0, + "content": "Figure A3: By composing an energy parameterized diffusion model with the negation of another energy", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 424, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 436 + ], + "score": 1.0, + "content": "parameterized diffusion model, we can render images in unusual configurations not typically found in the data.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 455, + 503, + 479 + ], + "lines": [ + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 465, + 469 + ], + "score": 1.0, + "content": "The first term lacks parameters and the final is approximately 0 from the convergence of the", + "type": "text" + }, + { + "bbox": [ + 465, + 459, + 471, + 468 + ], + "score": 0.79, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 455, + 505, + 469 + ], + "score": 1.0, + "content": "-process", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 467, + 241, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 241, + 480 + ], + "score": 1.0, + "content": "so we focus on the middle terms.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 107, + 484, + 503, + 507 + ], + "lines": [ + { + "bbox": [ + 106, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 187, + 496 + ], + "score": 1.0, + "content": "Typically, the model", + "type": "text" + }, + { + "bbox": [ + 188, + 484, + 233, + 496 + ], + "score": 0.92, + "content": "\\mu _ { \\theta } ( x _ { t - t } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 483, + 401, + 496 + ], + "score": 1.0, + "content": "is not parameterized to predict the mean of", + "type": "text" + }, + { + "bbox": [ + 401, + 486, + 411, + 495 + ], + "score": 0.82, + "content": "x _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 483, + 505, + 496 + ], + "score": 1.0, + "content": ". Instead it is parameter-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 495, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 239, + 508 + ], + "score": 1.0, + "content": "ized to predict the noise added to", + "type": "text" + }, + { + "bbox": [ + 239, + 497, + 250, + 506 + ], + "score": 0.86, + "content": "x _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 495, + 296, + 508 + ], + "score": 1.0, + "content": "to arrive at", + "type": "text" + }, + { + "bbox": [ + 296, + 496, + 316, + 507 + ], + "score": 0.89, + "content": "x _ { t - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 495, + 505, + 508 + ], + "score": 1.0, + "content": ". This motivates the following parameterization", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 512, + 398, + 540 + ], + "lines": [ + { + "bbox": [ + 212, + 512, + 398, + 540 + ], + "spans": [ + { + "bbox": [ + 212, + 512, + 398, + 540 + ], + "score": 0.95, + "content": "\\mu _ { \\theta } ( x _ { t } , t ) = \\frac { 1 } { \\sqrt { \\alpha _ { t } } } \\left( x _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - \\bar { \\alpha } _ { t } } } \\epsilon _ { \\theta } ( x _ { t } , t ) \\right) .", + "type": "interline_equation", + "image_path": "1a11ef45d0966957a9faa2f9d42efdc2b6ddbaa6ca0ccfddb45b948447d2db8b.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 212, + 512, + 398, + 540 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 551, + 372, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 550, + 372, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 372, + 565 + ], + "score": 1.0, + "content": "In this form we can rewrite the important terms in the objective as", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 568, + 477, + 583 + ], + "lines": [ + { + "bbox": [ + 115, + 568, + 477, + 583 + ], + "spans": [ + { + "bbox": [ + 115, + 568, + 477, + 583 + ], + "score": 0.91, + "content": "D _ { K L } ( q ( x _ { t } | x _ { t - 1 } , x _ { 0 } ) | | p ( x _ { t } | | x _ { t - 1 } ) ) = - C _ { t } \\mathbf { E } _ { q ( x _ { 0 } ) , N ( \\epsilon ; 0 , I ) } \\left[ | | \\epsilon - \\epsilon _ { \\theta } ( x _ { t } , t ) | | ^ { 2 } \\right] = C _ { t } \\mathcal { L } ( x , \\sigma )", + "type": "interline_equation", + "image_path": "a7dead8216e272210d78e19956f9dd833f6d55f20d1ba05877753bf6de9c5bf3.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 115, + 568, + 477, + 583 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 587, + 505, + 611 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 133, + 601 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 588, + 145, + 599 + ], + "score": 0.88, + "content": "C _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 586, + 505, + 601 + ], + "score": 1.0, + "content": "is a time-dependent constant. Typically these are dropped and all objectives are weighted", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 598, + 142, + 614 + ], + "spans": [ + { + "bbox": [ + 104, + 598, + 142, + 614 + ], + "score": 1.0, + "content": "equally.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 108, + 615, + 504, + 647 + ], + "lines": [ + { + "bbox": [ + 106, + 614, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 434, + 630 + ], + "score": 1.0, + "content": "Once we finish training, we can draw samples from our model by first sampling", + "type": "text" + }, + { + "bbox": [ + 435, + 616, + 486, + 628 + ], + "score": 0.93, + "content": "x _ { T } \\sim p ( x _ { T } )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 614, + 505, + 630 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 103, + 627, + 402, + 649 + ], + "spans": [ + { + "bbox": [ + 103, + 627, + 212, + 649 + ], + "score": 1.0, + "content": "then equentially sampling", + "type": "text" + }, + { + "bbox": [ + 212, + 627, + 317, + 648 + ], + "score": 0.94, + "content": "x _ { t } = \\mu _ { \\theta } ( x _ { t - 1 } , t ) + \\sqrt { \\tilde { \\beta } _ { t } } \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 627, + 344, + 649 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 345, + 632, + 395, + 645 + ], + "score": 0.9, + "content": "\\epsilon \\sim \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 627, + 402, + 649 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 108, + 661, + 274, + 675 + ], + "lines": [ + { + "bbox": [ + 105, + 660, + 275, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 275, + 678 + ], + "score": 1.0, + "content": "C MCMC SAMPLING DETAILS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 106, + 688, + 351, + 701 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 351, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 351, + 701 + ], + "score": 1.0, + "content": "C.1 HAMILTONIAN MONTE-CARLO AND ITS VARIANTS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "Hamiltonian Monte-Carlo (Neal, 1996) seeks to sample from an unnormalized probability distribution", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 720, + 504, + 733 + ], + "spans": [ + { + "bbox": [ + 107, + 720, + 208, + 732 + ], + "score": 0.93, + "content": "\\log p ( x ) = f ( x ) + \\log Z", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 720, + 392, + 733 + ], + "score": 1.0, + "content": ". To do this, we augment our distribution over", + "type": "text" + }, + { + "bbox": [ + 392, + 723, + 399, + 730 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 720, + 497, + 733 + ], + "score": 1.0, + "content": "with auxillury variables", + "type": "text" + }, + { + "bbox": [ + 497, + 723, + 504, + 730 + ], + "score": 0.69, + "content": "v", + "type": "inline_equation" + } + ], + "index": 19 + } + ], + "index": 18.5 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 80, + 503, + 405 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 80, + 503, + 405 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 80, + 503, + 405 + ], + "spans": [ + { + "bbox": [ + 112, + 80, + 503, + 405 + ], + "score": 0.978, + "type": "image", + "image_path": "97483bdb0cce6ccf6854f297ede1fbe5b048eaf3ebdac470949060c514476bad.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 80, + 503, + 188.33333333333331 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 188.33333333333331, + 503, + 296.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 296.66666666666663, + 503, + 404.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 413, + 505, + 435 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 412, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 505, + 427 + ], + "score": 1.0, + "content": "Figure A3: By composing an energy parameterized diffusion model with the negation of another energy", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 424, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 436 + ], + "score": 1.0, + "content": "parameterized diffusion model, we can render images in unusual configurations not typically found in the data.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 455, + 503, + 479 + ], + "lines": [ + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 465, + 469 + ], + "score": 1.0, + "content": "The first term lacks parameters and the final is approximately 0 from the convergence of the", + "type": "text" + }, + { + "bbox": [ + 465, + 459, + 471, + 468 + ], + "score": 0.79, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 455, + 505, + 469 + ], + "score": 1.0, + "content": "-process", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 467, + 241, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 241, + 480 + ], + "score": 1.0, + "content": "so we focus on the middle terms.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 105, + 455, + 505, + 480 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 484, + 503, + 507 + ], + "lines": [ + { + "bbox": [ + 106, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 187, + 496 + ], + "score": 1.0, + "content": "Typically, the model", + "type": "text" + }, + { + "bbox": [ + 188, + 484, + 233, + 496 + ], + "score": 0.92, + "content": "\\mu _ { \\theta } ( x _ { t - t } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 483, + 401, + 496 + ], + "score": 1.0, + "content": "is not parameterized to predict the mean of", + "type": "text" + }, + { + "bbox": [ + 401, + 486, + 411, + 495 + ], + "score": 0.82, + "content": "x _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 483, + 505, + 496 + ], + "score": 1.0, + "content": ". Instead it is parameter-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 495, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 239, + 508 + ], + "score": 1.0, + "content": "ized to predict the noise added to", + "type": "text" + }, + { + "bbox": [ + 239, + 497, + 250, + 506 + ], + "score": 0.86, + "content": "x _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 495, + 296, + 508 + ], + "score": 1.0, + "content": "to arrive at", + "type": "text" + }, + { + "bbox": [ + 296, + 496, + 316, + 507 + ], + "score": 0.89, + "content": "x _ { t - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 495, + 505, + 508 + ], + "score": 1.0, + "content": ". This motivates the following parameterization", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 483, + 505, + 508 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 512, + 398, + 540 + ], + "lines": [ + { + "bbox": [ + 212, + 512, + 398, + 540 + ], + "spans": [ + { + "bbox": [ + 212, + 512, + 398, + 540 + ], + "score": 0.95, + "content": "\\mu _ { \\theta } ( x _ { t } , t ) = \\frac { 1 } { \\sqrt { \\alpha _ { t } } } \\left( x _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - \\bar { \\alpha } _ { t } } } \\epsilon _ { \\theta } ( x _ { t } , t ) \\right) .", + "type": "interline_equation", + "image_path": "1a11ef45d0966957a9faa2f9d42efdc2b6ddbaa6ca0ccfddb45b948447d2db8b.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 212, + 512, + 398, + 540 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 551, + 372, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 550, + 372, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 372, + 565 + ], + "score": 1.0, + "content": "In this form we can rewrite the important terms in the objective as", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 550, + 372, + 565 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 568, + 477, + 583 + ], + "lines": [ + { + "bbox": [ + 115, + 568, + 477, + 583 + ], + "spans": [ + { + "bbox": [ + 115, + 568, + 477, + 583 + ], + "score": 0.91, + "content": "D _ { K L } ( q ( x _ { t } | x _ { t - 1 } , x _ { 0 } ) | | p ( x _ { t } | | x _ { t - 1 } ) ) = - C _ { t } \\mathbf { E } _ { q ( x _ { 0 } ) , N ( \\epsilon ; 0 , I ) } \\left[ | | \\epsilon - \\epsilon _ { \\theta } ( x _ { t } , t ) | | ^ { 2 } \\right] = C _ { t } \\mathcal { L } ( x , \\sigma )", + "type": "interline_equation", + "image_path": "a7dead8216e272210d78e19956f9dd833f6d55f20d1ba05877753bf6de9c5bf3.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 115, + 568, + 477, + 583 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 587, + 505, + 611 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 133, + 601 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 588, + 145, + 599 + ], + "score": 0.88, + "content": "C _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 586, + 505, + 601 + ], + "score": 1.0, + "content": "is a time-dependent constant. Typically these are dropped and all objectives are weighted", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 598, + 142, + 614 + ], + "spans": [ + { + "bbox": [ + 104, + 598, + 142, + 614 + ], + "score": 1.0, + "content": "equally.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 104, + 586, + 505, + 614 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 615, + 504, + 647 + ], + "lines": [ + { + "bbox": [ + 106, + 614, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 434, + 630 + ], + "score": 1.0, + "content": "Once we finish training, we can draw samples from our model by first sampling", + "type": "text" + }, + { + "bbox": [ + 435, + 616, + 486, + 628 + ], + "score": 0.93, + "content": "x _ { T } \\sim p ( x _ { T } )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 614, + 505, + 630 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 103, + 627, + 402, + 649 + ], + "spans": [ + { + "bbox": [ + 103, + 627, + 212, + 649 + ], + "score": 1.0, + "content": "then equentially sampling", + "type": "text" + }, + { + "bbox": [ + 212, + 627, + 317, + 648 + ], + "score": 0.94, + "content": "x _ { t } = \\mu _ { \\theta } ( x _ { t - 1 } , t ) + \\sqrt { \\tilde { \\beta } _ { t } } \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 627, + 344, + 649 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 345, + 632, + 395, + 645 + ], + "score": 0.9, + "content": "\\epsilon \\sim \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 627, + 402, + 649 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 103, + 614, + 505, + 649 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 661, + 274, + 675 + ], + "lines": [ + { + "bbox": [ + 105, + 660, + 275, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 275, + 678 + ], + "score": 1.0, + "content": "C MCMC SAMPLING DETAILS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 106, + 688, + 351, + 701 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 351, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 351, + 701 + ], + "score": 1.0, + "content": "C.1 HAMILTONIAN MONTE-CARLO AND ITS VARIANTS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "Hamiltonian Monte-Carlo (Neal, 1996) seeks to sample from an unnormalized probability distribution", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 720, + 504, + 733 + ], + "spans": [ + { + "bbox": [ + 107, + 720, + 208, + 732 + ], + "score": 0.93, + "content": "\\log p ( x ) = f ( x ) + \\log Z", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 720, + 392, + 733 + ], + "score": 1.0, + "content": ". To do this, we augment our distribution over", + "type": "text" + }, + { + "bbox": [ + 392, + 723, + 399, + 730 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 720, + 497, + 733 + ], + "score": 1.0, + "content": "with auxillury variables", + "type": "text" + }, + { + "bbox": [ + 497, + 723, + 504, + 730 + ], + "score": 0.69, + "content": "v", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 232, + 393 + ], + "score": 1.0, + "content": "and define the joint distribution", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 233, + 380, + 340, + 393 + ], + "score": 0.93, + "content": "p ( x , v ) = p ( x ) \\mathcal { N } ( v ; 0 , M )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 341, + 380, + 428, + 393 + ], + "score": 1.0, + "content": "where the covariance", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 428, + 381, + 441, + 390 + ], + "score": 0.76, + "content": "M", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 441, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "is known as the", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 286, + 404 + ], + "score": 1.0, + "content": "“mass-matrix.” We now seek to draw samples", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 286, + 392, + 344, + 403 + ], + "score": 0.91, + "content": "x , v \\sim p ( x , v )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 344, + 391, + 383, + 404 + ], + "score": 1.0, + "content": "and since", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 384, + 393, + 391, + 401 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 392, + 391, + 408, + 404 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 408, + 393, + 415, + 401 + ], + "score": 0.76, + "content": "v", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 415, + 391, + 505, + 404 + ], + "score": 1.0, + "content": "are independent under", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 402, + 453, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 269, + 415 + ], + "score": 1.0, + "content": "the joint, we can simply throw away our", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 269, + 405, + 276, + 412 + ], + "score": 0.75, + "content": "v", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 276, + 403, + 412, + 415 + ], + "score": 1.0, + "content": "samples leaving us with a sample", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 412, + 402, + 450, + 415 + ], + "score": 0.93, + "content": "x \\sim p ( x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 451, + 403, + 453, + 415 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 7 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 708, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 79, + 505, + 322 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 79, + 505, + 322 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 79, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 112, + 79, + 505, + 322 + ], + "score": 0.974, + "type": "image", + "image_path": "33bfde1f9d20ade462b5cc2b09f14a0a4964f4a1e41214ab90a135b401948f6d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 79, + 505, + 160.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 160.0, + 505, + 241.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 241.0, + 505, + 322.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 331, + 504, + 352 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 331, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 506, + 343 + ], + "score": 1.0, + "content": "Figure A4: By composing multiple energy parameterized diffusion models, we can more accurately render the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 341, + 252, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 252, + 352 + ], + "score": 1.0, + "content": "underlying number of objects in ascene.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 380, + 505, + 414 + ], + "lines": [ + { + "bbox": [ + 106, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 232, + 393 + ], + "score": 1.0, + "content": "and define the joint distribution", + "type": "text" + }, + { + "bbox": [ + 233, + 380, + 340, + 393 + ], + "score": 0.93, + "content": "p ( x , v ) = p ( x ) \\mathcal { N } ( v ; 0 , M )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 380, + 428, + 393 + ], + "score": 1.0, + "content": "where the covariance", + "type": "text" + }, + { + "bbox": [ + 428, + 381, + 441, + 390 + ], + "score": 0.76, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "is known as the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 286, + 404 + ], + "score": 1.0, + "content": "“mass-matrix.” We now seek to draw samples", + "type": "text" + }, + { + "bbox": [ + 286, + 392, + 344, + 403 + ], + "score": 0.91, + "content": "x , v \\sim p ( x , v )", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 391, + 383, + 404 + ], + "score": 1.0, + "content": "and since", + "type": "text" + }, + { + "bbox": [ + 384, + 393, + 391, + 401 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 391, + 408, + 404 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 408, + 393, + 415, + 401 + ], + "score": 0.76, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 391, + 505, + 404 + ], + "score": 1.0, + "content": "are independent under", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 402, + 453, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 269, + 415 + ], + "score": 1.0, + "content": "the joint, we can simply throw away our", + "type": "text" + }, + { + "bbox": [ + 269, + 405, + 276, + 412 + ], + "score": 0.75, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 403, + 412, + 415 + ], + "score": 1.0, + "content": "samples leaving us with a sample", + "type": "text" + }, + { + "bbox": [ + 412, + 402, + 450, + 415 + ], + "score": 0.93, + "content": "x \\sim p ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 403, + 453, + 415 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 419, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 504, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 356, + 432 + ], + "score": 1.0, + "content": "Like other MCMC methods we sequentially update a particle", + "type": "text" + }, + { + "bbox": [ + 356, + 419, + 386, + 431 + ], + "score": 0.92, + "content": "( x ^ { i } , v ^ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 418, + 473, + 432 + ], + "score": 1.0, + "content": "in such a way that as", + "type": "text" + }, + { + "bbox": [ + 474, + 420, + 504, + 430 + ], + "score": 0.88, + "content": "i \\to \\infty", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 227, + 442 + ], + "score": 1.0, + "content": "we arrive at a sample from", + "type": "text" + }, + { + "bbox": [ + 227, + 430, + 256, + 442 + ], + "score": 0.92, + "content": "p ( x , v )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 430, + 403, + 442 + ], + "score": 1.0, + "content": ". For a step of HMC, starting at", + "type": "text" + }, + { + "bbox": [ + 404, + 430, + 434, + 442 + ], + "score": 0.93, + "content": "( x ^ { i } , v ^ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 430, + 505, + 442 + ], + "score": 1.0, + "content": "we first sample", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 439, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 107, + 441, + 183, + 453 + ], + "score": 0.92, + "content": "v ^ { i \\prime } \\sim \\mathcal { N } ( v ^ { i } ; 0 , M )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 439, + 363, + 455 + ], + "score": 1.0, + "content": "since the target distribution factorizes and", + "type": "text" + }, + { + "bbox": [ + 364, + 442, + 383, + 453 + ], + "score": 0.91, + "content": "p ( v )", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 439, + 506, + 455 + ], + "score": 1.0, + "content": "is known and tractable. We", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 452, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 323, + 465 + ], + "score": 1.0, + "content": "then integrate a likelihood-conserving ODE defined on", + "type": "text" + }, + { + "bbox": [ + 323, + 453, + 340, + 464 + ], + "score": 0.84, + "content": "x , v", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 452, + 505, + 465 + ], + "score": 1.0, + "content": "known as “Hamiltoninan Dynamics.” We", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 463, + 507, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 507, + 476 + ], + "score": 1.0, + "content": "can use the likelihood-preserving leapfrog integrator which will guarentee that the transition distri-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 473, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 203, + 487 + ], + "score": 1.0, + "content": "bution is symmetric, i.e", + "type": "text" + }, + { + "bbox": [ + 203, + 474, + 322, + 486 + ], + "score": 0.93, + "content": "k ( x ^ { \\prime } , v ^ { \\prime } | \\bar { x } , v ) ^ { \\top } = \\bar { k } ( x , v | \\bar { x } ^ { \\prime } , v ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 473, + 505, + 487 + ], + "score": 1.0, + "content": ". Thus, the Metropolis acceptance probability", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 485, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 157, + 506 + ], + "score": 1.0, + "content": "simplifies to", + "type": "text" + }, + { + "bbox": [ + 157, + 486, + 227, + 505 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\operatorname* { m i n } \\left( 1 , \\frac { p ( x ^ { \\prime } , v ^ { \\prime } ) } { p ( x , v ) } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 487, + 506, + 502 + ], + "score": 1.0, + "content": ". An overview of the HMC algorithm can be found in Algorithm 2. We", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 503, + 429, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 429, + 516 + ], + "score": 1.0, + "content": "refer the reader to Neal (1996) for a more complete description of the algorithm.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 107, + 535, + 265, + 547 + ], + "lines": [ + { + "bbox": [ + 106, + 534, + 267, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 267, + 549 + ], + "score": 1.0, + "content": "Algorithm 2 Hamiltonian Monte-Carlo", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "table", + "bbox": [ + 112, + 551, + 504, + 683 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 112, + 551, + 504, + 683 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 551, + 504, + 683 + ], + "spans": [ + { + "bbox": [ + 112, + 551, + 504, + 683 + ], + "score": 0.777, + "html": "
Input: Initial state x°,Mass matrix M, Number of steps N,Number leapfrog steps L,step-size ∈
fori=1,...,N do
Sample vi ~ N(0, M)# Sample momentum
x',uu =Leapfrog(xi-1, ,U²;,L)#Integrate dynamics with stepsize e for L steps
α = min p(x,') 1, p(xi-1,0)# Compute acceptance probability
With probability a
set xi= x'
else
set xi=xi-1
", + "type": "table", + "image_path": "48dc517eac89cd9d7101dfa19c58791a61504b882a0895cd8414f6d2f9412f35.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 112, + 551, + 504, + 595.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 112, + 595.0, + 504, + 639.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 112, + 639.0, + 504, + 683.0 + ], + "spans": [], + "index": 19 + } + ] + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 149, + 723 + ], + "score": 1.0, + "content": "Since our", + "type": "text" + }, + { + "bbox": [ + 149, + 709, + 180, + 722 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "parameterized models do not admit an explicit likelihood function, we use an", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 720, + 447, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 447, + 733 + ], + "score": 1.0, + "content": "unadjusted variant of HMC (U-HMC) where the accept/reject step is simply ignored.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 112, + 79, + 505, + 322 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 79, + 505, + 322 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 79, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 112, + 79, + 505, + 322 + ], + "score": 0.974, + "type": "image", + "image_path": "33bfde1f9d20ade462b5cc2b09f14a0a4964f4a1e41214ab90a135b401948f6d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 79, + 505, + 160.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 160.0, + 505, + 241.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 241.0, + 505, + 322.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 331, + 504, + 352 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 331, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 506, + 343 + ], + "score": 1.0, + "content": "Figure A4: By composing multiple energy parameterized diffusion models, we can more accurately render the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 341, + 252, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 252, + 352 + ], + "score": 1.0, + "content": "underlying number of objects in ascene.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 380, + 505, + 414 + ], + "lines": [], + "index": 6, + "bbox_fs": [ + 105, + 380, + 505, + 415 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 419, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 504, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 356, + 432 + ], + "score": 1.0, + "content": "Like other MCMC methods we sequentially update a particle", + "type": "text" + }, + { + "bbox": [ + 356, + 419, + 386, + 431 + ], + "score": 0.92, + "content": "( x ^ { i } , v ^ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 418, + 473, + 432 + ], + "score": 1.0, + "content": "in such a way that as", + "type": "text" + }, + { + "bbox": [ + 474, + 420, + 504, + 430 + ], + "score": 0.88, + "content": "i \\to \\infty", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 430, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 227, + 442 + ], + "score": 1.0, + "content": "we arrive at a sample from", + "type": "text" + }, + { + "bbox": [ + 227, + 430, + 256, + 442 + ], + "score": 0.92, + "content": "p ( x , v )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 430, + 403, + 442 + ], + "score": 1.0, + "content": ". For a step of HMC, starting at", + "type": "text" + }, + { + "bbox": [ + 404, + 430, + 434, + 442 + ], + "score": 0.93, + "content": "( x ^ { i } , v ^ { i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 430, + 505, + 442 + ], + "score": 1.0, + "content": "we first sample", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 439, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 107, + 441, + 183, + 453 + ], + "score": 0.92, + "content": "v ^ { i \\prime } \\sim \\mathcal { N } ( v ^ { i } ; 0 , M )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 439, + 363, + 455 + ], + "score": 1.0, + "content": "since the target distribution factorizes and", + "type": "text" + }, + { + "bbox": [ + 364, + 442, + 383, + 453 + ], + "score": 0.91, + "content": "p ( v )", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 439, + 506, + 455 + ], + "score": 1.0, + "content": "is known and tractable. We", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 452, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 323, + 465 + ], + "score": 1.0, + "content": "then integrate a likelihood-conserving ODE defined on", + "type": "text" + }, + { + "bbox": [ + 323, + 453, + 340, + 464 + ], + "score": 0.84, + "content": "x , v", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 452, + 505, + 465 + ], + "score": 1.0, + "content": "known as “Hamiltoninan Dynamics.” We", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 463, + 507, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 507, + 476 + ], + "score": 1.0, + "content": "can use the likelihood-preserving leapfrog integrator which will guarentee that the transition distri-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 473, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 203, + 487 + ], + "score": 1.0, + "content": "bution is symmetric, i.e", + "type": "text" + }, + { + "bbox": [ + 203, + 474, + 322, + 486 + ], + "score": 0.93, + "content": "k ( x ^ { \\prime } , v ^ { \\prime } | \\bar { x } , v ) ^ { \\top } = \\bar { k } ( x , v | \\bar { x } ^ { \\prime } , v ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 473, + 505, + 487 + ], + "score": 1.0, + "content": ". Thus, the Metropolis acceptance probability", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 485, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 157, + 506 + ], + "score": 1.0, + "content": "simplifies to", + "type": "text" + }, + { + "bbox": [ + 157, + 486, + 227, + 505 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\operatorname* { m i n } \\left( 1 , \\frac { p ( x ^ { \\prime } , v ^ { \\prime } ) } { p ( x , v ) } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 487, + 506, + 502 + ], + "score": 1.0, + "content": ". An overview of the HMC algorithm can be found in Algorithm 2. We", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 503, + 429, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 429, + 516 + ], + "score": 1.0, + "content": "refer the reader to Neal (1996) for a more complete description of the algorithm.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 418, + 507, + 516 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 535, + 265, + 547 + ], + "lines": [ + { + "bbox": [ + 106, + 534, + 267, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 267, + 549 + ], + "score": 1.0, + "content": "Algorithm 2 Hamiltonian Monte-Carlo", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "table", + "bbox": [ + 112, + 551, + 504, + 683 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 112, + 551, + 504, + 683 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 551, + 504, + 683 + ], + "spans": [ + { + "bbox": [ + 112, + 551, + 504, + 683 + ], + "score": 0.777, + "html": "
Input: Initial state x°,Mass matrix M, Number of steps N,Number leapfrog steps L,step-size ∈
fori=1,...,N do
Sample vi ~ N(0, M)# Sample momentum
x',uu =Leapfrog(xi-1, ,U²;,L)#Integrate dynamics with stepsize e for L steps
α = min p(x,') 1, p(xi-1,0)# Compute acceptance probability
With probability a
set xi= x'
else
set xi=xi-1
", + "type": "table", + "image_path": "48dc517eac89cd9d7101dfa19c58791a61504b882a0895cd8414f6d2f9412f35.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 112, + 551, + 504, + 595.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 112, + 595.0, + 504, + 639.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 112, + 639.0, + 504, + 683.0 + ], + "spans": [], + "index": 19 + } + ] + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 149, + 723 + ], + "score": 1.0, + "content": "Since our", + "type": "text" + }, + { + "bbox": [ + 149, + 709, + 180, + 722 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "parameterized models do not admit an explicit likelihood function, we use an", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 720, + 447, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 447, + 733 + ], + "score": 1.0, + "content": "unadjusted variant of HMC (U-HMC) where the accept/reject step is simply ignored.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 708, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 192 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "We can see in Algorithm 2 that at every step, the momentum is re-sampled. This can be sub-optimal,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 311, + 105 + ], + "score": 1.0, + "content": "as the momentum determines the initial direction of", + "type": "text" + }, + { + "bbox": [ + 311, + 95, + 318, + 104 + ], + "score": 0.71, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 94, + 506, + 105 + ], + "score": 1.0, + "content": "’s movement and if a good direction is found, it", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "may be beneficial to continue in that direction. To deal with this, Neal (1996) proposes a variant of", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 223, + 128 + ], + "score": 1.0, + "content": "HMC where the momentum", + "type": "text" + }, + { + "bbox": [ + 224, + 117, + 231, + 125 + ], + "score": 0.72, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "is partially retained between sampling steps. We add an additional", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 504, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 181, + 139 + ], + "score": 1.0, + "content": "sampler parameter", + "type": "text" + }, + { + "bbox": [ + 182, + 127, + 221, + 138 + ], + "score": 0.92, + "content": "\\gamma \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 127, + 497, + 139 + ], + "score": 1.0, + "content": "(known as the “damping-factor”) which controls the amount to which", + "type": "text" + }, + { + "bbox": [ + 497, + 128, + 504, + 136 + ], + "score": 0.64, + "content": "v", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 179, + 150 + ], + "score": 1.0, + "content": "is retained. When", + "type": "text" + }, + { + "bbox": [ + 180, + 139, + 187, + 149 + ], + "score": 0.82, + "content": "\\gamma", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 137, + 240, + 150 + ], + "score": 1.0, + "content": "is close to 1,", + "type": "text" + }, + { + "bbox": [ + 241, + 139, + 247, + 147 + ], + "score": 0.58, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 137, + 395, + 150 + ], + "score": 1.0, + "content": "is mostly kept and when it is near 0,", + "type": "text" + }, + { + "bbox": [ + 395, + 139, + 402, + 147 + ], + "score": 0.51, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 137, + 506, + 150 + ], + "score": 1.0, + "content": "is mostly refreshed. This", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 506, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 506, + 161 + ], + "score": 1.0, + "content": "variant is summarized in Algorithm 3. The potentially confusing momentum negations ensure the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 173 + ], + "score": 1.0, + "content": "validity of the sampler. Intuitively, when the proposal is accepted, the momentum is retained and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 170, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 170, + 505, + 183 + ], + "score": 1.0, + "content": "when it is rejected the momentum is flipped. For this reason, one should maintain a reasonably high", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 182, + 272, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 272, + 194 + ], + "score": 1.0, + "content": "acceptance rate when using this approch.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 4.5 + }, + { + "type": "table", + "bbox": [ + 108, + 221, + 504, + 421 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 205, + 417, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 204, + 417, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 417, + 218 + ], + "score": 1.0, + "content": "Algorithm 3 Hamiltonian Monte-Carlo with Partial Momentum Refreshment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "table_body", + "bbox": [ + 108, + 221, + 504, + 421 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 223, + 500, + 421 + ], + "spans": [ + { + "bbox": [ + 108, + 223, + 500, + 421 + ], + "score": 0.787, + "html": "
damping-factor yInput: Initial state xo,Mass matrix M,Number of steps N,Number leapfrog steps L,step-size
Sample v° ~ N(0, M)#Sample initial momentum
for i= 1,...,N do
λ~ N(0,M)
u(i-1)'=γv-1+√1-γ²x#Partially refresh momentum
x',v'=Leapfrog(x𝑖-1,u(i-1)';∈,L)# Integrate dynamics with stepsize εfor L steps
u'=-v' # Negate momentum
a = min (1, p(x',u')
p(xi-1,u(i-1)1))# Compute acceptance probability
With probability a
set xi= x',v = U'
else
set xi= xi-1,v²= v(i-1)1
U²=-ui # Negate momentum
end for
xN
return
", + "type": "table", + "image_path": "d4fc6fb9b0b0032a23cc88aa11277a24167f400673729444c6a54a54eb4852fc.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 108, + 221, + 504, + 287.6666666666667 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 108, + 287.6666666666667, + 504, + 354.33333333333337 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 108, + 354.33333333333337, + 504, + 421.00000000000006 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 11.0 + }, + { + "type": "title", + "bbox": [ + 107, + 440, + 204, + 452 + ], + "lines": [ + { + "bbox": [ + 106, + 439, + 205, + 454 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 205, + 454 + ], + "score": 1.0, + "content": "C.2 MCMC TUNING", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 461, + 505, + 493 + ], + "lines": [ + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "A crucial component to ensure successful MCMC sampling in diffusion models is the choice of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 471, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 443, + 485 + ], + "score": 1.0, + "content": "step sizes for samplers. We initialize step sizes for all samplers at each distribution", + "type": "text" + }, + { + "bbox": [ + 443, + 473, + 448, + 482 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 471, + 505, + 485 + ], + "score": 1.0, + "content": "to be roughly", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 482, + 435, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 183, + 495 + ], + "score": 1.0, + "content": "proportional to the", + "type": "text" + }, + { + "bbox": [ + 183, + 483, + 194, + 494 + ], + "score": 0.88, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 482, + 330, + 495 + ], + "score": 1.0, + "content": "noise values added to distribution", + "type": "text" + }, + { + "bbox": [ + 331, + 484, + 336, + 492 + ], + "score": 0.64, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 482, + 435, + 495 + ], + "score": 1.0, + "content": "in the diffusion process.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 499, + 505, + 566 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "To tune step sizes across timesteps for both HMC and MALA samplers, to set step sizes at each", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 144, + 523 + ], + "score": 1.0, + "content": "timestep", + "type": "text" + }, + { + "bbox": [ + 144, + 512, + 149, + 520 + ], + "score": 0.64, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 511, + 271, + 523 + ], + "score": 1.0, + "content": "to be constant multiplied by", + "type": "text" + }, + { + "bbox": [ + 271, + 511, + 281, + 522 + ], + "score": 0.87, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 511, + 472, + 523 + ], + "score": 1.0, + "content": ". We searched different constants to multiply", + "type": "text" + }, + { + "bbox": [ + 473, + 511, + 483, + 522 + ], + "score": 0.87, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 511, + 505, + 523 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "chose a value so that the average acceptance rate of MALA and HMC samplers across timesteps is", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 532, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 167, + 546 + ], + "score": 1.0, + "content": "approximately", + "type": "text" + }, + { + "bbox": [ + 168, + 532, + 188, + 543 + ], + "score": 0.87, + "content": "60 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 532, + 206, + 546 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 206, + 533, + 226, + 543 + ], + "score": 0.88, + "content": "70 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 532, + 506, + 546 + ], + "score": 1.0, + "content": "respectively. For un-adjusted variants of these samplers, we set step", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "sizes to be the same as adjusted samplers, and found limited gains when step sizes were specifically", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 554, + 472, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 381, + 567 + ], + "score": 1.0, + "content": "tuned towards the un-adjusted samplers. We utilize a mass matrix of", + "type": "text" + }, + { + "bbox": [ + 381, + 555, + 391, + 566 + ], + "score": 0.88, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 554, + 472, + 567 + ], + "score": 1.0, + "content": "for HMC samplers.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 107, + 571, + 492, + 582 + ], + "lines": [ + { + "bbox": [ + 106, + 570, + 493, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 493, + 584 + ], + "score": 1.0, + "content": "Precise details on the exact MCMC steps sizes used in experiments can be detailed in Section G.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 109, + 595, + 286, + 606 + ], + "lines": [ + { + "bbox": [ + 107, + 595, + 288, + 608 + ], + "spans": [ + { + "bbox": [ + 107, + 595, + 288, + 608 + ], + "score": 1.0, + "content": "C.3 MCMC IMPLEMENTATION DETAILS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 615, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "score": 1.0, + "content": "When initially running MCMC sampling on diffusion models in the image domain, we found that", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 626, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 641 + ], + "score": 1.0, + "content": "our samplers tended to converge to images which had uniform textures. After experimentation, we", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 639, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 505, + 650 + ], + "score": 1.0, + "content": "found that the primary cause of this issue was fact that by default, typical implementations of the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 649, + 507, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 469, + 661 + ], + "score": 1.0, + "content": "reverse diffusion process clip samples at intermediate time-steps of sampling to be between", + "type": "text" + }, + { + "bbox": [ + 469, + 649, + 479, + 659 + ], + "score": 0.29, + "content": "^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 649, + 507, + 661 + ], + "score": 1.0, + "content": "and 1.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 659, + 507, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 507, + 673 + ], + "score": 1.0, + "content": "To enable proper MCMC sampling, we found that it was important to not clip intermediate values of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 669, + 186, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 186, + 685 + ], + "score": 1.0, + "content": "diffusion sampling.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "When running MCMC sampling on image domains, we further found that it was helpful for mixing", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "to run a single step of the reverse process to initialize MCMC sampling, before running many steps", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 709, + 507, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 259, + 723 + ], + "score": 1.0, + "content": "of MCMC sampling at each timestep", + "type": "text" + }, + { + "bbox": [ + 260, + 711, + 264, + 720 + ], + "score": 0.59, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 709, + 507, + 723 + ], + "score": 1.0, + "content": ", and in all MCMC sampling settings on the image domain,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "we run one step of the reverse process before running MCMC sampling. Such a MCMC sampling", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 192 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "We can see in Algorithm 2 that at every step, the momentum is re-sampled. This can be sub-optimal,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 311, + 105 + ], + "score": 1.0, + "content": "as the momentum determines the initial direction of", + "type": "text" + }, + { + "bbox": [ + 311, + 95, + 318, + 104 + ], + "score": 0.71, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 94, + 506, + 105 + ], + "score": 1.0, + "content": "’s movement and if a good direction is found, it", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "may be beneficial to continue in that direction. To deal with this, Neal (1996) proposes a variant of", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 223, + 128 + ], + "score": 1.0, + "content": "HMC where the momentum", + "type": "text" + }, + { + "bbox": [ + 224, + 117, + 231, + 125 + ], + "score": 0.72, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "is partially retained between sampling steps. We add an additional", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 504, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 181, + 139 + ], + "score": 1.0, + "content": "sampler parameter", + "type": "text" + }, + { + "bbox": [ + 182, + 127, + 221, + 138 + ], + "score": 0.92, + "content": "\\gamma \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 127, + 497, + 139 + ], + "score": 1.0, + "content": "(known as the “damping-factor”) which controls the amount to which", + "type": "text" + }, + { + "bbox": [ + 497, + 128, + 504, + 136 + ], + "score": 0.64, + "content": "v", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 179, + 150 + ], + "score": 1.0, + "content": "is retained. When", + "type": "text" + }, + { + "bbox": [ + 180, + 139, + 187, + 149 + ], + "score": 0.82, + "content": "\\gamma", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 137, + 240, + 150 + ], + "score": 1.0, + "content": "is close to 1,", + "type": "text" + }, + { + "bbox": [ + 241, + 139, + 247, + 147 + ], + "score": 0.58, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 137, + 395, + 150 + ], + "score": 1.0, + "content": "is mostly kept and when it is near 0,", + "type": "text" + }, + { + "bbox": [ + 395, + 139, + 402, + 147 + ], + "score": 0.51, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 137, + 506, + 150 + ], + "score": 1.0, + "content": "is mostly refreshed. This", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 506, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 506, + 161 + ], + "score": 1.0, + "content": "variant is summarized in Algorithm 3. The potentially confusing momentum negations ensure the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 173 + ], + "score": 1.0, + "content": "validity of the sampler. Intuitively, when the proposal is accepted, the momentum is retained and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 170, + 505, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 170, + 505, + 183 + ], + "score": 1.0, + "content": "when it is rejected the momentum is flipped. For this reason, one should maintain a reasonably high", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 182, + 272, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 272, + 194 + ], + "score": 1.0, + "content": "acceptance rate when using this approch.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 82, + 506, + 194 + ] + }, + { + "type": "table", + "bbox": [ + 108, + 221, + 504, + 421 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 205, + 417, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 204, + 417, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 417, + 218 + ], + "score": 1.0, + "content": "Algorithm 3 Hamiltonian Monte-Carlo with Partial Momentum Refreshment", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "table_body", + "bbox": [ + 108, + 221, + 504, + 421 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 223, + 500, + 421 + ], + "spans": [ + { + "bbox": [ + 108, + 223, + 500, + 421 + ], + "score": 0.787, + "html": "
damping-factor yInput: Initial state xo,Mass matrix M,Number of steps N,Number leapfrog steps L,step-size
Sample v° ~ N(0, M)#Sample initial momentum
for i= 1,...,N do
λ~ N(0,M)
u(i-1)'=γv-1+√1-γ²x#Partially refresh momentum
x',v'=Leapfrog(x𝑖-1,u(i-1)';∈,L)# Integrate dynamics with stepsize εfor L steps
u'=-v' # Negate momentum
a = min (1, p(x',u')
p(xi-1,u(i-1)1))# Compute acceptance probability
With probability a
set xi= x',v = U'
else
set xi= xi-1,v²= v(i-1)1
U²=-ui # Negate momentum
end for
xN
return
", + "type": "table", + "image_path": "d4fc6fb9b0b0032a23cc88aa11277a24167f400673729444c6a54a54eb4852fc.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 108, + 221, + 504, + 287.6666666666667 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 108, + 287.6666666666667, + 504, + 354.33333333333337 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 108, + 354.33333333333337, + 504, + 421.00000000000006 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 11.0 + }, + { + "type": "title", + "bbox": [ + 107, + 440, + 204, + 452 + ], + "lines": [ + { + "bbox": [ + 106, + 439, + 205, + 454 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 205, + 454 + ], + "score": 1.0, + "content": "C.2 MCMC TUNING", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 461, + 505, + 493 + ], + "lines": [ + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "A crucial component to ensure successful MCMC sampling in diffusion models is the choice of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 471, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 443, + 485 + ], + "score": 1.0, + "content": "step sizes for samplers. We initialize step sizes for all samplers at each distribution", + "type": "text" + }, + { + "bbox": [ + 443, + 473, + 448, + 482 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 471, + 505, + 485 + ], + "score": 1.0, + "content": "to be roughly", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 482, + 435, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 183, + 495 + ], + "score": 1.0, + "content": "proportional to the", + "type": "text" + }, + { + "bbox": [ + 183, + 483, + 194, + 494 + ], + "score": 0.88, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 482, + 330, + 495 + ], + "score": 1.0, + "content": "noise values added to distribution", + "type": "text" + }, + { + "bbox": [ + 331, + 484, + 336, + 492 + ], + "score": 0.64, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 482, + 435, + 495 + ], + "score": 1.0, + "content": "in the diffusion process.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 461, + 505, + 495 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 499, + 505, + 566 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "To tune step sizes across timesteps for both HMC and MALA samplers, to set step sizes at each", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 144, + 523 + ], + "score": 1.0, + "content": "timestep", + "type": "text" + }, + { + "bbox": [ + 144, + 512, + 149, + 520 + ], + "score": 0.64, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 511, + 271, + 523 + ], + "score": 1.0, + "content": "to be constant multiplied by", + "type": "text" + }, + { + "bbox": [ + 271, + 511, + 281, + 522 + ], + "score": 0.87, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 511, + 472, + 523 + ], + "score": 1.0, + "content": ". We searched different constants to multiply", + "type": "text" + }, + { + "bbox": [ + 473, + 511, + 483, + 522 + ], + "score": 0.87, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 511, + 505, + 523 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "chose a value so that the average acceptance rate of MALA and HMC samplers across timesteps is", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 532, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 167, + 546 + ], + "score": 1.0, + "content": "approximately", + "type": "text" + }, + { + "bbox": [ + 168, + 532, + 188, + 543 + ], + "score": 0.87, + "content": "60 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 532, + 206, + 546 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 206, + 533, + 226, + 543 + ], + "score": 0.88, + "content": "70 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 532, + 506, + 546 + ], + "score": 1.0, + "content": "respectively. For un-adjusted variants of these samplers, we set step", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "sizes to be the same as adjusted samplers, and found limited gains when step sizes were specifically", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 554, + 472, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 381, + 567 + ], + "score": 1.0, + "content": "tuned towards the un-adjusted samplers. We utilize a mass matrix of", + "type": "text" + }, + { + "bbox": [ + 381, + 555, + 391, + 566 + ], + "score": 0.88, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 554, + 472, + 567 + ], + "score": 1.0, + "content": "for HMC samplers.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 500, + 506, + 567 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 571, + 492, + 582 + ], + "lines": [ + { + "bbox": [ + 106, + 570, + 493, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 493, + 584 + ], + "score": 1.0, + "content": "Precise details on the exact MCMC steps sizes used in experiments can be detailed in Section G.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 106, + 570, + 493, + 584 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 595, + 286, + 606 + ], + "lines": [ + { + "bbox": [ + 107, + 595, + 288, + 608 + ], + "spans": [ + { + "bbox": [ + 107, + 595, + 288, + 608 + ], + "score": 1.0, + "content": "C.3 MCMC IMPLEMENTATION DETAILS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 615, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "score": 1.0, + "content": "When initially running MCMC sampling on diffusion models in the image domain, we found that", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 626, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 641 + ], + "score": 1.0, + "content": "our samplers tended to converge to images which had uniform textures. After experimentation, we", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 639, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 505, + 650 + ], + "score": 1.0, + "content": "found that the primary cause of this issue was fact that by default, typical implementations of the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 649, + 507, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 469, + 661 + ], + "score": 1.0, + "content": "reverse diffusion process clip samples at intermediate time-steps of sampling to be between", + "type": "text" + }, + { + "bbox": [ + 469, + 649, + 479, + 659 + ], + "score": 0.29, + "content": "^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 649, + 507, + 661 + ], + "score": 1.0, + "content": "and 1.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 659, + 507, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 507, + 673 + ], + "score": 1.0, + "content": "To enable proper MCMC sampling, we found that it was important to not clip intermediate values of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 669, + 186, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 186, + 685 + ], + "score": 1.0, + "content": "diffusion sampling.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 615, + 507, + 685 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "When running MCMC sampling on image domains, we further found that it was helpful for mixing", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "to run a single step of the reverse process to initialize MCMC sampling, before running many steps", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 709, + 507, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 259, + 723 + ], + "score": 1.0, + "content": "of MCMC sampling at each timestep", + "type": "text" + }, + { + "bbox": [ + 260, + 711, + 264, + 720 + ], + "score": 0.59, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 709, + 507, + 723 + ], + "score": 1.0, + "content": ", and in all MCMC sampling settings on the image domain,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "we run one step of the reverse process before running MCMC sampling. Such a MCMC sampling", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "procedure is similar to the predictor-corrector sampling procedure introduced in (Song et al., 2021)", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 448, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 448, + 106 + ], + "score": 1.0, + "content": "for alleviating discretization errors when sampling continuous time diffusion models.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 687, + 507, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "procedure is similar to the predictor-corrector sampling procedure introduced in (Song et al., 2021)", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 448, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 448, + 106 + ], + "score": 1.0, + "content": "for alleviating discretization errors when sampling continuous time diffusion models.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "title", + "bbox": [ + 107, + 122, + 282, + 135 + ], + "lines": [ + { + "bbox": [ + 105, + 121, + 283, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 283, + 138 + ], + "score": 1.0, + "content": "D COMPOSITIONAL DIFFUSIONS", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 150, + 505, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 149, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 400, + 164 + ], + "score": 1.0, + "content": "In Equations 11 and 12, we demonstrate that for diffused distributions", + "type": "text" + }, + { + "bbox": [ + 400, + 150, + 436, + 163 + ], + "score": 0.94, + "content": "\\{ q _ { t } ^ { i } ( x _ { t } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 149, + 465, + 164 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 466, + 150, + 505, + 163 + ], + "score": 0.91, + "content": "q _ { t } ^ { i } ( x _ { t } ) =", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 160, + 506, + 175 + ], + "spans": [ + { + "bbox": [ + 107, + 162, + 192, + 174 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\int q ^ { i } ( \\overline { { x } } _ { 0 } ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 160, + 328, + 175 + ], + "score": 1.0, + "content": ", the diffusion of the product of", + "type": "text" + }, + { + "bbox": [ + 329, + 162, + 338, + 174 + ], + "score": 0.87, + "content": "q ^ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 160, + 506, + 175 + ], + "score": 1.0, + "content": "’s is not the same as the product of the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 171, + 506, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 506, + 186 + ], + "score": 1.0, + "content": "diffusions, meaning plugging the product of diffusions into standard reverse diffusion sampling will", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 183, + 505, + 197 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 505, + 197 + ], + "score": 1.0, + "content": "not draw samples from the product model. We present similar results for tempering and predictive", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 195, + 187, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 187, + 207 + ], + "score": 1.0, + "content": "model composition.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 108, + 221, + 409, + 233 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 409, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 319, + 234 + ], + "score": 1.0, + "content": "D.1 SAMPLING FROM A TEMPERED VERSION OF", + "type": "text" + }, + { + "bbox": [ + 319, + 223, + 325, + 233 + ], + "score": 0.77, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 221, + 409, + 234 + ], + "score": 1.0, + "content": "USING DIFFUSION?", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 242, + 504, + 255 + ], + "lines": [ + { + "bbox": [ + 106, + 242, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 505, + 256 + ], + "score": 1.0, + "content": "It is tempting to believe that we can sample from a tempered/annealed version of the data distribution", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "interline_equation", + "bbox": [ + 274, + 260, + 336, + 276 + ], + "lines": [ + { + "bbox": [ + 274, + 260, + 336, + 276 + ], + "spans": [ + { + "bbox": [ + 274, + 260, + 336, + 276 + ], + "score": 0.91, + "content": "q ^ { \\lambda } ( x ) \\propto q ( x ) ^ { \\lambda }", + "type": "interline_equation", + "image_path": "bf49bf51ae55674a93ecd1574e44e55cb49ea06d43812dbce10bd87f7c7b5b92.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 274, + 260, + 336, + 276 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 282, + 504, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 282, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 288, + 295 + ], + "score": 1.0, + "content": "using the tempered diffused data distribution", + "type": "text" + }, + { + "bbox": [ + 289, + 283, + 345, + 295 + ], + "score": 0.9, + "content": "\\lambda \\nabla \\log _ { t } q ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 282, + 505, + 295 + ], + "score": 1.0, + "content": "but this is incorrect. For this procedure", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 294, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 255, + 306 + ], + "score": 1.0, + "content": "to be correct, we would need to have", + "type": "text" + }, + { + "bbox": [ + 255, + 294, + 376, + 306 + ], + "score": 0.72, + "content": "\\nabla \\log q _ { t } ^ { \\lambda } ( x _ { t } ) = \\lambda \\nabla \\log q _ { t } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 294, + 404, + 306 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 405, + 295, + 409, + 304 + ], + "score": 0.71, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 294, + 505, + 306 + ], + "score": 1.0, + "content": ". However, while we do", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 304, + 400, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 127, + 318 + ], + "score": 1.0, + "content": "have", + "type": "text" + }, + { + "bbox": [ + 128, + 304, + 252, + 317 + ], + "score": 0.93, + "content": "\\nabla \\log q _ { 0 } ^ { \\lambda } ( { x } _ { 0 } ) = \\lambda \\nabla \\log q _ { 0 } ( { x } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 304, + 377, + 318 + ], + "score": 1.0, + "content": ", this equality does not hold for", + "type": "text" + }, + { + "bbox": [ + 378, + 306, + 400, + 316 + ], + "score": 0.88, + "content": "t > 0", + "type": "inline_equation" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 324, + 403, + 392 + ], + "lines": [ + { + "bbox": [ + 207, + 324, + 403, + 392 + ], + "spans": [ + { + "bbox": [ + 207, + 324, + 403, + 392 + ], + "score": 0.94, + "content": "\\begin{array} { l c l } { \\nabla \\log q _ { t } ^ { \\lambda } ( x _ { t } ) } & { = } & { \\nabla \\log \\displaystyle \\int q ^ { \\lambda } ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } } \\\\ & { \\nmid } & { \\nmid } \\\\ & { \\nmid } & { \\times \\mathrm { ~ \\lambda ~ } \\nabla \\log \\displaystyle \\int q ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } } \\\\ & { = } & { \\lambda \\nabla \\log q _ { t } ( x _ { t } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "9070b961defca9693c7d77dce3b72a9ef6557d2fbf68131ed361bb36e4a05c16.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 207, + 324, + 403, + 341.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 207, + 341.0, + 403, + 358.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 207, + 358.0, + 403, + 375.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 207, + 375.0, + 403, + 392.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 403, + 181, + 415 + ], + "lines": [ + { + "bbox": [ + 105, + 402, + 182, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 182, + 417 + ], + "score": 1.0, + "content": "D.2 GUIDANCE", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 424, + 504, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "For conditional generation, we should use in the reverse diffusion the score of the diffused conditional", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 435, + 242, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 155, + 448 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 155, + 436, + 213, + 448 + ], + "score": 0.94, + "content": "\\nabla \\log q _ { t } ( x _ { t } | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 435, + 242, + 448 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 454, + 377, + 480 + ], + "lines": [ + { + "bbox": [ + 233, + 454, + 377, + 480 + ], + "spans": [ + { + "bbox": [ + 233, + 454, + 377, + 480 + ], + "score": 0.94, + "content": "q _ { t } ( x _ { t } | y ) = \\int q ( x _ { 0 } | y ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } .", + "type": "interline_equation", + "image_path": "259310af6179efb3c670bfd75342d46a5db69926d291dc5bf41887ab53f2911f.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 233, + 454, + 377, + 480 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 160, + 504 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 162, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 162, + 505 + ], + "score": 1.0, + "content": "We also have", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 209, + 505, + 402, + 519 + ], + "lines": [ + { + "bbox": [ + 209, + 505, + 402, + 519 + ], + "spans": [ + { + "bbox": [ + 209, + 505, + 402, + 519 + ], + "score": 0.89, + "content": "\\nabla \\log q _ { t } ( x _ { t } | y ) = \\nabla \\log q _ { t } ( x _ { t } ) + \\nabla \\log q _ { t } ( y | x _ { t } )", + "type": "interline_equation", + "image_path": "7948486016648fa1daeb9696cefab5dc1dddd161d8099b691f184103373ad6a0.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 209, + 505, + 402, + 519 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 524, + 135, + 536 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 136, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 136, + 536 + ], + "score": 1.0, + "content": "so that", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "interline_equation", + "bbox": [ + 208, + 537, + 403, + 551 + ], + "lines": [ + { + "bbox": [ + 208, + 537, + 403, + 551 + ], + "spans": [ + { + "bbox": [ + 208, + 537, + 403, + 551 + ], + "score": 0.88, + "content": "\\nabla \\log q _ { t } ( y | x _ { t } ) : = \\nabla \\log q _ { t } ( x _ { t } | y ) - \\nabla \\log q _ { t } ( x _ { t } )", + "type": "interline_equation", + "image_path": "3716962d2a91afbd9b781602aa2faf47a5824416053769654f4a2941d78f6f43.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 208, + 537, + 403, + 551 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 555, + 440, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 554, + 441, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 376, + 570 + ], + "score": 1.0, + "content": "allows you to do guidance without having to train say a classifier if", + "type": "text" + }, + { + "bbox": [ + 376, + 558, + 383, + 568 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 554, + 441, + 570 + ], + "score": 1.0, + "content": "is categorical.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 400, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 400, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 400, + 586 + ], + "score": 1.0, + "content": "In practice, it was found that using in the reverse time diffusion the score", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 591, + 369, + 605 + ], + "lines": [ + { + "bbox": [ + 241, + 591, + 369, + 605 + ], + "spans": [ + { + "bbox": [ + 241, + 591, + 369, + 605 + ], + "score": 0.92, + "content": "\\nabla \\log q _ { t } ( x _ { t } ) + \\lambda \\nabla \\log q _ { t } ( y | x _ { t } )", + "type": "interline_equation", + "image_path": "f03d9fd6a1f5d0fdcc7fc29e7b434f3349305f245741cc9f356ab8aeb5832301.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 241, + 591, + 369, + 605 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 612, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 104, + 611, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 104, + 611, + 243, + 626 + ], + "score": 1.0, + "content": "generates much nicer images for", + "type": "text" + }, + { + "bbox": [ + 243, + 613, + 269, + 623 + ], + "score": 0.9, + "content": "\\lambda > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 611, + 506, + 626 + ], + "score": 1.0, + "content": ". However, it is also often claimed that it samples from a", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "score": 1.0, + "content": "modified posterior where the likelihood has been annealed. This is incorrect. For a modified posterior", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 634, + 272, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 272, + 646 + ], + "score": 1.0, + "content": "with annealed likelihood, we would have", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "interline_equation", + "bbox": [ + 240, + 652, + 370, + 668 + ], + "lines": [ + { + "bbox": [ + 240, + 652, + 370, + 668 + ], + "spans": [ + { + "bbox": [ + 240, + 652, + 370, + 668 + ], + "score": 0.9, + "content": "q ^ { \\lambda } ( x _ { 0 } | y ) \\propto q ( x _ { 0 } | y ) \\left\\{ q ( y | x _ { 0 } ) \\right\\} ^ { \\lambda }", + "type": "interline_equation", + "image_path": "68b8942e646d867b9e7d8158a55580a4c07c23367a6cc8b1b5789a599f2173d2.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 240, + 652, + 370, + 668 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 675, + 215, + 686 + ], + "lines": [ + { + "bbox": [ + 106, + 675, + 215, + 687 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 215, + 687 + ], + "score": 1.0, + "content": "and it is not true again that", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "interline_equation", + "bbox": [ + 196, + 694, + 414, + 735 + ], + "lines": [ + { + "bbox": [ + 196, + 694, + 414, + 735 + ], + "spans": [ + { + "bbox": [ + 196, + 694, + 414, + 735 + ], + "score": 0.92, + "content": "\\begin{array} { r c l } { \\nabla \\log q _ { t } ^ { \\lambda } ( x _ { t } | y ) } & { = } & { \\nabla \\log \\displaystyle \\int q ^ { \\lambda } ( x _ { 0 } | y ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } } \\\\ & { \\neq } & { \\nabla \\log q _ { t } ( x _ { t } ) + \\lambda \\nabla \\log q _ { t } ( y | x _ { t } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "3937f9e229b0d48c348bcae9c906dd0018f06029758c93c684cc13083b083e74.jpg" + } + ] + } + ], + "index": 34.5, + "virtual_lines": [ + { + "bbox": [ + 196, + 694, + 414, + 714.5 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 196, + 714.5, + 414, + 735.0 + ], + "spans": [], + "index": 35 + } + ] + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 506, + 106 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 122, + 282, + 135 + ], + "lines": [ + { + "bbox": [ + 105, + 121, + 283, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 283, + 138 + ], + "score": 1.0, + "content": "D COMPOSITIONAL DIFFUSIONS", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 150, + 505, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 149, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 400, + 164 + ], + "score": 1.0, + "content": "In Equations 11 and 12, we demonstrate that for diffused distributions", + "type": "text" + }, + { + "bbox": [ + 400, + 150, + 436, + 163 + ], + "score": 0.94, + "content": "\\{ q _ { t } ^ { i } ( x _ { t } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 149, + 465, + 164 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 466, + 150, + 505, + 163 + ], + "score": 0.91, + "content": "q _ { t } ^ { i } ( x _ { t } ) =", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 160, + 506, + 175 + ], + "spans": [ + { + "bbox": [ + 107, + 162, + 192, + 174 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\int q ^ { i } ( \\overline { { x } } _ { 0 } ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 160, + 328, + 175 + ], + "score": 1.0, + "content": ", the diffusion of the product of", + "type": "text" + }, + { + "bbox": [ + 329, + 162, + 338, + 174 + ], + "score": 0.87, + "content": "q ^ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 160, + 506, + 175 + ], + "score": 1.0, + "content": "’s is not the same as the product of the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 171, + 506, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 506, + 186 + ], + "score": 1.0, + "content": "diffusions, meaning plugging the product of diffusions into standard reverse diffusion sampling will", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 183, + 505, + 197 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 505, + 197 + ], + "score": 1.0, + "content": "not draw samples from the product model. We present similar results for tempering and predictive", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 195, + 187, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 187, + 207 + ], + "score": 1.0, + "content": "model composition.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 149, + 506, + 207 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 221, + 409, + 233 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 409, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 319, + 234 + ], + "score": 1.0, + "content": "D.1 SAMPLING FROM A TEMPERED VERSION OF", + "type": "text" + }, + { + "bbox": [ + 319, + 223, + 325, + 233 + ], + "score": 0.77, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 221, + 409, + 234 + ], + "score": 1.0, + "content": "USING DIFFUSION?", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 221, + 409, + 234 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 242, + 504, + 255 + ], + "lines": [ + { + "bbox": [ + 106, + 242, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 505, + 256 + ], + "score": 1.0, + "content": "It is tempting to believe that we can sample from a tempered/annealed version of the data distribution", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 106, + 242, + 505, + 256 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 274, + 260, + 336, + 276 + ], + "lines": [ + { + "bbox": [ + 274, + 260, + 336, + 276 + ], + "spans": [ + { + "bbox": [ + 274, + 260, + 336, + 276 + ], + "score": 0.91, + "content": "q ^ { \\lambda } ( x ) \\propto q ( x ) ^ { \\lambda }", + "type": "interline_equation", + "image_path": "bf49bf51ae55674a93ecd1574e44e55cb49ea06d43812dbce10bd87f7c7b5b92.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 274, + 260, + 336, + 276 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 282, + 504, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 282, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 288, + 295 + ], + "score": 1.0, + "content": "using the tempered diffused data distribution", + "type": "text" + }, + { + "bbox": [ + 289, + 283, + 345, + 295 + ], + "score": 0.9, + "content": "\\lambda \\nabla \\log _ { t } q ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 282, + 505, + 295 + ], + "score": 1.0, + "content": "but this is incorrect. For this procedure", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 294, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 255, + 306 + ], + "score": 1.0, + "content": "to be correct, we would need to have", + "type": "text" + }, + { + "bbox": [ + 255, + 294, + 376, + 306 + ], + "score": 0.72, + "content": "\\nabla \\log q _ { t } ^ { \\lambda } ( x _ { t } ) = \\lambda \\nabla \\log q _ { t } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 294, + 404, + 306 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 405, + 295, + 409, + 304 + ], + "score": 0.71, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 294, + 505, + 306 + ], + "score": 1.0, + "content": ". However, while we do", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 304, + 400, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 127, + 318 + ], + "score": 1.0, + "content": "have", + "type": "text" + }, + { + "bbox": [ + 128, + 304, + 252, + 317 + ], + "score": 0.93, + "content": "\\nabla \\log q _ { 0 } ^ { \\lambda } ( { x } _ { 0 } ) = \\lambda \\nabla \\log q _ { 0 } ( { x } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 304, + 377, + 318 + ], + "score": 1.0, + "content": ", this equality does not hold for", + "type": "text" + }, + { + "bbox": [ + 378, + 306, + 400, + 316 + ], + "score": 0.88, + "content": "t > 0", + "type": "inline_equation" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 282, + 505, + 318 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 324, + 403, + 392 + ], + "lines": [ + { + "bbox": [ + 207, + 324, + 403, + 392 + ], + "spans": [ + { + "bbox": [ + 207, + 324, + 403, + 392 + ], + "score": 0.94, + "content": "\\begin{array} { l c l } { \\nabla \\log q _ { t } ^ { \\lambda } ( x _ { t } ) } & { = } & { \\nabla \\log \\displaystyle \\int q ^ { \\lambda } ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } } \\\\ & { \\nmid } & { \\nmid } \\\\ & { \\nmid } & { \\times \\mathrm { ~ \\lambda ~ } \\nabla \\log \\displaystyle \\int q ( x _ { 0 } ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } } \\\\ & { = } & { \\lambda \\nabla \\log q _ { t } ( x _ { t } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "9070b961defca9693c7d77dce3b72a9ef6557d2fbf68131ed361bb36e4a05c16.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 207, + 324, + 403, + 341.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 207, + 341.0, + 403, + 358.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 207, + 358.0, + 403, + 375.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 207, + 375.0, + 403, + 392.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "title", + "bbox": [ + 106, + 403, + 181, + 415 + ], + "lines": [ + { + "bbox": [ + 105, + 402, + 182, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 182, + 417 + ], + "score": 1.0, + "content": "D.2 GUIDANCE", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 424, + 504, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "For conditional generation, we should use in the reverse diffusion the score of the diffused conditional", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 435, + 242, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 155, + 448 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 155, + 436, + 213, + 448 + ], + "score": 0.94, + "content": "\\nabla \\log q _ { t } ( x _ { t } | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 435, + 242, + 448 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 424, + 505, + 448 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 454, + 377, + 480 + ], + "lines": [ + { + "bbox": [ + 233, + 454, + 377, + 480 + ], + "spans": [ + { + "bbox": [ + 233, + 454, + 377, + 480 + ], + "score": 0.94, + "content": "q _ { t } ( x _ { t } | y ) = \\int q ( x _ { 0 } | y ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } .", + "type": "interline_equation", + "image_path": "259310af6179efb3c670bfd75342d46a5db69926d291dc5bf41887ab53f2911f.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 233, + 454, + 377, + 480 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 160, + 504 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 162, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 162, + 505 + ], + "score": 1.0, + "content": "We also have", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 491, + 162, + 505 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 209, + 505, + 402, + 519 + ], + "lines": [ + { + "bbox": [ + 209, + 505, + 402, + 519 + ], + "spans": [ + { + "bbox": [ + 209, + 505, + 402, + 519 + ], + "score": 0.89, + "content": "\\nabla \\log q _ { t } ( x _ { t } | y ) = \\nabla \\log q _ { t } ( x _ { t } ) + \\nabla \\log q _ { t } ( y | x _ { t } )", + "type": "interline_equation", + "image_path": "7948486016648fa1daeb9696cefab5dc1dddd161d8099b691f184103373ad6a0.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 209, + 505, + 402, + 519 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 524, + 135, + 536 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 136, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 136, + 536 + ], + "score": 1.0, + "content": "so that", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 523, + 136, + 536 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 208, + 537, + 403, + 551 + ], + "lines": [ + { + "bbox": [ + 208, + 537, + 403, + 551 + ], + "spans": [ + { + "bbox": [ + 208, + 537, + 403, + 551 + ], + "score": 0.88, + "content": "\\nabla \\log q _ { t } ( y | x _ { t } ) : = \\nabla \\log q _ { t } ( x _ { t } | y ) - \\nabla \\log q _ { t } ( x _ { t } )", + "type": "interline_equation", + "image_path": "3716962d2a91afbd9b781602aa2faf47a5824416053769654f4a2941d78f6f43.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 208, + 537, + 403, + 551 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 555, + 440, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 554, + 441, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 376, + 570 + ], + "score": 1.0, + "content": "allows you to do guidance without having to train say a classifier if", + "type": "text" + }, + { + "bbox": [ + 376, + 558, + 383, + 568 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 554, + 441, + 570 + ], + "score": 1.0, + "content": "is categorical.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 554, + 441, + 570 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 400, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 400, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 400, + 586 + ], + "score": 1.0, + "content": "In practice, it was found that using in the reverse time diffusion the score", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 572, + 400, + 586 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 591, + 369, + 605 + ], + "lines": [ + { + "bbox": [ + 241, + 591, + 369, + 605 + ], + "spans": [ + { + "bbox": [ + 241, + 591, + 369, + 605 + ], + "score": 0.92, + "content": "\\nabla \\log q _ { t } ( x _ { t } ) + \\lambda \\nabla \\log q _ { t } ( y | x _ { t } )", + "type": "interline_equation", + "image_path": "f03d9fd6a1f5d0fdcc7fc29e7b434f3349305f245741cc9f356ab8aeb5832301.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 241, + 591, + 369, + 605 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 612, + 505, + 645 + ], + "lines": [ + { + "bbox": [ + 104, + 611, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 104, + 611, + 243, + 626 + ], + "score": 1.0, + "content": "generates much nicer images for", + "type": "text" + }, + { + "bbox": [ + 243, + 613, + 269, + 623 + ], + "score": 0.9, + "content": "\\lambda > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 611, + 506, + 626 + ], + "score": 1.0, + "content": ". However, it is also often claimed that it samples from a", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 506, + 635 + ], + "score": 1.0, + "content": "modified posterior where the likelihood has been annealed. This is incorrect. For a modified posterior", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 634, + 272, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 272, + 646 + ], + "score": 1.0, + "content": "with annealed likelihood, we would have", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 104, + 611, + 506, + 646 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 240, + 652, + 370, + 668 + ], + "lines": [ + { + "bbox": [ + 240, + 652, + 370, + 668 + ], + "spans": [ + { + "bbox": [ + 240, + 652, + 370, + 668 + ], + "score": 0.9, + "content": "q ^ { \\lambda } ( x _ { 0 } | y ) \\propto q ( x _ { 0 } | y ) \\left\\{ q ( y | x _ { 0 } ) \\right\\} ^ { \\lambda }", + "type": "interline_equation", + "image_path": "68b8942e646d867b9e7d8158a55580a4c07c23367a6cc8b1b5789a599f2173d2.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 240, + 652, + 370, + 668 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 675, + 215, + 686 + ], + "lines": [ + { + "bbox": [ + 106, + 675, + 215, + 687 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 215, + 687 + ], + "score": 1.0, + "content": "and it is not true again that", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 675, + 215, + 687 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 196, + 694, + 414, + 735 + ], + "lines": [ + { + "bbox": [ + 196, + 694, + 414, + 735 + ], + "spans": [ + { + "bbox": [ + 196, + 694, + 414, + 735 + ], + "score": 0.92, + "content": "\\begin{array} { r c l } { \\nabla \\log q _ { t } ^ { \\lambda } ( x _ { t } | y ) } & { = } & { \\nabla \\log \\displaystyle \\int q ^ { \\lambda } ( x _ { 0 } | y ) q ( x _ { t } | x _ { 0 } ) d x _ { 0 } } \\\\ & { \\neq } & { \\nabla \\log q _ { t } ( x _ { t } ) + \\lambda \\nabla \\log q _ { t } ( y | x _ { t } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "3937f9e229b0d48c348bcae9c906dd0018f06029758c93c684cc13083b083e74.jpg" + } + ] + } + ], + "index": 34.5, + "virtual_lines": [ + { + "bbox": [ + 196, + 694, + 414, + 714.5 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 196, + 714.5, + 414, + 735.0 + ], + "spans": [], + "index": 35 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 82, + 329, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 329, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 329, + 95 + ], + "score": 1.0, + "content": "D.3 SAMPLING FROM COMPOSED DISTRIBUTIONS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 506, + 180 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 506, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 506, + 116 + ], + "score": 1.0, + "content": "We can see that products, tempering, and guidance applied to diffused distributions do not give", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 114, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 114, + 505, + 126 + ], + "score": 1.0, + "content": "diffusions of the modified target distributions. Thus, we should not expect to arrive at our desired", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 125, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 506, + 138 + ], + "score": 1.0, + "content": "result by applying a sampling procedure which reverses a diffusion applied to the target distribution.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "Thankfully, as stated in Section 4, these operators do give us a sequence of distributions which anneals", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 146, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 129, + 159 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 129, + 147, + 162, + 159 + ], + "score": 0.92, + "content": "\\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 146, + 505, + 159 + ], + "score": 1.0, + "content": "to the composed target which means we can utilize the family of annealed MCMC", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 158, + 506, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 158, + 506, + 170 + ], + "score": 1.0, + "content": "sampling methods mentioned in Section 4.1 to draw samples from our composed models in all of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 169, + 336, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 336, + 180 + ], + "score": 1.0, + "content": "these settings, directly using the available score estimate.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4 + }, + { + "type": "title", + "bbox": [ + 107, + 195, + 325, + 208 + ], + "lines": [ + { + "bbox": [ + 105, + 194, + 326, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 326, + 210 + ], + "score": 1.0, + "content": "E ENERGY-BASED PARAMETERIZATIONS", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 506, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 221, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 282, + 234 + ], + "score": 1.0, + "content": "As mentioned in section 4.2, when using the", + "type": "text" + }, + { + "bbox": [ + 283, + 222, + 314, + 234 + ], + "score": 0.95, + "content": "\\epsilon _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 221, + 506, + 234 + ], + "score": 1.0, + "content": "parameterization, we can recover an estimate of", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 231, + 507, + 249 + ], + "spans": [ + { + "bbox": [ + 104, + 231, + 264, + 249 + ], + "score": 1.0, + "content": "the time-conditional score function with", + "type": "text" + }, + { + "bbox": [ + 265, + 233, + 364, + 249 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\nabla _ { \\boldsymbol { x } } \\log p _ { t } ( \\boldsymbol { x } ) \\approx - \\frac { \\epsilon _ { \\theta } ( \\boldsymbol { x } , t ) } { \\sigma _ { t } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 231, + 507, + 249 + ], + "score": 1.0, + "content": ". This estimate of the log-likelihood", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 246, + 507, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 507, + 261 + ], + "score": 1.0, + "content": "gradient can be used for MCMC sampling methods which only require the log-likelihood gradient –", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 258, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 506, + 271 + ], + "score": 1.0, + "content": "such as ULA or U-HMC. These methods can work well, but will never generate exact samples when", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 270, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 506, + 282 + ], + "score": 1.0, + "content": "using non-zero step-sizes. Exact samplers can be derived from approximate samplers like the above", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 281, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 505, + 292 + ], + "score": 1.0, + "content": "methods using Metropolis corrections. Unfortunately, even if the samplers’ transition distribution", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 291, + 444, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 134, + 303 + ], + "score": 0.92, + "content": "k _ { i } ( \\cdot , \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 291, + 201, + 304 + ], + "score": 1.0, + "content": "does not require", + "type": "text" + }, + { + "bbox": [ + 202, + 291, + 244, + 303 + ], + "score": 0.93, + "content": "\\log p _ { \\theta } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 291, + 444, + 304 + ], + "score": 1.0, + "content": "evaluation, the Metropolis correction probability:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 306, + 369, + 335 + ], + "lines": [ + { + "bbox": [ + 241, + 306, + 369, + 335 + ], + "spans": [ + { + "bbox": [ + 241, + 306, + 369, + 335 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } \\left( 1 , \\frac { e ^ { f _ { \\theta } ( \\hat { x } ) } } { e ^ { f _ { \\theta } ( x _ { t - 1 } ) } } \\frac { k ( x _ { t - 1 } | \\hat { x } ) } { k ( \\hat { x } | x _ { t - 1 } ) } \\right)", + "type": "interline_equation", + "image_path": "f3cd671a3bccb762f2aa859c270902622f22f26617edbbc1bbe918a382019d34.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 241, + 306, + 369, + 320.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 241, + 320.5, + 369, + 335.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 504, + 361 + ], + "lines": [ + { + "bbox": [ + 106, + 337, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 351 + ], + "score": 1.0, + "content": "does. Futhermore, when we only have an estimate of the score at our disposal, we are only able to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 349, + 239, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 239, + 361 + ], + "score": 1.0, + "content": "compose models using products.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 365, + 506, + 452 + ], + "lines": [ + { + "bbox": [ + 106, + 365, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 505, + 378 + ], + "score": 1.0, + "content": "To enable the use of Metropolis corrections and more compositional operators, we propose to change", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 375, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 104, + 375, + 394, + 390 + ], + "score": 1.0, + "content": "the parameterization of our diffusion model. Instead of using a neural net", + "type": "text" + }, + { + "bbox": [ + 395, + 376, + 502, + 389 + ], + "score": 0.91, + "content": "\\epsilon _ { \\theta } ( x , t ) : \\bar { \\{ \\mathbb { R } ^ { d } \\times \\mathbb { N } \\} } \\to \\bar { \\mathbb { R } ^ { d } }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 375, + 506, + 390 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 386, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 267, + 401 + ], + "score": 1.0, + "content": "we define a scalar-output neural network", + "type": "text" + }, + { + "bbox": [ + 267, + 387, + 371, + 400 + ], + "score": 0.93, + "content": "f _ { \\theta } ( x , t ) : \\{ \\mathbb { R } ^ { d } \\times \\bar { \\mathbb { N } } \\} \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 386, + 506, + 401 + ], + "score": 1.0, + "content": ". We then compute the gradient of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 397, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 201, + 412 + ], + "score": 1.0, + "content": "this function and define", + "type": "text" + }, + { + "bbox": [ + 202, + 398, + 290, + 411 + ], + "score": 0.93, + "content": "\\epsilon _ { \\theta } ( x , t ) = \\nabla _ { x } f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 397, + 460, + 412 + ], + "score": 1.0, + "content": ". From here, we use this implicitly-defined", + "type": "text" + }, + { + "bbox": [ + 460, + 399, + 483, + 411 + ], + "score": 0.92, + "content": "\\epsilon _ { \\theta } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 397, + 506, + 412 + ], + "score": 1.0, + "content": "as in", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 103, + 408, + 507, + 443 + ], + "spans": [ + { + "bbox": [ + 103, + 408, + 203, + 443 + ], + "score": 1.0, + "content": "standard diffusion modeare also able to recover", + "type": "text" + }, + { + "bbox": [ + 323, + 408, + 353, + 443 + ], + "score": 1.0, + "content": "recover which", + "type": "text" + }, + { + "bbox": [ + 353, + 410, + 452, + 426 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\nabla _ { \\boldsymbol { x } } \\log p _ { t } ( \\boldsymbol { x } ) \\approx - \\frac { \\epsilon _ { \\theta } ( \\boldsymbol { x } , t ) } { \\sigma _ { t } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 408, + 507, + 443 + ], + "score": 1.0, + "content": ", but now wef Metropolis", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 203, + 425, + 322, + 442 + ], + "spans": [ + { + "bbox": [ + 203, + 425, + 322, + 442 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\log p _ { t } ( x ) \\approx - \\frac { f _ { \\theta } ( x , t ) } { \\sigma _ { t } } + \\log Z } \\end{array}", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 438, + 187, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 187, + 453 + ], + "score": 1.0, + "content": "corrected sampling.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 505, + 524 + ], + "lines": [ + { + "bbox": [ + 106, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 284, + 469 + ], + "score": 1.0, + "content": "Much prior work on EBMs parameterizes", + "type": "text" + }, + { + "bbox": [ + 284, + 457, + 316, + 469 + ], + "score": 0.92, + "content": "f _ { \\boldsymbol { \\theta } } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 456, + 506, + 469 + ], + "score": 1.0, + "content": "using a feed-forward neural network, whose", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "final layer has a single output (Nijkamp et al., 2020; Du & Mordatch, 2019). Salimans & Ho", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 312, + 492 + ], + "score": 1.0, + "content": "(2021) compare this approach with the standard", + "type": "text" + }, + { + "bbox": [ + 313, + 479, + 344, + 491 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 478, + 473, + 492 + ], + "score": 1.0, + "content": "parameterization and find the", + "type": "text" + }, + { + "bbox": [ + 474, + 479, + 505, + 491 + ], + "score": 0.92, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 490, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 506, + 502 + ], + "score": 1.0, + "content": "parameterization to perform better for unconditional image generation. We believe this has to do", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "with the relative sparsity of the gradients of feed-forward neural networks. This can cause difficulties", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 511, + 410, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 410, + 524 + ], + "score": 1.0, + "content": "when training to optimize a function of their implicitly computed gradients.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 528, + 505, + 551 + ], + "lines": [ + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "score": 1.0, + "content": "Intriguingly, Salimans & Ho (2021) also explore a more structured energy function definition inspired", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 540, + 218, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 218, + 552 + ], + "score": 1.0, + "content": "by denoising autoencoders:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 554, + 376, + 578 + ], + "lines": [ + { + "bbox": [ + 235, + 554, + 376, + 578 + ], + "spans": [ + { + "bbox": [ + 235, + 554, + 376, + 578 + ], + "score": 0.94, + "content": "f _ { \\theta } ^ { D A E } ( x , t ) = - \\frac { 1 } { 2 } \\vert \\vert x - s _ { \\theta } ( x , t ) \\vert \\vert ^ { 2 }", + "type": "interline_equation", + "image_path": "aabac0896f3d77c3a820abb7085bb638992b5a9981788c8616bef0022630ffca.jpg" + } + ] + } + ], + "index": 35, + "virtual_lines": [ + { + "bbox": [ + 235, + 554, + 376, + 578 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 582, + 505, + 605 + ], + "lines": [ + { + "bbox": [ + 106, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 133, + 595 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 582, + 241, + 595 + ], + "score": 0.92, + "content": "s _ { \\theta } ( x , t ) : \\{ \\mathbb { R } ^ { d } \\times \\mathbb { N } \\} \\to \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 581, + 423, + 595 + ], + "score": 1.0, + "content": "is a neural network (identical to the standard", + "type": "text" + }, + { + "bbox": [ + 423, + 583, + 456, + 595 + ], + "score": 0.91, + "content": "\\epsilon _ { \\theta } ( x , t ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 581, + 505, + 595 + ], + "score": 1.0, + "content": ") model. We", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 594, + 343, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 343, + 605 + ], + "score": 1.0, + "content": "can simply evaluate the gradients of this function to obtain", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "interline_equation", + "bbox": [ + 182, + 608, + 429, + 623 + ], + "lines": [ + { + "bbox": [ + 182, + 608, + 429, + 623 + ], + "spans": [ + { + "bbox": [ + 182, + 608, + 429, + 623 + ], + "score": 0.9, + "content": "\\nabla _ { x } f _ { \\theta } ^ { D A E } ( x , t ) = ( x - s _ { \\theta } ( x , t ) ) - ( x - s _ { \\theta } ( x , t ) ) \\nabla _ { x } s _ { \\theta } ( x , t ) .", + "type": "interline_equation", + "image_path": "b5da07929912cf483c5404411586f425f7480a27b169b744de7b4fb288f147d8.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 182, + 608, + 429, + 623 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 442, + 639 + ], + "score": 1.0, + "content": "In their study, this parameterization was found to perform near identically to the", + "type": "text" + }, + { + "bbox": [ + 442, + 627, + 473, + 639 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "param-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "score": 1.0, + "content": "eterization while admitting an explicit energy function. We believe this energy parameterization", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 386, + 661 + ], + "score": 1.0, + "content": "performs better because its gradients include the feed-forward network", + "type": "text" + }, + { + "bbox": [ + 386, + 649, + 417, + 661 + ], + "score": 0.93, + "content": "s _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 648, + 506, + 661 + ], + "score": 1.0, + "content": ", making optimization", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 658, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 303, + 673 + ], + "score": 1.0, + "content": "easier. Salimans & Ho (2021) conclude that the", + "type": "text" + }, + { + "bbox": [ + 303, + 660, + 334, + 672 + ], + "score": 0.92, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 658, + 506, + 673 + ], + "score": 1.0, + "content": "parameterization should be favored since", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 101, + 665, + 509, + 686 + ], + "spans": [ + { + "bbox": [ + 101, + 665, + 166, + 686 + ], + "score": 1.0, + "content": "the computing", + "type": "text" + }, + { + "bbox": [ + 167, + 671, + 227, + 683 + ], + "score": 0.92, + "content": "\\nabla _ { x } f _ { \\theta } ^ { D A E } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 665, + 308, + 686 + ], + "score": 1.0, + "content": "requires computing", + "type": "text" + }, + { + "bbox": [ + 308, + 672, + 348, + 682 + ], + "score": 0.92, + "content": "\\nabla s _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 665, + 509, + 686 + ], + "score": 1.0, + "content": "which requires an extra backward pass", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 682, + 303, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 303, + 695 + ], + "score": 1.0, + "content": "through the neural network, increasing compute.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "We reexamine this energy parameterization and two other choices now that our application motivates", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "having access to an explicit energy function. The other parameterizations are based different transfor-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 167, + 733 + ], + "score": 1.0, + "content": "mations of the", + "type": "text" + }, + { + "bbox": [ + 167, + 721, + 198, + 732 + ], + "score": 0.92, + "content": "s _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "architecture; the negative L2 norm (L2) and an inner product (IP). They are", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46 + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 82, + 329, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 329, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 329, + 95 + ], + "score": 1.0, + "content": "D.3 SAMPLING FROM COMPOSED DISTRIBUTIONS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 102, + 506, + 180 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 506, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 506, + 116 + ], + "score": 1.0, + "content": "We can see that products, tempering, and guidance applied to diffused distributions do not give", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 114, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 114, + 505, + 126 + ], + "score": 1.0, + "content": "diffusions of the modified target distributions. Thus, we should not expect to arrive at our desired", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 125, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 506, + 138 + ], + "score": 1.0, + "content": "result by applying a sampling procedure which reverses a diffusion applied to the target distribution.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "Thankfully, as stated in Section 4, these operators do give us a sequence of distributions which anneals", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 146, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 129, + 159 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 129, + 147, + 162, + 159 + ], + "score": 0.92, + "content": "\\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 146, + 505, + 159 + ], + "score": 1.0, + "content": "to the composed target which means we can utilize the family of annealed MCMC", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 158, + 506, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 158, + 506, + 170 + ], + "score": 1.0, + "content": "sampling methods mentioned in Section 4.1 to draw samples from our composed models in all of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 169, + 336, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 336, + 180 + ], + "score": 1.0, + "content": "these settings, directly using the available score estimate.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 102, + 506, + 180 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 195, + 325, + 208 + ], + "lines": [ + { + "bbox": [ + 105, + 194, + 326, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 326, + 210 + ], + "score": 1.0, + "content": "E ENERGY-BASED PARAMETERIZATIONS", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 506, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 221, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 282, + 234 + ], + "score": 1.0, + "content": "As mentioned in section 4.2, when using the", + "type": "text" + }, + { + "bbox": [ + 283, + 222, + 314, + 234 + ], + "score": 0.95, + "content": "\\epsilon _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 221, + 506, + 234 + ], + "score": 1.0, + "content": "parameterization, we can recover an estimate of", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 231, + 507, + 249 + ], + "spans": [ + { + "bbox": [ + 104, + 231, + 264, + 249 + ], + "score": 1.0, + "content": "the time-conditional score function with", + "type": "text" + }, + { + "bbox": [ + 265, + 233, + 364, + 249 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\nabla _ { \\boldsymbol { x } } \\log p _ { t } ( \\boldsymbol { x } ) \\approx - \\frac { \\epsilon _ { \\theta } ( \\boldsymbol { x } , t ) } { \\sigma _ { t } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 231, + 507, + 249 + ], + "score": 1.0, + "content": ". This estimate of the log-likelihood", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 246, + 507, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 507, + 261 + ], + "score": 1.0, + "content": "gradient can be used for MCMC sampling methods which only require the log-likelihood gradient –", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 258, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 506, + 271 + ], + "score": 1.0, + "content": "such as ULA or U-HMC. These methods can work well, but will never generate exact samples when", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 270, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 506, + 282 + ], + "score": 1.0, + "content": "using non-zero step-sizes. Exact samplers can be derived from approximate samplers like the above", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 281, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 505, + 292 + ], + "score": 1.0, + "content": "methods using Metropolis corrections. Unfortunately, even if the samplers’ transition distribution", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 291, + 444, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 134, + 303 + ], + "score": 0.92, + "content": "k _ { i } ( \\cdot , \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 291, + 201, + 304 + ], + "score": 1.0, + "content": "does not require", + "type": "text" + }, + { + "bbox": [ + 202, + 291, + 244, + 303 + ], + "score": 0.93, + "content": "\\log p _ { \\theta } ( x _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 291, + 444, + 304 + ], + "score": 1.0, + "content": "evaluation, the Metropolis correction probability:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12, + "bbox_fs": [ + 104, + 221, + 507, + 304 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 306, + 369, + 335 + ], + "lines": [ + { + "bbox": [ + 241, + 306, + 369, + 335 + ], + "spans": [ + { + "bbox": [ + 241, + 306, + 369, + 335 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } \\left( 1 , \\frac { e ^ { f _ { \\theta } ( \\hat { x } ) } } { e ^ { f _ { \\theta } ( x _ { t - 1 } ) } } \\frac { k ( x _ { t - 1 } | \\hat { x } ) } { k ( \\hat { x } | x _ { t - 1 } ) } \\right)", + "type": "interline_equation", + "image_path": "f3cd671a3bccb762f2aa859c270902622f22f26617edbbc1bbe918a382019d34.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 241, + 306, + 369, + 320.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 241, + 320.5, + 369, + 335.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 504, + 361 + ], + "lines": [ + { + "bbox": [ + 106, + 337, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 351 + ], + "score": 1.0, + "content": "does. Futhermore, when we only have an estimate of the score at our disposal, we are only able to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 349, + 239, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 239, + 361 + ], + "score": 1.0, + "content": "compose models using products.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 106, + 337, + 505, + 361 + ] + }, + { + "type": "list", + "bbox": [ + 106, + 365, + 506, + 452 + ], + "lines": [ + { + "bbox": [ + 106, + 365, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 505, + 378 + ], + "score": 1.0, + "content": "To enable the use of Metropolis corrections and more compositional operators, we propose to change", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 375, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 104, + 375, + 394, + 390 + ], + "score": 1.0, + "content": "the parameterization of our diffusion model. Instead of using a neural net", + "type": "text" + }, + { + "bbox": [ + 395, + 376, + 502, + 389 + ], + "score": 0.91, + "content": "\\epsilon _ { \\theta } ( x , t ) : \\bar { \\{ \\mathbb { R } ^ { d } \\times \\mathbb { N } \\} } \\to \\bar { \\mathbb { R } ^ { d } }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 375, + 506, + 390 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 386, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 267, + 401 + ], + "score": 1.0, + "content": "we define a scalar-output neural network", + "type": "text" + }, + { + "bbox": [ + 267, + 387, + 371, + 400 + ], + "score": 0.93, + "content": "f _ { \\theta } ( x , t ) : \\{ \\mathbb { R } ^ { d } \\times \\bar { \\mathbb { N } } \\} \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 386, + 506, + 401 + ], + "score": 1.0, + "content": ". We then compute the gradient of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 397, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 201, + 412 + ], + "score": 1.0, + "content": "this function and define", + "type": "text" + }, + { + "bbox": [ + 202, + 398, + 290, + 411 + ], + "score": 0.93, + "content": "\\epsilon _ { \\theta } ( x , t ) = \\nabla _ { x } f _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 397, + 460, + 412 + ], + "score": 1.0, + "content": ". From here, we use this implicitly-defined", + "type": "text" + }, + { + "bbox": [ + 460, + 399, + 483, + 411 + ], + "score": 0.92, + "content": "\\epsilon _ { \\theta } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 397, + 506, + 412 + ], + "score": 1.0, + "content": "as in", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 103, + 408, + 507, + 443 + ], + "spans": [ + { + "bbox": [ + 103, + 408, + 203, + 443 + ], + "score": 1.0, + "content": "standard diffusion modeare also able to recover", + "type": "text" + }, + { + "bbox": [ + 323, + 408, + 353, + 443 + ], + "score": 1.0, + "content": "recover which", + "type": "text" + }, + { + "bbox": [ + 353, + 410, + 452, + 426 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\nabla _ { \\boldsymbol { x } } \\log p _ { t } ( \\boldsymbol { x } ) \\approx - \\frac { \\epsilon _ { \\theta } ( \\boldsymbol { x } , t ) } { \\sigma _ { t } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 408, + 507, + 443 + ], + "score": 1.0, + "content": ", but now wef Metropolis", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 203, + 425, + 322, + 442 + ], + "spans": [ + { + "bbox": [ + 203, + 425, + 322, + 442 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\log p _ { t } ( x ) \\approx - \\frac { f _ { \\theta } ( x , t ) } { \\sigma _ { t } } + \\log Z } \\end{array}", + "type": "inline_equation" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 438, + 187, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 187, + 453 + ], + "score": 1.0, + "content": "corrected sampling.", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 23, + "bbox_fs": [ + 103, + 365, + 507, + 453 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 505, + 524 + ], + "lines": [ + { + "bbox": [ + 106, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 284, + 469 + ], + "score": 1.0, + "content": "Much prior work on EBMs parameterizes", + "type": "text" + }, + { + "bbox": [ + 284, + 457, + 316, + 469 + ], + "score": 0.92, + "content": "f _ { \\boldsymbol { \\theta } } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 456, + 506, + 469 + ], + "score": 1.0, + "content": "using a feed-forward neural network, whose", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "final layer has a single output (Nijkamp et al., 2020; Du & Mordatch, 2019). Salimans & Ho", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 312, + 492 + ], + "score": 1.0, + "content": "(2021) compare this approach with the standard", + "type": "text" + }, + { + "bbox": [ + 313, + 479, + 344, + 491 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 478, + 473, + 492 + ], + "score": 1.0, + "content": "parameterization and find the", + "type": "text" + }, + { + "bbox": [ + 474, + 479, + 505, + 491 + ], + "score": 0.92, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 490, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 506, + 502 + ], + "score": 1.0, + "content": "parameterization to perform better for unconditional image generation. We believe this has to do", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "with the relative sparsity of the gradients of feed-forward neural networks. This can cause difficulties", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 511, + 410, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 410, + 524 + ], + "score": 1.0, + "content": "when training to optimize a function of their implicitly computed gradients.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 456, + 506, + 524 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 528, + 505, + 551 + ], + "lines": [ + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "score": 1.0, + "content": "Intriguingly, Salimans & Ho (2021) also explore a more structured energy function definition inspired", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 540, + 218, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 218, + 552 + ], + "score": 1.0, + "content": "by denoising autoencoders:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 528, + 505, + 552 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 235, + 554, + 376, + 578 + ], + "lines": [ + { + "bbox": [ + 235, + 554, + 376, + 578 + ], + "spans": [ + { + "bbox": [ + 235, + 554, + 376, + 578 + ], + "score": 0.94, + "content": "f _ { \\theta } ^ { D A E } ( x , t ) = - \\frac { 1 } { 2 } \\vert \\vert x - s _ { \\theta } ( x , t ) \\vert \\vert ^ { 2 }", + "type": "interline_equation", + "image_path": "aabac0896f3d77c3a820abb7085bb638992b5a9981788c8616bef0022630ffca.jpg" + } + ] + } + ], + "index": 35, + "virtual_lines": [ + { + "bbox": [ + 235, + 554, + 376, + 578 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 582, + 505, + 605 + ], + "lines": [ + { + "bbox": [ + 106, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 133, + 595 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 582, + 241, + 595 + ], + "score": 0.92, + "content": "s _ { \\theta } ( x , t ) : \\{ \\mathbb { R } ^ { d } \\times \\mathbb { N } \\} \\to \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 581, + 423, + 595 + ], + "score": 1.0, + "content": "is a neural network (identical to the standard", + "type": "text" + }, + { + "bbox": [ + 423, + 583, + 456, + 595 + ], + "score": 0.91, + "content": "\\epsilon _ { \\theta } ( x , t ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 581, + 505, + 595 + ], + "score": 1.0, + "content": ") model. We", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 594, + 343, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 343, + 605 + ], + "score": 1.0, + "content": "can simply evaluate the gradients of this function to obtain", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 581, + 505, + 605 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 182, + 608, + 429, + 623 + ], + "lines": [ + { + "bbox": [ + 182, + 608, + 429, + 623 + ], + "spans": [ + { + "bbox": [ + 182, + 608, + 429, + 623 + ], + "score": 0.9, + "content": "\\nabla _ { x } f _ { \\theta } ^ { D A E } ( x , t ) = ( x - s _ { \\theta } ( x , t ) ) - ( x - s _ { \\theta } ( x , t ) ) \\nabla _ { x } s _ { \\theta } ( x , t ) .", + "type": "interline_equation", + "image_path": "b5da07929912cf483c5404411586f425f7480a27b169b744de7b4fb288f147d8.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 182, + 608, + 429, + 623 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 626, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 442, + 639 + ], + "score": 1.0, + "content": "In their study, this parameterization was found to perform near identically to the", + "type": "text" + }, + { + "bbox": [ + 442, + 627, + 473, + 639 + ], + "score": 0.93, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "param-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "score": 1.0, + "content": "eterization while admitting an explicit energy function. We believe this energy parameterization", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 386, + 661 + ], + "score": 1.0, + "content": "performs better because its gradients include the feed-forward network", + "type": "text" + }, + { + "bbox": [ + 386, + 649, + 417, + 661 + ], + "score": 0.93, + "content": "s _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 648, + 506, + 661 + ], + "score": 1.0, + "content": ", making optimization", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 658, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 303, + 673 + ], + "score": 1.0, + "content": "easier. Salimans & Ho (2021) conclude that the", + "type": "text" + }, + { + "bbox": [ + 303, + 660, + 334, + 672 + ], + "score": 0.92, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 658, + 506, + 673 + ], + "score": 1.0, + "content": "parameterization should be favored since", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 101, + 665, + 509, + 686 + ], + "spans": [ + { + "bbox": [ + 101, + 665, + 166, + 686 + ], + "score": 1.0, + "content": "the computing", + "type": "text" + }, + { + "bbox": [ + 167, + 671, + 227, + 683 + ], + "score": 0.92, + "content": "\\nabla _ { x } f _ { \\theta } ^ { D A E } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 665, + 308, + 686 + ], + "score": 1.0, + "content": "requires computing", + "type": "text" + }, + { + "bbox": [ + 308, + 672, + 348, + 682 + ], + "score": 0.92, + "content": "\\nabla s _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 665, + 509, + 686 + ], + "score": 1.0, + "content": "which requires an extra backward pass", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 682, + 303, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 303, + 695 + ], + "score": 1.0, + "content": "through the neural network, increasing compute.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41.5, + "bbox_fs": [ + 101, + 626, + 509, + 695 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "We reexamine this energy parameterization and two other choices now that our application motivates", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "having access to an explicit energy function. The other parameterizations are based different transfor-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 167, + 733 + ], + "score": 1.0, + "content": "mations of the", + "type": "text" + }, + { + "bbox": [ + 167, + 721, + 198, + 732 + ], + "score": 0.92, + "content": "s _ { \\theta } ( x , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "architecture; the negative L2 norm (L2) and an inner product (IP). They are", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46, + "bbox_fs": [ + 105, + 698, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 151, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 153, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 153, + 95 + ], + "score": 1.0, + "content": "defined as:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 95, + 387, + 135 + ], + "lines": [ + { + "bbox": [ + 224, + 95, + 387, + 135 + ], + "spans": [ + { + "bbox": [ + 224, + 95, + 387, + 135 + ], + "score": 0.9, + "content": "\\begin{array} { r c l } { { f _ { \\theta } ^ { L 2 } ( x , t ) } } & { { = } } & { { \\displaystyle - \\frac { 1 } { 2 } | | s _ { \\theta } ( x , t ) | | ^ { 2 } } } \\\\ { { \\nabla _ { x } f ^ { L 2 } ( x , t ) } } & { { = } } & { { \\displaystyle - s _ { \\theta } ( x , t ) \\nabla _ { x } s _ { \\theta } ( x , t ) } } \\end{array}", + "type": "interline_equation", + "image_path": "308317851c04995c84d877779c6703a08df81fdfa71b93b066bc01481b437230.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 224, + 95, + 387, + 108.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 224, + 108.33333333333333, + 387, + 121.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 224, + 121.66666666666666, + 387, + 135.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 138, + 126, + 149 + ], + "lines": [ + { + "bbox": [ + 105, + 137, + 126, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 126, + 150 + ], + "score": 1.0, + "content": "and:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 150, + 397, + 183 + ], + "lines": [ + { + "bbox": [ + 214, + 150, + 397, + 183 + ], + "spans": [ + { + "bbox": [ + 214, + 150, + 397, + 183 + ], + "score": 0.91, + "content": "\\begin{array} { r c l } { { f _ { \\theta } ^ { I P } ( x , t ) } } & { { = } } & { { { x ^ { T } s _ { \\theta } ( x , t ) } } } \\\\ { { \\nabla _ { x } f ^ { I P } ( x , t ) } } & { { = } } & { { s _ { \\theta } ( x , t ) + x ^ { T } \\nabla _ { x } s _ { \\theta } ( x , t ) . } } \\end{array}", + "type": "interline_equation", + "image_path": "1c8bf03a53f8d9171b9535bdab83652cfd4067bb877a90c51dc2e5b6758c2ad3.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 214, + 150, + 397, + 166.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 214, + 166.5, + 397, + 183.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 190, + 506, + 257 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 203 + ], + "score": 1.0, + "content": "We train models with each parameterization on ImageNet and compare using FID for unconditional", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 202, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 505, + 214 + ], + "score": 1.0, + "content": "sampling. Results can be seen in Table A1. We see that L2 and Inner-Product perform the best, but", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 213, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 505, + 225 + ], + "score": 1.0, + "content": "are both outperformed by the standard parameterization. We initially experimented with these two", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 224, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 104, + 224, + 505, + 237 + ], + "score": 1.0, + "content": "parameterizations but found that the L2-norm parameterization to be more stable for compositional", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 235, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 247 + ], + "score": 1.0, + "content": "sampling due, we believe, to the fact that the energy-function is bounded above meaning that MCMC", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 246, + 385, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 385, + 258 + ], + "score": 1.0, + "content": "sampling is incapable of running off to infinity to increase likelihood.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9.5 + }, + { + "type": "table", + "bbox": [ + 208, + 267, + 402, + 310 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 208, + 267, + 402, + 310 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 208, + 267, + 402, + 310 + ], + "spans": [ + { + "bbox": [ + 208, + 267, + 402, + 310 + ], + "score": 0.966, + "html": "
Parameterization ∈θ(x,t)
(1) DAE 97.4(2)L2 Norm 91.5(3) Inner-Product | 90.9 86.7
", + "type": "table", + "image_path": "97d2d88b17d20d0c6716ab8c71cb269c16a92d1c369ae6787ecba5a32dda786c.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 208, + 267, + 402, + 281.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 208, + 281.3333333333333, + 402, + 295.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 208, + 295.66666666666663, + 402, + 309.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 113, + 317, + 493, + 329 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 316, + 494, + 330 + ], + "spans": [ + { + "bbox": [ + 116, + 316, + 494, + 330 + ], + "score": 1.0, + "content": "Table A1: FID (1k samples) of various energy-parameterizations on unconditional ImageNet generation.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + } + ], + "index": 15.0 + }, + { + "type": "title", + "bbox": [ + 106, + 349, + 348, + 363 + ], + "lines": [ + { + "bbox": [ + 104, + 348, + 349, + 365 + ], + "spans": [ + { + "bbox": [ + 104, + 348, + 349, + 365 + ], + "score": 1.0, + "content": "F SYNTHETIC DISTRIBUTION COMPOSITIONS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 105, + 375, + 504, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 374, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 145, + 387 + ], + "score": 1.0, + "content": "Mixture", + "type": "text" + }, + { + "bbox": [ + 152, + 375, + 506, + 388 + ], + "score": 1.0, + "content": "We provide additional 2D illustrations of mixtures of two diffusion models in Figure A5.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 384, + 491, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 491, + 399 + ], + "score": 1.0, + "content": "We find that HMC sampling enables more accurate mixtures of different synthetic distributions.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "image", + "bbox": [ + 167, + 400, + 445, + 490 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 167, + 400, + 445, + 490 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 167, + 400, + 445, + 490 + ], + "spans": [ + { + "bbox": [ + 167, + 400, + 445, + 490 + ], + "score": 0.973, + "type": "image", + "image_path": "338e6c6a7dd7c357631290644720c829e3d176fbbd00a5a22297c26173aa4e08.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 167, + 400, + 445, + 430.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 167, + 430.0, + 445, + 460.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 167, + 460.0, + 445, + 490.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 498, + 504, + 520 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 498, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 506, + 510 + ], + "score": 1.0, + "content": "Figure A5: Examples of mixture applied to diffusion models. Left to right: Component distributions, reverse", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 508, + 503, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 503, + 520 + ], + "score": 1.0, + "content": "diffusion, HMC sampling. Reverse diffusion fails to sample accurately from mixed distributions distributions.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + } + ], + "index": 22.25 + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 530, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 149, + 543 + ], + "score": 1.0, + "content": "Negation", + "type": "text" + }, + { + "bbox": [ + 154, + 530, + 506, + 543 + ], + "score": 1.0, + "content": "We provide additional 2D illustrations of negating two diffusion model with respect to", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "each other in Figure A7. We find that HMC sampling enables accurate negations of different sythetic", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 551, + 161, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 161, + 565 + ], + "score": 1.0, + "content": "distributions.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "image", + "bbox": [ + 167, + 565, + 445, + 655 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 167, + 565, + 445, + 655 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 167, + 565, + 445, + 655 + ], + "spans": [ + { + "bbox": [ + 167, + 565, + 445, + 655 + ], + "score": 0.969, + "type": "image", + "image_path": "55963e4abfacce6fab91050a8f0316825523a8bdca454b6206986a0879a3b063.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 167, + 565, + 445, + 595.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 167, + 595.0, + 445, + 625.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 167, + 625.0, + 445, + 655.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 664, + 504, + 686 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 664, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 506, + 675 + ], + "score": 1.0, + "content": "Figure A6: Examples of negation applied to diffusion models. Left to right: Component distributions, reverse", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 674, + 461, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 674, + 461, + 685 + ], + "score": 1.0, + "content": "diffusion, HMC sampling. Reverse diffusion fails to sample accurately from negated distributions.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + } + ], + "index": 30.25 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "Failure Cases Next we illustrate a failure case of composition using our approach in Figure A7.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "Our approach fails to generate the product of two distribution when they are disjoint with respect to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 718, + 153, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 153, + 732 + ], + "score": 1.0, + "content": "each other.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "19", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 151, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 153, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 153, + 95 + ], + "score": 1.0, + "content": "defined as:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 105, + 81, + 153, + 95 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 95, + 387, + 135 + ], + "lines": [ + { + "bbox": [ + 224, + 95, + 387, + 135 + ], + "spans": [ + { + "bbox": [ + 224, + 95, + 387, + 135 + ], + "score": 0.9, + "content": "\\begin{array} { r c l } { { f _ { \\theta } ^ { L 2 } ( x , t ) } } & { { = } } & { { \\displaystyle - \\frac { 1 } { 2 } | | s _ { \\theta } ( x , t ) | | ^ { 2 } } } \\\\ { { \\nabla _ { x } f ^ { L 2 } ( x , t ) } } & { { = } } & { { \\displaystyle - s _ { \\theta } ( x , t ) \\nabla _ { x } s _ { \\theta } ( x , t ) } } \\end{array}", + "type": "interline_equation", + "image_path": "308317851c04995c84d877779c6703a08df81fdfa71b93b066bc01481b437230.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 224, + 95, + 387, + 108.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 224, + 108.33333333333333, + 387, + 121.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 224, + 121.66666666666666, + 387, + 135.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 138, + 126, + 149 + ], + "lines": [ + { + "bbox": [ + 105, + 137, + 126, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 126, + 150 + ], + "score": 1.0, + "content": "and:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 137, + 126, + 150 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 150, + 397, + 183 + ], + "lines": [ + { + "bbox": [ + 214, + 150, + 397, + 183 + ], + "spans": [ + { + "bbox": [ + 214, + 150, + 397, + 183 + ], + "score": 0.91, + "content": "\\begin{array} { r c l } { { f _ { \\theta } ^ { I P } ( x , t ) } } & { { = } } & { { { x ^ { T } s _ { \\theta } ( x , t ) } } } \\\\ { { \\nabla _ { x } f ^ { I P } ( x , t ) } } & { { = } } & { { s _ { \\theta } ( x , t ) + x ^ { T } \\nabla _ { x } s _ { \\theta } ( x , t ) . } } \\end{array}", + "type": "interline_equation", + "image_path": "1c8bf03a53f8d9171b9535bdab83652cfd4067bb877a90c51dc2e5b6758c2ad3.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 214, + 150, + 397, + 166.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 214, + 166.5, + 397, + 183.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 190, + 506, + 257 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 203 + ], + "score": 1.0, + "content": "We train models with each parameterization on ImageNet and compare using FID for unconditional", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 202, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 505, + 214 + ], + "score": 1.0, + "content": "sampling. Results can be seen in Table A1. We see that L2 and Inner-Product perform the best, but", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 213, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 505, + 225 + ], + "score": 1.0, + "content": "are both outperformed by the standard parameterization. We initially experimented with these two", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 224, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 104, + 224, + 505, + 237 + ], + "score": 1.0, + "content": "parameterizations but found that the L2-norm parameterization to be more stable for compositional", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 235, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 247 + ], + "score": 1.0, + "content": "sampling due, we believe, to the fact that the energy-function is bounded above meaning that MCMC", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 246, + 385, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 385, + 258 + ], + "score": 1.0, + "content": "sampling is incapable of running off to infinity to increase likelihood.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9.5, + "bbox_fs": [ + 104, + 190, + 505, + 258 + ] + }, + { + "type": "table", + "bbox": [ + 208, + 267, + 402, + 310 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 208, + 267, + 402, + 310 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 208, + 267, + 402, + 310 + ], + "spans": [ + { + "bbox": [ + 208, + 267, + 402, + 310 + ], + "score": 0.966, + "html": "
Parameterization ∈θ(x,t)
(1) DAE 97.4(2)L2 Norm 91.5(3) Inner-Product | 90.9 86.7
", + "type": "table", + "image_path": "97d2d88b17d20d0c6716ab8c71cb269c16a92d1c369ae6787ecba5a32dda786c.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 208, + 267, + 402, + 281.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 208, + 281.3333333333333, + 402, + 295.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 208, + 295.66666666666663, + 402, + 309.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 113, + 317, + 493, + 329 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 316, + 494, + 330 + ], + "spans": [ + { + "bbox": [ + 116, + 316, + 494, + 330 + ], + "score": 1.0, + "content": "Table A1: FID (1k samples) of various energy-parameterizations on unconditional ImageNet generation.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + } + ], + "index": 15.0 + }, + { + "type": "title", + "bbox": [ + 106, + 349, + 348, + 363 + ], + "lines": [ + { + "bbox": [ + 104, + 348, + 349, + 365 + ], + "spans": [ + { + "bbox": [ + 104, + 348, + 349, + 365 + ], + "score": 1.0, + "content": "F SYNTHETIC DISTRIBUTION COMPOSITIONS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "list", + "bbox": [ + 105, + 375, + 504, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 374, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 145, + 387 + ], + "score": 1.0, + "content": "Mixture", + "type": "text" + }, + { + "bbox": [ + 152, + 375, + 506, + 388 + ], + "score": 1.0, + "content": "We provide additional 2D illustrations of mixtures of two diffusion models in Figure A5.", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 384, + 491, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 491, + 399 + ], + "score": 1.0, + "content": "We find that HMC sampling enables more accurate mixtures of different synthetic distributions.", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 374, + 506, + 399 + ] + }, + { + "type": "image", + "bbox": [ + 167, + 400, + 445, + 490 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 167, + 400, + 445, + 490 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 167, + 400, + 445, + 490 + ], + "spans": [ + { + "bbox": [ + 167, + 400, + 445, + 490 + ], + "score": 0.973, + "type": "image", + "image_path": "338e6c6a7dd7c357631290644720c829e3d176fbbd00a5a22297c26173aa4e08.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 167, + 400, + 445, + 430.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 167, + 430.0, + 445, + 460.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 167, + 460.0, + 445, + 490.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 498, + 504, + 520 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 498, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 506, + 510 + ], + "score": 1.0, + "content": "Figure A5: Examples of mixture applied to diffusion models. Left to right: Component distributions, reverse", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 508, + 503, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 503, + 520 + ], + "score": 1.0, + "content": "diffusion, HMC sampling. Reverse diffusion fails to sample accurately from mixed distributions distributions.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + } + ], + "index": 22.25 + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 530, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 149, + 543 + ], + "score": 1.0, + "content": "Negation", + "type": "text" + }, + { + "bbox": [ + 154, + 530, + 506, + 543 + ], + "score": 1.0, + "content": "We provide additional 2D illustrations of negating two diffusion model with respect to", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "each other in Figure A7. We find that HMC sampling enables accurate negations of different sythetic", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 551, + 161, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 161, + 565 + ], + "score": 1.0, + "content": "distributions.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 530, + 506, + 565 + ] + }, + { + "type": "image", + "bbox": [ + 167, + 565, + 445, + 655 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 167, + 565, + 445, + 655 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 167, + 565, + 445, + 655 + ], + "spans": [ + { + "bbox": [ + 167, + 565, + 445, + 655 + ], + "score": 0.969, + "type": "image", + "image_path": "55963e4abfacce6fab91050a8f0316825523a8bdca454b6206986a0879a3b063.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 167, + 565, + 445, + 595.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 167, + 595.0, + 445, + 625.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 167, + 625.0, + 445, + 655.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 664, + 504, + 686 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 664, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 506, + 675 + ], + "score": 1.0, + "content": "Figure A6: Examples of negation applied to diffusion models. Left to right: Component distributions, reverse", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 674, + 461, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 674, + 461, + 685 + ], + "score": 1.0, + "content": "diffusion, HMC sampling. Reverse diffusion fails to sample accurately from negated distributions.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + } + ], + "index": 30.25 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "Failure Cases Next we illustrate a failure case of composition using our approach in Figure A7.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "Our approach fails to generate the product of two distribution when they are disjoint with respect to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 718, + 153, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 153, + 732 + ], + "score": 1.0, + "content": "each other.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 698, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 166, + 70, + 446, + 163 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 166, + 70, + 446, + 163 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 166, + 70, + 446, + 163 + ], + "spans": [ + { + "bbox": [ + 166, + 70, + 446, + 163 + ], + "score": 0.97, + "type": "image", + "image_path": "29b13b8e41f9d2f5bb30dd2e97dc196d72cbe5f8948aea267dae7ef8bf47a6ad.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 166, + 70, + 446, + 101.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 166, + 101.0, + 446, + 132.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 166, + 132.0, + 446, + 163.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 172, + 504, + 193 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 171, + 506, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 506, + 183 + ], + "score": 1.0, + "content": "Figure A7: Failure Cases of Our Approach. Left to right: Component distributions, reverse diffusion, HMC", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 181, + 406, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 406, + 194 + ], + "score": 1.0, + "content": "sampling. Our approach fails to generate products of distributions with no overlap.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 108, + 213, + 258, + 226 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 259, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 259, + 228 + ], + "score": 1.0, + "content": "G EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 240, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 105, + 240, + 507, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 507, + 253 + ], + "score": 1.0, + "content": "We provide detailed experimental details including underlying quantitative metrics, training details,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 250, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 264 + ], + "score": 1.0, + "content": "and architectures on 2D synthetic, CLEVR, ImageNet, and test-to-image settings below. To enable", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 263, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 505, + 275 + ], + "score": 1.0, + "content": "stable training of energy-based diffusion models in image settings, we clip gradient norms to be less", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 273, + 459, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 459, + 286 + ], + "score": 1.0, + "content": "than 10, and initialize convolutional layers using zero-initialization (Zhang et al., 2019).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 107, + 290, + 505, + 345 + ], + "lines": [ + { + "bbox": [ + 106, + 290, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 505, + 302 + ], + "score": 1.0, + "content": "Synthetic Datasets For synthetic datasets, we train both score and energy based diffusion models", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 301, + 504, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 504, + 313 + ], + "score": 1.0, + "content": "using a small residual MLP model with 4 residual blocks, with a internal hidden dimension of 128", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 311, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 505, + 325 + ], + "score": 1.0, + "content": "dimensions. We train models for 15000 iterations (10 minutes on a 8 TPUv2 cores) using the Adam", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 336 + ], + "score": 1.0, + "content": "optimizer with learning rate of 1e-3, and train diffusion models on 100 discrete timesteps with linear", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 334, + 193, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 154, + 346 + ], + "score": 1.0, + "content": "schedule of", + "type": "text" + }, + { + "bbox": [ + 155, + 335, + 162, + 345 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 334, + 193, + 346 + ], + "score": 1.0, + "content": "values.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 351, + 505, + 440 + ], + "lines": [ + { + "bbox": [ + 105, + 350, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 505, + 364 + ], + "score": 1.0, + "content": "When evaluating product of diffusion models, we generate two separate distributions, where train", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 361, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 506, + 375 + ], + "score": 1.0, + "content": "two separate diffusion models. In our first distribution, we construct a GMM of 8 Gaussians in a ring", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 373, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 505, + 385 + ], + "score": 1.0, + "content": "of radius 0.5 around the origin, with each Gaussian having a standard deviation of 0.3. In our second", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 384, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 340, + 396 + ], + "score": 1.0, + "content": "dataset, we construct a uniform distribution of points with", + "type": "text" + }, + { + "bbox": [ + 340, + 386, + 347, + 394 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 384, + 451, + 396 + ], + "score": 1.0, + "content": "between -0.1 and 0.1 and", + "type": "text" + }, + { + "bbox": [ + 451, + 385, + 458, + 396 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 384, + 506, + 396 + ], + "score": 1.0, + "content": "between -1", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 394, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 407 + ], + "score": 1.0, + "content": "and 1. When evaluating mixture diffusion models, we generate one distribution consisting of a mixture", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 406, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 334, + 418 + ], + "score": 1.0, + "content": "of 3 Gaussian with standard deviation 0.03 and centers at", + "type": "text" + }, + { + "bbox": [ + 334, + 406, + 385, + 418 + ], + "score": 0.54, + "content": "\\left( - 0 . 2 5 , 0 . 5 \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 406, + 389, + 418 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 389, + 406, + 439, + 418 + ], + "score": 0.42, + "content": "( - 0 . 2 5 , 0 . 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 406, + 444, + 418 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 444, + 406, + 502, + 417 + ], + "score": 0.78, + "content": "\\left( - 0 . 2 5 , - 0 . 5 \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 406, + 506, + 418 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 416, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 506, + 428 + ], + "score": 1.0, + "content": "and another distribution consisting of a mixture of 3 Gaussian with standard deviation 0.03 and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 427, + 296, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 296, + 441 + ], + "score": 1.0, + "content": "centers at (0.25, 0.5), (0.25, 0.0), (0.25, −0.5).", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 107, + 444, + 505, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 443, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 458 + ], + "score": 1.0, + "content": "To construct MCMC samplers from models on synthetic datasets, we run 3 steps of HMC per timestep,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "with 3 leapfrog steps per step of HMC. We run 10 steps of MALA sampling per timestep. We found", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 467, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 505, + 479 + ], + "score": 1.0, + "content": "that MCMC performed robustly in the 2D dimensional setting and set the step size of MALA to be", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 477, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 506, + 490 + ], + "score": 1.0, + "content": "0.002 across all distributions and the step size of HMC to be 0.03 across all distributions (with a mass", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 488, + 157, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 157, + 501 + ], + "score": 1.0, + "content": "matrix of 1)", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 317, + 518 + ], + "score": 1.0, + "content": "CLEVR For CLEVR, we generated a dataset of", + "type": "text" + }, + { + "bbox": [ + 318, + 506, + 388, + 516 + ], + "score": 0.78, + "content": "2 0 0 , 0 0 0 \\ 6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 505, + 505, + 518 + ], + "score": 1.0, + "content": "images with between 1 to 5", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "different cubes using dataset generation code in (Liu et al., 2021). To evaluate the accuracy in which", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 527, + 507, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 507, + 542 + ], + "score": 1.0, + "content": "generated images had cubes at each specified position, we trained a binary classifier on these images,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "and marked a cube as correctly generated if the confidence of the binary confidence of classifier is", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 550, + 173, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 173, + 560 + ], + "score": 1.0, + "content": "greater than 0.5.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 565, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 566, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 505, + 578 + ], + "score": 1.0, + "content": "To parameterize our diffusion architecture, we follow the architecture of (Ho et al., 2020), where we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "score": 1.0, + "content": "use a base hidden dimension of 128, and multiply the hidden dimensions by [1, 2, 3, 4] at different", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "score": 1.0, + "content": "resolutions of the image. We utilize 3 residual blocks at each resolution of the image. We trained", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 598, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 356, + 611 + ], + "score": 1.0, + "content": "diffusion models with 100 discrete timesteps with a linear", + "type": "text" + }, + { + "bbox": [ + 357, + 600, + 365, + 610 + ], + "score": 0.83, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 598, + 506, + 611 + ], + "score": 1.0, + "content": "schedule. CLEVR models were", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 609, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 622 + ], + "score": 1.0, + "content": "trained for 20000 iterations with a batch size of 1024 using the Adam optimizer with step size 1e-4,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 621, + 319, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 319, + 634 + ], + "score": 1.0, + "content": "corresponding to roughly 8 hours on 8 TPUv2 cores.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "To initialize MCMC sampling on the CLEVR domain, at each timestep, before applying MCMC", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "sampling, we run one step of the reverse process in the trained diffusion model. We run 40 steps of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 658, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 104, + 658, + 506, + 674 + ], + "score": 1.0, + "content": "MCMC sampling per timestep for MALA samplers, and 13 steps of HMC sampling (with 3 leapfrog", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 670, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 381, + 683 + ], + "score": 1.0, + "content": "step per HMC step) (with the mass matrix of HMC samplers set to", + "type": "text" + }, + { + "bbox": [ + 382, + 671, + 389, + 682 + ], + "score": 0.82, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 670, + 506, + 683 + ], + "score": 1.0, + "content": "). We use HMC with partial", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 680, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 506, + 695 + ], + "score": 1.0, + "content": "momentum refreshment, and use a dampening coefficient of 0.9 across HMC iterations. MALA step", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 691, + 369, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 167, + 706 + ], + "score": 1.0, + "content": "sizes are set to", + "type": "text" + }, + { + "bbox": [ + 167, + 693, + 209, + 704 + ], + "score": 0.91, + "content": "0 . 0 3 5 * \\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 691, + 336, + 706 + ], + "score": 1.0, + "content": ", and HMC step sizes are set to", + "type": "text" + }, + { + "bbox": [ + 336, + 693, + 369, + 704 + ], + "score": 0.9, + "content": "0 . 1 * \\beta _ { t }", + "type": "inline_equation" + } + ], + "index": 44 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 390, + 722 + ], + "score": 1.0, + "content": "ImageNet For ImageNet, we train an unconditional diffusion model", + "type": "text" + }, + { + "bbox": [ + 391, + 710, + 434, + 720 + ], + "score": 0.88, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "images. We train", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "diffusion models for 1 million iterations of ImageNet with a batch size of 64 (3 days on 16 TPUv2", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 166, + 70, + 446, + 163 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 166, + 70, + 446, + 163 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 166, + 70, + 446, + 163 + ], + "spans": [ + { + "bbox": [ + 166, + 70, + 446, + 163 + ], + "score": 0.97, + "type": "image", + "image_path": "29b13b8e41f9d2f5bb30dd2e97dc196d72cbe5f8948aea267dae7ef8bf47a6ad.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 166, + 70, + 446, + 101.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 166, + 101.0, + 446, + 132.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 166, + 132.0, + 446, + 163.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 172, + 504, + 193 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 171, + 506, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 506, + 183 + ], + "score": 1.0, + "content": "Figure A7: Failure Cases of Our Approach. Left to right: Component distributions, reverse diffusion, HMC", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 181, + 406, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 406, + 194 + ], + "score": 1.0, + "content": "sampling. Our approach fails to generate products of distributions with no overlap.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 108, + 213, + 258, + 226 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 259, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 259, + 228 + ], + "score": 1.0, + "content": "G EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 240, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 105, + 240, + 507, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 507, + 253 + ], + "score": 1.0, + "content": "We provide detailed experimental details including underlying quantitative metrics, training details,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 250, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 264 + ], + "score": 1.0, + "content": "and architectures on 2D synthetic, CLEVR, ImageNet, and test-to-image settings below. To enable", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 263, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 505, + 275 + ], + "score": 1.0, + "content": "stable training of energy-based diffusion models in image settings, we clip gradient norms to be less", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 273, + 459, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 459, + 286 + ], + "score": 1.0, + "content": "than 10, and initialize convolutional layers using zero-initialization (Zhang et al., 2019).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 240, + 507, + 286 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 290, + 505, + 345 + ], + "lines": [ + { + "bbox": [ + 106, + 290, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 505, + 302 + ], + "score": 1.0, + "content": "Synthetic Datasets For synthetic datasets, we train both score and energy based diffusion models", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 301, + 504, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 504, + 313 + ], + "score": 1.0, + "content": "using a small residual MLP model with 4 residual blocks, with a internal hidden dimension of 128", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 311, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 505, + 325 + ], + "score": 1.0, + "content": "dimensions. We train models for 15000 iterations (10 minutes on a 8 TPUv2 cores) using the Adam", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 336 + ], + "score": 1.0, + "content": "optimizer with learning rate of 1e-3, and train diffusion models on 100 discrete timesteps with linear", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 334, + 193, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 154, + 346 + ], + "score": 1.0, + "content": "schedule of", + "type": "text" + }, + { + "bbox": [ + 155, + 335, + 162, + 345 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 334, + 193, + 346 + ], + "score": 1.0, + "content": "values.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 290, + 506, + 346 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 351, + 505, + 440 + ], + "lines": [ + { + "bbox": [ + 105, + 350, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 505, + 364 + ], + "score": 1.0, + "content": "When evaluating product of diffusion models, we generate two separate distributions, where train", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 361, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 506, + 375 + ], + "score": 1.0, + "content": "two separate diffusion models. In our first distribution, we construct a GMM of 8 Gaussians in a ring", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 373, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 505, + 385 + ], + "score": 1.0, + "content": "of radius 0.5 around the origin, with each Gaussian having a standard deviation of 0.3. In our second", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 384, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 340, + 396 + ], + "score": 1.0, + "content": "dataset, we construct a uniform distribution of points with", + "type": "text" + }, + { + "bbox": [ + 340, + 386, + 347, + 394 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 384, + 451, + 396 + ], + "score": 1.0, + "content": "between -0.1 and 0.1 and", + "type": "text" + }, + { + "bbox": [ + 451, + 385, + 458, + 396 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 384, + 506, + 396 + ], + "score": 1.0, + "content": "between -1", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 394, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 407 + ], + "score": 1.0, + "content": "and 1. When evaluating mixture diffusion models, we generate one distribution consisting of a mixture", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 406, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 334, + 418 + ], + "score": 1.0, + "content": "of 3 Gaussian with standard deviation 0.03 and centers at", + "type": "text" + }, + { + "bbox": [ + 334, + 406, + 385, + 418 + ], + "score": 0.54, + "content": "\\left( - 0 . 2 5 , 0 . 5 \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 406, + 389, + 418 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 389, + 406, + 439, + 418 + ], + "score": 0.42, + "content": "( - 0 . 2 5 , 0 . 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 406, + 444, + 418 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 444, + 406, + 502, + 417 + ], + "score": 0.78, + "content": "\\left( - 0 . 2 5 , - 0 . 5 \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 406, + 506, + 418 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 416, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 506, + 428 + ], + "score": 1.0, + "content": "and another distribution consisting of a mixture of 3 Gaussian with standard deviation 0.03 and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 427, + 296, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 296, + 441 + ], + "score": 1.0, + "content": "centers at (0.25, 0.5), (0.25, 0.0), (0.25, −0.5).", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 350, + 506, + 441 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 444, + 505, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 443, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 458 + ], + "score": 1.0, + "content": "To construct MCMC samplers from models on synthetic datasets, we run 3 steps of HMC per timestep,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "with 3 leapfrog steps per step of HMC. We run 10 steps of MALA sampling per timestep. We found", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 467, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 505, + 479 + ], + "score": 1.0, + "content": "that MCMC performed robustly in the 2D dimensional setting and set the step size of MALA to be", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 477, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 506, + 490 + ], + "score": 1.0, + "content": "0.002 across all distributions and the step size of HMC to be 0.03 across all distributions (with a mass", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 488, + 157, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 157, + 501 + ], + "score": 1.0, + "content": "matrix of 1)", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 443, + 506, + 501 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 317, + 518 + ], + "score": 1.0, + "content": "CLEVR For CLEVR, we generated a dataset of", + "type": "text" + }, + { + "bbox": [ + 318, + 506, + 388, + 516 + ], + "score": 0.78, + "content": "2 0 0 , 0 0 0 \\ 6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 505, + 505, + 518 + ], + "score": 1.0, + "content": "images with between 1 to 5", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "different cubes using dataset generation code in (Liu et al., 2021). To evaluate the accuracy in which", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 527, + 507, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 507, + 542 + ], + "score": 1.0, + "content": "generated images had cubes at each specified position, we trained a binary classifier on these images,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 551 + ], + "score": 1.0, + "content": "and marked a cube as correctly generated if the confidence of the binary confidence of classifier is", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 550, + 173, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 173, + 560 + ], + "score": 1.0, + "content": "greater than 0.5.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 505, + 507, + 560 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 565, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 566, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 505, + 578 + ], + "score": 1.0, + "content": "To parameterize our diffusion architecture, we follow the architecture of (Ho et al., 2020), where we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "score": 1.0, + "content": "use a base hidden dimension of 128, and multiply the hidden dimensions by [1, 2, 3, 4] at different", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 505, + 600 + ], + "score": 1.0, + "content": "resolutions of the image. We utilize 3 residual blocks at each resolution of the image. We trained", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 598, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 356, + 611 + ], + "score": 1.0, + "content": "diffusion models with 100 discrete timesteps with a linear", + "type": "text" + }, + { + "bbox": [ + 357, + 600, + 365, + 610 + ], + "score": 0.83, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 598, + 506, + 611 + ], + "score": 1.0, + "content": "schedule. CLEVR models were", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 609, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 622 + ], + "score": 1.0, + "content": "trained for 20000 iterations with a batch size of 1024 using the Adam optimizer with step size 1e-4,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 621, + 319, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 319, + 634 + ], + "score": 1.0, + "content": "corresponding to roughly 8 hours on 8 TPUv2 cores.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 566, + 506, + 634 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "To initialize MCMC sampling on the CLEVR domain, at each timestep, before applying MCMC", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "sampling, we run one step of the reverse process in the trained diffusion model. We run 40 steps of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 658, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 104, + 658, + 506, + 674 + ], + "score": 1.0, + "content": "MCMC sampling per timestep for MALA samplers, and 13 steps of HMC sampling (with 3 leapfrog", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 670, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 381, + 683 + ], + "score": 1.0, + "content": "step per HMC step) (with the mass matrix of HMC samplers set to", + "type": "text" + }, + { + "bbox": [ + 382, + 671, + 389, + 682 + ], + "score": 0.82, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 670, + 506, + 683 + ], + "score": 1.0, + "content": "). We use HMC with partial", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 680, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 506, + 695 + ], + "score": 1.0, + "content": "momentum refreshment, and use a dampening coefficient of 0.9 across HMC iterations. MALA step", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 691, + 369, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 167, + 706 + ], + "score": 1.0, + "content": "sizes are set to", + "type": "text" + }, + { + "bbox": [ + 167, + 693, + 209, + 704 + ], + "score": 0.91, + "content": "0 . 0 3 5 * \\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 691, + 336, + 706 + ], + "score": 1.0, + "content": ", and HMC step sizes are set to", + "type": "text" + }, + { + "bbox": [ + 336, + 693, + 369, + 704 + ], + "score": 0.9, + "content": "0 . 1 * \\beta _ { t }", + "type": "inline_equation" + } + ], + "index": 44 + } + ], + "index": 41.5, + "bbox_fs": [ + 104, + 637, + 506, + 706 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 390, + 722 + ], + "score": 1.0, + "content": "ImageNet For ImageNet, we train an unconditional diffusion model", + "type": "text" + }, + { + "bbox": [ + 391, + 710, + 434, + 720 + ], + "score": 0.88, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "images. We train", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "diffusion models for 1 million iterations of ImageNet with a batch size of 64 (3 days on 16 TPUv2", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "cores), using Adam optimizer with learning rate 1e-4, for 1 million iterations. We train diffusion", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 293, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 293, + 106 + ], + "score": 1.0, + "content": "models with 1000 discrete timesteps using the", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 45.5, + "bbox_fs": [ + 106, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "cores), using Adam optimizer with learning rate 1e-4, for 1 million iterations. We train diffusion", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 293, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 293, + 106 + ], + "score": 1.0, + "content": "models with 1000 discrete timesteps using the", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 110, + 505, + 154 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "score": 1.0, + "content": "On the ImageNet dataset, we report three seperate metrics. To report classifier accuracy, we feed", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 121, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 506, + 135 + ], + "score": 1.0, + "content": "generated sample into a ImageNet classifier trained on clean images, and label a image as correctly", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 464, + 145 + ], + "score": 1.0, + "content": "generated if the classifier of a generated image having the specified class is greater than", + "type": "text" + }, + { + "bbox": [ + 465, + 132, + 484, + 143 + ], + "score": 0.87, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 132, + 506, + 145 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 144, + 482, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 482, + 156 + ], + "score": 1.0, + "content": "further report the Inception Score and FID, which are calculated on 50000 generated samples.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 160, + 503, + 194 + ], + "lines": [ + { + "bbox": [ + 106, + 160, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 505, + 172 + ], + "score": 1.0, + "content": "We follow the architecture of (Ho et al., 2020), where we use a base hidden dimension of 128 and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 171, + 506, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 246, + 183 + ], + "score": 1.0, + "content": "multiply the hidden dimensions by", + "type": "text" + }, + { + "bbox": [ + 247, + 171, + 295, + 183 + ], + "score": 0.45, + "content": "[ 1 , 1 , 2 , 3 , 4 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 171, + 506, + 183 + ], + "score": 1.0, + "content": "at the different resolution of the image. We utilize 2", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 182, + 297, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 297, + 195 + ], + "score": 1.0, + "content": "residual blocks at each resolution of the image.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 199, + 505, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 198, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 212 + ], + "score": 1.0, + "content": "To initialize MCMC sampling on the ImageNet domain, at each timestep, before applying MCMC", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 210, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 505, + 222 + ], + "score": 1.0, + "content": "sampling, we run one step of the reverse process in the trained diffusion model. We run 6 steps of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 219, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 506, + 235 + ], + "score": 1.0, + "content": "MCMC sampling per timestep for MALA samplers, and 2 steps of HMC sampling (with 3 leapfrog", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 231, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 381, + 245 + ], + "score": 1.0, + "content": "steps per HMC step and with the mass matrix of HMC samplers set to", + "type": "text" + }, + { + "bbox": [ + 382, + 232, + 389, + 243 + ], + "score": 0.8, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 231, + 506, + 245 + ], + "score": 1.0, + "content": "). MALA step sizes are set to", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 241, + 309, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 139, + 255 + ], + "score": 0.9, + "content": "0 . 5 * \\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 241, + 266, + 258 + ], + "score": 1.0, + "content": ", and HMC step sizes are set to", + "type": "text" + }, + { + "bbox": [ + 266, + 243, + 300, + 255 + ], + "score": 0.83, + "content": "0 . 6 * \\beta _ { t } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 241, + 309, + 258 + ], + "score": 1.0, + "content": ".5", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 260, + 505, + 326 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 273 + ], + "score": 1.0, + "content": "Text-to-Image For text-to-image models, we train models for one week on an internal text/image", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "score": 1.0, + "content": "dataset consisting of 400 million images using 32 TPUv3 cores, with a training data batch size of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "256. We train our energy-based text-to-image model using a total of 1000 timesteps with a cosine", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "score": 1.0, + "content": "beta schedule. We follow the architecture of (Ho et al., 2020), where we use a base hidden dimension", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 300, + 316 + ], + "score": 1.0, + "content": "of 256, and multiply the hidden dimensions by", + "type": "text" + }, + { + "bbox": [ + 300, + 304, + 340, + 316 + ], + "score": 0.34, + "content": "[ 1 , 2 , 3 , 4 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "at different resolution of the image. We", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 313, + 331, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 331, + 328 + ], + "score": 1.0, + "content": "utilize 3 residual blocks at each resolution of the image.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 107, + 331, + 505, + 365 + ], + "lines": [ + { + "bbox": [ + 105, + 331, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 217, + 344 + ], + "score": 1.0, + "content": "To upsample images from", + "type": "text" + }, + { + "bbox": [ + 217, + 332, + 252, + 342 + ], + "score": 0.88, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 331, + 308, + 344 + ], + "score": 1.0, + "content": "resolution to", + "type": "text" + }, + { + "bbox": [ + 309, + 332, + 364, + 342 + ], + "score": 0.91, + "content": "1 0 2 4 \\times 1 0 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 331, + 506, + 344 + ], + "score": 1.0, + "content": "resolution, we utilize two trained", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 343, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 367, + 355 + ], + "score": 1.0, + "content": "unconditional diffusion models, one trained to upsample from", + "type": "text" + }, + { + "bbox": [ + 367, + 343, + 402, + 353 + ], + "score": 0.88, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 343, + 459, + 355 + ], + "score": 1.0, + "content": "resolution to", + "type": "text" + }, + { + "bbox": [ + 459, + 343, + 505, + 353 + ], + "score": 0.88, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 353, + 482, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 285, + 366 + ], + "score": 1.0, + "content": "resolution and one trained to upsample from", + "type": "text" + }, + { + "bbox": [ + 286, + 354, + 329, + 364 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 353, + 383, + 366 + ], + "score": 1.0, + "content": "resolution to", + "type": "text" + }, + { + "bbox": [ + 383, + 354, + 436, + 364 + ], + "score": 0.89, + "content": "1 0 2 4 \\times 1 0 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 353, + 482, + 366 + ], + "score": 1.0, + "content": "resolution.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 505, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "To initialize MCMC sampling on the text-to-image domain, at each timestep, before applying MCMC", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 381, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 394 + ], + "score": 1.0, + "content": "sampling, we run one step of the reverse process in the trained diffusion model. We ran 2 steps of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 391, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 505, + 406 + ], + "score": 1.0, + "content": "HMC sampling per timestep, with 3 leapfrog step per HMC step and a mass matrix of HMC samplers", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 130, + 416 + ], + "score": 1.0, + "content": "set to", + "type": "text" + }, + { + "bbox": [ + 130, + 404, + 138, + 415 + ], + "score": 0.8, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "). We use HMC with partial momentum refreshment, and use a dampening coefficient of 0.9", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 413, + 338, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 306, + 428 + ], + "score": 1.0, + "content": "across HMC iterations. HMC step sizes are set to", + "type": "text" + }, + { + "bbox": [ + 306, + 415, + 338, + 426 + ], + "score": 0.91, + "content": "0 . 1 * \\beta _ { t }", + "type": "inline_equation" + } + ], + "index": 27 + } + ], + "index": 25 + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 505, + 106 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 110, + 505, + 154 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "score": 1.0, + "content": "On the ImageNet dataset, we report three seperate metrics. To report classifier accuracy, we feed", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 121, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 506, + 135 + ], + "score": 1.0, + "content": "generated sample into a ImageNet classifier trained on clean images, and label a image as correctly", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 464, + 145 + ], + "score": 1.0, + "content": "generated if the classifier of a generated image having the specified class is greater than", + "type": "text" + }, + { + "bbox": [ + 465, + 132, + 484, + 143 + ], + "score": 0.87, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 132, + 506, + 145 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 144, + 482, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 482, + 156 + ], + "score": 1.0, + "content": "further report the Inception Score and FID, which are calculated on 50000 generated samples.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 110, + 506, + 156 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 160, + 503, + 194 + ], + "lines": [ + { + "bbox": [ + 106, + 160, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 505, + 172 + ], + "score": 1.0, + "content": "We follow the architecture of (Ho et al., 2020), where we use a base hidden dimension of 128 and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 171, + 506, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 246, + 183 + ], + "score": 1.0, + "content": "multiply the hidden dimensions by", + "type": "text" + }, + { + "bbox": [ + 247, + 171, + 295, + 183 + ], + "score": 0.45, + "content": "[ 1 , 1 , 2 , 3 , 4 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 171, + 506, + 183 + ], + "score": 1.0, + "content": "at the different resolution of the image. We utilize 2", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 182, + 297, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 297, + 195 + ], + "score": 1.0, + "content": "residual blocks at each resolution of the image.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 160, + 506, + 195 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 199, + 505, + 254 + ], + "lines": [ + { + "bbox": [ + 105, + 198, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 212 + ], + "score": 1.0, + "content": "To initialize MCMC sampling on the ImageNet domain, at each timestep, before applying MCMC", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 210, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 505, + 222 + ], + "score": 1.0, + "content": "sampling, we run one step of the reverse process in the trained diffusion model. We run 6 steps of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 219, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 506, + 235 + ], + "score": 1.0, + "content": "MCMC sampling per timestep for MALA samplers, and 2 steps of HMC sampling (with 3 leapfrog", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 231, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 381, + 245 + ], + "score": 1.0, + "content": "steps per HMC step and with the mass matrix of HMC samplers set to", + "type": "text" + }, + { + "bbox": [ + 382, + 232, + 389, + 243 + ], + "score": 0.8, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 231, + 506, + 245 + ], + "score": 1.0, + "content": "). MALA step sizes are set to", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 241, + 309, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 139, + 255 + ], + "score": 0.9, + "content": "0 . 5 * \\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 241, + 266, + 258 + ], + "score": 1.0, + "content": ", and HMC step sizes are set to", + "type": "text" + }, + { + "bbox": [ + 266, + 243, + 300, + 255 + ], + "score": 0.83, + "content": "0 . 6 * \\beta _ { t } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 241, + 309, + 258 + ], + "score": 1.0, + "content": ".5", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 198, + 506, + 258 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 260, + 505, + 326 + ], + "lines": [ + { + "bbox": [ + 105, + 259, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 273 + ], + "score": 1.0, + "content": "Text-to-Image For text-to-image models, we train models for one week on an internal text/image", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "score": 1.0, + "content": "dataset consisting of 400 million images using 32 TPUv3 cores, with a training data batch size of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "256. We train our energy-based text-to-image model using a total of 1000 timesteps with a cosine", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "score": 1.0, + "content": "beta schedule. We follow the architecture of (Ho et al., 2020), where we use a base hidden dimension", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 300, + 316 + ], + "score": 1.0, + "content": "of 256, and multiply the hidden dimensions by", + "type": "text" + }, + { + "bbox": [ + 300, + 304, + 340, + 316 + ], + "score": 0.34, + "content": "[ 1 , 2 , 3 , 4 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "at different resolution of the image. We", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 313, + 331, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 331, + 328 + ], + "score": 1.0, + "content": "utilize 3 residual blocks at each resolution of the image.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 259, + 506, + 328 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 331, + 505, + 365 + ], + "lines": [ + { + "bbox": [ + 105, + 331, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 217, + 344 + ], + "score": 1.0, + "content": "To upsample images from", + "type": "text" + }, + { + "bbox": [ + 217, + 332, + 252, + 342 + ], + "score": 0.88, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 331, + 308, + 344 + ], + "score": 1.0, + "content": "resolution to", + "type": "text" + }, + { + "bbox": [ + 309, + 332, + 364, + 342 + ], + "score": 0.91, + "content": "1 0 2 4 \\times 1 0 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 331, + 506, + 344 + ], + "score": 1.0, + "content": "resolution, we utilize two trained", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 343, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 367, + 355 + ], + "score": 1.0, + "content": "unconditional diffusion models, one trained to upsample from", + "type": "text" + }, + { + "bbox": [ + 367, + 343, + 402, + 353 + ], + "score": 0.88, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 343, + 459, + 355 + ], + "score": 1.0, + "content": "resolution to", + "type": "text" + }, + { + "bbox": [ + 459, + 343, + 505, + 353 + ], + "score": 0.88, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 353, + 482, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 285, + 366 + ], + "score": 1.0, + "content": "resolution and one trained to upsample from", + "type": "text" + }, + { + "bbox": [ + 286, + 354, + 329, + 364 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 353, + 383, + 366 + ], + "score": 1.0, + "content": "resolution to", + "type": "text" + }, + { + "bbox": [ + 383, + 354, + 436, + 364 + ], + "score": 0.89, + "content": "1 0 2 4 \\times 1 0 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 353, + 482, + 366 + ], + "score": 1.0, + "content": "resolution.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 331, + 506, + 366 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 505, + 426 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "To initialize MCMC sampling on the text-to-image domain, at each timestep, before applying MCMC", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 381, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 394 + ], + "score": 1.0, + "content": "sampling, we run one step of the reverse process in the trained diffusion model. We ran 2 steps of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 391, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 505, + 406 + ], + "score": 1.0, + "content": "HMC sampling per timestep, with 3 leapfrog step per HMC step and a mass matrix of HMC samplers", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 130, + 416 + ], + "score": 1.0, + "content": "set to", + "type": "text" + }, + { + "bbox": [ + 130, + 404, + 138, + 415 + ], + "score": 0.8, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "). We use HMC with partial momentum refreshment, and use a dampening coefficient of 0.9", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 413, + 338, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 306, + 428 + ], + "score": 1.0, + "content": "across HMC iterations. HMC step sizes are set to", + "type": "text" + }, + { + "bbox": [ + 306, + 415, + 338, + 426 + ], + "score": 0.91, + "content": "0 . 1 * \\beta _ { t }", + "type": "inline_equation" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 370, + 505, + 428 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/dev/T8wHz4rnuGL/T8wHz4rnuGL_origin.pdf b/parse/dev/T8wHz4rnuGL/T8wHz4rnuGL_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..0768692d4de6eb4a04b90e084645ca7e67e7d2f4 --- /dev/null +++ b/parse/dev/T8wHz4rnuGL/T8wHz4rnuGL_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4001d131bb65f697c80dbce07e120fa8744a5440d21907a43f09932dacfba5cd +size 2402868 diff --git a/parse/dev/UKd6dpVGdu/UKd6dpVGdu_origin.pdf b/parse/dev/UKd6dpVGdu/UKd6dpVGdu_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..2227461ce98fceff09480809432b1a9ee510476d --- /dev/null +++ b/parse/dev/UKd6dpVGdu/UKd6dpVGdu_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e89801fa12885f0e31ac2a44ae3465a1ace6901bf8c2581a41f87103f2f163a6 +size 3852709 diff --git a/parse/dev/UKd6dpVGdu/UKd6dpVGdu_span.pdf b/parse/dev/UKd6dpVGdu/UKd6dpVGdu_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..9ae4b87d2a079d2b75e1e63eb7772bb39397ef04 --- /dev/null +++ b/parse/dev/UKd6dpVGdu/UKd6dpVGdu_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d14f8cefd6dbe8c8f3e913580764651f772722effc6e5b275b96a8e5f44af3f0 +size 4048986 diff --git a/parse/dev/VYYf6S67pQc/VYYf6S67pQc_origin.pdf b/parse/dev/VYYf6S67pQc/VYYf6S67pQc_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..b5e6ec0e89ce5e9d57cab097a628fe2f2bc66783 --- /dev/null +++ b/parse/dev/VYYf6S67pQc/VYYf6S67pQc_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b3633b3dd0308c42d35d260c50864481580238d8f344f50e8ce2f14ee4be705 +size 639024 diff --git a/parse/dev/XSEBx0iSjFQ/XSEBx0iSjFQ_layout.pdf b/parse/dev/XSEBx0iSjFQ/XSEBx0iSjFQ_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..6d10ab25b3b04c1bbe628450e4826a319de4ebf1 --- /dev/null +++ b/parse/dev/XSEBx0iSjFQ/XSEBx0iSjFQ_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fabc06ca32bf43d4f25e9cce6bc76679ac90873e87d4bc23e9bdb8b724d52ca1 +size 30135842 diff --git a/parse/dev/ZlCpRiZN7n/images/102ad76a5c55204b11f3683212360fe910c8ebdb2ed54aa1530f47bf6a20c1b3.jpg b/parse/dev/ZlCpRiZN7n/images/102ad76a5c55204b11f3683212360fe910c8ebdb2ed54aa1530f47bf6a20c1b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a31f174fd6dc7a8b5e673b65c188f9d66c6b9042 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/102ad76a5c55204b11f3683212360fe910c8ebdb2ed54aa1530f47bf6a20c1b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60dc0c96c517f36bd4c040b3b111fa7fd2bbc904ce493f00eb89bb110efd1a94 +size 9181 diff --git a/parse/dev/ZlCpRiZN7n/images/1cbeeb6f22ecc7417becd52fbe23824931895f067aba20899f40ce406fce9521.jpg b/parse/dev/ZlCpRiZN7n/images/1cbeeb6f22ecc7417becd52fbe23824931895f067aba20899f40ce406fce9521.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9f5111c8a4177d0421d04242596d0a9d8141d315 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/1cbeeb6f22ecc7417becd52fbe23824931895f067aba20899f40ce406fce9521.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f65c6a40f45c57d7bb77558ef8bd5bb96155d51d37140a4dd097080c4f42d076 +size 16974 diff --git a/parse/dev/ZlCpRiZN7n/images/1cf6b003e9c6d20751c2aa0c45c9569c74226a20fb875407d84e6d762e676b95.jpg b/parse/dev/ZlCpRiZN7n/images/1cf6b003e9c6d20751c2aa0c45c9569c74226a20fb875407d84e6d762e676b95.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7297eb5ae9c0d52e65a86108e551a7dcb27e95e4 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/1cf6b003e9c6d20751c2aa0c45c9569c74226a20fb875407d84e6d762e676b95.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:794ab060b57f0c604ae901834ae06d786165feee340647542eaf7fc8a27ca657 +size 63862 diff --git a/parse/dev/ZlCpRiZN7n/images/38082fab1521aa3b75d46378d66aa3347f8dc11019d9aa1bc8138861c8d09894.jpg b/parse/dev/ZlCpRiZN7n/images/38082fab1521aa3b75d46378d66aa3347f8dc11019d9aa1bc8138861c8d09894.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2d7f34084b2dec4098fc579cb9be6626a2cdfa4f --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/38082fab1521aa3b75d46378d66aa3347f8dc11019d9aa1bc8138861c8d09894.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28630e6dde82907cac38dbdc78dd32659bdd8f5983a1b56131262126f5e25c44 +size 13586 diff --git a/parse/dev/ZlCpRiZN7n/images/390575267a73798ff734d2b3d84698a528d3b709a2de299e12ee8d07e73a7ae5.jpg b/parse/dev/ZlCpRiZN7n/images/390575267a73798ff734d2b3d84698a528d3b709a2de299e12ee8d07e73a7ae5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..49f95b8b3a658873a3ada9a859c46a29996bfe66 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/390575267a73798ff734d2b3d84698a528d3b709a2de299e12ee8d07e73a7ae5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc13dcff4eede2c4b1fdeb5a895e8ed3f84114cc316319dca226e5db128771ff +size 21169 diff --git a/parse/dev/ZlCpRiZN7n/images/6c23096473ed9ee8edb5e9728797f8df27b139789ffeeede3b47d9f93d1119a1.jpg b/parse/dev/ZlCpRiZN7n/images/6c23096473ed9ee8edb5e9728797f8df27b139789ffeeede3b47d9f93d1119a1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..64aa6e3a67a4a9e0f63d5f376e8f704645c96450 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/6c23096473ed9ee8edb5e9728797f8df27b139789ffeeede3b47d9f93d1119a1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9747218aaf9a773f38b08a3dafb6ef30f6d12795065c08c778c811e08688964 +size 30620 diff --git a/parse/dev/ZlCpRiZN7n/images/72b58e8021fb31ca374a7880e57d213bcf6587a1218cb94c376fe722d919fd44.jpg b/parse/dev/ZlCpRiZN7n/images/72b58e8021fb31ca374a7880e57d213bcf6587a1218cb94c376fe722d919fd44.jpg new file mode 100644 index 0000000000000000000000000000000000000000..10f097aab03765c661c340af98474ee356e7438e --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/72b58e8021fb31ca374a7880e57d213bcf6587a1218cb94c376fe722d919fd44.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:126796c787b91a7ffe05fb8ea7b3749a54289159fc68c3b383f540d4813ff5d3 +size 18990 diff --git a/parse/dev/ZlCpRiZN7n/images/820a62a7472bc2879030f709c5eb4cdebe398fb43b941bb5025f417d2c1fb7f4.jpg b/parse/dev/ZlCpRiZN7n/images/820a62a7472bc2879030f709c5eb4cdebe398fb43b941bb5025f417d2c1fb7f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ca740439053ec81bca24001ff96d1c8dbda557e2 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/820a62a7472bc2879030f709c5eb4cdebe398fb43b941bb5025f417d2c1fb7f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8471406ce760bcbaf93e00ac5aa6aeea2027b9c1e2c2a921718afe0468e4a51 +size 12881 diff --git a/parse/dev/ZlCpRiZN7n/images/89d55eee1275a008a78b4522afa6cc3a0e3e2f4d93c2b45d36eb42bf6092db5c.jpg b/parse/dev/ZlCpRiZN7n/images/89d55eee1275a008a78b4522afa6cc3a0e3e2f4d93c2b45d36eb42bf6092db5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9113ab7ac71157c4365681f867c73371af4845cf --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/89d55eee1275a008a78b4522afa6cc3a0e3e2f4d93c2b45d36eb42bf6092db5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2bca42388b6f3f6eff4f175a2d9700a3ea9471b5a40f2fdc73ec73cabf733fa3 +size 26500 diff --git a/parse/dev/ZlCpRiZN7n/images/8d0736cec61698f5eb785bcfe105d39c7f8f6cb80b39a5e5528ff789de4e520f.jpg b/parse/dev/ZlCpRiZN7n/images/8d0736cec61698f5eb785bcfe105d39c7f8f6cb80b39a5e5528ff789de4e520f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9601092b40ee669753d7b5ce9ad07d21c4d11643 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/8d0736cec61698f5eb785bcfe105d39c7f8f6cb80b39a5e5528ff789de4e520f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d67ebbba11de2977bf920db34071335ddd87a4131590c1c14af1c8341c4caf64 +size 12185 diff --git a/parse/dev/ZlCpRiZN7n/images/8f1c1484906c82d4a974e9ddf2876a5dc94d19989006a72748c54651ffd83775.jpg b/parse/dev/ZlCpRiZN7n/images/8f1c1484906c82d4a974e9ddf2876a5dc94d19989006a72748c54651ffd83775.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a9794f2373051fcd1e071e0811477954bba12bb9 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/8f1c1484906c82d4a974e9ddf2876a5dc94d19989006a72748c54651ffd83775.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b178ffad0887bbfbd124d19a82cf6e3bd72a6bc8f036fcce4911a59e5e80b9ac +size 108986 diff --git a/parse/dev/ZlCpRiZN7n/images/90af06c8aa07dd9793efa9608248faf521106cb9a0aeeaed71f628138518ca9c.jpg b/parse/dev/ZlCpRiZN7n/images/90af06c8aa07dd9793efa9608248faf521106cb9a0aeeaed71f628138518ca9c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cde3d7be3edfcb4a9278cf08b33db805d9ef2d71 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/90af06c8aa07dd9793efa9608248faf521106cb9a0aeeaed71f628138518ca9c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5003644758c9f4b5f0345b5a269c83d4b5134b832f2f15bd2e4a273e243e379 +size 61123 diff --git a/parse/dev/ZlCpRiZN7n/images/aaf7c2e95d889f6fbf0054398c3813793b963ca757b945d0169d1fc51728394f.jpg b/parse/dev/ZlCpRiZN7n/images/aaf7c2e95d889f6fbf0054398c3813793b963ca757b945d0169d1fc51728394f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c460364a874f18b60f0392db5c8cdd4973bf9d55 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/aaf7c2e95d889f6fbf0054398c3813793b963ca757b945d0169d1fc51728394f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c43ecbe85d0bebc177803f19e3ce2f716405ad1b3ecebc0def25bc80e9c6569 +size 96603 diff --git a/parse/dev/ZlCpRiZN7n/images/b2c261f100ec1a3d7e7e980ea7a952e2a9b8979657de0af6d6eea3cb78ad3c10.jpg b/parse/dev/ZlCpRiZN7n/images/b2c261f100ec1a3d7e7e980ea7a952e2a9b8979657de0af6d6eea3cb78ad3c10.jpg new file mode 100644 index 0000000000000000000000000000000000000000..220fbb9f65fad838bca88e4d0129bf17c9b9f3cb --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/b2c261f100ec1a3d7e7e980ea7a952e2a9b8979657de0af6d6eea3cb78ad3c10.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dba57c933cd654e75efd6755f4122c61bf7b58a5abeb4b663ca9500b8dab21d5 +size 12012 diff --git a/parse/dev/ZlCpRiZN7n/images/b80a2f0e0a5dfae44e78f609fb3e6ecd0d90057d4efa703a095b0127ad6de59c.jpg b/parse/dev/ZlCpRiZN7n/images/b80a2f0e0a5dfae44e78f609fb3e6ecd0d90057d4efa703a095b0127ad6de59c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..617a72b7623b8a9074abd355c86a1ebf8a212f29 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/b80a2f0e0a5dfae44e78f609fb3e6ecd0d90057d4efa703a095b0127ad6de59c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed100e8f07e164ebc8cd511b6f304e77a2c1ae395fc714ca3713baa46822635a +size 64626 diff --git a/parse/dev/ZlCpRiZN7n/images/bb157af8ace2fd0b9e7f8a17234c6666e151d7c1f7de059b24fa24663cdcff97.jpg b/parse/dev/ZlCpRiZN7n/images/bb157af8ace2fd0b9e7f8a17234c6666e151d7c1f7de059b24fa24663cdcff97.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eea5488589de8c8811ab417773f35b77a1410801 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/bb157af8ace2fd0b9e7f8a17234c6666e151d7c1f7de059b24fa24663cdcff97.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:544a1d3c3673a95f7c417466c1e5215bee8fcd4f9f89e2436b805fedef969997 +size 2833 diff --git a/parse/dev/ZlCpRiZN7n/images/bd4096a200c9dcd14be078819d4e6465ca902e387c516eaf9c48a56560061bfe.jpg b/parse/dev/ZlCpRiZN7n/images/bd4096a200c9dcd14be078819d4e6465ca902e387c516eaf9c48a56560061bfe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2038c4db9ee496efae8c34f66e2fba5da6a5a7ea --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/bd4096a200c9dcd14be078819d4e6465ca902e387c516eaf9c48a56560061bfe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2632c8673708354aea53b38a8774f51abeabc45a631153b66d35e61b1db6ed6 +size 4697 diff --git a/parse/dev/ZlCpRiZN7n/images/be91ce1475fe30e8720ce2e3621be48a9d4d55757f51e14c472bd8841ffc5a82.jpg b/parse/dev/ZlCpRiZN7n/images/be91ce1475fe30e8720ce2e3621be48a9d4d55757f51e14c472bd8841ffc5a82.jpg new file mode 100644 index 0000000000000000000000000000000000000000..085b3804c5a8fe181d6ca2ca383d123ea4605eff --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/be91ce1475fe30e8720ce2e3621be48a9d4d55757f51e14c472bd8841ffc5a82.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d3bc62433fd66ef68d2f4b558ae930b438678817eb66cf22665624f44b5bdb8 +size 3289 diff --git a/parse/dev/ZlCpRiZN7n/images/d0e403ac5a1cb8d414f11a4ee78ae3ec365d69c1a53c3b9c33ba40fa3b06f0dc.jpg b/parse/dev/ZlCpRiZN7n/images/d0e403ac5a1cb8d414f11a4ee78ae3ec365d69c1a53c3b9c33ba40fa3b06f0dc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c7dd77621a78c3bf0de40a530edeab9045bcd46d --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/d0e403ac5a1cb8d414f11a4ee78ae3ec365d69c1a53c3b9c33ba40fa3b06f0dc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:93927730540f210f6d9a6b8f8e8314a6ff5e966676f63c60f69e2502a7665eea +size 12188 diff --git a/parse/dev/ZlCpRiZN7n/images/d1cd2e263f10c48e42a7327440c9af567204f199a969ad3ef81f5e08bd3b082e.jpg b/parse/dev/ZlCpRiZN7n/images/d1cd2e263f10c48e42a7327440c9af567204f199a969ad3ef81f5e08bd3b082e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ce5003e2a91b9ca86363cf8a3d3f3a2cd317681d --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/d1cd2e263f10c48e42a7327440c9af567204f199a969ad3ef81f5e08bd3b082e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9aaa91fa335ed5e3e3169f5b3c060d2323336f1607cc1e8d3bde1faa6340de0d +size 59863 diff --git a/parse/dev/ZlCpRiZN7n/images/d5938c64280d6d34169a787c675d830390dd3f7d9107b41f341c64d6c1e9c21e.jpg b/parse/dev/ZlCpRiZN7n/images/d5938c64280d6d34169a787c675d830390dd3f7d9107b41f341c64d6c1e9c21e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5310b1b46846f26285c85fff2f0a32d14fd78ca6 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/d5938c64280d6d34169a787c675d830390dd3f7d9107b41f341c64d6c1e9c21e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bfbabe552ca5cb0e3adb456b0ffde147b7915ae4ade79ec8e8f0e0e4773f1298 +size 29303 diff --git a/parse/dev/ZlCpRiZN7n/images/fc086e291577768b8aa75ce0a4339383c25c4eaa1ecddbef725b3888b41628b1.jpg b/parse/dev/ZlCpRiZN7n/images/fc086e291577768b8aa75ce0a4339383c25c4eaa1ecddbef725b3888b41628b1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..45e03e905152605426f6e41c246765480f02b4fe --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/fc086e291577768b8aa75ce0a4339383c25c4eaa1ecddbef725b3888b41628b1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96440126621cc35a6f7b58d443fb50107ee1855cd18117db591331b1bed1f48e +size 30622 diff --git a/parse/dev/ZlCpRiZN7n/images/fc3cf7f7df000b260535a1863454ba8784982f78a1441f9827b4b6af50041b91.jpg b/parse/dev/ZlCpRiZN7n/images/fc3cf7f7df000b260535a1863454ba8784982f78a1441f9827b4b6af50041b91.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a05d3ef26b4445fc256ec91f52b36795d85b62a --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/fc3cf7f7df000b260535a1863454ba8784982f78a1441f9827b4b6af50041b91.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b247a260430dc4af6d0b84e09871d59e4acb4a43466c706ff105390dcdfcfdb +size 36528 diff --git a/parse/dev/ZlCpRiZN7n/images/ff21b27fe42af9751709b511a5069210b107ecbb0a16ef22534cde1673342297.jpg b/parse/dev/ZlCpRiZN7n/images/ff21b27fe42af9751709b511a5069210b107ecbb0a16ef22534cde1673342297.jpg new file mode 100644 index 0000000000000000000000000000000000000000..583334a39046aed9eff2773e56476ba6644484e1 --- /dev/null +++ b/parse/dev/ZlCpRiZN7n/images/ff21b27fe42af9751709b511a5069210b107ecbb0a16ef22534cde1673342297.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:def87c802579d68679b112bd61a246c255ab091003ecc28b12864985caa15609 +size 24431 diff --git a/parse/dev/cOtBRgsf2fO/cOtBRgsf2fO.md b/parse/dev/cOtBRgsf2fO/cOtBRgsf2fO.md new file mode 100644 index 0000000000000000000000000000000000000000..d5bbadff70626474b686684e67ae4df23f224497 --- /dev/null +++ b/parse/dev/cOtBRgsf2fO/cOtBRgsf2fO.md @@ -0,0 +1,546 @@ +# LABEL LEAKAGE AND PROTECTION IN TWO-PARTY SPLIT LEARNING + +Oscar $\mathbf { L i } ^ { 1 \dagger }$ ∗, Jiankai $\mathbf { S u n ^ { 2 \ S } }$ , $\mathbf { X i n \ Y a n g ^ { 2 } }$ , Weihao $\mathbf { G a o } ^ { 2 }$ , Hongyi Zhang2, Junyuan $\mathbf { X i e } ^ { 2 }$ , Virginia Smith1, Chong Wang2 + +1Carnegie Mellon University +2ByteDance Inc. + +# ABSTRACT + +Two-party split learning is a popular technique for learning a model across featurepartitioned data. In this work, we explore whether it is possible for one party to steal the private label information from the other party during split training, and whether there are methods that can protect against such attacks. Specifically, we first formulate a realistic threat model and propose a privacy loss metric to quantify label leakage in split learning. We then show that there exist two simple yet effective methods within the threat model that can allow one party to accurately recover private ground-truth labels owned by the other party. To combat these attacks, we propose several random perturbation techniques, including Marvell, an approach that strategically finds the structure of the noise perturbation by minimizing the amount of label leakage (measured through our quantification metric) of a worst-case adversary. We empirically1 demonstrate the effectiveness of our protection techniques against the identified attacks, and show that Marvell in particular has improved privacy-utility tradeoffs relative to baseline approaches. + +# 1 INTRODUCTION + +With increasing concerns over data privacy in machine learning, federated learning (FL) (McMahan et al., 2017) has become a promising direction of study. Based on how sensitive data are distributed among parties, FL can be classified into different categories, notable among which are horizontal $F L$ and vertical $F L$ (Yang et al., 2019). In contrast to horizontal FL where the data are partitioned by examples, vertical FL considers data partitioned by features (including labels). As a canonical example of vertical FL, consider an online media platform $A$ which displays advertisements from company $B$ to its users, and charges $B$ for each conversion (e.g., a user clicking the ad and buying the product). In this case, both parties have different features for each user: $A$ has features on the user’s media viewing records, while $B$ has the user’s conversion labels. $B$ ’s labels are not available to $A$ because each user’s purchase behaviors happen entirely on $B$ ’s website/app. + +If both parties want to jointly learn a model to predict conversion without data sharing, split learning (Gupta & Raskar, 2018; Vepakomma et al., 2018) can be used to split the execution of a deep network between the parties on a layer-wise basis. In vanilla split learning, before training begins, both parties use Private Set Intersection (PSI) protocols (Kolesnikov et al., 2016; Pinkas et al., 2018) to find the intersection of their data records and achieve an example ID alignment. This alignment paves the way for the split training phase. During training (Figure 1), the party without labels (nonlabel party) sends the intermediate layer (cut layer) outputs rather than the raw data to the party with labels (label party), and the label party completes the rest of the forward computation to obtain the training loss. To compute the gradients with respect to model parameters, the label party initiates backpropagation from its training loss and computes its own parameters’ gradients. To allow the non-label party to also compute gradients of its parameters, the label party also computes the gradients with respect to the cut layer outputs and communicates this information back to the non-label party. As a result of the ID alignment, despite not knowing the label party’s raw label data, the non-label party can identify the gradient value returned by the label party for each example. + +At first glance, the process of split learning appears privacy-preserving because only the intermediate computations of the cut layer—rather than raw features or labels—are communicated between the two parties. However, such “gradient sharing” schemes have been shown to be vulnerable to privacy leakage in horizontal FL settings (e.g., Zhu et al., 2019). In vertical FL (and specifically split learning), it remains unclear whether the raw data can similarly be leaked during communication. In particular, as the raw labels often contain highly sensitive information (e.g., what a user has purchased (in online advertising) or whether a user has a disease or not (in disease prediction) Vepakomma et al. (2018)), developing a rigorous understanding of the threat of label leakage and its protection is particularly important. Towards this goal, we make the following contributions: + +![](images/6d3962aa293ad26de1b3789a4155ec37fededb9027eaf28bf79d4454dafbce24.jpg) +Figure 1: Communication diagram of two-party split training for an example of online advertising. We study whether it is possible for the communicated gradient $g$ to leak private label information. + +1. We formalize a threat model for label leakage in two-party split learning in the context of binary classification (Section 3.1), and propose specific privacy quantification metrics to measure the severity of such threats (Section 3.2). +2. We identify two simple and realistic methods within this threat model which can accurately recover the label party’s private label information (Section 3.3). +3. We propose several random perturbation techniques to limit the label-stealing ability of the non-label party (Section 4). Among them, our principled approach Marvell directly searches for the optimal random perturbation noise structure to minimize label leakage (as measured via our quantification metric) against a worst-case adversarial non-label party. +4. We experimentally demonstrate the effectiveness of our protection techniques and MARVELL’s improved privacy-utility tradeoffs compared to other protection baselines (Section 5). + +# 2 RELATED WORK + +Privacy leakage in split learning. Although raw data is not shared in federated learning, sensitive information may still be leaked when gradients and/or model parameters are communicated between parties. In horizontal FL, Zhu et al. (2019) showed that an honest-but-curious server can uncover the raw features and labels of a device by knowing the model architecture, parameters, and communicated gradient of the loss on the device’s data. Based on their techniques, Zhao et al. (2020) showed that the ground truth label of an example can be extracted by exploiting the directions of the gradients of the weights connected to the logits of different classes. Here we study a different setting—two-party split learning (in vertical FL) (Yang et al., 2019), where no party has access to the model architecture or model parameters of the other party. In this setting, Vepakomma et al. (2019) studied how the forward communication of feature representations can leak the non-label party’s raw data to the label party. We instead study whether label information may be leaked from the label party to the non-label party during the backward communication. Despite the importance of maintaining the privacy of these labels, we are unaware of prior work that has studied this problem. + +Privacy protection and quantification. Techniques to protect communication privacy in FL generally fall into three categories: 1) cryptographic methods such as secure multi-party computation (e.g., Bonawitz et al., 2017); 2) system-based methods including trusted execution environments (Subramanyan et al., 2017); and 3) perturbation methods that shuffle or modify the communicated messages (e.g., Abadi et al., 2016; McMahan et al., 2018; Erlingsson et al., 2019; Cheu et al., 2019; Zhu et al., 2019). Our protection techniques belong to the third category, as we add random perturbations to the gradients to protect the labels. Many randomness-based protection methods have been proposed in the domain of horizontal FL. In this case, differential privacy (DP) (Dwork, 2006; Dwork et al., 2014) is commonly used to measure the proposed random mechanisms’ ability to anonymize the identity of any single participating example in the model iterates. However, in split learning, after PSI, both parties know exactly the identity of which example has participated in a given gradient update. As we explain in Section 3.1, the object we aim to protect (the communicated cut layer gradients), unlike the model iterates, is not an aggregate function of all the examples but are instead example-specific. As a result, DP and its variants (e.g. label DP (Chaudhuri & Hsu, 2011; Ghazi et al., 2021)) are not directly applicable metrics in our setting, and we instead propose a different metric (discussed in Section 3.2). + +# 3 LABEL LEAKAGE IN SPLIT LEARNING + +We first introduce the two-party split learning problem for binary classification, and then formally describe our threat model and privacy quantification metrics with two concrete attack examples. + +# 3.1 TWO-PARTY SPLIT LEARNING IN BINARY CLASSIFICATION + +Problem setup. Consider two parties learning a composition model $h \circ f$ jointly for a binary classification problem over the domain $\mathcal { X } \times \{ 0 , \bar { 1 } \}$ (Figure 1). The non-label party owns the representation function $f : \mathcal { X } \to \mathbb { R } ^ { d }$ and each example’s raw feature $X \in { \mathcal { X } }$ while the label party owns the logit function $h : \mathbb { R } ^ { d } \mathbb { R }$ and each example’s label $y \in \{ 0 , 1 \}$ . Let $\ell \ = \ h ( f ( X ) )$ be the logit of the positive class whose predicted probability is given through the sigmoid function: $\widetilde { p } _ { 1 } = 1 / ( 1 + \exp ( - \ell ) )$ . We measure the loss of such prediction through the cross entropy loss $L = \log ( 1 + \exp ( - \ell ) ) + ( 1 - y ) \ell$ . During model inference, the non-label party computes ${ \dot { \boldsymbol { f } } } ( { \boldsymbol { X } } )$ and sends it to the label party who will then execute the rest of forward computation in Figure 1. + +Model training (Figure 1: backward gradient computation). To train the model using gradient descent, the label party starts by first computing the gradient of the loss $L$ with respect to the logit $\begin{array} { r } { \frac { d L } { d \ell } \ = \ ( \widetilde { p } _ { 1 } - y ) } \end{array}$ . Using the chain rule, the label party can then compute the gradient of $L$ with erespect to its function $h$ ’s parameters and perform the gradient updates. To also allow the non-label party to learn its function $f$ , the label party needs to additionally compute the gradient with respect to cut layer feature $f ( X )$ and communicate it to the non-label party. We denote this gradient by $g : = \nabla _ { f ( X ) } L = ( \widetilde { p } _ { 1 } - y ) \nabla _ { z } h ( z ) | _ { z = f ( X ) } \in \mathbb { R } ^ { d }$ (by chain rule). After receiving $g$ , the non-label eparty continues the backpropagation towards $f$ ’s parameters and also perform the gradient updates. + +Why Not Differential Privacy? Note that for a given iteration, the non-label party randomly chooses $B$ example IDs to form a batch. Therefore, the identity of which examples are used is known to the non-label party by default. In addition, the communicated features $f ( X )$ and returned gradients $g$ will both be matrices in $\mathbb { R } ^ { B \times d }$ with each row belonging to a specific example in the batch. The different gradients (rows of the matrix) are not with respect to the same model parameters, but are instead with respect to different examples’ cut-layer features; thus, no averaging over or shuffling of the rows of the gradient matrix can be done prior to communication to ensure correct computation of $f$ ’s parameters on the non-label party side. This example-aware and example-specific nature of the communicated gradient matrix makes differential privacy (which focuses on anonymizing an example’s participation in an aggregate function) inapplicable for this problem (see also Section 2). + +# 3.2 THREAT MODEL AND PRIVACY QUANTIFICATION + +Below we specify several key aspects of our threat model, including the adversary’s objective and capabilities, our metric for quantifying privacy loss, and the possible inclusion of side information. + +Adversary’s objective. At a given moment in time during training (with $f$ and $h$ fixed), since the communicated cut layer gradient $g$ is a deterministic function of $y$ (see Section 3.1), we consider an adversarial non-label party whose objective is to recover the label party’s hidden label $y$ based on the information contained in $g$ for every training example. + +Adversary’s capability. We consider an honest-but-curious non-label party which cannot tamper with training by selecting which examples to include in a batch or sending incorrect features $f ( X )$ ; instead, we assume that the adversary follows the agreed-upon split training procedure while trying to guess the label $y$ . This can be viewed as a binary classification problem where the (input, output) distribution is the induced distribution of $( g , y )$ . We allow the adversary to use any binary classifier $q : \mathbb { R } ^ { d } \{ 0 , 1 \}$ to guess the labels. This classifier can be represented by a (scoring function $r$ , threshold $t$ ) tuple, where $r : \mathbb { R } ^ { d } \mathbb { R }$ maps an example’s cut layer gradient to a real-valued score and the threshold $t \in \mathbb { R }$ determines a cut-off so that $\bar { q ( g ) } = 1$ if $\overset { \cdot } { r } ( g ) > t$ and $q ( g ) = 0$ if $r ( g ) \leq t$ . Moving forward, we use this tuple representation to describe adversarial non-label party classifiers. + +Privacy loss quantification. As we consider binary classification, a natural metric to quantify the performance of an adverary’s scoring function $r$ is the AUC of its ROC curve. Denote the unperturbed class-conditional distributions of the cut-layer gradients by $P ^ { ( 1 ) }$ and $P ^ { ( 0 ) }$ for the positive and negative class, respectively. The ROC curve of a scoring function $r$ is a parametric curve $t \mapsto \mathsf { \bar { ( F P R } } _ { r } ( t ) , \mathsf { T P R } _ { r } ( t ) ) \in \mathsf { [ 0 , 1 ] } ^ { 2 }$ which maps a threshold value $t \in \mathbb { R }$ to the corresponding (False Positive Rate, True Positive Rate) tuple of the classifier represented by $( r , t )$ , with $\mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } ( \{ g : r ( g ) > t \} )$ and $\mathrm { T P R } _ { r } ( t ) : = P ^ { ( 1 ) } ( \{ g : r ( g ) > t \} )$ . The AUC of the ROC curve of a scoring function $r$ (denote by $\operatorname { A U C } ( r )$ ) can be expressed as an integral: + +![](images/abe502d96660f12b083760acfe08c4529f2a3f7382efbc95190a0a239c34db31.jpg) +Figure 2: Distributions of quantities discussed in Observations 1-4 after the first 100, 200, 300, 400, 500 steps of stochastic gradient descent training of the WDL model on Criteo (see experiments). + +$$ +\begin{array} { r } { \begin{array} { r l } { \mathrm { A U C } ( r ) = \int _ { \infty } ^ { - \infty } \mathrm { T P R } _ { r } ( t ) ~ d \mathrm { F P R } _ { r } ( t ) } & { \in [ 0 , 1 ] } \end{array} } \end{array} +$$ + +(Leak AUC) + +(more details on this expression see Appendix A.1.) We use this value as the privacy loss quantification metric for a specific adversary scoring function $r$ and refer to it as the leak AUC. This metric summarizes the predictive performance of all classifiers that can be constructed through all threshold values $t$ and removes the need to tune this classifier-specific hyperparameter. The leak AUC being close to 1 implies that the corresponding scoring function $r$ can very accurately recover the private label, whereas a value of around 0.5 means $r$ is non-informative in predicting the labels. In practice, during batch training, the leak AUC of $r$ can be estimated at every gradient update iteration using the minibatch of cut-layer gradients together with their labels. + +Side information. Among all the scoring functions within our threat model, it is conceivable that only some would recover the hidden labels accurately. Picking such effective ones would require the non-label party to have population-level side information specifically regarding the properties of (and distinction between) the positive and negative class’s cut-layer gradient distributions. Since we allow the adversary to pick any specific (measurable) scoring function, we implicitly allow for such population-level side information for the adversary. However, we assume the non-label party has no example-level side information that is different example by example. Thus we also don’t use local DP for privacy quantification (detailed explanation in Appendix A.8). Next we provide two example scoring functions which use population-level side-information to effectively recover the label. + +# 3.3 PRACTICAL ATTACK METHODS + +Attack 1: Norm-based scoring function. Note that $\| g \| _ { 2 } = | \widetilde p _ { 1 } - y | \cdot \| \nabla _ { a } h ( a ) | _ { a = \pmb { f } ( X ) } \| _ { 2 }$ . We make the following observations for $| \widetilde { p } _ { 1 } - y |$ and $\| \nabla _ { a } h ( a ) | _ { a = f ( X ) } \| _ { 2 }$ , which hold true for a wide range of real-world learning problems. + +• Observation 1: Throughout training, the model tends to be less confident about a positive example being positive than a negative example being negative. In other words, the confidence gap of a positive example $1 - \widetilde { p } _ { 1 } = \left| \widetilde { p } _ { 1 } - y \right|$ (when $y = 1$ ) is typically larger than the confidence gap of a negative example $1 - \widetilde { p } _ { 0 } = \widetilde { p } _ { 1 } = | \widetilde { p } _ { 1 } - y |$ (when $y = 0$ ) (see Figure 2(a)). This observation e e eis particularly true for problems like advertising conversion prediction and disease prediction, where there is inherently more ambiguity for the positive class than the negative. For example, in advertising, uninterested users of a product will never click on its ad and convert, but those interested, even after clicking, might make the purchase only a fraction of the time depending on time/money constraints. (See A.2 for such an ambiguity even for a class-balanced setting.) + +• Observation 2: Throughout training, the norm of the gradient vector $\| \nabla _ { z } h ( z ) \vert _ { z = f ( X ) } \| _ { 2 }$ is on the same order of magnitude (has similar distribution) for both the positive and negative examples (Figure 2(b)). This is natural because $\nabla _ { a } h ( a ) | _ { a = f ( X ) }$ is not a function of $y$ . + +As a consequence of Observation 1 and 2, the gradient norm $\| g \| _ { 2 }$ of the positive instances are generally larger than that of the negative ones (Figure 2(c)). Thus, the scoring function $r _ { n } ( g ) =$ $\| g \| _ { 2 }$ is a strong predictor of the unseen label $y$ . We name the privacy loss (leak AUC) measured against the attack $r _ { n }$ the norm leak AUC. In Figure 2(c), the norm leak AUCs are consistently above 0.9, signaling a high level of label leakage throughout training. + +Attack 2: Direction-based scoring function. We now show that the direction of $g$ (in addition to its magnitude) can also leak the label. For a pair of examples, $( X _ { a } , y _ { a } ) , ( X _ { b } , y _ { b } )$ , let their respective predicted positive class probability be $\widetilde { p } _ { 1 , a } , \widetilde { p } _ { 1 , b }$ and their communicated gradients be $g _ { a } , \ g _ { b }$ . Let co $\mathbf { \tau } _ { \mathrm { 3 } } : \mathbb { R } ^ { d } \times \mathbf { \bar { R } } ^ { d } \mathbb { R }$ e e denote the cosine similarity function $\cos ( g _ { a } , g _ { b } ) = g _ { a } ^ { T } g _ { b } \bar { / } ( \| g _ { a } \| _ { 2 } \| g _ { b } \| _ { 2 } )$ . It is easy to see that $\begin{array} { r } { \cos \bigl ( g _ { a } , g _ { b } \bigr ) = \mathrm { s g n } \bigl ( \widetilde { p } _ { 1 , a } - y _ { a } \bigr ) \cdot \mathrm { s g n } \bigl ( \widetilde { p } _ { 1 , a } - y _ { b } \bigr ) \cdot \cos \bigl ( \nabla _ { z } h ( z ) \bigr | _ { z = f ( X _ { a } ) } , \nabla _ { z } h ( z ) \bigr | _ { z = f ( X _ { b } ) } \bigr ) } \end{array}$ , where $\operatorname { s g n } ( \mathbf { x } )$ e e is the sign function which returns 1 if $x \geq 0$ , and $- 1$ if $x < 0$ . We highlight two additional observations that can allow us to use cosine similarity to recover the label. + +• Observation 3: When the examples $a , b$ are of different classes, the term $\mathrm { s g n } ( \widetilde { p } _ { 1 , a } - y _ { a } ) ~ .$ · $\operatorname { s g n } ( \widetilde { p } _ { 1 , a } - y _ { b } ) = - 1$ is negative. On the other hand, when examples $a , b$ eare of the same eclass (both positive/both negative), this product will have a value of 1 and thus be positive. • Observation 4: Throughout training, for any two examples $a , b$ , their gradients of the function $h$ always form an acute angle, i.e. $\cos ( \nabla _ { z } h ( z ) | _ { z = \pmb { f } ( X _ { a } ) } , \nabla _ { z } h ( z ) | _ { z = \pmb { f } ( X _ { b } ) } ) > 0$ (Figure 2(d)). For neural networks that use monotonically increasing activation functions (such as ReLU, sigmoid, tanh), this is caused by the fact that the gradients of these activation functions with respect to its inputs are coordinatewise nonnegative and thus always lie in the first closed hyperorthant. + +Since $\cos ( g _ { a } , g _ { b } )$ is the product of the terms from Observation 3 and 4, we see that for a given example, all the examples that are of the same class result in a positive cosine similarity, while all opposite class examples result in a negative cosine similarity. If the problem is class-imbalanced and the non-label party knows there are fewer positive examples than negative ones, it can thus determine the label of each example: the class is negative if more than half of the examples result in positive cosine similarity; otherwise it is positive. For many practical applications, the non-label party may reasonably guess which class has more examples in the dataset a priori without ever seeing any data—for example, in disease prediction, the percentage of the entire population having a certain disease is almost always much lower than $50 \%$ ; in online advertising conversion prediction, the conversion rate (fraction of positive examples) is rarely higher than $30 \%$ . Note that the non-label party doesn’t need knowledge of the exact sample proportion of each class for this method to work. + +To simplify this attack for evaluation, we consider an even worse oracle scenario where the non-label party knows the clean gradient of one positive example $g _ { + }$ . Unlike the aforementioned practical majority counting attack which needs to first figure out the direction of one positive gradient, this oracle scenario assumes the non-label party is directly given this information. Thus, any protection method capable of defending this oracle attack would also protect against the more practical one. With $g _ { + }$ given, the direction-based scoring function $r _ { d }$ is simply $r _ { d } ( g ) = \cos ( g , g _ { + } )$ . We name the privacy loss (leak AUC) against this oracle attack $r _ { d }$ the cosine leak $A U C$ . In practice, we randomly choose a positive class clean gradient from each batch as $g _ { + }$ for evaluation. For iterations in Figure 2(e), the cosine leak AUC all have the highest value of 1 (complete label leakage). + +# 4 LABEL LEAKAGE PROTECTION METHODS + +In this section, we first introduce a heuristic random perturbation approach designed to prevent the practical attacks identified in Section 3.3. We then propose a theoretically justified method that aims to protect against the entire class of scoring functions considered in our threat model (Section 3.2). + +# 4.1 A HEURISTIC PROTECTION APPROACH + +Random perturbation and the isotropic Gaussian baseline. To protect against label leakage, the label party should ideally communicate essential information about the gradient without communicating its actual value. Random perturbation methods generally aim to achieve this goal. One obvious consideration for random perturbation is to keep the perturbed gradients unbiased. In other words, suppose $\tilde { g }$ is the perturbed version of an example’s true gradient $g$ , then we want $\mathbb { E } [ \tilde { g } \mid g ] = g$ . By chain rule and linearity of expectation, this ensures the computed gradients of the non-label party’s parameters $f$ will also be unbiased, a desirable property for stochastic optimization. Among unbiased perturbation methods, a simple approach is to add iid isotropic Gaussian noise to every gradient to mix the positive and negative gradient distribution before sending to the non-label party. + +Although isotropic Gaussian noise is a valid option, it may not be optimal because 1) the gradients are vectors but not scalars, so the structure of the noise covariance matrix matters. Isotropic noise might neglect the direction information; 2) due to the asymmetry of the positive and negative gradient distribution, the label party could add noise with different distributions to each class’s gradients. + +Norm-alignment heuristic. We now introduce an improved heuristic approach of adding zeromean Gaussian noise with non-isotropic and example-dependent covariance. [Magnitude choice] As we have seen that $\| g \| _ { 2 }$ can be different for positive and negative examples and thus leak label information, this heuristic first aims to make the norm of each perturbed gradient indistinguishable from one another. Specifically, we want to match the expected squared 2-norm of every perturbed gradient in a mini-batch to the largest squared 2-norm in this batch (denote by $\| g _ { \operatorname* { m a x } } \| _ { 2 } ^ { 2 } )$ . [Direction choice] In addition, as we have seen empirically from Figure 2(e), the positive and negative gradients lie close to a one-dimensional line in $\mathbb { R } ^ { d }$ , with positive examples pointing in one direction and negative examples in the other. Thus we consider only adding noise (roughly speaking) along “this line”. More concretely, for a gradient $g _ { j }$ in the batch, we add a zero-mean Gaussian noise vector $\eta _ { j }$ supported only on the one-dimensional space along the line of $g _ { j }$ . In other words, the noise’s covariance is the rank-1 matrix $\mathrm { C o v } [ \eta _ { j } ] = \sigma _ { j } ^ { 2 } g _ { j } g _ { j } ^ { T }$ . To calculate $\sigma _ { j }$ , we aim to match $\mathbb { E } [ \| g _ { j } + \eta _ { j } \| _ { 2 } ^ { 2 } ] =$ $\| g _ { \operatorname* { m a x } } \| _ { 2 } ^ { 2 }$ . A simple calculation gives $\sigma _ { j } = \sqrt { \| g _ { \mathrm { { m a x } } } \| _ { 2 } ^ { 2 } / \| g _ { j } \| _ { 2 } ^ { 2 } - 1 }$ . Since we align to the maximum norm, we name this heuristic protection method max norm. The advantage of max norm is that it has no parameter to tune. Unfortunately, it does not have a strong theoretical motivation, cannot flexibly trade-off between model utility and privacy, and may be broken by some unknown attacks. + +# 4.2 OPTIMIZED PERTURBATION METHOD: MA R V E L L + +Motivated by the above issues of max norm, we next study how to achieve a more principled trade-off between model performance (utility) and label protection (privacy). To do so, we directly minimize the worst-case adversarial scoring function’s leak AUC under a utility constraint. We name this protection method Marvell (optiMized perturbAtion to pReVEnt Label Leakage). + +Noise perturbation structure. Due to the distribution difference between the positive and negative class’s cut layer gradients, we consider having the label party additively perturb the randomly sampled positive $\overset { \cdot } { g } ^ { ( 1 ) }$ and negative $g ^ { ( 0 ) }$ gradients with independent zero-mean random noise vectors $\eta ^ { ( 1 ) }$ and $\eta ^ { ( 0 ) }$ with possibly different distributions (denoted by $D ^ { ( 1 ) }$ and ${ \cal D } ^ { ( 0 ) }$ ). We use $\widetilde { P } ^ { ( 1 ) }$ and $\widetilde { P } ^ { ( 0 ) }$ to denote the induced perturbed positive and negative gradient distributions. Our goal is to find the optimal noise distributions $D ^ { ( 1 ) }$ and ${ \cal D } ^ { ( 0 ) }$ by optimizing our privacy objective described below. + +Privacy protection optimization objective. As the adversarial non-label party in our threat model is allowed to use any measurable scoring function $r$ for label recovery, we aim to protect against all such scoring functions by minimizing the privacy loss of the worst case scoring function measured through our leak AUC metric. Formally, our optimization objective is $\begin{array} { r } { \operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \operatorname* { m a x } _ { r } \mathrm { A U C } ( r ) } \end{array}$ . Here to compute $\operatorname { A U C } ( r )$ , the $\mathrm { F P R } _ { r } ( t )$ and $\mathrm { T P R } _ { r } ( t )$ needs to be computed using the perturbed distributions $\widetilde { P } ^ { ( 1 ) }$ and $\widetilde { P } ^ { ( 0 ) }$ instead of the unperturbed $P ^ { ( 1 ) }$ and $P ^ { ( 0 ) }$ (Section 3.2). Since AUC is difficult to directly optimize, we consider optimizing an upper bound through the following theorem: + +Theorem 1. For $0 \leq \epsilon < 4$ and any perturbed gradient distributions $\widetilde { P } ^ { ( 1 ) }$ and $\widetilde { P } ^ { ( 0 ) }$ that are absolutely continuous with respect to each other, + +$$ +\begin{array} { r } { { \bf K L } ( \widetilde { P } ^ { ( 1 ) } \parallel \widetilde { P } ^ { ( 0 ) } ) + { \bf K L } ( \widetilde { P } ^ { ( 0 ) } \parallel \widetilde { P } ^ { ( 1 ) } ) \le \epsilon \quad i m p l i e s \quad \operatorname* { m a x } _ { r } { \bf A U C } ( r ) \le \frac { 1 } { 2 } + \frac { \sqrt { \epsilon } } { 2 } - \frac { \epsilon } { 8 } . } \end{array} +$$ + +From Theorem 1 (proof in Appendix A.3), we see that as long as the sum KL divergence is below 4,√ the smaller sumKL is, the smaller $\operatorname* { m a x } _ { r } \mathbf { A U C } ( r )$ is. $( 1 / 2 + \overline { { \sqrt { \epsilon } } } / 2 - \epsilon / 8 $ decreases as $\epsilon$ decreases.) Thus we can instead minimize the sum KL divergence between the perturbed gradient distributions: + +$$ +\begin{array} { r } { \operatorname { s u m K L } ^ { * } : = \operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \operatorname { K L } ( \widetilde { P } ^ { ( 1 ) } \parallel \widetilde { P } ^ { ( 0 ) } ) + \operatorname { K L } ( \widetilde { P } ^ { ( 0 ) } \parallel \widetilde { P } ^ { ( 1 ) } ) . } \end{array} +$$ + +Utility constraint. In an extreme case, we could add infinite noise to both the negative and positive gradients. This would minimize (1) optimally to 0 and make the worst case leak AUC 0.5, which is equivalent to a random guess. However, stochastic gradient descent cannot converge under infinitely large noise, so it is necessary to control the variance of the added noise. We thus introduce the noise power constraint: $p \cdot \mathrm { t r } ( \mathrm { C o v } [ \eta ^ { ( 1 ) } ] ) + ( 1 - p ) \cdot \mathrm { t r } ( \mathrm { C o v } [ \eta ^ { ( 0 ) } ] ) \le ~ P$ , where $p$ is the fraction of positive examples (already known to the label party); $\mathrm { t r } ( \mathrm { C o v } [ \eta ^ { ( i ) } ] )$ denotes the trace of the covariance matrix of the random noise $\eta ^ { ( i ) }$ ; and the upper bound $P$ is a tunable hyperparameter to control the level of noise: larger $P$ would achieve a lower sumKL and thus lower worst-case leak AUC and better privacy; however, it would also add more noise to the gradients, leading to slower optimization convergence and possibly worse model utility. We weight each class’s noise level $\mathrm { t r } ( \mathrm { C o v } [ \eta ^ { ( i ) } ] )$ by its example proportion $\dot { p }$ or $1 - p )$ since, from an optimization perspective, we want to equally control every training example’s gradient noise. The constrained optimization problem becomes: + +$$ +\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \mathrm { K L } ( \widetilde P ^ { ( 1 ) } \parallel \widetilde P ^ { ( 0 ) } ) + \mathrm { K L } ( \widetilde P ^ { ( 0 ) } \parallel \widetilde P ^ { ( 1 ) } ) \mathrm { s . t . } p \cdot \mathrm { t r } ( \mathrm { C o v } [ \eta ^ { ( 1 ) } ] ) + ( 1 - p ) \cdot \mathrm { t r } ( \mathrm { C o v } [ \eta ^ { ( 0 ) } ] ) \le P . +$$ + +Optimizing the objective in practice. To solve the optimization problem we first introduce some modelling assumptions. We assume that the unperturbed gradient of each class follows a Gaussian distribution: $g ^ { ( 1 ) } \stackrel { \cdot } { \sim } \mathcal { N } ( \bar { g } ^ { ( 1 ) } , v I _ { d \times d } )$ and $g ^ { ( 0 ) } \sim \dot { \mathcal { N } } ( \bar { g } ^ { ( 0 ) } , u I _ { d \times d } )$ . Despite this being an approximation, as we see later in Section 5, it can achieve strong protection quality against our identified attacks. In addition, it makes the optimization easier (see below) and provides us with insight on the optimal noise structure. We also search for perturbation distributions that are Gaussian: $D ^ { \top } \ d _ { 1 } ) = \mathcal { N } ( 0 , \dot { \Sigma _ { 1 } } )$ and $D ^ { ( 0 ) } = \mathcal { N } ( 0 , \Sigma _ { 0 } )$ with commuting covariance matrices: $\Sigma _ { 1 } \Sigma _ { 0 } = \Sigma _ { 0 } \Sigma _ { 1 }$ . The commutative requirement slightly restricts our search space but also makes the optimization problem more tractable. Our goal is to solve for the optimal noise structure, i.e. the positive semidefinite covariance matrices $\Sigma _ { 0 }$ , $\Sigma _ { 1 }$ . Let $\Delta g : = \bar { g } ^ { ( 1 ) } - \bar { g } ^ { ( 0 ) }$ denote the difference between the positive and negative gradient’s mean vectors. We now have the following theorem (proof and interpretation in Appendix A.4): + +Theorem 2. The optimal $\Sigma _ { 1 } ^ { * }$ and $\Sigma _ { 0 } ^ { * }$ to (2) with the above assumptions have the form: + +$$ +\Sigma _ { 1 } ^ { * } = \frac { \lambda _ { 1 } ^ { ( 1 ) * } - \lambda _ { 2 } ^ { ( 1 ) * } } { \| \Delta g \| _ { 2 } ^ { 2 } } ( \Delta g ) ( \Delta g ) ^ { \top } + \lambda _ { 2 } ^ { ( 1 ) * } I _ { d } , \quad \Sigma _ { 0 } ^ { * } = \frac { \lambda _ { 1 } ^ { ( 0 ) * } - \dot { \lambda } _ { 2 } ^ { ( 0 ) * } } { \| \Delta g \| _ { 2 } ^ { 2 } } ( \Delta g ) ( \Delta g ) ^ { \top } + \lambda _ { 2 } ^ { ( 0 ) * } I _ { d } , +$$ + +$( \lambda _ { 1 } ^ { ( 0 ) * } , \lambda _ { 2 } ^ { ( 0 ) * } , \lambda _ { 1 } ^ { ( 1 ) * } , \lambda _ { 2 } ^ { ( 1 ) * } )$ is the solution to the following 4-variable optimization problem: + +$$ +\mathop { \operatorname* { m i n } } _ { \lambda _ { 1 } ^ { ( 0 ) } , \lambda _ { 1 } ^ { ( 0 ) } , \lambda _ { 2 } ^ { ( 1 ) } } ( d - 1 ) \frac { \lambda _ { 2 } ^ { ( 0 ) } + u } { \lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \frac { \lambda _ { 2 } ^ { ( 1 ) } + v } { \lambda _ { 2 } ^ { ( 0 ) } + u } + \frac { \lambda _ { 1 } ^ { ( 0 ) } + u + \| \Delta g \| _ { 2 } ^ { 2 } } { \lambda _ { 1 } ^ { ( 1 ) } + v } + \frac { \lambda _ { 1 } ^ { ( 1 ) } + v + \| \Delta g \| _ { 2 } ^ { 2 } } { \lambda _ { 1 } ^ { ( 0 ) } + u } +$$ + +$$ +\begin{array} { r l } { s . t . \quad p \lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) \lambda _ { 2 } ^ { ( 0 ) } \leq { \cal P } , } & \\ { - \lambda _ { 1 } ^ { ( 1 ) } \leq 0 , \quad - \lambda _ { 1 } ^ { ( 0 ) } \leq 0 , \quad - \lambda _ { 2 } ^ { ( 1 ) } \leq 0 , \quad - \lambda _ { 2 } ^ { ( 0 ) } \leq 0 , \quad \lambda _ { 2 } ^ { ( 1 ) } - \lambda _ { 1 } ^ { ( 1 ) } \leq 0 , \quad \lambda _ { 2 } ^ { ( 0 ) } - \lambda _ { 1 } ^ { ( 0 ) } \leq 0 } \end{array} +$$ + +Additional details of Marvell. By Theorem 2, our optimization problem over two positive semidefinite matrices is reduced to a much simpler 4-variable optimization problem. We include a detailed description of how the constants in the problem are estimated in practice and what solver we use in a full description of the Marvell algorithm in Appendix A.5. Beyond optimization details, it is worth noting how to set the power constraint hyperparameter $P$ in Equation 2 in practice. As directly choosing $P$ requires knowledge of the scale of the gradients in the specific application and the scale could also shrink as the optimization converges, we instead express $P = s \| \Delta g \| _ { 2 } ^ { 2 }$ , and tune for a fixed hyperparameter $s > 0$ . This alleviates the need to know the scale of the gradients in advance, and the resulting value of $P$ can also dynamically change throughout training as the distance between the two gradient distributions’ mean $\| \Delta g \| _ { 2 }$ changes. + +# 5 EXPERIMENTS + +In this section, we first describe our experiment setup and then demonstrate the label protection quality of Marvell as well as its privacy-utility trade-off relative to baseline approaches. + +Empirical Setup. We use three real-world binary classification datasets for evaluation: Criteo and Avazu, two online advertising prediction datasets with millions of examples; and ISIC, a healthcare image dataset for skin cancer prediction. All three datasets exhibit severe label leakage problem without protection. (see Appendix A.6.1 on dataset and preprocessing details). We defer similar results on Avazu to Appendix A.7 and focus on Criteo and ISIC in this section. For Criteo, we train a Wide&Deep model (Cheng et al., 2016) where the non-label party owns the embedding layers for input features and the first three 128-unit ReLU activated MLP layers (first half of the deep part) while the label party owns the remaining layers of the deep part and the entire wide part of the model3. For ISIC, we train a model with 6 convolutional layers each with 64 channels followed by a 64-unit ReLU MLP layer, and the cut layer is after the fourth convolutional layer. In this case, an example’s cut layer feature $f ( X )$ and gradient $g$ are both in $\mathbb { R } ^ { 5 \times 5 \times 6 4 }$ . We treat such tensors as vectors in $\mathbb { R } ^ { 1 6 0 0 }$ to fit into our analysis framework (for additional model architecture and training details see Appendix A.6.2, A.6.3). + +![](images/da7562e01beb63c35fea76cbea790a0088e2c1811a01fbe1e54c87a5bcc7c358.jpg) +Figure 3: Norm and cosine leak AUC (computed every batch) at the cut layer and at the first layer under no protection vs. Marvell with different scale hyperparameter $s$ throughout the ISIC training. + +# 5.1 LABEL LEAKAGE AND MA R V E L L’S STRONG AND FLEXIBLE PROTECTION + +We first evaluate the protection quality of Marvell against the norm and cosine attacks discussed in Section 3.3. We also compare against the leakage metrics when no protection is applied (no noise). As the results across the three datasets are highly similar, we use ISIC as an example (other datasets see Appendix A.7.1). We see in Figure 3(a)(b) that unlike no noise where the label information is completely leaked (leak $\mathbf { A U C } \approx 1$ ) throughout training, Marvell achieves a flexible degree of protection (by varying $s$ ) against both the norm 2(a) and direction attacks 2(b) on the cut layer gradients and has strong protection (leak AUC $\approx { \bf 0 . 5 }$ ) at $s = 4 . 0$ . Additionally, it is natural to ask whether the gradients of layers before the cut layer (on the non-label party side) can also leak the labels as the non-label party keeps back propagating towards the first layer. In Figure 3(c)(d), we compute the leak AUC values when using the non-label party’s first layer activation gradient as inputs to the scoring functions to predict $y$ . Without protection, the first layer gradient still leaks the label very consistently. In constrast, Marvell still achieves strong privacy protection at the first layer $s = 4 . 0$ ) despite the protection being analyzed at the cut layer. + +# 5.2 PRIVACY-UTILITY TRADE-OFF COMPARISON + +After showing Marvell can provide strong privacy protection against our identified attacks, we now see how well it can preserve utility by comparing its privacy-utility tradeoff against other protection baselines: no noise, isotropic Gaussian (iso), and our proposed heuristic max norm. Similar to how we allow Marvell to use a power constraint to depend on the current iteration’s gradient distribution through $P = s \| \Delta g \| _ { 2 } ^ { 2 }$ , we also allow iso to have such type of dependence— specifically, we add $\eta \sim \mathcal { N } ( \mathbf { 0 } , ( t / d ) \cdot \| g _ { \operatorname* { m a x } } \| _ { 2 } ^ { 2 } I _ { d \times d } )$ to every gradient in a batch with $t$ a tunable privacy hyperparameter to be fixed throughout training. To trace out the complete tradeoff curve for Marvell and iso, we conduct more than 20 training runs for each protection method with a different value of privacy hyperparameter $s$ for Marvell, $t$ for iso) in each run on every dataset. (Note that no noise and max norm do not have privacy hyperparameters.) + +We present the tradeoffs between privacy (measured through norm and cosine leak AUC at cut layer/first layer) and utility (measured using test loss and test AUC) in Figure 4. To summarize the leak AUC over a given training run, we pick the $9 5 \%$ quantile over the batch-computed leak AUCs throughout all training iterations. This quantile is chosen instead of the mean because we want to measure the most-leaked iteration’s privacy leakage (highest leak AUC across iterations) to ensure the labels are not leaked at any points during training. $9 5 \%$ quantile is chosen instead of the max $( 1 0 0 \% )$ ) as we want this privacy leak estimate to be robust against randomness of the training process. + +Privacy-Utility Tradeoff comparison results. In measuring the privacy-utility tradeoff, we aim to find a method that consistently achieves a lower leak AUC (better privacy) for the same utility value. + +• [Marvell vs iso] As shown in Figure 4, Marvell almost always achieves a better tradeoff than iso against both of our proposed attacks at both the cut layer and the first layer on both the ISIC and Criteo datasets. It is important to note that although the utility constraint is in terms of training loss optimization, Marvell’s better tradeoff still translates to the generalization performance when the utility is measured through test loss or test AUC. Additionally, despite achieving reasonable (though still worse than Marvell) privacy-utility tradeoff against the norm-based attack, iso performs much worse against the direction-based attack: on ISIC, even after applying a significant amount of isotropic noise (with $t > 2 0$ ), iso’s cosine leak AUC is still higher than 0.9 at the cut layer (Figure 4(b,f)). In contrast, Marvell is effective against this direction-based attack with a much lower cosine leak AUC $\approx 0 . 6$ . + +![](images/48b9f068b52ed0c35fe5504d6d05601a466864114d83bc64d2db5cda6f53531e.jpg) +Figure 4: Privacy (norm & cosine leak AUC) vs Utility (test loss & test AUC) trade-off of protection methods (Marvell, iso, no noise, max norm) at the cut and first layer on ISIC and Criteo. + +• [max norm heuristic] Beyond Marvell, we see that our heuristic approach max norm can match and sometimes achieve even lower (Figure 4(a,f,i)) leak AUC value than Marvell at the same utility level. We believe this specifically results from our norm and direction consideration when designing this heuristic. However, without a tunable hyperparameter, max norm cannot tradeoff between privacy and utility. Additionally, unlike Marvell which is designed to protect against the entire class of adversarial scoring functions, max norm might still fail to protect against other future attack methods beyond those considered here. + +In summary, our principled method Marvell significantly outperforms the isotropic Gaussian baseline, and our proposed max norm heuristic can also work particularly well against the norm- and direction-based attacks which we identified in Section 3.3. + +# 6 CONCLUSION + +In this paper, we formulate a label leakage threat model in the two-party split learning binary classification problem through a novel privacy loss quantification metric (leak AUC). Within this threat model, we provide two simple yet effective attack methods that can accurately uncover the private labels of the label party. To counter such attacks, we propose a heuristic random perturbation method max norm as well as a theoretically principled method Marvell which searches for the optimal noise distributions to protect against the worst-case adversaries in the threat model. We have conducted extensive experiments to demonstrate the effectiveness of Marvell and max norm over the isotropic Gaussian perturbation baseline iso. + +Open questions and future work. Our work is the first we are aware of to identify, rigorously quantify, and protect against the threat of label leakage in split-learning, and opens up a number of worthy directions of future study. In particular, as the model parameters are updated every batch in our problem setup, the true gradient of an example and the gradient distribution would both change. An interesting question is whether the adversarial non-label party can remember the stale gradient of the same example from past updates (possibly separated by hundreds of updates steps) in order to recover the label information in the current iteration in a more complex threat model. It would also be interesting to build on our results to study whether there exist attack methods when the classification problem is multiclass instead of binary, and when the split learning scenario involves more than two parties with possibly more complicated training communication protocols (e.g., Vepakomma et al., 2018). + +Ethics Statement. In our paper, we have identified a realistic threat of label leakage in the twoparty split learning binary classification problem. We aim to raise awareness about potential privacy issues in this problem domain, where many industrial applications have been deployed. Beyond making such threats clear, we have taken the first steps towards protection—we have proposed both heuristic and principled methods that can preserve label privacy. We hope our work will pave the way for future analyses that make the two-party split learning framework more effective and secure. + +Reproducibility Statement. To make our paper reproducible, we provide: + +• Proofs of our Theorem 1 and Theorem 2 in Appendix A.3,A.4; +• Detailed experiment description including 1) data preprocessing, 2) model architecture, 3) training algorithm and hyperparameters in Appendix A.6. +• Source code with running instructions (in README.md) at https://github.com/OscarcarLi/label-protection. + +# REFERENCES + +Martin Abadi, Andy Chu, Ian Goodfellow, H Brendan McMahan, Ilya Mironov, Kunal Talwar, and Li Zhang. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, pp. 308–318, 2016. + +Avazu. Avazu click-through rate prediction, 2015. URL https://www.kaggle.com/c/ avazu-ctr-prediction/data. + +Keith Bonawitz, Vladimir Ivanov, Ben Kreuter, Antonio Marcedone, H Brendan McMahan, Sarvar Patel, Daniel Ramage, Aaron Segal, and Karn Seth. Practical secure aggregation for privacypreserving machine learning. In Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security, pp. 1175–1191, 2017. + +Kamalika Chaudhuri and Daniel Hsu. Sample complexity bounds for differentially private learning. In Proceedings of the 24th Annual Conference on Learning Theory, pp. 155–186. JMLR Workshop and Conference Proceedings, 2011. + +Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al. Wide & deep learning for recommender systems. In Proceedings of the 1st workshop on deep learning for recommender systems, pp. 7–10, 2016. + +Albert Cheu, Adam Smith, Jonathan Ullman, David Zeber, and Maxim Zhilyaev. Distributed differential privacy via shuffling. In Annual International Conference on the Theory and Applications of Cryptographic Techniques, pp. 375–403. Springer, 2019. + +Criteo. Criteo display advertising challenge, 2014. URL https://www.kaggle.com/c/ criteo-display-ad-challenge/data. + +Cynthia Dwork. Differential privacy. In Michele Bugliesi, Bart Preneel, Vladimiro Sassone, and Ingo Wegener (eds.), Automata, Languages and Programming, pp. 1–12, Berlin, Heidelberg, 2006. Springer Berlin Heidelberg. ISBN 978-3-540-35908-1. + +Cynthia Dwork, Aaron Roth, et al. The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science, 9(3-4):211–407, 2014. + +Ulfar Erlingsson, Vitaly Feldman, Ilya Mironov, Ananth Raghunathan, Kunal Talwar, and ´ Abhradeep Thakurta. Amplification by shuffling: From local to central differential privacy via anonymity. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algorithms, pp. 2468–2479. SIAM, 2019. + +Badih Ghazi, Noah Golowich, Ravi Kumar, Pasin Manurangsi, and Chiyuan Zhang. On deep learning with label differential privacy. arXiv preprint arXiv:2102.06062, 2021. + +Otkrist Gupta and Ramesh Raskar. Distributed learning of deep neural network over multiple agents. Journal of Network and Computer Applications, 116:1–8, 2018. + +ISIC. Siim-isic melanoma classification, 2020. URL https://www.kaggle.com/c/ siim-isic-melanoma-classification/data. +Vladimir Kolesnikov, Ranjit Kumaresan, Mike Rosulek, and Ni Trieu. Efficient batched oblivious prf with applications to private set intersection. In Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, CCS ’16, pp. 818–829, 2016. +Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. Communication-efficient learning of deep networks from decentralized data. In Artificial Intelligence and Statistics, pp. 1273–1282. PMLR, 2017. +H. Brendan McMahan, Daniel Ramage, Kunal Talwar, and Li Zhang. Learning differentially private recurrent language models. In International Conference on Learning Representations, 2018. URL https://openreview.net/forum?id=BJ0hF1Z0b. +Benny Pinkas, Thomas Schneider, and Michael Zohner. Scalable private set intersection based on ot extension. ACM Transactions on Privacy and Security (TOPS), 21(2):1–35, 2018. +Pramod Subramanyan, Rohit Sinha, Ilia Lebedev, Srinivas Devadas, and Sanjit A Seshia. A formal foundation for secure remote execution of enclaves. In Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security, pp. 2435–2450, 2017. +Praneeth Vepakomma, Otkrist Gupta, Tristan Swedish, and Ramesh Raskar. Split learning for health: Distributed deep learning without sharing raw patient data. arXiv preprint arXiv:1812.00564, 2018. +Praneeth Vepakomma, Otkrist Gupta, Abhimanyu Dubey, and Ramesh Raskar. Reducing leakage in distributed deep learning for sensitive health data. arXiv preprint arXiv:1812.00564, 2019. +Qiang Yang, Yang Liu, Tianjian Chen, and Yongxin Tong. Federated machine learning: Concept and applications. ACM Transactions on Intelligent Systems and Technology (TIST), 10(2):1–19, 2019. +Bo Zhao, Konda Reddy Mopuri, and Hakan Bilen. idlg: Improved deep leakage from gradients. arXiv preprint arXiv:2001.02610, 2020. +Ligeng Zhu, Zhijian Liu, and Song Han. Deep leakage from gradients. In Advances in Neural Information Processing Systems, pp. 14774–14784, 2019. + +# A APPENDIX + +# Appendix Outline + +A.1 Expressing $\operatorname { A U C } ( r )$ as an integral +A.2 Toy Class-balanced Example of Positive Example Prediction Lacking Confidence +A.3 Proof of Theorem 1 +A.4 Proof and Interpretation of Theorem 2 +A.5 Marvell Algorithm Description A.5.1 Marvell Time Complexity A.5.2 Marvell Empirical Run Time +A.6 Data Setup and Experimental Details A.6.1 Dataset Preprocessing A.6.2 Model Architecture Details A.6.3 Model Training Details +A.7 Complete Experimental Results A.7.1 AUC Progression for Avazu and Criteo A.7.2 Complete Privacy-Utility Tradeoffs A.7.3 Impact of Activation Gradients’ Layer Location on the Corresponding Leak AUC +A.8 Reasons why we don’t use local DP in our setup + +A.1 EXPRESSING $\operatorname { A U C } ( r )$ AS AN INTEGRAL + +Recall that the ROC curve of a scoring function $r : \mathbb { R } ^ { d } \mathbb { R }$ is a parametric curve $c _ { r } : \mathbb { R } [ 0 , 1 ] ^ { 2 }$ such that $c _ { r } ( t ) = ( \mathrm { F P R } _ { r } ( t ) , \mathrm { T P R } _ { r } ( t ) )$ , with + +$$ +\begin{array} { r l } & { \mathrm { F P R } _ { r } : \mathbb { R } [ 0 , 1 ] , \mathrm { s u c h t h a t } \mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } ( \{ g \in \mathbb { R } ^ { d } : r ( g ) > t \} ) } \\ & { \mathrm { T P R } _ { r } : \mathbb { R } [ 0 , 1 ] , \mathrm { s u c h t h a t } \mathrm { T P R } _ { r } ( t ) : = P ^ { ( 1 ) } ( \{ g \in \mathbb { R } ^ { d } : r ( g ) > t \} ) . } \end{array} +$$ + +We notice that $\mathrm { F P R } _ { r }$ and $\mathrm { T P R } _ { r }$ are both monotonically decreasing functions in $t$ with + +$$ +\begin{array} { r l r } & { \underset { t - \infty } { \operatorname* { l i m } } \mathrm { F P R } _ { r } ( t ) = 1 , \ \underset { t \infty } { \operatorname* { l i m } } \mathrm { F P R } _ { r } ( t ) = 0 } & \\ & { \underset { t - \infty } { \operatorname* { l i m } } \mathrm { T P R } _ { r } ( t ) = 1 , \ \underset { t \infty } { \operatorname* { l i m } } \mathrm { T P R } _ { r } ( t ) = 0 . } & \end{array} +$$ + +However, $\mathrm { F P R } _ { r }$ and $\mathrm { T P R } _ { r }$ does not need to be differentiable everywhere with respect to $t$ . Thus we cannot simply express the area under the curve using $\begin{array} { r } { \int _ { \infty } ^ { - \infty } \mathrm { T P R } _ { r } ( t ) \mathrm { F P R } _ { r } ^ { \prime } ( t ) d t } \end{array}$ . On the other hand, because both $\mathrm { F P R } _ { r }$ and $\mathrm { T P R } _ { r }$ are functions of bounded variations, we can express the area under the parametric curve $c _ { r }$ through the Riemann-Stieltjes integral which involves integrating the function $\mathrm { T P R } _ { r }$ with respect to the function $\mathrm { F P R } _ { r }$ : + +$$ +\operatorname { A U C } ( r ) = \int _ { t = \infty } ^ { t = - \infty } \operatorname { T P R } _ { r } ( t ) d \operatorname { F P R } _ { r } ( t ) +$$ + +Here the integration boundary is from $\infty$ to $- \infty$ in order to ensure the integral evaluates to a positive value. + +# A.2 TOY CLASS-BALANCED EXAMPLE OF POSITIVE EXAMPLE PREDICTION LACKINGCONFIDENCE + +As mentioned in Observation 1 of norm-based attack in Section 3.3, for many practical applications, there is inherently more ambiguity for the positive class than the negative class. To make this more concrete, let’s consider a simple toy example. + +Suppose the binary classification problem is over the real line. Here, $50 \%$ of the data is positive and lies uniformly in the interval [0, 1]. On the other hand, the remaining negative data (the other $50 \%$ of the total data) is a mixture distribution: $10 \%$ of the negative data also lies uniformly in [0, 1], while the rest $90 \%$ of the negative data lies uniformly in [1, 2]. + +$$ +\begin{array} { r l r l } & { p ( x \mid y = 1 ) = \mathbb { 1 } ( x \in [ 0 , 1 ] ) , } & & { \qquad p ( y = 1 ) = 0 . 5 . } \\ & { p ( x \mid y = 0 ) = 0 . 1 \cdot \mathbb { 1 } ( x \in [ 0 , 1 ] ) + 0 . 9 \cdot \mathbb { 1 } ( x \in [ 1 , 2 ] ) , } & & { \qquad p ( y = 0 ) = 0 . 5 . } \end{array} +$$ + +This setup mirrors our online advertising example where for all the users interested in the product (with feature $x \in [ 0 , 1 ]$ ), only a part of them would actually make the purchase after clicking on its advertisement. + +In this case, the best possible probabilistic classifiers $C ^ { \mathrm { { o p t } } }$ that can be ever learned, by Bayes Rule, would predict any example in $[ 0 , 1 ]$ to be of the positive class with probability $\frac { 1 0 } { 1 1 }$ , while it would predict any example in [1, 2] to be of the negative class with probability 1: + +$$ +C ^ { \mathrm { o p t } } ( y = 1 \mid x ) = { \left\{ \begin{array} { l l } { { \frac { 1 0 } { 1 1 } } } & { { \mathrm { ~ i f ~ } } x \in [ 0 , 1 ] } \\ { 0 } & { { \mathrm { ~ i f ~ } } x \in [ 1 , 2 ] } \end{array} \right. } . +$$ + +Thus even for this best possible classifier $C ^ { \mathrm { { o p t } } }$ , every positive example would have a confidence gap of $\scriptstyle { \frac { 1 } { 1 1 } }$ while $9 0 \%$ of the negative examples (the ones in [1, 2]) would have a confidence gap of 0. Hence we see that in this toy example, our empirical observation of the lack of prediction confidence for positive examples would still hold true. + +Besides, it is important to notice that this lack of positive prediction confidence phenomenon happens even in this class-balanced toy example $( p ( y = 1 ) = p ( y = 0 ) = 0 . 5 )$ . Thus, Observation 1 does not require the data distribution to be class-imbalanced to hold, which further demonstrates the generality of our norm-based attack. + +# A.3 PROOF OF THEOREM 1 + +(Theorem 1). For $0 \leq \epsilon < 4$ and any perturbed gradient distributions $\widetilde { P } ^ { ( 1 ) }$ and $\widetilde { P } ^ { ( 0 ) }$ that are absolutely continuous with respect to each other, + +$$ +\mathbf { K L } \big ( \widetilde { P } ^ { ( 1 ) } \mid \big | \widetilde { P } ^ { ( 0 ) } \big ) + \mathbf { K L } \big ( \widetilde { P } ^ { ( 0 ) } \mid \big | \widetilde { P } ^ { ( 1 ) } \big ) \leq \epsilon \quad i m p l i e s \quad \operatorname* { m a x } _ { r } \mathbf { A } \mathbf { U C } ( r ) \leq \frac { 1 } { 2 } + \frac { \sqrt { \epsilon } } { 2 } - \frac { \epsilon } { 8 } . +$$ + +Proof of Theorem 1. Combining Pinsker’s inequality with Jensen’s inequality, we can obtain an upper bound of total variation distance by the symmetrized KL divergence (sumKL) for a pair of distributions $( P , Q )$ that are absolutely continuous with respect to each other: + +$$ +\mathbf { T } \mathbf { V } ( P , Q ) \leq { \frac { 1 } { 2 } } ( { \sqrt { \mathbf { K L } ( P \parallel Q ) / 2 } } + { \sqrt { \mathbf { K L } ( Q \parallel P ) / 2 } } ) \leq { \frac { 1 } { 2 } } { \sqrt { \mathbf { K L } ( P \parallel Q ) + \mathbf { K L } ( Q \parallel P ) } } . +$$ + +By our assumption, this implies that $\begin{array} { r } { \mathrm { T V } ( \widetilde { P } ^ { ( 1 ) } , \widetilde { P } ^ { ( 0 ) } ) \le \frac { \sqrt { \epsilon } } { 2 } } \end{array}$ . By the equivalent definition of total variation distance $\begin{array} { r } { \mathrm { T V } ( \widetilde P ^ { ( 1 ) } , \widetilde P ^ { ( 0 ) } ) ) = \operatorname* { m a x } _ { A \subset \mathbb { R } ^ { d } } [ \widetilde P ^ { ( 1 ) } ( A ) - \widetilde P ^ { ( 0 ) } ( A ) ] } \end{array}$ , we know that for any $A \subset$ $\mathbb { R } ^ { d }$ , $\begin{array} { r } { \widetilde { P } ^ { ( 1 ) } ( A ) - \widetilde { P } ^ { ( 0 ) } ( A ) \le \frac { \sqrt { \epsilon } } { 2 } } \end{array}$ . For any scoring function $r$ and any threshold value $t$ , let $A =$ $\{ g : r ( g ) > t \}$ , then we have $\begin{array} { r } { \bar { \mathrm { T P R } } _ { r } ( t ) - \mathrm { F P R } _ { r } ( t ) = \widetilde { P } ^ { ( 1 ) } ( A ) - \widetilde { P } ^ { ( 0 ) } ( A ) \leq \frac { \sqrt { \epsilon } } { 2 } . } \end{array}$ . + +Therefore, the AUC of the scoring function $r$ can be upper bounded in the following way: + +$$ +\begin{array} { l } { \displaystyle \mathrm { A U C } ( { \boldsymbol { r } } ) = \int _ { \infty } ^ { - \infty } { \mathrm { T P R } _ { r } ( t ) ~ d \mathrm { F P R } _ { r } ( t ) } } \\ { \displaystyle \leq \int _ { \infty } ^ { - \infty } \operatorname* { m i n } \bigg ( \mathrm { F P R } _ { r } ( t ) + \frac { \sqrt { \epsilon } } { 2 } , ~ 1 \bigg ) ~ d \mathrm { F P R } _ { r } ( t ) , } \end{array} +$$ + +where in (6) we use the additional fact that $\mathrm { T P R } _ { r } ( t ) \leq 1$ for all $t \in \mathbb { R }$ . + +When $\epsilon \in \ [ 0 , 4 )$ , we have $1 ~ - ~ \frac { \sqrt { \epsilon } } { 2 } ~ \in ~ ( 0 , 1 ]$ . As $\mathrm { F P R } _ { r } ( t )$ is a monotonically nonincreasing function in $t$ with range in $[ 0 , 1 ]$ , the set $\begin{array} { r } { \left\{ t : \mathrm { F P R } _ { r } ( t ) \leq 1 - \frac { \sqrt { \epsilon } } { 2 } \right\} \ \neq \ \phi } \end{array}$ is not empty. Let $\begin{array} { r } { k : = \operatorname* { i n f } \left\{ t : \mathrm { F P R } _ { r } ( t ) \leq 1 - \frac { \sqrt { \epsilon } } { 2 } \right\} } \end{array}$ . Again by $\mathrm { F P R } _ { r } ( t )$ being a monotonically nonincreasing function in $t$ , we can break the integration in Equation (6) into two terms: + +$$ +\begin{array} { r l } & { \quad \mathrm { A U C } ( r ) } \\ & { = \displaystyle \int _ { - \infty } ^ { k } \operatorname* { m i n } \left( \mathrm { F P R } _ { \mathrm { v } } ( t ) + \frac { \sqrt { \epsilon } } { 2 } , \mathbf { \Phi } _ { 1 } \right) d \mathrm { F P R } _ { v } ( t ) + \displaystyle \int _ { k } ^ { - \infty } \operatorname* { m i n } \left( \mathrm { F P R } _ { v } ( t ) + \frac { \sqrt { \epsilon } } { 2 } , \mathbf { \Phi } _ { 1 } \right) d \mathrm { F P R } _ { v } ( t ) } \\ & { \le \displaystyle \int _ { \infty } ^ { k } \left( \mathrm { F P R } _ { v } ( t ) + \frac { \sqrt { \epsilon } } { 2 } \right) d \mathrm { F P R } _ { v } ( t ) + \displaystyle \int _ { k } ^ { - \infty } 1 d \mathrm { F P R } _ { v } ( t ) } \\ & { = \left[ \frac { \left[ \mathrm { F P R } _ { r } ( v ) \right] ^ { 2 } } { 2 } + \frac { \sqrt { \epsilon } } { 2 } \mathrm { F P R } _ { v } ( v ) \right] \displaystyle _ { t = \infty } ^ { t - k } + \mathrm { F P R } _ { v } ( t ) \displaystyle _ { t = k } ^ { t - \infty } } \\ & { \le \frac { ( 1 - \frac { v \epsilon } { 2 } ) ^ { 2 } } { 2 } + \frac { \sqrt { \epsilon } } { 2 } \left( 1 - \frac { \sqrt { \epsilon } } { 2 } \right) + \left[ 1 - \left( 1 - \frac { \sqrt { \epsilon } } { 2 } \right) \right] } \\ & { = \frac { 1 } { 2 } + \frac { \sqrt { \epsilon } } { 2 } - \frac { \epsilon } { 8 } } \end{array} +$$ + +Since this inequality is true for any scoring function $r$ , it is true for the maximum value. Thus the proof is complete. □ + +# A.4 PROOF AND INTERPRETATION OF THEOREM 2 + +(Theorem 2). The optimal $\Sigma _ { 1 } ^ { * }$ and $\Sigma _ { 0 } ^ { * }$ to the following problem + +$\operatorname* { m i n } _ { \Sigma _ { 0 } , \Sigma _ { 1 } \in \mathbb S } K L ( \mathcal { N } ( \bar { g } ^ { ( 1 ) } , v I + \Sigma _ { 1 } ) \parallel \mathcal { N } ( \bar { g } ^ { ( 0 ) } , u I + \Sigma _ { 0 } ) ) + K L ( \mathcal { N } ( \bar { g } ^ { ( 0 ) } , u I + \Sigma _ { 0 } ) \parallel \mathcal { N } ( \bar { g } ^ { ( 1 ) } , v I + \Sigma _ { 1 } ) )$ subject to + +$$ +\begin{array} { c } { \Sigma _ { 0 } \Sigma _ { 1 } = \Sigma _ { 1 } \Sigma _ { 0 } , } \\ { p \cdot \operatorname { t r } ( \Sigma _ { 1 } ) + ( 1 - p ) \cdot \operatorname { t r } ( \Sigma _ { 0 } ) \leq P , } \\ { \Sigma _ { 1 } \succeq \mathbf { 0 } , } \\ { \Sigma _ { 0 } \succeq \mathbf { 0 } . } \end{array} +$$ + +have the form: + +$$ +\Sigma _ { 1 } ^ { * } = \frac { \tilde { \lambda } _ { 1 } ^ { ( 1 ) * } - \lambda _ { 2 } ^ { ( 1 ) * } } { \| \Delta g \| _ { 2 } ^ { 2 } } ( \Delta g ) ( \Delta g ) ^ { \top } + \lambda _ { 2 } ^ { ( 1 ) * } I _ { d } , \quad \Sigma _ { 0 } ^ { * } = \frac { \lambda _ { 1 } ^ { ( 0 ) * } - \lambda _ { 2 } ^ { ( 0 ) * } } { \| \Delta g \| _ { 2 } ^ { 2 } } ( \Delta g ) ( \Delta g ) ^ { \top } + \lambda _ { 2 } ^ { ( 0 ) * } I _ { d } , +$$ + +where $( \lambda _ { 1 } ^ { ( 0 ) * } , \lambda _ { 2 } ^ { ( 0 ) * } , \lambda _ { 1 } ^ { ( 1 ) * } , \lambda _ { 2 } ^ { ( 1 ) * } )$ is the solution to the following 4-variable optimization problem: + +$$ +\begin{array} { r l } & { \underset { \lambda _ { 1 } ^ { ( 0 ) } , \lambda _ { 1 } ^ { ( 0 ) } , \lambda _ { 2 } ^ { ( 1 ) } } { \operatorname* { m i n } } ( d - 1 ) \frac { \lambda _ { 2 } ^ { ( 0 ) } + u } { \lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \frac { \lambda _ { 2 } ^ { ( 1 ) } + v } { \lambda _ { 2 } ^ { ( 0 ) } + u } + \frac { \lambda _ { 1 } ^ { ( 0 ) } + u + \| \Delta g \| _ { 2 } ^ { 2 } } { \lambda _ { 1 } ^ { ( 1 ) } + v } + \frac { \lambda _ { 1 } ^ { ( 1 ) } + v + \| \Delta g \| _ { 2 } ^ { 2 } } { \lambda _ { 1 } ^ { ( 0 ) } + u } } \\ & { \mathrm { ~ s . t . ~ } p \lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) \lambda _ { 2 } ^ { ( 0 ) } \leq P , } \\ & { \mathrm { ~ \ ~ - ~ } \lambda _ { 1 } ^ { ( 1 ) } \leq 0 , \mathrm { ~ \ ~ - ~ } \lambda _ { 1 } ^ { ( 0 ) } \leq ~ 0 , \mathrm { ~ \ ~ - ~ } \lambda _ { 2 } ^ { ( 1 ) } \leq ~ 0 , \mathrm { ~ \ ~ - ~ } \lambda _ { 2 } ^ { ( 0 ) } \leq ~ 0 , } \\ & { \quad \quad \quad \lambda _ { 2 } ^ { ( 1 ) } - \lambda _ { 1 } ^ { ( 1 ) } \leq ~ 0 , \lambda _ { 2 } ^ { ( 0 ) } - \lambda _ { 1 } ^ { ( 0 ) } \leq 0 } \end{array} +$$ + +Proof of Theorem 2. By writing out the analytical close-form of the KL divergence between two Gaussian distributions, the optimization can be written as: + +$$ +\begin{array} { c } { { \displaystyle \operatorname* { m i n } _ { \Sigma _ { 0 } , \Sigma _ { 1 } \in \mathbb S } \mathrm { t r } \big ( ( \Sigma _ { 1 } + v I ) ^ { - 1 } ( \Sigma _ { 0 } + u I ) \big ) + \mathrm { t r } \big ( ( \Sigma _ { 0 } + u I ) ^ { - 1 } ( \Sigma _ { 1 } + v I ) \big ) + } } \\ { { \displaystyle \big ( \bar { g } ^ { ( 1 ) } - \bar { g } ^ { ( 0 ) } \big ) ^ { \top } \left( ( \Sigma _ { 1 } + v I ) ^ { - 1 } + ( \Sigma _ { 0 } + u I ) ^ { - 1 } \right) \left( \bar { g } ^ { ( 1 ) } - \bar { g } ^ { ( 0 ) } \right) } } \end{array} +$$ + +subject to + +$$ +\begin{array} { c } { \Sigma _ { 0 } \Sigma _ { 1 } = \Sigma _ { 1 } \Sigma _ { 0 } , } \\ { p \cdot \operatorname { t r } ( \Sigma _ { 1 } ) + ( 1 - p ) \cdot \operatorname { t r } ( \Sigma _ { 0 } ) \leq P , } \\ { \Sigma _ { 1 } \succeq \mathbf { 0 } , } \\ { \Sigma _ { 0 } \succeq \mathbf { 0 } . } \end{array} +$$ + +By the commutative constraint on the two positive semidefinite matrices $\Sigma _ { 1 }$ and $\Sigma _ { 0 }$ , we know that we can factor these two matrices using the same set of eigenvectors. We thus write: + +$$ +\begin{array} { r } { \Sigma _ { 0 } = \boldsymbol { Q } ^ { \top } \mathrm { d i a g } ( \lambda _ { 1 } ^ { ( 0 ) } , \ldots , \lambda _ { d } ^ { ( 0 ) } ) \boldsymbol { Q } , } \\ { \Sigma _ { 1 } = \boldsymbol { Q } ^ { \top } \mathrm { d i a g } ( \lambda _ { 1 } ^ { ( 1 ) } , \ldots , \lambda _ { d } ^ { ( 1 ) } ) \boldsymbol { Q } , } \end{array} +$$ + +where $Q \in \mathbb { R } ^ { d \times d }$ is an orthogonal matrix and the eigenvalues $\lambda _ { i } ^ { ( 0 ) } , \lambda _ { i } ^ { ( 1 ) }$ are nonnegative and decreasing in value. + +Using this alternative expression of $\Sigma _ { 1 }$ and $\Sigma _ { 0 }$ , we can express the optimization in terms of $\{ \lambda _ { i } ^ { ( 1 ) } \} , \{ \lambda _ { i } ^ { ( 0 ) } \} , Q$ : + +$$ +\begin{array} { r l } { \underset { \{ \lambda _ { i } ^ { ( 1 ) } \} , \{ \lambda _ { i } ^ { ( 0 ) } \} , Q } { \operatorname* { m i n } } } & { \displaystyle \sum _ { i = 1 } ^ { d } \frac { \lambda _ { i } ^ { ( 0 ) } + u } { \lambda _ { i } ^ { ( 1 ) } + v } + \sum _ { i = 1 } ^ { d } \frac { \lambda _ { i } ^ { ( 1 ) } + v } { \lambda _ { i } ^ { ( 0 ) } + u } + } \\ & { \displaystyle \left[ Q ( \bar { g } ^ { ( 1 ) } - \bar { g } ^ { ( 0 ) } ) \right] ^ { \top } \mathrm { d i a g } \left( \cdots , \frac { 1 } { \lambda _ { i } ^ { ( 0 ) } + u } + \frac { 1 } { \lambda _ { i } ^ { ( 1 ) } + v } , \ldots \right) Q ( \bar { g } ^ { ( 1 ) } - \bar { g } ^ { ( 0 ) } ) } \end{array} +$$ + +$$ +\begin{array} { r l } { \mathrm { s u b j e c t ~ t o ~ } } & { p ( \displaystyle \sum _ { i = 1 } ^ { d } \lambda _ { i } ^ { ( 1 ) } ) + ( 1 - p ) ( \displaystyle \sum _ { i = 1 } ^ { d } \lambda _ { i } ^ { ( 0 ) } ) \le P } \\ & { - \lambda _ { i } ^ { ( 1 ) } \le 0 , \forall i \in [ d ] } \\ & { - \lambda _ { i } ^ { ( 0 ) } \le 0 , \forall i \in [ d ] . } \\ & { \lambda _ { i } ^ { ( 1 ) } \ge \lambda _ { j } ^ { ( 1 ) } , \forall i < j . } \\ & { \lambda _ { i } ^ { ( 0 ) } \ge \lambda _ { j } ^ { ( 0 ) } , \forall i < j . } \\ & { Q \ \mathrm { o r t h o g o n a l } . } \end{array} +$$ + +For any fixed feasible $\{ \lambda _ { i } ^ { ( 1 ) } \} , \{ \lambda _ { i } ^ { ( 0 ) } \}$ , we see that the corresponding minimizing $Q$ will set its first row to be the unit vector in the direction of $\Delta g$ . Thus by first minimizing $Q$ , the optimization objective reduces to: + +$$ +\begin{array} { c c l } { \displaystyle \sum _ { i = 1 } ^ { d } \frac { \lambda _ { i } ^ { ( 0 ) } + u } { \lambda _ { i } ^ { ( 1 ) } + v } + \displaystyle \sum _ { i = 1 } ^ { d } \frac { \lambda _ { i } ^ { ( 1 ) } + v } { \lambda _ { i } ^ { ( 0 ) } + u } + \frac { g } { \lambda _ { 1 } ^ { ( 0 ) } + u } + \frac { g } { \lambda _ { 1 } ^ { ( 1 ) } + v } } \\ { = \displaystyle \sum _ { i = 2 } ^ { d } \frac { \lambda _ { i } ^ { ( 0 ) } + u } { \lambda _ { i } ^ { ( 1 ) } + v } + \displaystyle \sum _ { i = 2 } ^ { d } \frac { \lambda _ { i } ^ { ( 1 ) } + v } { \lambda _ { i } ^ { ( 0 ) } + u } + \frac { \lambda _ { 1 } ^ { ( 1 ) } + v + \| \Delta g \| _ { 2 } ^ { 2 } } { \lambda _ { 1 } ^ { ( 0 ) } + u } + \frac { \lambda _ { 1 } ^ { ( 0 ) } + u + \| \Delta g \| _ { 2 } ^ { 2 } } { \lambda _ { 1 } ^ { ( 1 ) } + v } } \end{array} +$$ + +We see that for the pair of variable $( \lambda _ { i } ^ { ( 1 ) } , \lambda _ { i } ^ { ( 0 ) } )$ $( i ~ \geq ~ 2 )$ , the function $\begin{array} { r } { \frac { \lambda _ { i } ^ { ( 0 ) } + u } { \lambda _ { i } ^ { ( 1 ) } + v } + \frac { \lambda _ { i } ^ { ( 1 ) } + v } { \lambda _ { 1 } ^ { ( 0 ) } + u } } \end{array}$ is strictly convex over the line segment $p \lambda _ { i } ^ { ( 1 ) } + ( 1 - p ) \lambda _ { i } ^ { ( 0 ) } = c$ for any nonnegative $c$ and attains the the minimum value at $\lambda _ { i } ^ { ( 1 ) } = 0$ when $u \ : < \ : v$ and $\lambda _ { i } ^ { ( 0 ) } = 0$ when $u \geq v$ . Suppose without loss of generality $u \geq v$ , then for the optimal solution we must have $\lambda _ { i } ^ { ( 0 ) } = 0$ for all $i \geq 2$ . Under this Thus for all condition, we notice that the function m(x) = ux+v $\left\{ \lambda _ { i } ^ { \left( 1 \right) } \right\}$ that satisfies $\Sigma _ { i = 2 } ^ { d } \lambda _ { i } ^ { ( 1 ) } = c$ is strictly c nonnegative $c$ vex on the positive reals., by Jensen inequality, we have + +$$ +\begin{array} { l } { \displaystyle \frac { 1 } { d - 1 } \sum _ { i = 2 } ^ { d } \left( \frac { u } { \lambda _ { i } ^ { ( 1 ) } + v } + \frac { \lambda _ { i } ^ { ( 1 ) } + v } { u } \right) } \\ { \displaystyle = \frac { 1 } { d - 1 } \sum _ { i = 2 } ^ { d } m ( \lambda _ { i } ^ { ( 1 ) } ) } \\ { \displaystyle \geq m \left( \frac { 1 } { d - 1 } \sum _ { i = 2 } ^ { d } \lambda _ { i } ^ { ( 1 ) } \right) } \\ { \displaystyle = m \left( \frac { c } { d - 1 } \right) . } \end{array} +$$ + +From this, we see that the optimal solution’s variables $\{ \lambda _ { i } ^ { ( 1 ) } \}$ must take on the same value $\scriptstyle { \big ( } { \frac { c } { d - 1 } } { \big ) }$ for all $i \geq 2$ . The case when $u \leq v$ is similar. As a result, we have proved that at the optimal solution, we must have: + +$$ +\lambda _ { i } ^ { ( 0 ) } = \lambda _ { j } ^ { ( 0 ) } \mathrm { a n d } \lambda _ { i } ^ { ( 1 ) } = \lambda _ { j } ^ { ( 1 ) } , \mathrm { f o r } i , j \ge 2 +$$ + +Hence, the optimization problem over the 2d variables nλ(1)i odi=1 $\left\{ \lambda _ { i } ^ { ( 1 ) } \right\} _ { i = 1 } ^ { d } \bigcup \left\{ \lambda _ { i } ^ { ( 0 ) } \right\} _ { i = 1 } ^ { d }$ can be reduced to an optimization problem over the four variables $\left\{ \lambda _ { 1 } ^ { ( 1 ) } , \lambda _ { 2 } ^ { ( 1 ) } , \lambda _ { 1 } ^ { ( 0 ) } , \lambda _ { 2 } ^ { ( 0 ) } \right\}$ : + +$$ +\operatorname* { m i n } _ { \lambda _ { 1 } ^ { ( 0 ) } , \lambda _ { 1 } ^ { ( 1 ) } , \lambda _ { 2 } ^ { ( 0 ) } , \lambda _ { 2 } ^ { ( 1 ) } } \left( d - 1 \right) \frac { \lambda _ { 2 } ^ { ( 0 ) } + u } { \lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \frac { \lambda _ { 2 } ^ { ( 1 ) } + v } { \lambda _ { 2 } ^ { ( 0 ) } + u } + \frac { \lambda _ { 1 } ^ { ( 0 ) } + u + \| \Delta g \| _ { 2 } ^ { 2 } } { \lambda _ { 1 } ^ { ( 1 ) } + v } + \frac { \lambda _ { 1 } ^ { ( 1 ) } + v + \| \Delta g \| _ { 2 } ^ { 2 } } { \lambda _ { 1 } ^ { ( 0 ) } + u } +$$ + +$$ +\begin{array} { r l } { \mathrm { s u b j e c t ~ t o ~ } } & { \ p \lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) ( \lambda _ { 2 } ^ { ( 0 ) } ) \leq P } \\ & { \quad \quad \quad - \lambda _ { 1 } ^ { ( 1 ) } \leq 0 } \\ & { \quad \quad \quad - \lambda _ { 0 } ^ { ( 0 ) } \leq 0 } \\ & { \quad \quad \quad - \lambda _ { 2 } ^ { ( 1 ) } \leq 0 } \\ & { \quad \quad \quad - \lambda _ { 2 } ^ { ( 0 ) } \leq 0 } \\ & { \quad \quad \quad \lambda _ { 2 } ^ { ( 1 ) } - \lambda _ { 1 } ^ { ( 1 ) } \leq 0 } \\ & { \quad \quad \quad \lambda _ { 2 } ^ { ( 0 ) } - \lambda _ { 1 } ^ { ( 1 ) } \leq 0 } \\ & { \quad \quad \quad \lambda _ { 2 } ^ { ( 0 ) } - \lambda _ { 1 } ^ { ( 0 ) } \leq 0 . } \end{array} +$$ + +Given the optimal solution to the above 4-variable problem $( \lambda _ { 1 } ^ { ( 0 ) * } , \lambda _ { 2 } ^ { ( 0 ) * } , \lambda _ { 1 } ^ { ( 1 ) * } , \lambda _ { 2 } ^ { ( 1 ) * } )$ , we can set $Q$ to be any orthogonal matrix whose first row is the vector k∆gk2 . Plugging this back into the expression of $\Sigma _ { 1 }$ and $\Sigma _ { 0 }$ in Equation (14) gives us the final result. + +Thus the proof is complete. + +Remark (Interpreting the optimal $\Sigma _ { 1 } ^ { * }$ and $\Sigma _ { 0 } ^ { * }$ ). From the form of the optimal solution in (12), we see that the optimal covariance matrices are both linear combinations of two terms: a rank one matrix $( \Delta g ) ( \dot { \Delta } g ) ^ { \top }$ and the identity matrix $I _ { d }$ . Because a zero-mean Gaussian random vector with convariance matrix $( A + B )$ can be constructed as the sum of two independent zero-mean Gaussian random vectors with covariance matrices $A$ and $B$ respectively, we see that the optimal additive noise random variables $\eta ^ { ( 1 ) }$ and $\eta ^ { ( 0 ) }$ each consist of two independent components: one random component lies along the line that connects the positive and negative gradient mean vectors (whose covariance matrix is proportional to $\Delta g \Delta g ^ { \top } )$ ); the other component is sampled from an isotropic Gaussian. The use of the first random directional component and the fact that the isotropic Gaussian component have different variance scaling for the positive and negative class clearly distinguishes Marvell from the isotropic Gaussian baseline iso. + +Remark (How to solve). By analyzing the KKT condition of this four variable problem we can find that the optimal solution must exactly lie on the hyperplane $p \lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \bar { \lambda _ { 2 } ^ { ( 1 ) } } + ( 1 - p ) \lambda _ { 1 } ^ { ( 0 ) } +$ $( 1 - p ) ( d - 1 ) \lambda _ { 2 } ^ { ( 0 ) } = P$ . From the proof above we additionally know that $\lambda _ { 2 } ^ { ( 1 ) * } = 0$ if $u < v$ and $\lambda _ { 2 } ^ { ( 0 ) * } = 0$ if $u \geq v$ . Thus the problem is further reduced to a 3-variable problem. If we consider keeping one of the 3 remaining variables’ values fixed, then the feasible region becomes restricted to a line segment. We can simply perform a line search optimization of the convex objective and find the optimal values for the remaining two free variables. We can then alternate over which one of the three variables to fix and optimize over the other two. This optimization procedure would eventually converge and give us the optimal solution. This approach mimics the Sequential Minimal Optimization technique used to solve the dual of the support vector machine (SVM). + +# A.5 MA R V E L L ALGORITHM DESCRIPTION + +We use $\mathbf { 1 } \in \mathbb { R } ^ { d }$ to denote the vector with 1 in each coordinate. We use $g [ j ]$ to denote the $j$ -th row of the matrix $g$ and $y [ j ]$ to denote the $j$ -th coordinate of the vector $y$ . + +Algorithm 1: Marvell algorithm input : g ∈ RB×d, a size- $. B$ batch of unperturbed gradients $y \in \left\{ 0 , 1 \right\} ^ { B }$ , the label for each example in the batch $s$ , privacy hyperparameter for the power constraint $P$ output: g˜ ∈ RB×d, batch of perturbed gradients // Step 1: estimate the optimization constants from the batch gradients using maximum likelihood estimation (MLE) +1 $p \gets \frac { \mathbf { 1 } ^ { T } y } { d }$ 1 T y ; /\* positive fraction \*/ +2 g¯(1) $\begin{array} { r } { \bar { g } ^ { ( 1 ) } \frac { 1 } { \mathbf { 1 } ^ { T } y } g ^ { T } y } \end{array}$ ; $/ \star$ positive mean \*/ +3 $\begin{array} { r } { \bar { g } ^ { ( 0 ) } \frac { 1 } { B - \mathbf { 1 } ^ { T } y } g ^ { T } ( \mathbf { 1 } - y ) } \end{array}$ ; /\* negative mean \*/ +4 $\Delta g = \bar { g } ^ { ( 1 ) } - \bar { g } ^ { ( 0 ) }$ ; +5 $\begin{array} { r } { v \frac { 1 } { d \cdot ( \mathbf { 1 } ^ { T } y ) } \sum _ { j = 1 } ^ { B } y [ j ] \cdot g [ j ] - \bar { g } ^ { ( 1 ) } _ { 2 } ^ { 2 } } \end{array}$ ; /\* positive convariance \*/ +6 $\begin{array} { r } { u \frac { 1 } { d \cdot ( B - \mathbf { 1 } ^ { T } y ) } \sum _ { j = 1 } ^ { B } ( 1 - y [ j ] ) \cdot g [ j ] - \bar { g } ^ { ( 0 ) } _ { 2 } ^ { 2 } } \end{array}$ ; /\* negative convariance $\star /$ +7 $P \gets s \cdot \| \Delta g \| _ { 2 } ^ { 2 }$ ; /\* power constraint hyperparameter $\star /$ // Step 2: optimize the four-variable problem +8 if $u < v$ then +9 $\lambda _ { 2 } ^ { ( 1 ) } 0$ ; /\* this variable is optimal at $0 ~ \star /$ +10 Randomly initialize the optimization variables $\lambda _ { 1 } ^ { ( 1 ) }$ , $\lambda _ { 1 } ^ { ( 0 ) }$ , $\lambda _ { 2 } ^ { ( 0 ) }$ in the feasible region in (15). +11 else +12 13 $\lambda _ { 2 } ^ { ( 0 ) } 0 \ : ;$ initialize the optimization varia $/ \star$ le is optimal at 0 in the feasible region in (1 $\star /$ $\lambda _ { 1 } ^ { ( 1 ) } , \lambda _ { 2 } ^ { ( 1 ) } , \lambda _ { 1 } ^ { ( 0 ) }$ +14 end +15 while not converged do +16 Fix one of the newly updated optimization variables; // The 4-variable optimal solution lies on a hyperplane in $\mathbb { R } ^ { 4 }$ (see Appendix A.4 Remark) so fixing two variables gives us a line-segment +17 Update the remaining two optimization variables by performing 1-d line-search minimization of the convex function (15) while satisfying the constraints; + +# 18 end + +// Step 3: compute the optimal covariance matrices +19 $\begin{array} { r } { \Sigma _ { 1 } ^ { * } \frac { \lambda _ { 1 } ^ { ( 1 ) } - \lambda _ { 2 } ^ { ( 1 ) } } { \| \Delta g \| _ { 2 } ^ { 2 } } ( \Delta g ) ( \Delta g ) ^ { \top } + \lambda _ { 2 } ^ { ( 1 ) } I _ { d } ; } \end{array}$ ; +20 $\begin{array} { r } { \Sigma _ { 0 } ^ { * } \frac { \lambda _ { 1 } ^ { ( 0 ) } - \lambda _ { 2 } ^ { ( 0 ) } } { \| \Delta g \| _ { 2 } ^ { 2 } } ( \Delta g ) ( \Delta g ) ^ { \top } + \lambda _ { 2 } ^ { ( 0 ) } I _ { d } ; } \end{array}$ ; +// Step 4: perturb the gradients +21 $\tilde { g } \mathbf { 0 } _ { B \times d }$ ; $/ \star$ an empty matrix to store the perturbed gradients \*/ +22 for $j 1$ to $B$ do +23 if $y [ j ] = 1$ then +24 $\tilde { g } [ j ] g [ j ] + \eta ^ { ( 1 ) }$ , where $\boldsymbol \eta ^ { ( 1 ) } \sim \mathcal { N } ( \mathbf 0 , \Sigma _ { 1 } ^ { * } )$ ; +25 else $\begin{array} { r l } { \langle \mathscr { I } / } & { { } y [ j ] = 0 } \end{array}$ +26 $\begin{array} { r l } { \vert } & { { } \tilde { g } [ j ] g [ j ] + \eta ^ { ( 0 ) } } \end{array}$ , where $\boldsymbol \eta ^ { ( 0 ) } \sim \mathcal { N } ( \mathbf 0 , \Sigma _ { 0 } ^ { * } )$ ; +27 end +28 end + +A.5.1 MA R V E L L TIME COMPLEXITY + +The Marvell algorithm in 1 consists of the following steps: + +Step 1. Compute the positive and negative gradient mean and their difference (line 1 - 4). This amounts to averaging over at most $B$ numbers over each of the d gradient dimensions. Thus would have a time complexity of $O ( B d )$ . +Step 2. Compute the positive and negative covariance constants $u$ and $v$ (line 5, 6). This operation also takes $O ( \bar { B } d )$ as it averages over squares of coordinate differences. +Step 3. Use the results from step 1 and 2, solve the 4-variable optimization problem (line 7-18). Here because the constant size of the number of optimization variables. Solving this problem up to a fixed precision takes constant time $O ( 1 )$ . +Step 4. Perform the actual random perturbation (line $1 9 - 2 8 $ ). Because of the additive structure of each class’s covariance matrix, for every example’s gradient in the batch, we can independently sample one random Gaussian vector with rank-1 covariance and also another spherical Gaussian random vector and add both vectors to this gradient for perturbation. This step would take $O ( d )$ for each example and thus takes $O ( B d )$ in total. + +As a result, the entire Marvell algorithm has a time complexity of $O ( B d )$ . This can be further sped up through parallel computation using multi-threading/multi-core. Considering backpropagation through the cut layer would also require $O ( B d )$ time complexity, the Marvell algorithm would not slow down the split-learning process in any significant way at all. + +# A.5.2 MA R V E L L EMPIRICAL RUN TIME + +Empirically, in Table 1, we present the average amount of time it takes to run Marvell with the privacy hyperparameter value $s = 4$ for the three models (one for each dataset) we considered in our experiments (s value is chosen as it achieves good privacy protection as shown with purple line in Figure 3, 5, 6). We also compare it to the average time it takes to run one update iteration. We additionally include the $9 5 \%$ confidence interval for the mean estimator of the run time. Here we notice that the total run time of Marvell only takes up a very small amount of the total training time for each method, further corroborating our algorithm’s time efficiency advantage. + +$$ +\frac | \begin{array} { l c } { { \mathrm { b a t c h } } } \\ { { \mathrm { s i z e } } } \\ { { \frac { B } { 1 . 8 1 \times } } } \end{array} | \begin{array} { l c } { { \mathrm { c u t ~ l a y e r ~ f e a t u r e } } } \\ { { \mathrm { d i m e n s i o n } } } \\ { { \mathrm { ~ d i o n } } } \end{array} | \begin{array} { l l } { { \mathrm { ~ a n v e r ~ f e n ~ t i m e } } } \\ { { \mathrm { M a r v e l ~ r u n ~ t i m e } } } \\ { { \mathrm { ~ 1 . 7 9 \times ~ l 0 ^ { - 2 } ~ } } } \\ { { \mathrm { ~ 2 . 7 . 0 3 \times ~ 1 0 ^ { - 5 } ~ } } } \end{array} | \begin{array} { l l } { { \mathrm { ~ a v e r a g e } } } \\ { { \mathrm { ~ ( s e c o n d s / ~ t i m e ) } } } \\ { { \mathrm { ~ 3 . 9 4 \times ~ 1 0 ^ { - 1 } ~ } } } \\ { { \mathrm { ~ 4 . 0 5 \times ~ 1 0 ^ { - 3 } ~ } } } \\ { { \mathrm { ~ 1 . 6 8 \times ~ 1 0 ^ { - 2 } ~ } } } \end{array} | \begin{array} { l l } { { \mathrm { a v e r a g e } } } \\ { { \mathrm { ~ 4 . 2 8 \times ~ 1 0 ^ { - 1 } ~ } } } \\ { { \mathrm { ~ 4 . 8 4 ~ } } } \end{array} +$$ + +Table 1: Average run time of Marvell $s = 4$ ) and average total update time per iteration for models trained on ISIC, Criteo, and Avazu. + +# A.6 DATA SETUP AND EXPERIMENTAL DETAILS + +We first describe how we preprocess each of the datasets in A.6.1. We then describe the model architecture used for each dataset in A.6.2. Finally, we describe what the training hyperparameters are used for each dataset/model combination and the amount of compute used for the experiments in A.6.3. + +# A.6.1 DATASET PREPROCESSING + +[Criteo] Every record of Criteo has 27 categorical input features and 14 real-valued input features. We first replace all the NA values in categorical features with a single new category (which we represent using the empty string) and replace all the NA values in real-valued features with 0. For each categorical feature, we convert each of its possible value uniquely to an integer between 0 (inclusive) and the total number of unique categories (exclusive). For each real-valued feature, we linearly normalize it into [0, 1]. We then randomly sample $1 0 \%$ of the entire Criteo publicly provided training set as our entire dataset (for faster training to generate privacy-utility trade-off comparision) and further make the subsampled dataset into a $90 \%$ - $10 \%$ train-test split. + +[Avazu] Unlike Criteo, each record in Avazu only has categorical input features. We similarly replace all NA value with a single new category (the empty string), and for each categorical feature, we convert each of its possible value uniquely to an integer between 0 (inclusive) and the total number of unique categories (exclusive). We use all the records in provided in Avazu and randomly split it into $90 \%$ for training and $10 \%$ for test. + +[ISIC] The official SIIM-ISIC Melanoma Classification dataset has a total 33126 of skin lesion images with less than $2 \%$ positive examples. Because for image classification model training it is desirable to use a batch size of $\sim 1 0 ^ { 2 }$ , it is highly likely that there won’t be any positive examples sampled in a batch of such size. Thus to make the label leakage problem more severe, we modify the dataset by retaining all the 584 positive examples and randomly choosing $5 8 4 \times 9$ examples out of all the negative examples. By doing this, we enforce that there are $10 \%$ positive examples in this modified dataset. We randomly split these 5840 examples into a $80 \% - 2 0 \%$ training and test split. We also resize the images to size $8 4 \times 8 4$ for efficient model training. + +# A.6.2 MODEL ARCHITECTURE DETAILS + +[Criteo, Avazu] We use a popular deep learning model architecture WDL (Cheng et al., 2016) for online advertising. Here the deep and wide part each first processes the categorical features in a given record by applying an embedding lookup for every categorical feature’s value. We use an embedding dimension of 4 for the deep part and embedding dimension of 1 for the wide part. After the lookup, the deep/wide embeddings are then concatenated with the continuous features to form the raw input vectors for both the deep part and wide part respectively. (This step is skipped for Avazu as it has no continuous features.) Then the wide part computes the wide part logit value through a real-valued linear function (with bias) of its raw input vectors, while the deep part processes its raw input features using 6 ReLU-activated 128-unit MLP layers before producing a single deep part logit. The two logits are summed up to form the final logic value. The cut layer is after the output of the 3rd ReLU layer on the deep part. + +[ISIC] Every input image after resizing is of size $8 4 \times 8 4 \times 3$ . We use a convolutional model with 6 convolutional layers each with 64 channels $3 \times 3$ filter size with $1 \times 1$ stride size. Each convolutional layer is followed by a ReLU activation function whose output is then max pooled with $2 \times 2$ window and stride size $2 \times 2$ . The max-pooled output of the 6th layer is then flattened and pass into a 64-unit ReLU-activated MLP layer before finally being linearly transformed into a single logit score. The cut layer is after the output of the 4th max pool layer. Thus the cut layer feature and gradient are both of shape $5 \times 5 \times 6 4$ . + +# A.6.3 MODEL TRAINING DETAILS + +Because the protection mechanism requires adding noise to the cut layer gradient, the induced variance of the gradients of non-label party’s $f$ -parameters becomes larger. Thus to ensure smooth optimization and sufficient training loss minimization, we use a slightly smaller learning rate than what is normally used. + +[Criteo] We use the Adam optimizer with a batch size of 1024 and a learning rate of $1 e { - 4 }$ throughout the entire training of 5 epochs (approximately $2 0 \mathrm { k }$ stochastic gradient updates). + +[ISIC] We use the Adam optimizer with a batch size of 128 and a learning rate of $1 e { - } 5$ throughout the entire training of 1000 epochs (approximately $3 5 \mathrm { k }$ stochastic gradient updates). + +[Avazu] We use the Adam optimizer with a batch size of 32768 and a learning rate of $1 e { - 4 }$ throughout the entire training of 5 epochs (approximately $5 . 5 \mathrm { k }$ stochastic gradient updates). + +We conduct our experiments over 16 Nvidia 1080Ti GPU card. Each run of Avazu takes about 11 hours to finish on a single GPU card occupying 8GB of GPU RAM. Each run of Criteo takes about 37 hours to finish on a single GPU card using 5 GB of GPU RAM. Each run of ISIC takes about 12 hours to finish on a single GPU card occupying 4GB of GPU RAM. + +# A.7 COMPLETE EXPERIMENTAL RESULTS + +# A.7.1 LEAK AUC PROGRESSION FOR AVAZU AND CRITEO + +In addition to the leak AUC progression on ISIC shown in Figure 3 in the main paper, we also show the leak AUC progression on the Avazu and Criteo datasets throughout training here in Figure 5 and Figure 6. We similarly compare Marvell with different levels of protection strength (s values) against the no protection baseline no noise. As we can see, Marvell still achieves strong and flexible privacy protection on these two datasets against our label attacks at different model layers. + +![](images/fc64579aad78e2fa08ffb6045f49d1a0fde107f7a7dc254d23753e69f6c1753e.jpg) +Figure 5: Norm and cosine leak AUC (computed every batch) at the cut layer and at the first layer of no noise (no protection) vs. Marvell with different scale hyperparameter $s$ throughout the Avazu training. + +![](images/7998fbd902b18742af0791b87deac7e2a9cb14ec0faab7692e4248043d10fa75.jpg) +Figure 6: Norm and cosine leak AUC (computed every batch) at the cut layer and at the first layer of no noise (no protection) vs. Marvell with different scale hyperparameter s throughout the Criteo training. + +# A.7.2 COMPLETE PRIVACY-UTILITY TRADEOFFS + +We show additional Privacy-Utility tradeoff results for all three datasets considered in this paper. (Some of the plots have already been shown in the main paper but we still include them here for completeness and ease of reference.) For each dataset, we compare the privacy-utility tradeoff over multiple measures of privacy and utility: + +# Privacy + +We consider our introduced privacy metrics using the activation gradient from the cut layer and the first layer of the non-label party : + +• $9 5 \%$ norm leak AUC at cut layer • $9 5 \%$ cosine leak AUC at cut layer • $9 5 \%$ norm leak AUC at first layer • $9 5 \%$ cosine leak AUC at first layer + +# + +We consider three metrics of utility: + +• training loss (train loss): the lowest loss achieved on the training set throughout training. This directly measures how much the random protection perturbation influences the optimization. • test loss. Because we only control the training optimization stochastic gradient’s variance, measuring test loss directly tells us how much impact the training optimization random perturbation influences beyond optimization but on the learned model’s generalization ability. • test AUC. As we are dealing with binary classification problem (where performance is commonly measured through test AUC), we also naturally consider it as a utility metric. + +As shown in Figure 7,8,9, Marvell consistently outperforms the isotropic Gaussian baseline over all the different privacy-utility definitions. In addition, our proposed heuristic max norm is also particularly effective against our identified norm and direction-based attacks. + +![](images/cfa83c054f2045646208abae64f5e5969c5183c3b6697991cd25d9d7634217b2.jpg) +Figure 7: Privacy (norm and cosine leak AUC) vs Utility (train loss, test loss, and test AUC) trade-off of protection methods (Marvell, iso, no noise, max norm) at the cut layer and first layer on Avazu. + +![](images/f9b37071e310cce66c89d73ed7335527adaa39502b1e85c60894e68a77ab15fc.jpg) +Figure 8: Privacy (norm and cosine leak AUC) vs Utility (train loss, test loss, and test AUC) trade-off of protection methods (Marvell, iso, no noise, max norm) at the cut layer and first layer on Criteo. + +![](images/6397972f53ef2b60ce1c15ca92d0c6d5017d66a8e99b3d79b27208a0b0d8b18f.jpg) +Figure 9: Privacy (norm and cosine leak AUC) vs Utility (train loss, test loss, and test AUC) trade-off of protection methods (Marvell, iso, no noise, max norm) at the cut layer and first layer on ISIC. + +A.7.3 IMPACT OF ACTIVATION GRADIENTS’ LAYER LOCATION ON THE CORRESPONDINGLEAK AUC + +In Section 5.1, we have shown that the activation gradients of not only the cut layer but also the first layer can leak the label if no protection is applied. In this section, we analyze the effect of this layer location on the degree of label leakage under our proposed attacks measured through leak AUC. Here, for the convolutional neural network trained on ISIC, we have allocated 4 Conv-ReLUMaxPool layers on the non-label party side. Here each of these 4 layers’ activation (from ReLU output) gradients can be used to infer the label. In Figure 10, we plot the progression of norm and cosine leak AUC for layer 2 and 3 in addition to the first layer (layer 1) and the cut layer (layer 4) when applying no protection or Marvell with $s = 4$ . We notice that when no protection is applied, the norm and cosine leak AUC are approximately the same across the layers and both at a very high level, indicating a strong degree of label leakage. In contrast, when Marvell is applied, the leak AUCs are much lowered to a reasonable level of around 0.6 throughout all the layers. Empirically, we notice that the leak AUCs for a given training iteration are close among all the layers with a maximum difference of around 0.05. Here it is unclear whether the earlier layers’ gradients would leak the labels more than the cut layer or not — for norm leak AUC, the earlier layers have a higher AUC values (more leakage) than the cut layer, while for cosine leak AUC, the earlier layers have a lower AUC values (less leakage) than the cut layer. Further understanding the relationship between these leak AUC values across different layers is an open question motivated by our observations here. + +![](images/56ed010866be0a341dc9aca02c15880bab8d679f13f2a8fb1257db4f8e6de8e9.jpg) +ISIC, non-label party's convolution network f with four Conv-ReLU-MaxPool layers +Figure 10: For a non-label party’s four-layer convolutional architecture trained on ISIC, we plot the progression of the norm or cosine leak AUC computed using the activation gradients from each of the four layers throughout training in (a), (b) when no random protection is applied and in (c), (d) when using Marvell with privacy hyperparameter $s = 4$ at the cut layer (layer 4). The four curves overlap in (b) at the value of 1.0. For each layer’s data in each figure, a 1-d Gaussian kernel with standard deviation of 5 is convolved with the raw 1-d array of leak AUC values to smooth out the fluctuations and make the different layers’ degree of leakage more visually distinct. + +Beyond the four layer convolutional neural network trained on ISIC, we also train a new MLP model architecture on Criteo. Here instead of having 3 128-unit ReLU-activated MLP layers for the nonlabel party’s function $f$ as described in Section A.6.2, we use 8 such layers for $f$ while keeping every other architectural component the same. We plot the progression of leak AUC values on each of the eight layers in Figure 11. Here the observations are very similar to what we see in Figure 10, confirming our aforementioned conclusions in this section. + +![](images/02c92cdadf09570bf4017dec0ff34b8c73f98c3538f14856b782c4449ed8450b.jpg) +Figure 11: For a non-label party’s eight-layer MLP architecture trained on Criteo, we plot the progression of the norm or cosine leak AUC computed using the activation gradients from each of the eight layers throughout training in (a), (b) when no random protection is applied and in (c), (d) when using Marvell with privacy hyperparameter $s = 4$ at the cut layer (layer 4). The eight curves overlap in (b) at the value of 1.0. For each layer’s data in each figure, a 1-d Gaussian kernel with standard deviation of 100 is convolved with the raw 1-d array of leak AUC values to smooth out the fluctuations and make the different layers’ degree of leakage more visually distinct. + +It is possible to use a local differential privacy definition to analyze our problem setup if we treat any single training example as a dataset and two examples (dataset) are adjacent if they share the same $X$ but have the opposite label $y$ . Then one can analyze the activation gradient computation function’s local DP properties. However, we choose not to use local DP in our setup for the following two reasons: + +1. We are not aware of any attacks associated with local DP that are suitable for practical use in our problem setup. In contrast, our leak AUC privacy quantification metrics have corresponding concrete, realistic attack methods that the non-label party can use in practice. 2. Even if there exists a practical attack associated with local DP, this attack would require example-level side information as the activation gradient distributions would be examplespecific. Distinguishing between such gradient distributions for a pair of adjacent examples (with same $X$ but opposite label $y$ ) would require knowledge about the two distributions specific to this example pair itself. As we have mentioned in Section 3.2, we do not assume the non-label party would realistically have access to such fine-grained knowledge. Instead, in our setup, the non-label party uses a scoring function that only takes in the communicated cut layer gradient without any additional information. \ No newline at end of file diff --git a/parse/dev/cOtBRgsf2fO/cOtBRgsf2fO_middle.json b/parse/dev/cOtBRgsf2fO/cOtBRgsf2fO_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..6e36c9f5f11b02638dbe4ea37f6ec3860dbf0b7c --- /dev/null +++ b/parse/dev/cOtBRgsf2fO/cOtBRgsf2fO_middle.json @@ -0,0 +1,68334 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 392, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 79, + 393, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 79, + 393, + 96 + ], + "score": 1.0, + "content": "LABEL LEAKAGE AND PROTECTION IN", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 97, + 330, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 97, + 330, + 117 + ], + "score": 1.0, + "content": "TWO-PARTY SPLIT LEARNING", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 113, + 134, + 359, + 157 + ], + "lines": [ + { + "bbox": [ + 112, + 133, + 317, + 147 + ], + "spans": [ + { + "bbox": [ + 112, + 133, + 137, + 147 + ], + "score": 1.0, + "content": "Oscar", + "type": "text" + }, + { + "bbox": [ + 137, + 134, + 154, + 146 + ], + "score": 0.84, + "content": "\\mathbf { L i } ^ { 1 \\dagger }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 133, + 188, + 147 + ], + "score": 1.0, + "content": "∗, Jiankai", + "type": "text" + }, + { + "bbox": [ + 189, + 134, + 212, + 146 + ], + "score": 0.83, + "content": "\\mathbf { S u n ^ { 2 \\ S } }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 133, + 216, + 147 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 216, + 134, + 257, + 146 + ], + "score": 0.29, + "content": "\\mathbf { X i n \\ Y a n g ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 133, + 291, + 147 + ], + "score": 1.0, + "content": ", Weihao", + "type": "text" + }, + { + "bbox": [ + 291, + 135, + 312, + 145 + ], + "score": 0.31, + "content": "\\mathbf { G a o } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 133, + 317, + 147 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 145, + 358, + 159 + ], + "spans": [ + { + "bbox": [ + 116, + 145, + 216, + 159 + ], + "score": 1.0, + "content": "Hongyi Zhang2, Junyuan", + "type": "text" + }, + { + "bbox": [ + 217, + 146, + 235, + 156 + ], + "score": 0.64, + "content": "\\mathbf { X i e } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 145, + 358, + 159 + ], + "score": 1.0, + "content": ", Virginia Smith1, Chong Wang2", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 389, + 136, + 493, + 157 + ], + "lines": [ + { + "bbox": [ + 388, + 134, + 494, + 148 + ], + "spans": [ + { + "bbox": [ + 388, + 134, + 494, + 148 + ], + "score": 1.0, + "content": "1Carnegie Mellon University", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 430, + 145, + 494, + 159 + ], + "spans": [ + { + "bbox": [ + 430, + 145, + 494, + 159 + ], + "score": 1.0, + "content": "2ByteDance Inc.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.0 + }, + { + "type": "title", + "bbox": [ + 278, + 172, + 333, + 184 + ], + "lines": [ + { + "bbox": [ + 276, + 171, + 336, + 186 + ], + "spans": [ + { + "bbox": [ + 276, + 171, + 336, + 186 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 143, + 190, + 468, + 343 + ], + "lines": [ + { + "bbox": [ + 141, + 190, + 469, + 201 + ], + "spans": [ + { + "bbox": [ + 141, + 190, + 469, + 201 + ], + "score": 1.0, + "content": "Two-party split learning is a popular technique for learning a model across feature-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 140, + 200, + 470, + 214 + ], + "spans": [ + { + "bbox": [ + 140, + 200, + 470, + 214 + ], + "score": 1.0, + "content": "partitioned data. In this work, we explore whether it is possible for one party to", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 212, + 470, + 225 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 470, + 225 + ], + "score": 1.0, + "content": "steal the private label information from the other party during split training, and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 223, + 469, + 235 + ], + "spans": [ + { + "bbox": [ + 141, + 223, + 469, + 235 + ], + "score": 1.0, + "content": "whether there are methods that can protect against such attacks. Specifically, we", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 233, + 469, + 247 + ], + "spans": [ + { + "bbox": [ + 141, + 233, + 469, + 247 + ], + "score": 1.0, + "content": "first formulate a realistic threat model and propose a privacy loss metric to quan-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "score": 1.0, + "content": "tify label leakage in split learning. We then show that there exist two simple yet", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 255, + 469, + 268 + ], + "spans": [ + { + "bbox": [ + 141, + 255, + 469, + 268 + ], + "score": 1.0, + "content": "effective methods within the threat model that can allow one party to accurately", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 267, + 469, + 278 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 469, + 278 + ], + "score": 1.0, + "content": "recover private ground-truth labels owned by the other party. To combat these at-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 278, + 469, + 290 + ], + "spans": [ + { + "bbox": [ + 142, + 278, + 469, + 290 + ], + "score": 1.0, + "content": "tacks, we propose several random perturbation techniques, including Marvell,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 289, + 469, + 301 + ], + "spans": [ + { + "bbox": [ + 142, + 289, + 469, + 301 + ], + "score": 1.0, + "content": "an approach that strategically finds the structure of the noise perturbation by min-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 300, + 469, + 311 + ], + "spans": [ + { + "bbox": [ + 142, + 300, + 469, + 311 + ], + "score": 1.0, + "content": "imizing the amount of label leakage (measured through our quantification metric)", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 310, + 470, + 323 + ], + "spans": [ + { + "bbox": [ + 142, + 310, + 470, + 323 + ], + "score": 1.0, + "content": "of a worst-case adversary. We empirically1 demonstrate the effectiveness of our", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 322, + 469, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 322, + 469, + 334 + ], + "score": 1.0, + "content": "protection techniques against the identified attacks, and show that Marvell in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 332, + 463, + 346 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 463, + 346 + ], + "score": 1.0, + "content": "particular has improved privacy-utility tradeoffs relative to baseline approaches.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 13.5 + }, + { + "type": "title", + "bbox": [ + 108, + 356, + 206, + 368 + ], + "lines": [ + { + "bbox": [ + 105, + 354, + 208, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 208, + 371 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 372, + 505, + 482 + ], + "lines": [ + { + "bbox": [ + 106, + 371, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 506, + 384 + ], + "score": 1.0, + "content": "With increasing concerns over data privacy in machine learning, federated learning (FL) (McMahan", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 382, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 505, + 395 + ], + "score": 1.0, + "content": "et al., 2017) has become a promising direction of study. Based on how sensitive data are distributed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "among parties, FL can be classified into different categories, notable among which are horizontal", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 404, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 120, + 414 + ], + "score": 0.74, + "content": "F L", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 404, + 171, + 417 + ], + "score": 1.0, + "content": "and vertical", + "type": "text" + }, + { + "bbox": [ + 172, + 405, + 185, + 415 + ], + "score": 0.61, + "content": "F L", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 404, + 505, + 417 + ], + "score": 1.0, + "content": "(Yang et al., 2019). In contrast to horizontal FL where the data are partitioned", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 414, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 427 + ], + "score": 1.0, + "content": "by examples, vertical FL considers data partitioned by features (including labels). As a canonical", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 347, + 438 + ], + "score": 1.0, + "content": "example of vertical FL, consider an online media platform", + "type": "text" + }, + { + "bbox": [ + 347, + 427, + 356, + 436 + ], + "score": 0.68, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "which displays advertisements from", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 436, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 104, + 436, + 145, + 451 + ], + "score": 1.0, + "content": "company", + "type": "text" + }, + { + "bbox": [ + 146, + 438, + 155, + 447 + ], + "score": 0.82, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 436, + 254, + 451 + ], + "score": 1.0, + "content": "to its users, and charges", + "type": "text" + }, + { + "bbox": [ + 255, + 438, + 264, + 447 + ], + "score": 0.82, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 436, + 506, + 451 + ], + "score": 1.0, + "content": "for each conversion (e.g., a user clicking the ad and buying", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 416, + 460 + ], + "score": 1.0, + "content": "the product). In this case, both parties have different features for each user:", + "type": "text" + }, + { + "bbox": [ + 416, + 448, + 425, + 458 + ], + "score": 0.6, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "has features on the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 252, + 471 + ], + "score": 1.0, + "content": "user’s media viewing records, while", + "type": "text" + }, + { + "bbox": [ + 253, + 460, + 262, + 469 + ], + "score": 0.78, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 460, + 394, + 471 + ], + "score": 1.0, + "content": "has the user’s conversion labels.", + "type": "text" + }, + { + "bbox": [ + 394, + 459, + 403, + 469 + ], + "score": 0.78, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 460, + 505, + 471 + ], + "score": 1.0, + "content": "’s labels are not available", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 469, + 432, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 117, + 483 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 471, + 126, + 480 + ], + "score": 0.7, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 469, + 362, + 483 + ], + "score": 1.0, + "content": "because each user’s purchase behaviors happen entirely on", + "type": "text" + }, + { + "bbox": [ + 363, + 471, + 372, + 480 + ], + "score": 0.76, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 469, + 432, + 483 + ], + "score": 1.0, + "content": "’s website/app.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 107, + 487, + 505, + 640 + ], + "lines": [ + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "score": 1.0, + "content": "If both parties want to jointly learn a model to predict conversion without data sharing, split learn-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 497, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 505, + 511 + ], + "score": 1.0, + "content": "ing (Gupta & Raskar, 2018; Vepakomma et al., 2018) can be used to split the execution of a deep", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "network between the parties on a layer-wise basis. In vanilla split learning, before training begins,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 519, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 532 + ], + "score": 1.0, + "content": "both parties use Private Set Intersection (PSI) protocols (Kolesnikov et al., 2016; Pinkas et al., 2018)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 531, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 542 + ], + "score": 1.0, + "content": "to find the intersection of their data records and achieve an example ID alignment. This alignment", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 542, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 104, + 542, + 506, + 555 + ], + "score": 1.0, + "content": "paves the way for the split training phase. During training (Figure 1), the party without labels (non-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "label party) sends the intermediate layer (cut layer) outputs rather than the raw data to the party with", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "labels (label party), and the label party completes the rest of the forward computation to obtain the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 505, + 587 + ], + "score": 1.0, + "content": "training loss. To compute the gradients with respect to model parameters, the label party initiates", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 586, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 598 + ], + "score": 1.0, + "content": "backpropagation from its training loss and computes its own parameters’ gradients. To allow the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 596, + 504, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 504, + 608 + ], + "score": 1.0, + "content": "non-label party to also compute gradients of its parameters, the label party also computes the gradi-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "ents with respect to the cut layer outputs and communicates this information back to the non-label", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 618, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 506, + 631 + ], + "score": 1.0, + "content": "party. As a result of the ID alignment, despite not knowing the label party’s raw label data, the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 629, + 473, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 473, + 642 + ], + "score": 1.0, + "content": "non-label party can identify the gradient value returned by the label party for each example.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 108, + 646, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "At first glance, the process of split learning appears privacy-preserving because only the intermedi-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 657, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 505, + 670 + ], + "score": 1.0, + "content": "ate computations of the cut layer—rather than raw features or labels—are communicated between", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 668, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 668, + 505, + 680 + ], + "score": 1.0, + "content": "the two parties. However, such “gradient sharing” schemes have been shown to be vulnerable to", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 679, + 504, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 504, + 692 + ], + "score": 1.0, + "content": "privacy leakage in horizontal FL settings (e.g., Zhu et al., 2019). In vertical FL (and specifically", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 690, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 690, + 505, + 703 + ], + "score": 1.0, + "content": "split learning), it remains unclear whether the raw data can similarly be leaked during communi-", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 117, + 710, + 471, + 731 + ], + "lines": [ + { + "bbox": [ + 117, + 708, + 471, + 723 + ], + "spans": [ + { + "bbox": [ + 117, + 708, + 471, + 723 + ], + "score": 1.0, + "content": "∗work done during internship at ByteDance Inc. †oscarli@cmu.edu §jiankai.sun@bytedance.com", + "type": "text" + } + ] + }, + { + "bbox": [ + 119, + 719, + 356, + 733 + ], + "spans": [ + { + "bbox": [ + 119, + 719, + 356, + 733 + ], + "score": 1.0, + "content": "1Code available at https://github.com/OscarcarLi/label-protection", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 392, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 79, + 393, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 79, + 393, + 96 + ], + "score": 1.0, + "content": "LABEL LEAKAGE AND PROTECTION IN", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 97, + 330, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 97, + 330, + 117 + ], + "score": 1.0, + "content": "TWO-PARTY SPLIT LEARNING", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 113, + 134, + 359, + 157 + ], + "lines": [ + { + "bbox": [ + 112, + 133, + 317, + 147 + ], + "spans": [ + { + "bbox": [ + 112, + 133, + 137, + 147 + ], + "score": 1.0, + "content": "Oscar", + "type": "text" + }, + { + "bbox": [ + 137, + 134, + 154, + 146 + ], + "score": 0.84, + "content": "\\mathbf { L i } ^ { 1 \\dagger }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 133, + 188, + 147 + ], + "score": 1.0, + "content": "∗, Jiankai", + "type": "text" + }, + { + "bbox": [ + 189, + 134, + 212, + 146 + ], + "score": 0.83, + "content": "\\mathbf { S u n ^ { 2 \\ S } }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 133, + 216, + 147 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 216, + 134, + 257, + 146 + ], + "score": 0.29, + "content": "\\mathbf { X i n \\ Y a n g ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 133, + 291, + 147 + ], + "score": 1.0, + "content": ", Weihao", + "type": "text" + }, + { + "bbox": [ + 291, + 135, + 312, + 145 + ], + "score": 0.31, + "content": "\\mathbf { G a o } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 133, + 317, + 147 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 145, + 358, + 159 + ], + "spans": [ + { + "bbox": [ + 116, + 145, + 216, + 159 + ], + "score": 1.0, + "content": "Hongyi Zhang2, Junyuan", + "type": "text" + }, + { + "bbox": [ + 217, + 146, + 235, + 156 + ], + "score": 0.64, + "content": "\\mathbf { X i e } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 145, + 358, + 159 + ], + "score": 1.0, + "content": ", Virginia Smith1, Chong Wang2", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.0, + "bbox_fs": [ + 112, + 133, + 358, + 159 + ] + }, + { + "type": "index", + "bbox": [ + 389, + 136, + 493, + 157 + ], + "lines": [ + { + "bbox": [ + 388, + 134, + 494, + 148 + ], + "spans": [ + { + "bbox": [ + 388, + 134, + 494, + 148 + ], + "score": 1.0, + "content": "1Carnegie Mellon University", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 430, + 145, + 494, + 159 + ], + "spans": [ + { + "bbox": [ + 430, + 145, + 494, + 159 + ], + "score": 1.0, + "content": "2ByteDance Inc.", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + } + ], + "index": 4.0, + "bbox_fs": [ + 388, + 134, + 494, + 159 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 172, + 333, + 184 + ], + "lines": [ + { + "bbox": [ + 276, + 171, + 336, + 186 + ], + "spans": [ + { + "bbox": [ + 276, + 171, + 336, + 186 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 143, + 190, + 468, + 343 + ], + "lines": [ + { + "bbox": [ + 141, + 190, + 469, + 201 + ], + "spans": [ + { + "bbox": [ + 141, + 190, + 469, + 201 + ], + "score": 1.0, + "content": "Two-party split learning is a popular technique for learning a model across feature-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 140, + 200, + 470, + 214 + ], + "spans": [ + { + "bbox": [ + 140, + 200, + 470, + 214 + ], + "score": 1.0, + "content": "partitioned data. In this work, we explore whether it is possible for one party to", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 212, + 470, + 225 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 470, + 225 + ], + "score": 1.0, + "content": "steal the private label information from the other party during split training, and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 223, + 469, + 235 + ], + "spans": [ + { + "bbox": [ + 141, + 223, + 469, + 235 + ], + "score": 1.0, + "content": "whether there are methods that can protect against such attacks. Specifically, we", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 233, + 469, + 247 + ], + "spans": [ + { + "bbox": [ + 141, + 233, + 469, + 247 + ], + "score": 1.0, + "content": "first formulate a realistic threat model and propose a privacy loss metric to quan-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "score": 1.0, + "content": "tify label leakage in split learning. We then show that there exist two simple yet", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 255, + 469, + 268 + ], + "spans": [ + { + "bbox": [ + 141, + 255, + 469, + 268 + ], + "score": 1.0, + "content": "effective methods within the threat model that can allow one party to accurately", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 267, + 469, + 278 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 469, + 278 + ], + "score": 1.0, + "content": "recover private ground-truth labels owned by the other party. To combat these at-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 278, + 469, + 290 + ], + "spans": [ + { + "bbox": [ + 142, + 278, + 469, + 290 + ], + "score": 1.0, + "content": "tacks, we propose several random perturbation techniques, including Marvell,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 289, + 469, + 301 + ], + "spans": [ + { + "bbox": [ + 142, + 289, + 469, + 301 + ], + "score": 1.0, + "content": "an approach that strategically finds the structure of the noise perturbation by min-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 300, + 469, + 311 + ], + "spans": [ + { + "bbox": [ + 142, + 300, + 469, + 311 + ], + "score": 1.0, + "content": "imizing the amount of label leakage (measured through our quantification metric)", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 310, + 470, + 323 + ], + "spans": [ + { + "bbox": [ + 142, + 310, + 470, + 323 + ], + "score": 1.0, + "content": "of a worst-case adversary. We empirically1 demonstrate the effectiveness of our", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 322, + 469, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 322, + 469, + 334 + ], + "score": 1.0, + "content": "protection techniques against the identified attacks, and show that Marvell in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 332, + 463, + 346 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 463, + 346 + ], + "score": 1.0, + "content": "particular has improved privacy-utility tradeoffs relative to baseline approaches.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 13.5, + "bbox_fs": [ + 140, + 190, + 470, + 346 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 356, + 206, + 368 + ], + "lines": [ + { + "bbox": [ + 105, + 354, + 208, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 208, + 371 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 372, + 505, + 482 + ], + "lines": [ + { + "bbox": [ + 106, + 371, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 506, + 384 + ], + "score": 1.0, + "content": "With increasing concerns over data privacy in machine learning, federated learning (FL) (McMahan", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 382, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 505, + 395 + ], + "score": 1.0, + "content": "et al., 2017) has become a promising direction of study. Based on how sensitive data are distributed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "among parties, FL can be classified into different categories, notable among which are horizontal", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 404, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 120, + 414 + ], + "score": 0.74, + "content": "F L", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 404, + 171, + 417 + ], + "score": 1.0, + "content": "and vertical", + "type": "text" + }, + { + "bbox": [ + 172, + 405, + 185, + 415 + ], + "score": 0.61, + "content": "F L", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 404, + 505, + 417 + ], + "score": 1.0, + "content": "(Yang et al., 2019). In contrast to horizontal FL where the data are partitioned", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 414, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 427 + ], + "score": 1.0, + "content": "by examples, vertical FL considers data partitioned by features (including labels). As a canonical", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 347, + 438 + ], + "score": 1.0, + "content": "example of vertical FL, consider an online media platform", + "type": "text" + }, + { + "bbox": [ + 347, + 427, + 356, + 436 + ], + "score": 0.68, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "which displays advertisements from", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 436, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 104, + 436, + 145, + 451 + ], + "score": 1.0, + "content": "company", + "type": "text" + }, + { + "bbox": [ + 146, + 438, + 155, + 447 + ], + "score": 0.82, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 436, + 254, + 451 + ], + "score": 1.0, + "content": "to its users, and charges", + "type": "text" + }, + { + "bbox": [ + 255, + 438, + 264, + 447 + ], + "score": 0.82, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 436, + 506, + 451 + ], + "score": 1.0, + "content": "for each conversion (e.g., a user clicking the ad and buying", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 416, + 460 + ], + "score": 1.0, + "content": "the product). In this case, both parties have different features for each user:", + "type": "text" + }, + { + "bbox": [ + 416, + 448, + 425, + 458 + ], + "score": 0.6, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "has features on the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 252, + 471 + ], + "score": 1.0, + "content": "user’s media viewing records, while", + "type": "text" + }, + { + "bbox": [ + 253, + 460, + 262, + 469 + ], + "score": 0.78, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 460, + 394, + 471 + ], + "score": 1.0, + "content": "has the user’s conversion labels.", + "type": "text" + }, + { + "bbox": [ + 394, + 459, + 403, + 469 + ], + "score": 0.78, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 460, + 505, + 471 + ], + "score": 1.0, + "content": "’s labels are not available", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 469, + 432, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 117, + 483 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 471, + 126, + 480 + ], + "score": 0.7, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 469, + 362, + 483 + ], + "score": 1.0, + "content": "because each user’s purchase behaviors happen entirely on", + "type": "text" + }, + { + "bbox": [ + 363, + 471, + 372, + 480 + ], + "score": 0.76, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 469, + 432, + 483 + ], + "score": 1.0, + "content": "’s website/app.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26.5, + "bbox_fs": [ + 104, + 371, + 506, + 483 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 487, + 505, + 640 + ], + "lines": [ + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "score": 1.0, + "content": "If both parties want to jointly learn a model to predict conversion without data sharing, split learn-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 497, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 505, + 511 + ], + "score": 1.0, + "content": "ing (Gupta & Raskar, 2018; Vepakomma et al., 2018) can be used to split the execution of a deep", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "network between the parties on a layer-wise basis. In vanilla split learning, before training begins,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 519, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 532 + ], + "score": 1.0, + "content": "both parties use Private Set Intersection (PSI) protocols (Kolesnikov et al., 2016; Pinkas et al., 2018)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 531, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 542 + ], + "score": 1.0, + "content": "to find the intersection of their data records and achieve an example ID alignment. This alignment", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 542, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 104, + 542, + 506, + 555 + ], + "score": 1.0, + "content": "paves the way for the split training phase. During training (Figure 1), the party without labels (non-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "label party) sends the intermediate layer (cut layer) outputs rather than the raw data to the party with", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "labels (label party), and the label party completes the rest of the forward computation to obtain the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 505, + 587 + ], + "score": 1.0, + "content": "training loss. To compute the gradients with respect to model parameters, the label party initiates", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 586, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 598 + ], + "score": 1.0, + "content": "backpropagation from its training loss and computes its own parameters’ gradients. To allow the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 596, + 504, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 504, + 608 + ], + "score": 1.0, + "content": "non-label party to also compute gradients of its parameters, the label party also computes the gradi-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "ents with respect to the cut layer outputs and communicates this information back to the non-label", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 618, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 506, + 631 + ], + "score": 1.0, + "content": "party. As a result of the ID alignment, despite not knowing the label party’s raw label data, the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 629, + 473, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 473, + 642 + ], + "score": 1.0, + "content": "non-label party can identify the gradient value returned by the label party for each example.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 38.5, + "bbox_fs": [ + 104, + 486, + 506, + 642 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 646, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "At first glance, the process of split learning appears privacy-preserving because only the intermedi-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 657, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 505, + 670 + ], + "score": 1.0, + "content": "ate computations of the cut layer—rather than raw features or labels—are communicated between", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 668, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 668, + 505, + 680 + ], + "score": 1.0, + "content": "the two parties. However, such “gradient sharing” schemes have been shown to be vulnerable to", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 679, + 504, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 504, + 692 + ], + "score": 1.0, + "content": "privacy leakage in horizontal FL settings (e.g., Zhu et al., 2019). In vertical FL (and specifically", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 690, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 690, + 505, + 703 + ], + "score": 1.0, + "content": "split learning), it remains unclear whether the raw data can similarly be leaked during communi-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "cation. In particular, as the raw labels often contain highly sensitive information (e.g., what a user", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 218, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 505, + 230 + ], + "score": 1.0, + "content": "has purchased (in online advertising) or whether a user has a disease or not (in disease predic-", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 228, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 505, + 242 + ], + "score": 1.0, + "content": "tion) Vepakomma et al. (2018)), developing a rigorous understanding of the threat of label leakage", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 241, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 252 + ], + "score": 1.0, + "content": "and its protection is particularly important. Towards this goal, we make the following contributions:", + "type": "text", + "cross_page": true + } + ], + "index": 8 + } + ], + "index": 48, + "bbox_fs": [ + 105, + 646, + 505, + 703 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 78, + 498, + 185 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 78, + 498, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 78, + 498, + 183 + ], + "spans": [ + { + "bbox": [ + 110, + 78, + 498, + 183 + ], + "score": 0.967, + "type": "image", + "image_path": "6d3962aa293ad26de1b3789a4155ec37fededb9027eaf28bf79d4454dafbce24.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 78, + 498, + 113.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 113.66666666666666, + 498, + 149.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 149.33333333333331, + 498, + 184.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 183, + 502, + 204 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 182, + 504, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 504, + 195 + ], + "score": 1.0, + "content": "Figure 1: Communication diagram of two-party split training for an example of online advertising. We study", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 192, + 424, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 297, + 204 + ], + "score": 1.0, + "content": "whether it is possible for the communicated gradient", + "type": "text" + }, + { + "bbox": [ + 297, + 195, + 303, + 203 + ], + "score": 0.81, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 192, + 424, + 204 + ], + "score": 1.0, + "content": "to leak private label information.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 207, + 505, + 252 + ], + "lines": [ + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "cation. In particular, as the raw labels often contain highly sensitive information (e.g., what a user", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 218, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 505, + 230 + ], + "score": 1.0, + "content": "has purchased (in online advertising) or whether a user has a disease or not (in disease predic-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 228, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 505, + 242 + ], + "score": 1.0, + "content": "tion) Vepakomma et al. (2018)), developing a rigorous understanding of the threat of label leakage", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 241, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 252 + ], + "score": 1.0, + "content": "and its protection is particularly important. Towards this goal, we make the following contributions:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 113, + 257, + 506, + 384 + ], + "lines": [ + { + "bbox": [ + 113, + 257, + 504, + 270 + ], + "spans": [ + { + "bbox": [ + 113, + 257, + 504, + 270 + ], + "score": 1.0, + "content": "1. We formalize a threat model for label leakage in two-party split learning in the context of binary", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 125, + 267, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 125, + 267, + 505, + 281 + ], + "score": 1.0, + "content": "classification (Section 3.1), and propose specific privacy quantification metrics to measure the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 279, + 277, + 292 + ], + "spans": [ + { + "bbox": [ + 126, + 279, + 277, + 292 + ], + "score": 1.0, + "content": "severity of such threats (Section 3.2).", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 114, + 292, + 504, + 304 + ], + "spans": [ + { + "bbox": [ + 114, + 294, + 124, + 303 + ], + "score": 1.0, + "content": "2.", + "type": "text" + }, + { + "bbox": [ + 126, + 292, + 504, + 304 + ], + "score": 1.0, + "content": "We identify two simple and realistic methods within this threat model which can accurately", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 126, + 303, + 382, + 316 + ], + "spans": [ + { + "bbox": [ + 126, + 303, + 382, + 316 + ], + "score": 1.0, + "content": "recover the label party’s private label information (Section 3.3).", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 317, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 114, + 317, + 123, + 327 + ], + "score": 1.0, + "content": "3.", + "type": "text" + }, + { + "bbox": [ + 126, + 317, + 505, + 328 + ], + "score": 1.0, + "content": "We propose several random perturbation techniques to limit the label-stealing ability of the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 126, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 126, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "non-label party (Section 4). Among them, our principled approach Marvell directly searches", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 126, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 126, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "for the optimal random perturbation noise structure to minimize label leakage (as measured via", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 125, + 348, + 425, + 363 + ], + "spans": [ + { + "bbox": [ + 125, + 348, + 425, + 363 + ], + "score": 1.0, + "content": "our quantification metric) against a worst-case adversarial non-label party.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 114, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 114, + 363, + 124, + 372 + ], + "score": 1.0, + "content": "4.", + "type": "text" + }, + { + "bbox": [ + 125, + 362, + 506, + 375 + ], + "score": 1.0, + "content": "We experimentally demonstrate the effectiveness of our protection techniques and MARVELL’s", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 126, + 374, + 469, + 385 + ], + "spans": [ + { + "bbox": [ + 126, + 374, + 469, + 385 + ], + "score": 1.0, + "content": "improved privacy-utility tradeoffs compared to other protection baselines (Section 5).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 393, + 212, + 406 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 213, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 213, + 408 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 413, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "score": 1.0, + "content": "Privacy leakage in split learning. Although raw data is not shared in federated learning, sensitive", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 423, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 505, + 436 + ], + "score": 1.0, + "content": "information may still be leaked when gradients and/or model parameters are communicated between", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 434, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 104, + 434, + 506, + 448 + ], + "score": 1.0, + "content": "parties. In horizontal FL, Zhu et al. (2019) showed that an honest-but-curious server can uncover", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 446, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 505, + 457 + ], + "score": 1.0, + "content": "the raw features and labels of a device by knowing the model architecture, parameters, and com-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 457, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 505, + 469 + ], + "score": 1.0, + "content": "municated gradient of the loss on the device’s data. Based on their techniques, Zhao et al. (2020)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 468, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 505, + 479 + ], + "score": 1.0, + "content": "showed that the ground truth label of an example can be extracted by exploiting the directions of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 478, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 506, + 491 + ], + "score": 1.0, + "content": "the gradients of the weights connected to the logits of different classes. Here we study a different", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "setting—two-party split learning (in vertical FL) (Yang et al., 2019), where no party has access to the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 500, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 505, + 513 + ], + "score": 1.0, + "content": "model architecture or model parameters of the other party. In this setting, Vepakomma et al. (2019)", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "score": 1.0, + "content": "studied how the forward communication of feature representations can leak the non-label party’s", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 523, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 534 + ], + "score": 1.0, + "content": "raw data to the label party. We instead study whether label information may be leaked from the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 534, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 545 + ], + "score": 1.0, + "content": "label party to the non-label party during the backward communication. Despite the importance of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 544, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 557 + ], + "score": 1.0, + "content": "maintaining the privacy of these labels, we are unaware of prior work that has studied this problem.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 505, + 736 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "Privacy protection and quantification. Techniques to protect communication privacy in FL gen-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 571, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 506, + 585 + ], + "score": 1.0, + "content": "erally fall into three categories: 1) cryptographic methods such as secure multi-party computation", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "(e.g., Bonawitz et al., 2017); 2) system-based methods including trusted execution environments", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 595, + 504, + 605 + ], + "spans": [ + { + "bbox": [ + 107, + 595, + 504, + 605 + ], + "score": 1.0, + "content": "(Subramanyan et al., 2017); and 3) perturbation methods that shuffle or modify the communicated", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "messages (e.g., Abadi et al., 2016; McMahan et al., 2018; Erlingsson et al., 2019; Cheu et al., 2019;", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 615, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 628 + ], + "score": 1.0, + "content": "Zhu et al., 2019). Our protection techniques belong to the third category, as we add random per-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "turbations to the gradients to protect the labels. Many randomness-based protection methods have", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "been proposed in the domain of horizontal FL. In this case, differential privacy (DP) (Dwork, 2006;", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "Dwork et al., 2014) is commonly used to measure the proposed random mechanisms’ ability to", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "anonymize the identity of any single participating example in the model iterates. However, in split", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "score": 1.0, + "content": "learning, after PSI, both parties know exactly the identity of which example has participated in a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 682, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 505, + 694 + ], + "score": 1.0, + "content": "given gradient update. As we explain in Section 3.1, the object we aim to protect (the communi-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 693, + 504, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 504, + 704 + ], + "score": 1.0, + "content": "cated cut layer gradients), unlike the model iterates, is not an aggregate function of all the examples", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 703, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 106, + 703, + 505, + 716 + ], + "score": 1.0, + "content": "but are instead example-specific. As a result, DP and its variants (e.g. label DP (Chaudhuri &", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 714, + 504, + 727 + ], + "spans": [ + { + "bbox": [ + 106, + 714, + 504, + 727 + ], + "score": 1.0, + "content": "Hsu, 2011; Ghazi et al., 2021)) are not directly applicable metrics in our setting, and we instead", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 726, + 326, + 737 + ], + "spans": [ + { + "bbox": [ + 105, + 726, + 326, + 737 + ], + "score": 1.0, + "content": "propose a different metric (discussed in Section 3.2).", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 41.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 78, + 498, + 185 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 78, + 498, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 78, + 498, + 183 + ], + "spans": [ + { + "bbox": [ + 110, + 78, + 498, + 183 + ], + "score": 0.967, + "type": "image", + "image_path": "6d3962aa293ad26de1b3789a4155ec37fededb9027eaf28bf79d4454dafbce24.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 78, + 498, + 113.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 113.66666666666666, + 498, + 149.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 149.33333333333331, + 498, + 184.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 183, + 502, + 204 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 182, + 504, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 504, + 195 + ], + "score": 1.0, + "content": "Figure 1: Communication diagram of two-party split training for an example of online advertising. We study", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 192, + 424, + 204 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 297, + 204 + ], + "score": 1.0, + "content": "whether it is possible for the communicated gradient", + "type": "text" + }, + { + "bbox": [ + 297, + 195, + 303, + 203 + ], + "score": 0.81, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 192, + 424, + 204 + ], + "score": 1.0, + "content": "to leak private label information.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 207, + 505, + 252 + ], + "lines": [], + "index": 6.5, + "bbox_fs": [ + 105, + 207, + 505, + 252 + ], + "lines_deleted": true + }, + { + "type": "list", + "bbox": [ + 113, + 257, + 506, + 384 + ], + "lines": [ + { + "bbox": [ + 113, + 257, + 504, + 270 + ], + "spans": [ + { + "bbox": [ + 113, + 257, + 504, + 270 + ], + "score": 1.0, + "content": "1. We formalize a threat model for label leakage in two-party split learning in the context of binary", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 267, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 125, + 267, + 505, + 281 + ], + "score": 1.0, + "content": "classification (Section 3.1), and propose specific privacy quantification metrics to measure the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 279, + 277, + 292 + ], + "spans": [ + { + "bbox": [ + 126, + 279, + 277, + 292 + ], + "score": 1.0, + "content": "severity of such threats (Section 3.2).", + "type": "text" + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 114, + 292, + 504, + 304 + ], + "spans": [ + { + "bbox": [ + 114, + 294, + 124, + 303 + ], + "score": 1.0, + "content": "2.", + "type": "text" + }, + { + "bbox": [ + 126, + 292, + 504, + 304 + ], + "score": 1.0, + "content": "We identify two simple and realistic methods within this threat model which can accurately", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 303, + 382, + 316 + ], + "spans": [ + { + "bbox": [ + 126, + 303, + 382, + 316 + ], + "score": 1.0, + "content": "recover the label party’s private label information (Section 3.3).", + "type": "text" + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 114, + 317, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 114, + 317, + 123, + 327 + ], + "score": 1.0, + "content": "3.", + "type": "text" + }, + { + "bbox": [ + 126, + 317, + 505, + 328 + ], + "score": 1.0, + "content": "We propose several random perturbation techniques to limit the label-stealing ability of the", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 126, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "non-label party (Section 4). Among them, our principled approach Marvell directly searches", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 126, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 126, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "for the optimal random perturbation noise structure to minimize label leakage (as measured via", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 125, + 348, + 425, + 363 + ], + "spans": [ + { + "bbox": [ + 125, + 348, + 425, + 363 + ], + "score": 1.0, + "content": "our quantification metric) against a worst-case adversarial non-label party.", + "type": "text" + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 114, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 114, + 363, + 124, + 372 + ], + "score": 1.0, + "content": "4.", + "type": "text" + }, + { + "bbox": [ + 125, + 362, + 506, + 375 + ], + "score": 1.0, + "content": "We experimentally demonstrate the effectiveness of our protection techniques and MARVELL’s", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 374, + 469, + 385 + ], + "spans": [ + { + "bbox": [ + 126, + 374, + 469, + 385 + ], + "score": 1.0, + "content": "improved privacy-utility tradeoffs compared to other protection baselines (Section 5).", + "type": "text" + } + ], + "index": 19, + "is_list_end_line": true + } + ], + "index": 14, + "bbox_fs": [ + 113, + 257, + 506, + 385 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 393, + 212, + 406 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 213, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 213, + 408 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 413, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "score": 1.0, + "content": "Privacy leakage in split learning. Although raw data is not shared in federated learning, sensitive", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 423, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 505, + 436 + ], + "score": 1.0, + "content": "information may still be leaked when gradients and/or model parameters are communicated between", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 434, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 104, + 434, + 506, + 448 + ], + "score": 1.0, + "content": "parties. In horizontal FL, Zhu et al. (2019) showed that an honest-but-curious server can uncover", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 446, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 505, + 457 + ], + "score": 1.0, + "content": "the raw features and labels of a device by knowing the model architecture, parameters, and com-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 457, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 505, + 469 + ], + "score": 1.0, + "content": "municated gradient of the loss on the device’s data. Based on their techniques, Zhao et al. (2020)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 468, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 505, + 479 + ], + "score": 1.0, + "content": "showed that the ground truth label of an example can be extracted by exploiting the directions of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 478, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 506, + 491 + ], + "score": 1.0, + "content": "the gradients of the weights connected to the logits of different classes. Here we study a different", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "setting—two-party split learning (in vertical FL) (Yang et al., 2019), where no party has access to the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 500, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 505, + 513 + ], + "score": 1.0, + "content": "model architecture or model parameters of the other party. In this setting, Vepakomma et al. (2019)", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "score": 1.0, + "content": "studied how the forward communication of feature representations can leak the non-label party’s", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 523, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 534 + ], + "score": 1.0, + "content": "raw data to the label party. We instead study whether label information may be leaked from the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 534, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 545 + ], + "score": 1.0, + "content": "label party to the non-label party during the backward communication. Despite the importance of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 544, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 557 + ], + "score": 1.0, + "content": "maintaining the privacy of these labels, we are unaware of prior work that has studied this problem.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 27, + "bbox_fs": [ + 104, + 413, + 506, + 557 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 505, + 736 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "Privacy protection and quantification. Techniques to protect communication privacy in FL gen-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 571, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 506, + 585 + ], + "score": 1.0, + "content": "erally fall into three categories: 1) cryptographic methods such as secure multi-party computation", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "(e.g., Bonawitz et al., 2017); 2) system-based methods including trusted execution environments", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 595, + 504, + 605 + ], + "spans": [ + { + "bbox": [ + 107, + 595, + 504, + 605 + ], + "score": 1.0, + "content": "(Subramanyan et al., 2017); and 3) perturbation methods that shuffle or modify the communicated", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "messages (e.g., Abadi et al., 2016; McMahan et al., 2018; Erlingsson et al., 2019; Cheu et al., 2019;", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 615, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 628 + ], + "score": 1.0, + "content": "Zhu et al., 2019). Our protection techniques belong to the third category, as we add random per-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "turbations to the gradients to protect the labels. Many randomness-based protection methods have", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "been proposed in the domain of horizontal FL. In this case, differential privacy (DP) (Dwork, 2006;", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "Dwork et al., 2014) is commonly used to measure the proposed random mechanisms’ ability to", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "anonymize the identity of any single participating example in the model iterates. However, in split", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "score": 1.0, + "content": "learning, after PSI, both parties know exactly the identity of which example has participated in a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 682, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 505, + 694 + ], + "score": 1.0, + "content": "given gradient update. As we explain in Section 3.1, the object we aim to protect (the communi-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 693, + 504, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 504, + 704 + ], + "score": 1.0, + "content": "cated cut layer gradients), unlike the model iterates, is not an aggregate function of all the examples", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 703, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 106, + 703, + 505, + 716 + ], + "score": 1.0, + "content": "but are instead example-specific. As a result, DP and its variants (e.g. label DP (Chaudhuri &", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 714, + 504, + 727 + ], + "spans": [ + { + "bbox": [ + 106, + 714, + 504, + 727 + ], + "score": 1.0, + "content": "Hsu, 2011; Ghazi et al., 2021)) are not directly applicable metrics in our setting, and we instead", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 726, + 326, + 737 + ], + "spans": [ + { + "bbox": [ + 105, + 726, + 326, + 737 + ], + "score": 1.0, + "content": "propose a different metric (discussed in Section 3.2).", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 561, + 506, + 737 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 320, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 321, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 321, + 96 + ], + "score": 1.0, + "content": "3 LABEL LEAKAGE IN SPLIT LEARNING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 102, + 504, + 125 + ], + "lines": [ + { + "bbox": [ + 106, + 102, + 505, + 115 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 115 + ], + "score": 1.0, + "content": "We first introduce the two-party split learning problem for binary classification, and then formally", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 113, + 490, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 490, + 127 + ], + "score": 1.0, + "content": "describe our threat model and privacy quantification metrics with two concrete attack examples.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "title", + "bbox": [ + 107, + 135, + 382, + 146 + ], + "lines": [ + { + "bbox": [ + 105, + 135, + 383, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 383, + 148 + ], + "score": 1.0, + "content": "3.1 TWO-PARTY SPLIT LEARNING IN BINARY CLASSIFICATION", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 152, + 505, + 244 + ], + "lines": [ + { + "bbox": [ + 104, + 151, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 104, + 151, + 396, + 166 + ], + "score": 1.0, + "content": "Problem setup. Consider two parties learning a composition model", + "type": "text" + }, + { + "bbox": [ + 396, + 153, + 421, + 164 + ], + "score": 0.9, + "content": "h \\circ f", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 151, + 505, + 166 + ], + "score": 1.0, + "content": "jointly for a binary", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 162, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 264, + 177 + ], + "score": 1.0, + "content": "classification problem over the domain", + "type": "text" + }, + { + "bbox": [ + 265, + 163, + 311, + 176 + ], + "score": 0.93, + "content": "\\mathcal { X } \\times \\{ 0 , \\bar { 1 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 162, + 505, + 177 + ], + "score": 1.0, + "content": "(Figure 1). The non-label party owns the repre-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 174, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 104, + 174, + 181, + 189 + ], + "score": 1.0, + "content": "sentation function", + "type": "text" + }, + { + "bbox": [ + 181, + 174, + 234, + 187 + ], + "score": 0.92, + "content": "f : \\mathcal { X } \\to \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 174, + 363, + 189 + ], + "score": 1.0, + "content": "and each example’s raw feature", + "type": "text" + }, + { + "bbox": [ + 364, + 176, + 396, + 187 + ], + "score": 0.9, + "content": "X \\in { \\mathcal { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 174, + 505, + 189 + ], + "score": 1.0, + "content": "while the label party owns", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 187, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 181, + 201 + ], + "score": 1.0, + "content": "the logit function", + "type": "text" + }, + { + "bbox": [ + 182, + 187, + 242, + 199 + ], + "score": 0.89, + "content": "h : \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 187, + 352, + 201 + ], + "score": 1.0, + "content": "and each example’s label", + "type": "text" + }, + { + "bbox": [ + 353, + 188, + 404, + 201 + ], + "score": 0.93, + "content": "y \\in \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 187, + 430, + 201 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 430, + 188, + 490, + 201 + ], + "score": 0.92, + "content": "\\ell \\ = \\ h ( f ( X ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 187, + 505, + 201 + ], + "score": 1.0, + "content": "be", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 199, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 505, + 213 + ], + "score": 1.0, + "content": "the logit of the positive class whose predicted probability is given through the sigmoid function:", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 210, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 205, + 222 + ], + "score": 0.88, + "content": "\\widetilde { p } _ { 1 } = 1 / ( 1 + \\exp ( - \\ell ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 210, + 506, + 224 + ], + "score": 1.0, + "content": ". We measure the loss of such prediction through the cross entropy loss", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 221, + 504, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 222, + 252, + 234 + ], + "score": 0.87, + "content": "L = \\log ( 1 + \\exp ( - \\ell ) ) + ( 1 - y ) \\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 221, + 480, + 235 + ], + "score": 1.0, + "content": ". During model inference, the non-label party computes", + "type": "text" + }, + { + "bbox": [ + 480, + 221, + 504, + 234 + ], + "score": 0.91, + "content": "{ \\dot { \\boldsymbol { f } } } ( { \\boldsymbol { X } } )", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 233, + 493, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 493, + 245 + ], + "score": 1.0, + "content": "and sends it to the label party who will then execute the rest of forward computation in Figure 1.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 249, + 505, + 340 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 506, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 506, + 263 + ], + "score": 1.0, + "content": "Model training (Figure 1: backward gradient computation). To train the model using gradient", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 260, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 397, + 273 + ], + "score": 1.0, + "content": "descent, the label party starts by first computing the gradient of the loss", + "type": "text" + }, + { + "bbox": [ + 398, + 261, + 406, + 270 + ], + "score": 0.72, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 260, + 505, + 273 + ], + "score": 1.0, + "content": "with respect to the logit", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 107, + 270, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 107, + 271, + 173, + 284 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\frac { d L } { d \\ell } \\ = \\ ( \\widetilde { p } _ { 1 } - y ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 270, + 474, + 285 + ], + "score": 1.0, + "content": ". Using the chain rule, the label party can then compute the gradient of", + "type": "text" + }, + { + "bbox": [ + 474, + 272, + 483, + 281 + ], + "score": 0.79, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 270, + 505, + 285 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 283, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 195, + 294 + ], + "score": 1.0, + "content": "erespect to its function", + "type": "text" + }, + { + "bbox": [ + 195, + 283, + 202, + 292 + ], + "score": 0.79, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 283, + 505, + 294 + ], + "score": 1.0, + "content": "’s parameters and perform the gradient updates. To also allow the non-label", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 293, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 104, + 293, + 210, + 306 + ], + "score": 1.0, + "content": "party to learn its function", + "type": "text" + }, + { + "bbox": [ + 210, + 294, + 217, + 304 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 293, + 506, + 306 + ], + "score": 1.0, + "content": ", the label party needs to additionally compute the gradient with respect", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 187, + 317 + ], + "score": 1.0, + "content": "to cut layer feature", + "type": "text" + }, + { + "bbox": [ + 187, + 304, + 211, + 316 + ], + "score": 0.92, + "content": "f ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 304, + 505, + 317 + ], + "score": 1.0, + "content": "and communicate it to the non-label party. We denote this gradient by", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 314, + 507, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 304, + 330 + ], + "score": 0.86, + "content": "g : = \\nabla _ { f ( X ) } L = ( \\widetilde { p } _ { 1 } - y ) \\nabla _ { z } h ( z ) | _ { z = f ( X ) } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 314, + 438, + 330 + ], + "score": 1.0, + "content": "(by chain rule). After receiving", + "type": "text" + }, + { + "bbox": [ + 438, + 318, + 445, + 329 + ], + "score": 0.7, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 314, + 507, + 330 + ], + "score": 1.0, + "content": ", the non-label", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 287, + 341 + ], + "score": 1.0, + "content": "eparty continues the backpropagation towards", + "type": "text" + }, + { + "bbox": [ + 288, + 329, + 294, + 339 + ], + "score": 0.81, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 328, + 505, + 341 + ], + "score": 1.0, + "content": "’s parameters and also perform the gradient updates.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 344, + 505, + 457 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "score": 1.0, + "content": "Why Not Differential Privacy? Note that for a given iteration, the non-label party randomly", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 356, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 141, + 367 + ], + "score": 1.0, + "content": "chooses", + "type": "text" + }, + { + "bbox": [ + 142, + 356, + 151, + 366 + ], + "score": 0.77, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 356, + 505, + 367 + ], + "score": 1.0, + "content": "example IDs to form a batch. Therefore, the identity of which examples are used is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 366, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 427, + 380 + ], + "score": 1.0, + "content": "known to the non-label party by default. In addition, the communicated features", + "type": "text" + }, + { + "bbox": [ + 428, + 367, + 451, + 379 + ], + "score": 0.95, + "content": "f ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 366, + 505, + 380 + ], + "score": 1.0, + "content": "and returned", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 103, + 375, + 507, + 394 + ], + "spans": [ + { + "bbox": [ + 103, + 375, + 145, + 394 + ], + "score": 1.0, + "content": "gradients", + "type": "text" + }, + { + "bbox": [ + 145, + 381, + 151, + 391 + ], + "score": 0.78, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 375, + 245, + 394 + ], + "score": 1.0, + "content": "will both be matrices in", + "type": "text" + }, + { + "bbox": [ + 246, + 378, + 271, + 390 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { B \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 375, + 507, + 394 + ], + "score": 1.0, + "content": "with each row belonging to a specific example in the batch.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 389, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 506, + 403 + ], + "score": 1.0, + "content": "The different gradients (rows of the matrix) are not with respect to the same model parameters, but", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "score": 1.0, + "content": "are instead with respect to different examples’ cut-layer features; thus, no averaging over or shuffling", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 411, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 505, + 426 + ], + "score": 1.0, + "content": "of the rows of the gradient matrix can be done prior to communication to ensure correct computa-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 422, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 135, + 437 + ], + "score": 1.0, + "content": "tion of", + "type": "text" + }, + { + "bbox": [ + 135, + 424, + 142, + 435 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 422, + 505, + 437 + ], + "score": 1.0, + "content": "’s parameters on the non-label party side. This example-aware and example-specific nature", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "of the communicated gradient matrix makes differential privacy (which focuses on anonymizing an", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 445, + 504, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 504, + 458 + ], + "score": 1.0, + "content": "example’s participation in an aggregate function) inapplicable for this problem (see also Section 2).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 107, + 465, + 340, + 476 + ], + "lines": [ + { + "bbox": [ + 105, + 464, + 342, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 342, + 478 + ], + "score": 1.0, + "content": "3.2 THREAT MODEL AND PRIVACY QUANTIFICATION", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 482, + 504, + 505 + ], + "lines": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "Below we specify several key aspects of our threat model, including the adversary’s objective and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 493, + 502, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 502, + 506 + ], + "score": 1.0, + "content": "capabilities, our metric for quantifying privacy loss, and the possible inclusion of side information.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 554 + ], + "lines": [ + { + "bbox": [ + 106, + 510, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 403, + 522 + ], + "score": 1.0, + "content": "Adversary’s objective. At a given moment in time during training (with", + "type": "text" + }, + { + "bbox": [ + 404, + 511, + 411, + 522 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 510, + 429, + 522 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 430, + 510, + 437, + 520 + ], + "score": 0.75, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 510, + 505, + 522 + ], + "score": 1.0, + "content": "fixed), since the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 521, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 241, + 533 + ], + "score": 1.0, + "content": "communicated cut layer gradient", + "type": "text" + }, + { + "bbox": [ + 241, + 523, + 248, + 532 + ], + "score": 0.81, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 521, + 373, + 533 + ], + "score": 1.0, + "content": "is a deterministic function of", + "type": "text" + }, + { + "bbox": [ + 374, + 523, + 380, + 532 + ], + "score": 0.77, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 521, + 505, + 533 + ], + "score": 1.0, + "content": "(see Section 3.1), we consider", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 460, + 545 + ], + "score": 1.0, + "content": "an adversarial non-label party whose objective is to recover the label party’s hidden label", + "type": "text" + }, + { + "bbox": [ + 460, + 533, + 467, + 544 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "based on", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 543, + 340, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 222, + 555 + ], + "score": 1.0, + "content": "the information contained in", + "type": "text" + }, + { + "bbox": [ + 222, + 545, + 228, + 555 + ], + "score": 0.8, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 543, + 340, + 555 + ], + "score": 1.0, + "content": "for every training example.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 106, + 560, + 505, + 663 + ], + "lines": [ + { + "bbox": [ + 106, + 559, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 559, + 505, + 573 + ], + "score": 1.0, + "content": "Adversary’s capability. We consider an honest-but-curious non-label party which cannot tamper", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 571, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 477, + 584 + ], + "score": 1.0, + "content": "with training by selecting which examples to include in a batch or sending incorrect features", + "type": "text" + }, + { + "bbox": [ + 477, + 571, + 501, + 583 + ], + "score": 0.92, + "content": "f ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 571, + 505, + 584 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 580, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 505, + 596 + ], + "score": 1.0, + "content": "instead, we assume that the adversary follows the agreed-upon split training procedure while trying", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 592, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 177, + 606 + ], + "score": 1.0, + "content": "to guess the label", + "type": "text" + }, + { + "bbox": [ + 178, + 594, + 184, + 604 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 592, + 505, + 606 + ], + "score": 1.0, + "content": ". This can be viewed as a binary classification problem where the (input, output)", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 272, + 617 + ], + "score": 1.0, + "content": "distribution is the induced distribution of", + "type": "text" + }, + { + "bbox": [ + 272, + 604, + 295, + 616 + ], + "score": 0.91, + "content": "( g , y )", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 603, + 505, + 617 + ], + "score": 1.0, + "content": ". We allow the adversary to use any binary classifier", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 614, + 507, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 178, + 628 + ], + "score": 0.93, + "content": "q : \\mathbb { R } ^ { d } \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 614, + 495, + 631 + ], + "score": 1.0, + "content": "to guess the labels. This classifier can be represented by a (scoring function", + "type": "text" + }, + { + "bbox": [ + 495, + 619, + 501, + 627 + ], + "score": 0.63, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 614, + 507, + 631 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 627, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 146, + 642 + ], + "score": 1.0, + "content": "threshold", + "type": "text" + }, + { + "bbox": [ + 146, + 630, + 152, + 640 + ], + "score": 0.62, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 628, + 209, + 642 + ], + "score": 1.0, + "content": ") tuple, where", + "type": "text" + }, + { + "bbox": [ + 209, + 627, + 261, + 640 + ], + "score": 0.91, + "content": "r : \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 628, + 505, + 642 + ], + "score": 1.0, + "content": "maps an example’s cut layer gradient to a real-valued score", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 178, + 653 + ], + "score": 1.0, + "content": "and the threshold", + "type": "text" + }, + { + "bbox": [ + 178, + 640, + 203, + 650 + ], + "score": 0.91, + "content": "t \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 640, + 316, + 653 + ], + "score": 1.0, + "content": "determines a cut-off so that", + "type": "text" + }, + { + "bbox": [ + 316, + 640, + 353, + 652 + ], + "score": 0.92, + "content": "\\bar { q ( g ) } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 640, + 363, + 653 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 363, + 640, + 399, + 652 + ], + "score": 0.93, + "content": "\\overset { \\cdot } { r } ( g ) > t", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 640, + 417, + 653 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 418, + 640, + 455, + 652 + ], + "score": 0.93, + "content": "q ( g ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 640, + 465, + 653 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 465, + 640, + 501, + 652 + ], + "score": 0.92, + "content": "r ( g ) \\leq t", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 640, + 505, + 653 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 651, + 504, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 504, + 663 + ], + "score": 1.0, + "content": "Moving forward, we use this tuple representation to describe adversarial non-label party classifiers.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 668, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "score": 1.0, + "content": "Privacy loss quantification. As we consider binary classification, a natural metric to quantify", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 680, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 317, + 691 + ], + "score": 1.0, + "content": "the performance of an adverary’s scoring function", + "type": "text" + }, + { + "bbox": [ + 317, + 681, + 324, + 689 + ], + "score": 0.75, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 680, + 505, + 691 + ], + "score": 1.0, + "content": "is the AUC of its ROC curve. Denote the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 690, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 409, + 704 + ], + "score": 1.0, + "content": "unperturbed class-conditional distributions of the cut-layer gradients by", + "type": "text" + }, + { + "bbox": [ + 410, + 690, + 429, + 701 + ], + "score": 0.9, + "content": "P ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 690, + 451, + 704 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 451, + 690, + 471, + 701 + ], + "score": 0.9, + "content": "P ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 690, + 505, + 704 + ], + "score": 1.0, + "content": "for the", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 712, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 117, + 708, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 117, + 708, + 506, + 725 + ], + "score": 1.0, + "content": "2To simplify notation, we assume no additional features in the label party to compute the logit. The data", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "leakage problem still holds true for other more complicated settings (see WDL experiment setting in Section 5).", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 320, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 79, + 321, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 79, + 321, + 96 + ], + "score": 1.0, + "content": "3 LABEL LEAKAGE IN SPLIT LEARNING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 102, + 504, + 125 + ], + "lines": [ + { + "bbox": [ + 106, + 102, + 505, + 115 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 505, + 115 + ], + "score": 1.0, + "content": "We first introduce the two-party split learning problem for binary classification, and then formally", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 113, + 490, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 490, + 127 + ], + "score": 1.0, + "content": "describe our threat model and privacy quantification metrics with two concrete attack examples.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 102, + 505, + 127 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 135, + 382, + 146 + ], + "lines": [ + { + "bbox": [ + 105, + 135, + 383, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 383, + 148 + ], + "score": 1.0, + "content": "3.1 TWO-PARTY SPLIT LEARNING IN BINARY CLASSIFICATION", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 152, + 505, + 244 + ], + "lines": [ + { + "bbox": [ + 104, + 151, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 104, + 151, + 396, + 166 + ], + "score": 1.0, + "content": "Problem setup. Consider two parties learning a composition model", + "type": "text" + }, + { + "bbox": [ + 396, + 153, + 421, + 164 + ], + "score": 0.9, + "content": "h \\circ f", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 151, + 505, + 166 + ], + "score": 1.0, + "content": "jointly for a binary", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 162, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 264, + 177 + ], + "score": 1.0, + "content": "classification problem over the domain", + "type": "text" + }, + { + "bbox": [ + 265, + 163, + 311, + 176 + ], + "score": 0.93, + "content": "\\mathcal { X } \\times \\{ 0 , \\bar { 1 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 162, + 505, + 177 + ], + "score": 1.0, + "content": "(Figure 1). The non-label party owns the repre-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 174, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 104, + 174, + 181, + 189 + ], + "score": 1.0, + "content": "sentation function", + "type": "text" + }, + { + "bbox": [ + 181, + 174, + 234, + 187 + ], + "score": 0.92, + "content": "f : \\mathcal { X } \\to \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 174, + 363, + 189 + ], + "score": 1.0, + "content": "and each example’s raw feature", + "type": "text" + }, + { + "bbox": [ + 364, + 176, + 396, + 187 + ], + "score": 0.9, + "content": "X \\in { \\mathcal { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 174, + 505, + 189 + ], + "score": 1.0, + "content": "while the label party owns", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 187, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 181, + 201 + ], + "score": 1.0, + "content": "the logit function", + "type": "text" + }, + { + "bbox": [ + 182, + 187, + 242, + 199 + ], + "score": 0.89, + "content": "h : \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 187, + 352, + 201 + ], + "score": 1.0, + "content": "and each example’s label", + "type": "text" + }, + { + "bbox": [ + 353, + 188, + 404, + 201 + ], + "score": 0.93, + "content": "y \\in \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 187, + 430, + 201 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 430, + 188, + 490, + 201 + ], + "score": 0.92, + "content": "\\ell \\ = \\ h ( f ( X ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 187, + 505, + 201 + ], + "score": 1.0, + "content": "be", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 199, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 505, + 213 + ], + "score": 1.0, + "content": "the logit of the positive class whose predicted probability is given through the sigmoid function:", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 210, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 205, + 222 + ], + "score": 0.88, + "content": "\\widetilde { p } _ { 1 } = 1 / ( 1 + \\exp ( - \\ell ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 210, + 506, + 224 + ], + "score": 1.0, + "content": ". We measure the loss of such prediction through the cross entropy loss", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 221, + 504, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 222, + 252, + 234 + ], + "score": 0.87, + "content": "L = \\log ( 1 + \\exp ( - \\ell ) ) + ( 1 - y ) \\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 221, + 480, + 235 + ], + "score": 1.0, + "content": ". During model inference, the non-label party computes", + "type": "text" + }, + { + "bbox": [ + 480, + 221, + 504, + 234 + ], + "score": 0.91, + "content": "{ \\dot { \\boldsymbol { f } } } ( { \\boldsymbol { X } } )", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 233, + 493, + 245 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 493, + 245 + ], + "score": 1.0, + "content": "and sends it to the label party who will then execute the rest of forward computation in Figure 1.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 7.5, + "bbox_fs": [ + 104, + 151, + 506, + 245 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 249, + 505, + 340 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 506, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 506, + 263 + ], + "score": 1.0, + "content": "Model training (Figure 1: backward gradient computation). To train the model using gradient", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 260, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 397, + 273 + ], + "score": 1.0, + "content": "descent, the label party starts by first computing the gradient of the loss", + "type": "text" + }, + { + "bbox": [ + 398, + 261, + 406, + 270 + ], + "score": 0.72, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 260, + 505, + 273 + ], + "score": 1.0, + "content": "with respect to the logit", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 107, + 270, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 107, + 271, + 173, + 284 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\frac { d L } { d \\ell } \\ = \\ ( \\widetilde { p } _ { 1 } - y ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 270, + 474, + 285 + ], + "score": 1.0, + "content": ". Using the chain rule, the label party can then compute the gradient of", + "type": "text" + }, + { + "bbox": [ + 474, + 272, + 483, + 281 + ], + "score": 0.79, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 270, + 505, + 285 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 283, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 195, + 294 + ], + "score": 1.0, + "content": "erespect to its function", + "type": "text" + }, + { + "bbox": [ + 195, + 283, + 202, + 292 + ], + "score": 0.79, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 283, + 505, + 294 + ], + "score": 1.0, + "content": "’s parameters and perform the gradient updates. To also allow the non-label", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 293, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 104, + 293, + 210, + 306 + ], + "score": 1.0, + "content": "party to learn its function", + "type": "text" + }, + { + "bbox": [ + 210, + 294, + 217, + 304 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 293, + 506, + 306 + ], + "score": 1.0, + "content": ", the label party needs to additionally compute the gradient with respect", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 304, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 187, + 317 + ], + "score": 1.0, + "content": "to cut layer feature", + "type": "text" + }, + { + "bbox": [ + 187, + 304, + 211, + 316 + ], + "score": 0.92, + "content": "f ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 304, + 505, + 317 + ], + "score": 1.0, + "content": "and communicate it to the non-label party. We denote this gradient by", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 314, + 507, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 304, + 330 + ], + "score": 0.86, + "content": "g : = \\nabla _ { f ( X ) } L = ( \\widetilde { p } _ { 1 } - y ) \\nabla _ { z } h ( z ) | _ { z = f ( X ) } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 314, + 438, + 330 + ], + "score": 1.0, + "content": "(by chain rule). After receiving", + "type": "text" + }, + { + "bbox": [ + 438, + 318, + 445, + 329 + ], + "score": 0.7, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 314, + 507, + 330 + ], + "score": 1.0, + "content": ", the non-label", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 287, + 341 + ], + "score": 1.0, + "content": "eparty continues the backpropagation towards", + "type": "text" + }, + { + "bbox": [ + 288, + 329, + 294, + 339 + ], + "score": 0.81, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 328, + 505, + 341 + ], + "score": 1.0, + "content": "’s parameters and also perform the gradient updates.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5, + "bbox_fs": [ + 104, + 248, + 507, + 341 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 344, + 505, + 457 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "score": 1.0, + "content": "Why Not Differential Privacy? Note that for a given iteration, the non-label party randomly", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 356, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 141, + 367 + ], + "score": 1.0, + "content": "chooses", + "type": "text" + }, + { + "bbox": [ + 142, + 356, + 151, + 366 + ], + "score": 0.77, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 356, + 505, + 367 + ], + "score": 1.0, + "content": "example IDs to form a batch. Therefore, the identity of which examples are used is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 366, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 427, + 380 + ], + "score": 1.0, + "content": "known to the non-label party by default. In addition, the communicated features", + "type": "text" + }, + { + "bbox": [ + 428, + 367, + 451, + 379 + ], + "score": 0.95, + "content": "f ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 366, + 505, + 380 + ], + "score": 1.0, + "content": "and returned", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 103, + 375, + 507, + 394 + ], + "spans": [ + { + "bbox": [ + 103, + 375, + 145, + 394 + ], + "score": 1.0, + "content": "gradients", + "type": "text" + }, + { + "bbox": [ + 145, + 381, + 151, + 391 + ], + "score": 0.78, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 375, + 245, + 394 + ], + "score": 1.0, + "content": "will both be matrices in", + "type": "text" + }, + { + "bbox": [ + 246, + 378, + 271, + 390 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { B \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 375, + 507, + 394 + ], + "score": 1.0, + "content": "with each row belonging to a specific example in the batch.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 389, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 506, + 403 + ], + "score": 1.0, + "content": "The different gradients (rows of the matrix) are not with respect to the same model parameters, but", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "score": 1.0, + "content": "are instead with respect to different examples’ cut-layer features; thus, no averaging over or shuffling", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 411, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 505, + 426 + ], + "score": 1.0, + "content": "of the rows of the gradient matrix can be done prior to communication to ensure correct computa-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 422, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 135, + 437 + ], + "score": 1.0, + "content": "tion of", + "type": "text" + }, + { + "bbox": [ + 135, + 424, + 142, + 435 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 422, + 505, + 437 + ], + "score": 1.0, + "content": "’s parameters on the non-label party side. This example-aware and example-specific nature", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "of the communicated gradient matrix makes differential privacy (which focuses on anonymizing an", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 445, + 504, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 504, + 458 + ], + "score": 1.0, + "content": "example’s participation in an aggregate function) inapplicable for this problem (see also Section 2).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5, + "bbox_fs": [ + 103, + 344, + 507, + 458 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 465, + 340, + 476 + ], + "lines": [ + { + "bbox": [ + 105, + 464, + 342, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 342, + 478 + ], + "score": 1.0, + "content": "3.2 THREAT MODEL AND PRIVACY QUANTIFICATION", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 482, + 504, + 505 + ], + "lines": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "Below we specify several key aspects of our threat model, including the adversary’s objective and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 493, + 502, + 506 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 502, + 506 + ], + "score": 1.0, + "content": "capabilities, our metric for quantifying privacy loss, and the possible inclusion of side information.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 106, + 482, + 505, + 506 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 554 + ], + "lines": [ + { + "bbox": [ + 106, + 510, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 403, + 522 + ], + "score": 1.0, + "content": "Adversary’s objective. At a given moment in time during training (with", + "type": "text" + }, + { + "bbox": [ + 404, + 511, + 411, + 522 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 510, + 429, + 522 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 430, + 510, + 437, + 520 + ], + "score": 0.75, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 510, + 505, + 522 + ], + "score": 1.0, + "content": "fixed), since the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 521, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 241, + 533 + ], + "score": 1.0, + "content": "communicated cut layer gradient", + "type": "text" + }, + { + "bbox": [ + 241, + 523, + 248, + 532 + ], + "score": 0.81, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 521, + 373, + 533 + ], + "score": 1.0, + "content": "is a deterministic function of", + "type": "text" + }, + { + "bbox": [ + 374, + 523, + 380, + 532 + ], + "score": 0.77, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 521, + 505, + 533 + ], + "score": 1.0, + "content": "(see Section 3.1), we consider", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 460, + 545 + ], + "score": 1.0, + "content": "an adversarial non-label party whose objective is to recover the label party’s hidden label", + "type": "text" + }, + { + "bbox": [ + 460, + 533, + 467, + 544 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "based on", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 543, + 340, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 222, + 555 + ], + "score": 1.0, + "content": "the information contained in", + "type": "text" + }, + { + "bbox": [ + 222, + 545, + 228, + 555 + ], + "score": 0.8, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 543, + 340, + 555 + ], + "score": 1.0, + "content": "for every training example.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 510, + 505, + 555 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 560, + 505, + 663 + ], + "lines": [ + { + "bbox": [ + 106, + 559, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 559, + 505, + 573 + ], + "score": 1.0, + "content": "Adversary’s capability. We consider an honest-but-curious non-label party which cannot tamper", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 571, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 477, + 584 + ], + "score": 1.0, + "content": "with training by selecting which examples to include in a batch or sending incorrect features", + "type": "text" + }, + { + "bbox": [ + 477, + 571, + 501, + 583 + ], + "score": 0.92, + "content": "f ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 571, + 505, + 584 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 580, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 505, + 596 + ], + "score": 1.0, + "content": "instead, we assume that the adversary follows the agreed-upon split training procedure while trying", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 592, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 177, + 606 + ], + "score": 1.0, + "content": "to guess the label", + "type": "text" + }, + { + "bbox": [ + 178, + 594, + 184, + 604 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 592, + 505, + 606 + ], + "score": 1.0, + "content": ". This can be viewed as a binary classification problem where the (input, output)", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 272, + 617 + ], + "score": 1.0, + "content": "distribution is the induced distribution of", + "type": "text" + }, + { + "bbox": [ + 272, + 604, + 295, + 616 + ], + "score": 0.91, + "content": "( g , y )", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 603, + 505, + 617 + ], + "score": 1.0, + "content": ". We allow the adversary to use any binary classifier", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 614, + 507, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 178, + 628 + ], + "score": 0.93, + "content": "q : \\mathbb { R } ^ { d } \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 614, + 495, + 631 + ], + "score": 1.0, + "content": "to guess the labels. This classifier can be represented by a (scoring function", + "type": "text" + }, + { + "bbox": [ + 495, + 619, + 501, + 627 + ], + "score": 0.63, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 614, + 507, + 631 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 627, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 146, + 642 + ], + "score": 1.0, + "content": "threshold", + "type": "text" + }, + { + "bbox": [ + 146, + 630, + 152, + 640 + ], + "score": 0.62, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 628, + 209, + 642 + ], + "score": 1.0, + "content": ") tuple, where", + "type": "text" + }, + { + "bbox": [ + 209, + 627, + 261, + 640 + ], + "score": 0.91, + "content": "r : \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 628, + 505, + 642 + ], + "score": 1.0, + "content": "maps an example’s cut layer gradient to a real-valued score", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 178, + 653 + ], + "score": 1.0, + "content": "and the threshold", + "type": "text" + }, + { + "bbox": [ + 178, + 640, + 203, + 650 + ], + "score": 0.91, + "content": "t \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 640, + 316, + 653 + ], + "score": 1.0, + "content": "determines a cut-off so that", + "type": "text" + }, + { + "bbox": [ + 316, + 640, + 353, + 652 + ], + "score": 0.92, + "content": "\\bar { q ( g ) } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 640, + 363, + 653 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 363, + 640, + 399, + 652 + ], + "score": 0.93, + "content": "\\overset { \\cdot } { r } ( g ) > t", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 640, + 417, + 653 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 418, + 640, + 455, + 652 + ], + "score": 0.93, + "content": "q ( g ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 640, + 465, + 653 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 465, + 640, + 501, + 652 + ], + "score": 0.92, + "content": "r ( g ) \\leq t", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 640, + 505, + 653 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 651, + 504, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 504, + 663 + ], + "score": 1.0, + "content": "Moving forward, we use this tuple representation to describe adversarial non-label party classifiers.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 559, + 507, + 663 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 668, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "score": 1.0, + "content": "Privacy loss quantification. As we consider binary classification, a natural metric to quantify", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 680, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 317, + 691 + ], + "score": 1.0, + "content": "the performance of an adverary’s scoring function", + "type": "text" + }, + { + "bbox": [ + 317, + 681, + 324, + 689 + ], + "score": 0.75, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 680, + 505, + 691 + ], + "score": 1.0, + "content": "is the AUC of its ROC curve. Denote the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 690, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 409, + 704 + ], + "score": 1.0, + "content": "unperturbed class-conditional distributions of the cut-layer gradients by", + "type": "text" + }, + { + "bbox": [ + 410, + 690, + 429, + 701 + ], + "score": 0.9, + "content": "P ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 690, + 451, + 704 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 451, + 690, + 471, + 701 + ], + "score": 0.9, + "content": "P ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 690, + 505, + 704 + ], + "score": 1.0, + "content": "for the", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 439, + 284 + ], + "score": 1.0, + "content": "positive and negative class, respectively. The ROC curve of a scoring function", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 439, + 274, + 446, + 281 + ], + "score": 0.78, + "content": "r", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 446, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "is a paramet-", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 281, + 504, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 146, + 295 + ], + "score": 1.0, + "content": "ric curve", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 146, + 282, + 291, + 294 + ], + "score": 0.89, + "content": "t \\mapsto \\mathsf { \\bar { ( F P R } } _ { r } ( t ) , \\mathsf { T P R } _ { r } ( t ) ) \\in \\mathsf { [ 0 , 1 ] } ^ { 2 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 292, + 281, + 420, + 295 + ], + "score": 1.0, + "content": "which maps a threshold value", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 420, + 283, + 449, + 293 + ], + "score": 0.91, + "content": "t \\in \\mathbb { R }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 450, + 281, + 504, + 295 + ], + "score": 1.0, + "content": "to the corre-", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 292, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 460, + 307 + ], + "score": 1.0, + "content": "sponding (False Positive Rate, True Positive Rate) tuple of the classifier represented by", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 460, + 293, + 480, + 305 + ], + "score": 0.91, + "content": "( r , t )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 480, + 292, + 506, + 307 + ], + "score": 1.0, + "content": ", with", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 303, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 244, + 318 + ], + "score": 0.92, + "content": "\\mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } ( \\{ g : r ( g ) > t \\} )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 245, + 303, + 265, + 320 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 265, + 304, + 403, + 318 + ], + "score": 0.92, + "content": "\\mathrm { T P R } _ { r } ( t ) : = P ^ { ( 1 ) } ( \\{ g : r ( g ) > t \\} )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 403, + 303, + 506, + 320 + ], + "score": 1.0, + "content": ". The AUC of the ROC", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 316, + 435, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 216, + 330 + ], + "score": 1.0, + "content": "curve of a scoring function", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 217, + 319, + 222, + 326 + ], + "score": 0.74, + "content": "r", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 222, + 316, + 268, + 330 + ], + "score": 1.0, + "content": "(denote by", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 269, + 317, + 302, + 329 + ], + "score": 0.65, + "content": "\\operatorname { A U C } ( r )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 303, + 316, + 435, + 330 + ], + "score": 1.0, + "content": ") can be expressed as an integral:", + "type": "text", + "cross_page": true + } + ], + "index": 9 + } + ], + "index": 47, + "bbox_fs": [ + 105, + 667, + 505, + 704 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 79, + 491, + 239 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 79, + 491, + 239 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 79, + 491, + 239 + ], + "spans": [ + { + "bbox": [ + 107, + 79, + 491, + 239 + ], + "score": 0.974, + "type": "image", + "image_path": "abe502d96660f12b083760acfe08c4529f2a3f7382efbc95190a0a239c34db31.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 79, + 491, + 132.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 132.33333333333334, + 491, + 185.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 185.66666666666669, + 491, + 239.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 244, + 504, + 265 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 244, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 505, + 257 + ], + "score": 1.0, + "content": "Figure 2: Distributions of quantities discussed in Observations 1-4 after the first 100, 200, 300, 400, 500 steps", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 255, + 418, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 418, + 266 + ], + "score": 1.0, + "content": "of stochastic gradient descent training of the WDL model on Criteo (see experiments).", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 271, + 505, + 328 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 439, + 284 + ], + "score": 1.0, + "content": "positive and negative class, respectively. The ROC curve of a scoring function", + "type": "text" + }, + { + "bbox": [ + 439, + 274, + 446, + 281 + ], + "score": 0.78, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "is a paramet-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 281, + 504, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 146, + 295 + ], + "score": 1.0, + "content": "ric curve", + "type": "text" + }, + { + "bbox": [ + 146, + 282, + 291, + 294 + ], + "score": 0.89, + "content": "t \\mapsto \\mathsf { \\bar { ( F P R } } _ { r } ( t ) , \\mathsf { T P R } _ { r } ( t ) ) \\in \\mathsf { [ 0 , 1 ] } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 281, + 420, + 295 + ], + "score": 1.0, + "content": "which maps a threshold value", + "type": "text" + }, + { + "bbox": [ + 420, + 283, + 449, + 293 + ], + "score": 0.91, + "content": "t \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 281, + 504, + 295 + ], + "score": 1.0, + "content": "to the corre-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 292, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 460, + 307 + ], + "score": 1.0, + "content": "sponding (False Positive Rate, True Positive Rate) tuple of the classifier represented by", + "type": "text" + }, + { + "bbox": [ + 460, + 293, + 480, + 305 + ], + "score": 0.91, + "content": "( r , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 292, + 506, + 307 + ], + "score": 1.0, + "content": ", with", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 303, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 244, + 318 + ], + "score": 0.92, + "content": "\\mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } ( \\{ g : r ( g ) > t \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 303, + 265, + 320 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 265, + 304, + 403, + 318 + ], + "score": 0.92, + "content": "\\mathrm { T P R } _ { r } ( t ) : = P ^ { ( 1 ) } ( \\{ g : r ( g ) > t \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 303, + 506, + 320 + ], + "score": 1.0, + "content": ". The AUC of the ROC", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 316, + 435, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 216, + 330 + ], + "score": 1.0, + "content": "curve of a scoring function", + "type": "text" + }, + { + "bbox": [ + 217, + 319, + 222, + 326 + ], + "score": 0.74, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 316, + 268, + 330 + ], + "score": 1.0, + "content": "(denote by", + "type": "text" + }, + { + "bbox": [ + 269, + 317, + 302, + 329 + ], + "score": 0.65, + "content": "\\operatorname { A U C } ( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 316, + 435, + 330 + ], + "score": 1.0, + "content": ") can be expressed as an integral:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 213, + 329, + 398, + 345 + ], + "lines": [ + { + "bbox": [ + 213, + 329, + 398, + 345 + ], + "spans": [ + { + "bbox": [ + 213, + 329, + 398, + 345 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\begin{array} { r l } { \\mathrm { A U C } ( r ) = \\int _ { \\infty } ^ { - \\infty } \\mathrm { T P R } _ { r } ( t ) ~ d \\mathrm { F P R } _ { r } ( t ) } & { \\in [ 0 , 1 ] } \\end{array} } \\end{array}", + "type": "interline_equation", + "image_path": "68d0383ca478985c6e65689e00531f27f0cc7638222d2600d346fc31680de898.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 213, + 329, + 398, + 345 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 451, + 331, + 504, + 343 + ], + "lines": [ + { + "bbox": [ + 450, + 330, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 450, + 330, + 505, + 346 + ], + "score": 1.0, + "content": "(Leak AUC)", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 346, + 505, + 434 + ], + "lines": [ + { + "bbox": [ + 106, + 345, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 505, + 359 + ], + "score": 1.0, + "content": "(more details on this expression see Appendix A.1.) We use this value as the privacy loss quantifi-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 357, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 324, + 369 + ], + "score": 1.0, + "content": "cation metric for a specific adversary scoring function", + "type": "text" + }, + { + "bbox": [ + 324, + 359, + 330, + 367 + ], + "score": 0.78, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 357, + 505, + 369 + ], + "score": 1.0, + "content": "and refer to it as the leak AUC. This metric", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 368, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 505, + 380 + ], + "score": 1.0, + "content": "summarizes the predictive performance of all classifiers that can be constructed through all threshold", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 378, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 104, + 378, + 134, + 392 + ], + "score": 1.0, + "content": "values", + "type": "text" + }, + { + "bbox": [ + 135, + 380, + 140, + 389 + ], + "score": 0.69, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 378, + 505, + 392 + ], + "score": 1.0, + "content": "and removes the need to tune this classifier-specific hyperparameter. The leak AUC being", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 389, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 340, + 402 + ], + "score": 1.0, + "content": "close to 1 implies that the corresponding scoring function", + "type": "text" + }, + { + "bbox": [ + 340, + 392, + 347, + 399 + ], + "score": 0.67, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 389, + 505, + 402 + ], + "score": 1.0, + "content": "can very accurately recover the private", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 279, + 414 + ], + "score": 1.0, + "content": "label, whereas a value of around 0.5 means", + "type": "text" + }, + { + "bbox": [ + 279, + 403, + 285, + 411 + ], + "score": 0.76, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 399, + 505, + 414 + ], + "score": 1.0, + "content": "is non-informative in predicting the labels. In practice,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 411, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 104, + 411, + 266, + 425 + ], + "score": 1.0, + "content": "during batch training, the leak AUC of", + "type": "text" + }, + { + "bbox": [ + 266, + 413, + 273, + 421 + ], + "score": 0.72, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 411, + 505, + 425 + ], + "score": 1.0, + "content": "can be estimated at every gradient update iteration using", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 423, + 355, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 355, + 435 + ], + "score": 1.0, + "content": "the minibatch of cut-layer gradients together with their labels.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 439, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "Side information. Among all the scoring functions within our threat model, it is conceivable that", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "only some would recover the hidden labels accurately. Picking such effective ones would require", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "score": 1.0, + "content": "the non-label party to have population-level side information specifically regarding the properties of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 471, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 506, + 486 + ], + "score": 1.0, + "content": "(and distinction between) the positive and negative class’s cut-layer gradient distributions. Since we", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "allow the adversary to pick any specific (measurable) scoring function, we implicitly allow for such", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 494, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 507 + ], + "score": 1.0, + "content": "population-level side information for the adversary. However, we assume the non-label party has no", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "score": 1.0, + "content": "example-level side information that is different example by example. Thus we also don’t use local", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "DP for privacy quantification (detailed explanation in Appendix A.8). Next we provide two example", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 527, + 483, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 483, + 540 + ], + "score": 1.0, + "content": "scoring functions which use population-level side-information to effectively recover the label.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 108, + 546, + 264, + 557 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 266, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 266, + 558 + ], + "score": 1.0, + "content": "3.3 PRACTICAL ATTACK METHODS", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 561, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 105, + 559, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 326, + 576 + ], + "score": 1.0, + "content": "Attack 1: Norm-based scoring function. Note that", + "type": "text" + }, + { + "bbox": [ + 326, + 561, + 482, + 574 + ], + "score": 0.93, + "content": "\\| g \\| _ { 2 } = | \\widetilde p _ { 1 } - y | \\cdot \\| \\nabla _ { a } h ( a ) | _ { a = \\pmb { f } ( X ) } \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 559, + 506, + 576 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 572, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 256, + 587 + ], + "score": 1.0, + "content": "make the following observations for", + "type": "text" + }, + { + "bbox": [ + 257, + 573, + 290, + 586 + ], + "score": 0.93, + "content": "| \\widetilde { p } _ { 1 } - y |", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 572, + 309, + 587 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 310, + 574, + 389, + 586 + ], + "score": 0.92, + "content": "\\| \\nabla _ { a } h ( a ) | _ { a = f ( X ) } \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 572, + 505, + 587 + ], + "score": 1.0, + "content": ", which hold true for a wide", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 585, + 264, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 264, + 597 + ], + "score": 1.0, + "content": "range of real-world learning problems.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 112, + 596, + 505, + 696 + ], + "lines": [ + { + "bbox": [ + 113, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 113, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "• Observation 1: Throughout training, the model tends to be less confident about a positive ex-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 120, + 607, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 120, + 607, + 505, + 621 + ], + "score": 1.0, + "content": "ample being positive than a negative example being negative. In other words, the confidence gap", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 120, + 618, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 120, + 618, + 208, + 632 + ], + "score": 1.0, + "content": "of a positive example", + "type": "text" + }, + { + "bbox": [ + 209, + 619, + 280, + 630 + ], + "score": 0.91, + "content": "1 - \\widetilde { p } _ { 1 } = \\left| \\widetilde { p } _ { 1 } - y \\right|", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 618, + 308, + 632 + ], + "score": 1.0, + "content": "(when", + "type": "text" + }, + { + "bbox": [ + 308, + 619, + 334, + 630 + ], + "score": 0.9, + "content": "y = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 618, + 505, + 632 + ], + "score": 1.0, + "content": ") is typically larger than the confidence gap", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 120, + 629, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 120, + 629, + 210, + 644 + ], + "score": 1.0, + "content": "of a negative example", + "type": "text" + }, + { + "bbox": [ + 211, + 630, + 306, + 641 + ], + "score": 0.9, + "content": "1 - \\widetilde { p } _ { 0 } = \\widetilde { p } _ { 1 } = | \\widetilde { p } _ { 1 } - y |", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 629, + 335, + 644 + ], + "score": 1.0, + "content": "(when", + "type": "text" + }, + { + "bbox": [ + 335, + 631, + 360, + 641 + ], + "score": 0.89, + "content": "y = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 629, + 506, + 644 + ], + "score": 1.0, + "content": ") (see Figure 2(a)). This observation", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 120, + 641, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 120, + 641, + 505, + 653 + ], + "score": 1.0, + "content": "e e eis particularly true for problems like advertising conversion prediction and disease prediction,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 120, + 651, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 120, + 651, + 505, + 664 + ], + "score": 1.0, + "content": "where there is inherently more ambiguity for the positive class than the negative. For example,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 120, + 663, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 120, + 663, + 506, + 675 + ], + "score": 1.0, + "content": "in advertising, uninterested users of a product will never click on its ad and convert, but those", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 120, + 673, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 120, + 673, + 505, + 686 + ], + "score": 1.0, + "content": "interested, even after clicking, might make the purchase only a fraction of the time depending", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 121, + 685, + 497, + 697 + ], + "spans": [ + { + "bbox": [ + 121, + 685, + 497, + 697 + ], + "score": 1.0, + "content": "on time/money constraints. (See A.2 for such an ambiguity even for a class-balanced setting.)", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 113, + 699, + 505, + 733 + ], + "lines": [ + { + "bbox": [ + 111, + 697, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 111, + 697, + 414, + 713 + ], + "score": 1.0, + "content": "• Observation 2: Throughout training, the norm of the gradient vector", + "type": "text" + }, + { + "bbox": [ + 415, + 698, + 493, + 712 + ], + "score": 0.93, + "content": "\\| \\nabla _ { z } h ( z ) \\vert _ { z = f ( X ) } \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 697, + 506, + 713 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 121, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 121, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "on the same order of magnitude (has similar distribution) for both the positive and negative", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 120, + 720, + 468, + 735 + ], + "spans": [ + { + "bbox": [ + 120, + 720, + 313, + 735 + ], + "score": 1.0, + "content": "examples (Figure 2(b)). This is natural because", + "type": "text" + }, + { + "bbox": [ + 313, + 721, + 377, + 733 + ], + "score": 0.93, + "content": "\\nabla _ { a } h ( a ) | _ { a = f ( X ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 720, + 456, + 735 + ], + "score": 1.0, + "content": "is not a function of", + "type": "text" + }, + { + "bbox": [ + 457, + 723, + 463, + 732 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 720, + 468, + 735 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 79, + 491, + 239 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 79, + 491, + 239 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 79, + 491, + 239 + ], + "spans": [ + { + "bbox": [ + 107, + 79, + 491, + 239 + ], + "score": 0.974, + "type": "image", + "image_path": "abe502d96660f12b083760acfe08c4529f2a3f7382efbc95190a0a239c34db31.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 79, + 491, + 132.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 132.33333333333334, + 491, + 185.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 185.66666666666669, + 491, + 239.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 244, + 504, + 265 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 244, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 505, + 257 + ], + "score": 1.0, + "content": "Figure 2: Distributions of quantities discussed in Observations 1-4 after the first 100, 200, 300, 400, 500 steps", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 255, + 418, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 418, + 266 + ], + "score": 1.0, + "content": "of stochastic gradient descent training of the WDL model on Criteo (see experiments).", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 271, + 505, + 328 + ], + "lines": [], + "index": 7, + "bbox_fs": [ + 105, + 271, + 506, + 330 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 213, + 329, + 398, + 345 + ], + "lines": [ + { + "bbox": [ + 213, + 329, + 398, + 345 + ], + "spans": [ + { + "bbox": [ + 213, + 329, + 398, + 345 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\begin{array} { r l } { \\mathrm { A U C } ( r ) = \\int _ { \\infty } ^ { - \\infty } \\mathrm { T P R } _ { r } ( t ) ~ d \\mathrm { F P R } _ { r } ( t ) } & { \\in [ 0 , 1 ] } \\end{array} } \\end{array}", + "type": "interline_equation", + "image_path": "68d0383ca478985c6e65689e00531f27f0cc7638222d2600d346fc31680de898.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 213, + 329, + 398, + 345 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 451, + 331, + 504, + 343 + ], + "lines": [ + { + "bbox": [ + 450, + 330, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 450, + 330, + 505, + 346 + ], + "score": 1.0, + "content": "(Leak AUC)", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 450, + 330, + 505, + 346 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 346, + 505, + 434 + ], + "lines": [ + { + "bbox": [ + 106, + 345, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 505, + 359 + ], + "score": 1.0, + "content": "(more details on this expression see Appendix A.1.) We use this value as the privacy loss quantifi-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 357, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 324, + 369 + ], + "score": 1.0, + "content": "cation metric for a specific adversary scoring function", + "type": "text" + }, + { + "bbox": [ + 324, + 359, + 330, + 367 + ], + "score": 0.78, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 357, + 505, + 369 + ], + "score": 1.0, + "content": "and refer to it as the leak AUC. This metric", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 368, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 505, + 380 + ], + "score": 1.0, + "content": "summarizes the predictive performance of all classifiers that can be constructed through all threshold", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 378, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 104, + 378, + 134, + 392 + ], + "score": 1.0, + "content": "values", + "type": "text" + }, + { + "bbox": [ + 135, + 380, + 140, + 389 + ], + "score": 0.69, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 378, + 505, + 392 + ], + "score": 1.0, + "content": "and removes the need to tune this classifier-specific hyperparameter. The leak AUC being", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 389, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 340, + 402 + ], + "score": 1.0, + "content": "close to 1 implies that the corresponding scoring function", + "type": "text" + }, + { + "bbox": [ + 340, + 392, + 347, + 399 + ], + "score": 0.67, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 389, + 505, + 402 + ], + "score": 1.0, + "content": "can very accurately recover the private", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 279, + 414 + ], + "score": 1.0, + "content": "label, whereas a value of around 0.5 means", + "type": "text" + }, + { + "bbox": [ + 279, + 403, + 285, + 411 + ], + "score": 0.76, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 399, + 505, + 414 + ], + "score": 1.0, + "content": "is non-informative in predicting the labels. In practice,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 411, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 104, + 411, + 266, + 425 + ], + "score": 1.0, + "content": "during batch training, the leak AUC of", + "type": "text" + }, + { + "bbox": [ + 266, + 413, + 273, + 421 + ], + "score": 0.72, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 411, + 505, + 425 + ], + "score": 1.0, + "content": "can be estimated at every gradient update iteration using", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 423, + 355, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 355, + 435 + ], + "score": 1.0, + "content": "the minibatch of cut-layer gradients together with their labels.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5, + "bbox_fs": [ + 104, + 345, + 505, + 435 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 439, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "Side information. Among all the scoring functions within our threat model, it is conceivable that", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "only some would recover the hidden labels accurately. Picking such effective ones would require", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "score": 1.0, + "content": "the non-label party to have population-level side information specifically regarding the properties of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 471, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 506, + 486 + ], + "score": 1.0, + "content": "(and distinction between) the positive and negative class’s cut-layer gradient distributions. Since we", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "allow the adversary to pick any specific (measurable) scoring function, we implicitly allow for such", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 494, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 507 + ], + "score": 1.0, + "content": "population-level side information for the adversary. However, we assume the non-label party has no", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "score": 1.0, + "content": "example-level side information that is different example by example. Thus we also don’t use local", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 529 + ], + "score": 1.0, + "content": "DP for privacy quantification (detailed explanation in Appendix A.8). Next we provide two example", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 527, + 483, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 483, + 540 + ], + "score": 1.0, + "content": "scoring functions which use population-level side-information to effectively recover the label.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 438, + 506, + 540 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 546, + 264, + 557 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 266, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 266, + 558 + ], + "score": 1.0, + "content": "3.3 PRACTICAL ATTACK METHODS", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 561, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 105, + 559, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 326, + 576 + ], + "score": 1.0, + "content": "Attack 1: Norm-based scoring function. Note that", + "type": "text" + }, + { + "bbox": [ + 326, + 561, + 482, + 574 + ], + "score": 0.93, + "content": "\\| g \\| _ { 2 } = | \\widetilde p _ { 1 } - y | \\cdot \\| \\nabla _ { a } h ( a ) | _ { a = \\pmb { f } ( X ) } \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 559, + 506, + 576 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 572, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 256, + 587 + ], + "score": 1.0, + "content": "make the following observations for", + "type": "text" + }, + { + "bbox": [ + 257, + 573, + 290, + 586 + ], + "score": 0.93, + "content": "| \\widetilde { p } _ { 1 } - y |", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 572, + 309, + 587 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 310, + 574, + 389, + 586 + ], + "score": 0.92, + "content": "\\| \\nabla _ { a } h ( a ) | _ { a = f ( X ) } \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 572, + 505, + 587 + ], + "score": 1.0, + "content": ", which hold true for a wide", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 585, + 264, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 264, + 597 + ], + "score": 1.0, + "content": "range of real-world learning problems.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 559, + 506, + 597 + ] + }, + { + "type": "text", + "bbox": [ + 112, + 596, + 505, + 696 + ], + "lines": [ + { + "bbox": [ + 113, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 113, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "• Observation 1: Throughout training, the model tends to be less confident about a positive ex-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 120, + 607, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 120, + 607, + 505, + 621 + ], + "score": 1.0, + "content": "ample being positive than a negative example being negative. In other words, the confidence gap", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 120, + 618, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 120, + 618, + 208, + 632 + ], + "score": 1.0, + "content": "of a positive example", + "type": "text" + }, + { + "bbox": [ + 209, + 619, + 280, + 630 + ], + "score": 0.91, + "content": "1 - \\widetilde { p } _ { 1 } = \\left| \\widetilde { p } _ { 1 } - y \\right|", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 618, + 308, + 632 + ], + "score": 1.0, + "content": "(when", + "type": "text" + }, + { + "bbox": [ + 308, + 619, + 334, + 630 + ], + "score": 0.9, + "content": "y = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 618, + 505, + 632 + ], + "score": 1.0, + "content": ") is typically larger than the confidence gap", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 120, + 629, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 120, + 629, + 210, + 644 + ], + "score": 1.0, + "content": "of a negative example", + "type": "text" + }, + { + "bbox": [ + 211, + 630, + 306, + 641 + ], + "score": 0.9, + "content": "1 - \\widetilde { p } _ { 0 } = \\widetilde { p } _ { 1 } = | \\widetilde { p } _ { 1 } - y |", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 629, + 335, + 644 + ], + "score": 1.0, + "content": "(when", + "type": "text" + }, + { + "bbox": [ + 335, + 631, + 360, + 641 + ], + "score": 0.89, + "content": "y = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 629, + 506, + 644 + ], + "score": 1.0, + "content": ") (see Figure 2(a)). This observation", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 120, + 641, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 120, + 641, + 505, + 653 + ], + "score": 1.0, + "content": "e e eis particularly true for problems like advertising conversion prediction and disease prediction,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 120, + 651, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 120, + 651, + 505, + 664 + ], + "score": 1.0, + "content": "where there is inherently more ambiguity for the positive class than the negative. For example,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 120, + 663, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 120, + 663, + 506, + 675 + ], + "score": 1.0, + "content": "in advertising, uninterested users of a product will never click on its ad and convert, but those", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 120, + 673, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 120, + 673, + 505, + 686 + ], + "score": 1.0, + "content": "interested, even after clicking, might make the purchase only a fraction of the time depending", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 121, + 685, + 497, + 697 + ], + "spans": [ + { + "bbox": [ + 121, + 685, + 497, + 697 + ], + "score": 1.0, + "content": "on time/money constraints. (See A.2 for such an ambiguity even for a class-balanced setting.)", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37, + "bbox_fs": [ + 113, + 596, + 506, + 697 + ] + }, + { + "type": "text", + "bbox": [ + 113, + 699, + 505, + 733 + ], + "lines": [ + { + "bbox": [ + 111, + 697, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 111, + 697, + 414, + 713 + ], + "score": 1.0, + "content": "• Observation 2: Throughout training, the norm of the gradient vector", + "type": "text" + }, + { + "bbox": [ + 415, + 698, + 493, + 712 + ], + "score": 0.93, + "content": "\\| \\nabla _ { z } h ( z ) \\vert _ { z = f ( X ) } \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 697, + 506, + 713 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 121, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 121, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "on the same order of magnitude (has similar distribution) for both the positive and negative", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 120, + 720, + 468, + 735 + ], + "spans": [ + { + "bbox": [ + 120, + 720, + 313, + 735 + ], + "score": 1.0, + "content": "examples (Figure 2(b)). This is natural because", + "type": "text" + }, + { + "bbox": [ + 313, + 721, + 377, + 733 + ], + "score": 0.93, + "content": "\\nabla _ { a } h ( a ) | _ { a = f ( X ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 720, + 456, + 735 + ], + "score": 1.0, + "content": "is not a function of", + "type": "text" + }, + { + "bbox": [ + 457, + 723, + 463, + 732 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 720, + 468, + 735 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 111, + 697, + 506, + 735 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 363, + 96 + ], + "score": 1.0, + "content": "As a consequence of Observation 1 and 2, the gradient norm", + "type": "text" + }, + { + "bbox": [ + 364, + 82, + 385, + 95 + ], + "score": 0.91, + "content": "\\| g \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "of the positive instances are", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 468, + 108 + ], + "score": 1.0, + "content": "generally larger than that of the negative ones (Figure 2(c)). Thus, the scoring function", + "type": "text" + }, + { + "bbox": [ + 468, + 93, + 505, + 106 + ], + "score": 0.91, + "content": "r _ { n } ( g ) =", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 107, + 105, + 127, + 117 + ], + "score": 0.9, + "content": "\\| g \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 104, + 294, + 117 + ], + "score": 1.0, + "content": "is a strong predictor of the unseen label", + "type": "text" + }, + { + "bbox": [ + 294, + 106, + 300, + 116 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 104, + 505, + 117 + ], + "score": 1.0, + "content": ". We name the privacy loss (leak AUC) measured", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 177, + 129 + ], + "score": 1.0, + "content": "against the attack", + "type": "text" + }, + { + "bbox": [ + 177, + 117, + 189, + 127 + ], + "score": 0.86, + "content": "r _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 114, + 506, + 129 + ], + "score": 1.0, + "content": "the norm leak AUC. In Figure 2(c), the norm leak AUCs are consistently above", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 359, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 359, + 140 + ], + "score": 1.0, + "content": "0.9, signaling a high level of label leakage throughout training.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 143, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 106, + 143, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 437, + 155 + ], + "score": 1.0, + "content": "Attack 2: Direction-based scoring function. We now show that the direction of", + "type": "text" + }, + { + "bbox": [ + 438, + 145, + 444, + 155 + ], + "score": 0.81, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 143, + 505, + 155 + ], + "score": 1.0, + "content": "(in addition to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 352, + 167 + ], + "score": 1.0, + "content": "its magnitude) can also leak the label. For a pair of examples,", + "type": "text" + }, + { + "bbox": [ + 353, + 154, + 426, + 166 + ], + "score": 0.88, + "content": "( X _ { a } , y _ { a } ) , ( X _ { b } , y _ { b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 154, + 505, + 167 + ], + "score": 1.0, + "content": ", let their respective", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 263, + 178 + ], + "score": 1.0, + "content": "predicted positive class probability be", + "type": "text" + }, + { + "bbox": [ + 264, + 166, + 300, + 177 + ], + "score": 0.41, + "content": "\\widetilde { p } _ { 1 , a } , \\widetilde { p } _ { 1 , b }", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 165, + 456, + 178 + ], + "score": 1.0, + "content": "and their communicated gradients be", + "type": "text" + }, + { + "bbox": [ + 456, + 167, + 482, + 177 + ], + "score": 0.51, + "content": "g _ { a } , \\ g _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 165, + 506, + 178 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 174, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 104, + 174, + 118, + 190 + ], + "score": 1.0, + "content": "co", + "type": "text" + }, + { + "bbox": [ + 118, + 176, + 182, + 186 + ], + "score": 0.73, + "content": "\\mathbf { \\tau } _ { \\mathrm { 3 } } : \\mathbb { R } ^ { d } \\times \\mathbf { \\bar { R } } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 174, + 333, + 190 + ], + "score": 1.0, + "content": "e e denote the cosine similarity function", + "type": "text" + }, + { + "bbox": [ + 333, + 176, + 461, + 187 + ], + "score": 0.9, + "content": "\\cos ( g _ { a } , g _ { b } ) = g _ { a } ^ { T } g _ { b } \\bar { / } ( \\| g _ { a } \\| _ { 2 } \\| g _ { b } \\| _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 174, + 506, + 190 + ], + "score": 1.0, + "content": ". It is easy", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 186, + 506, + 201 + ], + "spans": [ + { + "bbox": [ + 104, + 186, + 149, + 201 + ], + "score": 1.0, + "content": "to see that", + "type": "text" + }, + { + "bbox": [ + 150, + 188, + 474, + 199 + ], + "score": 0.79, + "content": "\\begin{array} { r } { \\cos \\bigl ( g _ { a } , g _ { b } \\bigr ) = \\mathrm { s g n } \\bigl ( \\widetilde { p } _ { 1 , a } - y _ { a } \\bigr ) \\cdot \\mathrm { s g n } \\bigl ( \\widetilde { p } _ { 1 , a } - y _ { b } \\bigr ) \\cdot \\cos \\bigl ( \\nabla _ { z } h ( z ) \\bigr | _ { z = f ( X _ { a } ) } , \\nabla _ { z } h ( z ) \\bigr | _ { z = f ( X _ { b } ) } \\bigr ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 186, + 506, + 201 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 133, + 209 + ], + "score": 0.82, + "content": "\\operatorname { s g n } ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 198, + 288, + 210 + ], + "score": 1.0, + "content": "e e is the sign function which returns 1 if", + "type": "text" + }, + { + "bbox": [ + 288, + 199, + 314, + 209 + ], + "score": 0.89, + "content": "x \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 198, + 335, + 210 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 335, + 199, + 349, + 209 + ], + "score": 0.81, + "content": "- 1", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 198, + 359, + 210 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 360, + 199, + 385, + 208 + ], + "score": 0.86, + "content": "x < 0", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 198, + 505, + 210 + ], + "score": 1.0, + "content": ". We highlight two additional", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 404, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 404, + 221 + ], + "score": 1.0, + "content": "observations that can allow us to use cosine similarity to recover the label.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 112, + 222, + 505, + 315 + ], + "lines": [ + { + "bbox": [ + 111, + 221, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 111, + 221, + 277, + 236 + ], + "score": 1.0, + "content": "• Observation 3: When the examples", + "type": "text" + }, + { + "bbox": [ + 277, + 223, + 293, + 234 + ], + "score": 0.89, + "content": "a , b", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 221, + 434, + 236 + ], + "score": 1.0, + "content": "are of different classes, the term", + "type": "text" + }, + { + "bbox": [ + 435, + 222, + 502, + 235 + ], + "score": 0.88, + "content": "\\mathrm { s g n } ( \\widetilde { p } _ { 1 , a } - y _ { a } ) ~ .", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 221, + 505, + 236 + ], + "score": 1.0, + "content": "·", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 121, + 234, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 121, + 234, + 214, + 246 + ], + "score": 0.9, + "content": "\\operatorname { s g n } ( \\widetilde { p } _ { 1 , a } - y _ { b } ) = - 1", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 234, + 418, + 247 + ], + "score": 1.0, + "content": "is negative. On the other hand, when examples", + "type": "text" + }, + { + "bbox": [ + 418, + 234, + 436, + 244 + ], + "score": 0.38, + "content": "a , b", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 234, + 506, + 247 + ], + "score": 1.0, + "content": "eare of the same", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 120, + 245, + 484, + 257 + ], + "spans": [ + { + "bbox": [ + 120, + 245, + 484, + 257 + ], + "score": 1.0, + "content": "eclass (both positive/both negative), this product will have a value of 1 and thus be positive.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 113, + 258, + 504, + 272 + ], + "spans": [ + { + "bbox": [ + 113, + 258, + 358, + 272 + ], + "score": 1.0, + "content": "• Observation 4: Throughout training, for any two examples", + "type": "text" + }, + { + "bbox": [ + 359, + 260, + 374, + 270 + ], + "score": 0.84, + "content": "a , b", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 258, + 497, + 272 + ], + "score": 1.0, + "content": ", their gradients of the function", + "type": "text" + }, + { + "bbox": [ + 497, + 259, + 504, + 269 + ], + "score": 0.76, + "content": "h", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 119, + 268, + 506, + 285 + ], + "spans": [ + { + "bbox": [ + 119, + 268, + 251, + 285 + ], + "score": 1.0, + "content": "always form an acute angle, i.e.", + "type": "text" + }, + { + "bbox": [ + 251, + 270, + 430, + 284 + ], + "score": 0.88, + "content": "\\cos ( \\nabla _ { z } h ( z ) | _ { z = \\pmb { f } ( X _ { a } ) } , \\nabla _ { z } h ( z ) | _ { z = \\pmb { f } ( X _ { b } ) } ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 268, + 506, + 285 + ], + "score": 1.0, + "content": "(Figure 2(d)). For", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 120, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 120, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "neural networks that use monotonically increasing activation functions (such as ReLU, sigmoid,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 121, + 292, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 121, + 292, + 505, + 305 + ], + "score": 1.0, + "content": "tanh), this is caused by the fact that the gradients of these activation functions with respect to its", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 121, + 303, + 481, + 316 + ], + "spans": [ + { + "bbox": [ + 121, + 303, + 481, + 316 + ], + "score": 1.0, + "content": "inputs are coordinatewise nonnegative and thus always lie in the first closed hyperorthant.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 505, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 132, + 330 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 316, + 177, + 329 + ], + "score": 0.92, + "content": "\\cos ( g _ { a } , g _ { b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "is the product of the terms from Observation 3 and 4, we see that for a given", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "example, all the examples that are of the same class result in a positive cosine similarity, while all", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 339, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 506, + 351 + ], + "score": 1.0, + "content": "opposite class examples result in a negative cosine similarity. If the problem is class-imbalanced", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 351, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 505, + 362 + ], + "score": 1.0, + "content": "and the non-label party knows there are fewer positive examples than negative ones, it can thus", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 361, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 505, + 373 + ], + "score": 1.0, + "content": "determine the label of each example: the class is negative if more than half of the examples result", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 372, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 506, + 384 + ], + "score": 1.0, + "content": "in positive cosine similarity; otherwise it is positive. For many practical applications, the non-label", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 104, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "party may reasonably guess which class has more examples in the dataset a priori without ever", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 392, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 104, + 392, + 506, + 408 + ], + "score": 1.0, + "content": "seeing any data—for example, in disease prediction, the percentage of the entire population having", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 405, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 308, + 417 + ], + "score": 1.0, + "content": "a certain disease is almost always much lower than", + "type": "text" + }, + { + "bbox": [ + 308, + 405, + 327, + 415 + ], + "score": 0.87, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 405, + 505, + 417 + ], + "score": 1.0, + "content": "; in online advertising conversion prediction,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 416, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 387, + 428 + ], + "score": 1.0, + "content": "the conversion rate (fraction of positive examples) is rarely higher than", + "type": "text" + }, + { + "bbox": [ + 388, + 416, + 407, + 426 + ], + "score": 0.86, + "content": "30 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 416, + 506, + 428 + ], + "score": 1.0, + "content": ". Note that the non-label", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 427, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 505, + 438 + ], + "score": 1.0, + "content": "party doesn’t need knowledge of the exact sample proportion of each class for this method to work.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 542 + ], + "lines": [ + { + "bbox": [ + 105, + 443, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "To simplify this attack for evaluation, we consider an even worse oracle scenario where the non-label", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 454, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 336, + 468 + ], + "score": 1.0, + "content": "party knows the clean gradient of one positive example", + "type": "text" + }, + { + "bbox": [ + 336, + 456, + 349, + 466 + ], + "score": 0.87, + "content": "g _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 454, + 506, + 468 + ], + "score": 1.0, + "content": ". Unlike the aforementioned practical", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "majority counting attack which needs to first figure out the direction of one positive gradient, this", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 476, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 505, + 489 + ], + "score": 1.0, + "content": "oracle scenario assumes the non-label party is directly given this information. Thus, any protection", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "score": 1.0, + "content": "method capable of defending this oracle attack would also protect against the more practical one.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 497, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 129, + 513 + ], + "score": 1.0, + "content": "With", + "type": "text" + }, + { + "bbox": [ + 129, + 500, + 142, + 510 + ], + "score": 0.85, + "content": "g _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 497, + 321, + 513 + ], + "score": 1.0, + "content": "given, the direction-based scoring function", + "type": "text" + }, + { + "bbox": [ + 322, + 500, + 333, + 509 + ], + "score": 0.84, + "content": "r _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 497, + 375, + 513 + ], + "score": 1.0, + "content": "is simply", + "type": "text" + }, + { + "bbox": [ + 375, + 498, + 457, + 510 + ], + "score": 0.91, + "content": "r _ { d } ( g ) = \\cos ( g , g _ { + } )", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 497, + 506, + 513 + ], + "score": 1.0, + "content": ". We name", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 333, + 522 + ], + "score": 1.0, + "content": "the privacy loss (leak AUC) against this oracle attack", + "type": "text" + }, + { + "bbox": [ + 333, + 511, + 344, + 520 + ], + "score": 0.84, + "content": "r _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 509, + 412, + 522 + ], + "score": 1.0, + "content": "the cosine leak", + "type": "text" + }, + { + "bbox": [ + 413, + 510, + 435, + 519 + ], + "score": 0.58, + "content": "A U C", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 509, + 506, + 522 + ], + "score": 1.0, + "content": ". In practice, we", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 520, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 374, + 533 + ], + "score": 1.0, + "content": "randomly choose a positive class clean gradient from each batch as", + "type": "text" + }, + { + "bbox": [ + 375, + 522, + 387, + 532 + ], + "score": 0.86, + "content": "g _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 520, + 506, + 533 + ], + "score": 1.0, + "content": "for evaluation. For iterations", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 532, + 475, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 475, + 543 + ], + "score": 1.0, + "content": "in Figure 2(e), the cosine leak AUC all have the highest value of 1 (complete label leakage).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 107, + 552, + 339, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 340, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 340, + 567 + ], + "score": 1.0, + "content": "4 LABEL LEAKAGE PROTECTION METHODS", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 573, + 505, + 606 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "In this section, we first introduce a heuristic random perturbation approach designed to prevent the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 584, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 505, + 596 + ], + "score": 1.0, + "content": "practical attacks identified in Section 3.3. We then propose a theoretically justified method that aims", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 595, + 503, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 503, + 608 + ], + "score": 1.0, + "content": "to protect against the entire class of scoring functions considered in our threat model (Section 3.2).", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "title", + "bbox": [ + 107, + 612, + 297, + 623 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 298, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 298, + 625 + ], + "score": 1.0, + "content": "4.1 A HEURISTIC PROTECTION APPROACH", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 107, + 632, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "Random perturbation and the isotropic Gaussian baseline. To protect against label leakage, the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "score": 1.0, + "content": "label party should ideally communicate essential information about the gradient without commu-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "nicating its actual value. Random perturbation methods generally aim to achieve this goal. One", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "obvious consideration for random perturbation is to keep the perturbed gradients unbiased. In other", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 676, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 169, + 691 + ], + "score": 1.0, + "content": "words, suppose", + "type": "text" + }, + { + "bbox": [ + 169, + 677, + 176, + 689 + ], + "score": 0.85, + "content": "\\tilde { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 676, + 387, + 691 + ], + "score": 1.0, + "content": "is the perturbed version of an example’s true gradient", + "type": "text" + }, + { + "bbox": [ + 387, + 679, + 393, + 688 + ], + "score": 0.74, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 676, + 451, + 691 + ], + "score": 1.0, + "content": ", then we want", + "type": "text" + }, + { + "bbox": [ + 451, + 677, + 501, + 689 + ], + "score": 0.92, + "content": "\\mathbb { E } [ \\tilde { g } \\mid g ] = g", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 676, + 506, + 691 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "By chain rule and linearity of expectation, this ensures the computed gradients of the non-label", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 182, + 713 + ], + "score": 1.0, + "content": "party’s parameters", + "type": "text" + }, + { + "bbox": [ + 182, + 699, + 190, + 711 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "will also be unbiased, a desirable property for stochastic optimization. Among", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 707, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 505, + 724 + ], + "score": 1.0, + "content": "unbiased perturbation methods, a simple approach is to add iid isotropic Gaussian noise to every", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 721, + 504, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 504, + 734 + ], + "score": 1.0, + "content": "gradient to mix the positive and negative gradient distribution before sending to the non-label party.", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 49 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 363, + 96 + ], + "score": 1.0, + "content": "As a consequence of Observation 1 and 2, the gradient norm", + "type": "text" + }, + { + "bbox": [ + 364, + 82, + 385, + 95 + ], + "score": 0.91, + "content": "\\| g \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "of the positive instances are", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 468, + 108 + ], + "score": 1.0, + "content": "generally larger than that of the negative ones (Figure 2(c)). Thus, the scoring function", + "type": "text" + }, + { + "bbox": [ + 468, + 93, + 505, + 106 + ], + "score": 0.91, + "content": "r _ { n } ( g ) =", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 107, + 105, + 127, + 117 + ], + "score": 0.9, + "content": "\\| g \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 104, + 294, + 117 + ], + "score": 1.0, + "content": "is a strong predictor of the unseen label", + "type": "text" + }, + { + "bbox": [ + 294, + 106, + 300, + 116 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 104, + 505, + 117 + ], + "score": 1.0, + "content": ". We name the privacy loss (leak AUC) measured", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 177, + 129 + ], + "score": 1.0, + "content": "against the attack", + "type": "text" + }, + { + "bbox": [ + 177, + 117, + 189, + 127 + ], + "score": 0.86, + "content": "r _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 114, + 506, + 129 + ], + "score": 1.0, + "content": "the norm leak AUC. In Figure 2(c), the norm leak AUCs are consistently above", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 359, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 359, + 140 + ], + "score": 1.0, + "content": "0.9, signaling a high level of label leakage throughout training.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 82, + 506, + 140 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 143, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 106, + 143, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 437, + 155 + ], + "score": 1.0, + "content": "Attack 2: Direction-based scoring function. We now show that the direction of", + "type": "text" + }, + { + "bbox": [ + 438, + 145, + 444, + 155 + ], + "score": 0.81, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 143, + 505, + 155 + ], + "score": 1.0, + "content": "(in addition to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 352, + 167 + ], + "score": 1.0, + "content": "its magnitude) can also leak the label. For a pair of examples,", + "type": "text" + }, + { + "bbox": [ + 353, + 154, + 426, + 166 + ], + "score": 0.88, + "content": "( X _ { a } , y _ { a } ) , ( X _ { b } , y _ { b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 154, + 505, + 167 + ], + "score": 1.0, + "content": ", let their respective", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 263, + 178 + ], + "score": 1.0, + "content": "predicted positive class probability be", + "type": "text" + }, + { + "bbox": [ + 264, + 166, + 300, + 177 + ], + "score": 0.41, + "content": "\\widetilde { p } _ { 1 , a } , \\widetilde { p } _ { 1 , b }", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 165, + 456, + 178 + ], + "score": 1.0, + "content": "and their communicated gradients be", + "type": "text" + }, + { + "bbox": [ + 456, + 167, + 482, + 177 + ], + "score": 0.51, + "content": "g _ { a } , \\ g _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 165, + 506, + 178 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 174, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 104, + 174, + 118, + 190 + ], + "score": 1.0, + "content": "co", + "type": "text" + }, + { + "bbox": [ + 118, + 176, + 182, + 186 + ], + "score": 0.73, + "content": "\\mathbf { \\tau } _ { \\mathrm { 3 } } : \\mathbb { R } ^ { d } \\times \\mathbf { \\bar { R } } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 174, + 333, + 190 + ], + "score": 1.0, + "content": "e e denote the cosine similarity function", + "type": "text" + }, + { + "bbox": [ + 333, + 176, + 461, + 187 + ], + "score": 0.9, + "content": "\\cos ( g _ { a } , g _ { b } ) = g _ { a } ^ { T } g _ { b } \\bar { / } ( \\| g _ { a } \\| _ { 2 } \\| g _ { b } \\| _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 174, + 506, + 190 + ], + "score": 1.0, + "content": ". It is easy", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 186, + 506, + 201 + ], + "spans": [ + { + "bbox": [ + 104, + 186, + 149, + 201 + ], + "score": 1.0, + "content": "to see that", + "type": "text" + }, + { + "bbox": [ + 150, + 188, + 474, + 199 + ], + "score": 0.79, + "content": "\\begin{array} { r } { \\cos \\bigl ( g _ { a } , g _ { b } \\bigr ) = \\mathrm { s g n } \\bigl ( \\widetilde { p } _ { 1 , a } - y _ { a } \\bigr ) \\cdot \\mathrm { s g n } \\bigl ( \\widetilde { p } _ { 1 , a } - y _ { b } \\bigr ) \\cdot \\cos \\bigl ( \\nabla _ { z } h ( z ) \\bigr | _ { z = f ( X _ { a } ) } , \\nabla _ { z } h ( z ) \\bigr | _ { z = f ( X _ { b } ) } \\bigr ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 186, + 506, + 201 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 133, + 209 + ], + "score": 0.82, + "content": "\\operatorname { s g n } ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 198, + 288, + 210 + ], + "score": 1.0, + "content": "e e is the sign function which returns 1 if", + "type": "text" + }, + { + "bbox": [ + 288, + 199, + 314, + 209 + ], + "score": 0.89, + "content": "x \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 198, + 335, + 210 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 335, + 199, + 349, + 209 + ], + "score": 0.81, + "content": "- 1", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 198, + 359, + 210 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 360, + 199, + 385, + 208 + ], + "score": 0.86, + "content": "x < 0", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 198, + 505, + 210 + ], + "score": 1.0, + "content": ". We highlight two additional", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 404, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 404, + 221 + ], + "score": 1.0, + "content": "observations that can allow us to use cosine similarity to recover the label.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8, + "bbox_fs": [ + 104, + 143, + 506, + 221 + ] + }, + { + "type": "text", + "bbox": [ + 112, + 222, + 505, + 315 + ], + "lines": [ + { + "bbox": [ + 111, + 221, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 111, + 221, + 277, + 236 + ], + "score": 1.0, + "content": "• Observation 3: When the examples", + "type": "text" + }, + { + "bbox": [ + 277, + 223, + 293, + 234 + ], + "score": 0.89, + "content": "a , b", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 221, + 434, + 236 + ], + "score": 1.0, + "content": "are of different classes, the term", + "type": "text" + }, + { + "bbox": [ + 435, + 222, + 502, + 235 + ], + "score": 0.88, + "content": "\\mathrm { s g n } ( \\widetilde { p } _ { 1 , a } - y _ { a } ) ~ .", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 221, + 505, + 236 + ], + "score": 1.0, + "content": "·", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 121, + 234, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 121, + 234, + 214, + 246 + ], + "score": 0.9, + "content": "\\operatorname { s g n } ( \\widetilde { p } _ { 1 , a } - y _ { b } ) = - 1", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 234, + 418, + 247 + ], + "score": 1.0, + "content": "is negative. On the other hand, when examples", + "type": "text" + }, + { + "bbox": [ + 418, + 234, + 436, + 244 + ], + "score": 0.38, + "content": "a , b", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 234, + 506, + 247 + ], + "score": 1.0, + "content": "eare of the same", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 120, + 245, + 484, + 257 + ], + "spans": [ + { + "bbox": [ + 120, + 245, + 484, + 257 + ], + "score": 1.0, + "content": "eclass (both positive/both negative), this product will have a value of 1 and thus be positive.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 113, + 258, + 504, + 272 + ], + "spans": [ + { + "bbox": [ + 113, + 258, + 358, + 272 + ], + "score": 1.0, + "content": "• Observation 4: Throughout training, for any two examples", + "type": "text" + }, + { + "bbox": [ + 359, + 260, + 374, + 270 + ], + "score": 0.84, + "content": "a , b", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 258, + 497, + 272 + ], + "score": 1.0, + "content": ", their gradients of the function", + "type": "text" + }, + { + "bbox": [ + 497, + 259, + 504, + 269 + ], + "score": 0.76, + "content": "h", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 119, + 268, + 506, + 285 + ], + "spans": [ + { + "bbox": [ + 119, + 268, + 251, + 285 + ], + "score": 1.0, + "content": "always form an acute angle, i.e.", + "type": "text" + }, + { + "bbox": [ + 251, + 270, + 430, + 284 + ], + "score": 0.88, + "content": "\\cos ( \\nabla _ { z } h ( z ) | _ { z = \\pmb { f } ( X _ { a } ) } , \\nabla _ { z } h ( z ) | _ { z = \\pmb { f } ( X _ { b } ) } ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 268, + 506, + 285 + ], + "score": 1.0, + "content": "(Figure 2(d)). For", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 120, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 120, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "neural networks that use monotonically increasing activation functions (such as ReLU, sigmoid,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 121, + 292, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 121, + 292, + 505, + 305 + ], + "score": 1.0, + "content": "tanh), this is caused by the fact that the gradients of these activation functions with respect to its", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 121, + 303, + 481, + 316 + ], + "spans": [ + { + "bbox": [ + 121, + 303, + 481, + 316 + ], + "score": 1.0, + "content": "inputs are coordinatewise nonnegative and thus always lie in the first closed hyperorthant.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5, + "bbox_fs": [ + 111, + 221, + 506, + 316 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 505, + 438 + ], + "lines": [ + { + "bbox": [ + 106, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 132, + 330 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 316, + 177, + 329 + ], + "score": 0.92, + "content": "\\cos ( g _ { a } , g _ { b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "is the product of the terms from Observation 3 and 4, we see that for a given", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "example, all the examples that are of the same class result in a positive cosine similarity, while all", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 339, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 506, + 351 + ], + "score": 1.0, + "content": "opposite class examples result in a negative cosine similarity. If the problem is class-imbalanced", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 351, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 505, + 362 + ], + "score": 1.0, + "content": "and the non-label party knows there are fewer positive examples than negative ones, it can thus", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 361, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 505, + 373 + ], + "score": 1.0, + "content": "determine the label of each example: the class is negative if more than half of the examples result", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 372, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 506, + 384 + ], + "score": 1.0, + "content": "in positive cosine similarity; otherwise it is positive. For many practical applications, the non-label", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 104, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "party may reasonably guess which class has more examples in the dataset a priori without ever", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 392, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 104, + 392, + 506, + 408 + ], + "score": 1.0, + "content": "seeing any data—for example, in disease prediction, the percentage of the entire population having", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 405, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 308, + 417 + ], + "score": 1.0, + "content": "a certain disease is almost always much lower than", + "type": "text" + }, + { + "bbox": [ + 308, + 405, + 327, + 415 + ], + "score": 0.87, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 405, + 505, + 417 + ], + "score": 1.0, + "content": "; in online advertising conversion prediction,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 416, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 387, + 428 + ], + "score": 1.0, + "content": "the conversion rate (fraction of positive examples) is rarely higher than", + "type": "text" + }, + { + "bbox": [ + 388, + 416, + 407, + 426 + ], + "score": 0.86, + "content": "30 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 416, + 506, + 428 + ], + "score": 1.0, + "content": ". Note that the non-label", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 427, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 505, + 438 + ], + "score": 1.0, + "content": "party doesn’t need knowledge of the exact sample proportion of each class for this method to work.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25, + "bbox_fs": [ + 104, + 316, + 506, + 438 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 443, + 505, + 542 + ], + "lines": [ + { + "bbox": [ + 105, + 443, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "To simplify this attack for evaluation, we consider an even worse oracle scenario where the non-label", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 454, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 336, + 468 + ], + "score": 1.0, + "content": "party knows the clean gradient of one positive example", + "type": "text" + }, + { + "bbox": [ + 336, + 456, + 349, + 466 + ], + "score": 0.87, + "content": "g _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 454, + 506, + 468 + ], + "score": 1.0, + "content": ". Unlike the aforementioned practical", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "majority counting attack which needs to first figure out the direction of one positive gradient, this", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 476, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 505, + 489 + ], + "score": 1.0, + "content": "oracle scenario assumes the non-label party is directly given this information. Thus, any protection", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "score": 1.0, + "content": "method capable of defending this oracle attack would also protect against the more practical one.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 497, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 129, + 513 + ], + "score": 1.0, + "content": "With", + "type": "text" + }, + { + "bbox": [ + 129, + 500, + 142, + 510 + ], + "score": 0.85, + "content": "g _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 497, + 321, + 513 + ], + "score": 1.0, + "content": "given, the direction-based scoring function", + "type": "text" + }, + { + "bbox": [ + 322, + 500, + 333, + 509 + ], + "score": 0.84, + "content": "r _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 497, + 375, + 513 + ], + "score": 1.0, + "content": "is simply", + "type": "text" + }, + { + "bbox": [ + 375, + 498, + 457, + 510 + ], + "score": 0.91, + "content": "r _ { d } ( g ) = \\cos ( g , g _ { + } )", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 497, + 506, + 513 + ], + "score": 1.0, + "content": ". We name", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 333, + 522 + ], + "score": 1.0, + "content": "the privacy loss (leak AUC) against this oracle attack", + "type": "text" + }, + { + "bbox": [ + 333, + 511, + 344, + 520 + ], + "score": 0.84, + "content": "r _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 509, + 412, + 522 + ], + "score": 1.0, + "content": "the cosine leak", + "type": "text" + }, + { + "bbox": [ + 413, + 510, + 435, + 519 + ], + "score": 0.58, + "content": "A U C", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 509, + 506, + 522 + ], + "score": 1.0, + "content": ". In practice, we", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 520, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 374, + 533 + ], + "score": 1.0, + "content": "randomly choose a positive class clean gradient from each batch as", + "type": "text" + }, + { + "bbox": [ + 375, + 522, + 387, + 532 + ], + "score": 0.86, + "content": "g _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 520, + 506, + 533 + ], + "score": 1.0, + "content": "for evaluation. For iterations", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 532, + 475, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 475, + 543 + ], + "score": 1.0, + "content": "in Figure 2(e), the cosine leak AUC all have the highest value of 1 (complete label leakage).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 443, + 506, + 543 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 552, + 339, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 340, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 340, + 567 + ], + "score": 1.0, + "content": "4 LABEL LEAKAGE PROTECTION METHODS", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 573, + 505, + 606 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "In this section, we first introduce a heuristic random perturbation approach designed to prevent the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 584, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 505, + 596 + ], + "score": 1.0, + "content": "practical attacks identified in Section 3.3. We then propose a theoretically justified method that aims", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 595, + 503, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 503, + 608 + ], + "score": 1.0, + "content": "to protect against the entire class of scoring functions considered in our threat model (Section 3.2).", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 572, + 505, + 608 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 612, + 297, + 623 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 298, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 298, + 625 + ], + "score": 1.0, + "content": "4.1 A HEURISTIC PROTECTION APPROACH", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 107, + 632, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "Random perturbation and the isotropic Gaussian baseline. To protect against label leakage, the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "score": 1.0, + "content": "label party should ideally communicate essential information about the gradient without commu-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "nicating its actual value. Random perturbation methods generally aim to achieve this goal. One", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "obvious consideration for random perturbation is to keep the perturbed gradients unbiased. In other", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 676, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 169, + 691 + ], + "score": 1.0, + "content": "words, suppose", + "type": "text" + }, + { + "bbox": [ + 169, + 677, + 176, + 689 + ], + "score": 0.85, + "content": "\\tilde { g }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 676, + 387, + 691 + ], + "score": 1.0, + "content": "is the perturbed version of an example’s true gradient", + "type": "text" + }, + { + "bbox": [ + 387, + 679, + 393, + 688 + ], + "score": 0.74, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 676, + 451, + 691 + ], + "score": 1.0, + "content": ", then we want", + "type": "text" + }, + { + "bbox": [ + 451, + 677, + 501, + 689 + ], + "score": 0.92, + "content": "\\mathbb { E } [ \\tilde { g } \\mid g ] = g", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 676, + 506, + 691 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "By chain rule and linearity of expectation, this ensures the computed gradients of the non-label", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 182, + 713 + ], + "score": 1.0, + "content": "party’s parameters", + "type": "text" + }, + { + "bbox": [ + 182, + 699, + 190, + 711 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "will also be unbiased, a desirable property for stochastic optimization. Among", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 707, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 505, + 724 + ], + "score": 1.0, + "content": "unbiased perturbation methods, a simple approach is to add iid isotropic Gaussian noise to every", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 721, + 504, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 504, + 734 + ], + "score": 1.0, + "content": "gradient to mix the positive and negative gradient distribution before sending to the non-label party.", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 49, + "bbox_fs": [ + 105, + 632, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 126 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Although isotropic Gaussian noise is a valid option, it may not be optimal because 1) the gradients", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "are vectors but not scalars, so the structure of the noise covariance matrix matters. Isotropic noise", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "might neglect the direction information; 2) due to the asymmetry of the positive and negative gradi-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "ent distribution, the label party could add noise with different distributions to each class’s gradients.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 505, + 314 + ], + "lines": [ + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "Norm-alignment heuristic. We now introduce an improved heuristic approach of adding zero-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "mean Gaussian noise with non-isotropic and example-dependent covariance. [Magnitude choice]", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 195, + 166 + ], + "score": 1.0, + "content": "As we have seen that", + "type": "text" + }, + { + "bbox": [ + 196, + 154, + 216, + 166 + ], + "score": 0.92, + "content": "\\| g \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 155, + 505, + 166 + ], + "score": 1.0, + "content": "can be different for positive and negative examples and thus leak label", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 166, + 504, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 504, + 177 + ], + "score": 1.0, + "content": "information, this heuristic first aims to make the norm of each perturbed gradient indistinguishable", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 177, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 505, + 188 + ], + "score": 1.0, + "content": "from one another. Specifically, we want to match the expected squared 2-norm of every perturbed", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 419, + 200 + ], + "score": 1.0, + "content": "gradient in a mini-batch to the largest squared 2-norm in this batch (denote by", + "type": "text" + }, + { + "bbox": [ + 419, + 187, + 455, + 199 + ], + "score": 0.93, + "content": "\\| g _ { \\operatorname* { m a x } } \\| _ { 2 } ^ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 187, + 505, + 200 + ], + "score": 1.0, + "content": ". [Direction", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "choice] In addition, as we have seen empirically from Figure 2(e), the positive and negative gradients", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 265, + 224 + ], + "score": 1.0, + "content": "lie close to a one-dimensional line in", + "type": "text" + }, + { + "bbox": [ + 265, + 209, + 278, + 221 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 209, + 505, + 224 + ], + "score": 1.0, + "content": ", with positive examples pointing in one direction and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "score": 1.0, + "content": "negative examples in the other. Thus we consider only adding noise (roughly speaking) along “this", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 506, + 246 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 263, + 246 + ], + "score": 1.0, + "content": "line”. More concretely, for a gradient", + "type": "text" + }, + { + "bbox": [ + 263, + 234, + 273, + 245 + ], + "score": 0.86, + "content": "g _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 231, + 506, + 246 + ], + "score": 1.0, + "content": "in the batch, we add a zero-mean Gaussian noise vector", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 244, + 506, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 117, + 256 + ], + "score": 0.83, + "content": "\\eta _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 244, + 376, + 256 + ], + "score": 1.0, + "content": "supported only on the one-dimensional space along the line of", + "type": "text" + }, + { + "bbox": [ + 377, + 245, + 387, + 256 + ], + "score": 0.86, + "content": "g _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 244, + 506, + 256 + ], + "score": 1.0, + "content": ". In other words, the noise’s", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 255, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 231, + 269 + ], + "score": 1.0, + "content": "covariance is the rank-1 matrix", + "type": "text" + }, + { + "bbox": [ + 231, + 255, + 308, + 269 + ], + "score": 0.92, + "content": "\\mathrm { C o v } [ \\eta _ { j } ] = \\sigma _ { j } ^ { 2 } g _ { j } g _ { j } ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 255, + 361, + 269 + ], + "score": 1.0, + "content": ". To calculate", + "type": "text" + }, + { + "bbox": [ + 362, + 257, + 372, + 268 + ], + "score": 0.85, + "content": "\\sigma _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 255, + 443, + 269 + ], + "score": 1.0, + "content": ", we aim to match", + "type": "text" + }, + { + "bbox": [ + 443, + 255, + 505, + 268 + ], + "score": 0.91, + "content": "\\mathbb { E } [ \\| g _ { j } + \\eta _ { j } \\| _ { 2 } ^ { 2 } ] =", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 268, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 107, + 268, + 139, + 281 + ], + "score": 0.92, + "content": "\\| g _ { \\operatorname* { m a x } } \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 268, + 257, + 283 + ], + "score": 1.0, + "content": ". A simple calculation gives", + "type": "text" + }, + { + "bbox": [ + 257, + 269, + 365, + 281 + ], + "score": 0.9, + "content": "\\sigma _ { j } = \\sqrt { \\| g _ { \\mathrm { { m a x } } } \\| _ { 2 } ^ { 2 } / \\| g _ { j } \\| _ { 2 } ^ { 2 } - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 268, + 506, + 283 + ], + "score": 1.0, + "content": ". Since we align to the maximum", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 280, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 293 + ], + "score": 1.0, + "content": "norm, we name this heuristic protection method max norm. The advantage of max norm is that", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "score": 1.0, + "content": "it has no parameter to tune. Unfortunately, it does not have a strong theoretical motivation, cannot", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 301, + 503, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 503, + 315 + ], + "score": 1.0, + "content": "flexibly trade-off between model utility and privacy, and may be broken by some unknown attacks.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 106, + 319, + 338, + 330 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 340, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 340, + 331 + ], + "score": 1.0, + "content": "4.2 OPTIMIZED PERTURBATION METHOD: MA R V E L L", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 505, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 344 + ], + "score": 1.0, + "content": "Motivated by the above issues of max norm, we next study how to achieve a more principled", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "trade-off between model performance (utility) and label protection (privacy). To do so, we directly", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 355, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 505, + 366 + ], + "score": 1.0, + "content": "minimize the worst-case adversarial scoring function’s leak AUC under a utility constraint. We", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 366, + 487, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 487, + 377 + ], + "score": 1.0, + "content": "name this protection method Marvell (optiMized perturbAtion to pReVEnt Label Leakage).", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 381, + 505, + 450 + ], + "lines": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "score": 1.0, + "content": "Noise perturbation structure. Due to the distribution difference between the positive and nega-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "tive class’s cut layer gradients, we consider having the label party additively perturb the randomly", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 403, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 175, + 419 + ], + "score": 1.0, + "content": "sampled positive", + "type": "text" + }, + { + "bbox": [ + 176, + 404, + 192, + 417 + ], + "score": 0.91, + "content": "\\overset { \\cdot } { g } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 403, + 245, + 419 + ], + "score": 1.0, + "content": "and negative", + "type": "text" + }, + { + "bbox": [ + 246, + 404, + 262, + 417 + ], + "score": 0.91, + "content": "g ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 403, + 506, + 419 + ], + "score": 1.0, + "content": "gradients with independent zero-mean random noise vectors", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 415, + 503, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 121, + 427 + ], + "score": 0.88, + "content": "\\eta ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 121, + 415, + 140, + 428 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 141, + 417, + 155, + 427 + ], + "score": 0.88, + "content": "\\eta ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 415, + 355, + 428 + ], + "score": 1.0, + "content": "with possibly different distributions (denoted by", + "type": "text" + }, + { + "bbox": [ + 356, + 416, + 372, + 426 + ], + "score": 0.87, + "content": "D ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 415, + 392, + 428 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 392, + 416, + 409, + 426 + ], + "score": 0.86, + "content": "{ \\cal D } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 415, + 451, + 428 + ], + "score": 1.0, + "content": "). We use", + "type": "text" + }, + { + "bbox": [ + 451, + 416, + 468, + 426 + ], + "score": 0.86, + "content": "\\widetilde { P } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 415, + 487, + 428 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 487, + 416, + 503, + 426 + ], + "score": 0.85, + "content": "\\widetilde { P } ^ { ( 0 ) }", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "to denote the induced perturbed positive and negative gradient distributions. Our goal is to find the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 438, + 482, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 215, + 451 + ], + "score": 1.0, + "content": "optimal noise distributions", + "type": "text" + }, + { + "bbox": [ + 215, + 438, + 232, + 448 + ], + "score": 0.87, + "content": "D ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 438, + 250, + 451 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 251, + 438, + 267, + 448 + ], + "score": 0.87, + "content": "{ \\cal D } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 438, + 482, + 451 + ], + "score": 1.0, + "content": "by optimizing our privacy objective described below.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 454, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 455, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 505, + 467 + ], + "score": 1.0, + "content": "Privacy protection optimization objective. As the adversarial non-label party in our threat model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 307, + 479 + ], + "score": 1.0, + "content": "is allowed to use any measurable scoring function", + "type": "text" + }, + { + "bbox": [ + 307, + 468, + 313, + 476 + ], + "score": 0.72, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "for label recovery, we aim to protect against all", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 477, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 489 + ], + "score": 1.0, + "content": "such scoring functions by minimizing the privacy loss of the worst case scoring function measured", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 487, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 104, + 487, + 396, + 502 + ], + "score": 1.0, + "content": "through our leak AUC metric. Formally, our optimization objective is", + "type": "text" + }, + { + "bbox": [ + 397, + 488, + 501, + 501 + ], + "score": 0.81, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \\operatorname* { m a x } _ { r } \\mathrm { A U C } ( r ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 487, + 505, + 502 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 499, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 179, + 513 + ], + "score": 1.0, + "content": "Here to compute", + "type": "text" + }, + { + "bbox": [ + 180, + 500, + 213, + 511 + ], + "score": 0.87, + "content": "\\operatorname { A U C } ( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 499, + 235, + 513 + ], + "score": 1.0, + "content": ", the", + "type": "text" + }, + { + "bbox": [ + 235, + 499, + 270, + 512 + ], + "score": 0.92, + "content": "\\mathrm { F P R } _ { r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 499, + 290, + 513 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 291, + 500, + 326, + 512 + ], + "score": 0.92, + "content": "\\mathrm { T P R } _ { r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 499, + 506, + 513 + ], + "score": 1.0, + "content": "needs to be computed using the perturbed", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 160, + 523 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 160, + 511, + 176, + 521 + ], + "score": 0.87, + "content": "\\widetilde { P } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 511, + 197, + 523 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 197, + 512, + 213, + 521 + ], + "score": 0.88, + "content": "\\widetilde { P } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 511, + 327, + 523 + ], + "score": 1.0, + "content": "instead of the unperturbed", + "type": "text" + }, + { + "bbox": [ + 328, + 511, + 344, + 521 + ], + "score": 0.88, + "content": "P ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 511, + 365, + 523 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 365, + 511, + 381, + 521 + ], + "score": 0.87, + "content": "P ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 511, + 505, + 523 + ], + "score": 1.0, + "content": "(Section 3.2). Since AUC is", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "difficult to directly optimize, we consider optimizing an upper bound through the following theorem:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 535, + 503, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 534, + 505, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 177, + 548 + ], + "score": 1.0, + "content": "Theorem 1. For", + "type": "text" + }, + { + "bbox": [ + 178, + 535, + 223, + 546 + ], + "score": 0.94, + "content": "0 \\leq \\epsilon < 4", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 534, + 391, + 548 + ], + "score": 1.0, + "content": "and any perturbed gradient distributions", + "type": "text" + }, + { + "bbox": [ + 392, + 535, + 408, + 545 + ], + "score": 0.87, + "content": "\\widetilde { P } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 534, + 428, + 548 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 428, + 535, + 444, + 545 + ], + "score": 0.88, + "content": "\\widetilde { P } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 534, + 505, + 548 + ], + "score": 1.0, + "content": "that are abso-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 546, + 284, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 284, + 559 + ], + "score": 1.0, + "content": "lutely continuous with respect to each other,", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "interline_equation", + "bbox": [ + 126, + 561, + 449, + 577 + ], + "lines": [ + { + "bbox": [ + 126, + 561, + 449, + 577 + ], + "spans": [ + { + "bbox": [ + 126, + 561, + 449, + 577 + ], + "score": 0.88, + "content": "\\begin{array} { r } { { \\bf K L } ( \\widetilde { P } ^ { ( 1 ) } \\parallel \\widetilde { P } ^ { ( 0 ) } ) + { \\bf K L } ( \\widetilde { P } ^ { ( 0 ) } \\parallel \\widetilde { P } ^ { ( 1 ) } ) \\le \\epsilon \\quad i m p l i e s \\quad \\operatorname* { m a x } _ { r } { \\bf A U C } ( r ) \\le \\frac { 1 } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } - \\frac { \\epsilon } { 8 } . } \\end{array}", + "type": "interline_equation", + "image_path": "ae6b61eb0cff1a125dab320998ed1be5faaadccef2a976c140928310b25f3090.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 126, + 561, + 449, + 577 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 615 + ], + "lines": [ + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "score": 1.0, + "content": "From Theorem 1 (proof in Appendix A.3), we see that as long as the sum KL divergence is below 4,√", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 593, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 245, + 606 + ], + "score": 1.0, + "content": "the smaller sumKL is, the smaller", + "type": "text" + }, + { + "bbox": [ + 245, + 593, + 303, + 605 + ], + "score": 0.84, + "content": "\\operatorname* { m a x } _ { r } \\mathbf { A U C } ( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 593, + 318, + 606 + ], + "score": 1.0, + "content": "is.", + "type": "text" + }, + { + "bbox": [ + 318, + 593, + 399, + 605 + ], + "score": 0.92, + "content": "( 1 / 2 + \\overline { { \\sqrt { \\epsilon } } } / 2 - \\epsilon / 8 ", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 593, + 451, + 606 + ], + "score": 1.0, + "content": "decreases as", + "type": "text" + }, + { + "bbox": [ + 452, + 596, + 457, + 603 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 593, + 505, + 606 + ], + "score": 1.0, + "content": "decreases.)", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 603, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 616 + ], + "score": 1.0, + "content": "Thus we can instead minimize the sum KL divergence between the perturbed gradient distributions:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 617, + 425, + 632 + ], + "lines": [ + { + "bbox": [ + 185, + 617, + 425, + 632 + ], + "spans": [ + { + "bbox": [ + 185, + 617, + 425, + 632 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\operatorname { s u m K L } ^ { * } : = \\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \\operatorname { K L } ( \\widetilde { P } ^ { ( 1 ) } \\parallel \\widetilde { P } ^ { ( 0 ) } ) + \\operatorname { K L } ( \\widetilde { P } ^ { ( 0 ) } \\parallel \\widetilde { P } ^ { ( 1 ) } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "51e42a6f3f13d695e9edbed1d48782e49acf777dff2421afed5b2bfd672ac43f.jpg" + } + ] + } + ], + "index": 44, + "virtual_lines": [ + { + "bbox": [ + 185, + 617, + 425, + 632 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "Utility constraint. In an extreme case, we could add infinite noise to both the negative and positive", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 646, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 506, + 660 + ], + "score": 1.0, + "content": "gradients. This would minimize (1) optimally to 0 and make the worst case leak AUC 0.5, which is", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 657, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 505, + 670 + ], + "score": 1.0, + "content": "equivalent to a random guess. However, stochastic gradient descent cannot converge under infinitely", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 666, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 680 + ], + "score": 1.0, + "content": "large noise, so it is necessary to control the variance of the added noise. We thus introduce the noise", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 179, + 692 + ], + "score": 1.0, + "content": "power constraint:", + "type": "text" + }, + { + "bbox": [ + 180, + 678, + 363, + 690 + ], + "score": 0.89, + "content": "p \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 1 ) } ] ) + ( 1 - p ) \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 0 ) } ] ) \\le ~ P", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 676, + 394, + 692 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 394, + 680, + 401, + 690 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 676, + 506, + 692 + ], + "score": 1.0, + "content": "is the fraction of positive", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 689, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 289, + 701 + ], + "score": 1.0, + "content": "examples (already known to the label party);", + "type": "text" + }, + { + "bbox": [ + 289, + 691, + 332, + 700 + ], + "score": 0.87, + "content": "\\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( i ) } ] )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 689, + 505, + 701 + ], + "score": 1.0, + "content": "denotes the trace of the covariance matrix", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 699, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 191, + 714 + ], + "score": 1.0, + "content": "of the random noise", + "type": "text" + }, + { + "bbox": [ + 191, + 700, + 207, + 712 + ], + "score": 0.91, + "content": "\\eta ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 699, + 299, + 714 + ], + "score": 1.0, + "content": "; and the upper bound", + "type": "text" + }, + { + "bbox": [ + 300, + 702, + 308, + 711 + ], + "score": 0.83, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 699, + 506, + 714 + ], + "score": 1.0, + "content": "is a tunable hyperparameter to control the level", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 711, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 172, + 722 + ], + "score": 1.0, + "content": "of noise: larger", + "type": "text" + }, + { + "bbox": [ + 172, + 711, + 182, + 721 + ], + "score": 0.82, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 711, + 505, + 722 + ], + "score": 1.0, + "content": "would achieve a lower sumKL and thus lower worst-case leak AUC and better", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "privacy; however, it would also add more noise to the gradients, leading to slower optimization", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 49 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 126 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Although isotropic Gaussian noise is a valid option, it may not be optimal because 1) the gradients", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "are vectors but not scalars, so the structure of the noise covariance matrix matters. Isotropic noise", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "might neglect the direction information; 2) due to the asymmetry of the positive and negative gradi-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "ent distribution, the label party could add noise with different distributions to each class’s gradients.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 82, + 505, + 128 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 505, + 314 + ], + "lines": [ + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 505, + 145 + ], + "score": 1.0, + "content": "Norm-alignment heuristic. We now introduce an improved heuristic approach of adding zero-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "mean Gaussian noise with non-isotropic and example-dependent covariance. [Magnitude choice]", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 195, + 166 + ], + "score": 1.0, + "content": "As we have seen that", + "type": "text" + }, + { + "bbox": [ + 196, + 154, + 216, + 166 + ], + "score": 0.92, + "content": "\\| g \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 155, + 505, + 166 + ], + "score": 1.0, + "content": "can be different for positive and negative examples and thus leak label", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 166, + 504, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 504, + 177 + ], + "score": 1.0, + "content": "information, this heuristic first aims to make the norm of each perturbed gradient indistinguishable", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 177, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 505, + 188 + ], + "score": 1.0, + "content": "from one another. Specifically, we want to match the expected squared 2-norm of every perturbed", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 419, + 200 + ], + "score": 1.0, + "content": "gradient in a mini-batch to the largest squared 2-norm in this batch (denote by", + "type": "text" + }, + { + "bbox": [ + 419, + 187, + 455, + 199 + ], + "score": 0.93, + "content": "\\| g _ { \\operatorname* { m a x } } \\| _ { 2 } ^ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 187, + 505, + 200 + ], + "score": 1.0, + "content": ". [Direction", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "choice] In addition, as we have seen empirically from Figure 2(e), the positive and negative gradients", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 265, + 224 + ], + "score": 1.0, + "content": "lie close to a one-dimensional line in", + "type": "text" + }, + { + "bbox": [ + 265, + 209, + 278, + 221 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 209, + 505, + 224 + ], + "score": 1.0, + "content": ", with positive examples pointing in one direction and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "score": 1.0, + "content": "negative examples in the other. Thus we consider only adding noise (roughly speaking) along “this", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 506, + 246 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 263, + 246 + ], + "score": 1.0, + "content": "line”. More concretely, for a gradient", + "type": "text" + }, + { + "bbox": [ + 263, + 234, + 273, + 245 + ], + "score": 0.86, + "content": "g _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 231, + 506, + 246 + ], + "score": 1.0, + "content": "in the batch, we add a zero-mean Gaussian noise vector", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 244, + 506, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 117, + 256 + ], + "score": 0.83, + "content": "\\eta _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 244, + 376, + 256 + ], + "score": 1.0, + "content": "supported only on the one-dimensional space along the line of", + "type": "text" + }, + { + "bbox": [ + 377, + 245, + 387, + 256 + ], + "score": 0.86, + "content": "g _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 244, + 506, + 256 + ], + "score": 1.0, + "content": ". In other words, the noise’s", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 255, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 231, + 269 + ], + "score": 1.0, + "content": "covariance is the rank-1 matrix", + "type": "text" + }, + { + "bbox": [ + 231, + 255, + 308, + 269 + ], + "score": 0.92, + "content": "\\mathrm { C o v } [ \\eta _ { j } ] = \\sigma _ { j } ^ { 2 } g _ { j } g _ { j } ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 255, + 361, + 269 + ], + "score": 1.0, + "content": ". To calculate", + "type": "text" + }, + { + "bbox": [ + 362, + 257, + 372, + 268 + ], + "score": 0.85, + "content": "\\sigma _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 255, + 443, + 269 + ], + "score": 1.0, + "content": ", we aim to match", + "type": "text" + }, + { + "bbox": [ + 443, + 255, + 505, + 268 + ], + "score": 0.91, + "content": "\\mathbb { E } [ \\| g _ { j } + \\eta _ { j } \\| _ { 2 } ^ { 2 } ] =", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 268, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 107, + 268, + 139, + 281 + ], + "score": 0.92, + "content": "\\| g _ { \\operatorname* { m a x } } \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 268, + 257, + 283 + ], + "score": 1.0, + "content": ". A simple calculation gives", + "type": "text" + }, + { + "bbox": [ + 257, + 269, + 365, + 281 + ], + "score": 0.9, + "content": "\\sigma _ { j } = \\sqrt { \\| g _ { \\mathrm { { m a x } } } \\| _ { 2 } ^ { 2 } / \\| g _ { j } \\| _ { 2 } ^ { 2 } - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 268, + 506, + 283 + ], + "score": 1.0, + "content": ". Since we align to the maximum", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 280, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 293 + ], + "score": 1.0, + "content": "norm, we name this heuristic protection method max norm. The advantage of max norm is that", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "score": 1.0, + "content": "it has no parameter to tune. Unfortunately, it does not have a strong theoretical motivation, cannot", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 301, + 503, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 503, + 315 + ], + "score": 1.0, + "content": "flexibly trade-off between model utility and privacy, and may be broken by some unknown attacks.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 132, + 506, + 315 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 319, + 338, + 330 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 340, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 340, + 331 + ], + "score": 1.0, + "content": "4.2 OPTIMIZED PERTURBATION METHOD: MA R V E L L", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 332, + 505, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 344 + ], + "score": 1.0, + "content": "Motivated by the above issues of max norm, we next study how to achieve a more principled", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "trade-off between model performance (utility) and label protection (privacy). To do so, we directly", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 355, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 505, + 366 + ], + "score": 1.0, + "content": "minimize the worst-case adversarial scoring function’s leak AUC under a utility constraint. We", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 366, + 487, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 487, + 377 + ], + "score": 1.0, + "content": "name this protection method Marvell (optiMized perturbAtion to pReVEnt Label Leakage).", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 333, + 505, + 377 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 381, + 505, + 450 + ], + "lines": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "score": 1.0, + "content": "Noise perturbation structure. Due to the distribution difference between the positive and nega-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "tive class’s cut layer gradients, we consider having the label party additively perturb the randomly", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 403, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 175, + 419 + ], + "score": 1.0, + "content": "sampled positive", + "type": "text" + }, + { + "bbox": [ + 176, + 404, + 192, + 417 + ], + "score": 0.91, + "content": "\\overset { \\cdot } { g } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 403, + 245, + 419 + ], + "score": 1.0, + "content": "and negative", + "type": "text" + }, + { + "bbox": [ + 246, + 404, + 262, + 417 + ], + "score": 0.91, + "content": "g ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 403, + 506, + 419 + ], + "score": 1.0, + "content": "gradients with independent zero-mean random noise vectors", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 415, + 503, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 121, + 427 + ], + "score": 0.88, + "content": "\\eta ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 121, + 415, + 140, + 428 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 141, + 417, + 155, + 427 + ], + "score": 0.88, + "content": "\\eta ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 415, + 355, + 428 + ], + "score": 1.0, + "content": "with possibly different distributions (denoted by", + "type": "text" + }, + { + "bbox": [ + 356, + 416, + 372, + 426 + ], + "score": 0.87, + "content": "D ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 415, + 392, + 428 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 392, + 416, + 409, + 426 + ], + "score": 0.86, + "content": "{ \\cal D } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 415, + 451, + 428 + ], + "score": 1.0, + "content": "). We use", + "type": "text" + }, + { + "bbox": [ + 451, + 416, + 468, + 426 + ], + "score": 0.86, + "content": "\\widetilde { P } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 415, + 487, + 428 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 487, + 416, + 503, + 426 + ], + "score": 0.85, + "content": "\\widetilde { P } ^ { ( 0 ) }", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "to denote the induced perturbed positive and negative gradient distributions. Our goal is to find the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 438, + 482, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 215, + 451 + ], + "score": 1.0, + "content": "optimal noise distributions", + "type": "text" + }, + { + "bbox": [ + 215, + 438, + 232, + 448 + ], + "score": 0.87, + "content": "D ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 438, + 250, + 451 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 251, + 438, + 267, + 448 + ], + "score": 0.87, + "content": "{ \\cal D } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 438, + 482, + 451 + ], + "score": 1.0, + "content": "by optimizing our privacy objective described below.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 381, + 506, + 451 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 454, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 455, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 505, + 467 + ], + "score": 1.0, + "content": "Privacy protection optimization objective. As the adversarial non-label party in our threat model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 307, + 479 + ], + "score": 1.0, + "content": "is allowed to use any measurable scoring function", + "type": "text" + }, + { + "bbox": [ + 307, + 468, + 313, + 476 + ], + "score": 0.72, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "for label recovery, we aim to protect against all", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 477, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 489 + ], + "score": 1.0, + "content": "such scoring functions by minimizing the privacy loss of the worst case scoring function measured", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 487, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 104, + 487, + 396, + 502 + ], + "score": 1.0, + "content": "through our leak AUC metric. Formally, our optimization objective is", + "type": "text" + }, + { + "bbox": [ + 397, + 488, + 501, + 501 + ], + "score": 0.81, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \\operatorname* { m a x } _ { r } \\mathrm { A U C } ( r ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 487, + 505, + 502 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 499, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 179, + 513 + ], + "score": 1.0, + "content": "Here to compute", + "type": "text" + }, + { + "bbox": [ + 180, + 500, + 213, + 511 + ], + "score": 0.87, + "content": "\\operatorname { A U C } ( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 499, + 235, + 513 + ], + "score": 1.0, + "content": ", the", + "type": "text" + }, + { + "bbox": [ + 235, + 499, + 270, + 512 + ], + "score": 0.92, + "content": "\\mathrm { F P R } _ { r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 499, + 290, + 513 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 291, + 500, + 326, + 512 + ], + "score": 0.92, + "content": "\\mathrm { T P R } _ { r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 499, + 506, + 513 + ], + "score": 1.0, + "content": "needs to be computed using the perturbed", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 160, + 523 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 160, + 511, + 176, + 521 + ], + "score": 0.87, + "content": "\\widetilde { P } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 511, + 197, + 523 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 197, + 512, + 213, + 521 + ], + "score": 0.88, + "content": "\\widetilde { P } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 511, + 327, + 523 + ], + "score": 1.0, + "content": "instead of the unperturbed", + "type": "text" + }, + { + "bbox": [ + 328, + 511, + 344, + 521 + ], + "score": 0.88, + "content": "P ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 511, + 365, + 523 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 365, + 511, + 381, + 521 + ], + "score": 0.87, + "content": "P ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 511, + 505, + 523 + ], + "score": 1.0, + "content": "(Section 3.2). Since AUC is", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "difficult to directly optimize, we consider optimizing an upper bound through the following theorem:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34, + "bbox_fs": [ + 104, + 455, + 506, + 534 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 535, + 503, + 557 + ], + "lines": [ + { + "bbox": [ + 105, + 534, + 505, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 177, + 548 + ], + "score": 1.0, + "content": "Theorem 1. For", + "type": "text" + }, + { + "bbox": [ + 178, + 535, + 223, + 546 + ], + "score": 0.94, + "content": "0 \\leq \\epsilon < 4", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 534, + 391, + 548 + ], + "score": 1.0, + "content": "and any perturbed gradient distributions", + "type": "text" + }, + { + "bbox": [ + 392, + 535, + 408, + 545 + ], + "score": 0.87, + "content": "\\widetilde { P } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 534, + 428, + 548 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 428, + 535, + 444, + 545 + ], + "score": 0.88, + "content": "\\widetilde { P } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 534, + 505, + 548 + ], + "score": 1.0, + "content": "that are abso-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 546, + 284, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 284, + 559 + ], + "score": 1.0, + "content": "lutely continuous with respect to each other,", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 534, + 505, + 559 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 126, + 561, + 449, + 577 + ], + "lines": [ + { + "bbox": [ + 126, + 561, + 449, + 577 + ], + "spans": [ + { + "bbox": [ + 126, + 561, + 449, + 577 + ], + "score": 0.88, + "content": "\\begin{array} { r } { { \\bf K L } ( \\widetilde { P } ^ { ( 1 ) } \\parallel \\widetilde { P } ^ { ( 0 ) } ) + { \\bf K L } ( \\widetilde { P } ^ { ( 0 ) } \\parallel \\widetilde { P } ^ { ( 1 ) } ) \\le \\epsilon \\quad i m p l i e s \\quad \\operatorname* { m a x } _ { r } { \\bf A U C } ( r ) \\le \\frac { 1 } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } - \\frac { \\epsilon } { 8 } . } \\end{array}", + "type": "interline_equation", + "image_path": "ae6b61eb0cff1a125dab320998ed1be5faaadccef2a976c140928310b25f3090.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 126, + 561, + 449, + 577 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 582, + 505, + 615 + ], + "lines": [ + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "score": 1.0, + "content": "From Theorem 1 (proof in Appendix A.3), we see that as long as the sum KL divergence is below 4,√", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 593, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 245, + 606 + ], + "score": 1.0, + "content": "the smaller sumKL is, the smaller", + "type": "text" + }, + { + "bbox": [ + 245, + 593, + 303, + 605 + ], + "score": 0.84, + "content": "\\operatorname* { m a x } _ { r } \\mathbf { A U C } ( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 593, + 318, + 606 + ], + "score": 1.0, + "content": "is.", + "type": "text" + }, + { + "bbox": [ + 318, + 593, + 399, + 605 + ], + "score": 0.92, + "content": "( 1 / 2 + \\overline { { \\sqrt { \\epsilon } } } / 2 - \\epsilon / 8 ", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 593, + 451, + 606 + ], + "score": 1.0, + "content": "decreases as", + "type": "text" + }, + { + "bbox": [ + 452, + 596, + 457, + 603 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 593, + 505, + 606 + ], + "score": 1.0, + "content": "decreases.)", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 603, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 616 + ], + "score": 1.0, + "content": "Thus we can instead minimize the sum KL divergence between the perturbed gradient distributions:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 582, + 505, + 616 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 617, + 425, + 632 + ], + "lines": [ + { + "bbox": [ + 185, + 617, + 425, + 632 + ], + "spans": [ + { + "bbox": [ + 185, + 617, + 425, + 632 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\operatorname { s u m K L } ^ { * } : = \\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \\operatorname { K L } ( \\widetilde { P } ^ { ( 1 ) } \\parallel \\widetilde { P } ^ { ( 0 ) } ) + \\operatorname { K L } ( \\widetilde { P } ^ { ( 0 ) } \\parallel \\widetilde { P } ^ { ( 1 ) } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "51e42a6f3f13d695e9edbed1d48782e49acf777dff2421afed5b2bfd672ac43f.jpg" + } + ] + } + ], + "index": 44, + "virtual_lines": [ + { + "bbox": [ + 185, + 617, + 425, + 632 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 649 + ], + "score": 1.0, + "content": "Utility constraint. In an extreme case, we could add infinite noise to both the negative and positive", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 646, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 506, + 660 + ], + "score": 1.0, + "content": "gradients. This would minimize (1) optimally to 0 and make the worst case leak AUC 0.5, which is", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 657, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 505, + 670 + ], + "score": 1.0, + "content": "equivalent to a random guess. However, stochastic gradient descent cannot converge under infinitely", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 666, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 680 + ], + "score": 1.0, + "content": "large noise, so it is necessary to control the variance of the added noise. We thus introduce the noise", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 676, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 179, + 692 + ], + "score": 1.0, + "content": "power constraint:", + "type": "text" + }, + { + "bbox": [ + 180, + 678, + 363, + 690 + ], + "score": 0.89, + "content": "p \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 1 ) } ] ) + ( 1 - p ) \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 0 ) } ] ) \\le ~ P", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 676, + 394, + 692 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 394, + 680, + 401, + 690 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 676, + 506, + 692 + ], + "score": 1.0, + "content": "is the fraction of positive", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 689, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 289, + 701 + ], + "score": 1.0, + "content": "examples (already known to the label party);", + "type": "text" + }, + { + "bbox": [ + 289, + 691, + 332, + 700 + ], + "score": 0.87, + "content": "\\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( i ) } ] )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 689, + 505, + 701 + ], + "score": 1.0, + "content": "denotes the trace of the covariance matrix", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 699, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 191, + 714 + ], + "score": 1.0, + "content": "of the random noise", + "type": "text" + }, + { + "bbox": [ + 191, + 700, + 207, + 712 + ], + "score": 0.91, + "content": "\\eta ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 699, + 299, + 714 + ], + "score": 1.0, + "content": "; and the upper bound", + "type": "text" + }, + { + "bbox": [ + 300, + 702, + 308, + 711 + ], + "score": 0.83, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 699, + 506, + 714 + ], + "score": 1.0, + "content": "is a tunable hyperparameter to control the level", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 711, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 172, + 722 + ], + "score": 1.0, + "content": "of noise: larger", + "type": "text" + }, + { + "bbox": [ + 172, + 711, + 182, + 721 + ], + "score": 0.82, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 711, + 505, + 722 + ], + "score": 1.0, + "content": "would achieve a lower sumKL and thus lower worst-case leak AUC and better", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "privacy; however, it would also add more noise to the gradients, leading to slower optimization", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 434, + 95 + ], + "score": 1.0, + "content": "convergence and possibly worse model utility. We weight each class’s noise level", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 435, + 83, + 479, + 94 + ], + "score": 0.91, + "content": "\\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( i ) } ] )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 479, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "by its", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 190, + 105 + ], + "score": 1.0, + "content": "example proportion", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 190, + 94, + 197, + 104 + ], + "score": 0.66, + "content": "\\dot { p }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 197, + 93, + 208, + 105 + ], + "score": 1.0, + "content": "or", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 208, + 93, + 234, + 104 + ], + "score": 0.88, + "content": "1 - p )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 234, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "since, from an optimization perspective, we want to equally control", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 464, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 464, + 115 + ], + "score": 1.0, + "content": "every training example’s gradient noise. The constrained optimization problem becomes:", + "type": "text", + "cross_page": true + } + ], + "index": 2 + } + ], + "index": 49, + "bbox_fs": [ + 105, + 637, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 505, + 114 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 434, + 95 + ], + "score": 1.0, + "content": "convergence and possibly worse model utility. We weight each class’s noise level", + "type": "text" + }, + { + "bbox": [ + 435, + 83, + 479, + 94 + ], + "score": 0.91, + "content": "\\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( i ) } ] )", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "by its", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 190, + 105 + ], + "score": 1.0, + "content": "example proportion", + "type": "text" + }, + { + "bbox": [ + 190, + 94, + 197, + 104 + ], + "score": 0.66, + "content": "\\dot { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 93, + 208, + 105 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 208, + 93, + 234, + 104 + ], + "score": 0.88, + "content": "1 - p )", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "since, from an optimization perspective, we want to equally control", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 464, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 464, + 115 + ], + "score": 1.0, + "content": "every training example’s gradient noise. The constrained optimization problem becomes:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 108, + 115, + 490, + 135 + ], + "lines": [ + { + "bbox": [ + 108, + 115, + 490, + 135 + ], + "spans": [ + { + "bbox": [ + 108, + 115, + 490, + 135 + ], + "score": 0.5, + "content": "\\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \\mathrm { K L } ( \\widetilde P ^ { ( 1 ) } \\parallel \\widetilde P ^ { ( 0 ) } ) + \\mathrm { K L } ( \\widetilde P ^ { ( 0 ) } \\parallel \\widetilde P ^ { ( 1 ) } ) \\mathrm { s . t . } p \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 1 ) } ] ) + ( 1 - p ) \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 0 ) } ] ) \\le P .", + "type": "interline_equation", + "image_path": "ad48e8bb31c0afb74128c8485025b715d1ae2320b5914fbd36274cee23da97fb.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 108, + 115, + 490, + 135 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 142, + 505, + 265 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "Optimizing the objective in practice. To solve the optimization problem we first introduce some", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "modelling assumptions. We assume that the unperturbed gradient of each class follows a Gaussian", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 163, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 158, + 178 + ], + "score": 1.0, + "content": "distribution:", + "type": "text" + }, + { + "bbox": [ + 159, + 164, + 244, + 177 + ], + "score": 0.95, + "content": "g ^ { ( 1 ) } \\stackrel { \\cdot } { \\sim } \\mathcal { N } ( \\bar { g } ^ { ( 1 ) } , v I _ { d \\times d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 163, + 263, + 178 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 263, + 164, + 350, + 177 + ], + "score": 0.92, + "content": "g ^ { ( 0 ) } \\sim \\dot { \\mathcal { N } } ( \\bar { g } ^ { ( 0 ) } , u I _ { d \\times d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 163, + 505, + 178 + ], + "score": 1.0, + "content": ". Despite this being an approximation,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "as we see later in Section 5, it can achieve strong protection quality against our identified attacks. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 187, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 198 + ], + "score": 1.0, + "content": "addition, it makes the optimization easier (see below) and provides us with insight on the optimal", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 196, + 506, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 420, + 210 + ], + "score": 1.0, + "content": "noise structure. We also search for perturbation distributions that are Gaussian:", + "type": "text" + }, + { + "bbox": [ + 421, + 197, + 487, + 209 + ], + "score": 0.92, + "content": "D ^ { \\top } \\ d _ { 1 } ) = \\mathcal { N } ( 0 , \\dot { \\Sigma _ { 1 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 196, + 506, + 210 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 206, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 107, + 208, + 175, + 221 + ], + "score": 0.92, + "content": "D ^ { ( 0 ) } = \\mathcal { N } ( 0 , \\Sigma _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 206, + 333, + 224 + ], + "score": 1.0, + "content": "with commuting covariance matrices:", + "type": "text" + }, + { + "bbox": [ + 333, + 209, + 391, + 220 + ], + "score": 0.92, + "content": "\\Sigma _ { 1 } \\Sigma _ { 0 } = \\Sigma _ { 0 } \\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 206, + 506, + 224 + ], + "score": 1.0, + "content": ". The commutative require-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "score": 1.0, + "content": "ment slightly restricts our search space but also makes the optimization problem more tractable. Our", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 230, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 488, + 245 + ], + "score": 1.0, + "content": "goal is to solve for the optimal noise structure, i.e. the positive semidefinite covariance matrices", + "type": "text" + }, + { + "bbox": [ + 489, + 231, + 501, + 242 + ], + "score": 0.86, + "content": "\\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 230, + 506, + 245 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 241, + 506, + 256 + ], + "spans": [ + { + "bbox": [ + 107, + 243, + 119, + 254 + ], + "score": 0.86, + "content": "\\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 241, + 140, + 256 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 140, + 241, + 214, + 254 + ], + "score": 0.93, + "content": "\\Delta g : = \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 241, + 506, + 256 + ], + "score": 1.0, + "content": "denote the difference between the positive and negative gradient’s mean", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 254, + 463, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 463, + 267 + ], + "score": 1.0, + "content": "vectors. We now have the following theorem (proof and interpretation in Appendix A.4):", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 112, + 266, + 453, + 277 + ], + "lines": [ + { + "bbox": [ + 110, + 264, + 454, + 280 + ], + "spans": [ + { + "bbox": [ + 110, + 264, + 210, + 280 + ], + "score": 1.0, + "content": "Theorem 2. The optimal", + "type": "text" + }, + { + "bbox": [ + 211, + 266, + 223, + 277 + ], + "score": 0.87, + "content": "\\Sigma _ { 1 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 264, + 242, + 280 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 242, + 266, + 255, + 278 + ], + "score": 0.86, + "content": "\\Sigma _ { 0 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 264, + 454, + 280 + ], + "score": 1.0, + "content": "to (2) with the above assumptions have the form:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "interline_equation", + "bbox": [ + 136, + 276, + 474, + 304 + ], + "lines": [ + { + "bbox": [ + 136, + 276, + 474, + 304 + ], + "spans": [ + { + "bbox": [ + 136, + 276, + 474, + 304 + ], + "score": 0.91, + "content": "\\Sigma _ { 1 } ^ { * } = \\frac { \\lambda _ { 1 } ^ { ( 1 ) * } - \\lambda _ { 2 } ^ { ( 1 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 1 ) * } I _ { d } , \\quad \\Sigma _ { 0 } ^ { * } = \\frac { \\lambda _ { 1 } ^ { ( 0 ) * } - \\dot { \\lambda } _ { 2 } ^ { ( 0 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 0 ) * } I _ { d } ,", + "type": "interline_equation", + "image_path": "f675e848147dd1440c17ca80be4ace89ae8bbfa40dea00fb26351567eba5422e.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 136, + 276, + 474, + 285.3333333333333 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 136, + 285.3333333333333, + 474, + 294.66666666666663 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 136, + 294.66666666666663, + 474, + 303.99999999999994 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 110, + 306, + 496, + 321 + ], + "lines": [ + { + "bbox": [ + 136, + 301, + 502, + 326 + ], + "spans": [ + { + "bbox": [ + 136, + 306, + 240, + 321 + ], + "score": 0.88, + "content": "( \\lambda _ { 1 } ^ { ( 0 ) * } , \\lambda _ { 2 } ^ { ( 0 ) * } , \\lambda _ { 1 } ^ { ( 1 ) * } , \\lambda _ { 2 } ^ { ( 1 ) * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 301, + 502, + 326 + ], + "score": 1.0, + "content": "is the solution to the following 4-variable optimization problem:", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 320, + 504, + 348 + ], + "lines": [ + { + "bbox": [ + 133, + 320, + 504, + 348 + ], + "spans": [ + { + "bbox": [ + 133, + 320, + 504, + 348 + ], + "score": 0.66, + "content": "\\mathop { \\operatorname* { m i n } } _ { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u }", + "type": "interline_equation", + "image_path": "8da4cd6894e52291dfc051925d299f28978fae853cc99911fe481e8666667963.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 133, + 320, + 504, + 329.3333333333333 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 133, + 329.3333333333333, + 504, + 338.66666666666663 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 133, + 338.66666666666663, + 504, + 347.99999999999994 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 123, + 351, + 478, + 383 + ], + "lines": [ + { + "bbox": [ + 123, + 351, + 478, + 383 + ], + "spans": [ + { + "bbox": [ + 123, + 351, + 478, + 383 + ], + "score": 0.75, + "content": "\\begin{array} { r l } { s . t . \\quad p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) \\lambda _ { 2 } ^ { ( 0 ) } \\leq { \\cal P } , } & \\\\ { - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 , \\quad - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 , \\quad - \\lambda _ { 2 } ^ { ( 1 ) } \\leq 0 , \\quad - \\lambda _ { 2 } ^ { ( 0 ) } \\leq 0 , \\quad \\lambda _ { 2 } ^ { ( 1 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 , \\quad \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 } \\end{array}", + "type": "interline_equation", + "image_path": "97fbec2854168347c45f978448b8f7aa9dc22d0f2e286e0d75b796dbde8e8fdc.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 123, + 351, + 478, + 361.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 123, + 361.6666666666667, + 478, + 372.33333333333337 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 123, + 372.33333333333337, + 478, + 383.00000000000006 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 386, + 505, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 505, + 399 + ], + "score": 1.0, + "content": "Additional details of Marvell. By Theorem 2, our optimization problem over two positive", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 398, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 505, + 410 + ], + "score": 1.0, + "content": "semidefinite matrices is reduced to a much simpler 4-variable optimization problem. We include", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 408, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 104, + 408, + 505, + 421 + ], + "score": 1.0, + "content": "a detailed description of how the constants in the problem are estimated in practice and what solver", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "we use in a full description of the Marvell algorithm in Appendix A.5. Beyond optimization de-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 430, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 391, + 444 + ], + "score": 1.0, + "content": "tails, it is worth noting how to set the power constraint hyperparameter", + "type": "text" + }, + { + "bbox": [ + 392, + 431, + 401, + 441 + ], + "score": 0.84, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 430, + 506, + 444 + ], + "score": 1.0, + "content": "in Equation 2 in practice.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 442, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 192, + 453 + ], + "score": 1.0, + "content": "As directly choosing", + "type": "text" + }, + { + "bbox": [ + 193, + 442, + 202, + 452 + ], + "score": 0.8, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 442, + 505, + 453 + ], + "score": 1.0, + "content": "requires knowledge of the scale of the gradients in the specific application", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 453, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 428, + 469 + ], + "score": 1.0, + "content": "and the scale could also shrink as the optimization converges, we instead express", + "type": "text" + }, + { + "bbox": [ + 428, + 453, + 484, + 467 + ], + "score": 0.93, + "content": "P = s \\| \\Delta g \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 453, + 506, + 469 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 236, + 477 + ], + "score": 1.0, + "content": "tune for a fixed hyperparameter", + "type": "text" + }, + { + "bbox": [ + 237, + 466, + 262, + 475 + ], + "score": 0.89, + "content": "s > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 466, + 505, + 477 + ], + "score": 1.0, + "content": ". This alleviates the need to know the scale of the gradients", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 264, + 489 + ], + "score": 1.0, + "content": "in advance, and the resulting value of", + "type": "text" + }, + { + "bbox": [ + 264, + 477, + 274, + 486 + ], + "score": 0.81, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 475, + 505, + 489 + ], + "score": 1.0, + "content": "can also dynamically change throughout training as the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 486, + 391, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 324, + 501 + ], + "score": 1.0, + "content": "distance between the two gradient distributions’ mean", + "type": "text" + }, + { + "bbox": [ + 324, + 487, + 353, + 500 + ], + "score": 0.93, + "content": "\\| \\Delta g \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 486, + 391, + 501 + ], + "score": 1.0, + "content": "changes.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5 + }, + { + "type": "title", + "bbox": [ + 108, + 510, + 200, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 510, + 201, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 201, + 524 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 108, + 531, + 504, + 554 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "In this section, we first describe our experiment setup and then demonstrate the label protection", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 542, + 468, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 468, + 555 + ], + "score": 1.0, + "content": "quality of Marvell as well as its privacy-utility trade-off relative to baseline approaches.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 559, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 559, + 505, + 570 + ], + "score": 1.0, + "content": "Empirical Setup. We use three real-world binary classification datasets for evaluation: Criteo and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "score": 1.0, + "content": "Avazu, two online advertising prediction datasets with millions of examples; and ISIC, a healthcare", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 581, + 504, + 593 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 504, + 593 + ], + "score": 1.0, + "content": "image dataset for skin cancer prediction. All three datasets exhibit severe label leakage problem", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "score": 1.0, + "content": "without protection. (see Appendix A.6.1 on dataset and preprocessing details). We defer similar", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 603, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 615 + ], + "score": 1.0, + "content": "results on Avazu to Appendix A.7 and focus on Criteo and ISIC in this section. For Criteo, we train", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 613, + 505, + 626 + ], + "spans": [ + { + "bbox": [ + 104, + 613, + 505, + 626 + ], + "score": 1.0, + "content": "a Wide&Deep model (Cheng et al., 2016) where the non-label party owns the embedding layers for", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "input features and the first three 128-unit ReLU activated MLP layers (first half of the deep part)", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "score": 1.0, + "content": "while the label party owns the remaining layers of the deep part and the entire wide part of the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 104, + 645, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 645, + 505, + 659 + ], + "score": 1.0, + "content": "model3. For ISIC, we train a model with 6 convolutional layers each with 64 channels followed by", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "a 64-unit ReLU MLP layer, and the cut layer is after the fourth convolutional layer. In this case,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 103, + 666, + 507, + 684 + ], + "spans": [ + { + "bbox": [ + 103, + 666, + 230, + 684 + ], + "score": 1.0, + "content": "an example’s cut layer feature", + "type": "text" + }, + { + "bbox": [ + 230, + 669, + 254, + 682 + ], + "score": 0.93, + "content": "f ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 666, + 308, + 684 + ], + "score": 1.0, + "content": "and gradient", + "type": "text" + }, + { + "bbox": [ + 308, + 672, + 315, + 681 + ], + "score": 0.79, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 666, + 362, + 684 + ], + "score": 1.0, + "content": "are both in", + "type": "text" + }, + { + "bbox": [ + 363, + 668, + 399, + 680 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { 5 \\times 5 \\times 6 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 666, + 507, + 684 + ], + "score": 1.0, + "content": ". We treat such tensors as", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 104, + 678, + 507, + 696 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 149, + 696 + ], + "score": 1.0, + "content": "vectors in", + "type": "text" + }, + { + "bbox": [ + 149, + 680, + 174, + 692 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { 1 6 0 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 678, + 507, + 696 + ], + "score": 1.0, + "content": "to fit into our analysis framework (for additional model architecture and training", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 692, + 249, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 249, + 705 + ], + "score": 1.0, + "content": "details see Appendix A.6.2, A.6.3).", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 45 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 711, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 118, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 118, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "3In this setting, the label party will also process input features (through the wide part) just like the non-label", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 721, + 346, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 346, + 732 + ], + "score": 1.0, + "content": "party, further relaxing our formal split learning setup in Section 3.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 505, + 114 + ], + "lines": [], + "index": 1, + "bbox_fs": [ + 105, + 82, + 505, + 115 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 108, + 115, + 490, + 135 + ], + "lines": [ + { + "bbox": [ + 108, + 115, + 490, + 135 + ], + "spans": [ + { + "bbox": [ + 108, + 115, + 490, + 135 + ], + "score": 0.5, + "content": "\\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \\mathrm { K L } ( \\widetilde P ^ { ( 1 ) } \\parallel \\widetilde P ^ { ( 0 ) } ) + \\mathrm { K L } ( \\widetilde P ^ { ( 0 ) } \\parallel \\widetilde P ^ { ( 1 ) } ) \\mathrm { s . t . } p \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 1 ) } ] ) + ( 1 - p ) \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 0 ) } ] ) \\le P .", + "type": "interline_equation", + "image_path": "ad48e8bb31c0afb74128c8485025b715d1ae2320b5914fbd36274cee23da97fb.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 108, + 115, + 490, + 135 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 142, + 505, + 265 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "Optimizing the objective in practice. To solve the optimization problem we first introduce some", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "modelling assumptions. We assume that the unperturbed gradient of each class follows a Gaussian", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 163, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 158, + 178 + ], + "score": 1.0, + "content": "distribution:", + "type": "text" + }, + { + "bbox": [ + 159, + 164, + 244, + 177 + ], + "score": 0.95, + "content": "g ^ { ( 1 ) } \\stackrel { \\cdot } { \\sim } \\mathcal { N } ( \\bar { g } ^ { ( 1 ) } , v I _ { d \\times d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 163, + 263, + 178 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 263, + 164, + 350, + 177 + ], + "score": 0.92, + "content": "g ^ { ( 0 ) } \\sim \\dot { \\mathcal { N } } ( \\bar { g } ^ { ( 0 ) } , u I _ { d \\times d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 163, + 505, + 178 + ], + "score": 1.0, + "content": ". Despite this being an approximation,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "as we see later in Section 5, it can achieve strong protection quality against our identified attacks. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 187, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 198 + ], + "score": 1.0, + "content": "addition, it makes the optimization easier (see below) and provides us with insight on the optimal", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 196, + 506, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 420, + 210 + ], + "score": 1.0, + "content": "noise structure. We also search for perturbation distributions that are Gaussian:", + "type": "text" + }, + { + "bbox": [ + 421, + 197, + 487, + 209 + ], + "score": 0.92, + "content": "D ^ { \\top } \\ d _ { 1 } ) = \\mathcal { N } ( 0 , \\dot { \\Sigma _ { 1 } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 196, + 506, + 210 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 206, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 107, + 208, + 175, + 221 + ], + "score": 0.92, + "content": "D ^ { ( 0 ) } = \\mathcal { N } ( 0 , \\Sigma _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 206, + 333, + 224 + ], + "score": 1.0, + "content": "with commuting covariance matrices:", + "type": "text" + }, + { + "bbox": [ + 333, + 209, + 391, + 220 + ], + "score": 0.92, + "content": "\\Sigma _ { 1 } \\Sigma _ { 0 } = \\Sigma _ { 0 } \\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 206, + 506, + 224 + ], + "score": 1.0, + "content": ". The commutative require-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "score": 1.0, + "content": "ment slightly restricts our search space but also makes the optimization problem more tractable. Our", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 230, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 488, + 245 + ], + "score": 1.0, + "content": "goal is to solve for the optimal noise structure, i.e. the positive semidefinite covariance matrices", + "type": "text" + }, + { + "bbox": [ + 489, + 231, + 501, + 242 + ], + "score": 0.86, + "content": "\\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 230, + 506, + 245 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 241, + 506, + 256 + ], + "spans": [ + { + "bbox": [ + 107, + 243, + 119, + 254 + ], + "score": 0.86, + "content": "\\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 241, + 140, + 256 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 140, + 241, + 214, + 254 + ], + "score": 0.93, + "content": "\\Delta g : = \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 241, + 506, + 256 + ], + "score": 1.0, + "content": "denote the difference between the positive and negative gradient’s mean", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 254, + 463, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 463, + 267 + ], + "score": 1.0, + "content": "vectors. We now have the following theorem (proof and interpretation in Appendix A.4):", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 142, + 506, + 267 + ] + }, + { + "type": "text", + "bbox": [ + 112, + 266, + 453, + 277 + ], + "lines": [ + { + "bbox": [ + 110, + 264, + 454, + 280 + ], + "spans": [ + { + "bbox": [ + 110, + 264, + 210, + 280 + ], + "score": 1.0, + "content": "Theorem 2. The optimal", + "type": "text" + }, + { + "bbox": [ + 211, + 266, + 223, + 277 + ], + "score": 0.87, + "content": "\\Sigma _ { 1 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 264, + 242, + 280 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 242, + 266, + 255, + 278 + ], + "score": 0.86, + "content": "\\Sigma _ { 0 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 264, + 454, + 280 + ], + "score": 1.0, + "content": "to (2) with the above assumptions have the form:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 110, + 264, + 454, + 280 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 136, + 276, + 474, + 304 + ], + "lines": [ + { + "bbox": [ + 136, + 276, + 474, + 304 + ], + "spans": [ + { + "bbox": [ + 136, + 276, + 474, + 304 + ], + "score": 0.91, + "content": "\\Sigma _ { 1 } ^ { * } = \\frac { \\lambda _ { 1 } ^ { ( 1 ) * } - \\lambda _ { 2 } ^ { ( 1 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 1 ) * } I _ { d } , \\quad \\Sigma _ { 0 } ^ { * } = \\frac { \\lambda _ { 1 } ^ { ( 0 ) * } - \\dot { \\lambda } _ { 2 } ^ { ( 0 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 0 ) * } I _ { d } ,", + "type": "interline_equation", + "image_path": "f675e848147dd1440c17ca80be4ace89ae8bbfa40dea00fb26351567eba5422e.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 136, + 276, + 474, + 285.3333333333333 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 136, + 285.3333333333333, + 474, + 294.66666666666663 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 136, + 294.66666666666663, + 474, + 303.99999999999994 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 110, + 306, + 496, + 321 + ], + "lines": [ + { + "bbox": [ + 136, + 301, + 502, + 326 + ], + "spans": [ + { + "bbox": [ + 136, + 306, + 240, + 321 + ], + "score": 0.88, + "content": "( \\lambda _ { 1 } ^ { ( 0 ) * } , \\lambda _ { 2 } ^ { ( 0 ) * } , \\lambda _ { 1 } ^ { ( 1 ) * } , \\lambda _ { 2 } ^ { ( 1 ) * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 301, + 502, + 326 + ], + "score": 1.0, + "content": "is the solution to the following 4-variable optimization problem:", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 136, + 301, + 502, + 326 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 320, + 504, + 348 + ], + "lines": [ + { + "bbox": [ + 133, + 320, + 504, + 348 + ], + "spans": [ + { + "bbox": [ + 133, + 320, + 504, + 348 + ], + "score": 0.66, + "content": "\\mathop { \\operatorname* { m i n } } _ { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u }", + "type": "interline_equation", + "image_path": "8da4cd6894e52291dfc051925d299f28978fae853cc99911fe481e8666667963.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 133, + 320, + 504, + 329.3333333333333 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 133, + 329.3333333333333, + 504, + 338.66666666666663 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 133, + 338.66666666666663, + 504, + 347.99999999999994 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 123, + 351, + 478, + 383 + ], + "lines": [ + { + "bbox": [ + 123, + 351, + 478, + 383 + ], + "spans": [ + { + "bbox": [ + 123, + 351, + 478, + 383 + ], + "score": 0.75, + "content": "\\begin{array} { r l } { s . t . \\quad p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) \\lambda _ { 2 } ^ { ( 0 ) } \\leq { \\cal P } , } & \\\\ { - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 , \\quad - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 , \\quad - \\lambda _ { 2 } ^ { ( 1 ) } \\leq 0 , \\quad - \\lambda _ { 2 } ^ { ( 0 ) } \\leq 0 , \\quad \\lambda _ { 2 } ^ { ( 1 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 , \\quad \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 } \\end{array}", + "type": "interline_equation", + "image_path": "97fbec2854168347c45f978448b8f7aa9dc22d0f2e286e0d75b796dbde8e8fdc.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 123, + 351, + 478, + 361.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 123, + 361.6666666666667, + 478, + 372.33333333333337 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 123, + 372.33333333333337, + 478, + 383.00000000000006 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 386, + 505, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 505, + 399 + ], + "score": 1.0, + "content": "Additional details of Marvell. By Theorem 2, our optimization problem over two positive", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 398, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 505, + 410 + ], + "score": 1.0, + "content": "semidefinite matrices is reduced to a much simpler 4-variable optimization problem. We include", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 408, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 104, + 408, + 505, + 421 + ], + "score": 1.0, + "content": "a detailed description of how the constants in the problem are estimated in practice and what solver", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "we use in a full description of the Marvell algorithm in Appendix A.5. Beyond optimization de-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 430, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 391, + 444 + ], + "score": 1.0, + "content": "tails, it is worth noting how to set the power constraint hyperparameter", + "type": "text" + }, + { + "bbox": [ + 392, + 431, + 401, + 441 + ], + "score": 0.84, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 430, + 506, + 444 + ], + "score": 1.0, + "content": "in Equation 2 in practice.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 442, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 192, + 453 + ], + "score": 1.0, + "content": "As directly choosing", + "type": "text" + }, + { + "bbox": [ + 193, + 442, + 202, + 452 + ], + "score": 0.8, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 442, + 505, + 453 + ], + "score": 1.0, + "content": "requires knowledge of the scale of the gradients in the specific application", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 453, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 428, + 469 + ], + "score": 1.0, + "content": "and the scale could also shrink as the optimization converges, we instead express", + "type": "text" + }, + { + "bbox": [ + 428, + 453, + 484, + 467 + ], + "score": 0.93, + "content": "P = s \\| \\Delta g \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 453, + 506, + 469 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 236, + 477 + ], + "score": 1.0, + "content": "tune for a fixed hyperparameter", + "type": "text" + }, + { + "bbox": [ + 237, + 466, + 262, + 475 + ], + "score": 0.89, + "content": "s > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 466, + 505, + 477 + ], + "score": 1.0, + "content": ". This alleviates the need to know the scale of the gradients", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 264, + 489 + ], + "score": 1.0, + "content": "in advance, and the resulting value of", + "type": "text" + }, + { + "bbox": [ + 264, + 477, + 274, + 486 + ], + "score": 0.81, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 475, + 505, + 489 + ], + "score": 1.0, + "content": "can also dynamically change throughout training as the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 486, + 391, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 324, + 501 + ], + "score": 1.0, + "content": "distance between the two gradient distributions’ mean", + "type": "text" + }, + { + "bbox": [ + 324, + 487, + 353, + 500 + ], + "score": 0.93, + "content": "\\| \\Delta g \\| _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 486, + 391, + 501 + ], + "score": 1.0, + "content": "changes.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5, + "bbox_fs": [ + 104, + 386, + 506, + 501 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 510, + 200, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 510, + 201, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 201, + 524 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 108, + 531, + 504, + 554 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "In this section, we first describe our experiment setup and then demonstrate the label protection", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 542, + 468, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 468, + 555 + ], + "score": 1.0, + "content": "quality of Marvell as well as its privacy-utility trade-off relative to baseline approaches.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 531, + 505, + 555 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 559, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 559, + 505, + 570 + ], + "score": 1.0, + "content": "Empirical Setup. We use three real-world binary classification datasets for evaluation: Criteo and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "score": 1.0, + "content": "Avazu, two online advertising prediction datasets with millions of examples; and ISIC, a healthcare", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 581, + 504, + 593 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 504, + 593 + ], + "score": 1.0, + "content": "image dataset for skin cancer prediction. All three datasets exhibit severe label leakage problem", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 505, + 604 + ], + "score": 1.0, + "content": "without protection. (see Appendix A.6.1 on dataset and preprocessing details). We defer similar", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 603, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 615 + ], + "score": 1.0, + "content": "results on Avazu to Appendix A.7 and focus on Criteo and ISIC in this section. For Criteo, we train", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 613, + 505, + 626 + ], + "spans": [ + { + "bbox": [ + 104, + 613, + 505, + 626 + ], + "score": 1.0, + "content": "a Wide&Deep model (Cheng et al., 2016) where the non-label party owns the embedding layers for", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "input features and the first three 128-unit ReLU activated MLP layers (first half of the deep part)", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "score": 1.0, + "content": "while the label party owns the remaining layers of the deep part and the entire wide part of the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 104, + 645, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 645, + 505, + 659 + ], + "score": 1.0, + "content": "model3. For ISIC, we train a model with 6 convolutional layers each with 64 channels followed by", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "a 64-unit ReLU MLP layer, and the cut layer is after the fourth convolutional layer. In this case,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 103, + 666, + 507, + 684 + ], + "spans": [ + { + "bbox": [ + 103, + 666, + 230, + 684 + ], + "score": 1.0, + "content": "an example’s cut layer feature", + "type": "text" + }, + { + "bbox": [ + 230, + 669, + 254, + 682 + ], + "score": 0.93, + "content": "f ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 666, + 308, + 684 + ], + "score": 1.0, + "content": "and gradient", + "type": "text" + }, + { + "bbox": [ + 308, + 672, + 315, + 681 + ], + "score": 0.79, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 666, + 362, + 684 + ], + "score": 1.0, + "content": "are both in", + "type": "text" + }, + { + "bbox": [ + 363, + 668, + 399, + 680 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { 5 \\times 5 \\times 6 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 666, + 507, + 684 + ], + "score": 1.0, + "content": ". We treat such tensors as", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 104, + 678, + 507, + 696 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 149, + 696 + ], + "score": 1.0, + "content": "vectors in", + "type": "text" + }, + { + "bbox": [ + 149, + 680, + 174, + 692 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { 1 6 0 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 678, + 507, + 696 + ], + "score": 1.0, + "content": "to fit into our analysis framework (for additional model architecture and training", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 692, + 249, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 249, + 705 + ], + "score": 1.0, + "content": "details see Appendix A.6.2, A.6.3).", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 45, + "bbox_fs": [ + 103, + 559, + 507, + 705 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 78, + 505, + 168 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 78, + 505, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 78, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 505, + 168 + ], + "score": 0.963, + "type": "image", + "image_path": "da7562e01beb63c35fea76cbea790a0088e2c1811a01fbe1e54c87a5bcc7c358.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 78, + 505, + 108.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 108.0, + 505, + 138.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 138.0, + 505, + 168.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 169, + 503, + 189 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 167, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 505, + 180 + ], + "score": 1.0, + "content": "Figure 3: Norm and cosine leak AUC (computed every batch) at the cut layer and at the first layer under no", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 177, + 442, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 327, + 191 + ], + "score": 1.0, + "content": "protection vs. Marvell with different scale hyperparameter", + "type": "text" + }, + { + "bbox": [ + 328, + 180, + 334, + 187 + ], + "score": 0.54, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 177, + 442, + 191 + ], + "score": 1.0, + "content": "throughout the ISIC training.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 106, + 200, + 442, + 212 + ], + "lines": [ + { + "bbox": [ + 106, + 200, + 443, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 443, + 213 + ], + "score": 1.0, + "content": "5.1 LABEL LEAKAGE AND MA R V E L L’S STRONG AND FLEXIBLE PROTECTION", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 505, + 364 + ], + "lines": [ + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "score": 1.0, + "content": "We first evaluate the protection quality of Marvell against the norm and cosine attacks dis-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "score": 1.0, + "content": "cussed in Section 3.3. We also compare against the leakage metrics when no protection is applied", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "(no noise). As the results across the three datasets are highly similar, we use ISIC as an example", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 254, + 504, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 504, + 266 + ], + "score": 1.0, + "content": "(other datasets see Appendix A.7.1). We see in Figure 3(a)(b) that unlike no noise where the label", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 264, + 278 + ], + "score": 1.0, + "content": "information is completely leaked (leak", + "type": "text" + }, + { + "bbox": [ + 265, + 265, + 306, + 276 + ], + "score": 0.5, + "content": "\\mathbf { A U C } \\approx 1", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 265, + 506, + 278 + ], + "score": 1.0, + "content": ") throughout training, Marvell achieves a flex-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 264, + 289 + ], + "score": 1.0, + "content": "ible degree of protection (by varying", + "type": "text" + }, + { + "bbox": [ + 265, + 277, + 271, + 286 + ], + "score": 0.25, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 276, + 505, + 289 + ], + "score": 1.0, + "content": ") against both the norm 2(a) and direction attacks 2(b)", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 374, + 300 + ], + "score": 1.0, + "content": "on the cut layer gradients and has strong protection (leak AUC", + "type": "text" + }, + { + "bbox": [ + 374, + 287, + 399, + 298 + ], + "score": 0.45, + "content": "\\approx { \\bf 0 . 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 287, + 414, + 300 + ], + "score": 1.0, + "content": ") at", + "type": "text" + }, + { + "bbox": [ + 414, + 287, + 446, + 298 + ], + "score": 0.89, + "content": "s = 4 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 287, + 505, + 300 + ], + "score": 1.0, + "content": ". Additionally,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 297, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 505, + 311 + ], + "score": 1.0, + "content": "it is natural to ask whether the gradients of layers before the cut layer (on the non-label party side)", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 308, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 104, + 308, + 505, + 323 + ], + "score": 1.0, + "content": "can also leak the labels as the non-label party keeps back propagating towards the first layer. In Fig-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "score": 1.0, + "content": "ure 3(c)(d), we compute the leak AUC values when using the non-label party’s first layer activation", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 330, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 104, + 330, + 320, + 344 + ], + "score": 1.0, + "content": "gradient as inputs to the scoring functions to predict", + "type": "text" + }, + { + "bbox": [ + 320, + 333, + 326, + 342 + ], + "score": 0.75, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 330, + 505, + 344 + ], + "score": 1.0, + "content": ". Without protection, the first layer gradient", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 341, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 355 + ], + "score": 1.0, + "content": "still leaks the label very consistently. In constrast, Marvell still achieves strong privacy protection", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 353, + 425, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 174, + 366 + ], + "score": 1.0, + "content": "at the first layer", + "type": "text" + }, + { + "bbox": [ + 174, + 353, + 207, + 364 + ], + "score": 0.85, + "content": "s = 4 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 353, + 425, + 366 + ], + "score": 1.0, + "content": ") despite the protection being analyzed at the cut layer.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 107, + 378, + 322, + 389 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 322, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 322, + 390 + ], + "score": 1.0, + "content": "5.2 PRIVACY-UTILITY TRADE-OFF COMPARISON", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 399, + 505, + 509 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 412 + ], + "score": 1.0, + "content": "After showing Marvell can provide strong privacy protection against our identified attacks, we", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 409, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 423 + ], + "score": 1.0, + "content": "now see how well it can preserve utility by comparing its privacy-utility tradeoff against other pro-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 421, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 433 + ], + "score": 1.0, + "content": "tection baselines: no noise, isotropic Gaussian (iso), and our proposed heuristic max norm.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "score": 1.0, + "content": "Similar to how we allow Marvell to use a power constraint to depend on the current iteration’s", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 442, + 501, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 224, + 455 + ], + "score": 1.0, + "content": "gradient distribution through", + "type": "text" + }, + { + "bbox": [ + 225, + 442, + 280, + 454 + ], + "score": 0.92, + "content": "P = s \\| \\Delta g \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 442, + 501, + 455 + ], + "score": 1.0, + "content": ", we also allow iso to have such type of dependence—", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 191, + 466 + ], + "score": 1.0, + "content": "specifically, we add", + "type": "text" + }, + { + "bbox": [ + 191, + 454, + 316, + 465 + ], + "score": 0.89, + "content": "\\eta \\sim \\mathcal { N } ( \\mathbf { 0 } , ( t / d ) \\cdot \\| g _ { \\operatorname* { m a x } } \\| _ { 2 } ^ { 2 } I _ { d \\times d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 453, + 457, + 466 + ], + "score": 1.0, + "content": "to every gradient in a batch with", + "type": "text" + }, + { + "bbox": [ + 458, + 455, + 463, + 464 + ], + "score": 0.66, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "a tunable", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "privacy hyperparameter to be fixed throughout training. To trace out the complete tradeoff curve", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "score": 1.0, + "content": "for Marvell and iso, we conduct more than 20 training runs for each protection method with a", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 277, + 500 + ], + "score": 1.0, + "content": "different value of privacy hyperparameter", + "type": "text" + }, + { + "bbox": [ + 278, + 488, + 284, + 496 + ], + "score": 0.4, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 486, + 345, + 500 + ], + "score": 1.0, + "content": "for Marvell,", + "type": "text" + }, + { + "bbox": [ + 345, + 487, + 350, + 497 + ], + "score": 0.66, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 486, + 505, + 500 + ], + "score": 1.0, + "content": "for iso) in each run on every dataset.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 498, + 417, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 417, + 510 + ], + "score": 1.0, + "content": "(Note that no noise and max norm do not have privacy hyperparameters.)", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 107, + 514, + 505, + 592 + ], + "lines": [ + { + "bbox": [ + 106, + 514, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 505, + 527 + ], + "score": 1.0, + "content": "We present the tradeoffs between privacy (measured through norm and cosine leak AUC at cut", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 526, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 505, + 537 + ], + "score": 1.0, + "content": "layer/first layer) and utility (measured using test loss and test AUC) in Figure 4. To summarize the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 535, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 301, + 550 + ], + "score": 1.0, + "content": "leak AUC over a given training run, we pick the", + "type": "text" + }, + { + "bbox": [ + 301, + 537, + 321, + 547 + ], + "score": 0.87, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 535, + 506, + 550 + ], + "score": 1.0, + "content": "quantile over the batch-computed leak AUCs", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 547, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 505, + 560 + ], + "score": 1.0, + "content": "throughout all training iterations. This quantile is chosen instead of the mean because we want to", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 558, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 505, + 571 + ], + "score": 1.0, + "content": "measure the most-leaked iteration’s privacy leakage (highest leak AUC across iterations) to ensure", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 569, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 330, + 582 + ], + "score": 1.0, + "content": "the labels are not leaked at any points during training.", + "type": "text" + }, + { + "bbox": [ + 330, + 569, + 350, + 580 + ], + "score": 0.87, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 569, + 505, + 582 + ], + "score": 1.0, + "content": "quantile is chosen instead of the max", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 108, + 579, + 506, + 593 + ], + "spans": [ + { + "bbox": [ + 108, + 580, + 136, + 591 + ], + "score": 0.81, + "content": "( 1 0 0 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 579, + 506, + 593 + ], + "score": 1.0, + "content": ") as we want this privacy leak estimate to be robust against randomness of the training process.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 597, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "score": 1.0, + "content": "Privacy-Utility Tradeoff comparison results. In measuring the privacy-utility tradeoff, we aim to", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 607, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 505, + 621 + ], + "score": 1.0, + "content": "find a method that consistently achieves a lower leak AUC (better privacy) for the same utility value.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 112, + 622, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 112, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 112, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "• [Marvell vs iso] As shown in Figure 4, Marvell almost always achieves a better trade-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 120, + 633, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 120, + 633, + 506, + 646 + ], + "score": 1.0, + "content": "off than iso against both of our proposed attacks at both the cut layer and the first layer on", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 120, + 643, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 120, + 643, + 506, + 656 + ], + "score": 1.0, + "content": "both the ISIC and Criteo datasets. It is important to note that although the utility constraint is", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 120, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 120, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "in terms of training loss optimization, Marvell’s better tradeoff still translates to the gener-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 120, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 120, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "alization performance when the utility is measured through test loss or test AUC. Additionally,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 120, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 120, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "despite achieving reasonable (though still worse than Marvell) privacy-utility tradeoff against", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 120, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 120, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "the norm-based attack, iso performs much worse against the direction-based attack: on ISIC,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 120, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 120, + 699, + 387, + 711 + ], + "score": 1.0, + "content": "even after applying a significant amount of isotropic noise (with", + "type": "text" + }, + { + "bbox": [ + 387, + 699, + 418, + 709 + ], + "score": 0.87, + "content": "t > 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "), iso’s cosine leak", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 120, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 120, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "AUC is still higher than 0.9 at the cut layer (Figure 4(b,f)). In contrast, Marvell is effective", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 120, + 721, + 435, + 732 + ], + "spans": [ + { + "bbox": [ + 120, + 721, + 405, + 732 + ], + "score": 1.0, + "content": "against this direction-based attack with a much lower cosine leak AUC", + "type": "text" + }, + { + "bbox": [ + 405, + 721, + 430, + 731 + ], + "score": 0.73, + "content": "\\approx 0 . 6", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 721, + 435, + 732 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 43.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 78, + 505, + 168 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 78, + 505, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 78, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 505, + 168 + ], + "score": 0.963, + "type": "image", + "image_path": "da7562e01beb63c35fea76cbea790a0088e2c1811a01fbe1e54c87a5bcc7c358.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 78, + 505, + 108.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 108.0, + 505, + 138.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 138.0, + 505, + 168.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 169, + 503, + 189 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 167, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 505, + 180 + ], + "score": 1.0, + "content": "Figure 3: Norm and cosine leak AUC (computed every batch) at the cut layer and at the first layer under no", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 177, + 442, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 327, + 191 + ], + "score": 1.0, + "content": "protection vs. Marvell with different scale hyperparameter", + "type": "text" + }, + { + "bbox": [ + 328, + 180, + 334, + 187 + ], + "score": 0.54, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 177, + 442, + 191 + ], + "score": 1.0, + "content": "throughout the ISIC training.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 106, + 200, + 442, + 212 + ], + "lines": [ + { + "bbox": [ + 106, + 200, + 443, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 443, + 213 + ], + "score": 1.0, + "content": "5.1 LABEL LEAKAGE AND MA R V E L L’S STRONG AND FLEXIBLE PROTECTION", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 505, + 364 + ], + "lines": [ + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 233 + ], + "score": 1.0, + "content": "We first evaluate the protection quality of Marvell against the norm and cosine attacks dis-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "score": 1.0, + "content": "cussed in Section 3.3. We also compare against the leakage metrics when no protection is applied", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "(no noise). As the results across the three datasets are highly similar, we use ISIC as an example", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 254, + 504, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 504, + 266 + ], + "score": 1.0, + "content": "(other datasets see Appendix A.7.1). We see in Figure 3(a)(b) that unlike no noise where the label", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 264, + 278 + ], + "score": 1.0, + "content": "information is completely leaked (leak", + "type": "text" + }, + { + "bbox": [ + 265, + 265, + 306, + 276 + ], + "score": 0.5, + "content": "\\mathbf { A U C } \\approx 1", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 265, + 506, + 278 + ], + "score": 1.0, + "content": ") throughout training, Marvell achieves a flex-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 276, + 264, + 289 + ], + "score": 1.0, + "content": "ible degree of protection (by varying", + "type": "text" + }, + { + "bbox": [ + 265, + 277, + 271, + 286 + ], + "score": 0.25, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 276, + 505, + 289 + ], + "score": 1.0, + "content": ") against both the norm 2(a) and direction attacks 2(b)", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 374, + 300 + ], + "score": 1.0, + "content": "on the cut layer gradients and has strong protection (leak AUC", + "type": "text" + }, + { + "bbox": [ + 374, + 287, + 399, + 298 + ], + "score": 0.45, + "content": "\\approx { \\bf 0 . 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 287, + 414, + 300 + ], + "score": 1.0, + "content": ") at", + "type": "text" + }, + { + "bbox": [ + 414, + 287, + 446, + 298 + ], + "score": 0.89, + "content": "s = 4 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 287, + 505, + 300 + ], + "score": 1.0, + "content": ". Additionally,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 297, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 505, + 311 + ], + "score": 1.0, + "content": "it is natural to ask whether the gradients of layers before the cut layer (on the non-label party side)", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 308, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 104, + 308, + 505, + 323 + ], + "score": 1.0, + "content": "can also leak the labels as the non-label party keeps back propagating towards the first layer. In Fig-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "score": 1.0, + "content": "ure 3(c)(d), we compute the leak AUC values when using the non-label party’s first layer activation", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 330, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 104, + 330, + 320, + 344 + ], + "score": 1.0, + "content": "gradient as inputs to the scoring functions to predict", + "type": "text" + }, + { + "bbox": [ + 320, + 333, + 326, + 342 + ], + "score": 0.75, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 330, + 505, + 344 + ], + "score": 1.0, + "content": ". Without protection, the first layer gradient", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 341, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 355 + ], + "score": 1.0, + "content": "still leaks the label very consistently. In constrast, Marvell still achieves strong privacy protection", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 353, + 425, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 174, + 366 + ], + "score": 1.0, + "content": "at the first layer", + "type": "text" + }, + { + "bbox": [ + 174, + 353, + 207, + 364 + ], + "score": 0.85, + "content": "s = 4 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 353, + 425, + 366 + ], + "score": 1.0, + "content": ") despite the protection being analyzed at the cut layer.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 12, + "bbox_fs": [ + 104, + 221, + 506, + 366 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 378, + 322, + 389 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 322, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 322, + 390 + ], + "score": 1.0, + "content": "5.2 PRIVACY-UTILITY TRADE-OFF COMPARISON", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 399, + 505, + 509 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 412 + ], + "score": 1.0, + "content": "After showing Marvell can provide strong privacy protection against our identified attacks, we", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 409, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 423 + ], + "score": 1.0, + "content": "now see how well it can preserve utility by comparing its privacy-utility tradeoff against other pro-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 421, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 433 + ], + "score": 1.0, + "content": "tection baselines: no noise, isotropic Gaussian (iso), and our proposed heuristic max norm.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "score": 1.0, + "content": "Similar to how we allow Marvell to use a power constraint to depend on the current iteration’s", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 442, + 501, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 224, + 455 + ], + "score": 1.0, + "content": "gradient distribution through", + "type": "text" + }, + { + "bbox": [ + 225, + 442, + 280, + 454 + ], + "score": 0.92, + "content": "P = s \\| \\Delta g \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 442, + 501, + 455 + ], + "score": 1.0, + "content": ", we also allow iso to have such type of dependence—", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 191, + 466 + ], + "score": 1.0, + "content": "specifically, we add", + "type": "text" + }, + { + "bbox": [ + 191, + 454, + 316, + 465 + ], + "score": 0.89, + "content": "\\eta \\sim \\mathcal { N } ( \\mathbf { 0 } , ( t / d ) \\cdot \\| g _ { \\operatorname* { m a x } } \\| _ { 2 } ^ { 2 } I _ { d \\times d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 453, + 457, + 466 + ], + "score": 1.0, + "content": "to every gradient in a batch with", + "type": "text" + }, + { + "bbox": [ + 458, + 455, + 463, + 464 + ], + "score": 0.66, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "a tunable", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "privacy hyperparameter to be fixed throughout training. To trace out the complete tradeoff curve", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "score": 1.0, + "content": "for Marvell and iso, we conduct more than 20 training runs for each protection method with a", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 277, + 500 + ], + "score": 1.0, + "content": "different value of privacy hyperparameter", + "type": "text" + }, + { + "bbox": [ + 278, + 488, + 284, + 496 + ], + "score": 0.4, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 486, + 345, + 500 + ], + "score": 1.0, + "content": "for Marvell,", + "type": "text" + }, + { + "bbox": [ + 345, + 487, + 350, + 497 + ], + "score": 0.66, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 486, + 505, + 500 + ], + "score": 1.0, + "content": "for iso) in each run on every dataset.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 498, + 417, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 417, + 510 + ], + "score": 1.0, + "content": "(Note that no noise and max norm do not have privacy hyperparameters.)", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 398, + 506, + 510 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 514, + 505, + 592 + ], + "lines": [ + { + "bbox": [ + 106, + 514, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 505, + 527 + ], + "score": 1.0, + "content": "We present the tradeoffs between privacy (measured through norm and cosine leak AUC at cut", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 526, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 505, + 537 + ], + "score": 1.0, + "content": "layer/first layer) and utility (measured using test loss and test AUC) in Figure 4. To summarize the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 535, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 301, + 550 + ], + "score": 1.0, + "content": "leak AUC over a given training run, we pick the", + "type": "text" + }, + { + "bbox": [ + 301, + 537, + 321, + 547 + ], + "score": 0.87, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 535, + 506, + 550 + ], + "score": 1.0, + "content": "quantile over the batch-computed leak AUCs", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 547, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 505, + 560 + ], + "score": 1.0, + "content": "throughout all training iterations. This quantile is chosen instead of the mean because we want to", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 558, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 505, + 571 + ], + "score": 1.0, + "content": "measure the most-leaked iteration’s privacy leakage (highest leak AUC across iterations) to ensure", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 569, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 330, + 582 + ], + "score": 1.0, + "content": "the labels are not leaked at any points during training.", + "type": "text" + }, + { + "bbox": [ + 330, + 569, + 350, + 580 + ], + "score": 0.87, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 569, + 505, + 582 + ], + "score": 1.0, + "content": "quantile is chosen instead of the max", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 108, + 579, + 506, + 593 + ], + "spans": [ + { + "bbox": [ + 108, + 580, + 136, + 591 + ], + "score": 0.81, + "content": "( 1 0 0 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 579, + 506, + 593 + ], + "score": 1.0, + "content": ") as we want this privacy leak estimate to be robust against randomness of the training process.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 514, + 506, + 593 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 597, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "score": 1.0, + "content": "Privacy-Utility Tradeoff comparison results. In measuring the privacy-utility tradeoff, we aim to", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 607, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 505, + 621 + ], + "score": 1.0, + "content": "find a method that consistently achieves a lower leak AUC (better privacy) for the same utility value.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 596, + 505, + 621 + ] + }, + { + "type": "text", + "bbox": [ + 112, + 622, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 112, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 112, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "• [Marvell vs iso] As shown in Figure 4, Marvell almost always achieves a better trade-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 120, + 633, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 120, + 633, + 506, + 646 + ], + "score": 1.0, + "content": "off than iso against both of our proposed attacks at both the cut layer and the first layer on", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 120, + 643, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 120, + 643, + 506, + 656 + ], + "score": 1.0, + "content": "both the ISIC and Criteo datasets. It is important to note that although the utility constraint is", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 120, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 120, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "in terms of training loss optimization, Marvell’s better tradeoff still translates to the gener-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 120, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 120, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "alization performance when the utility is measured through test loss or test AUC. Additionally,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 120, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 120, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "despite achieving reasonable (though still worse than Marvell) privacy-utility tradeoff against", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 120, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 120, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "the norm-based attack, iso performs much worse against the direction-based attack: on ISIC,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 120, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 120, + 699, + 387, + 711 + ], + "score": 1.0, + "content": "even after applying a significant amount of isotropic noise (with", + "type": "text" + }, + { + "bbox": [ + 387, + 699, + 418, + 709 + ], + "score": 0.87, + "content": "t > 2 0", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "), iso’s cosine leak", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 120, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 120, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "AUC is still higher than 0.9 at the cut layer (Figure 4(b,f)). In contrast, Marvell is effective", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 120, + 721, + 435, + 732 + ], + "spans": [ + { + "bbox": [ + 120, + 721, + 405, + 732 + ], + "score": 1.0, + "content": "against this direction-based attack with a much lower cosine leak AUC", + "type": "text" + }, + { + "bbox": [ + 405, + 721, + 430, + 731 + ], + "score": 0.73, + "content": "\\approx 0 . 6", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 721, + 435, + 732 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 43.5, + "bbox_fs": [ + 112, + 621, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 78, + 500, + 320 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 78, + 500, + 320 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 78, + 500, + 320 + ], + "spans": [ + { + "bbox": [ + 110, + 78, + 500, + 320 + ], + "score": 0.973, + "type": "image", + "image_path": "48b9f068b52ed0c35fe5504d6d05601a466864114d83bc64d2db5cda6f53531e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 78, + 500, + 158.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 158.66666666666669, + 500, + 239.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 239.33333333333337, + 500, + 320.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 324, + 503, + 344 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 324, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 505, + 335 + ], + "score": 1.0, + "content": "Figure 4: Privacy (norm & cosine leak AUC) vs Utility (test loss & test AUC) trade-off of protection methods", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 333, + 423, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 423, + 345 + ], + "score": 1.0, + "content": "(Marvell, iso, no noise, max norm) at the cut and first layer on ISIC and Criteo.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 113, + 359, + 505, + 436 + ], + "lines": [ + { + "bbox": [ + 112, + 359, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 112, + 359, + 505, + 371 + ], + "score": 1.0, + "content": "• [max norm heuristic] Beyond Marvell, we see that our heuristic approach max norm can", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 121, + 370, + 504, + 381 + ], + "spans": [ + { + "bbox": [ + 121, + 370, + 504, + 381 + ], + "score": 1.0, + "content": "match and sometimes achieve even lower (Figure 4(a,f,i)) leak AUC value than Marvell at the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 120, + 381, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 120, + 381, + 505, + 393 + ], + "score": 1.0, + "content": "same utility level. We believe this specifically results from our norm and direction consideration", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 122, + 393, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 122, + 393, + 505, + 404 + ], + "score": 1.0, + "content": "when designing this heuristic. However, without a tunable hyperparameter, max norm cannot", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 120, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 120, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "tradeoff between privacy and utility. Additionally, unlike Marvell which is designed to protect", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 120, + 414, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 120, + 414, + 506, + 426 + ], + "score": 1.0, + "content": "against the entire class of adversarial scoring functions, max norm might still fail to protect", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 120, + 425, + 388, + 437 + ], + "spans": [ + { + "bbox": [ + 120, + 425, + 388, + 437 + ], + "score": 1.0, + "content": "against other future attack methods beyond those considered here.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 108, + 446, + 505, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 445, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 505, + 459 + ], + "score": 1.0, + "content": "In summary, our principled method Marvell significantly outperforms the isotropic Gaussian", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 457, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 505, + 469 + ], + "score": 1.0, + "content": "baseline, and our proposed max norm heuristic can also work particularly well against the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 468, + 397, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 397, + 479 + ], + "score": 1.0, + "content": "norm- and direction-based attacks which we identified in Section 3.3.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "title", + "bbox": [ + 107, + 491, + 195, + 504 + ], + "lines": [ + { + "bbox": [ + 105, + 489, + 198, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 198, + 507 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 513, + 505, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "score": 1.0, + "content": "In this paper, we formulate a label leakage threat model in the two-party split learning binary clas-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 523, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 506, + 536 + ], + "score": 1.0, + "content": "sification problem through a novel privacy loss quantification metric (leak AUC). Within this threat", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "model, we provide two simple yet effective attack methods that can accurately uncover the private", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "score": 1.0, + "content": "labels of the label party. To counter such attacks, we propose a heuristic random perturbation method", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "max norm as well as a theoretically principled method Marvell which searches for the optimal", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 568, + 505, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 505, + 580 + ], + "score": 1.0, + "content": "noise distributions to protect against the worst-case adversaries in the threat model. We have con-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 579, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 505, + 591 + ], + "score": 1.0, + "content": "ducted extensive experiments to demonstrate the effectiveness of Marvell and max norm over", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 590, + 307, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 307, + 603 + ], + "score": 1.0, + "content": "the isotropic Gaussian perturbation baseline iso.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 505, + 727 + ], + "lines": [ + { + "bbox": [ + 106, + 606, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 505, + 619 + ], + "score": 1.0, + "content": "Open questions and future work. Our work is the first we are aware of to identify, rigorously", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 618, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 506, + 630 + ], + "score": 1.0, + "content": "quantify, and protect against the threat of label leakage in split-learning, and opens up a number of", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 628, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 506, + 642 + ], + "score": 1.0, + "content": "worthy directions of future study. In particular, as the model parameters are updated every batch in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 639, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 104, + 639, + 505, + 653 + ], + "score": 1.0, + "content": "our problem setup, the true gradient of an example and the gradient distribution would both change.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 651, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 506, + 663 + ], + "score": 1.0, + "content": "An interesting question is whether the adversarial non-label party can remember the stale gradient", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 662, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 674 + ], + "score": 1.0, + "content": "of the same example from past updates (possibly separated by hundreds of updates steps) in order to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 672, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 505, + 684 + ], + "score": 1.0, + "content": "recover the label information in the current iteration in a more complex threat model. It would also be", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 684, + 505, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 505, + 695 + ], + "score": 1.0, + "content": "interesting to build on our results to study whether there exist attack methods when the classification", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 694, + 505, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 505, + 706 + ], + "score": 1.0, + "content": "problem is multiclass instead of binary, and when the split learning scenario involves more than two", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 705, + 505, + 718 + ], + "spans": [ + { + "bbox": [ + 105, + 705, + 505, + 718 + ], + "score": 1.0, + "content": "parties with possibly more complicated training communication protocols (e.g., Vepakomma et al.,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 714, + 136, + 729 + ], + "spans": [ + { + "bbox": [ + 105, + 714, + 136, + 729 + ], + "score": 1.0, + "content": "2018).", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 78, + 500, + 320 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 78, + 500, + 320 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 78, + 500, + 320 + ], + "spans": [ + { + "bbox": [ + 110, + 78, + 500, + 320 + ], + "score": 0.973, + "type": "image", + "image_path": "48b9f068b52ed0c35fe5504d6d05601a466864114d83bc64d2db5cda6f53531e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 78, + 500, + 158.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 158.66666666666669, + 500, + 239.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 239.33333333333337, + 500, + 320.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 324, + 503, + 344 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 324, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 505, + 335 + ], + "score": 1.0, + "content": "Figure 4: Privacy (norm & cosine leak AUC) vs Utility (test loss & test AUC) trade-off of protection methods", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 333, + 423, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 423, + 345 + ], + "score": 1.0, + "content": "(Marvell, iso, no noise, max norm) at the cut and first layer on ISIC and Criteo.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 113, + 359, + 505, + 436 + ], + "lines": [ + { + "bbox": [ + 112, + 359, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 112, + 359, + 505, + 371 + ], + "score": 1.0, + "content": "• [max norm heuristic] Beyond Marvell, we see that our heuristic approach max norm can", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 121, + 370, + 504, + 381 + ], + "spans": [ + { + "bbox": [ + 121, + 370, + 504, + 381 + ], + "score": 1.0, + "content": "match and sometimes achieve even lower (Figure 4(a,f,i)) leak AUC value than Marvell at the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 120, + 381, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 120, + 381, + 505, + 393 + ], + "score": 1.0, + "content": "same utility level. We believe this specifically results from our norm and direction consideration", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 122, + 393, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 122, + 393, + 505, + 404 + ], + "score": 1.0, + "content": "when designing this heuristic. However, without a tunable hyperparameter, max norm cannot", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 120, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 120, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "tradeoff between privacy and utility. Additionally, unlike Marvell which is designed to protect", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 120, + 414, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 120, + 414, + 506, + 426 + ], + "score": 1.0, + "content": "against the entire class of adversarial scoring functions, max norm might still fail to protect", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 120, + 425, + 388, + 437 + ], + "spans": [ + { + "bbox": [ + 120, + 425, + 388, + 437 + ], + "score": 1.0, + "content": "against other future attack methods beyond those considered here.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8, + "bbox_fs": [ + 112, + 359, + 506, + 437 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 446, + 505, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 445, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 505, + 459 + ], + "score": 1.0, + "content": "In summary, our principled method Marvell significantly outperforms the isotropic Gaussian", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 457, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 505, + 469 + ], + "score": 1.0, + "content": "baseline, and our proposed max norm heuristic can also work particularly well against the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 468, + 397, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 397, + 479 + ], + "score": 1.0, + "content": "norm- and direction-based attacks which we identified in Section 3.3.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 106, + 445, + 505, + 479 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 491, + 195, + 504 + ], + "lines": [ + { + "bbox": [ + 105, + 489, + 198, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 198, + 507 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 513, + 505, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "score": 1.0, + "content": "In this paper, we formulate a label leakage threat model in the two-party split learning binary clas-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 523, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 506, + 536 + ], + "score": 1.0, + "content": "sification problem through a novel privacy loss quantification metric (leak AUC). Within this threat", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "model, we provide two simple yet effective attack methods that can accurately uncover the private", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "score": 1.0, + "content": "labels of the label party. To counter such attacks, we propose a heuristic random perturbation method", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "max norm as well as a theoretically principled method Marvell which searches for the optimal", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 568, + 505, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 505, + 580 + ], + "score": 1.0, + "content": "noise distributions to protect against the worst-case adversaries in the threat model. We have con-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 579, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 505, + 591 + ], + "score": 1.0, + "content": "ducted extensive experiments to demonstrate the effectiveness of Marvell and max norm over", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 590, + 307, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 307, + 603 + ], + "score": 1.0, + "content": "the isotropic Gaussian perturbation baseline iso.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 512, + 506, + 603 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 505, + 727 + ], + "lines": [ + { + "bbox": [ + 106, + 606, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 505, + 619 + ], + "score": 1.0, + "content": "Open questions and future work. Our work is the first we are aware of to identify, rigorously", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 618, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 506, + 630 + ], + "score": 1.0, + "content": "quantify, and protect against the threat of label leakage in split-learning, and opens up a number of", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 628, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 506, + 642 + ], + "score": 1.0, + "content": "worthy directions of future study. In particular, as the model parameters are updated every batch in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 639, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 104, + 639, + 505, + 653 + ], + "score": 1.0, + "content": "our problem setup, the true gradient of an example and the gradient distribution would both change.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 651, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 506, + 663 + ], + "score": 1.0, + "content": "An interesting question is whether the adversarial non-label party can remember the stale gradient", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 662, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 674 + ], + "score": 1.0, + "content": "of the same example from past updates (possibly separated by hundreds of updates steps) in order to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 672, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 505, + 684 + ], + "score": 1.0, + "content": "recover the label information in the current iteration in a more complex threat model. It would also be", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 684, + 505, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 505, + 695 + ], + "score": 1.0, + "content": "interesting to build on our results to study whether there exist attack methods when the classification", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 694, + 505, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 505, + 706 + ], + "score": 1.0, + "content": "problem is multiclass instead of binary, and when the split learning scenario involves more than two", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 705, + 505, + 718 + ], + "spans": [ + { + "bbox": [ + 105, + 705, + 505, + 718 + ], + "score": 1.0, + "content": "parties with possibly more complicated training communication protocols (e.g., Vepakomma et al.,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 714, + 136, + 729 + ], + "spans": [ + { + "bbox": [ + 105, + 714, + 136, + 729 + ], + "score": 1.0, + "content": "2018).", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29, + "bbox_fs": [ + 104, + 606, + 506, + 729 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Ethics Statement. In our paper, we have identified a realistic threat of label leakage in the two-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "score": 1.0, + "content": "party split learning binary classification problem. We aim to raise awareness about potential privacy", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "issues in this problem domain, where many industrial applications have been deployed. Beyond", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "making such threats clear, we have taken the first steps towards protection—we have proposed both", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "score": 1.0, + "content": "heuristic and principled methods that can preserve label privacy. We hope our work will pave the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "way for future analyses that make the two-party split learning framework more effective and secure.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 161, + 412, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 415, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 415, + 176 + ], + "score": 1.0, + "content": "Reproducibility Statement. To make our paper reproducible, we provide:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 134, + 178, + 505, + 233 + ], + "lines": [ + { + "bbox": [ + 132, + 177, + 396, + 190 + ], + "spans": [ + { + "bbox": [ + 132, + 177, + 396, + 190 + ], + "score": 1.0, + "content": "• Proofs of our Theorem 1 and Theorem 2 in Appendix A.3,A.4;", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 132, + 188, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 132, + 188, + 505, + 201 + ], + "score": 1.0, + "content": "• Detailed experiment description including 1) data preprocessing, 2) model architecture, 3)", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 200, + 374, + 212 + ], + "spans": [ + { + "bbox": [ + 141, + 200, + 374, + 212 + ], + "score": 1.0, + "content": "training algorithm and hyperparameters in Appendix A.6.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 134, + 210, + 380, + 223 + ], + "spans": [ + { + "bbox": [ + 134, + 210, + 380, + 223 + ], + "score": 1.0, + "content": "• Source code with running instructions (in README.md) at", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 221, + 332, + 234 + ], + "spans": [ + { + "bbox": [ + 141, + 221, + 332, + 234 + ], + "score": 1.0, + "content": "https://github.com/OscarcarLi/label-protection.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 107, + 250, + 175, + 262 + ], + "lines": [ + { + "bbox": [ + 106, + 249, + 176, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 176, + 264 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 109, + 268, + 503, + 302 + ], + "lines": [ + { + "bbox": [ + 107, + 267, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 107, + 267, + 505, + 281 + ], + "score": 1.0, + "content": "Martin Abadi, Andy Chu, Ian Goodfellow, H Brendan McMahan, Ilya Mironov, Kunal Talwar, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 279, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 115, + 279, + 505, + 293 + ], + "score": 1.0, + "content": "Li Zhang. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 117, + 290, + 424, + 304 + ], + "spans": [ + { + "bbox": [ + 117, + 290, + 424, + 304 + ], + "score": 1.0, + "content": "Conference on Computer and Communications Security, pp. 308–318, 2016.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 309, + 503, + 332 + ], + "lines": [ + { + "bbox": [ + 105, + 307, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 505, + 322 + ], + "score": 1.0, + "content": "Avazu. Avazu click-through rate prediction, 2015. URL https://www.kaggle.com/c/", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 321, + 271, + 332 + ], + "spans": [ + { + "bbox": [ + 116, + 321, + 271, + 332 + ], + "score": 1.0, + "content": "avazu-ctr-prediction/data.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 107, + 339, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 340, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 505, + 351 + ], + "score": 1.0, + "content": "Keith Bonawitz, Vladimir Ivanov, Ben Kreuter, Antonio Marcedone, H Brendan McMahan, Sarvar", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 350, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 115, + 350, + 505, + 363 + ], + "score": 1.0, + "content": "Patel, Daniel Ramage, Aaron Segal, and Karn Seth. Practical secure aggregation for privacy-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 361, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 115, + 361, + 506, + 375 + ], + "score": 1.0, + "content": "preserving machine learning. In Proceedings of the 2017 ACM SIGSAC Conference on Computer", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 372, + 331, + 385 + ], + "spans": [ + { + "bbox": [ + 115, + 372, + 331, + 385 + ], + "score": 1.0, + "content": "and Communications Security, pp. 1175–1191, 2017.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 109, + 391, + 504, + 425 + ], + "lines": [ + { + "bbox": [ + 105, + 389, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 505, + 405 + ], + "score": 1.0, + "content": "Kamalika Chaudhuri and Daniel Hsu. Sample complexity bounds for differentially private learn-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 402, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 115, + 402, + 505, + 416 + ], + "score": 1.0, + "content": "ing. In Proceedings of the 24th Annual Conference on Learning Theory, pp. 155–186. JMLR", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 413, + 303, + 426 + ], + "spans": [ + { + "bbox": [ + 116, + 413, + 303, + 426 + ], + "score": 1.0, + "content": "Workshop and Conference Proceedings, 2011.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 432, + 505, + 477 + ], + "lines": [ + { + "bbox": [ + 105, + 431, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 505, + 446 + ], + "score": 1.0, + "content": "Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 442, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 115, + 442, + 505, + 456 + ], + "score": 1.0, + "content": "Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al. Wide & deep learning for recom-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 454, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 115, + 454, + 506, + 468 + ], + "score": 1.0, + "content": "mender systems. In Proceedings of the 1st workshop on deep learning for recommender systems,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 466, + 182, + 477 + ], + "spans": [ + { + "bbox": [ + 115, + 466, + 182, + 477 + ], + "score": 1.0, + "content": "pp. 7–10, 2016.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 484, + 504, + 518 + ], + "lines": [ + { + "bbox": [ + 106, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "Albert Cheu, Adam Smith, Jonathan Ullman, David Zeber, and Maxim Zhilyaev. Distributed differ-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 495, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 115, + 495, + 505, + 508 + ], + "score": 1.0, + "content": "ential privacy via shuffling. In Annual International Conference on the Theory and Applications", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 507, + 355, + 519 + ], + "spans": [ + { + "bbox": [ + 116, + 507, + 355, + 519 + ], + "score": 1.0, + "content": "of Cryptographic Techniques, pp. 375–403. Springer, 2019.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 503, + 548 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 504, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 504, + 538 + ], + "score": 1.0, + "content": "Criteo. Criteo display advertising challenge, 2014. URL https://www.kaggle.com/c/", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 117, + 537, + 314, + 549 + ], + "spans": [ + { + "bbox": [ + 117, + 537, + 314, + 549 + ], + "score": 1.0, + "content": "criteo-display-ad-challenge/data.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 107, + 555, + 504, + 589 + ], + "lines": [ + { + "bbox": [ + 106, + 555, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 568 + ], + "score": 1.0, + "content": "Cynthia Dwork. Differential privacy. In Michele Bugliesi, Bart Preneel, Vladimiro Sassone, and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 566, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 115, + 566, + 506, + 580 + ], + "score": 1.0, + "content": "Ingo Wegener (eds.), Automata, Languages and Programming, pp. 1–12, Berlin, Heidelberg,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 577, + 365, + 590 + ], + "spans": [ + { + "bbox": [ + 116, + 577, + 365, + 590 + ], + "score": 1.0, + "content": "2006. Springer Berlin Heidelberg. ISBN 978-3-540-35908-1.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 104, + 596, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "score": 1.0, + "content": "Cynthia Dwork, Aaron Roth, et al. The algorithmic foundations of differential privacy. Foundations", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 607, + 392, + 620 + ], + "spans": [ + { + "bbox": [ + 115, + 607, + 392, + 620 + ], + "score": 1.0, + "content": "and Trends in Theoretical Computer Science, 9(3-4):211–407, 2014.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 627, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "Ulfar Erlingsson, Vitaly Feldman, Ilya Mironov, Ananth Raghunathan, Kunal Talwar, and ´", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 117, + 639, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 117, + 639, + 505, + 651 + ], + "score": 1.0, + "content": "Abhradeep Thakurta. Amplification by shuffling: From local to central differential privacy via", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 650, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 116, + 650, + 505, + 662 + ], + "score": 1.0, + "content": "anonymity. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algo-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 661, + 267, + 673 + ], + "spans": [ + { + "bbox": [ + 116, + 661, + 267, + 673 + ], + "score": 1.0, + "content": "rithms, pp. 2468–2479. SIAM, 2019.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 106, + 679, + 502, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 504, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 504, + 693 + ], + "score": 1.0, + "content": "Badih Ghazi, Noah Golowich, Ravi Kumar, Pasin Manurangsi, and Chiyuan Zhang. On deep learn-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 691, + 417, + 703 + ], + "spans": [ + { + "bbox": [ + 115, + 691, + 417, + 703 + ], + "score": 1.0, + "content": "ing with label differential privacy. arXiv preprint arXiv:2102.06062, 2021.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 503, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 503, + 722 + ], + "score": 1.0, + "content": "Otkrist Gupta and Ramesh Raskar. Distributed learning of deep neural network over multiple agents.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 720, + 374, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 374, + 732 + ], + "score": 1.0, + "content": "Journal of Network and Computer Applications, 116:1–8, 2018.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Ethics Statement. In our paper, we have identified a realistic threat of label leakage in the two-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "score": 1.0, + "content": "party split learning binary classification problem. We aim to raise awareness about potential privacy", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "issues in this problem domain, where many industrial applications have been deployed. Beyond", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "making such threats clear, we have taken the first steps towards protection—we have proposed both", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "score": 1.0, + "content": "heuristic and principled methods that can preserve label privacy. We hope our work will pave the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "way for future analyses that make the two-party split learning framework more effective and secure.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 82, + 505, + 150 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 161, + 412, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 415, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 415, + 176 + ], + "score": 1.0, + "content": "Reproducibility Statement. To make our paper reproducible, we provide:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 158, + 415, + 176 + ] + }, + { + "type": "list", + "bbox": [ + 134, + 178, + 505, + 233 + ], + "lines": [ + { + "bbox": [ + 132, + 177, + 396, + 190 + ], + "spans": [ + { + "bbox": [ + 132, + 177, + 396, + 190 + ], + "score": 1.0, + "content": "• Proofs of our Theorem 1 and Theorem 2 in Appendix A.3,A.4;", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 188, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 132, + 188, + 505, + 201 + ], + "score": 1.0, + "content": "• Detailed experiment description including 1) data preprocessing, 2) model architecture, 3)", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 200, + 374, + 212 + ], + "spans": [ + { + "bbox": [ + 141, + 200, + 374, + 212 + ], + "score": 1.0, + "content": "training algorithm and hyperparameters in Appendix A.6.", + "type": "text" + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 134, + 210, + 380, + 223 + ], + "spans": [ + { + "bbox": [ + 134, + 210, + 380, + 223 + ], + "score": 1.0, + "content": "• Source code with running instructions (in README.md) at", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 141, + 221, + 332, + 234 + ], + "spans": [ + { + "bbox": [ + 141, + 221, + 332, + 234 + ], + "score": 1.0, + "content": "https://github.com/OscarcarLi/label-protection.", + "type": "text" + } + ], + "index": 11, + "is_list_end_line": true + } + ], + "index": 9, + "bbox_fs": [ + 132, + 177, + 505, + 234 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 250, + 175, + 262 + ], + "lines": [ + { + "bbox": [ + 106, + 249, + 176, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 176, + 264 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 109, + 268, + 503, + 302 + ], + "lines": [ + { + "bbox": [ + 107, + 267, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 107, + 267, + 505, + 281 + ], + "score": 1.0, + "content": "Martin Abadi, Andy Chu, Ian Goodfellow, H Brendan McMahan, Ilya Mironov, Kunal Talwar, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 279, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 115, + 279, + 505, + 293 + ], + "score": 1.0, + "content": "Li Zhang. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 117, + 290, + 424, + 304 + ], + "spans": [ + { + "bbox": [ + 117, + 290, + 424, + 304 + ], + "score": 1.0, + "content": "Conference on Computer and Communications Security, pp. 308–318, 2016.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 107, + 267, + 505, + 304 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 309, + 503, + 332 + ], + "lines": [ + { + "bbox": [ + 105, + 307, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 505, + 322 + ], + "score": 1.0, + "content": "Avazu. Avazu click-through rate prediction, 2015. URL https://www.kaggle.com/c/", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 321, + 271, + 332 + ], + "spans": [ + { + "bbox": [ + 116, + 321, + 271, + 332 + ], + "score": 1.0, + "content": "avazu-ctr-prediction/data.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 307, + 505, + 332 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 339, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 340, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 505, + 351 + ], + "score": 1.0, + "content": "Keith Bonawitz, Vladimir Ivanov, Ben Kreuter, Antonio Marcedone, H Brendan McMahan, Sarvar", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 350, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 115, + 350, + 505, + 363 + ], + "score": 1.0, + "content": "Patel, Daniel Ramage, Aaron Segal, and Karn Seth. Practical secure aggregation for privacy-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 361, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 115, + 361, + 506, + 375 + ], + "score": 1.0, + "content": "preserving machine learning. In Proceedings of the 2017 ACM SIGSAC Conference on Computer", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 372, + 331, + 385 + ], + "spans": [ + { + "bbox": [ + 115, + 372, + 331, + 385 + ], + "score": 1.0, + "content": "and Communications Security, pp. 1175–1191, 2017.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 340, + 506, + 385 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 391, + 504, + 425 + ], + "lines": [ + { + "bbox": [ + 105, + 389, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 505, + 405 + ], + "score": 1.0, + "content": "Kamalika Chaudhuri and Daniel Hsu. Sample complexity bounds for differentially private learn-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 402, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 115, + 402, + 505, + 416 + ], + "score": 1.0, + "content": "ing. In Proceedings of the 24th Annual Conference on Learning Theory, pp. 155–186. JMLR", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 413, + 303, + 426 + ], + "spans": [ + { + "bbox": [ + 116, + 413, + 303, + 426 + ], + "score": 1.0, + "content": "Workshop and Conference Proceedings, 2011.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 389, + 505, + 426 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 432, + 505, + 477 + ], + "lines": [ + { + "bbox": [ + 105, + 431, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 505, + 446 + ], + "score": 1.0, + "content": "Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 442, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 115, + 442, + 505, + 456 + ], + "score": 1.0, + "content": "Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al. Wide & deep learning for recom-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 454, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 115, + 454, + 506, + 468 + ], + "score": 1.0, + "content": "mender systems. In Proceedings of the 1st workshop on deep learning for recommender systems,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 466, + 182, + 477 + ], + "spans": [ + { + "bbox": [ + 115, + 466, + 182, + 477 + ], + "score": 1.0, + "content": "pp. 7–10, 2016.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 431, + 506, + 477 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 484, + 504, + 518 + ], + "lines": [ + { + "bbox": [ + 106, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "Albert Cheu, Adam Smith, Jonathan Ullman, David Zeber, and Maxim Zhilyaev. Distributed differ-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 495, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 115, + 495, + 505, + 508 + ], + "score": 1.0, + "content": "ential privacy via shuffling. In Annual International Conference on the Theory and Applications", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 507, + 355, + 519 + ], + "spans": [ + { + "bbox": [ + 116, + 507, + 355, + 519 + ], + "score": 1.0, + "content": "of Cryptographic Techniques, pp. 375–403. Springer, 2019.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 484, + 505, + 519 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 503, + 548 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 504, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 504, + 538 + ], + "score": 1.0, + "content": "Criteo. Criteo display advertising challenge, 2014. URL https://www.kaggle.com/c/", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 117, + 537, + 314, + 549 + ], + "spans": [ + { + "bbox": [ + 117, + 537, + 314, + 549 + ], + "score": 1.0, + "content": "criteo-display-ad-challenge/data.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 523, + 504, + 549 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 555, + 504, + 589 + ], + "lines": [ + { + "bbox": [ + 106, + 555, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 568 + ], + "score": 1.0, + "content": "Cynthia Dwork. Differential privacy. In Michele Bugliesi, Bart Preneel, Vladimiro Sassone, and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 566, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 115, + 566, + 506, + 580 + ], + "score": 1.0, + "content": "Ingo Wegener (eds.), Automata, Languages and Programming, pp. 1–12, Berlin, Heidelberg,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 577, + 365, + 590 + ], + "spans": [ + { + "bbox": [ + 116, + 577, + 365, + 590 + ], + "score": 1.0, + "content": "2006. Springer Berlin Heidelberg. ISBN 978-3-540-35908-1.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 106, + 555, + 506, + 590 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 596, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 610 + ], + "score": 1.0, + "content": "Cynthia Dwork, Aaron Roth, et al. The algorithmic foundations of differential privacy. Foundations", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 607, + 392, + 620 + ], + "spans": [ + { + "bbox": [ + 115, + 607, + 392, + 620 + ], + "score": 1.0, + "content": "and Trends in Theoretical Computer Science, 9(3-4):211–407, 2014.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 595, + 505, + 620 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 627, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "Ulfar Erlingsson, Vitaly Feldman, Ilya Mironov, Ananth Raghunathan, Kunal Talwar, and ´", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 117, + 639, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 117, + 639, + 505, + 651 + ], + "score": 1.0, + "content": "Abhradeep Thakurta. Amplification by shuffling: From local to central differential privacy via", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 650, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 116, + 650, + 505, + 662 + ], + "score": 1.0, + "content": "anonymity. In Proceedings of the Thirtieth Annual ACM-SIAM Symposium on Discrete Algo-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 661, + 267, + 673 + ], + "spans": [ + { + "bbox": [ + 116, + 661, + 267, + 673 + ], + "score": 1.0, + "content": "rithms, pp. 2468–2479. SIAM, 2019.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5, + "bbox_fs": [ + 106, + 627, + 506, + 673 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 679, + 502, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 678, + 504, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 504, + 693 + ], + "score": 1.0, + "content": "Badih Ghazi, Noah Golowich, Ravi Kumar, Pasin Manurangsi, and Chiyuan Zhang. On deep learn-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 691, + 417, + 703 + ], + "spans": [ + { + "bbox": [ + 115, + 691, + 417, + 703 + ], + "score": 1.0, + "content": "ing with label differential privacy. arXiv preprint arXiv:2102.06062, 2021.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 678, + 504, + 703 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 503, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 503, + 722 + ], + "score": 1.0, + "content": "Otkrist Gupta and Ramesh Raskar. Distributed learning of deep neural network over multiple agents.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 720, + 374, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 374, + 732 + ], + "score": 1.0, + "content": "Journal of Network and Computer Applications, 116:1–8, 2018.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 708, + 503, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 70, + 506, + 479 + ], + "lines": [ + { + "bbox": [ + 104, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "ISIC. Siim-isic melanoma classification, 2020. URL https://www.kaggle.com/c/", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 350, + 107 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 350, + 107 + ], + "score": 1.0, + "content": "siim-isic-melanoma-classification/data.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 112, + 506, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 506, + 126 + ], + "score": 1.0, + "content": "Vladimir Kolesnikov, Ranjit Kumaresan, Mike Rosulek, and Ni Trieu. Efficient batched oblivi-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 506, + 136 + ], + "score": 1.0, + "content": "ous prf with applications to private set intersection. In Proceedings of the 2016 ACM SIGSAC", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 117, + 135, + 464, + 147 + ], + "spans": [ + { + "bbox": [ + 117, + 135, + 464, + 147 + ], + "score": 1.0, + "content": "Conference on Computer and Communications Security, CCS ’16, pp. 818–829, 2016.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 152, + 504, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 152, + 504, + 167 + ], + "score": 1.0, + "content": "Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 164, + 504, + 177 + ], + "spans": [ + { + "bbox": [ + 115, + 164, + 504, + 177 + ], + "score": 1.0, + "content": "Communication-efficient learning of deep networks from decentralized data. In Artificial Intelli-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 323, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 323, + 188 + ], + "score": 1.0, + "content": "gence and Statistics, pp. 1273–1282. PMLR, 2017.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 194, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 505, + 208 + ], + "score": 1.0, + "content": "H. Brendan McMahan, Daniel Ramage, Kunal Talwar, and Li Zhang. Learning differentially private", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 205, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 115, + 205, + 505, + 218 + ], + "score": 1.0, + "content": "recurrent language models. In International Conference on Learning Representations, 2018. URL", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 216, + 366, + 229 + ], + "spans": [ + { + "bbox": [ + 115, + 216, + 366, + 229 + ], + "score": 1.0, + "content": "https://openreview.net/forum?id=BJ0hF1Z0b.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 235, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 248 + ], + "score": 1.0, + "content": "Benny Pinkas, Thomas Schneider, and Michael Zohner. Scalable private set intersection based on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 246, + 456, + 258 + ], + "spans": [ + { + "bbox": [ + 115, + 246, + 456, + 258 + ], + "score": 1.0, + "content": "ot extension. ACM Transactions on Privacy and Security (TOPS), 21(2):1–35, 2018.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "score": 1.0, + "content": "Pramod Subramanyan, Rohit Sinha, Ilia Lebedev, Srinivas Devadas, and Sanjit A Seshia. A formal", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 277, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 116, + 277, + 505, + 288 + ], + "score": 1.0, + "content": "foundation for secure remote execution of enclaves. In Proceedings of the 2017 ACM SIGSAC", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 287, + 434, + 300 + ], + "spans": [ + { + "bbox": [ + 116, + 287, + 434, + 300 + ], + "score": 1.0, + "content": "Conference on Computer and Communications Security, pp. 2435–2450, 2017.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 305, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 505, + 319 + ], + "score": 1.0, + "content": "Praneeth Vepakomma, Otkrist Gupta, Tristan Swedish, and Ramesh Raskar. Split learning for health:", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 115, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "Distributed deep learning without sharing raw patient data. arXiv preprint arXiv:1812.00564,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 327, + 142, + 340 + ], + "spans": [ + { + "bbox": [ + 116, + 327, + 142, + 340 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 345, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 506, + 360 + ], + "score": 1.0, + "content": "Praneeth Vepakomma, Otkrist Gupta, Abhimanyu Dubey, and Ramesh Raskar. Reducing leakage in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 358, + 480, + 370 + ], + "spans": [ + { + "bbox": [ + 115, + 358, + 480, + 370 + ], + "score": 1.0, + "content": "distributed deep learning for sensitive health data. arXiv preprint arXiv:1812.00564, 2019.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 375, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 506, + 390 + ], + "score": 1.0, + "content": "Qiang Yang, Yang Liu, Tianjian Chen, and Yongxin Tong. Federated machine learning: Concept", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 114, + 387, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 114, + 387, + 506, + 401 + ], + "score": 1.0, + "content": "and applications. ACM Transactions on Intelligent Systems and Technology (TIST), 10(2):1–19,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 398, + 142, + 411 + ], + "spans": [ + { + "bbox": [ + 116, + 398, + 142, + 411 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 416, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 104, + 416, + 505, + 431 + ], + "score": 1.0, + "content": "Bo Zhao, Konda Reddy Mopuri, and Hakan Bilen. idlg: Improved deep leakage from gradients.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 428, + 279, + 441 + ], + "spans": [ + { + "bbox": [ + 115, + 428, + 279, + 441 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2001.02610, 2020.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "score": 1.0, + "content": "Ligeng Zhu, Zhijian Liu, and Song Han. Deep leakage from gradients. In Advances in Neural", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 458, + 350, + 472 + ], + "spans": [ + { + "bbox": [ + 115, + 458, + 350, + 472 + ], + "score": 1.0, + "content": "Information Processing Systems, pp. 14774–14784, 2019.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 13.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 104, + 70, + 506, + 479 + ], + "lines": [ + { + "bbox": [ + 104, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "ISIC. Siim-isic melanoma classification, 2020. URL https://www.kaggle.com/c/", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 93, + 350, + 107 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 350, + 107 + ], + "score": 1.0, + "content": "siim-isic-melanoma-classification/data.", + "type": "text" + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 112, + 506, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 506, + 126 + ], + "score": 1.0, + "content": "Vladimir Kolesnikov, Ranjit Kumaresan, Mike Rosulek, and Ni Trieu. Efficient batched oblivi-", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 123, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 506, + 136 + ], + "score": 1.0, + "content": "ous prf with applications to private set intersection. In Proceedings of the 2016 ACM SIGSAC", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 117, + 135, + 464, + 147 + ], + "spans": [ + { + "bbox": [ + 117, + 135, + 464, + 147 + ], + "score": 1.0, + "content": "Conference on Computer and Communications Security, CCS ’16, pp. 818–829, 2016.", + "type": "text" + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 152, + 504, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 152, + 504, + 167 + ], + "score": 1.0, + "content": "Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas.", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 164, + 504, + 177 + ], + "spans": [ + { + "bbox": [ + 115, + 164, + 504, + 177 + ], + "score": 1.0, + "content": "Communication-efficient learning of deep networks from decentralized data. In Artificial Intelli-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 323, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 323, + 188 + ], + "score": 1.0, + "content": "gence and Statistics, pp. 1273–1282. PMLR, 2017.", + "type": "text" + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 194, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 505, + 208 + ], + "score": 1.0, + "content": "H. Brendan McMahan, Daniel Ramage, Kunal Talwar, and Li Zhang. Learning differentially private", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 205, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 115, + 205, + 505, + 218 + ], + "score": 1.0, + "content": "recurrent language models. In International Conference on Learning Representations, 2018. URL", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 216, + 366, + 229 + ], + "spans": [ + { + "bbox": [ + 115, + 216, + 366, + 229 + ], + "score": 1.0, + "content": "https://openreview.net/forum?id=BJ0hF1Z0b.", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 235, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 248 + ], + "score": 1.0, + "content": "Benny Pinkas, Thomas Schneider, and Michael Zohner. Scalable private set intersection based on", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 246, + 456, + 258 + ], + "spans": [ + { + "bbox": [ + 115, + 246, + 456, + 258 + ], + "score": 1.0, + "content": "ot extension. ACM Transactions on Privacy and Security (TOPS), 21(2):1–35, 2018.", + "type": "text" + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 506, + 278 + ], + "score": 1.0, + "content": "Pramod Subramanyan, Rohit Sinha, Ilia Lebedev, Srinivas Devadas, and Sanjit A Seshia. A formal", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 277, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 116, + 277, + 505, + 288 + ], + "score": 1.0, + "content": "foundation for secure remote execution of enclaves. In Proceedings of the 2017 ACM SIGSAC", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 287, + 434, + 300 + ], + "spans": [ + { + "bbox": [ + 116, + 287, + 434, + 300 + ], + "score": 1.0, + "content": "Conference on Computer and Communications Security, pp. 2435–2450, 2017.", + "type": "text" + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 305, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 505, + 319 + ], + "score": 1.0, + "content": "Praneeth Vepakomma, Otkrist Gupta, Tristan Swedish, and Ramesh Raskar. Split learning for health:", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 115, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "Distributed deep learning without sharing raw patient data. arXiv preprint arXiv:1812.00564,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 327, + 142, + 340 + ], + "spans": [ + { + "bbox": [ + 116, + 327, + 142, + 340 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 345, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 506, + 360 + ], + "score": 1.0, + "content": "Praneeth Vepakomma, Otkrist Gupta, Abhimanyu Dubey, and Ramesh Raskar. Reducing leakage in", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 358, + 480, + 370 + ], + "spans": [ + { + "bbox": [ + 115, + 358, + 480, + 370 + ], + "score": 1.0, + "content": "distributed deep learning for sensitive health data. arXiv preprint arXiv:1812.00564, 2019.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 375, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 506, + 390 + ], + "score": 1.0, + "content": "Qiang Yang, Yang Liu, Tianjian Chen, and Yongxin Tong. Federated machine learning: Concept", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 114, + 387, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 114, + 387, + 506, + 401 + ], + "score": 1.0, + "content": "and applications. ACM Transactions on Intelligent Systems and Technology (TIST), 10(2):1–19,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 398, + 142, + 411 + ], + "spans": [ + { + "bbox": [ + 116, + 398, + 142, + 411 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 416, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 104, + 416, + 505, + 431 + ], + "score": 1.0, + "content": "Bo Zhao, Konda Reddy Mopuri, and Hakan Bilen. idlg: Improved deep leakage from gradients.", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 428, + 279, + 441 + ], + "spans": [ + { + "bbox": [ + 115, + 428, + 279, + 441 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2001.02610, 2020.", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "score": 1.0, + "content": "Ligeng Zhu, Zhijian Liu, and Song Han. Deep leakage from gradients. In Advances in Neural", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 458, + 350, + 472 + ], + "spans": [ + { + "bbox": [ + 115, + 458, + 350, + 472 + ], + "score": 1.0, + "content": "Information Processing Systems, pp. 14774–14784, 2019.", + "type": "text" + } + ], + "index": 27, + "is_list_end_line": true + } + ], + "index": 13.5, + "bbox_fs": [ + 104, + 81, + 506, + 472 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 183, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 185, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 185, + 97 + ], + "score": 1.0, + "content": "A APPENDIX", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 106, + 185, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 185, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 185, + 120 + ], + "score": 1.0, + "content": "Appendix Outline", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 105, + 122, + 483, + 391 + ], + "lines": [ + { + "bbox": [ + 106, + 122, + 266, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 176, + 136 + ], + "score": 1.0, + "content": "A.1 Expressing", + "type": "text" + }, + { + "bbox": [ + 176, + 123, + 210, + 135 + ], + "score": 0.58, + "content": "\\operatorname { A U C } ( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 122, + 266, + 136 + ], + "score": 1.0, + "content": "as an integral", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 139, + 456, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 456, + 153 + ], + "score": 1.0, + "content": "A.2 Toy Class-balanced Example of Positive Example Prediction Lacking Confidence", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 156, + 209, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 209, + 168 + ], + "score": 1.0, + "content": "A.3 Proof of Theorem 1", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 173, + 284, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 173, + 284, + 186 + ], + "score": 1.0, + "content": "A.4 Proof and Interpretation of Theorem 2", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 190, + 266, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 266, + 203 + ], + "score": 1.0, + "content": "A.5 Marvell Algorithm Description", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 125, + 207, + 272, + 220 + ], + "spans": [ + { + "bbox": [ + 125, + 207, + 272, + 220 + ], + "score": 1.0, + "content": "A.5.1 Marvell Time Complexity", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 126, + 224, + 284, + 237 + ], + "spans": [ + { + "bbox": [ + 126, + 224, + 284, + 237 + ], + "score": 1.0, + "content": "A.5.2 Marvell Empirical Run Time", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 242, + 279, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 279, + 254 + ], + "score": 1.0, + "content": "A.6 Data Setup and Experimental Details", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 126, + 258, + 242, + 272 + ], + "spans": [ + { + "bbox": [ + 126, + 258, + 242, + 272 + ], + "score": 1.0, + "content": "A.6.1 Dataset Preprocessing", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 275, + 262, + 287 + ], + "spans": [ + { + "bbox": [ + 126, + 275, + 262, + 287 + ], + "score": 1.0, + "content": "A.6.2 Model Architecture Details", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 293, + 247, + 304 + ], + "spans": [ + { + "bbox": [ + 126, + 293, + 247, + 304 + ], + "score": 1.0, + "content": "A.6.3 Model Training Details", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 309, + 257, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 257, + 321 + ], + "score": 1.0, + "content": "A.7 Complete Experimental Results", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 126, + 326, + 314, + 338 + ], + "spans": [ + { + "bbox": [ + 126, + 326, + 314, + 338 + ], + "score": 1.0, + "content": "A.7.1 AUC Progression for Avazu and Criteo", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 125, + 342, + 300, + 356 + ], + "spans": [ + { + "bbox": [ + 125, + 342, + 300, + 356 + ], + "score": 1.0, + "content": "A.7.2 Complete Privacy-Utility Tradeoffs", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 126, + 360, + 483, + 372 + ], + "spans": [ + { + "bbox": [ + 126, + 360, + 483, + 372 + ], + "score": 1.0, + "content": "A.7.3 Impact of Activation Gradients’ Layer Location on the Corresponding Leak AUC", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 376, + 324, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 324, + 391 + ], + "score": 1.0, + "content": "A.8 Reasons why we don’t use local DP in our setup", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 9.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 183, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 185, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 185, + 97 + ], + "score": 1.0, + "content": "A APPENDIX", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 106, + 185, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 185, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 185, + 120 + ], + "score": 1.0, + "content": "Appendix Outline", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "list", + "bbox": [ + 105, + 122, + 483, + 391 + ], + "lines": [ + { + "bbox": [ + 106, + 122, + 266, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 176, + 136 + ], + "score": 1.0, + "content": "A.1 Expressing", + "type": "text" + }, + { + "bbox": [ + 176, + 123, + 210, + 135 + ], + "score": 0.58, + "content": "\\operatorname { A U C } ( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 122, + 266, + 136 + ], + "score": 1.0, + "content": "as an integral", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 139, + 456, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 456, + 153 + ], + "score": 1.0, + "content": "A.2 Toy Class-balanced Example of Positive Example Prediction Lacking Confidence", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 156, + 209, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 209, + 168 + ], + "score": 1.0, + "content": "A.3 Proof of Theorem 1", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 173, + 284, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 173, + 284, + 186 + ], + "score": 1.0, + "content": "A.4 Proof and Interpretation of Theorem 2", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 190, + 266, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 266, + 203 + ], + "score": 1.0, + "content": "A.5 Marvell Algorithm Description", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 125, + 207, + 272, + 220 + ], + "spans": [ + { + "bbox": [ + 125, + 207, + 272, + 220 + ], + "score": 1.0, + "content": "A.5.1 Marvell Time Complexity", + "type": "text" + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 126, + 224, + 284, + 237 + ], + "spans": [ + { + "bbox": [ + 126, + 224, + 284, + 237 + ], + "score": 1.0, + "content": "A.5.2 Marvell Empirical Run Time", + "type": "text" + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 242, + 279, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 279, + 254 + ], + "score": 1.0, + "content": "A.6 Data Setup and Experimental Details", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 126, + 258, + 242, + 272 + ], + "spans": [ + { + "bbox": [ + 126, + 258, + 242, + 272 + ], + "score": 1.0, + "content": "A.6.1 Dataset Preprocessing", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 126, + 275, + 262, + 287 + ], + "spans": [ + { + "bbox": [ + 126, + 275, + 262, + 287 + ], + "score": 1.0, + "content": "A.6.2 Model Architecture Details", + "type": "text" + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 126, + 293, + 247, + 304 + ], + "spans": [ + { + "bbox": [ + 126, + 293, + 247, + 304 + ], + "score": 1.0, + "content": "A.6.3 Model Training Details", + "type": "text" + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 309, + 257, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 257, + 321 + ], + "score": 1.0, + "content": "A.7 Complete Experimental Results", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 126, + 326, + 314, + 338 + ], + "spans": [ + { + "bbox": [ + 126, + 326, + 314, + 338 + ], + "score": 1.0, + "content": "A.7.1 AUC Progression for Avazu and Criteo", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 125, + 342, + 300, + 356 + ], + "spans": [ + { + "bbox": [ + 125, + 342, + 300, + 356 + ], + "score": 1.0, + "content": "A.7.2 Complete Privacy-Utility Tradeoffs", + "type": "text" + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 126, + 360, + 483, + 372 + ], + "spans": [ + { + "bbox": [ + 126, + 360, + 483, + 372 + ], + "score": 1.0, + "content": "A.7.3 Impact of Activation Gradients’ Layer Location on the Corresponding Leak AUC", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 376, + 324, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 324, + 391 + ], + "score": 1.0, + "content": "A.8 Reasons why we don’t use local DP in our setup", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 122, + 483, + 391 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 300, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 302, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 190, + 96 + ], + "score": 1.0, + "content": "A.1 EXPRESSING", + "type": "text" + }, + { + "bbox": [ + 190, + 82, + 225, + 95 + ], + "score": 0.61, + "content": "\\operatorname { A U C } ( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 81, + 302, + 96 + ], + "score": 1.0, + "content": "AS AN INTEGRAL", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 103, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 504, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 300, + 119 + ], + "score": 1.0, + "content": "Recall that the ROC curve of a scoring function", + "type": "text" + }, + { + "bbox": [ + 300, + 103, + 350, + 115 + ], + "score": 0.91, + "content": "r : \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 102, + 437, + 119 + ], + "score": 1.0, + "content": "is a parametric curve", + "type": "text" + }, + { + "bbox": [ + 438, + 104, + 504, + 117 + ], + "score": 0.91, + "content": "c _ { r } : \\mathbb { R } [ 0 , 1 ] ^ { 2 }", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 115, + 283, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 145, + 128 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 145, + 115, + 258, + 128 + ], + "score": 0.9, + "content": "c _ { r } ( t ) = ( \\mathrm { F P R } _ { r } ( t ) , \\mathrm { T P R } _ { r } ( t ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 115, + 283, + 128 + ], + "score": 1.0, + "content": ", with", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "interline_equation", + "bbox": [ + 158, + 127, + 451, + 172 + ], + "lines": [ + { + "bbox": [ + 158, + 127, + 451, + 172 + ], + "spans": [ + { + "bbox": [ + 158, + 127, + 451, + 172 + ], + "score": 0.75, + "content": "\\begin{array} { r l } & { \\mathrm { F P R } _ { r } : \\mathbb { R } [ 0 , 1 ] , \\mathrm { s u c h t h a t } \\mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } ( \\{ g \\in \\mathbb { R } ^ { d } : r ( g ) > t \\} ) } \\\\ & { \\mathrm { T P R } _ { r } : \\mathbb { R } [ 0 , 1 ] , \\mathrm { s u c h t h a t } \\mathrm { T P R } _ { r } ( t ) : = P ^ { ( 1 ) } ( \\{ g \\in \\mathbb { R } ^ { d } : r ( g ) > t \\} ) . } \\end{array}", + "type": "interline_equation", + "image_path": "398d48be374fbf75b272d5833a2b7009430aaac2d8f6846a8ac3c5b9839fa28c.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 158, + 127, + 451, + 142.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 158, + 142.0, + 451, + 157.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 158, + 157.0, + 451, + 172.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 112, + 171, + 450, + 183 + ], + "lines": [ + { + "bbox": [ + 110, + 170, + 449, + 184 + ], + "spans": [ + { + "bbox": [ + 110, + 170, + 166, + 184 + ], + "score": 1.0, + "content": "We notice that", + "type": "text" + }, + { + "bbox": [ + 167, + 171, + 189, + 182 + ], + "score": 0.89, + "content": "\\mathrm { F P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 170, + 208, + 184 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 208, + 171, + 231, + 182 + ], + "score": 0.88, + "content": "\\mathrm { T P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 170, + 423, + 184 + ], + "score": 1.0, + "content": "are both monotonically decreasing functions in", + "type": "text" + }, + { + "bbox": [ + 423, + 172, + 428, + 181 + ], + "score": 0.78, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 170, + 449, + 184 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 183, + 389, + 222 + ], + "lines": [ + { + "bbox": [ + 221, + 183, + 389, + 222 + ], + "spans": [ + { + "bbox": [ + 221, + 183, + 389, + 222 + ], + "score": 0.89, + "content": "\\begin{array} { r l r } & { \\underset { t - \\infty } { \\operatorname* { l i m } } \\mathrm { F P R } _ { r } ( t ) = 1 , \\ \\underset { t \\infty } { \\operatorname* { l i m } } \\mathrm { F P R } _ { r } ( t ) = 0 } & \\\\ & { \\underset { t - \\infty } { \\operatorname* { l i m } } \\mathrm { T P R } _ { r } ( t ) = 1 , \\ \\underset { t \\infty } { \\operatorname* { l i m } } \\mathrm { T P R } _ { r } ( t ) = 0 . } & \\end{array}", + "type": "interline_equation", + "image_path": "72fd6c727901124c0bcf7f4c384ead33167ef3f57ad105c734292e48ea21bf04.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 221, + 183, + 389, + 196.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 221, + 196.0, + 389, + 209.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 221, + 209.0, + 389, + 222.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 235, + 506, + 293 + ], + "lines": [ + { + "bbox": [ + 105, + 235, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 146, + 249 + ], + "score": 1.0, + "content": "However,", + "type": "text" + }, + { + "bbox": [ + 147, + 237, + 170, + 248 + ], + "score": 0.9, + "content": "\\mathrm { F P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 235, + 188, + 249 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 189, + 237, + 212, + 248 + ], + "score": 0.89, + "content": "\\mathrm { T P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 235, + 459, + 249 + ], + "score": 1.0, + "content": "does not need to be differentiable everywhere with respect to", + "type": "text" + }, + { + "bbox": [ + 459, + 237, + 464, + 246 + ], + "score": 0.75, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 235, + 506, + 249 + ], + "score": 1.0, + "content": ". Thus we", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 247, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 106, + 250, + 320, + 262 + ], + "score": 1.0, + "content": "cannot simply express the area under the curve using", + "type": "text" + }, + { + "bbox": [ + 321, + 248, + 424, + 262 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\int _ { \\infty } ^ { - \\infty } \\mathrm { T P R } _ { r } ( t ) \\mathrm { F P R } _ { r } ^ { \\prime } ( t ) d t } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 247, + 505, + 263 + ], + "score": 1.0, + "content": ". On the other hand,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 160, + 273 + ], + "score": 1.0, + "content": "because both", + "type": "text" + }, + { + "bbox": [ + 160, + 261, + 183, + 271 + ], + "score": 0.89, + "content": "\\mathrm { F P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 261, + 201, + 273 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 201, + 261, + 225, + 271 + ], + "score": 0.89, + "content": "\\mathrm { T P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "are functions of bounded variations, we can express the area under the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 176, + 284 + ], + "score": 1.0, + "content": "parametric curve", + "type": "text" + }, + { + "bbox": [ + 177, + 273, + 187, + 282 + ], + "score": 0.86, + "content": "c _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "through the Riemann-Stieltjes integral which involves integrating the function", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 281, + 271, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 130, + 293 + ], + "score": 0.88, + "content": "\\mathrm { T P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 281, + 243, + 295 + ], + "score": 1.0, + "content": "with respect to the function", + "type": "text" + }, + { + "bbox": [ + 243, + 282, + 266, + 293 + ], + "score": 0.88, + "content": "\\mathrm { F P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 281, + 271, + 295 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 294, + 385, + 322 + ], + "lines": [ + { + "bbox": [ + 225, + 294, + 385, + 322 + ], + "spans": [ + { + "bbox": [ + 225, + 294, + 385, + 322 + ], + "score": 0.94, + "content": "\\operatorname { A U C } ( r ) = \\int _ { t = \\infty } ^ { t = - \\infty } \\operatorname { T P R } _ { r } ( t ) d \\operatorname { F P R } _ { r } ( t )", + "type": "interline_equation", + "image_path": "a46f1458ad3c56830431028e4b0da5af05186ad48bb9e6e064a9f3810f074147.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 294, + 385, + 308.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 225, + 308.0, + 385, + 322.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 322, + 503, + 344 + ], + "lines": [ + { + "bbox": [ + 105, + 321, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 257, + 334 + ], + "score": 1.0, + "content": "Here the integration boundary is from", + "type": "text" + }, + { + "bbox": [ + 258, + 324, + 268, + 332 + ], + "score": 0.81, + "content": "\\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 321, + 279, + 334 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 280, + 324, + 298, + 332 + ], + "score": 0.88, + "content": "- \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 321, + 505, + 334 + ], + "score": 1.0, + "content": "in order to ensure the integral evaluates to a positive", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 332, + 134, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 134, + 346 + ], + "score": 1.0, + "content": "value.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 106, + 357, + 473, + 379 + ], + "lines": [ + { + "bbox": [ + 106, + 356, + 474, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 474, + 370 + ], + "score": 1.0, + "content": "A.2 TOY CLASS-BALANCED EXAMPLE OF POSITIVE EXAMPLE PREDICTION LACKING", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 132, + 369, + 191, + 380 + ], + "spans": [ + { + "bbox": [ + 132, + 369, + 191, + 380 + ], + "score": 1.0, + "content": "CONFIDENCE", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 108, + 388, + 505, + 422 + ], + "lines": [ + { + "bbox": [ + 106, + 387, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 505, + 402 + ], + "score": 1.0, + "content": "As mentioned in Observation 1 of norm-based attack in Section 3.3, for many practical applications,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "there is inherently more ambiguity for the positive class than the negative class. To make this more", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 411, + 289, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 289, + 423 + ], + "score": 1.0, + "content": "concrete, let’s consider a simple toy example.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 427, + 505, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 428, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 380, + 439 + ], + "score": 1.0, + "content": "Suppose the binary classification problem is over the real line. Here,", + "type": "text" + }, + { + "bbox": [ + 380, + 428, + 400, + 438 + ], + "score": 0.86, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 428, + 505, + 439 + ], + "score": 1.0, + "content": "of the data is positive and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 438, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 474, + 451 + ], + "score": 1.0, + "content": "lies uniformly in the interval [0, 1]. On the other hand, the remaining negative data (the other", + "type": "text" + }, + { + "bbox": [ + 474, + 439, + 493, + 449 + ], + "score": 0.87, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 438, + 506, + 451 + ], + "score": 1.0, + "content": "of", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 266, + 461 + ], + "score": 1.0, + "content": "the total data) is a mixture distribution:", + "type": "text" + }, + { + "bbox": [ + 266, + 450, + 286, + 460 + ], + "score": 0.86, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "of the negative data also lies uniformly in [0, 1], while", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 460, + 333, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 137, + 473 + ], + "score": 1.0, + "content": "the rest", + "type": "text" + }, + { + "bbox": [ + 138, + 460, + 158, + 470 + ], + "score": 0.85, + "content": "90 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 460, + 333, + 473 + ], + "score": 1.0, + "content": "of the negative data lies uniformly in [1, 2].", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 471, + 469, + 501 + ], + "lines": [ + { + "bbox": [ + 142, + 471, + 469, + 501 + ], + "spans": [ + { + "bbox": [ + 142, + 471, + 469, + 501 + ], + "score": 0.86, + "content": "\\begin{array} { r l r l } & { p ( x \\mid y = 1 ) = \\mathbb { 1 } ( x \\in [ 0 , 1 ] ) , } & & { \\qquad p ( y = 1 ) = 0 . 5 . } \\\\ & { p ( x \\mid y = 0 ) = 0 . 1 \\cdot \\mathbb { 1 } ( x \\in [ 0 , 1 ] ) + 0 . 9 \\cdot \\mathbb { 1 } ( x \\in [ 1 , 2 ] ) , } & & { \\qquad p ( y = 0 ) = 0 . 5 . } \\end{array}", + "type": "interline_equation", + "image_path": "e08d16bb7b179939dbcce420287fa0464f154ec001dbf10e295b29befaac8350.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 142, + 471, + 469, + 481.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 142, + 481.0, + 469, + 491.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 142, + 491.0, + 469, + 501.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 505, + 547 + ], + "lines": [ + { + "bbox": [ + 106, + 514, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 505, + 526 + ], + "score": 1.0, + "content": "This setup mirrors our online advertising example where for all the users interested in the product", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 525, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 160, + 538 + ], + "score": 1.0, + "content": "(with feature", + "type": "text" + }, + { + "bbox": [ + 160, + 525, + 200, + 537 + ], + "score": 0.91, + "content": "x \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 525, + 505, + 538 + ], + "score": 1.0, + "content": "), only a part of them would actually make the purchase after clicking on its", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 536, + 168, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 168, + 548 + ], + "score": 1.0, + "content": "advertisement.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 553, + 505, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 321, + 567 + ], + "score": 1.0, + "content": "In this case, the best possible probabilistic classifiers", + "type": "text" + }, + { + "bbox": [ + 321, + 554, + 339, + 564 + ], + "score": 0.88, + "content": "C ^ { \\mathrm { { o p t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 551, + 505, + 567 + ], + "score": 1.0, + "content": "that can be ever learned, by Bayes Rule,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 559, + 507, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 230, + 581 + ], + "score": 1.0, + "content": "would predict any example in", + "type": "text" + }, + { + "bbox": [ + 231, + 564, + 251, + 576 + ], + "score": 0.5, + "content": "[ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 559, + 428, + 581 + ], + "score": 1.0, + "content": "to be of the positive class with probability", + "type": "text" + }, + { + "bbox": [ + 428, + 563, + 439, + 578 + ], + "score": 0.88, + "content": "\\frac { 1 0 } { 1 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 559, + 507, + 581 + ], + "score": 1.0, + "content": ", while it would", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 575, + 404, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 404, + 589 + ], + "score": 1.0, + "content": "predict any example in [1, 2] to be of the negative class with probability 1:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 588, + 386, + 616 + ], + "lines": [ + { + "bbox": [ + 225, + 588, + 386, + 616 + ], + "spans": [ + { + "bbox": [ + 225, + 588, + 386, + 616 + ], + "score": 0.94, + "content": "C ^ { \\mathrm { o p t } } ( y = 1 \\mid x ) = { \\left\\{ \\begin{array} { l l } { { \\frac { 1 0 } { 1 1 } } } & { { \\mathrm { ~ i f ~ } } x \\in [ 0 , 1 ] } \\\\ { 0 } & { { \\mathrm { ~ i f ~ } } x \\in [ 1 , 2 ] } \\end{array} \\right. } .", + "type": "interline_equation", + "image_path": "6f7905bc0b20a6d065f5592b4422f0042bce3b5aed035e86664e9097b76898a7.jpg" + } + ] + } + ], + "index": 37.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 588, + 386, + 602.0 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 225, + 602.0, + 386, + 616.0 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 280, + 650 + ], + "score": 1.0, + "content": "Thus even for this best possible classifier", + "type": "text" + }, + { + "bbox": [ + 280, + 638, + 299, + 648 + ], + "score": 0.87, + "content": "C ^ { \\mathrm { { o p t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 637, + 506, + 650 + ], + "score": 1.0, + "content": ", every positive example would have a confidence", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 647, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 104, + 647, + 134, + 663 + ], + "score": 1.0, + "content": "gap of", + "type": "text" + }, + { + "bbox": [ + 135, + 648, + 145, + 662 + ], + "score": 0.9, + "content": "\\scriptstyle { \\frac { 1 } { 1 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 647, + 171, + 663 + ], + "score": 1.0, + "content": "while", + "type": "text" + }, + { + "bbox": [ + 171, + 649, + 191, + 659 + ], + "score": 0.89, + "content": "9 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 647, + 506, + 663 + ], + "score": 1.0, + "content": "of the negative examples (the ones in [1, 2]) would have a confidence gap of 0.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "Hence we see that in this toy example, our empirical observation of the lack of prediction confidence", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 671, + 280, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 280, + 683 + ], + "score": 1.0, + "content": "for positive examples would still hold true.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "Besides, it is important to notice that this lack of positive prediction confidence phenomenon hap-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 290, + 712 + ], + "score": 1.0, + "content": "pens even in this class-balanced toy example", + "type": "text" + }, + { + "bbox": [ + 290, + 699, + 411, + 711 + ], + "score": 0.89, + "content": "( p ( y = 1 ) = p ( y = 0 ) = 0 . 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 699, + 506, + 712 + ], + "score": 1.0, + "content": ". Thus, Observation 1", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "does not require the data distribution to be class-imbalanced to hold, which further demonstrates", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 268, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 268, + 733 + ], + "score": 1.0, + "content": "the generality of our norm-based attack.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 300, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 302, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 190, + 96 + ], + "score": 1.0, + "content": "A.1 EXPRESSING", + "type": "text" + }, + { + "bbox": [ + 190, + 82, + 225, + 95 + ], + "score": 0.61, + "content": "\\operatorname { A U C } ( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 81, + 302, + 96 + ], + "score": 1.0, + "content": "AS AN INTEGRAL", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 106, + 81, + 302, + 96 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 103, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 504, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 300, + 119 + ], + "score": 1.0, + "content": "Recall that the ROC curve of a scoring function", + "type": "text" + }, + { + "bbox": [ + 300, + 103, + 350, + 115 + ], + "score": 0.91, + "content": "r : \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 102, + 437, + 119 + ], + "score": 1.0, + "content": "is a parametric curve", + "type": "text" + }, + { + "bbox": [ + 438, + 104, + 504, + 117 + ], + "score": 0.91, + "content": "c _ { r } : \\mathbb { R } [ 0 , 1 ] ^ { 2 }", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 115, + 283, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 145, + 128 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 145, + 115, + 258, + 128 + ], + "score": 0.9, + "content": "c _ { r } ( t ) = ( \\mathrm { F P R } _ { r } ( t ) , \\mathrm { T P R } _ { r } ( t ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 115, + 283, + 128 + ], + "score": 1.0, + "content": ", with", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 102, + 504, + 128 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 158, + 127, + 451, + 172 + ], + "lines": [ + { + "bbox": [ + 158, + 127, + 451, + 172 + ], + "spans": [ + { + "bbox": [ + 158, + 127, + 451, + 172 + ], + "score": 0.75, + "content": "\\begin{array} { r l } & { \\mathrm { F P R } _ { r } : \\mathbb { R } [ 0 , 1 ] , \\mathrm { s u c h t h a t } \\mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } ( \\{ g \\in \\mathbb { R } ^ { d } : r ( g ) > t \\} ) } \\\\ & { \\mathrm { T P R } _ { r } : \\mathbb { R } [ 0 , 1 ] , \\mathrm { s u c h t h a t } \\mathrm { T P R } _ { r } ( t ) : = P ^ { ( 1 ) } ( \\{ g \\in \\mathbb { R } ^ { d } : r ( g ) > t \\} ) . } \\end{array}", + "type": "interline_equation", + "image_path": "398d48be374fbf75b272d5833a2b7009430aaac2d8f6846a8ac3c5b9839fa28c.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 158, + 127, + 451, + 142.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 158, + 142.0, + 451, + 157.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 158, + 157.0, + 451, + 172.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 112, + 171, + 450, + 183 + ], + "lines": [ + { + "bbox": [ + 110, + 170, + 449, + 184 + ], + "spans": [ + { + "bbox": [ + 110, + 170, + 166, + 184 + ], + "score": 1.0, + "content": "We notice that", + "type": "text" + }, + { + "bbox": [ + 167, + 171, + 189, + 182 + ], + "score": 0.89, + "content": "\\mathrm { F P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 170, + 208, + 184 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 208, + 171, + 231, + 182 + ], + "score": 0.88, + "content": "\\mathrm { T P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 170, + 423, + 184 + ], + "score": 1.0, + "content": "are both monotonically decreasing functions in", + "type": "text" + }, + { + "bbox": [ + 423, + 172, + 428, + 181 + ], + "score": 0.78, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 170, + 449, + 184 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 110, + 170, + 449, + 184 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 183, + 389, + 222 + ], + "lines": [ + { + "bbox": [ + 221, + 183, + 389, + 222 + ], + "spans": [ + { + "bbox": [ + 221, + 183, + 389, + 222 + ], + "score": 0.89, + "content": "\\begin{array} { r l r } & { \\underset { t - \\infty } { \\operatorname* { l i m } } \\mathrm { F P R } _ { r } ( t ) = 1 , \\ \\underset { t \\infty } { \\operatorname* { l i m } } \\mathrm { F P R } _ { r } ( t ) = 0 } & \\\\ & { \\underset { t - \\infty } { \\operatorname* { l i m } } \\mathrm { T P R } _ { r } ( t ) = 1 , \\ \\underset { t \\infty } { \\operatorname* { l i m } } \\mathrm { T P R } _ { r } ( t ) = 0 . } & \\end{array}", + "type": "interline_equation", + "image_path": "72fd6c727901124c0bcf7f4c384ead33167ef3f57ad105c734292e48ea21bf04.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 221, + 183, + 389, + 196.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 221, + 196.0, + 389, + 209.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 221, + 209.0, + 389, + 222.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 235, + 506, + 293 + ], + "lines": [ + { + "bbox": [ + 105, + 235, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 146, + 249 + ], + "score": 1.0, + "content": "However,", + "type": "text" + }, + { + "bbox": [ + 147, + 237, + 170, + 248 + ], + "score": 0.9, + "content": "\\mathrm { F P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 235, + 188, + 249 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 189, + 237, + 212, + 248 + ], + "score": 0.89, + "content": "\\mathrm { T P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 235, + 459, + 249 + ], + "score": 1.0, + "content": "does not need to be differentiable everywhere with respect to", + "type": "text" + }, + { + "bbox": [ + 459, + 237, + 464, + 246 + ], + "score": 0.75, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 235, + 506, + 249 + ], + "score": 1.0, + "content": ". Thus we", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 247, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 106, + 250, + 320, + 262 + ], + "score": 1.0, + "content": "cannot simply express the area under the curve using", + "type": "text" + }, + { + "bbox": [ + 321, + 248, + 424, + 262 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\int _ { \\infty } ^ { - \\infty } \\mathrm { T P R } _ { r } ( t ) \\mathrm { F P R } _ { r } ^ { \\prime } ( t ) d t } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 247, + 505, + 263 + ], + "score": 1.0, + "content": ". On the other hand,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 160, + 273 + ], + "score": 1.0, + "content": "because both", + "type": "text" + }, + { + "bbox": [ + 160, + 261, + 183, + 271 + ], + "score": 0.89, + "content": "\\mathrm { F P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 261, + 201, + 273 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 201, + 261, + 225, + 271 + ], + "score": 0.89, + "content": "\\mathrm { T P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "are functions of bounded variations, we can express the area under the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 176, + 284 + ], + "score": 1.0, + "content": "parametric curve", + "type": "text" + }, + { + "bbox": [ + 177, + 273, + 187, + 282 + ], + "score": 0.86, + "content": "c _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "through the Riemann-Stieltjes integral which involves integrating the function", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 281, + 271, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 130, + 293 + ], + "score": 0.88, + "content": "\\mathrm { T P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 281, + 243, + 295 + ], + "score": 1.0, + "content": "with respect to the function", + "type": "text" + }, + { + "bbox": [ + 243, + 282, + 266, + 293 + ], + "score": 0.88, + "content": "\\mathrm { F P R } _ { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 281, + 271, + 295 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 235, + 506, + 295 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 294, + 385, + 322 + ], + "lines": [ + { + "bbox": [ + 225, + 294, + 385, + 322 + ], + "spans": [ + { + "bbox": [ + 225, + 294, + 385, + 322 + ], + "score": 0.94, + "content": "\\operatorname { A U C } ( r ) = \\int _ { t = \\infty } ^ { t = - \\infty } \\operatorname { T P R } _ { r } ( t ) d \\operatorname { F P R } _ { r } ( t )", + "type": "interline_equation", + "image_path": "a46f1458ad3c56830431028e4b0da5af05186ad48bb9e6e064a9f3810f074147.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 294, + 385, + 308.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 225, + 308.0, + 385, + 322.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 322, + 503, + 344 + ], + "lines": [ + { + "bbox": [ + 105, + 321, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 257, + 334 + ], + "score": 1.0, + "content": "Here the integration boundary is from", + "type": "text" + }, + { + "bbox": [ + 258, + 324, + 268, + 332 + ], + "score": 0.81, + "content": "\\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 321, + 279, + 334 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 280, + 324, + 298, + 332 + ], + "score": 0.88, + "content": "- \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 321, + 505, + 334 + ], + "score": 1.0, + "content": "in order to ensure the integral evaluates to a positive", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 332, + 134, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 134, + 346 + ], + "score": 1.0, + "content": "value.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 321, + 505, + 346 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 357, + 473, + 379 + ], + "lines": [ + { + "bbox": [ + 106, + 356, + 474, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 474, + 370 + ], + "score": 1.0, + "content": "A.2 TOY CLASS-BALANCED EXAMPLE OF POSITIVE EXAMPLE PREDICTION LACKING", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 132, + 369, + 191, + 380 + ], + "spans": [ + { + "bbox": [ + 132, + 369, + 191, + 380 + ], + "score": 1.0, + "content": "CONFIDENCE", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 108, + 388, + 505, + 422 + ], + "lines": [ + { + "bbox": [ + 106, + 387, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 505, + 402 + ], + "score": 1.0, + "content": "As mentioned in Observation 1 of norm-based attack in Section 3.3, for many practical applications,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "there is inherently more ambiguity for the positive class than the negative class. To make this more", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 411, + 289, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 289, + 423 + ], + "score": 1.0, + "content": "concrete, let’s consider a simple toy example.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 387, + 505, + 423 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 427, + 505, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 428, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 380, + 439 + ], + "score": 1.0, + "content": "Suppose the binary classification problem is over the real line. Here,", + "type": "text" + }, + { + "bbox": [ + 380, + 428, + 400, + 438 + ], + "score": 0.86, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 428, + 505, + 439 + ], + "score": 1.0, + "content": "of the data is positive and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 438, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 474, + 451 + ], + "score": 1.0, + "content": "lies uniformly in the interval [0, 1]. On the other hand, the remaining negative data (the other", + "type": "text" + }, + { + "bbox": [ + 474, + 439, + 493, + 449 + ], + "score": 0.87, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 438, + 506, + 451 + ], + "score": 1.0, + "content": "of", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 266, + 461 + ], + "score": 1.0, + "content": "the total data) is a mixture distribution:", + "type": "text" + }, + { + "bbox": [ + 266, + 450, + 286, + 460 + ], + "score": 0.86, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "of the negative data also lies uniformly in [0, 1], while", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 460, + 333, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 137, + 473 + ], + "score": 1.0, + "content": "the rest", + "type": "text" + }, + { + "bbox": [ + 138, + 460, + 158, + 470 + ], + "score": 0.85, + "content": "90 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 460, + 333, + 473 + ], + "score": 1.0, + "content": "of the negative data lies uniformly in [1, 2].", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 428, + 506, + 473 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 471, + 469, + 501 + ], + "lines": [ + { + "bbox": [ + 142, + 471, + 469, + 501 + ], + "spans": [ + { + "bbox": [ + 142, + 471, + 469, + 501 + ], + "score": 0.86, + "content": "\\begin{array} { r l r l } & { p ( x \\mid y = 1 ) = \\mathbb { 1 } ( x \\in [ 0 , 1 ] ) , } & & { \\qquad p ( y = 1 ) = 0 . 5 . } \\\\ & { p ( x \\mid y = 0 ) = 0 . 1 \\cdot \\mathbb { 1 } ( x \\in [ 0 , 1 ] ) + 0 . 9 \\cdot \\mathbb { 1 } ( x \\in [ 1 , 2 ] ) , } & & { \\qquad p ( y = 0 ) = 0 . 5 . } \\end{array}", + "type": "interline_equation", + "image_path": "e08d16bb7b179939dbcce420287fa0464f154ec001dbf10e295b29befaac8350.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 142, + 471, + 469, + 481.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 142, + 481.0, + 469, + 491.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 142, + 491.0, + 469, + 501.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 505, + 547 + ], + "lines": [ + { + "bbox": [ + 106, + 514, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 505, + 526 + ], + "score": 1.0, + "content": "This setup mirrors our online advertising example where for all the users interested in the product", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 525, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 160, + 538 + ], + "score": 1.0, + "content": "(with feature", + "type": "text" + }, + { + "bbox": [ + 160, + 525, + 200, + 537 + ], + "score": 0.91, + "content": "x \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 525, + 505, + 538 + ], + "score": 1.0, + "content": "), only a part of them would actually make the purchase after clicking on its", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 536, + 168, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 168, + 548 + ], + "score": 1.0, + "content": "advertisement.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 514, + 505, + 548 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 553, + 505, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 321, + 567 + ], + "score": 1.0, + "content": "In this case, the best possible probabilistic classifiers", + "type": "text" + }, + { + "bbox": [ + 321, + 554, + 339, + 564 + ], + "score": 0.88, + "content": "C ^ { \\mathrm { { o p t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 551, + 505, + 567 + ], + "score": 1.0, + "content": "that can be ever learned, by Bayes Rule,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 559, + 507, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 230, + 581 + ], + "score": 1.0, + "content": "would predict any example in", + "type": "text" + }, + { + "bbox": [ + 231, + 564, + 251, + 576 + ], + "score": 0.5, + "content": "[ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 559, + 428, + 581 + ], + "score": 1.0, + "content": "to be of the positive class with probability", + "type": "text" + }, + { + "bbox": [ + 428, + 563, + 439, + 578 + ], + "score": 0.88, + "content": "\\frac { 1 0 } { 1 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 559, + 507, + 581 + ], + "score": 1.0, + "content": ", while it would", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 575, + 404, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 404, + 589 + ], + "score": 1.0, + "content": "predict any example in [1, 2] to be of the negative class with probability 1:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 551, + 507, + 589 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 588, + 386, + 616 + ], + "lines": [ + { + "bbox": [ + 225, + 588, + 386, + 616 + ], + "spans": [ + { + "bbox": [ + 225, + 588, + 386, + 616 + ], + "score": 0.94, + "content": "C ^ { \\mathrm { o p t } } ( y = 1 \\mid x ) = { \\left\\{ \\begin{array} { l l } { { \\frac { 1 0 } { 1 1 } } } & { { \\mathrm { ~ i f ~ } } x \\in [ 0 , 1 ] } \\\\ { 0 } & { { \\mathrm { ~ i f ~ } } x \\in [ 1 , 2 ] } \\end{array} \\right. } .", + "type": "interline_equation", + "image_path": "6f7905bc0b20a6d065f5592b4422f0042bce3b5aed035e86664e9097b76898a7.jpg" + } + ] + } + ], + "index": 37.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 588, + 386, + 602.0 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 225, + 602.0, + 386, + 616.0 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 280, + 650 + ], + "score": 1.0, + "content": "Thus even for this best possible classifier", + "type": "text" + }, + { + "bbox": [ + 280, + 638, + 299, + 648 + ], + "score": 0.87, + "content": "C ^ { \\mathrm { { o p t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 637, + 506, + 650 + ], + "score": 1.0, + "content": ", every positive example would have a confidence", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 647, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 104, + 647, + 134, + 663 + ], + "score": 1.0, + "content": "gap of", + "type": "text" + }, + { + "bbox": [ + 135, + 648, + 145, + 662 + ], + "score": 0.9, + "content": "\\scriptstyle { \\frac { 1 } { 1 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 647, + 171, + 663 + ], + "score": 1.0, + "content": "while", + "type": "text" + }, + { + "bbox": [ + 171, + 649, + 191, + 659 + ], + "score": 0.89, + "content": "9 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 647, + 506, + 663 + ], + "score": 1.0, + "content": "of the negative examples (the ones in [1, 2]) would have a confidence gap of 0.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "Hence we see that in this toy example, our empirical observation of the lack of prediction confidence", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 671, + 280, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 280, + 683 + ], + "score": 1.0, + "content": "for positive examples would still hold true.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5, + "bbox_fs": [ + 104, + 637, + 506, + 683 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "Besides, it is important to notice that this lack of positive prediction confidence phenomenon hap-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 290, + 712 + ], + "score": 1.0, + "content": "pens even in this class-balanced toy example", + "type": "text" + }, + { + "bbox": [ + 290, + 699, + 411, + 711 + ], + "score": 0.89, + "content": "( p ( y = 1 ) = p ( y = 0 ) = 0 . 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 699, + 506, + 712 + ], + "score": 1.0, + "content": ". Thus, Observation 1", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "does not require the data distribution to be class-imbalanced to hold, which further demonstrates", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 268, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 268, + 733 + ], + "score": 1.0, + "content": "the generality of our norm-based attack.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 687, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 230, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 231, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 231, + 94 + ], + "score": 1.0, + "content": "A.3 PROOF OF THEOREM 1", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 102, + 505, + 126 + ], + "lines": [ + { + "bbox": [ + 109, + 101, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 109, + 101, + 189, + 117 + ], + "score": 1.0, + "content": "(Theorem 1). For", + "type": "text" + }, + { + "bbox": [ + 190, + 103, + 241, + 114 + ], + "score": 0.9, + "content": "0 \\leq \\epsilon < 4", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 101, + 414, + 117 + ], + "score": 1.0, + "content": "and any perturbed gradient distributions", + "type": "text" + }, + { + "bbox": [ + 414, + 104, + 430, + 113 + ], + "score": 0.87, + "content": "\\widetilde { P } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 101, + 451, + 117 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 452, + 104, + 468, + 113 + ], + "score": 0.87, + "content": "\\widetilde { P } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 101, + 506, + 117 + ], + "score": 1.0, + "content": "that are", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 114, + 302, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 114, + 302, + 127 + ], + "score": 1.0, + "content": "absolutely continuous with respect to each other,", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 136, + 477, + 162 + ], + "lines": [ + { + "bbox": [ + 133, + 136, + 477, + 162 + ], + "spans": [ + { + "bbox": [ + 133, + 136, + 477, + 162 + ], + "score": 0.88, + "content": "\\mathbf { K L } \\big ( \\widetilde { P } ^ { ( 1 ) } \\mid \\big | \\widetilde { P } ^ { ( 0 ) } \\big ) + \\mathbf { K L } \\big ( \\widetilde { P } ^ { ( 0 ) } \\mid \\big | \\widetilde { P } ^ { ( 1 ) } \\big ) \\leq \\epsilon \\quad i m p l i e s \\quad \\operatorname* { m a x } _ { r } \\mathbf { A } \\mathbf { U C } ( r ) \\leq \\frac { 1 } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } - \\frac { \\epsilon } { 8 } .", + "type": "interline_equation", + "image_path": "b7809dfa2abfe8e2446262709f9a2dd9d5a0428e5e9bb7a530c2c32bc0ef83a1.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 133, + 136, + 477, + 162 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 172, + 505, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "score": 1.0, + "content": "Proof of Theorem 1. Combining Pinsker’s inequality with Jensen’s inequality, we can obtain an up-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 183, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 506, + 196 + ], + "score": 1.0, + "content": "per bound of total variation distance by the symmetrized KL divergence (sumKL) for a pair of", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 194, + 417, + 208 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 159, + 208 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 159, + 194, + 186, + 207 + ], + "score": 0.93, + "content": "( P , Q )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 194, + 417, + 208 + ], + "score": 1.0, + "content": "that are absolutely continuous with respect to each other:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 120, + 210, + 479, + 235 + ], + "lines": [ + { + "bbox": [ + 120, + 210, + 479, + 235 + ], + "spans": [ + { + "bbox": [ + 120, + 210, + 479, + 235 + ], + "score": 0.93, + "content": "\\mathbf { T } \\mathbf { V } ( P , Q ) \\leq { \\frac { 1 } { 2 } } ( { \\sqrt { \\mathbf { K L } ( P \\parallel Q ) / 2 } } + { \\sqrt { \\mathbf { K L } ( Q \\parallel P ) / 2 } } ) \\leq { \\frac { 1 } { 2 } } { \\sqrt { \\mathbf { K L } ( P \\parallel Q ) + \\mathbf { K L } ( Q \\parallel P ) } } .", + "type": "interline_equation", + "image_path": "80c5e26ed530cd2e26696516bb82f8950db2a9538afd1400fb90f3646707e767.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 120, + 210, + 479, + 235 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 240, + 505, + 300 + ], + "lines": [ + { + "bbox": [ + 104, + 239, + 507, + 258 + ], + "spans": [ + { + "bbox": [ + 104, + 239, + 258, + 258 + ], + "score": 1.0, + "content": "By our assumption, this implies that", + "type": "text" + }, + { + "bbox": [ + 258, + 240, + 350, + 255 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\mathrm { T V } ( \\widetilde { P } ^ { ( 1 ) } , \\widetilde { P } ^ { ( 0 ) } ) \\le \\frac { \\sqrt { \\epsilon } } { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 239, + 507, + 258 + ], + "score": 1.0, + "content": ". By the equivalent definition of total", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 254, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 104, + 254, + 180, + 272 + ], + "score": 1.0, + "content": "variation distance", + "type": "text" + }, + { + "bbox": [ + 180, + 255, + 390, + 270 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathrm { T V } ( \\widetilde P ^ { ( 1 ) } , \\widetilde P ^ { ( 0 ) } ) ) = \\operatorname* { m a x } _ { A \\subset \\mathbb { R } ^ { d } } [ \\widetilde P ^ { ( 1 ) } ( A ) - \\widetilde P ^ { ( 0 ) } ( A ) ] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 254, + 483, + 272 + ], + "score": 1.0, + "content": ", we know that for any", + "type": "text" + }, + { + "bbox": [ + 484, + 257, + 505, + 269 + ], + "score": 0.84, + "content": "A \\subset", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 267, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 119, + 282 + ], + "score": 0.79, + "content": "\\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 267, + 133, + 286 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 133, + 269, + 244, + 285 + ], + "score": 0.86, + "content": "\\begin{array} { r } { \\widetilde { P } ^ { ( 1 ) } ( A ) - \\widetilde { P } ^ { ( 0 ) } ( A ) \\le \\frac { \\sqrt { \\epsilon } } { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 267, + 353, + 286 + ], + "score": 1.0, + "content": ". For any scoring function", + "type": "text" + }, + { + "bbox": [ + 353, + 274, + 360, + 281 + ], + "score": 0.72, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 267, + 460, + 286 + ], + "score": 1.0, + "content": "and any threshold value", + "type": "text" + }, + { + "bbox": [ + 461, + 272, + 466, + 282 + ], + "score": 0.71, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 267, + 483, + 286 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 483, + 271, + 505, + 283 + ], + "score": 0.83, + "content": "A =", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 282, + 428, + 302 + ], + "spans": [ + { + "bbox": [ + 107, + 286, + 165, + 299 + ], + "score": 0.92, + "content": "\\{ g : r ( g ) > t \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 282, + 223, + 302 + ], + "score": 1.0, + "content": ", then we have", + "type": "text" + }, + { + "bbox": [ + 224, + 284, + 424, + 300 + ], + "score": 0.85, + "content": "\\begin{array} { r } { \\bar { \\mathrm { T P R } } _ { r } ( t ) - \\mathrm { F P R } _ { r } ( t ) = \\widetilde { P } ^ { ( 1 ) } ( A ) - \\widetilde { P } ^ { ( 0 ) } ( A ) \\leq \\frac { \\sqrt { \\epsilon } } { 2 } . } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 282, + 428, + 302 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 103, + 303, + 467, + 316 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 467, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 281, + 317 + ], + "score": 1.0, + "content": "Therefore, the AUC of the scoring function", + "type": "text" + }, + { + "bbox": [ + 281, + 306, + 288, + 313 + ], + "score": 0.77, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 302, + 467, + 317 + ], + "score": 1.0, + "content": "can be upper bounded in the following way:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 189, + 320, + 421, + 378 + ], + "lines": [ + { + "bbox": [ + 189, + 320, + 421, + 378 + ], + "spans": [ + { + "bbox": [ + 189, + 320, + 421, + 378 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle \\mathrm { A U C } ( { \\boldsymbol { r } } ) = \\int _ { \\infty } ^ { - \\infty } { \\mathrm { T P R } _ { r } ( t ) ~ d \\mathrm { F P R } _ { r } ( t ) } } \\\\ { \\displaystyle \\leq \\int _ { \\infty } ^ { - \\infty } \\operatorname* { m i n } \\bigg ( \\mathrm { F P R } _ { r } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } , ~ 1 \\bigg ) ~ d \\mathrm { F P R } _ { r } ( t ) , } \\end{array}", + "type": "interline_equation", + "image_path": "ba8c8d0ee2d8f4faa55ba028d41a5b23ac0488964a59711dbccd1403cdd0075e.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 189, + 320, + 421, + 339.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 189, + 339.3333333333333, + 421, + 358.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 189, + 358.66666666666663, + 421, + 377.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 381, + 388, + 394 + ], + "lines": [ + { + "bbox": [ + 106, + 381, + 389, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 279, + 395 + ], + "score": 1.0, + "content": "where in (6) we use the additional fact that", + "type": "text" + }, + { + "bbox": [ + 280, + 382, + 333, + 394 + ], + "score": 0.94, + "content": "\\mathrm { T P R } _ { r } ( t ) \\leq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 381, + 361, + 395 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 361, + 383, + 385, + 393 + ], + "score": 0.9, + "content": "t \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 381, + 389, + 395 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 399, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 135, + 417 + ], + "score": 1.0, + "content": "When", + "type": "text" + }, + { + "bbox": [ + 135, + 402, + 182, + 414 + ], + "score": 0.91, + "content": "\\epsilon \\in \\ [ 0 , 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 398, + 229, + 417 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 230, + 399, + 306, + 416 + ], + "score": 0.9, + "content": "1 ~ - ~ \\frac { \\sqrt { \\epsilon } } { 2 } ~ \\in ~ ( 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 398, + 333, + 417 + ], + "score": 1.0, + "content": ". As", + "type": "text" + }, + { + "bbox": [ + 334, + 402, + 369, + 414 + ], + "score": 0.88, + "content": "\\mathrm { F P R } _ { r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 398, + 506, + 417 + ], + "score": 1.0, + "content": "is a monotonically nonincreas-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 102, + 412, + 508, + 437 + ], + "spans": [ + { + "bbox": [ + 102, + 412, + 172, + 437 + ], + "score": 1.0, + "content": "ing function in", + "type": "text" + }, + { + "bbox": [ + 172, + 419, + 177, + 428 + ], + "score": 0.73, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 412, + 239, + 437 + ], + "score": 1.0, + "content": "with range in", + "type": "text" + }, + { + "bbox": [ + 240, + 417, + 260, + 430 + ], + "score": 0.68, + "content": "[ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 412, + 298, + 437 + ], + "score": 1.0, + "content": ", the set", + "type": "text" + }, + { + "bbox": [ + 298, + 414, + 425, + 434 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\left\\{ t : \\mathrm { F P R } _ { r } ( t ) \\leq 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right\\} \\ \\neq \\ \\phi } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 412, + 508, + 437 + ], + "score": 1.0, + "content": "is not empty. Let", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 433, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 244, + 453 + ], + "score": 0.91, + "content": "\\begin{array} { r } { k : = \\operatorname* { i n f } \\left\\{ t : \\mathrm { F P R } _ { r } ( t ) \\leq 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right\\} } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 433, + 286, + 452 + ], + "score": 1.0, + "content": ". Again by", + "type": "text" + }, + { + "bbox": [ + 286, + 437, + 321, + 449 + ], + "score": 0.9, + "content": "\\mathrm { F P R } _ { r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 433, + 506, + 452 + ], + "score": 1.0, + "content": "being a monotonically nonincreasing function", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 450, + 367, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 117, + 464 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 117, + 453, + 122, + 461 + ], + "score": 0.75, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 450, + 367, + 464 + ], + "score": 1.0, + "content": ", we can break the integration in Equation (6) into two terms:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 465, + 481, + 636 + ], + "lines": [ + { + "bbox": [ + 114, + 465, + 481, + 636 + ], + "spans": [ + { + "bbox": [ + 114, + 465, + 481, + 636 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\quad \\mathrm { A U C } ( r ) } \\\\ & { = \\displaystyle \\int _ { - \\infty } ^ { k } \\operatorname* { m i n } \\left( \\mathrm { F P R } _ { \\mathrm { v } } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } , \\mathbf { \\Phi } _ { 1 } \\right) d \\mathrm { F P R } _ { v } ( t ) + \\displaystyle \\int _ { k } ^ { - \\infty } \\operatorname* { m i n } \\left( \\mathrm { F P R } _ { v } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } , \\mathbf { \\Phi } _ { 1 } \\right) d \\mathrm { F P R } _ { v } ( t ) } \\\\ & { \\le \\displaystyle \\int _ { \\infty } ^ { k } \\left( \\mathrm { F P R } _ { v } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } \\right) d \\mathrm { F P R } _ { v } ( t ) + \\displaystyle \\int _ { k } ^ { - \\infty } 1 d \\mathrm { F P R } _ { v } ( t ) } \\\\ & { = \\left[ \\frac { \\left[ \\mathrm { F P R } _ { r } ( v ) \\right] ^ { 2 } } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } \\mathrm { F P R } _ { v } ( v ) \\right] \\displaystyle _ { t = \\infty } ^ { t - k } + \\mathrm { F P R } _ { v } ( t ) \\displaystyle _ { t = k } ^ { t - \\infty } } \\\\ & { \\le \\frac { ( 1 - \\frac { v \\epsilon } { 2 } ) ^ { 2 } } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } \\left( 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right) + \\left[ 1 - \\left( 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right) \\right] } \\\\ & { = \\frac { 1 } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } - \\frac { \\epsilon } { 8 } } \\end{array}", + "type": "interline_equation", + "image_path": "c37fca5f32294ed20c297a9758cfee8014b0b38c95a97a532212844660b914d2.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 114, + 465, + 481, + 522.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 114, + 522.0, + 481, + 579.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 114, + 579.0, + 481, + 636.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 639, + 505, + 663 + ], + "lines": [ + { + "bbox": [ + 105, + 638, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 320, + 652 + ], + "score": 1.0, + "content": "Since this inequality is true for any scoring function", + "type": "text" + }, + { + "bbox": [ + 320, + 642, + 325, + 649 + ], + "score": 0.74, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 638, + 506, + 652 + ], + "score": 1.0, + "content": ", it is true for the maximum value. Thus the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 650, + 504, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 182, + 664 + ], + "score": 1.0, + "content": "proof is complete.", + "type": "text" + }, + { + "bbox": [ + 496, + 653, + 504, + 660 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 230, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 231, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 231, + 94 + ], + "score": 1.0, + "content": "A.3 PROOF OF THEOREM 1", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 102, + 505, + 126 + ], + "lines": [ + { + "bbox": [ + 109, + 101, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 109, + 101, + 189, + 117 + ], + "score": 1.0, + "content": "(Theorem 1). For", + "type": "text" + }, + { + "bbox": [ + 190, + 103, + 241, + 114 + ], + "score": 0.9, + "content": "0 \\leq \\epsilon < 4", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 101, + 414, + 117 + ], + "score": 1.0, + "content": "and any perturbed gradient distributions", + "type": "text" + }, + { + "bbox": [ + 414, + 104, + 430, + 113 + ], + "score": 0.87, + "content": "\\widetilde { P } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 101, + 451, + 117 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 452, + 104, + 468, + 113 + ], + "score": 0.87, + "content": "\\widetilde { P } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 101, + 506, + 117 + ], + "score": 1.0, + "content": "that are", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 114, + 302, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 114, + 302, + 127 + ], + "score": 1.0, + "content": "absolutely continuous with respect to each other,", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 106, + 101, + 506, + 127 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 136, + 477, + 162 + ], + "lines": [ + { + "bbox": [ + 133, + 136, + 477, + 162 + ], + "spans": [ + { + "bbox": [ + 133, + 136, + 477, + 162 + ], + "score": 0.88, + "content": "\\mathbf { K L } \\big ( \\widetilde { P } ^ { ( 1 ) } \\mid \\big | \\widetilde { P } ^ { ( 0 ) } \\big ) + \\mathbf { K L } \\big ( \\widetilde { P } ^ { ( 0 ) } \\mid \\big | \\widetilde { P } ^ { ( 1 ) } \\big ) \\leq \\epsilon \\quad i m p l i e s \\quad \\operatorname* { m a x } _ { r } \\mathbf { A } \\mathbf { U C } ( r ) \\leq \\frac { 1 } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } - \\frac { \\epsilon } { 8 } .", + "type": "interline_equation", + "image_path": "b7809dfa2abfe8e2446262709f9a2dd9d5a0428e5e9bb7a530c2c32bc0ef83a1.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 133, + 136, + 477, + 162 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 172, + 505, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 505, + 186 + ], + "score": 1.0, + "content": "Proof of Theorem 1. Combining Pinsker’s inequality with Jensen’s inequality, we can obtain an up-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 183, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 506, + 196 + ], + "score": 1.0, + "content": "per bound of total variation distance by the symmetrized KL divergence (sumKL) for a pair of", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 194, + 417, + 208 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 159, + 208 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 159, + 194, + 186, + 207 + ], + "score": 0.93, + "content": "( P , Q )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 194, + 417, + 208 + ], + "score": 1.0, + "content": "that are absolutely continuous with respect to each other:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 171, + 506, + 208 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 120, + 210, + 479, + 235 + ], + "lines": [ + { + "bbox": [ + 120, + 210, + 479, + 235 + ], + "spans": [ + { + "bbox": [ + 120, + 210, + 479, + 235 + ], + "score": 0.93, + "content": "\\mathbf { T } \\mathbf { V } ( P , Q ) \\leq { \\frac { 1 } { 2 } } ( { \\sqrt { \\mathbf { K L } ( P \\parallel Q ) / 2 } } + { \\sqrt { \\mathbf { K L } ( Q \\parallel P ) / 2 } } ) \\leq { \\frac { 1 } { 2 } } { \\sqrt { \\mathbf { K L } ( P \\parallel Q ) + \\mathbf { K L } ( Q \\parallel P ) } } .", + "type": "interline_equation", + "image_path": "80c5e26ed530cd2e26696516bb82f8950db2a9538afd1400fb90f3646707e767.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 120, + 210, + 479, + 235 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 240, + 505, + 300 + ], + "lines": [ + { + "bbox": [ + 104, + 239, + 507, + 258 + ], + "spans": [ + { + "bbox": [ + 104, + 239, + 258, + 258 + ], + "score": 1.0, + "content": "By our assumption, this implies that", + "type": "text" + }, + { + "bbox": [ + 258, + 240, + 350, + 255 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\mathrm { T V } ( \\widetilde { P } ^ { ( 1 ) } , \\widetilde { P } ^ { ( 0 ) } ) \\le \\frac { \\sqrt { \\epsilon } } { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 239, + 507, + 258 + ], + "score": 1.0, + "content": ". By the equivalent definition of total", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 254, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 104, + 254, + 180, + 272 + ], + "score": 1.0, + "content": "variation distance", + "type": "text" + }, + { + "bbox": [ + 180, + 255, + 390, + 270 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathrm { T V } ( \\widetilde P ^ { ( 1 ) } , \\widetilde P ^ { ( 0 ) } ) ) = \\operatorname* { m a x } _ { A \\subset \\mathbb { R } ^ { d } } [ \\widetilde P ^ { ( 1 ) } ( A ) - \\widetilde P ^ { ( 0 ) } ( A ) ] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 254, + 483, + 272 + ], + "score": 1.0, + "content": ", we know that for any", + "type": "text" + }, + { + "bbox": [ + 484, + 257, + 505, + 269 + ], + "score": 0.84, + "content": "A \\subset", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 267, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 119, + 282 + ], + "score": 0.79, + "content": "\\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 267, + 133, + 286 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 133, + 269, + 244, + 285 + ], + "score": 0.86, + "content": "\\begin{array} { r } { \\widetilde { P } ^ { ( 1 ) } ( A ) - \\widetilde { P } ^ { ( 0 ) } ( A ) \\le \\frac { \\sqrt { \\epsilon } } { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 267, + 353, + 286 + ], + "score": 1.0, + "content": ". For any scoring function", + "type": "text" + }, + { + "bbox": [ + 353, + 274, + 360, + 281 + ], + "score": 0.72, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 267, + 460, + 286 + ], + "score": 1.0, + "content": "and any threshold value", + "type": "text" + }, + { + "bbox": [ + 461, + 272, + 466, + 282 + ], + "score": 0.71, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 267, + 483, + 286 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 483, + 271, + 505, + 283 + ], + "score": 0.83, + "content": "A =", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 282, + 428, + 302 + ], + "spans": [ + { + "bbox": [ + 107, + 286, + 165, + 299 + ], + "score": 0.92, + "content": "\\{ g : r ( g ) > t \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 282, + 223, + 302 + ], + "score": 1.0, + "content": ", then we have", + "type": "text" + }, + { + "bbox": [ + 224, + 284, + 424, + 300 + ], + "score": 0.85, + "content": "\\begin{array} { r } { \\bar { \\mathrm { T P R } } _ { r } ( t ) - \\mathrm { F P R } _ { r } ( t ) = \\widetilde { P } ^ { ( 1 ) } ( A ) - \\widetilde { P } ^ { ( 0 ) } ( A ) \\leq \\frac { \\sqrt { \\epsilon } } { 2 } . } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 282, + 428, + 302 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 104, + 239, + 507, + 302 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 303, + 467, + 316 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 467, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 281, + 317 + ], + "score": 1.0, + "content": "Therefore, the AUC of the scoring function", + "type": "text" + }, + { + "bbox": [ + 281, + 306, + 288, + 313 + ], + "score": 0.77, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 302, + 467, + 317 + ], + "score": 1.0, + "content": "can be upper bounded in the following way:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 302, + 467, + 317 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 189, + 320, + 421, + 378 + ], + "lines": [ + { + "bbox": [ + 189, + 320, + 421, + 378 + ], + "spans": [ + { + "bbox": [ + 189, + 320, + 421, + 378 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle \\mathrm { A U C } ( { \\boldsymbol { r } } ) = \\int _ { \\infty } ^ { - \\infty } { \\mathrm { T P R } _ { r } ( t ) ~ d \\mathrm { F P R } _ { r } ( t ) } } \\\\ { \\displaystyle \\leq \\int _ { \\infty } ^ { - \\infty } \\operatorname* { m i n } \\bigg ( \\mathrm { F P R } _ { r } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } , ~ 1 \\bigg ) ~ d \\mathrm { F P R } _ { r } ( t ) , } \\end{array}", + "type": "interline_equation", + "image_path": "ba8c8d0ee2d8f4faa55ba028d41a5b23ac0488964a59711dbccd1403cdd0075e.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 189, + 320, + 421, + 339.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 189, + 339.3333333333333, + 421, + 358.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 189, + 358.66666666666663, + 421, + 377.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 381, + 388, + 394 + ], + "lines": [ + { + "bbox": [ + 106, + 381, + 389, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 279, + 395 + ], + "score": 1.0, + "content": "where in (6) we use the additional fact that", + "type": "text" + }, + { + "bbox": [ + 280, + 382, + 333, + 394 + ], + "score": 0.94, + "content": "\\mathrm { T P R } _ { r } ( t ) \\leq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 381, + 361, + 395 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 361, + 383, + 385, + 393 + ], + "score": 0.9, + "content": "t \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 381, + 389, + 395 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 106, + 381, + 389, + 395 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 399, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 135, + 417 + ], + "score": 1.0, + "content": "When", + "type": "text" + }, + { + "bbox": [ + 135, + 402, + 182, + 414 + ], + "score": 0.91, + "content": "\\epsilon \\in \\ [ 0 , 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 398, + 229, + 417 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 230, + 399, + 306, + 416 + ], + "score": 0.9, + "content": "1 ~ - ~ \\frac { \\sqrt { \\epsilon } } { 2 } ~ \\in ~ ( 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 398, + 333, + 417 + ], + "score": 1.0, + "content": ". As", + "type": "text" + }, + { + "bbox": [ + 334, + 402, + 369, + 414 + ], + "score": 0.88, + "content": "\\mathrm { F P R } _ { r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 398, + 506, + 417 + ], + "score": 1.0, + "content": "is a monotonically nonincreas-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 102, + 412, + 508, + 437 + ], + "spans": [ + { + "bbox": [ + 102, + 412, + 172, + 437 + ], + "score": 1.0, + "content": "ing function in", + "type": "text" + }, + { + "bbox": [ + 172, + 419, + 177, + 428 + ], + "score": 0.73, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 412, + 239, + 437 + ], + "score": 1.0, + "content": "with range in", + "type": "text" + }, + { + "bbox": [ + 240, + 417, + 260, + 430 + ], + "score": 0.68, + "content": "[ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 412, + 298, + 437 + ], + "score": 1.0, + "content": ", the set", + "type": "text" + }, + { + "bbox": [ + 298, + 414, + 425, + 434 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\left\\{ t : \\mathrm { F P R } _ { r } ( t ) \\leq 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right\\} \\ \\neq \\ \\phi } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 412, + 508, + 437 + ], + "score": 1.0, + "content": "is not empty. Let", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 433, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 244, + 453 + ], + "score": 0.91, + "content": "\\begin{array} { r } { k : = \\operatorname* { i n f } \\left\\{ t : \\mathrm { F P R } _ { r } ( t ) \\leq 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right\\} } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 433, + 286, + 452 + ], + "score": 1.0, + "content": ". Again by", + "type": "text" + }, + { + "bbox": [ + 286, + 437, + 321, + 449 + ], + "score": 0.9, + "content": "\\mathrm { F P R } _ { r } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 433, + 506, + 452 + ], + "score": 1.0, + "content": "being a monotonically nonincreasing function", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 450, + 367, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 117, + 464 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 117, + 453, + 122, + 461 + ], + "score": 0.75, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 450, + 367, + 464 + ], + "score": 1.0, + "content": ", we can break the integration in Equation (6) into two terms:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 102, + 398, + 508, + 464 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 465, + 481, + 636 + ], + "lines": [ + { + "bbox": [ + 114, + 465, + 481, + 636 + ], + "spans": [ + { + "bbox": [ + 114, + 465, + 481, + 636 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\quad \\mathrm { A U C } ( r ) } \\\\ & { = \\displaystyle \\int _ { - \\infty } ^ { k } \\operatorname* { m i n } \\left( \\mathrm { F P R } _ { \\mathrm { v } } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } , \\mathbf { \\Phi } _ { 1 } \\right) d \\mathrm { F P R } _ { v } ( t ) + \\displaystyle \\int _ { k } ^ { - \\infty } \\operatorname* { m i n } \\left( \\mathrm { F P R } _ { v } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } , \\mathbf { \\Phi } _ { 1 } \\right) d \\mathrm { F P R } _ { v } ( t ) } \\\\ & { \\le \\displaystyle \\int _ { \\infty } ^ { k } \\left( \\mathrm { F P R } _ { v } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } \\right) d \\mathrm { F P R } _ { v } ( t ) + \\displaystyle \\int _ { k } ^ { - \\infty } 1 d \\mathrm { F P R } _ { v } ( t ) } \\\\ & { = \\left[ \\frac { \\left[ \\mathrm { F P R } _ { r } ( v ) \\right] ^ { 2 } } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } \\mathrm { F P R } _ { v } ( v ) \\right] \\displaystyle _ { t = \\infty } ^ { t - k } + \\mathrm { F P R } _ { v } ( t ) \\displaystyle _ { t = k } ^ { t - \\infty } } \\\\ & { \\le \\frac { ( 1 - \\frac { v \\epsilon } { 2 } ) ^ { 2 } } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } \\left( 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right) + \\left[ 1 - \\left( 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right) \\right] } \\\\ & { = \\frac { 1 } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } - \\frac { \\epsilon } { 8 } } \\end{array}", + "type": "interline_equation", + "image_path": "c37fca5f32294ed20c297a9758cfee8014b0b38c95a97a532212844660b914d2.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 114, + 465, + 481, + 522.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 114, + 522.0, + 481, + 579.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 114, + 579.0, + 481, + 636.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 639, + 505, + 663 + ], + "lines": [ + { + "bbox": [ + 105, + 638, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 320, + 652 + ], + "score": 1.0, + "content": "Since this inequality is true for any scoring function", + "type": "text" + }, + { + "bbox": [ + 320, + 642, + 325, + 649 + ], + "score": 0.74, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 638, + 506, + 652 + ], + "score": 1.0, + "content": ", it is true for the maximum value. Thus the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 650, + 504, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 182, + 664 + ], + "score": 1.0, + "content": "proof is complete.", + "type": "text" + }, + { + "bbox": [ + 496, + 653, + 504, + 660 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 638, + 506, + 664 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 329, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 329, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 329, + 94 + ], + "score": 1.0, + "content": "A.4 PROOF AND INTERPRETATION OF THEOREM 2", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 102, + 366, + 115 + ], + "lines": [ + { + "bbox": [ + 109, + 101, + 366, + 117 + ], + "spans": [ + { + "bbox": [ + 109, + 101, + 219, + 117 + ], + "score": 1.0, + "content": "(Theorem 2). The optimal", + "type": "text" + }, + { + "bbox": [ + 220, + 104, + 232, + 115 + ], + "score": 0.89, + "content": "\\Sigma _ { 1 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 101, + 251, + 117 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 252, + 104, + 264, + 115 + ], + "score": 0.9, + "content": "\\Sigma _ { 0 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 101, + 366, + 117 + ], + "score": 1.0, + "content": "to the following problem", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 118, + 514, + 153 + ], + "lines": [ + { + "bbox": [ + 118, + 118, + 512, + 138 + ], + "spans": [ + { + "bbox": [ + 118, + 118, + 512, + 138 + ], + "score": 0.89, + "content": "\\operatorname* { m i n } _ { \\Sigma _ { 0 } , \\Sigma _ { 1 } \\in \\mathbb S } K L ( \\mathcal { N } ( \\bar { g } ^ { ( 1 ) } , v I + \\Sigma _ { 1 } ) \\parallel \\mathcal { N } ( \\bar { g } ^ { ( 0 ) } , u I + \\Sigma _ { 0 } ) ) + K L ( \\mathcal { N } ( \\bar { g } ^ { ( 0 ) } , u I + \\Sigma _ { 0 } ) \\parallel \\mathcal { N } ( \\bar { g } ^ { ( 1 ) } , v I + \\Sigma _ { 1 } ) )", + "type": "inline_equation", + "image_path": "e796f999c6f8db73395f19307a6bc4aa91ee32f833aa00cfd239ab209f659492.jpg" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 139, + 154, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 139, + 154, + 155 + ], + "score": 1.0, + "content": "subject to", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "interline_equation", + "bbox": [ + 226, + 159, + 384, + 217 + ], + "lines": [ + { + "bbox": [ + 226, + 159, + 384, + 217 + ], + "spans": [ + { + "bbox": [ + 226, + 159, + 384, + 217 + ], + "score": 0.92, + "content": "\\begin{array} { c } { \\Sigma _ { 0 } \\Sigma _ { 1 } = \\Sigma _ { 1 } \\Sigma _ { 0 } , } \\\\ { p \\cdot \\operatorname { t r } ( \\Sigma _ { 1 } ) + ( 1 - p ) \\cdot \\operatorname { t r } ( \\Sigma _ { 0 } ) \\leq P , } \\\\ { \\Sigma _ { 1 } \\succeq \\mathbf { 0 } , } \\\\ { \\Sigma _ { 0 } \\succeq \\mathbf { 0 } . } \\end{array}", + "type": "interline_equation", + "image_path": "dc292def3437841d19baa0b9d23595c69b19d84032986df03d9daeda9c526603.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 226, + 159, + 384, + 178.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 226, + 178.33333333333334, + 384, + 197.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 226, + 197.66666666666669, + 384, + 217.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 218, + 166, + 230 + ], + "lines": [ + { + "bbox": [ + 105, + 218, + 168, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 168, + 231 + ], + "score": 1.0, + "content": "have the form:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 228, + 478, + 261 + ], + "lines": [ + { + "bbox": [ + 115, + 228, + 478, + 261 + ], + "spans": [ + { + "bbox": [ + 115, + 228, + 478, + 261 + ], + "score": 0.92, + "content": "\\Sigma _ { 1 } ^ { * } = \\frac { \\tilde { \\lambda } _ { 1 } ^ { ( 1 ) * } - \\lambda _ { 2 } ^ { ( 1 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 1 ) * } I _ { d } , \\quad \\Sigma _ { 0 } ^ { * } = \\frac { \\lambda _ { 1 } ^ { ( 0 ) * } - \\lambda _ { 2 } ^ { ( 0 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 0 ) * } I _ { d } ,", + "type": "interline_equation", + "image_path": "e2caa62cbc2222993059ebf827bb1ef6c7d107cc64ed8463fbe21d20f1f105d7.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 115, + 228, + 478, + 239.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 115, + 239.0, + 478, + 250.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 115, + 250.0, + 478, + 261.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 264, + 495, + 279 + ], + "lines": [ + { + "bbox": [ + 102, + 260, + 501, + 284 + ], + "spans": [ + { + "bbox": [ + 102, + 260, + 133, + 284 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 263, + 237, + 279 + ], + "score": 0.93, + "content": "( \\lambda _ { 1 } ^ { ( 0 ) * } , \\lambda _ { 2 } ^ { ( 0 ) * } , \\lambda _ { 1 } ^ { ( 1 ) * } , \\lambda _ { 2 } ^ { ( 1 ) * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 260, + 501, + 284 + ], + "score": 1.0, + "content": "is the solution to the following 4-variable optimization problem:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 282, + 501, + 370 + ], + "lines": [ + { + "bbox": [ + 111, + 282, + 501, + 370 + ], + "spans": [ + { + "bbox": [ + 111, + 282, + 501, + 370 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\underset { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } { \\operatorname* { m i n } } ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u } } \\\\ & { \\mathrm { ~ s . t . ~ } p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) \\lambda _ { 2 } ^ { ( 0 ) } \\leq P , } \\\\ & { \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 , \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 1 } ^ { ( 0 ) } \\leq ~ 0 , \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 2 } ^ { ( 1 ) } \\leq ~ 0 , \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 2 } ^ { ( 0 ) } \\leq ~ 0 , } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 1 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq ~ 0 , \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 } \\end{array}", + "type": "interline_equation", + "image_path": "39c2d9dbc0dccd7e6d54af37b083ce9f0ba4f411e91f17143a1e07b418c60dd8.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 111, + 282, + 501, + 311.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 111, + 311.3333333333333, + 501, + 340.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 111, + 340.66666666666663, + 501, + 369.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 391, + 505, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 505, + 406 + ], + "score": 1.0, + "content": "Proof of Theorem 2. By writing out the analytical close-form of the KL divergence between two", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 402, + 340, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 340, + 416 + ], + "score": 1.0, + "content": "Gaussian distributions, the optimization can be written as:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 416, + 468, + 455 + ], + "lines": [ + { + "bbox": [ + 145, + 416, + 468, + 455 + ], + "spans": [ + { + "bbox": [ + 145, + 416, + 468, + 455 + ], + "score": 0.88, + "content": "\\begin{array} { c } { { \\displaystyle \\operatorname* { m i n } _ { \\Sigma _ { 0 } , \\Sigma _ { 1 } \\in \\mathbb S } \\mathrm { t r } \\big ( ( \\Sigma _ { 1 } + v I ) ^ { - 1 } ( \\Sigma _ { 0 } + u I ) \\big ) + \\mathrm { t r } \\big ( ( \\Sigma _ { 0 } + u I ) ^ { - 1 } ( \\Sigma _ { 1 } + v I ) \\big ) + } } \\\\ { { \\displaystyle \\big ( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } \\big ) ^ { \\top } \\left( ( \\Sigma _ { 1 } + v I ) ^ { - 1 } + ( \\Sigma _ { 0 } + u I ) ^ { - 1 } \\right) \\left( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } \\right) } } \\end{array}", + "type": "interline_equation", + "image_path": "ee62646b8070c24ddfb6ceb375beed0eb93db91e867078f6c09ac3815840ee7d.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 145, + 416, + 468, + 429.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 145, + 429.0, + 468, + 442.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 145, + 442.0, + 468, + 455.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 145, + 456, + 192, + 467 + ], + "lines": [ + { + "bbox": [ + 145, + 455, + 193, + 469 + ], + "spans": [ + { + "bbox": [ + 145, + 455, + 193, + 469 + ], + "score": 1.0, + "content": "subject to", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 226, + 471, + 385, + 529 + ], + "lines": [ + { + "bbox": [ + 226, + 471, + 385, + 529 + ], + "spans": [ + { + "bbox": [ + 226, + 471, + 385, + 529 + ], + "score": 0.93, + "content": "\\begin{array} { c } { \\Sigma _ { 0 } \\Sigma _ { 1 } = \\Sigma _ { 1 } \\Sigma _ { 0 } , } \\\\ { p \\cdot \\operatorname { t r } ( \\Sigma _ { 1 } ) + ( 1 - p ) \\cdot \\operatorname { t r } ( \\Sigma _ { 0 } ) \\leq P , } \\\\ { \\Sigma _ { 1 } \\succeq \\mathbf { 0 } , } \\\\ { \\Sigma _ { 0 } \\succeq \\mathbf { 0 } . } \\end{array}", + "type": "interline_equation", + "image_path": "08418693ea8bfdd01a7a6a1c7ddf637978c73c26e217e60ed73645fa0564c850.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 226, + 471, + 385, + 490.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 226, + 490.3333333333333, + 385, + 509.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 226, + 509.66666666666663, + 385, + 529.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 533, + 504, + 556 + ], + "lines": [ + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 399, + 546 + ], + "score": 1.0, + "content": "By the commutative constraint on the two positive semidefinite matrices", + "type": "text" + }, + { + "bbox": [ + 400, + 534, + 412, + 545 + ], + "score": 0.89, + "content": "\\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 534, + 431, + 546 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 432, + 534, + 444, + 545 + ], + "score": 0.89, + "content": "\\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 534, + 505, + 546 + ], + "score": 1.0, + "content": ", we know that", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 545, + 441, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 441, + 557 + ], + "score": 1.0, + "content": "we can factor these two matrices using the same set of eigenvectors. We thus write:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "interline_equation", + "bbox": [ + 240, + 559, + 370, + 595 + ], + "lines": [ + { + "bbox": [ + 240, + 559, + 370, + 595 + ], + "spans": [ + { + "bbox": [ + 240, + 559, + 370, + 595 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\Sigma _ { 0 } = \\boldsymbol { Q } ^ { \\top } \\mathrm { d i a g } ( \\lambda _ { 1 } ^ { ( 0 ) } , \\ldots , \\lambda _ { d } ^ { ( 0 ) } ) \\boldsymbol { Q } , } \\\\ { \\Sigma _ { 1 } = \\boldsymbol { Q } ^ { \\top } \\mathrm { d i a g } ( \\lambda _ { 1 } ^ { ( 1 ) } , \\ldots , \\lambda _ { d } ^ { ( 1 ) } ) \\boldsymbol { Q } , } \\end{array}", + "type": "interline_equation", + "image_path": "b474dca7717260cb89d217095f4862721dc5a64fe6ba226f4a2d37963156c21d.jpg" + } + ] + } + ], + "index": 26.5, + "virtual_lines": [ + { + "bbox": [ + 240, + 559, + 370, + 577.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 240, + 577.0, + 370, + 595.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 598, + 502, + 623 + ], + "lines": [ + { + "bbox": [ + 104, + 594, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 104, + 594, + 134, + 615 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 598, + 181, + 612 + ], + "score": 0.93, + "content": "Q \\in \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 594, + 364, + 615 + ], + "score": 1.0, + "content": "is an orthogonal matrix and the eigenvalues", + "type": "text" + }, + { + "bbox": [ + 365, + 597, + 403, + 612 + ], + "score": 0.93, + "content": "\\lambda _ { i } ^ { ( 0 ) } , \\lambda _ { i } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 594, + 506, + 615 + ], + "score": 1.0, + "content": "are nonnegative and de-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 610, + 179, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 179, + 623 + ], + "score": 1.0, + "content": "creasing in value.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 627, + 504, + 654 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 260, + 641 + ], + "score": 1.0, + "content": "Using this alternative expression of", + "type": "text" + }, + { + "bbox": [ + 260, + 628, + 273, + 639 + ], + "score": 0.89, + "content": "\\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 626, + 295, + 641 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 295, + 628, + 308, + 639 + ], + "score": 0.88, + "content": "\\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 626, + 506, + 641 + ], + "score": 1.0, + "content": ", we can express the optimization in terms of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 637, + 183, + 656 + ], + "spans": [ + { + "bbox": [ + 107, + 639, + 178, + 654 + ], + "score": 0.93, + "content": "\\{ \\lambda _ { i } ^ { ( 1 ) } \\} , \\{ \\lambda _ { i } ^ { ( 0 ) } \\} , Q", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 637, + 183, + 656 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "interline_equation", + "bbox": [ + 123, + 662, + 492, + 732 + ], + "lines": [ + { + "bbox": [ + 123, + 662, + 492, + 732 + ], + "spans": [ + { + "bbox": [ + 123, + 662, + 492, + 732 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\underset { \\{ \\lambda _ { i } ^ { ( 1 ) } \\} , \\{ \\lambda _ { i } ^ { ( 0 ) } \\} , Q } { \\operatorname* { m i n } } } & { \\displaystyle \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { i } ^ { ( 0 ) } + u } + } \\\\ & { \\displaystyle \\left[ Q ( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } ) \\right] ^ { \\top } \\mathrm { d i a g } \\left( \\cdots , \\frac { 1 } { \\lambda _ { i } ^ { ( 0 ) } + u } + \\frac { 1 } { \\lambda _ { i } ^ { ( 1 ) } + v } , \\ldots \\right) Q ( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } ) } \\end{array}", + "type": "interline_equation", + "image_path": "5ad55577ab10a26de34f4e01be3083928b03f499e7e0cb3f91405f36297aafe8.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 123, + 662, + 492, + 685.3333333333334 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 123, + 685.3333333333334, + 492, + 708.6666666666667 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 123, + 708.6666666666667, + 492, + 732.0000000000001 + ], + "spans": [], + "index": 34 + } + ] + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 329, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 329, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 329, + 94 + ], + "score": 1.0, + "content": "A.4 PROOF AND INTERPRETATION OF THEOREM 2", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 102, + 366, + 115 + ], + "lines": [ + { + "bbox": [ + 109, + 101, + 366, + 117 + ], + "spans": [ + { + "bbox": [ + 109, + 101, + 219, + 117 + ], + "score": 1.0, + "content": "(Theorem 2). The optimal", + "type": "text" + }, + { + "bbox": [ + 220, + 104, + 232, + 115 + ], + "score": 0.89, + "content": "\\Sigma _ { 1 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 101, + 251, + 117 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 252, + 104, + 264, + 115 + ], + "score": 0.9, + "content": "\\Sigma _ { 0 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 101, + 366, + 117 + ], + "score": 1.0, + "content": "to the following problem", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 109, + 101, + 366, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 118, + 514, + 153 + ], + "lines": [ + { + "bbox": [ + 118, + 118, + 512, + 138 + ], + "spans": [ + { + "bbox": [ + 118, + 118, + 512, + 138 + ], + "score": 0.89, + "content": "\\operatorname* { m i n } _ { \\Sigma _ { 0 } , \\Sigma _ { 1 } \\in \\mathbb S } K L ( \\mathcal { N } ( \\bar { g } ^ { ( 1 ) } , v I + \\Sigma _ { 1 } ) \\parallel \\mathcal { N } ( \\bar { g } ^ { ( 0 ) } , u I + \\Sigma _ { 0 } ) ) + K L ( \\mathcal { N } ( \\bar { g } ^ { ( 0 ) } , u I + \\Sigma _ { 0 } ) \\parallel \\mathcal { N } ( \\bar { g } ^ { ( 1 ) } , v I + \\Sigma _ { 1 } ) )", + "type": "inline_equation", + "image_path": "e796f999c6f8db73395f19307a6bc4aa91ee32f833aa00cfd239ab209f659492.jpg" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 139, + 154, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 139, + 154, + 155 + ], + "score": 1.0, + "content": "subject to", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 104, + 118, + 512, + 155 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 226, + 159, + 384, + 217 + ], + "lines": [ + { + "bbox": [ + 226, + 159, + 384, + 217 + ], + "spans": [ + { + "bbox": [ + 226, + 159, + 384, + 217 + ], + "score": 0.92, + "content": "\\begin{array} { c } { \\Sigma _ { 0 } \\Sigma _ { 1 } = \\Sigma _ { 1 } \\Sigma _ { 0 } , } \\\\ { p \\cdot \\operatorname { t r } ( \\Sigma _ { 1 } ) + ( 1 - p ) \\cdot \\operatorname { t r } ( \\Sigma _ { 0 } ) \\leq P , } \\\\ { \\Sigma _ { 1 } \\succeq \\mathbf { 0 } , } \\\\ { \\Sigma _ { 0 } \\succeq \\mathbf { 0 } . } \\end{array}", + "type": "interline_equation", + "image_path": "dc292def3437841d19baa0b9d23595c69b19d84032986df03d9daeda9c526603.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 226, + 159, + 384, + 178.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 226, + 178.33333333333334, + 384, + 197.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 226, + 197.66666666666669, + 384, + 217.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 218, + 166, + 230 + ], + "lines": [ + { + "bbox": [ + 105, + 218, + 168, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 168, + 231 + ], + "score": 1.0, + "content": "have the form:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 218, + 168, + 231 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 228, + 478, + 261 + ], + "lines": [ + { + "bbox": [ + 115, + 228, + 478, + 261 + ], + "spans": [ + { + "bbox": [ + 115, + 228, + 478, + 261 + ], + "score": 0.92, + "content": "\\Sigma _ { 1 } ^ { * } = \\frac { \\tilde { \\lambda } _ { 1 } ^ { ( 1 ) * } - \\lambda _ { 2 } ^ { ( 1 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 1 ) * } I _ { d } , \\quad \\Sigma _ { 0 } ^ { * } = \\frac { \\lambda _ { 1 } ^ { ( 0 ) * } - \\lambda _ { 2 } ^ { ( 0 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 0 ) * } I _ { d } ,", + "type": "interline_equation", + "image_path": "e2caa62cbc2222993059ebf827bb1ef6c7d107cc64ed8463fbe21d20f1f105d7.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 115, + 228, + 478, + 239.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 115, + 239.0, + 478, + 250.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 115, + 250.0, + 478, + 261.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 264, + 495, + 279 + ], + "lines": [ + { + "bbox": [ + 102, + 260, + 501, + 284 + ], + "spans": [ + { + "bbox": [ + 102, + 260, + 133, + 284 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 263, + 237, + 279 + ], + "score": 0.93, + "content": "( \\lambda _ { 1 } ^ { ( 0 ) * } , \\lambda _ { 2 } ^ { ( 0 ) * } , \\lambda _ { 1 } ^ { ( 1 ) * } , \\lambda _ { 2 } ^ { ( 1 ) * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 260, + 501, + 284 + ], + "score": 1.0, + "content": "is the solution to the following 4-variable optimization problem:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 102, + 260, + 501, + 284 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 282, + 501, + 370 + ], + "lines": [ + { + "bbox": [ + 111, + 282, + 501, + 370 + ], + "spans": [ + { + "bbox": [ + 111, + 282, + 501, + 370 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\underset { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } { \\operatorname* { m i n } } ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u } } \\\\ & { \\mathrm { ~ s . t . ~ } p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) \\lambda _ { 2 } ^ { ( 0 ) } \\leq P , } \\\\ & { \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 , \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 1 } ^ { ( 0 ) } \\leq ~ 0 , \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 2 } ^ { ( 1 ) } \\leq ~ 0 , \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 2 } ^ { ( 0 ) } \\leq ~ 0 , } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 1 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq ~ 0 , \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 } \\end{array}", + "type": "interline_equation", + "image_path": "39c2d9dbc0dccd7e6d54af37b083ce9f0ba4f411e91f17143a1e07b418c60dd8.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 111, + 282, + 501, + 311.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 111, + 311.3333333333333, + 501, + 340.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 111, + 340.66666666666663, + 501, + 369.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 391, + 505, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 505, + 406 + ], + "score": 1.0, + "content": "Proof of Theorem 2. By writing out the analytical close-form of the KL divergence between two", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 402, + 340, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 340, + 416 + ], + "score": 1.0, + "content": "Gaussian distributions, the optimization can be written as:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 391, + 505, + 416 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 416, + 468, + 455 + ], + "lines": [ + { + "bbox": [ + 145, + 416, + 468, + 455 + ], + "spans": [ + { + "bbox": [ + 145, + 416, + 468, + 455 + ], + "score": 0.88, + "content": "\\begin{array} { c } { { \\displaystyle \\operatorname* { m i n } _ { \\Sigma _ { 0 } , \\Sigma _ { 1 } \\in \\mathbb S } \\mathrm { t r } \\big ( ( \\Sigma _ { 1 } + v I ) ^ { - 1 } ( \\Sigma _ { 0 } + u I ) \\big ) + \\mathrm { t r } \\big ( ( \\Sigma _ { 0 } + u I ) ^ { - 1 } ( \\Sigma _ { 1 } + v I ) \\big ) + } } \\\\ { { \\displaystyle \\big ( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } \\big ) ^ { \\top } \\left( ( \\Sigma _ { 1 } + v I ) ^ { - 1 } + ( \\Sigma _ { 0 } + u I ) ^ { - 1 } \\right) \\left( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } \\right) } } \\end{array}", + "type": "interline_equation", + "image_path": "ee62646b8070c24ddfb6ceb375beed0eb93db91e867078f6c09ac3815840ee7d.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 145, + 416, + 468, + 429.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 145, + 429.0, + 468, + 442.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 145, + 442.0, + 468, + 455.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 145, + 456, + 192, + 467 + ], + "lines": [ + { + "bbox": [ + 145, + 455, + 193, + 469 + ], + "spans": [ + { + "bbox": [ + 145, + 455, + 193, + 469 + ], + "score": 1.0, + "content": "subject to", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 145, + 455, + 193, + 469 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 226, + 471, + 385, + 529 + ], + "lines": [ + { + "bbox": [ + 226, + 471, + 385, + 529 + ], + "spans": [ + { + "bbox": [ + 226, + 471, + 385, + 529 + ], + "score": 0.93, + "content": "\\begin{array} { c } { \\Sigma _ { 0 } \\Sigma _ { 1 } = \\Sigma _ { 1 } \\Sigma _ { 0 } , } \\\\ { p \\cdot \\operatorname { t r } ( \\Sigma _ { 1 } ) + ( 1 - p ) \\cdot \\operatorname { t r } ( \\Sigma _ { 0 } ) \\leq P , } \\\\ { \\Sigma _ { 1 } \\succeq \\mathbf { 0 } , } \\\\ { \\Sigma _ { 0 } \\succeq \\mathbf { 0 } . } \\end{array}", + "type": "interline_equation", + "image_path": "08418693ea8bfdd01a7a6a1c7ddf637978c73c26e217e60ed73645fa0564c850.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 226, + 471, + 385, + 490.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 226, + 490.3333333333333, + 385, + 509.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 226, + 509.66666666666663, + 385, + 529.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 533, + 504, + 556 + ], + "lines": [ + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 399, + 546 + ], + "score": 1.0, + "content": "By the commutative constraint on the two positive semidefinite matrices", + "type": "text" + }, + { + "bbox": [ + 400, + 534, + 412, + 545 + ], + "score": 0.89, + "content": "\\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 534, + 431, + 546 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 432, + 534, + 444, + 545 + ], + "score": 0.89, + "content": "\\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 534, + 505, + 546 + ], + "score": 1.0, + "content": ", we know that", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 545, + 441, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 441, + 557 + ], + "score": 1.0, + "content": "we can factor these two matrices using the same set of eigenvectors. We thus write:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 106, + 534, + 505, + 557 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 240, + 559, + 370, + 595 + ], + "lines": [ + { + "bbox": [ + 240, + 559, + 370, + 595 + ], + "spans": [ + { + "bbox": [ + 240, + 559, + 370, + 595 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\Sigma _ { 0 } = \\boldsymbol { Q } ^ { \\top } \\mathrm { d i a g } ( \\lambda _ { 1 } ^ { ( 0 ) } , \\ldots , \\lambda _ { d } ^ { ( 0 ) } ) \\boldsymbol { Q } , } \\\\ { \\Sigma _ { 1 } = \\boldsymbol { Q } ^ { \\top } \\mathrm { d i a g } ( \\lambda _ { 1 } ^ { ( 1 ) } , \\ldots , \\lambda _ { d } ^ { ( 1 ) } ) \\boldsymbol { Q } , } \\end{array}", + "type": "interline_equation", + "image_path": "b474dca7717260cb89d217095f4862721dc5a64fe6ba226f4a2d37963156c21d.jpg" + } + ] + } + ], + "index": 26.5, + "virtual_lines": [ + { + "bbox": [ + 240, + 559, + 370, + 577.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 240, + 577.0, + 370, + 595.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 598, + 502, + 623 + ], + "lines": [ + { + "bbox": [ + 104, + 594, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 104, + 594, + 134, + 615 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 598, + 181, + 612 + ], + "score": 0.93, + "content": "Q \\in \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 594, + 364, + 615 + ], + "score": 1.0, + "content": "is an orthogonal matrix and the eigenvalues", + "type": "text" + }, + { + "bbox": [ + 365, + 597, + 403, + 612 + ], + "score": 0.93, + "content": "\\lambda _ { i } ^ { ( 0 ) } , \\lambda _ { i } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 594, + 506, + 615 + ], + "score": 1.0, + "content": "are nonnegative and de-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 610, + 179, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 179, + 623 + ], + "score": 1.0, + "content": "creasing in value.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 104, + 594, + 506, + 623 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 627, + 504, + 654 + ], + "lines": [ + { + "bbox": [ + 105, + 626, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 260, + 641 + ], + "score": 1.0, + "content": "Using this alternative expression of", + "type": "text" + }, + { + "bbox": [ + 260, + 628, + 273, + 639 + ], + "score": 0.89, + "content": "\\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 626, + 295, + 641 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 295, + 628, + 308, + 639 + ], + "score": 0.88, + "content": "\\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 626, + 506, + 641 + ], + "score": 1.0, + "content": ", we can express the optimization in terms of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 637, + 183, + 656 + ], + "spans": [ + { + "bbox": [ + 107, + 639, + 178, + 654 + ], + "score": 0.93, + "content": "\\{ \\lambda _ { i } ^ { ( 1 ) } \\} , \\{ \\lambda _ { i } ^ { ( 0 ) } \\} , Q", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 637, + 183, + 656 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 626, + 506, + 656 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 123, + 662, + 492, + 732 + ], + "lines": [ + { + "bbox": [ + 123, + 662, + 492, + 732 + ], + "spans": [ + { + "bbox": [ + 123, + 662, + 492, + 732 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\underset { \\{ \\lambda _ { i } ^ { ( 1 ) } \\} , \\{ \\lambda _ { i } ^ { ( 0 ) } \\} , Q } { \\operatorname* { m i n } } } & { \\displaystyle \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { i } ^ { ( 0 ) } + u } + } \\\\ & { \\displaystyle \\left[ Q ( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } ) \\right] ^ { \\top } \\mathrm { d i a g } \\left( \\cdots , \\frac { 1 } { \\lambda _ { i } ^ { ( 0 ) } + u } + \\frac { 1 } { \\lambda _ { i } ^ { ( 1 ) } + v } , \\ldots \\right) Q ( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } ) } \\end{array}", + "type": "interline_equation", + "image_path": "5ad55577ab10a26de34f4e01be3083928b03f499e7e0cb3f91405f36297aafe8.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 123, + 662, + 492, + 685.3333333333334 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 123, + 685.3333333333334, + 492, + 708.6666666666667 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 123, + 708.6666666666667, + 492, + 732.0000000000001 + ], + "spans": [], + "index": 34 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "interline_equation", + "bbox": [ + 201, + 79, + 408, + 200 + ], + "lines": [ + { + "bbox": [ + 201, + 79, + 408, + 200 + ], + "spans": [ + { + "bbox": [ + 201, + 79, + 408, + 200 + ], + "score": 0.93, + "content": "\\begin{array} { r l } { \\mathrm { s u b j e c t ~ t o ~ } } & { p ( \\displaystyle \\sum _ { i = 1 } ^ { d } \\lambda _ { i } ^ { ( 1 ) } ) + ( 1 - p ) ( \\displaystyle \\sum _ { i = 1 } ^ { d } \\lambda _ { i } ^ { ( 0 ) } ) \\le P } \\\\ & { - \\lambda _ { i } ^ { ( 1 ) } \\le 0 , \\forall i \\in [ d ] } \\\\ & { - \\lambda _ { i } ^ { ( 0 ) } \\le 0 , \\forall i \\in [ d ] . } \\\\ & { \\lambda _ { i } ^ { ( 1 ) } \\ge \\lambda _ { j } ^ { ( 1 ) } , \\forall i < j . } \\\\ & { \\lambda _ { i } ^ { ( 0 ) } \\ge \\lambda _ { j } ^ { ( 0 ) } , \\forall i < j . } \\\\ & { Q \\ \\mathrm { o r t h o g o n a l } . } \\end{array}", + "type": "interline_equation", + "image_path": "658a524177670a85c5c1bc827631eb1439bbe9a4b9abd4f734f229d947058e98.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 201, + 79, + 408, + 99.16666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 201, + 99.16666666666667, + 408, + 119.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 201, + 119.33333333333334, + 408, + 139.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 201, + 139.5, + 408, + 159.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 201, + 159.66666666666666, + 408, + 179.83333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 201, + 179.83333333333331, + 408, + 199.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 227, + 506, + 264 + ], + "lines": [ + { + "bbox": [ + 104, + 226, + 507, + 245 + ], + "spans": [ + { + "bbox": [ + 104, + 226, + 197, + 245 + ], + "score": 1.0, + "content": "For any fixed feasible", + "type": "text" + }, + { + "bbox": [ + 198, + 227, + 255, + 243 + ], + "score": 0.94, + "content": "\\{ \\lambda _ { i } ^ { ( 1 ) } \\} , \\{ \\lambda _ { i } ^ { ( 0 ) } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 226, + 431, + 245 + ], + "score": 1.0, + "content": ", we see that the corresponding minimizing", + "type": "text" + }, + { + "bbox": [ + 432, + 231, + 441, + 241 + ], + "score": 0.85, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 226, + 507, + 245 + ], + "score": 1.0, + "content": "will set its first", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 104, + 240, + 292, + 254 + ], + "score": 1.0, + "content": "row to be the unit vector in the direction of", + "type": "text" + }, + { + "bbox": [ + 293, + 241, + 307, + 253 + ], + "score": 0.87, + "content": "\\Delta g", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 240, + 421, + 254 + ], + "score": 1.0, + "content": ". Thus by first minimizing", + "type": "text" + }, + { + "bbox": [ + 422, + 242, + 431, + 253 + ], + "score": 0.84, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 240, + 506, + 254 + ], + "score": 1.0, + "content": ", the optimization", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 252, + 191, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 252, + 191, + 265 + ], + "score": 1.0, + "content": "objective reduces to:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 284, + 460, + 354 + ], + "lines": [ + { + "bbox": [ + 150, + 284, + 460, + 354 + ], + "spans": [ + { + "bbox": [ + 150, + 284, + 460, + 354 + ], + "score": 0.93, + "content": "\\begin{array} { c c l } { \\displaystyle \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\displaystyle \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { i } ^ { ( 0 ) } + u } + \\frac { g } { \\lambda _ { 1 } ^ { ( 0 ) } + u } + \\frac { g } { \\lambda _ { 1 } ^ { ( 1 ) } + v } } \\\\ { = \\displaystyle \\sum _ { i = 2 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\displaystyle \\sum _ { i = 2 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { i } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } } \\end{array}", + "type": "interline_equation", + "image_path": "706550a04bb8352b793ee0ef90e1e490bae791f4828804effa9828257d1beb44.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 150, + 284, + 460, + 307.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 150, + 307.3333333333333, + 460, + 330.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 150, + 330.66666666666663, + 460, + 353.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 374, + 506, + 481 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 254, + 398 + ], + "score": 1.0, + "content": "We see that for the pair of variable", + "type": "text" + }, + { + "bbox": [ + 255, + 376, + 300, + 391 + ], + "score": 0.82, + "content": "( \\lambda _ { i } ^ { ( 1 ) } , \\lambda _ { i } ^ { ( 0 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 378, + 335, + 390 + ], + "score": 0.6, + "content": "( i ~ \\geq ~ 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 368, + 394, + 398 + ], + "score": 1.0, + "content": ", the function", + "type": "text" + }, + { + "bbox": [ + 394, + 374, + 462, + 396 + ], + "score": 0.95, + "content": "\\begin{array} { r } { \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { 1 } ^ { ( 0 ) } + u } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 370, + 506, + 400 + ], + "score": 1.0, + "content": "is strictly", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 393, + 508, + 412 + ], + "spans": [ + { + "bbox": [ + 104, + 393, + 228, + 412 + ], + "score": 1.0, + "content": "convex over the line segment", + "type": "text" + }, + { + "bbox": [ + 229, + 395, + 334, + 410 + ], + "score": 0.92, + "content": "p \\lambda _ { i } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { i } ^ { ( 0 ) } = c", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 393, + 420, + 412 + ], + "score": 1.0, + "content": "for any nonnegative", + "type": "text" + }, + { + "bbox": [ + 420, + 400, + 426, + 408 + ], + "score": 0.71, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 393, + 508, + 412 + ], + "score": 1.0, + "content": "and attains the the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 407, + 508, + 427 + ], + "spans": [ + { + "bbox": [ + 104, + 407, + 184, + 427 + ], + "score": 1.0, + "content": "minimum value at", + "type": "text" + }, + { + "bbox": [ + 184, + 409, + 224, + 424 + ], + "score": 0.94, + "content": "\\lambda _ { i } ^ { ( 1 ) } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 407, + 251, + 427 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 251, + 413, + 281, + 423 + ], + "score": 0.86, + "content": "u \\ : < \\ : v", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 407, + 300, + 427 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 300, + 410, + 340, + 424 + ], + "score": 0.92, + "content": "\\lambda _ { i } ^ { ( 0 ) } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 407, + 367, + 427 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 367, + 412, + 397, + 423 + ], + "score": 0.88, + "content": "u \\geq v", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 407, + 508, + 427 + ], + "score": 1.0, + "content": ". Suppose without loss of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 103, + 422, + 507, + 439 + ], + "spans": [ + { + "bbox": [ + 103, + 422, + 149, + 439 + ], + "score": 1.0, + "content": "generality", + "type": "text" + }, + { + "bbox": [ + 150, + 427, + 177, + 438 + ], + "score": 0.9, + "content": "u \\geq v", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 422, + 360, + 439 + ], + "score": 1.0, + "content": ", then for the optimal solution we must have", + "type": "text" + }, + { + "bbox": [ + 360, + 424, + 399, + 439 + ], + "score": 0.94, + "content": "\\lambda _ { i } ^ { ( 0 ) } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 422, + 428, + 439 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 428, + 426, + 454, + 438 + ], + "score": 0.89, + "content": "i \\geq 2", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 422, + 507, + 439 + ], + "score": 1.0, + "content": ". Under this", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 102, + 434, + 510, + 473 + ], + "spans": [ + { + "bbox": [ + 102, + 435, + 156, + 473 + ], + "score": 1.0, + "content": "Thus for all", + "type": "text" + }, + { + "bbox": [ + 103, + 434, + 321, + 455 + ], + "score": 1.0, + "content": "condition, we notice that the function m(x) = ux+v", + "type": "text" + }, + { + "bbox": [ + 156, + 451, + 186, + 471 + ], + "score": 0.94, + "content": "\\left\\{ \\lambda _ { i } ^ { \\left( 1 \\right) } \\right\\}", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 435, + 239, + 473 + ], + "score": 1.0, + "content": "that satisfies", + "type": "text" + }, + { + "bbox": [ + 240, + 452, + 301, + 468 + ], + "score": 0.93, + "content": "\\Sigma _ { i = 2 } ^ { d } \\lambda _ { i } ^ { ( 1 ) } = c", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 435, + 396, + 473 + ], + "score": 1.0, + "content": "is strictly c nonnegative", + "type": "text" + }, + { + "bbox": [ + 396, + 458, + 401, + 465 + ], + "score": 0.7, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 435, + 510, + 473 + ], + "score": 1.0, + "content": "vex on the positive reals., by Jensen inequality, we", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 467, + 129, + 483 + ], + "spans": [ + { + "bbox": [ + 104, + 467, + 129, + 483 + ], + "score": 1.0, + "content": "have", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 499, + 382, + 632 + ], + "lines": [ + { + "bbox": [ + 225, + 499, + 382, + 632 + ], + "spans": [ + { + "bbox": [ + 225, + 499, + 382, + 632 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle \\frac { 1 } { d - 1 } \\sum _ { i = 2 } ^ { d } \\left( \\frac { u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { u } \\right) } \\\\ { \\displaystyle = \\frac { 1 } { d - 1 } \\sum _ { i = 2 } ^ { d } m ( \\lambda _ { i } ^ { ( 1 ) } ) } \\\\ { \\displaystyle \\geq m \\left( \\frac { 1 } { d - 1 } \\sum _ { i = 2 } ^ { d } \\lambda _ { i } ^ { ( 1 ) } \\right) } \\\\ { \\displaystyle = m \\left( \\frac { c } { d - 1 } \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "16c9236733afaaf4e375a5a8e5a3607ba4a0d5d0e2a6b8857db7936e46f5e51a.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 225, + 499, + 382, + 518.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 225, + 518.0, + 382, + 537.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 225, + 537.0, + 382, + 556.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 225, + 556.0, + 382, + 575.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 225, + 575.0, + 382, + 594.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 225, + 594.0, + 382, + 613.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 225, + 613.0, + 382, + 632.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 658, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 104, + 656, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 104, + 656, + 322, + 675 + ], + "score": 1.0, + "content": "From this, we see that the optimal solution’s variables", + "type": "text" + }, + { + "bbox": [ + 323, + 658, + 350, + 673 + ], + "score": 0.94, + "content": "\\{ \\lambda _ { i } ^ { ( 1 ) } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 656, + 465, + 675 + ], + "score": 1.0, + "content": "must take on the same value", + "type": "text" + }, + { + "bbox": [ + 466, + 661, + 489, + 674 + ], + "score": 0.89, + "content": "\\scriptstyle { \\big ( } { \\frac { c } { d - 1 } } { \\big ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 656, + 506, + 675 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 671, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 119, + 685 + ], + "score": 1.0, + "content": "all", + "type": "text" + }, + { + "bbox": [ + 119, + 672, + 142, + 684 + ], + "score": 0.89, + "content": "i \\geq 2", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 671, + 209, + 685 + ], + "score": 1.0, + "content": ". The case when", + "type": "text" + }, + { + "bbox": [ + 209, + 673, + 235, + 684 + ], + "score": 0.9, + "content": "u \\leq v", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 671, + 505, + 685 + ], + "score": 1.0, + "content": "is similar. As a result, we have proved that at the optimal solution,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 684, + 166, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 684, + 166, + 695 + ], + "score": 1.0, + "content": "we must have:", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 713, + 386, + 731 + ], + "lines": [ + { + "bbox": [ + 221, + 713, + 386, + 731 + ], + "spans": [ + { + "bbox": [ + 221, + 713, + 386, + 731 + ], + "score": 0.69, + "content": "\\lambda _ { i } ^ { ( 0 ) } = \\lambda _ { j } ^ { ( 0 ) } \\mathrm { a n d } \\lambda _ { i } ^ { ( 1 ) } = \\lambda _ { j } ^ { ( 1 ) } , \\mathrm { f o r } i , j \\ge 2", + "type": "interline_equation", + "image_path": "c624dd0c01c5067fc20e00deddb8495d04b412abcb7df06ecc95d02e4dc89254.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 221, + 713, + 386, + 731 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "interline_equation", + "bbox": [ + 201, + 79, + 408, + 200 + ], + "lines": [ + { + "bbox": [ + 201, + 79, + 408, + 200 + ], + "spans": [ + { + "bbox": [ + 201, + 79, + 408, + 200 + ], + "score": 0.93, + "content": "\\begin{array} { r l } { \\mathrm { s u b j e c t ~ t o ~ } } & { p ( \\displaystyle \\sum _ { i = 1 } ^ { d } \\lambda _ { i } ^ { ( 1 ) } ) + ( 1 - p ) ( \\displaystyle \\sum _ { i = 1 } ^ { d } \\lambda _ { i } ^ { ( 0 ) } ) \\le P } \\\\ & { - \\lambda _ { i } ^ { ( 1 ) } \\le 0 , \\forall i \\in [ d ] } \\\\ & { - \\lambda _ { i } ^ { ( 0 ) } \\le 0 , \\forall i \\in [ d ] . } \\\\ & { \\lambda _ { i } ^ { ( 1 ) } \\ge \\lambda _ { j } ^ { ( 1 ) } , \\forall i < j . } \\\\ & { \\lambda _ { i } ^ { ( 0 ) } \\ge \\lambda _ { j } ^ { ( 0 ) } , \\forall i < j . } \\\\ & { Q \\ \\mathrm { o r t h o g o n a l } . } \\end{array}", + "type": "interline_equation", + "image_path": "658a524177670a85c5c1bc827631eb1439bbe9a4b9abd4f734f229d947058e98.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 201, + 79, + 408, + 99.16666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 201, + 99.16666666666667, + 408, + 119.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 201, + 119.33333333333334, + 408, + 139.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 201, + 139.5, + 408, + 159.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 201, + 159.66666666666666, + 408, + 179.83333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 201, + 179.83333333333331, + 408, + 199.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 227, + 506, + 264 + ], + "lines": [ + { + "bbox": [ + 104, + 226, + 507, + 245 + ], + "spans": [ + { + "bbox": [ + 104, + 226, + 197, + 245 + ], + "score": 1.0, + "content": "For any fixed feasible", + "type": "text" + }, + { + "bbox": [ + 198, + 227, + 255, + 243 + ], + "score": 0.94, + "content": "\\{ \\lambda _ { i } ^ { ( 1 ) } \\} , \\{ \\lambda _ { i } ^ { ( 0 ) } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 226, + 431, + 245 + ], + "score": 1.0, + "content": ", we see that the corresponding minimizing", + "type": "text" + }, + { + "bbox": [ + 432, + 231, + 441, + 241 + ], + "score": 0.85, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 226, + 507, + 245 + ], + "score": 1.0, + "content": "will set its first", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 104, + 240, + 292, + 254 + ], + "score": 1.0, + "content": "row to be the unit vector in the direction of", + "type": "text" + }, + { + "bbox": [ + 293, + 241, + 307, + 253 + ], + "score": 0.87, + "content": "\\Delta g", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 240, + 421, + 254 + ], + "score": 1.0, + "content": ". Thus by first minimizing", + "type": "text" + }, + { + "bbox": [ + 422, + 242, + 431, + 253 + ], + "score": 0.84, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 240, + 506, + 254 + ], + "score": 1.0, + "content": ", the optimization", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 252, + 191, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 252, + 191, + 265 + ], + "score": 1.0, + "content": "objective reduces to:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 104, + 226, + 507, + 265 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 284, + 460, + 354 + ], + "lines": [ + { + "bbox": [ + 150, + 284, + 460, + 354 + ], + "spans": [ + { + "bbox": [ + 150, + 284, + 460, + 354 + ], + "score": 0.93, + "content": "\\begin{array} { c c l } { \\displaystyle \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\displaystyle \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { i } ^ { ( 0 ) } + u } + \\frac { g } { \\lambda _ { 1 } ^ { ( 0 ) } + u } + \\frac { g } { \\lambda _ { 1 } ^ { ( 1 ) } + v } } \\\\ { = \\displaystyle \\sum _ { i = 2 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\displaystyle \\sum _ { i = 2 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { i } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } } \\end{array}", + "type": "interline_equation", + "image_path": "706550a04bb8352b793ee0ef90e1e490bae791f4828804effa9828257d1beb44.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 150, + 284, + 460, + 307.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 150, + 307.3333333333333, + 460, + 330.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 150, + 330.66666666666663, + 460, + 353.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 374, + 506, + 481 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 254, + 398 + ], + "score": 1.0, + "content": "We see that for the pair of variable", + "type": "text" + }, + { + "bbox": [ + 255, + 376, + 300, + 391 + ], + "score": 0.82, + "content": "( \\lambda _ { i } ^ { ( 1 ) } , \\lambda _ { i } ^ { ( 0 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 378, + 335, + 390 + ], + "score": 0.6, + "content": "( i ~ \\geq ~ 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 368, + 394, + 398 + ], + "score": 1.0, + "content": ", the function", + "type": "text" + }, + { + "bbox": [ + 394, + 374, + 462, + 396 + ], + "score": 0.95, + "content": "\\begin{array} { r } { \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { 1 } ^ { ( 0 ) } + u } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 370, + 506, + 400 + ], + "score": 1.0, + "content": "is strictly", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 393, + 508, + 412 + ], + "spans": [ + { + "bbox": [ + 104, + 393, + 228, + 412 + ], + "score": 1.0, + "content": "convex over the line segment", + "type": "text" + }, + { + "bbox": [ + 229, + 395, + 334, + 410 + ], + "score": 0.92, + "content": "p \\lambda _ { i } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { i } ^ { ( 0 ) } = c", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 393, + 420, + 412 + ], + "score": 1.0, + "content": "for any nonnegative", + "type": "text" + }, + { + "bbox": [ + 420, + 400, + 426, + 408 + ], + "score": 0.71, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 393, + 508, + 412 + ], + "score": 1.0, + "content": "and attains the the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 407, + 508, + 427 + ], + "spans": [ + { + "bbox": [ + 104, + 407, + 184, + 427 + ], + "score": 1.0, + "content": "minimum value at", + "type": "text" + }, + { + "bbox": [ + 184, + 409, + 224, + 424 + ], + "score": 0.94, + "content": "\\lambda _ { i } ^ { ( 1 ) } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 407, + 251, + 427 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 251, + 413, + 281, + 423 + ], + "score": 0.86, + "content": "u \\ : < \\ : v", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 407, + 300, + 427 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 300, + 410, + 340, + 424 + ], + "score": 0.92, + "content": "\\lambda _ { i } ^ { ( 0 ) } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 407, + 367, + 427 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 367, + 412, + 397, + 423 + ], + "score": 0.88, + "content": "u \\geq v", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 407, + 508, + 427 + ], + "score": 1.0, + "content": ". Suppose without loss of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 103, + 422, + 507, + 439 + ], + "spans": [ + { + "bbox": [ + 103, + 422, + 149, + 439 + ], + "score": 1.0, + "content": "generality", + "type": "text" + }, + { + "bbox": [ + 150, + 427, + 177, + 438 + ], + "score": 0.9, + "content": "u \\geq v", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 422, + 360, + 439 + ], + "score": 1.0, + "content": ", then for the optimal solution we must have", + "type": "text" + }, + { + "bbox": [ + 360, + 424, + 399, + 439 + ], + "score": 0.94, + "content": "\\lambda _ { i } ^ { ( 0 ) } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 422, + 428, + 439 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 428, + 426, + 454, + 438 + ], + "score": 0.89, + "content": "i \\geq 2", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 422, + 507, + 439 + ], + "score": 1.0, + "content": ". Under this", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 102, + 434, + 510, + 473 + ], + "spans": [ + { + "bbox": [ + 102, + 435, + 156, + 473 + ], + "score": 1.0, + "content": "Thus for all", + "type": "text" + }, + { + "bbox": [ + 103, + 434, + 321, + 455 + ], + "score": 1.0, + "content": "condition, we notice that the function m(x) = ux+v", + "type": "text" + }, + { + "bbox": [ + 156, + 451, + 186, + 471 + ], + "score": 0.94, + "content": "\\left\\{ \\lambda _ { i } ^ { \\left( 1 \\right) } \\right\\}", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 435, + 239, + 473 + ], + "score": 1.0, + "content": "that satisfies", + "type": "text" + }, + { + "bbox": [ + 240, + 452, + 301, + 468 + ], + "score": 0.93, + "content": "\\Sigma _ { i = 2 } ^ { d } \\lambda _ { i } ^ { ( 1 ) } = c", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 435, + 396, + 473 + ], + "score": 1.0, + "content": "is strictly c nonnegative", + "type": "text" + }, + { + "bbox": [ + 396, + 458, + 401, + 465 + ], + "score": 0.7, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 435, + 510, + 473 + ], + "score": 1.0, + "content": "vex on the positive reals., by Jensen inequality, we", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 467, + 129, + 483 + ], + "spans": [ + { + "bbox": [ + 104, + 467, + 129, + 483 + ], + "score": 1.0, + "content": "have", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5, + "bbox_fs": [ + 102, + 368, + 510, + 483 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 499, + 382, + 632 + ], + "lines": [ + { + "bbox": [ + 225, + 499, + 382, + 632 + ], + "spans": [ + { + "bbox": [ + 225, + 499, + 382, + 632 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle \\frac { 1 } { d - 1 } \\sum _ { i = 2 } ^ { d } \\left( \\frac { u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { u } \\right) } \\\\ { \\displaystyle = \\frac { 1 } { d - 1 } \\sum _ { i = 2 } ^ { d } m ( \\lambda _ { i } ^ { ( 1 ) } ) } \\\\ { \\displaystyle \\geq m \\left( \\frac { 1 } { d - 1 } \\sum _ { i = 2 } ^ { d } \\lambda _ { i } ^ { ( 1 ) } \\right) } \\\\ { \\displaystyle = m \\left( \\frac { c } { d - 1 } \\right) . } \\end{array}", + "type": "interline_equation", + "image_path": "16c9236733afaaf4e375a5a8e5a3607ba4a0d5d0e2a6b8857db7936e46f5e51a.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 225, + 499, + 382, + 518.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 225, + 518.0, + 382, + 537.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 225, + 537.0, + 382, + 556.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 225, + 556.0, + 382, + 575.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 225, + 575.0, + 382, + 594.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 225, + 594.0, + 382, + 613.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 225, + 613.0, + 382, + 632.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 658, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 104, + 656, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 104, + 656, + 322, + 675 + ], + "score": 1.0, + "content": "From this, we see that the optimal solution’s variables", + "type": "text" + }, + { + "bbox": [ + 323, + 658, + 350, + 673 + ], + "score": 0.94, + "content": "\\{ \\lambda _ { i } ^ { ( 1 ) } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 656, + 465, + 675 + ], + "score": 1.0, + "content": "must take on the same value", + "type": "text" + }, + { + "bbox": [ + 466, + 661, + 489, + 674 + ], + "score": 0.89, + "content": "\\scriptstyle { \\big ( } { \\frac { c } { d - 1 } } { \\big ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 656, + 506, + 675 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 671, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 119, + 685 + ], + "score": 1.0, + "content": "all", + "type": "text" + }, + { + "bbox": [ + 119, + 672, + 142, + 684 + ], + "score": 0.89, + "content": "i \\geq 2", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 671, + 209, + 685 + ], + "score": 1.0, + "content": ". The case when", + "type": "text" + }, + { + "bbox": [ + 209, + 673, + 235, + 684 + ], + "score": 0.9, + "content": "u \\leq v", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 671, + 505, + 685 + ], + "score": 1.0, + "content": "is similar. As a result, we have proved that at the optimal solution,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 684, + 166, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 684, + 166, + 695 + ], + "score": 1.0, + "content": "we must have:", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 104, + 656, + 506, + 695 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 713, + 386, + 731 + ], + "lines": [ + { + "bbox": [ + 221, + 713, + 386, + 731 + ], + "spans": [ + { + "bbox": [ + 221, + 713, + 386, + 731 + ], + "score": 0.69, + "content": "\\lambda _ { i } ^ { ( 0 ) } = \\lambda _ { j } ^ { ( 0 ) } \\mathrm { a n d } \\lambda _ { i } ^ { ( 1 ) } = \\lambda _ { j } ^ { ( 1 ) } , \\mathrm { f o r } i , j \\ge 2", + "type": "interline_equation", + "image_path": "c624dd0c01c5067fc20e00deddb8495d04b412abcb7df06ecc95d02e4dc89254.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 221, + 713, + 386, + 731 + ], + "spans": [], + "index": 28 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 505, + 122 + ], + "lines": [ + { + "bbox": [ + 101, + 77, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 101, + 77, + 374, + 107 + ], + "score": 1.0, + "content": "Hence, the optimization problem over the 2d variables nλ(1)i odi=1", + "type": "text" + }, + { + "bbox": [ + 330, + 80, + 429, + 103 + ], + "score": 0.92, + "content": "\\left\\{ \\lambda _ { i } ^ { ( 1 ) } \\right\\} _ { i = 1 } ^ { d } \\bigcup \\left\\{ \\lambda _ { i } ^ { ( 0 ) } \\right\\} _ { i = 1 } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 84, + 505, + 99 + ], + "score": 1.0, + "content": "can be reduced to", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 100, + 92, + 401, + 127 + ], + "spans": [ + { + "bbox": [ + 100, + 92, + 299, + 127 + ], + "score": 1.0, + "content": "an optimization problem over the four variables", + "type": "text" + }, + { + "bbox": [ + 299, + 102, + 393, + 122 + ], + "score": 0.92, + "content": "\\left\\{ \\lambda _ { 1 } ^ { ( 1 ) } , \\lambda _ { 2 } ^ { ( 1 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 0 ) } \\right\\}", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 92, + 401, + 127 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 145, + 502, + 178 + ], + "lines": [ + { + "bbox": [ + 111, + 145, + 502, + 178 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 502, + 178 + ], + "score": 0.9, + "content": "\\operatorname* { m i n } _ { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 1 ) } , \\lambda _ { 2 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } \\left( d - 1 \\right) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u }", + "type": "interline_equation", + "image_path": "392eaafae911b71d9320531eb80e120dc504b366f393c8126bb2ca26ee47d57b.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 111, + 145, + 502, + 156.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 156.0, + 502, + 167.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 111, + 167.0, + 502, + 178.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 190, + 439, + 315 + ], + "lines": [ + { + "bbox": [ + 110, + 190, + 439, + 315 + ], + "spans": [ + { + "bbox": [ + 110, + 190, + 439, + 315 + ], + "score": 0.92, + "content": "\\begin{array} { r l } { \\mathrm { s u b j e c t ~ t o ~ } } & { \\ p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) ( \\lambda _ { 2 } ^ { ( 0 ) } ) \\leq P } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 0 } ^ { ( 0 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 2 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 2 } ^ { ( 0 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 1 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 . } \\end{array}", + "type": "interline_equation", + "image_path": "626e8277767b3aa353997a4442dc09040ee7ea6301db2958964b53d95f556601.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 110, + 190, + 439, + 231.66666666666666 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 110, + 231.66666666666666, + 439, + 273.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 110, + 273.3333333333333, + 439, + 315.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 343, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 106, + 338, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 338, + 351, + 367 + ], + "score": 1.0, + "content": "Given the optimal solution to the above 4-variable problem", + "type": "text" + }, + { + "bbox": [ + 351, + 344, + 455, + 359 + ], + "score": 0.89, + "content": "( \\lambda _ { 1 } ^ { ( 0 ) * } , \\lambda _ { 2 } ^ { ( 0 ) * } , \\lambda _ { 1 } ^ { ( 1 ) * } , \\lambda _ { 2 } ^ { ( 1 ) * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 338, + 505, + 367 + ], + "score": 1.0, + "content": ", we can set", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 357, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 116, + 371 + ], + "score": 0.83, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 359, + 358, + 373 + ], + "score": 1.0, + "content": "to be any orthogonal matrix whose first row is the vector", + "type": "text" + }, + { + "bbox": [ + 354, + 359, + 392, + 380 + ], + "score": 1.0, + "content": "k∆gk2 .", + "type": "text" + }, + { + "bbox": [ + 388, + 357, + 506, + 375 + ], + "score": 1.0, + "content": "Plugging this back into the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 373, + 372, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 162, + 385 + ], + "score": 1.0, + "content": "expression of", + "type": "text" + }, + { + "bbox": [ + 163, + 373, + 175, + 384 + ], + "score": 0.9, + "content": "\\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 373, + 193, + 385 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 373, + 206, + 384 + ], + "score": 0.88, + "content": "\\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 373, + 372, + 385 + ], + "score": 1.0, + "content": "in Equation (14) gives us the final result.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 390, + 217, + 402 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 218, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 218, + 403 + ], + "score": 1.0, + "content": "Thus the proof is complete.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 467, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 243, + 482 + ], + "score": 1.0, + "content": "Remark (Interpreting the optimal", + "type": "text" + }, + { + "bbox": [ + 243, + 469, + 256, + 480 + ], + "score": 0.89, + "content": "\\Sigma _ { 1 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 468, + 276, + 482 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 276, + 469, + 289, + 480 + ], + "score": 0.88, + "content": "\\Sigma _ { 0 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 468, + 506, + 482 + ], + "score": 1.0, + "content": "). From the form of the optimal solution in (12), we", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "see that the optimal covariance matrices are both linear combinations of two terms: a rank one", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 135, + 502 + ], + "score": 1.0, + "content": "matrix", + "type": "text" + }, + { + "bbox": [ + 136, + 489, + 185, + 502 + ], + "score": 0.93, + "content": "( \\Delta g ) ( \\dot { \\Delta } g ) ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 490, + 281, + 502 + ], + "score": 1.0, + "content": "and the identity matrix", + "type": "text" + }, + { + "bbox": [ + 281, + 491, + 291, + 501 + ], + "score": 0.88, + "content": "I _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 490, + 505, + 502 + ], + "score": 1.0, + "content": ". Because a zero-mean Gaussian random vector with", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 185, + 513 + ], + "score": 1.0, + "content": "convariance matrix", + "type": "text" + }, + { + "bbox": [ + 185, + 501, + 220, + 513 + ], + "score": 0.92, + "content": "( A + B )", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "can be constructed as the sum of two independent zero-mean Gaussian", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 277, + 524 + ], + "score": 1.0, + "content": "random vectors with covariance matrices", + "type": "text" + }, + { + "bbox": [ + 277, + 512, + 286, + 522 + ], + "score": 0.78, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 511, + 306, + 524 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 307, + 512, + 316, + 522 + ], + "score": 0.79, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 511, + 505, + 524 + ], + "score": 1.0, + "content": "respectively, we see that the optimal additive", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 522, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 104, + 522, + 204, + 537 + ], + "score": 1.0, + "content": "noise random variables", + "type": "text" + }, + { + "bbox": [ + 205, + 523, + 222, + 535 + ], + "score": 0.91, + "content": "\\eta ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 522, + 242, + 537 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 242, + 523, + 259, + 536 + ], + "score": 0.91, + "content": "\\eta ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 522, + 506, + 537 + ], + "score": 1.0, + "content": "each consist of two independent components: one random", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "component lies along the line that connects the positive and negative gradient mean vectors (whose", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 545, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 255, + 559 + ], + "score": 1.0, + "content": "covariance matrix is proportional to", + "type": "text" + }, + { + "bbox": [ + 255, + 545, + 291, + 558 + ], + "score": 0.91, + "content": "\\Delta g \\Delta g ^ { \\top } )", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "); the other component is sampled from an isotropic", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "Gaussian. The use of the first random directional component and the fact that the isotropic Gaussian", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "score": 1.0, + "content": "component have different variance scaling for the positive and negative class clearly distinguishes", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 578, + 322, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 322, + 592 + ], + "score": 1.0, + "content": "Marvell from the isotropic Gaussian baseline iso.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 612, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 506, + 626 + ], + "score": 1.0, + "content": "Remark (How to solve). By analyzing the KKT condition of this four variable problem we can find", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 102, + 619, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 102, + 619, + 347, + 644 + ], + "score": 1.0, + "content": "that the optimal solution must exactly lie on the hyperplane", + "type": "text" + }, + { + "bbox": [ + 347, + 624, + 505, + 639 + ], + "score": 0.91, + "content": "p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\bar { \\lambda _ { 2 } ^ { ( 1 ) } } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } +", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 632, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 107, + 638, + 207, + 653 + ], + "score": 0.91, + "content": "( 1 - p ) ( d - 1 ) \\lambda _ { 2 } ^ { ( 0 ) } = P", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 632, + 409, + 660 + ], + "score": 1.0, + "content": ". From the proof above we additionally know that", + "type": "text" + }, + { + "bbox": [ + 409, + 639, + 450, + 653 + ], + "score": 0.94, + "content": "\\lambda _ { 2 } ^ { ( 1 ) * } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 632, + 460, + 660 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 460, + 642, + 487, + 651 + ], + "score": 0.83, + "content": "u < v", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 632, + 506, + 660 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 651, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 652, + 149, + 667 + ], + "score": 0.91, + "content": "\\lambda _ { 2 } ^ { ( 0 ) * } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 651, + 159, + 669 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 160, + 655, + 187, + 666 + ], + "score": 0.88, + "content": "u \\geq v", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 651, + 506, + 669 + ], + "score": 1.0, + "content": ". Thus the problem is further reduced to a 3-variable problem. If we consider", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 666, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 506, + 679 + ], + "score": 1.0, + "content": "keeping one of the 3 remaining variables’ values fixed, then the feasible region becomes restricted", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 677, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 689 + ], + "score": 1.0, + "content": "to a line segment. We can simply perform a line search optimization of the convex objective and", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "find the optimal values for the remaining two free variables. We can then alternate over which one", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "of the three variables to fix and optimize over the other two. This optimization procedure would", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "eventually converge and give us the optimal solution. This approach mimics the Sequential Minimal", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 720, + 446, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 446, + 733 + ], + "score": 1.0, + "content": "Optimization technique used to solve the dual of the support vector machine (SVM).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 27.5 + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "17", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 390, + 505, + 401 + ], + "lines": [ + { + "bbox": [ + 496, + 392, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 496, + 392, + 505, + 401 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 505, + 122 + ], + "lines": [ + { + "bbox": [ + 101, + 77, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 101, + 77, + 374, + 107 + ], + "score": 1.0, + "content": "Hence, the optimization problem over the 2d variables nλ(1)i odi=1", + "type": "text" + }, + { + "bbox": [ + 330, + 80, + 429, + 103 + ], + "score": 0.92, + "content": "\\left\\{ \\lambda _ { i } ^ { ( 1 ) } \\right\\} _ { i = 1 } ^ { d } \\bigcup \\left\\{ \\lambda _ { i } ^ { ( 0 ) } \\right\\} _ { i = 1 } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 84, + 505, + 99 + ], + "score": 1.0, + "content": "can be reduced to", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 100, + 92, + 401, + 127 + ], + "spans": [ + { + "bbox": [ + 100, + 92, + 299, + 127 + ], + "score": 1.0, + "content": "an optimization problem over the four variables", + "type": "text" + }, + { + "bbox": [ + 299, + 102, + 393, + 122 + ], + "score": 0.92, + "content": "\\left\\{ \\lambda _ { 1 } ^ { ( 1 ) } , \\lambda _ { 2 } ^ { ( 1 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 0 ) } \\right\\}", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 92, + 401, + 127 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 100, + 77, + 505, + 127 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 145, + 502, + 178 + ], + "lines": [ + { + "bbox": [ + 111, + 145, + 502, + 178 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 502, + 178 + ], + "score": 0.9, + "content": "\\operatorname* { m i n } _ { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 1 ) } , \\lambda _ { 2 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } \\left( d - 1 \\right) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u }", + "type": "interline_equation", + "image_path": "392eaafae911b71d9320531eb80e120dc504b366f393c8126bb2ca26ee47d57b.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 111, + 145, + 502, + 156.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 156.0, + 502, + 167.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 111, + 167.0, + 502, + 178.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 110, + 190, + 439, + 315 + ], + "lines": [ + { + "bbox": [ + 110, + 190, + 439, + 315 + ], + "spans": [ + { + "bbox": [ + 110, + 190, + 439, + 315 + ], + "score": 0.92, + "content": "\\begin{array} { r l } { \\mathrm { s u b j e c t ~ t o ~ } } & { \\ p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) ( \\lambda _ { 2 } ^ { ( 0 ) } ) \\leq P } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 0 } ^ { ( 0 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 2 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 2 } ^ { ( 0 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 1 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 . } \\end{array}", + "type": "interline_equation", + "image_path": "626e8277767b3aa353997a4442dc09040ee7ea6301db2958964b53d95f556601.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 110, + 190, + 439, + 231.66666666666666 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 110, + 231.66666666666666, + 439, + 273.3333333333333 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 110, + 273.3333333333333, + 439, + 315.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 343, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 106, + 338, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 338, + 351, + 367 + ], + "score": 1.0, + "content": "Given the optimal solution to the above 4-variable problem", + "type": "text" + }, + { + "bbox": [ + 351, + 344, + 455, + 359 + ], + "score": 0.89, + "content": "( \\lambda _ { 1 } ^ { ( 0 ) * } , \\lambda _ { 2 } ^ { ( 0 ) * } , \\lambda _ { 1 } ^ { ( 1 ) * } , \\lambda _ { 2 } ^ { ( 1 ) * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 338, + 505, + 367 + ], + "score": 1.0, + "content": ", we can set", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 357, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 116, + 371 + ], + "score": 0.83, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 359, + 358, + 373 + ], + "score": 1.0, + "content": "to be any orthogonal matrix whose first row is the vector", + "type": "text" + }, + { + "bbox": [ + 354, + 359, + 392, + 380 + ], + "score": 1.0, + "content": "k∆gk2 .", + "type": "text" + }, + { + "bbox": [ + 388, + 357, + 506, + 375 + ], + "score": 1.0, + "content": "Plugging this back into the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 373, + 372, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 162, + 385 + ], + "score": 1.0, + "content": "expression of", + "type": "text" + }, + { + "bbox": [ + 163, + 373, + 175, + 384 + ], + "score": 0.9, + "content": "\\Sigma _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 373, + 193, + 385 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 373, + 206, + 384 + ], + "score": 0.88, + "content": "\\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 373, + 372, + 385 + ], + "score": 1.0, + "content": "in Equation (14) gives us the final result.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 338, + 506, + 385 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 390, + 217, + 402 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 218, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 218, + 403 + ], + "score": 1.0, + "content": "Thus the proof is complete.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 388, + 218, + 403 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 467, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 243, + 482 + ], + "score": 1.0, + "content": "Remark (Interpreting the optimal", + "type": "text" + }, + { + "bbox": [ + 243, + 469, + 256, + 480 + ], + "score": 0.89, + "content": "\\Sigma _ { 1 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 468, + 276, + 482 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 276, + 469, + 289, + 480 + ], + "score": 0.88, + "content": "\\Sigma _ { 0 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 468, + 506, + 482 + ], + "score": 1.0, + "content": "). From the form of the optimal solution in (12), we", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "see that the optimal covariance matrices are both linear combinations of two terms: a rank one", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 135, + 502 + ], + "score": 1.0, + "content": "matrix", + "type": "text" + }, + { + "bbox": [ + 136, + 489, + 185, + 502 + ], + "score": 0.93, + "content": "( \\Delta g ) ( \\dot { \\Delta } g ) ^ { \\top }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 490, + 281, + 502 + ], + "score": 1.0, + "content": "and the identity matrix", + "type": "text" + }, + { + "bbox": [ + 281, + 491, + 291, + 501 + ], + "score": 0.88, + "content": "I _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 490, + 505, + 502 + ], + "score": 1.0, + "content": ". Because a zero-mean Gaussian random vector with", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 185, + 513 + ], + "score": 1.0, + "content": "convariance matrix", + "type": "text" + }, + { + "bbox": [ + 185, + 501, + 220, + 513 + ], + "score": 0.92, + "content": "( A + B )", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "can be constructed as the sum of two independent zero-mean Gaussian", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 277, + 524 + ], + "score": 1.0, + "content": "random vectors with covariance matrices", + "type": "text" + }, + { + "bbox": [ + 277, + 512, + 286, + 522 + ], + "score": 0.78, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 511, + 306, + 524 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 307, + 512, + 316, + 522 + ], + "score": 0.79, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 511, + 505, + 524 + ], + "score": 1.0, + "content": "respectively, we see that the optimal additive", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 522, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 104, + 522, + 204, + 537 + ], + "score": 1.0, + "content": "noise random variables", + "type": "text" + }, + { + "bbox": [ + 205, + 523, + 222, + 535 + ], + "score": 0.91, + "content": "\\eta ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 522, + 242, + 537 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 242, + 523, + 259, + 536 + ], + "score": 0.91, + "content": "\\eta ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 522, + 506, + 537 + ], + "score": 1.0, + "content": "each consist of two independent components: one random", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "component lies along the line that connects the positive and negative gradient mean vectors (whose", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 545, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 255, + 559 + ], + "score": 1.0, + "content": "covariance matrix is proportional to", + "type": "text" + }, + { + "bbox": [ + 255, + 545, + 291, + 558 + ], + "score": 0.91, + "content": "\\Delta g \\Delta g ^ { \\top } )", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "); the other component is sampled from an isotropic", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "Gaussian. The use of the first random directional component and the fact that the isotropic Gaussian", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "score": 1.0, + "content": "component have different variance scaling for the positive and negative class clearly distinguishes", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 578, + 322, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 322, + 592 + ], + "score": 1.0, + "content": "Marvell from the isotropic Gaussian baseline iso.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 17, + "bbox_fs": [ + 104, + 468, + 506, + 592 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 612, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 506, + 626 + ], + "score": 1.0, + "content": "Remark (How to solve). By analyzing the KKT condition of this four variable problem we can find", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 102, + 619, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 102, + 619, + 347, + 644 + ], + "score": 1.0, + "content": "that the optimal solution must exactly lie on the hyperplane", + "type": "text" + }, + { + "bbox": [ + 347, + 624, + 505, + 639 + ], + "score": 0.91, + "content": "p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\bar { \\lambda _ { 2 } ^ { ( 1 ) } } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } +", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 632, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 107, + 638, + 207, + 653 + ], + "score": 0.91, + "content": "( 1 - p ) ( d - 1 ) \\lambda _ { 2 } ^ { ( 0 ) } = P", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 632, + 409, + 660 + ], + "score": 1.0, + "content": ". From the proof above we additionally know that", + "type": "text" + }, + { + "bbox": [ + 409, + 639, + 450, + 653 + ], + "score": 0.94, + "content": "\\lambda _ { 2 } ^ { ( 1 ) * } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 632, + 460, + 660 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 460, + 642, + 487, + 651 + ], + "score": 0.83, + "content": "u < v", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 632, + 506, + 660 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 651, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 652, + 149, + 667 + ], + "score": 0.91, + "content": "\\lambda _ { 2 } ^ { ( 0 ) * } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 651, + 159, + 669 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 160, + 655, + 187, + 666 + ], + "score": 0.88, + "content": "u \\geq v", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 651, + 506, + 669 + ], + "score": 1.0, + "content": ". Thus the problem is further reduced to a 3-variable problem. If we consider", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 666, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 506, + 679 + ], + "score": 1.0, + "content": "keeping one of the 3 remaining variables’ values fixed, then the feasible region becomes restricted", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 677, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 689 + ], + "score": 1.0, + "content": "to a line segment. We can simply perform a line search optimization of the convex objective and", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "find the optimal values for the remaining two free variables. We can then alternate over which one", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "of the three variables to fix and optimize over the other two. This optimization procedure would", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "eventually converge and give us the optimal solution. This approach mimics the Sequential Minimal", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 720, + 446, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 446, + 733 + ], + "score": 1.0, + "content": "Optimization technique used to solve the dual of the support vector machine (SVM).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 27.5, + "bbox_fs": [ + 102, + 613, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 290, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 291, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 291, + 95 + ], + "score": 1.0, + "content": "A.5 MA R V E L L ALGORITHM DESCRIPTION", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 102, + 504, + 126 + ], + "lines": [ + { + "bbox": [ + 106, + 101, + 506, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 101, + 137, + 116 + ], + "score": 1.0, + "content": "We use", + "type": "text" + }, + { + "bbox": [ + 137, + 101, + 168, + 113 + ], + "score": 0.94, + "content": "\\mathbf { 1 } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 101, + 387, + 116 + ], + "score": 1.0, + "content": "to denote the vector with 1 in each coordinate. We use", + "type": "text" + }, + { + "bbox": [ + 387, + 103, + 403, + 115 + ], + "score": 0.89, + "content": "g [ j ]", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 101, + 459, + 116 + ], + "score": 1.0, + "content": "to denote the", + "type": "text" + }, + { + "bbox": [ + 459, + 104, + 464, + 115 + ], + "score": 0.82, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 101, + 506, + 116 + ], + "score": 1.0, + "content": "-th row of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 112, + 372, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 149, + 128 + ], + "score": 1.0, + "content": "the matrix", + "type": "text" + }, + { + "bbox": [ + 150, + 116, + 156, + 126 + ], + "score": 0.81, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 112, + 174, + 128 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 174, + 114, + 191, + 126 + ], + "score": 0.9, + "content": "y [ j ]", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 112, + 246, + 128 + ], + "score": 1.0, + "content": "to denote the", + "type": "text" + }, + { + "bbox": [ + 246, + 115, + 252, + 126 + ], + "score": 0.82, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 112, + 361, + 128 + ], + "score": 1.0, + "content": "-th coordinate of the vector", + "type": "text" + }, + { + "bbox": [ + 362, + 117, + 368, + 126 + ], + "score": 0.77, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 112, + 372, + 128 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 107, + 136, + 249, + 147 + ], + "lines": [ + { + "bbox": [ + 107, + 136, + 249, + 147 + ], + "spans": [], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 96, + 140, + 491, + 503 + ], + "lines": [ + { + "bbox": [ + 106, + 135, + 250, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 250, + 149 + ], + "score": 1.0, + "content": "Algorithm 1: Marvell algorithm", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 146, + 358, + 162 + ], + "spans": [ + { + "bbox": [ + 104, + 146, + 187, + 162 + ], + "score": 1.0, + "content": "input : g ∈ RB×d,", + "type": "text" + }, + { + "bbox": [ + 195, + 149, + 223, + 162 + ], + "score": 1.0, + "content": "a size-", + "type": "text" + }, + { + "bbox": [ + 223, + 150, + 232, + 159 + ], + "score": 0.76, + "content": ". B", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 149, + 358, + 162 + ], + "score": 1.0, + "content": "batch of unperturbed gradients", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 140, + 157, + 358, + 176 + ], + "spans": [ + { + "bbox": [ + 140, + 161, + 190, + 174 + ], + "score": 0.88, + "content": "y \\in \\left\\{ 0 , 1 \\right\\} ^ { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 157, + 358, + 176 + ], + "score": 1.0, + "content": ", the label for each example in the batch", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 173, + 401, + 186 + ], + "spans": [ + { + "bbox": [ + 141, + 176, + 146, + 183 + ], + "score": 0.35, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 174, + 151, + 186 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 196, + 173, + 392, + 185 + ], + "score": 1.0, + "content": "privacy hyperparameter for the power constraint", + "type": "text" + }, + { + "bbox": [ + 392, + 174, + 401, + 183 + ], + "score": 0.7, + "content": "P", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 183, + 313, + 198 + ], + "spans": [ + { + "bbox": [ + 104, + 183, + 189, + 198 + ], + "score": 1.0, + "content": "output: g˜ ∈ RB×d,", + "type": "text" + }, + { + "bbox": [ + 195, + 184, + 313, + 198 + ], + "score": 1.0, + "content": "batch of perturbed gradients", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 196, + 444, + 210 + ], + "spans": [ + { + "bbox": [ + 107, + 196, + 444, + 210 + ], + "score": 1.0, + "content": "// Step 1: estimate the optimization constants from the batch", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 123, + 208, + 401, + 220 + ], + "spans": [ + { + "bbox": [ + 123, + 208, + 401, + 220 + ], + "score": 1.0, + "content": "gradients using maximum likelihood estimation (MLE)", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 95, + 216, + 490, + 235 + ], + "spans": [ + { + "bbox": [ + 95, + 223, + 106, + 234 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 106, + 219, + 145, + 235 + ], + "score": 0.9, + "content": "p \\gets \\frac { \\mathbf { 1 } ^ { T } y } { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 216, + 153, + 232 + ], + "score": 1.0, + "content": "1 T y ;", + "type": "text" + }, + { + "bbox": [ + 361, + 222, + 490, + 235 + ], + "score": 1.0, + "content": "/* positive fraction */", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 93, + 232, + 491, + 250 + ], + "spans": [ + { + "bbox": [ + 93, + 232, + 131, + 249 + ], + "score": 1.0, + "content": "2 g¯(1)", + "type": "text" + }, + { + "bbox": [ + 107, + 235, + 174, + 250 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\bar { g } ^ { ( 1 ) } \\frac { 1 } { \\mathbf { 1 } ^ { T } y } g ^ { T } y } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 237, + 180, + 248 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 383, + 236, + 397, + 247 + ], + "score": 0.34, + "content": "/ \\star", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 235, + 491, + 249 + ], + "score": 1.0, + "content": "positive mean */", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 94, + 247, + 490, + 264 + ], + "spans": [ + { + "bbox": [ + 94, + 247, + 106, + 264 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 106, + 250, + 212, + 264 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\bar { g } ^ { ( 0 ) } \\frac { 1 } { B - \\mathbf { 1 } ^ { T } y } g ^ { T } ( \\mathbf { 1 } - y ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 248, + 219, + 264 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 383, + 250, + 490, + 263 + ], + "score": 1.0, + "content": "/* negative mean */", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 94, + 260, + 184, + 280 + ], + "spans": [ + { + "bbox": [ + 94, + 260, + 106, + 280 + ], + "score": 1.0, + "content": "4", + "type": "text" + }, + { + "bbox": [ + 106, + 264, + 177, + 277 + ], + "score": 0.85, + "content": "\\Delta g = \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 260, + 184, + 280 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 96, + 275, + 490, + 298 + ], + "spans": [ + { + "bbox": [ + 96, + 275, + 105, + 298 + ], + "score": 1.0, + "content": "5", + "type": "text" + }, + { + "bbox": [ + 106, + 277, + 270, + 294 + ], + "score": 0.73, + "content": "\\begin{array} { r } { v \\frac { 1 } { d \\cdot ( \\mathbf { 1 } ^ { T } y ) } \\sum _ { j = 1 } ^ { B } y [ j ] \\cdot g [ j ] - \\bar { g } ^ { ( 1 ) } _ { 2 } ^ { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 275, + 276, + 298 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 345, + 280, + 490, + 293 + ], + "score": 1.0, + "content": "/* positive convariance */", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 91, + 291, + 489, + 315 + ], + "spans": [ + { + "bbox": [ + 91, + 291, + 105, + 315 + ], + "score": 1.0, + "content": "6", + "type": "text" + }, + { + "bbox": [ + 105, + 296, + 306, + 312 + ], + "score": 0.73, + "content": "\\begin{array} { r } { u \\frac { 1 } { d \\cdot ( B - \\mathbf { 1 } ^ { T } y ) } \\sum _ { j = 1 } ^ { B } ( 1 - y [ j ] ) \\cdot g [ j ] - \\bar { g } ^ { ( 0 ) } _ { 2 } ^ { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 291, + 315, + 315 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 345, + 297, + 477, + 310 + ], + "score": 1.0, + "content": "/* negative convariance", + "type": "text" + }, + { + "bbox": [ + 477, + 298, + 489, + 308 + ], + "score": 0.26, + "content": "\\star /", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 94, + 309, + 489, + 326 + ], + "spans": [ + { + "bbox": [ + 94, + 309, + 106, + 326 + ], + "score": 1.0, + "content": "7", + "type": "text" + }, + { + "bbox": [ + 106, + 312, + 172, + 324 + ], + "score": 0.7, + "content": "P \\gets s \\cdot \\| \\Delta g \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 309, + 179, + 326 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 285, + 310, + 477, + 325 + ], + "score": 1.0, + "content": "/* power constraint hyperparameter", + "type": "text" + }, + { + "bbox": [ + 477, + 313, + 489, + 322 + ], + "score": 0.32, + "content": "\\star /", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 321, + 359, + 335 + ], + "spans": [ + { + "bbox": [ + 104, + 321, + 165, + 335 + ], + "score": 1.0, + "content": "// Step 2:", + "type": "text" + }, + { + "bbox": [ + 172, + 324, + 359, + 335 + ], + "score": 1.0, + "content": "optimize the four-variable problem", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 96, + 334, + 164, + 345 + ], + "spans": [ + { + "bbox": [ + 96, + 334, + 115, + 345 + ], + "score": 1.0, + "content": "8 if", + "type": "text" + }, + { + "bbox": [ + 115, + 335, + 141, + 344 + ], + "score": 0.86, + "content": "u < v", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 334, + 164, + 345 + ], + "score": 1.0, + "content": "then", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 97, + 341, + 489, + 364 + ], + "spans": [ + { + "bbox": [ + 97, + 349, + 105, + 359 + ], + "score": 1.0, + "content": "9", + "type": "text" + }, + { + "bbox": [ + 122, + 345, + 162, + 360 + ], + "score": 0.64, + "content": "\\lambda _ { 2 } ^ { ( 1 ) } 0", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 341, + 171, + 364 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 298, + 347, + 466, + 359 + ], + "score": 1.0, + "content": "/* this variable is optimal at", + "type": "text" + }, + { + "bbox": [ + 466, + 348, + 489, + 358 + ], + "score": 0.6, + "content": "0 ~ \\star /", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 94, + 355, + 489, + 379 + ], + "spans": [ + { + "bbox": [ + 94, + 363, + 105, + 374 + ], + "score": 1.0, + "content": "10", + "type": "text" + }, + { + "bbox": [ + 118, + 355, + 308, + 379 + ], + "score": 1.0, + "content": "Randomly initialize the optimization variables", + "type": "text" + }, + { + "bbox": [ + 309, + 359, + 326, + 374 + ], + "score": 0.81, + "content": "\\lambda _ { 1 } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 355, + 330, + 379 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 330, + 359, + 348, + 374 + ], + "score": 0.65, + "content": "\\lambda _ { 1 } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 355, + 351, + 379 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 352, + 359, + 369, + 374 + ], + "score": 0.85, + "content": "\\lambda _ { 2 } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 355, + 489, + 379 + ], + "score": 1.0, + "content": "in the feasible region in (15).", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 92, + 374, + 126, + 386 + ], + "spans": [ + { + "bbox": [ + 92, + 374, + 126, + 386 + ], + "score": 1.0, + "content": "11 else", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 93, + 384, + 489, + 423 + ], + "spans": [ + { + "bbox": [ + 93, + 388, + 106, + 415 + ], + "score": 1.0, + "content": "12 13", + "type": "text" + }, + { + "bbox": [ + 115, + 384, + 120, + 423 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 121, + 384, + 166, + 401 + ], + "score": 0.81, + "content": "\\lambda _ { 2 } ^ { ( 0 ) } 0 \\ : ;", + "type": "inline_equation", + "image_path": "983a21a35c931c70d2225b6e46518a843d5009549160e42b646e0975a396d632.jpg" + }, + { + "bbox": [ + 166, + 384, + 292, + 423 + ], + "score": 1.0, + "content": "initialize the optimization varia", + "type": "text" + }, + { + "bbox": [ + 293, + 388, + 307, + 398 + ], + "score": 0.32, + "content": "/ \\star", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 384, + 477, + 423 + ], + "score": 1.0, + "content": "le is optimal at 0 in the feasible region in (1", + "type": "text" + }, + { + "bbox": [ + 477, + 389, + 489, + 398 + ], + "score": 0.35, + "content": "\\star /", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 310, + 399, + 370, + 415 + ], + "spans": [ + { + "bbox": [ + 310, + 399, + 370, + 415 + ], + "score": 0.36, + "content": "\\lambda _ { 1 } ^ { ( 1 ) } , \\lambda _ { 2 } ^ { ( 1 ) } , \\lambda _ { 1 } ^ { ( 0 ) }", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 92, + 415, + 126, + 426 + ], + "spans": [ + { + "bbox": [ + 92, + 415, + 126, + 426 + ], + "score": 1.0, + "content": "14 end", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 92, + 425, + 204, + 437 + ], + "spans": [ + { + "bbox": [ + 92, + 425, + 204, + 437 + ], + "score": 1.0, + "content": "15 while not converged do", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 93, + 435, + 335, + 448 + ], + "spans": [ + { + "bbox": [ + 93, + 437, + 105, + 448 + ], + "score": 1.0, + "content": "16", + "type": "text" + }, + { + "bbox": [ + 120, + 435, + 335, + 448 + ], + "score": 1.0, + "content": "Fix one of the newly updated optimization variables;", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 120, + 444, + 482, + 461 + ], + "spans": [ + { + "bbox": [ + 120, + 444, + 441, + 461 + ], + "score": 1.0, + "content": "// The 4-variable optimal solution lies on a hyperplane in", + "type": "text" + }, + { + "bbox": [ + 441, + 447, + 453, + 457 + ], + "score": 0.84, + "content": "\\mathbb { R } ^ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 444, + 482, + 461 + ], + "score": 1.0, + "content": "(see", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 139, + 459, + 439, + 470 + ], + "spans": [ + { + "bbox": [ + 139, + 459, + 439, + 470 + ], + "score": 1.0, + "content": "Appendix A.4 Remark) so fixing two variables gives us a", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 139, + 469, + 207, + 481 + ], + "spans": [ + { + "bbox": [ + 139, + 469, + 207, + 481 + ], + "score": 1.0, + "content": "line-segment", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 94, + 480, + 441, + 492 + ], + "spans": [ + { + "bbox": [ + 94, + 480, + 105, + 492 + ], + "score": 1.0, + "content": "17", + "type": "text" + }, + { + "bbox": [ + 121, + 480, + 441, + 492 + ], + "score": 1.0, + "content": "Update the remaining two optimization variables by performing 1-d line-search", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 125, + 491, + 423, + 504 + ], + "spans": [ + { + "bbox": [ + 125, + 491, + 423, + 504 + ], + "score": 1.0, + "content": "minimization of the convex function (15) while satisfying the constraints;", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 95, + 503, + 124, + 512 + ], + "lines": [ + { + "bbox": [ + 93, + 502, + 124, + 513 + ], + "spans": [ + { + "bbox": [ + 93, + 502, + 124, + 513 + ], + "score": 1.0, + "content": "18 end", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 94, + 510, + 492, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 512, + 385, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 166, + 526 + ], + "score": 1.0, + "content": "// Step 3:", + "type": "text" + }, + { + "bbox": [ + 172, + 513, + 385, + 525 + ], + "score": 1.0, + "content": "compute the optimal covariance matrices", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 91, + 521, + 265, + 549 + ], + "spans": [ + { + "bbox": [ + 91, + 526, + 106, + 545 + ], + "score": 1.0, + "content": "19", + "type": "text" + }, + { + "bbox": [ + 106, + 524, + 259, + 545 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\Sigma _ { 1 } ^ { * } \\frac { \\lambda _ { 1 } ^ { ( 1 ) } - \\lambda _ { 2 } ^ { ( 1 ) } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 1 ) } I _ { d } ; } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 521, + 265, + 549 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 91, + 541, + 264, + 569 + ], + "spans": [ + { + "bbox": [ + 91, + 546, + 106, + 565 + ], + "score": 1.0, + "content": "20", + "type": "text" + }, + { + "bbox": [ + 106, + 545, + 260, + 566 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\Sigma _ { 0 } ^ { * } \\frac { \\lambda _ { 1 } ^ { ( 0 ) } - \\lambda _ { 2 } ^ { ( 0 ) } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 0 ) } I _ { d } ; } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 541, + 264, + 569 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 563, + 289, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 167, + 578 + ], + "score": 1.0, + "content": "// Step 4:", + "type": "text" + }, + { + "bbox": [ + 171, + 565, + 289, + 577 + ], + "score": 1.0, + "content": "perturb the gradients", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 91, + 574, + 490, + 589 + ], + "spans": [ + { + "bbox": [ + 91, + 574, + 106, + 589 + ], + "score": 1.0, + "content": "21", + "type": "text" + }, + { + "bbox": [ + 106, + 576, + 151, + 587 + ], + "score": 0.88, + "content": "\\tilde { g } \\mathbf { 0 } _ { B \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 574, + 159, + 589 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 195, + 576, + 209, + 586 + ], + "score": 0.46, + "content": "/ \\star", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 574, + 490, + 588 + ], + "score": 1.0, + "content": "an empty matrix to store the perturbed gradients */", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 92, + 586, + 184, + 599 + ], + "spans": [ + { + "bbox": [ + 92, + 586, + 121, + 599 + ], + "score": 1.0, + "content": "22 for", + "type": "text" + }, + { + "bbox": [ + 122, + 588, + 148, + 598 + ], + "score": 0.89, + "content": "j 1", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 586, + 159, + 599 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 160, + 587, + 169, + 597 + ], + "score": 0.73, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 586, + 184, + 599 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 92, + 597, + 188, + 610 + ], + "spans": [ + { + "bbox": [ + 92, + 597, + 106, + 610 + ], + "score": 1.0, + "content": "23", + "type": "text" + }, + { + "bbox": [ + 120, + 597, + 130, + 609 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 131, + 598, + 165, + 609 + ], + "score": 0.89, + "content": "y [ j ] = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 597, + 188, + 609 + ], + "score": 1.0, + "content": "then", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 92, + 606, + 317, + 624 + ], + "spans": [ + { + "bbox": [ + 92, + 610, + 106, + 622 + ], + "score": 1.0, + "content": "24", + "type": "text" + }, + { + "bbox": [ + 133, + 606, + 135, + 624 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 135, + 609, + 213, + 622 + ], + "score": 0.76, + "content": "\\tilde { g } [ j ] g [ j ] + \\eta ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 606, + 243, + 624 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 244, + 609, + 313, + 622 + ], + "score": 0.92, + "content": "\\boldsymbol \\eta ^ { ( 1 ) } \\sim \\mathcal { N } ( \\mathbf 0 , \\Sigma _ { 1 } ^ { * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 606, + 317, + 624 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 92, + 620, + 194, + 633 + ], + "spans": [ + { + "bbox": [ + 92, + 621, + 106, + 633 + ], + "score": 1.0, + "content": "25", + "type": "text" + }, + { + "bbox": [ + 120, + 620, + 143, + 633 + ], + "score": 1.0, + "content": "else", + "type": "text" + }, + { + "bbox": [ + 143, + 622, + 194, + 632 + ], + "score": 0.76, + "content": "\\begin{array} { r l } { \\langle \\mathscr { I } / } & { { } y [ j ] = 0 } \\end{array}", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 92, + 630, + 317, + 647 + ], + "spans": [ + { + "bbox": [ + 92, + 633, + 106, + 645 + ], + "score": 1.0, + "content": "26", + "type": "text" + }, + { + "bbox": [ + 125, + 632, + 212, + 645 + ], + "score": 0.77, + "content": "\\begin{array} { r l } { \\vert } & { { } \\tilde { g } [ j ] g [ j ] + \\eta ^ { ( 0 ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 630, + 243, + 647 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 244, + 632, + 313, + 645 + ], + "score": 0.9, + "content": "\\boldsymbol \\eta ^ { ( 0 ) } \\sim \\mathcal { N } ( \\mathbf 0 , \\Sigma _ { 0 } ^ { * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 630, + 317, + 647 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 92, + 644, + 141, + 656 + ], + "spans": [ + { + "bbox": [ + 92, + 644, + 106, + 656 + ], + "score": 1.0, + "content": "27", + "type": "text" + }, + { + "bbox": [ + 119, + 644, + 141, + 655 + ], + "score": 1.0, + "content": "end", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 91, + 654, + 127, + 668 + ], + "spans": [ + { + "bbox": [ + 91, + 654, + 127, + 668 + ], + "score": 1.0, + "content": "28 end", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 39.5 + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 290, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 291, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 291, + 95 + ], + "score": 1.0, + "content": "A.5 MA R V E L L ALGORITHM DESCRIPTION", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 102, + 504, + 126 + ], + "lines": [ + { + "bbox": [ + 106, + 101, + 506, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 101, + 137, + 116 + ], + "score": 1.0, + "content": "We use", + "type": "text" + }, + { + "bbox": [ + 137, + 101, + 168, + 113 + ], + "score": 0.94, + "content": "\\mathbf { 1 } \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 101, + 387, + 116 + ], + "score": 1.0, + "content": "to denote the vector with 1 in each coordinate. We use", + "type": "text" + }, + { + "bbox": [ + 387, + 103, + 403, + 115 + ], + "score": 0.89, + "content": "g [ j ]", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 101, + 459, + 116 + ], + "score": 1.0, + "content": "to denote the", + "type": "text" + }, + { + "bbox": [ + 459, + 104, + 464, + 115 + ], + "score": 0.82, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 101, + 506, + 116 + ], + "score": 1.0, + "content": "-th row of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 112, + 372, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 149, + 128 + ], + "score": 1.0, + "content": "the matrix", + "type": "text" + }, + { + "bbox": [ + 150, + 116, + 156, + 126 + ], + "score": 0.81, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 112, + 174, + 128 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 174, + 114, + 191, + 126 + ], + "score": 0.9, + "content": "y [ j ]", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 112, + 246, + 128 + ], + "score": 1.0, + "content": "to denote the", + "type": "text" + }, + { + "bbox": [ + 246, + 115, + 252, + 126 + ], + "score": 0.82, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 112, + 361, + 128 + ], + "score": 1.0, + "content": "-th coordinate of the vector", + "type": "text" + }, + { + "bbox": [ + 362, + 117, + 368, + 126 + ], + "score": 0.77, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 112, + 372, + 128 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 101, + 506, + 128 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 136, + 249, + 147 + ], + "lines": [ + { + "bbox": [ + 107, + 136, + 249, + 147 + ], + "spans": [], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 107, + 136, + 249, + 147 + ] + }, + { + "type": "list", + "bbox": [ + 96, + 140, + 491, + 503 + ], + "lines": [ + { + "bbox": [ + 106, + 135, + 250, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 250, + 149 + ], + "score": 1.0, + "content": "Algorithm 1: Marvell algorithm", + "type": "text" + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 146, + 358, + 162 + ], + "spans": [ + { + "bbox": [ + 104, + 146, + 187, + 162 + ], + "score": 1.0, + "content": "input : g ∈ RB×d,", + "type": "text" + }, + { + "bbox": [ + 195, + 149, + 223, + 162 + ], + "score": 1.0, + "content": "a size-", + "type": "text" + }, + { + "bbox": [ + 223, + 150, + 232, + 159 + ], + "score": 0.76, + "content": ". B", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 149, + 358, + 162 + ], + "score": 1.0, + "content": "batch of unperturbed gradients", + "type": "text" + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 140, + 157, + 358, + 176 + ], + "spans": [ + { + "bbox": [ + 140, + 161, + 190, + 174 + ], + "score": 0.88, + "content": "y \\in \\left\\{ 0 , 1 \\right\\} ^ { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 157, + 358, + 176 + ], + "score": 1.0, + "content": ", the label for each example in the batch", + "type": "text" + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 141, + 173, + 401, + 186 + ], + "spans": [ + { + "bbox": [ + 141, + 176, + 146, + 183 + ], + "score": 0.35, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 174, + 151, + 186 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 196, + 173, + 392, + 185 + ], + "score": 1.0, + "content": "privacy hyperparameter for the power constraint", + "type": "text" + }, + { + "bbox": [ + 392, + 174, + 401, + 183 + ], + "score": 0.7, + "content": "P", + "type": "inline_equation" + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 183, + 313, + 198 + ], + "spans": [ + { + "bbox": [ + 104, + 183, + 189, + 198 + ], + "score": 1.0, + "content": "output: g˜ ∈ RB×d,", + "type": "text" + }, + { + "bbox": [ + 195, + 184, + 313, + 198 + ], + "score": 1.0, + "content": "batch of perturbed gradients", + "type": "text" + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 196, + 444, + 210 + ], + "spans": [ + { + "bbox": [ + 107, + 196, + 444, + 210 + ], + "score": 1.0, + "content": "// Step 1: estimate the optimization constants from the batch", + "type": "text" + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 123, + 208, + 401, + 220 + ], + "spans": [ + { + "bbox": [ + 123, + 208, + 401, + 220 + ], + "score": 1.0, + "content": "gradients using maximum likelihood estimation (MLE)", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 95, + 216, + 490, + 235 + ], + "spans": [ + { + "bbox": [ + 95, + 223, + 106, + 234 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 106, + 219, + 145, + 235 + ], + "score": 0.9, + "content": "p \\gets \\frac { \\mathbf { 1 } ^ { T } y } { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 216, + 153, + 232 + ], + "score": 1.0, + "content": "1 T y ;", + "type": "text" + }, + { + "bbox": [ + 361, + 222, + 490, + 235 + ], + "score": 1.0, + "content": "/* positive fraction */", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 232, + 491, + 250 + ], + "spans": [ + { + "bbox": [ + 93, + 232, + 131, + 249 + ], + "score": 1.0, + "content": "2 g¯(1)", + "type": "text" + }, + { + "bbox": [ + 107, + 235, + 174, + 250 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\bar { g } ^ { ( 1 ) } \\frac { 1 } { \\mathbf { 1 } ^ { T } y } g ^ { T } y } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 237, + 180, + 248 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 383, + 236, + 397, + 247 + ], + "score": 0.34, + "content": "/ \\star", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 235, + 491, + 249 + ], + "score": 1.0, + "content": "positive mean */", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 94, + 247, + 490, + 264 + ], + "spans": [ + { + "bbox": [ + 94, + 247, + 106, + 264 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 106, + 250, + 212, + 264 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\bar { g } ^ { ( 0 ) } \\frac { 1 } { B - \\mathbf { 1 } ^ { T } y } g ^ { T } ( \\mathbf { 1 } - y ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 248, + 219, + 264 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 383, + 250, + 490, + 263 + ], + "score": 1.0, + "content": "/* negative mean */", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 94, + 260, + 184, + 280 + ], + "spans": [ + { + "bbox": [ + 94, + 260, + 106, + 280 + ], + "score": 1.0, + "content": "4", + "type": "text" + }, + { + "bbox": [ + 106, + 264, + 177, + 277 + ], + "score": 0.85, + "content": "\\Delta g = \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 260, + 184, + 280 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 96, + 275, + 490, + 298 + ], + "spans": [ + { + "bbox": [ + 96, + 275, + 105, + 298 + ], + "score": 1.0, + "content": "5", + "type": "text" + }, + { + "bbox": [ + 106, + 277, + 270, + 294 + ], + "score": 0.73, + "content": "\\begin{array} { r } { v \\frac { 1 } { d \\cdot ( \\mathbf { 1 } ^ { T } y ) } \\sum _ { j = 1 } ^ { B } y [ j ] \\cdot g [ j ] - \\bar { g } ^ { ( 1 ) } _ { 2 } ^ { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 275, + 276, + 298 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 345, + 280, + 490, + 293 + ], + "score": 1.0, + "content": "/* positive convariance */", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 291, + 489, + 315 + ], + "spans": [ + { + "bbox": [ + 91, + 291, + 105, + 315 + ], + "score": 1.0, + "content": "6", + "type": "text" + }, + { + "bbox": [ + 105, + 296, + 306, + 312 + ], + "score": 0.73, + "content": "\\begin{array} { r } { u \\frac { 1 } { d \\cdot ( B - \\mathbf { 1 } ^ { T } y ) } \\sum _ { j = 1 } ^ { B } ( 1 - y [ j ] ) \\cdot g [ j ] - \\bar { g } ^ { ( 0 ) } _ { 2 } ^ { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 291, + 315, + 315 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 345, + 297, + 477, + 310 + ], + "score": 1.0, + "content": "/* negative convariance", + "type": "text" + }, + { + "bbox": [ + 477, + 298, + 489, + 308 + ], + "score": 0.26, + "content": "\\star /", + "type": "inline_equation" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 94, + 309, + 489, + 326 + ], + "spans": [ + { + "bbox": [ + 94, + 309, + 106, + 326 + ], + "score": 1.0, + "content": "7", + "type": "text" + }, + { + "bbox": [ + 106, + 312, + 172, + 324 + ], + "score": 0.7, + "content": "P \\gets s \\cdot \\| \\Delta g \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 309, + 179, + 326 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 285, + 310, + 477, + 325 + ], + "score": 1.0, + "content": "/* power constraint hyperparameter", + "type": "text" + }, + { + "bbox": [ + 477, + 313, + 489, + 322 + ], + "score": 0.32, + "content": "\\star /", + "type": "inline_equation" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 104, + 321, + 359, + 335 + ], + "spans": [ + { + "bbox": [ + 104, + 321, + 165, + 335 + ], + "score": 1.0, + "content": "// Step 2:", + "type": "text" + }, + { + "bbox": [ + 172, + 324, + 359, + 335 + ], + "score": 1.0, + "content": "optimize the four-variable problem", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 96, + 334, + 164, + 345 + ], + "spans": [ + { + "bbox": [ + 96, + 334, + 115, + 345 + ], + "score": 1.0, + "content": "8 if", + "type": "text" + }, + { + "bbox": [ + 115, + 335, + 141, + 344 + ], + "score": 0.86, + "content": "u < v", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 334, + 164, + 345 + ], + "score": 1.0, + "content": "then", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 97, + 341, + 489, + 364 + ], + "spans": [ + { + "bbox": [ + 97, + 349, + 105, + 359 + ], + "score": 1.0, + "content": "9", + "type": "text" + }, + { + "bbox": [ + 122, + 345, + 162, + 360 + ], + "score": 0.64, + "content": "\\lambda _ { 2 } ^ { ( 1 ) } 0", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 341, + 171, + 364 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 298, + 347, + 466, + 359 + ], + "score": 1.0, + "content": "/* this variable is optimal at", + "type": "text" + }, + { + "bbox": [ + 466, + 348, + 489, + 358 + ], + "score": 0.6, + "content": "0 ~ \\star /", + "type": "inline_equation" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 94, + 355, + 489, + 379 + ], + "spans": [ + { + "bbox": [ + 94, + 363, + 105, + 374 + ], + "score": 1.0, + "content": "10", + "type": "text" + }, + { + "bbox": [ + 118, + 355, + 308, + 379 + ], + "score": 1.0, + "content": "Randomly initialize the optimization variables", + "type": "text" + }, + { + "bbox": [ + 309, + 359, + 326, + 374 + ], + "score": 0.81, + "content": "\\lambda _ { 1 } ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 355, + 330, + 379 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 330, + 359, + 348, + 374 + ], + "score": 0.65, + "content": "\\lambda _ { 1 } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 355, + 351, + 379 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 352, + 359, + 369, + 374 + ], + "score": 0.85, + "content": "\\lambda _ { 2 } ^ { ( 0 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 355, + 489, + 379 + ], + "score": 1.0, + "content": "in the feasible region in (15).", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 374, + 126, + 386 + ], + "spans": [ + { + "bbox": [ + 92, + 374, + 126, + 386 + ], + "score": 1.0, + "content": "11 else", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 93, + 384, + 489, + 423 + ], + "spans": [ + { + "bbox": [ + 93, + 388, + 106, + 415 + ], + "score": 1.0, + "content": "12 13", + "type": "text" + }, + { + "bbox": [ + 115, + 384, + 120, + 423 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 121, + 384, + 166, + 401 + ], + "score": 0.81, + "content": "\\lambda _ { 2 } ^ { ( 0 ) } 0 \\ : ;", + "type": "inline_equation", + "image_path": "983a21a35c931c70d2225b6e46518a843d5009549160e42b646e0975a396d632.jpg" + }, + { + "bbox": [ + 166, + 384, + 292, + 423 + ], + "score": 1.0, + "content": "initialize the optimization varia", + "type": "text" + }, + { + "bbox": [ + 293, + 388, + 307, + 398 + ], + "score": 0.32, + "content": "/ \\star", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 384, + 477, + 423 + ], + "score": 1.0, + "content": "le is optimal at 0 in the feasible region in (1", + "type": "text" + }, + { + "bbox": [ + 477, + 389, + 489, + 398 + ], + "score": 0.35, + "content": "\\star /", + "type": "inline_equation" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 310, + 399, + 370, + 415 + ], + "spans": [ + { + "bbox": [ + 310, + 399, + 370, + 415 + ], + "score": 0.36, + "content": "\\lambda _ { 1 } ^ { ( 1 ) } , \\lambda _ { 2 } ^ { ( 1 ) } , \\lambda _ { 1 } ^ { ( 0 ) }", + "type": "inline_equation" + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 92, + 415, + 126, + 426 + ], + "spans": [ + { + "bbox": [ + 92, + 415, + 126, + 426 + ], + "score": 1.0, + "content": "14 end", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 92, + 425, + 204, + 437 + ], + "spans": [ + { + "bbox": [ + 92, + 425, + 204, + 437 + ], + "score": 1.0, + "content": "15 while not converged do", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 93, + 435, + 335, + 448 + ], + "spans": [ + { + "bbox": [ + 93, + 437, + 105, + 448 + ], + "score": 1.0, + "content": "16", + "type": "text" + }, + { + "bbox": [ + 120, + 435, + 335, + 448 + ], + "score": 1.0, + "content": "Fix one of the newly updated optimization variables;", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 120, + 444, + 482, + 461 + ], + "spans": [ + { + "bbox": [ + 120, + 444, + 441, + 461 + ], + "score": 1.0, + "content": "// The 4-variable optimal solution lies on a hyperplane in", + "type": "text" + }, + { + "bbox": [ + 441, + 447, + 453, + 457 + ], + "score": 0.84, + "content": "\\mathbb { R } ^ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 444, + 482, + 461 + ], + "score": 1.0, + "content": "(see", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 139, + 459, + 439, + 470 + ], + "spans": [ + { + "bbox": [ + 139, + 459, + 439, + 470 + ], + "score": 1.0, + "content": "Appendix A.4 Remark) so fixing two variables gives us a", + "type": "text" + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 139, + 469, + 207, + 481 + ], + "spans": [ + { + "bbox": [ + 139, + 469, + 207, + 481 + ], + "score": 1.0, + "content": "line-segment", + "type": "text" + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 94, + 480, + 441, + 492 + ], + "spans": [ + { + "bbox": [ + 94, + 480, + 105, + 492 + ], + "score": 1.0, + "content": "17", + "type": "text" + }, + { + "bbox": [ + 121, + 480, + 441, + 492 + ], + "score": 1.0, + "content": "Update the remaining two optimization variables by performing 1-d line-search", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 125, + 491, + 423, + 504 + ], + "spans": [ + { + "bbox": [ + 125, + 491, + 423, + 504 + ], + "score": 1.0, + "content": "minimization of the convex function (15) while satisfying the constraints;", + "type": "text" + } + ], + "index": 32, + "is_list_end_line": true + } + ], + "index": 18, + "bbox_fs": [ + 91, + 135, + 491, + 504 + ] + }, + { + "type": "title", + "bbox": [ + 95, + 503, + 124, + 512 + ], + "lines": [ + { + "bbox": [ + 93, + 502, + 124, + 513 + ], + "spans": [ + { + "bbox": [ + 93, + 502, + 124, + 513 + ], + "score": 1.0, + "content": "18 end", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "index", + "bbox": [ + 94, + 510, + 492, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 512, + 385, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 166, + 526 + ], + "score": 1.0, + "content": "// Step 3:", + "type": "text" + }, + { + "bbox": [ + 172, + 513, + 385, + 525 + ], + "score": 1.0, + "content": "compute the optimal covariance matrices", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 521, + 265, + 549 + ], + "spans": [ + { + "bbox": [ + 91, + 526, + 106, + 545 + ], + "score": 1.0, + "content": "19", + "type": "text" + }, + { + "bbox": [ + 106, + 524, + 259, + 545 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\Sigma _ { 1 } ^ { * } \\frac { \\lambda _ { 1 } ^ { ( 1 ) } - \\lambda _ { 2 } ^ { ( 1 ) } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 1 ) } I _ { d } ; } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 521, + 265, + 549 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 541, + 264, + 569 + ], + "spans": [ + { + "bbox": [ + 91, + 546, + 106, + 565 + ], + "score": 1.0, + "content": "20", + "type": "text" + }, + { + "bbox": [ + 106, + 545, + 260, + 566 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\Sigma _ { 0 } ^ { * } \\frac { \\lambda _ { 1 } ^ { ( 0 ) } - \\lambda _ { 2 } ^ { ( 0 ) } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 0 ) } I _ { d } ; } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 541, + 264, + 569 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 563, + 289, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 167, + 578 + ], + "score": 1.0, + "content": "// Step 4:", + "type": "text" + }, + { + "bbox": [ + 171, + 565, + 289, + 577 + ], + "score": 1.0, + "content": "perturb the gradients", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 574, + 490, + 589 + ], + "spans": [ + { + "bbox": [ + 91, + 574, + 106, + 589 + ], + "score": 1.0, + "content": "21", + "type": "text" + }, + { + "bbox": [ + 106, + 576, + 151, + 587 + ], + "score": 0.88, + "content": "\\tilde { g } \\mathbf { 0 } _ { B \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 574, + 159, + 589 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 195, + 576, + 209, + 586 + ], + "score": 0.46, + "content": "/ \\star", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 574, + 490, + 588 + ], + "score": 1.0, + "content": "an empty matrix to store the perturbed gradients */", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 586, + 184, + 599 + ], + "spans": [ + { + "bbox": [ + 92, + 586, + 121, + 599 + ], + "score": 1.0, + "content": "22 for", + "type": "text" + }, + { + "bbox": [ + 122, + 588, + 148, + 598 + ], + "score": 0.89, + "content": "j 1", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 586, + 159, + 599 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 160, + 587, + 169, + 597 + ], + "score": 0.73, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 586, + 184, + 599 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 597, + 188, + 610 + ], + "spans": [ + { + "bbox": [ + 92, + 597, + 106, + 610 + ], + "score": 1.0, + "content": "23", + "type": "text" + }, + { + "bbox": [ + 120, + 597, + 130, + 609 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 131, + 598, + 165, + 609 + ], + "score": 0.89, + "content": "y [ j ] = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 597, + 188, + 609 + ], + "score": 1.0, + "content": "then", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 606, + 317, + 624 + ], + "spans": [ + { + "bbox": [ + 92, + 610, + 106, + 622 + ], + "score": 1.0, + "content": "24", + "type": "text" + }, + { + "bbox": [ + 133, + 606, + 135, + 624 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 135, + 609, + 213, + 622 + ], + "score": 0.76, + "content": "\\tilde { g } [ j ] g [ j ] + \\eta ^ { ( 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 606, + 243, + 624 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 244, + 609, + 313, + 622 + ], + "score": 0.92, + "content": "\\boldsymbol \\eta ^ { ( 1 ) } \\sim \\mathcal { N } ( \\mathbf 0 , \\Sigma _ { 1 } ^ { * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 606, + 317, + 624 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 620, + 194, + 633 + ], + "spans": [ + { + "bbox": [ + 92, + 621, + 106, + 633 + ], + "score": 1.0, + "content": "25", + "type": "text" + }, + { + "bbox": [ + 120, + 620, + 143, + 633 + ], + "score": 1.0, + "content": "else", + "type": "text" + }, + { + "bbox": [ + 143, + 622, + 194, + 632 + ], + "score": 0.76, + "content": "\\begin{array} { r l } { \\langle \\mathscr { I } / } & { { } y [ j ] = 0 } \\end{array}", + "type": "inline_equation" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 630, + 317, + 647 + ], + "spans": [ + { + "bbox": [ + 92, + 633, + 106, + 645 + ], + "score": 1.0, + "content": "26", + "type": "text" + }, + { + "bbox": [ + 125, + 632, + 212, + 645 + ], + "score": 0.77, + "content": "\\begin{array} { r l } { \\vert } & { { } \\tilde { g } [ j ] g [ j ] + \\eta ^ { ( 0 ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 630, + 243, + 647 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 244, + 632, + 313, + 645 + ], + "score": 0.9, + "content": "\\boldsymbol \\eta ^ { ( 0 ) } \\sim \\mathcal { N } ( \\mathbf 0 , \\Sigma _ { 0 } ^ { * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 630, + 317, + 647 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 92, + 644, + 141, + 656 + ], + "spans": [ + { + "bbox": [ + 92, + 644, + 106, + 656 + ], + "score": 1.0, + "content": "27", + "type": "text" + }, + { + "bbox": [ + 119, + 644, + 141, + 655 + ], + "score": 1.0, + "content": "end", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 654, + 127, + 668 + ], + "spans": [ + { + "bbox": [ + 91, + 654, + 127, + 668 + ], + "score": 1.0, + "content": "28 end", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + } + ], + "index": 39.5, + "bbox_fs": [ + 91, + 512, + 490, + 668 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 83, + 267, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 268, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 268, + 95 + ], + "score": 1.0, + "content": "A.5.1 MA R V E L L TIME COMPLEXITY", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 101, + 352, + 113 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 354, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 354, + 115 + ], + "score": 1.0, + "content": "The Marvell algorithm in 1 consists of the following steps:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 108, + 118, + 506, + 262 + ], + "lines": [ + { + "bbox": [ + 109, + 119, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 109, + 119, + 506, + 132 + ], + "score": 1.0, + "content": "Step 1. Compute the positive and negative gradient mean and their difference (line 1 - 4). This", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 141, + 130, + 505, + 143 + ], + "spans": [ + { + "bbox": [ + 141, + 130, + 289, + 143 + ], + "score": 1.0, + "content": "amounts to averaging over at most", + "type": "text" + }, + { + "bbox": [ + 289, + 130, + 298, + 140 + ], + "score": 0.78, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 130, + 505, + 143 + ], + "score": 1.0, + "content": "numbers over each of the d gradient dimensions.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 142, + 140, + 331, + 153 + ], + "spans": [ + { + "bbox": [ + 142, + 140, + 298, + 153 + ], + "score": 1.0, + "content": "Thus would have a time complexity of", + "type": "text" + }, + { + "bbox": [ + 298, + 141, + 328, + 153 + ], + "score": 0.93, + "content": "O ( B d )", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 140, + 331, + 153 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 109, + 151, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 109, + 151, + 367, + 164 + ], + "score": 1.0, + "content": "Step 2. Compute the positive and negative covariance constants", + "type": "text" + }, + { + "bbox": [ + 367, + 154, + 374, + 162 + ], + "score": 0.76, + "content": "u", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 151, + 392, + 164 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 392, + 154, + 399, + 162 + ], + "score": 0.71, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 151, + 505, + 164 + ], + "score": 1.0, + "content": "(line 5, 6). This operation", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 162, + 428, + 176 + ], + "spans": [ + { + "bbox": [ + 141, + 162, + 183, + 176 + ], + "score": 1.0, + "content": "also takes", + "type": "text" + }, + { + "bbox": [ + 184, + 162, + 214, + 174 + ], + "score": 0.93, + "content": "O ( \\bar { B } d )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 162, + 428, + 176 + ], + "score": 1.0, + "content": "as it averages over squares of coordinate differences.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 109, + 173, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 109, + 173, + 505, + 186 + ], + "score": 1.0, + "content": "Step 3. Use the results from step 1 and 2, solve the 4-variable optimization problem (line 7-18).", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 183, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 141, + 183, + 505, + 198 + ], + "score": 1.0, + "content": "Here because the constant size of the number of optimization variables. Solving this prob-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 142, + 195, + 354, + 207 + ], + "spans": [ + { + "bbox": [ + 142, + 196, + 329, + 207 + ], + "score": 1.0, + "content": "lem up to a fixed precision takes constant time", + "type": "text" + }, + { + "bbox": [ + 330, + 195, + 351, + 207 + ], + "score": 0.89, + "content": "O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 196, + 354, + 207 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 109, + 206, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 109, + 206, + 328, + 219 + ], + "score": 1.0, + "content": "Step 4. Perform the actual random perturbation (line", + "type": "text" + }, + { + "bbox": [ + 329, + 207, + 360, + 217 + ], + "score": 0.5, + "content": "1 9 - 2 8 ", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 206, + 505, + 219 + ], + "score": 1.0, + "content": "). Because of the additive structure", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 141, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "of each class’s covariance matrix, for every example’s gradient in the batch, we can in-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 228, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 141, + 228, + 505, + 241 + ], + "score": 1.0, + "content": "dependently sample one random Gaussian vector with rank-1 covariance and also another", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 239, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 141, + 239, + 505, + 252 + ], + "score": 1.0, + "content": "spherical Gaussian random vector and add both vectors to this gradient for perturbation.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 250, + 444, + 262 + ], + "spans": [ + { + "bbox": [ + 142, + 251, + 227, + 262 + ], + "score": 1.0, + "content": "This step would take", + "type": "text" + }, + { + "bbox": [ + 227, + 250, + 249, + 262 + ], + "score": 0.91, + "content": "O ( d )", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 251, + 380, + 262 + ], + "score": 1.0, + "content": "for each example and thus takes", + "type": "text" + }, + { + "bbox": [ + 381, + 250, + 410, + 262 + ], + "score": 0.93, + "content": "O ( B d )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 251, + 444, + 262 + ], + "score": 1.0, + "content": "in total.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 108, + 267, + 505, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 267, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 371, + 280 + ], + "score": 1.0, + "content": "As a result, the entire Marvell algorithm has a time complexity of", + "type": "text" + }, + { + "bbox": [ + 371, + 267, + 401, + 279 + ], + "score": 0.92, + "content": "O ( B d )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 267, + 506, + 280 + ], + "score": 1.0, + "content": ". This can be further sped", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "score": 1.0, + "content": "up through parallel computation using multi-threading/multi-core. Considering backpropagation", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 270, + 302 + ], + "score": 1.0, + "content": "through the cut layer would also require", + "type": "text" + }, + { + "bbox": [ + 270, + 289, + 300, + 301 + ], + "score": 0.92, + "content": "O ( B d )", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "time complexity, the Marvell algorithm would not", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 300, + 369, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 369, + 312 + ], + "score": 1.0, + "content": "slow down the split-learning process in any significant way at all.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 108, + 324, + 282, + 335 + ], + "lines": [ + { + "bbox": [ + 106, + 323, + 284, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 284, + 336 + ], + "score": 1.0, + "content": "A.5.2 MA R V E L L EMPIRICAL RUN TIME", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 342, + 505, + 420 + ], + "lines": [ + { + "bbox": [ + 106, + 342, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 506, + 356 + ], + "score": 1.0, + "content": "Empirically, in Table 1, we present the average amount of time it takes to run Marvell with the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 354, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 228, + 366 + ], + "score": 1.0, + "content": "privacy hyperparameter value", + "type": "text" + }, + { + "bbox": [ + 229, + 354, + 256, + 364 + ], + "score": 0.89, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 354, + 505, + 366 + ], + "score": 1.0, + "content": "for the three models (one for each dataset) we considered in", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 365, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 505, + 377 + ], + "score": 1.0, + "content": "our experiments (s value is chosen as it achieves good privacy protection as shown with purple line", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "score": 1.0, + "content": "in Figure 3, 5, 6). We also compare it to the average time it takes to run one update iteration. We", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 386, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 204, + 399 + ], + "score": 1.0, + "content": "additionally include the", + "type": "text" + }, + { + "bbox": [ + 204, + 387, + 225, + 397 + ], + "score": 0.87, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 386, + 506, + 399 + ], + "score": 1.0, + "content": "confidence interval for the mean estimator of the run time. Here we", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 398, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 410 + ], + "score": 1.0, + "content": "notice that the total run time of Marvell only takes up a very small amount of the total training time", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 408, + 432, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 432, + 422 + ], + "score": 1.0, + "content": "for each method, further corroborating our algorithm’s time efficiency advantage.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 131, + 426, + 478, + 538 + ], + "lines": [ + { + "bbox": [ + 131, + 426, + 478, + 538 + ], + "spans": [ + { + "bbox": [ + 131, + 426, + 478, + 538 + ], + "score": 0.47, + "content": "\\frac | \\begin{array} { l c } { { \\mathrm { b a t c h } } } \\\\ { { \\mathrm { s i z e } } } \\\\ { { \\frac { B } { 1 . 8 1 \\times } } } \\end{array} | \\begin{array} { l c } { { \\mathrm { c u t ~ l a y e r ~ f e a t u r e } } } \\\\ { { \\mathrm { d i m e n s i o n } } } \\\\ { { \\mathrm { ~ d i o n } } } \\end{array} | \\begin{array} { l l } { { \\mathrm { ~ a n v e r ~ f e n ~ t i m e } } } \\\\ { { \\mathrm { M a r v e l ~ r u n ~ t i m e } } } \\\\ { { \\mathrm { ~ 1 . 7 9 \\times ~ l 0 ^ { - 2 } ~ } } } \\\\ { { \\mathrm { ~ 2 . 7 . 0 3 \\times ~ 1 0 ^ { - 5 } ~ } } } \\end{array} | \\begin{array} { l l } { { \\mathrm { ~ a v e r a g e } } } \\\\ { { \\mathrm { ~ ( s e c o n d s / ~ t i m e ) } } } \\\\ { { \\mathrm { ~ 3 . 9 4 \\times ~ 1 0 ^ { - 1 } ~ } } } \\\\ { { \\mathrm { ~ 4 . 0 5 \\times ~ 1 0 ^ { - 3 } ~ } } } \\\\ { { \\mathrm { ~ 1 . 6 8 \\times ~ 1 0 ^ { - 2 } ~ } } } \\end{array} | \\begin{array} { l l } { { \\mathrm { a v e r a g e } } } \\\\ { { \\mathrm { ~ 4 . 2 8 \\times ~ 1 0 ^ { - 1 } ~ } } } \\\\ { { \\mathrm { ~ 4 . 8 4 ~ } } } \\end{array}", + "type": "interline_equation", + "image_path": "3c3a81d3a11c92453f592de42dae85a09e01dba4e1a20b9f31ee7a206368aa7f.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 131, + 426, + 478, + 463.3333333333333 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 131, + 463.3333333333333, + 478, + 500.66666666666663 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 131, + 500.66666666666663, + 478, + 538.0 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "table", + "bbox": [], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 545, + 505, + 565 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 255, + 556 + ], + "score": 1.0, + "content": "Table 1: Average run time of Marvell", + "type": "text" + }, + { + "bbox": [ + 255, + 545, + 279, + 555 + ], + "score": 0.84, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 543, + 505, + 556 + ], + "score": 1.0, + "content": ") and average total update time per iteration for models trained", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 554, + 209, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 209, + 566 + ], + "score": 1.0, + "content": "on ISIC, Criteo, and Avazu.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + } + ], + "index": 30.5 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "19", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 83, + 267, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 268, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 268, + 95 + ], + "score": 1.0, + "content": "A.5.1 MA R V E L L TIME COMPLEXITY", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 106, + 82, + 268, + 95 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 101, + 352, + 113 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 354, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 354, + 115 + ], + "score": 1.0, + "content": "The Marvell algorithm in 1 consists of the following steps:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 100, + 354, + 115 + ] + }, + { + "type": "list", + "bbox": [ + 108, + 118, + 506, + 262 + ], + "lines": [ + { + "bbox": [ + 109, + 119, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 109, + 119, + 506, + 132 + ], + "score": 1.0, + "content": "Step 1. Compute the positive and negative gradient mean and their difference (line 1 - 4). This", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 130, + 505, + 143 + ], + "spans": [ + { + "bbox": [ + 141, + 130, + 289, + 143 + ], + "score": 1.0, + "content": "amounts to averaging over at most", + "type": "text" + }, + { + "bbox": [ + 289, + 130, + 298, + 140 + ], + "score": 0.78, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 130, + 505, + 143 + ], + "score": 1.0, + "content": "numbers over each of the d gradient dimensions.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 142, + 140, + 331, + 153 + ], + "spans": [ + { + "bbox": [ + 142, + 140, + 298, + 153 + ], + "score": 1.0, + "content": "Thus would have a time complexity of", + "type": "text" + }, + { + "bbox": [ + 298, + 141, + 328, + 153 + ], + "score": 0.93, + "content": "O ( B d )", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 140, + 331, + 153 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 151, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 109, + 151, + 367, + 164 + ], + "score": 1.0, + "content": "Step 2. Compute the positive and negative covariance constants", + "type": "text" + }, + { + "bbox": [ + 367, + 154, + 374, + 162 + ], + "score": 0.76, + "content": "u", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 151, + 392, + 164 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 392, + 154, + 399, + 162 + ], + "score": 0.71, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 151, + 505, + 164 + ], + "score": 1.0, + "content": "(line 5, 6). This operation", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 162, + 428, + 176 + ], + "spans": [ + { + "bbox": [ + 141, + 162, + 183, + 176 + ], + "score": 1.0, + "content": "also takes", + "type": "text" + }, + { + "bbox": [ + 184, + 162, + 214, + 174 + ], + "score": 0.93, + "content": "O ( \\bar { B } d )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 162, + 428, + 176 + ], + "score": 1.0, + "content": "as it averages over squares of coordinate differences.", + "type": "text" + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 173, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 109, + 173, + 505, + 186 + ], + "score": 1.0, + "content": "Step 3. Use the results from step 1 and 2, solve the 4-variable optimization problem (line 7-18).", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 183, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 141, + 183, + 505, + 198 + ], + "score": 1.0, + "content": "Here because the constant size of the number of optimization variables. Solving this prob-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 142, + 195, + 354, + 207 + ], + "spans": [ + { + "bbox": [ + 142, + 196, + 329, + 207 + ], + "score": 1.0, + "content": "lem up to a fixed precision takes constant time", + "type": "text" + }, + { + "bbox": [ + 330, + 195, + 351, + 207 + ], + "score": 0.89, + "content": "O ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 196, + 354, + 207 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 206, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 109, + 206, + 328, + 219 + ], + "score": 1.0, + "content": "Step 4. Perform the actual random perturbation (line", + "type": "text" + }, + { + "bbox": [ + 329, + 207, + 360, + 217 + ], + "score": 0.5, + "content": "1 9 - 2 8 ", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 206, + 505, + 219 + ], + "score": 1.0, + "content": "). Because of the additive structure", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 141, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "of each class’s covariance matrix, for every example’s gradient in the batch, we can in-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 228, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 141, + 228, + 505, + 241 + ], + "score": 1.0, + "content": "dependently sample one random Gaussian vector with rank-1 covariance and also another", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 239, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 141, + 239, + 505, + 252 + ], + "score": 1.0, + "content": "spherical Gaussian random vector and add both vectors to this gradient for perturbation.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 250, + 444, + 262 + ], + "spans": [ + { + "bbox": [ + 142, + 251, + 227, + 262 + ], + "score": 1.0, + "content": "This step would take", + "type": "text" + }, + { + "bbox": [ + 227, + 250, + 249, + 262 + ], + "score": 0.91, + "content": "O ( d )", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 251, + 380, + 262 + ], + "score": 1.0, + "content": "for each example and thus takes", + "type": "text" + }, + { + "bbox": [ + 381, + 250, + 410, + 262 + ], + "score": 0.93, + "content": "O ( B d )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 251, + 444, + 262 + ], + "score": 1.0, + "content": "in total.", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + } + ], + "index": 8, + "bbox_fs": [ + 109, + 119, + 506, + 262 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 267, + 505, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 267, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 371, + 280 + ], + "score": 1.0, + "content": "As a result, the entire Marvell algorithm has a time complexity of", + "type": "text" + }, + { + "bbox": [ + 371, + 267, + 401, + 279 + ], + "score": 0.92, + "content": "O ( B d )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 267, + 506, + 280 + ], + "score": 1.0, + "content": ". This can be further sped", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 291 + ], + "score": 1.0, + "content": "up through parallel computation using multi-threading/multi-core. Considering backpropagation", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 270, + 302 + ], + "score": 1.0, + "content": "through the cut layer would also require", + "type": "text" + }, + { + "bbox": [ + 270, + 289, + 300, + 301 + ], + "score": 0.92, + "content": "O ( B d )", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "time complexity, the Marvell algorithm would not", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 300, + 369, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 369, + 312 + ], + "score": 1.0, + "content": "slow down the split-learning process in any significant way at all.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 267, + 506, + 312 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 324, + 282, + 335 + ], + "lines": [ + { + "bbox": [ + 106, + 323, + 284, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 284, + 336 + ], + "score": 1.0, + "content": "A.5.2 MA R V E L L EMPIRICAL RUN TIME", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 342, + 505, + 420 + ], + "lines": [ + { + "bbox": [ + 106, + 342, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 506, + 356 + ], + "score": 1.0, + "content": "Empirically, in Table 1, we present the average amount of time it takes to run Marvell with the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 354, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 228, + 366 + ], + "score": 1.0, + "content": "privacy hyperparameter value", + "type": "text" + }, + { + "bbox": [ + 229, + 354, + 256, + 364 + ], + "score": 0.89, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 354, + 505, + 366 + ], + "score": 1.0, + "content": "for the three models (one for each dataset) we considered in", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 365, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 505, + 377 + ], + "score": 1.0, + "content": "our experiments (s value is chosen as it achieves good privacy protection as shown with purple line", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "score": 1.0, + "content": "in Figure 3, 5, 6). We also compare it to the average time it takes to run one update iteration. We", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 386, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 204, + 399 + ], + "score": 1.0, + "content": "additionally include the", + "type": "text" + }, + { + "bbox": [ + 204, + 387, + 225, + 397 + ], + "score": 0.87, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 386, + 506, + 399 + ], + "score": 1.0, + "content": "confidence interval for the mean estimator of the run time. Here we", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 398, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 410 + ], + "score": 1.0, + "content": "notice that the total run time of Marvell only takes up a very small amount of the total training time", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 408, + 432, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 432, + 422 + ], + "score": 1.0, + "content": "for each method, further corroborating our algorithm’s time efficiency advantage.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 342, + 506, + 422 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 131, + 426, + 478, + 538 + ], + "lines": [ + { + "bbox": [ + 131, + 426, + 478, + 538 + ], + "spans": [ + { + "bbox": [ + 131, + 426, + 478, + 538 + ], + "score": 0.47, + "content": "\\frac | \\begin{array} { l c } { { \\mathrm { b a t c h } } } \\\\ { { \\mathrm { s i z e } } } \\\\ { { \\frac { B } { 1 . 8 1 \\times } } } \\end{array} | \\begin{array} { l c } { { \\mathrm { c u t ~ l a y e r ~ f e a t u r e } } } \\\\ { { \\mathrm { d i m e n s i o n } } } \\\\ { { \\mathrm { ~ d i o n } } } \\end{array} | \\begin{array} { l l } { { \\mathrm { ~ a n v e r ~ f e n ~ t i m e } } } \\\\ { { \\mathrm { M a r v e l ~ r u n ~ t i m e } } } \\\\ { { \\mathrm { ~ 1 . 7 9 \\times ~ l 0 ^ { - 2 } ~ } } } \\\\ { { \\mathrm { ~ 2 . 7 . 0 3 \\times ~ 1 0 ^ { - 5 } ~ } } } \\end{array} | \\begin{array} { l l } { { \\mathrm { ~ a v e r a g e } } } \\\\ { { \\mathrm { ~ ( s e c o n d s / ~ t i m e ) } } } \\\\ { { \\mathrm { ~ 3 . 9 4 \\times ~ 1 0 ^ { - 1 } ~ } } } \\\\ { { \\mathrm { ~ 4 . 0 5 \\times ~ 1 0 ^ { - 3 } ~ } } } \\\\ { { \\mathrm { ~ 1 . 6 8 \\times ~ 1 0 ^ { - 2 } ~ } } } \\end{array} | \\begin{array} { l l } { { \\mathrm { a v e r a g e } } } \\\\ { { \\mathrm { ~ 4 . 2 8 \\times ~ 1 0 ^ { - 1 } ~ } } } \\\\ { { \\mathrm { ~ 4 . 8 4 ~ } } } \\end{array}", + "type": "interline_equation", + "image_path": "3c3a81d3a11c92453f592de42dae85a09e01dba4e1a20b9f31ee7a206368aa7f.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 131, + 426, + 478, + 463.3333333333333 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 131, + 463.3333333333333, + 478, + 500.66666666666663 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 131, + 500.66666666666663, + 478, + 538.0 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "table", + "bbox": [], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 545, + 505, + 565 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 255, + 556 + ], + "score": 1.0, + "content": "Table 1: Average run time of Marvell", + "type": "text" + }, + { + "bbox": [ + 255, + 545, + 279, + 555 + ], + "score": 0.84, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 543, + 505, + 556 + ], + "score": 1.0, + "content": ") and average total update time per iteration for models trained", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 554, + 209, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 209, + 566 + ], + "score": 1.0, + "content": "on ISIC, Criteo, and Avazu.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + } + ], + "index": 30.5 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 109, + 82, + 318, + 93 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 320, + 95 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 320, + 95 + ], + "score": 1.0, + "content": "A.6 DATA SETUP AND EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 103, + 505, + 147 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 505, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 505, + 115 + ], + "score": 1.0, + "content": "We first describe how we preprocess each of the datasets in A.6.1. We then describe the model", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 113, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 505, + 127 + ], + "score": 1.0, + "content": "architecture used for each dataset in A.6.2. Finally, we describe what the training hyperparameters", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 505, + 137 + ], + "score": 1.0, + "content": "are used for each dataset/model combination and the amount of compute used for the experiments", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 135, + 144, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 144, + 148 + ], + "score": 1.0, + "content": "in A.6.3.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 160, + 253, + 171 + ], + "lines": [ + { + "bbox": [ + 106, + 160, + 254, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 254, + 172 + ], + "score": 1.0, + "content": "A.6.1 DATASET PREPROCESSING", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 178, + 505, + 267 + ], + "lines": [ + { + "bbox": [ + 106, + 178, + 505, + 192 + ], + "spans": [ + { + "bbox": [ + 106, + 178, + 505, + 192 + ], + "score": 1.0, + "content": "[Criteo] Every record of Criteo has 27 categorical input features and 14 real-valued input features.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 190, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 203 + ], + "score": 1.0, + "content": "We first replace all the NA values in categorical features with a single new category (which we", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 201, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 213 + ], + "score": 1.0, + "content": "represent using the empty string) and replace all the NA values in real-valued features with 0. For", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 212, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 506, + 225 + ], + "score": 1.0, + "content": "each categorical feature, we convert each of its possible value uniquely to an integer between 0", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 222, + 506, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 506, + 236 + ], + "score": 1.0, + "content": "(inclusive) and the total number of unique categories (exclusive). For each real-valued feature, we", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 234, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 336, + 247 + ], + "score": 1.0, + "content": "linearly normalize it into [0, 1]. We then randomly sample", + "type": "text" + }, + { + "bbox": [ + 336, + 234, + 356, + 245 + ], + "score": 0.89, + "content": "1 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 234, + 506, + 247 + ], + "score": 1.0, + "content": "of the entire Criteo publicly provided", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 246, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 505, + 258 + ], + "score": 1.0, + "content": "training set as our entire dataset (for faster training to generate privacy-utility trade-off comparision)", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 256, + 398, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 296, + 268 + ], + "score": 1.0, + "content": "and further make the subsampled dataset into a", + "type": "text" + }, + { + "bbox": [ + 296, + 256, + 316, + 267 + ], + "score": 0.65, + "content": "90 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 256, + 318, + 268 + ], + "score": 1.0, + "content": "-", + "type": "text" + }, + { + "bbox": [ + 318, + 256, + 337, + 267 + ], + "score": 0.37, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 256, + 398, + 268 + ], + "score": 1.0, + "content": "train-test split.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 280, + 504, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "[Avazu] Unlike Criteo, each record in Avazu only has categorical input features. We similarly", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "score": 1.0, + "content": "replace all NA value with a single new category (the empty string), and for each categorical feature,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 302, + 506, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 314 + ], + "score": 1.0, + "content": "we convert each of its possible value uniquely to an integer between 0 (inclusive) and the total", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 312, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 505, + 326 + ], + "score": 1.0, + "content": "number of unique categories (exclusive). We use all the records in provided in Avazu and randomly", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 324, + 291, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 152, + 336 + ], + "score": 1.0, + "content": "split it into", + "type": "text" + }, + { + "bbox": [ + 152, + 324, + 172, + 334 + ], + "score": 0.86, + "content": "90 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 324, + 237, + 336 + ], + "score": 1.0, + "content": "for training and", + "type": "text" + }, + { + "bbox": [ + 238, + 324, + 257, + 334 + ], + "score": 0.87, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 324, + 291, + 336 + ], + "score": 1.0, + "content": "for test.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 347, + 505, + 435 + ], + "lines": [ + { + "bbox": [ + 106, + 347, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 505, + 359 + ], + "score": 1.0, + "content": "[ISIC] The official SIIM-ISIC Melanoma Classification dataset has a total 33126 of skin lesion", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 358, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 196, + 370 + ], + "score": 1.0, + "content": "images with less than", + "type": "text" + }, + { + "bbox": [ + 196, + 358, + 212, + 369 + ], + "score": 0.86, + "content": "2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 358, + 505, + 370 + ], + "score": 1.0, + "content": "positive examples. Because for image classification model training it is", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 369, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 232, + 382 + ], + "score": 1.0, + "content": "desirable to use a batch size of", + "type": "text" + }, + { + "bbox": [ + 233, + 369, + 259, + 380 + ], + "score": 0.91, + "content": "\\sim 1 0 ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 369, + 505, + 382 + ], + "score": 1.0, + "content": ", it is highly likely that there won’t be any positive examples", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "sampled in a batch of such size. Thus to make the label leakage problem more severe, we modify", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 392, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 415, + 404 + ], + "score": 1.0, + "content": "the dataset by retaining all the 584 positive examples and randomly choosing", + "type": "text" + }, + { + "bbox": [ + 415, + 392, + 449, + 402 + ], + "score": 0.9, + "content": "5 8 4 \\times 9", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 392, + 505, + 404 + ], + "score": 1.0, + "content": "examples out", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 383, + 415 + ], + "score": 1.0, + "content": "of all the negative examples. By doing this, we enforce that there are", + "type": "text" + }, + { + "bbox": [ + 384, + 402, + 403, + 413 + ], + "score": 0.87, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "positive examples in this", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 414, + 504, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 371, + 426 + ], + "score": 1.0, + "content": "modified dataset. We randomly split these 5840 examples into a", + "type": "text" + }, + { + "bbox": [ + 371, + 414, + 412, + 424 + ], + "score": 0.75, + "content": "80 \\% - 2 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 414, + 504, + 426 + ], + "score": 1.0, + "content": "training and test split.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 423, + 387, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 240, + 438 + ], + "score": 1.0, + "content": "We also resize the images to size", + "type": "text" + }, + { + "bbox": [ + 240, + 425, + 273, + 435 + ], + "score": 0.9, + "content": "8 4 \\times 8 4", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 423, + 387, + 438 + ], + "score": 1.0, + "content": "for efficient model training.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22.5 + }, + { + "type": "title", + "bbox": [ + 108, + 448, + 282, + 459 + ], + "lines": [ + { + "bbox": [ + 106, + 448, + 284, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 284, + 460 + ], + "score": 1.0, + "content": "A.6.2 MODEL ARCHITECTURE DETAILS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 467, + 505, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 466, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 505, + 480 + ], + "score": 1.0, + "content": "[Criteo, Avazu] We use a popular deep learning model architecture WDL (Cheng et al., 2016)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "for online advertising. Here the deep and wide part each first processes the categorical features in", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 489, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 104, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "a given record by applying an embedding lookup for every categorical feature’s value. We use an", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 500, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 505, + 513 + ], + "score": 1.0, + "content": "embedding dimension of 4 for the deep part and embedding dimension of 1 for the wide part. After", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "score": 1.0, + "content": "the lookup, the deep/wide embeddings are then concatenated with the continuous features to form the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 522, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 535 + ], + "score": 1.0, + "content": "raw input vectors for both the deep part and wide part respectively. (This step is skipped for Avazu", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 533, + 504, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 504, + 546 + ], + "score": 1.0, + "content": "as it has no continuous features.) Then the wide part computes the wide part logit value through", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 544, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 104, + 544, + 505, + 557 + ], + "score": 1.0, + "content": "a real-valued linear function (with bias) of its raw input vectors, while the deep part processes its", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "score": 1.0, + "content": "raw input features using 6 ReLU-activated 128-unit MLP layers before producing a single deep part", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "score": 1.0, + "content": "logit. The two logits are summed up to form the final logic value. The cut layer is after the output", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 576, + 266, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 266, + 590 + ], + "score": 1.0, + "content": "of the 3rd ReLU layer on the deep part.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 677 + ], + "lines": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 308, + 613 + ], + "score": 1.0, + "content": "[ISIC] Every input image after resizing is of size", + "type": "text" + }, + { + "bbox": [ + 309, + 601, + 353, + 611 + ], + "score": 0.91, + "content": "8 4 \\times 8 4 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 600, + 506, + 613 + ], + "score": 1.0, + "content": ". We use a convolutional model with 6", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 612, + 504, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 278, + 623 + ], + "score": 1.0, + "content": "convolutional layers each with 64 channels", + "type": "text" + }, + { + "bbox": [ + 279, + 612, + 300, + 622 + ], + "score": 0.9, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 612, + 358, + 623 + ], + "score": 1.0, + "content": "filter size with", + "type": "text" + }, + { + "bbox": [ + 359, + 612, + 380, + 622 + ], + "score": 0.9, + "content": "1 \\times 1", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 612, + 504, + 623 + ], + "score": 1.0, + "content": "stride size. Each convolutional", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 448, + 635 + ], + "score": 1.0, + "content": "layer is followed by a ReLU activation function whose output is then max pooled with", + "type": "text" + }, + { + "bbox": [ + 449, + 623, + 470, + 633 + ], + "score": 0.89, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "window", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 634, + 504, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 165, + 645 + ], + "score": 1.0, + "content": "and stride size", + "type": "text" + }, + { + "bbox": [ + 165, + 634, + 187, + 644 + ], + "score": 0.89, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 634, + 504, + 645 + ], + "score": 1.0, + "content": ". The max-pooled output of the 6th layer is then flattened and pass into a 64-unit", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "ReLU-activated MLP layer before finally being linearly transformed into a single logit score. The", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 656, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 505, + 668 + ], + "score": 1.0, + "content": "cut layer is after the output of the 4th max pool layer. Thus the cut layer feature and gradient are", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 667, + 212, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 162, + 679 + ], + "score": 1.0, + "content": "both of shape", + "type": "text" + }, + { + "bbox": [ + 163, + 667, + 208, + 677 + ], + "score": 0.91, + "content": "5 \\times 5 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 667, + 212, + 679 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42 + }, + { + "type": "title", + "bbox": [ + 108, + 690, + 258, + 701 + ], + "lines": [ + { + "bbox": [ + 106, + 689, + 260, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 260, + 703 + ], + "score": 1.0, + "content": "A.6.3 MODEL TRAINING DETAILS", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Because the protection mechanism requires adding noise to the cut layer gradient, the induced vari-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 281, + 733 + ], + "score": 1.0, + "content": "ance of the gradients of non-label party’s", + "type": "text" + }, + { + "bbox": [ + 281, + 721, + 288, + 732 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "-parameters becomes larger. Thus to ensure smooth", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5 + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 109, + 82, + 318, + 93 + ], + "lines": [ + { + "bbox": [ + 107, + 82, + 320, + 95 + ], + "spans": [ + { + "bbox": [ + 107, + 82, + 320, + 95 + ], + "score": 1.0, + "content": "A.6 DATA SETUP AND EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 103, + 505, + 147 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 505, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 505, + 115 + ], + "score": 1.0, + "content": "We first describe how we preprocess each of the datasets in A.6.1. We then describe the model", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 113, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 505, + 127 + ], + "score": 1.0, + "content": "architecture used for each dataset in A.6.2. Finally, we describe what the training hyperparameters", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 505, + 137 + ], + "score": 1.0, + "content": "are used for each dataset/model combination and the amount of compute used for the experiments", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 135, + 144, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 144, + 148 + ], + "score": 1.0, + "content": "in A.6.3.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 102, + 505, + 148 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 160, + 253, + 171 + ], + "lines": [ + { + "bbox": [ + 106, + 160, + 254, + 172 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 254, + 172 + ], + "score": 1.0, + "content": "A.6.1 DATASET PREPROCESSING", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 178, + 505, + 267 + ], + "lines": [ + { + "bbox": [ + 106, + 178, + 505, + 192 + ], + "spans": [ + { + "bbox": [ + 106, + 178, + 505, + 192 + ], + "score": 1.0, + "content": "[Criteo] Every record of Criteo has 27 categorical input features and 14 real-valued input features.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 190, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 203 + ], + "score": 1.0, + "content": "We first replace all the NA values in categorical features with a single new category (which we", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 201, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 213 + ], + "score": 1.0, + "content": "represent using the empty string) and replace all the NA values in real-valued features with 0. For", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 212, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 506, + 225 + ], + "score": 1.0, + "content": "each categorical feature, we convert each of its possible value uniquely to an integer between 0", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 222, + 506, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 506, + 236 + ], + "score": 1.0, + "content": "(inclusive) and the total number of unique categories (exclusive). For each real-valued feature, we", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 234, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 336, + 247 + ], + "score": 1.0, + "content": "linearly normalize it into [0, 1]. We then randomly sample", + "type": "text" + }, + { + "bbox": [ + 336, + 234, + 356, + 245 + ], + "score": 0.89, + "content": "1 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 234, + 506, + 247 + ], + "score": 1.0, + "content": "of the entire Criteo publicly provided", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 246, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 505, + 258 + ], + "score": 1.0, + "content": "training set as our entire dataset (for faster training to generate privacy-utility trade-off comparision)", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 256, + 398, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 296, + 268 + ], + "score": 1.0, + "content": "and further make the subsampled dataset into a", + "type": "text" + }, + { + "bbox": [ + 296, + 256, + 316, + 267 + ], + "score": 0.65, + "content": "90 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 256, + 318, + 268 + ], + "score": 1.0, + "content": "-", + "type": "text" + }, + { + "bbox": [ + 318, + 256, + 337, + 267 + ], + "score": 0.37, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 256, + 398, + 268 + ], + "score": 1.0, + "content": "train-test split.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 178, + 506, + 268 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 280, + 504, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "[Avazu] Unlike Criteo, each record in Avazu only has categorical input features. We similarly", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "score": 1.0, + "content": "replace all NA value with a single new category (the empty string), and for each categorical feature,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 302, + 506, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 314 + ], + "score": 1.0, + "content": "we convert each of its possible value uniquely to an integer between 0 (inclusive) and the total", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 312, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 505, + 326 + ], + "score": 1.0, + "content": "number of unique categories (exclusive). We use all the records in provided in Avazu and randomly", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 324, + 291, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 152, + 336 + ], + "score": 1.0, + "content": "split it into", + "type": "text" + }, + { + "bbox": [ + 152, + 324, + 172, + 334 + ], + "score": 0.86, + "content": "90 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 324, + 237, + 336 + ], + "score": 1.0, + "content": "for training and", + "type": "text" + }, + { + "bbox": [ + 238, + 324, + 257, + 334 + ], + "score": 0.87, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 324, + 291, + 336 + ], + "score": 1.0, + "content": "for test.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 279, + 506, + 336 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 347, + 505, + 435 + ], + "lines": [ + { + "bbox": [ + 106, + 347, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 505, + 359 + ], + "score": 1.0, + "content": "[ISIC] The official SIIM-ISIC Melanoma Classification dataset has a total 33126 of skin lesion", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 358, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 196, + 370 + ], + "score": 1.0, + "content": "images with less than", + "type": "text" + }, + { + "bbox": [ + 196, + 358, + 212, + 369 + ], + "score": 0.86, + "content": "2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 358, + 505, + 370 + ], + "score": 1.0, + "content": "positive examples. Because for image classification model training it is", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 369, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 232, + 382 + ], + "score": 1.0, + "content": "desirable to use a batch size of", + "type": "text" + }, + { + "bbox": [ + 233, + 369, + 259, + 380 + ], + "score": 0.91, + "content": "\\sim 1 0 ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 369, + 505, + 382 + ], + "score": 1.0, + "content": ", it is highly likely that there won’t be any positive examples", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "sampled in a batch of such size. Thus to make the label leakage problem more severe, we modify", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 392, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 415, + 404 + ], + "score": 1.0, + "content": "the dataset by retaining all the 584 positive examples and randomly choosing", + "type": "text" + }, + { + "bbox": [ + 415, + 392, + 449, + 402 + ], + "score": 0.9, + "content": "5 8 4 \\times 9", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 392, + 505, + 404 + ], + "score": 1.0, + "content": "examples out", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 383, + 415 + ], + "score": 1.0, + "content": "of all the negative examples. By doing this, we enforce that there are", + "type": "text" + }, + { + "bbox": [ + 384, + 402, + 403, + 413 + ], + "score": 0.87, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "positive examples in this", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 414, + 504, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 371, + 426 + ], + "score": 1.0, + "content": "modified dataset. We randomly split these 5840 examples into a", + "type": "text" + }, + { + "bbox": [ + 371, + 414, + 412, + 424 + ], + "score": 0.75, + "content": "80 \\% - 2 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 414, + 504, + 426 + ], + "score": 1.0, + "content": "training and test split.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 423, + 387, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 240, + 438 + ], + "score": 1.0, + "content": "We also resize the images to size", + "type": "text" + }, + { + "bbox": [ + 240, + 425, + 273, + 435 + ], + "score": 0.9, + "content": "8 4 \\times 8 4", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 423, + 387, + 438 + ], + "score": 1.0, + "content": "for efficient model training.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 347, + 505, + 438 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 448, + 282, + 459 + ], + "lines": [ + { + "bbox": [ + 106, + 448, + 284, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 284, + 460 + ], + "score": 1.0, + "content": "A.6.2 MODEL ARCHITECTURE DETAILS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 467, + 505, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 466, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 505, + 480 + ], + "score": 1.0, + "content": "[Criteo, Avazu] We use a popular deep learning model architecture WDL (Cheng et al., 2016)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "for online advertising. Here the deep and wide part each first processes the categorical features in", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 489, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 104, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "a given record by applying an embedding lookup for every categorical feature’s value. We use an", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 500, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 505, + 513 + ], + "score": 1.0, + "content": "embedding dimension of 4 for the deep part and embedding dimension of 1 for the wide part. After", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 524 + ], + "score": 1.0, + "content": "the lookup, the deep/wide embeddings are then concatenated with the continuous features to form the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 522, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 535 + ], + "score": 1.0, + "content": "raw input vectors for both the deep part and wide part respectively. (This step is skipped for Avazu", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 533, + 504, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 504, + 546 + ], + "score": 1.0, + "content": "as it has no continuous features.) Then the wide part computes the wide part logit value through", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 544, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 104, + 544, + 505, + 557 + ], + "score": 1.0, + "content": "a real-valued linear function (with bias) of its raw input vectors, while the deep part processes its", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 568 + ], + "score": 1.0, + "content": "raw input features using 6 ReLU-activated 128-unit MLP layers before producing a single deep part", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 505, + 579 + ], + "score": 1.0, + "content": "logit. The two logits are summed up to form the final logic value. The cut layer is after the output", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 576, + 266, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 266, + 590 + ], + "score": 1.0, + "content": "of the 3rd ReLU layer on the deep part.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33, + "bbox_fs": [ + 104, + 466, + 506, + 590 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 677 + ], + "lines": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 308, + 613 + ], + "score": 1.0, + "content": "[ISIC] Every input image after resizing is of size", + "type": "text" + }, + { + "bbox": [ + 309, + 601, + 353, + 611 + ], + "score": 0.91, + "content": "8 4 \\times 8 4 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 600, + 506, + 613 + ], + "score": 1.0, + "content": ". We use a convolutional model with 6", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 612, + 504, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 278, + 623 + ], + "score": 1.0, + "content": "convolutional layers each with 64 channels", + "type": "text" + }, + { + "bbox": [ + 279, + 612, + 300, + 622 + ], + "score": 0.9, + "content": "3 \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 612, + 358, + 623 + ], + "score": 1.0, + "content": "filter size with", + "type": "text" + }, + { + "bbox": [ + 359, + 612, + 380, + 622 + ], + "score": 0.9, + "content": "1 \\times 1", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 612, + 504, + 623 + ], + "score": 1.0, + "content": "stride size. Each convolutional", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 448, + 635 + ], + "score": 1.0, + "content": "layer is followed by a ReLU activation function whose output is then max pooled with", + "type": "text" + }, + { + "bbox": [ + 449, + 623, + 470, + 633 + ], + "score": 0.89, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "window", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 634, + 504, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 165, + 645 + ], + "score": 1.0, + "content": "and stride size", + "type": "text" + }, + { + "bbox": [ + 165, + 634, + 187, + 644 + ], + "score": 0.89, + "content": "2 \\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 634, + 504, + 645 + ], + "score": 1.0, + "content": ". The max-pooled output of the 6th layer is then flattened and pass into a 64-unit", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "ReLU-activated MLP layer before finally being linearly transformed into a single logit score. The", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 656, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 505, + 668 + ], + "score": 1.0, + "content": "cut layer is after the output of the 4th max pool layer. Thus the cut layer feature and gradient are", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 667, + 212, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 162, + 679 + ], + "score": 1.0, + "content": "both of shape", + "type": "text" + }, + { + "bbox": [ + 163, + 667, + 208, + 677 + ], + "score": 0.91, + "content": "5 \\times 5 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 667, + 212, + 679 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 600, + 506, + 679 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 690, + 258, + 701 + ], + "lines": [ + { + "bbox": [ + 106, + 689, + 260, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 260, + 703 + ], + "score": 1.0, + "content": "A.6.3 MODEL TRAINING DETAILS", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Because the protection mechanism requires adding noise to the cut layer gradient, the induced vari-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 281, + 733 + ], + "score": 1.0, + "content": "ance of the gradients of non-label party’s", + "type": "text" + }, + { + "bbox": [ + 281, + 721, + 288, + 732 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "-parameters becomes larger. Thus to ensure smooth", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "optimization and sufficient training loss minimization, we use a slightly smaller learning rate than", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 199, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 199, + 105 + ], + "score": 1.0, + "content": "what is normally used.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 47.5, + "bbox_fs": [ + 106, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "optimization and sufficient training loss minimization, we use a slightly smaller learning rate than", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 199, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 199, + 105 + ], + "score": 1.0, + "content": "what is normally used.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 117, + 503, + 140 + ], + "lines": [ + { + "bbox": [ + 105, + 115, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 444, + 130 + ], + "score": 1.0, + "content": "[Criteo] We use the Adam optimizer with a batch size of 1024 and a learning rate of", + "type": "text" + }, + { + "bbox": [ + 444, + 118, + 468, + 128 + ], + "score": 0.47, + "content": "1 e { - 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 115, + 505, + 130 + ], + "score": 1.0, + "content": "through-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 128, + 439, + 141 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 306, + 141 + ], + "score": 1.0, + "content": "out the entire training of 5 epochs (approximately", + "type": "text" + }, + { + "bbox": [ + 307, + 128, + 324, + 139 + ], + "score": 0.31, + "content": "2 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 128, + 439, + 141 + ], + "score": 1.0, + "content": "stochastic gradient updates).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 105, + 151, + 504, + 174 + ], + "lines": [ + { + "bbox": [ + 106, + 151, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 151, + 434, + 164 + ], + "score": 1.0, + "content": "[ISIC] We use the Adam optimizer with a batch size of 128 and a learning rate of", + "type": "text" + }, + { + "bbox": [ + 435, + 152, + 458, + 162 + ], + "score": 0.58, + "content": "1 e { - } 5", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 151, + 505, + 164 + ], + "score": 1.0, + "content": "throughout", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 163, + 439, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 306, + 176 + ], + "score": 1.0, + "content": "the entire training of 1000 epochs (approximately", + "type": "text" + }, + { + "bbox": [ + 307, + 163, + 323, + 173 + ], + "score": 0.64, + "content": "3 5 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 163, + 439, + 176 + ], + "score": 1.0, + "content": "stochastic gradient updates).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 186, + 504, + 209 + ], + "lines": [ + { + "bbox": [ + 106, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 480, + 199 + ], + "score": 1.0, + "content": "[Avazu] We use the Adam optimizer with a batch size of 32768 and a learning rate of", + "type": "text" + }, + { + "bbox": [ + 480, + 186, + 505, + 197 + ], + "score": 0.46, + "content": "1 e { - 4 }", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 198, + 473, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 337, + 210 + ], + "score": 1.0, + "content": "throughout the entire training of 5 epochs (approximately", + "type": "text" + }, + { + "bbox": [ + 338, + 198, + 357, + 208 + ], + "score": 0.54, + "content": "5 . 5 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 198, + 473, + 210 + ], + "score": 1.0, + "content": "stochastic gradient updates).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 214, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 106, + 214, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 505, + 226 + ], + "score": 1.0, + "content": "We conduct our experiments over 16 Nvidia 1080Ti GPU card. Each run of Avazu takes about 11", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 225, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 237 + ], + "score": 1.0, + "content": "hours to finish on a single GPU card occupying 8GB of GPU RAM. Each run of Criteo takes about", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 235, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 506, + 249 + ], + "score": 1.0, + "content": "37 hours to finish on a single GPU card using 5 GB of GPU RAM. Each run of ISIC takes about 12", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 248, + 379, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 379, + 259 + ], + "score": 1.0, + "content": "hours to finish on a single GPU card occupying 4GB of GPU RAM.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 310, + 760 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 81, + 505, + 105 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 117, + 503, + 140 + ], + "lines": [ + { + "bbox": [ + 105, + 115, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 444, + 130 + ], + "score": 1.0, + "content": "[Criteo] We use the Adam optimizer with a batch size of 1024 and a learning rate of", + "type": "text" + }, + { + "bbox": [ + 444, + 118, + 468, + 128 + ], + "score": 0.47, + "content": "1 e { - 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 115, + 505, + 130 + ], + "score": 1.0, + "content": "through-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 128, + 439, + 141 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 306, + 141 + ], + "score": 1.0, + "content": "out the entire training of 5 epochs (approximately", + "type": "text" + }, + { + "bbox": [ + 307, + 128, + 324, + 139 + ], + "score": 0.31, + "content": "2 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 128, + 439, + 141 + ], + "score": 1.0, + "content": "stochastic gradient updates).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 115, + 505, + 141 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 151, + 504, + 174 + ], + "lines": [ + { + "bbox": [ + 106, + 151, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 151, + 434, + 164 + ], + "score": 1.0, + "content": "[ISIC] We use the Adam optimizer with a batch size of 128 and a learning rate of", + "type": "text" + }, + { + "bbox": [ + 435, + 152, + 458, + 162 + ], + "score": 0.58, + "content": "1 e { - } 5", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 151, + 505, + 164 + ], + "score": 1.0, + "content": "throughout", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 163, + 439, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 306, + 176 + ], + "score": 1.0, + "content": "the entire training of 1000 epochs (approximately", + "type": "text" + }, + { + "bbox": [ + 307, + 163, + 323, + 173 + ], + "score": 0.64, + "content": "3 5 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 163, + 439, + 176 + ], + "score": 1.0, + "content": "stochastic gradient updates).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 106, + 151, + 505, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 186, + 504, + 209 + ], + "lines": [ + { + "bbox": [ + 106, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 480, + 199 + ], + "score": 1.0, + "content": "[Avazu] We use the Adam optimizer with a batch size of 32768 and a learning rate of", + "type": "text" + }, + { + "bbox": [ + 480, + 186, + 505, + 197 + ], + "score": 0.46, + "content": "1 e { - 4 }", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 198, + 473, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 337, + 210 + ], + "score": 1.0, + "content": "throughout the entire training of 5 epochs (approximately", + "type": "text" + }, + { + "bbox": [ + 338, + 198, + 357, + 208 + ], + "score": 0.54, + "content": "5 . 5 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 198, + 473, + 210 + ], + "score": 1.0, + "content": "stochastic gradient updates).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 186, + 505, + 210 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 214, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 106, + 214, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 505, + 226 + ], + "score": 1.0, + "content": "We conduct our experiments over 16 Nvidia 1080Ti GPU card. Each run of Avazu takes about 11", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 225, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 237 + ], + "score": 1.0, + "content": "hours to finish on a single GPU card occupying 8GB of GPU RAM. Each run of Criteo takes about", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 235, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 506, + 249 + ], + "score": 1.0, + "content": "37 hours to finish on a single GPU card using 5 GB of GPU RAM. Each run of ISIC takes about 12", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 248, + 379, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 379, + 259 + ], + "score": 1.0, + "content": "hours to finish on a single GPU card occupying 4GB of GPU RAM.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 214, + 506, + 259 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 292, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 294, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 294, + 95 + ], + "score": 1.0, + "content": "A.7 COMPLETE EXPERIMENTAL RESULTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 102, + 362, + 114 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 363, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 363, + 115 + ], + "score": 1.0, + "content": "A.7.1 LEAK AUC PROGRESSION FOR AVAZU AND CRITEO", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 121, + 505, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 505, + 135 + ], + "score": 1.0, + "content": "In addition to the leak AUC progression on ISIC shown in Figure 3 in the main paper, we also show", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 133, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 505, + 145 + ], + "score": 1.0, + "content": "the leak AUC progression on the Avazu and Criteo datasets throughout training here in Figure 5 and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 145, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 505, + 156 + ], + "score": 1.0, + "content": "Figure 6. We similarly compare Marvell with different levels of protection strength (s values)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 155, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 104, + 155, + 506, + 168 + ], + "score": 1.0, + "content": "against the no protection baseline no noise. As we can see, Marvell still achieves strong and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 165, + 501, + 179 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 501, + 179 + ], + "score": 1.0, + "content": "flexible privacy protection on these two datasets against our label attacks at different model layers.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4 + }, + { + "type": "image", + "bbox": [ + 106, + 186, + 505, + 276 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 186, + 505, + 276 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 186, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 505, + 276 + ], + "score": 0.963, + "type": "image", + "image_path": "fc64579aad78e2fa08ffb6045f49d1a0fde107f7a7dc254d23753e69f6c1753e.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 106, + 186, + 505, + 216.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 106, + 216.0, + 505, + 246.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 106, + 246.0, + 505, + 276.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 277, + 505, + 297 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 276, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 506, + 287 + ], + "score": 1.0, + "content": "Figure 5: Norm and cosine leak AUC (computed every batch) at the cut layer and at the first layer of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 284, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 104, + 284, + 387, + 298 + ], + "score": 1.0, + "content": "no noise (no protection) vs. Marvell with different scale hyperparameter", + "type": "text" + }, + { + "bbox": [ + 387, + 288, + 393, + 294 + ], + "score": 0.61, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 284, + 505, + 298 + ], + "score": 1.0, + "content": "throughout the Avazu training.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + } + ], + "index": 9.25 + }, + { + "type": "image", + "bbox": [ + 106, + 316, + 505, + 407 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 316, + 505, + 407 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 316, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 505, + 407 + ], + "score": 0.967, + "type": "image", + "image_path": "7998fbd902b18742af0791b87deac7e2a9cb14ec0faab7692e4248043d10fa75.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 106, + 316, + 505, + 346.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 106, + 346.3333333333333, + 505, + 376.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 376.66666666666663, + 505, + 406.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 408, + 505, + 428 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 406, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 505, + 417 + ], + "score": 1.0, + "content": "Figure 6: Norm and cosine leak AUC (computed every batch) at the cut layer and at the first layer of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 415, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 104, + 415, + 506, + 429 + ], + "score": 1.0, + "content": "no noise (no protection) vs. Marvell with different scale hyperparameter s throughout the Criteo training.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + } + ], + "index": 14.25 + }, + { + "type": "title", + "bbox": [ + 107, + 446, + 322, + 457 + ], + "lines": [ + { + "bbox": [ + 106, + 446, + 323, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 323, + 459 + ], + "score": 1.0, + "content": "A.7.2 COMPLETE PRIVACY-UTILITY TRADEOFFS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 465, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "We show additional Privacy-Utility tradeoff results for all three datasets considered in this paper.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 476, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 506, + 489 + ], + "score": 1.0, + "content": "(Some of the plots have already been shown in the main paper but we still include them here for", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 488, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 505, + 501 + ], + "score": 1.0, + "content": "completeness and ease of reference.) For each dataset, we compare the privacy-utility tradeoff over", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 498, + 270, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 270, + 511 + ], + "score": 1.0, + "content": "multiple measures of privacy and utility:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "title", + "bbox": [ + 107, + 515, + 141, + 527 + ], + "lines": [ + { + "bbox": [ + 105, + 513, + 142, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 142, + 531 + ], + "score": 1.0, + "content": "Privacy", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 532, + 504, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "We consider our introduced privacy metrics using the activation gradient from the cut layer and the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 542, + 241, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 241, + 556 + ], + "score": 1.0, + "content": "first layer of the non-label party :", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 134, + 560, + 282, + 604 + ], + "lines": [ + { + "bbox": [ + 133, + 559, + 276, + 572 + ], + "spans": [ + { + "bbox": [ + 133, + 559, + 142, + 572 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 142, + 560, + 162, + 570 + ], + "score": 0.85, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 559, + 276, + 572 + ], + "score": 1.0, + "content": "norm leak AUC at cut layer", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 132, + 570, + 280, + 583 + ], + "spans": [ + { + "bbox": [ + 132, + 570, + 142, + 583 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 142, + 571, + 162, + 581 + ], + "score": 0.84, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 570, + 280, + 583 + ], + "score": 1.0, + "content": "cosine leak AUC at cut layer", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 133, + 581, + 279, + 594 + ], + "spans": [ + { + "bbox": [ + 133, + 581, + 142, + 594 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 142, + 582, + 162, + 592 + ], + "score": 0.84, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 581, + 279, + 594 + ], + "score": 1.0, + "content": "norm leak AUC at first layer", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 133, + 591, + 284, + 605 + ], + "spans": [ + { + "bbox": [ + 133, + 591, + 141, + 605 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 142, + 593, + 162, + 604 + ], + "score": 0.84, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 591, + 284, + 605 + ], + "score": 1.0, + "content": "cosine leak AUC at first layer", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "title", + "bbox": [ + 107, + 610, + 135, + 621 + ], + "lines": [ + { + "bbox": [ + 107, + 610, + 135, + 621 + ], + "spans": [], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 108, + 627, + 252, + 638 + ], + "lines": [ + { + "bbox": [ + 106, + 624, + 252, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 252, + 640 + ], + "score": 1.0, + "content": "We consider three metrics of utility:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 133, + 644, + 507, + 732 + ], + "lines": [ + { + "bbox": [ + 134, + 642, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 134, + 642, + 505, + 658 + ], + "score": 1.0, + "content": "• training loss (train loss): the lowest loss achieved on the training set throughout training.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 141, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 141, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "This directly measures how much the random protection perturbation influences the opti-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 141, + 665, + 181, + 677 + ], + "spans": [ + { + "bbox": [ + 141, + 665, + 181, + 677 + ], + "score": 1.0, + "content": "mization.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 135, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 135, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "• test loss. Because we only control the training optimization stochastic gradient’s variance,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 141, + 688, + 504, + 700 + ], + "spans": [ + { + "bbox": [ + 141, + 688, + 504, + 700 + ], + "score": 1.0, + "content": "measuring test loss directly tells us how much impact the training optimization random per-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 141, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 141, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "turbation influences beyond optimization but on the learned model’s generalization ability.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 132, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 132, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "• test AUC. As we are dealing with binary classification problem (where performance is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 142, + 721, + 496, + 733 + ], + "spans": [ + { + "bbox": [ + 142, + 721, + 496, + 733 + ], + "score": 1.0, + "content": "commonly measured through test AUC), we also naturally consider it as a utility metric.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 34.5 + } + ], + "page_idx": 21, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 292, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 294, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 294, + 95 + ], + "score": 1.0, + "content": "A.7 COMPLETE EXPERIMENTAL RESULTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 102, + 362, + 114 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 363, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 363, + 115 + ], + "score": 1.0, + "content": "A.7.1 LEAK AUC PROGRESSION FOR AVAZU AND CRITEO", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 121, + 505, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 505, + 135 + ], + "score": 1.0, + "content": "In addition to the leak AUC progression on ISIC shown in Figure 3 in the main paper, we also show", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 133, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 505, + 145 + ], + "score": 1.0, + "content": "the leak AUC progression on the Avazu and Criteo datasets throughout training here in Figure 5 and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 145, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 505, + 156 + ], + "score": 1.0, + "content": "Figure 6. We similarly compare Marvell with different levels of protection strength (s values)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 155, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 104, + 155, + 506, + 168 + ], + "score": 1.0, + "content": "against the no protection baseline no noise. As we can see, Marvell still achieves strong and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 165, + 501, + 179 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 501, + 179 + ], + "score": 1.0, + "content": "flexible privacy protection on these two datasets against our label attacks at different model layers.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4, + "bbox_fs": [ + 104, + 122, + 506, + 179 + ] + }, + { + "type": "image", + "bbox": [ + 106, + 186, + 505, + 276 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 186, + 505, + 276 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 186, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 505, + 276 + ], + "score": 0.963, + "type": "image", + "image_path": "fc64579aad78e2fa08ffb6045f49d1a0fde107f7a7dc254d23753e69f6c1753e.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 106, + 186, + 505, + 216.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 106, + 216.0, + 505, + 246.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 106, + 246.0, + 505, + 276.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 277, + 505, + 297 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 276, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 506, + 287 + ], + "score": 1.0, + "content": "Figure 5: Norm and cosine leak AUC (computed every batch) at the cut layer and at the first layer of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 284, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 104, + 284, + 387, + 298 + ], + "score": 1.0, + "content": "no noise (no protection) vs. Marvell with different scale hyperparameter", + "type": "text" + }, + { + "bbox": [ + 387, + 288, + 393, + 294 + ], + "score": 0.61, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 284, + 505, + 298 + ], + "score": 1.0, + "content": "throughout the Avazu training.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + } + ], + "index": 9.25 + }, + { + "type": "image", + "bbox": [ + 106, + 316, + 505, + 407 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 316, + 505, + 407 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 316, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 505, + 407 + ], + "score": 0.967, + "type": "image", + "image_path": "7998fbd902b18742af0791b87deac7e2a9cb14ec0faab7692e4248043d10fa75.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 106, + 316, + 505, + 346.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 106, + 346.3333333333333, + 505, + 376.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 106, + 376.66666666666663, + 505, + 406.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 408, + 505, + 428 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 406, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 505, + 417 + ], + "score": 1.0, + "content": "Figure 6: Norm and cosine leak AUC (computed every batch) at the cut layer and at the first layer of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 415, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 104, + 415, + 506, + 429 + ], + "score": 1.0, + "content": "no noise (no protection) vs. Marvell with different scale hyperparameter s throughout the Criteo training.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + } + ], + "index": 14.25 + }, + { + "type": "title", + "bbox": [ + 107, + 446, + 322, + 457 + ], + "lines": [ + { + "bbox": [ + 106, + 446, + 323, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 323, + 459 + ], + "score": 1.0, + "content": "A.7.2 COMPLETE PRIVACY-UTILITY TRADEOFFS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 465, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "We show additional Privacy-Utility tradeoff results for all three datasets considered in this paper.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 476, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 506, + 489 + ], + "score": 1.0, + "content": "(Some of the plots have already been shown in the main paper but we still include them here for", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 488, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 505, + 501 + ], + "score": 1.0, + "content": "completeness and ease of reference.) For each dataset, we compare the privacy-utility tradeoff over", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 498, + 270, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 270, + 511 + ], + "score": 1.0, + "content": "multiple measures of privacy and utility:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 464, + 506, + 511 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 515, + 141, + 527 + ], + "lines": [ + { + "bbox": [ + 105, + 513, + 142, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 142, + 531 + ], + "score": 1.0, + "content": "Privacy", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 532, + 504, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "We consider our introduced privacy metrics using the activation gradient from the cut layer and the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 542, + 241, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 241, + 556 + ], + "score": 1.0, + "content": "first layer of the non-label party :", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 532, + 505, + 556 + ] + }, + { + "type": "text", + "bbox": [ + 134, + 560, + 282, + 604 + ], + "lines": [ + { + "bbox": [ + 133, + 559, + 276, + 572 + ], + "spans": [ + { + "bbox": [ + 133, + 559, + 142, + 572 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 142, + 560, + 162, + 570 + ], + "score": 0.85, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 559, + 276, + 572 + ], + "score": 1.0, + "content": "norm leak AUC at cut layer", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 132, + 570, + 280, + 583 + ], + "spans": [ + { + "bbox": [ + 132, + 570, + 142, + 583 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 142, + 571, + 162, + 581 + ], + "score": 0.84, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 570, + 280, + 583 + ], + "score": 1.0, + "content": "cosine leak AUC at cut layer", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 133, + 581, + 279, + 594 + ], + "spans": [ + { + "bbox": [ + 133, + 581, + 142, + 594 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 142, + 582, + 162, + 592 + ], + "score": 0.84, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 581, + 279, + 594 + ], + "score": 1.0, + "content": "norm leak AUC at first layer", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 133, + 591, + 284, + 605 + ], + "spans": [ + { + "bbox": [ + 133, + 591, + 141, + 605 + ], + "score": 1.0, + "content": "•", + "type": "text" + }, + { + "bbox": [ + 142, + 593, + 162, + 604 + ], + "score": 0.84, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 591, + 284, + 605 + ], + "score": 1.0, + "content": "cosine leak AUC at first layer", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 132, + 559, + 284, + 605 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 610, + 135, + 621 + ], + "lines": [ + { + "bbox": [ + 107, + 610, + 135, + 621 + ], + "spans": [], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 108, + 627, + 252, + 638 + ], + "lines": [ + { + "bbox": [ + 106, + 624, + 252, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 252, + 640 + ], + "score": 1.0, + "content": "We consider three metrics of utility:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 624, + 252, + 640 + ] + }, + { + "type": "text", + "bbox": [ + 133, + 644, + 507, + 732 + ], + "lines": [ + { + "bbox": [ + 134, + 642, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 134, + 642, + 505, + 658 + ], + "score": 1.0, + "content": "• training loss (train loss): the lowest loss achieved on the training set throughout training.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 141, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 141, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "This directly measures how much the random protection perturbation influences the opti-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 141, + 665, + 181, + 677 + ], + "spans": [ + { + "bbox": [ + 141, + 665, + 181, + 677 + ], + "score": 1.0, + "content": "mization.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 135, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 135, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "• test loss. Because we only control the training optimization stochastic gradient’s variance,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 141, + 688, + 504, + 700 + ], + "spans": [ + { + "bbox": [ + 141, + 688, + 504, + 700 + ], + "score": 1.0, + "content": "measuring test loss directly tells us how much impact the training optimization random per-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 141, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 141, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "turbation influences beyond optimization but on the learned model’s generalization ability.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 132, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 132, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "• test AUC. As we are dealing with binary classification problem (where performance is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 142, + 721, + 496, + 733 + ], + "spans": [ + { + "bbox": [ + 142, + 721, + 496, + 733 + ], + "score": 1.0, + "content": "commonly measured through test AUC), we also naturally consider it as a utility metric.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 34.5, + "bbox_fs": [ + 132, + 642, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "As shown in Figure 7,8,9, Marvell consistently outperforms the isotropic Gaussian baseline over", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "all the different privacy-utility definitions. In addition, our proposed heuristic max norm is also", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 416, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 416, + 116 + ], + "score": 1.0, + "content": "particularly effective against our identified norm and direction-based attacks.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "image", + "bbox": [ + 110, + 124, + 501, + 374 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 124, + 501, + 374 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 124, + 501, + 374 + ], + "spans": [ + { + "bbox": [ + 110, + 124, + 501, + 374 + ], + "score": 0.973, + "type": "image", + "image_path": "cfa83c054f2045646208abae64f5e5969c5183c3b6697991cd25d9d7634217b2.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 110, + 124, + 501, + 207.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 110, + 207.33333333333331, + 501, + 290.66666666666663 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 110, + 290.66666666666663, + 501, + 373.99999999999994 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 381, + 505, + 403 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 382, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 382, + 505, + 393 + ], + "score": 1.0, + "content": "Figure 7: Privacy (norm and cosine leak AUC) vs Utility (train loss, test loss, and test AUC) trade-off of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 392, + 480, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 480, + 404 + ], + "score": 1.0, + "content": "protection methods (Marvell, iso, no noise, max norm) at the cut layer and first layer on Avazu.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + } + ], + "index": 5.25 + }, + { + "type": "image", + "bbox": [ + 110, + 418, + 501, + 670 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 418, + 501, + 670 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 418, + 501, + 670 + ], + "spans": [ + { + "bbox": [ + 110, + 418, + 501, + 670 + ], + "score": 0.971, + "type": "image", + "image_path": "f9b37071e310cce66c89d73ed7335527adaa39502b1e85c60894e68a77ab15fc.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 110, + 418, + 501, + 502.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 110, + 502.0, + 501, + 586.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 110, + 586.0, + 501, + 670.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 678, + 505, + 699 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 689 + ], + "score": 1.0, + "content": "Figure 8: Privacy (norm and cosine leak AUC) vs Utility (train loss, test loss, and test AUC) trade-off of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 689, + 479, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 479, + 700 + ], + "score": 1.0, + "content": "protection methods (Marvell, iso, no noise, max norm) at the cut layer and first layer on Criteo.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + } + ], + "index": 10.25 + } + ], + "page_idx": 22, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "As shown in Figure 7,8,9, Marvell consistently outperforms the isotropic Gaussian baseline over", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "all the different privacy-utility definitions. In addition, our proposed heuristic max norm is also", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 416, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 416, + 116 + ], + "score": 1.0, + "content": "particularly effective against our identified norm and direction-based attacks.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 82, + 506, + 116 + ] + }, + { + "type": "image", + "bbox": [ + 110, + 124, + 501, + 374 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 124, + 501, + 374 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 124, + 501, + 374 + ], + "spans": [ + { + "bbox": [ + 110, + 124, + 501, + 374 + ], + "score": 0.973, + "type": "image", + "image_path": "cfa83c054f2045646208abae64f5e5969c5183c3b6697991cd25d9d7634217b2.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 110, + 124, + 501, + 207.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 110, + 207.33333333333331, + 501, + 290.66666666666663 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 110, + 290.66666666666663, + 501, + 373.99999999999994 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 381, + 505, + 403 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 382, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 382, + 505, + 393 + ], + "score": 1.0, + "content": "Figure 7: Privacy (norm and cosine leak AUC) vs Utility (train loss, test loss, and test AUC) trade-off of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 392, + 480, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 480, + 404 + ], + "score": 1.0, + "content": "protection methods (Marvell, iso, no noise, max norm) at the cut layer and first layer on Avazu.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + } + ], + "index": 5.25 + }, + { + "type": "image", + "bbox": [ + 110, + 418, + 501, + 670 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 418, + 501, + 670 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 418, + 501, + 670 + ], + "spans": [ + { + "bbox": [ + 110, + 418, + 501, + 670 + ], + "score": 0.971, + "type": "image", + "image_path": "f9b37071e310cce66c89d73ed7335527adaa39502b1e85c60894e68a77ab15fc.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 110, + 418, + 501, + 502.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 110, + 502.0, + 501, + 586.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 110, + 586.0, + 501, + 670.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 678, + 505, + 699 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 689 + ], + "score": 1.0, + "content": "Figure 8: Privacy (norm and cosine leak AUC) vs Utility (train loss, test loss, and test AUC) trade-off of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 689, + 479, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 479, + 700 + ], + "score": 1.0, + "content": "protection methods (Marvell, iso, no noise, max norm) at the cut layer and first layer on Criteo.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + } + ], + "index": 10.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 78, + 500, + 331 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 78, + 500, + 331 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 78, + 500, + 331 + ], + "spans": [ + { + "bbox": [ + 110, + 78, + 500, + 331 + ], + "score": 0.975, + "type": "image", + "image_path": "6397972f53ef2b60ce1c15ca92d0c6d5017d66a8e99b3d79b27208a0b0d8b18f.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 78, + 500, + 162.33333333333331 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 162.33333333333331, + 500, + 246.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 246.66666666666663, + 500, + 330.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 338, + 504, + 360 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 338, + 506, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 506, + 350 + ], + "score": 1.0, + "content": "Figure 9: Privacy (norm and cosine leak AUC) vs Utility (train loss, test loss, and test AUC) trade-off of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 348, + 474, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 474, + 360 + ], + "score": 1.0, + "content": "protection methods (Marvell, iso, no noise, max norm) at the cut layer and first layer on ISIC.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + } + ], + "page_idx": 23, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 78, + 500, + 331 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 78, + 500, + 331 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 78, + 500, + 331 + ], + "spans": [ + { + "bbox": [ + 110, + 78, + 500, + 331 + ], + "score": 0.975, + "type": "image", + "image_path": "6397972f53ef2b60ce1c15ca92d0c6d5017d66a8e99b3d79b27208a0b0d8b18f.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 78, + 500, + 162.33333333333331 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 162.33333333333331, + 500, + 246.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 246.66666666666663, + 500, + 330.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 338, + 504, + 360 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 338, + 506, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 506, + 350 + ], + "score": 1.0, + "content": "Figure 9: Privacy (norm and cosine leak AUC) vs Utility (train loss, test loss, and test AUC) trade-off of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 348, + 474, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 474, + 360 + ], + "score": 1.0, + "content": "protection methods (Marvell, iso, no noise, max norm) at the cut layer and first layer on ISIC.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 492, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 494, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 494, + 95 + ], + "score": 1.0, + "content": "A.7.3 IMPACT OF ACTIVATION GRADIENTS’ LAYER LOCATION ON THE CORRESPONDING", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 140, + 92, + 193, + 106 + ], + "spans": [ + { + "bbox": [ + 140, + 92, + 193, + 106 + ], + "score": 1.0, + "content": "LEAK AUC", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 112, + 506, + 310 + ], + "lines": [ + { + "bbox": [ + 106, + 113, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 113, + 505, + 124 + ], + "score": 1.0, + "content": "In Section 5.1, we have shown that the activation gradients of not only the cut layer but also the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 124, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 506, + 136 + ], + "score": 1.0, + "content": "first layer can leak the label if no protection is applied. In this section, we analyze the effect of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 135, + 506, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 147 + ], + "score": 1.0, + "content": "this layer location on the degree of label leakage under our proposed attacks measured through leak", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 146, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 505, + 157 + ], + "score": 1.0, + "content": "AUC. Here, for the convolutional neural network trained on ISIC, we have allocated 4 Conv-ReLU-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 156, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 506, + 169 + ], + "score": 1.0, + "content": "MaxPool layers on the non-label party side. Here each of these 4 layers’ activation (from ReLU", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "score": 1.0, + "content": "output) gradients can be used to infer the label. In Figure 10, we plot the progression of norm and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 178, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 505, + 191 + ], + "score": 1.0, + "content": "cosine leak AUC for layer 2 and 3 in addition to the first layer (layer 1) and the cut layer (layer", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 189, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 315, + 202 + ], + "score": 1.0, + "content": "4) when applying no protection or Marvell with", + "type": "text" + }, + { + "bbox": [ + 315, + 190, + 342, + 200 + ], + "score": 0.89, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 189, + 506, + 202 + ], + "score": 1.0, + "content": ". We notice that when no protection is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "score": 1.0, + "content": "applied, the norm and cosine leak AUC are approximately the same across the layers and both", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 212, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 505, + 224 + ], + "score": 1.0, + "content": "at a very high level, indicating a strong degree of label leakage. In contrast, when Marvell is", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "score": 1.0, + "content": "applied, the leak AUCs are much lowered to a reasonable level of around 0.6 throughout all the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "score": 1.0, + "content": "layers. Empirically, we notice that the leak AUCs for a given training iteration are close among all", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 245, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 505, + 256 + ], + "score": 1.0, + "content": "the layers with a maximum difference of around 0.05. Here it is unclear whether the earlier layers’", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 255, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 506, + 267 + ], + "score": 1.0, + "content": "gradients would leak the labels more than the cut layer or not — for norm leak AUC, the earlier", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 266, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 505, + 279 + ], + "score": 1.0, + "content": "layers have a higher AUC values (more leakage) than the cut layer, while for cosine leak AUC, the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "earlier layers have a lower AUC values (less leakage) than the cut layer. Further understanding the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 288, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 301 + ], + "score": 1.0, + "content": "relationship between these leak AUC values across different layers is an open question motivated by", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 299, + 197, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 197, + 311 + ], + "score": 1.0, + "content": "our observations here.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 10.5 + }, + { + "type": "image", + "bbox": [ + 106, + 327, + 506, + 442 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 177, + 318, + 443, + 327 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 179, + 317, + 443, + 328 + ], + "spans": [ + { + "bbox": [ + 179, + 317, + 443, + 328 + ], + "score": 1.0, + "content": "ISIC, non-label party's convolution network f with four Conv-ReLU-MaxPool layers", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "image_body", + "bbox": [ + 106, + 327, + 506, + 442 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 327, + 506, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 506, + 442 + ], + "score": 0.961, + "type": "image", + "image_path": "56ed010866be0a341dc9aca02c15880bab8d679f13f2a8fb1257db4f8e6de8e9.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 106, + 327, + 506, + 365.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 106, + 365.3333333333333, + 506, + 403.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 106, + 403.66666666666663, + 506, + 441.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 450, + 505, + 520 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "Figure 10: For a non-label party’s four-layer convolutional architecture trained on ISIC, we plot the progression", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 460, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 506, + 472 + ], + "score": 1.0, + "content": "of the norm or cosine leak AUC computed using the activation gradients from each of the four layers throughout", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "training in (a), (b) when no random protection is applied and in (c), (d) when using Marvell with privacy", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 480, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 165, + 491 + ], + "score": 1.0, + "content": "hyperparameter", + "type": "text" + }, + { + "bbox": [ + 165, + 481, + 190, + 490 + ], + "score": 0.9, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 480, + 505, + 491 + ], + "score": 1.0, + "content": "at the cut layer (layer 4). The four curves overlap in (b) at the value of 1.0. For each", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 490, + 504, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 504, + 501 + ], + "score": 1.0, + "content": "layer’s data in each figure, a 1-d Gaussian kernel with standard deviation of 5 is convolved with the raw 1-d", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "score": 1.0, + "content": "array of leak AUC values to smooth out the fluctuations and make the different layers’ degree of leakage more", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 510, + 168, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 168, + 522 + ], + "score": 1.0, + "content": "visually distinct.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 535, + 505, + 601 + ], + "lines": [ + { + "bbox": [ + 106, + 536, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 505, + 547 + ], + "score": 1.0, + "content": "Beyond the four layer convolutional neural network trained on ISIC, we also train a new MLP model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "architecture on Criteo. Here instead of having 3 128-unit ReLU-activated MLP layers for the non-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 555, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 555, + 196, + 570 + ], + "score": 1.0, + "content": "label party’s function", + "type": "text" + }, + { + "bbox": [ + 196, + 558, + 204, + 568 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 555, + 436, + 570 + ], + "score": 1.0, + "content": "as described in Section A.6.2, we use 8 such layers for", + "type": "text" + }, + { + "bbox": [ + 436, + 558, + 443, + 568 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 555, + 506, + 570 + ], + "score": 1.0, + "content": "while keeping", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 568, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 505, + 581 + ], + "score": 1.0, + "content": "every other architectural component the same. We plot the progression of leak AUC values on each", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "of the eight layers in Figure 11. Here the observations are very similar to what we see in Figure 10,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 591, + 343, + 602 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 343, + 602 + ], + "score": 1.0, + "content": "confirming our aforementioned conclusions in this section.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33.5 + } + ], + "page_idx": 24, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 492, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 494, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 494, + 95 + ], + "score": 1.0, + "content": "A.7.3 IMPACT OF ACTIVATION GRADIENTS’ LAYER LOCATION ON THE CORRESPONDING", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 140, + 92, + 193, + 106 + ], + "spans": [ + { + "bbox": [ + 140, + 92, + 193, + 106 + ], + "score": 1.0, + "content": "LEAK AUC", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 494, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 112, + 506, + 310 + ], + "lines": [ + { + "bbox": [ + 106, + 113, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 113, + 505, + 124 + ], + "score": 1.0, + "content": "In Section 5.1, we have shown that the activation gradients of not only the cut layer but also the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 124, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 506, + 136 + ], + "score": 1.0, + "content": "first layer can leak the label if no protection is applied. In this section, we analyze the effect of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 135, + 506, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 147 + ], + "score": 1.0, + "content": "this layer location on the degree of label leakage under our proposed attacks measured through leak", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 146, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 505, + 157 + ], + "score": 1.0, + "content": "AUC. Here, for the convolutional neural network trained on ISIC, we have allocated 4 Conv-ReLU-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 156, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 506, + 169 + ], + "score": 1.0, + "content": "MaxPool layers on the non-label party side. Here each of these 4 layers’ activation (from ReLU", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "score": 1.0, + "content": "output) gradients can be used to infer the label. In Figure 10, we plot the progression of norm and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 178, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 505, + 191 + ], + "score": 1.0, + "content": "cosine leak AUC for layer 2 and 3 in addition to the first layer (layer 1) and the cut layer (layer", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 189, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 315, + 202 + ], + "score": 1.0, + "content": "4) when applying no protection or Marvell with", + "type": "text" + }, + { + "bbox": [ + 315, + 190, + 342, + 200 + ], + "score": 0.89, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 189, + 506, + 202 + ], + "score": 1.0, + "content": ". We notice that when no protection is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "score": 1.0, + "content": "applied, the norm and cosine leak AUC are approximately the same across the layers and both", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 212, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 505, + 224 + ], + "score": 1.0, + "content": "at a very high level, indicating a strong degree of label leakage. In contrast, when Marvell is", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 506, + 235 + ], + "score": 1.0, + "content": "applied, the leak AUCs are much lowered to a reasonable level of around 0.6 throughout all the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "score": 1.0, + "content": "layers. Empirically, we notice that the leak AUCs for a given training iteration are close among all", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 245, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 505, + 256 + ], + "score": 1.0, + "content": "the layers with a maximum difference of around 0.05. Here it is unclear whether the earlier layers’", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 255, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 506, + 267 + ], + "score": 1.0, + "content": "gradients would leak the labels more than the cut layer or not — for norm leak AUC, the earlier", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 266, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 505, + 279 + ], + "score": 1.0, + "content": "layers have a higher AUC values (more leakage) than the cut layer, while for cosine leak AUC, the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "earlier layers have a lower AUC values (less leakage) than the cut layer. Further understanding the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 288, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 301 + ], + "score": 1.0, + "content": "relationship between these leak AUC values across different layers is an open question motivated by", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 299, + 197, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 197, + 311 + ], + "score": 1.0, + "content": "our observations here.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 113, + 506, + 311 + ] + }, + { + "type": "image", + "bbox": [ + 106, + 327, + 506, + 442 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 177, + 318, + 443, + 327 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 179, + 317, + 443, + 328 + ], + "spans": [ + { + "bbox": [ + 179, + 317, + 443, + 328 + ], + "score": 1.0, + "content": "ISIC, non-label party's convolution network f with four Conv-ReLU-MaxPool layers", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "image_body", + "bbox": [ + 106, + 327, + 506, + 442 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 327, + 506, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 506, + 442 + ], + "score": 0.961, + "type": "image", + "image_path": "56ed010866be0a341dc9aca02c15880bab8d679f13f2a8fb1257db4f8e6de8e9.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 106, + 327, + 506, + 365.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 106, + 365.3333333333333, + 506, + 403.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 106, + 403.66666666666663, + 506, + 441.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 450, + 505, + 520 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "Figure 10: For a non-label party’s four-layer convolutional architecture trained on ISIC, we plot the progression", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 460, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 460, + 506, + 472 + ], + "score": 1.0, + "content": "of the norm or cosine leak AUC computed using the activation gradients from each of the four layers throughout", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "training in (a), (b) when no random protection is applied and in (c), (d) when using Marvell with privacy", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 480, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 165, + 491 + ], + "score": 1.0, + "content": "hyperparameter", + "type": "text" + }, + { + "bbox": [ + 165, + 481, + 190, + 490 + ], + "score": 0.9, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 480, + 505, + 491 + ], + "score": 1.0, + "content": "at the cut layer (layer 4). The four curves overlap in (b) at the value of 1.0. For each", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 490, + 504, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 504, + 501 + ], + "score": 1.0, + "content": "layer’s data in each figure, a 1-d Gaussian kernel with standard deviation of 5 is convolved with the raw 1-d", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "score": 1.0, + "content": "array of leak AUC values to smooth out the fluctuations and make the different layers’ degree of leakage more", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 510, + 168, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 168, + 522 + ], + "score": 1.0, + "content": "visually distinct.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 535, + 505, + 601 + ], + "lines": [ + { + "bbox": [ + 106, + 536, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 505, + 547 + ], + "score": 1.0, + "content": "Beyond the four layer convolutional neural network trained on ISIC, we also train a new MLP model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "architecture on Criteo. Here instead of having 3 128-unit ReLU-activated MLP layers for the non-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 555, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 555, + 196, + 570 + ], + "score": 1.0, + "content": "label party’s function", + "type": "text" + }, + { + "bbox": [ + 196, + 558, + 204, + 568 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 555, + 436, + 570 + ], + "score": 1.0, + "content": "as described in Section A.6.2, we use 8 such layers for", + "type": "text" + }, + { + "bbox": [ + 436, + 558, + 443, + 568 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 555, + 506, + 570 + ], + "score": 1.0, + "content": "while keeping", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 568, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 505, + 581 + ], + "score": 1.0, + "content": "every other architectural component the same. We plot the progression of leak AUC values on each", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "of the eight layers in Figure 11. Here the observations are very similar to what we see in Figure 10,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 591, + 343, + 602 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 343, + 602 + ], + "score": 1.0, + "content": "confirming our aforementioned conclusions in this section.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33.5, + "bbox_fs": [ + 104, + 536, + 506, + 602 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 88, + 505, + 207 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 88, + 505, + 207 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 88, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 88, + 505, + 207 + ], + "score": 0.93, + "type": "image", + "image_path": "02c92cdadf09570bf4017dec0ff34b8c73f98c3538f14856b782c4449ed8450b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 88, + 505, + 127.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 127.66666666666666, + 505, + 167.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 167.33333333333331, + 505, + 206.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 214, + 506, + 285 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 214, + 506, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 227 + ], + "score": 1.0, + "content": "Figure 11: For a non-label party’s eight-layer MLP architecture trained on Criteo, we plot the progression of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 225, + 506, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 236 + ], + "score": 1.0, + "content": "the norm or cosine leak AUC computed using the activation gradients from each of the eight layers throughout", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 234, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 247 + ], + "score": 1.0, + "content": "training in (a), (b) when no random protection is applied and in (c), (d) when using Marvell with privacy", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 245, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 165, + 256 + ], + "score": 1.0, + "content": "hyperparameter", + "type": "text" + }, + { + "bbox": [ + 165, + 245, + 189, + 254 + ], + "score": 0.89, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 245, + 505, + 256 + ], + "score": 1.0, + "content": "at the cut layer (layer 4). The eight curves overlap in (b) at the value of 1.0. For each", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 255, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 266 + ], + "score": 1.0, + "content": "layer’s data in each figure, a 1-d Gaussian kernel with standard deviation of 100 is convolved with the raw 1-d", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 263, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 277 + ], + "score": 1.0, + "content": "array of leak AUC values to smooth out the fluctuations and make the different layers’ degree of leakage more", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 274, + 169, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 169, + 287 + ], + "score": 1.0, + "content": "visually distinct.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + } + ], + "page_idx": 25, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "26", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 170, + 79, + 447, + 88 + ], + "lines": [ + { + "bbox": [ + 171, + 78, + 448, + 90 + ], + "spans": [ + { + "bbox": [ + 171, + 78, + 448, + 90 + ], + "score": 1.0, + "content": "Criteo, non-label party's MLP network f with eight 128-unit ReLU-activated MLP layers", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 88, + 505, + 207 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 88, + 505, + 207 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 88, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 88, + 505, + 207 + ], + "score": 0.93, + "type": "image", + "image_path": "02c92cdadf09570bf4017dec0ff34b8c73f98c3538f14856b782c4449ed8450b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 88, + 505, + 127.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 127.66666666666666, + 505, + 167.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 167.33333333333331, + 505, + 206.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 214, + 506, + 285 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 214, + 506, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 227 + ], + "score": 1.0, + "content": "Figure 11: For a non-label party’s eight-layer MLP architecture trained on Criteo, we plot the progression of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 225, + 506, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 236 + ], + "score": 1.0, + "content": "the norm or cosine leak AUC computed using the activation gradients from each of the eight layers throughout", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 234, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 247 + ], + "score": 1.0, + "content": "training in (a), (b) when no random protection is applied and in (c), (d) when using Marvell with privacy", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 245, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 165, + 256 + ], + "score": 1.0, + "content": "hyperparameter", + "type": "text" + }, + { + "bbox": [ + 165, + 245, + 189, + 254 + ], + "score": 0.89, + "content": "s = 4", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 245, + 505, + 256 + ], + "score": 1.0, + "content": "at the cut layer (layer 4). The eight curves overlap in (b) at the value of 1.0. For each", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 255, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 266 + ], + "score": 1.0, + "content": "layer’s data in each figure, a 1-d Gaussian kernel with standard deviation of 100 is convolved with the raw 1-d", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 263, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 277 + ], + "score": 1.0, + "content": "array of leak AUC values to smooth out the fluctuations and make the different layers’ degree of leakage more", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 274, + 169, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 169, + 287 + ], + "score": 1.0, + "content": "visually distinct.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 102, + 505, + 158 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 505, + 116 + ], + "score": 1.0, + "content": "It is possible to use a local differential privacy definition to analyze our problem setup if we treat any", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 114, + 504, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 114, + 493, + 126 + ], + "score": 1.0, + "content": "single training example as a dataset and two examples (dataset) are adjacent if they share the same", + "type": "text" + }, + { + "bbox": [ + 494, + 114, + 504, + 124 + ], + "score": 0.79, + "content": "X", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 216, + 137 + ], + "score": 1.0, + "content": "but have the opposite label", + "type": "text" + }, + { + "bbox": [ + 216, + 127, + 223, + 136 + ], + "score": 0.7, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 125, + 505, + 137 + ], + "score": 1.0, + "content": ". Then one can analyze the activation gradient computation function’s", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 104, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "local DP properties. However, we choose not to use local DP in our setup for the following two", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 148, + 141, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 141, + 159 + ], + "score": 1.0, + "content": "reasons:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 130, + 164, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 130, + 164, + 504, + 176 + ], + "spans": [ + { + "bbox": [ + 130, + 164, + 504, + 176 + ], + "score": 1.0, + "content": "1. We are not aware of any attacks associated with local DP that are suitable for practical", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 142, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 142, + 175, + 505, + 187 + ], + "score": 1.0, + "content": "use in our problem setup. In contrast, our leak AUC privacy quantification metrics have", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 142, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 142, + 186, + 505, + 199 + ], + "score": 1.0, + "content": "corresponding concrete, realistic attack methods that the non-label party can use in practice.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 195, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 130, + 198, + 140, + 207 + ], + "score": 1.0, + "content": "2.", + "type": "text" + }, + { + "bbox": [ + 140, + 195, + 505, + 210 + ], + "score": 1.0, + "content": "Even if there exists a practical attack associated with local DP, this attack would require", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 208, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 141, + 208, + 505, + 221 + ], + "score": 1.0, + "content": "example-level side information as the activation gradient distributions would be example-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 218, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 141, + 218, + 505, + 231 + ], + "score": 1.0, + "content": "specific. Distinguishing between such gradient distributions for a pair of adjacent examples", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 142, + 230, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 142, + 230, + 190, + 243 + ], + "score": 1.0, + "content": "(with same", + "type": "text" + }, + { + "bbox": [ + 190, + 230, + 200, + 240 + ], + "score": 0.8, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 230, + 276, + 243 + ], + "score": 1.0, + "content": "but opposite label", + "type": "text" + }, + { + "bbox": [ + 277, + 231, + 284, + 241 + ], + "score": 0.55, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 230, + 505, + 243 + ], + "score": 1.0, + "content": ") would require knowledge about the two distributions", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 240, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 141, + 240, + 505, + 254 + ], + "score": 1.0, + "content": "specific to this example pair itself. As we have mentioned in Section 3.2, we do not assume", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 142, + 252, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 142, + 252, + 505, + 264 + ], + "score": 1.0, + "content": "the non-label party would realistically have access to such fine-grained knowledge.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 262, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 141, + 262, + 505, + 276 + ], + "score": 1.0, + "content": "Instead, in our setup, the non-label party uses a scoring function that only takes in the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 274, + 420, + 286 + ], + "spans": [ + { + "bbox": [ + 142, + 274, + 420, + 286 + ], + "score": 1.0, + "content": "communicated cut layer gradient without any additional information.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10 + } + ], + "page_idx": 26, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 111, + 83, + 373, + 93 + ], + "lines": [ + { + "bbox": [ + 109, + 82, + 374, + 95 + ], + "spans": [ + { + "bbox": [ + 109, + 82, + 374, + 95 + ], + "score": 1.0, + "content": "A.8 REASONS WHY WE DON’T USE LOCAL DP IN OUR SETUP", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 102, + 505, + 158 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 505, + 116 + ], + "score": 1.0, + "content": "It is possible to use a local differential privacy definition to analyze our problem setup if we treat any", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 114, + 504, + 126 + ], + "spans": [ + { + "bbox": [ + 106, + 114, + 493, + 126 + ], + "score": 1.0, + "content": "single training example as a dataset and two examples (dataset) are adjacent if they share the same", + "type": "text" + }, + { + "bbox": [ + 494, + 114, + 504, + 124 + ], + "score": 0.79, + "content": "X", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 216, + 137 + ], + "score": 1.0, + "content": "but have the opposite label", + "type": "text" + }, + { + "bbox": [ + 216, + 127, + 223, + 136 + ], + "score": 0.7, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 125, + 505, + 137 + ], + "score": 1.0, + "content": ". Then one can analyze the activation gradient computation function’s", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 104, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "local DP properties. However, we choose not to use local DP in our setup for the following two", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 148, + 141, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 141, + 159 + ], + "score": 1.0, + "content": "reasons:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 104, + 102, + 506, + 159 + ] + }, + { + "type": "text", + "bbox": [ + 130, + 164, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 130, + 164, + 504, + 176 + ], + "spans": [ + { + "bbox": [ + 130, + 164, + 504, + 176 + ], + "score": 1.0, + "content": "1. We are not aware of any attacks associated with local DP that are suitable for practical", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 142, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 142, + 175, + 505, + 187 + ], + "score": 1.0, + "content": "use in our problem setup. In contrast, our leak AUC privacy quantification metrics have", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 142, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 142, + 186, + 505, + 199 + ], + "score": 1.0, + "content": "corresponding concrete, realistic attack methods that the non-label party can use in practice.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 195, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 130, + 198, + 140, + 207 + ], + "score": 1.0, + "content": "2.", + "type": "text" + }, + { + "bbox": [ + 140, + 195, + 505, + 210 + ], + "score": 1.0, + "content": "Even if there exists a practical attack associated with local DP, this attack would require", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 208, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 141, + 208, + 505, + 221 + ], + "score": 1.0, + "content": "example-level side information as the activation gradient distributions would be example-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 218, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 141, + 218, + 505, + 231 + ], + "score": 1.0, + "content": "specific. Distinguishing between such gradient distributions for a pair of adjacent examples", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 142, + 230, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 142, + 230, + 190, + 243 + ], + "score": 1.0, + "content": "(with same", + "type": "text" + }, + { + "bbox": [ + 190, + 230, + 200, + 240 + ], + "score": 0.8, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 230, + 276, + 243 + ], + "score": 1.0, + "content": "but opposite label", + "type": "text" + }, + { + "bbox": [ + 277, + 231, + 284, + 241 + ], + "score": 0.55, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 230, + 505, + 243 + ], + "score": 1.0, + "content": ") would require knowledge about the two distributions", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 240, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 141, + 240, + 505, + 254 + ], + "score": 1.0, + "content": "specific to this example pair itself. As we have mentioned in Section 3.2, we do not assume", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 142, + 252, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 142, + 252, + 505, + 264 + ], + "score": 1.0, + "content": "the non-label party would realistically have access to such fine-grained knowledge.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 262, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 141, + 262, + 505, + 276 + ], + "score": 1.0, + "content": "Instead, in our setup, the non-label party uses a scoring function that only takes in the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 274, + 420, + 286 + ], + "spans": [ + { + "bbox": [ + 142, + 274, + 420, + 286 + ], + "score": 1.0, + "content": "communicated cut layer gradient without any additional information.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10, + "bbox_fs": [ + 130, + 164, + 505, + 286 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/dev/cOtBRgsf2fO/cOtBRgsf2fO_model.json b/parse/dev/cOtBRgsf2fO/cOtBRgsf2fO_model.json new file mode 100644 index 0000000000000000000000000000000000000000..ff3b754069fda21926240bd1e59f02ba76e9b25e --- /dev/null +++ b/parse/dev/cOtBRgsf2fO/cOtBRgsf2fO_model.json @@ -0,0 +1,53269 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 529, + 1302, + 529, + 1302, + 955, + 398, + 955 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1355, + 1403, + 1355, + 1403, + 1780, + 298, + 1780 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1034, + 1404, + 1034, + 1404, + 1339, + 298, + 1339 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 300, + 1797, + 1400, + 1797, + 1400, + 1950, + 300, + 1950 + ], + "score": 0.973 + }, + { + "category_id": 0, + "poly": [ + 298, + 220, + 1090, + 220, + 1090, + 323, + 298, + 323 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 315, + 373, + 999, + 373, + 999, + 438, + 315, + 438 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 1081, + 378, + 1372, + 378, + 1372, + 437, + 1081, + 437 + ], + "score": 0.925 + }, + { + "category_id": 0, + "poly": [ + 301, + 990, + 574, + 990, + 574, + 1023, + 301, + 1023 + ], + "score": 0.899 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 816, + 76, + 816, + 104, + 300, + 104 + ], + "score": 0.88 + }, + { + "category_id": 0, + "poly": [ + 773, + 479, + 927, + 479, + 927, + 512, + 773, + 512 + ], + "score": 0.831 + }, + { + "category_id": 2, + "poly": [ + 326, + 1973, + 1310, + 1973, + 1310, + 2033, + 326, + 2033 + ], + "score": 0.696 + }, + { + "category_id": 2, + "poly": [ + 842, + 2089, + 857, + 2089, + 857, + 2112, + 842, + 2112 + ], + "score": 0.677 + }, + { + "category_id": 2, + "poly": [ + 326, + 1973, + 1309, + 1973, + 1309, + 2033, + 326, + 2033 + ], + "score": 0.503 + }, + { + "category_id": 13, + "poly": [ + 382, + 374, + 430, + 374, + 430, + 406, + 382, + 406 + ], + "score": 0.84, + "latex": "\\mathbf { L i } ^ { 1 \\dagger }" + }, + { + "category_id": 13, + "poly": [ + 525, + 374, + 590, + 374, + 590, + 406, + 525, + 406 + ], + "score": 0.83, + "latex": "\\mathbf { S u n ^ { 2 \\ S } }" + }, + { + "category_id": 13, + "poly": [ + 709, + 1217, + 735, + 1217, + 735, + 1243, + 709, + 1243 + ], + "score": 0.82, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 406, + 1217, + 432, + 1217, + 432, + 1243, + 406, + 1243 + ], + "score": 0.82, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 1096, + 1277, + 1122, + 1277, + 1122, + 1304, + 1096, + 1304 + ], + "score": 0.78, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 703, + 1278, + 729, + 1278, + 729, + 1305, + 703, + 1305 + ], + "score": 0.78, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 1009, + 1309, + 1035, + 1309, + 1035, + 1335, + 1009, + 1335 + ], + "score": 0.76, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 297, + 1126, + 334, + 1126, + 334, + 1152, + 297, + 1152 + ], + "score": 0.74, + "latex": "F L" + }, + { + "category_id": 13, + "poly": [ + 326, + 1309, + 350, + 1309, + 350, + 1335, + 326, + 1335 + ], + "score": 0.7, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 965, + 1187, + 989, + 1187, + 989, + 1213, + 965, + 1213 + ], + "score": 0.68, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 603, + 407, + 653, + 407, + 653, + 436, + 603, + 436 + ], + "score": 0.64, + "latex": "\\mathbf { X i e } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 478, + 1126, + 515, + 1126, + 515, + 1153, + 478, + 1153 + ], + "score": 0.61, + "latex": "F L" + }, + { + "category_id": 13, + "poly": [ + 1157, + 1247, + 1182, + 1247, + 1182, + 1274, + 1157, + 1274 + ], + "score": 0.6, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 811, + 375, + 869, + 375, + 869, + 405, + 811, + 405 + ], + "score": 0.31, + "latex": "\\mathbf { G a o } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 601, + 374, + 714, + 374, + 714, + 407, + 601, + 407 + ], + "score": 0.29, + "latex": "\\mathbf { X i n \\ Y a n g ^ { 2 } }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 220.0, + 1092.0, + 220.0, + 1092.0, + 268.0, + 295.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 271.0, + 917.0, + 271.0, + 917.0, + 327.0, + 296.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 986.0, + 579.0, + 986.0, + 579.0, + 1031.0, + 293.0, + 1031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 475.0, + 934.0, + 475.0, + 934.0, + 518.0, + 768.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1967.0, + 1309.0, + 1967.0, + 1309.0, + 2009.0, + 326.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 2000.0, + 990.0, + 2000.0, + 990.0, + 2034.0, + 332.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 2088.0, + 860.0, + 2088.0, + 860.0, + 2118.0, + 841.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1967.0, + 1309.0, + 1967.0, + 1309.0, + 2009.0, + 326.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1999.0, + 991.0, + 1999.0, + 991.0, + 2037.0, + 331.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 529.0, + 1305.0, + 529.0, + 1305.0, + 561.0, + 394.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 558.0, + 1306.0, + 558.0, + 1306.0, + 596.0, + 391.0, + 596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 590.0, + 1306.0, + 590.0, + 1306.0, + 625.0, + 393.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 621.0, + 1305.0, + 621.0, + 1305.0, + 655.0, + 394.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 648.0, + 1305.0, + 648.0, + 1305.0, + 687.0, + 394.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 681.0, + 1305.0, + 681.0, + 1305.0, + 715.0, + 395.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 711.0, + 1304.0, + 711.0, + 1304.0, + 745.0, + 394.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 744.0, + 1305.0, + 744.0, + 1305.0, + 774.0, + 393.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 774.0, + 1304.0, + 774.0, + 1304.0, + 806.0, + 395.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 805.0, + 1305.0, + 805.0, + 1305.0, + 837.0, + 395.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 834.0, + 1304.0, + 834.0, + 1304.0, + 866.0, + 395.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 863.0, + 1306.0, + 863.0, + 1306.0, + 898.0, + 395.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 895.0, + 1305.0, + 895.0, + 1305.0, + 928.0, + 393.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 924.0, + 1288.0, + 924.0, + 1288.0, + 962.0, + 392.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1352.0, + 1404.0, + 1352.0, + 1404.0, + 1390.0, + 293.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1382.0, + 1405.0, + 1382.0, + 1405.0, + 1420.0, + 296.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1414.0, + 1407.0, + 1414.0, + 1407.0, + 1451.0, + 292.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1444.0, + 1403.0, + 1444.0, + 1403.0, + 1479.0, + 294.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1477.0, + 1404.0, + 1477.0, + 1404.0, + 1508.0, + 294.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1506.0, + 1407.0, + 1506.0, + 1407.0, + 1542.0, + 291.0, + 1542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1535.0, + 1404.0, + 1535.0, + 1404.0, + 1570.0, + 294.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1565.0, + 1405.0, + 1565.0, + 1405.0, + 1602.0, + 293.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1599.0, + 1403.0, + 1599.0, + 1403.0, + 1631.0, + 296.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1628.0, + 1405.0, + 1628.0, + 1405.0, + 1663.0, + 294.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1658.0, + 1400.0, + 1658.0, + 1400.0, + 1690.0, + 294.0, + 1690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1689.0, + 1405.0, + 1689.0, + 1405.0, + 1723.0, + 293.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1718.0, + 1406.0, + 1718.0, + 1406.0, + 1754.0, + 292.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1749.0, + 1316.0, + 1749.0, + 1316.0, + 1785.0, + 294.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1032.0, + 1406.0, + 1032.0, + 1406.0, + 1068.0, + 296.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1062.0, + 1405.0, + 1062.0, + 1405.0, + 1098.0, + 293.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1094.0, + 1406.0, + 1094.0, + 1406.0, + 1128.0, + 292.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1123.0, + 477.0, + 1123.0, + 477.0, + 1159.0, + 335.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1123.0, + 1405.0, + 1123.0, + 1405.0, + 1159.0, + 516.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1152.0, + 1405.0, + 1152.0, + 1405.0, + 1188.0, + 294.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1185.0, + 964.0, + 1185.0, + 964.0, + 1219.0, + 292.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1185.0, + 1404.0, + 1185.0, + 1404.0, + 1219.0, + 990.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1213.0, + 405.0, + 1213.0, + 405.0, + 1254.0, + 291.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1213.0, + 708.0, + 1213.0, + 708.0, + 1254.0, + 433.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1213.0, + 1406.0, + 1213.0, + 1406.0, + 1254.0, + 736.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1247.0, + 1156.0, + 1247.0, + 1156.0, + 1279.0, + 296.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1247.0, + 1405.0, + 1247.0, + 1405.0, + 1279.0, + 1183.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1278.0, + 702.0, + 1278.0, + 702.0, + 1309.0, + 296.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1278.0, + 1095.0, + 1278.0, + 1095.0, + 1309.0, + 730.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1278.0, + 1405.0, + 1278.0, + 1405.0, + 1309.0, + 1123.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1305.0, + 325.0, + 1305.0, + 325.0, + 1343.0, + 293.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1305.0, + 1008.0, + 1305.0, + 1008.0, + 1343.0, + 351.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1305.0, + 1202.0, + 1305.0, + 1202.0, + 1343.0, + 1036.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1796.0, + 1403.0, + 1796.0, + 1403.0, + 1833.0, + 295.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1825.0, + 1405.0, + 1825.0, + 1405.0, + 1863.0, + 293.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1858.0, + 1404.0, + 1858.0, + 1404.0, + 1891.0, + 296.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1887.0, + 1402.0, + 1887.0, + 1402.0, + 1924.0, + 294.0, + 1924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1917.0, + 1404.0, + 1917.0, + 1404.0, + 1954.0, + 295.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 370.0, + 381.0, + 370.0, + 381.0, + 410.0, + 313.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 370.0, + 524.0, + 370.0, + 524.0, + 410.0, + 431.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 370.0, + 600.0, + 370.0, + 600.0, + 410.0, + 591.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 370.0, + 810.0, + 370.0, + 810.0, + 410.0, + 715.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 370.0, + 881.0, + 370.0, + 881.0, + 410.0, + 870.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 404.0, + 602.0, + 404.0, + 602.0, + 442.0, + 324.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 404.0, + 997.0, + 404.0, + 997.0, + 442.0, + 654.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 373.0, + 1374.0, + 373.0, + 1374.0, + 412.0, + 1079.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 403.0, + 1374.0, + 403.0, + 1374.0, + 443.0, + 1197.0, + 443.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1561, + 1404, + 1561, + 1404, + 2047, + 297, + 2047 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1148, + 1405, + 1148, + 1405, + 1544, + 297, + 1544 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 576, + 1403, + 576, + 1403, + 700, + 299, + 700 + ], + "score": 0.975 + }, + { + "category_id": 3, + "poly": [ + 307, + 218, + 1386, + 218, + 1386, + 509, + 307, + 509 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 316, + 716, + 1406, + 716, + 1406, + 1069, + 316, + 1069 + ], + "score": 0.958 + }, + { + "category_id": 4, + "poly": [ + 295, + 511, + 1396, + 511, + 1396, + 567, + 295, + 567 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 300, + 1093, + 589, + 1093, + 589, + 1128, + 300, + 1128 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.689 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.346 + }, + { + "category_id": 13, + "poly": [ + 827, + 542, + 844, + 542, + 844, + 566, + 827, + 566 + ], + "score": 0.81, + "latex": "g" + }, + { + "category_id": 14, + "poly": [ + 828, + 440, + 1368, + 440, + 1368, + 515, + 828, + 515 + ], + "score": 0.33, + "latex": "\\begin{array} { l } { \\displaystyle { g : = \\nabla _ { f ( X ) } L \\frac { \\mathsf { L } } { \\mathsf { \\Gamma } _ { z } [ ( z ) ] _ { z = f ( X ) } } \\frac { L } { d \\ell } = ( \\widetilde { p } _ { 1 } - y ) \\qquad - \\frac { d L } { d L } = 1 } } \\\\ { \\displaystyle { \\quad = ( \\widetilde { p } _ { 1 } - y ) \\nabla _ { z } [ ( z ) ] _ { z = f ( X ) } \\qquad } } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 223.0, + 564.0, + 223.0, + 564.0, + 251.0, + 411.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 228.0, + 810.0, + 228.0, + 810.0, + 261.0, + 714.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 216.0, + 1190.0, + 216.0, + 1190.0, + 254.0, + 1075.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 266.0, + 434.0, + 266.0, + 434.0, + 309.0, + 338.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 256.0, + 524.0, + 256.0, + 524.0, + 276.0, + 504.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 268.0, + 661.0, + 268.0, + 661.0, + 299.0, + 578.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 265.0, + 842.0, + 265.0, + 842.0, + 306.0, + 675.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 278.0, + 948.0, + 278.0, + 948.0, + 309.0, + 893.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 266.0, + 1032.0, + 266.0, + 1032.0, + 293.0, + 1007.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 250.0, + 1215.0, + 250.0, + 1215.0, + 311.0, + 1064.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 252.0, + 1383.0, + 252.0, + 1383.0, + 306.0, + 1246.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 307.0, + 526.0, + 307.0, + 526.0, + 394.0, + 498.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 315.0, + 769.0, + 315.0, + 769.0, + 329.0, + 754.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 305.0, + 936.0, + 305.0, + 936.0, + 387.0, + 904.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 310.0, + 1316.0, + 310.0, + 1316.0, + 329.0, + 1294.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 340.0, + 401.0, + 340.0, + 401.0, + 365.0, + 374.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 340.0, + 833.0, + 340.0, + 833.0, + 388.0, + 692.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 340.0, + 1147.0, + 340.0, + 1147.0, + 371.0, + 1118.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 340.0, + 1321.0, + 340.0, + 1321.0, + 388.0, + 1169.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 411.0, + 403.0, + 411.0, + 403.0, + 498.0, + 316.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 400.0, + 560.0, + 400.0, + 560.0, + 440.0, + 463.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 374.0, + 827.0, + 374.0, + 827.0, + 506.0, + 587.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 381.0, + 1218.0, + 381.0, + 1218.0, + 427.0, + 1177.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 388.0, + 1269.0, + 388.0, + 1269.0, + 415.0, + 1228.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 451.0, + 1243.0, + 451.0, + 1243.0, + 502.0, + 1085.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 506.0, + 1400.0, + 506.0, + 1400.0, + 543.0, + 294.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 535.0, + 826.0, + 535.0, + 826.0, + 569.0, + 295.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 535.0, + 1180.0, + 535.0, + 1180.0, + 569.0, + 845.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1090.0, + 592.0, + 1090.0, + 592.0, + 1134.0, + 292.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1559.0, + 1404.0, + 1559.0, + 1404.0, + 1595.0, + 294.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1588.0, + 1406.0, + 1588.0, + 1406.0, + 1627.0, + 294.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1622.0, + 1404.0, + 1622.0, + 1404.0, + 1654.0, + 297.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1653.0, + 1402.0, + 1653.0, + 1402.0, + 1683.0, + 299.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1679.0, + 1405.0, + 1679.0, + 1405.0, + 1716.0, + 292.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1710.0, + 1404.0, + 1710.0, + 1404.0, + 1747.0, + 295.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1743.0, + 1405.0, + 1743.0, + 1405.0, + 1777.0, + 295.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1772.0, + 1404.0, + 1772.0, + 1404.0, + 1805.0, + 294.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1804.0, + 1404.0, + 1804.0, + 1404.0, + 1838.0, + 295.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1833.0, + 1405.0, + 1833.0, + 1405.0, + 1867.0, + 295.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1866.0, + 1405.0, + 1866.0, + 1405.0, + 1897.0, + 296.0, + 1897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1896.0, + 1404.0, + 1896.0, + 1404.0, + 1929.0, + 293.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1927.0, + 1402.0, + 1927.0, + 1402.0, + 1958.0, + 296.0, + 1958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1955.0, + 1405.0, + 1955.0, + 1405.0, + 1990.0, + 295.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1986.0, + 1402.0, + 1986.0, + 1402.0, + 2020.0, + 295.0, + 2020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2018.0, + 908.0, + 2018.0, + 908.0, + 2049.0, + 293.0, + 2049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1148.0, + 1405.0, + 1148.0, + 1405.0, + 1181.0, + 296.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1177.0, + 1403.0, + 1177.0, + 1403.0, + 1212.0, + 294.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1207.0, + 1408.0, + 1207.0, + 1408.0, + 1245.0, + 291.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1240.0, + 1403.0, + 1240.0, + 1403.0, + 1272.0, + 296.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1271.0, + 1403.0, + 1271.0, + 1403.0, + 1303.0, + 295.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1301.0, + 1405.0, + 1301.0, + 1405.0, + 1333.0, + 296.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1329.0, + 1406.0, + 1329.0, + 1406.0, + 1365.0, + 295.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1360.0, + 1405.0, + 1360.0, + 1405.0, + 1396.0, + 292.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1390.0, + 1403.0, + 1390.0, + 1403.0, + 1425.0, + 292.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1421.0, + 1405.0, + 1421.0, + 1405.0, + 1457.0, + 294.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1453.0, + 1403.0, + 1453.0, + 1403.0, + 1486.0, + 295.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1484.0, + 1405.0, + 1484.0, + 1405.0, + 1516.0, + 296.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1513.0, + 1403.0, + 1513.0, + 1403.0, + 1549.0, + 295.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 575.0, + 1405.0, + 575.0, + 1405.0, + 610.0, + 294.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 608.0, + 1403.0, + 608.0, + 1403.0, + 640.0, + 295.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 636.0, + 1404.0, + 636.0, + 1404.0, + 673.0, + 293.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 670.0, + 1403.0, + 670.0, + 1403.0, + 702.0, + 295.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 714.0, + 1401.0, + 714.0, + 1401.0, + 752.0, + 316.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 744.0, + 1405.0, + 744.0, + 1405.0, + 782.0, + 349.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 777.0, + 771.0, + 777.0, + 771.0, + 813.0, + 350.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 817.0, + 345.0, + 817.0, + 345.0, + 844.0, + 318.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 813.0, + 1400.0, + 813.0, + 1400.0, + 847.0, + 350.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 844.0, + 1063.0, + 844.0, + 1063.0, + 880.0, + 350.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 882.0, + 344.0, + 882.0, + 344.0, + 909.0, + 318.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 881.0, + 1404.0, + 881.0, + 1404.0, + 913.0, + 352.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 909.0, + 1404.0, + 909.0, + 1404.0, + 945.0, + 350.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 940.0, + 1405.0, + 940.0, + 1405.0, + 976.0, + 352.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 967.0, + 1182.0, + 967.0, + 1182.0, + 1010.0, + 348.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1010.0, + 345.0, + 1010.0, + 345.0, + 1035.0, + 318.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1006.0, + 1406.0, + 1006.0, + 1406.0, + 1042.0, + 348.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1039.0, + 1304.0, + 1039.0, + 1304.0, + 1071.0, + 352.0, + 1071.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 958, + 1404, + 958, + 1404, + 1270, + 297, + 1270 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 296, + 424, + 1404, + 424, + 1404, + 679, + 296, + 679 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 694, + 1403, + 694, + 1403, + 945, + 297, + 945 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1556, + 1404, + 1556, + 1404, + 1842, + 297, + 1842 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1418, + 1403, + 1418, + 1403, + 1541, + 299, + 1541 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 298, + 1857, + 1403, + 1857, + 1403, + 1952, + 298, + 1952 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 293, + 286, + 1400, + 286, + 1400, + 349, + 293, + 349 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 298, + 1340, + 1401, + 1340, + 1401, + 1403, + 298, + 1403 + ], + "score": 0.944 + }, + { + "category_id": 2, + "poly": [ + 297, + 1978, + 1403, + 1978, + 1403, + 2034, + 297, + 2034 + ], + "score": 0.94 + }, + { + "category_id": 0, + "poly": [ + 299, + 226, + 890, + 226, + 890, + 262, + 299, + 262 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 104, + 300, + 104 + ], + "score": 0.886 + }, + { + "category_id": 0, + "poly": [ + 298, + 1293, + 947, + 1293, + 947, + 1324, + 298, + 1324 + ], + "score": 0.86 + }, + { + "category_id": 0, + "poly": [ + 299, + 377, + 1062, + 377, + 1062, + 408, + 299, + 408 + ], + "score": 0.753 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.674 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.314 + }, + { + "category_id": 1, + "poly": [ + 299, + 377, + 1062, + 377, + 1062, + 408, + 299, + 408 + ], + "score": 0.141 + }, + { + "category_id": 13, + "poly": [ + 1189, + 1020, + 1255, + 1020, + 1255, + 1054, + 1189, + 1054 + ], + "score": 0.95, + "latex": "f ( X )" + }, + { + "category_id": 13, + "poly": [ + 737, + 455, + 865, + 455, + 865, + 489, + 737, + 489 + ], + "score": 0.93, + "latex": "\\mathcal { X } \\times \\{ 0 , \\bar { 1 } \\}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1709, + 496, + 1709, + 496, + 1747, + 297, + 1747 + ], + "score": 0.93, + "latex": "q : \\mathbb { R } ^ { d } \\{ 0 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 981, + 524, + 1124, + 524, + 1124, + 559, + 981, + 559 + ], + "score": 0.93, + "latex": "y \\in \\{ 0 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 1162, + 1779, + 1266, + 1779, + 1266, + 1812, + 1162, + 1812 + ], + "score": 0.93, + "latex": "q ( g ) = 0" + }, + { + "category_id": 13, + "poly": [ + 1010, + 1779, + 1111, + 1779, + 1111, + 1812, + 1010, + 1812 + ], + "score": 0.93, + "latex": "\\overset { \\cdot } { r } ( g ) > t" + }, + { + "category_id": 13, + "poly": [ + 1197, + 524, + 1363, + 524, + 1363, + 559, + 1197, + 559 + ], + "score": 0.92, + "latex": "\\ell \\ = \\ h ( f ( X ) )" + }, + { + "category_id": 13, + "poly": [ + 879, + 1779, + 982, + 1779, + 982, + 1812, + 879, + 1812 + ], + "score": 0.92, + "latex": "\\bar { q ( g ) } = 1" + }, + { + "category_id": 13, + "poly": [ + 299, + 753, + 482, + 753, + 482, + 791, + 299, + 791 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\frac { d L } { d \\ell } \\ = \\ ( \\widetilde { p } _ { 1 } - y ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1294, + 1779, + 1394, + 1779, + 1394, + 1812, + 1294, + 1812 + ], + "score": 0.92, + "latex": "r ( g ) \\leq t" + }, + { + "category_id": 13, + "poly": [ + 522, + 846, + 588, + 846, + 588, + 880, + 522, + 880 + ], + "score": 0.92, + "latex": "f ( X )" + }, + { + "category_id": 13, + "poly": [ + 505, + 486, + 650, + 486, + 650, + 521, + 505, + 521 + ], + "score": 0.92, + "latex": "f : \\mathcal { X } \\to \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1327, + 1587, + 1392, + 1587, + 1392, + 1621, + 1327, + 1621 + ], + "score": 0.92, + "latex": "f ( X )" + }, + { + "category_id": 13, + "poly": [ + 1335, + 616, + 1400, + 616, + 1400, + 650, + 1335, + 650 + ], + "score": 0.91, + "latex": "{ \\dot { \\boldsymbol { f } } } ( { \\boldsymbol { X } } )" + }, + { + "category_id": 13, + "poly": [ + 758, + 1678, + 822, + 1678, + 822, + 1712, + 758, + 1712 + ], + "score": 0.91, + "latex": "( g , y )" + }, + { + "category_id": 13, + "poly": [ + 583, + 1744, + 727, + 1744, + 727, + 1778, + 583, + 1778 + ], + "score": 0.91, + "latex": "r : \\mathbb { R } ^ { d } \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 684, + 1051, + 755, + 1051, + 755, + 1084, + 684, + 1084 + ], + "score": 0.91, + "latex": "\\mathbb { R } ^ { B \\times d }" + }, + { + "category_id": 13, + "poly": [ + 496, + 1780, + 564, + 1780, + 564, + 1808, + 496, + 1808 + ], + "score": 0.91, + "latex": "t \\in \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 1012, + 490, + 1100, + 490, + 1100, + 520, + 1012, + 520 + ], + "score": 0.9, + "latex": "X \\in { \\mathcal { X } }" + }, + { + "category_id": 13, + "poly": [ + 1101, + 426, + 1170, + 426, + 1170, + 457, + 1101, + 457 + ], + "score": 0.9, + "latex": "h \\circ f" + }, + { + "category_id": 13, + "poly": [ + 1255, + 1917, + 1309, + 1917, + 1309, + 1948, + 1255, + 1948 + ], + "score": 0.9, + "latex": "P ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 1139, + 1917, + 1193, + 1917, + 1193, + 1948, + 1139, + 1948 + ], + "score": 0.9, + "latex": "P ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 506, + 522, + 674, + 522, + 674, + 554, + 506, + 554 + ], + "score": 0.89, + "latex": "h : \\mathbb { R } ^ { d } \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 297, + 586, + 571, + 586, + 571, + 618, + 297, + 618 + ], + "score": 0.88, + "latex": "\\widetilde { p } _ { 1 } = 1 / ( 1 + \\exp ( - \\ell ) )" + }, + { + "category_id": 13, + "poly": [ + 297, + 619, + 700, + 619, + 700, + 650, + 297, + 650 + ], + "score": 0.87, + "latex": "L = \\log ( 1 + \\exp ( - \\ell ) ) + ( 1 - y ) \\ell" + }, + { + "category_id": 13, + "poly": [ + 296, + 878, + 847, + 878, + 847, + 918, + 296, + 918 + ], + "score": 0.86, + "latex": "g : = \\nabla _ { f ( X ) } L = ( \\widetilde { p } _ { 1 } - y ) \\nabla _ { z } h ( z ) | _ { z = f ( X ) } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1123, + 1420, + 1143, + 1420, + 1143, + 1451, + 1123, + 1451 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 586, + 817, + 605, + 817, + 605, + 847, + 586, + 847 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 377, + 1179, + 397, + 1179, + 397, + 1210, + 377, + 1210 + ], + "score": 0.83, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 800, + 916, + 819, + 916, + 819, + 944, + 800, + 944 + ], + "score": 0.81, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 671, + 1453, + 689, + 1453, + 689, + 1480, + 671, + 1480 + ], + "score": 0.81, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 618, + 1515, + 636, + 1515, + 636, + 1542, + 618, + 1542 + ], + "score": 0.8, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 544, + 787, + 563, + 787, + 563, + 813, + 544, + 813 + ], + "score": 0.79, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 1279, + 1483, + 1298, + 1483, + 1298, + 1512, + 1279, + 1512 + ], + "score": 0.79, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1319, + 756, + 1342, + 756, + 1342, + 782, + 1319, + 782 + ], + "score": 0.79, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 495, + 1651, + 513, + 1651, + 513, + 1679, + 495, + 1679 + ], + "score": 0.79, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 404, + 1061, + 422, + 1061, + 422, + 1088, + 404, + 1088 + ], + "score": 0.78, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 1039, + 1453, + 1058, + 1453, + 1058, + 1480, + 1039, + 1480 + ], + "score": 0.77, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 395, + 991, + 421, + 991, + 421, + 1017, + 395, + 1017 + ], + "score": 0.77, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 883, + 1894, + 900, + 1894, + 900, + 1915, + 883, + 1915 + ], + "score": 0.75, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 1195, + 1419, + 1215, + 1419, + 1215, + 1447, + 1195, + 1447 + ], + "score": 0.75, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 1106, + 726, + 1128, + 726, + 1128, + 752, + 1106, + 752 + ], + "score": 0.72, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 1219, + 886, + 1238, + 886, + 1238, + 914, + 1219, + 914 + ], + "score": 0.7, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 1377, + 1721, + 1394, + 1721, + 1394, + 1742, + 1377, + 1742 + ], + "score": 0.63, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 408, + 1752, + 423, + 1752, + 423, + 1778, + 408, + 1778 + ], + "score": 0.62, + "latex": "t" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1968.0, + 1408.0, + 1968.0, + 1408.0, + 2016.0, + 325.0, + 2016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2001.0, + 1405.0, + 2001.0, + 1405.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 222.0, + 894.0, + 222.0, + 894.0, + 267.0, + 291.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 816.0, + 72.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1290.0, + 950.0, + 1290.0, + 950.0, + 1330.0, + 293.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 375.0, + 1065.0, + 375.0, + 1065.0, + 412.0, + 294.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 958.0, + 1404.0, + 958.0, + 1404.0, + 994.0, + 296.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 989.0, + 394.0, + 989.0, + 394.0, + 1021.0, + 296.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 989.0, + 1404.0, + 989.0, + 1404.0, + 1021.0, + 422.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1019.0, + 1188.0, + 1019.0, + 1188.0, + 1056.0, + 292.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 1019.0, + 1405.0, + 1019.0, + 1405.0, + 1056.0, + 1256.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1042.0, + 403.0, + 1042.0, + 403.0, + 1097.0, + 287.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1042.0, + 683.0, + 1042.0, + 683.0, + 1097.0, + 423.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 1042.0, + 1410.0, + 1042.0, + 1410.0, + 1097.0, + 756.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1083.0, + 1406.0, + 1083.0, + 1406.0, + 1120.0, + 295.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1114.0, + 1405.0, + 1114.0, + 1405.0, + 1155.0, + 292.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1144.0, + 1404.0, + 1144.0, + 1404.0, + 1184.0, + 294.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1174.0, + 376.0, + 1174.0, + 376.0, + 1214.0, + 294.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1174.0, + 1405.0, + 1174.0, + 1405.0, + 1214.0, + 398.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1207.0, + 1405.0, + 1207.0, + 1405.0, + 1244.0, + 294.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1238.0, + 1402.0, + 1238.0, + 1402.0, + 1274.0, + 295.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 422.0, + 1100.0, + 422.0, + 1100.0, + 463.0, + 291.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 422.0, + 1404.0, + 422.0, + 1404.0, + 463.0, + 1171.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 452.0, + 736.0, + 452.0, + 736.0, + 492.0, + 294.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 452.0, + 1404.0, + 452.0, + 1404.0, + 492.0, + 866.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 486.0, + 504.0, + 486.0, + 504.0, + 527.0, + 291.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 486.0, + 1011.0, + 486.0, + 1011.0, + 527.0, + 651.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 486.0, + 1405.0, + 486.0, + 1405.0, + 527.0, + 1101.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 521.0, + 505.0, + 521.0, + 505.0, + 561.0, + 293.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 521.0, + 980.0, + 521.0, + 980.0, + 561.0, + 675.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 521.0, + 1196.0, + 521.0, + 1196.0, + 561.0, + 1125.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1364.0, + 521.0, + 1405.0, + 521.0, + 1405.0, + 561.0, + 1364.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 553.0, + 1405.0, + 553.0, + 1405.0, + 592.0, + 293.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 584.0, + 296.0, + 584.0, + 296.0, + 624.0, + 290.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 584.0, + 1406.0, + 584.0, + 1406.0, + 624.0, + 572.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 616.0, + 1334.0, + 616.0, + 1334.0, + 654.0, + 701.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 648.0, + 1371.0, + 648.0, + 1371.0, + 682.0, + 296.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 690.0, + 1407.0, + 690.0, + 1407.0, + 731.0, + 292.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 724.0, + 1105.0, + 724.0, + 1105.0, + 759.0, + 295.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 724.0, + 1405.0, + 724.0, + 1405.0, + 759.0, + 1129.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 752.0, + 1318.0, + 752.0, + 1318.0, + 794.0, + 483.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.0, + 752.0, + 1405.0, + 752.0, + 1405.0, + 794.0, + 1343.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 787.0, + 543.0, + 787.0, + 543.0, + 818.0, + 292.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 787.0, + 1405.0, + 787.0, + 1405.0, + 818.0, + 564.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 815.0, + 585.0, + 815.0, + 585.0, + 851.0, + 291.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 815.0, + 1406.0, + 815.0, + 1406.0, + 851.0, + 606.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 845.0, + 521.0, + 845.0, + 521.0, + 883.0, + 292.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 845.0, + 1405.0, + 845.0, + 1405.0, + 883.0, + 589.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 874.0, + 295.0, + 874.0, + 295.0, + 919.0, + 287.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 874.0, + 1218.0, + 874.0, + 1218.0, + 919.0, + 848.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 874.0, + 1410.0, + 874.0, + 1410.0, + 919.0, + 1239.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 912.0, + 799.0, + 912.0, + 799.0, + 949.0, + 292.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 912.0, + 1403.0, + 912.0, + 1403.0, + 949.0, + 820.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1554.0, + 1405.0, + 1554.0, + 1405.0, + 1593.0, + 295.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1587.0, + 1326.0, + 1587.0, + 1326.0, + 1624.0, + 294.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1393.0, + 1587.0, + 1404.0, + 1587.0, + 1404.0, + 1624.0, + 1393.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1613.0, + 1405.0, + 1613.0, + 1405.0, + 1656.0, + 292.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1646.0, + 494.0, + 1646.0, + 494.0, + 1685.0, + 292.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1646.0, + 1405.0, + 1646.0, + 1405.0, + 1685.0, + 514.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1676.0, + 757.0, + 1676.0, + 757.0, + 1715.0, + 292.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1676.0, + 1405.0, + 1676.0, + 1405.0, + 1715.0, + 823.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1706.0, + 296.0, + 1706.0, + 296.0, + 1753.0, + 290.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 1706.0, + 1376.0, + 1706.0, + 1376.0, + 1753.0, + 497.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1706.0, + 1409.0, + 1706.0, + 1409.0, + 1753.0, + 1395.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1745.0, + 407.0, + 1745.0, + 407.0, + 1786.0, + 292.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1745.0, + 582.0, + 1745.0, + 582.0, + 1786.0, + 424.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1745.0, + 1405.0, + 1745.0, + 1405.0, + 1786.0, + 728.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1779.0, + 495.0, + 1779.0, + 495.0, + 1816.0, + 294.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 1779.0, + 878.0, + 1779.0, + 878.0, + 1816.0, + 565.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1779.0, + 1009.0, + 1779.0, + 1009.0, + 1816.0, + 983.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1779.0, + 1161.0, + 1779.0, + 1161.0, + 1816.0, + 1112.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 1779.0, + 1293.0, + 1779.0, + 1293.0, + 1816.0, + 1267.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1779.0, + 1405.0, + 1779.0, + 1405.0, + 1816.0, + 1395.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1810.0, + 1402.0, + 1810.0, + 1402.0, + 1844.0, + 295.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1419.0, + 1122.0, + 1419.0, + 1122.0, + 1452.0, + 297.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1419.0, + 1194.0, + 1419.0, + 1194.0, + 1452.0, + 1144.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1419.0, + 1404.0, + 1419.0, + 1404.0, + 1452.0, + 1216.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1448.0, + 670.0, + 1448.0, + 670.0, + 1481.0, + 295.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 1448.0, + 1038.0, + 1448.0, + 1038.0, + 1481.0, + 690.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1448.0, + 1404.0, + 1448.0, + 1404.0, + 1481.0, + 1059.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1480.0, + 1278.0, + 1480.0, + 1278.0, + 1514.0, + 292.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1480.0, + 1404.0, + 1480.0, + 1404.0, + 1514.0, + 1299.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1511.0, + 617.0, + 1511.0, + 617.0, + 1543.0, + 295.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1511.0, + 945.0, + 1511.0, + 945.0, + 1543.0, + 637.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1853.0, + 1404.0, + 1853.0, + 1404.0, + 1896.0, + 293.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1889.0, + 882.0, + 1889.0, + 882.0, + 1920.0, + 294.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 1889.0, + 1404.0, + 1889.0, + 1404.0, + 1920.0, + 901.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1917.0, + 1138.0, + 1917.0, + 1138.0, + 1957.0, + 293.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1917.0, + 1254.0, + 1917.0, + 1254.0, + 1957.0, + 1194.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 1917.0, + 1405.0, + 1917.0, + 1405.0, + 1957.0, + 1310.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 286.0, + 1403.0, + 286.0, + 1403.0, + 321.0, + 296.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 315.0, + 1362.0, + 315.0, + 1362.0, + 353.0, + 293.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1340.0, + 1405.0, + 1340.0, + 1405.0, + 1375.0, + 296.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1372.0, + 1396.0, + 1372.0, + 1396.0, + 1407.0, + 296.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 375.0, + 1065.0, + 375.0, + 1065.0, + 412.0, + 294.0, + 412.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1221, + 1405, + 1221, + 1405, + 1498, + 296, + 1498 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 296, + 962, + 1404, + 962, + 1404, + 1208, + 296, + 1208 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 754, + 1403, + 754, + 1403, + 912, + 297, + 912 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 299, + 221, + 1364, + 221, + 1364, + 665, + 299, + 665 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 1559, + 1404, + 1559, + 1404, + 1656, + 299, + 1656 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 595, + 917, + 1105, + 917, + 1105, + 959, + 595, + 959 + ], + "score": 0.931 + }, + { + "category_id": 4, + "poly": [ + 294, + 680, + 1401, + 680, + 1401, + 738, + 294, + 738 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 313, + 1658, + 1403, + 1658, + 1403, + 1935, + 313, + 1935 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.892 + }, + { + "category_id": 0, + "poly": [ + 300, + 1518, + 735, + 1518, + 735, + 1549, + 300, + 1549 + ], + "score": 0.885 + }, + { + "category_id": 1, + "poly": [ + 316, + 1942, + 1404, + 1942, + 1404, + 2037, + 316, + 2037 + ], + "score": 0.877 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2111, + 841, + 2111 + ], + "score": 0.761 + }, + { + "category_id": 1, + "poly": [ + 1253, + 921, + 1400, + 921, + 1400, + 953, + 1253, + 953 + ], + "score": 0.701 + }, + { + "category_id": 13, + "poly": [ + 714, + 1594, + 806, + 1594, + 806, + 1628, + 714, + 1628 + ], + "score": 0.93, + "latex": "| \\widetilde { p } _ { 1 } - y |" + }, + { + "category_id": 13, + "poly": [ + 871, + 2003, + 1049, + 2003, + 1049, + 2038, + 871, + 2038 + ], + "score": 0.93, + "latex": "\\nabla _ { a } h ( a ) | _ { a = f ( X ) }" + }, + { + "category_id": 13, + "poly": [ + 908, + 1559, + 1341, + 1559, + 1341, + 1597, + 908, + 1597 + ], + "score": 0.93, + "latex": "\\| g \\| _ { 2 } = | \\widetilde p _ { 1 } - y | \\cdot \\| \\nabla _ { a } h ( a ) | _ { a = \\pmb { f } ( X ) } \\| _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1153, + 1941, + 1372, + 1941, + 1372, + 1978, + 1153, + 1978 + ], + "score": 0.93, + "latex": "\\| \\nabla _ { z } h ( z ) \\vert _ { z = f ( X ) } \\| _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 847, + 680, + 847, + 680, + 884, + 297, + 884 + ], + "score": 0.92, + "latex": "\\mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } ( \\{ g : r ( g ) > t \\} )" + }, + { + "category_id": 13, + "poly": [ + 738, + 846, + 1121, + 846, + 1121, + 884, + 738, + 884 + ], + "score": 0.92, + "latex": "\\mathrm { T P R } _ { r } ( t ) : = P ^ { ( 1 ) } ( \\{ g : r ( g ) > t \\} )" + }, + { + "category_id": 13, + "poly": [ + 862, + 1595, + 1081, + 1595, + 1081, + 1630, + 862, + 1630 + ], + "score": 0.92, + "latex": "\\| \\nabla _ { a } h ( a ) | _ { a = f ( X ) } \\| _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 581, + 1720, + 778, + 1720, + 778, + 1751, + 581, + 1751 + ], + "score": 0.91, + "latex": "1 - \\widetilde { p } _ { 1 } = \\left| \\widetilde { p } _ { 1 } - y \\right|" + }, + { + "category_id": 13, + "poly": [ + 1279, + 816, + 1335, + 816, + 1335, + 849, + 1279, + 849 + ], + "score": 0.91, + "latex": "( r , t )" + }, + { + "category_id": 13, + "poly": [ + 1168, + 787, + 1249, + 787, + 1249, + 814, + 1168, + 814 + ], + "score": 0.91, + "latex": "t \\in \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 858, + 1721, + 928, + 1721, + 928, + 1751, + 858, + 1751 + ], + "score": 0.9, + "latex": "y = 1" + }, + { + "category_id": 13, + "poly": [ + 587, + 1752, + 852, + 1752, + 852, + 1783, + 587, + 1783 + ], + "score": 0.9, + "latex": "1 - \\widetilde { p } _ { 0 } = \\widetilde { p } _ { 1 } = | \\widetilde { p } _ { 1 } - y |" + }, + { + "category_id": 14, + "poly": [ + 592, + 915, + 1107, + 915, + 1107, + 959, + 592, + 959 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\begin{array} { r l } { \\mathrm { A U C } ( r ) = \\int _ { \\infty } ^ { - \\infty } \\mathrm { T P R } _ { r } ( t ) ~ d \\mathrm { F P R } _ { r } ( t ) } & { \\in [ 0 , 1 ] } \\end{array} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 408, + 785, + 811, + 785, + 811, + 818, + 408, + 818 + ], + "score": 0.89, + "latex": "t \\mapsto \\mathsf { \\bar { ( F P R } } _ { r } ( t ) , \\mathsf { T P R } _ { r } ( t ) ) \\in \\mathsf { [ 0 , 1 ] } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 932, + 1753, + 1001, + 1753, + 1001, + 1783, + 932, + 1783 + ], + "score": 0.89, + "latex": "y = 0" + }, + { + "category_id": 13, + "poly": [ + 1270, + 2009, + 1288, + 2009, + 1288, + 2035, + 1270, + 2035 + ], + "score": 0.78, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1222, + 762, + 1240, + 762, + 1240, + 783, + 1222, + 783 + ], + "score": 0.78, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 901, + 999, + 918, + 999, + 918, + 1020, + 901, + 1020 + ], + "score": 0.78, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 777, + 1120, + 794, + 1120, + 794, + 1142, + 777, + 1142 + ], + "score": 0.76, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 603, + 888, + 618, + 888, + 618, + 908, + 603, + 908 + ], + "score": 0.74, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 741, + 1149, + 759, + 1149, + 759, + 1172, + 741, + 1172 + ], + "score": 0.72, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 375, + 1056, + 389, + 1056, + 389, + 1081, + 375, + 1081 + ], + "score": 0.69, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 946, + 1090, + 964, + 1090, + 964, + 1111, + 946, + 1111 + ], + "score": 0.67, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 748, + 883, + 841, + 883, + 841, + 914, + 748, + 914 + ], + "score": 0.65, + "latex": "\\operatorname { A U C } ( r )" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 240.0, + 344.0, + 240.0, + 344.0, + 335.0, + 324.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 233.0, + 643.0, + 233.0, + 643.0, + 257.0, + 622.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 237.0, + 698.0, + 237.0, + 698.0, + 334.0, + 675.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 234.0, + 993.0, + 234.0, + 993.0, + 243.0, + 975.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 240.0, + 1054.0, + 240.0, + 1054.0, + 335.0, + 1034.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 314.0, + 836.0, + 314.0, + 836.0, + 331.0, + 822.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 322.0, + 362.0, + 322.0, + 362.0, + 340.0, + 338.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 322.0, + 716.0, + 322.0, + 716.0, + 339.0, + 692.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.0, + 322.0, + 1072.0, + 322.0, + 1072.0, + 340.0, + 1049.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 336.0, + 362.0, + 336.0, + 362.0, + 354.0, + 338.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 336.0, + 717.0, + 336.0, + 717.0, + 353.0, + 693.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 336.0, + 1072.0, + 336.0, + 1072.0, + 354.0, + 1048.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 349.0, + 363.0, + 349.0, + 363.0, + 369.0, + 338.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 349.0, + 717.0, + 349.0, + 717.0, + 368.0, + 692.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 351.0, + 1072.0, + 351.0, + 1072.0, + 369.0, + 1048.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 364.0, + 363.0, + 364.0, + 363.0, + 383.0, + 338.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 364.0, + 717.0, + 364.0, + 717.0, + 382.0, + 692.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 363.0, + 1074.0, + 363.0, + 1074.0, + 385.0, + 1045.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 376.0, + 364.0, + 376.0, + 364.0, + 398.0, + 335.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 376.0, + 720.0, + 376.0, + 720.0, + 398.0, + 691.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 379.0, + 1073.0, + 379.0, + 1073.0, + 397.0, + 1046.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 399.0, + 409.0, + 399.0, + 409.0, + 420.0, + 386.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 401.0, + 454.0, + 401.0, + 454.0, + 420.0, + 431.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 399.0, + 499.0, + 399.0, + 499.0, + 420.0, + 477.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 401.0, + 545.0, + 401.0, + 545.0, + 420.0, + 521.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 401.0, + 590.0, + 401.0, + 590.0, + 420.0, + 567.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 399.0, + 635.0, + 399.0, + 635.0, + 420.0, + 613.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 399.0, + 751.0, + 399.0, + 751.0, + 420.0, + 728.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 397.0, + 785.0, + 397.0, + 785.0, + 420.0, + 757.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 401.0, + 817.0, + 401.0, + 817.0, + 419.0, + 793.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 394.0, + 948.0, + 394.0, + 948.0, + 422.0, + 819.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 401.0, + 976.0, + 401.0, + 976.0, + 419.0, + 949.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 399.0, + 1006.0, + 399.0, + 1006.0, + 419.0, + 983.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 398.0, + 1091.0, + 398.0, + 1091.0, + 421.0, + 1060.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 401.0, + 1120.0, + 401.0, + 1120.0, + 420.0, + 1095.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 401.0, + 1154.0, + 401.0, + 1154.0, + 420.0, + 1131.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 401.0, + 1191.0, + 401.0, + 1191.0, + 420.0, + 1165.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 401.0, + 1223.0, + 401.0, + 1223.0, + 419.0, + 1200.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 401.0, + 1259.0, + 401.0, + 1259.0, + 420.0, + 1234.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 401.0, + 1293.0, + 401.0, + 1293.0, + 420.0, + 1268.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 401.0, + 1328.0, + 401.0, + 1328.0, + 420.0, + 1304.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 401.0, + 1361.0, + 401.0, + 1361.0, + 419.0, + 1339.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 416.0, + 327.0, + 416.0, + 327.0, + 444.0, + 295.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 413.0, + 541.0, + 413.0, + 541.0, + 442.0, + 474.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 414.0, + 683.0, + 414.0, + 683.0, + 443.0, + 648.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 796.0, + 409.0, + 924.0, + 409.0, + 924.0, + 441.0, + 796.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 415.0, + 1038.0, + 415.0, + 1038.0, + 444.0, + 1004.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 411.0, + 1242.0, + 411.0, + 1242.0, + 442.0, + 1191.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 436.0, + 539.0, + 436.0, + 539.0, + 530.0, + 520.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 454.0, + 840.0, + 454.0, + 840.0, + 473.0, + 636.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 435.0, + 893.0, + 435.0, + 893.0, + 532.0, + 870.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 459.0, + 1080.0, + 459.0, + 1080.0, + 472.0, + 1056.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 470.0, + 1081.0, + 470.0, + 1081.0, + 485.0, + 1053.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 496.0, + 557.0, + 496.0, + 557.0, + 516.0, + 534.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 500.0, + 910.0, + 500.0, + 910.0, + 515.0, + 890.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 527.0, + 557.0, + 527.0, + 557.0, + 544.0, + 533.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 527.0, + 913.0, + 527.0, + 913.0, + 544.0, + 889.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 555.0, + 557.0, + 555.0, + 557.0, + 572.0, + 533.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 556.0, + 913.0, + 556.0, + 913.0, + 573.0, + 889.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 585.0, + 556.0, + 585.0, + 556.0, + 598.0, + 535.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 585.0, + 910.0, + 585.0, + 910.0, + 600.0, + 890.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 611.0, + 558.0, + 611.0, + 558.0, + 629.0, + 533.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 613.0, + 913.0, + 613.0, + 913.0, + 630.0, + 887.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 641.0, + 524.0, + 641.0, + 524.0, + 670.0, + 489.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 623.0, + 878.0, + 623.0, + 878.0, + 669.0, + 562.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 628.0, + 970.0, + 628.0, + 970.0, + 646.0, + 942.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 625.0, + 1118.0, + 625.0, + 1118.0, + 666.0, + 993.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 626.0, + 1169.0, + 626.0, + 1169.0, + 646.0, + 1147.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 678.0, + 1404.0, + 678.0, + 1404.0, + 714.0, + 294.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 709.0, + 1163.0, + 709.0, + 1163.0, + 740.0, + 296.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 294.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1516.0, + 739.0, + 1516.0, + 739.0, + 1552.0, + 295.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1219.0, + 1407.0, + 1219.0, + 1407.0, + 1257.0, + 294.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1250.0, + 1403.0, + 1250.0, + 1403.0, + 1286.0, + 294.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1285.0, + 1405.0, + 1285.0, + 1405.0, + 1318.0, + 296.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1311.0, + 1406.0, + 1311.0, + 1406.0, + 1350.0, + 293.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1345.0, + 1406.0, + 1345.0, + 1406.0, + 1382.0, + 294.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1374.0, + 1405.0, + 1374.0, + 1405.0, + 1411.0, + 293.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1405.0, + 1405.0, + 1405.0, + 1405.0, + 1439.0, + 293.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1434.0, + 1405.0, + 1434.0, + 1405.0, + 1471.0, + 293.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1466.0, + 1343.0, + 1466.0, + 1343.0, + 1500.0, + 293.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 961.0, + 1404.0, + 961.0, + 1404.0, + 998.0, + 295.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 993.0, + 900.0, + 993.0, + 900.0, + 1026.0, + 294.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 993.0, + 1405.0, + 993.0, + 1405.0, + 1026.0, + 919.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1024.0, + 1405.0, + 1024.0, + 1405.0, + 1056.0, + 293.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1050.0, + 374.0, + 1050.0, + 374.0, + 1091.0, + 291.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1050.0, + 1405.0, + 1050.0, + 1405.0, + 1091.0, + 390.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1083.0, + 945.0, + 1083.0, + 945.0, + 1119.0, + 293.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1083.0, + 1404.0, + 1083.0, + 1404.0, + 1119.0, + 965.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1111.0, + 776.0, + 1111.0, + 776.0, + 1151.0, + 293.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1111.0, + 1404.0, + 1111.0, + 1404.0, + 1151.0, + 795.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1143.0, + 740.0, + 1143.0, + 740.0, + 1182.0, + 291.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 1143.0, + 1405.0, + 1143.0, + 1405.0, + 1182.0, + 760.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1176.0, + 988.0, + 1176.0, + 988.0, + 1210.0, + 295.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 754.0, + 1221.0, + 754.0, + 1221.0, + 791.0, + 295.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 754.0, + 1404.0, + 754.0, + 1404.0, + 791.0, + 1241.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 782.0, + 407.0, + 782.0, + 407.0, + 820.0, + 293.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 782.0, + 1167.0, + 782.0, + 1167.0, + 820.0, + 812.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 782.0, + 1401.0, + 782.0, + 1401.0, + 820.0, + 1250.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 813.0, + 1278.0, + 813.0, + 1278.0, + 853.0, + 292.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 813.0, + 1406.0, + 813.0, + 1406.0, + 853.0, + 1336.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 844.0, + 296.0, + 844.0, + 296.0, + 889.0, + 292.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 844.0, + 737.0, + 844.0, + 737.0, + 889.0, + 681.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 844.0, + 1408.0, + 844.0, + 1408.0, + 889.0, + 1122.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 880.0, + 602.0, + 880.0, + 602.0, + 918.0, + 295.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 880.0, + 747.0, + 880.0, + 747.0, + 918.0, + 619.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 880.0, + 1211.0, + 880.0, + 1211.0, + 918.0, + 842.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1555.0, + 907.0, + 1555.0, + 907.0, + 1600.0, + 292.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1555.0, + 1408.0, + 1555.0, + 1408.0, + 1600.0, + 1342.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1590.0, + 713.0, + 1590.0, + 713.0, + 1632.0, + 293.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1590.0, + 861.0, + 1590.0, + 861.0, + 1632.0, + 807.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 1590.0, + 1405.0, + 1590.0, + 1405.0, + 1632.0, + 1082.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1625.0, + 735.0, + 1625.0, + 735.0, + 1659.0, + 295.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 1656.0, + 1403.0, + 1656.0, + 1403.0, + 1693.0, + 314.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1688.0, + 1405.0, + 1688.0, + 1405.0, + 1727.0, + 335.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1718.0, + 580.0, + 1718.0, + 580.0, + 1758.0, + 335.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1718.0, + 857.0, + 1718.0, + 857.0, + 1758.0, + 779.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 1718.0, + 1405.0, + 1718.0, + 1405.0, + 1758.0, + 929.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1749.0, + 586.0, + 1749.0, + 586.0, + 1789.0, + 335.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1749.0, + 931.0, + 1749.0, + 931.0, + 1789.0, + 853.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1749.0, + 1406.0, + 1749.0, + 1406.0, + 1789.0, + 1002.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1782.0, + 1403.0, + 1782.0, + 1403.0, + 1815.0, + 335.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1811.0, + 1405.0, + 1811.0, + 1405.0, + 1847.0, + 335.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1842.0, + 1406.0, + 1842.0, + 1406.0, + 1875.0, + 335.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1870.0, + 1405.0, + 1870.0, + 1405.0, + 1908.0, + 334.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1904.0, + 1382.0, + 1904.0, + 1382.0, + 1937.0, + 337.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 1937.0, + 1152.0, + 1937.0, + 1152.0, + 1981.0, + 309.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 1937.0, + 1407.0, + 1937.0, + 1407.0, + 1981.0, + 1373.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2007.0, + 337.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 2000.0, + 870.0, + 2000.0, + 870.0, + 2042.0, + 335.0, + 2042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 2000.0, + 1269.0, + 2000.0, + 1269.0, + 2042.0, + 1050.0, + 2042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 2000.0, + 1301.0, + 2000.0, + 1301.0, + 2042.0, + 1289.0, + 2042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 918.0, + 1405.0, + 918.0, + 1405.0, + 962.0, + 1251.0, + 962.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1758, + 1404, + 1758, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1231, + 1404, + 1231, + 1404, + 1508, + 298, + 1508 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 881, + 1405, + 881, + 1405, + 1217, + 297, + 1217 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1592, + 1403, + 1592, + 1403, + 1685, + 298, + 1685 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1403, + 229, + 1403, + 385, + 298, + 385 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 298, + 399, + 1404, + 399, + 1404, + 614, + 298, + 614 + ], + "score": 0.969 + }, + { + "category_id": 0, + "poly": [ + 299, + 1536, + 944, + 1536, + 944, + 1572, + 299, + 1572 + ], + "score": 0.914 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.897 + }, + { + "category_id": 0, + "poly": [ + 299, + 1702, + 825, + 1702, + 825, + 1733, + 299, + 1733 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.719 + }, + { + "category_id": 1, + "poly": [ + 313, + 619, + 1404, + 619, + 1404, + 876, + 313, + 876 + ], + "score": 0.505 + }, + { + "category_id": 13, + "poly": [ + 1255, + 1881, + 1394, + 1881, + 1394, + 1915, + 1255, + 1915 + ], + "score": 0.92, + "latex": "\\mathbb { E } [ \\tilde { g } \\mid g ] = g" + }, + { + "category_id": 13, + "poly": [ + 369, + 880, + 494, + 880, + 494, + 915, + 369, + 915 + ], + "score": 0.92, + "latex": "\\cos ( g _ { a } , g _ { b } )" + }, + { + "category_id": 13, + "poly": [ + 1044, + 1385, + 1270, + 1385, + 1270, + 1418, + 1044, + 1418 + ], + "score": 0.91, + "latex": "r _ { d } ( g ) = \\cos ( g , g _ { + } )" + }, + { + "category_id": 13, + "poly": [ + 1012, + 230, + 1070, + 230, + 1070, + 264, + 1012, + 264 + ], + "score": 0.91, + "latex": "\\| g \\| _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1301, + 261, + 1403, + 261, + 1403, + 295, + 1301, + 295 + ], + "score": 0.91, + "latex": "r _ { n } ( g ) =" + }, + { + "category_id": 13, + "poly": [ + 298, + 292, + 355, + 292, + 355, + 325, + 298, + 325 + ], + "score": 0.9, + "latex": "\\| g \\| _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 338, + 650, + 596, + 650, + 596, + 684, + 338, + 684 + ], + "score": 0.9, + "latex": "\\operatorname { s g n } ( \\widetilde { p } _ { 1 , a } - y _ { b } ) = - 1" + }, + { + "category_id": 13, + "poly": [ + 926, + 490, + 1282, + 490, + 1282, + 522, + 926, + 522 + ], + "score": 0.9, + "latex": "\\cos ( g _ { a } , g _ { b } ) = g _ { a } ^ { T } g _ { b } \\bar { / } ( \\| g _ { a } \\| _ { 2 } \\| g _ { b } \\| _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 771, + 620, + 814, + 620, + 814, + 651, + 771, + 651 + ], + "score": 0.89, + "latex": "a , b" + }, + { + "category_id": 13, + "poly": [ + 802, + 555, + 874, + 555, + 874, + 582, + 802, + 582 + ], + "score": 0.89, + "latex": "x \\geq 0" + }, + { + "category_id": 13, + "poly": [ + 699, + 752, + 1196, + 752, + 1196, + 789, + 699, + 789 + ], + "score": 0.88, + "latex": "\\cos ( \\nabla _ { z } h ( z ) | _ { z = \\pmb { f } ( X _ { a } ) } , \\nabla _ { z } h ( z ) | _ { z = \\pmb { f } ( X _ { b } ) } ) > 0" + }, + { + "category_id": 13, + "poly": [ + 1209, + 619, + 1395, + 619, + 1395, + 653, + 1209, + 653 + ], + "score": 0.88, + "latex": "\\mathrm { s g n } ( \\widetilde { p } _ { 1 , a } - y _ { a } ) ~ ." + }, + { + "category_id": 13, + "poly": [ + 981, + 429, + 1185, + 429, + 1185, + 463, + 981, + 463 + ], + "score": 0.88, + "latex": "( X _ { a } , y _ { a } ) , ( X _ { b } , y _ { b } )" + }, + { + "category_id": 13, + "poly": [ + 857, + 1126, + 911, + 1126, + 911, + 1155, + 857, + 1155 + ], + "score": 0.87, + "latex": "50 \\%" + }, + { + "category_id": 13, + "poly": [ + 935, + 1267, + 971, + 1267, + 971, + 1296, + 935, + 1296 + ], + "score": 0.87, + "latex": "g _ { + }" + }, + { + "category_id": 13, + "poly": [ + 1042, + 1451, + 1077, + 1451, + 1077, + 1479, + 1042, + 1479 + ], + "score": 0.86, + "latex": "g _ { + }" + }, + { + "category_id": 13, + "poly": [ + 1000, + 555, + 1072, + 555, + 1072, + 580, + 1000, + 580 + ], + "score": 0.86, + "latex": "x < 0" + }, + { + "category_id": 13, + "poly": [ + 493, + 325, + 525, + 325, + 525, + 353, + 493, + 353 + ], + "score": 0.86, + "latex": "r _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1078, + 1156, + 1132, + 1156, + 1132, + 1185, + 1078, + 1185 + ], + "score": 0.86, + "latex": "30 \\%" + }, + { + "category_id": 13, + "poly": [ + 508, + 1944, + 528, + 1944, + 528, + 1975, + 508, + 1975 + ], + "score": 0.86, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 360, + 1389, + 396, + 1389, + 396, + 1418, + 360, + 1418 + ], + "score": 0.85, + "latex": "g _ { + }" + }, + { + "category_id": 13, + "poly": [ + 472, + 1883, + 490, + 1883, + 490, + 1914, + 472, + 1914 + ], + "score": 0.85, + "latex": "\\tilde { g }" + }, + { + "category_id": 13, + "poly": [ + 927, + 1420, + 957, + 1420, + 957, + 1446, + 927, + 1446 + ], + "score": 0.84, + "latex": "r _ { d }" + }, + { + "category_id": 13, + "poly": [ + 895, + 1389, + 925, + 1389, + 925, + 1416, + 895, + 1416 + ], + "score": 0.84, + "latex": "r _ { d }" + }, + { + "category_id": 13, + "poly": [ + 998, + 723, + 1040, + 723, + 1040, + 752, + 998, + 752 + ], + "score": 0.84, + "latex": "a , b" + }, + { + "category_id": 13, + "poly": [ + 297, + 553, + 372, + 553, + 372, + 583, + 297, + 583 + ], + "score": 0.82, + "latex": "\\operatorname { s g n } ( \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 933, + 555, + 972, + 555, + 972, + 581, + 933, + 581 + ], + "score": 0.81, + "latex": "- 1" + }, + { + "category_id": 13, + "poly": [ + 1217, + 405, + 1235, + 405, + 1235, + 431, + 1217, + 431 + ], + "score": 0.81, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 417, + 523, + 1317, + 523, + 1317, + 555, + 417, + 555 + ], + "score": 0.79, + "latex": "\\begin{array} { r } { \\cos \\bigl ( g _ { a } , g _ { b } \\bigr ) = \\mathrm { s g n } \\bigl ( \\widetilde { p } _ { 1 , a } - y _ { a } \\bigr ) \\cdot \\mathrm { s g n } \\bigl ( \\widetilde { p } _ { 1 , a } - y _ { b } \\bigr ) \\cdot \\cos \\bigl ( \\nabla _ { z } h ( z ) \\bigr | _ { z = f ( X _ { a } ) } , \\nabla _ { z } h ( z ) \\bigr | _ { z = f ( X _ { b } ) } \\bigr ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1382, + 722, + 1402, + 722, + 1402, + 749, + 1382, + 749 + ], + "score": 0.76, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 818, + 296, + 836, + 296, + 836, + 324, + 818, + 324 + ], + "score": 0.76, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1077, + 1887, + 1094, + 1887, + 1094, + 1913, + 1077, + 1913 + ], + "score": 0.74, + "latex": "g" + }, + { + "category_id": 13, + "poly": [ + 329, + 490, + 507, + 490, + 507, + 519, + 329, + 519 + ], + "score": 0.73, + "latex": "\\mathbf { \\tau } _ { \\mathrm { 3 } } : \\mathbb { R } ^ { d } \\times \\mathbf { \\bar { R } } ^ { d } \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 734, + 462, + 779, + 462, + 779, + 493, + 734, + 493 + ], + "score": 0.62, + "latex": "\\widetilde { p } _ { 1 , a }" + }, + { + "category_id": 13, + "poly": [ + 1148, + 1418, + 1210, + 1418, + 1210, + 1444, + 1148, + 1444 + ], + "score": 0.58, + "latex": "A U C" + }, + { + "category_id": 13, + "poly": [ + 792, + 463, + 835, + 463, + 835, + 493, + 792, + 493 + ], + "score": 0.57, + "latex": "\\widetilde { p } _ { 1 , b }" + }, + { + "category_id": 13, + "poly": [ + 1269, + 465, + 1340, + 465, + 1340, + 493, + 1269, + 493 + ], + "score": 0.51, + "latex": "g _ { a } , \\ g _ { b }" + }, + { + "category_id": 13, + "poly": [ + 1313, + 466, + 1340, + 466, + 1340, + 493, + 1313, + 493 + ], + "score": 0.42, + "latex": "g _ { b }" + }, + { + "category_id": 13, + "poly": [ + 734, + 462, + 836, + 462, + 836, + 493, + 734, + 493 + ], + "score": 0.41, + "latex": "\\widetilde { p } _ { 1 , a } , \\widetilde { p } _ { 1 , b }" + }, + { + "category_id": 13, + "poly": [ + 1163, + 652, + 1213, + 652, + 1213, + 680, + 1163, + 680 + ], + "score": 0.38, + "latex": "a , b" + }, + { + "category_id": 13, + "poly": [ + 1196, + 652, + 1213, + 652, + 1213, + 678, + 1196, + 678 + ], + "score": 0.33, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1269, + 466, + 1299, + 466, + 1299, + 493, + 1269, + 493 + ], + "score": 0.32, + "latex": "g _ { a }" + }, + { + "category_id": 13, + "poly": [ + 1163, + 656, + 1180, + 656, + 1180, + 678, + 1163, + 678 + ], + "score": 0.25, + "latex": "a" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1535.0, + 946.0, + 1535.0, + 946.0, + 1576.0, + 293.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1700.0, + 828.0, + 1700.0, + 828.0, + 1737.0, + 294.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1758.0, + 1404.0, + 1758.0, + 1404.0, + 1793.0, + 296.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1786.0, + 1404.0, + 1786.0, + 1404.0, + 1826.0, + 293.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1822.0, + 1404.0, + 1822.0, + 1404.0, + 1854.0, + 296.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 1405.0, + 1849.0, + 1405.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1879.0, + 471.0, + 1879.0, + 471.0, + 1921.0, + 292.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1879.0, + 1076.0, + 1879.0, + 1076.0, + 1921.0, + 491.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1879.0, + 1254.0, + 1879.0, + 1254.0, + 1921.0, + 1095.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1879.0, + 1406.0, + 1879.0, + 1406.0, + 1921.0, + 1395.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1910.0, + 1405.0, + 1910.0, + 1405.0, + 1946.0, + 293.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1942.0, + 507.0, + 1942.0, + 507.0, + 1982.0, + 293.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 1942.0, + 1408.0, + 1942.0, + 1408.0, + 1982.0, + 529.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1966.0, + 1405.0, + 1966.0, + 1405.0, + 2012.0, + 292.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 1402.0, + 2003.0, + 1402.0, + 2040.0, + 294.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1231.0, + 1406.0, + 1231.0, + 1406.0, + 1266.0, + 294.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1262.0, + 934.0, + 1262.0, + 934.0, + 1300.0, + 292.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 1262.0, + 1406.0, + 1262.0, + 1406.0, + 1300.0, + 972.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1294.0, + 1405.0, + 1294.0, + 1405.0, + 1329.0, + 293.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1324.0, + 1404.0, + 1324.0, + 1404.0, + 1360.0, + 294.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1352.0, + 1406.0, + 1352.0, + 1406.0, + 1390.0, + 292.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1381.0, + 359.0, + 1381.0, + 359.0, + 1425.0, + 293.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1381.0, + 894.0, + 1381.0, + 894.0, + 1425.0, + 397.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1381.0, + 1043.0, + 1381.0, + 1043.0, + 1425.0, + 926.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1381.0, + 1406.0, + 1381.0, + 1406.0, + 1425.0, + 1271.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1415.0, + 926.0, + 1415.0, + 926.0, + 1452.0, + 294.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1415.0, + 1147.0, + 1415.0, + 1147.0, + 1452.0, + 958.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1415.0, + 1406.0, + 1415.0, + 1406.0, + 1452.0, + 1211.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1445.0, + 1041.0, + 1445.0, + 1041.0, + 1482.0, + 293.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1445.0, + 1406.0, + 1445.0, + 1406.0, + 1482.0, + 1078.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1478.0, + 1320.0, + 1478.0, + 1320.0, + 1511.0, + 296.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 879.0, + 368.0, + 879.0, + 368.0, + 918.0, + 295.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 879.0, + 1406.0, + 879.0, + 1406.0, + 918.0, + 495.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 913.0, + 1405.0, + 913.0, + 1405.0, + 945.0, + 294.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 943.0, + 1406.0, + 943.0, + 1406.0, + 977.0, + 292.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 975.0, + 1403.0, + 975.0, + 1403.0, + 1006.0, + 296.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1003.0, + 1405.0, + 1003.0, + 1405.0, + 1038.0, + 295.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1034.0, + 1406.0, + 1034.0, + 1406.0, + 1069.0, + 294.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1064.0, + 1406.0, + 1064.0, + 1406.0, + 1101.0, + 291.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1090.0, + 1406.0, + 1090.0, + 1406.0, + 1136.0, + 291.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1126.0, + 856.0, + 1126.0, + 856.0, + 1160.0, + 294.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 1126.0, + 1405.0, + 1126.0, + 1405.0, + 1160.0, + 912.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1156.0, + 1077.0, + 1156.0, + 1077.0, + 1189.0, + 294.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1156.0, + 1406.0, + 1156.0, + 1406.0, + 1189.0, + 1133.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1187.0, + 1403.0, + 1187.0, + 1403.0, + 1219.0, + 292.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1590.0, + 1404.0, + 1590.0, + 1404.0, + 1626.0, + 293.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1623.0, + 1405.0, + 1623.0, + 1405.0, + 1657.0, + 294.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1653.0, + 1398.0, + 1653.0, + 1398.0, + 1689.0, + 292.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 230.0, + 1011.0, + 230.0, + 1011.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 230.0, + 1407.0, + 230.0, + 1407.0, + 267.0, + 1071.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 259.0, + 1300.0, + 259.0, + 1300.0, + 300.0, + 292.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 290.0, + 817.0, + 290.0, + 817.0, + 327.0, + 356.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 290.0, + 1405.0, + 290.0, + 1405.0, + 327.0, + 837.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 319.0, + 492.0, + 319.0, + 492.0, + 359.0, + 292.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 319.0, + 1407.0, + 319.0, + 1407.0, + 359.0, + 526.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 350.0, + 999.0, + 350.0, + 999.0, + 391.0, + 293.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 398.0, + 1216.0, + 398.0, + 1216.0, + 433.0, + 296.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 398.0, + 1405.0, + 398.0, + 1405.0, + 433.0, + 1236.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 428.0, + 980.0, + 428.0, + 980.0, + 465.0, + 293.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 428.0, + 1404.0, + 428.0, + 1404.0, + 465.0, + 1186.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 461.0, + 733.0, + 461.0, + 733.0, + 495.0, + 294.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 461.0, + 1268.0, + 461.0, + 1268.0, + 495.0, + 837.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1341.0, + 461.0, + 1406.0, + 461.0, + 1406.0, + 495.0, + 1341.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 484.0, + 328.0, + 484.0, + 328.0, + 530.0, + 291.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 484.0, + 925.0, + 484.0, + 925.0, + 530.0, + 508.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 484.0, + 1406.0, + 484.0, + 1406.0, + 530.0, + 1283.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 519.0, + 416.0, + 519.0, + 416.0, + 561.0, + 291.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 519.0, + 1406.0, + 519.0, + 1406.0, + 561.0, + 1318.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 551.0, + 296.0, + 551.0, + 296.0, + 585.0, + 292.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 551.0, + 801.0, + 551.0, + 801.0, + 585.0, + 373.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 551.0, + 932.0, + 551.0, + 932.0, + 585.0, + 875.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 551.0, + 999.0, + 551.0, + 999.0, + 585.0, + 973.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 551.0, + 1405.0, + 551.0, + 1405.0, + 585.0, + 1073.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 582.0, + 1124.0, + 582.0, + 1124.0, + 616.0, + 294.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 615.0, + 770.0, + 615.0, + 770.0, + 656.0, + 310.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 615.0, + 1208.0, + 615.0, + 1208.0, + 656.0, + 815.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1396.0, + 615.0, + 1403.0, + 615.0, + 1403.0, + 656.0, + 1396.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 650.0, + 1162.0, + 650.0, + 1162.0, + 688.0, + 597.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 650.0, + 1406.0, + 650.0, + 1406.0, + 688.0, + 1214.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 681.0, + 1345.0, + 681.0, + 1345.0, + 715.0, + 336.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 718.0, + 997.0, + 718.0, + 997.0, + 757.0, + 314.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 718.0, + 1381.0, + 718.0, + 1381.0, + 757.0, + 1041.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 747.0, + 698.0, + 747.0, + 698.0, + 793.0, + 332.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 747.0, + 1408.0, + 747.0, + 1408.0, + 793.0, + 1197.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 782.0, + 1403.0, + 782.0, + 1403.0, + 817.0, + 335.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 813.0, + 1404.0, + 813.0, + 1404.0, + 848.0, + 337.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 844.0, + 1337.0, + 844.0, + 1337.0, + 879.0, + 337.0, + 879.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 367, + 1404, + 367, + 1404, + 873, + 297, + 873 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1770, + 1403, + 1770, + 1403, + 2035, + 297, + 2035 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1061, + 1403, + 1061, + 1403, + 1250, + 297, + 1250 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 296, + 1263, + 1404, + 1263, + 1404, + 1482, + 296, + 1482 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 229, + 1402, + 229, + 1402, + 352, + 299, + 352 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 924, + 1404, + 924, + 1404, + 1047, + 298, + 1047 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 299, + 1618, + 1403, + 1618, + 1403, + 1709, + 299, + 1709 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 516, + 1714, + 1180, + 1714, + 1180, + 1756, + 516, + 1756 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 298, + 1487, + 1399, + 1487, + 1399, + 1548, + 298, + 1548 + ], + "score": 0.92 + }, + { + "category_id": 8, + "poly": [ + 349, + 1562, + 1248, + 1562, + 1248, + 1604, + 349, + 1604 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 297, + 887, + 941, + 887, + 941, + 918, + 297, + 918 + ], + "score": 0.906 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1369, + 1724, + 1400, + 1724, + 1400, + 1751, + 1369, + 1751 + ], + "score": 0.864 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 858, + 2089, + 858, + 2112, + 840, + 2112 + ], + "score": 0.785 + }, + { + "category_id": 13, + "poly": [ + 495, + 1488, + 621, + 1488, + 621, + 1518, + 495, + 1518 + ], + "score": 0.94, + "latex": "0 \\leq \\epsilon < 4" + }, + { + "category_id": 13, + "poly": [ + 1166, + 521, + 1265, + 521, + 1265, + 554, + 1166, + 554 + ], + "score": 0.93, + "latex": "\\| g _ { \\operatorname* { m a x } } \\| _ { 2 } ^ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 643, + 709, + 857, + 709, + 857, + 748, + 643, + 748 + ], + "score": 0.92, + "latex": "\\mathrm { C o v } [ \\eta _ { j } ] = \\sigma _ { j } ^ { 2 } g _ { j } g _ { j } ^ { T }" + }, + { + "category_id": 13, + "poly": [ + 885, + 1648, + 1109, + 1648, + 1109, + 1683, + 885, + 1683 + ], + "score": 0.92, + "latex": "( 1 / 2 + \\overline { { \\sqrt { \\epsilon } } } / 2 - \\epsilon / 8 " + }, + { + "category_id": 13, + "poly": [ + 654, + 1388, + 750, + 1388, + 750, + 1423, + 654, + 1423 + ], + "score": 0.92, + "latex": "\\mathrm { F P R } _ { r } ( t )" + }, + { + "category_id": 13, + "poly": [ + 545, + 429, + 601, + 429, + 601, + 463, + 545, + 463 + ], + "score": 0.92, + "latex": "\\| g \\| _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 809, + 1389, + 906, + 1389, + 906, + 1423, + 809, + 1423 + ], + "score": 0.92, + "latex": "\\mathrm { T P R } _ { r } ( t )" + }, + { + "category_id": 13, + "poly": [ + 298, + 747, + 388, + 747, + 388, + 782, + 298, + 782 + ], + "score": 0.92, + "latex": "\\| g _ { \\operatorname* { m a x } } \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 489, + 1123, + 535, + 1123, + 535, + 1159, + 489, + 1159 + ], + "score": 0.91, + "latex": "\\overset { \\cdot } { g } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 1233, + 710, + 1405, + 710, + 1405, + 746, + 1233, + 746 + ], + "score": 0.91, + "latex": "\\mathbb { E } [ \\| g _ { j } + \\eta _ { j } \\| _ { 2 } ^ { 2 } ] =" + }, + { + "category_id": 13, + "poly": [ + 532, + 1945, + 576, + 1945, + 576, + 1979, + 532, + 1979 + ], + "score": 0.91, + "latex": "\\eta ^ { ( i ) }" + }, + { + "category_id": 13, + "poly": [ + 684, + 1123, + 730, + 1123, + 730, + 1159, + 684, + 1159 + ], + "score": 0.91, + "latex": "g ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 716, + 748, + 1016, + 748, + 1016, + 783, + 716, + 783 + ], + "score": 0.9, + "latex": "\\sigma _ { j } = \\sqrt { \\| g _ { \\mathrm { { m a x } } } \\| _ { 2 } ^ { 2 } / \\| g _ { j } \\| _ { 2 } ^ { 2 } - 1 }" + }, + { + "category_id": 13, + "poly": [ + 500, + 1885, + 1009, + 1885, + 1009, + 1919, + 500, + 1919 + ], + "score": 0.89, + "latex": "p \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 1 ) } ] ) + ( 1 - p ) \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 0 ) } ] ) \\le ~ P" + }, + { + "category_id": 13, + "poly": [ + 392, + 1159, + 432, + 1159, + 432, + 1187, + 392, + 1187 + ], + "score": 0.88, + "latex": "\\eta ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 738, + 581, + 774, + 581, + 774, + 615, + 738, + 615 + ], + "score": 0.88, + "latex": "\\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 912, + 1422, + 956, + 1422, + 956, + 1448, + 912, + 1448 + ], + "score": 0.88, + "latex": "P ^ { ( 1 ) }" + }, + { + "category_id": 14, + "poly": [ + 350, + 1561, + 1248, + 1561, + 1248, + 1604, + 350, + 1604 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { { \\bf K L } ( \\widetilde { P } ^ { ( 1 ) } \\parallel \\widetilde { P } ^ { ( 0 ) } ) + { \\bf K L } ( \\widetilde { P } ^ { ( 0 ) } \\parallel \\widetilde { P } ^ { ( 1 ) } ) \\le \\epsilon \\quad i m p l i e s \\quad \\operatorname* { m a x } _ { r } { \\bf A U C } ( r ) \\le \\frac { 1 } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } - \\frac { \\epsilon } { 8 } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1158, + 337, + 1158, + 337, + 1187, + 297, + 1187 + ], + "score": 0.88, + "latex": "\\eta ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 1190, + 1488, + 1234, + 1488, + 1234, + 1515, + 1190, + 1515 + ], + "score": 0.88, + "latex": "\\widetilde { P } ^ { ( 0 ) }" + }, + { + "category_id": 14, + "poly": [ + 515, + 1714, + 1181, + 1714, + 1181, + 1758, + 515, + 1758 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { \\operatorname { s u m K L } ^ { * } : = \\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \\operatorname { K L } ( \\widetilde { P } ^ { ( 1 ) } \\parallel \\widetilde { P } ^ { ( 0 ) } ) + \\operatorname { K L } ( \\widetilde { P } ^ { ( 0 ) } \\parallel \\widetilde { P } ^ { ( 1 ) } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 549, + 1423, + 593, + 1423, + 593, + 1448, + 549, + 1448 + ], + "score": 0.88, + "latex": "\\widetilde { P } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 1089, + 1488, + 1134, + 1488, + 1134, + 1515, + 1089, + 1515 + ], + "score": 0.87, + "latex": "\\widetilde { P } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 698, + 1219, + 744, + 1219, + 744, + 1246, + 698, + 1246 + ], + "score": 0.87, + "latex": "{ \\cal D } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 599, + 1219, + 645, + 1219, + 645, + 1245, + 599, + 1245 + ], + "score": 0.87, + "latex": "D ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 989, + 1158, + 1036, + 1158, + 1036, + 1185, + 989, + 1185 + ], + "score": 0.87, + "latex": "D ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 805, + 1920, + 924, + 1920, + 924, + 1946, + 805, + 1946 + ], + "score": 0.87, + "latex": "\\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( i ) } ] )" + }, + { + "category_id": 13, + "poly": [ + 446, + 1422, + 490, + 1422, + 490, + 1448, + 446, + 1448 + ], + "score": 0.87, + "latex": "\\widetilde { P } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 1015, + 1422, + 1060, + 1422, + 1060, + 1448, + 1015, + 1448 + ], + "score": 0.87, + "latex": "P ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 500, + 1389, + 594, + 1389, + 594, + 1422, + 500, + 1422 + ], + "score": 0.87, + "latex": "\\operatorname { A U C } ( r )" + }, + { + "category_id": 13, + "poly": [ + 1255, + 1158, + 1300, + 1158, + 1300, + 1185, + 1255, + 1185 + ], + "score": 0.86, + "latex": "\\widetilde { P } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 1090, + 1158, + 1137, + 1158, + 1137, + 1185, + 1090, + 1185 + ], + "score": 0.86, + "latex": "{ \\cal D } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 1048, + 681, + 1076, + 681, + 1076, + 712, + 1048, + 712 + ], + "score": 0.86, + "latex": "g _ { j }" + }, + { + "category_id": 13, + "poly": [ + 732, + 651, + 760, + 651, + 760, + 681, + 732, + 681 + ], + "score": 0.86, + "latex": "g _ { j }" + }, + { + "category_id": 13, + "poly": [ + 1006, + 716, + 1036, + 716, + 1036, + 746, + 1006, + 746 + ], + "score": 0.85, + "latex": "\\sigma _ { j }" + }, + { + "category_id": 13, + "poly": [ + 1355, + 1158, + 1399, + 1158, + 1399, + 1184, + 1355, + 1184 + ], + "score": 0.85, + "latex": "\\widetilde { P } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 682, + 1649, + 844, + 1649, + 844, + 1683, + 682, + 1683 + ], + "score": 0.84, + "latex": "\\operatorname* { m a x } _ { r } \\mathbf { A U C } ( r )" + }, + { + "category_id": 13, + "poly": [ + 297, + 683, + 326, + 683, + 326, + 712, + 297, + 712 + ], + "score": 0.83, + "latex": "\\eta _ { j }" + }, + { + "category_id": 13, + "poly": [ + 834, + 1950, + 857, + 1950, + 857, + 1975, + 834, + 1975 + ], + "score": 0.83, + "latex": "P" + }, + { + "category_id": 13, + "poly": [ + 480, + 1977, + 506, + 1977, + 506, + 2003, + 480, + 2003 + ], + "score": 0.82, + "latex": "P" + }, + { + "category_id": 13, + "poly": [ + 1103, + 1358, + 1394, + 1358, + 1394, + 1393, + 1103, + 1393 + ], + "score": 0.81, + "latex": "\\begin{array} { r } { \\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \\operatorname* { m a x } _ { r } \\mathrm { A U C } ( r ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1096, + 1891, + 1115, + 1891, + 1115, + 1918, + 1096, + 1918 + ], + "score": 0.79, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1256, + 1656, + 1272, + 1656, + 1272, + 1677, + 1256, + 1677 + ], + "score": 0.74, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 855, + 1302, + 872, + 1302, + 872, + 1323, + 855, + 1323 + ], + "score": 0.72, + "latex": "r" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 885.0, + 946.0, + 885.0, + 946.0, + 922.0, + 294.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 368.0, + 1404.0, + 368.0, + 1404.0, + 403.0, + 295.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 399.0, + 1404.0, + 399.0, + 1404.0, + 434.0, + 294.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 431.0, + 544.0, + 431.0, + 544.0, + 462.0, + 296.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 431.0, + 1404.0, + 431.0, + 1404.0, + 462.0, + 602.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 462.0, + 1402.0, + 462.0, + 1402.0, + 494.0, + 296.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 492.0, + 1404.0, + 492.0, + 1404.0, + 524.0, + 296.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 520.0, + 1165.0, + 520.0, + 1165.0, + 558.0, + 292.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 520.0, + 1405.0, + 520.0, + 1405.0, + 558.0, + 1266.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 552.0, + 1404.0, + 552.0, + 1404.0, + 587.0, + 295.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 581.0, + 737.0, + 581.0, + 737.0, + 623.0, + 292.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 581.0, + 1405.0, + 581.0, + 1405.0, + 623.0, + 775.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 616.0, + 1404.0, + 616.0, + 1404.0, + 652.0, + 295.0, + 652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 643.0, + 731.0, + 643.0, + 731.0, + 684.0, + 294.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 643.0, + 1406.0, + 643.0, + 1406.0, + 684.0, + 761.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 678.0, + 296.0, + 678.0, + 296.0, + 713.0, + 292.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 678.0, + 1047.0, + 678.0, + 1047.0, + 713.0, + 327.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 678.0, + 1407.0, + 678.0, + 1407.0, + 713.0, + 1077.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 710.0, + 642.0, + 710.0, + 642.0, + 749.0, + 294.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 710.0, + 1005.0, + 710.0, + 1005.0, + 749.0, + 858.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 710.0, + 1232.0, + 710.0, + 1232.0, + 749.0, + 1037.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 746.0, + 297.0, + 746.0, + 297.0, + 787.0, + 292.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 746.0, + 715.0, + 746.0, + 715.0, + 787.0, + 389.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 746.0, + 1406.0, + 746.0, + 1406.0, + 787.0, + 1017.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 779.0, + 1406.0, + 779.0, + 1406.0, + 814.0, + 294.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 809.0, + 1406.0, + 809.0, + 1406.0, + 845.0, + 294.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 838.0, + 1399.0, + 838.0, + 1399.0, + 876.0, + 294.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1772.0, + 1405.0, + 1772.0, + 1405.0, + 1804.0, + 296.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1797.0, + 1406.0, + 1797.0, + 1406.0, + 1834.0, + 293.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1826.0, + 1405.0, + 1826.0, + 1405.0, + 1863.0, + 294.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1852.0, + 1405.0, + 1852.0, + 1405.0, + 1891.0, + 293.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1880.0, + 499.0, + 1880.0, + 499.0, + 1923.0, + 292.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1880.0, + 1095.0, + 1880.0, + 1095.0, + 1923.0, + 1010.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 1880.0, + 1407.0, + 1880.0, + 1407.0, + 1923.0, + 1116.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1914.0, + 804.0, + 1914.0, + 804.0, + 1949.0, + 295.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1914.0, + 1405.0, + 1914.0, + 1405.0, + 1949.0, + 925.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1943.0, + 531.0, + 1943.0, + 531.0, + 1984.0, + 292.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1943.0, + 833.0, + 1943.0, + 833.0, + 1984.0, + 577.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 1943.0, + 1408.0, + 1943.0, + 1408.0, + 1984.0, + 858.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1976.0, + 479.0, + 1976.0, + 479.0, + 2008.0, + 296.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 1976.0, + 1405.0, + 1976.0, + 1405.0, + 2008.0, + 507.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2001.0, + 1406.0, + 2001.0, + 1406.0, + 2039.0, + 292.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1059.0, + 1403.0, + 1059.0, + 1403.0, + 1099.0, + 294.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1092.0, + 1403.0, + 1092.0, + 1403.0, + 1130.0, + 294.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1121.0, + 488.0, + 1121.0, + 488.0, + 1165.0, + 292.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1121.0, + 683.0, + 1121.0, + 683.0, + 1165.0, + 536.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 1121.0, + 1408.0, + 1121.0, + 1408.0, + 1165.0, + 731.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1154.0, + 296.0, + 1154.0, + 296.0, + 1190.0, + 293.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1154.0, + 391.0, + 1154.0, + 391.0, + 1190.0, + 338.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1154.0, + 988.0, + 1154.0, + 988.0, + 1190.0, + 433.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1154.0, + 1089.0, + 1154.0, + 1089.0, + 1190.0, + 1037.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 1154.0, + 1254.0, + 1154.0, + 1254.0, + 1190.0, + 1138.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1154.0, + 1354.0, + 1154.0, + 1354.0, + 1190.0, + 1301.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1154.0, + 1403.0, + 1154.0, + 1403.0, + 1190.0, + 1400.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1186.0, + 1405.0, + 1186.0, + 1405.0, + 1221.0, + 293.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1217.0, + 598.0, + 1217.0, + 598.0, + 1254.0, + 295.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 1217.0, + 697.0, + 1217.0, + 697.0, + 1254.0, + 646.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1217.0, + 1341.0, + 1217.0, + 1341.0, + 1254.0, + 745.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1264.0, + 1405.0, + 1264.0, + 1405.0, + 1299.0, + 294.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1296.0, + 854.0, + 1296.0, + 854.0, + 1331.0, + 293.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 1296.0, + 1405.0, + 1296.0, + 1405.0, + 1331.0, + 873.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1326.0, + 1405.0, + 1326.0, + 1405.0, + 1361.0, + 294.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1355.0, + 1102.0, + 1355.0, + 1102.0, + 1396.0, + 291.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1355.0, + 1405.0, + 1355.0, + 1405.0, + 1396.0, + 1395.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1387.0, + 499.0, + 1387.0, + 499.0, + 1426.0, + 293.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1387.0, + 653.0, + 1387.0, + 653.0, + 1426.0, + 595.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1387.0, + 808.0, + 1387.0, + 808.0, + 1426.0, + 751.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 1387.0, + 1406.0, + 1387.0, + 1406.0, + 1426.0, + 907.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1420.0, + 445.0, + 1420.0, + 445.0, + 1455.0, + 294.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1420.0, + 548.0, + 1420.0, + 548.0, + 1455.0, + 491.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 1420.0, + 911.0, + 1420.0, + 911.0, + 1455.0, + 594.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1420.0, + 1014.0, + 1420.0, + 1014.0, + 1455.0, + 957.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 1420.0, + 1405.0, + 1420.0, + 1405.0, + 1455.0, + 1061.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1451.0, + 1404.0, + 1451.0, + 1404.0, + 1486.0, + 295.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 230.0, + 1404.0, + 230.0, + 1404.0, + 266.0, + 295.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 261.0, + 1403.0, + 261.0, + 1403.0, + 294.0, + 294.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 293.0, + 1404.0, + 293.0, + 1404.0, + 325.0, + 295.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 320.0, + 1404.0, + 320.0, + 1404.0, + 357.0, + 293.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 925.0, + 1405.0, + 925.0, + 1405.0, + 958.0, + 294.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 954.0, + 1405.0, + 954.0, + 1405.0, + 990.0, + 293.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 987.0, + 1405.0, + 987.0, + 1405.0, + 1019.0, + 294.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1017.0, + 1355.0, + 1017.0, + 1355.0, + 1049.0, + 294.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1617.0, + 1405.0, + 1617.0, + 1405.0, + 1654.0, + 294.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1648.0, + 681.0, + 1648.0, + 681.0, + 1684.0, + 294.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1648.0, + 884.0, + 1648.0, + 884.0, + 1684.0, + 845.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 1648.0, + 1255.0, + 1648.0, + 1255.0, + 1684.0, + 1110.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 1648.0, + 1405.0, + 1648.0, + 1405.0, + 1684.0, + 1273.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1675.0, + 1403.0, + 1675.0, + 1403.0, + 1712.0, + 294.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1484.0, + 494.0, + 1484.0, + 494.0, + 1523.0, + 294.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 1484.0, + 1088.0, + 1484.0, + 1088.0, + 1523.0, + 622.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 1484.0, + 1189.0, + 1484.0, + 1189.0, + 1523.0, + 1135.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 1484.0, + 1405.0, + 1484.0, + 1405.0, + 1523.0, + 1235.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1517.0, + 790.0, + 1517.0, + 790.0, + 1553.0, + 296.0, + 1553.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1553, + 1404, + 1553, + 1404, + 1958, + 297, + 1958 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 397, + 1405, + 397, + 1405, + 738, + 297, + 738 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 1074, + 1405, + 1074, + 1405, + 1387, + 297, + 1387 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 301, + 228, + 1404, + 228, + 1404, + 318, + 301, + 318 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 300, + 1476, + 1400, + 1476, + 1400, + 1539, + 300, + 1539 + ], + "score": 0.951 + }, + { + "category_id": 2, + "poly": [ + 297, + 1977, + 1402, + 1977, + 1402, + 2034, + 297, + 2034 + ], + "score": 0.926 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.904 + }, + { + "category_id": 0, + "poly": [ + 300, + 1418, + 558, + 1418, + 558, + 1453, + 300, + 1453 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1369, + 793, + 1400, + 793, + 1400, + 821, + 1369, + 821 + ], + "score": 0.845 + }, + { + "category_id": 8, + "poly": [ + 369, + 772, + 1315, + 772, + 1315, + 845, + 369, + 845 + ], + "score": 0.828 + }, + { + "category_id": 9, + "poly": [ + 1371, + 333, + 1400, + 333, + 1400, + 361, + 1371, + 361 + ], + "score": 0.789 + }, + { + "category_id": 1, + "poly": [ + 312, + 741, + 1259, + 741, + 1259, + 771, + 312, + 771 + ], + "score": 0.764 + }, + { + "category_id": 2, + "poly": [ + 842, + 2087, + 858, + 2087, + 858, + 2111, + 842, + 2111 + ], + "score": 0.706 + }, + { + "category_id": 8, + "poly": [ + 310, + 323, + 1337, + 323, + 1337, + 381, + 310, + 381 + ], + "score": 0.691 + }, + { + "category_id": 1, + "poly": [ + 308, + 850, + 1380, + 850, + 1380, + 893, + 308, + 893 + ], + "score": 0.502 + }, + { + "category_id": 8, + "poly": [ + 338, + 893, + 1398, + 893, + 1398, + 1061, + 338, + 1061 + ], + "score": 0.471 + }, + { + "category_id": 1, + "poly": [ + 310, + 323, + 1337, + 323, + 1337, + 381, + 310, + 381 + ], + "score": 0.207 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 859, + 2087, + 859, + 2111, + 841, + 2111 + ], + "score": 0.133 + }, + { + "category_id": 13, + "poly": [ + 442, + 458, + 680, + 458, + 680, + 492, + 442, + 492 + ], + "score": 0.95, + "latex": "g ^ { ( 1 ) } \\stackrel { \\cdot } { \\sim } \\mathcal { N } ( \\bar { g } ^ { ( 1 ) } , v I _ { d \\times d } )" + }, + { + "category_id": 13, + "poly": [ + 641, + 1861, + 707, + 1861, + 707, + 1895, + 641, + 1895 + ], + "score": 0.93, + "latex": "f ( X )" + }, + { + "category_id": 13, + "poly": [ + 1191, + 1259, + 1347, + 1259, + 1347, + 1298, + 1191, + 1298 + ], + "score": 0.93, + "latex": "P = s \\| \\Delta g \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 902, + 1354, + 982, + 1354, + 982, + 1389, + 902, + 1389 + ], + "score": 0.93, + "latex": "\\| \\Delta g \\| _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 391, + 672, + 595, + 672, + 595, + 708, + 391, + 708 + ], + "score": 0.93, + "latex": "\\Delta g : = \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 733, + 458, + 973, + 458, + 973, + 492, + 733, + 492 + ], + "score": 0.92, + "latex": "g ^ { ( 0 ) } \\sim \\dot { \\mathcal { N } } ( \\bar { g } ^ { ( 0 ) } , u I _ { d \\times d } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 580, + 488, + 580, + 488, + 614, + 298, + 614 + ], + "score": 0.92, + "latex": "D ^ { ( 0 ) } = \\mathcal { N } ( 0 , \\Sigma _ { 0 } )" + }, + { + "category_id": 13, + "poly": [ + 1170, + 549, + 1354, + 549, + 1354, + 583, + 1170, + 583 + ], + "score": 0.92, + "latex": "D ^ { \\top } \\ d _ { 1 } ) = \\mathcal { N } ( 0 , \\dot { \\Sigma _ { 1 } } )" + }, + { + "category_id": 13, + "poly": [ + 926, + 583, + 1088, + 583, + 1088, + 612, + 926, + 612 + ], + "score": 0.92, + "latex": "\\Sigma _ { 1 } \\Sigma _ { 0 } = \\Sigma _ { 0 } \\Sigma _ { 1 }" + }, + { + "category_id": 14, + "poly": [ + 380, + 768, + 1317, + 768, + 1317, + 846, + 380, + 846 + ], + "score": 0.91, + "latex": "\\Sigma _ { 1 } ^ { * } = \\frac { \\lambda _ { 1 } ^ { ( 1 ) * } - \\lambda _ { 2 } ^ { ( 1 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 1 ) * } I _ { d } , \\quad \\Sigma _ { 0 } ^ { * } = \\frac { \\lambda _ { 1 } ^ { ( 0 ) * } - \\dot { \\lambda } _ { 2 } ^ { ( 0 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 0 ) * } I _ { d } ," + }, + { + "category_id": 13, + "poly": [ + 1209, + 232, + 1332, + 232, + 1332, + 262, + 1209, + 262 + ], + "score": 0.91, + "latex": "\\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( i ) } ] )" + }, + { + "category_id": 13, + "poly": [ + 1009, + 1857, + 1111, + 1857, + 1111, + 1890, + 1009, + 1890 + ], + "score": 0.9, + "latex": "\\mathbb { R } ^ { 5 \\times 5 \\times 6 4 }" + }, + { + "category_id": 13, + "poly": [ + 415, + 1891, + 484, + 1891, + 484, + 1924, + 415, + 1924 + ], + "score": 0.9, + "latex": "\\mathbb { R } ^ { 1 6 0 0 }" + }, + { + "category_id": 13, + "poly": [ + 659, + 1295, + 730, + 1295, + 730, + 1322, + 659, + 1322 + ], + "score": 0.89, + "latex": "s > 0" + }, + { + "category_id": 13, + "poly": [ + 379, + 851, + 667, + 851, + 667, + 893, + 379, + 893 + ], + "score": 0.88, + "latex": "( \\lambda _ { 1 } ^ { ( 0 ) * } , \\lambda _ { 2 } ^ { ( 0 ) * } , \\lambda _ { 1 } ^ { ( 1 ) * } , \\lambda _ { 2 } ^ { ( 1 ) * } )" + }, + { + "category_id": 13, + "poly": [ + 580, + 261, + 651, + 261, + 651, + 290, + 580, + 290 + ], + "score": 0.88, + "latex": "1 - p )" + }, + { + "category_id": 13, + "poly": [ + 587, + 741, + 621, + 741, + 621, + 772, + 587, + 772 + ], + "score": 0.87, + "latex": "\\Sigma _ { 1 } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 298, + 676, + 332, + 676, + 332, + 707, + 298, + 707 + ], + "score": 0.86, + "latex": "\\Sigma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 674, + 741, + 710, + 741, + 710, + 774, + 674, + 774 + ], + "score": 0.86, + "latex": "\\Sigma _ { 0 } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1359, + 644, + 1393, + 644, + 1393, + 674, + 1359, + 674 + ], + "score": 0.86, + "latex": "\\Sigma _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1089, + 1199, + 1114, + 1199, + 1114, + 1225, + 1089, + 1225 + ], + "score": 0.84, + "latex": "P" + }, + { + "category_id": 13, + "poly": [ + 736, + 1326, + 762, + 1326, + 762, + 1352, + 736, + 1352 + ], + "score": 0.81, + "latex": "P" + }, + { + "category_id": 13, + "poly": [ + 537, + 1229, + 562, + 1229, + 562, + 1256, + 537, + 1256 + ], + "score": 0.8, + "latex": "P" + }, + { + "category_id": 13, + "poly": [ + 312, + 323, + 795, + 323, + 795, + 382, + 312, + 382 + ], + "score": 0.8, + "latex": "\\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } { \\mathrm { K L } } ( \\widetilde { P } ^ { ( 1 ) } \\parallel \\widetilde { P } ^ { ( 0 ) } ) + { \\mathrm { K L } } ( \\widetilde { P } ^ { ( 0 ) } \\parallel \\widetilde { P } ^ { ( 1 ) } )" + }, + { + "category_id": 13, + "poly": [ + 858, + 1867, + 877, + 1867, + 877, + 1894, + 858, + 1894 + ], + "score": 0.79, + "latex": "g" + }, + { + "category_id": 14, + "poly": [ + 343, + 976, + 1330, + 976, + 1330, + 1064, + 343, + 1064 + ], + "score": 0.75, + "latex": "\\begin{array} { r l } { s . t . \\quad p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) \\lambda _ { 2 } ^ { ( 0 ) } \\leq { \\cal P } , } & \\\\ { - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 , \\quad - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 , \\quad - \\lambda _ { 2 } ^ { ( 1 ) } \\leq 0 , \\quad - \\lambda _ { 2 } ^ { ( 0 ) } \\leq 0 , \\quad \\lambda _ { 2 } ^ { ( 1 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 , \\quad \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 530, + 263, + 548, + 263, + 548, + 290, + 530, + 290 + ], + "score": 0.66, + "latex": "\\dot { p }" + }, + { + "category_id": 14, + "poly": [ + 370, + 889, + 1400, + 889, + 1400, + 967, + 370, + 967 + ], + "score": 0.66, + "latex": "\\mathop { \\operatorname* { m i n } } _ { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u }" + }, + { + "category_id": 14, + "poly": [ + 843, + 324, + 1358, + 324, + 1358, + 365, + 843, + 365 + ], + "score": 0.64, + "latex": "p \\cdot \\operatorname { t r } ( \\operatorname { C o v } [ \\eta ^ { ( 1 ) } ] ) + ( 1 - p ) \\cdot \\operatorname { t r } ( \\operatorname { C o v } [ \\eta ^ { ( 0 ) } ] ) \\leq ~ P ." + }, + { + "category_id": 14, + "poly": [ + 302, + 322, + 1362, + 322, + 1362, + 376, + 302, + 376 + ], + "score": 0.5, + "latex": "\\operatorname* { m i n } _ { D ^ { ( 1 ) } , D ^ { ( 0 ) } } \\mathrm { K L } ( \\widetilde P ^ { ( 1 ) } \\parallel \\widetilde P ^ { ( 0 ) } ) + \\mathrm { K L } ( \\widetilde P ^ { ( 0 ) } \\parallel \\widetilde P ^ { ( 1 ) } ) \\mathrm { s . t . } p \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 1 ) } ] ) + ( 1 - p ) \\cdot \\mathrm { t r } ( \\mathrm { C o v } [ \\eta ^ { ( 0 ) } ] ) \\le P ." + }, + { + "category_id": 13, + "poly": [ + 388, + 889, + 1399, + 889, + 1399, + 968, + 388, + 968 + ], + "score": 0.48, + "latex": "\\mathop { \\operatorname* { m i n } } _ { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u }" + }, + { + "category_id": 13, + "poly": [ + 844, + 324, + 1360, + 324, + 1360, + 365, + 844, + 365 + ], + "score": 0.45, + "latex": "p \\cdot \\operatorname { t r } ( \\operatorname { C o v } [ \\eta ^ { ( 1 ) } ] ) + ( 1 - p ) \\cdot \\operatorname { t r } ( \\operatorname { C o v } [ \\eta ^ { ( 0 ) } ] ) \\leq ~ P ." + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2010.0, + 329.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2004.0, + 962.0, + 2004.0, + 962.0, + 2036.0, + 293.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1418.0, + 560.0, + 1418.0, + 560.0, + 1458.0, + 293.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1555.0, + 1405.0, + 1555.0, + 1405.0, + 1586.0, + 296.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1584.0, + 1405.0, + 1584.0, + 1405.0, + 1618.0, + 296.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1615.0, + 1402.0, + 1615.0, + 1402.0, + 1649.0, + 295.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1645.0, + 1405.0, + 1645.0, + 1405.0, + 1679.0, + 295.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1675.0, + 1404.0, + 1675.0, + 1404.0, + 1709.0, + 294.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1705.0, + 1405.0, + 1705.0, + 1405.0, + 1740.0, + 291.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1735.0, + 1404.0, + 1735.0, + 1404.0, + 1771.0, + 294.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1765.0, + 1405.0, + 1765.0, + 1405.0, + 1802.0, + 294.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1793.0, + 1405.0, + 1793.0, + 1405.0, + 1833.0, + 291.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1825.0, + 1406.0, + 1825.0, + 1406.0, + 1865.0, + 292.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1850.0, + 640.0, + 1850.0, + 640.0, + 1900.0, + 288.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 1850.0, + 857.0, + 1850.0, + 857.0, + 1900.0, + 708.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 1850.0, + 1008.0, + 1850.0, + 1008.0, + 1900.0, + 878.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1850.0, + 1411.0, + 1850.0, + 1411.0, + 1900.0, + 1112.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1884.0, + 414.0, + 1884.0, + 414.0, + 1936.0, + 289.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 1884.0, + 1410.0, + 1884.0, + 1410.0, + 1936.0, + 485.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1924.0, + 693.0, + 1924.0, + 693.0, + 1961.0, + 294.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 396.0, + 1405.0, + 396.0, + 1405.0, + 433.0, + 296.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 430.0, + 1405.0, + 430.0, + 1405.0, + 465.0, + 294.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 455.0, + 441.0, + 455.0, + 441.0, + 497.0, + 292.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 455.0, + 732.0, + 455.0, + 732.0, + 497.0, + 681.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 455.0, + 1405.0, + 455.0, + 1405.0, + 497.0, + 974.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 489.0, + 1406.0, + 489.0, + 1406.0, + 527.0, + 292.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 521.0, + 1403.0, + 521.0, + 1403.0, + 552.0, + 296.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 547.0, + 1169.0, + 547.0, + 1169.0, + 586.0, + 292.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 547.0, + 1407.0, + 547.0, + 1407.0, + 586.0, + 1355.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 574.0, + 297.0, + 574.0, + 297.0, + 623.0, + 291.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 574.0, + 925.0, + 574.0, + 925.0, + 623.0, + 489.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 574.0, + 1407.0, + 574.0, + 1407.0, + 623.0, + 1089.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 611.0, + 1405.0, + 611.0, + 1405.0, + 646.0, + 294.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 639.0, + 1358.0, + 639.0, + 1358.0, + 682.0, + 292.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 639.0, + 1407.0, + 639.0, + 1407.0, + 682.0, + 1394.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 670.0, + 297.0, + 670.0, + 297.0, + 712.0, + 294.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 670.0, + 390.0, + 670.0, + 390.0, + 712.0, + 333.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 670.0, + 1407.0, + 670.0, + 1407.0, + 712.0, + 596.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 707.0, + 1288.0, + 707.0, + 1288.0, + 742.0, + 295.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1073.0, + 1405.0, + 1073.0, + 1405.0, + 1110.0, + 294.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1108.0, + 1403.0, + 1108.0, + 1403.0, + 1140.0, + 295.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1135.0, + 1405.0, + 1135.0, + 1405.0, + 1172.0, + 291.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1167.0, + 1403.0, + 1167.0, + 1403.0, + 1203.0, + 294.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1196.0, + 1088.0, + 1196.0, + 1088.0, + 1236.0, + 292.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 1196.0, + 1406.0, + 1196.0, + 1406.0, + 1236.0, + 1115.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1229.0, + 536.0, + 1229.0, + 536.0, + 1261.0, + 296.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 1229.0, + 1403.0, + 1229.0, + 1403.0, + 1261.0, + 563.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1259.0, + 1190.0, + 1259.0, + 1190.0, + 1303.0, + 292.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1259.0, + 1407.0, + 1259.0, + 1407.0, + 1303.0, + 1348.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1295.0, + 658.0, + 1295.0, + 658.0, + 1327.0, + 296.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 1295.0, + 1403.0, + 1295.0, + 1403.0, + 1327.0, + 731.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1322.0, + 735.0, + 1322.0, + 735.0, + 1361.0, + 292.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 1322.0, + 1405.0, + 1322.0, + 1405.0, + 1361.0, + 763.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1352.0, + 901.0, + 1352.0, + 901.0, + 1392.0, + 294.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1352.0, + 1088.0, + 1352.0, + 1088.0, + 1392.0, + 983.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 1208.0, + 229.0, + 1208.0, + 266.0, + 296.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 266.0, + 1333.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 259.0, + 529.0, + 259.0, + 529.0, + 292.0, + 295.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 259.0, + 579.0, + 259.0, + 579.0, + 292.0, + 549.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 259.0, + 1405.0, + 259.0, + 1405.0, + 292.0, + 652.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 286.0, + 1290.0, + 286.0, + 1290.0, + 322.0, + 294.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1476.0, + 1405.0, + 1476.0, + 1405.0, + 1511.0, + 295.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1506.0, + 1302.0, + 1506.0, + 1302.0, + 1543.0, + 294.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 736.0, + 586.0, + 736.0, + 586.0, + 778.0, + 306.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 736.0, + 673.0, + 736.0, + 673.0, + 778.0, + 622.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 736.0, + 1262.0, + 736.0, + 1262.0, + 778.0, + 711.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 838.0, + 378.0, + 838.0, + 378.0, + 908.0, + 291.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 838.0, + 1397.0, + 838.0, + 1397.0, + 908.0, + 668.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 337.5, + 404.0, + 337.5, + 404.0, + 355.5, + 390.0, + 355.5 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 616, + 1404, + 616, + 1404, + 1013, + 297, + 1013 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1109, + 1404, + 1109, + 1404, + 1415, + 298, + 1415 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1430, + 1403, + 1430, + 1403, + 1645, + 298, + 1645 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 312, + 1728, + 1404, + 1728, + 1404, + 2034, + 312, + 2034 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 296, + 217, + 1403, + 217, + 1403, + 468, + 296, + 468 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 298, + 1659, + 1400, + 1659, + 1400, + 1722, + 298, + 1722 + ], + "score": 0.942 + }, + { + "category_id": 0, + "poly": [ + 298, + 1050, + 895, + 1050, + 895, + 1082, + 298, + 1082 + ], + "score": 0.907 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 815, + 76, + 815, + 104, + 298, + 104 + ], + "score": 0.886 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2112, + 841, + 2112 + ], + "score": 0.79 + }, + { + "category_id": 0, + "poly": [ + 297, + 558, + 1230, + 558, + 1230, + 589, + 297, + 589 + ], + "score": 0.611 + }, + { + "category_id": 4, + "poly": [ + 297, + 471, + 1398, + 471, + 1398, + 527, + 297, + 527 + ], + "score": 0.498 + }, + { + "category_id": 4, + "poly": [ + 294, + 471, + 1402, + 471, + 1402, + 527, + 294, + 527 + ], + "score": 0.299 + }, + { + "category_id": 1, + "poly": [ + 297, + 558, + 1230, + 558, + 1230, + 589, + 297, + 589 + ], + "score": 0.098 + }, + { + "category_id": 13, + "poly": [ + 625, + 1230, + 778, + 1230, + 778, + 1263, + 625, + 1263 + ], + "score": 0.92, + "latex": "P = s \\| \\Delta g \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 532, + 1263, + 880, + 1263, + 880, + 1294, + 532, + 1294 + ], + "score": 0.89, + "latex": "\\eta \\sim \\mathcal { N } ( \\mathbf { 0 } , ( t / d ) \\cdot \\| g _ { \\operatorname* { m a x } } \\| _ { 2 } ^ { 2 } I _ { d \\times d } )" + }, + { + "category_id": 13, + "poly": [ + 1151, + 799, + 1240, + 799, + 1240, + 828, + 1151, + 828 + ], + "score": 0.89, + "latex": "s = 4 . 0" + }, + { + "category_id": 13, + "poly": [ + 1076, + 1943, + 1162, + 1943, + 1162, + 1972, + 1076, + 1972 + ], + "score": 0.87, + "latex": "t > 2 0" + }, + { + "category_id": 13, + "poly": [ + 838, + 1492, + 893, + 1492, + 893, + 1521, + 838, + 1521 + ], + "score": 0.87, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 919, + 1582, + 974, + 1582, + 974, + 1613, + 919, + 1613 + ], + "score": 0.87, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 486, + 983, + 576, + 983, + 576, + 1012, + 486, + 1012 + ], + "score": 0.85, + "latex": "s = 4 . 0" + }, + { + "category_id": 13, + "poly": [ + 302, + 1612, + 379, + 1612, + 379, + 1643, + 302, + 1643 + ], + "score": 0.81, + "latex": "( 1 0 0 \\% )" + }, + { + "category_id": 13, + "poly": [ + 890, + 925, + 908, + 925, + 908, + 952, + 890, + 952 + ], + "score": 0.75, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1127, + 2004, + 1196, + 2004, + 1196, + 2032, + 1127, + 2032 + ], + "score": 0.73, + "latex": "\\approx 0 . 6" + }, + { + "category_id": 13, + "poly": [ + 1273, + 1264, + 1288, + 1264, + 1288, + 1289, + 1273, + 1289 + ], + "score": 0.66, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 960, + 1355, + 974, + 1355, + 974, + 1381, + 960, + 1381 + ], + "score": 0.66, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 912, + 502, + 928, + 502, + 928, + 522, + 912, + 522 + ], + "score": 0.54, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 737, + 738, + 851, + 738, + 851, + 767, + 737, + 767 + ], + "score": 0.5, + "latex": "\\mathbf { A U C } \\approx 1" + }, + { + "category_id": 13, + "poly": [ + 849, + 453, + 896, + 453, + 896, + 469, + 849, + 469 + ], + "score": 0.49, + "latex": "s { = } 0 . 2 5" + }, + { + "category_id": 13, + "poly": [ + 1041, + 799, + 1111, + 799, + 1111, + 828, + 1041, + 828 + ], + "score": 0.45, + "latex": "\\approx { \\bf 0 . 5 }" + }, + { + "category_id": 13, + "poly": [ + 773, + 1358, + 789, + 1358, + 789, + 1380, + 773, + 1380 + ], + "score": 0.4, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 798, + 739, + 850, + 739, + 850, + 767, + 798, + 767 + ], + "score": 0.33, + "latex": "\\approx 1" + }, + { + "category_id": 13, + "poly": [ + 702, + 452, + 742, + 452, + 742, + 469, + 702, + 469 + ], + "score": 0.32, + "latex": "{ \\mathsf { S } } { = } 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 1004, + 452, + 1045, + 452, + 1045, + 469, + 1004, + 469 + ], + "score": 0.3, + "latex": "{ \\sf s } { = } 1 . 0" + }, + { + "category_id": 13, + "poly": [ + 737, + 772, + 753, + 772, + 753, + 796, + 737, + 796 + ], + "score": 0.25, + "latex": "s" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 222.0, + 523.0, + 222.0, + 523.0, + 245.0, + 391.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 222.0, + 799.0, + 222.0, + 799.0, + 245.0, + 667.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 222.0, + 1077.0, + 222.0, + 1077.0, + 245.0, + 941.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 222.0, + 1354.0, + 222.0, + 1354.0, + 245.0, + 1217.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 241.0, + 344.0, + 241.0, + 344.0, + 263.0, + 308.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 241.0, + 621.0, + 241.0, + 621.0, + 263.0, + 584.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 241.0, + 899.0, + 241.0, + 899.0, + 263.0, + 861.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 241.0, + 1175.0, + 241.0, + 1175.0, + 263.0, + 1136.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 280.0, + 258.0, + 355.0, + 258.0, + 355.0, + 392.0, + 280.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 258.0, + 632.0, + 258.0, + 632.0, + 405.0, + 558.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 257.0, + 908.0, + 257.0, + 908.0, + 393.0, + 834.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 258.0, + 1184.0, + 258.0, + 1184.0, + 397.0, + 1110.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 336.0, + 614.0, + 336.0, + 614.0, + 358.0, + 583.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 343.0, + 922.0, + 343.0, + 922.0, + 353.0, + 911.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 336.0, + 1168.0, + 336.0, + 1168.0, + 358.0, + 1136.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 371.0, + 339.0, + 371.0, + 339.0, + 389.0, + 309.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 356.0, + 646.0, + 356.0, + 646.0, + 366.0, + 633.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 368.0, + 892.0, + 368.0, + 892.0, + 391.0, + 860.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 364.0, + 1170.0, + 364.0, + 1170.0, + 392.0, + 1135.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 361.0, + 1194.0, + 361.0, + 1194.0, + 369.0, + 1183.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 402.0, + 337.0, + 402.0, + 337.0, + 420.0, + 309.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 400.0, + 614.0, + 400.0, + 614.0, + 423.0, + 583.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 400.0, + 891.0, + 400.0, + 891.0, + 423.0, + 860.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 401.0, + 932.0, + 401.0, + 932.0, + 410.0, + 917.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 400.0, + 1168.0, + 400.0, + 1168.0, + 423.0, + 1136.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 417.0, + 359.0, + 417.0, + 359.0, + 434.0, + 344.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 414.0, + 436.0, + 414.0, + 436.0, + 435.0, + 384.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 416.0, + 545.0, + 416.0, + 545.0, + 435.0, + 438.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 416.0, + 636.0, + 416.0, + 636.0, + 434.0, + 621.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 413.0, + 822.0, + 413.0, + 822.0, + 436.0, + 660.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 417.0, + 912.0, + 417.0, + 912.0, + 434.0, + 896.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 413.0, + 1098.0, + 413.0, + 1098.0, + 436.0, + 935.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 414.0, + 1190.0, + 414.0, + 1190.0, + 435.0, + 1171.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 416.0, + 1373.0, + 416.0, + 1373.0, + 435.0, + 1213.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 431.0, + 534.0, + 431.0, + 534.0, + 455.0, + 379.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 433.0, + 808.0, + 433.0, + 808.0, + 452.0, + 656.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 433.0, + 1083.0, + 433.0, + 1083.0, + 452.0, + 931.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 433.0, + 1361.0, + 433.0, + 1361.0, + 452.0, + 1209.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 452.0, + 598.0, + 452.0, + 598.0, + 472.0, + 533.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 451.0, + 701.0, + 451.0, + 701.0, + 470.0, + 647.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 451.0, + 848.0, + 451.0, + 848.0, + 470.0, + 793.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 451.0, + 900.0, + 451.0, + 900.0, + 470.0, + 897.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 451.0, + 1003.0, + 451.0, + 1003.0, + 470.0, + 949.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 451.0, + 1197.0, + 451.0, + 1197.0, + 470.0, + 1095.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1050.0, + 897.0, + 1050.0, + 897.0, + 1086.0, + 295.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 70.0, + 818.0, + 70.0, + 818.0, + 110.0, + 294.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 557.0, + 1233.0, + 557.0, + 1233.0, + 592.0, + 295.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 466.0, + 1404.0, + 466.0, + 1404.0, + 500.0, + 295.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 494.0, + 911.0, + 494.0, + 911.0, + 532.0, + 293.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 494.0, + 1228.0, + 494.0, + 1228.0, + 532.0, + 929.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 466.0, + 1405.0, + 466.0, + 1405.0, + 500.0, + 294.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 494.0, + 911.0, + 494.0, + 911.0, + 532.0, + 293.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 494.0, + 1227.0, + 494.0, + 1227.0, + 532.0, + 929.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 616.0, + 1404.0, + 616.0, + 1404.0, + 649.0, + 297.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 645.0, + 1405.0, + 645.0, + 1405.0, + 681.0, + 294.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 676.0, + 1405.0, + 676.0, + 1405.0, + 711.0, + 294.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 708.0, + 1402.0, + 708.0, + 1402.0, + 741.0, + 296.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 737.0, + 736.0, + 737.0, + 736.0, + 773.0, + 294.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 737.0, + 1406.0, + 737.0, + 1406.0, + 773.0, + 852.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 767.0, + 736.0, + 767.0, + 736.0, + 803.0, + 294.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 767.0, + 1404.0, + 767.0, + 1404.0, + 803.0, + 754.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 798.0, + 1040.0, + 798.0, + 1040.0, + 834.0, + 294.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 798.0, + 1150.0, + 798.0, + 1150.0, + 834.0, + 1112.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 798.0, + 1405.0, + 798.0, + 1405.0, + 834.0, + 1241.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 827.0, + 1405.0, + 827.0, + 1405.0, + 864.0, + 292.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 856.0, + 1405.0, + 856.0, + 1405.0, + 899.0, + 290.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 890.0, + 1404.0, + 890.0, + 1404.0, + 925.0, + 292.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 919.0, + 889.0, + 919.0, + 889.0, + 957.0, + 291.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 919.0, + 1405.0, + 919.0, + 1405.0, + 957.0, + 909.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 949.0, + 1405.0, + 949.0, + 1405.0, + 987.0, + 292.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 981.0, + 485.0, + 981.0, + 485.0, + 1018.0, + 294.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 981.0, + 1181.0, + 981.0, + 1181.0, + 1018.0, + 577.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1107.0, + 1405.0, + 1107.0, + 1405.0, + 1145.0, + 293.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1138.0, + 1403.0, + 1138.0, + 1403.0, + 1176.0, + 292.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1172.0, + 1404.0, + 1172.0, + 1404.0, + 1204.0, + 294.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1199.0, + 1406.0, + 1199.0, + 1406.0, + 1235.0, + 293.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1229.0, + 624.0, + 1229.0, + 624.0, + 1266.0, + 292.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1229.0, + 1394.0, + 1229.0, + 1394.0, + 1266.0, + 779.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1261.0, + 531.0, + 1261.0, + 531.0, + 1296.0, + 293.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1261.0, + 1272.0, + 1261.0, + 1272.0, + 1296.0, + 881.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 1261.0, + 1405.0, + 1261.0, + 1405.0, + 1296.0, + 1289.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1291.0, + 1406.0, + 1291.0, + 1406.0, + 1327.0, + 293.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1322.0, + 1406.0, + 1322.0, + 1406.0, + 1357.0, + 293.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1350.0, + 772.0, + 1350.0, + 772.0, + 1389.0, + 292.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1350.0, + 959.0, + 1350.0, + 959.0, + 1389.0, + 790.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 1350.0, + 1405.0, + 1350.0, + 1405.0, + 1389.0, + 975.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1385.0, + 1160.0, + 1385.0, + 1160.0, + 1417.0, + 296.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1430.0, + 1404.0, + 1430.0, + 1404.0, + 1465.0, + 296.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1462.0, + 1404.0, + 1462.0, + 1404.0, + 1493.0, + 296.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1488.0, + 837.0, + 1488.0, + 837.0, + 1528.0, + 292.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1488.0, + 1407.0, + 1488.0, + 1407.0, + 1528.0, + 894.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1520.0, + 1405.0, + 1520.0, + 1405.0, + 1557.0, + 293.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1551.0, + 1405.0, + 1551.0, + 1405.0, + 1587.0, + 292.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1581.0, + 918.0, + 1581.0, + 918.0, + 1617.0, + 293.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 1581.0, + 1405.0, + 1581.0, + 1405.0, + 1617.0, + 975.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1610.0, + 301.0, + 1610.0, + 301.0, + 1649.0, + 293.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 1610.0, + 1407.0, + 1610.0, + 1407.0, + 1649.0, + 380.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 1727.0, + 1403.0, + 1727.0, + 1403.0, + 1763.0, + 313.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1759.0, + 1406.0, + 1759.0, + 1406.0, + 1795.0, + 335.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1787.0, + 1406.0, + 1787.0, + 1406.0, + 1824.0, + 334.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1819.0, + 1404.0, + 1819.0, + 1404.0, + 1856.0, + 334.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1849.0, + 1404.0, + 1849.0, + 1404.0, + 1885.0, + 336.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1881.0, + 1406.0, + 1881.0, + 1406.0, + 1917.0, + 336.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1910.0, + 1403.0, + 1910.0, + 1403.0, + 1946.0, + 336.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1942.0, + 1075.0, + 1942.0, + 1075.0, + 1976.0, + 334.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1163.0, + 1942.0, + 1403.0, + 1942.0, + 1403.0, + 1976.0, + 1163.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1970.0, + 1407.0, + 1970.0, + 1407.0, + 2007.0, + 335.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 2003.0, + 1126.0, + 2003.0, + 1126.0, + 2036.0, + 335.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 2003.0, + 1209.0, + 2003.0, + 1209.0, + 2036.0, + 1197.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1656.0, + 1405.0, + 1656.0, + 1405.0, + 1697.0, + 292.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1688.0, + 1404.0, + 1688.0, + 1404.0, + 1726.0, + 293.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 557.0, + 1233.0, + 557.0, + 1233.0, + 592.0, + 295.0, + 592.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1686, + 1404, + 1686, + 1404, + 2020, + 297, + 2020 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1425, + 1404, + 1425, + 1404, + 1670, + 298, + 1670 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 314, + 998, + 1403, + 998, + 1403, + 1213, + 314, + 1213 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 307, + 219, + 1390, + 219, + 1390, + 891, + 307, + 891 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 301, + 1239, + 1404, + 1239, + 1404, + 1331, + 301, + 1331 + ], + "score": 0.97 + }, + { + "category_id": 4, + "poly": [ + 295, + 900, + 1399, + 900, + 1399, + 957, + 295, + 957 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 299, + 1366, + 544, + 1366, + 544, + 1401, + 299, + 1401 + ], + "score": 0.899 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.884 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.788 + }, + { + "category_id": 15, + "poly": [ + 283.0, + 205.0, + 381.0, + 205.0, + 381.0, + 413.0, + 283.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 222.0, + 527.0, + 222.0, + 527.0, + 245.0, + 407.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 223.0, + 796.0, + 223.0, + 796.0, + 243.0, + 677.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 223.0, + 1074.0, + 223.0, + 1074.0, + 243.0, + 945.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 197.0, + 1202.0, + 197.0, + 1202.0, + 455.0, + 1099.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 221.0, + 1348.0, + 221.0, + 1348.0, + 244.0, + 1213.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 233.0, + 463.0, + 233.0, + 463.0, + 276.0, + 325.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 249.0, + 489.0, + 249.0, + 489.0, + 278.0, + 475.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 245.0, + 547.0, + 245.0, + 547.0, + 281.0, + 495.0, + 281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 234.0, + 776.0, + 234.0, + 776.0, + 274.0, + 595.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 240.0, + 921.0, + 240.0, + 921.0, + 308.0, + 867.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 250.0, + 1028.0, + 250.0, + 1028.0, + 278.0, + 1019.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 246.0, + 1090.0, + 246.0, + 1090.0, + 282.0, + 1037.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 237.0, + 1194.0, + 237.0, + 1194.0, + 289.0, + 1138.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 250.0, + 1300.0, + 250.0, + 1300.0, + 278.0, + 1290.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1308.0, + 246.0, + 1360.0, + 246.0, + 1360.0, + 282.0, + 1308.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 281.0, + 364.0, + 281.0, + 364.0, + 303.0, + 325.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 278.0, + 422.0, + 278.0, + 422.0, + 288.0, + 412.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 282.0, + 466.0, + 282.0, + 466.0, + 293.0, + 455.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 284.0, + 487.0, + 284.0, + 487.0, + 294.0, + 476.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 280.0, + 568.0, + 280.0, + 568.0, + 299.0, + 496.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 284.0, + 1028.0, + 284.0, + 1028.0, + 294.0, + 1019.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 280.0, + 1109.0, + 280.0, + 1109.0, + 299.0, + 1038.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 273.0, + 1179.0, + 273.0, + 1179.0, + 298.0, + 1137.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 284.0, + 1198.0, + 284.0, + 1198.0, + 308.0, + 1181.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 284.0, + 1254.0, + 284.0, + 1254.0, + 298.0, + 1239.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 284.0, + 1301.0, + 284.0, + 1301.0, + 296.0, + 1290.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1308.0, + 280.0, + 1380.0, + 280.0, + 1380.0, + 299.0, + 1308.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 296.0, + 427.0, + 296.0, + 427.0, + 307.0, + 415.0, + 307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 299.0, + 489.0, + 299.0, + 489.0, + 314.0, + 474.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 296.0, + 557.0, + 296.0, + 557.0, + 315.0, + 496.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 293.0, + 729.0, + 293.0, + 729.0, + 304.0, + 718.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 292.0, + 906.0, + 292.0, + 906.0, + 317.0, + 866.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 299.0, + 1031.0, + 299.0, + 1031.0, + 314.0, + 1015.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 296.0, + 1098.0, + 296.0, + 1098.0, + 315.0, + 1037.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 293.0, + 1260.0, + 293.0, + 1260.0, + 308.0, + 1244.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1288.0, + 300.0, + 1300.0, + 300.0, + 1300.0, + 313.0, + 1288.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1308.0, + 296.0, + 1370.0, + 296.0, + 1370.0, + 316.0, + 1308.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 315.0, + 470.0, + 315.0, + 470.0, + 326.0, + 459.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 311.0, + 1178.0, + 311.0, + 1178.0, + 337.0, + 1137.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1274.0, + 315.0, + 1282.0, + 315.0, + 1282.0, + 325.0, + 1274.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 329.0, + 908.0, + 329.0, + 908.0, + 355.0, + 866.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 331.0, + 1301.0, + 331.0, + 1301.0, + 340.0, + 1292.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 339.0, + 457.0, + 339.0, + 457.0, + 345.0, + 450.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 337.0, + 480.0, + 337.0, + 480.0, + 349.0, + 469.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 339.0, + 661.0, + 339.0, + 661.0, + 354.0, + 646.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 335.0, + 694.0, + 335.0, + 694.0, + 354.0, + 670.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 356.0, + 660.0, + 356.0, + 660.0, + 368.0, + 649.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 351.0, + 720.0, + 351.0, + 720.0, + 370.0, + 669.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 351.0, + 1178.0, + 351.0, + 1178.0, + 373.0, + 1138.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 350.0, + 1318.0, + 350.0, + 1318.0, + 362.0, + 1307.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 386.0, + 332.0, + 386.0, + 332.0, + 426.0, + 311.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 396.0, + 568.0, + 396.0, + 568.0, + 409.0, + 548.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 373.0, + 661.0, + 373.0, + 661.0, + 403.0, + 646.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 368.0, + 739.0, + 368.0, + 739.0, + 404.0, + 667.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 377.0, + 842.0, + 377.0, + 842.0, + 412.0, + 810.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 386.0, + 874.0, + 386.0, + 874.0, + 426.0, + 852.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 392.0, + 1015.0, + 392.0, + 1015.0, + 408.0, + 1000.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 397.0, + 1032.0, + 397.0, + 1032.0, + 408.0, + 1020.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 397.0, + 1046.0, + 397.0, + 1046.0, + 408.0, + 1035.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 397.0, + 1112.0, + 397.0, + 1112.0, + 411.0, + 1096.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 395.0, + 1249.0, + 395.0, + 1249.0, + 409.0, + 1231.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 374.0, + 1381.0, + 374.0, + 1381.0, + 386.0, + 1370.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 409.0, + 420.0, + 409.0, + 420.0, + 426.0, + 390.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 409.0, + 470.0, + 409.0, + 470.0, + 426.0, + 439.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 409.0, + 521.0, + 409.0, + 521.0, + 426.0, + 491.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 408.0, + 571.0, + 408.0, + 571.0, + 426.0, + 540.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 408.0, + 691.0, + 408.0, + 691.0, + 426.0, + 661.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 408.0, + 741.0, + 408.0, + 741.0, + 426.0, + 710.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 409.0, + 792.0, + 409.0, + 792.0, + 426.0, + 762.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 408.0, + 841.0, + 408.0, + 841.0, + 426.0, + 812.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 409.0, + 968.0, + 409.0, + 968.0, + 426.0, + 904.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 409.0, + 1005.0, + 409.0, + 1005.0, + 426.0, + 972.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 409.0, + 1038.0, + 409.0, + 1038.0, + 426.0, + 1006.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 409.0, + 1106.0, + 409.0, + 1106.0, + 426.0, + 1041.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 408.0, + 1239.0, + 408.0, + 1239.0, + 426.0, + 1174.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 409.0, + 1275.0, + 409.0, + 1275.0, + 426.0, + 1243.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 409.0, + 1308.0, + 409.0, + 1308.0, + 426.0, + 1277.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 409.0, + 1376.0, + 409.0, + 1376.0, + 426.0, + 1312.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 417.0, + 500.0, + 417.0, + 500.0, + 441.0, + 434.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 417.0, + 771.0, + 417.0, + 771.0, + 441.0, + 705.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 417.0, + 1041.0, + 417.0, + 1041.0, + 441.0, + 975.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 420.0, + 1312.0, + 420.0, + 1312.0, + 440.0, + 1247.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 431.0, + 528.0, + 431.0, + 528.0, + 460.0, + 404.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 433.0, + 796.0, + 433.0, + 796.0, + 456.0, + 677.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 434.0, + 1076.0, + 434.0, + 1076.0, + 457.0, + 942.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 432.0, + 1348.0, + 432.0, + 1348.0, + 459.0, + 1211.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 450.0, + 367.0, + 450.0, + 367.0, + 472.0, + 325.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 461.0, + 391.0, + 461.0, + 391.0, + 477.0, + 375.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 459.0, + 425.0, + 459.0, + 425.0, + 478.0, + 399.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 463.0, + 554.0, + 463.0, + 554.0, + 473.0, + 543.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 448.0, + 638.0, + 448.0, + 638.0, + 473.0, + 595.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 460.0, + 728.0, + 460.0, + 728.0, + 474.0, + 713.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 457.0, + 747.0, + 457.0, + 747.0, + 469.0, + 736.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 456.0, + 768.0, + 456.0, + 768.0, + 468.0, + 757.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 455.0, + 837.0, + 455.0, + 837.0, + 490.0, + 804.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 461.0, + 932.0, + 461.0, + 932.0, + 477.0, + 918.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 459.0, + 966.0, + 459.0, + 966.0, + 478.0, + 940.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 457.0, + 1112.0, + 457.0, + 1112.0, + 492.0, + 1083.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 461.0, + 1203.0, + 461.0, + 1203.0, + 477.0, + 1189.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 459.0, + 1237.0, + 459.0, + 1237.0, + 478.0, + 1211.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1369.0, + 456.0, + 1378.0, + 456.0, + 1378.0, + 467.0, + 1369.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 478.0, + 391.0, + 478.0, + 391.0, + 492.0, + 375.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 474.0, + 449.0, + 474.0, + 449.0, + 495.0, + 397.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 477.0, + 559.0, + 477.0, + 559.0, + 487.0, + 550.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 467.0, + 708.0, + 467.0, + 708.0, + 495.0, + 659.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 479.0, + 930.0, + 479.0, + 930.0, + 490.0, + 920.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 474.0, + 990.0, + 474.0, + 990.0, + 495.0, + 940.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 479.0, + 1201.0, + 479.0, + 1201.0, + 491.0, + 1191.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 474.0, + 1263.0, + 474.0, + 1263.0, + 495.0, + 1211.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 492.0, + 367.0, + 492.0, + 367.0, + 518.0, + 324.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 493.0, + 391.0, + 493.0, + 391.0, + 509.0, + 375.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 492.0, + 466.0, + 492.0, + 466.0, + 512.0, + 399.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 490.0, + 545.0, + 490.0, + 545.0, + 501.0, + 535.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 496.0, + 560.0, + 496.0, + 560.0, + 506.0, + 550.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 493.0, + 932.0, + 493.0, + 932.0, + 509.0, + 918.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 492.0, + 1009.0, + 492.0, + 1009.0, + 512.0, + 940.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 484.0, + 1102.0, + 484.0, + 1102.0, + 509.0, + 1079.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 485.0, + 1179.0, + 485.0, + 1179.0, + 512.0, + 1137.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 495.0, + 1203.0, + 495.0, + 1203.0, + 509.0, + 1189.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 492.0, + 1297.0, + 492.0, + 1297.0, + 512.0, + 1211.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 512.0, + 391.0, + 512.0, + 391.0, + 526.0, + 375.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 509.0, + 458.0, + 509.0, + 458.0, + 528.0, + 397.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 506.0, + 779.0, + 506.0, + 779.0, + 516.0, + 768.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 504.0, + 908.0, + 504.0, + 908.0, + 530.0, + 866.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 514.0, + 931.0, + 514.0, + 931.0, + 525.0, + 919.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 508.0, + 1000.0, + 508.0, + 1000.0, + 528.0, + 940.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 502.0, + 1100.0, + 502.0, + 1100.0, + 526.0, + 1074.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1190.0, + 513.0, + 1201.0, + 513.0, + 1201.0, + 525.0, + 1190.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 509.0, + 1271.0, + 509.0, + 1271.0, + 528.0, + 1211.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 507.0, + 1293.0, + 507.0, + 1293.0, + 519.0, + 1282.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 527.0, + 568.0, + 527.0, + 568.0, + 538.0, + 556.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 525.0, + 763.0, + 525.0, + 763.0, + 533.0, + 754.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 528.0, + 786.0, + 528.0, + 786.0, + 543.0, + 771.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 526.0, + 1176.0, + 526.0, + 1176.0, + 548.0, + 1138.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 527.0, + 1275.0, + 527.0, + 1275.0, + 538.0, + 1264.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 542.0, + 906.0, + 542.0, + 906.0, + 567.0, + 866.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 542.0, + 1265.0, + 542.0, + 1265.0, + 554.0, + 1255.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 536.0, + 1336.0, + 536.0, + 1336.0, + 555.0, + 1322.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 549.0, + 508.0, + 549.0, + 508.0, + 560.0, + 497.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 550.0, + 559.0, + 550.0, + 559.0, + 561.0, + 547.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 553.0, + 757.0, + 553.0, + 757.0, + 565.0, + 746.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 548.0, + 793.0, + 548.0, + 793.0, + 567.0, + 767.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 562.0, + 545.0, + 562.0, + 545.0, + 578.0, + 531.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 569.0, + 757.0, + 569.0, + 757.0, + 578.0, + 747.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 563.0, + 819.0, + 563.0, + 819.0, + 583.0, + 767.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 563.0, + 1176.0, + 563.0, + 1176.0, + 585.0, + 1138.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 561.0, + 1251.0, + 561.0, + 1251.0, + 576.0, + 1237.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 582.0, + 481.0, + 582.0, + 481.0, + 613.0, + 439.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 590.0, + 500.0, + 590.0, + 500.0, + 604.0, + 485.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 586.0, + 757.0, + 586.0, + 757.0, + 596.0, + 747.0, + 596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 582.0, + 837.0, + 582.0, + 837.0, + 601.0, + 767.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 586.0, + 1190.0, + 586.0, + 1190.0, + 655.0, + 1115.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 602.0, + 431.0, + 602.0, + 431.0, + 622.0, + 360.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 601.0, + 760.0, + 601.0, + 760.0, + 615.0, + 745.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 597.0, + 828.0, + 597.0, + 828.0, + 617.0, + 767.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 598.0, + 874.0, + 598.0, + 874.0, + 638.0, + 852.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 609.0, + 921.0, + 609.0, + 921.0, + 624.0, + 904.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 609.0, + 982.0, + 609.0, + 982.0, + 621.0, + 957.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 608.0, + 995.0, + 608.0, + 995.0, + 620.0, + 984.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 621.0, + 379.0, + 621.0, + 379.0, + 638.0, + 347.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 621.0, + 416.0, + 621.0, + 416.0, + 638.0, + 384.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 621.0, + 498.0, + 621.0, + 498.0, + 653.0, + 426.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 621.0, + 535.0, + 621.0, + 535.0, + 638.0, + 503.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 621.0, + 572.0, + 621.0, + 572.0, + 638.0, + 542.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 621.0, + 649.0, + 621.0, + 649.0, + 638.0, + 618.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 621.0, + 687.0, + 621.0, + 687.0, + 638.0, + 656.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 621.0, + 769.0, + 621.0, + 769.0, + 653.0, + 698.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 621.0, + 805.0, + 621.0, + 805.0, + 638.0, + 775.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 621.0, + 844.0, + 621.0, + 844.0, + 638.0, + 813.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 621.0, + 938.0, + 621.0, + 938.0, + 638.0, + 905.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 621.0, + 973.0, + 621.0, + 973.0, + 638.0, + 941.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 621.0, + 1044.0, + 621.0, + 1044.0, + 653.0, + 976.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 621.0, + 1116.0, + 621.0, + 1116.0, + 638.0, + 1085.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 621.0, + 1208.0, + 621.0, + 1208.0, + 638.0, + 1178.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 621.0, + 1243.0, + 621.0, + 1243.0, + 638.0, + 1211.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 621.0, + 1314.0, + 621.0, + 1314.0, + 653.0, + 1248.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 624.0, + 1349.0, + 624.0, + 1349.0, + 637.0, + 1320.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 621.0, + 1386.0, + 621.0, + 1386.0, + 638.0, + 1356.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 673.0, + 523.0, + 673.0, + 523.0, + 696.0, + 402.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 672.0, + 603.0, + 672.0, + 603.0, + 794.0, + 576.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 673.0, + 796.0, + 673.0, + 796.0, + 696.0, + 674.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 672.0, + 1076.0, + 672.0, + 1076.0, + 699.0, + 935.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 672.0, + 1147.0, + 672.0, + 1147.0, + 886.0, + 1116.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 672.0, + 1345.0, + 672.0, + 1345.0, + 699.0, + 1207.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 690.0, + 363.0, + 690.0, + 363.0, + 756.0, + 322.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 701.0, + 388.0, + 701.0, + 388.0, + 748.0, + 372.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 699.0, + 464.0, + 699.0, + 464.0, + 752.0, + 392.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 696.0, + 565.0, + 696.0, + 565.0, + 752.0, + 528.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 689.0, + 632.0, + 689.0, + 632.0, + 759.0, + 590.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 726.0, + 724.0, + 726.0, + 724.0, + 741.0, + 709.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 718.0, + 744.0, + 718.0, + 744.0, + 727.0, + 733.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 709.0, + 765.0, + 709.0, + 765.0, + 721.0, + 755.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 692.0, + 840.0, + 692.0, + 840.0, + 752.0, + 782.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 706.0, + 901.0, + 706.0, + 901.0, + 727.0, + 863.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 702.0, + 930.0, + 702.0, + 930.0, + 749.0, + 914.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 697.0, + 1006.0, + 697.0, + 1006.0, + 750.0, + 935.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 683.0, + 1116.0, + 683.0, + 1116.0, + 749.0, + 1075.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 689.0, + 1175.0, + 689.0, + 1175.0, + 750.0, + 1133.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 702.0, + 1201.0, + 702.0, + 1201.0, + 749.0, + 1185.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 699.0, + 1277.0, + 699.0, + 1277.0, + 750.0, + 1206.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 732.0, + 1322.0, + 732.0, + 1322.0, + 754.0, + 1304.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 719.0, + 1333.0, + 719.0, + 1333.0, + 731.0, + 1323.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 694.0, + 1380.0, + 694.0, + 1380.0, + 753.0, + 1346.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 748.0, + 455.0, + 748.0, + 455.0, + 879.0, + 305.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 777.0, + 506.0, + 777.0, + 506.0, + 789.0, + 495.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 771.0, + 518.0, + 771.0, + 518.0, + 777.0, + 512.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 785.0, + 630.0, + 785.0, + 630.0, + 882.0, + 577.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 756.0, + 679.0, + 756.0, + 679.0, + 863.0, + 645.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 791.0, + 759.0, + 791.0, + 759.0, + 853.0, + 743.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 772.0, + 832.0, + 772.0, + 832.0, + 858.0, + 762.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 754.0, + 926.0, + 754.0, + 926.0, + 864.0, + 901.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 748.0, + 998.0, + 748.0, + 998.0, + 863.0, + 936.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 810.0, + 1057.0, + 810.0, + 1057.0, + 843.0, + 1040.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 748.0, + 1091.0, + 748.0, + 1091.0, + 810.0, + 1059.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 752.0, + 1200.0, + 752.0, + 1200.0, + 861.0, + 1132.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 748.0, + 1272.0, + 748.0, + 1272.0, + 864.0, + 1207.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 754.0, + 1308.0, + 754.0, + 1308.0, + 837.0, + 1279.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 784.0, + 1334.0, + 784.0, + 1334.0, + 814.0, + 1314.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 762.0, + 1349.0, + 762.0, + 1349.0, + 773.0, + 1336.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 861.0, + 375.0, + 861.0, + 375.0, + 878.0, + 343.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 861.0, + 412.0, + 861.0, + 412.0, + 879.0, + 381.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 861.0, + 453.0, + 861.0, + 453.0, + 878.0, + 423.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 861.0, + 491.0, + 861.0, + 491.0, + 878.0, + 460.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 861.0, + 532.0, + 861.0, + 532.0, + 878.0, + 500.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 861.0, + 570.0, + 861.0, + 570.0, + 878.0, + 539.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 861.0, + 646.0, + 861.0, + 646.0, + 878.0, + 614.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 861.0, + 683.0, + 861.0, + 683.0, + 878.0, + 653.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 861.0, + 729.0, + 861.0, + 729.0, + 878.0, + 694.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 861.0, + 762.0, + 861.0, + 762.0, + 878.0, + 731.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 861.0, + 802.0, + 861.0, + 802.0, + 878.0, + 771.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 861.0, + 840.0, + 861.0, + 840.0, + 878.0, + 809.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 861.0, + 969.0, + 861.0, + 969.0, + 878.0, + 901.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 861.0, + 1005.0, + 861.0, + 1005.0, + 878.0, + 974.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 861.0, + 1040.0, + 861.0, + 1040.0, + 878.0, + 1009.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 861.0, + 1076.0, + 861.0, + 1076.0, + 878.0, + 1046.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 861.0, + 1113.0, + 861.0, + 1113.0, + 878.0, + 1081.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 860.0, + 1242.0, + 860.0, + 1242.0, + 881.0, + 1173.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 861.0, + 1276.0, + 861.0, + 1276.0, + 878.0, + 1245.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 860.0, + 1312.0, + 860.0, + 1312.0, + 881.0, + 1277.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 861.0, + 1349.0, + 861.0, + 1349.0, + 878.0, + 1317.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 861.0, + 1383.0, + 861.0, + 1383.0, + 878.0, + 1351.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 872.0, + 495.0, + 872.0, + 495.0, + 893.0, + 431.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 870.0, + 766.0, + 870.0, + 766.0, + 893.0, + 701.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 870.0, + 1037.0, + 870.0, + 1037.0, + 893.0, + 974.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 870.0, + 1308.0, + 870.0, + 1308.0, + 893.0, + 1243.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.25, + 251.0, + 819.25, + 251.0, + 819.25, + 277.5, + 783.25, + 277.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 255.5, + 1229.0, + 255.5, + 1229.0, + 272.5, + 1204.0, + 272.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.75, + 334.0, + 795.75, + 334.0, + 795.75, + 354.0, + 738.75, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 354.5, + 1208.0, + 354.5, + 1208.0, + 366.0, + 1204.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 362.0, + 539.0, + 362.0, + 539.0, + 397.0, + 468.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 391.5, + 988.0, + 391.5, + 988.0, + 405.0, + 957.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 486.5, + 667.0, + 486.5, + 667.0, + 502.5, + 627.0, + 502.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 480.0, + 1376.0, + 480.0, + 1376.0, + 487.0, + 1342.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 496.5, + 1350.0, + 496.5, + 1350.0, + 521.5, + 1346.0, + 521.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.25, + 506.0, + 553.25, + 506.0, + 553.25, + 524.0, + 531.25, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 513.0, + 1090.0, + 513.0, + 1090.0, + 559.5, + 1071.0, + 559.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.25, + 546.0, + 734.25, + 546.0, + 734.25, + 577.0, + 730.25, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.25, + 574.0, + 728.25, + 574.0, + 728.25, + 585.0, + 695.25, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 595.5, + 714.0, + 595.5, + 714.0, + 620.0, + 640.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 602.0, + 1026.0, + 602.0, + 1026.0, + 611.0, + 1017.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 604.5, + 1255.0, + 604.5, + 1255.0, + 622.5, + 1223.0, + 622.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 571.0, + 1343.0, + 571.0, + 1343.0, + 590.0, + 1256.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 621.5, + 1081.0, + 621.5, + 1081.0, + 638.0, + 1046.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.25, + 719.5, + 1079.25, + 719.5, + 1079.25, + 748.0, + 1075.25, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.75, + 742.0, + 713.75, + 742.0, + 713.75, + 759.0, + 671.75, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 750.5, + 1350.0, + 750.5, + 1350.0, + 763.5, + 1345.0, + 763.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 774.5, + 1080.0, + 774.5, + 1080.0, + 800.0, + 1073.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.75, + 769.5, + 660.75, + 769.5, + 660.75, + 787.0, + 625.75, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.25, + 782.0, + 1060.25, + 782.0, + 1060.25, + 811.0, + 1054.25, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 798.0, + 502.0, + 798.0, + 502.0, + 820.0, + 453.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.75, + 815.5, + 463.75, + 815.5, + 463.75, + 837.5, + 451.75, + 837.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.75, + 840.0, + 700.75, + 840.0, + 700.75, + 853.5, + 674.75, + 853.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.25, + 826.0, + 732.25, + 826.0, + 732.25, + 839.0, + 693.25, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 827.0, + 1053.0, + 827.0, + 1053.0, + 839.0, + 1020.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 846.5, + 681.0, + 846.5, + 681.0, + 860.0, + 654.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 900.0, + 1403.0, + 900.0, + 1403.0, + 931.0, + 295.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 925.0, + 1177.0, + 925.0, + 1177.0, + 960.0, + 294.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1359.0, + 550.0, + 1359.0, + 550.0, + 1410.0, + 292.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1684.0, + 1405.0, + 1684.0, + 1405.0, + 1722.0, + 295.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1717.0, + 1406.0, + 1717.0, + 1406.0, + 1751.0, + 293.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1746.0, + 1406.0, + 1746.0, + 1406.0, + 1784.0, + 294.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1775.0, + 1405.0, + 1775.0, + 1405.0, + 1815.0, + 291.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1809.0, + 1406.0, + 1809.0, + 1406.0, + 1843.0, + 294.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1839.0, + 1406.0, + 1839.0, + 1406.0, + 1874.0, + 294.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1868.0, + 1404.0, + 1868.0, + 1404.0, + 1902.0, + 294.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1901.0, + 1404.0, + 1901.0, + 1404.0, + 1932.0, + 296.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1929.0, + 1405.0, + 1929.0, + 1405.0, + 1963.0, + 294.0, + 1963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1960.0, + 1404.0, + 1960.0, + 1404.0, + 1995.0, + 292.0, + 1995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1986.0, + 378.0, + 1986.0, + 378.0, + 2026.0, + 292.0, + 2026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1424.0, + 1405.0, + 1424.0, + 1405.0, + 1462.0, + 292.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1455.0, + 1407.0, + 1455.0, + 1407.0, + 1491.0, + 292.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1488.0, + 1405.0, + 1488.0, + 1405.0, + 1522.0, + 293.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1516.0, + 1405.0, + 1516.0, + 1405.0, + 1553.0, + 293.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1549.0, + 1404.0, + 1549.0, + 1404.0, + 1582.0, + 293.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1579.0, + 1405.0, + 1579.0, + 1405.0, + 1612.0, + 293.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1609.0, + 1405.0, + 1609.0, + 1405.0, + 1642.0, + 294.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1639.0, + 855.0, + 1639.0, + 855.0, + 1675.0, + 293.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 998.0, + 1405.0, + 998.0, + 1405.0, + 1033.0, + 312.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1030.0, + 1402.0, + 1030.0, + 1402.0, + 1061.0, + 337.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1060.0, + 1405.0, + 1060.0, + 1405.0, + 1093.0, + 335.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1092.0, + 1405.0, + 1092.0, + 1405.0, + 1123.0, + 340.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1119.0, + 1405.0, + 1119.0, + 1405.0, + 1153.0, + 336.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1151.0, + 1406.0, + 1151.0, + 1406.0, + 1185.0, + 336.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1182.0, + 1079.0, + 1182.0, + 1079.0, + 1216.0, + 335.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1238.0, + 1404.0, + 1238.0, + 1404.0, + 1275.0, + 295.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1271.0, + 1404.0, + 1271.0, + 1404.0, + 1304.0, + 296.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1301.0, + 1103.0, + 1301.0, + 1103.0, + 1332.0, + 295.0, + 1332.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 229, + 1403, + 229, + 1403, + 414, + 298, + 414 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 373, + 495, + 1404, + 495, + 1404, + 649, + 373, + 649 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 298, + 448, + 1147, + 448, + 1147, + 480, + 298, + 480 + ], + "score": 0.886 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.883 + }, + { + "category_id": 2, + "poly": [ + 837, + 2088, + 865, + 2088, + 865, + 2112, + 837, + 2112 + ], + "score": 0.822 + }, + { + "category_id": 1, + "poly": [ + 299, + 1459, + 1399, + 1459, + 1399, + 1524, + 299, + 1524 + ], + "score": 0.785 + }, + { + "category_id": 1, + "poly": [ + 289, + 1657, + 1401, + 1657, + 1401, + 1721, + 289, + 1721 + ], + "score": 0.781 + }, + { + "category_id": 1, + "poly": [ + 298, + 943, + 1405, + 943, + 1405, + 1068, + 298, + 1068 + ], + "score": 0.766 + }, + { + "category_id": 1, + "poly": [ + 305, + 747, + 1399, + 747, + 1399, + 840, + 305, + 840 + ], + "score": 0.756 + }, + { + "category_id": 1, + "poly": [ + 295, + 859, + 1399, + 859, + 1399, + 923, + 295, + 923 + ], + "score": 0.754 + }, + { + "category_id": 1, + "poly": [ + 303, + 1087, + 1402, + 1087, + 1402, + 1182, + 303, + 1182 + ], + "score": 0.75 + }, + { + "category_id": 1, + "poly": [ + 298, + 1543, + 1401, + 1543, + 1401, + 1637, + 298, + 1637 + ], + "score": 0.747 + }, + { + "category_id": 1, + "poly": [ + 296, + 1201, + 1403, + 1201, + 1403, + 1325, + 296, + 1325 + ], + "score": 0.717 + }, + { + "category_id": 0, + "poly": [ + 299, + 697, + 488, + 697, + 488, + 729, + 299, + 729 + ], + "score": 0.701 + }, + { + "category_id": 1, + "poly": [ + 299, + 1972, + 1396, + 1972, + 1396, + 2034, + 299, + 2034 + ], + "score": 0.621 + }, + { + "category_id": 1, + "poly": [ + 297, + 1743, + 1403, + 1743, + 1403, + 1868, + 297, + 1868 + ], + "score": 0.606 + }, + { + "category_id": 1, + "poly": [ + 297, + 1346, + 1401, + 1346, + 1401, + 1441, + 297, + 1441 + ], + "score": 0.582 + }, + { + "category_id": 1, + "poly": [ + 296, + 1888, + 1397, + 1888, + 1397, + 1951, + 296, + 1951 + ], + "score": 0.58 + }, + { + "category_id": 1, + "poly": [ + 299, + 697, + 488, + 697, + 488, + 729, + 299, + 729 + ], + "score": 0.173 + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 694.0, + 491.0, + 694.0, + 491.0, + 734.0, + 296.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 230.0, + 1404.0, + 230.0, + 1404.0, + 266.0, + 294.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 260.0, + 1405.0, + 260.0, + 1405.0, + 300.0, + 293.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 291.0, + 1405.0, + 291.0, + 1405.0, + 327.0, + 294.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 321.0, + 1405.0, + 321.0, + 1405.0, + 357.0, + 294.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 352.0, + 1405.0, + 352.0, + 1405.0, + 388.0, + 294.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 381.0, + 1405.0, + 381.0, + 1405.0, + 417.0, + 296.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 493.0, + 1101.0, + 493.0, + 1101.0, + 529.0, + 368.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 523.0, + 1405.0, + 523.0, + 1405.0, + 561.0, + 367.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 557.0, + 1040.0, + 557.0, + 1040.0, + 589.0, + 393.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 586.0, + 1056.0, + 586.0, + 1056.0, + 620.0, + 373.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 615.0, + 924.0, + 615.0, + 924.0, + 650.0, + 394.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 441.0, + 1153.0, + 441.0, + 1153.0, + 491.0, + 293.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1455.0, + 1402.0, + 1455.0, + 1402.0, + 1495.0, + 294.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1493.0, + 873.0, + 1493.0, + 873.0, + 1525.0, + 325.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1653.0, + 1404.0, + 1653.0, + 1404.0, + 1697.0, + 293.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1688.0, + 1091.0, + 1688.0, + 1091.0, + 1724.0, + 322.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 945.0, + 1405.0, + 945.0, + 1405.0, + 977.0, + 296.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 973.0, + 1403.0, + 973.0, + 1403.0, + 1011.0, + 320.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1005.0, + 1406.0, + 1005.0, + 1406.0, + 1042.0, + 322.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1035.0, + 922.0, + 1035.0, + 922.0, + 1070.0, + 322.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 744.0, + 1405.0, + 744.0, + 1405.0, + 783.0, + 299.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 777.0, + 1405.0, + 777.0, + 1405.0, + 814.0, + 321.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 808.0, + 1179.0, + 808.0, + 1179.0, + 845.0, + 325.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 855.0, + 1403.0, + 855.0, + 1403.0, + 897.0, + 294.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 892.0, + 755.0, + 892.0, + 755.0, + 924.0, + 323.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1083.0, + 1403.0, + 1083.0, + 1403.0, + 1126.0, + 294.0, + 1126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1119.0, + 1405.0, + 1119.0, + 1405.0, + 1156.0, + 322.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1148.0, + 844.0, + 1148.0, + 844.0, + 1184.0, + 323.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1544.0, + 1405.0, + 1544.0, + 1405.0, + 1578.0, + 297.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1573.0, + 1406.0, + 1573.0, + 1406.0, + 1612.0, + 321.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1603.0, + 1015.0, + 1603.0, + 1015.0, + 1641.0, + 323.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1199.0, + 1404.0, + 1199.0, + 1404.0, + 1239.0, + 293.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1230.0, + 1404.0, + 1230.0, + 1404.0, + 1269.0, + 320.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1262.0, + 1407.0, + 1262.0, + 1407.0, + 1300.0, + 320.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1295.0, + 507.0, + 1295.0, + 507.0, + 1327.0, + 320.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1969.0, + 1399.0, + 1969.0, + 1399.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 2002.0, + 1039.0, + 2002.0, + 1039.0, + 2036.0, + 323.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1742.0, + 1407.0, + 1742.0, + 1407.0, + 1779.0, + 296.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1776.0, + 1405.0, + 1776.0, + 1405.0, + 1809.0, + 325.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1807.0, + 1403.0, + 1807.0, + 1403.0, + 1839.0, + 324.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1837.0, + 742.0, + 1837.0, + 742.0, + 1870.0, + 324.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1345.0, + 1405.0, + 1345.0, + 1405.0, + 1383.0, + 295.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1375.0, + 1403.0, + 1375.0, + 1403.0, + 1413.0, + 322.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1409.0, + 988.0, + 1409.0, + 988.0, + 1442.0, + 323.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1886.0, + 1401.0, + 1886.0, + 1401.0, + 1925.0, + 294.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1920.0, + 1160.0, + 1920.0, + 1160.0, + 1953.0, + 321.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 694.0, + 491.0, + 694.0, + 491.0, + 734.0, + 296.0, + 734.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 816, + 75, + 816, + 105, + 299, + 105 + ], + "score": 0.87 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2113, + 836, + 2113 + ], + "score": 0.83 + }, + { + "category_id": 1, + "poly": [ + 289, + 197, + 1408, + 197, + 1408, + 1332, + 289, + 1332 + ], + "score": 0.638 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 227.0, + 1403.0, + 227.0, + 1403.0, + 268.0, + 291.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 261.0, + 974.0, + 261.0, + 974.0, + 298.0, + 323.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 312.0, + 1406.0, + 312.0, + 1406.0, + 350.0, + 295.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 344.0, + 1407.0, + 344.0, + 1407.0, + 380.0, + 322.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 375.0, + 1289.0, + 375.0, + 1289.0, + 411.0, + 325.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 424.0, + 1402.0, + 424.0, + 1402.0, + 466.0, + 294.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 457.0, + 1402.0, + 457.0, + 1402.0, + 493.0, + 322.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 489.0, + 898.0, + 489.0, + 898.0, + 523.0, + 321.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 539.0, + 1405.0, + 539.0, + 1405.0, + 578.0, + 294.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 570.0, + 1405.0, + 570.0, + 1405.0, + 606.0, + 322.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 602.0, + 1019.0, + 602.0, + 1019.0, + 638.0, + 322.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 653.0, + 1405.0, + 653.0, + 1405.0, + 689.0, + 294.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 685.0, + 1267.0, + 685.0, + 1267.0, + 719.0, + 320.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 737.0, + 1406.0, + 737.0, + 1406.0, + 774.0, + 294.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 770.0, + 1405.0, + 770.0, + 1405.0, + 802.0, + 323.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 798.0, + 1208.0, + 798.0, + 1208.0, + 834.0, + 324.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 849.0, + 1405.0, + 849.0, + 1405.0, + 888.0, + 292.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 878.0, + 1406.0, + 878.0, + 1406.0, + 919.0, + 320.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 911.0, + 395.0, + 911.0, + 395.0, + 946.0, + 323.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 961.0, + 1406.0, + 961.0, + 1406.0, + 1001.0, + 294.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 995.0, + 1335.0, + 995.0, + 1335.0, + 1029.0, + 322.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1044.0, + 1406.0, + 1044.0, + 1406.0, + 1086.0, + 294.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1076.0, + 1406.0, + 1076.0, + 1406.0, + 1114.0, + 319.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1107.0, + 395.0, + 1107.0, + 395.0, + 1143.0, + 323.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1157.0, + 1403.0, + 1157.0, + 1403.0, + 1199.0, + 291.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1190.0, + 775.0, + 1190.0, + 775.0, + 1226.0, + 321.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1242.0, + 1407.0, + 1242.0, + 1407.0, + 1282.0, + 292.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1273.0, + 973.0, + 1273.0, + 973.0, + 1313.0, + 320.0, + 1313.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 294, + 341, + 1344, + 341, + 1344, + 1087, + 294, + 1087 + ], + "score": 0.953 + }, + { + "category_id": 0, + "poly": [ + 298, + 295, + 514, + 295, + 514, + 329, + 298, + 329 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 816, + 74, + 816, + 106, + 299, + 106 + ], + "score": 0.88 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 866, + 2087, + 866, + 2113, + 836, + 2113 + ], + "score": 0.82 + }, + { + "category_id": 0, + "poly": [ + 300, + 226, + 511, + 226, + 511, + 262, + 300, + 262 + ], + "score": 0.627 + }, + { + "category_id": 1, + "poly": [ + 300, + 226, + 511, + 226, + 511, + 262, + 300, + 262 + ], + "score": 0.257 + }, + { + "category_id": 13, + "poly": [ + 490, + 343, + 584, + 343, + 584, + 377, + 490, + 377 + ], + "score": 0.58, + "latex": "\\operatorname { A U C } ( r )" + }, + { + "category_id": 13, + "poly": [ + 549, + 344, + 584, + 344, + 584, + 377, + 549, + 377 + ], + "score": 0.37, + "latex": "( r )" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 292.0, + 516.0, + 292.0, + 516.0, + 334.0, + 294.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 109.0, + 295.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 220.0, + 514.0, + 220.0, + 514.0, + 271.0, + 294.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 341.0, + 489.0, + 341.0, + 489.0, + 379.0, + 296.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 341.0, + 740.0, + 341.0, + 740.0, + 379.0, + 585.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 387.0, + 1269.0, + 387.0, + 1269.0, + 427.0, + 293.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 435.0, + 583.0, + 435.0, + 583.0, + 469.0, + 294.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 482.0, + 791.0, + 482.0, + 791.0, + 519.0, + 294.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 529.0, + 740.0, + 529.0, + 740.0, + 566.0, + 294.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 577.0, + 757.0, + 577.0, + 757.0, + 613.0, + 349.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 624.0, + 791.0, + 624.0, + 791.0, + 661.0, + 351.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 673.0, + 775.0, + 673.0, + 775.0, + 707.0, + 296.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 717.0, + 674.0, + 717.0, + 674.0, + 756.0, + 351.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 766.0, + 730.0, + 766.0, + 730.0, + 798.0, + 351.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 814.0, + 687.0, + 814.0, + 687.0, + 847.0, + 352.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 861.0, + 716.0, + 861.0, + 716.0, + 894.0, + 296.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 907.0, + 874.0, + 907.0, + 874.0, + 941.0, + 351.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 952.0, + 835.0, + 952.0, + 835.0, + 989.0, + 349.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1000.0, + 1342.0, + 1000.0, + 1342.0, + 1034.0, + 351.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1046.0, + 902.0, + 1046.0, + 902.0, + 1087.0, + 293.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 220.0, + 514.0, + 220.0, + 514.0, + 271.0, + 294.0, + 271.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 655, + 1406, + 655, + 1406, + 816, + 296, + 816 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1770, + 1404, + 1770, + 1404, + 1897, + 297, + 1897 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1187, + 1405, + 1187, + 1405, + 1312, + 297, + 1312 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1537, + 1404, + 1537, + 1404, + 1632, + 297, + 1632 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 1910, + 1403, + 1910, + 1403, + 2035, + 297, + 2035 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 296, + 1429, + 1405, + 1429, + 1405, + 1522, + 296, + 1522 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 301, + 1079, + 1404, + 1079, + 1404, + 1174, + 301, + 1174 + ], + "score": 0.963 + }, + { + "category_id": 8, + "poly": [ + 624, + 1637, + 1070, + 1637, + 1070, + 1713, + 624, + 1713 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 616, + 512, + 1079, + 512, + 1079, + 615, + 616, + 615 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 628, + 821, + 1071, + 821, + 1071, + 895, + 628, + 895 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 298, + 896, + 1399, + 896, + 1399, + 958, + 298, + 958 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 299, + 288, + 1404, + 288, + 1404, + 355, + 299, + 355 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 299, + 73, + 816, + 73, + 816, + 106, + 299, + 106 + ], + "score": 0.914 + }, + { + "category_id": 8, + "poly": [ + 393, + 1316, + 1302, + 1316, + 1302, + 1394, + 393, + 1394 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.867 + }, + { + "category_id": 1, + "poly": [ + 437, + 359, + 1253, + 359, + 1253, + 472, + 437, + 472 + ], + "score": 0.815 + }, + { + "category_id": 1, + "poly": [ + 313, + 475, + 1250, + 475, + 1250, + 509, + 313, + 509 + ], + "score": 0.649 + }, + { + "category_id": 0, + "poly": [ + 297, + 993, + 1314, + 993, + 1314, + 1054, + 297, + 1054 + ], + "score": 0.544 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 836, + 228, + 836, + 263, + 298, + 263 + ], + "score": 0.506 + }, + { + "category_id": 1, + "poly": [ + 297, + 993, + 1314, + 993, + 1314, + 1054, + 297, + 1054 + ], + "score": 0.476 + }, + { + "category_id": 0, + "poly": [ + 298, + 228, + 836, + 228, + 836, + 263, + 298, + 263 + ], + "score": 0.462 + }, + { + "category_id": 14, + "poly": [ + 626, + 1634, + 1073, + 1634, + 1073, + 1713, + 626, + 1713 + ], + "score": 0.94, + "latex": "C ^ { \\mathrm { o p t } } ( y = 1 \\mid x ) = { \\left\\{ \\begin{array} { l l } { { \\frac { 1 0 } { 1 1 } } } & { { \\mathrm { ~ i f ~ } } x \\in [ 0 , 1 ] } \\\\ { 0 } & { { \\mathrm { ~ i f ~ } } x \\in [ 1 , 2 ] } \\end{array} \\right. } ." + }, + { + "category_id": 14, + "poly": [ + 627, + 818, + 1072, + 818, + 1072, + 897, + 627, + 897 + ], + "score": 0.94, + "latex": "\\operatorname { A U C } ( r ) = \\int _ { t = \\infty } ^ { t = - \\infty } \\operatorname { T P R } _ { r } ( t ) d \\operatorname { F P R } _ { r } ( t )" + }, + { + "category_id": 13, + "poly": [ + 447, + 1461, + 558, + 1461, + 558, + 1494, + 447, + 1494 + ], + "score": 0.91, + "latex": "x \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 1217, + 291, + 1400, + 291, + 1400, + 326, + 1217, + 326 + ], + "score": 0.91, + "latex": "c _ { r } : \\mathbb { R } [ 0 , 1 ] ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 835, + 288, + 973, + 288, + 973, + 320, + 835, + 320 + ], + "score": 0.91, + "latex": "r : \\mathbb { R } ^ { d } \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 892, + 689, + 1178, + 689, + 1178, + 730, + 892, + 730 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\int _ { \\infty } ^ { - \\infty } \\mathrm { T P R } _ { r } ( t ) \\mathrm { F P R } _ { r } ^ { \\prime } ( t ) d t } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 409, + 659, + 474, + 659, + 474, + 690, + 409, + 690 + ], + "score": 0.9, + "latex": "\\mathrm { F P R } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 375, + 1801, + 405, + 1801, + 405, + 1840, + 375, + 1840 + ], + "score": 0.9, + "latex": "\\scriptstyle { \\frac { 1 } { 1 1 } }" + }, + { + "category_id": 13, + "poly": [ + 404, + 322, + 718, + 322, + 718, + 357, + 404, + 357 + ], + "score": 0.9, + "latex": "c _ { r } ( t ) = ( \\mathrm { F P R } _ { r } ( t ) , \\mathrm { T P R } _ { r } ( t ) )" + }, + { + "category_id": 13, + "poly": [ + 464, + 477, + 527, + 477, + 527, + 508, + 464, + 508 + ], + "score": 0.89, + "latex": "\\mathrm { F P R } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 560, + 725, + 625, + 725, + 625, + 755, + 560, + 755 + ], + "score": 0.89, + "latex": "\\mathrm { T P R } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 477, + 1803, + 532, + 1803, + 532, + 1833, + 477, + 1833 + ], + "score": 0.89, + "latex": "9 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 525, + 659, + 591, + 659, + 591, + 690, + 525, + 690 + ], + "score": 0.89, + "latex": "\\mathrm { T P R } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 446, + 725, + 510, + 725, + 510, + 755, + 446, + 755 + ], + "score": 0.89, + "latex": "\\mathrm { F P R } _ { r }" + }, + { + "category_id": 14, + "poly": [ + 614, + 509, + 1082, + 509, + 1082, + 617, + 614, + 617 + ], + "score": 0.89, + "latex": "\\begin{array} { r l r } & { \\underset { t - \\infty } { \\operatorname* { l i m } } \\mathrm { F P R } _ { r } ( t ) = 1 , \\ \\underset { t \\infty } { \\operatorname* { l i m } } \\mathrm { F P R } _ { r } ( t ) = 0 } & \\\\ & { \\underset { t - \\infty } { \\operatorname* { l i m } } \\mathrm { T P R } _ { r } ( t ) = 1 , \\ \\underset { t \\infty } { \\operatorname* { l i m } } \\mathrm { T P R } _ { r } ( t ) = 0 . } & \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 808, + 1942, + 1144, + 1942, + 1144, + 1976, + 808, + 1976 + ], + "score": 0.89, + "latex": "( p ( y = 1 ) = p ( y = 0 ) = 0 . 5 )" + }, + { + "category_id": 13, + "poly": [ + 778, + 901, + 829, + 901, + 829, + 924, + 778, + 924 + ], + "score": 0.88, + "latex": "- \\infty" + }, + { + "category_id": 13, + "poly": [ + 1191, + 1566, + 1221, + 1566, + 1221, + 1606, + 1191, + 1606 + ], + "score": 0.88, + "latex": "\\frac { 1 0 } { 1 1 }" + }, + { + "category_id": 13, + "poly": [ + 677, + 786, + 740, + 786, + 740, + 816, + 677, + 816 + ], + "score": 0.88, + "latex": "\\mathrm { F P R } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 579, + 477, + 644, + 477, + 644, + 508, + 579, + 508 + ], + "score": 0.88, + "latex": "\\mathrm { T P R } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 296, + 786, + 363, + 786, + 363, + 816, + 296, + 816 + ], + "score": 0.88, + "latex": "\\mathrm { T P R } _ { r }" + }, + { + "category_id": 13, + "poly": [ + 894, + 1539, + 944, + 1539, + 944, + 1567, + 894, + 1567 + ], + "score": 0.88, + "latex": "C ^ { \\mathrm { { o p t } } }" + }, + { + "category_id": 13, + "poly": [ + 780, + 1774, + 831, + 1774, + 831, + 1802, + 780, + 1802 + ], + "score": 0.87, + "latex": "C ^ { \\mathrm { { o p t } } }" + }, + { + "category_id": 13, + "poly": [ + 1318, + 1220, + 1372, + 1220, + 1372, + 1248, + 1318, + 1248 + ], + "score": 0.87, + "latex": "50 \\%" + }, + { + "category_id": 13, + "poly": [ + 1058, + 1189, + 1113, + 1189, + 1113, + 1218, + 1058, + 1218 + ], + "score": 0.86, + "latex": "50 \\%" + }, + { + "category_id": 13, + "poly": [ + 492, + 760, + 520, + 760, + 520, + 786, + 492, + 786 + ], + "score": 0.86, + "latex": "c _ { r }" + }, + { + "category_id": 13, + "poly": [ + 741, + 1250, + 795, + 1250, + 795, + 1279, + 741, + 1279 + ], + "score": 0.86, + "latex": "10 \\%" + }, + { + "category_id": 14, + "poly": [ + 395, + 1310, + 1304, + 1310, + 1304, + 1393, + 395, + 1393 + ], + "score": 0.86, + "latex": "\\begin{array} { r l r l } & { p ( x \\mid y = 1 ) = \\mathbb { 1 } ( x \\in [ 0 , 1 ] ) , } & & { \\qquad p ( y = 1 ) = 0 . 5 . } \\\\ & { p ( x \\mid y = 0 ) = 0 . 1 \\cdot \\mathbb { 1 } ( x \\in [ 0 , 1 ] ) + 0 . 9 \\cdot \\mathbb { 1 } ( x \\in [ 1 , 2 ] ) , } & & { \\qquad p ( y = 0 ) = 0 . 5 . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 384, + 1280, + 439, + 1280, + 439, + 1308, + 384, + 1308 + ], + "score": 0.85, + "latex": "90 \\%" + }, + { + "category_id": 13, + "poly": [ + 776, + 420, + 1253, + 420, + 1253, + 477, + 776, + 477 + ], + "score": 0.84, + "latex": "\\mathrm { T P R } _ { r } ( t ) : = P ^ { ( 1 ) } \\left( \\left\\{ g \\in \\mathbb { R } ^ { d } : r ( g ) > t \\right\\} \\right) ." + }, + { + "category_id": 13, + "poly": [ + 717, + 902, + 747, + 902, + 747, + 923, + 717, + 923 + ], + "score": 0.81, + "latex": "\\infty" + }, + { + "category_id": 13, + "poly": [ + 443, + 369, + 650, + 369, + 650, + 404, + 443, + 404 + ], + "score": 0.79, + "latex": "\\mathrm { F P R } _ { r } : \\mathbb { R } \\to [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 1176, + 480, + 1189, + 480, + 1189, + 503, + 1176, + 503 + ], + "score": 0.78, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 441, + 430, + 650, + 430, + 650, + 465, + 441, + 465 + ], + "score": 0.77, + "latex": "\\mathrm { T P R } _ { r } : \\mathbb { R } [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 1277, + 661, + 1290, + 661, + 1290, + 685, + 1277, + 685 + ], + "score": 0.75, + "latex": "t" + }, + { + "category_id": 14, + "poly": [ + 439, + 355, + 1255, + 355, + 1255, + 480, + 439, + 480 + ], + "score": 0.75, + "latex": "\\begin{array} { r l } & { \\mathrm { F P R } _ { r } : \\mathbb { R } [ 0 , 1 ] , \\mathrm { s u c h t h a t } \\mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } ( \\{ g \\in \\mathbb { R } ^ { d } : r ( g ) > t \\} ) } \\\\ & { \\mathrm { T P R } _ { r } : \\mathbb { R } [ 0 , 1 ] , \\mathrm { s u c h t h a t } \\mathrm { T P R } _ { r } ( t ) : = P ^ { ( 1 ) } ( \\{ g \\in \\mathbb { R } ^ { d } : r ( g ) > t \\} ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 529, + 230, + 626, + 230, + 626, + 264, + 529, + 264 + ], + "score": 0.61, + "latex": "\\operatorname { A U C } ( r )" + }, + { + "category_id": 13, + "poly": [ + 775, + 357, + 1248, + 357, + 1248, + 417, + 775, + 417 + ], + "score": 0.6, + "latex": "\\mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } \\left( \\left\\{ g \\in \\mathbb { R } ^ { d } : r ( g ) > t \\right\\} \\right)" + }, + { + "category_id": 13, + "poly": [ + 642, + 1569, + 699, + 1569, + 699, + 1601, + 642, + 1601 + ], + "score": 0.5, + "latex": "[ 0 , 1 ]" + }, + { + "category_id": 14, + "poly": [ + 772, + 357, + 1246, + 357, + 1246, + 416, + 772, + 416 + ], + "score": 0.26, + "latex": "\\mathrm { F P R } _ { r } ( t ) : = P ^ { ( 0 ) } \\left( \\left\\{ g \\in \\mathbb { R } ^ { d } : r ( g ) > t \\right\\} \\right)" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 990.0, + 1317.0, + 990.0, + 1317.0, + 1029.0, + 295.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1025.0, + 531.0, + 1025.0, + 531.0, + 1056.0, + 368.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 227.0, + 528.0, + 227.0, + 528.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 227.0, + 839.0, + 227.0, + 839.0, + 267.0, + 627.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 654.0, + 408.0, + 654.0, + 408.0, + 693.0, + 293.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 654.0, + 524.0, + 654.0, + 524.0, + 693.0, + 475.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 654.0, + 1276.0, + 654.0, + 1276.0, + 693.0, + 592.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 654.0, + 1406.0, + 654.0, + 1406.0, + 693.0, + 1291.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 696.0, + 891.0, + 696.0, + 891.0, + 730.0, + 295.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 688.0, + 1404.0, + 688.0, + 1404.0, + 732.0, + 1179.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 725.0, + 445.0, + 725.0, + 445.0, + 759.0, + 295.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 725.0, + 559.0, + 725.0, + 559.0, + 759.0, + 511.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 725.0, + 1403.0, + 725.0, + 1403.0, + 759.0, + 626.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 753.0, + 491.0, + 753.0, + 491.0, + 790.0, + 293.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 753.0, + 1404.0, + 753.0, + 1404.0, + 790.0, + 521.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 783.0, + 676.0, + 783.0, + 676.0, + 821.0, + 364.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 783.0, + 753.0, + 783.0, + 753.0, + 821.0, + 741.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 689.0, + 956.0, + 689.0, + 956.0, + 717.0, + 928.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1771.0, + 779.0, + 1771.0, + 779.0, + 1807.0, + 295.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1771.0, + 1406.0, + 1771.0, + 1406.0, + 1807.0, + 832.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1799.0, + 374.0, + 1799.0, + 374.0, + 1844.0, + 291.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1799.0, + 476.0, + 1799.0, + 476.0, + 1844.0, + 406.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1799.0, + 1407.0, + 1799.0, + 1407.0, + 1844.0, + 533.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1834.0, + 1405.0, + 1834.0, + 1405.0, + 1870.0, + 294.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1865.0, + 779.0, + 1865.0, + 779.0, + 1899.0, + 295.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1189.0, + 1057.0, + 1189.0, + 1057.0, + 1221.0, + 296.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 1189.0, + 1405.0, + 1189.0, + 1405.0, + 1221.0, + 1114.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1217.0, + 1317.0, + 1217.0, + 1317.0, + 1254.0, + 292.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 1217.0, + 1406.0, + 1217.0, + 1406.0, + 1254.0, + 1373.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1249.0, + 740.0, + 1249.0, + 740.0, + 1282.0, + 295.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 796.0, + 1249.0, + 1405.0, + 1249.0, + 1405.0, + 1282.0, + 796.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1279.0, + 383.0, + 1279.0, + 383.0, + 1315.0, + 294.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1279.0, + 927.0, + 1279.0, + 927.0, + 1315.0, + 440.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1533.0, + 893.0, + 1533.0, + 893.0, + 1575.0, + 292.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 1533.0, + 1405.0, + 1533.0, + 1405.0, + 1575.0, + 945.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1555.0, + 641.0, + 1555.0, + 641.0, + 1615.0, + 294.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1555.0, + 1190.0, + 1555.0, + 1190.0, + 1615.0, + 700.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 1555.0, + 1411.0, + 1555.0, + 1411.0, + 1615.0, + 1222.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1598.0, + 1123.0, + 1598.0, + 1123.0, + 1637.0, + 292.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1909.0, + 1403.0, + 1909.0, + 1403.0, + 1947.0, + 292.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1942.0, + 807.0, + 1942.0, + 807.0, + 1978.0, + 292.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 1942.0, + 1407.0, + 1942.0, + 1407.0, + 1978.0, + 1145.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1974.0, + 1404.0, + 1974.0, + 1404.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2002.0, + 747.0, + 2002.0, + 747.0, + 2038.0, + 292.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1430.0, + 1405.0, + 1430.0, + 1405.0, + 1463.0, + 295.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1461.0, + 446.0, + 1461.0, + 446.0, + 1495.0, + 295.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 1461.0, + 1403.0, + 1461.0, + 1403.0, + 1495.0, + 559.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1490.0, + 467.0, + 1490.0, + 467.0, + 1524.0, + 294.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1075.0, + 1404.0, + 1075.0, + 1404.0, + 1118.0, + 295.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1111.0, + 1403.0, + 1111.0, + 1403.0, + 1145.0, + 296.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1142.0, + 805.0, + 1142.0, + 805.0, + 1176.0, + 297.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 894.0, + 716.0, + 894.0, + 716.0, + 930.0, + 294.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 894.0, + 777.0, + 894.0, + 777.0, + 930.0, + 748.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 894.0, + 1404.0, + 894.0, + 1404.0, + 930.0, + 830.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 923.0, + 374.0, + 923.0, + 374.0, + 962.0, + 292.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 284.0, + 834.0, + 284.0, + 834.0, + 331.0, + 292.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 284.0, + 1216.0, + 284.0, + 1216.0, + 331.0, + 974.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 284.0, + 1404.0, + 284.0, + 1404.0, + 331.0, + 1401.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 320.0, + 403.0, + 320.0, + 403.0, + 357.0, + 294.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 320.0, + 787.0, + 320.0, + 787.0, + 357.0, + 719.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 344.0, + 438.0, + 344.0, + 438.0, + 420.0, + 431.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 408.0, + 438.0, + 408.0, + 438.0, + 479.0, + 432.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 473.0, + 463.0, + 473.0, + 463.0, + 512.0, + 307.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 473.0, + 578.0, + 473.0, + 578.0, + 512.0, + 528.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 473.0, + 1175.0, + 473.0, + 1175.0, + 512.0, + 645.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1190.0, + 473.0, + 1249.0, + 473.0, + 1249.0, + 512.0, + 1190.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 227.0, + 528.0, + 227.0, + 528.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 227.0, + 839.0, + 227.0, + 839.0, + 267.0, + 627.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 990.0, + 1317.0, + 990.0, + 1317.0, + 1029.0, + 295.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1025.0, + 531.0, + 1025.0, + 531.0, + 1056.0, + 368.0, + 1056.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 478, + 1405, + 478, + 1405, + 575, + 298, + 575 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 295, + 1110, + 1405, + 1110, + 1405, + 1289, + 295, + 1289 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 296, + 668, + 1405, + 668, + 1405, + 834, + 296, + 834 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 1775, + 1404, + 1775, + 1404, + 1842, + 298, + 1842 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 292, + 285, + 1403, + 285, + 1403, + 352, + 292, + 352 + ], + "score": 0.937 + }, + { + "category_id": 8, + "poly": [ + 340, + 589, + 1329, + 589, + 1329, + 654, + 340, + 654 + ], + "score": 0.931 + }, + { + "category_id": 8, + "poly": [ + 369, + 386, + 1325, + 386, + 1325, + 449, + 369, + 449 + ], + "score": 0.916 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 816, + 73, + 816, + 106, + 298, + 106 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 296, + 1061, + 1080, + 1061, + 1080, + 1097, + 296, + 1097 + ], + "score": 0.908 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1718, + 1400, + 1718, + 1400, + 1749, + 1352, + 1749 + ], + "score": 0.89 + }, + { + "category_id": 1, + "poly": [ + 287, + 842, + 1298, + 842, + 1298, + 879, + 287, + 879 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1640, + 1400, + 1640, + 1400, + 1671, + 1351, + 1671 + ], + "score": 0.879 + }, + { + "category_id": 9, + "poly": [ + 1365, + 916, + 1401, + 916, + 1401, + 947, + 1365, + 947 + ], + "score": 0.876 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1452, + 1401, + 1452, + 1401, + 1482, + 1365, + 1482 + ], + "score": 0.868 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.865 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1545, + 1401, + 1545, + 1401, + 1574, + 1365, + 1574 + ], + "score": 0.864 + }, + { + "category_id": 9, + "poly": [ + 1365, + 996, + 1401, + 996, + 1401, + 1026, + 1365, + 1026 + ], + "score": 0.859 + }, + { + "category_id": 9, + "poly": [ + 1365, + 607, + 1401, + 607, + 1401, + 638, + 1365, + 638 + ], + "score": 0.845 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1371, + 1401, + 1371, + 1401, + 1401, + 1366, + 1401 + ], + "score": 0.834 + }, + { + "category_id": 0, + "poly": [ + 300, + 228, + 640, + 228, + 640, + 262, + 300, + 262 + ], + "score": 0.72 + }, + { + "category_id": 8, + "poly": [ + 319, + 1429, + 975, + 1429, + 975, + 1504, + 319, + 1504 + ], + "score": 0.639 + }, + { + "category_id": 8, + "poly": [ + 320, + 1509, + 950, + 1509, + 950, + 1608, + 320, + 1608 + ], + "score": 0.605 + }, + { + "category_id": 8, + "poly": [ + 528, + 893, + 1173, + 893, + 1173, + 1050, + 528, + 1050 + ], + "score": 0.48 + }, + { + "category_id": 8, + "poly": [ + 526, + 894, + 956, + 894, + 956, + 968, + 526, + 968 + ], + "score": 0.467 + }, + { + "category_id": 8, + "poly": [ + 326, + 1347, + 1336, + 1347, + 1336, + 1424, + 326, + 1424 + ], + "score": 0.467 + }, + { + "category_id": 8, + "poly": [ + 626, + 973, + 1170, + 973, + 1170, + 1048, + 626, + 1048 + ], + "score": 0.447 + }, + { + "category_id": 8, + "poly": [ + 318, + 1303, + 1343, + 1303, + 1343, + 1425, + 318, + 1425 + ], + "score": 0.323 + }, + { + "category_id": 8, + "poly": [ + 319, + 1613, + 929, + 1613, + 929, + 1694, + 319, + 1694 + ], + "score": 0.309 + }, + { + "category_id": 1, + "poly": [ + 300, + 228, + 640, + 228, + 640, + 262, + 300, + 262 + ], + "score": 0.248 + }, + { + "category_id": 8, + "poly": [ + 320, + 1303, + 985, + 1303, + 985, + 1767, + 320, + 1767 + ], + "score": 0.105 + }, + { + "category_id": 8, + "poly": [ + 319, + 1611, + 927, + 1611, + 927, + 1767, + 319, + 1767 + ], + "score": 0.104 + }, + { + "category_id": 13, + "poly": [ + 778, + 1062, + 926, + 1062, + 926, + 1097, + 778, + 1097 + ], + "score": 0.94, + "latex": "\\mathrm { T P R } _ { r } ( t ) \\leq 1" + }, + { + "category_id": 14, + "poly": [ + 317, + 1294, + 1338, + 1294, + 1338, + 1768, + 317, + 1768 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\quad \\mathrm { A U C } ( r ) } \\\\ & { = \\displaystyle \\int _ { - \\infty } ^ { k } \\operatorname* { m i n } \\left( \\mathrm { F P R } _ { \\mathrm { v } } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } , \\mathbf { \\Phi } _ { 1 } \\right) d \\mathrm { F P R } _ { v } ( t ) + \\displaystyle \\int _ { k } ^ { - \\infty } \\operatorname* { m i n } \\left( \\mathrm { F P R } _ { v } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } , \\mathbf { \\Phi } _ { 1 } \\right) d \\mathrm { F P R } _ { v } ( t ) } \\\\ & { \\le \\displaystyle \\int _ { \\infty } ^ { k } \\left( \\mathrm { F P R } _ { v } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } \\right) d \\mathrm { F P R } _ { v } ( t ) + \\displaystyle \\int _ { k } ^ { - \\infty } 1 d \\mathrm { F P R } _ { v } ( t ) } \\\\ & { = \\left[ \\frac { \\left[ \\mathrm { F P R } _ { r } ( v ) \\right] ^ { 2 } } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } \\mathrm { F P R } _ { v } ( v ) \\right] \\displaystyle _ { t = \\infty } ^ { t - k } + \\mathrm { F P R } _ { v } ( t ) \\displaystyle _ { t = k } ^ { t - \\infty } } \\\\ & { \\le \\frac { ( 1 - \\frac { v \\epsilon } { 2 } ) ^ { 2 } } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } \\left( 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right) + \\left[ 1 - \\left( 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right) \\right] } \\\\ & { = \\frac { 1 } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } - \\frac { \\epsilon } { 8 } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 525, + 890, + 1172, + 890, + 1172, + 1052, + 525, + 1052 + ], + "score": 0.93, + "latex": "\\begin{array} { l } { \\displaystyle \\mathrm { A U C } ( { \\boldsymbol { r } } ) = \\int _ { \\infty } ^ { - \\infty } { \\mathrm { T P R } _ { r } ( t ) ~ d \\mathrm { F P R } _ { r } ( t ) } } \\\\ { \\displaystyle \\leq \\int _ { \\infty } ^ { - \\infty } \\operatorname* { m i n } \\bigg ( \\mathrm { F P R } _ { r } ( t ) + \\frac { \\sqrt { \\epsilon } } { 2 } , ~ 1 \\bigg ) ~ d \\mathrm { F P R } _ { r } ( t ) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 830, + 1152, + 1182, + 1152, + 1182, + 1207, + 830, + 1207 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\left\\{ t : \\mathrm { F P R } _ { r } ( t ) \\leq 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right\\} \\ \\neq \\ \\phi } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 334, + 586, + 1332, + 586, + 1332, + 655, + 334, + 655 + ], + "score": 0.93, + "latex": "\\mathbf { T } \\mathbf { V } ( P , Q ) \\leq { \\frac { 1 } { 2 } } ( { \\sqrt { \\mathbf { K L } ( P \\parallel Q ) / 2 } } + { \\sqrt { \\mathbf { K L } ( Q \\parallel P ) / 2 } } ) \\leq { \\frac { 1 } { 2 } } { \\sqrt { \\mathbf { K L } ( P \\parallel Q ) + \\mathbf { K L } ( Q \\parallel P ) } } ." + }, + { + "category_id": 13, + "poly": [ + 444, + 541, + 519, + 541, + 519, + 575, + 444, + 575 + ], + "score": 0.93, + "latex": "( P , Q )" + }, + { + "category_id": 13, + "poly": [ + 298, + 795, + 460, + 795, + 460, + 832, + 298, + 832 + ], + "score": 0.92, + "latex": "\\{ g : r ( g ) > t \\}" + }, + { + "category_id": 13, + "poly": [ + 296, + 1203, + 679, + 1203, + 679, + 1260, + 296, + 1260 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { k : = \\operatorname* { i n f } \\left\\{ t : \\mathrm { F P R } _ { r } ( t ) \\leq 1 - \\frac { \\sqrt { \\epsilon } } { 2 } \\right\\} } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 376, + 1118, + 508, + 1118, + 508, + 1152, + 376, + 1152 + ], + "score": 0.91, + "latex": "\\epsilon \\in \\ [ 0 , 4 )" + }, + { + "category_id": 13, + "poly": [ + 639, + 1110, + 852, + 1110, + 852, + 1156, + 639, + 1156 + ], + "score": 0.9, + "latex": "1 ~ - ~ \\frac { \\sqrt { \\epsilon } } { 2 } ~ \\in ~ ( 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 796, + 1214, + 892, + 1214, + 892, + 1249, + 796, + 1249 + ], + "score": 0.9, + "latex": "\\mathrm { F P R } _ { r } ( t )" + }, + { + "category_id": 13, + "poly": [ + 719, + 668, + 973, + 668, + 973, + 711, + 719, + 711 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\mathrm { T V } ( \\widetilde { P } ^ { ( 1 ) } , \\widetilde { P } ^ { ( 0 ) } ) \\le \\frac { \\sqrt { \\epsilon } } { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 528, + 288, + 671, + 288, + 671, + 319, + 528, + 319 + ], + "score": 0.9, + "latex": "0 \\leq \\epsilon < 4" + }, + { + "category_id": 13, + "poly": [ + 1004, + 1064, + 1070, + 1064, + 1070, + 1092, + 1004, + 1092 + ], + "score": 0.9, + "latex": "t \\in \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 501, + 711, + 1084, + 711, + 1084, + 750, + 501, + 750 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\mathrm { T V } ( \\widetilde P ^ { ( 1 ) } , \\widetilde P ^ { ( 0 ) } ) ) = \\operatorname* { m a x } _ { A \\subset \\mathbb { R } ^ { d } } [ \\widetilde P ^ { ( 1 ) } ( A ) - \\widetilde P ^ { ( 0 ) } ( A ) ] } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 372, + 380, + 1326, + 380, + 1326, + 452, + 372, + 452 + ], + "score": 0.88, + "latex": "\\mathbf { K L } \\big ( \\widetilde { P } ^ { ( 1 ) } \\mid \\big | \\widetilde { P } ^ { ( 0 ) } \\big ) + \\mathbf { K L } \\big ( \\widetilde { P } ^ { ( 0 ) } \\mid \\big | \\widetilde { P } ^ { ( 1 ) } \\big ) \\leq \\epsilon \\quad i m p l i e s \\quad \\operatorname* { m a x } _ { r } \\mathbf { A } \\mathbf { U C } ( r ) \\leq \\frac { 1 } { 2 } + \\frac { \\sqrt { \\epsilon } } { 2 } - \\frac { \\epsilon } { 8 } ." + }, + { + "category_id": 13, + "poly": [ + 928, + 1117, + 1026, + 1117, + 1026, + 1152, + 928, + 1152 + ], + "score": 0.88, + "latex": "\\mathrm { F P R } _ { r } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1151, + 289, + 1195, + 289, + 1195, + 315, + 1151, + 315 + ], + "score": 0.87, + "latex": "\\widetilde { P } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 1256, + 289, + 1300, + 289, + 1300, + 315, + 1256, + 315 + ], + "score": 0.87, + "latex": "\\widetilde { P } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 371, + 749, + 678, + 749, + 678, + 792, + 371, + 792 + ], + "score": 0.86, + "latex": "\\begin{array} { r } { \\widetilde { P } ^ { ( 1 ) } ( A ) - \\widetilde { P } ^ { ( 0 ) } ( A ) \\le \\frac { \\sqrt { \\epsilon } } { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 623, + 789, + 1179, + 789, + 1179, + 834, + 623, + 834 + ], + "score": 0.85, + "latex": "\\begin{array} { r } { \\bar { \\mathrm { T P R } } _ { r } ( t ) - \\mathrm { F P R } _ { r } ( t ) = \\widetilde { P } ^ { ( 1 ) } ( A ) - \\widetilde { P } ^ { ( 0 ) } ( A ) \\leq \\frac { \\sqrt { \\epsilon } } { 2 } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1345, + 715, + 1404, + 715, + 1404, + 748, + 1345, + 748 + ], + "score": 0.84, + "latex": "A \\subset" + }, + { + "category_id": 13, + "poly": [ + 1344, + 755, + 1405, + 755, + 1405, + 787, + 1344, + 787 + ], + "score": 0.83, + "latex": "A =" + }, + { + "category_id": 13, + "poly": [ + 297, + 750, + 333, + 750, + 333, + 785, + 297, + 785 + ], + "score": 0.79, + "latex": "\\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 783, + 852, + 800, + 852, + 800, + 872, + 783, + 872 + ], + "score": 0.77, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 326, + 1260, + 339, + 1260, + 339, + 1282, + 326, + 1282 + ], + "score": 0.75, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 890, + 1785, + 905, + 1785, + 905, + 1805, + 890, + 1805 + ], + "score": 0.74, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 479, + 1166, + 494, + 1166, + 494, + 1190, + 479, + 1190 + ], + "score": 0.73, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 982, + 763, + 1000, + 763, + 1000, + 783, + 982, + 783 + ], + "score": 0.72, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 1281, + 758, + 1295, + 758, + 1295, + 784, + 1281, + 784 + ], + "score": 0.71, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 667, + 1161, + 724, + 1161, + 724, + 1196, + 667, + 1196 + ], + "score": 0.68, + "latex": "[ 0 , 1 ]" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2122.0, + 832.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 229.0, + 642.0, + 229.0, + 642.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 476.0, + 1403.0, + 476.0, + 1403.0, + 518.0, + 292.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 509.0, + 1407.0, + 509.0, + 1407.0, + 546.0, + 292.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 540.0, + 443.0, + 540.0, + 443.0, + 578.0, + 295.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 540.0, + 1161.0, + 540.0, + 1161.0, + 578.0, + 520.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1106.0, + 375.0, + 1106.0, + 375.0, + 1160.0, + 292.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 1106.0, + 638.0, + 1106.0, + 638.0, + 1160.0, + 509.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1106.0, + 927.0, + 1106.0, + 927.0, + 1160.0, + 853.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 1106.0, + 1407.0, + 1106.0, + 1407.0, + 1160.0, + 1027.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1145.0, + 478.0, + 1145.0, + 478.0, + 1214.0, + 285.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 1145.0, + 666.0, + 1145.0, + 666.0, + 1214.0, + 495.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 1145.0, + 829.0, + 1145.0, + 829.0, + 1214.0, + 725.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1145.0, + 1413.0, + 1145.0, + 1413.0, + 1214.0, + 1183.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1203.0, + 295.0, + 1203.0, + 295.0, + 1258.0, + 292.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1203.0, + 795.0, + 1203.0, + 795.0, + 1258.0, + 680.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1203.0, + 1407.0, + 1203.0, + 1407.0, + 1258.0, + 893.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1252.0, + 325.0, + 1252.0, + 325.0, + 1290.0, + 293.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1252.0, + 1020.0, + 1252.0, + 1020.0, + 1290.0, + 340.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 664.0, + 718.0, + 664.0, + 718.0, + 717.0, + 290.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 664.0, + 1410.0, + 664.0, + 1410.0, + 717.0, + 974.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 707.0, + 500.0, + 707.0, + 500.0, + 756.0, + 291.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 707.0, + 1344.0, + 707.0, + 1344.0, + 756.0, + 1085.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 707.0, + 1408.0, + 707.0, + 1408.0, + 756.0, + 1405.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 744.0, + 296.0, + 744.0, + 296.0, + 796.0, + 290.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 744.0, + 370.0, + 744.0, + 370.0, + 796.0, + 334.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 744.0, + 981.0, + 744.0, + 981.0, + 796.0, + 679.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 744.0, + 1280.0, + 744.0, + 1280.0, + 796.0, + 1001.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 744.0, + 1343.0, + 744.0, + 1343.0, + 796.0, + 1296.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 784.0, + 297.0, + 784.0, + 297.0, + 840.0, + 291.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 784.0, + 622.0, + 784.0, + 622.0, + 840.0, + 461.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 784.0, + 1189.0, + 784.0, + 1189.0, + 840.0, + 1180.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1773.0, + 889.0, + 1773.0, + 889.0, + 1812.0, + 293.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 1773.0, + 1406.0, + 1773.0, + 1406.0, + 1812.0, + 906.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1806.0, + 506.0, + 1806.0, + 506.0, + 1845.0, + 292.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 1814.0, + 1402.0, + 1814.0, + 1402.0, + 1835.0, + 1379.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 282.0, + 527.0, + 282.0, + 527.0, + 325.0, + 304.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 282.0, + 1150.0, + 282.0, + 1150.0, + 325.0, + 672.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 282.0, + 1255.0, + 282.0, + 1255.0, + 325.0, + 1196.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 282.0, + 1406.0, + 282.0, + 1406.0, + 325.0, + 1301.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 319.0, + 841.0, + 319.0, + 841.0, + 355.0, + 295.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1060.0, + 777.0, + 1060.0, + 777.0, + 1099.0, + 296.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1060.0, + 1003.0, + 1060.0, + 1003.0, + 1099.0, + 927.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 1060.0, + 1082.0, + 1060.0, + 1082.0, + 1099.0, + 1071.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 839.0, + 782.0, + 839.0, + 782.0, + 883.0, + 294.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 839.0, + 1298.0, + 839.0, + 1298.0, + 883.0, + 801.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 229.0, + 642.0, + 229.0, + 642.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 345, + 1842, + 1361, + 1842, + 1361, + 2032, + 345, + 2032 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 296, + 330, + 1429, + 330, + 1429, + 425, + 296, + 425 + ], + "score": 0.961 + }, + { + "category_id": 8, + "poly": [ + 626, + 445, + 1068, + 445, + 1068, + 599, + 626, + 599 + ], + "score": 0.961 + }, + { + "category_id": 8, + "poly": [ + 627, + 1309, + 1068, + 1309, + 1068, + 1464, + 627, + 1464 + ], + "score": 0.96 + }, + { + "category_id": 8, + "poly": [ + 666, + 1556, + 1032, + 1556, + 1032, + 1652, + 666, + 1652 + ], + "score": 0.958 + }, + { + "category_id": 8, + "poly": [ + 303, + 784, + 1395, + 784, + 1395, + 1027, + 303, + 1027 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 295, + 1483, + 1401, + 1483, + 1401, + 1546, + 295, + 1546 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 299, + 1743, + 1400, + 1743, + 1400, + 1817, + 299, + 1817 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 294, + 1088, + 1404, + 1088, + 1404, + 1152, + 294, + 1152 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 300, + 1662, + 1396, + 1662, + 1396, + 1732, + 300, + 1732 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 405, + 1159, + 1299, + 1159, + 1299, + 1263, + 405, + 1263 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 816, + 73, + 816, + 106, + 298, + 106 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 296, + 608, + 463, + 608, + 463, + 639, + 296, + 639 + ], + "score": 0.913 + }, + { + "category_id": 8, + "poly": [ + 324, + 638, + 1332, + 638, + 1332, + 727, + 324, + 727 + ], + "score": 0.907 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1373, + 1401, + 1373, + 1401, + 1404, + 1351, + 1404 + ], + "score": 0.902 + }, + { + "category_id": 0, + "poly": [ + 299, + 227, + 914, + 227, + 914, + 263, + 299, + 263 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1614, + 1400, + 1614, + 1400, + 1645, + 1352, + 1645 + ], + "score": 0.896 + }, + { + "category_id": 1, + "poly": [ + 302, + 284, + 1019, + 284, + 1019, + 321, + 302, + 321 + ], + "score": 0.895 + }, + { + "category_id": 1, + "poly": [ + 300, + 735, + 1376, + 735, + 1376, + 777, + 300, + 777 + ], + "score": 0.891 + }, + { + "category_id": 1, + "poly": [ + 405, + 1268, + 534, + 1268, + 534, + 1299, + 405, + 1299 + ], + "score": 0.879 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.864 + }, + { + "category_id": 9, + "poly": [ + 1351, + 665, + 1401, + 665, + 1401, + 699, + 1351, + 699 + ], + "score": 0.847 + }, + { + "category_id": 14, + "poly": [ + 344, + 1839, + 1367, + 1839, + 1367, + 2034, + 344, + 2034 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } { \\underset { \\{ \\lambda _ { i } ^ { ( 1 ) } \\} , \\{ \\lambda _ { i } ^ { ( 0 ) } \\} , Q } { \\operatorname* { m i n } } } & { \\displaystyle \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { i } ^ { ( 0 ) } + u } + } \\\\ & { \\displaystyle \\left[ Q ( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } ) \\right] ^ { \\top } \\mathrm { d i a g } \\left( \\cdots , \\frac { 1 } { \\lambda _ { i } ^ { ( 0 ) } + u } + \\frac { 1 } { \\lambda _ { i } ^ { ( 1 ) } + v } , \\ldots \\right) Q ( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } ) } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 310, + 784, + 1394, + 784, + 1394, + 1028, + 310, + 1028 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\underset { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } { \\operatorname* { m i n } } ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u } } \\\\ & { \\mathrm { ~ s . t . ~ } p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) \\lambda _ { 2 } ^ { ( 0 ) } \\leq P , } \\\\ & { \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 , \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 1 } ^ { ( 0 ) } \\leq ~ 0 , \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 2 } ^ { ( 1 ) } \\leq ~ 0 , \\mathrm { ~ \\ ~ - ~ } \\lambda _ { 2 } ^ { ( 0 ) } \\leq ~ 0 , } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 1 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq ~ 0 , \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1014, + 1659, + 1120, + 1659, + 1120, + 1702, + 1014, + 1702 + ], + "score": 0.93, + "latex": "\\lambda _ { i } ^ { ( 0 ) } , \\lambda _ { i } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 374, + 1662, + 504, + 1662, + 504, + 1700, + 374, + 1700 + ], + "score": 0.93, + "latex": "Q \\in \\mathbb { R } ^ { d \\times d }" + }, + { + "category_id": 14, + "poly": [ + 629, + 1309, + 1070, + 1309, + 1070, + 1470, + 629, + 1470 + ], + "score": 0.93, + "latex": "\\begin{array} { c } { \\Sigma _ { 0 } \\Sigma _ { 1 } = \\Sigma _ { 1 } \\Sigma _ { 0 } , } \\\\ { p \\cdot \\operatorname { t r } ( \\Sigma _ { 1 } ) + ( 1 - p ) \\cdot \\operatorname { t r } ( \\Sigma _ { 0 } ) \\leq P , } \\\\ { \\Sigma _ { 1 } \\succeq \\mathbf { 0 } , } \\\\ { \\Sigma _ { 0 } \\succeq \\mathbf { 0 } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 1775, + 495, + 1775, + 495, + 1817, + 298, + 1817 + ], + "score": 0.93, + "latex": "\\{ \\lambda _ { i } ^ { ( 1 ) } \\} , \\{ \\lambda _ { i } ^ { ( 0 ) } \\} , Q" + }, + { + "category_id": 13, + "poly": [ + 372, + 733, + 660, + 733, + 660, + 777, + 372, + 777 + ], + "score": 0.93, + "latex": "( \\lambda _ { 1 } ^ { ( 0 ) * } , \\lambda _ { 2 } ^ { ( 0 ) * } , \\lambda _ { 1 } ^ { ( 1 ) * } , \\lambda _ { 2 } ^ { ( 1 ) * } )" + }, + { + "category_id": 14, + "poly": [ + 321, + 636, + 1329, + 636, + 1329, + 726, + 321, + 726 + ], + "score": 0.92, + "latex": "\\Sigma _ { 1 } ^ { * } = \\frac { \\tilde { \\lambda } _ { 1 } ^ { ( 1 ) * } - \\lambda _ { 2 } ^ { ( 1 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 1 ) * } I _ { d } , \\quad \\Sigma _ { 0 } ^ { * } = \\frac { \\lambda _ { 1 } ^ { ( 0 ) * } - \\lambda _ { 2 } ^ { ( 0 ) * } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 0 ) * } I _ { d } ," + }, + { + "category_id": 14, + "poly": [ + 630, + 444, + 1069, + 444, + 1069, + 604, + 630, + 604 + ], + "score": 0.92, + "latex": "\\begin{array} { c } { \\Sigma _ { 0 } \\Sigma _ { 1 } = \\Sigma _ { 1 } \\Sigma _ { 0 } , } \\\\ { p \\cdot \\operatorname { t r } ( \\Sigma _ { 1 } ) + ( 1 - p ) \\cdot \\operatorname { t r } ( \\Sigma _ { 0 } ) \\leq P , } \\\\ { \\Sigma _ { 1 } \\succeq \\mathbf { 0 } , } \\\\ { \\Sigma _ { 0 } \\succeq \\mathbf { 0 } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 668, + 1553, + 1030, + 1553, + 1030, + 1655, + 668, + 1655 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\Sigma _ { 0 } = \\boldsymbol { Q } ^ { \\top } \\mathrm { d i a g } ( \\lambda _ { 1 } ^ { ( 0 ) } , \\ldots , \\lambda _ { d } ^ { ( 0 ) } ) \\boldsymbol { Q } , } \\\\ { \\Sigma _ { 1 } = \\boldsymbol { Q } ^ { \\top } \\mathrm { d i a g } ( \\lambda _ { 1 } ^ { ( 1 ) } , \\ldots , \\lambda _ { d } ^ { ( 1 ) } ) \\boldsymbol { Q } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 700, + 289, + 735, + 289, + 735, + 322, + 700, + 322 + ], + "score": 0.9, + "latex": "\\Sigma _ { 0 } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 612, + 289, + 646, + 289, + 646, + 321, + 612, + 321 + ], + "score": 0.89, + "latex": "\\Sigma _ { 1 } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1112, + 1486, + 1147, + 1486, + 1147, + 1515, + 1112, + 1515 + ], + "score": 0.89, + "latex": "\\Sigma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1200, + 1486, + 1235, + 1486, + 1235, + 1515, + 1200, + 1515 + ], + "score": 0.89, + "latex": "\\Sigma _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 724, + 1746, + 759, + 1746, + 759, + 1776, + 724, + 1776 + ], + "score": 0.89, + "latex": "\\Sigma _ { 1 }" + }, + { + "category_id": 14, + "poly": [ + 330, + 328, + 1424, + 328, + 1424, + 386, + 330, + 386 + ], + "score": 0.89, + "latex": "\\operatorname* { m i n } _ { \\Sigma _ { 0 } , \\Sigma _ { 1 } \\in \\mathbb S } K L ( \\mathcal { N } ( \\bar { g } ^ { ( 1 ) } , v I + \\Sigma _ { 1 } ) \\parallel \\mathcal { N } ( \\bar { g } ^ { ( 0 ) } , u I + \\Sigma _ { 0 } ) ) + K L ( \\mathcal { N } ( \\bar { g } ^ { ( 0 ) } , u I + \\Sigma _ { 0 } ) \\parallel \\mathcal { N } ( \\bar { g } ^ { ( 1 ) } , v I + \\Sigma _ { 1 } ) )" + }, + { + "category_id": 13, + "poly": [ + 821, + 1746, + 857, + 1746, + 857, + 1776, + 821, + 1776 + ], + "score": 0.88, + "latex": "\\Sigma _ { 0 }" + }, + { + "category_id": 14, + "poly": [ + 405, + 1156, + 1300, + 1156, + 1300, + 1266, + 405, + 1266 + ], + "score": 0.88, + "latex": "\\begin{array} { c } { { \\displaystyle \\operatorname* { m i n } _ { \\Sigma _ { 0 } , \\Sigma _ { 1 } \\in \\mathbb S } \\mathrm { t r } \\big ( ( \\Sigma _ { 1 } + v I ) ^ { - 1 } ( \\Sigma _ { 0 } + u I ) \\big ) + \\mathrm { t r } \\big ( ( \\Sigma _ { 0 } + u I ) ^ { - 1 } ( \\Sigma _ { 1 } + v I ) \\big ) + } } \\\\ { { \\displaystyle \\big ( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } \\big ) ^ { \\top } \\left( ( \\Sigma _ { 1 } + v I ) ^ { - 1 } + ( \\Sigma _ { 0 } + u I ) ^ { - 1 } \\right) \\left( \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) } \\right) } } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 230.0, + 915.0, + 230.0, + 915.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 326.0, + 1429.0, + 326.0, + 1429.0, + 374.0, + 1425.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 387.0, + 430.0, + 387.0, + 430.0, + 431.0, + 291.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1485.0, + 1111.0, + 1485.0, + 1111.0, + 1517.0, + 297.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 1485.0, + 1199.0, + 1485.0, + 1199.0, + 1517.0, + 1148.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1485.0, + 1403.0, + 1485.0, + 1403.0, + 1517.0, + 1236.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1516.0, + 1225.0, + 1516.0, + 1225.0, + 1548.0, + 297.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1741.0, + 723.0, + 1741.0, + 723.0, + 1783.0, + 294.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 1741.0, + 820.0, + 1741.0, + 820.0, + 1783.0, + 760.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 1741.0, + 1406.0, + 1741.0, + 1406.0, + 1783.0, + 858.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1770.0, + 297.0, + 1770.0, + 297.0, + 1823.0, + 294.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 1770.0, + 509.0, + 1770.0, + 509.0, + 1823.0, + 496.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1087.0, + 1405.0, + 1087.0, + 1405.0, + 1128.0, + 293.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1119.0, + 946.0, + 1119.0, + 946.0, + 1156.0, + 294.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1650.0, + 373.0, + 1650.0, + 373.0, + 1710.0, + 289.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1650.0, + 1013.0, + 1650.0, + 1013.0, + 1710.0, + 505.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1650.0, + 1406.0, + 1650.0, + 1406.0, + 1710.0, + 1121.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1697.0, + 499.0, + 1697.0, + 499.0, + 1733.0, + 293.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 606.0, + 467.0, + 606.0, + 467.0, + 643.0, + 294.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 283.0, + 611.0, + 283.0, + 611.0, + 327.0, + 303.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 283.0, + 699.0, + 283.0, + 699.0, + 327.0, + 647.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 283.0, + 1018.0, + 283.0, + 1018.0, + 327.0, + 736.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 724.0, + 371.0, + 724.0, + 371.0, + 790.0, + 286.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 724.0, + 1392.0, + 724.0, + 1392.0, + 790.0, + 661.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1264.0, + 538.0, + 1264.0, + 538.0, + 1304.0, + 403.0, + 1304.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 1040, + 1407, + 1040, + 1407, + 1338, + 295, + 1338 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 1829, + 1405, + 1829, + 1405, + 1933, + 296, + 1933 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 296, + 631, + 1406, + 631, + 1406, + 734, + 296, + 734 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 419, + 790, + 1277, + 790, + 1277, + 986, + 419, + 986 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 629, + 1387, + 1063, + 1387, + 1063, + 1755, + 629, + 1755 + ], + "score": 0.961 + }, + { + "category_id": 8, + "poly": [ + 558, + 223, + 1140, + 223, + 1140, + 558, + 558, + 558 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 617, + 1980, + 1083, + 1980, + 1083, + 2032, + 617, + 2032 + ], + "score": 0.946 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 816, + 74, + 816, + 106, + 297, + 106 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 866, + 2087, + 866, + 2114, + 836, + 2114 + ], + "score": 0.855 + }, + { + "category_id": 13, + "poly": [ + 1096, + 1039, + 1286, + 1039, + 1286, + 1102, + 1096, + 1102 + ], + "score": 0.95, + "latex": "\\begin{array} { r } { \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { 1 } ^ { ( 0 ) } + u } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1002, + 1179, + 1109, + 1179, + 1109, + 1220, + 1002, + 1220 + ], + "score": 0.94, + "latex": "\\lambda _ { i } ^ { ( 0 ) } = 0" + }, + { + "category_id": 13, + "poly": [ + 513, + 1138, + 624, + 1138, + 624, + 1180, + 513, + 1180 + ], + "score": 0.94, + "latex": "\\lambda _ { i } ^ { ( 1 ) } = 0" + }, + { + "category_id": 13, + "poly": [ + 550, + 632, + 710, + 632, + 710, + 675, + 550, + 675 + ], + "score": 0.94, + "latex": "\\{ \\lambda _ { i } ^ { ( 1 ) } \\} , \\{ \\lambda _ { i } ^ { ( 0 ) } \\}" + }, + { + "category_id": 13, + "poly": [ + 435, + 1253, + 519, + 1253, + 519, + 1311, + 435, + 1311 + ], + "score": 0.94, + "latex": "\\left\\{ \\lambda _ { i } ^ { \\left( 1 \\right) } \\right\\}" + }, + { + "category_id": 13, + "poly": [ + 898, + 1828, + 974, + 1828, + 974, + 1871, + 898, + 1871 + ], + "score": 0.94, + "latex": "\\{ \\lambda _ { i } ^ { ( 1 ) } \\}" + }, + { + "category_id": 13, + "poly": [ + 726, + 1217, + 962, + 1217, + 962, + 1257, + 726, + 1257 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { m ( x ) = \\frac { u } { x + v } + \\frac { x + v } { u } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 417, + 789, + 1278, + 789, + 1278, + 986, + 417, + 986 + ], + "score": 0.93, + "latex": "\\begin{array} { c c l } { \\displaystyle \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\displaystyle \\sum _ { i = 1 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { i } ^ { ( 0 ) } + u } + \\frac { g } { \\lambda _ { 1 } ^ { ( 0 ) } + u } + \\frac { g } { \\lambda _ { 1 } ^ { ( 1 ) } + v } } \\\\ { = \\displaystyle \\sum _ { i = 2 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 0 ) } + u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\displaystyle \\sum _ { i = 2 } ^ { d } \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { \\lambda _ { i } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 560, + 221, + 1134, + 221, + 1134, + 556, + 560, + 556 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } { \\mathrm { s u b j e c t ~ t o ~ } } & { p ( \\displaystyle \\sum _ { i = 1 } ^ { d } \\lambda _ { i } ^ { ( 1 ) } ) + ( 1 - p ) ( \\displaystyle \\sum _ { i = 1 } ^ { d } \\lambda _ { i } ^ { ( 0 ) } ) \\le P } \\\\ & { - \\lambda _ { i } ^ { ( 1 ) } \\le 0 , \\forall i \\in [ d ] } \\\\ & { - \\lambda _ { i } ^ { ( 0 ) } \\le 0 , \\forall i \\in [ d ] . } \\\\ & { \\lambda _ { i } ^ { ( 1 ) } \\ge \\lambda _ { j } ^ { ( 1 ) } , \\forall i < j . } \\\\ & { \\lambda _ { i } ^ { ( 0 ) } \\ge \\lambda _ { j } ^ { ( 0 ) } , \\forall i < j . } \\\\ & { Q \\ \\mathrm { o r t h o g o n a l } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 627, + 1387, + 1063, + 1387, + 1063, + 1758, + 627, + 1758 + ], + "score": 0.93, + "latex": "\\begin{array} { l } { \\displaystyle \\frac { 1 } { d - 1 } \\sum _ { i = 2 } ^ { d } \\left( \\frac { u } { \\lambda _ { i } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { i } ^ { ( 1 ) } + v } { u } \\right) } \\\\ { \\displaystyle = \\frac { 1 } { d - 1 } \\sum _ { i = 2 } ^ { d } m ( \\lambda _ { i } ^ { ( 1 ) } ) } \\\\ { \\displaystyle \\geq m \\left( \\frac { 1 } { d - 1 } \\sum _ { i = 2 } ^ { d } \\lambda _ { i } ^ { ( 1 ) } \\right) } \\\\ { \\displaystyle = m \\left( \\frac { c } { d - 1 } \\right) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 667, + 1258, + 837, + 1258, + 837, + 1301, + 667, + 1301 + ], + "score": 0.93, + "latex": "\\Sigma _ { i = 2 } ^ { d } \\lambda _ { i } ^ { ( 1 ) } = c" + }, + { + "category_id": 13, + "poly": [ + 836, + 1140, + 947, + 1140, + 947, + 1180, + 836, + 1180 + ], + "score": 0.92, + "latex": "\\lambda _ { i } ^ { ( 0 ) } = 0" + }, + { + "category_id": 13, + "poly": [ + 637, + 1098, + 929, + 1098, + 929, + 1141, + 637, + 1141 + ], + "score": 0.92, + "latex": "p \\lambda _ { i } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { i } ^ { ( 0 ) } = c" + }, + { + "category_id": 13, + "poly": [ + 417, + 1187, + 494, + 1187, + 494, + 1217, + 417, + 1217 + ], + "score": 0.9, + "latex": "u \\geq v" + }, + { + "category_id": 13, + "poly": [ + 583, + 1871, + 655, + 1871, + 655, + 1900, + 583, + 1900 + ], + "score": 0.9, + "latex": "u \\leq v" + }, + { + "category_id": 13, + "poly": [ + 332, + 1869, + 397, + 1869, + 397, + 1900, + 332, + 1900 + ], + "score": 0.89, + "latex": "i \\geq 2" + }, + { + "category_id": 13, + "poly": [ + 1191, + 1185, + 1262, + 1185, + 1262, + 1217, + 1191, + 1217 + ], + "score": 0.89, + "latex": "i \\geq 2" + }, + { + "category_id": 13, + "poly": [ + 1295, + 1837, + 1361, + 1837, + 1361, + 1873, + 1295, + 1873 + ], + "score": 0.89, + "latex": "\\scriptstyle { \\big ( } { \\frac { c } { d - 1 } } { \\big ) }" + }, + { + "category_id": 13, + "poly": [ + 1022, + 1147, + 1103, + 1147, + 1103, + 1177, + 1022, + 1177 + ], + "score": 0.88, + "latex": "u \\geq v" + }, + { + "category_id": 13, + "poly": [ + 814, + 672, + 855, + 672, + 855, + 704, + 814, + 704 + ], + "score": 0.87, + "latex": "\\Delta g" + }, + { + "category_id": 13, + "poly": [ + 699, + 1148, + 781, + 1148, + 781, + 1176, + 699, + 1176 + ], + "score": 0.86, + "latex": "u \\ : < \\ : v" + }, + { + "category_id": 13, + "poly": [ + 1200, + 642, + 1226, + 642, + 1226, + 672, + 1200, + 672 + ], + "score": 0.85, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 1173, + 673, + 1198, + 673, + 1198, + 704, + 1173, + 704 + ], + "score": 0.84, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 709, + 1045, + 835, + 1045, + 835, + 1088, + 709, + 1088 + ], + "score": 0.82, + "latex": "( \\lambda _ { i } ^ { ( 1 ) } , \\lambda _ { i } ^ { ( 0 ) } )" + }, + { + "category_id": 13, + "poly": [ + 799, + 1983, + 930, + 1983, + 930, + 2030, + 799, + 2030 + ], + "score": 0.81, + "latex": "\\lambda _ { i } ^ { ( 1 ) } = \\lambda _ { j } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 986, + 1991, + 1074, + 1991, + 1074, + 2024, + 986, + 2024 + ], + "score": 0.79, + "latex": "i , j \\geq 2" + }, + { + "category_id": 13, + "poly": [ + 1168, + 1113, + 1185, + 1113, + 1185, + 1134, + 1168, + 1134 + ], + "score": 0.71, + "latex": "c" + }, + { + "category_id": 13, + "poly": [ + 1101, + 1273, + 1116, + 1273, + 1116, + 1293, + 1101, + 1293 + ], + "score": 0.7, + "latex": "c" + }, + { + "category_id": 14, + "poly": [ + 614, + 1981, + 1074, + 1981, + 1074, + 2031, + 614, + 2031 + ], + "score": 0.69, + "latex": "\\lambda _ { i } ^ { ( 0 ) } = \\lambda _ { j } ^ { ( 0 ) } \\mathrm { a n d } \\lambda _ { i } ^ { ( 1 ) } = \\lambda _ { j } ^ { ( 1 ) } , \\mathrm { f o r } i , j \\ge 2" + }, + { + "category_id": 13, + "poly": [ + 848, + 1050, + 931, + 1050, + 931, + 1086, + 848, + 1086 + ], + "score": 0.6, + "latex": "( i ~ \\geq ~ 2 )" + }, + { + "category_id": 13, + "poly": [ + 616, + 1982, + 748, + 1982, + 748, + 2030, + 616, + 2030 + ], + "score": 0.54, + "latex": "\\lambda _ { i } ^ { ( 0 ) } = \\lambda _ { j } ^ { ( 0 ) }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 109.0, + 296.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1023.0, + 708.0, + 1023.0, + 708.0, + 1107.0, + 293.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1023.0, + 847.0, + 1023.0, + 847.0, + 1107.0, + 836.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 1023.0, + 1095.0, + 1023.0, + 1095.0, + 1107.0, + 932.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 1030.0, + 1407.0, + 1030.0, + 1407.0, + 1112.0, + 1287.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1093.0, + 636.0, + 1093.0, + 636.0, + 1147.0, + 289.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1093.0, + 1167.0, + 1093.0, + 1167.0, + 1147.0, + 930.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 1093.0, + 1412.0, + 1093.0, + 1412.0, + 1147.0, + 1186.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1132.0, + 512.0, + 1132.0, + 512.0, + 1187.0, + 290.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 1132.0, + 698.0, + 1132.0, + 698.0, + 1187.0, + 625.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 1132.0, + 835.0, + 1132.0, + 835.0, + 1187.0, + 782.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 1132.0, + 1021.0, + 1132.0, + 1021.0, + 1187.0, + 948.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 1132.0, + 1413.0, + 1132.0, + 1413.0, + 1187.0, + 1104.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1173.0, + 416.0, + 1173.0, + 416.0, + 1222.0, + 288.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 1173.0, + 1001.0, + 1173.0, + 1001.0, + 1222.0, + 495.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 1173.0, + 1190.0, + 1173.0, + 1190.0, + 1222.0, + 1110.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 1173.0, + 1411.0, + 1173.0, + 1411.0, + 1222.0, + 1263.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1209.0, + 434.0, + 1209.0, + 434.0, + 1316.0, + 285.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 1209.0, + 666.0, + 1209.0, + 666.0, + 1316.0, + 520.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1209.0, + 1100.0, + 1209.0, + 1100.0, + 1316.0, + 963.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 1209.0, + 1417.0, + 1209.0, + 1417.0, + 1316.0, + 1117.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1299.0, + 361.0, + 1299.0, + 361.0, + 1343.0, + 291.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1206.0, + 893.0, + 1206.0, + 893.0, + 1264.0, + 287.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1824.0, + 897.0, + 1824.0, + 897.0, + 1877.0, + 290.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 1824.0, + 1294.0, + 1824.0, + 1294.0, + 1877.0, + 975.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1362.0, + 1824.0, + 1408.0, + 1824.0, + 1408.0, + 1877.0, + 1362.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1866.0, + 331.0, + 1866.0, + 331.0, + 1905.0, + 293.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1866.0, + 582.0, + 1866.0, + 582.0, + 1905.0, + 398.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1866.0, + 1405.0, + 1866.0, + 1405.0, + 1905.0, + 656.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1901.0, + 463.0, + 1901.0, + 463.0, + 1933.0, + 294.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 629.0, + 549.0, + 629.0, + 549.0, + 681.0, + 289.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 629.0, + 1199.0, + 629.0, + 1199.0, + 681.0, + 711.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 629.0, + 1411.0, + 629.0, + 1411.0, + 681.0, + 1227.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 667.0, + 813.0, + 667.0, + 813.0, + 708.0, + 291.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 667.0, + 1172.0, + 667.0, + 1172.0, + 708.0, + 856.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 667.0, + 1406.0, + 667.0, + 1406.0, + 708.0, + 1199.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 700.0, + 532.0, + 700.0, + 532.0, + 737.0, + 295.0, + 737.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1299, + 1405, + 1299, + 1405, + 1640, + 297, + 1640 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 296, + 1701, + 1405, + 1701, + 1405, + 2035, + 296, + 2035 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 955, + 1404, + 955, + 1404, + 1071, + 298, + 1071 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 296, + 224, + 1405, + 224, + 1405, + 339, + 296, + 339 + ], + "score": 0.968 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 816, + 73, + 816, + 106, + 298, + 106 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 299, + 1084, + 603, + 1084, + 603, + 1117, + 299, + 1117 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 864, + 2087, + 864, + 2112, + 836, + 2112 + ], + "score": 0.855 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1085, + 1403, + 1085, + 1403, + 1114, + 1374, + 1114 + ], + "score": 0.791 + }, + { + "category_id": 8, + "poly": [ + 300, + 402, + 1400, + 402, + 1400, + 875, + 300, + 875 + ], + "score": 0.768 + }, + { + "category_id": 9, + "poly": [ + 1351, + 494, + 1402, + 494, + 1402, + 524, + 1351, + 524 + ], + "score": 0.518 + }, + { + "category_id": 8, + "poly": [ + 304, + 531, + 1247, + 531, + 1247, + 876, + 304, + 876 + ], + "score": 0.165 + }, + { + "category_id": 8, + "poly": [ + 302, + 403, + 1394, + 403, + 1394, + 500, + 302, + 500 + ], + "score": 0.108 + }, + { + "category_id": 13, + "poly": [ + 1138, + 1775, + 1252, + 1775, + 1252, + 1815, + 1138, + 1815 + ], + "score": 0.94, + "latex": "\\lambda _ { 2 } ^ { ( 1 ) * } = 0" + }, + { + "category_id": 13, + "poly": [ + 378, + 1360, + 515, + 1360, + 515, + 1396, + 378, + 1396 + ], + "score": 0.93, + "latex": "( \\Delta g ) ( \\dot { \\Delta } g ) ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 996, + 997, + 1066, + 997, + 1066, + 1043, + 996, + 1043 + ], + "score": 0.93, + "latex": "\\frac { \\Delta g } { \\| \\Delta g \\| _ { 2 } }" + }, + { + "category_id": 14, + "poly": [ + 306, + 530, + 1221, + 530, + 1221, + 877, + 306, + 877 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } { \\mathrm { s u b j e c t ~ t o ~ } } & { \\ p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\lambda _ { 2 } ^ { ( 1 ) } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } + ( 1 - p ) ( d - 1 ) ( \\lambda _ { 2 } ^ { ( 0 ) } ) \\leq P } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 0 } ^ { ( 0 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 2 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad - \\lambda _ { 2 } ^ { ( 0 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 1 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 1 ) } \\leq 0 } \\\\ & { \\quad \\quad \\quad \\lambda _ { 2 } ^ { ( 0 ) } - \\lambda _ { 1 } ^ { ( 0 ) } \\leq 0 . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 516, + 1393, + 613, + 1393, + 613, + 1425, + 516, + 1425 + ], + "score": 0.92, + "latex": "( A + B )" + }, + { + "category_id": 13, + "poly": [ + 833, + 286, + 1092, + 286, + 1092, + 341, + 833, + 341 + ], + "score": 0.92, + "latex": "\\left\\{ \\lambda _ { 1 } ^ { ( 1 ) } , \\lambda _ { 2 } ^ { ( 1 ) } , \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 0 ) } \\right\\}" + }, + { + "category_id": 13, + "poly": [ + 919, + 223, + 1193, + 223, + 1193, + 288, + 919, + 288 + ], + "score": 0.92, + "latex": "\\left\\{ \\lambda _ { i } ^ { ( 1 ) } \\right\\} _ { i = 1 } ^ { d } \\bigcup \\left\\{ \\lambda _ { i } ^ { ( 0 ) } \\right\\} _ { i = 1 } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 674, + 1453, + 721, + 1453, + 721, + 1489, + 674, + 1489 + ], + "score": 0.91, + "latex": "\\eta ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 966, + 1734, + 1404, + 1734, + 1404, + 1776, + 966, + 1776 + ], + "score": 0.91, + "latex": "p \\lambda _ { 1 } ^ { ( 1 ) } + p ( d - 1 ) \\bar { \\lambda _ { 2 } ^ { ( 1 ) } } + ( 1 - p ) \\lambda _ { 1 } ^ { ( 0 ) } +" + }, + { + "category_id": 13, + "poly": [ + 298, + 1773, + 577, + 1773, + 577, + 1816, + 298, + 1816 + ], + "score": 0.91, + "latex": "( 1 - p ) ( d - 1 ) \\lambda _ { 2 } ^ { ( 0 ) } = P" + }, + { + "category_id": 13, + "poly": [ + 711, + 1515, + 811, + 1515, + 811, + 1550, + 711, + 1550 + ], + "score": 0.91, + "latex": "\\Delta g \\Delta g ^ { \\top } )" + }, + { + "category_id": 13, + "poly": [ + 570, + 1453, + 617, + 1453, + 617, + 1488, + 570, + 1488 + ], + "score": 0.91, + "latex": "\\eta ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1813, + 415, + 1813, + 415, + 1854, + 297, + 1854 + ], + "score": 0.91, + "latex": "\\lambda _ { 2 } ^ { ( 0 ) * } = 0" + }, + { + "category_id": 14, + "poly": [ + 310, + 404, + 1395, + 404, + 1395, + 496, + 310, + 496 + ], + "score": 0.9, + "latex": "\\operatorname* { m i n } _ { \\lambda _ { 1 } ^ { ( 0 ) } , \\lambda _ { 1 } ^ { ( 1 ) } , \\lambda _ { 2 } ^ { ( 0 ) } , \\lambda _ { 2 } ^ { ( 1 ) } } \\left( d - 1 \\right) \\frac { \\lambda _ { 2 } ^ { ( 0 ) } + u } { \\lambda _ { 2 } ^ { ( 1 ) } + v } + ( d - 1 ) \\frac { \\lambda _ { 2 } ^ { ( 1 ) } + v } { \\lambda _ { 2 } ^ { ( 0 ) } + u } + \\frac { \\lambda _ { 1 } ^ { ( 0 ) } + u + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 1 ) } + v } + \\frac { \\lambda _ { 1 } ^ { ( 1 ) } + v + \\| \\Delta g \\| _ { 2 } ^ { 2 } } { \\lambda _ { 1 } ^ { ( 0 ) } + u }" + }, + { + "category_id": 13, + "poly": [ + 453, + 1038, + 487, + 1038, + 487, + 1068, + 453, + 1068 + ], + "score": 0.9, + "latex": "\\Sigma _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 976, + 956, + 1264, + 956, + 1264, + 999, + 976, + 999 + ], + "score": 0.89, + "latex": "( \\lambda _ { 1 } ^ { ( 0 ) * } , \\lambda _ { 2 } ^ { ( 0 ) * } , \\lambda _ { 1 } ^ { ( 1 ) * } , \\lambda _ { 2 } ^ { ( 1 ) * } )" + }, + { + "category_id": 13, + "poly": [ + 677, + 1303, + 713, + 1303, + 713, + 1334, + 677, + 1334 + ], + "score": 0.89, + "latex": "\\Sigma _ { 1 } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 539, + 1037, + 574, + 1037, + 574, + 1069, + 539, + 1069 + ], + "score": 0.88, + "latex": "\\Sigma _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 445, + 1821, + 521, + 1821, + 521, + 1851, + 445, + 1851 + ], + "score": 0.88, + "latex": "u \\geq v" + }, + { + "category_id": 13, + "poly": [ + 768, + 1303, + 805, + 1303, + 805, + 1335, + 768, + 1335 + ], + "score": 0.88, + "latex": "\\Sigma _ { 0 } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 783, + 1364, + 811, + 1364, + 811, + 1393, + 783, + 1393 + ], + "score": 0.88, + "latex": "I _ { d }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1000, + 323, + 1000, + 323, + 1033, + 297, + 1033 + ], + "score": 0.83, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 1280, + 1784, + 1354, + 1784, + 1354, + 1811, + 1280, + 1811 + ], + "score": 0.83, + "latex": "u < v" + }, + { + "category_id": 13, + "poly": [ + 853, + 1424, + 879, + 1424, + 879, + 1452, + 853, + 1452 + ], + "score": 0.79, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 774, + 242, + 806, + 242, + 806, + 269, + 774, + 269 + ], + "score": 0.78, + "latex": "2 d" + }, + { + "category_id": 13, + "poly": [ + 772, + 1424, + 797, + 1424, + 797, + 1451, + 772, + 1451 + ], + "score": 0.78, + "latex": "A" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1089.0, + 1403.0, + 1089.0, + 1403.0, + 1115.0, + 1378.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1300.0, + 676.0, + 1300.0, + 676.0, + 1339.0, + 294.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1300.0, + 767.0, + 1300.0, + 767.0, + 1339.0, + 714.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 1300.0, + 1406.0, + 1300.0, + 1406.0, + 1339.0, + 806.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1333.0, + 1405.0, + 1333.0, + 1405.0, + 1364.0, + 294.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1362.0, + 377.0, + 1362.0, + 377.0, + 1397.0, + 294.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1362.0, + 782.0, + 1362.0, + 782.0, + 1397.0, + 516.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1362.0, + 1405.0, + 1362.0, + 1405.0, + 1397.0, + 812.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1393.0, + 515.0, + 1393.0, + 515.0, + 1427.0, + 292.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1393.0, + 1405.0, + 1393.0, + 1405.0, + 1427.0, + 614.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1422.0, + 771.0, + 1422.0, + 771.0, + 1457.0, + 294.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 1422.0, + 852.0, + 1422.0, + 852.0, + 1457.0, + 798.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 1422.0, + 1403.0, + 1422.0, + 1403.0, + 1457.0, + 880.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1451.0, + 569.0, + 1451.0, + 569.0, + 1493.0, + 291.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 1451.0, + 673.0, + 1451.0, + 673.0, + 1493.0, + 618.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1451.0, + 1407.0, + 1451.0, + 1407.0, + 1493.0, + 722.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1487.0, + 1405.0, + 1487.0, + 1405.0, + 1522.0, + 295.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1518.0, + 710.0, + 1518.0, + 710.0, + 1553.0, + 295.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1518.0, + 1405.0, + 1518.0, + 1405.0, + 1553.0, + 812.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1546.0, + 1405.0, + 1546.0, + 1405.0, + 1582.0, + 294.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1577.0, + 1406.0, + 1577.0, + 1406.0, + 1615.0, + 292.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1607.0, + 895.0, + 1607.0, + 895.0, + 1645.0, + 292.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1703.0, + 1406.0, + 1703.0, + 1406.0, + 1741.0, + 295.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1721.0, + 965.0, + 1721.0, + 965.0, + 1790.0, + 284.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1757.0, + 297.0, + 1757.0, + 297.0, + 1835.0, + 292.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1757.0, + 1137.0, + 1757.0, + 1137.0, + 1835.0, + 578.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 1757.0, + 1279.0, + 1757.0, + 1279.0, + 1835.0, + 1253.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1757.0, + 1406.0, + 1757.0, + 1406.0, + 1835.0, + 1355.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1811.0, + 444.0, + 1811.0, + 444.0, + 1859.0, + 416.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1811.0, + 1407.0, + 1811.0, + 1407.0, + 1859.0, + 522.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1850.0, + 1407.0, + 1850.0, + 1407.0, + 1888.0, + 291.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1406.0, + 1882.0, + 1406.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 1406.0, + 1911.0, + 1406.0, + 1946.0, + 294.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2002.0, + 1240.0, + 2002.0, + 1240.0, + 2037.0, + 296.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 939.0, + 975.0, + 939.0, + 975.0, + 1021.0, + 296.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 939.0, + 1405.0, + 939.0, + 1405.0, + 1021.0, + 1265.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 998.0, + 995.0, + 998.0, + 995.0, + 1037.0, + 324.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 994.0, + 1406.0, + 994.0, + 1406.0, + 1042.0, + 1079.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1037.0, + 452.0, + 1037.0, + 452.0, + 1072.0, + 293.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 1037.0, + 538.0, + 1037.0, + 538.0, + 1072.0, + 488.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1037.0, + 1036.0, + 1037.0, + 1036.0, + 1072.0, + 575.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 990.5, + 1052.0, + 990.5, + 1052.0, + 1022.0, + 1015.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.75, + 999.5, + 1089.75, + 999.5, + 1089.75, + 1056.0, + 983.75, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 236.0, + 1404.0, + 236.0, + 1404.0, + 276.0, + 1194.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 280.0, + 258.0, + 832.0, + 258.0, + 832.0, + 354.0, + 280.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 258.0, + 1116.0, + 258.0, + 1116.0, + 354.0, + 1093.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 260.0, + 1197.0, + 260.0, + 1197.0, + 292.0, + 1194.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 216.0, + 1039.0, + 216.0, + 1039.0, + 298.0, + 282.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 212.0, + 1196.0, + 212.0, + 1196.0, + 294.0, + 1023.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1079.0, + 607.0, + 1079.0, + 607.0, + 1122.0, + 296.0, + 1122.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 293, + 285, + 1402, + 285, + 1402, + 352, + 293, + 352 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 267, + 390, + 1366, + 390, + 1366, + 1399, + 267, + 1399 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 262, + 1417, + 1367, + 1417, + 1367, + 1864, + 262, + 1864 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 816, + 74, + 816, + 106, + 299, + 106 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.856 + }, + { + "category_id": 1, + "poly": [ + 299, + 380, + 693, + 380, + 693, + 411, + 299, + 411 + ], + "score": 0.606 + }, + { + "category_id": 0, + "poly": [ + 265, + 1399, + 346, + 1399, + 346, + 1424, + 265, + 1424 + ], + "score": 0.6 + }, + { + "category_id": 0, + "poly": [ + 300, + 230, + 808, + 230, + 808, + 262, + 300, + 262 + ], + "score": 0.584 + }, + { + "category_id": 13, + "poly": [ + 383, + 282, + 469, + 282, + 469, + 316, + 383, + 316 + ], + "score": 0.94, + "latex": "\\mathbf { 1 } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 296, + 1458, + 722, + 1458, + 722, + 1516, + 296, + 1516 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\Sigma _ { 1 } ^ { * } \\frac { \\lambda _ { 1 } ^ { ( 1 ) } - \\lambda _ { 2 } ^ { ( 1 ) } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 1 ) } I _ { d } ; } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 678, + 1692, + 870, + 1692, + 870, + 1729, + 678, + 1729 + ], + "score": 0.92, + "latex": "\\boldsymbol \\eta ^ { ( 1 ) } \\sim \\mathcal { N } ( \\mathbf 0 , \\Sigma _ { 1 } ^ { * } )" + }, + { + "category_id": 13, + "poly": [ + 296, + 1516, + 723, + 1516, + 723, + 1573, + 296, + 1573 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\Sigma _ { 0 } ^ { * } \\frac { \\lambda _ { 1 } ^ { ( 0 ) } - \\lambda _ { 2 } ^ { ( 0 ) } } { \\| \\Delta g \\| _ { 2 } ^ { 2 } } ( \\Delta g ) ( \\Delta g ) ^ { \\top } + \\lambda _ { 2 } ^ { ( 0 ) } I _ { d } ; } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 678, + 1756, + 870, + 1756, + 870, + 1794, + 678, + 1794 + ], + "score": 0.9, + "latex": "\\boldsymbol \\eta ^ { ( 0 ) } \\sim \\mathcal { N } ( \\mathbf 0 , \\Sigma _ { 0 } ^ { * } )" + }, + { + "category_id": 13, + "poly": [ + 297, + 610, + 405, + 610, + 405, + 655, + 297, + 655 + ], + "score": 0.9, + "latex": "p \\gets \\frac { \\mathbf { 1 } ^ { T } y } { d }" + }, + { + "category_id": 13, + "poly": [ + 392, + 512, + 511, + 512, + 511, + 550, + 392, + 550 + ], + "score": 0.9, + "latex": "\\tilde { g } \\in \\mathbb R ^ { B \\times d }" + }, + { + "category_id": 13, + "poly": [ + 485, + 319, + 531, + 319, + 531, + 352, + 485, + 352 + ], + "score": 0.9, + "latex": "y [ j ]" + }, + { + "category_id": 13, + "poly": [ + 339, + 1634, + 412, + 1634, + 412, + 1663, + 339, + 1663 + ], + "score": 0.89, + "latex": "j 1" + }, + { + "category_id": 13, + "poly": [ + 364, + 1663, + 460, + 1663, + 460, + 1694, + 364, + 1694 + ], + "score": 0.89, + "latex": "y [ j ] = 1" + }, + { + "category_id": 13, + "poly": [ + 1077, + 287, + 1122, + 287, + 1122, + 321, + 1077, + 321 + ], + "score": 0.89, + "latex": "g [ j ]" + }, + { + "category_id": 13, + "poly": [ + 297, + 695, + 590, + 695, + 590, + 736, + 297, + 736 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\bar { g } ^ { ( 0 ) } \\frac { 1 } { B - \\mathbf { 1 } ^ { T } y } g ^ { T } ( \\mathbf { 1 } - y ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 296, + 1602, + 421, + 1602, + 421, + 1633, + 296, + 1633 + ], + "score": 0.88, + "latex": "\\tilde { g } \\mathbf { 0 } _ { B \\times d }" + }, + { + "category_id": 13, + "poly": [ + 391, + 448, + 529, + 448, + 529, + 486, + 391, + 486 + ], + "score": 0.88, + "latex": "y \\in \\left\\{ 0 , 1 \\right\\} ^ { B }" + }, + { + "category_id": 13, + "poly": [ + 298, + 654, + 484, + 654, + 484, + 695, + 298, + 695 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { \\bar { g } ^ { ( 1 ) } \\frac { 1 } { \\mathbf { 1 } ^ { T } y } g ^ { T } y } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 321, + 931, + 392, + 931, + 392, + 957, + 321, + 957 + ], + "score": 0.86, + "latex": "u < v" + }, + { + "category_id": 13, + "poly": [ + 391, + 411, + 511, + 411, + 511, + 449, + 391, + 449 + ], + "score": 0.86, + "latex": "g \\in \\mathbb { R } ^ { B \\times d }" + }, + { + "category_id": 13, + "poly": [ + 297, + 735, + 494, + 735, + 494, + 771, + 297, + 771 + ], + "score": 0.85, + "latex": "\\Delta g = \\bar { g } ^ { ( 1 ) } - \\bar { g } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 978, + 999, + 1027, + 999, + 1027, + 1041, + 978, + 1041 + ], + "score": 0.85, + "latex": "\\lambda _ { 2 } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 1226, + 1243, + 1260, + 1243, + 1260, + 1271, + 1226, + 1271 + ], + "score": 0.84, + "latex": "\\mathbb { R } ^ { 4 }" + }, + { + "category_id": 13, + "poly": [ + 1276, + 290, + 1291, + 290, + 1291, + 320, + 1276, + 320 + ], + "score": 0.82, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 685, + 321, + 701, + 321, + 701, + 351, + 685, + 351 + ], + "score": 0.82, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 859, + 999, + 907, + 999, + 907, + 1041, + 859, + 1041 + ], + "score": 0.81, + "latex": "\\lambda _ { 1 } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 417, + 324, + 436, + 324, + 436, + 351, + 417, + 351 + ], + "score": 0.81, + "latex": "g" + }, + { + "category_id": 14, + "poly": [ + 337, + 1068, + 462, + 1068, + 462, + 1114, + 337, + 1114 + ], + "score": 0.81, + "latex": "\\lambda _ { 2 } ^ { ( 0 ) } 0 \\ : ;" + }, + { + "category_id": 13, + "poly": [ + 349, + 1758, + 590, + 1758, + 590, + 1793, + 349, + 1793 + ], + "score": 0.77, + "latex": "\\begin{array} { r l } { \\vert } & { { } \\tilde { g } [ j ] g [ j ] + \\eta ^ { ( 0 ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1006, + 326, + 1023, + 326, + 1023, + 351, + 1006, + 351 + ], + "score": 0.77, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 399, + 1728, + 539, + 1728, + 539, + 1758, + 399, + 1758 + ], + "score": 0.76, + "latex": "\\begin{array} { r l } { \\langle \\mathscr { I } / } & { { } y [ j ] = 0 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 377, + 1693, + 592, + 1693, + 592, + 1728, + 377, + 1728 + ], + "score": 0.76, + "latex": "\\tilde { g } [ j ] g [ j ] + \\eta ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 621, + 417, + 647, + 417, + 647, + 444, + 621, + 444 + ], + "score": 0.76, + "latex": ". B" + }, + { + "category_id": 13, + "poly": [ + 445, + 1633, + 471, + 1633, + 471, + 1659, + 445, + 1659 + ], + "score": 0.73, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 294, + 823, + 852, + 823, + 852, + 867, + 294, + 867 + ], + "score": 0.73, + "latex": "\\begin{array} { r } { u \\frac { 1 } { d \\cdot ( B - \\mathbf { 1 } ^ { T } y ) } \\sum _ { j = 1 } ^ { B } ( 1 - y [ j ] ) \\cdot g [ j ] - \\bar { g } ^ { ( 0 ) } _ { 2 } ^ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 295, + 772, + 752, + 772, + 752, + 819, + 295, + 819 + ], + "score": 0.73, + "latex": "\\begin{array} { r } { v \\frac { 1 } { d \\cdot ( \\mathbf { 1 } ^ { T } y ) } \\sum _ { j = 1 } ^ { B } y [ j ] \\cdot g [ j ] - \\bar { g } ^ { ( 1 ) } _ { 2 } ^ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 859, + 1112, + 907, + 1112, + 907, + 1153, + 859, + 1153 + ], + "score": 0.72, + "latex": "\\lambda _ { 1 } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 978, + 1112, + 1027, + 1112, + 1027, + 1153, + 978, + 1153 + ], + "score": 0.72, + "latex": "\\lambda _ { 1 } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 297, + 867, + 479, + 867, + 479, + 901, + 297, + 901 + ], + "score": 0.7, + "latex": "P \\gets s \\cdot \\| \\Delta g \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1090, + 484, + 1115, + 484, + 1115, + 510, + 1090, + 510 + ], + "score": 0.7, + "latex": "P" + }, + { + "category_id": 13, + "poly": [ + 918, + 998, + 967, + 998, + 967, + 1041, + 918, + 1041 + ], + "score": 0.65, + "latex": "\\lambda _ { 1 } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 339, + 959, + 450, + 959, + 450, + 1001, + 339, + 1001 + ], + "score": 0.64, + "latex": "\\lambda _ { 2 } ^ { ( 1 ) } 0" + }, + { + "category_id": 13, + "poly": [ + 1296, + 968, + 1361, + 968, + 1361, + 997, + 1296, + 997 + ], + "score": 0.6, + "latex": "0 ~ \\star /" + }, + { + "category_id": 13, + "poly": [ + 919, + 1112, + 967, + 1112, + 967, + 1153, + 919, + 1153 + ], + "score": 0.53, + "latex": "\\lambda _ { 2 } ^ { ( 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 543, + 1602, + 581, + 1602, + 581, + 1630, + 543, + 1630 + ], + "score": 0.46, + "latex": "/ \\star" + }, + { + "category_id": 13, + "poly": [ + 862, + 1111, + 1028, + 1111, + 1028, + 1154, + 862, + 1154 + ], + "score": 0.36, + "latex": "\\lambda _ { 1 } ^ { ( 1 ) } , \\lambda _ { 2 } ^ { ( 1 ) } , \\lambda _ { 1 } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 1326, + 1081, + 1361, + 1081, + 1361, + 1108, + 1326, + 1108 + ], + "score": 0.35, + "latex": "\\star /" + }, + { + "category_id": 13, + "poly": [ + 392, + 491, + 407, + 491, + 407, + 511, + 392, + 511 + ], + "score": 0.35, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 1065, + 658, + 1104, + 658, + 1104, + 687, + 1065, + 687 + ], + "score": 0.34, + "latex": "/ \\star" + }, + { + "category_id": 13, + "poly": [ + 814, + 1080, + 853, + 1080, + 853, + 1108, + 814, + 1108 + ], + "score": 0.32, + "latex": "/ \\star" + }, + { + "category_id": 13, + "poly": [ + 1326, + 870, + 1360, + 870, + 1360, + 897, + 1326, + 897 + ], + "score": 0.32, + "latex": "\\star /" + }, + { + "category_id": 13, + "poly": [ + 1326, + 830, + 1361, + 830, + 1361, + 857, + 1326, + 857 + ], + "score": 0.26, + "latex": "\\star /" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 817.0, + 72.0, + 817.0, + 109.0, + 295.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 261.0, + 1395.0, + 347.0, + 1395.0, + 347.0, + 1426.0, + 261.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 810.0, + 229.0, + 810.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 281.0, + 382.0, + 281.0, + 382.0, + 324.0, + 295.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 281.0, + 1076.0, + 281.0, + 1076.0, + 324.0, + 470.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 281.0, + 1275.0, + 281.0, + 1275.0, + 324.0, + 1123.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 281.0, + 1407.0, + 281.0, + 1407.0, + 324.0, + 1292.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 313.0, + 416.0, + 313.0, + 416.0, + 358.0, + 293.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 313.0, + 484.0, + 313.0, + 484.0, + 358.0, + 437.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 313.0, + 684.0, + 313.0, + 684.0, + 358.0, + 532.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 313.0, + 1005.0, + 313.0, + 1005.0, + 358.0, + 702.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 313.0, + 1034.0, + 313.0, + 1034.0, + 358.0, + 1024.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 381.0, + 694.0, + 381.0, + 694.0, + 412.0, + 297.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 414.0, + 620.0, + 414.0, + 620.0, + 450.0, + 543.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 414.0, + 995.0, + 414.0, + 995.0, + 450.0, + 648.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 437.0, + 390.0, + 437.0, + 390.0, + 490.0, + 377.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 437.0, + 995.0, + 437.0, + 995.0, + 490.0, + 530.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 486.0, + 391.0, + 486.0, + 391.0, + 517.0, + 388.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 486.0, + 422.0, + 486.0, + 422.0, + 517.0, + 408.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 482.0, + 1089.0, + 482.0, + 1089.0, + 514.0, + 545.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 482.0, + 1120.0, + 482.0, + 1120.0, + 514.0, + 1116.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 537.0, + 519.0, + 537.0, + 519.0, + 548.0, + 512.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 513.0, + 870.0, + 513.0, + 870.0, + 552.0, + 542.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 545.0, + 1234.0, + 545.0, + 1234.0, + 584.0, + 298.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 580.0, + 1114.0, + 580.0, + 1114.0, + 612.0, + 342.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 621.0, + 296.0, + 621.0, + 296.0, + 651.0, + 265.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 638.0, + 419.0, + 638.0, + 419.0, + 648.0, + 409.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 617.0, + 1362.0, + 617.0, + 1362.0, + 653.0, + 1005.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 659.0, + 500.0, + 659.0, + 500.0, + 690.0, + 485.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 654.0, + 1364.0, + 654.0, + 1364.0, + 692.0, + 1105.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 263.0, + 687.0, + 296.0, + 687.0, + 296.0, + 736.0, + 263.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 690.0, + 609.0, + 690.0, + 609.0, + 735.0, + 591.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 695.0, + 1362.0, + 695.0, + 1362.0, + 732.0, + 1065.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 262.0, + 724.0, + 296.0, + 724.0, + 296.0, + 778.0, + 262.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 724.0, + 513.0, + 724.0, + 513.0, + 778.0, + 495.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 268.0, + 764.0, + 294.0, + 764.0, + 294.0, + 828.0, + 268.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 764.0, + 769.0, + 764.0, + 769.0, + 828.0, + 753.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 778.0, + 1362.0, + 778.0, + 1362.0, + 814.0, + 961.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 811.0, + 293.0, + 811.0, + 293.0, + 877.0, + 253.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 811.0, + 876.0, + 811.0, + 876.0, + 877.0, + 853.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 825.0, + 1325.0, + 825.0, + 1325.0, + 862.0, + 960.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 263.0, + 859.0, + 296.0, + 859.0, + 296.0, + 907.0, + 263.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 859.0, + 499.0, + 859.0, + 499.0, + 907.0, + 480.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 863.0, + 1325.0, + 863.0, + 1325.0, + 904.0, + 794.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 863.0, + 1364.0, + 863.0, + 1364.0, + 904.0, + 1361.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 894.0, + 461.0, + 894.0, + 461.0, + 933.0, + 291.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 901.0, + 999.0, + 901.0, + 999.0, + 932.0, + 478.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 268.0, + 928.0, + 320.0, + 928.0, + 320.0, + 960.0, + 268.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 928.0, + 457.0, + 928.0, + 457.0, + 960.0, + 393.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 271.0, + 970.0, + 294.0, + 970.0, + 294.0, + 998.0, + 271.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 949.0, + 338.0, + 949.0, + 338.0, + 1012.0, + 329.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 949.0, + 476.0, + 949.0, + 476.0, + 1012.0, + 451.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 965.0, + 1295.0, + 965.0, + 1295.0, + 998.0, + 828.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 262.0, + 1010.0, + 294.0, + 1010.0, + 294.0, + 1041.0, + 262.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 987.0, + 858.0, + 987.0, + 858.0, + 1054.0, + 329.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 987.0, + 917.0, + 987.0, + 917.0, + 1054.0, + 908.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 987.0, + 977.0, + 987.0, + 977.0, + 1054.0, + 968.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 987.0, + 1360.0, + 987.0, + 1360.0, + 1054.0, + 1028.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 257.0, + 1040.0, + 351.0, + 1040.0, + 351.0, + 1074.0, + 257.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 1078.0, + 296.0, + 1078.0, + 296.0, + 1153.0, + 259.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1069.0, + 336.0, + 1069.0, + 336.0, + 1177.0, + 321.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1069.0, + 813.0, + 1069.0, + 813.0, + 1177.0, + 463.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1069.0, + 858.0, + 1069.0, + 858.0, + 1177.0, + 854.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 1069.0, + 1325.0, + 1069.0, + 1325.0, + 1177.0, + 1029.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 1154.0, + 350.0, + 1154.0, + 350.0, + 1184.0, + 258.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 257.0, + 1183.0, + 568.0, + 1183.0, + 568.0, + 1215.0, + 257.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 1215.0, + 294.0, + 1215.0, + 294.0, + 1246.0, + 259.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1210.0, + 933.0, + 1210.0, + 933.0, + 1247.0, + 335.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1236.0, + 1225.0, + 1236.0, + 1225.0, + 1281.0, + 334.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1236.0, + 1341.0, + 1236.0, + 1341.0, + 1281.0, + 1261.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 1277.0, + 1221.0, + 1277.0, + 1221.0, + 1308.0, + 387.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 1305.0, + 576.0, + 1305.0, + 576.0, + 1337.0, + 388.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 262.0, + 1336.0, + 294.0, + 1336.0, + 294.0, + 1367.0, + 262.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1334.0, + 1225.0, + 1334.0, + 1225.0, + 1367.0, + 338.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1364.0, + 1176.0, + 1364.0, + 1176.0, + 1400.0, + 348.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 407.5, + 521.0, + 407.5, + 521.0, + 450.5, + 290.0, + 450.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 509.0, + 526.0, + 509.0, + 526.0, + 552.5, + 290.0, + 552.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 600.5, + 426.0, + 600.5, + 426.0, + 645.0, + 351.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 260.0, + 645.0, + 364.0, + 645.0, + 364.0, + 692.0, + 260.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 688.0, + 519.0, + 688.0, + 519.0, + 746.0, + 369.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1061.0, + 391.0, + 1061.0, + 391.0, + 1122.0, + 344.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 1412.0, + 285.0, + 1412.0, + 285.0, + 1422.0, + 275.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1423.0, + 463.0, + 1423.0, + 463.0, + 1463.0, + 295.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1426.0, + 1072.0, + 1426.0, + 1072.0, + 1460.0, + 479.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 1463.0, + 295.0, + 1463.0, + 295.0, + 1516.0, + 255.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1449.0, + 738.0, + 1449.0, + 738.0, + 1525.0, + 723.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 1518.0, + 295.0, + 1518.0, + 295.0, + 1571.0, + 255.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1505.0, + 736.0, + 1505.0, + 736.0, + 1581.0, + 724.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1566.0, + 466.0, + 1566.0, + 466.0, + 1607.0, + 293.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1572.0, + 803.0, + 1572.0, + 803.0, + 1603.0, + 476.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 1596.0, + 295.0, + 1596.0, + 295.0, + 1638.0, + 255.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 1596.0, + 444.0, + 1596.0, + 444.0, + 1638.0, + 422.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 1597.0, + 1362.0, + 1597.0, + 1362.0, + 1636.0, + 582.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 257.0, + 1630.0, + 338.0, + 1630.0, + 338.0, + 1665.0, + 257.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1630.0, + 444.0, + 1630.0, + 444.0, + 1665.0, + 413.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 1630.0, + 513.0, + 1630.0, + 513.0, + 1665.0, + 472.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 1661.0, + 296.0, + 1661.0, + 296.0, + 1695.0, + 258.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 335.0, + 1661.0, + 363.0, + 1661.0, + 363.0, + 1693.0, + 335.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 1661.0, + 523.0, + 1661.0, + 523.0, + 1693.0, + 461.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 1695.0, + 297.0, + 1695.0, + 297.0, + 1728.0, + 258.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1686.0, + 376.0, + 1686.0, + 376.0, + 1735.0, + 370.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 1686.0, + 677.0, + 1686.0, + 677.0, + 1735.0, + 593.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1686.0, + 882.0, + 1686.0, + 882.0, + 1735.0, + 871.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 1726.0, + 296.0, + 1726.0, + 296.0, + 1760.0, + 258.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1724.0, + 398.0, + 1724.0, + 398.0, + 1759.0, + 336.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1724.0, + 545.0, + 1724.0, + 545.0, + 1759.0, + 540.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 1760.0, + 297.0, + 1760.0, + 297.0, + 1793.0, + 258.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 1750.0, + 677.0, + 1750.0, + 677.0, + 1799.0, + 591.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1750.0, + 882.0, + 1750.0, + 882.0, + 1799.0, + 871.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 1789.0, + 296.0, + 1789.0, + 296.0, + 1824.0, + 258.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1789.0, + 394.0, + 1789.0, + 394.0, + 1821.0, + 332.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1819.0, + 353.0, + 1819.0, + 353.0, + 1858.0, + 253.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1510.0, + 434.0, + 1510.0, + 434.0, + 1545.0, + 363.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 377.0, + 695.0, + 377.0, + 695.0, + 416.0, + 296.0, + 416.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 952, + 1405, + 952, + 1405, + 1169, + 297, + 1169 + ], + "score": 0.979 + }, + { + "category_id": 5, + "poly": [ + 378, + 1194, + 1323, + 1194, + 1323, + 1494, + 378, + 1494 + ], + "score": 0.974, + "html": "
batchsizeBcut layer featuredimensiondaverageMarvell run time(seconds /run)averageupdate time(seconds /iteration)
ISIC1281600=5×5×641.79 × 10-2± 7.03 × 10-53.94×10-1± 4.05 × 10-3
Criteo10241281.68×10-2± 8.67 × 10-54.84± 3.28 × 10-1
Avazu327681281.75 ×10-2± 1.47 × 10-46.22± 3.71 × 10-1
" + }, + { + "category_id": 1, + "poly": [ + 300, + 743, + 1405, + 743, + 1405, + 866, + 300, + 866 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 298, + 282, + 980, + 282, + 980, + 315, + 298, + 315 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.842 + }, + { + "category_id": 0, + "poly": [ + 300, + 900, + 786, + 900, + 786, + 931, + 300, + 931 + ], + "score": 0.751 + }, + { + "category_id": 6, + "poly": [ + 294, + 1514, + 1405, + 1514, + 1405, + 1572, + 294, + 1572 + ], + "score": 0.676 + }, + { + "category_id": 1, + "poly": [ + 301, + 330, + 1406, + 330, + 1406, + 728, + 301, + 728 + ], + "score": 0.628 + }, + { + "category_id": 1, + "poly": [ + 300, + 231, + 743, + 231, + 743, + 262, + 300, + 262 + ], + "score": 0.472 + }, + { + "category_id": 0, + "poly": [ + 300, + 231, + 743, + 231, + 743, + 262, + 300, + 262 + ], + "score": 0.391 + }, + { + "category_id": 4, + "poly": [ + 294, + 1514, + 1405, + 1514, + 1405, + 1572, + 294, + 1572 + ], + "score": 0.164 + }, + { + "category_id": 1, + "poly": [ + 300, + 900, + 786, + 900, + 786, + 931, + 300, + 931 + ], + "score": 0.141 + }, + { + "category_id": 13, + "poly": [ + 512, + 452, + 595, + 452, + 595, + 486, + 512, + 486 + ], + "score": 0.93, + "latex": "O ( \\bar { B } d )" + }, + { + "category_id": 13, + "poly": [ + 1059, + 696, + 1141, + 696, + 1141, + 730, + 1059, + 730 + ], + "score": 0.93, + "latex": "O ( B d )" + }, + { + "category_id": 13, + "poly": [ + 830, + 392, + 912, + 392, + 912, + 425, + 830, + 425 + ], + "score": 0.93, + "latex": "O ( B d )" + }, + { + "category_id": 13, + "poly": [ + 1033, + 743, + 1115, + 743, + 1115, + 776, + 1033, + 776 + ], + "score": 0.92, + "latex": "O ( B d )" + }, + { + "category_id": 13, + "poly": [ + 752, + 804, + 835, + 804, + 835, + 838, + 752, + 838 + ], + "score": 0.92, + "latex": "O ( B d )" + }, + { + "category_id": 13, + "poly": [ + 633, + 696, + 694, + 696, + 694, + 730, + 633, + 730 + ], + "score": 0.91, + "latex": "O ( d )" + }, + { + "category_id": 13, + "poly": [ + 637, + 985, + 712, + 985, + 712, + 1012, + 637, + 1012 + ], + "score": 0.89, + "latex": "s = 4" + }, + { + "category_id": 13, + "poly": [ + 917, + 544, + 976, + 544, + 976, + 576, + 917, + 576 + ], + "score": 0.89, + "latex": "O ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 569, + 1076, + 625, + 1076, + 625, + 1105, + 569, + 1105 + ], + "score": 0.87, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 711, + 1516, + 776, + 1516, + 776, + 1542, + 711, + 1542 + ], + "score": 0.84, + "latex": "s = 4" + }, + { + "category_id": 13, + "poly": [ + 805, + 363, + 830, + 363, + 830, + 389, + 805, + 389 + ], + "score": 0.78, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 1022, + 429, + 1041, + 429, + 1041, + 450, + 1022, + 450 + ], + "score": 0.76, + "latex": "u" + }, + { + "category_id": 13, + "poly": [ + 1091, + 428, + 1109, + 428, + 1109, + 450, + 1091, + 450 + ], + "score": 0.71, + "latex": "v" + }, + { + "category_id": 13, + "poly": [ + 869, + 1393, + 1042, + 1393, + 1042, + 1421, + 869, + 1421 + ], + "score": 0.59, + "latex": "\\pm 8 . 6 7 \\times 1 0 ^ { - 5 }" + }, + { + "category_id": 13, + "poly": [ + 684, + 1258, + 704, + 1258, + 704, + 1286, + 684, + 1286 + ], + "score": 0.54, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 612, + 1327, + 773, + 1327, + 773, + 1358, + 612, + 1358 + ], + "score": 0.53, + "latex": "= 5 \\times 5 \\times 6 4" + }, + { + "category_id": 13, + "poly": [ + 914, + 577, + 1001, + 577, + 1001, + 605, + 914, + 605 + ], + "score": 0.5, + "latex": "1 9 - 2 8 " + }, + { + "category_id": 13, + "poly": [ + 868, + 1328, + 1042, + 1328, + 1042, + 1357, + 868, + 1357 + ], + "score": 0.49, + "latex": "\\pm 7 . 0 3 \\times 1 0 ^ { - 5 }" + }, + { + "category_id": 13, + "poly": [ + 828, + 1362, + 969, + 1362, + 969, + 1390, + 828, + 1390 + ], + "score": 0.48, + "latex": "1 . 6 8 \\times 1 0 ^ { - 2 }" + }, + { + "category_id": 14, + "poly": [ + 365, + 1186, + 1328, + 1186, + 1328, + 1497, + 365, + 1497 + ], + "score": 0.47, + "latex": "\\frac | \\begin{array} { l c } { { \\mathrm { b a t c h } } } \\\\ { { \\mathrm { s i z e } } } \\\\ { { \\frac { B } { 1 . 8 1 \\times } } } \\end{array} | \\begin{array} { l c } { { \\mathrm { c u t ~ l a y e r ~ f e a t u r e } } } \\\\ { { \\mathrm { d i m e n s i o n } } } \\\\ { { \\mathrm { ~ d i o n } } } \\end{array} | \\begin{array} { l l } { { \\mathrm { ~ a n v e r ~ f e n ~ t i m e } } } \\\\ { { \\mathrm { M a r v e l ~ r u n ~ t i m e } } } \\\\ { { \\mathrm { ~ 1 . 7 9 \\times ~ l 0 ^ { - 2 } ~ } } } \\\\ { { \\mathrm { ~ 2 . 7 . 0 3 \\times ~ 1 0 ^ { - 5 } ~ } } } \\end{array} | \\begin{array} { l l } { { \\mathrm { ~ a v e r a g e } } } \\\\ { { \\mathrm { ~ ( s e c o n d s / ~ t i m e ) } } } \\\\ { { \\mathrm { ~ 3 . 9 4 \\times ~ 1 0 ^ { - 1 } ~ } } } \\\\ { { \\mathrm { ~ 4 . 0 5 \\times ~ 1 0 ^ { - 3 } ~ } } } \\\\ { { \\mathrm { ~ 1 . 6 8 \\times ~ 1 0 ^ { - 2 } ~ } } } \\end{array} | \\begin{array} { l l } { { \\mathrm { a v e r a g e } } } \\\\ { { \\mathrm { ~ 4 . 2 8 \\times ~ 1 0 ^ { - 1 } ~ } } } \\\\ { { \\mathrm { ~ 4 . 8 4 ~ } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 523, + 1258, + 549, + 1258, + 549, + 1285, + 523, + 1285 + ], + "score": 0.4, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 828, + 1296, + 969, + 1296, + 969, + 1325, + 828, + 1325 + ], + "score": 0.4, + "latex": "\\overline { { 1 . 7 9 \\times 1 0 ^ { - 2 } } }" + }, + { + "category_id": 13, + "poly": [ + 821, + 1295, + 1045, + 1295, + 1045, + 1357, + 821, + 1357 + ], + "score": 0.3, + "latex": "\\begin{array} { c } { { \\overline { { 1 . 7 9 \\times 1 0 ^ { - 2 } } } } } \\\\ { { \\pm 7 . 0 3 \\times 1 0 ^ { - 5 } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 821, + 1425, + 1045, + 1425, + 1045, + 1493, + 821, + 1493 + ], + "score": 0.29, + "latex": "\\begin{array} { c } { { 1 . 7 5 \\times 1 0 ^ { - 2 } } } \\\\ { { \\pm 1 . 4 7 \\times 1 0 ^ { - 4 } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1129, + 1326, + 1301, + 1326, + 1301, + 1358, + 1129, + 1358 + ], + "score": 0.28, + "latex": "\\pm 4 . 0 5 \\times 1 0 ^ { - 3 }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 899.0, + 789.0, + 899.0, + 789.0, + 934.0, + 296.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1511.0, + 710.0, + 1511.0, + 710.0, + 1547.0, + 293.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 1511.0, + 1405.0, + 1511.0, + 1405.0, + 1547.0, + 777.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1539.0, + 581.0, + 1539.0, + 581.0, + 1574.0, + 293.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 745.0, + 229.0, + 745.0, + 264.0, + 296.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1511.0, + 710.0, + 1511.0, + 710.0, + 1547.0, + 293.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 1511.0, + 1405.0, + 1511.0, + 1405.0, + 1547.0, + 777.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1539.0, + 581.0, + 1539.0, + 581.0, + 1574.0, + 293.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 952.0, + 1406.0, + 952.0, + 1406.0, + 990.0, + 295.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 984.0, + 636.0, + 984.0, + 636.0, + 1017.0, + 294.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 984.0, + 1405.0, + 984.0, + 1405.0, + 1017.0, + 713.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1015.0, + 1405.0, + 1015.0, + 1405.0, + 1049.0, + 295.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1047.0, + 1403.0, + 1047.0, + 1403.0, + 1078.0, + 296.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1074.0, + 568.0, + 1074.0, + 568.0, + 1110.0, + 294.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 1074.0, + 1406.0, + 1074.0, + 1406.0, + 1110.0, + 626.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1106.0, + 1405.0, + 1106.0, + 1405.0, + 1141.0, + 294.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1135.0, + 1202.0, + 1135.0, + 1202.0, + 1173.0, + 294.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 742.0, + 1032.0, + 742.0, + 1032.0, + 778.0, + 295.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 742.0, + 1406.0, + 742.0, + 1406.0, + 778.0, + 1116.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 773.0, + 1405.0, + 773.0, + 1405.0, + 809.0, + 294.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 805.0, + 751.0, + 805.0, + 751.0, + 841.0, + 295.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 805.0, + 1405.0, + 805.0, + 1405.0, + 841.0, + 836.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 836.0, + 1027.0, + 836.0, + 1027.0, + 868.0, + 296.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 279.0, + 984.0, + 279.0, + 984.0, + 322.0, + 294.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 331.0, + 1406.0, + 331.0, + 1406.0, + 367.0, + 305.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 362.0, + 804.0, + 362.0, + 804.0, + 398.0, + 394.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 362.0, + 1405.0, + 362.0, + 1405.0, + 398.0, + 831.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 390.0, + 829.0, + 390.0, + 829.0, + 427.0, + 395.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 390.0, + 921.0, + 390.0, + 921.0, + 427.0, + 913.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 420.0, + 1021.0, + 420.0, + 1021.0, + 458.0, + 305.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 420.0, + 1090.0, + 420.0, + 1090.0, + 458.0, + 1042.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 420.0, + 1403.0, + 420.0, + 1403.0, + 458.0, + 1110.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 451.0, + 511.0, + 451.0, + 511.0, + 489.0, + 393.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 451.0, + 1190.0, + 451.0, + 1190.0, + 489.0, + 596.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 482.0, + 1403.0, + 482.0, + 1403.0, + 519.0, + 304.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 511.0, + 1405.0, + 511.0, + 1405.0, + 551.0, + 393.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 545.0, + 916.0, + 545.0, + 916.0, + 577.0, + 395.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 545.0, + 985.0, + 545.0, + 985.0, + 577.0, + 977.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 574.0, + 913.0, + 574.0, + 913.0, + 609.0, + 305.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 574.0, + 1405.0, + 574.0, + 1405.0, + 609.0, + 1002.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 603.0, + 1403.0, + 603.0, + 1403.0, + 639.0, + 394.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 635.0, + 1405.0, + 635.0, + 1405.0, + 672.0, + 394.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 665.0, + 1405.0, + 665.0, + 1405.0, + 700.0, + 393.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 698.0, + 632.0, + 698.0, + 632.0, + 730.0, + 395.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 698.0, + 1058.0, + 698.0, + 1058.0, + 730.0, + 695.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 698.0, + 1236.0, + 698.0, + 1236.0, + 730.0, + 1142.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 745.0, + 229.0, + 745.0, + 264.0, + 296.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 899.0, + 789.0, + 899.0, + 789.0, + 934.0, + 296.0, + 934.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1299, + 1404, + 1299, + 1404, + 1635, + 298, + 1635 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 497, + 1404, + 497, + 1404, + 743, + 298, + 743 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1668, + 1404, + 1668, + 1404, + 1883, + 298, + 1883 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 965, + 1403, + 965, + 1403, + 1211, + 298, + 1211 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 778, + 1402, + 778, + 1402, + 931, + 298, + 931 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 287, + 1403, + 287, + 1403, + 409, + 299, + 409 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 299, + 1972, + 1399, + 1972, + 1399, + 2034, + 299, + 2034 + ], + "score": 0.945 + }, + { + "category_id": 0, + "poly": [ + 301, + 1246, + 785, + 1246, + 785, + 1276, + 301, + 1276 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.884 + }, + { + "category_id": 0, + "poly": [ + 301, + 445, + 704, + 445, + 704, + 476, + 301, + 476 + ], + "score": 0.883 + }, + { + "category_id": 0, + "poly": [ + 301, + 1918, + 718, + 1918, + 718, + 1949, + 301, + 1949 + ], + "score": 0.877 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.848 + }, + { + "category_id": 0, + "poly": [ + 304, + 230, + 885, + 230, + 885, + 261, + 304, + 261 + ], + "score": 0.839 + }, + { + "category_id": 13, + "poly": [ + 648, + 1026, + 722, + 1026, + 722, + 1056, + 648, + 1056 + ], + "score": 0.91, + "latex": "\\sim 1 0 ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 453, + 1853, + 579, + 1853, + 579, + 1882, + 453, + 1882 + ], + "score": 0.91, + "latex": "5 \\times 5 \\times 6 4" + }, + { + "category_id": 13, + "poly": [ + 859, + 1670, + 983, + 1670, + 983, + 1699, + 859, + 1699 + ], + "score": 0.91, + "latex": "8 4 \\times 8 4 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 668, + 1181, + 761, + 1181, + 761, + 1209, + 668, + 1209 + ], + "score": 0.9, + "latex": "8 4 \\times 8 4" + }, + { + "category_id": 13, + "poly": [ + 998, + 1701, + 1057, + 1701, + 1057, + 1729, + 998, + 1729 + ], + "score": 0.9, + "latex": "1 \\times 1" + }, + { + "category_id": 13, + "poly": [ + 775, + 1701, + 834, + 1701, + 834, + 1730, + 775, + 1730 + ], + "score": 0.9, + "latex": "3 \\times 3" + }, + { + "category_id": 13, + "poly": [ + 1155, + 1089, + 1249, + 1089, + 1249, + 1118, + 1155, + 1118 + ], + "score": 0.9, + "latex": "5 8 4 \\times 9" + }, + { + "category_id": 13, + "poly": [ + 460, + 1762, + 521, + 1762, + 521, + 1791, + 460, + 1791 + ], + "score": 0.89, + "latex": "2 \\times 2" + }, + { + "category_id": 13, + "poly": [ + 935, + 651, + 989, + 651, + 989, + 681, + 935, + 681 + ], + "score": 0.89, + "latex": "1 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 1248, + 1731, + 1307, + 1731, + 1307, + 1760, + 1248, + 1760 + ], + "score": 0.89, + "latex": "2 \\times 2" + }, + { + "category_id": 13, + "poly": [ + 1067, + 1119, + 1121, + 1119, + 1121, + 1148, + 1067, + 1148 + ], + "score": 0.87, + "latex": "10 \\%" + }, + { + "category_id": 13, + "poly": [ + 662, + 901, + 715, + 901, + 715, + 930, + 662, + 930 + ], + "score": 0.87, + "latex": "10 \\%" + }, + { + "category_id": 13, + "poly": [ + 424, + 901, + 478, + 901, + 478, + 930, + 424, + 930 + ], + "score": 0.86, + "latex": "90 \\%" + }, + { + "category_id": 13, + "poly": [ + 547, + 996, + 589, + 996, + 589, + 1026, + 547, + 1026 + ], + "score": 0.86, + "latex": "2 \\%" + }, + { + "category_id": 13, + "poly": [ + 782, + 2004, + 801, + 2004, + 801, + 2035, + 782, + 2035 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1033, + 1150, + 1147, + 1150, + 1147, + 1179, + 1033, + 1179 + ], + "score": 0.75, + "latex": "80 \\% - 2 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 824, + 713, + 878, + 713, + 878, + 742, + 824, + 742 + ], + "score": 0.65, + "latex": "90 \\%" + }, + { + "category_id": 13, + "poly": [ + 885, + 713, + 938, + 713, + 938, + 742, + 885, + 742 + ], + "score": 0.37, + "latex": "10 \\%" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1245.0, + 789.0, + 1245.0, + 789.0, + 1278.0, + 297.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 445.0, + 708.0, + 445.0, + 708.0, + 478.0, + 297.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1915.0, + 724.0, + 1915.0, + 724.0, + 1953.0, + 296.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2123.0, + 829.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 228.0, + 890.0, + 228.0, + 890.0, + 265.0, + 298.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1297.0, + 1405.0, + 1297.0, + 1405.0, + 1336.0, + 294.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1330.0, + 1405.0, + 1330.0, + 1405.0, + 1364.0, + 294.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1359.0, + 1406.0, + 1359.0, + 1406.0, + 1396.0, + 291.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1391.0, + 1405.0, + 1391.0, + 1405.0, + 1426.0, + 294.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1421.0, + 1404.0, + 1421.0, + 1404.0, + 1456.0, + 294.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1452.0, + 1405.0, + 1452.0, + 1405.0, + 1487.0, + 293.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1482.0, + 1402.0, + 1482.0, + 1402.0, + 1517.0, + 294.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1512.0, + 1404.0, + 1512.0, + 1404.0, + 1548.0, + 291.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1540.0, + 1405.0, + 1540.0, + 1405.0, + 1580.0, + 292.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1571.0, + 1404.0, + 1571.0, + 1404.0, + 1611.0, + 293.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1602.0, + 740.0, + 1602.0, + 740.0, + 1639.0, + 293.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 497.0, + 1403.0, + 497.0, + 1403.0, + 534.0, + 295.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 530.0, + 1405.0, + 530.0, + 1405.0, + 564.0, + 296.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 561.0, + 1405.0, + 561.0, + 1405.0, + 592.0, + 293.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 591.0, + 1406.0, + 591.0, + 1406.0, + 625.0, + 294.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 618.0, + 1406.0, + 618.0, + 1406.0, + 657.0, + 292.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 650.0, + 934.0, + 650.0, + 934.0, + 687.0, + 292.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 650.0, + 1406.0, + 650.0, + 1406.0, + 687.0, + 990.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 684.0, + 1404.0, + 684.0, + 1404.0, + 717.0, + 294.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 713.0, + 823.0, + 713.0, + 823.0, + 746.0, + 294.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 713.0, + 884.0, + 713.0, + 884.0, + 746.0, + 879.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 713.0, + 1107.0, + 713.0, + 1107.0, + 746.0, + 939.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1668.0, + 858.0, + 1668.0, + 858.0, + 1703.0, + 293.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1668.0, + 1406.0, + 1668.0, + 1406.0, + 1703.0, + 984.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1700.0, + 774.0, + 1700.0, + 774.0, + 1731.0, + 296.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 1700.0, + 997.0, + 1700.0, + 997.0, + 1731.0, + 835.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1700.0, + 1402.0, + 1700.0, + 1402.0, + 1731.0, + 1058.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1730.0, + 1247.0, + 1730.0, + 1247.0, + 1764.0, + 294.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1308.0, + 1730.0, + 1404.0, + 1730.0, + 1404.0, + 1764.0, + 1308.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1763.0, + 459.0, + 1763.0, + 459.0, + 1794.0, + 294.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1763.0, + 1402.0, + 1763.0, + 1402.0, + 1794.0, + 522.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1791.0, + 1404.0, + 1791.0, + 1404.0, + 1826.0, + 293.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1823.0, + 1405.0, + 1823.0, + 1405.0, + 1858.0, + 294.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1853.0, + 452.0, + 1853.0, + 452.0, + 1887.0, + 294.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1853.0, + 591.0, + 1853.0, + 591.0, + 1887.0, + 580.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 965.0, + 1405.0, + 965.0, + 1405.0, + 999.0, + 296.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 997.0, + 546.0, + 997.0, + 546.0, + 1030.0, + 294.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 997.0, + 1405.0, + 997.0, + 1405.0, + 1030.0, + 590.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1026.0, + 647.0, + 1026.0, + 647.0, + 1062.0, + 293.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1026.0, + 1405.0, + 1026.0, + 1405.0, + 1062.0, + 723.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1058.0, + 1403.0, + 1058.0, + 1403.0, + 1092.0, + 294.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1089.0, + 1154.0, + 1089.0, + 1154.0, + 1123.0, + 294.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1089.0, + 1405.0, + 1089.0, + 1405.0, + 1123.0, + 1250.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1119.0, + 1066.0, + 1119.0, + 1066.0, + 1153.0, + 294.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1119.0, + 1405.0, + 1119.0, + 1405.0, + 1153.0, + 1122.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1150.0, + 1032.0, + 1150.0, + 1032.0, + 1184.0, + 294.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 1150.0, + 1402.0, + 1150.0, + 1402.0, + 1184.0, + 1148.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1177.0, + 667.0, + 1177.0, + 667.0, + 1217.0, + 295.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1177.0, + 1075.0, + 1177.0, + 1075.0, + 1217.0, + 762.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 777.0, + 1406.0, + 777.0, + 1406.0, + 814.0, + 294.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 807.0, + 1406.0, + 807.0, + 1406.0, + 845.0, + 293.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 839.0, + 1406.0, + 839.0, + 1406.0, + 874.0, + 293.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 869.0, + 1404.0, + 869.0, + 1404.0, + 906.0, + 294.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 902.0, + 423.0, + 902.0, + 423.0, + 935.0, + 296.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 902.0, + 661.0, + 902.0, + 661.0, + 935.0, + 479.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 902.0, + 809.0, + 902.0, + 809.0, + 935.0, + 716.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 286.0, + 1405.0, + 286.0, + 1405.0, + 321.0, + 294.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 316.0, + 1405.0, + 316.0, + 1405.0, + 353.0, + 294.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 348.0, + 1405.0, + 348.0, + 1405.0, + 382.0, + 293.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 377.0, + 402.0, + 377.0, + 402.0, + 412.0, + 292.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1971.0, + 1403.0, + 1971.0, + 1403.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 781.0, + 2002.0, + 781.0, + 2037.0, + 295.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 2002.0, + 1404.0, + 2002.0, + 1404.0, + 2037.0, + 802.0, + 2037.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 595, + 1403, + 595, + 1403, + 719, + 299, + 719 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 294, + 421, + 1401, + 421, + 1401, + 485, + 294, + 485 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 297, + 325, + 1399, + 325, + 1399, + 389, + 297, + 389 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 299, + 518, + 1400, + 518, + 1400, + 581, + 299, + 581 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 295, + 229, + 1401, + 229, + 1401, + 292, + 295, + 292 + ], + "score": 0.942 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 835, + 2089, + 863, + 2089, + 863, + 2112, + 835, + 2112 + ], + "score": 0.83 + }, + { + "category_id": 13, + "poly": [ + 853, + 454, + 899, + 454, + 899, + 483, + 853, + 483 + ], + "score": 0.64, + "latex": "3 5 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 1209, + 424, + 1274, + 424, + 1274, + 452, + 1209, + 452 + ], + "score": 0.58, + "latex": "1 e { - } 5" + }, + { + "category_id": 13, + "poly": [ + 939, + 550, + 993, + 550, + 993, + 579, + 939, + 579 + ], + "score": 0.54, + "latex": "5 . 5 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 1235, + 328, + 1300, + 328, + 1300, + 356, + 1235, + 356 + ], + "score": 0.47, + "latex": "1 e { - 4 }" + }, + { + "category_id": 13, + "poly": [ + 1335, + 519, + 1403, + 519, + 1403, + 549, + 1335, + 549 + ], + "score": 0.46, + "latex": "1 e { - 4 }" + }, + { + "category_id": 13, + "poly": [ + 853, + 358, + 900, + 358, + 900, + 387, + 853, + 387 + ], + "score": 0.31, + "latex": "2 0 \\mathrm { k }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 596.0, + 1403.0, + 596.0, + 1403.0, + 629.0, + 297.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 626.0, + 1407.0, + 626.0, + 1407.0, + 660.0, + 292.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 655.0, + 1407.0, + 655.0, + 1407.0, + 692.0, + 293.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 689.0, + 1055.0, + 689.0, + 1055.0, + 721.0, + 294.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 421.0, + 1208.0, + 421.0, + 1208.0, + 457.0, + 296.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 421.0, + 1404.0, + 421.0, + 1404.0, + 457.0, + 1275.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 453.0, + 852.0, + 453.0, + 852.0, + 489.0, + 296.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 453.0, + 1222.0, + 453.0, + 1222.0, + 489.0, + 900.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 322.0, + 1234.0, + 322.0, + 1234.0, + 363.0, + 293.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 322.0, + 1403.0, + 322.0, + 1403.0, + 363.0, + 1301.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 357.0, + 852.0, + 357.0, + 852.0, + 393.0, + 295.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 357.0, + 1222.0, + 357.0, + 1222.0, + 393.0, + 901.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 518.0, + 1334.0, + 518.0, + 1334.0, + 553.0, + 295.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 550.0, + 938.0, + 550.0, + 938.0, + 585.0, + 295.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 550.0, + 1316.0, + 550.0, + 1316.0, + 585.0, + 994.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 227.0, + 1405.0, + 227.0, + 1405.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 262.0, + 554.0, + 262.0, + 554.0, + 294.0, + 297.0, + 294.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1294, + 1404, + 1294, + 1404, + 1417, + 298, + 1417 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 338, + 1404, + 338, + 1404, + 494, + 297, + 494 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 296, + 880, + 1403, + 880, + 1403, + 1132, + 296, + 1132 + ], + "score": 0.967 + }, + { + "category_id": 3, + "poly": [ + 297, + 518, + 1403, + 518, + 1403, + 769, + 297, + 769 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 373, + 1557, + 786, + 1557, + 786, + 1680, + 373, + 1680 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 296, + 1479, + 1402, + 1479, + 1402, + 1543, + 296, + 1543 + ], + "score": 0.944 + }, + { + "category_id": 4, + "poly": [ + 292, + 770, + 1405, + 770, + 1405, + 825, + 292, + 825 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 301, + 1742, + 700, + 1742, + 700, + 1774, + 301, + 1774 + ], + "score": 0.926 + }, + { + "category_id": 0, + "poly": [ + 299, + 1241, + 896, + 1241, + 896, + 1272, + 299, + 1272 + ], + "score": 0.912 + }, + { + "category_id": 0, + "poly": [ + 298, + 1432, + 392, + 1432, + 392, + 1464, + 298, + 1464 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 299, + 286, + 1007, + 286, + 1007, + 319, + 299, + 319 + ], + "score": 0.897 + }, + { + "category_id": 0, + "poly": [ + 299, + 1695, + 376, + 1695, + 376, + 1727, + 299, + 1727 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.894 + }, + { + "category_id": 1, + "poly": [ + 370, + 1789, + 1409, + 1789, + 1409, + 2035, + 370, + 2035 + ], + "score": 0.853 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.852 + }, + { + "category_id": 0, + "poly": [ + 300, + 229, + 812, + 229, + 812, + 262, + 300, + 262 + ], + "score": 0.602 + }, + { + "category_id": 4, + "poly": [ + 293, + 1134, + 1403, + 1134, + 1403, + 1189, + 293, + 1189 + ], + "score": 0.502 + }, + { + "category_id": 1, + "poly": [ + 293, + 1134, + 1403, + 1134, + 1403, + 1189, + 293, + 1189 + ], + "score": 0.483 + }, + { + "category_id": 1, + "poly": [ + 300, + 229, + 812, + 229, + 812, + 262, + 300, + 262 + ], + "score": 0.321 + }, + { + "category_id": 13, + "poly": [ + 396, + 1558, + 451, + 1558, + 451, + 1586, + 396, + 1586 + ], + "score": 0.85, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 396, + 1588, + 451, + 1588, + 451, + 1616, + 396, + 1616 + ], + "score": 0.84, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 395, + 1649, + 451, + 1649, + 451, + 1678, + 395, + 1678 + ], + "score": 0.84, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 396, + 1619, + 451, + 1619, + 451, + 1647, + 396, + 1647 + ], + "score": 0.84, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 849, + 752, + 896, + 752, + 896, + 768, + 849, + 768 + ], + "score": 0.69, + "latex": "{ \\sf S } = 0 . 2 5" + }, + { + "category_id": 13, + "poly": [ + 1076, + 800, + 1092, + 800, + 1092, + 819, + 1076, + 819 + ], + "score": 0.61, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 702, + 751, + 741, + 751, + 741, + 768, + 702, + 768 + ], + "score": 0.6, + "latex": "{ \\tt S } = 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 1005, + 751, + 1045, + 751, + 1045, + 768, + 1005, + 768 + ], + "score": 0.43, + "latex": "{ \\sf s } = { \\sf 1 } . 0" + }, + { + "category_id": 13, + "poly": [ + 699, + 1115, + 739, + 1115, + 739, + 1132, + 699, + 1132 + ], + "score": 0.27, + "latex": "{ \\tt S } = 0 . 1" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 886.0, + 529.0, + 886.0, + 529.0, + 909.0, + 382.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 884.0, + 803.0, + 884.0, + 803.0, + 910.0, + 656.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 885.0, + 1080.0, + 885.0, + 1080.0, + 910.0, + 928.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 885.0, + 1355.0, + 885.0, + 1355.0, + 910.0, + 1202.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 905.0, + 344.0, + 905.0, + 344.0, + 927.0, + 308.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 905.0, + 614.0, + 905.0, + 614.0, + 927.0, + 582.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 905.0, + 892.0, + 905.0, + 892.0, + 927.0, + 857.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 905.0, + 1163.0, + 905.0, + 1163.0, + 927.0, + 1131.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 279.0, + 928.0, + 354.0, + 928.0, + 354.0, + 1056.0, + 279.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 938.0, + 614.0, + 938.0, + 614.0, + 1048.0, + 569.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 928.0, + 904.0, + 928.0, + 904.0, + 1056.0, + 829.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 938.0, + 1163.0, + 938.0, + 1163.0, + 1048.0, + 1117.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 1011.0, + 613.0, + 1011.0, + 613.0, + 1033.0, + 582.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1010.0, + 1161.0, + 1010.0, + 1161.0, + 1033.0, + 1130.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1063.0, + 339.0, + 1063.0, + 339.0, + 1085.0, + 308.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 1063.0, + 613.0, + 1063.0, + 613.0, + 1085.0, + 582.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1063.0, + 888.0, + 1063.0, + 888.0, + 1085.0, + 856.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1063.0, + 1163.0, + 1063.0, + 1163.0, + 1085.0, + 1131.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1079.0, + 359.0, + 1079.0, + 359.0, + 1096.0, + 344.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1078.0, + 582.0, + 1078.0, + 582.0, + 1098.0, + 383.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1078.0, + 636.0, + 1078.0, + 636.0, + 1099.0, + 616.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 1078.0, + 856.0, + 1078.0, + 856.0, + 1098.0, + 657.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1078.0, + 910.0, + 1078.0, + 910.0, + 1099.0, + 891.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 1078.0, + 1130.0, + 1078.0, + 1130.0, + 1098.0, + 931.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 1078.0, + 1184.0, + 1078.0, + 1184.0, + 1099.0, + 1165.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1078.0, + 1404.0, + 1078.0, + 1404.0, + 1098.0, + 1205.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 1094.0, + 533.0, + 1094.0, + 533.0, + 1116.0, + 378.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1094.0, + 807.0, + 1094.0, + 807.0, + 1117.0, + 653.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 1096.0, + 1080.0, + 1096.0, + 1080.0, + 1116.0, + 929.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1095.0, + 1352.0, + 1095.0, + 1352.0, + 1115.0, + 1202.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1115.0, + 596.0, + 1115.0, + 596.0, + 1134.0, + 531.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 1114.0, + 698.0, + 1114.0, + 698.0, + 1133.0, + 643.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1114.0, + 743.0, + 1114.0, + 743.0, + 1133.0, + 740.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 1114.0, + 896.0, + 1114.0, + 896.0, + 1133.0, + 789.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 1114.0, + 1043.0, + 1114.0, + 1043.0, + 1133.0, + 945.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1111.0, + 1191.0, + 1111.0, + 1191.0, + 1134.0, + 1087.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 520.0, + 531.0, + 520.0, + 531.0, + 544.0, + 383.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 518.0, + 809.0, + 518.0, + 809.0, + 545.0, + 658.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 518.0, + 1086.0, + 518.0, + 1086.0, + 545.0, + 932.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 518.0, + 1363.0, + 518.0, + 1363.0, + 545.0, + 1207.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 540.0, + 340.0, + 540.0, + 340.0, + 562.0, + 307.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 540.0, + 618.0, + 540.0, + 618.0, + 562.0, + 584.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 550.0, + 636.0, + 550.0, + 636.0, + 559.0, + 626.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 540.0, + 891.0, + 540.0, + 891.0, + 563.0, + 860.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 540.0, + 1174.0, + 540.0, + 1174.0, + 562.0, + 1136.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 556.0, + 354.0, + 556.0, + 354.0, + 691.0, + 278.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 556.0, + 630.0, + 556.0, + 630.0, + 696.0, + 556.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 556.0, + 907.0, + 556.0, + 907.0, + 692.0, + 834.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 557.0, + 1184.0, + 557.0, + 1184.0, + 698.0, + 1110.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 667.0, + 340.0, + 667.0, + 340.0, + 690.0, + 307.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 667.0, + 616.0, + 667.0, + 616.0, + 690.0, + 584.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 667.0, + 893.0, + 667.0, + 893.0, + 690.0, + 860.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 667.0, + 1169.0, + 667.0, + 1169.0, + 690.0, + 1136.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 700.0, + 339.0, + 700.0, + 339.0, + 721.0, + 307.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 700.0, + 615.0, + 700.0, + 615.0, + 721.0, + 584.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 700.0, + 891.0, + 700.0, + 891.0, + 721.0, + 860.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 698.0, + 1168.0, + 698.0, + 1168.0, + 723.0, + 1136.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 715.0, + 360.0, + 715.0, + 360.0, + 732.0, + 344.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 713.0, + 448.0, + 713.0, + 448.0, + 735.0, + 407.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 713.0, + 523.0, + 713.0, + 523.0, + 735.0, + 481.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 715.0, + 636.0, + 715.0, + 636.0, + 732.0, + 620.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 711.0, + 726.0, + 711.0, + 726.0, + 736.0, + 680.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 713.0, + 800.0, + 713.0, + 800.0, + 735.0, + 757.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 715.0, + 912.0, + 715.0, + 912.0, + 732.0, + 896.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 713.0, + 1001.0, + 713.0, + 1001.0, + 735.0, + 959.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 713.0, + 1076.0, + 713.0, + 1076.0, + 735.0, + 1033.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 715.0, + 1188.0, + 715.0, + 1188.0, + 732.0, + 1173.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 713.0, + 1277.0, + 713.0, + 1277.0, + 735.0, + 1234.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 713.0, + 1353.0, + 713.0, + 1353.0, + 735.0, + 1310.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 731.0, + 532.0, + 731.0, + 532.0, + 751.0, + 381.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 730.0, + 810.0, + 730.0, + 810.0, + 753.0, + 655.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 730.0, + 1086.0, + 730.0, + 1086.0, + 753.0, + 932.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 731.0, + 1361.0, + 731.0, + 1361.0, + 751.0, + 1209.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 751.0, + 597.0, + 751.0, + 597.0, + 770.0, + 532.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 750.0, + 701.0, + 750.0, + 701.0, + 769.0, + 646.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 750.0, + 746.0, + 750.0, + 746.0, + 769.0, + 742.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 750.0, + 848.0, + 750.0, + 848.0, + 769.0, + 793.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 750.0, + 900.0, + 750.0, + 900.0, + 769.0, + 897.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 750.0, + 1004.0, + 750.0, + 1004.0, + 770.0, + 946.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 750.0, + 1049.0, + 750.0, + 1049.0, + 770.0, + 1046.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 750.0, + 1197.0, + 750.0, + 1197.0, + 769.0, + 1096.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 767.0, + 1406.0, + 767.0, + 1406.0, + 798.0, + 295.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 790.0, + 1075.0, + 790.0, + 1075.0, + 830.0, + 291.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 790.0, + 1404.0, + 790.0, + 1404.0, + 830.0, + 1093.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1239.0, + 898.0, + 1239.0, + 898.0, + 1275.0, + 297.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1425.0, + 397.0, + 1425.0, + 397.0, + 1475.0, + 293.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 285.0, + 1009.0, + 285.0, + 1009.0, + 322.0, + 294.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1690.0, + 381.0, + 1690.0, + 381.0, + 1748.0, + 294.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 818.0, + 229.0, + 818.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1130.0, + 1404.0, + 1130.0, + 1404.0, + 1160.0, + 295.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1155.0, + 1406.0, + 1155.0, + 1406.0, + 1194.0, + 291.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1290.0, + 1405.0, + 1290.0, + 1405.0, + 1330.0, + 293.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1324.0, + 1406.0, + 1324.0, + 1406.0, + 1360.0, + 294.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1356.0, + 1405.0, + 1356.0, + 1405.0, + 1392.0, + 294.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1385.0, + 752.0, + 1385.0, + 752.0, + 1422.0, + 295.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 339.0, + 1405.0, + 339.0, + 1405.0, + 376.0, + 294.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 372.0, + 1404.0, + 372.0, + 1404.0, + 405.0, + 295.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 403.0, + 1404.0, + 403.0, + 1404.0, + 436.0, + 295.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 431.0, + 1406.0, + 431.0, + 1406.0, + 469.0, + 291.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 461.0, + 1392.0, + 461.0, + 1392.0, + 499.0, + 295.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1553.0, + 395.0, + 1553.0, + 395.0, + 1591.0, + 370.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1553.0, + 768.0, + 1553.0, + 768.0, + 1591.0, + 452.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1584.0, + 395.0, + 1584.0, + 395.0, + 1621.0, + 369.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1584.0, + 780.0, + 1584.0, + 780.0, + 1621.0, + 452.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1615.0, + 395.0, + 1615.0, + 395.0, + 1650.0, + 370.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1615.0, + 777.0, + 1615.0, + 777.0, + 1650.0, + 452.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1644.0, + 394.0, + 1644.0, + 394.0, + 1683.0, + 371.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1644.0, + 789.0, + 1644.0, + 789.0, + 1683.0, + 452.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1478.0, + 1404.0, + 1478.0, + 1404.0, + 1514.0, + 296.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1507.0, + 671.0, + 1507.0, + 671.0, + 1547.0, + 295.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1736.0, + 702.0, + 1736.0, + 702.0, + 1780.0, + 296.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1786.0, + 1403.0, + 1786.0, + 1403.0, + 1828.0, + 373.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1818.0, + 1403.0, + 1818.0, + 1403.0, + 1857.0, + 392.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1848.0, + 505.0, + 1848.0, + 505.0, + 1883.0, + 393.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 1881.0, + 1406.0, + 1881.0, + 1406.0, + 1918.0, + 377.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1913.0, + 1402.0, + 1913.0, + 1402.0, + 1947.0, + 394.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1979.0, + 392.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2008.0, + 369.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 2005.0, + 1378.0, + 2005.0, + 1378.0, + 2037.0, + 395.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1130.0, + 1404.0, + 1130.0, + 1404.0, + 1160.0, + 295.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1155.0, + 1406.0, + 1155.0, + 1406.0, + 1194.0, + 291.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 818.0, + 229.0, + 818.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 21, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 307, + 347, + 1392, + 347, + 1392, + 1040, + 307, + 1040 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 306, + 1162, + 1392, + 1162, + 1392, + 1863, + 306, + 1863 + ], + "score": 0.971 + }, + { + "category_id": 4, + "poly": [ + 296, + 1061, + 1403, + 1061, + 1403, + 1121, + 296, + 1121 + ], + "score": 0.93 + }, + { + "category_id": 4, + "poly": [ + 295, + 1884, + 1403, + 1884, + 1403, + 1943, + 295, + 1943 + ], + "score": 0.922 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.894 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 1402, + 229, + 1402, + 323, + 297, + 323 + ], + "score": 0.878 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.855 + }, + { + "category_id": 15, + "poly": [ + 282.0, + 328.0, + 386.0, + 328.0, + 386.0, + 571.0, + 282.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 350.0, + 526.0, + 350.0, + 526.0, + 370.0, + 398.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 325.0, + 658.0, + 325.0, + 658.0, + 575.0, + 558.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 350.0, + 798.0, + 350.0, + 798.0, + 370.0, + 668.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 334.0, + 924.0, + 334.0, + 924.0, + 479.0, + 835.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 350.0, + 1070.0, + 350.0, + 1070.0, + 370.0, + 938.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 334.0, + 1191.0, + 334.0, + 1191.0, + 435.0, + 1109.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 349.0, + 1342.0, + 349.0, + 1342.0, + 369.0, + 1207.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 363.0, + 388.0, + 363.0, + 388.0, + 411.0, + 325.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 378.0, + 482.0, + 378.0, + 482.0, + 420.0, + 471.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 373.0, + 559.0, + 373.0, + 559.0, + 425.0, + 490.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 399.0, + 632.0, + 399.0, + 632.0, + 421.0, + 593.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 380.0, + 664.0, + 380.0, + 664.0, + 433.0, + 640.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 396.0, + 725.0, + 396.0, + 725.0, + 415.0, + 710.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 400.0, + 937.0, + 400.0, + 937.0, + 420.0, + 912.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 376.0, + 1025.0, + 376.0, + 1025.0, + 420.0, + 1010.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 373.0, + 1100.0, + 373.0, + 1100.0, + 425.0, + 1032.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 369.0, + 1198.0, + 369.0, + 1198.0, + 428.0, + 1174.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 378.0, + 1295.0, + 378.0, + 1295.0, + 420.0, + 1283.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 387.0, + 1370.0, + 387.0, + 1370.0, + 425.0, + 1303.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 415.0, + 927.0, + 415.0, + 927.0, + 433.0, + 914.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 415.0, + 945.0, + 415.0, + 945.0, + 428.0, + 935.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 425.0, + 363.0, + 425.0, + 363.0, + 508.0, + 324.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 430.0, + 420.0, + 430.0, + 420.0, + 494.0, + 370.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 504.0, + 452.0, + 504.0, + 452.0, + 518.0, + 442.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 511.0, + 484.0, + 511.0, + 484.0, + 526.0, + 466.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 421.0, + 562.0, + 421.0, + 562.0, + 532.0, + 490.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 432.0, + 633.0, + 432.0, + 633.0, + 526.0, + 593.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 447.0, + 668.0, + 447.0, + 668.0, + 458.0, + 658.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 515.0, + 712.0, + 515.0, + 712.0, + 531.0, + 695.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 462.0, + 754.0, + 462.0, + 754.0, + 525.0, + 740.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 447.0, + 827.0, + 447.0, + 827.0, + 527.0, + 761.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 425.0, + 933.0, + 425.0, + 933.0, + 551.0, + 837.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 427.0, + 961.0, + 427.0, + 961.0, + 468.0, + 937.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 496.0, + 996.0, + 496.0, + 996.0, + 527.0, + 970.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 423.0, + 1026.0, + 423.0, + 1026.0, + 438.0, + 1010.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 421.0, + 1104.0, + 421.0, + 1104.0, + 534.0, + 1032.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 442.0, + 1182.0, + 442.0, + 1182.0, + 551.0, + 1106.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 422.0, + 1222.0, + 422.0, + 1222.0, + 477.0, + 1184.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 430.0, + 1268.0, + 430.0, + 1268.0, + 531.0, + 1238.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 426.0, + 1296.0, + 426.0, + 1296.0, + 477.0, + 1278.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 421.0, + 1374.0, + 421.0, + 1374.0, + 505.0, + 1303.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 530.0, + 446.0, + 530.0, + 446.0, + 547.0, + 348.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 531.0, + 477.0, + 531.0, + 477.0, + 545.0, + 448.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 530.0, + 577.0, + 530.0, + 577.0, + 546.0, + 479.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 529.0, + 715.0, + 529.0, + 715.0, + 549.0, + 619.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 530.0, + 750.0, + 530.0, + 750.0, + 546.0, + 717.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 530.0, + 847.0, + 530.0, + 847.0, + 546.0, + 751.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 531.0, + 920.0, + 531.0, + 920.0, + 545.0, + 893.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 530.0, + 954.0, + 530.0, + 954.0, + 546.0, + 921.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 531.0, + 985.0, + 531.0, + 985.0, + 545.0, + 957.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 530.0, + 1020.0, + 530.0, + 1020.0, + 546.0, + 988.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 530.0, + 1119.0, + 530.0, + 1119.0, + 546.0, + 1021.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 529.0, + 1256.0, + 529.0, + 1256.0, + 549.0, + 1161.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 530.0, + 1388.0, + 530.0, + 1388.0, + 546.0, + 1258.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 541.0, + 496.0, + 541.0, + 496.0, + 561.0, + 426.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 541.0, + 767.0, + 541.0, + 767.0, + 561.0, + 698.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 541.0, + 1038.0, + 541.0, + 1038.0, + 561.0, + 968.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 538.0, + 1309.0, + 538.0, + 1309.0, + 561.0, + 1239.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 582.0, + 366.0, + 582.0, + 366.0, + 792.0, + 308.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 629.0, + 392.0, + 629.0, + 392.0, + 649.0, + 369.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 582.0, + 548.0, + 582.0, + 548.0, + 643.0, + 400.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 580.0, + 636.0, + 580.0, + 636.0, + 794.0, + 579.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 613.0, + 664.0, + 613.0, + 664.0, + 651.0, + 643.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 583.0, + 801.0, + 583.0, + 801.0, + 650.0, + 674.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 587.0, + 927.0, + 587.0, + 927.0, + 656.0, + 853.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 582.0, + 1090.0, + 582.0, + 1090.0, + 645.0, + 931.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 558.0, + 1201.0, + 558.0, + 1201.0, + 780.0, + 1100.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 583.0, + 1361.0, + 583.0, + 1361.0, + 643.0, + 1211.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 644.0, + 394.0, + 644.0, + 394.0, + 667.0, + 373.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 641.0, + 566.0, + 641.0, + 566.0, + 658.0, + 498.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 643.0, + 665.0, + 643.0, + 665.0, + 667.0, + 646.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 651.0, + 754.0, + 651.0, + 754.0, + 667.0, + 740.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 648.0, + 874.0, + 648.0, + 874.0, + 696.0, + 853.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 639.0, + 930.0, + 639.0, + 930.0, + 667.0, + 915.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 651.0, + 947.0, + 651.0, + 947.0, + 661.0, + 938.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 641.0, + 1107.0, + 641.0, + 1107.0, + 658.0, + 1040.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1288.0, + 643.0, + 1303.0, + 643.0, + 1303.0, + 658.0, + 1288.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 641.0, + 1378.0, + 641.0, + 1378.0, + 658.0, + 1310.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 660.0, + 364.0, + 660.0, + 364.0, + 682.0, + 325.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 658.0, + 398.0, + 658.0, + 398.0, + 722.0, + 368.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 661.0, + 488.0, + 661.0, + 488.0, + 672.0, + 477.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 656.0, + 558.0, + 656.0, + 558.0, + 676.0, + 496.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 663.0, + 759.0, + 663.0, + 759.0, + 680.0, + 745.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 660.0, + 906.0, + 660.0, + 906.0, + 682.0, + 866.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 666.0, + 930.0, + 666.0, + 930.0, + 677.0, + 917.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 666.0, + 948.0, + 666.0, + 948.0, + 672.0, + 942.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 663.0, + 1030.0, + 663.0, + 1030.0, + 672.0, + 1019.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 656.0, + 1099.0, + 656.0, + 1099.0, + 676.0, + 1038.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 657.0, + 1206.0, + 657.0, + 1206.0, + 682.0, + 1188.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 660.0, + 1303.0, + 660.0, + 1303.0, + 674.0, + 1286.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 656.0, + 1370.0, + 656.0, + 1370.0, + 676.0, + 1309.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 690.0, + 367.0, + 690.0, + 367.0, + 779.0, + 324.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 681.0, + 424.0, + 681.0, + 424.0, + 732.0, + 396.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 743.0, + 456.0, + 743.0, + 456.0, + 757.0, + 446.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 748.0, + 489.0, + 748.0, + 489.0, + 767.0, + 468.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 755.0, + 521.0, + 755.0, + 521.0, + 768.0, + 511.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 669.0, + 637.0, + 669.0, + 637.0, + 767.0, + 594.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 682.0, + 674.0, + 682.0, + 674.0, + 695.0, + 663.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 754.0, + 720.0, + 754.0, + 720.0, + 770.0, + 703.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 700.0, + 761.0, + 700.0, + 761.0, + 762.0, + 746.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 684.0, + 836.0, + 684.0, + 836.0, + 767.0, + 766.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 674.0, + 931.0, + 674.0, + 931.0, + 798.0, + 838.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 726.0, + 943.0, + 726.0, + 943.0, + 736.0, + 932.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 676.0, + 966.0, + 676.0, + 966.0, + 705.0, + 946.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 716.0, + 1000.0, + 716.0, + 1000.0, + 752.0, + 972.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 749.0, + 1030.0, + 749.0, + 1030.0, + 764.0, + 1014.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 755.0, + 1066.0, + 755.0, + 1066.0, + 772.0, + 1052.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 759.0, + 1109.0, + 759.0, + 1109.0, + 769.0, + 1099.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 701.0, + 1177.0, + 701.0, + 1177.0, + 791.0, + 1122.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1203.0, + 695.0, + 1225.0, + 695.0, + 1225.0, + 713.0, + 1203.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 669.0, + 1270.0, + 669.0, + 1270.0, + 770.0, + 1246.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 690.0, + 1301.0, + 690.0, + 1301.0, + 716.0, + 1282.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 717.0, + 1336.0, + 717.0, + 1336.0, + 733.0, + 1322.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1369.0, + 729.0, + 1380.0, + 729.0, + 1380.0, + 742.0, + 1369.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 769.0, + 383.0, + 769.0, + 383.0, + 786.0, + 351.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 770.0, + 417.0, + 770.0, + 417.0, + 786.0, + 384.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 769.0, + 487.0, + 769.0, + 487.0, + 786.0, + 419.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 770.0, + 521.0, + 770.0, + 521.0, + 786.0, + 488.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 770.0, + 554.0, + 770.0, + 554.0, + 786.0, + 522.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 769.0, + 724.0, + 769.0, + 724.0, + 786.0, + 622.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 770.0, + 757.0, + 770.0, + 757.0, + 786.0, + 725.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 770.0, + 791.0, + 770.0, + 791.0, + 786.0, + 759.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 769.0, + 825.0, + 769.0, + 825.0, + 786.0, + 794.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 769.0, + 924.0, + 769.0, + 924.0, + 786.0, + 893.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 770.0, + 958.0, + 770.0, + 958.0, + 786.0, + 926.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 770.0, + 994.0, + 770.0, + 994.0, + 786.0, + 961.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 769.0, + 1027.0, + 769.0, + 1027.0, + 786.0, + 995.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 770.0, + 1062.0, + 770.0, + 1062.0, + 786.0, + 1030.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 770.0, + 1096.0, + 770.0, + 1096.0, + 786.0, + 1064.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 769.0, + 1228.0, + 769.0, + 1228.0, + 786.0, + 1164.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 770.0, + 1265.0, + 770.0, + 1265.0, + 786.0, + 1232.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 770.0, + 1299.0, + 770.0, + 1299.0, + 786.0, + 1267.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 770.0, + 1367.0, + 770.0, + 1367.0, + 786.0, + 1300.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 777.0, + 499.0, + 777.0, + 499.0, + 802.0, + 433.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 780.0, + 768.0, + 780.0, + 768.0, + 800.0, + 704.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 780.0, + 1041.0, + 780.0, + 1041.0, + 800.0, + 975.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 777.0, + 1313.0, + 777.0, + 1313.0, + 802.0, + 1246.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 824.0, + 366.0, + 824.0, + 366.0, + 1029.0, + 308.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 850.0, + 392.0, + 850.0, + 392.0, + 866.0, + 376.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 821.0, + 532.0, + 821.0, + 532.0, + 866.0, + 396.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 858.0, + 545.0, + 858.0, + 545.0, + 869.0, + 534.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 855.0, + 559.0, + 855.0, + 559.0, + 867.0, + 550.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 817.0, + 638.0, + 817.0, + 638.0, + 1036.0, + 578.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 824.0, + 803.0, + 824.0, + 803.0, + 843.0, + 674.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 858.0, + 810.0, + 858.0, + 810.0, + 864.0, + 804.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 825.0, + 909.0, + 825.0, + 909.0, + 898.0, + 852.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 850.0, + 932.0, + 850.0, + 932.0, + 866.0, + 917.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 824.0, + 1084.0, + 824.0, + 1084.0, + 868.0, + 938.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 798.0, + 1204.0, + 798.0, + 1204.0, + 1023.0, + 1085.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 821.0, + 1348.0, + 821.0, + 1348.0, + 866.0, + 1210.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 842.0, + 1383.0, + 842.0, + 1383.0, + 862.0, + 1354.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 868.0, + 366.0, + 868.0, + 366.0, + 924.0, + 324.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 866.0, + 392.0, + 866.0, + 392.0, + 915.0, + 376.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 862.0, + 467.0, + 862.0, + 467.0, + 917.0, + 396.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 869.0, + 552.0, + 869.0, + 552.0, + 918.0, + 511.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 872.0, + 636.0, + 872.0, + 636.0, + 933.0, + 595.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 905.0, + 700.0, + 905.0, + 700.0, + 918.0, + 689.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 894.0, + 714.0, + 894.0, + 714.0, + 905.0, + 703.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 887.0, + 788.0, + 887.0, + 788.0, + 907.0, + 772.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 868.0, + 915.0, + 868.0, + 915.0, + 1006.0, + 837.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 866.0, + 932.0, + 866.0, + 932.0, + 914.0, + 917.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 862.0, + 1010.0, + 862.0, + 1010.0, + 917.0, + 940.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 867.0, + 1091.0, + 867.0, + 1091.0, + 933.0, + 1035.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 866.0, + 1204.0, + 866.0, + 1204.0, + 915.0, + 1188.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 862.0, + 1293.0, + 862.0, + 1293.0, + 926.0, + 1210.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 866.0, + 1325.0, + 866.0, + 1325.0, + 920.0, + 1306.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 861.0, + 1353.0, + 861.0, + 1353.0, + 899.0, + 1326.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 930.0, + 366.0, + 930.0, + 366.0, + 956.0, + 324.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 945.0, + 636.0, + 945.0, + 636.0, + 971.0, + 594.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 943.0, + 651.0, + 943.0, + 651.0, + 954.0, + 641.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 925.0, + 677.0, + 925.0, + 677.0, + 938.0, + 666.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 923.0, + 761.0, + 923.0, + 761.0, + 969.0, + 745.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 936.0, + 820.0, + 936.0, + 820.0, + 972.0, + 767.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 933.0, + 906.0, + 933.0, + 906.0, + 955.0, + 866.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 928.0, + 1057.0, + 928.0, + 1057.0, + 948.0, + 1027.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 940.0, + 1177.0, + 940.0, + 1177.0, + 962.0, + 1136.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 959.0, + 1217.0, + 959.0, + 1217.0, + 971.0, + 1207.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 930.0, + 1258.0, + 930.0, + 1258.0, + 954.0, + 1232.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 965.0, + 392.0, + 965.0, + 392.0, + 1019.0, + 325.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 996.0, + 406.0, + 996.0, + 406.0, + 1006.0, + 395.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 992.0, + 447.0, + 992.0, + 447.0, + 1002.0, + 432.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 982.0, + 658.0, + 982.0, + 658.0, + 1012.0, + 595.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 974.0, + 758.0, + 974.0, + 758.0, + 1002.0, + 747.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 969.0, + 838.0, + 969.0, + 838.0, + 1006.0, + 767.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 977.0, + 932.0, + 977.0, + 932.0, + 1035.0, + 843.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 997.0, + 951.0, + 997.0, + 951.0, + 1011.0, + 935.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 990.0, + 988.0, + 990.0, + 988.0, + 998.0, + 975.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 991.0, + 1021.0, + 991.0, + 1021.0, + 1003.0, + 1010.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 991.0, + 1145.0, + 991.0, + 1145.0, + 1032.0, + 1122.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 970.0, + 1193.0, + 970.0, + 1193.0, + 982.0, + 1183.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1009.0, + 390.0, + 1009.0, + 390.0, + 1027.0, + 355.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1009.0, + 435.0, + 1009.0, + 435.0, + 1027.0, + 398.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 1009.0, + 480.0, + 1009.0, + 480.0, + 1027.0, + 445.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1009.0, + 526.0, + 1009.0, + 526.0, + 1027.0, + 490.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1009.0, + 573.0, + 1009.0, + 573.0, + 1027.0, + 536.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 1005.0, + 661.0, + 1005.0, + 661.0, + 1027.0, + 625.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1009.0, + 705.0, + 1009.0, + 705.0, + 1027.0, + 669.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1009.0, + 752.0, + 1009.0, + 752.0, + 1027.0, + 714.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1009.0, + 798.0, + 1009.0, + 798.0, + 1027.0, + 761.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 1009.0, + 843.0, + 1009.0, + 843.0, + 1027.0, + 806.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1009.0, + 933.0, + 1009.0, + 933.0, + 1027.0, + 896.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1009.0, + 977.0, + 1009.0, + 977.0, + 1027.0, + 940.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1009.0, + 1024.0, + 1009.0, + 1024.0, + 1027.0, + 987.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1009.0, + 1069.0, + 1009.0, + 1069.0, + 1027.0, + 1032.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1009.0, + 1115.0, + 1009.0, + 1115.0, + 1027.0, + 1077.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 1007.0, + 1204.0, + 1007.0, + 1204.0, + 1027.0, + 1167.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1011.0, + 1247.0, + 1011.0, + 1247.0, + 1027.0, + 1211.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1009.0, + 1294.0, + 1009.0, + 1294.0, + 1027.0, + 1257.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1009.0, + 1340.0, + 1009.0, + 1340.0, + 1027.0, + 1303.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1009.0, + 1385.0, + 1009.0, + 1385.0, + 1027.0, + 1348.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1021.0, + 498.0, + 1021.0, + 498.0, + 1040.0, + 435.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1018.0, + 770.0, + 1018.0, + 770.0, + 1042.0, + 706.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1018.0, + 1040.0, + 1018.0, + 1040.0, + 1042.0, + 976.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1018.0, + 1311.0, + 1018.0, + 1311.0, + 1042.0, + 1246.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.75, + 365.0, + 694.75, + 365.0, + 694.75, + 390.0, + 628.75, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.75, + 368.5, + 924.75, + 368.5, + 924.75, + 389.5, + 902.75, + 389.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 383.5, + 1226.0, + 383.5, + 1226.0, + 399.5, + 1196.0, + 399.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 388.5, + 708.0, + 388.5, + 708.0, + 401.5, + 702.0, + 401.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.75, + 372.5, + 1326.75, + 372.5, + 1326.75, + 391.0, + 1302.75, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.75, + 386.5, + 938.75, + 386.5, + 938.75, + 405.0, + 899.75, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 411.5, + 393.0, + 411.5, + 393.0, + 425.0, + 368.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 414.0, + 762.0, + 414.0, + 762.0, + 445.5, + 727.0, + 445.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 484.0, + 680.0, + 484.0, + 680.0, + 494.0, + 675.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 490.5, + 1223.0, + 490.5, + 1223.0, + 500.5, + 1214.0, + 500.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.75, + 475.5, + 987.75, + 475.5, + 987.75, + 503.0, + 961.75, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.75, + 497.5, + 437.75, + 497.5, + 437.75, + 506.5, + 431.75, + 506.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 509.0, + 1023.0, + 509.0, + 1023.0, + 524.0, + 1009.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 520.5, + 438.0, + 520.5, + 438.0, + 529.5, + 429.0, + 529.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.25, + 599.5, + 697.25, + 599.5, + 697.25, + 624.5, + 631.25, + 624.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.25, + 599.5, + 390.25, + 599.5, + 390.25, + 633.5, + 357.25, + 633.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.75, + 618.5, + 1232.75, + 618.5, + 1232.75, + 634.5, + 1199.75, + 634.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.25, + 647.5, + 1263.25, + 647.5, + 1263.25, + 676.0, + 1235.25, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 720.5, + 683.0, + 720.5, + 683.0, + 730.5, + 677.0, + 730.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 687.5, + 941.0, + 687.5, + 941.0, + 711.0, + 916.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 729.0, + 1229.0, + 729.0, + 1229.0, + 740.0, + 1217.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 736.5, + 441.0, + 736.5, + 441.0, + 745.5, + 435.0, + 745.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.75, + 841.0, + 572.75, + 841.0, + 572.75, + 859.0, + 543.75, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 836.5, + 847.0, + 836.5, + 847.0, + 865.0, + 750.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 845.5, + 1349.0, + 845.5, + 1349.0, + 859.5, + 1319.0, + 859.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 861.5, + 1080.0, + 861.5, + 1080.0, + 871.0, + 1057.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 884.5, + 540.0, + 884.5, + 540.0, + 903.5, + 519.0, + 903.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 865.0, + 760.0, + 865.0, + 760.0, + 879.0, + 714.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.25, + 902.5, + 1333.25, + 902.5, + 1333.25, + 912.5, + 1312.25, + 912.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 935.0, + 1305.0, + 935.0, + 1305.0, + 952.5, + 1277.0, + 952.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.25, + 958.5, + 733.25, + 958.5, + 733.25, + 968.5, + 711.25, + 968.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.25, + 969.5, + 722.25, + 969.5, + 722.25, + 983.5, + 686.25, + 983.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.75, + 965.5, + 1276.75, + 965.5, + 1276.75, + 975.5, + 1255.75, + 975.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 944.0, + 547.0, + 944.0, + 547.0, + 968.0, + 442.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 993.0, + 1205.0, + 993.0, + 1205.0, + 1011.0, + 1173.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1147.0, + 377.0, + 1147.0, + 377.0, + 1355.0, + 285.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 1165.0, + 567.0, + 1165.0, + 567.0, + 1242.0, + 401.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1161.0, + 659.0, + 1161.0, + 659.0, + 1345.0, + 577.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1166.0, + 837.0, + 1166.0, + 837.0, + 1242.0, + 672.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 1167.0, + 924.0, + 1167.0, + 924.0, + 1288.0, + 844.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1166.0, + 1108.0, + 1166.0, + 1108.0, + 1242.0, + 940.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1184.0, + 1198.0, + 1184.0, + 1198.0, + 1268.0, + 1137.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 1165.0, + 1379.0, + 1165.0, + 1379.0, + 1242.0, + 1209.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1235.0, + 366.0, + 1235.0, + 366.0, + 1261.0, + 323.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1242.0, + 489.0, + 1242.0, + 489.0, + 1257.0, + 475.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 1240.0, + 557.0, + 1240.0, + 557.0, + 1260.0, + 496.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1256.0, + 634.0, + 1256.0, + 634.0, + 1278.0, + 596.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1238.0, + 716.0, + 1238.0, + 716.0, + 1250.0, + 704.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1245.0, + 758.0, + 1245.0, + 758.0, + 1256.0, + 748.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1240.0, + 828.0, + 1240.0, + 828.0, + 1260.0, + 766.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1231.0, + 938.0, + 1231.0, + 938.0, + 1352.0, + 838.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1245.0, + 1029.0, + 1245.0, + 1029.0, + 1255.0, + 1017.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 1240.0, + 1099.0, + 1240.0, + 1099.0, + 1260.0, + 1038.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1253.0, + 1178.0, + 1253.0, + 1178.0, + 1276.0, + 1136.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1235.0, + 1236.0, + 1235.0, + 1236.0, + 1263.0, + 1216.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1266.0, + 1254.0, + 1266.0, + 1254.0, + 1281.0, + 1238.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 1242.0, + 1302.0, + 1242.0, + 1302.0, + 1257.0, + 1286.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 1240.0, + 1370.0, + 1240.0, + 1370.0, + 1260.0, + 1309.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1272.0, + 366.0, + 1272.0, + 366.0, + 1298.0, + 324.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1273.0, + 759.0, + 1273.0, + 759.0, + 1285.0, + 750.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 1273.0, + 1207.0, + 1273.0, + 1207.0, + 1295.0, + 1189.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 1277.0, + 1258.0, + 1277.0, + 1258.0, + 1290.0, + 1247.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1292.0, + 637.0, + 1292.0, + 637.0, + 1319.0, + 595.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1291.0, + 779.0, + 1291.0, + 779.0, + 1306.0, + 764.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 1298.0, + 949.0, + 1298.0, + 949.0, + 1316.0, + 935.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1291.0, + 1178.0, + 1291.0, + 1178.0, + 1314.0, + 1137.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 1298.0, + 1284.0, + 1298.0, + 1284.0, + 1310.0, + 1272.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1326.0, + 333.0, + 1326.0, + 333.0, + 1370.0, + 310.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1335.0, + 444.0, + 1335.0, + 444.0, + 1345.0, + 430.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1336.0, + 472.0, + 1336.0, + 472.0, + 1351.0, + 457.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1339.0, + 507.0, + 1339.0, + 507.0, + 1354.0, + 478.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 1340.0, + 571.0, + 1340.0, + 571.0, + 1355.0, + 554.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1324.0, + 643.0, + 1324.0, + 643.0, + 1382.0, + 574.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 1337.0, + 706.0, + 1337.0, + 706.0, + 1352.0, + 690.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1317.0, + 838.0, + 1317.0, + 838.0, + 1330.0, + 828.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1308.0, + 906.0, + 1308.0, + 906.0, + 1372.0, + 851.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 1311.0, + 937.0, + 1311.0, + 937.0, + 1337.0, + 922.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1315.0, + 961.0, + 1315.0, + 961.0, + 1335.0, + 944.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1340.0, + 1013.0, + 1340.0, + 1013.0, + 1351.0, + 1001.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 1339.0, + 1049.0, + 1339.0, + 1049.0, + 1354.0, + 1019.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1341.0, + 1108.0, + 1341.0, + 1108.0, + 1354.0, + 1099.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1329.0, + 1178.0, + 1329.0, + 1178.0, + 1350.0, + 1137.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 1339.0, + 1248.0, + 1339.0, + 1248.0, + 1354.0, + 1231.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 1306.0, + 1301.0, + 1306.0, + 1301.0, + 1317.0, + 1291.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 1320.0, + 1318.0, + 1320.0, + 1318.0, + 1330.0, + 1309.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 1336.0, + 1380.0, + 1336.0, + 1380.0, + 1347.0, + 1370.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1351.0, + 428.0, + 1351.0, + 428.0, + 1369.0, + 363.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 1352.0, + 497.0, + 1352.0, + 497.0, + 1369.0, + 431.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1352.0, + 564.0, + 1352.0, + 564.0, + 1369.0, + 499.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 1352.0, + 698.0, + 1352.0, + 698.0, + 1369.0, + 635.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1354.0, + 733.0, + 1354.0, + 733.0, + 1367.0, + 704.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1352.0, + 769.0, + 1352.0, + 769.0, + 1369.0, + 735.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1351.0, + 834.0, + 1351.0, + 834.0, + 1369.0, + 770.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 1350.0, + 970.0, + 1350.0, + 970.0, + 1370.0, + 903.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1350.0, + 1107.0, + 1350.0, + 1107.0, + 1370.0, + 973.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 1351.0, + 1242.0, + 1351.0, + 1242.0, + 1369.0, + 1176.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 1352.0, + 1377.0, + 1352.0, + 1377.0, + 1369.0, + 1243.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 1362.0, + 502.0, + 1362.0, + 502.0, + 1384.0, + 431.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1362.0, + 772.0, + 1362.0, + 772.0, + 1384.0, + 702.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 1361.0, + 1043.0, + 1361.0, + 1043.0, + 1385.0, + 973.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 1362.0, + 1315.0, + 1362.0, + 1315.0, + 1384.0, + 1243.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1388.0, + 386.0, + 1388.0, + 386.0, + 1595.0, + 285.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1406.0, + 567.0, + 1406.0, + 567.0, + 1500.0, + 402.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1402.0, + 651.0, + 1402.0, + 651.0, + 1585.0, + 578.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1406.0, + 835.0, + 1406.0, + 835.0, + 1499.0, + 674.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 1407.0, + 932.0, + 1407.0, + 932.0, + 1546.0, + 844.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1405.0, + 1108.0, + 1405.0, + 1108.0, + 1499.0, + 939.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1422.0, + 1198.0, + 1422.0, + 1198.0, + 1509.0, + 1137.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 1406.0, + 1378.0, + 1406.0, + 1378.0, + 1502.0, + 1207.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1494.0, + 635.0, + 1494.0, + 635.0, + 1520.0, + 595.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1497.0, + 742.0, + 1497.0, + 742.0, + 1509.0, + 730.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 1491.0, + 934.0, + 1491.0, + 934.0, + 1594.0, + 839.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1494.0, + 1178.0, + 1494.0, + 1178.0, + 1516.0, + 1136.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1506.0, + 1254.0, + 1506.0, + 1254.0, + 1521.0, + 1239.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1512.0, + 366.0, + 1512.0, + 366.0, + 1539.0, + 324.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1514.0, + 759.0, + 1514.0, + 759.0, + 1525.0, + 749.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1516.0, + 1206.0, + 1516.0, + 1206.0, + 1533.0, + 1191.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1516.0, + 1260.0, + 1516.0, + 1260.0, + 1531.0, + 1246.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1533.0, + 637.0, + 1533.0, + 637.0, + 1559.0, + 595.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1533.0, + 776.0, + 1533.0, + 776.0, + 1545.0, + 765.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1531.0, + 1178.0, + 1531.0, + 1178.0, + 1554.0, + 1137.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 1539.0, + 950.0, + 1539.0, + 950.0, + 1556.0, + 935.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 1539.0, + 1284.0, + 1539.0, + 1284.0, + 1550.0, + 1272.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 1546.0, + 1301.0, + 1546.0, + 1301.0, + 1558.0, + 1291.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1568.0, + 333.0, + 1568.0, + 333.0, + 1609.0, + 310.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1575.0, + 443.0, + 1575.0, + 443.0, + 1585.0, + 430.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1576.0, + 472.0, + 1576.0, + 472.0, + 1591.0, + 457.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 1579.0, + 507.0, + 1579.0, + 507.0, + 1594.0, + 477.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1579.0, + 571.0, + 1579.0, + 571.0, + 1595.0, + 555.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1564.0, + 643.0, + 1564.0, + 643.0, + 1623.0, + 572.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1578.0, + 706.0, + 1578.0, + 706.0, + 1593.0, + 691.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1558.0, + 839.0, + 1558.0, + 839.0, + 1570.0, + 828.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1567.0, + 875.0, + 1567.0, + 875.0, + 1609.0, + 852.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 1553.0, + 935.0, + 1553.0, + 935.0, + 1578.0, + 922.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1555.0, + 963.0, + 1555.0, + 963.0, + 1575.0, + 944.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1575.0, + 985.0, + 1575.0, + 985.0, + 1585.0, + 971.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1580.0, + 1013.0, + 1580.0, + 1013.0, + 1591.0, + 1001.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 1581.0, + 1032.0, + 1581.0, + 1032.0, + 1591.0, + 1019.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1579.0, + 1048.0, + 1579.0, + 1048.0, + 1594.0, + 1033.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1580.0, + 1108.0, + 1580.0, + 1108.0, + 1594.0, + 1099.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1569.0, + 1178.0, + 1569.0, + 1178.0, + 1590.0, + 1137.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 1579.0, + 1248.0, + 1579.0, + 1248.0, + 1594.0, + 1231.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1560.0, + 1317.0, + 1560.0, + 1317.0, + 1570.0, + 1307.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 1576.0, + 1380.0, + 1576.0, + 1380.0, + 1588.0, + 1370.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1591.0, + 426.0, + 1591.0, + 426.0, + 1609.0, + 362.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1593.0, + 462.0, + 1593.0, + 462.0, + 1609.0, + 430.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 1593.0, + 496.0, + 1593.0, + 496.0, + 1609.0, + 464.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1591.0, + 564.0, + 1591.0, + 564.0, + 1609.0, + 498.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 1591.0, + 697.0, + 1591.0, + 697.0, + 1609.0, + 633.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1591.0, + 733.0, + 1591.0, + 733.0, + 1609.0, + 701.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 1590.0, + 835.0, + 1590.0, + 835.0, + 1610.0, + 734.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 1590.0, + 969.0, + 1590.0, + 969.0, + 1610.0, + 901.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1591.0, + 1105.0, + 1591.0, + 1105.0, + 1609.0, + 971.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 1591.0, + 1241.0, + 1591.0, + 1241.0, + 1609.0, + 1175.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1242.0, + 1591.0, + 1310.0, + 1591.0, + 1310.0, + 1609.0, + 1242.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1311.0, + 1593.0, + 1375.0, + 1593.0, + 1375.0, + 1609.0, + 1311.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1600.0, + 500.0, + 1600.0, + 500.0, + 1624.0, + 433.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1600.0, + 770.0, + 1600.0, + 770.0, + 1624.0, + 704.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1600.0, + 1041.0, + 1600.0, + 1041.0, + 1624.0, + 974.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1600.0, + 1313.0, + 1600.0, + 1313.0, + 1624.0, + 1246.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1628.0, + 377.0, + 1628.0, + 377.0, + 1835.0, + 285.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1646.0, + 530.0, + 1646.0, + 530.0, + 1666.0, + 403.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1640.0, + 608.0, + 1640.0, + 608.0, + 1825.0, + 576.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1646.0, + 802.0, + 1646.0, + 802.0, + 1666.0, + 674.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1646.0, + 1078.0, + 1646.0, + 1078.0, + 1666.0, + 940.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1646.0, + 1346.0, + 1646.0, + 1346.0, + 1666.0, + 1213.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1678.0, + 366.0, + 1678.0, + 366.0, + 1704.0, + 323.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1673.0, + 391.0, + 1673.0, + 391.0, + 1720.0, + 376.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1668.0, + 467.0, + 1668.0, + 467.0, + 1723.0, + 396.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1668.0, + 567.0, + 1668.0, + 567.0, + 1720.0, + 540.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1658.0, + 638.0, + 1658.0, + 638.0, + 1723.0, + 595.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 1673.0, + 662.0, + 1673.0, + 662.0, + 1720.0, + 646.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1669.0, + 755.0, + 1669.0, + 755.0, + 1723.0, + 669.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 1689.0, + 771.0, + 1689.0, + 771.0, + 1708.0, + 756.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1665.0, + 839.0, + 1665.0, + 839.0, + 1680.0, + 821.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 1674.0, + 931.0, + 1674.0, + 931.0, + 1771.0, + 846.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1668.0, + 1010.0, + 1668.0, + 1010.0, + 1723.0, + 938.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1670.0, + 1111.0, + 1670.0, + 1111.0, + 1728.0, + 1075.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1696.0, + 1178.0, + 1696.0, + 1178.0, + 1719.0, + 1137.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 1673.0, + 1204.0, + 1673.0, + 1204.0, + 1720.0, + 1188.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1669.0, + 1280.0, + 1669.0, + 1280.0, + 1723.0, + 1210.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 1689.0, + 1337.0, + 1689.0, + 1337.0, + 1717.0, + 1314.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1717.0, + 365.0, + 1717.0, + 365.0, + 1739.0, + 324.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 1722.0, + 391.0, + 1722.0, + 391.0, + 1738.0, + 375.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1719.0, + 458.0, + 1719.0, + 458.0, + 1739.0, + 395.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1713.0, + 546.0, + 1713.0, + 546.0, + 1730.0, + 533.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 1723.0, + 661.0, + 1723.0, + 661.0, + 1738.0, + 646.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1719.0, + 730.0, + 1719.0, + 730.0, + 1739.0, + 667.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1718.0, + 751.0, + 1718.0, + 751.0, + 1730.0, + 740.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1725.0, + 931.0, + 1725.0, + 931.0, + 1735.0, + 919.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1719.0, + 999.0, + 1719.0, + 999.0, + 1739.0, + 938.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1713.0, + 1097.0, + 1713.0, + 1097.0, + 1765.0, + 1064.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 1722.0, + 1204.0, + 1722.0, + 1204.0, + 1738.0, + 1188.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 1720.0, + 1271.0, + 1720.0, + 1271.0, + 1738.0, + 1208.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1713.0, + 1321.0, + 1713.0, + 1321.0, + 1743.0, + 1299.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1722.0, + 1357.0, + 1722.0, + 1357.0, + 1737.0, + 1342.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1753.0, + 366.0, + 1753.0, + 366.0, + 1779.0, + 323.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1734.0, + 637.0, + 1734.0, + 637.0, + 1760.0, + 595.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 1737.0, + 735.0, + 1737.0, + 735.0, + 1767.0, + 712.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1754.0, + 1070.0, + 1754.0, + 1070.0, + 1783.0, + 1056.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 1734.0, + 1094.0, + 1734.0, + 1094.0, + 1752.0, + 1079.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1734.0, + 1178.0, + 1734.0, + 1178.0, + 1757.0, + 1137.0, + 1757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1742.0, + 1300.0, + 1742.0, + 1300.0, + 1775.0, + 1281.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1755.0, + 1337.0, + 1755.0, + 1337.0, + 1772.0, + 1322.0, + 1772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1733.0, + 1351.0, + 1733.0, + 1351.0, + 1745.0, + 1339.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1773.0, + 637.0, + 1773.0, + 637.0, + 1799.0, + 595.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1770.0, + 709.0, + 1770.0, + 709.0, + 1787.0, + 695.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 1768.0, + 1080.0, + 1768.0, + 1080.0, + 1780.0, + 1069.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1772.0, + 1176.0, + 1772.0, + 1176.0, + 1794.0, + 1137.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 1777.0, + 1278.0, + 1777.0, + 1278.0, + 1792.0, + 1262.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 1788.0, + 906.0, + 1788.0, + 906.0, + 1810.0, + 865.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1788.0, + 1267.0, + 1788.0, + 1267.0, + 1797.0, + 1257.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 1797.0, + 650.0, + 1797.0, + 650.0, + 1812.0, + 635.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1799.0, + 1044.0, + 1799.0, + 1044.0, + 1820.0, + 1030.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 1799.0, + 1061.0, + 1799.0, + 1061.0, + 1815.0, + 1047.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1800.0, + 1248.0, + 1800.0, + 1248.0, + 1810.0, + 1239.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1807.0, + 333.0, + 1807.0, + 333.0, + 1850.0, + 310.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1819.0, + 378.0, + 1819.0, + 378.0, + 1835.0, + 363.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 1818.0, + 456.0, + 1818.0, + 456.0, + 1835.0, + 414.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1804.0, + 644.0, + 1804.0, + 644.0, + 1863.0, + 572.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 1819.0, + 921.0, + 1819.0, + 921.0, + 1835.0, + 905.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1819.0, + 981.0, + 1819.0, + 981.0, + 1833.0, + 958.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 1822.0, + 995.0, + 1822.0, + 995.0, + 1832.0, + 985.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1809.0, + 1190.0, + 1809.0, + 1190.0, + 1830.0, + 1136.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 1818.0, + 1253.0, + 1818.0, + 1253.0, + 1835.0, + 1225.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1832.0, + 397.0, + 1832.0, + 397.0, + 1849.0, + 363.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1832.0, + 431.0, + 1832.0, + 431.0, + 1849.0, + 398.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1832.0, + 467.0, + 1832.0, + 467.0, + 1849.0, + 436.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 1832.0, + 502.0, + 1832.0, + 502.0, + 1849.0, + 471.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 1832.0, + 539.0, + 1832.0, + 539.0, + 1849.0, + 507.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1832.0, + 574.0, + 1832.0, + 574.0, + 1849.0, + 543.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 1832.0, + 669.0, + 1832.0, + 669.0, + 1849.0, + 635.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1832.0, + 701.0, + 1832.0, + 701.0, + 1849.0, + 670.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1832.0, + 736.0, + 1832.0, + 736.0, + 1847.0, + 704.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1827.0, + 808.0, + 1827.0, + 808.0, + 1852.0, + 740.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1829.0, + 847.0, + 1829.0, + 847.0, + 1850.0, + 812.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 1832.0, + 939.0, + 1832.0, + 939.0, + 1849.0, + 906.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1832.0, + 974.0, + 1832.0, + 974.0, + 1849.0, + 940.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 1832.0, + 1008.0, + 1832.0, + 1008.0, + 1850.0, + 977.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1832.0, + 1044.0, + 1832.0, + 1044.0, + 1849.0, + 1012.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.0, + 1832.0, + 1080.0, + 1832.0, + 1080.0, + 1849.0, + 1049.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1832.0, + 1116.0, + 1832.0, + 1116.0, + 1849.0, + 1084.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 1832.0, + 1209.0, + 1832.0, + 1209.0, + 1849.0, + 1176.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1832.0, + 1243.0, + 1832.0, + 1243.0, + 1849.0, + 1211.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 1832.0, + 1315.0, + 1832.0, + 1315.0, + 1849.0, + 1248.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 1832.0, + 1351.0, + 1832.0, + 1351.0, + 1849.0, + 1320.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 1832.0, + 1388.0, + 1832.0, + 1388.0, + 1849.0, + 1356.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1840.0, + 500.0, + 1840.0, + 500.0, + 1865.0, + 435.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1840.0, + 770.0, + 1840.0, + 770.0, + 1865.0, + 705.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1840.0, + 1041.0, + 1840.0, + 1041.0, + 1865.0, + 976.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 1840.0, + 1311.0, + 1840.0, + 1311.0, + 1865.0, + 1247.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.25, + 1204.0, + 685.25, + 1204.0, + 685.25, + 1217.5, + 674.25, + 1217.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1206.0, + 1204.0, + 1206.0, + 1204.0, + 1217.0, + 1200.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.25, + 1298.0, + 666.25, + 1298.0, + 666.25, + 1310.5, + 661.25, + 1310.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 1256.5, + 738.0, + 1256.5, + 738.0, + 1268.5, + 732.0, + 1268.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1311.5, + 1215.0, + 1311.5, + 1215.0, + 1324.5, + 1202.0, + 1324.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 1439.0, + 687.0, + 1439.0, + 687.0, + 1457.0, + 664.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1445.0, + 1204.0, + 1445.0, + 1204.0, + 1460.0, + 1197.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.25, + 1538.5, + 667.25, + 1538.5, + 667.25, + 1550.5, + 661.25, + 1550.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.75, + 1552.5, + 1211.75, + 1552.5, + 1211.75, + 1564.5, + 1202.75, + 1564.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.75, + 1672.5, + 808.75, + 1672.5, + 808.75, + 1689.5, + 757.75, + 1689.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1700.0, + 837.0, + 1700.0, + 837.0, + 1710.0, + 788.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 1688.0, + 1390.0, + 1688.0, + 1390.0, + 1700.0, + 1333.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1746.0, + 1082.0, + 1746.0, + 1082.0, + 1771.0, + 1078.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1775.5, + 1052.0, + 1775.5, + 1052.0, + 1821.5, + 1044.0, + 1821.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.75, + 1760.5, + 575.75, + 1760.5, + 575.75, + 1781.0, + 479.75, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.75, + 1818.5, + 710.75, + 1818.5, + 710.75, + 1832.0, + 684.75, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1062.0, + 1405.0, + 1062.0, + 1405.0, + 1093.0, + 296.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1090.0, + 1335.0, + 1090.0, + 1335.0, + 1123.0, + 293.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1885.0, + 1405.0, + 1885.0, + 1405.0, + 1916.0, + 295.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1914.0, + 1333.0, + 1914.0, + 1333.0, + 1945.0, + 295.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 294.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 1407.0, + 229.0, + 1407.0, + 266.0, + 295.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 261.0, + 1406.0, + 261.0, + 1406.0, + 295.0, + 295.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 292.0, + 1156.0, + 292.0, + 1156.0, + 324.0, + 292.0, + 324.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 22, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 307, + 217, + 1391, + 217, + 1391, + 920, + 307, + 920 + ], + "score": 0.975 + }, + { + "category_id": 4, + "poly": [ + 294, + 941, + 1401, + 941, + 1401, + 1000, + 294, + 1000 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.884 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.848 + }, + { + "category_id": 15, + "poly": [ + 308.0, + 222.0, + 367.0, + 222.0, + 367.0, + 340.0, + 308.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 221.0, + 527.0, + 221.0, + 527.0, + 265.0, + 406.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 197.0, + 657.0, + 197.0, + 657.0, + 419.0, + 556.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 220.0, + 797.0, + 220.0, + 797.0, + 265.0, + 678.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 223.0, + 909.0, + 223.0, + 909.0, + 336.0, + 851.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 222.0, + 1069.0, + 222.0, + 1069.0, + 266.0, + 948.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 221.0, + 1341.0, + 221.0, + 1341.0, + 292.0, + 1123.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 267.0, + 488.0, + 267.0, + 488.0, + 277.0, + 477.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 261.0, + 548.0, + 261.0, + 548.0, + 281.0, + 496.0, + 281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 262.0, + 715.0, + 262.0, + 715.0, + 274.0, + 703.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 271.0, + 975.0, + 271.0, + 975.0, + 284.0, + 964.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 271.0, + 1030.0, + 271.0, + 1030.0, + 282.0, + 1019.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 269.0, + 1246.0, + 269.0, + 1246.0, + 281.0, + 1235.0, + 281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 281.0, + 364.0, + 281.0, + 364.0, + 304.0, + 325.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 277.0, + 433.0, + 277.0, + 433.0, + 289.0, + 422.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 281.0, + 468.0, + 281.0, + 468.0, + 296.0, + 453.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 284.0, + 488.0, + 284.0, + 488.0, + 295.0, + 475.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 281.0, + 565.0, + 281.0, + 565.0, + 297.0, + 497.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 281.0, + 907.0, + 281.0, + 907.0, + 304.0, + 867.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 289.0, + 985.0, + 289.0, + 985.0, + 300.0, + 974.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 286.0, + 1256.0, + 286.0, + 1256.0, + 297.0, + 1245.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 295.0, + 442.0, + 295.0, + 442.0, + 306.0, + 432.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 301.0, + 488.0, + 301.0, + 488.0, + 311.0, + 477.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 296.0, + 558.0, + 296.0, + 558.0, + 316.0, + 496.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 294.0, + 742.0, + 294.0, + 742.0, + 305.0, + 729.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 315.0, + 474.0, + 315.0, + 474.0, + 325.0, + 464.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 325.0, + 1012.0, + 325.0, + 1012.0, + 335.0, + 1001.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 315.0, + 1048.0, + 315.0, + 1048.0, + 328.0, + 1036.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 319.0, + 1288.0, + 319.0, + 1288.0, + 335.0, + 1272.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 334.0, + 484.0, + 334.0, + 484.0, + 351.0, + 458.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 340.0, + 659.0, + 340.0, + 659.0, + 351.0, + 648.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 335.0, + 695.0, + 335.0, + 695.0, + 354.0, + 669.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 340.0, + 930.0, + 340.0, + 930.0, + 351.0, + 919.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 335.0, + 966.0, + 335.0, + 966.0, + 354.0, + 940.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 340.0, + 1202.0, + 340.0, + 1202.0, + 351.0, + 1189.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 335.0, + 1237.0, + 335.0, + 1237.0, + 354.0, + 1212.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 356.0, + 659.0, + 356.0, + 659.0, + 368.0, + 649.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 350.0, + 721.0, + 350.0, + 721.0, + 370.0, + 669.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 356.0, + 930.0, + 356.0, + 930.0, + 368.0, + 919.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 350.0, + 992.0, + 350.0, + 992.0, + 372.0, + 939.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 355.0, + 1202.0, + 355.0, + 1202.0, + 368.0, + 1191.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 350.0, + 1262.0, + 350.0, + 1262.0, + 372.0, + 1210.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 374.0, + 475.0, + 374.0, + 475.0, + 383.0, + 464.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 373.0, + 659.0, + 373.0, + 659.0, + 384.0, + 649.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 368.0, + 739.0, + 368.0, + 739.0, + 388.0, + 669.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 373.0, + 930.0, + 373.0, + 930.0, + 384.0, + 918.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 368.0, + 1019.0, + 368.0, + 1019.0, + 388.0, + 939.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 373.0, + 1202.0, + 373.0, + 1202.0, + 384.0, + 1191.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 366.0, + 1281.0, + 366.0, + 1281.0, + 387.0, + 1210.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 387.0, + 604.0, + 387.0, + 604.0, + 429.0, + 581.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 390.0, + 659.0, + 390.0, + 659.0, + 402.0, + 648.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 385.0, + 729.0, + 385.0, + 729.0, + 405.0, + 668.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 379.0, + 834.0, + 379.0, + 834.0, + 408.0, + 807.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 390.0, + 930.0, + 390.0, + 930.0, + 402.0, + 921.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 385.0, + 1001.0, + 385.0, + 1001.0, + 405.0, + 939.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1062.0, + 384.0, + 1080.0, + 384.0, + 1080.0, + 398.0, + 1062.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 390.0, + 1202.0, + 390.0, + 1202.0, + 402.0, + 1191.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 385.0, + 1271.0, + 385.0, + 1271.0, + 405.0, + 1209.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 387.0, + 1382.0, + 387.0, + 1382.0, + 409.0, + 1350.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 408.0, + 384.0, + 408.0, + 384.0, + 425.0, + 354.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 409.0, + 427.0, + 409.0, + 427.0, + 425.0, + 395.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 408.0, + 469.0, + 408.0, + 469.0, + 425.0, + 437.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 409.0, + 511.0, + 409.0, + 511.0, + 425.0, + 479.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 410.0, + 551.0, + 410.0, + 551.0, + 424.0, + 523.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 408.0, + 654.0, + 408.0, + 654.0, + 425.0, + 626.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 408.0, + 697.0, + 408.0, + 697.0, + 427.0, + 666.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 407.0, + 739.0, + 407.0, + 739.0, + 425.0, + 704.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 410.0, + 780.0, + 410.0, + 780.0, + 424.0, + 752.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 409.0, + 823.0, + 409.0, + 823.0, + 427.0, + 792.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 408.0, + 925.0, + 408.0, + 925.0, + 425.0, + 896.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 407.0, + 970.0, + 407.0, + 970.0, + 428.0, + 935.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 407.0, + 1012.0, + 407.0, + 1012.0, + 425.0, + 974.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 409.0, + 1053.0, + 409.0, + 1053.0, + 425.0, + 1020.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 410.0, + 1092.0, + 410.0, + 1092.0, + 424.0, + 1065.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 408.0, + 1197.0, + 408.0, + 1197.0, + 425.0, + 1167.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 408.0, + 1240.0, + 408.0, + 1240.0, + 425.0, + 1208.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 408.0, + 1282.0, + 408.0, + 1282.0, + 425.0, + 1249.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 409.0, + 1324.0, + 409.0, + 1324.0, + 425.0, + 1291.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 410.0, + 1363.0, + 410.0, + 1363.0, + 424.0, + 1336.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 416.0, + 503.0, + 416.0, + 503.0, + 441.0, + 431.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 419.0, + 774.0, + 419.0, + 774.0, + 441.0, + 702.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 419.0, + 1044.0, + 419.0, + 1044.0, + 441.0, + 974.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 419.0, + 1315.0, + 419.0, + 1315.0, + 441.0, + 1245.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 463.0, + 547.0, + 463.0, + 547.0, + 579.0, + 310.0, + 579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 461.0, + 652.0, + 461.0, + 652.0, + 656.0, + 560.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 461.0, + 797.0, + 461.0, + 797.0, + 515.0, + 678.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 462.0, + 1071.0, + 462.0, + 1071.0, + 523.0, + 866.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 461.0, + 1342.0, + 461.0, + 1342.0, + 532.0, + 1121.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 522.0, + 364.0, + 522.0, + 364.0, + 545.0, + 325.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 517.0, + 422.0, + 517.0, + 422.0, + 530.0, + 412.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 522.0, + 467.0, + 522.0, + 467.0, + 535.0, + 457.0, + 535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 525.0, + 486.0, + 525.0, + 486.0, + 535.0, + 477.0, + 535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 521.0, + 567.0, + 521.0, + 567.0, + 537.0, + 497.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 522.0, + 906.0, + 522.0, + 906.0, + 545.0, + 867.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 523.0, + 1241.0, + 523.0, + 1241.0, + 538.0, + 1225.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 530.0, + 967.0, + 530.0, + 967.0, + 540.0, + 956.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 536.0, + 426.0, + 536.0, + 426.0, + 547.0, + 415.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 541.0, + 486.0, + 541.0, + 486.0, + 552.0, + 477.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 536.0, + 558.0, + 536.0, + 558.0, + 556.0, + 496.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 535.0, + 728.0, + 535.0, + 728.0, + 545.0, + 718.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 550.0, + 377.0, + 550.0, + 377.0, + 646.0, + 295.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 553.0, + 470.0, + 553.0, + 470.0, + 566.0, + 459.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 557.0, + 733.0, + 557.0, + 733.0, + 569.0, + 724.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 565.0, + 999.0, + 565.0, + 999.0, + 576.0, + 990.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 556.0, + 1039.0, + 556.0, + 1039.0, + 567.0, + 1029.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 558.0, + 1276.0, + 558.0, + 1276.0, + 575.0, + 1260.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 576.0, + 480.0, + 576.0, + 480.0, + 589.0, + 469.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 570.0, + 636.0, + 570.0, + 636.0, + 592.0, + 594.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 580.0, + 659.0, + 580.0, + 659.0, + 591.0, + 648.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 576.0, + 695.0, + 576.0, + 695.0, + 595.0, + 670.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 580.0, + 930.0, + 580.0, + 930.0, + 591.0, + 919.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 575.0, + 966.0, + 575.0, + 966.0, + 595.0, + 940.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 580.0, + 1202.0, + 580.0, + 1202.0, + 591.0, + 1189.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 576.0, + 1237.0, + 576.0, + 1237.0, + 595.0, + 1212.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 596.0, + 659.0, + 596.0, + 659.0, + 607.0, + 649.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 590.0, + 721.0, + 590.0, + 721.0, + 611.0, + 669.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 596.0, + 930.0, + 596.0, + 930.0, + 607.0, + 919.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 590.0, + 992.0, + 590.0, + 992.0, + 611.0, + 940.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 596.0, + 1202.0, + 596.0, + 1202.0, + 607.0, + 1191.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 590.0, + 1263.0, + 590.0, + 1263.0, + 611.0, + 1210.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 611.0, + 662.0, + 611.0, + 662.0, + 626.0, + 647.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 609.0, + 739.0, + 609.0, + 739.0, + 629.0, + 669.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 612.0, + 930.0, + 612.0, + 930.0, + 624.0, + 918.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 609.0, + 1009.0, + 609.0, + 1009.0, + 629.0, + 940.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 606.0, + 1076.0, + 606.0, + 1076.0, + 624.0, + 1046.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 611.0, + 1203.0, + 611.0, + 1203.0, + 626.0, + 1188.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 609.0, + 1281.0, + 609.0, + 1281.0, + 629.0, + 1210.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 622.0, + 332.0, + 622.0, + 332.0, + 666.0, + 310.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 629.0, + 604.0, + 629.0, + 604.0, + 670.0, + 581.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 630.0, + 659.0, + 630.0, + 659.0, + 641.0, + 648.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 625.0, + 729.0, + 625.0, + 729.0, + 645.0, + 669.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 617.0, + 843.0, + 617.0, + 843.0, + 653.0, + 812.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 625.0, + 874.0, + 625.0, + 874.0, + 665.0, + 851.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 630.0, + 930.0, + 630.0, + 930.0, + 641.0, + 919.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 625.0, + 1001.0, + 625.0, + 1001.0, + 645.0, + 940.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 617.0, + 1099.0, + 617.0, + 1099.0, + 626.0, + 1088.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 617.0, + 1180.0, + 617.0, + 1180.0, + 644.0, + 1136.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 630.0, + 1202.0, + 630.0, + 1202.0, + 641.0, + 1191.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 625.0, + 1272.0, + 625.0, + 1272.0, + 645.0, + 1210.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 624.0, + 1384.0, + 624.0, + 1384.0, + 652.0, + 1354.0, + 652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 635.0, + 567.0, + 635.0, + 567.0, + 649.0, + 544.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 648.0, + 421.0, + 648.0, + 421.0, + 666.0, + 390.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 648.0, + 470.0, + 648.0, + 470.0, + 666.0, + 440.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 648.0, + 521.0, + 648.0, + 521.0, + 666.0, + 490.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 648.0, + 570.0, + 648.0, + 570.0, + 666.0, + 541.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 645.0, + 694.0, + 645.0, + 694.0, + 668.0, + 659.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 645.0, + 743.0, + 645.0, + 743.0, + 668.0, + 707.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 649.0, + 792.0, + 649.0, + 792.0, + 666.0, + 761.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 648.0, + 842.0, + 648.0, + 842.0, + 665.0, + 812.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 646.0, + 965.0, + 646.0, + 965.0, + 668.0, + 929.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 645.0, + 1013.0, + 645.0, + 1013.0, + 668.0, + 980.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 649.0, + 1064.0, + 649.0, + 1064.0, + 666.0, + 1032.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 648.0, + 1113.0, + 648.0, + 1113.0, + 666.0, + 1083.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 648.0, + 1233.0, + 648.0, + 1233.0, + 666.0, + 1202.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 645.0, + 1284.0, + 645.0, + 1284.0, + 668.0, + 1250.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 648.0, + 1334.0, + 648.0, + 1334.0, + 666.0, + 1303.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 648.0, + 1383.0, + 648.0, + 1383.0, + 666.0, + 1353.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 659.0, + 500.0, + 659.0, + 500.0, + 680.0, + 433.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 659.0, + 770.0, + 659.0, + 770.0, + 680.0, + 705.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 659.0, + 1040.0, + 659.0, + 1040.0, + 680.0, + 975.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 659.0, + 1313.0, + 659.0, + 1313.0, + 680.0, + 1246.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 683.0, + 387.0, + 683.0, + 387.0, + 896.0, + 289.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 703.0, + 523.0, + 703.0, + 523.0, + 723.0, + 410.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 701.0, + 796.0, + 701.0, + 796.0, + 725.0, + 679.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 703.0, + 1070.0, + 703.0, + 1070.0, + 723.0, + 948.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 693.0, + 1187.0, + 693.0, + 1187.0, + 750.0, + 1114.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 703.0, + 1339.0, + 703.0, + 1339.0, + 723.0, + 1220.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 718.0, + 367.0, + 718.0, + 367.0, + 784.0, + 324.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 732.0, + 391.0, + 732.0, + 391.0, + 794.0, + 375.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 727.0, + 467.0, + 727.0, + 467.0, + 797.0, + 398.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 758.0, + 546.0, + 758.0, + 546.0, + 768.0, + 536.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 744.0, + 560.0, + 744.0, + 560.0, + 773.0, + 548.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 733.0, + 708.0, + 733.0, + 708.0, + 763.0, + 658.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 727.0, + 728.0, + 727.0, + 728.0, + 743.0, + 713.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 725.0, + 748.0, + 725.0, + 748.0, + 738.0, + 737.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 722.0, + 780.0, + 722.0, + 780.0, + 786.0, + 755.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 720.0, + 840.0, + 720.0, + 840.0, + 737.0, + 801.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 716.0, + 909.0, + 716.0, + 909.0, + 784.0, + 863.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 732.0, + 930.0, + 732.0, + 930.0, + 792.0, + 919.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 727.0, + 1011.0, + 727.0, + 1011.0, + 797.0, + 939.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 745.0, + 1091.0, + 745.0, + 1091.0, + 779.0, + 1072.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 732.0, + 1145.0, + 732.0, + 1145.0, + 772.0, + 1124.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 767.0, + 1252.0, + 767.0, + 1252.0, + 789.0, + 1225.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 755.0, + 1267.0, + 755.0, + 1267.0, + 768.0, + 1256.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 742.0, + 1292.0, + 742.0, + 1292.0, + 758.0, + 1277.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 737.0, + 1310.0, + 737.0, + 1310.0, + 749.0, + 1299.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 720.0, + 1383.0, + 720.0, + 1383.0, + 777.0, + 1326.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 794.0, + 645.0, + 794.0, + 645.0, + 890.0, + 557.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 798.0, + 782.0, + 798.0, + 782.0, + 808.0, + 773.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 796.0, + 1191.0, + 796.0, + 1191.0, + 893.0, + 1109.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 786.0, + 1224.0, + 786.0, + 1224.0, + 804.0, + 1201.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1308.0, + 797.0, + 1328.0, + 797.0, + 1328.0, + 817.0, + 1308.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 803.0, + 1053.0, + 803.0, + 1053.0, + 818.0, + 1038.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 816.0, + 509.0, + 816.0, + 509.0, + 828.0, + 497.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 817.0, + 558.0, + 817.0, + 558.0, + 828.0, + 547.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 809.0, + 634.0, + 809.0, + 634.0, + 832.0, + 595.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 819.0, + 758.0, + 819.0, + 758.0, + 832.0, + 747.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 816.0, + 795.0, + 816.0, + 795.0, + 834.0, + 768.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 821.0, + 1300.0, + 821.0, + 1300.0, + 832.0, + 1289.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 816.0, + 1336.0, + 816.0, + 1336.0, + 834.0, + 1309.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 832.0, + 542.0, + 832.0, + 542.0, + 842.0, + 532.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 831.0, + 819.0, + 831.0, + 819.0, + 851.0, + 766.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 831.0, + 1361.0, + 831.0, + 1361.0, + 852.0, + 1309.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 837.0, + 758.0, + 837.0, + 758.0, + 847.0, + 748.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 836.0, + 1299.0, + 836.0, + 1299.0, + 847.0, + 1289.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 852.0, + 758.0, + 852.0, + 758.0, + 863.0, + 748.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 847.0, + 837.0, + 847.0, + 837.0, + 867.0, + 766.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 848.0, + 980.0, + 848.0, + 980.0, + 868.0, + 905.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 850.0, + 1001.0, + 850.0, + 1001.0, + 863.0, + 985.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 853.0, + 1055.0, + 853.0, + 1055.0, + 865.0, + 1043.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 852.0, + 1300.0, + 852.0, + 1300.0, + 865.0, + 1289.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 848.0, + 1379.0, + 848.0, + 1379.0, + 868.0, + 1309.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 865.0, + 332.0, + 865.0, + 332.0, + 906.0, + 310.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 872.0, + 393.0, + 872.0, + 393.0, + 890.0, + 364.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 865.0, + 458.0, + 865.0, + 458.0, + 878.0, + 441.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 860.0, + 497.0, + 860.0, + 497.0, + 871.0, + 486.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 871.0, + 604.0, + 871.0, + 604.0, + 911.0, + 581.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 870.0, + 758.0, + 870.0, + 758.0, + 882.0, + 747.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 865.0, + 828.0, + 865.0, + 828.0, + 885.0, + 766.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 866.0, + 874.0, + 866.0, + 874.0, + 906.0, + 851.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 860.0, + 1176.0, + 860.0, + 1176.0, + 912.0, + 1123.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 868.0, + 1302.0, + 868.0, + 1302.0, + 883.0, + 1287.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 867.0, + 1369.0, + 867.0, + 1369.0, + 883.0, + 1310.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 888.0, + 378.0, + 888.0, + 378.0, + 906.0, + 348.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 888.0, + 416.0, + 888.0, + 416.0, + 906.0, + 385.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 888.0, + 456.0, + 888.0, + 456.0, + 906.0, + 425.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 888.0, + 496.0, + 888.0, + 496.0, + 906.0, + 464.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 888.0, + 535.0, + 888.0, + 535.0, + 906.0, + 502.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 886.0, + 574.0, + 886.0, + 574.0, + 907.0, + 539.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 888.0, + 649.0, + 888.0, + 649.0, + 906.0, + 618.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 888.0, + 687.0, + 888.0, + 687.0, + 906.0, + 657.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 888.0, + 728.0, + 888.0, + 728.0, + 906.0, + 696.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 888.0, + 766.0, + 888.0, + 766.0, + 906.0, + 734.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 888.0, + 806.0, + 888.0, + 806.0, + 906.0, + 774.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 887.0, + 844.0, + 887.0, + 844.0, + 906.0, + 813.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 888.0, + 919.0, + 888.0, + 919.0, + 906.0, + 890.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 888.0, + 959.0, + 888.0, + 959.0, + 906.0, + 928.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 888.0, + 998.0, + 888.0, + 998.0, + 906.0, + 966.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 888.0, + 1036.0, + 888.0, + 1036.0, + 906.0, + 1006.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 888.0, + 1076.0, + 888.0, + 1076.0, + 906.0, + 1045.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 888.0, + 1115.0, + 888.0, + 1115.0, + 906.0, + 1083.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 888.0, + 1192.0, + 888.0, + 1192.0, + 906.0, + 1160.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 888.0, + 1230.0, + 888.0, + 1230.0, + 906.0, + 1198.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 888.0, + 1270.0, + 888.0, + 1270.0, + 906.0, + 1237.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 888.0, + 1307.0, + 888.0, + 1307.0, + 906.0, + 1276.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 888.0, + 1349.0, + 888.0, + 1349.0, + 906.0, + 1315.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 888.0, + 1385.0, + 888.0, + 1385.0, + 906.0, + 1355.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 897.0, + 498.0, + 897.0, + 498.0, + 921.0, + 434.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 897.0, + 769.0, + 897.0, + 769.0, + 921.0, + 705.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 897.0, + 1039.0, + 897.0, + 1039.0, + 921.0, + 977.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 897.0, + 1311.0, + 897.0, + 1311.0, + 921.0, + 1247.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 232.5, + 455.0, + 232.5, + 455.0, + 262.5, + 358.0, + 262.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.25, + 233.0, + 1019.25, + 233.0, + 1019.25, + 261.5, + 899.25, + 261.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1327.0, + 264.0, + 1338.0, + 264.0, + 1338.0, + 278.5, + 1327.0, + 278.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1358.25, + 298.5, + 1362.25, + 298.5, + 1362.25, + 315.5, + 1358.25, + 315.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 325.0, + 801.0, + 325.0, + 801.0, + 359.0, + 746.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 354.5, + 1344.0, + 354.5, + 1344.0, + 374.5, + 1290.0, + 374.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 367.5, + 542.0, + 367.5, + 542.0, + 393.0, + 477.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 343.5, + 1033.0, + 343.5, + 1033.0, + 358.5, + 1018.0, + 358.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.25, + 349.0, + 1064.25, + 349.0, + 1064.25, + 363.0, + 1050.25, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 365.5, + 1048.0, + 365.5, + 1048.0, + 375.0, + 1039.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.25, + 367.0, + 1110.25, + 367.0, + 1110.25, + 388.5, + 1061.25, + 388.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 536.5, + 1360.0, + 536.5, + 1360.0, + 563.0, + 1335.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.25, + 572.0, + 798.25, + 572.0, + 798.25, + 595.0, + 737.25, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.25, + 593.5, + 1342.25, + 593.5, + 1342.25, + 617.5, + 1278.25, + 617.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 591.0, + 1052.0, + 591.0, + 1052.0, + 604.0, + 1048.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 602.5, + 1038.0, + 602.5, + 1038.0, + 611.0, + 1024.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.25, + 619.0, + 1071.25, + 619.0, + 1071.25, + 636.0, + 1049.25, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 634.0, + 1106.0, + 634.0, + 1106.0, + 648.0, + 1088.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.75, + 716.5, + 1095.75, + 716.5, + 1095.75, + 750.5, + 1092.75, + 750.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 730.0, + 814.0, + 730.0, + 814.0, + 756.5, + 805.0, + 756.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 723.0, + 569.0, + 723.0, + 569.0, + 737.0, + 535.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.75, + 756.0, + 664.75, + 756.0, + 664.75, + 768.0, + 630.75, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 774.0, + 554.0, + 774.0, + 554.0, + 793.0, + 530.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 795.0, + 1207.0, + 795.0, + 1207.0, + 813.5, + 1172.0, + 813.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 798.0, + 768.0, + 798.0, + 768.0, + 806.0, + 740.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.75, + 850.5, + 476.75, + 850.5, + 476.75, + 877.0, + 452.75, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 829.0, + 755.0, + 829.0, + 755.0, + 848.5, + 690.0, + 848.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 829.5, + 1051.0, + 829.5, + 1051.0, + 854.5, + 980.0, + 854.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 842.0, + 499.0, + 842.0, + 499.0, + 854.0, + 459.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.75, + 859.0, + 1261.75, + 859.0, + 1261.75, + 867.5, + 1240.75, + 867.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.75, + 864.0, + 713.75, + 864.0, + 713.75, + 889.0, + 641.75, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.25, + 874.0, + 975.25, + 874.0, + 975.25, + 890.5, + 916.25, + 890.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 868.5, + 1250.0, + 868.5, + 1250.0, + 889.5, + 1185.0, + 889.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 939.0, + 1407.0, + 939.0, + 1407.0, + 974.0, + 294.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 969.0, + 1319.0, + 969.0, + 1319.0, + 1002.0, + 293.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 23, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 312, + 1406, + 312, + 1406, + 862, + 297, + 862 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 1488, + 1403, + 1488, + 1403, + 1672, + 297, + 1672 + ], + "score": 0.977 + }, + { + "category_id": 4, + "poly": [ + 296, + 1252, + 1405, + 1252, + 1405, + 1446, + 296, + 1446 + ], + "score": 0.963 + }, + { + "category_id": 3, + "poly": [ + 295, + 911, + 1406, + 911, + 1406, + 1229, + 295, + 1229 + ], + "score": 0.961 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.848 + }, + { + "category_id": 4, + "poly": [ + 494, + 885, + 1231, + 885, + 1231, + 909, + 494, + 909 + ], + "score": 0.826 + }, + { + "category_id": 1, + "poly": [ + 294, + 230, + 1369, + 230, + 1369, + 292, + 294, + 292 + ], + "score": 0.702 + }, + { + "category_id": 0, + "poly": [ + 294, + 230, + 1369, + 230, + 1369, + 292, + 294, + 292 + ], + "score": 0.232 + }, + { + "category_id": 13, + "poly": [ + 461, + 1337, + 529, + 1337, + 529, + 1362, + 461, + 1362 + ], + "score": 0.9, + "latex": "s = 4" + }, + { + "category_id": 13, + "poly": [ + 876, + 528, + 952, + 528, + 952, + 556, + 876, + 556 + ], + "score": 0.89, + "latex": "s = 4" + }, + { + "category_id": 13, + "poly": [ + 1213, + 1550, + 1233, + 1550, + 1233, + 1580, + 1213, + 1580 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 547, + 1550, + 568, + 1550, + 568, + 1580, + 547, + 1580 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1043, + 916, + 1100, + 916, + 1100, + 938, + 1043, + 938 + ], + "score": 0.38, + "latex": "{ 5 } \\mathrm { = } 4 . 0" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1252.0, + 1405.0, + 1252.0, + 1405.0, + 1285.0, + 295.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1279.0, + 1406.0, + 1279.0, + 1406.0, + 1312.0, + 295.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1307.0, + 1403.0, + 1307.0, + 1403.0, + 1340.0, + 295.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1336.0, + 460.0, + 1336.0, + 460.0, + 1366.0, + 293.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 1336.0, + 1405.0, + 1336.0, + 1405.0, + 1366.0, + 530.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1363.0, + 1402.0, + 1363.0, + 1402.0, + 1392.0, + 295.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1388.0, + 1405.0, + 1388.0, + 1405.0, + 1425.0, + 294.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1417.0, + 469.0, + 1417.0, + 469.0, + 1451.0, + 295.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 913.0, + 525.0, + 913.0, + 525.0, + 942.0, + 405.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 913.0, + 803.0, + 913.0, + 803.0, + 942.0, + 681.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 911.0, + 1042.0, + 911.0, + 1042.0, + 943.0, + 934.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 911.0, + 1104.0, + 911.0, + 1104.0, + 943.0, + 1101.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 912.0, + 1383.0, + 912.0, + 1383.0, + 939.0, + 1211.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 945.0, + 383.0, + 945.0, + 383.0, + 1131.0, + 278.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 947.0, + 661.0, + 947.0, + 661.0, + 1138.0, + 551.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 950.0, + 914.0, + 950.0, + 914.0, + 977.0, + 866.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 956.0, + 1192.0, + 956.0, + 1192.0, + 984.0, + 1142.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1020.0, + 360.0, + 1020.0, + 360.0, + 1050.0, + 311.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1019.0, + 639.0, + 1019.0, + 639.0, + 1050.0, + 587.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 980.0, + 912.0, + 980.0, + 912.0, + 1102.0, + 845.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 977.0, + 1191.0, + 977.0, + 1191.0, + 1110.0, + 1125.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1057.0, + 360.0, + 1057.0, + 360.0, + 1087.0, + 311.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1056.0, + 637.0, + 1056.0, + 637.0, + 1088.0, + 588.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 1074.0, + 912.0, + 1074.0, + 912.0, + 1100.0, + 864.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1071.0, + 1191.0, + 1071.0, + 1191.0, + 1099.0, + 1142.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1095.0, + 359.0, + 1095.0, + 359.0, + 1122.0, + 312.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1095.0, + 637.0, + 1095.0, + 637.0, + 1122.0, + 588.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 1131.0, + 357.0, + 1131.0, + 357.0, + 1159.0, + 311.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1131.0, + 636.0, + 1131.0, + 636.0, + 1159.0, + 588.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1129.0, + 1191.0, + 1129.0, + 1191.0, + 1156.0, + 1142.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1151.0, + 379.0, + 1151.0, + 379.0, + 1169.0, + 362.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1150.0, + 546.0, + 1150.0, + 546.0, + 1171.0, + 398.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 1151.0, + 656.0, + 1151.0, + 656.0, + 1169.0, + 640.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1150.0, + 823.0, + 1150.0, + 823.0, + 1171.0, + 674.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1151.0, + 934.0, + 1151.0, + 934.0, + 1169.0, + 917.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 1150.0, + 1100.0, + 1150.0, + 1100.0, + 1171.0, + 951.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1151.0, + 1210.0, + 1151.0, + 1210.0, + 1169.0, + 1194.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 1150.0, + 1379.0, + 1150.0, + 1379.0, + 1171.0, + 1230.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1165.0, + 557.0, + 1165.0, + 557.0, + 1193.0, + 371.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 1165.0, + 834.0, + 1165.0, + 834.0, + 1193.0, + 648.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1168.0, + 1112.0, + 1168.0, + 1112.0, + 1193.0, + 926.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 1168.0, + 1389.0, + 1168.0, + 1389.0, + 1193.0, + 1204.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1202.0, + 748.0, + 1202.0, + 748.0, + 1227.0, + 598.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1199.0, + 874.0, + 1199.0, + 874.0, + 1229.0, + 808.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1199.0, + 997.0, + 1199.0, + 997.0, + 1229.0, + 930.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1198.0, + 1202.0, + 1198.0, + 1202.0, + 1227.0, + 1054.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.25, + 991.0, + 543.25, + 991.0, + 543.25, + 1020.0, + 443.25, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 882.0, + 1233.0, + 882.0, + 1233.0, + 913.0, + 498.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 1374.0, + 228.0, + 1374.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 258.0, + 538.0, + 258.0, + 538.0, + 295.0, + 390.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 315.0, + 1404.0, + 315.0, + 1404.0, + 347.0, + 295.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 345.0, + 1407.0, + 345.0, + 1407.0, + 380.0, + 294.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 375.0, + 1406.0, + 375.0, + 1406.0, + 411.0, + 294.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 407.0, + 1404.0, + 407.0, + 1404.0, + 438.0, + 296.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 436.0, + 1406.0, + 436.0, + 1406.0, + 471.0, + 294.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 469.0, + 1404.0, + 469.0, + 1404.0, + 501.0, + 296.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 496.0, + 1404.0, + 496.0, + 1404.0, + 532.0, + 292.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 527.0, + 875.0, + 527.0, + 875.0, + 563.0, + 294.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 527.0, + 1406.0, + 527.0, + 1406.0, + 563.0, + 953.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 557.0, + 1406.0, + 557.0, + 1406.0, + 592.0, + 294.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 591.0, + 1404.0, + 591.0, + 1404.0, + 623.0, + 295.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 619.0, + 1406.0, + 619.0, + 1406.0, + 653.0, + 294.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 649.0, + 1404.0, + 649.0, + 1404.0, + 685.0, + 295.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 681.0, + 1403.0, + 681.0, + 1403.0, + 713.0, + 296.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 711.0, + 1406.0, + 711.0, + 1406.0, + 744.0, + 294.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 739.0, + 1404.0, + 739.0, + 1404.0, + 775.0, + 295.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 772.0, + 1403.0, + 772.0, + 1403.0, + 804.0, + 296.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 800.0, + 1404.0, + 800.0, + 1404.0, + 838.0, + 292.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 833.0, + 548.0, + 833.0, + 548.0, + 865.0, + 296.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1489.0, + 1404.0, + 1489.0, + 1404.0, + 1520.0, + 296.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1517.0, + 1405.0, + 1517.0, + 1405.0, + 1553.0, + 294.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1544.0, + 546.0, + 1544.0, + 546.0, + 1586.0, + 291.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 1544.0, + 1212.0, + 1544.0, + 1212.0, + 1586.0, + 569.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 1544.0, + 1407.0, + 1544.0, + 1407.0, + 1586.0, + 1234.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1579.0, + 1405.0, + 1579.0, + 1405.0, + 1614.0, + 295.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1609.0, + 1405.0, + 1609.0, + 1405.0, + 1645.0, + 295.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1642.0, + 953.0, + 1642.0, + 953.0, + 1674.0, + 296.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 1374.0, + 228.0, + 1374.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 258.0, + 538.0, + 258.0, + 538.0, + 295.0, + 390.0, + 295.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 24, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 295, + 245, + 1404, + 245, + 1404, + 575, + 295, + 575 + ], + "score": 0.93 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.886 + }, + { + "category_id": 4, + "poly": [ + 295, + 597, + 1406, + 597, + 1406, + 794, + 295, + 794 + ], + "score": 0.886 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.846 + }, + { + "category_id": 4, + "poly": [ + 295, + 598, + 1406, + 598, + 1406, + 794, + 295, + 794 + ], + "score": 0.708 + }, + { + "category_id": 2, + "poly": [ + 474, + 222, + 1242, + 222, + 1242, + 246, + 474, + 246 + ], + "score": 0.283 + }, + { + "category_id": 4, + "poly": [ + 474, + 222, + 1242, + 222, + 1242, + 246, + 474, + 246 + ], + "score": 0.203 + }, + { + "category_id": 13, + "poly": [ + 461, + 683, + 527, + 683, + 527, + 707, + 461, + 707 + ], + "score": 0.89, + "latex": "s = 4" + }, + { + "category_id": 13, + "poly": [ + 1038, + 253, + 1095, + 253, + 1095, + 275, + 1038, + 275 + ], + "score": 0.3, + "latex": "s { = } 4 . 0" + }, + { + "category_id": 13, + "poly": [ + 1318, + 253, + 1374, + 253, + 1374, + 276, + 1318, + 276 + ], + "score": 0.25, + "latex": "s { = } 4 . 0" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 251.0, + 518.0, + 251.0, + 518.0, + 279.0, + 399.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 249.0, + 799.0, + 249.0, + 799.0, + 281.0, + 677.0, + 281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 251.0, + 1037.0, + 251.0, + 1037.0, + 275.0, + 930.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 251.0, + 1100.0, + 251.0, + 1100.0, + 275.0, + 1096.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 251.0, + 1317.0, + 251.0, + 1317.0, + 277.0, + 1208.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 251.0, + 1378.0, + 251.0, + 1378.0, + 277.0, + 1375.0, + 277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 270.0, + 366.0, + 270.0, + 366.0, + 295.0, + 312.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 267.0, + 635.0, + 267.0, + 635.0, + 293.0, + 580.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 287.0, + 915.0, + 287.0, + 915.0, + 312.0, + 867.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 320.0, + 322.0, + 320.0, + 322.0, + 439.0, + 293.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 307.0, + 635.0, + 307.0, + 635.0, + 448.0, + 562.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 323.0, + 658.0, + 323.0, + 658.0, + 340.0, + 644.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 318.0, + 878.0, + 318.0, + 878.0, + 441.0, + 849.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 300.0, + 1192.0, + 300.0, + 1192.0, + 445.0, + 1129.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 335.0, + 357.0, + 335.0, + 357.0, + 361.0, + 310.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 348.0, + 635.0, + 348.0, + 635.0, + 373.0, + 580.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 334.0, + 915.0, + 334.0, + 915.0, + 362.0, + 866.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 344.0, + 1192.0, + 344.0, + 1192.0, + 370.0, + 1145.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1203.0, + 378.0, + 1215.0, + 378.0, + 1215.0, + 388.0, + 1203.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 401.0, + 360.0, + 401.0, + 360.0, + 430.0, + 310.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 397.0, + 433.0, + 397.0, + 433.0, + 409.0, + 411.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 384.0, + 637.0, + 384.0, + 637.0, + 415.0, + 575.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 383.0, + 915.0, + 383.0, + 915.0, + 409.0, + 867.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 389.0, + 1192.0, + 389.0, + 1192.0, + 416.0, + 1145.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 426.0, + 637.0, + 426.0, + 637.0, + 455.0, + 579.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 432.0, + 914.0, + 432.0, + 914.0, + 458.0, + 867.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 438.0, + 1193.0, + 438.0, + 1193.0, + 463.0, + 1145.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 470.0, + 356.0, + 470.0, + 356.0, + 492.0, + 312.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 469.0, + 635.0, + 469.0, + 635.0, + 493.0, + 579.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 486.0, + 380.0, + 486.0, + 380.0, + 507.0, + 361.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 486.0, + 570.0, + 486.0, + 570.0, + 505.0, + 394.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 487.0, + 657.0, + 487.0, + 657.0, + 504.0, + 640.0, + 504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 486.0, + 848.0, + 486.0, + 848.0, + 505.0, + 671.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 485.0, + 935.0, + 485.0, + 935.0, + 507.0, + 916.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 483.0, + 1040.0, + 483.0, + 1040.0, + 508.0, + 989.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 483.0, + 1128.0, + 483.0, + 1128.0, + 508.0, + 1077.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 486.0, + 1215.0, + 486.0, + 1215.0, + 507.0, + 1195.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 483.0, + 1318.0, + 483.0, + 1318.0, + 508.0, + 1268.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 483.0, + 1406.0, + 483.0, + 1406.0, + 508.0, + 1356.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 504.0, + 551.0, + 504.0, + 551.0, + 527.0, + 366.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 502.0, + 829.0, + 502.0, + 829.0, + 527.0, + 643.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 504.0, + 1106.0, + 504.0, + 1106.0, + 527.0, + 923.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 504.0, + 1385.0, + 504.0, + 1385.0, + 527.0, + 1201.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 526.0, + 721.0, + 526.0, + 721.0, + 549.0, + 571.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 525.0, + 845.0, + 525.0, + 845.0, + 550.0, + 776.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 524.0, + 968.0, + 524.0, + 968.0, + 552.0, + 901.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 526.0, + 1088.0, + 526.0, + 1088.0, + 549.0, + 1022.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 544.0, + 637.0, + 544.0, + 637.0, + 574.0, + 570.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 546.0, + 846.0, + 546.0, + 846.0, + 572.0, + 776.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 546.0, + 968.0, + 546.0, + 968.0, + 572.0, + 899.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 546.0, + 1171.0, + 546.0, + 1171.0, + 572.0, + 1023.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 818.0, + 72.0, + 818.0, + 110.0, + 295.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 597.0, + 1406.0, + 597.0, + 1406.0, + 631.0, + 294.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 625.0, + 1406.0, + 625.0, + 1406.0, + 658.0, + 294.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 653.0, + 1403.0, + 653.0, + 1403.0, + 686.0, + 294.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 681.0, + 460.0, + 681.0, + 460.0, + 713.0, + 294.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 681.0, + 1404.0, + 681.0, + 1404.0, + 713.0, + 528.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 710.0, + 1404.0, + 710.0, + 1404.0, + 739.0, + 295.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 733.0, + 1404.0, + 733.0, + 1404.0, + 770.0, + 293.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 762.0, + 470.0, + 762.0, + 470.0, + 799.0, + 295.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2121.0, + 830.0, + 2121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 597.0, + 1408.0, + 597.0, + 1408.0, + 634.0, + 294.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 627.0, + 1404.0, + 627.0, + 1404.0, + 656.0, + 295.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 651.0, + 1403.0, + 651.0, + 1403.0, + 687.0, + 293.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 681.0, + 460.0, + 681.0, + 460.0, + 712.0, + 293.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 681.0, + 1404.0, + 681.0, + 1404.0, + 712.0, + 528.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 709.0, + 1403.0, + 709.0, + 1403.0, + 738.0, + 294.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 734.0, + 1406.0, + 734.0, + 1406.0, + 771.0, + 293.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 765.0, + 470.0, + 765.0, + 470.0, + 798.0, + 295.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 219.0, + 1246.0, + 219.0, + 1246.0, + 250.0, + 475.0, + 250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 219.0, + 1246.0, + 219.0, + 1246.0, + 250.0, + 475.0, + 250.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 25, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 362, + 457, + 1403, + 457, + 1403, + 792, + 362, + 792 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 300, + 286, + 1403, + 286, + 1403, + 439, + 300, + 439 + ], + "score": 0.956 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.883 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2112, + 835, + 2112 + ], + "score": 0.843 + }, + { + "category_id": 2, + "poly": [ + 311, + 231, + 1037, + 231, + 1037, + 261, + 311, + 261 + ], + "score": 0.317 + }, + { + "category_id": 13, + "poly": [ + 529, + 641, + 558, + 641, + 558, + 668, + 529, + 668 + ], + "score": 0.8, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 1373, + 319, + 1401, + 319, + 1401, + 346, + 1373, + 346 + ], + "score": 0.79, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 602, + 353, + 621, + 353, + 621, + 380, + 602, + 380 + ], + "score": 0.7, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 770, + 644, + 790, + 644, + 790, + 671, + 770, + 671 + ], + "score": 0.55, + "latex": "y" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 230.0, + 1040.0, + 230.0, + 1040.0, + 264.0, + 305.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 458.0, + 1402.0, + 458.0, + 1402.0, + 489.0, + 363.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 487.0, + 1404.0, + 487.0, + 1404.0, + 522.0, + 395.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 518.0, + 1404.0, + 518.0, + 1404.0, + 553.0, + 395.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 550.0, + 389.0, + 550.0, + 389.0, + 576.0, + 363.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 544.0, + 1404.0, + 544.0, + 1404.0, + 585.0, + 391.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 578.0, + 1404.0, + 578.0, + 1404.0, + 615.0, + 393.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 608.0, + 1405.0, + 608.0, + 1405.0, + 644.0, + 393.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 641.0, + 528.0, + 641.0, + 528.0, + 675.0, + 395.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 641.0, + 769.0, + 641.0, + 769.0, + 675.0, + 559.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 641.0, + 1404.0, + 641.0, + 1404.0, + 675.0, + 791.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 669.0, + 1405.0, + 669.0, + 1405.0, + 706.0, + 392.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 700.0, + 1403.0, + 700.0, + 1403.0, + 734.0, + 395.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 730.0, + 1404.0, + 730.0, + 1404.0, + 767.0, + 394.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 762.0, + 1167.0, + 762.0, + 1167.0, + 796.0, + 395.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 285.0, + 1403.0, + 285.0, + 1403.0, + 324.0, + 293.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 319.0, + 1372.0, + 319.0, + 1372.0, + 352.0, + 296.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 349.0, + 601.0, + 349.0, + 601.0, + 382.0, + 295.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 349.0, + 1404.0, + 349.0, + 1404.0, + 382.0, + 622.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 376.0, + 1407.0, + 376.0, + 1407.0, + 415.0, + 291.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 413.0, + 394.0, + 413.0, + 394.0, + 443.0, + 294.0, + 443.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 26, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/dev/f-fVCElZ-G1/images/21f35abab8d7dfbf34e6d2cf5a916fb89ebde9ac663be31184d5f2b6c04070e1.jpg b/parse/dev/f-fVCElZ-G1/images/21f35abab8d7dfbf34e6d2cf5a916fb89ebde9ac663be31184d5f2b6c04070e1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f1a79c340e81bbbfa16a9d052f505fd25cc500c --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/21f35abab8d7dfbf34e6d2cf5a916fb89ebde9ac663be31184d5f2b6c04070e1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca2990fe8be2dbcf721d4630a1c67f05b3875eeee0f99d45d954c3db62e724d5 +size 32595 diff --git a/parse/dev/f-fVCElZ-G1/images/35e7e2998b6025ee1b32ba988ce167ff17010d058caa489e86fac774cf092c38.jpg b/parse/dev/f-fVCElZ-G1/images/35e7e2998b6025ee1b32ba988ce167ff17010d058caa489e86fac774cf092c38.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5b2ae29e575f28f55c0cd3303cb226e900cea8c2 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/35e7e2998b6025ee1b32ba988ce167ff17010d058caa489e86fac774cf092c38.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f3335de407e7c4e1744472e62b6d3a21acd9ada75ef06b53f82b8b2984e74f6 +size 84641 diff --git a/parse/dev/f-fVCElZ-G1/images/4dc1348ec4ca9564e57c925fd05b8df8405884858e6719c2085062fbde315918.jpg b/parse/dev/f-fVCElZ-G1/images/4dc1348ec4ca9564e57c925fd05b8df8405884858e6719c2085062fbde315918.jpg new file mode 100644 index 0000000000000000000000000000000000000000..62b2c2c6855728a34289d4d8095907768745b97a --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/4dc1348ec4ca9564e57c925fd05b8df8405884858e6719c2085062fbde315918.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bdf8261936c8b1bc338b5c32cb5f239ec19159a9ee96ee9f3109164cc7bff6ae +size 7706 diff --git a/parse/dev/f-fVCElZ-G1/images/51eb615b84a5a50a35001b81a501b0be8c9c2012d35ca6ca15983af0ddb7bda3.jpg b/parse/dev/f-fVCElZ-G1/images/51eb615b84a5a50a35001b81a501b0be8c9c2012d35ca6ca15983af0ddb7bda3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5251505073687a3462c11385e62da52f0f23a933 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/51eb615b84a5a50a35001b81a501b0be8c9c2012d35ca6ca15983af0ddb7bda3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6381919d5f8c83171ac51dd85cf08a2a84798aa40f1bdb8e7a8a1c27e6bdb761 +size 14812 diff --git a/parse/dev/f-fVCElZ-G1/images/52f2631ffb9d46bab4bc1c10472f6d8fb72739fb1547a27b2d6447f405665435.jpg b/parse/dev/f-fVCElZ-G1/images/52f2631ffb9d46bab4bc1c10472f6d8fb72739fb1547a27b2d6447f405665435.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3061a08cc418b004934f68f1a5b02fd5e34dc269 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/52f2631ffb9d46bab4bc1c10472f6d8fb72739fb1547a27b2d6447f405665435.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae9a81bddbd3f2a58ce8fc916faa1baba138e023f0f09294e53903ec6af8e139 +size 50440 diff --git a/parse/dev/f-fVCElZ-G1/images/5c42bc1fee147731d46d90ce3b65c338ffa2f0a9ec56d91148bd0be4feeb1bb4.jpg b/parse/dev/f-fVCElZ-G1/images/5c42bc1fee147731d46d90ce3b65c338ffa2f0a9ec56d91148bd0be4feeb1bb4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fd52cc140e54289f78b0c84d6b0328830c088219 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/5c42bc1fee147731d46d90ce3b65c338ffa2f0a9ec56d91148bd0be4feeb1bb4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:677232efb2b15231e2ba797c4241ac91a1e4d5e405a7b3666820a0c5d9791821 +size 51381 diff --git a/parse/dev/f-fVCElZ-G1/images/5f3f136bd119381779c24ce59f09719d375d8d63a0a580af36010e79fcb258d1.jpg b/parse/dev/f-fVCElZ-G1/images/5f3f136bd119381779c24ce59f09719d375d8d63a0a580af36010e79fcb258d1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..da4cb411148e9838770652273b1a97b1ea1d3965 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/5f3f136bd119381779c24ce59f09719d375d8d63a0a580af36010e79fcb258d1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:131abc34809d1ccfbb73a89cba128c97950adaac7471d21acc3e2274c492cb16 +size 51888 diff --git a/parse/dev/f-fVCElZ-G1/images/63f77665b438db357963aba154d7f4ae3d9e19d0b1c56b5c253caa349baf73f1.jpg b/parse/dev/f-fVCElZ-G1/images/63f77665b438db357963aba154d7f4ae3d9e19d0b1c56b5c253caa349baf73f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bee0906e9195a41eea8dc1599dcb4fe7ccf6ade9 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/63f77665b438db357963aba154d7f4ae3d9e19d0b1c56b5c253caa349baf73f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9699ce0383fea9823e3c388e6007983202847d865b9f6ecdd7a973eacfd44b60 +size 77749 diff --git a/parse/dev/f-fVCElZ-G1/images/a2dca122815f99ddb040d073632b2d126537cbeebb4ffb4d6b36b423bca5a255.jpg b/parse/dev/f-fVCElZ-G1/images/a2dca122815f99ddb040d073632b2d126537cbeebb4ffb4d6b36b423bca5a255.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4bec92a05577b956052afce109687e844295a022 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/a2dca122815f99ddb040d073632b2d126537cbeebb4ffb4d6b36b423bca5a255.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60332e3e0dedb0e97f09d710d199140d75e7d8cd666d35fc4d5fbbc4ea46bb87 +size 57956 diff --git a/parse/dev/f-fVCElZ-G1/images/aa57b112b2da0bf75996dd274dafd3945fd4e9d8639a886788ddc48c6442430e.jpg b/parse/dev/f-fVCElZ-G1/images/aa57b112b2da0bf75996dd274dafd3945fd4e9d8639a886788ddc48c6442430e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3d60ef9408a322a3c329561c21620f7344c4a3c1 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/aa57b112b2da0bf75996dd274dafd3945fd4e9d8639a886788ddc48c6442430e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b4104e899a8780ed991fbfb16f7c2b9eb868c85f2f49fc4cc9f1fc47904a1a1 +size 16092 diff --git a/parse/dev/f-fVCElZ-G1/images/b7ccdd045c7c258ea47e1bc5b9ae3005151b5c70d2d84b32df1fe7bed308ab3f.jpg b/parse/dev/f-fVCElZ-G1/images/b7ccdd045c7c258ea47e1bc5b9ae3005151b5c70d2d84b32df1fe7bed308ab3f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c4c1370103dae04e8f7b4707a3d382998978a0be --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/b7ccdd045c7c258ea47e1bc5b9ae3005151b5c70d2d84b32df1fe7bed308ab3f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1b5009e1507100b2fd42750b4e3d75017f31d8ac0c5351987ead7252f32ca99 +size 21109 diff --git a/parse/dev/f-fVCElZ-G1/images/e003d37e3d482fcd798a0d8e7d1be94febcf5048398aca175c49b7a253357071.jpg b/parse/dev/f-fVCElZ-G1/images/e003d37e3d482fcd798a0d8e7d1be94febcf5048398aca175c49b7a253357071.jpg new file mode 100644 index 0000000000000000000000000000000000000000..59d5c8ba332032b67df95f1ebcd1132139c8ed9e --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/e003d37e3d482fcd798a0d8e7d1be94febcf5048398aca175c49b7a253357071.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5837b95bc2ab1bd694b990b3ce7c1bb3a192364c4063b0c3ebf142ea0ff2c3a1 +size 20943 diff --git a/parse/dev/f-fVCElZ-G1/images/e1035b012a5333f4a730c48890985a9581d33c82e401db1200e513dc168af1f9.jpg b/parse/dev/f-fVCElZ-G1/images/e1035b012a5333f4a730c48890985a9581d33c82e401db1200e513dc168af1f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fe926f23e8d22ec3fec5377d64fdd08b1e1f1442 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/e1035b012a5333f4a730c48890985a9581d33c82e401db1200e513dc168af1f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7e9557333747a4654d11902f3857ca472473445cd2158f02ace55085efb22f6 +size 5811 diff --git a/parse/dev/f-fVCElZ-G1/images/f1291b3aec0429be45d4ac807c5c02f4bb3eced1d829c0ebd9852ea83612fd08.jpg b/parse/dev/f-fVCElZ-G1/images/f1291b3aec0429be45d4ac807c5c02f4bb3eced1d829c0ebd9852ea83612fd08.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e49b3c74e5f6998e3b936793583f677b35d91458 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/f1291b3aec0429be45d4ac807c5c02f4bb3eced1d829c0ebd9852ea83612fd08.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:20195f202a50501cb0974ce7d9747d7946af435bff21f3818305e18b5b6f4dc6 +size 31071 diff --git a/parse/dev/f-fVCElZ-G1/images/f3b76b6d91260f2c1b03c248b1a4d692c58238d585e68698e5fe1ed437aefc7d.jpg b/parse/dev/f-fVCElZ-G1/images/f3b76b6d91260f2c1b03c248b1a4d692c58238d585e68698e5fe1ed437aefc7d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..550f86ae9c5b672c178d211408e6d322809d4bf8 --- /dev/null +++ b/parse/dev/f-fVCElZ-G1/images/f3b76b6d91260f2c1b03c248b1a4d692c58238d585e68698e5fe1ed437aefc7d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e77d783d03d8f5168750e2feecd5d34012ef78bbe222bf32e6c79b4dd46afcba +size 37417 diff --git a/parse/dev/gmL46YMpu2J/gmL46YMpu2J_content_list.json b/parse/dev/gmL46YMpu2J/gmL46YMpu2J_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..cf8983f7fa14299b58863170c178b8f42df7769e --- /dev/null +++ b/parse/dev/gmL46YMpu2J/gmL46YMpu2J_content_list.json @@ -0,0 +1,2237 @@ +[ + { + "type": "text", + "text": "PROMPTAGATOR FEW-SHOT DENSE RETRIEVAL FROM 8 EXAMPLES ", + "text_level": 1, + "bbox": [ + 173, + 106, + 823, + 156 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Zhuyun Dai∗†, Vincent Y. Zhao∗†, Ji $\\mathbf { M } \\mathbf { a } ^ { * \\dagger }$ , Yi Luan∗†, Jianmo Ni, Jing Lu, Anton Bakalov, \nKelvin Guu, Keith B. Hall and Ming-Wei Chang† \nGoogle Research \n{zhuyundai, vzhao, maji, luanyi, mingweichang}@google.com \n∗equal contributions †corresponding authors ", + "bbox": [ + 184, + 178, + 816, + 251 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 287, + 544, + 303 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Much recent research on information retrieval has focused on how to transfer from one task (typically with abundant supervised data) to various other retrieval tasks where supervision is limited, with the implicit assumption that it is possible to generalize from one task to all the rest. However, this overlooks the fact that there are many diverse and unique retrieval problems, each targeting different search intents, queries, and search domains. In this paper, we suggest to work on Few-shot Dense Retrieval, a setting where each task comes with a short description and a few examples. To address this, we introduce Prompt-based Query Generation for Retrieval (PROMPTAGATOR ): for each task, we feed the few-shot examples to a large language model (LLM) and prompt it to behave as a task-specific query generator. Using this, we can synthetically generate a large number of relevant queries for any document, yielding abundant data for training task-specific retrievers — with no reliance on traditional resources such as Natural Questions (Kwiatkowski et al., 2019) or MS MARCO (Nguyen et al., 2016). Surprisingly, PROMPTAGATOR using only 8 annotated examples enables efficient dual encoder retrievers to outperform computationally more expensive models trained on MS MARCO such as ColBERT v2 (Santhanam et al., 2022) by more than 1.2 points $\\mathrm { n D C G } @ 1 0$ on average on 11 retrieval sets. Further training standard-size rerankers using the same generated data yields another 5.0 points $\\mathrm { n D C G } @ 1 0$ improvement. Our studies show that synthetic query generation can be far more effective than previously observed, especially when a small amount of task-specific knowledge is given. ", + "bbox": [ + 233, + 318, + 766, + 609 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 635, + 336, + 651 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Significant progress has been made on neural retrieval models such as dual encoders, which can search over a large collection of documents containing millions to billions of passages (Yih et al., 2011; Lee et al., 2019; Karpukhin et al., 2020). However, Thakur et al. (2021) recently proposed the BEIR heterogeneous retrieval benchmark, and showed that it is still difficult for neural retrievers to perform well on a wide variety of retrieval tasks that lack dedicated training data. To address this problem, many previous approaches focus on transferring knowledge from high-resource question answering (QA) datasets such as MS MARCO (Nguyen et al., 2016), and propose architectures that possess good inductive biases, such as models that allow fine-grained token-level interaction (e.g., ColBERT (Khattab & Zaharia, 2020; Santhanam et al., 2022) and SPLADE (Formal et al., 2021)) which often come with higher inference cost. Data augmentation via synthetic query generation has previously been explored (Ma et al., 2021; Shakeri et al., 2020), but these question generators are learned from high-resource QA datasets, and often cannot generalize well to new retrieval tasks. ", + "bbox": [ + 174, + 666, + 825, + 833 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We argue that it is hard to anticipate models based on one or two QA datasets to perform well across all retrieval tasks. First, different retrieval tasks have very different search intents; in other words, different definitions of “relevance”. For example, consider Figure 1(a): both DbpediaEntity (Hasibi et al., 2017) and FEVER (Thorne et al., 2018) are tasks to retrieve documents from Wikipedia. However, Dbpedia-Entity is a task to retrieve entities that are mentioned in the query, while FEVER is a task to find evidence that either supports or refutes a given statement. Which document is relevant to the query can be very different from one task to another task even if they share the same domain. Moreover, different tasks have distinct distributions of queries even when their search intents are similar. For example, queries in HotpotQA (Yang et al., 2018) are long compositional questions, while queries in FiQA (Maia et al., 2018) are short financial questions. ", + "bbox": [ + 174, + 840, + 825, + 924 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/ee0c43785d29c1f62b49c195278b442952aaf7a8f03a8a4f888a0e81832b1f12.jpg", + "image_caption": [ + "Figure 1: Few-shot retrieval with PROMPTAGATOR. Left (a): Retrieval tasks from BEIR differ in query distribution, retrieval corpus, and search intents. Middle (b): Most prior work uses supervised setting (2) which trains model on a large QA retrieval datasets and transfer to other retrieval tasks. Right (c): Few-shot PROMPTAGATOR performance. Average nDCG $@ 1 0$ on 11 datasets from BEIR from our PROMPTAGATOR models and previously MS MARCO-supervised models (SPLADE v2). " + ], + "image_footnote": [], + "bbox": [ + 215, + 102, + 750, + 238 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 329, + 825, + 386 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Motivated by these observations, we advocate to work on the setting of Few-shot Retrieval for diverse retrieval tasks (§2), where each task comes with a short description and a few annotated examples to clearly illustrate the search intent. To address this challenge, we propose Prompt-based Query Generation for Retrieval (PROMPTAGATOR) (§3): for each new retrieval task, we feed the few-shot examples to a large language model (LLM) such as FLAN1 (Wei et al., 2022a) and prompt it to perform doc-to-query generation. Importantly, the few-shot examples ensure that we capture the specific search intent of that task. Using this query generator, we can synthetically generate a large number of relevant queries for any document, yielding abundant data for training any retriever, including highly efficient dual encoder models. ", + "bbox": [ + 173, + 392, + 825, + 517 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "We find that our few-shot LLM query generator can produce good queries without any fine-tuning (§3.1). In fact, as shown in Figure 1(b), our synthetically generated data is strong enough to completely forego using annotated query-document pairs from traditional high-resource datasets such as Natural Questions (Kwiatkowski et al., 2019) or MS MARCO (Nguyen et al., 2016). ", + "bbox": [ + 174, + 525, + 825, + 580 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "While PROMPTAGATOR is not the first application of LLMs for retrieval, prior work did not explore task-specific few-shot adaptation, and often came with high inference cost. Neelakantan et al. (2022) proposes to use GPT-3 (Brown et al., 2020) in dual encoders. However, their embedding dimension is $1 2 \\mathrm { k }$ , which makes the search index footprint and inference cost prohibitively high for many applications. Sachan et al. (2022) and Bonifacio et al. (2022) prompt LLMs for question generation, but did not explore the idea of using task-specific few-shot prompts for rapid task adaptation.2 They also focus primarily on models that rerank top retrievals from an existing retriever, rather than directly adapting the underlying retriever which must efficiently search over millions or billions of documents. ", + "bbox": [ + 174, + 588, + 825, + 699 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "To summarize, the contributions of the paper are as follows: ", + "bbox": [ + 174, + 705, + 566, + 720 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• We highlight previously overlooked differences across retrieval tasks (e.g., search intent and query distribution), and propose a Few-Shot Retrieval evaluation for the BEIR dataset. We propose PROMPTAGATOR, a simple recipe for few-shot retrieval by prompting an LLM to generate synthetic task-specific training data. For the first time, we can train fully neural retrievers and rerankers solely based on a few supervised examples. Our results show that, surprisingly, PROMPTAGATOR with two-to-eight examples produces significantly better retrievers than recent models trained on MS MARCO or NQ that have over 500k human annotated examples (Figure 1(c)) and utilize more expensive architectures: PROMPTAGATOR outperforms ColBERT v2 and SPLADE v2 on 11 retrieval tasks we tested, while reranking boosts results by another 5 points on standard retrieval evaluation metric. ", + "bbox": [ + 215, + 727, + 825, + 876 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 FEW-SHOT RETRIEVAL TASK ", + "text_level": 1, + "bbox": [ + 176, + 102, + 444, + 118 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In this section, we first introduce the definition of a few-shot retrieval task and discuss the differences among tasks. We then propose a new Few-Shot Retrieval setting for the BEIR benchmark. ", + "bbox": [ + 174, + 132, + 823, + 161 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.1 RETRIEVAL TASK ", + "text_level": 1, + "bbox": [ + 174, + 178, + 338, + 193 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Given a large corpus, a retrieval model is responsible for finding the documents that are most relevant to a provided query $q$ according to a pre-defined notion of relevance. Formally, a retrieval task is: ", + "bbox": [ + 173, + 204, + 823, + 232 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/2225e499b8d5f500e4746b4fd596ef178e268172bd0c43057333ad50527457be.jpg", + "text": "$$\nT = \\{ \\mathcal { D } , \\mathcal { Q } , \\mathcal { Z } \\} ,\n$$", + "text_format": "latex", + "bbox": [ + 444, + 237, + 552, + 253 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\mathcal { D } = \\{ d _ { 1 } , d _ { 2 } , . . . , d _ { n } \\}$ is a large corpus of documents for retrieval, $\\mathcal { Q }$ is a query distribution, and $\\mathcal { T }$ is the underlying search intent for the task. Depending on the task, $\\mathcal { D }$ can be any document collection, such as the web or Wikipedia. $Q$ also varies across tasks, e.g., short keyword search queries, questions, arguments, etc. If $\\textstyle { \\mathcal { I } } ( q , d ) = 1$ , it means the search intent of $q$ has been satisfied by the document $d$ . For example, in a question answering task $\\mathcal { T } _ { \\mathrm { Q A } } ( q , d ) = 1$ if $d$ answers $q$ . For the same $( q , d )$ pair, relevance may be either 1 or 0 depending on the search intent. For example, some argument retrieval tasks only seek to retrieve supporting arguments, while others aim to retrieve counterarguments. ", + "bbox": [ + 174, + 257, + 825, + 369 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In this work, we assume a target retrieval corpus $\\mathcal { D } _ { \\mathcal { T } }$ is given, but the amount of annotated querydocument pairs for the new task is limited. Most prior research efforts focused on adapting retrievers to a new corpus $\\mathcal { D } _ { \\mathcal { T } }$ , but didn’t fully account for divergence in queries $\\mathcal { Q } _ { T }$ or intents $\\mathcal { T } _ { T }$ . Next, we explore how a search intent can be expressed with a short description and very few examples. ", + "bbox": [ + 174, + 377, + 825, + 433 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.2 FEW-SHOT BEIR SETTING ", + "text_level": 1, + "bbox": [ + 176, + 449, + 398, + 463 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Intuitively, a person can understand a retrieval task by reading a short prompt and going over a few examples. In this work, we ask if a few (8 or fewer) examples are sufficient to learn a task-specific retriever. To facilitate our study and future research on few-shot retrieval, we define a new few-shot retrieval evaluation built upon the BEIR heterogeneous retrieval benchmark (Thakur et al., 2021). ", + "bbox": [ + 174, + 474, + 825, + 531 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "BEIR has 18 information retrieval datasets across 9 domains, including Bio-Medical, Finance, News, Twitter, Wikipedia, StackExchange, Quora, Scientific, and Misc. These datasets also cover a diverse range of search intents: QA retrieval (question-to-document), duplicate question discovery (questionto-question), fact checking (claim-to-document), etc. Following Santhanam et al. (2022) and Formal et al. (2021), we narrow our focus to the publicly-available datasets in BEIR. The original BEIR evaluation used a zero-shot setup, where no queries or relevant query-document pairs from the evaluation datasets can be used for training. ", + "bbox": [ + 174, + 537, + 825, + 636 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We relax BEIR to the few-shot setting by randomly taking a few (2 to 8) in-domain relevant querydocument examples as task-specific supervision — in realistic applications, this number of examples is almost always possible to obtain. The examples are sampled from the development set when it is available. For BEIR tasks which only have a test set, we use samples from the test data. To be fair when evaluating our models, we always mark these test-set examples as ‘failed‘: we remove the documents from our retrieved results when computing metrics, even if they are correctly retrieved (the worst possible outcome). The prompts and few-shot examples will be released to the public. ", + "bbox": [ + 174, + 642, + 825, + 741 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 PROMPTAGATOR ", + "text_level": 1, + "bbox": [ + 176, + 760, + 341, + 775 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The key idea of PROMPTAGATOR is to transform a few examples into many more examples by prompting an LLM to generate more data, instead of using them to train a retriever directly. ", + "bbox": [ + 173, + 791, + 821, + 819 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "PROMPTAGATOR consists of three components: prompt-based query generation, consistency filtering, and retriever training. During prompt-based query generation, a task-specific prompt will be combined with a large language model to produce relevant queries for all documents in $\\mathcal { D } _ { T }$ . Then, a filtering step cleans the generated data based on round-trip consistency. Surprisingly, we found that a retriever trained only on our synthetic data can be used to filter the synthetic data. Finally, a retriever (in this paper, dual encoders) and a cross attention reranker are trained based on the filtered data. Figure 5 in Appendix shows the overall procedure. ", + "bbox": [ + 174, + 825, + 825, + 924 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.1 PROMPT-BASED QUERY GENERATION", + "text_level": 1, + "bbox": [ + 176, + 103, + 475, + 118 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In this first step, we feed our task-specific few-shot examples into a large language model (LLM) and prompt it to perform document-to-query generation. More precisely, let $\\{ ( q _ { i } , - d _ { i } ) \\bar \\} ^ { k }$ be the $k$ few-shot examples, where each example is a query $( q _ { i } \\sim \\mathcal { Q } _ { T } )$ and a document relevant to that query $( d _ { i } \\in \\mathcal { D } _ { T }$ ) according to the target task $T$ $( \\mathbb { Z } _ { T } ( q _ { i } , d _ { i } ) = 1 )$ ). ", + "bbox": [ + 174, + 128, + 825, + 186 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "ollowing FLAN (Wei et al., 2022a), we instruction-prompt the LLM with the following string prefix: ", + "bbox": [ + 186, + 191, + 818, + 207 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/77edbfd39c8ed331e944862e98cc5b7f043a5c42888d92b5f52df985950e093a.jpg", + "text": "$$\ne _ { d o c } ( d _ { i } ) \\diamond e _ { q u e r y } ( q _ { 1 } ) \\diamond . . . \\diamond e _ { d o c } ( d _ { k } ) \\diamond e _ { q u e r y } ( q _ { k } ) \\diamond e _ { d o c } ( d ) ,\n$$", + "text_format": "latex", + "bbox": [ + 297, + 212, + 699, + 229 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $\\diamond$ is a separator token, $e _ { d o c } ( d )$ and $e _ { q u e r y } ( q )$ are task-specific document and query descriptions respectively, and $d$ is a new document presented at inference time. Using the ArguAna task as an example, we set $e _ { d o c } ( d ) = \\mathrm { \\ \" { s } ~ } \\mathrm { \\bar { s } ~ } _ { \\mathrm { \\bar { d } r g u m e n t : } } \\left\\{ \\mathrm { \\Omega } \\right.$ $\\{ d \\}$ ” and $e _ { q u e r y } =$ “Counter Argument: $\\{ { q \\} } ^ { , }$ to inform the LLM to generate counterarguments 3. The LLM is expected to generate $e _ { q u e r y } ( \\hat { q } )$ . We consider it a generation failure if the query description does not precede the actual query; otherwise, we accept $\\hat { q }$ and form a synthetic relevant example $( \\hat { q } , d )$ . ", + "bbox": [ + 174, + 233, + 825, + 320 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Running the prompt on all documents from $\\mathcal { D } _ { T }$ , we obtain a large set of synthetic $( \\hat { q } , d )$ examples, amplifying the information from a few examples into a large synthetic dataset whose query distribution is similar to the true task distribution $\\mathcal { Q } _ { T }$ and whose query-document pairs convey the true search intent $\\mathcal { I } _ { T }$ . This form of few-shot data extrapolation is similar to Lee et al. (2021). ", + "bbox": [ + 174, + 325, + 825, + 382 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We use FLAN (Wei et al., 2022a) as our LLM, and refer to our query generator as $p _ { \\mathrm { F L A N } } ( q | d )$ . FLAN is trained on a collection of tasks described via instructions and was shown to have good zero/few-shot performance on unseen tasks. We use the 137B parameter checkpoint. During prompt engineering, we use at most 8 examples, and reduce the number if they exceed the input length limit of FLAN. We also manually truncate individual queries and documents in the examples if they are too long. We randomly sample up to 1 million documents from each corpus and generate 8 questions per document using sampling-based decoding with temperature 0.7. ", + "bbox": [ + 174, + 388, + 825, + 487 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2 ROUND-TRIP FILTERING GENERATED DATA ", + "text_level": 1, + "bbox": [ + 178, + 503, + 508, + 517 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We employ round-trip filtering (Alberti et al., 2019; Lewis et al., 2021) to improve the quality of our synthetic data. The main intuition is that for any synthetic query $\\hat { q }$ generated from passage $d$ , a good $\\hat { q }$ should also retrieve its original passage $d$ . In other words, the original $d$ should have high probability under some retriever, $p ( d | \\hat { q } )$ (the reverse direction of query generation). If not, then we filter out $\\hat { q }$ . ", + "bbox": [ + 174, + 529, + 825, + 585 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Round-trip filtering has been very effective for synthetic question generation on QA tasks. However, these techniques typically rely on a question-answering model for the reverse direction filter. Since not all retrieval tasks resemble question-answering, this will not suffice in our setting. ", + "bbox": [ + 176, + 592, + 823, + 635 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Instead, we train an initial retriever from the unfiltered synthetic data, and then use it to filter the synthetic data. This works surprisingly well over the different search intents observed in BEIR. More precisely, given a synthetic query-document pair $( \\hat { q } , d )$ , we use the initial retriever to predict the most relevant passages for $\\hat { q }$ . We keep $\\hat { q }$ only when $d$ occurs among the Top- $K$ passages returned by the retriever. We show this filter substantially reduces the number of synthetic queries and significantly improves retrieval performance. In Appendix F, we provide more insight into why this can work by viewing our synthetic queries as latent variables. ", + "bbox": [ + 174, + 641, + 825, + 739 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.3 FEW-SHOT PROMPTAGATOR RETRIEVER ", + "text_level": 1, + "bbox": [ + 176, + 756, + 480, + 770 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Our synthetically generated data allows training task-specific neutral retrievers for tasks where indomain fine-tuning is challenging due to data scarcity. In this work, we use a standard dual encoder retrieval architecture and we propose a simple pretrain-finetune recipe. ", + "bbox": [ + 174, + 781, + 825, + 824 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We pretrain our retriever on C4 with the independent cropping task from Contriever (Izacard et al., 2022a), where we treat two random crops from the same document as an artificial positive (query, document) pair and train with a cross-entropy loss over in-batch random negatives. Next, we fine-tune the dual encoder on $( \\hat { q } , d )$ pairs from our prompt-based query generation, again with in-batch random negatives. After training for a set number of epochs, we apply round-trip filtering on our synthetic ", + "bbox": [ + 174, + 830, + 825, + 901 + ], + "page_idx": 3 + }, + { + "type": "table", + "img_path": "images/8ff4d4a1a185520110f6ba990d81554139ae3920f0b15e79743d8d4b38954ea1.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
Training RecipeRetrieval ArchitectureQGen
Retrieval SupervisionCross-Attn DistillationRetrieverToken-level Retrieval#Reranking Doc.Serving Model SizeModel
ContrieverNAself110M
GTR-XXLMS MARCO (500K)self6B
Splade v2MS MARCO (500K)//vvself√110M
ColBERT v2MS MARCO (500K)self√110M
GenQMS MARCO (500K)self110MT5 (MS MARCO)
GPLMS MARCO (500K)self110MT5 (MS MARCO)
MonoT5MS MARCO (500K)BM25√10003B
InParsFew (3 from MS MARCO)BM25√10003BGPT-3
UPRNAContriever1000110M+3BT0*
PROMPTAGATORFew (0-8)self110MFLAN
PROMPTAGATOR++Few (0-8)PROMPTAGATOR200110M+125MFLAN
", + "bbox": [ + 173, + 99, + 821, + 242 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Table 1: Comparison of different retrieval frameworks. Our serving models are just a 110M-size dual encoder PROMPTAGATOR and a 125M-size reranker PROMPTAGAT $\\mathrm { \\Phi _ { O R + + } }$ , as good quality generated data allows simple models/pipeline to achieve strong performance.4 InPars’s few-shot examples are from MS MARCO and is task-independent. See text for more details for UPR’s QGen model5. ", + "bbox": [ + 173, + 252, + 825, + 308 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "data as described in $\\ S 3 . 2$ using this initial dual encoder, and then continue to fine-tune the dual encoder on the filtered data. ", + "bbox": [ + 174, + 325, + 821, + 353 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We also propose PROMPTAGATOR $^ { + + }$ , a reranker trained on the same synthetic data generated from our prompt-based QGen, which refines the retrieved candidates using a slower but more accurate cross-attention model. We train the reranker using a cross-entropy loss with 31 sampled negatives from top 200 passages retrieved by the PROMPTAGATOR retriever, which approximates the inference time distribution (reranking top 200 from the retriever). ", + "bbox": [ + 174, + 361, + 825, + 430 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Zero-shot PROMPTAGATOR Our prompt-based query generation can run in a zero-shot manner, where we universally apply the following prompt irrespective of the target task: ’{d} Read the passage and generate a query.’. Here {d} denotes the document text. Training retrievers and rerankers on this data leads to zero-shot PROMPTAGATOR and zero-shot PROMPTAGATOR $^ { + + }$ . This recipe serves as a baseline to show the benefits of adapting the few-shot prompt to the target task. ", + "bbox": [ + 174, + 446, + 825, + 517 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.4 COMPARISON WITH PRIOR METHODS ", + "text_level": 1, + "bbox": [ + 178, + 536, + 468, + 549 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Table 1 compares the setting of PROMPTAGATOR to some recently proposed approaches. Several dimensions of our recipe are simpler: our dual encoder does not employ hard negative mining, distillation from a cross-attention teacher or token-level retrieval. Also, our 125M parameter reranker is smaller than most other rerankers. We aim to show that even simpler and smaller architectures can achieve excellent results if trained with synthetic data that has been few-shot adapted $( \\ S 4 . 3 )$ . Compared to InPars (Bonifacio et al., 2022) and UPR (Sachan et al., 2022), our approach employs task-specific few-shot adaption, while InPars and UPR’s prompts are task-independent and thus bear the same limitation of previous query generation approach (Ma et al., 2021; Wang et al., 2022). Another key difference is that prior works focused on reranking, while we enable few-shot learning for both reranking and retrieval. ", + "bbox": [ + 174, + 563, + 826, + 702 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4 EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 176, + 723, + 326, + 738 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We evaluate PROMPTAGATOR on the BEIR benchmark. We then dive deeper into the results through ablation studies and qualitative analysis. ", + "bbox": [ + 176, + 756, + 823, + 784 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.1 IMPLEMENTATION ", + "text_level": 1, + "bbox": [ + 176, + 803, + 341, + 816 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The original FLAN training set overlapped with 2 datasets in BEIR: NQ and Quora6. Most existing systems use MS MARCO for fully supervised learning and therefore do not report few or zeroshot results on MS MARCO. Therefore we exclude MS MARCO, NQ and Quora from our main evaluations. We report $\\mathrm { n D C G } @ 1 0$ , the standard retrieval evaluation metric on BEIR. ", + "bbox": [ + 176, + 828, + 825, + 857 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/5cba16e99adaaf919181c599ebac6a2a3d27e4cc047248bc0479afc5aa2c6859.jpg", + "table_caption": [ + "Table 2: Main Results. $\\mathrm { n D C G } @ 1 0$ on BEIR. Retriever Comparisons (Upper Half): Among the various kind of retrievers, both zero-shot and few-shot PROMPTAGATOR produce strong results. Retriever+Reranker Comparisons (Lower Half): In the scenario where speed is not a concern, reranker is often used. We train PROMPTAGATOR $^ { + + }$ use the same generated data and get significant improvement. See text for more details for Climate-FEVER and Webis-Touché2020. " + ], + "table_footnote": [], + "table_body": "
argtouchécovidnfchotpotdbpclimatefeverscifactscidocsfiqaAVG.
Unsupervised
BM2531.536.765.632.560.331.321.375.366.515.823.641.8
Contriever37.919.327.431.7 48.129.215.568.264.914.924.534.7
Supervised by MS MARCO16.1
GTR-XXL54.025.650.134.259.940.826.774.066.246.744.9
SPLADE v247.927.271.033.468.443.523.578.669.315.833.646.6
ColBERTv246.326.373.833.866.744.617.678.569.315.435.646.2
GenQ49.318.261.9 70.031.953.432.817.566.964.414.330.840.1
GPL55.725.534.558.238.423.575.967.416.934.445.5
PROMPTAGATOR (110M)
Zero-shot53.826.672.733.460.436.421.476.262.316.340.445.5
Few-shot59.434.575.633.461.438.016.8 (24.0*)77.065.018.446.247.8
Retriever+ Reranker
Unsupervised
UPR (3B) InPars (3B)50.321.360.4 33.372.233.89.557.3 169.617.345.042.7
178.411111
Supervised by MS MARCO
monoT5 (220M)13.227.777.835.769.541.924.580.273.616.541.445.6
monoT5 (3B)28.820.079.538.475.947.828.085.077.719.751.451.1
PROMPTAGATOR++(110M+ 125M)
Zero-shot52.127.876.036.071.241.322.683.8 86.673.2 73.119.1 20.145.949.9
Few-shot63.038.176.237.073.643.420.3 (24.1*)49.452.8
", + "bbox": [ + 174, + 101, + 823, + 388 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 176, + 483, + 820, + 512 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "For PROMPTAGATOR’s query generation, we sample questions from FLAN with a temperature of 0.7. For round-trip filtering, we find that setting filtering threshold $K$ to 1 gives the best results on MS MARCO and thus use 1 for all BEIR datasets, We implement PROMPTAGATOR’s dual encoders following GTR (Ni et al., 2021).To ensure efficiency, we use the T5-base encoder architecture consisting of 110M parameters. For PROMPTAGATOR $^ { + + }$ reranking models, we initialize from a T5- base version 1.1 encoder checkpoint which has 125M parameters . More details of the reranker implementation can be found in Appendix C. At inference time, we rerank the top 200 candidates retrieved from the PROMPTAGATOR dual encoder. ", + "bbox": [ + 174, + 518, + 825, + 630 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We mostly follow the hyperparameters used in Ni et al. (2021). By default, we use batch size 6k; however, some of the corpora in BEIR contain only a few thousand documents, making multiple relevant documents appear in the same batch, which interacts negatively with our in-batch softmax loss. To address this issue, we split all datasets into three groups based on corpus size: small $( < 5 0 \\mathrm { k } )$ , medium $( 5 0 \\mathbf { k } { - } 5 0 0 \\mathbf { k } )$ and large $( > 5 0 0 \\mathrm { k } )$ . For dual encoder training, we use 128 batch size for small datasets and 6k for others. We finetune for 5k steps for large datasets and 1k for others. For ranking models, we use batch size 64 for all datasets and finetune large datasets for $2 0 \\mathrm { k }$ steps, $5 \\mathrm { k }$ for others. ", + "bbox": [ + 174, + 637, + 825, + 734 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.2 MAIN RESULTS ", + "text_level": 1, + "bbox": [ + 176, + 752, + 323, + 765 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Table 2 shows the experimental results. We first notice that zero-shot PROMPTAGATOR already serves as a strong baseline, comparing favorably to other retrieval baselines trained on $\\mathcal { O } ( 1 0 0 \\mathrm { K } )$ examples from MS MARCO. Nonetheless, few-shot PROMPTAGATOR markedly improves upon zeroshot PROMPTAGATOR, increasing average nDCG $@ 1 0$ by over 2 points, which highlights the impact of few-shot learning. Few-shot PROMPTAGATOR, despite having a simple training procedure and model architecture, outperforms strong baselines such as GenQ (Thakur et al., 2021) and GPL (Wang et al., 2022) which also use query generation to augment training data, as well as ColBERT v2 (Santhanam et al., 2022) and SPLADE v2 (Formal et al., 2021) which rely on token level interaction architectures and distillation recipes. ", + "bbox": [ + 173, + 777, + 825, + 876 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/8fd7de3bf06be111ea0a3c358116bf0648d406528f07ae2fda10ca3417e0c05f.jpg", + "image_caption": [ + "Figure 2: Left (a). Delta in $\\mathrm { n D C G } @ 1 0$ between few-shot PROMPTAGATOR with and without filtering. Middle (b): Comparing the effect of the generated data versus the number of supervised data on MS MARCO. PROMPTAGATOR with 8 examples can catch up with 50k labeled examples, when simple dual encoders are used. Right (c): Ablation on query generation model. GenQ is a prior system from Thakur et al. (2021), while NQ-QGen is our in-house NQ-trained T5 query generation model. Other than the generated data, NQ-QGen and PROMPTAGATOR uses the same hyper parameters. " + ], + "image_footnote": [], + "bbox": [ + 192, + 103, + 805, + 251 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 380, + 825, + 409 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Our reranker PROMPTAGATOR $^ { + + }$ boosts performance by another 5 points on nDCG $@ 1 0$ . It significantly outperforms UPR (Sachan et al., 2022) whose reranker uses T0 (Sanh et al., 2022), an instruction tuned LLM similar to FLAN. It also outperforms monoT5-3B (Nogueira et al., 2020), which achieved previous state-of-the-art reranking performance on BEIR in a recent study (Rosa et al., 2022). Note that most of these reranker approaches use a large 3B parameter model for better generalization, while PROMPTAGATOR $^ { + + }$ uses a standard $1 2 5 \\mathbf { M }$ reranker. ", + "bbox": [ + 174, + 415, + 825, + 498 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Comparing few-shot PROMPTAGATOR to baselines, the biggest improvement is on Webis-Touché2020 (touché), followed by ArguAna (arg) . Webis-Touché2020’s goal is to retrieve documents for a controversial topic, e.g., “should felons who have completed their sentence be allowed to vote?”. ArguAna’s goal is to find the counter-arguments that oppose the input argument, and the input arguments are often several-sentence long. Both tasks are extremely different from traditional QA retrieval data that other models use, which are dominated by factoid questions. On the other hand, few-shot PROMPTAGATOR can successfully adapt to this task with a few examples. ", + "bbox": [ + 174, + 506, + 825, + 603 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.3 ANALYSIS ", + "text_level": 1, + "bbox": [ + 174, + 625, + 285, + 638 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Impact of round-trip filtering. In Figure 2(a), we show quality differences between few-shot PROMPTAGATOR with and without filtering. Filtering improves performance on 8 out of 11 datasets and leads to 2.5 points improvement on average, demonstrating the effectiveness of our filtering strategy. Nonetheless, filtering hurts model quality on NFCorpus and SciFact. These are the smallest datasets in terms of generated queries and may indicate overfitting of our retrievers. ", + "bbox": [ + 174, + 651, + 825, + 722 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We find that the majority of filtered examples are either queries that are too generic that match many documents, or queries that contain additional terms that are irrelevant to the document. Examples are in Fig. 6 in the Appendix. There are also cases where high quality data was incorrectly removed. We suspect that designing dynamic filtering thresholds would help, and leave it to future exploration. ", + "bbox": [ + 174, + 728, + 823, + 784 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Can generated queries replace human annotated queries? In Figure 2(b), we evaluate 8-shot PROMPTAGATOR on MS MARCO, comparing it against dual encoders trained on MS MARCO’s supervised data. Note that we did not add other components, to make the comparison simple. We chose MS MARCO as there are enough labeled data for this task and neither FLAN nor our models are trained on MS MARCO examples. The results show that eight examples plus an LLM can replace a significant portion of supervised examples. ", + "bbox": [ + 174, + 791, + 825, + 875 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "How does PROMPTAGATOR compare to other query generation approaches? Figure 2(c) compares zero-shot PROMPTAGATOR to two other query generation approaches: GenQ is prior system from Thakur et al. (2021) using a MS MARCO trained T5 query generation model, and NQ-QGen is our in-house T5 QGen model finetuned on NQ. The figure shows the advantages of zero-shot PROMPTAGATOR, outperforming both baselines by large margins. Importantly, NQ-QGen uses the same filtering, dual-encoder training, batch sizes and training steps as PROMPTAGATOR, providing a faircomparison of query generators. This indicates that the main contributing factor to PROMPTAGATOR is better queries from prompting an LLM, not the specific training recipe or hyperparameters. ", + "bbox": [ + 176, + 882, + 825, + 924 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/bccd345f37d9223171375909f0efc19c6e2c652516c3eb8eeffe8a7e9a90c4ac.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
argtouchecovidnfchotpotdbpclimatefeverscifactscidocsfiqaAVG.
FLAN original59.434.575.633.461.438.0(24.0*)77.065.018.446.248.5
FLAN w/o NQ and Quora58.833.370.233.761.734.4(23.5*)76.263.818.343.047.0
", + "bbox": [ + 173, + 102, + 821, + 147 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/6b0deda12c0ebe78b5a86065c0e4184c528a916a77041c1ebe8bcd8119ad8c08.jpg", + "image_caption": [ + "Table 3: Impact of different FLAN versions. This study uses Fever prompt for Climate Fever (§4.3) ", + "Figure 3: Top first word distribution on queries generated from different models in the ArguAna dataset. Left (a)(b)(c): Compare gold queries (a) and generated queries (b)(c). Queries generated by few-shot models has closer distribution to the gold queries, while the NQ-QGen queries are mostly questions. Right (d): The few shot FLAN can generate diverse queries even though there are only 4 examples in the prompt. Statistics of more datasets are available in the Appendix (Figure 4). " + ], + "image_footnote": [], + "bbox": [ + 184, + 185, + 816, + 258 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 366, + 825, + 436 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Does few-shot always improve over zero-shot? As shown in Table 2, few-shot PROMPTAGATOR almost always outperforms zero-shot PROMPTAGATOR except for Climate-FEVER. The original Climate-FEVER dataset uses one of three tags to annotate a query-document pair, namely “supports”, “refutes”, or “not enough info”. However, BEIR treats all these three annotations as relevant, which is problematic. Using query-document pairs annotated “not enough info” in our prompt could be detrimental to generation quality. Therefore, we tried switching to FEVER’s few-shot prompt, as the two datasets share same corpus and similar search intents. With the better annotated examples, few-shot PROMPTAGATOR indeed surpass zero-shot. This result provides some evidence that low quality few-shot examples negatively affect PROMPTAGATOR. ", + "bbox": [ + 173, + 443, + 825, + 569 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Impact of FLAN Versions FLAN was trained on a collection of datasets which have some overlap with BEIR; specifically, it includes Natural Questions (NQ) and Quora. It was not trained on querydocument pairs from NQ or Quora; however, in order to determine whether the inclusion of this data biased the results on the final retrieval evaluation, we designed an additional ablation experiment. Following the original FLAN recipe (Wei et al., 2022a), we trained an additional LLM excluding both the NQ and Quora datasets. Table 4 shows the results. While the accuracy drops slightly, the overall performance still outperform prior retrievers. ", + "bbox": [ + 173, + 575, + 825, + 672 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Qualitative Analysis In order to understand the advantages of few-shot PROMPTAGATOR, we analyze the distribution of queries generated by different query generation methods for ArguAna in Figure 3. To easily see the differences, for each distribution we plot the histogram of each query’s first word. Note that the distribution of few-shot PROMPTAGATOR (Fig. 3b) is much closer to the real distribution (Fig. 3a) while the NQ-QGen (Fig. 3c) mostly generated questions even when the queries in this task are generally arguments, not questions. More examples are showcased in Table 6 in the Appendix. ", + "bbox": [ + 173, + 689, + 825, + 772 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Use few-shot examples directly To study the effect of using the few-shot examples directly, we conduct a study of fine-tuning task-specific models using the GTR dual encoder (110M) (Ni et al., 2021) with the few-shot examples in appendix A. As expected, it does not provide a large amount of impact for GTR base, where the average nDCG decreases from 40.4 to 38.7. ", + "bbox": [ + 174, + 787, + 825, + 843 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5 RELATED WORK ", + "text_level": 1, + "bbox": [ + 176, + 863, + 344, + 880 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Neural Retrieval and Reranking Models The majority of neural retrievers today employ a dual encoder architecture that encodes queries and documents independently into dense vectors and retrieves documents using maximum inner product search (MIPS). Recent research has primarily focused on the following aspects: developing better pre-training tasks (Lee et al., 2019; Chang et al., 2020; Izacard et al., 2022a; Gao & Callan, 2021; Oguz et al., 2022), improving contrastive negatives (Qu et al., 2021; Xiong et al., 2021; Lu et al., 2021), and improving generalization across different domains (Thakur et al., 2021; Ren et al., 2022). ", + "bbox": [ + 174, + 895, + 825, + 924 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 103, + 825, + 174 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Although dual encoders enable fast retrieval, their expressivity is limited due to the fact that their score is just a dot-product between a query vector and a document vector. A common solution is to use a cross-attention model to rerank retrieved candidates (Nogueira & Cho, 2019; Nogueira et al., 2020), as cross-attention rerankers can explicitly model the interaction between query and document tokens. Distilling cross-attention models into dual encoders has been effective in closing the gap between the two (Hofstätter et al., 2020; Ren et al., 2021; Reddi et al., 2021; Zhang et al., 2022). ", + "bbox": [ + 174, + 180, + 825, + 263 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Neural Retrieval with Fine-Grained Interactions An alternative for bridging dense retrievers and cross-attention models is to allow some amount of fine-grained query-document interactions in the retriever. Humeau et al. (2020) and Luan et al. (2021) represent and retrieve queries and documents with multiple vectors instead of a single vector. ColBERT (Khattab & Zaharia, 2020), COIL (Gao et al., 2021a) and SPLADE (Formal et al., 2021) take this further by using token-level interactions between queries and documents. Because these models are not just modeling a dot product, MIPS algorithms cannot be used directly. Hence, these models usually have much higher inference/serving cost compared to dual encoders. ", + "bbox": [ + 174, + 270, + 825, + 382 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Prompt-based Query Generation The idea of using prompted LLMs for query generation has previously been proposed for improving retrieval reranking. UPR (Sachan et al., 2022) proposed to use prompted LLMs to rerank passages directly. InPars (Bonifacio et al., 2022) is probably the most closely related work to ours. They proposed to use few-shot prompting with GPT-3 to generate synthetic data for training a T5-based reranker. Though InPars was tested on multiple retrieval datasets, they used a task-independent prompt constructed from MS MARCO and did not explore task-specific few-shot learning. They also focused exclusively on reranking, whereas we also address full-scale retrieval. ", + "bbox": [ + 174, + 388, + 825, + 500 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Few-shot Learning pre-trained LLMs have significantly advanced few-shot learning, thanks to prompting strategies such as in-context learning and instruction-prompting (Brown et al., 2020; Wei et al., 2022b). Some approaches fine-tune LLMs specifically for few-shot learning (Schick & Schütze, 2021a;b;c; Gao et al., 2021b; Logan IV et al., 2022; Izacard et al., 2022b) while others do not (Brown et al., 2020; Bonifacio et al., 2022). We use LLMs as few-shot data generators — this form of data augmentation via few-shot example extrapolation is similar to Lee et al. (2021). ", + "bbox": [ + 174, + 507, + 825, + 592 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "6 CONCLUSION AND DISCUSSIONS ", + "text_level": 1, + "bbox": [ + 176, + 630, + 478, + 646 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In this paper, we present PROMPTAGATOR, a novel approach to few-shot retrieval. We showed that it is possible to create task-specific retrievers and rerankers with only a few annotated examples. The few-shot examples, amplified by prompt-based LLM query generation, simplifies the complexity of training neural retrievers for new tasks and leads to promising performance gains. It hopefully inspires future research towards generalizable retrieval systems that can seamlessly and efficiently adapt to many tasks. ", + "bbox": [ + 174, + 672, + 825, + 756 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "While we demonstrate that LLM-based query generation can be very effective, many questions remain. One of the key issue requiring further investigation is on the generated data efficiency. We have not yet explored exactly how many query-document pairs are needed for each task, or how to use these generated examples more efficiently. Another issue is the sensitivity of the final retriever’s performance with respect to the prompt. Finally, we would like to draw a connection from PROMPTAGATOR to distillation, as the final dual encoders indirectly “learn” from the LLM. Analyzing the headroom and understanding how we can better transfer knowledge from LLMs to retrievers would be a critical topic for the future. ", + "bbox": [ + 174, + 762, + 825, + 875 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "As mentioned in $\\ S 6$ , PROMPTAGATOR can be viewed as distilling LLM to standard-sized dual encoders via prompt-based query generation. While the distillation process is computationally expensive, it significantly reduces cost for inference. ", + "bbox": [ + 176, + 882, + 823, + 924 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "ACKNOWLEDGEMENTS ", + "text_level": 1, + "bbox": [ + 176, + 103, + 367, + 117 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We thank Kenton Lee, Tom Kwiatkowski, and Daniel Gillick for technical discussion and providing feedback on our manuscript. We thank Alex Salcianu for developing a bulk inference pipeline for large language models. ", + "bbox": [ + 174, + 133, + 825, + 176 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 176, + 199, + 285, + 214 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Chris Alberti, Daniel Andor, Emily Pitler, Jacob Devlin, and Michael Collins. Synthetic QA corpora generation with roundtrip consistency. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 6168–6173, Florence, Italy, July 2019. Association for Computational Linguistics. doi: 10.18653/v1/P19-1620. URL https://aclanthology. org/P19-1620. ", + "bbox": [ + 174, + 223, + 826, + 292 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Luiz Bonifacio, Hugo Abonizio, Marzieh Fadaee, and Rodrigo Nogueira. Inpars: Unsupervised dataset generation for information retrieval. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ’22, pp. 2387–2392, New York, NY, USA, 2022. Association for Computing Machinery. ISBN 9781450387323. doi: 10.1145/3477495.3531863. URL https://doi.org/10.1145/3477495.3531863. ", + "bbox": [ + 173, + 305, + 826, + 376 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, and Dario Amodei. Language models are few-shot learners. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin (eds.), Advances in Neural Information Processing Systems, volume 33, pp. 1877–1901. Curran Associates, Inc., 2020. URL https://proceedings.neurips. cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf. ", + "bbox": [ + 174, + 387, + 828, + 513 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Wei-Cheng Chang, Felix X. Yu, Yin-Wen Chang, Yiming Yang, and Sanjiv Kumar. Pre-training tasks for embedding-based large-scale retrieval. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id $=$ rkg-mA4FDr. ", + "bbox": [ + 174, + 525, + 826, + 568 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "A. P. Dempster, N. M. Laird, and D. B. Rubin. Maximum likelihood from incomplete data via the em algorithm. JOURNAL OF THE ROYAL STATISTICAL SOCIETY, SERIES B, 39(1):1–38, 1977. ", + "bbox": [ + 169, + 580, + 823, + 608 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Thibault Formal, Carlos Lassance, Benjamin Piwowarski, and Stéphane Clinchant. SPLADE v2: Sparse lexical and expansion model for information retrieval. CoRR, abs/2109.10086, 2021. URL https://arxiv.org/abs/2109.10086. ", + "bbox": [ + 174, + 621, + 825, + 662 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Luyu Gao and Jamie Callan. Condenser: a pre-training architecture for dense retrieval. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp. 981–993, Online and Punta Cana, Dominican Republic, November 2021. Association for Computational Linguistics. doi: 10.18653/v1/2021.emnlp-main.75. URL https://aclanthology.org/ 2021.emnlp-main.75. ", + "bbox": [ + 173, + 675, + 826, + 744 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Luyu Gao, Zhuyun Dai, and Jamie Callan. COIL: Revisit exact lexical match in information retrieval with contextualized inverted list. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 3030–3042, Online, June 2021a. Association for Computational Linguistics. doi: 10.18653/v1/ 2021.naacl-main.241. URL https://aclanthology.org/2021.naacl-main.241. ", + "bbox": [ + 173, + 757, + 826, + 828 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Tianyu Gao, Adam Fisch, and Danqi Chen. Making pre-trained language models better few-shot learners. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pp. 3816–3830, Online, August 2021b. Association for Computational Linguistics. doi: 10.18653/v1/2021.acl-long.295. URL https://aclanthology.org/2021.acl-long. 295. ", + "bbox": [ + 174, + 839, + 826, + 922 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Faegheh Hasibi, Fedor Nikolaev, Chenyan Xiong, Krisztian Balog, Svein Erik Bratsberg, Alexander Kotov, and Jamie Callan. Dbpedia-entity v2: A test collection for entity search. In Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ’17, pp. 1265–1268, New York, NY, USA, 2017. ISBN 9781450350228. URL https://doi.org/10.1145/3077136.3080751. ", + "bbox": [ + 174, + 103, + 826, + 174 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Sebastian Hofstätter, Sophia Althammer, Michael Schröder, Mete Sertkan, and Allan Hanbury. Improving efficient neural ranking models with cross-architecture knowledge distillation. $A r X i \\nu$ , abs/2010.02666, 2020. URL https://arxiv.org/abs/2010.02666. ", + "bbox": [ + 174, + 183, + 826, + 226 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Samuel Humeau, Kurt Shuster, Marie-Anne Lachaux, and Jason Weston. Poly-encoders: Architectures and pre-training strategies for fast and accurate multi-sentence scoring. In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020. OpenReview.net, 2020. URL https://openreview.net/forum?id $=$ SkxgnnNFvH. ", + "bbox": [ + 174, + 234, + 826, + 291 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Gautier Izacard, Mathilde Caron, Lucas Hosseini, Sebastian Riedel, Piotr Bojanowski, Armand Joulin, and Edouard Grave. Unsupervised dense information retrieval with contrastive learning. Transactions on Machine Learning Research, 2022a. URL https://openreview.net/ forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ jKN1pXi7b0. ", + "bbox": [ + 173, + 300, + 826, + 357 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi-Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave. Few-shot learning with retrieval augmented language models, 2022b. URL https://arxiv.org/abs/2208.03299. ", + "bbox": [ + 176, + 366, + 823, + 409 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. Dense passage retrieval for open-domain question answering. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 6769–6781, Online, November 2020. Association for Computational Linguistics. doi: 10.18653/v1/ 2020.emnlp-main.550. URL https://aclanthology.org/2020.emnlp-main.550. ", + "bbox": [ + 174, + 417, + 826, + 488 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Omar Khattab and Matei Zaharia. ColBERT: Efficient and effective passage search via contextualized late interaction over BERT. In Jimmy Huang, Yi Chang, Xueqi Cheng, Jaap Kamps, Vanessa Murdock, Ji-Rong Wen, and Yiqun Liu (eds.), Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval, SIGIR 2020, Virtual Event, China, July 25-30, 2020, pp. 39–48. ACM, 2020. doi: 10.1145/3397271.3401075. URL https: //doi.org/10.1145/3397271.3401075. ", + "bbox": [ + 173, + 497, + 826, + 582 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, Kristina Toutanova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew M. Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. Natural questions: A benchmark for question answering research. Transactions of the Association for Computational Linguistics, 7:452–466, March 2019. doi: 10.1162/tacl_a_00276. URL https://aclanthology.org/Q19-1026. ", + "bbox": [ + 174, + 590, + 826, + 675 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Kenton Lee, Ming-Wei Chang, and Kristina Toutanova. Latent retrieval for weakly supervised open domain question answering. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 6086–6096, Florence, Italy, July 2019. Association for Computational Linguistics. doi: 10.18653/v1/P19-1612. URL https://aclanthology.org/P19-1612. ", + "bbox": [ + 174, + 684, + 826, + 741 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Kenton Lee, Kelvin Guu, Luheng He, Tim Dozat, and Hyung Won Chung. Neural data augmentation via example extrapolation. CoRR, abs/2102.01335, 2021. URL https://arxiv.org/abs/ 2102.01335. ", + "bbox": [ + 173, + 750, + 826, + 792 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Patrick Lewis, Yuxiang Wu, Linqing Liu, Pasquale Minervini, Heinrich Küttler, Aleksandra Piktus, Pontus Stenetorp, and Sebastian Riedel. PAQ: 65 million probably-asked questions and what you can do with them. Transactions of the Association for Computational Linguistics, 9:1098–1115, 2021. doi: 10.1162/tacl_a_00415. URL https://aclanthology.org/2021.tacl-1. 65. ", + "bbox": [ + 174, + 801, + 826, + 871 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Robert Logan IV, Ivana Balazevic, Eric Wallace, Fabio Petroni, Sameer Singh, and Sebastian Riedel. Cutting down on prompts and parameters: Simple few-shot learning with language models. In Findings of the Association for Computational Linguistics: ACL 2022, pp. 2824–2835, Dublin, ", + "bbox": [ + 176, + 881, + 825, + 924 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ireland, May 2022. Association for Computational Linguistics. doi: 10.18653/v1/2022.findings-acl. \n222. URL https://aclanthology.org/2022.findings-acl.222. ", + "bbox": [ + 181, + 103, + 825, + 133 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Jing Lu, Gustavo Hernandez Abrego, Ji Ma, Jianmo Ni, and Yinfei Yang. Multi-stage training with improved negative contrast for neural passage retrieval. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp. 6091–6103, Online and Punta Cana, Dominican Republic, November 2021. Association for Computational Linguistics. doi: 10.18653/v1/ 2021.emnlp-main.492. URL https://aclanthology.org/2021.emnlp-main.492. ", + "bbox": [ + 174, + 140, + 826, + 210 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Yi Luan, Jacob Eisenstein, Kristina Toutanova, and Michael Collins. Sparse, Dense, and Attentional Representations for Text Retrieval. Transactions of the Association for Computational Linguistics, 9:329–345, 04 2021. doi: 10.1162/tacl_a_00369. URL https://doi.org/10.1162/ tacl_a_00369. ", + "bbox": [ + 173, + 218, + 826, + 275 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Ji Ma, Ivan Korotkov, Yinfei Yang, Keith Hall, and Ryan McDonald. Zero-shot neural passage retrieval via domain-targeted synthetic question generation. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pp. 1075–1088, Online, April 2021. Association for Computational Linguistics. doi: 10.18653/v1/ 2021.eacl-main.92. URL https://aclanthology.org/2021.eacl-main.92. ", + "bbox": [ + 173, + 282, + 826, + 353 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Macedo Maia, Siegfried Handschuh, André Freitas, Brian Davis, Ross McDermott, Manel Zarrouk, and Alexandra Balahur. Www’18 open challenge: financial opinion mining and question answering. In Companion proceedings of the the web conference 2018, pp. 1941–1942, 2018. URL https: //doi.org/10.1145/3184558.3192301. ", + "bbox": [ + 173, + 361, + 826, + 417 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Arvind Neelakantan, Tao Xu, Raul Puri, Alec Radford, Jesse Michael Han, Jerry Tworek, Qiming Yuan, Nikolas A. Tezak, Jong Wook Kim, Chris Hallacy, Johannes Heidecke, Pranav Shyam, Boris Power, Tyna Eloundou Nekoul, Girish Sastry, Gretchen Krueger, David P. Schnurr, Felipe Petroski Such, Kenny Sai-Kin Hsu, Madeleine Thompson, Tabarak Khan, Toki Sherbakov, Joanne Jang, Peter Welinder, and Lilian Weng. Text and code embeddings by contrastive pre-training. ArXiv, abs/2201.10005, 2022. URL https://arxiv.org/abs/2201.10005. ", + "bbox": [ + 174, + 425, + 826, + 510 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Tri Nguyen, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Majumder, and Li Deng. Ms marco: A human generated machine reading comprehension dataset. In CoCo@NIPS, 2016. URL http://ceur-ws.org/Vol-1773/CoCoNIPS_2016_paper9.pdf. ", + "bbox": [ + 174, + 517, + 825, + 560 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Jianmo Ni, Chen Qu, Jing Lu, Zhuyun Dai, Gustavo Hernández Ábrego, Ji Ma, Vincent Y. Zhao, Yi Luan, Keith B. Hall, Ming-Wei Chang, and Yinfei Yang. Large dual encoders are generalizable retrievers. CoRR, abs/2112.07899, 2021. URL https://arxiv.org/abs/2112.07899. ", + "bbox": [ + 174, + 568, + 826, + 611 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Rodrigo Nogueira and Kyunghyun Cho. Passage re-ranking with bert. arXiv, 2019. URL https: //arxiv.org/abs/1901.04085. ", + "bbox": [ + 169, + 618, + 825, + 647 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Rodrigo Nogueira, Zhiying Jiang, Ronak Pradeep, and Jimmy Lin. Document ranking with a pretrained sequence-to-sequence model. In Trevor Cohn, Yulan He, and Yang Liu (eds.), Findings of the Association for Computational Linguistics: EMNLP 2020, Online Event, 16-20 November 2020, volume EMNLP 2020 of Findings of ACL, pp. 708–718. Association for Computational Linguistics, 2020. doi: 10.18653/v1/2020.findings-emnlp.63. URL https://doi.org/10. 18653/v1/2020.findings-emnlp.63. ", + "bbox": [ + 173, + 655, + 826, + 739 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Barlas Oguz, Kushal Lakhotia, Anchit Gupta, Patrick Lewis, Vladimir Karpukhin, Aleksandra Piktus, Xilun Chen, Sebastian Riedel, Scott Yih, Sonal Gupta, and Yashar Mehdad. Domain-matched pre-training tasks for dense retrieval. In Findings of the Association for Computational Linguistics: NAACL 2022, pp. 1524–1534, Seattle, United States, July 2022. Association for Computational Linguistics. doi: 10.18653/v1/2022.findings-naacl.114. URL https://aclanthology. org/2022.findings-naacl.114. ", + "bbox": [ + 174, + 747, + 826, + 832 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Yingqi Qu, Yuchen Ding, Jing Liu, Kai Liu, Ruiyang Ren, Wayne Xin Zhao, Daxiang Dong, Hua Wu, and Haifeng Wang. RocketQA: An optimized training approach to dense passage retrieval for open-domain question answering. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 5835–5847, June 2021. doi: 10.18653/v1/2021.naacl-main.466. URL https: //aclanthology.org/2021.naacl-main.466. ", + "bbox": [ + 176, + 840, + 826, + 924 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Sashank J. Reddi, Rama Kumar Pasumarthi, Aditya Krishna Menon, Ankit Singh Rawat, Felix X. Yu, Seungyeon Kim, Andreas Veit, and Sanjiv Kumar. Rankdistil: Knowledge distillation for ranking. In AISTATS, pp. 2368–2376, 2021. URL http://proceedings.mlr.press/ v130/reddi21a.html. ", + "bbox": [ + 174, + 103, + 826, + 159 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Ruiyang Ren, Shangwen Lv, Yingqi Qu, Jing Liu, Wayne Xin Zhao, QiaoQiao She, Hua Wu, Haifeng Wang, and Ji-Rong Wen. PAIR: Leveraging passage-centric similarity relation for improving dense passage retrieval. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pp. 2173–2183, Online, August 2021. Association for Computational Linguistics. doi: 10.18653/ v1/2021.findings-acl.191. URL https://aclanthology.org/2021.findings-acl. 191. ", + "bbox": [ + 173, + 169, + 826, + 252 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Ruiyang Ren, Yingqi Qu, Jing Liu, Wayne Xin Zhao, Qifei Wu, Yuchen Ding, Hua Wu, Haifeng Wang, and Ji-Rong Wen. A thorough examination on zero-shot dense retrieval, 2022. URL https://arxiv.org/abs/2204.12755. ", + "bbox": [ + 174, + 261, + 826, + 304 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Guilherme Moraes Rosa, Luiz Henrique Bonifacio, Vitor Jeronymo, Hugo Abonizio, Marzieh Fadaee, Roberto de Alencar Lotufo, and Rodrigo Nogueira. No parameter left behind: How distillation and model size affect zero-shot retrieval. CoRR, abs/2206.02873, 2022. doi: 10.48550/arXiv.2206. 02873. URL https://doi.org/10.48550/arXiv.2206.02873. ", + "bbox": [ + 174, + 314, + 826, + 369 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Devendra Singh Sachan, Mike Lewis, Mandar Joshi, Armen Aghajanyan, Wen-tau Yih, Joelle Pineau, and Luke Zettlemoyer. Improving passage retrieval with zero-shot question generation. arXiv, 2022. URL https://arxiv.org/abs/2204.07496. ", + "bbox": [ + 176, + 378, + 825, + 421 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Victor Sanh, Albert Webson, Colin Raffel, Stephen Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Arun Raja, Manan Dey, M Saiful Bari, Canwen Xu, Urmish Thakker, Shanya Sharma Sharma, Eliza Szczechla, Taewoon Kim, Gunjan Chhablani, Nihal V. Nayak, Debajyoti Datta, Jonathan Chang, Mike Tian-Jian Jiang, Han Wang, Matteo Manica, Sheng Shen, Zheng Xin Yong, Harshit Pandey, Rachel Bawden, Thomas Wang, Trishala Neeraj, Jos Rozen, Abheesht Sharma, Andrea Santilli, Thibault Févry, Jason Alan Fries, Ryan Teehan, Teven Le Scao, Stella Biderman, Leo Gao, Thomas Wolf, and Alexander M. Rush. Multitask prompted training enables zero-shot task generalization. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net, 2022. URL https://openreview.net/forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ 9Vrb9D0WI4. ", + "bbox": [ + 174, + 430, + 826, + 569 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Keshav Santhanam, Omar Khattab, Jon Saad-Falcon, Christopher Potts, and Matei Zaharia. ColBERTv2: Effective and efficient retrieval via lightweight late interaction. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 3715–3734, July 2022. URL https://aclanthology.org/2022.naacl-main.272. ", + "bbox": [ + 174, + 579, + 826, + 648 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Timo Schick and Hinrich Schütze. Exploiting cloze-questions for few-shot text classification and natural language inference. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pp. 255–269, Online, April 2021a. Association for Computational Linguistics. doi: 10.18653/v1/2021.eacl-main.20. URL https://aclanthology.org/2021.eacl-main.20. ", + "bbox": [ + 174, + 657, + 825, + 728 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Timo Schick and Hinrich Schütze. It’s not just size that matters: Small language models are also few-shot learners. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 2339–2352, Online, June 2021b. Association for Computational Linguistics. doi: 10.18653/v1/2021.naacl-main. 185. URL https://aclanthology.org/2021.naacl-main.185. ", + "bbox": [ + 174, + 737, + 826, + 808 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Timo Schick and Hinrich Schütze. Few-shot text generation with natural language instructions. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp. 390–402, Online and Punta Cana, Dominican Republic, November 2021c. Association for Computational Linguistics. doi: 10.18653/v1/2021.emnlp-main.32. URL https: //aclanthology.org/2021.emnlp-main.32. ", + "bbox": [ + 174, + 816, + 826, + 887 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Siamak Shakeri, Cicero Nogueira dos Santos, Henghui Zhu, Patrick Ng, Feng Nan, Zhiguo Wang, Ramesh Nallapati, and Bing Xiang. End-to-end synthetic data generation for domain adaptation of question answering systems. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 5445–5460. Association for Computational Linguistics, 2020. URL https://aclanthology.org/2020.emnlp-main.439. ", + "bbox": [ + 176, + 895, + 825, + 924 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "", + "bbox": [ + 187, + 103, + 823, + 146 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna Gurevych. BEIR: A heterogeneous benchmark for zero-shot evaluation of information retrieval models. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2), 2021. URL https://openreview.net/forum?id ${ . } = { }$ wCu6T5xFjeJ. ", + "bbox": [ + 176, + 154, + 826, + 210 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, YaGuang Li, Hongrae Lee, Huaixiu Steven Zheng, Amin Ghafouri, Marcelo Menegali, Yanping Huang, Maxim Krikun, Dmitry Lepikhin, James Qin, Dehao Chen, Yuanzhong Xu, Zhifeng Chen, Adam Roberts, Maarten Bosma, Yanqi Zhou, Chung-Ching Chang, Igor Krivokon, Will Rusch, Marc Pickett, Kathleen S. Meier-Hellstern, Meredith Ringel Morris, Tulsee Doshi, Renelito Delos Santos, Toju Duke, Johnny Soraker, Ben Zevenbergen, Vinodkumar Prabhakaran, Mark Diaz, Ben Hutchinson, Kristen Olson, Alejandra Molina, Erin Hoffman-John, Josh Lee, Lora Aroyo, Ravi Rajakumar, Alena Butryna, Matthew Lamm, Viktoriya Kuzmina, Joe Fenton, Aaron Cohen, Rachel Bernstein, Ray Kurzweil, Blaise Aguera-Arcas, Claire Cui, Marian Croak, Ed H. Chi, and Quoc Le. Lamda: Language models for dialog applications. CoRR, abs/2201.08239, 2022. URL https://arxiv.org/abs/2201. 08239. ", + "bbox": [ + 176, + 219, + 826, + 385 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "James Thorne, Andreas Vlachos, Christos Christodoulopoulos, and Arpit Mittal. FEVER: a large-scale dataset for fact extraction and VERification. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pp. 809–819, New Orleans, Louisiana, June 2018. Association for Computational Linguistics. doi: 10.18653/v1/N18-1074. URL https://aclanthology.org/N18-1074. ", + "bbox": [ + 173, + 395, + 826, + 479 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Kexin Wang, Nandan Thakur, Nils Reimers, and Iryna Gurevych. GPL: Generative pseudo labeling for unsupervised domain adaptation of dense retrieval. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 2345–2360, Seattle, United States, July 2022. Association for Computational Linguistics. doi: 10.18653/v1/2022.naacl-main.168. URL https://aclanthology.org/ 2022.naacl-main.168. ", + "bbox": [ + 174, + 487, + 826, + 570 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Jason Wei, Maarten Paul Bosma, Vincent Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew Mingbo Dai, and Quoc V. Le. Finetuned language models are zero-shot learners. In International Conference on Learning Representations, 2022a. URL https://openreview. net/forum?id ${ . } = { }$ gEZrGCozdqR. ", + "bbox": [ + 174, + 580, + 826, + 636 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus. Emergent abilities of large language models. Transactions on Machine Learning Research, 2022b. URL https://openreview.net/forum? id $=$ yzkSU5zdwD. ", + "bbox": [ + 173, + 645, + 826, + 715 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Lee Xiong, Chenyan Xiong, Ye Li, Kwok-Fung Tang, Jialin Liu, Paul N. Bennett, Junaid Ahmed, and Arnold Overwijk. Approximate nearest neighbor negative contrastive learning for dense text retrieval. In International Conference on Learning Representations, 2021. URL https: //openreview.net/forum?id $\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =$ zeFrfgyZln. ", + "bbox": [ + 173, + 723, + 826, + 780 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William Cohen, Ruslan Salakhutdinov, and Christopher D. Manning. HotpotQA: A dataset for diverse, explainable multi-hop question answering. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pp. 2369–2380, Brussels, Belgium, October-November 2018. Association for Computational Linguistics. doi: 10.18653/v1/D18-1259. URL https://aclanthology.org/D18-1259. ", + "bbox": [ + 173, + 789, + 826, + 859 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Wen-tau Yih, Kristina Toutanova, John C. Platt, and Christopher Meek. Learning discriminative projections for text similarity measures. In Proceedings of the Fifteenth Conference on Computational Natural Language Learning, pp. 247–256, Portland, Oregon, USA, June 2011. Association for Computational Linguistics. URL https://aclanthology.org/W11-0329. ", + "bbox": [ + 176, + 867, + 825, + 924 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Hang Zhang, Yeyun Gong, Yelong Shen, Jiancheng Lv, Nan Duan, and Weizhu Chen. Adversarial retriever-ranker for dense text retrieval. In International Conference on Learning Representations, 2022. URL https://openreview.net/forum?id $=$ MR7XubKUFB. ", + "bbox": [ + 178, + 103, + 823, + 146 + ], + "page_idx": 14 + }, + { + "type": "table", + "img_path": "images/2a622920e3d3b733604a54f0ce2a62c04b5d743f465b2a2a9020ca7dc28aadea.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
argtouchécovidnfchotpotdbpclimatefeverscifactscidocsfiqaAVG.
GTR-base51.120.553.930.853.534.7(24.1*)66.060.014.934.940.4
GTR-base with 8 examples51.924.156.230.523.435.4(25.4*)68.260.114.835.338.7
", + "bbox": [ + 173, + 102, + 823, + 146 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Table 4: Directly using eight examples might not improve the results. To showcase this, we fine-tuned a GTR-based model with the same few-shot examples used for tuning. ", + "bbox": [ + 173, + 157, + 821, + 185 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "COMPUTE USAGE AND ENVIRONMENTAL IMPACT ", + "text_level": 1, + "bbox": [ + 174, + 217, + 588, + 233 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We used the 137B FLAN, which is based on LaMDA (Thoppilan et al., 2022). LaMDA was pretrained on a large corpus consisting of 1.56T words, costing 451 MWh energy and $2 5 . 2 \\mathrm { t C O } 2 \\mathrm { e }$ carbon footprint. In PROMPTAGATOR, we generated 29.23M queries $^ { * } 2$ prompts $= 5 8 . 4 6 \\mathrm { M }$ queries, for a total of 610M words. ", + "bbox": [ + 173, + 251, + 826, + 308 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "A FINE-TUNING GTR WITH FEW-SHOT EXAMPLES ", + "text_level": 1, + "bbox": [ + 174, + 334, + 611, + 351 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We study the effect of directly adding few-shot (no more than 8 examples) on top of a 110M dual encoder, GTR-base Ni et al. (2021). To maximum the utilities of the 8 examples, we associated each positive example with 32 random negative examples from the target corpus to construct a batch, and ran 50 steps of fine-tuning for each task. As expected, the 8 examples did not provide a large amount of impact for GTR base, where the average ndcg go from 40.4 to 38.7 after fine-tuning using few-shot examples. ", + "bbox": [ + 173, + 369, + 825, + 454 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "B ANALYSIS ON PROMPTS ", + "text_level": 1, + "bbox": [ + 176, + 481, + 408, + 496 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Table 5 shows the list of prompt templates on different BEIR datasets. In order to further analysis the difference between zero-shot and few-shot prompts, we compare the few-shot and zero-shot generated queries given the same paragraph, randomly sampled from three datasets in Table 6. We observe that in general, the few-shot generated queries are closer to the original queries, while zero-shot queries are mostly questions. For example, in the ArguAna dataset, the few-shot queries are in general longer and more claim-like. In contrary, the zero-shot queries are most short question-like queries. Interestingly, for the HotpotQA dataset, even though both few-shot and zero-shot queries are generating questions-like queries, few-shot queries sometimes generate multi-hop questions, while zero-shot mostly generates single-hop questions. We further conduct first word distribution across different generation models for all datasets in Figure 4. ", + "bbox": [ + 173, + 515, + 825, + 655 + ], + "page_idx": 15 + }, + { + "type": "table", + "img_path": "images/c7d0140b3c5270503bce236139785a1d1cb7ba02047f95016c115ee5a6951b3d.jpg", + "table_caption": [ + "Table 5: Prompt template for each dataset. " + ], + "table_footnote": [], + "table_body": "
DatasetPrompt
ArguAna0 Argument: passage X 1 Counter argument: query X
FiQA0 passage X 1 query X
HotpotQAl0 Evidence: passage X 1 Vexed question: query X
DBPedia-Entity0 entity: passage X 1 query: query X
NFCorpusl0 Article: passage X 1 Query: query X
Toucheé-2020|0 passage X 1 Debate: query X
TREC-Covidl0 passage X 1 Question: query X
SciFactl0 passage X 1 Finding: query X
SCIDOCSl0 passage X 1 The passage is about query X
FEVER|0 passage X 1 Is it true that query X
", + "bbox": [ + 174, + 674, + 821, + 886 + ], + "page_idx": 15 + }, + { + "type": "table", + "img_path": "images/7523c20451a4d992dd99f0a9c9156bfffa39d4dab07823282a702a6743df62f1.jpg", + "table_caption": [ + "Table 6: Few-shot and zero-shot generated queries randomly sampled from ArguAna, FiQA and HotpotQA dataset. " + ], + "table_footnote": [], + "table_body": "
ParagraphFew-shotZero-shotAnalysis In ArguAna dataset,
ArguAna: Banning skin whiten- ing on such a basis also requires the acceptance of the racial over- tones.Some form of tan is popular in many societies of people of Eu- ropean ethnic origins- that is not a racial matter, it is more based on economic social perceptions (that of holidays to warmer climates). Ascribing a racial element to ev- erything to do with skin tone is at best a lazy analysis.Irrespective of issues of race and perceptions of ethnic origins,and its intersec- tion with beauty standards,some people will be given advantages in life due to their appearance.Ban- ninga certain form of cosmetic, even if it can have some racial and ethnic undertones,won't change that.Query 1:1 Banning skin whitening on such a ba- sis also requires the accep- tance of the racial overtones. However,the racial over- tones are simply a side ef- fect of the product being used incorrectly in order to attain a lighter skin tone than they are naturally born with. The product itself is not racist, but rather the peo- ple who use it. For exam- ple,a person of Asian de- scent may use a skin whiten- ing cream to attain a lighter skin tone so that they may fit into a more white-dominant society. The cream itself is not racist,but rather the fact that the person has used it to attain a lighter skin tone so that they may fit into a moreQuery 1: What did they try to ban? Query 2: Is skin whitening racist? Query 3:Banning skin whiten- ing on such a basis also re- quires the acceptance of the racial overtones.Some form of tan is popular in many so- cieties of people of European ethnic origins- that is not a racial matter, it is more based on economic social perceptions (that of holidays to warmer cli- mates). Query 4: What are some of the racial overtones in skin- whitening? Query 5: Are tans a sign of af- fluence?the few-shot ex- amplesaremore statement-likeand are longer than zero-shot.
white-dominant society. Touché-2020:Psychologically, diet-wiseandacademically. Housewives can be both educated and uneducated but the fact that they spend more time with their children means that they put in a greater effort and since they are not preoccupied with their job(read:cellphone); they_pay the close attention needed to keep their children out of trouble. Think of housewives as inspectors in your home...Query 1: Housewives are better parents Query 2: Should house- wives be considered better parents?Query 1: diet-wise and aca- demically. Query 2: People want to be relaxed all the time(At home AND at work) that is why all the research on work-life bal- ance. Query 3: has shown that home- makers are more relaxed, less stressed,healthier,happier - the list goes on.The few-shot gener- atesargument-like queries that are more controversial, while zero-shot generate randomstatements that sometimes even have grammar problems.
HotpotQA: The Yung See San Fong House,or Young's Home in the Heart of the Hills, is a house in Los Gatos in the U.S.state of California.It is a combination of Asian decorative motifs and pagoda roofs together with West- ern massing and layout.It was completed in 1917 forRuth Com- fort Mitchell Young,a writer,and her husband, Sanborn Young,a gentleman farmer,conservationist and later California State Senator.Query 1: Which author was the husband of Ruth Com- fort Mitchell Young? Query 2: Which is known as the Yount-Lee House Query 3:Who is the archi- tect of the Yung See San Fong House? Query 4:The Yung See San Fong House,or Young's Home in the Heart of the Hills;isa house in which U.S. state?Query 1: Yung See San Fong House was completed in 1917 for Ruth Comfort Mitchell Young,a writer,and her hus- band, Sanborn Young. Query 2: How old is the Yung See San Fong House? Query 3: Yung See San Fong House was completed in 1917 for Ruth Comfort Mitchell Young,a writer,and her hus- band, Sanborn Young.Few-shot examples sometimes will create multihop questions indicated in blue, whichrarelyhap- pensin zero-shot examples.
", + "bbox": [ + 178, + 147, + 916, + 839 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "C DETAILED IMPLEMENTATION ", + "text_level": 1, + "bbox": [ + 176, + 102, + 450, + 118 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Figure 5 shows the overall process of PROMPTAGATOR $^ { + + }$ , the details of which are in Section 3. ", + "bbox": [ + 171, + 135, + 789, + 150 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Our cross attention reranker is a listwise model based on T5. Specifically, it takes a list of documents given a query as the input. We represent each query-document pair as “Query: {q} Document: $\\{ \\mathrm { d } \\} ^ { \\prime \\prime }$ and feed it into the encoder of a T5 model. We then apply a projection layer on the output encodings of the first token and use the output as the ranking score. We optimize the model using softmax cross entropy loss over over a ranking list consisting of a positive $( q , \\bar { d } ^ { + } )$ pair and 31 sampled negative $( q , d ^ { - } )$ pairs. Unlike monoT5 (Nogueira et al., 2020), which is pointwise reranker that uses an encoder-decoder model and is trained to generate a relevance label, our model is a listwise reranker and is directly optimizes for ranking performance. ", + "bbox": [ + 173, + 156, + 826, + 268 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "D QUERY GENERATION STATISTICS ", + "text_level": 1, + "bbox": [ + 174, + 290, + 486, + 308 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "In Table 7, we analyze the length of the generated questions by different query generation systems. Note that NQ-QGen always generates short queries due to the query generation models being fine-tuned on the NQ dataset, and all of the generated questions have similar length to those questions of NQ. Interestingly, zero-shot PROMPTAGATOR already obtains more variance in terms of length compared to NQ-QGen. Finally, few-shot PROMPTAGATOR offers significantly more variance in terms of the length of generated queries. ", + "bbox": [ + 174, + 324, + 531, + 463 + ], + "page_idx": 17 + }, + { + "type": "table", + "img_path": "images/607aaa880eaa37538b6b6ae40ba3ed1c16b6ea8fd6721d98b983f199ff40283a.jpg", + "table_caption": [ + "Table 7: Average query length. " + ], + "table_footnote": [], + "table_body": "
Few-shotZero-shotNQ QGen
ArguAna98.226.09.7
Touche-20207.813.49.8
TREC-Covid10.811.410.2
NFCorpus8.311.510.3
HotpotQA11.212.28.8
DBPedia-Entity8.213.88.8
Fever12.110.78.8
Climate-Fever12.910.78.8
SciFact12.612.410.0
SCIDOCS7.415.710.7
FiQA-201812.510.19.5
AVG.17.813.59.6
", + "bbox": [ + 547, + 340, + 821, + 489 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "E ROUND-TRIP FILTERING EXAMPLES ", + "text_level": 1, + "bbox": [ + 174, + 486, + 511, + 502 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Figure 6 shows examples of queries from few-shot PROMPTAGATOR that were removed by round-trip filtering. ", + "bbox": [ + 173, + 520, + 531, + 563 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "F ROUND-TRIP FILTERING AS LATENT VARIABLE MODELING ", + "text_level": 1, + "bbox": [ + 173, + 587, + 503, + 618 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "This section aims to share some insight into why our round-trip filtering method is effective, by viewing queries as latent variables that we estimate. First, consider a hypothetical graphical model in which each query $q$ is a latent variable and the documents retrieved for that query are observed variables following some distribution, $p ( d | q , \\theta ^ { * } )$ , where $\\theta ^ { * }$ represents the parameters of a hypothetical “optimal” retriever that always selects the “best” documents for any query. For synthetic data generation, we make it our goal to sample queries from the posterior, $p ( q | d , \\theta ^ { * } )$ , which according to Bayes rule is: ", + "bbox": [ + 173, + 636, + 825, + 733 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/47340a0a39eb4815fe4ed83dc4cd18c83604a39cc536eede976cb37cef826a8a.jpg", + "text": "$$\np ( q | d , \\theta ^ { * } ) = \\frac { p ( d | q , \\theta ^ { * } ) p ( q | \\theta ^ { * } ) } { \\sum _ { q ^ { \\prime } } p ( d | q ^ { \\prime } , \\theta ^ { * } ) p ( q ^ { \\prime } | \\theta ^ { * } ) }\n$$", + "text_format": "latex", + "bbox": [ + 374, + 734, + 622, + 772 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "where $p ( q | \\theta ^ { * } )$ is a prior over queries. We will assume it is an “uninformative prior”that is uniform over all $q$ and later write it as just $p ( q )$ . If we knew $\\theta ^ { * }$ (the parameters of an optimal retriever), we could just directly compute the above expression. But in practice we do not, so we will estimate $\\theta$ using expectation maximization (EM) (Dempster et al., 1977), which will turn out to mirror our round-trip filtering algorithm. EM and other latent variable learning methods have long been used to impute missing data with great success. ", + "bbox": [ + 174, + 777, + 825, + 863 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "In EM, we estimate $\\theta$ by approximately maximizing the marginal likelihood of the observed variables: ", + "bbox": [ + 171, + 869, + 823, + 885 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/93497dc542a530bc5303deb8ed564d31fe0adaf4686b894aadca35ac3b694810.jpg", + "text": "$$\n\\hat { \\theta } = \\arg \\operatorname* { m a x } _ { \\theta } \\prod _ { d \\in \\mathcal { D } _ { T } } p ( d ) = \\prod _ { d \\in \\mathcal { D } _ { T } } \\sum _ { q } p ( d | q , \\theta ) p ( q )\n$$", + "text_format": "latex", + "bbox": [ + 331, + 892, + 665, + 928 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "The first step in EM is to make an initial estimate of $p ( q | d )$ for every document $d$ . We use our FLAN query generator as our initial estimate: $p _ { \\mathrm { F L A N } } ( q | d )$ . We then proceed to the M-step of EM, which computes: ", + "bbox": [ + 174, + 102, + 825, + 145 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/33cb3a068e7f236a26898c28d50f0d920f6de695761fe1f698257527c81b5466.jpg", + "text": "$$\n\\hat { \\theta } = \\arg \\operatorname* { m a x } _ { \\theta } \\sum _ { d \\in \\mathcal { D } _ { T } } \\sum _ { q } p _ { \\mathrm { F L A N } } ( q | d ) p ( d | q , \\theta )\n$$", + "text_format": "latex", + "bbox": [ + 354, + 142, + 642, + 179 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "This is equivalent to training an initial retriever $p ( d | q , \\hat { \\theta } )$ on documents from $\\mathcal { D } _ { T }$ that have been paired with our FLAN-generated queries (what we do). Finally, we proceed to the E-step of EM, which estimates $p ( q | d , \\hat { \\theta } )$ for every document $d$ : ", + "bbox": [ + 173, + 184, + 826, + 229 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/db8aee6b4697a8923a2a4290a9f38e32913122de5b420e46ae9c0995aa1a552c.jpg", + "text": "$$\np ( q | d , \\hat { \\theta } ) \\propto p ( d | q , \\hat { \\theta } ) p ( q )\n$$", + "text_format": "latex", + "bbox": [ + 415, + 236, + 583, + 256 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "From this, we see that the highest probability queries under $p ( q | d , \\hat { \\theta } )$ are the ones with the highest probability of retrieving $d$ under our initial retriever (since $p ( q )$ is an uninformative prior that is uniform over all $q$ ). ", + "bbox": [ + 176, + 263, + 826, + 308 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "To sample from $p ( q | d , \\hat { \\theta } )$ , we could employ importance sampling. In importance sampling, we first sample $q$ from any proposal distribution that we choose, ${ \\bar { p } } _ { \\mathrm { p r o p } } { \\bar { ( } } q | d )$ . We would then weight that sample by the ratio $p ( q | d , \\hat { \\theta } ) / p _ { \\mathrm { p r o p } } ( q | d )$ . We choose $p _ { \\mathrm { F L A N } } ( q | d )$ as our proposal distribution. Then, instead of actually importance-weighting each sample, our filtering procedure just discards samples with a low value of $p ( d | q , \\hat { \\theta } )$ , which is proportional to the numerator in the importance weight. This reveals both the differences and connections between our method and EM. ", + "bbox": [ + 173, + 314, + 826, + 405 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Although we could repeat this EM-like procedure until convergence, we found that a single round of filtering yielded sufficient quality gains. It is also worth noting that the training of our final retriever could be viewed as another M-step. ", + "bbox": [ + 174, + 411, + 825, + 454 + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/f2f805a5569b3d0696d3404b4ca626a70ebadd14470da677cc8df0ac5633e8f4.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 179, + 112, + 816, + 945 + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/c557b60c369f7c8ad18b9f3ad02e06ce7bb6d6e40b24a596c8fcb9c6da9a6602.jpg", + "image_caption": [ + "Figure 4: Top first word distribution on queries generated from different models in all other BEIR datasets. " + ], + "image_footnote": [], + "bbox": [ + 181, + 176, + 818, + 558 + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/9e9f4dad11d0e8ff0ddf2ffc474843fddec973536abf8c874edfde444b984b73.jpg", + "image_caption": [ + "Figure 5: PROMPTAGATOR $^ { + + }$ Training pipeline. " + ], + "image_footnote": [], + "bbox": [ + 179, + 704, + 823, + 844 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Passage: As the COVID-19 pandemic sweeps the globe, evolving containment measures have created an unprecedented need for rapid and effective science communication that is able to engage the public in behavioural change on a mass scale. Public health bodies, governments, and media outlets have turned to comics in this time of need and found a natural and capable medium for responding to the challenge... ", + "bbox": [ + 181, + 185, + 883, + 241 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Query: What is the authors’ purpose for writing this article? ", + "bbox": [ + 192, + 241, + 580, + 255 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Remarks: The query lacks context. This query is also generic and can be used to all articles ", + "bbox": [ + 186, + 255, + 790, + 268 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Passage: Big Bad Love is a 2001 film directed by Arliss Howard , who co-wrote the script with his brother , James Howard , based on a collection of short stories of the same name by Larry Brown . The story recounts an episode in the life of an alcoholic Vietnam veteran and struggling writer named Leon Barlow , who is played by Arliss Howard , and his wife , played by Howard ’s wife Debra Winger . The soundtrack includes music by Tom Verlaine , the Kronos Quartet , and R. L. Burnside . ", + "bbox": [ + 183, + 281, + 883, + 351 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Query: music artist is from the United States ", + "bbox": [ + 187, + 352, + 482, + 364 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Remarks: The query lacks context. ", + "bbox": [ + 186, + 366, + 423, + 378 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Passage: Aminopeptidase N (APN) is the major cell surface receptor for group 1 coronaviruses. In this study, we have isolated and characterized a feline APN cDNA and shown that the transfection of human embryonic kidney cells with this cDNA renders them susceptible to infection with the feline coronavirus feline infectious peritonitis virus, the human coronavirus (HCV) 229E and the porcine coronavirus porcine transmissible gastroenteritis virus. ... ", + "bbox": [ + 183, + 392, + 883, + 462 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Query: What is the function of Aminopeptidase N? ", + "bbox": [ + 186, + 462, + 522, + 476 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Remarks: The query is too general and can match many documents that mentions APN. ", + "bbox": [ + 179, + 477, + 761, + 489 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Passage: Selena Danielle Coppa (born February 25, 1983) was a military intelligence Sergeant in the United States Army. She is primarily notable for her organizing and activism against the US Occupation of Iraq while serving as an active duty military member, including serving on the Executive Board of Iraq Veterans Against the War. In 2009 it was announced that she was heading a committee responsible for gaining and training more active duty anti-war soldiers. ", + "bbox": [ + 181, + 503, + 883, + 573 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Query: who was the first president of the senate? ", + "bbox": [ + 186, + 574, + 506, + 588 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Remarks: Hallucination. The passage didn’t mention president of the senate. ", + "bbox": [ + 184, + 588, + 691, + 602 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Passage: Porcine deltacoronavirus (PDCoV) is an emerging swine coronavirus that causes severe diarrhea, resulting in high mortality in neonatal piglets. Despite widespread outbreaks in many countries, no effective PDCoV vaccines are currently available. Here, we generated, for the first time, a full-length infectious cDNA clone of PDCoV. We further manipulated the infectious clone by replacing the NS6 gene with a green fluorescent protein (GFP) to generate rPDCoV-NS6-GFP... ", + "bbox": [ + 183, + 614, + 882, + 684 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Query: What virus is the first one that we’ve been able to create a vaccine for? ", + "bbox": [ + 186, + 684, + 707, + 698 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Remarks: Hallucination. The passage didn’t mention that Porcine deltacoronavirus is the first virus one can create a vaccine. ", + "bbox": [ + 186, + 699, + 882, + 726 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Passage: Nacoleia rectistrialis is a moth in the Crambidae family. It was described by Hampson in 1912. It is found on the Key Islands. \nQuery: Who composed the music for the movie the ring? \nRemarks: Hallucination. The passage is about moth and is unrelated to “music for the movie the ring”. ", + "bbox": [ + 176, + 738, + 882, + 796 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Figure 6: Examples of queries from few-shot PROMPTAGATOR’s prompt-based query generation that were removed by round-trip filtering. ", + "bbox": [ + 173, + 809, + 826, + 838 + ], + "page_idx": 21 + } +] \ No newline at end of file diff --git a/parse/dev/jlAjNL8z5cs/images/0ff873c8d4b4ff8ed4f02cc0d7813660f2045240a021b72602e386dfa691d646.jpg b/parse/dev/jlAjNL8z5cs/images/0ff873c8d4b4ff8ed4f02cc0d7813660f2045240a021b72602e386dfa691d646.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c41fbdd00085a5b606eab5fa12f7339ef36b3e00 --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/0ff873c8d4b4ff8ed4f02cc0d7813660f2045240a021b72602e386dfa691d646.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf8746b9f5bd7e69e12c457fefdddccc7553ba6320de30cf9f31867b14796a56 +size 62786 diff --git a/parse/dev/jlAjNL8z5cs/images/1bbbf377c769e34e7b316e942e6531612f15df31dc112659ffe2dbade22555ce.jpg b/parse/dev/jlAjNL8z5cs/images/1bbbf377c769e34e7b316e942e6531612f15df31dc112659ffe2dbade22555ce.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ec4c36c96fa6c86e43d0e7807e28cc3f5a8d70ff --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/1bbbf377c769e34e7b316e942e6531612f15df31dc112659ffe2dbade22555ce.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac28a3af1feb42c57a2ada9d40546345f267c3d5def831c656ed388b50096412 +size 51694 diff --git a/parse/dev/jlAjNL8z5cs/images/1e7caffc21ea861414376baa5c3b218c252a5a508bde3c37ebcbc56f50784613.jpg b/parse/dev/jlAjNL8z5cs/images/1e7caffc21ea861414376baa5c3b218c252a5a508bde3c37ebcbc56f50784613.jpg new file mode 100644 index 0000000000000000000000000000000000000000..53c7f54cbacdc6538a0cfc89e12179cbb2b723d5 --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/1e7caffc21ea861414376baa5c3b218c252a5a508bde3c37ebcbc56f50784613.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8089d8f237b0833ef2c02c92c2e2453c65eb83c76c901ff85569bcb9dcd9c550 +size 44575 diff --git a/parse/dev/jlAjNL8z5cs/images/684aca7763a161329faf7297c0df23c07de3c6d3572299b549ab198db1c30c45.jpg b/parse/dev/jlAjNL8z5cs/images/684aca7763a161329faf7297c0df23c07de3c6d3572299b549ab198db1c30c45.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cc8a4b9e9b85819d052b61defc31ec778a459b5e --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/684aca7763a161329faf7297c0df23c07de3c6d3572299b549ab198db1c30c45.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:73711f26a68fe3114659a97b7285c279bbbe5f6c573bc2c86edef22e16e63e13 +size 86471 diff --git a/parse/dev/jlAjNL8z5cs/images/6993c67a3bf8f097288dfc04dece000c7cf61c6be28fed8df55c7251dd26fd24.jpg b/parse/dev/jlAjNL8z5cs/images/6993c67a3bf8f097288dfc04dece000c7cf61c6be28fed8df55c7251dd26fd24.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d418907874071fa82f8130f9a3f80649bea6632f --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/6993c67a3bf8f097288dfc04dece000c7cf61c6be28fed8df55c7251dd26fd24.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80a9898b44b2f6ca1427a1dcca790106e21dcfdae21f9c9f2e43fcca7ba9061a +size 23575 diff --git a/parse/dev/jlAjNL8z5cs/images/71057e9d734a0301371192686f151cbc2ca7f46eb4e564728a8311505158e7fd.jpg b/parse/dev/jlAjNL8z5cs/images/71057e9d734a0301371192686f151cbc2ca7f46eb4e564728a8311505158e7fd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..82bc481afb18099632200dd09ad5c9a47aba8b3d --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/71057e9d734a0301371192686f151cbc2ca7f46eb4e564728a8311505158e7fd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9ce79477b23a7c1a30649403e493797c2c5518c08cb64aabeefeb89af34ca42 +size 43334 diff --git a/parse/dev/jlAjNL8z5cs/images/8a6728b345f3d79f6102f60c9abd115a52d268aadd10225cc0a7d28b55ee48c7.jpg b/parse/dev/jlAjNL8z5cs/images/8a6728b345f3d79f6102f60c9abd115a52d268aadd10225cc0a7d28b55ee48c7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..67fb7b340fd63a3f3e513031a8794117073f1bed --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/8a6728b345f3d79f6102f60c9abd115a52d268aadd10225cc0a7d28b55ee48c7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa6993ec97e26e4a2fc6e42c3087d3fa93c9d2540e0f7be73eef2028a262e90d +size 5092 diff --git a/parse/dev/jlAjNL8z5cs/images/951d650b619aefba4324e4695cea151f2c31798b9c17112a96d2877fef1c3dec.jpg b/parse/dev/jlAjNL8z5cs/images/951d650b619aefba4324e4695cea151f2c31798b9c17112a96d2877fef1c3dec.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ef5c3adcd0bb70896d9ef96260141b095eb1f684 --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/951d650b619aefba4324e4695cea151f2c31798b9c17112a96d2877fef1c3dec.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:34150422e7f5136371e9fc3a0393704da961e6840a406d160f31acda5fdba30c +size 88828 diff --git a/parse/dev/jlAjNL8z5cs/images/9d35980d1366486b0383f5b291a446478cd342db60192416269eef54a257d2f5.jpg b/parse/dev/jlAjNL8z5cs/images/9d35980d1366486b0383f5b291a446478cd342db60192416269eef54a257d2f5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..62546e093cf299a57c477294d03d71446c45aacd --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/9d35980d1366486b0383f5b291a446478cd342db60192416269eef54a257d2f5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4ff1725b2c4bcc37b225d7b0008817466cea1f710652fedaa008733618939fd +size 28054 diff --git a/parse/dev/jlAjNL8z5cs/images/afc75d3d75c0cc4348369793b7fc9c20e93e3869f28194838851bf7935fa5825.jpg b/parse/dev/jlAjNL8z5cs/images/afc75d3d75c0cc4348369793b7fc9c20e93e3869f28194838851bf7935fa5825.jpg new file mode 100644 index 0000000000000000000000000000000000000000..05e8e424038913f23a6e7b3002c941308b9ebaaa --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/afc75d3d75c0cc4348369793b7fc9c20e93e3869f28194838851bf7935fa5825.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11c394e39487de34c96f98f7edea2a58928b4f2a324d19139e479b7039d12581 +size 32341 diff --git a/parse/dev/jlAjNL8z5cs/images/bde69ced4d52cf5f242378cfc5ec164346c4fa12119d5dc17b6ff034c7112f2c.jpg b/parse/dev/jlAjNL8z5cs/images/bde69ced4d52cf5f242378cfc5ec164346c4fa12119d5dc17b6ff034c7112f2c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7ac241cd4d0221316e409b97149faccb543d7468 --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/bde69ced4d52cf5f242378cfc5ec164346c4fa12119d5dc17b6ff034c7112f2c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2cc3c628e705f076dcd89ffe354fffc511751d8b7e3fffcf4467abf6b63ea2bb +size 21073 diff --git a/parse/dev/jlAjNL8z5cs/images/ca3f96d9efd58a2ebeb276ab999910b520c01b2660bd1c28307215341bb78df6.jpg b/parse/dev/jlAjNL8z5cs/images/ca3f96d9efd58a2ebeb276ab999910b520c01b2660bd1c28307215341bb78df6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..446927e6788b131b83658850eda056a5bd540f09 --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/ca3f96d9efd58a2ebeb276ab999910b520c01b2660bd1c28307215341bb78df6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:708f0bd542095f1a5390ddea374d217f4844c1e3c29b7aff74a674f8c3e42b4b +size 2773 diff --git a/parse/dev/jlAjNL8z5cs/images/fc20a4a17769825aa14fdbba20f3f023668f243e0c7574e4417c6f1b48b175e2.jpg b/parse/dev/jlAjNL8z5cs/images/fc20a4a17769825aa14fdbba20f3f023668f243e0c7574e4417c6f1b48b175e2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ad94526a14d205fa33326631675bc9f4ba6b38fb --- /dev/null +++ b/parse/dev/jlAjNL8z5cs/images/fc20a4a17769825aa14fdbba20f3f023668f243e0c7574e4417c6f1b48b175e2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7915b30ee2efe016485b7e4535fcc5349d80173c7b79d5a7bc5f4ecc83299a4 +size 34237 diff --git a/parse/dev/ksVGCOlOEba/ksVGCOlOEba.md b/parse/dev/ksVGCOlOEba/ksVGCOlOEba.md new file mode 100644 index 0000000000000000000000000000000000000000..34a6dddbd21723001c60179af0e9e3a7b93b5ccf --- /dev/null +++ b/parse/dev/ksVGCOlOEba/ksVGCOlOEba.md @@ -0,0 +1,282 @@ +# Optimal Brain Compression: A Framework for Accurate Post-Training Quantization and Pruning + +Elias Frantar ∗ IST Austria elias.frantar@ist.ac.at + +Sidak Pal Singh ETH Zurich sidak.singh@inf.ethz.ch + +Dan Alistarh IST Austria & Neural Magic dan.alistarh@ist.ac.at + +# Abstract + +We consider the problem of model compression for deep neural networks (DNNs) in the challenging one-shot/post-training setting, in which we are given an accurate trained model, and must compress it without any retraining, based only on a small amount of calibration input data. This problem has become popular in view of the emerging software and hardware support for executing models compressed via pruning and/or quantization with speedup, and well-performing solutions have been proposed independently for both compression approaches. In this paper, we introduce a new compression framework which covers both weight pruning and quantization in a unified setting, is time- and space-efficient, and considerably improves upon the practical performance of existing post-training methods. At the technical level, our approach is based on an exact and efficient realization of the classical Optimal Brain Surgeon (OBS) framework of [LeCun, Denker, and Solla, 1990] extended to also cover weight quantization at the scale of modern DNNs. From the practical perspective, our experimental results show that it can improve significantly upon the compression-accuracy trade-offs of existing post-training methods, and that it can enable the accurate compound application of both pruning and quantization in a post-training setting. + +# 1 Introduction + +The impressive recent progress of deep learning for solving challenging tasks across several domains has been accompanied by a significant increase in parameter counts and computational costs for executing such models. A natural consequence has been a growing effort to reduce such costs via model compression, and the two most popular approaches for model compression are pruning— removing neural network weights by setting them to zero—and quantization, reducing the precision at which neural network weights and activations are stored and manipulated. Hundreds of such pruning and quantization approaches have been proposed and analyzed [17, 11], with the general goal of obtaining efficient variants of deep neural nets (DNNs) which would preserve accuracy while maximizing compression. Despite impressive progress, compression is still a laborious process: the pruning and quantization stages are often done independently, and recovering model accuracy after compression often requires partial or even full retraining of the compressed model. + +An alternative but challenging scenario is the post-training compression setup [31, 24, 19, 25], in which we are given a trained but uncompressed model, together with a small amount of calibration data, and must produce an accurate compressed model in one shot, i.e. a single compression step, without retraining, and with limited computational costs. This is motivated by practical scenarios such as the MLPerf Inference Benchmark [37], and is the setting we focus on in this paper. + +Compression via weight pruning started with seminal work by LeCun et al. [23], complemented by Hassibi and Stork [13], who proposed a mathematical framework called the Optimal Brain Surgeon (OBS), for choosing the set of weights to remove from a trained neural network, by leveraging secondorder information. (We describe their approach in Section 3.) Recent advances, e.g. [6, 40, 39, 10] showed that OBS can lead to state-of-the-art compression at DNN scale, by introducing numerical methods which can approximate the second-order information needed by OBS at the massive parameter counts of modern models. However, these approaches do not apply to the post-training setting, as they require gradual pruning, as well as significant retraining, to recover good accuracy. + +An alternative approach, which is standard in the context of post-training compression, has been to break the compression task into layer-wise sub-problems, identifying a compressed weight approximation for each layer, given a sub-sample of the layer’s inputs and outputs based on calibration data. This line of work, e.g. [42, 31, 19], introduced elegant solvers for the resulting layer-wise weight quantization problem, which achieve state-of-the-art results for post-training quantization. Recently, AdaPrune [18] showed that this approach can also be effective for post-training weight pruning. + +In this context, a natural question is whether existing approaches for pruning and quantization can be unified in order to cover both types of compression in the post-training setting, thus making DNN compression simpler and, hopefully, more accurate. This question is also of practical importance, since both GPU and CPU platforms now jointly support sparse and quantized formats [30, 35], and, as we illustrate experimentally, the resulting models could be executed with compound speedups. + +Contribution. In this paper, we provide a mathematical framework for compression via pruning or quantization, which leads to state-of-the-art accuracy-versus-compression trade-offs in the challenging post-training compression setup. Our framework starts from the layer-wise compression problem described above, by which the global compression task, defined either for pruning or quantization, is first split into layer-wise sub-problems, based on the layer behavior on the calibration data. Specifically, given a layer $\ell$ defined by weights $\mathbf { W } _ { \ell }$ , and layer inputs $\mathbf { X } _ { \ell }$ , we aim to find a compressed version of the weights $\widehat { \mathbf { W } } _ { \ell }$ which minimizes the output difference relative to the uncompressed layer, measured via the squared error between the original and compressed layer, acting on the sample input $\| { \mathbf { W } } _ { \ell } { \mathbf { X } } _ { \ell } - \widehat { { \mathbf { W } } } _ { \ell } { \mathbf { X } } _ { \ell } \| _ { 2 } ^ { 2 }$ , under a fixed compression constraint on $\widehat { \mathbf { W } } _ { \ell }$ . + +Although solving this problem optimally for sparsity or quantization constraints is NP-hard [2, 31], it is a key step in all state-of-the-art post-training compression methods, both for pruning [18, 9] and for quantization [31, 18, 24]. Once this is solved per layer, a solution to the global problem can be obtained by combining layer-wise solutions, which is handy especially for non-uniform compression, e.g. [15, 9]. Thus, several approximations for this problem have been proposed [31, 19, 18]. + +We show that there still is significant room for improvement when solving the layer-wise compression problem. Roughly, our approach is to specialize the OBS framework to the squared error formulation above: in this case, the framework can in theory produce an exact greedy solution, but a direct implementation would have infeasible $\Theta ( d ^ { 4 } )$ computational cost, where $d$ is the layer dimension. Our main technical contribution is a series of algorithms which reduce this computational cost, without any approximations, to $O ( d \cdot d _ { c o l } ^ { 2 } )$ where $d _ { c o l }$ is the column dimension of the weight matrix. In practice, these improvements are significant enough to allow us to implement the exact OBS greedy solution, which prunes one weight at a time, and updates all remaining weights after each step, at the scale of modern DNNs with tens of millions of parameters, within reasonable time, on a single GPU. We provide efficient implementations of our methods at https://github.com/IST-DASLab/OBC. + +In turn, this algorithmic development allows us to apply the OBS approach to quantization. The resulting algorithm, called the Optimal Brain Quantizer (OBQ), quantizes weights iteratively one-ata-time, depending on their impact on the loss increase, after which it applies a closed-form update to the remaining unquantized weights, further reducing the loss. This solves the two problems efficiently, and in a unified manner—we call the unified framework the Optimal Brain Compressor (OBC). + +Experimental Results. We apply OBC to standard tasks and models covering image classification, object detection, and language modelling applications. We first show that our framework yields significantly better solutions for the layer-wise compression problem, which leads to higher-accuracy end-to-end compressed models for both pruning and quantization, relative to the corresponding stateof-the-art techniques, often by significant margins. Second, we show that our pruning and quantization approaches can be compounded, with surprisingly strong results: we obtain a $1 2 \times$ reduction in theoretical operations with a $2 \%$ accuracy drop for GPU-supported compound compression [30], and a $4 \times$ speedup in actual runtime with only $1 \%$ accuracy drop for a CPU-based sparsity-aware runtime [35]. Together, these results suggest for the first time that post-training compression can be competitive with full retraining. + +# 2 Related Work + +Optimal Brain Surgeon (OBS). The classic OBS framework [23, 13] was originally applied to networks with hundreds of weights; more recently, methods such as WoodFisher [39] rendered the approach computationally feasible for DNNs by using a block-diagonal Fisher approximation of the Hessian, while follow-up methods introduced more efficient and general algorithms for handling the inverse Fisher matrix, or customize this approximation to specific model families [21]. Earlier work called Layer-wise OBS (L-OBS) [6] was inspired by the K-FAC approximation [29, 12]: L-OBS approximates the OBS framework not for the global objective, but for a quadratic per-layer loss, while also pruning all weights based on a single Hessian computation. At a high level, our approach is similar, in that we apply OBS layer-wise; however, we apply OBS exactly, that is, pruning one weight at a time, and exactly recomputing the Hessian after every pruning step. This is made computationally tractable by several new algorithmic ideas, and yields significantly improved results relative to L-OBS. This prior work on pruning considered settings with extensive finetuning. By contrast, we will focus on the post-training setting, where only a small amount of calibration data is available. + +Post-Training Quantization. This setting has been primarily considered for quantization, and most state-of-the-art methods work by performing layer-wise compression. Specifically, BitSplit [6] optimizes the quantized weights bit by bit, while AdaRound [31] finds a weight rounding policy through gradient based optimization with an annealed penalty term that encourages weights to move towards points on the quantization grid. AdaQuant [19] relaxes the AdaRound constraint, allowing weights to change during quantization-aware optimization, via straight-through estimation [33]. BRECQ [24] suggested that accuracy can be improved further by integrating second-order information into the layer-wise losses and by jointly optimizing hand-crafted blocks of related layers. + +A key step of AdaRound, AdaQuant and BRECQ is to quantize layers incrementally, in sequential order, so that errors accumulated in earlier layers can be compensated by weight adjustments in later ones. This significantly improves performance, but reduces flexibility, as the entire process may need to be re-done whenever we wish to change compression parameters of one layer. We instead target independent compression of each layer, allowing the end model to be simply “stitched” together from layer-wise results. Despite operating independently on each layer, we find that, after correcting basic statistics such as batchnorm, our method performs on par to sequential ones for uniform quantization. + +Post-Training Sparsification. The layer-wise approach was shown to also be effective for posttraining pruning by AdaPrune [18], which pruned weights to the GPU-supported N:M pattern [45]. AdaPrune first drops parameters according to their magnitude [46] and then reoptimizes the remaining weights to reconstruct the pre-compression calibration set output. This is similar to [16, 8] which also perform layer-wise reoptimization of the remaining weights. Follow-up work [10] noted that the results of AdaPrune can be improved further by performing more frequent pruning/optimization steps. Our algorithm pushes this idea to the limit, performing full reoptimization after every single pruned weight, while remaining computationally tractable. We further use a more sophisticated weight selection metric which incorporates second-order information. Finally, [10] also introduces global AdaPrune, a more expensive global optimization step applied on top of the layer-wise AdaPrune results, which can bring additional accuracy gains. This can also be applied to our pruned models. + +Non-Uniform Compression. An orthogonal practical question is how to compress different layers to maximize accuracy under a given resource constraint, such as latency or energy consumption. Existing methods can be roughly categorized into search-based and solver-based approaches. The former, e.g. AMC [15] or HAQ [41], search for a layer-wise compression policy directly via, for example, reinforcement learning or genetic programming [43], whereas the latter, e.g. HAWQv3 [44] or AdaQuant [19], construct a relaxed version of the overall problem that is then solved exactly. We focus here on solver-based approaches, as they can rapidly adapt to different scenarios when combined with accurate independent layer-wise compression schemes; however, our techniques could be of interest for search-based methods as well. Concretely, we use the problem formulation of AdaQuant [19] to which we apply the DP algorithm of SPDY [10] to achieve fast solving times even with a large number of possible choices per layer. + +# 3 Problem Definition and Background + +The Layerwise Compression Problem. Following prior work on post-training compression, e.g. [31, 19], we define the problem as follows. Mathematically, we model a layer $\ell$ as a function $f _ { \ell } ( X _ { \ell } , W _ { \ell } )$ acting on inputs $X _ { \ell }$ , parametrized by weights $W _ { \ell }$ . The goal of layer-wise compression is to find a “compressed” version of $W _ { \ell }$ that performs as similarly as possible to the original weights. More formally, the compressed weights $\widehat { W } _ { \ell }$ should minimize the expected layer output change as measured by some loss $\mathcal { L }$ while at the same time satisfying a generic compression constraint, which we denote by ${ \mathcal { C } } ( \widehat { W } _ { \ell } ) > C$ , which will be customized depending on the compression type: + +$$ +\mathrm { a r g m i n } _ { \widehat { W } _ { \ell } } \quad \mathbb { E } _ { X _ { \ell } } \mathcal { L } \big ( f _ { \ell } ( X _ { \ell } , W _ { \ell } ) , f _ { \ell } ( X _ { \ell } , \widehat { W } _ { \ell } ) \big ) \quad \mathrm { s u b j e c t ~ t o } \quad \mathcal { C } ( \widehat { W } _ { \ell } ) > C . +$$ + +The expectation over the layer inputs $X _ { \ell }$ is usually approximated by taking the mean over a small set of $N$ input samples. This low-data setting is one of the primary applications of layer-wise compression. Further, most works [42, 31, 19] focus on compressing linear and convolutional layers, which can be unfolded into linear ones, as these are prevalent in practice, and use the squared loss to measure the approximation error. This definition of the loss can be motivated, via a sequence of approximations, from second-order information: please see [31] for a precise derivation. Furthermore, this approximation approach has been shown to work well in many applications [31, 19, 9]. + +We follow these conventions as well, and work with the specific layer-wise compression problem stated formally below, where the weights $\mathbf { W } _ { \ell }$ are a $d _ { \mathrm { r o w } } \times d _ { \mathrm { c o l } }$ matrix (for conv-layers $d _ { \mathrm { c o l } }$ corresponds to the total number of weights in a single filter), and the input $\mathbf { X } _ { \ell }$ has dimensions $d _ { \mathrm { c o l } } \times N$ . + +$$ +\operatorname { a r g m i n } _ { \widehat { \mathbf { W } } _ { \ell } } \quad | | \mathbf { W } _ { \ell } \mathbf { X } _ { \ell } - \widehat { \mathbf { W } } _ { \ell } \mathbf { X } _ { \ell } | | _ { 2 } ^ { 2 } \quad \mathrm { s . t . } \quad \mathcal { C } ( \widehat { \mathbf { W } } _ { \ell } ) > C . +$$ + +The Optimal Brain Surgeon (OBS) Framework. The OBS framework [23, 13] considers the problem of accurately pruning a trained dense neural network. It starts from the Taylor approximation at the given point (assumed to have negligible gradient), and provides explicit formulas for the optimal single weight to remove, as well as the optimal update of the remaining weights which would compensate for the removal. More precisely, let $\mathbf { H }$ denote the Hessian matrix of the loss at the given (dense) model. Then the weight to prune $w _ { p }$ which incurs the minimal increase in loss and the corresponding update of the remaining weights $\delta _ { p }$ can be calculated as follows: + +$$ +w _ { p } = \operatorname * { a r g m i n } _ { w _ { p } } \frac { w _ { p } ^ { 2 } } { [ \mathbf { H } ^ { - 1 } ] _ { p p } } , \quad \delta _ { p } = - \frac { w _ { p } } { [ \mathbf { H } ^ { - 1 } ] _ { p p } } \cdot \mathbf { H } _ { : , p } ^ { - 1 } , +$$ + +where $[ \mathbf { H } ^ { - 1 } ] _ { p p }$ denotes the $p$ th diagonal entry of the inverse Hessian, and $\mathbf { H } _ { : , p } ^ { - 1 }$ is its $p$ th column. + +OBS for Layer-Wise Pruning. We will now instantiate this framework for the layer-wise pruning problem, defined above. First, the loss in equation (2) is quadratic and since our starting point is given by the dense weights achieving the minimal loss of 0, the assumptions of the OBS framework are fully met, meaning that its formulas are exact for this specific problem formulation. Thus, iterating the OBS framework to remove one weight at a time would yield an exact greedy solution for the layer-wise pruning problem, as it takes the (locally) optimal decision at each step. While this greedy approach does not guarantee convergence to a global optimum, such approaches can be very effective for dealing with problem instances that are too large to be handled by exact methods. + +# 4 An Optimal Greedy Solver for Sparsity + +The obvious challenge is that applying the OBS framework in its true form, i.e. pruning a single weight at a time using the exact formulas in (3), is computationally very demanding. The Hessian $\mathbf { H }$ is a $d { \times } d$ matrix where $d = d _ { \mathrm { r o w } } \cdot d _ { \mathrm { c o l } }$ , which is already expensive to store and compute with. Additionally, this matrix needs to be updated and inverted at each of the $O ( d )$ steps with a computational complexity of $\Theta ( d ^ { 3 } )$ . Clearly, an $O ( d ^ { 4 } )$ total runtime is too inefficient for pruning most layers of modern neural networks, as $d$ is usually $\geq 1 0 ^ { 5 }$ or even $\geq 1 0 ^ { 6 }$ for several layers. However, as we will now show, it is actually possible to reduce the overall costs of this process to $O ( d _ { \mathrm { r o w } } \cdot d _ { \mathrm { c o l } } ^ { 3 } )$ time and $\Theta ( d _ { \mathrm { c o l } } ^ { 2 } )$ memory, making it efficient enough to prune e.g. all layers of a medium-sized model such as ResNet50 in a bit more than one hour on a single NVIDIA RTX 3090 GPU. We emphasize that the techniques we introduce are exact; unlike prior work [6, 39], we do not rely on any approximations. + +The ExactOBS Algorithm. In the following, we introduce our efficient instantiation of the OBS framework, for the layer-wise compression problem, which we call ExactOBS, in step-by-step fashion. We start by rewriting the matrix squared error in (2) as the sum of the squared errors for each row in the weight matrix. As we are always dealing with a fixed layer $\ell .$ , we drop the subscript $\ell$ to simplify notation. The objective is then equivalent to $\begin{array} { r } { \sum _ { i = 1 } ^ { d _ { \mathrm { r o w } } } | | \mathbf { W } _ { \mathbf { i } , : } \mathbf { \widehat X } - \widehat { \mathbf { W } } _ { \mathbf { i } , : } \mathbf { X } | | _ { 2 } ^ { 2 } } \end{array}$ . + +This way of writing the error makes it clear that removing a single weight $[ \mathbf { W } ] _ { i j }$ only affects the error of the corresponding output row $\mathbf { Y _ { i , : } } = \mathbf { W _ { i , : } } \mathbf { X }$ . Hence, there is no Hessian interaction between different rows and so it suffices to work only with the individual $d _ { \mathrm { c o l } } \times d _ { \mathrm { c o l } }$ Hessians corresponding to each of the $d _ { \mathrm { r o w } }$ rows. Further, as the dense layer output $\mathbf { Y } = \mathbf { W } \mathbf { X }$ is fixed, the objective for each row has standard least squares form and its Hessian is given by $\mathbf { H } = \mathbf { 2 } \mathbf { X } \mathbf { X } ^ { \top }$ . + +Although this observation already reduces computational complexity, two key challenges remain: (a) applying OBS to each row still costs $O ( d _ { \mathrm { c o l } } \cdot \bar { d } _ { \mathrm { c o l } } ^ { 3 } )$ time, which is too slow for large layers, and (b) we need fast access to the Hessian inverses of all $d _ { \mathrm { r o w } }$ rows, since we want to prune the minimum score weight of the whole matrix rather than just per row in each step. In particular, (b) requires $O ( d _ { \mathrm { r o w } } \cdot \bar { d _ { \mathrm { c o l } } ^ { 2 } } )$ GPU memory, which is likely to be infeasible. + +Step 1: Handling a Single Row. We first describe how to efficiently prune weights from a single row with $d _ { \mathrm { c o l } }$ parameters. For simplicity, we denote such a row by w with corresponding Hessian $\mathbf { H }$ . The full algorithm for this procedure is given in Algorithm 1; in the following, we provide a detailed description. The key idea is to avoid having to do the full $\Theta ( N \cdot d _ { \mathrm { c o l } } ^ { 2 } )$ calculation and $\Theta ( d _ { \mathrm { c o l } } ^ { 3 } )$ inversion of $\mathbf { H }$ in each step. The former is easy, as the weights themselves do not enter the calculation of $\mathbf { H } = 2 \mathbf { X } \mathbf { X } ^ { \top }$ , and the Hessian for the weights with pruning mask $M$ denoted by ${ \bf { H } } _ { M }$ is thus simply comprised of the corresponding rows and columns in the fully dense version $\mathbf { H }$ . Hence, we only have to compute $\mathbf { H }$ (which is actually the same for all rows) once, from which we can then extract the rows and columns corresponding to $M$ as needed. + +Critically, this trick is not applicable to the inverse, as $( \mathbf { H } _ { M } ) ^ { - 1 } \neq ( \mathbf { H } ^ { - 1 } ) _ { M }$ . However, using the fact that the removal of one parameter $p$ simply drops the corresponding row and column from $\mathbf { H }$ , we can actually update the inverse to remove parameter $p$ directly using a single step of Gaussian elimination, with cost $\Theta ( d _ { c o l } ^ { 2 } )$ . The following result, whose proof is in the Appendix, formalizes this. + +Lemma 1 (Row & Column Removal). Given an invertible $d _ { c o l } \times d _ { c o l }$ matrix $\mathbf { H }$ and its inverse ${ \bf H } ^ { - 1 }$ , we want to efficiently compute the inverse of $\mathbf { H }$ with row and column $p$ removed, which we denote by $\mathbf { H } _ { - p }$ . This can be accomplished through the following formula: + +$$ +{ \bf H } _ { - p } ^ { - 1 } = \Big ( { \bf H } ^ { - 1 } - \frac { 1 } { [ { \bf H } ^ { - 1 } ] _ { p p } } { \bf H } _ { : , p } ^ { - 1 } { \bf H } _ { p , : } ^ { - 1 } \Big ) _ { - p } , +$$ + +which corresponds to performing Gaussian elimination of row and column $p$ in ${ \bf H } ^ { - 1 }$ followed by dropping them completely. This has $\Theta ( d _ { c o l } ^ { 2 } )$ time complexity. + +The resulting pseudocode is shown in Algorithm 1, where we avoid constantly resizing ${ \bf H } ^ { - 1 }$ (and correspondingly changing indices) by utilizing the fact that row and column $p$ have no effect on any future calculations after they have been eliminated by Lemma 1 as they are 0 (and the non-zero diagonal element is never accessed again). One can check that this algorithm applies OBS to a single row of W with a per-step cost of $\Theta ( \bar { d } _ { \mathrm { c o l } } ^ { 2 } )$ , and thus $\Theta ( k \cdot d _ { \mathrm { c o l } } ^ { 2 } )$ overall time for pruning $k$ weights. + +
Algorithm1Prune k ≤ dcol weights from row w with inverse Hessian H-1 = (2XXT)-1 according to OBS in O(k · d²ol) time.
M = {1,...,dco1} fori=1,...,k do
p ← argminp∈M 1 [H-1]
pp w←w-H-1 1 .Wp
:,p [H-1] pp H-1←H-1 1 H-11H-1 p,
+ +![](images/420513c26a2b97b398b281b3e7ef3eb773ec0bd92c2b0cc71eac615d9769ed42.jpg) +Figure 1: Efficient global OBS using the rowwise results. + +Step 2: Jointly Considering All Rows. Applying the OBS framework to the full weight matrix W rather than just to each row independently requires fast access to all $d _ { \mathrm { r o w } }$ row-wise inverse Hessians, in order to select the weight with the smallest overall pruning score in each step. However, storing $d _ { \mathrm { r o w } }$ matrices of size $d _ { \mathrm { c o l } } \times d _ { \mathrm { c o l } }$ each in GPU memory can be too expensive; while it would be possible to offload some Hessians to main memory, this could result in a large number of expensive memory transfers. However, since there is no Hessian interaction between rows, the final compressed weights of each row only depend on the total number of parameters that were pruned in it. Similarly, the change in loss incurred by pruning some weight only depends on the previously pruned weights in the same row, which also means that the order in which weights are pruned in each row is fixed. + +The consequence of these insights is that we can process each row independently, pruning all weights in order while always recording the corresponding change in loss $\delta \bar { \mathcal { L } _ { p } } = w _ { p } ^ { 2 } \big / [ \bar { \mathbf { H } } ^ { - 1 } ] _ { p p }$ . At the end, we know $\delta { \mathcal { L } } _ { p }$ for all $d$ weights and can then simply determine the global mask that would be chosen by OBS on the full matrix by selecting the weights with the lowest values in order, requiring only $\Theta ( d )$ extra memory. We note that once the per-row masks $M _ { i }$ are known, we can directly solve for the optimal update of the remaining weights via the corresponding group OBS formula [21] $\delta _ { M _ { i } } = \bar { \mathbf { H } } _ { : , M _ { i } } ^ { - 1 } ( ( \bar { \mathbf { H } } ^ { - 1 } ) _ { M _ { i } } ) ^ { - 1 } \mathbf { w } _ { M _ { i } }$ . This will be considerably faster in practice than simply rerunning the iterative pruning process in Algorithm 1. Alternatively, if enough CPU memory is available, one can keep the full pruning trace of each row, that is, the full weight vector after every individual pruning step, in CPU memory and ultimately simply reload the entries corresponding to the global mask. This requires $O ( d _ { \mathrm { r o w } } \cdot d _ { \mathrm { c o l } } ^ { 2 } )$ extra CPU memory but avoids a second computation pass to reconstruct the not pruned weights and will therefore be faster. Figure 1 visualizes both options just discussed. + +Implementation Details. In practice, the matrix $\mathbf { H }$ might not always be invertible for reasons such as using too few data samples or dead / linearly dependent inputs. The former can usually be addressed by extending the calibration dataset with augmentations (additional augmented samples only need to be accumulated into the Hessian once and are thus very cheap to include) and the latter can be prevented by adding a small diagonal dampening term to the Hessian before inverting it. Second, a direct GPU implementation of Algorithm 1 will perform a large number of small CUDA calls, which can be expensive. This overhead can be removed by using batch operations to process multiple matrix rows simultaneously—for more details please see our sample implementation. Finally, when applied to an already sparse weight matrix, the complexity of our algorithm can scale cubicly with the row-density by working with a dense version of the weights / Hessians consisting only of the non-zero elements and mapping the pruning result back at the end. + +N:M Sparsity. Our method can be easily extended to various forms of semi-structured sparsity. This includes, for example, the N:M sparsity pattern [45], which enforces exactly $N$ non-zero values in each block of $M$ consecutive weights, and is becoming popular due to support on newer NVIDIA hardware [30]. Adapting our algorithm to this pattern requires only one simple change: instead of selecting the weight with the smallest change in loss, we select the weight with the smallest change in loss that is in a block with $< N$ pruned weights. We note that all rows have exactly the same sparsity $1 - N / M$ in the N:M pattern and so we can terminate per-row pruning as soon as this target sparsity value is reached. For the same reason, there is no need for the global mask selection step described earlier. Thus, our method will be even more efficient in this case. + +Block-Sparsity. Another practically relevant pruning pattern, particularly in the context of CPU acceleration [7, 22], is block-pruning, where zeros appear only in consecutive blocks of size $c$ , which is typically a small number like 4 or 8. We follow recent work [21] that extends the OBS framework to pruning small groups of connected weights in order to account for the correlation between them, using the following formulas for the target block and weight update, respectively: + +$$ +\begin{array} { r } { { \bf w } _ { P } = \operatorname * { a r g m i n } _ { { \bf w } _ { P } } { \bf w } _ { P } ^ { \top } ( ( { \bf H } ^ { - 1 } ) _ { P } ) ^ { - 1 } { \bf w } _ { P } , \quad \delta _ { P } = - { \bf H } _ { : , P } ^ { - 1 } ( ( { \bf H } ^ { - 1 } ) _ { P } ) ^ { - 1 } { \bf w } _ { P } , } \end{array} +$$ + +where $P$ denotes the set of indices corresponding to one block. Algorithm 1 can easily be adapted to operate on blocks using the above equations and applying the update of ${ \bf H } ^ { - 1 }$ via Lemma 1 successively for all $p \in P$ . Although there are now only $d _ { c o l } / c$ steps per row, each update of ${ \bf H } ^ { - 1 }$ also takes ${ \dot { O } } ( c \cdot d _ { \mathrm { c o l } } ^ { 2 } )$ time and so the overall asymptotic runtime stays the same. Additional practical overhead only comes from the extra $O ( c ^ { 2 } \cdot d _ { \mathrm { c o l } } ^ { 2 } )$ terms that are the result of computing and multiplying with the $c \times c$ matrices $\big ( ( \mathbf { H } ^ { - 1 } \big ) _ { P } \big ) ^ { - 1 }$ . + +# 5 The Optimal Brain Quantizer (OBQ) + +Although the classical OBS framework [23, 13] has inspired a long line of work on pruning methods for DNNs [39, 10, 27], so far it has not been used for quantization. We now show that our results from the previous section can in fact be extended to quantization in an effective and accurate way, via a method which we call the Optimal Brain Quantizer (OBQ), in the spirit of [23, 13]. + +The Quantization Order and Update Derivations. Under the standard assumption that the gradient at the current point w is negligible, the OBS formulas for the optimal weight to be pruned $w _ { p }$ and the corresponding update $\delta _ { p }$ can be derived by writing the locally quadratic problem under the constraint that element $p$ of $\delta _ { p }$ is equal to $- w _ { p }$ , which means that $w _ { p }$ is zero after applying the update to w. This problem has the following Lagrangian: + +$$ +\begin{array} { r } { L ( \delta _ { p } , \lambda ) = \delta _ { \mathbf { p } } ^ { \top } \mathbf H \delta _ { p } + \lambda ( \mathbf e _ { \mathbf { p } } ^ { \top } \delta _ { p } - ( - w _ { p } ) ) , } \end{array} +$$ + +where $\mathbf { H }$ denotes the Hessian at w and $\mathbf { e _ { p } }$ is the pth canonical basis vector. The optimal solution is then derived by first finding the optimal solution to $\delta _ { p }$ via setting the derivative $\partial L / \partial \delta _ { p }$ to zero and then substituting this solution back into $L$ and solving for $\lambda$ ; please see e.g. [13, 39] for examples. + +Assume a setting in which we are looking to quantize the weights in a layer on a fixed grid of width $\Delta$ while minimizing the loss. To map OBS to a quantized projection, we can set the target of the Lagrangian constraint in (6) to ( ${ \mathfrak { q u a n t } } ( w _ { p } ) - w _ { p } )$ , where quant $( w _ { p } )$ is the weight rounding given by quantization; then $w _ { p } = { \tt q u a n t } ( w _ { p } )$ after the update. + +Assuming we wish to quantize weights iteratively, one-at-a-time, we can derive formulas for the “optimal” weight to quantize at a step, in terms of minimizing the loss increase, and for the corresponding optimal update to the unquantized weights, in similar fashion as discussed above: + +$$ +\begin{array} { r } { w _ { p } = \operatorname * { a r g m i n } _ { w _ { p } } \frac { ( \mathrm { q u a n t } ( w _ { p } ) - w _ { p } ) ^ { 2 } } { [ \mathbf { H } ^ { - 1 } ] _ { p p } } , \quad \delta _ { p } = - \frac { w _ { p } - \mathrm { q u a n t } ( w _ { p } ) } { [ \mathbf { H } ^ { - 1 } ] _ { p p } } \cdot \mathbf { H } _ { : , p } ^ { - 1 } . } \end{array} +$$ + +In fact, since $- w _ { p }$ is a constant during all derivations, we can just substitute it with $( \mathrm { q u a n t } ( w _ { p } ) - w _ { p } )$ in the final result. We note that the resulting formulas are a generalization of standard OBS for pruning, if quant $( \cdot )$ always “quantizes” a weight to 0, then we recover the original form. + +Quantizing Full Layers. At first glance, OBQ might appear curious since one usually quantizes all weights in a layer, leaving no more weights to update. At the same time, the weight selection metric influences only the quantization order, but not the quantization value. However, this view changes when considering OBQ in the context of our efficient one-weight-at-a-time pruning algorithm described in the previous section. Specifically, using OBQ, we can greedily quantize the currently “easiest” weight by the above metric, and then adjust all the remaining unquantized weights to compensate for this loss of precision, thus changing their value. We then choose the next weight to quantize, and so on. This can result in quantization assignments that are different from the ones that would have been chosen by rounding initially, and in better overall quantization results. Concretely, to realize this, we can plug (7) into Algorithm 1 in order to iteratively quantize weights for a given layer, leading to the similar Algorithm in the Appendix, thus essentially unifying pruning and quantization. + +Quantization Outliers. One practical issue with this greedy scheme can occur especially when applied to quantization grids that permit some outliers in order to achieve a lower error on the majority of weights, which are currently standard [4, 34]. Since these outliers can have high quantization error, they will usually be quantized last, when there are only few other unquantized weights available that may be adjusted to compensate for the large error incurred by quantizing the outliers. This effect can become worse when some weights are pushed even further outside the grid by intermediate updates. We prevent this with a simple but effective heuristic: we quantize outliers, e.g. weights with a quantization error $> \Delta / 2$ where $\Delta$ is the distance between quantized values, as soon as they appear (which typically happens only a few times per layer). With this heuristic, OBQ yields a highly effective layer-wise quantization scheme, as our experiments in the next section demonstrate. Finally, we note that the OBQ version of the techniques discussed in Section 4 has all the same runtime and memory characteristics (barring the global step in Figure 1, which is unnecessary for quantization). + +# 6 Experiments + +Objectives, Models & Datasets. To demonstrate the effectiveness and flexibility of our method, we consider several different standard post-training compression scenarios [31, 19, 18]. We begin with settings where only a single type of compression is applied: concretely, we consider unstructured pruning for given FLOP targets, global 2:4 and 4:8 pruning, as well as uniform weight quantization. Additionally, we also study two practical tasks that feature joint pruning and quantization: a GPU scenario where quantization and N:M pruning are combined, as well as a CPU scenario combining quantization and block pruning. We work with variants of the following models and tasks: ResNet [14] for image classification on Imagenet [38], YOLOv5 [20] for object detection on COCO [26] and BERT [5] for question answering on SQuAD [36]. Our smaller BERT models denoted by BERT3 and BERT6 correspond to the smaller 3 and 6 layer variants of BERT-base, respectively, trained by [21]. The Appendix contains additional experiments as well as runtime information of our algorithms. + +Experimental Setup. All of our calibration datasets consist of 1024 random training samples. For ImageNet, where we use roughly $0 . 1 \%$ of the training data, we additionally apply standard flipping and cropping augmentations to artificially increase the size of this dataset by $1 0 \times$ ; other tasks do not use any augmentations. While the effect of augmentations is typically minor, they are very cheap to include for our method. For ResNet models, batchnorm statistics are reset using 100 batches of 128 samples from the calibration set with standard augmentations. For other models, we apply mean and variance correction [32, 1] after all normalization layers (so that the correction parameters can be easily merged and incur no extra cost) on a single batch of samples of size 128 (for YOLO) and 512 (for BERT). We found this to be more effective than batchnorm tuning for YOLO, and the BERT models have no batchnorm layers. + +When compressing to a given FLOP or timing constraint, we need to solve the problem of identifying per-layer compression targets, which match the constraint, while maximizing accuracy. To identify these non-uniform targets, we follow the approach of [10]: we first collect a “model database” containing for each compression level (e.g. bit-width or sparsity setting) the corresponding (independently) compressed version of each layer. For building a joint sparse and quantized database we simply sparsify layers first and then apply quantization to the remaining weights. Next, similarly to [19], we compute the layer-wise calibration losses (without augmentations) for all compression levels, corresponding to the models with exactly one layer compressed to a certain level. Then, given layer-wise FLOP or timing information, we set up a constrained layer-wise compression problem of the form described in AdaQuant [19] and solve it with the dynamic programming algorithm of SPDY [10]. This returns an optimal per-layer assignment of compression levels, for which we can then easily produce the corresponding model, via a two-step process: we first stitch together layers at the corresponding compression levels from the database, and then perform the discussed statistics correction to recover some extra accuracy [19]. + +Unstructured Sparsity. We begin our experiments with unstructured sparsity, comparing against global magnitude pruning (GMP) [46], the approximate layer-wise OBS method L-OBS [6], and the post-training pruning state-of-the-art method AdaPrune [18]. As a sanity check, we examine in Figure 2 whether our method provides better results in terms of layer-wise squared error, pruning the first layer of a ResNet18 (RN18) model to several sparsities. In this metric, ExactOBS performs best by a wide margin ahead of AdaPrune, which significantly outperforms the other two methods. + +Next, in Table 1, we turn our attention to the practical problem of pruning various models to achieve a given FLOP reduction of $2 \times - 4 \times$ , applying the per-layer target sparsity optimization technique described above. Our ExactOBS generally performs best (except for YOLOv5l $2 \times$ where all methods perform similarly in terms of mAP $@ 0 . 5$ ) and at $4 \times$ FLOP reduction even with a $> 1 \%$ gap to the next best method. Interestingly, on the hard-to-prune BERT model, ExactOBS appears to be the only method which still produces reasonable results at higher reduction targets. For BERT $3 \times$ and $4 \times$ , where the performance drop of all methods is $> 2 \%$ , we additionally assess the compatibility of our results with the more powerful (but also more expensive) post processing method global AdaPrune [10]. While this global optimization technique is able to recover lost accuracy, the ExactOBS models still maintain $a > 0 . { \bar { 5 } } \%$ and $> 2 \%$ F1 advantage, respectively (see Table 5). + +N:M Sparsity. Next, we study the performance of our method for semi-structured sparsity via the N:M pattern. Specifically, we compare against the 4:8 results of AdaPrune with batchnorm tuning [18] on ResNet models (see Table 2) in addition to a 2:4 comparison on BERT models (see Table 3). We highlight that ExactOBS matches or even slightly exceeds the 4:8 results of AdaPrune with the considerably more stringent 2:4 pattern, which is already well supported on NVIDIA hardware. Furthermore, in a 2:4 comparison on BERT models, ExactOBS achieves $1 { - } 2 \%$ higher F1 scores. + +![](images/e63c4464a79a22fecd310d811360589f3a30affa49cccf71cee521d9e2137d65.jpg) + +Table 1: Unstructured pruning for different FLOP reduction targets. + +
MethodResNet50 -76.13YOLOv5l-66.97BERT-88.53
2×3×4×2×3×4×2×3×4×
GMP74.8671.4464.8465.8362.3055.0965.6412.5209.23
L-OBS75.4873.7371.2466.2164.4761.1577.673.626.63
AdaPrune75.5374.4772.3966.0064.8862.7187.1270.3218.75
ExactOBS75.6475.0174.0566.1465.3564.0587.8185.8782.10
+ +Figure 2: RN18 squared error. +Table 2: Semi-structured N:M pruning ( $^ +$ batchnorm tuning) of all layers except the first and the last. + +
ModelDenseAdaPrune 4:8ExactOBS 2:4 4:8
ResNet1869.7668.6368.8169.18
ResNet3473.3172.3672.6672.95
ResNet5076.1374.7574.7175.20
+ +Table 3: Semi-structured 2:4 pruning of all layers except the embeddings. + +
ModelDenseAdaPruneExactOBS
BERT384.6682.7583.54
BERT688.3385.0286.97
BERT88.5385.2486.77
+ +Quantization. Additionally, we compare OBQ’s independent performance (after batchnorm tuning) with the state-of-the-art sequential post-training methods AdaQuant [19], AdaRound [31] and BRECQ [24]. We perform standard asymmetric per-channel quantization of all weights, using the authors’ implementations. We rerun all methods on Torchvision [28] ResNets to ensure a uniform baseline. The quantization grids for OBQ as well as AdaRound are determined with the same LAPQ [34] procedure that is used by BRECQ. Surprisingly, we find that, despite optimizing layers independently, OBQ achieves very similar (sometimes even slightly better) accuracies as existing non-independent methods for 4 and 3 bits. This suggests that it should be well-suited for mixed precision applications where one needs to quickly generate many non-uniform models optimized for different constraints. (However, we note that ExactOBS can also be applied sequentially; see Appendix.) + +Table 5: Further improving results in Table 1 with $5 \ 3 \%$ performance drops through more expensive post-processing via global AdaPrune (gAP). + +
MethodsBERT 3× 4×
gAP + AdaPrune86.9984.10
gAP +ExactOBS87.5786.42
+ +Table 4: Comparison with state-of-the-art post-training methods for asymmetric per-channel weight quantization of all layers. We mark whether methods are Layer-wise (Lw.) or Independent (Ind.). + +
MethodLw.Ind.ResNet18-69.76ResNet50-76.13
4bit3bit2bit4bit3bit2bit
AdaRound AdaQuant BRECQyesno69.3468.3763.3775.8475.1471.58
yesno68.1259.2100.1074.6864.9800.10
nono69.3768.4764.7075.8875.3272.41
OBQ(ours)yesyes69.5668.6964.0475.7275.2470.71
+ +BOP-Constrained Mixed GPU Compression. We now consider a practical setting where we are given a trained model together with some calibration data and want to compress this model for efficient inference on an NVIDIA GPU which supports 8-bit and 4-bit arithmetic, also in combination with 2:4 sparsity. Thus, there are 4 possible compression choices per layer: 8bit weights $+ ~ 8 \mathrm { b i t }$ activations (8w8a), 4w4a, $8 \mathrm { w } 8 \mathrm { a } + 2 { : } 4$ and $4 \mathrm { w } 4 \mathrm { a } + 2 { : } 4$ . Unlike in the previous section, we do symmetric per-channel quantization of the weights as it has better hardware support; activations are quantized asymmetrically per-tensor. We then generate mixed precision configurations for various BOP (number of bits times FLOPs) reduction targets and visualize the resulting compression-accuracy trade-off curves in Figure 3. In summary, at the cost of a $\approx 2 . 5 \%$ relative performance drop, we can achieve a $1 2 - 1 4 \times$ BOP reduction for ResNets and a $7 - 8 \times$ reduction for the more challenging YOLO and BERT models (relative to the compute in compressible layers). To the best of our knowledge, we are the first to consider joint N:M pruning and quantization in a post-training setting. Recent work [3] also studies joint $4 \mathrm { w } 4 \mathrm { a } + 2 { : } 4$ compression for ResNet18 but with 90 epochs of (sparse) Quantization-Aware Training (QAT) on the full dataset and report $6 7 . 3 3 \%$ accuracy. Although not perfectly comparable (we keep the first layer dense and their dense baseline has $0 . 9 4 \%$ higher accuracy and uses 4:8 sparse activations), we achieve similar $6 7 . 2 0 \%$ accuracy for $4 \mathrm { w } 4 \mathrm { a } + 2 { : } 4$ post training, which emphasizes the effectiveness of our methods for joint sparsification and quantization. + +Time-Constrained CPU Compression. Lastly, we explore a similar scenario, but targeting actual CPU inference speedup on a 12-core Intel Xeon Silver 4214 CPU using the DeepSparse inference engine [35, 22], which provides acceleration for joint 8-bit quantization and block-sparsity with blocksize 4. In this case, we work with real layer-wise timing data (for batchsize 64), as in [9]. There are 30 available block-sparsity targets per-layer, in steps of pruning $10 \%$ of the remaining weights, all of which are further quantized to 8 bits. The base acceleration of the dense 8 bit model is $\approx 2 . 7 \times$ on top of which sparsity speedup acts roughly multiplicatively. Figure 2d shows results for ResNet50 and several (real-time) speedup targets—we achieve $4 \times$ and $5 \times$ (actual) speedup with $1 \%$ and $2 \%$ accuracy loss, respectively. These are the first full post-training results in this setting (the authors of [10] only performed 4-block pruning post-training, followed by 5 epochs of QAT on the entire ImageNet dataset), and they show very encouraging accuracy-speedup trade-offs. + +![](images/48b7ae4ee49d35553186d141fe1a71bab7edce99ae5239859f187d9fe93726c1.jpg) +Mixed quantization pruning +Mixed quantization pruning +Figure 3: (a) to (c) Mixed quantization and 2:4 pruning for various BOP reduction targets. (d) Joint block-pruning and quantization for CPU inference time speedups. + +# 7 Conclusions & Future Work + +We have presented a new efficient and accurate approach for solving the layer-wise compression problem, and built on it to obtain state-of-the-art post-training compression solutions for both pruning and quantization. Our framework should be naturally extensible to structured pruning, which in fact should allow for further optimizations, and should also be compatible with further compression via unstructured pruning and quantization. Our results suggest that post-training compression may be able to reach comparable accuracies to much more expensive retraining methods. We plan to investigate this in future work, in particular in the context of more resource-intensive models, such as very large-scale language models. + +# 8 Acknowledgements + +We gratefully acknowledge funding from the European Research Council (ERC) under the European Union’s Horizon 2020 programme (grant agreement No 805223 ScaleML), as well as computational support from AWS EC2. We thank Eldar Kurtic for providing us BERT code and pretrained models, and the Neural Magic Team, notably Michael Goin and Mark Kurtz, for support with their software. + +References +[1] Ron Banner, Yury Nahshan, and Daniel Soudry. Post training 4-bit quantization of convolutional networks for rapid-deployment. In Conference on Neural Information Processing Systems (NeurIPS), 2019. +[2] Thomas Blumensath and Mike E Davies. Iterative thresholding for sparse approximations. Journal of Fourier Analysis and Applications, 14(5-6):629–654, 2008. +[3] Brian Chmiel, Itay Hubara, Ron Banner, and Daniel Soudry. Optimal fine-grained N:M sparsity for activations and neural gradients. arXiv preprint arXiv:2203.10991, 2022. +[4] Yoni Choukroun, Eli Kravchik, Fan Yang, and Pavel Kisilev. Low-bit quantization of neural networks for efficient inference. In International Conference on Computer Vision Workshop (ICCVW), 2019. +[5] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. In North American Chapter of the Association for Computational Linguistics (NAACL), 2019. +[6] Xin Dong, Shangyu Chen, and Sinno Jialin Pan. Learning to prune deep neural networks via layer-wise optimal brain surgeon. In Conference on Neural Information Processing Systems (NeurIPS), 2017. +[7] Erich Elsen, Marat Dukhan, Trevor Gale, and Karen Simonyan. Fast sparse convnets. In Conference on Computer Vision and Pattern Recognition (CVPR), 2020. +[8] Utku Evci, Nicolas Le Roux, Pablo Castro, and Leon Bottou. Mean replacement pruning. 2018. +[9] Elias Frantar and Dan Alistarh. SPDY: Accurate pruning with speedup guarantees. arXiv preprint arXiv:2201.13096, 2022. +[10] Elias Frantar, Eldar Kurtic, and Dan Alistarh. M-FAC: Efficient matrix-free approximations of second-order information. In Conference on Neural Information Processing Systems (NeurIPS), 2021. +[11] Amir Gholami, Sehoon Kim, Zhen Dong, Zhewei Yao, Michael W Mahoney, and Kurt Keutzer. A survey of quantization methods for efficient neural network inference. arXiv preprint arXiv:2103.13630, 2021. +[12] Roger Grosse and James Martens. A Kronecker-factored approximate Fisher matrix for convolution layers. In International Conference on Machine Learning (ICML), 2016. +[13] Babak Hassibi, David G Stork, and Gregory J Wolff. Optimal brain surgeon and general network pruning. In IEEE International Conference on Neural Networks, 1993. +[14] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Conference on Computer Vision and Pattern Recognition (CVPR), 2016. +[15] Yihui He, Ji Lin, Zhijian Liu, Hanrui Wang, Li-Jia Li, and Song Han. AMC: AutoML for model compression and acceleration on mobile devices. In European Conference on Computer Vision (ECCV), 2018. +[16] Yihui He, Xiangyu Zhang, and Jian Sun. Channel pruning for accelerating very deep neural networks. In International Conference on Computer Vision (ICCV), 2017. +[17] Torsten Hoefler, Dan Alistarh, Tal Ben-Nun, Nikoli Dryden, and Alexandra Peste. Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks. arXiv preprint arXiv:2102.00554, 2021. +[18] Itay Hubara, Brian Chmiel, Moshe Island, Ron Banner, Seffi Naor, and Daniel Soudry. Accelerated sparse neural training: A provable and efficient method to find N:M transposable masks. In Conference on Neural Information Processing Systems (NeurIPS), 2021. + +[19] Itay Hubara, Yury Nahshan, Yair Hanani, Ron Banner, and Daniel Soudry. Accurate post training quantization with small calibration sets. In International Conference on Machine Learning (ICML), 2021. + +[20] Glenn Jocher. YOLOv5. https://github.com/ultralytics/yolov5, 2022. +[21] Eldar Kurtic, Daniel Campos, Tuan Nguyen, Elias Frantar, Mark Kurtz, Benjamin Fineran, Michael Goin, and Dan Alistarh. The Optimal BERT Surgeon: Scalable and accurate secondorder pruning for large language models. arXiv preprint arXiv:2203.07259, 2022. +[22] Mark Kurtz, Justin Kopinsky, Rati Gelashvili, Alexander Matveev, John Carr, Michael Goin, William Leiserson, Sage Moore, Bill Nell, Nir Shavit, and Dan Alistarh. Inducing and exploiting activation sparsity for fast inference on deep neural networks. In International Conference on Machine Learning (ICML), 2020. +[23] Yann LeCun, John S Denker, and Sara A Solla. Optimal brain damage. In Conference on Neural Information Processing Systems (NeurIPS), 1990. +[24] Yuhang Li, Ruihao Gong, Xu Tan, Yang Yang, Peng Hu, Qi Zhang, Fengwei Yu, Wei Wang, and Shi Gu. BRECQ: Pushing the limit of post-training quantization by block reconstruction. In International Conference on Learning Representations (ICLR), 2021. +[25] Tailin Liang, John Glossner, Lei Wang, Shaobo Shi, and Xiaotong Zhang. Pruning and quantization for deep neural network acceleration: A survey. Neurocomputing, 461:370–403, 2021. +[26] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft COCO: Common objects in context. In European Conference on Computer Vision (ECCV), 2014. +[27] Liyang Liu, Shilong Zhang, Zhanghui Kuang, Aojun Zhou, Jing-Hao Xue, Xinjiang Wang, Yimin Chen, Wenming Yang, Qingmin Liao, and Wayne Zhang. Group fisher pruning for practical network compression. In International Conference on Machine Learning (ICML), 2021. +[28] Sébastien Marcel and Yann Rodriguez. Torchvision the machine-vision package of torch. In ACM International Conference on Multimedia, 2010. +[29] James Martens and Roger Grosse. Optimizing neural networks with kronecker-factored approximate curvature. In International Conference on Machine Learning (ICML), 2015. +[30] Asit Mishra, Jorge Albericio Latorre, Jeff Pool, Darko Stosic, Dusan Stosic, Ganesh Venkatesh, Chong Yu, and Paulius Micikevicius. Accelerating sparse deep neural networks. arXiv preprint arXiv:2104.08378, 2021. +[31] Markus Nagel, Rana Ali Amjad, Mart Van Baalen, Christos Louizos, and Tijmen Blankevoort. Up or down? Adaptive rounding for post-training quantization. In International Conference on Machine Learning (ICML), 2020. +[32] Markus Nagel, Mart van Baalen, Tijmen Blankevoort, and Max Welling. Data-free quantization through weight equalization and bias correction. In International Conference on Computer Vision (ICCV), 2019. +[33] Markus Nagel, Marios Fournarakis, Rana Ali Amjad, Yelysei Bondarenko, Mart van Baalen, and Tijmen Blankevoort. A white paper on neural network quantization. arXiv preprint arXiv:2106.08295, 2021. +[34] Yury Nahshan, Brian Chmiel, Chaim Baskin, Evgenii Zheltonozhskii, Ron Banner, Alex M Bronstein, and Avi Mendelson. Loss aware post-training quantization. Machine Learning, 110(11):3245–3262, 2021. +[35] NeuralMagic. The DeepSparse Inference Engine. https://github.com/neuralmagic/deepsparse, 2022. +[36] Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. SQuAD: $^ { 1 0 0 , 0 0 0 + }$ questions for machine comprehension of text. In Conference on Empirical Methods in Natural Language Processing (EMNLP), 2016. +[37] Vijay Janapa Reddi, Christine Cheng, David Kanter, Peter Mattson, Guenther Schmuelling, Carole-Jean Wu, Brian Anderson, Maximilien Breughe, Mark Charlebois, William Chou, et al. Mlperf inference benchmark. In 2020 ACM/IEEE 47th Annual International Symposium on Computer Architecture (ISCA), pages 446–459. IEEE, 2020. +[38] Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge. International Journal of Computer Vision, 115(3):211–252, 2015. +[39] Sidak Pal Singh and Dan Alistarh. WoodFisher: Efficient second-order approximation for neural network compression. In Conference on Neural Information Processing Systems (NeurIPS), 2020. +[40] Chaoqi Wang, Roger Grosse, Sanja Fidler, and Guodong Zhang. Eigendamage: Structured pruning in the Kronecker-factored eigenbasis. In International Conference on Machine Learning (ICML), 2019. +[41] Kuan Wang, Zhijian Liu, Yujun Lin, Ji Lin, and Song Han. HAQ: Hardware-aware automated quantization with mixed precision. In Conference on Computer Vision and Pattern Recognition (CVPR), 2019. +[42] Peisong Wang, Qiang Chen, Xiangyu He, and Jian Cheng. Towards accurate post-training network quantization via bit-split and stitching. In International Conference on Machine Learning (ICML), 2020. +[43] Haichuan Yang, Shupeng Gui, Yuhao Zhu, and Ji Liu. Automatic neural network compression by sparsity-quantization joint learning: A constrained optimization-based approach. In Conference on Computer Vision and Pattern Recognition (CVPR), 2020. +[44] Zhewei Yao, Zhen Dong, Zhangcheng Zheng, Amir Gholami, Jiali Yu, Eric Tan, Leyuan Wang, Qijing Huang, Yida Wang, Michael Mahoney, et al. HAWQ-v3: Dyadic neural network quantization. In International Conference on Machine Learning (ICML), 2021. +[45] Aojun Zhou, Yukun Ma, Junnan Zhu, Jianbo Liu, Zhijie Zhang, Kun Yuan, Wenxiu Sun, and Hongsheng Li. Learning N:M fine-grained structured sparse neural networks from scratch. In International Conference on Learning Representations (ICLR), 2021. +[46] Michael Zhu and Suyog Gupta. To prune, or not to prune: exploring the efficacy of pruning for model compression. arXiv preprint arXiv:1710.01878, 2017. + +# Checklist + +1. For all authors... + +(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? [Yes] +(b) Did you describe the limitations of your work? [Yes] +(c) Did you discuss any potential negative societal impacts of your work? [N/A] +(d) Have you read the ethics review guidelines and ensured that your paper conforms to them? [Yes] + +2. If you are including theoretical results... + +(a) Did you state the full set of assumptions of all theoretical results? [Yes] (b) Did you include complete proofs of all theoretical results? [Yes] + +3. If you ran experiments... + +(a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] +(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] +(c) Did you report error bars (e.g., with respect to the random seed after running experiments multiple times)? [Yes] +(d) Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] + +4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets... + +(a) If your work uses existing assets, did you cite the creators? [Yes] +(b) Did you mention the license of the assets? [N/A] +(c) Did you include any new assets either in the supplemental material or as a URL? [No] Assets are open-source with standard permissive licenses. +(d) Did you discuss whether and how consent was obtained from people whose data you’re using/curating? [N/A] +(e) Did you discuss whether the data you are using/curating contains personally identifiable information or offensive content? [N/A] + +5. If you used crowdsourcing or conducted research with human subjects... + +(a) Did you include the full text of instructions given to participants and screenshots, if applicable? [N/A] +(b) Did you describe any potential participant risks, with links to Institutional Review Board (IRB) approvals, if applicable? [N/A] +(c) Did you include the estimated hourly wage paid to participants and the total amount spent on participant compensation? [N/A] \ No newline at end of file diff --git a/parse/dev/nrksGSRT7kX/images/1b1d9721867cb60538e16634cfa70e10ace7d370403db3db388fb3277a7358af.jpg b/parse/dev/nrksGSRT7kX/images/1b1d9721867cb60538e16634cfa70e10ace7d370403db3db388fb3277a7358af.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6bc7cde220230610bc17c53ea6018207b5d4f2bb --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/1b1d9721867cb60538e16634cfa70e10ace7d370403db3db388fb3277a7358af.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8ac26d645f9581f50173cbf57d3326d2dab67e6a43254e7612f31848178c3e8 +size 7033 diff --git a/parse/dev/nrksGSRT7kX/images/33246d70376ca9748239423f153f8657d6c0e9d252dde15a5276ce9601fd2cd3.jpg b/parse/dev/nrksGSRT7kX/images/33246d70376ca9748239423f153f8657d6c0e9d252dde15a5276ce9601fd2cd3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..24790afa2cb3f59de5f271007be63c72aa8b5eac --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/33246d70376ca9748239423f153f8657d6c0e9d252dde15a5276ce9601fd2cd3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5b07bbc868bd2e37194b54bef4d5db4f2050dd38dd2f589b73448bcb3ee3a0d +size 3060 diff --git a/parse/dev/nrksGSRT7kX/images/5233ee4e813ae4ab1f1a12cf9d1d67d68c9f4afd399d1d5e5b89be29418a02e0.jpg b/parse/dev/nrksGSRT7kX/images/5233ee4e813ae4ab1f1a12cf9d1d67d68c9f4afd399d1d5e5b89be29418a02e0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e62bc451a19e8e968852733fb3a8e72985fc6ff0 --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/5233ee4e813ae4ab1f1a12cf9d1d67d68c9f4afd399d1d5e5b89be29418a02e0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:79ab93d8783c6697ebafe9034c4c8604a55c6a51d98ab7fd59431a0579137cf0 +size 7497 diff --git a/parse/dev/nrksGSRT7kX/images/56fd88b4db91e208a1ffc8d901212f0014dd628b14e998f2f854a7f933b139e6.jpg b/parse/dev/nrksGSRT7kX/images/56fd88b4db91e208a1ffc8d901212f0014dd628b14e998f2f854a7f933b139e6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..08d652e9d101dfa9e8319d57cd9175465b2a384b --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/56fd88b4db91e208a1ffc8d901212f0014dd628b14e998f2f854a7f933b139e6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6094ea14674ab0dedc6aa4c461f7f453da0715da945555b56600a5cc5416c331 +size 11626 diff --git a/parse/dev/nrksGSRT7kX/images/7771d0cde31e799d3324c011028e34be987ccaa970d4f6fb85f9cd5ec50b81fc.jpg b/parse/dev/nrksGSRT7kX/images/7771d0cde31e799d3324c011028e34be987ccaa970d4f6fb85f9cd5ec50b81fc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..593cc059f7a2e2320443ac440d5c3b1d6c10f054 --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/7771d0cde31e799d3324c011028e34be987ccaa970d4f6fb85f9cd5ec50b81fc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef5684354e410bcb8f63b701e4f37bddea77767f23af8aee2b255431518fe3d6 +size 10512 diff --git a/parse/dev/nrksGSRT7kX/images/82079e130502e99af45a8c66cbfb8c8b75ae64f3a1308daee15562b6526754a5.jpg b/parse/dev/nrksGSRT7kX/images/82079e130502e99af45a8c66cbfb8c8b75ae64f3a1308daee15562b6526754a5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7fce353738091f372964e14f55a5d0ebfd886b5b --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/82079e130502e99af45a8c66cbfb8c8b75ae64f3a1308daee15562b6526754a5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49d4a5b950fccd2ff2ef7b04bb361af459f0416898953461249019741ef18b75 +size 10517 diff --git a/parse/dev/nrksGSRT7kX/images/865ac85a87414b43e2a869c7219361fbbe0e2ddb0700f829cd704fd52730cc93.jpg b/parse/dev/nrksGSRT7kX/images/865ac85a87414b43e2a869c7219361fbbe0e2ddb0700f829cd704fd52730cc93.jpg new file mode 100644 index 0000000000000000000000000000000000000000..df10e3f136b986f65cd7a9d05342aa34f38e0f16 --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/865ac85a87414b43e2a869c7219361fbbe0e2ddb0700f829cd704fd52730cc93.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:892ea766dc8ad9cfc517252cab13809578e3aae6bf85d8f9f9a744c801a39deb +size 4706 diff --git a/parse/dev/nrksGSRT7kX/images/86e5d7a10e7dcb5bdee486b0e197cc7f036aa5d6249f362a7cc2168321c28e0d.jpg b/parse/dev/nrksGSRT7kX/images/86e5d7a10e7dcb5bdee486b0e197cc7f036aa5d6249f362a7cc2168321c28e0d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..55e82987a53e0c0fd5a1e2f5c122b3169e2216dc --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/86e5d7a10e7dcb5bdee486b0e197cc7f036aa5d6249f362a7cc2168321c28e0d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7051685b30f466ad9e8875a156b9ff5217a912abf7b4ee73aaaa2da883df6312 +size 4173 diff --git a/parse/dev/nrksGSRT7kX/images/8b689c68e8ab8947e74d86ee4e3ad0adc3a0304330d236abe24584e76b30211e.jpg b/parse/dev/nrksGSRT7kX/images/8b689c68e8ab8947e74d86ee4e3ad0adc3a0304330d236abe24584e76b30211e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cee9454d2e783b5d5f4adf1a62d2282f2daed42d --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/8b689c68e8ab8947e74d86ee4e3ad0adc3a0304330d236abe24584e76b30211e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f366363ede6e4f33b2ed8f2753451292d358102790ebd1deda58efc97a83ee72 +size 10212 diff --git a/parse/dev/nrksGSRT7kX/images/9529d743ff4787c1fb6a32149289d5ec62c8b4518a9c148498ebe5f1dafd14a8.jpg b/parse/dev/nrksGSRT7kX/images/9529d743ff4787c1fb6a32149289d5ec62c8b4518a9c148498ebe5f1dafd14a8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d33dd638b765abcae6d93e354ce253dc40bc5f9b --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/9529d743ff4787c1fb6a32149289d5ec62c8b4518a9c148498ebe5f1dafd14a8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2763481ed4ddf3acccf34c495725bfb70152c2cad17b4553dbbdc6fa75064cba +size 9574 diff --git a/parse/dev/nrksGSRT7kX/images/ac84323481fe6c04c61be46f9a3893ddacf3303c7ef34906494f3c0edd8d084d.jpg b/parse/dev/nrksGSRT7kX/images/ac84323481fe6c04c61be46f9a3893ddacf3303c7ef34906494f3c0edd8d084d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b0c423888477f31971dd408c363554caf7aa12b6 --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/ac84323481fe6c04c61be46f9a3893ddacf3303c7ef34906494f3c0edd8d084d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa40930e225ca0b54e8067beab8ecb4f620f94ae0395d912f6c5f5d376f9daf7 +size 5474 diff --git a/parse/dev/nrksGSRT7kX/images/b9c266237b592a84fb667c6d9f0b17a17a8853e1660580310318ac0d0ce7b9f4.jpg b/parse/dev/nrksGSRT7kX/images/b9c266237b592a84fb667c6d9f0b17a17a8853e1660580310318ac0d0ce7b9f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..75a98e8e8f37583e15cc7d0a50ce7f4342eddb87 --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/b9c266237b592a84fb667c6d9f0b17a17a8853e1660580310318ac0d0ce7b9f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f922b3b7ea3459a96db467cd1317ccc1b73aa2e0d7ce7434b468d24a65beecf3 +size 171474 diff --git a/parse/dev/nrksGSRT7kX/images/f1fed017dcc24f7c8cd56eff01f719cbaf4cd08334f02a7068b0e6a1bbcd6b4a.jpg b/parse/dev/nrksGSRT7kX/images/f1fed017dcc24f7c8cd56eff01f719cbaf4cd08334f02a7068b0e6a1bbcd6b4a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4cdb0dc81332be5d260ddb93b87ce5a5e6e49fd0 --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/f1fed017dcc24f7c8cd56eff01f719cbaf4cd08334f02a7068b0e6a1bbcd6b4a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:706a04021995e0bf123d3e41b91aada427de81e125a8bab7ae9df12c4e219366 +size 7212 diff --git a/parse/dev/nrksGSRT7kX/images/f21e700f0c60b5897124b3a670b14f3ae226a597eeb8c70b1ba31dc99624eda2.jpg b/parse/dev/nrksGSRT7kX/images/f21e700f0c60b5897124b3a670b14f3ae226a597eeb8c70b1ba31dc99624eda2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b12682bfbee7781c0e54f79b2bc74616ca5b76c6 --- /dev/null +++ b/parse/dev/nrksGSRT7kX/images/f21e700f0c60b5897124b3a670b14f3ae226a597eeb8c70b1ba31dc99624eda2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ba82918cc0580bc0a59460b7e0643a26c8e2a31856f1a736265de01b5052959 +size 8363 diff --git a/parse/dev/o4neHaKMlse/images/037dbf79e243a4d3cb550e85077d658d5b3d82c68467ce3242dca9c39da14692.jpg b/parse/dev/o4neHaKMlse/images/037dbf79e243a4d3cb550e85077d658d5b3d82c68467ce3242dca9c39da14692.jpg new file mode 100644 index 0000000000000000000000000000000000000000..85c1accdb77256abd97acfab87b59df8151ba64e --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/037dbf79e243a4d3cb550e85077d658d5b3d82c68467ce3242dca9c39da14692.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09ab40550e3f452720da882ec8e059413272f16948ae195ecafe558760486960 +size 57171 diff --git a/parse/dev/o4neHaKMlse/images/13c6fad22e1d7cc8f86ea316c9a86715b07e520fa826ba36909e966fd4dac362.jpg b/parse/dev/o4neHaKMlse/images/13c6fad22e1d7cc8f86ea316c9a86715b07e520fa826ba36909e966fd4dac362.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a6cc5303ef88513ddd1fa5d9db7470fda030a872 --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/13c6fad22e1d7cc8f86ea316c9a86715b07e520fa826ba36909e966fd4dac362.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2101e44d7df6cb16fcbf48983630b5f8fcbb9b8594cd95c47aeb95d48c06e0f +size 113723 diff --git a/parse/dev/o4neHaKMlse/images/218abfcb7bd765b1713f3f79c3bf274aa825cb24799eb579a317663d24e4f436.jpg b/parse/dev/o4neHaKMlse/images/218abfcb7bd765b1713f3f79c3bf274aa825cb24799eb579a317663d24e4f436.jpg new file mode 100644 index 0000000000000000000000000000000000000000..966e84d3329559226a9c1ce7e79dc14d93060ee3 --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/218abfcb7bd765b1713f3f79c3bf274aa825cb24799eb579a317663d24e4f436.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7d0d816f2fc93e79b7bec64f6b6f4750b867e6fe682298b9ca5c61c712df98f +size 8530 diff --git a/parse/dev/o4neHaKMlse/images/3fadeb580476fb1227942093e53b97f626cde8f0ad9ef984a5d257476c567fb3.jpg b/parse/dev/o4neHaKMlse/images/3fadeb580476fb1227942093e53b97f626cde8f0ad9ef984a5d257476c567fb3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5717e0f4d3daa8a5434c27773172731c19d04463 --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/3fadeb580476fb1227942093e53b97f626cde8f0ad9ef984a5d257476c567fb3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2586ab2f39aef85eb98c994200e5c1b5e93f5abe34f71fdca3530573f86d8927 +size 72387 diff --git a/parse/dev/o4neHaKMlse/images/41a8b93dc045d81402659ae1325af9c40db37a5e03fb463aad1332c5cf46aa93.jpg b/parse/dev/o4neHaKMlse/images/41a8b93dc045d81402659ae1325af9c40db37a5e03fb463aad1332c5cf46aa93.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0ce1b637b78f81dced3c5acf56ea97d7b265b930 --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/41a8b93dc045d81402659ae1325af9c40db37a5e03fb463aad1332c5cf46aa93.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb417772102d0725c8bdb4b9133ac27861f6ee62f7174a66832e50aa0711d758 +size 87496 diff --git a/parse/dev/o4neHaKMlse/images/56ac8524c9003e78c420110abadb90484e4955fd886d3631310593a524cc0c16.jpg b/parse/dev/o4neHaKMlse/images/56ac8524c9003e78c420110abadb90484e4955fd886d3631310593a524cc0c16.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e2deb2422641203dbeceb2d791a1a12100ba259c --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/56ac8524c9003e78c420110abadb90484e4955fd886d3631310593a524cc0c16.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cde3a49db7ecc49327b029cd8fdebee82ece9df30152e35e43a94eb22e592dfe +size 66971 diff --git a/parse/dev/o4neHaKMlse/images/59ab2a9918c66b9e49af9c3bd096b5bae7c30e64401511f747661fde80d3d915.jpg b/parse/dev/o4neHaKMlse/images/59ab2a9918c66b9e49af9c3bd096b5bae7c30e64401511f747661fde80d3d915.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cdb7f2dcefea72b18f2df74a55ffbdf177a6ce7e --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/59ab2a9918c66b9e49af9c3bd096b5bae7c30e64401511f747661fde80d3d915.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27153b31f022aa6ee7ae18812ab5b83cdf2d8f01d0ecd6fdf467019a9ddc27cb +size 33401 diff --git a/parse/dev/o4neHaKMlse/images/5d2d42f1068123eead76a9dd306fcbce49165a2f6f3af4a899aadd4103b80f2b.jpg b/parse/dev/o4neHaKMlse/images/5d2d42f1068123eead76a9dd306fcbce49165a2f6f3af4a899aadd4103b80f2b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..357007261f53b94852334e789831fc25b98fdb90 --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/5d2d42f1068123eead76a9dd306fcbce49165a2f6f3af4a899aadd4103b80f2b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e8b0ae0f9d86cc5c257339e3d6cff26bf3ae25044883a982ec8b62f80044eab +size 8725 diff --git a/parse/dev/o4neHaKMlse/images/6f874a920aa1b0ad5bc3edaa3520dfdb73f51ef8fff81394a975d72eb478cec8.jpg b/parse/dev/o4neHaKMlse/images/6f874a920aa1b0ad5bc3edaa3520dfdb73f51ef8fff81394a975d72eb478cec8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..39756b19853b8787d6d6b7b1d417df614eb6c86d --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/6f874a920aa1b0ad5bc3edaa3520dfdb73f51ef8fff81394a975d72eb478cec8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b073fa1a6de9293e0d62c645e6db324df248a428a50ba70899be1a7db775985 +size 72330 diff --git a/parse/dev/o4neHaKMlse/images/7181a3b812919811201dfad5015043981b35bac92b9f41f10431467f9e3803ce.jpg b/parse/dev/o4neHaKMlse/images/7181a3b812919811201dfad5015043981b35bac92b9f41f10431467f9e3803ce.jpg new file mode 100644 index 0000000000000000000000000000000000000000..640918353fd4ad01a071f208f2aa5888550ba1d1 --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/7181a3b812919811201dfad5015043981b35bac92b9f41f10431467f9e3803ce.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4191c19be7fe2778c7014efd256b92c2a0041674b6be83358250f9b72d596ca6 +size 120970 diff --git a/parse/dev/o4neHaKMlse/images/8918d3f1c84a88ec6c6ec7ade53beed6ebaa62fc1173069e2e7802e7485a9ccf.jpg b/parse/dev/o4neHaKMlse/images/8918d3f1c84a88ec6c6ec7ade53beed6ebaa62fc1173069e2e7802e7485a9ccf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8e6e66af04385050b99c802b6d36ea151551056f --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/8918d3f1c84a88ec6c6ec7ade53beed6ebaa62fc1173069e2e7802e7485a9ccf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c239eaa3ead9bacdf11c5fd3b1655fefc09cba2c2ccb2544d7f9b38a26b9a917 +size 20483 diff --git a/parse/dev/o4neHaKMlse/images/8afe3400f4173ca1b0f613fa777f89bb7d2728cb877a2293a26fa7db7f619fb8.jpg b/parse/dev/o4neHaKMlse/images/8afe3400f4173ca1b0f613fa777f89bb7d2728cb877a2293a26fa7db7f619fb8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c96c019deb91c9ba820fd6f0dc4635915d0de392 --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/8afe3400f4173ca1b0f613fa777f89bb7d2728cb877a2293a26fa7db7f619fb8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:580a5ac0785f479fa45e8bbdde762938ab18f8ebe9daed444bf8177e323923c3 +size 94564 diff --git a/parse/dev/o4neHaKMlse/images/9f15e12e946c7e98a661c6f229461dc50231eba805caf50b8753e903cd947328.jpg b/parse/dev/o4neHaKMlse/images/9f15e12e946c7e98a661c6f229461dc50231eba805caf50b8753e903cd947328.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8679501b11943235323258316d21e8d61fef5234 --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/9f15e12e946c7e98a661c6f229461dc50231eba805caf50b8753e903cd947328.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8b3e0ad5feabe01baafb12affe78e594d09cfc31aa217d42cdbcf46e9b1fb47 +size 15317 diff --git a/parse/dev/o4neHaKMlse/images/d0b9e66ed1dc113ec499f2b29821d6ccdaa4937e652e0097138479e7966e9a87.jpg b/parse/dev/o4neHaKMlse/images/d0b9e66ed1dc113ec499f2b29821d6ccdaa4937e652e0097138479e7966e9a87.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6bf363141d1e60840b64cd454cd262e344ff33a4 --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/d0b9e66ed1dc113ec499f2b29821d6ccdaa4937e652e0097138479e7966e9a87.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:917a695895c46e4f80edfb419b44093c53237a0757f681e6f4da0fd0eaa812c8 +size 72419 diff --git a/parse/dev/o4neHaKMlse/images/fa28bd4542719e59db24d0b57b24f7833ac26b73e3cc1c80771962138fa20c51.jpg b/parse/dev/o4neHaKMlse/images/fa28bd4542719e59db24d0b57b24f7833ac26b73e3cc1c80771962138fa20c51.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f385f53670873303bd575416783ca6295d6b168d --- /dev/null +++ b/parse/dev/o4neHaKMlse/images/fa28bd4542719e59db24d0b57b24f7833ac26b73e3cc1c80771962138fa20c51.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e13cd31e0dfba0515f6b19fe0d2965b798adca5435e67ffebfd0424e233a24f4 +size 18566 diff --git a/parse/dev/r5hq-Ooh_Ba/r5hq-Ooh_Ba.md b/parse/dev/r5hq-Ooh_Ba/r5hq-Ooh_Ba.md new file mode 100644 index 0000000000000000000000000000000000000000..1993b657bd4caf02586b29fc1737576662de137d --- /dev/null +++ b/parse/dev/r5hq-Ooh_Ba/r5hq-Ooh_Ba.md @@ -0,0 +1,373 @@ +# MUTEXMATCH: SEMI-SUPERVISED LEARNING WITH MUTEX-BASED CONSISTENCY REGULARIZATION + +Anonymous authors Paper under double-blind review + +# ABSTRACT + +The core issue in semi-supervised learning (SSL) lies in how to effectively leverage unlabeled data, whereas most existing methods usually concentrate on the utilization of high-confidence samples yet seldom fully explore the usage of lowconfidence samples. Early SSL methods mostly require low-confidence samples to optimize the same loss function as high-confidence samples, but this setting might largely challenge the low-confidence samples especially at the early training stage. In this paper, we aim to utilize low-confidence samples in a novel way, which is realized by our proposed mutex-based consistency regularization, namely MutexMatch. To be specific, the high-confidence samples are required to exactly predict “What it is” by conventional True-Positive Classifier, while the low-confidence samples, for a much simpler goal, are employed to predict “What it is not” by True-Negative Classifier with ease. In this way, we not only mitigate the pseudo-labeling errors but also make full use of the low-confidence unlabeled data in the training stage. The proposed MutexMatch achieves superior performance on multiple benchmark datasets, i.e., CIFAR-10, CIFAR-100, SVHN, and STL-10. Particularly, our method shows further superiority under few quantities of labeled data, e.g., $9 1 . 7 7 \%$ accuracy with only 20 labeled data on CIFAR-10. + +# 1 INTRODUCTION + +Aiming to escape from time-consuming and laborious labeling tasks, semi-supervised learning (SSL) (Chapelle et al., 2009; Zhu, 2017) has been a longstanding yet important direction to leverage a large quantity of unlabeled data along with few labeled data during training. Recent SSL models could be categorized into consistency regularization based or entropy minimization based methods where the utilization of unlabeled data is crucial in both. In particular, consistency regularization based methods like (Laine & Aila, 2016; Tarvainen & Valpola, 2017) intend to utilize all unlabeled data together with the supervision on labeled data, which is at the risk of strong confirmation bias. Although recent holistic methods such as Sohn et al. (2020) realize consistency regularization by combining entropy minimization via pseudo labeling, a fatal limitation is that they set a confidence threshold to control whether unlabeled data should participate in training, preventing low-confidence unlabeled data from being effectively involved. Different from consistency based methods, recent entropy minimization based methods Rizve et al. (2021) employ pseudo labeling to iteratively incorporate a part of low-confidence samples into training process. However, in this way, in addition to possible error accumulation, some low-confidence unlabeled samples are still neglected. In a nutshell, the waste of unlabeled samples with low confidence causes the model difficult to learn the potential pattern from all unlabeled data, which might deteriorate the final performance. + +Being aware of the aforementioned limitations, we try to answer — if we could treat these lowconfidence unlabeled samples in a novel way? As shown in Figure 1, imaging a low-confidence sample with its actual label of a “horse”, it might be hard for a trained model to propose an accurate prediction, i.e., “it is a horse”. On the contrary, it can be much easier for the model to “guess what it is not, e.g., it is not a cat”. This drives us to consider a straightforward yet feasible direction — for low-confidence samples, can we design a paradigm to exclude “what it is not” to benefit the learning of “what it is”. Intuitively, by introducing this paradigm, the space of searching the optimal classifier could be largely reduced since the most impossible classes are initially excluded. + +That is to say, for a low-confidence image, it is unnecessary to get a certain class, and its complementary pseudo-label is easier to obtain. Thus, we could learn less error information when using unlabeled data with low confidence. + +As aforementioned, to leverage all unlabeled data in a novel way, we propose MutexMatch, a new framework of SSL using mutex-based consistency regularization. We utilize True-Positive Classifier (TPC): to predict “What it is”, and TrueNegative Classifier (TNC): to predict “What it is not”, which is designed to learn feature representation of unlabeled data from a mutex perspective. An improvement of MutexMatch compared with the existing methods is that it allows low-confidence unlabeled samples to participate in training by optimizing a much simpler objective compared with that of high-confidence unlabeled samples. Inspired by FixMatch (Sohn et al., 2020), the weakly-augmented unlabeled samples are used to generate pseudo-labels, and we use RandAugument (Cubuk et al., 2020) for strong augmentation. We set a threshold to control the high-confidence portion and low-confidence portion of pseudo-labels. In high-confidence portion and low-confidence portion, we enforce the consistency regularization on the output of TPC and TNC, respectively. A diagram of MutexMatch is shown in Figure 2. + +![](images/207647779515a6cfe4c418f405c73582ec86d916211dba5834302b336e5fbcd8.jpg) +Figure 1: Graphical explanations of the “Exclusion Method” for classification task. For an image, the proposed method excludes some wrong classes via different images of the same class, so as to eliminate the wrong answers. + +In this work, the key contributions include three aspects: (1) We propose mutex-based consistency regularization for SSL, which can make full use of unlabeled data in a more effective way; (2) We use two classifiers (TPC and TNC) to construct MutexMatch, a novel framework using pseudolabel and complementary label to learn an informative representation of unlabeled samples; (3) By exploiting all unlabeled data, we can obtain better classification results in a label-scarce setting than recently-proposed SSL algorithm. For example, on the most commonly-studied SSL benchmark CIFAR-10, the accuracy of MutexMatch using only 20 labels can reach up to $9 1 . 7 7 { \pm } 2 . 6 0 \%$ . + +# 2 RELATED WORK + +Consistency regularization is a significant branch of recent state-of-the-art (SOTA) SSL methods, which is proposed in Bachman et al. (2014). Such methods encourage the classifier to output same class probability distribution after different versions of augmentation for the same unlabeled data. Generally, the consistency regularization based models are trained with unlabeled data using the loss function: $| | p ( y | \alpha ( x ) \dot { - } p \bar { ( y | \alpha ( x ) ) ) } | | _ { 2 } ^ { 2 }$ , where $x$ is the input image and $\alpha ( \cdot )$ is an kind of transformation that does not change the image label. Particularly, $\alpha ( \cdot )$ can adopt different augmentation methods, e.g., Mixup (Zhang et al., 2017) in Berthelot et al. (2019), RandAugment (Sohn et al., 2020) in Sohn et al. (2020) and CTAugment in Berthelot et al. (2020). Laine & Aila (2016) enforces a loss of consistency on the predictions of two augmented variants of unlabeled data. In Tarvainen & Valpola (2017), a teacher model is maintained to generate more stable targets for unlabeled data, and the mean squared error is used to encourage same predictions of the student and teacher models. Xie et al. (2020) adopts automatic augmentation for data perturbation and enforces a loss of consistency by the KL divergence. Recently, some holistic methods (Sohn et al., 2020; Berthelot et al., 2020) have been proposed to combine consistency regularization with pseudo-labeling for better SSL performance. Differently, in MutexMatch, in addition to enforcing prediction consistency on TPC, we propose a novel mutex-based consistency to effectively leverage all unlabeled samples. + +Pseudo-labeling is widely leveraged for entropy minimization by constructing one-hot labels from predictions of unlabeled data with high-confidence and makes use of them based on cross-entropy loss (Lee et al., 2013; Shi et al., 2018; Sohn et al., 2020; Xie et al., 2020). However, these methods have a significant limitation, i.e., using confidence thresholds to select pseudo-labels results in that all unlabeled data is not sufficiently exploited. Recent pseudo-labeling based method (Rizve et al., 2021) proposes an uncertainty-aware pseudo-label selection framework to use both high and lowconfidence samples. However, it introduces two thresholds to control the pseudo-label generation, thus some unlabeled samples are still not utilized. + +![](images/ea87b65a4a75f58c4f5b1c9dcf561cf90691b4f26b6f6e39703b4922c331498a.jpg) +Figure 2: Diagram of the proposed MutexMatch. Given a batch of unlabeled samples, TPC $\mathcal { P }$ uses their weakly-augmented variants to generate pseudo-labels. Then we adopt the classes with the lowest confidence as the complementary labels to train TNC $\mathcal { N }$ separately. Meanwhile, TPC and TNC are used for mutex-based consistency regularization in the high and low-confidence portion of TPC’s predictions respectively. $f$ denotes output features and $p , r$ denote predictions of TPC and TNC. Superscripts $w$ and $s$ represent corresponding outputs for the weakly-augmented variant and strongly-augmented variant, respectively. + +Complementary label is used to help the model learn which class the input image does not belong to (Ishida et al., 2017; 2018; Yu et al., 2018). Considering a $c$ -class classification task, we denote $x \in \mathcal { X }$ as an input image and $y \in \mathcal { Y } = \{ 1 , . . . , c \}$ as its label. Complementary label $\overline { y }$ is generated by select from $\mathcal { V } \backslash \{ \boldsymbol { y } \}$ at random. Inspired by Rizve et al. (2021) and Kim et al. $( 2 0 1 9 ) ^ { 1 }$ , in MutexMatch, we design a novel way (detailed in Section 3.2) to propose complementary labels, so as to ensure their effectiveness in semi supervised learning. Experiments about using standard complementary label selections are discussed in Section 5.2. + +# 3 MUTEXMATCH + +# 3.1 OVERVIEW + +Different from existing SSL approaches, in addition to a feature extractor $\theta ( \cdot )$ , MutexMatch jointly trains two distinct classifiers, a True-Positive Classifier (TPC) $\mathcal { P } ( \cdot )$ and a True-Negative Classifier (TNC) $\mathcal { N } ( \cdot )$ . To be specific, TPC is used to predict which class the instance belongs to (i.e., true positive), while TNC is employed to indicate which class the instance is not (i.e., true negative). To mitigate pseudo-labeling errors, a pre-defined high-confidence threshold $\tau$ is utilized to split the unlabeled data into high-confidence and low-confidence portions. Besides training TPC on the highconfidence portion, we explore complementary labels on low-confidence samples to train TNC. In this way, all the unlabeled data could be effectively exploited. + +In a mini-batch, we have $B$ labeled data $\mathcal { X } ~ = ~ \{ ( x _ { b } ^ { l b } , y _ { b } ^ { l b } ) \} _ { b = 1 } ^ { B }$ and $\mu B$ unlabeled data $u \ =$ $\{ ( x _ { b } ^ { u l b } , y _ { b } ^ { u l b } ) \} _ { b = 1 } ^ { \mu B }$ , where $\mu$ represents the relative size of $\mathcal { X }$ and $\mathcal { U }$ . Following (Sohn et al., 2020), we perform weak and strong augmentations for data perturbations, denoted by $\alpha _ { w } ( \cdot )$ and $\alpha _ { s } ( \cdot )$ , respectively. Given weakly-augmented instance $x ^ { w }$ and strongly-augmented instance $x ^ { s }$ , MutexMatch simultaneously optimizes four losses: the supervised loss $\mathcal { L } _ { s u p }$ , the separated negative loss $\mathcal { L } _ { s e p }$ , the positive consistency loss ${ \mathcal { L } } _ { p }$ and the negative consistency loss ${ \mathcal { L } } _ { n }$ . In summary, the total loss is + +$$ +\mathcal { L } = \mathcal { L } _ { s u p } + \lambda _ { s e p } \mathcal { L } _ { s e p } + \lambda _ { p } \mathcal { L } _ { p } + \lambda _ { n } \mathcal { L } _ { n } , +$$ + +where $\lambda _ { s e p }$ , $\lambda _ { p }$ and $\lambda _ { n }$ are scalar hyper-parameters to adjust the relative importance of corresponding losses. The supervised loss $\mathcal { L } _ { s u p }$ is simply defined as the cross-entropy between $y ^ { l b }$ and the + +predictions of TPC on labeled data $x ^ { l b }$ , calculated as follows: + +$$ +\mathcal { L } _ { s u p } = \frac { 1 } { B } \sum _ { n = 1 } ^ { B } H ( y _ { n } ^ { l b } , \mathcal { P } ( \theta ( \alpha _ { w } ( x _ { n } ^ { l b } ) ) ) ) , +$$ + +where $H ( p , q )$ denotes the standard cross-entropy loss between distribution $q$ and $p$ + +# 3.2 TRUE-NEGATIVE CLASSIFIER + +In multi-class classification tasks, for a specific instance, it is easier to predict which class it does not belong to than to know which class it exactly is. For example, given an image of airplane in CIFAR-10, we can predict which class it does not belong to with a probability $90 \%$ at random, whereas we only have the probability of $10 \%$ to correctly predict it is an airplane. To this end, we design a True-Negative Classifier to predict which class it is not. Compared to + +![](images/6de18e33fa1a4e3e4c06c6e0e1140d78d074d9ef3c09a6db8bcbe63da0b747b8.jpg) +Figure 3: Training of TNC + +TPC, it is much easier to obtain correct labels for TNC. Thus we exploit TNC to provide more guidance information on unlabeled data. We then propose a mutex-based prediction consistency on TPC and TNC to make full use of unlabeled data, which is described in Section 3.3. The high-level training process of TNC is shown in Figure 3. Unlike the standard complementary label generations (Ishida et al., 2017; Yu et al., 2018), we use the class with the lowest confidence in TPC’s predictions as the complementary label to train TNC. The training loss $\mathcal { L } _ { s e p }$ can be calculated as + +$$ +\mathcal { L } _ { s e p } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } H ( \arg \operatorname* { m i n } ( \mathcal { P } ( \theta ( x _ { n } ^ { w } ) ) , \mathcal { N } ( \hat { \theta } ( x _ { n } ^ { w } ) ) ) , +$$ + +where $\hat { \theta }$ represents that $\theta$ is considered constant for the generation of this loss, i.e., stop backpropagating gradients. Since our downstream task is to accurately classify images, we adopt such gradient-blocking operation to ensure that the feature extractor will not be affected by the training of TNC. We extensively investigate the effectiveness of TNC in Section 4.3. + +# 3.3 MUTEX-BASED CONSISTENCY REGULARIZATION + +In recent consistency-regularization based SSL methods, only samples with high-confidence predictions are leveraged to train models. However, it could lead to inefficient utilization of unlabeled data, especially at the early stage of the training process. Differently, MutexMatch can also effectively exploit low-confidence unlabeled samples via introducing a novel mutex-based consistency regularization. A high-confidence threshold $\tau$ on TPC’s predictions is defined to split the unlabeled samples into two portions with mutex confidence intervals, i.e., the high-confidence one $( > \tau )$ and the low-confidence one $( \leq \tau )$ . In the high-confidence portion, we use TPC to learn what the unlabeled data is, while in the low-confidence portion, we employ TNC to learn what it is not, because it is difficult for us to obtain its real class information. + +On the one hand, we use weakly-augmented example $x ^ { w }$ to generate pseduo-labels from TPC and enforce positive consistency against its corresponding strongly-augmented variant $x ^ { s }$ . We can then obtain their predictions, $p ^ { \bar { w } } \ \stackrel { \cdot } { = } \ \mathcal { P } ( \theta ( x ^ { w } ) )$ and $p ^ { s } \doteq \mathcal { P } ( \bar { \theta ( } \bar { x ^ { s } } ) )$ . Let $\hat { p } ^ { w } = \arg \operatorname* { m a x } ( p ^ { w } )$ , such consistency can be achieved by minimizing the loss ${ \mathcal { L } } _ { p }$ : + +$$ +\mathcal { L } _ { p } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } \mathbb { 1 } ( \operatorname* { m a x } ( p _ { n } ^ { w } ) \geq \tau ) H ( \hat { p } _ { n } ^ { w } , p _ { n } ^ { s } ) , +$$ + +where $\mathbb { 1 } ( \operatorname* { m a x } ( p ^ { w } ) > \tau )$ retains the predictions whose maximum probabilities are larger than $\tau$ . On the other hand, for these low-confidence samples, we enforce consistency regularization against TNC’s predictions by minimizing the ${ \mathcal { L } } _ { n }$ : + +$$ +\mathcal { L } _ { n } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } \mathbb { 1 } ( \operatorname* { m a x } ( p _ { n } ^ { w } ) < \tau ) H ( r _ { n } ^ { w } , r _ { n } ^ { s } ) , +$$ + +Table 1: Accuracy for CIFAR-10, CIFAR100 and SVHN averaged on 5 different folds. Results with ∗ were reported in CoMatch (Li et al., 2020), while results with † are using our own reimplementation. Other results were reported in FixMatch (Sohn et al., 2020). Results with DA are achieved by combining the distribution alignment technique (Berthelot et al., 2020). + +
MethodCIFAR-10CIFAR-100SVHN
10 labels20 labels40 labels80 labels200 labels400 labels2500 labels40 labels250 labels
UDA=70.95±5.93==40.72±0.8866.87±0.2247.37±20.5194.31±2.76
MixMatch27.84±10.63*52.46±11.5080.79±1.28*33.39±1.3260.06±0.3757.45±14.5396.02±0.23
ReMixMatch w. DA80.90±9.64=55.72±2.0672.57±0.3196.66±0.2097.08±0.48
FixMatch64.08±20.33†82.32±9.77*88.61±3.3592.06±0.88*38.87±2.50†51.15±1.7571.71±0.1196.04±2.1797.52±0.38
FixMatch w. DA*83.81±9.3586.98±3.4092.29±0.86=
CoMatch*69.87±11.82†87.67±8.4793.09±1.3993.97±0.6296.47 ± 1.29†97.75± 0.19†
MutexMatch78.73±11.2191.77±2.6093.49±0.2294.34±0.8140.38±2.3656.14±1.4671.80±0.2397.19±0.2697.73±0.18
+ +where $r ^ { w } \ = \ \mathcal { N } ( \theta ( x ^ { w } ) )$ and $r ^ { s } = \mathcal { N } ( \theta ( x ^ { s } ) )$ are the label predictions of TNC for $x ^ { w }$ and $x ^ { s }$ , respectively. For the purpose of entropy minimization (Lee et al., 2013), we adopt hard pseudolabel $\hat { p } ^ { w }$ to enforce the consistency regularization on TPC. Differently, we use soft pseudo-label $r ^ { w }$ for consistency regularization of TNC, so as MutexMatch can know more information of impossible class for classification. We further discuss this soft-label setting in Section 5.2. The whole algorithm is presented in Section 1 of Appendix. + +# 4 EXPERIMENTS + +Following Tarvainen & Valpola (2017); Sohn et al. (2020), we perform evaluation on four benchmark datasets, including STL-10, CIFAR-10/100 and SVHN. We also conduct ablation studies in Section 5 to investigate the efficacy of MutexMatch. Other experiments, e.g., the impact of learning rate and hyper-parameters, are shown in Section D in the Appendix. + +# 4.1 CIFAR-10, CIFAR-100 AND SVHN + +We evaluate our method and baselines on three widely used SSL datasets: (1) CIFAR-10, consisting of 50,000 images from 10 classes, (2) CIFAR-100, consisting of 50,000 images from 100 classes, and (3) SVHN, consisting of more than 70,000 street view house number images from 10 classes. + +Baselines. We introduce recent state-of-the-art SSL methods, i.e., CoMatch (Li et al., 2020), FixMatch (Sohn et al., 2020) and FixMatch with distribution alignment (Berthelot et al., 2020) to compare with MutexMatch. Moreover, we compare our method with SSL methods such as UDA (Xie et al., 2020), MixMatch (Berthelot et al., 2019) and ReMixMatch (Berthelot et al., 2020). + +Settings. For all experiments, in MutexMatch, we adopt Wide ReseNet (Zagoruyko & Komodakis, 2016) as the backbone (WRN-28-2 for CIFAR-10, SVHN and WRN-28-8 for CIFAR100) following Sohn et al. (2020). In our implementation, TNC is the same two-layer MLP as TPC. For fair comparison, We follow these baseline methods (Sohn et al., 2020; Li et al., 2020) using SGD with a momentum of 0.9 and a weight decay of 0.0005 during training. Also, we train the model for 1024 epochs, using a learning rate of 0.03 without the decay schedule for CIFAR-10, and with cosine decay schedule for CIFAR-100 and SVHN. For hyper-parameters in MutexMatch, we set $\tau = 0 . 9 5 , \mu = 7 , B = 6 4$ for all experiments. Particularly, we set $\tau = 0 . 5$ on CIFAR-10 with 80 labels and train the model with cosine decay schedule for learning rate. In our method, RandAugment (Cubuk et al., 2020) is used for strong augmentation. Also, $\lambda _ { s e p }$ , $\lambda _ { p }$ and $\lambda _ { n }$ are set to 1 for simplicity. To reduce the influence from random data partition, we report the mean and variance of accuracy on five different folds of labeled/unlabeled data. + +Results. Table 1 shows the comparison between MutexMatch and baselines. With only 4 labeled data per class, MutexMatch achieves an accuracy of $9 3 . 4 9 { \pm } 0 . 2 2 \%$ on CIFAR-10, $5 6 . 1 4 { \pm } 1 . 4 6 \%$ on CIFAR-100 and $9 7 . 1 9 { \pm } 0 . 2 6 \%$ on SVHN, yielding improvement over prior SSL results. Especially, we demonstrate the superiority of MutexMatch under the extremely label-scarce setting. e.g., achieving an average accuracy of $9 1 . 7 7 \%$ on CIFAR-10 with only 20 labels, $4 0 . 3 8 \%$ on CIFAR-100 with 200 labels. In addition, details on barely supervised learning can be found in + +Section B of Appendix. The fewer labels will lead to the accumulation of more noise pseudo-labels in training, whereas MutexMatch uses all unlabeled data while introducing little error information. + +Moreover, we report additional results on CIFAR-10 with different backbone CNN-13 and more available labels. We compare MutexMatch with MT (Tarvainen & Valpola, 2017), ICT (Verma et al., 2019), DualStudent (Ke et al., 2019) and UPS Rizve et al. (2021). We conduct experiments using the same setting as CIFAR-10 with 80 labels. In Table 2, we find that MutexMatch is not backbone dependent, and achieves performance improvement when more labels are given, outperforming all baseline methods. More discussion of experimental results can be found in Section 4.3. + +Table 2: Accuracy on CIFAR-10 with larger amounts of labels and CNN-13 backbone. Results of baseline methods are reported in UPS (Rizve et al., 2021). + +
CIFAR-10Method1000 labels 4000 labels
1000 labels4000 labels
MT80.96±0:5188.59±0.25
ICT84.52±0.7892.71±0.02
DualStudent85.83±0.3891.11±0.09
UPS91.82±0.1593.61±0.02
MutexMatch 93.01±0.32 94.10±0.24
+ +Table 3: Accuracy on STL-10 averaged on 5 pre-defined folds with ResNet-18 backbone. Results of baseline methods are reported in CoMatch (Li et al., 2020). + +
MethodSTL-10 ResNet-18
MixMatch38.02±8.29
FixMatch65.38±0.42
FixMatch w.DA66.53±0.39
CoMatch79.80±0.38
MutexMatch83.36±0.22
+ +# 4.2 STL-10 + +STL-10 contains 10 classes of 5,000 labeled and 100,000 unlabeled images extracted from a similar but broader distribution. The challenge of STL-10 lies in other unlabeled images contains out of distribution images and this distribution shift enables us to test the robustness of SSL algorithm. + +Settings. For STL-10, we evaluate MutexMatch on the 5 pre-defined folds. Each fold contains 1,000 labeled data and 100,000 unlabeled data. Therefore, we trained five models and averaged their performance as the final result. Following Li et al. (2020), we use ResNet-18 as the backbone because it consumes less computing resources than WRN-28-8 used in Sohn et al. (2020). We use the same hyperparameters and learning rate as CIFAR-100 in Section 4.1, and train the models using SGD with a momentum of 0.9 and a weight decay of 0.0005. + +Results. Table 3 shows the results, averaged on all 5 runs. In this setting, MutexMatch achieves accuracy improvement from $7 9 . 8 0 { \pm } 0 . 3 8 \%$ to $8 3 . 3 6 { \pm } 0 . 2 2 \%$ compared with CoMatch. The performance of MutexMatch on STL-10 is much better than that of the existing methods, showing TNC is less sensitive to data distribution shift between labeled and unlabeled data, so that MutexMatch can maintain robust performance like on other datasets. + +# 4.3 EFFECTIVENESS ANALYSIS OF TNC + +As shown in Figure 1, the TNC of MutexMatch uses exclusion method to help the model deal with unlabeled data with low confidence. Ideally, we think that for one class, the distribution of complementary pseudo-labels from TNC should be evenly dispersed or diverse unlike pseudo-label from TPC, so MutexMatch can exclude more error classes as much as possible. We conduct experiments on CIFAR-10 with 40 labels using the same setting as in Section 4.1. We observe that the class prediction from TNC is indeed generally consistent with our hypothesis. As shown in Figure 4, during training, for each class of CIFAR-10, the prediction of TNC is gradually dispersed to several classes (i.e., far away from the main diagonal of heat map), instead of gathering at a single class, indicating that TNC could play the role of exclusion method. On the contrary, the prediction outputted by TPC is gradually concentrated to the correct class (i.e., gathered to the main diagonal of the heat map). Note that, TNC uses soft labels for consistency regularization, which can explore more complementary information to help TPC classify correctly. + +Compared with other baseline methods shown in Table 1, MutexMatch performs better on CIFAR10 with extremely scarce labels. We believe that confirmation bias (Yu et al., 2018) leads to the poor performance of other methods. Fewer labels will introduce more noisy pseudo-labeled examples to participate in the learning process. Nevertheless, MutexMatch utilizes the unlabeled samples with low confidence, in an exclusive manner by TNC, introducing few noisy pseudo-labels. As shown in Figure 5, our experiments on CIFAR-10 show MutexMatch produces more accurate pseudo-labels than FixMatch (Sohn et al., 2020), especially when there are very few labeled samples. In this figure, $M$ indicates the results of MutexMatch and $F$ indicates the results of FixMatch. + +![](images/c0040f1531bae28167b29b35bea5073b61fa823340a41d44c23ce2c3ec2a7714.jpg) +Figure 4: The rate $( \% )$ of each class (column in heat map) in the pseudo-label and complementary pseudo-label outputted by TPC and TNC respectively corresponding to each class (row in heat map) in CIFAR-10. The darker, the higher. Results are reported in a run on CIFAR-10 with 40 labels. + +The accuracy of complementary pseudo-label is very crucial. An important premise for MutexMatch to work is that the complementary pseudolabel outputted by TNC is easy to predict, so it will introduce less error information into the model. Figure 5 shows that the complementary pseudolabels outputted by TNC achieves high accuracy. Compared with the pseudo-label outputted by TPC, complementary label is more insensitive to the change of the number of labels. Even with only one label per class, it can maintain a high accuracy. + +![](images/bbe2a02403308f6a14471776d3983c2f8c4c669db531d3edc246903e5eaf57fe.jpg) +Figure 5: Accuracy of pseudo-label and complementary label on CIFAR-10 with different amount of labeled data. + +# 5 ABLATION STUDY + +We conduct an extensive ablation study to verify the effectiveness of MutexMatch. The experiments are mainly conducted on CIFAR-10 and SVHN using four labels per class, where MutexMatch achieves $9 3 . 4 9 { \pm } 0 . 2 2 \%$ and $9 7 . 1 9 { \pm } 0 . 2 6 \%$ accuracy using default setting. In the following experiments, we keep the supervised loss as Equation (2) and positive consistency loss as Equation (4). + +# 5.1 UTILIZATION OF LOW-CONFIDENCE SAMPLES + +In order to fairly verify the effectiveness of TNC, we use the same settings as Section 4.1. We believe that the reason why the performance of MutexMatch is better than other earlier SSL algorithms is that the existence of TNC enables the model to learn from all unlabeled data. For example, in + +FixMatch, with a predefined confidence threshold, the unlabeled samples whose confidence is less than this threshold will not participate in the training. Therefore, we use the three most intuitive ways to use all the unlabeled data. We first use TPC to compute prediction $p ^ { w } = \mathcal { P } ( x ^ { w } )$ of weaklyaugmented unlabeled data $x ^ { w }$ and then: + +(i) We use $\hat { p } ^ { w } = \arg \operatorname* { m a x } ( p ^ { w } )$ as a hard pseudo-label, and enforce the cross-entropy loss against the model’s prediction $p ^ { s } = \mathcal { P } ( x ^ { s } )$ of strongly-augmented unlabeled data $x ^ { s }$ : + +$$ +\mathcal { L } _ { a b 1 } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } \mathbb { 1 } ( \operatorname* { m a x } ( p _ { n } ^ { w } ) < \tau ) H ( \hat { p } _ { n } ^ { w } , p _ { n } ^ { s } ) . +$$ + +(ii) We use $p ^ { w }$ as a soft pseudo-label and enforce the cross-entropy loss against $p ^ { s }$ : + +$$ +\mathcal { L } _ { a b 1 } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } \mathbb { 1 } ( \operatorname* { m a x } ( p _ { n } ^ { w } ) < \tau ) H ( p _ { n } ^ { w } , p _ { n } ^ { s } ) . +$$ + +(iii) We use the features $f ^ { w } = \theta ( x ^ { w } )$ of weakly-augmented image and the features $f ^ { s } = \theta ( x ^ { s } )$ of strongly-augmented image extracted by the feature extractor: + +$$ +\mathcal { L } _ { a b 1 } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } \mathbb { 1 } ( \operatorname* { m a x } ( p _ { n } ^ { w } ) < \tau ) E ( f _ { n } ^ { w } , f _ { n } ^ { s } ) , +$$ + +where $E ( p , q )$ denotes the mean squared loss between two distributions $p$ and $q$ + +The loss given above minimized by experiments is simply $\mathcal { L } _ { s u p } + \mathcal { L } _ { p } + \mathcal { L } _ { a b 1 }$ . All models are trained on SVHN using four labels per class and we show the results of all experiments in Figure 6. In this figure, the FULL indicates the setting of (i), the $S O F T$ indicates the setting of (ii) and the MSE indicates the setting of (iii). On this dataset, the default MutexMatch achieves an accuracy of $9 7 . 1 9 { \pm } 0 . 2 6 \%$ , outperforming all other experiments. Other ways using low-confidence samples will introduce more noisy pseudo-labels, resulting in the decline and instability of the accuracy of pseudo-label, which is not conducive to consistency regularization. + +![](images/81b9511dbb909e0cbda8589a39505652619bd86c557ca4be519f147d7cd842e8.jpg) +Figure 6: The learning curve of ablation study on SVHN. The $\mathbf { X }$ -axis represents the training epoch and y-axis represents the test accuracy in (a) and the pseudo-label accuracy in (b). + +# 5.2 EVALUATION ON LEARNING SCHEME OF TNC + +The learning of TNC in MutexMatch is very important. In default MutexMatch, we use hard complementary pseudo-label $\hat { q } ^ { w } = \arg \operatorname* { m i n } ( p ^ { w } )$ to train TNC separately when stopping gradient back propagation on the feature extractor, and enforce consistency regularization against soft pseudolabel $\bar { r ^ { w } } = \mathcal { N } ( x ^ { w } )$ in the low-confidence portion of $p ^ { w } \leq \tau$ . In order to validate the effectiveness of learning scheme of TNC in MutexMatch, we use three changed learning schemes for experiments: + +(i) We use hard pseudo-label $\hat { q } ^ { w } = \arg \operatorname* { m i n } ( p ^ { w } )$ to train TNC separately while stopping gradient back propagation on $\theta$ , and enforce consistency regularization against hard complementary pseudo-label: + +$$ +\mathcal { L } _ { s e p } = \frac { 1 } { B } \sum _ { n = 1 } ^ { B } H ( \hat { q } _ { n } ^ { w } , r _ { n } ^ { w } ) , \mathcal { L } _ { a b 2 } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } \mathbb { 1 } ( \operatorname* { m a x } ( p _ { n } ^ { w } ) < \tau ) H ( \hat { r } _ { n } ^ { w } , r _ { n } ^ { s } ) , +$$ + +where $\hat { r } ^ { w } = \arg \operatorname* { m a x } ( r ^ { w } )$ and $r ^ { s } = \mathcal { N } ( { x } ^ { s } )$ . + +(ii) We use hard complementary pseudo-label $\hat { \gamma } ^ { w }$ , which is generated via randomly selecting the class without the highest confidence from $p ^ { w }$ (just like the standard complementary label selection) to train TNC separately, while stopping gradient back propagation on $\theta$ , and enforce consistency regularization against soft complementary pseudo-label: + +$$ +\mathcal { L } _ { s e p } = \frac { 1 } { B } \sum _ { n = 1 } ^ { B } H ( \hat { \gamma } _ { n } ^ { w } , r _ { n } ^ { w } ) , L _ { a b 2 } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } \mathbb { 1 } ( \operatorname* { m a x } ( p _ { n } ^ { w } ) < \tau ) H ( r _ { n } ^ { w } , r _ { n } ^ { s } ) . +$$ + +(iii) We remove the separately training part of TNC. The complementary pseudo-label for TNC is obtained directly by $q ^ { w } \ = \ \mathtt { N o r m } ( 1 \ - \ p ^ { w } )$ where $\operatorname { N o r m } ( \cdot )$ is operation normalizing $q ^ { w }$ into interval $[ 0 , 1 ]$ . We enforce consistency regularization against soft complementary pseudo-label: + +$$ +\mathcal { L } _ { a b 2 } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } \mathbb { 1 } ( \operatorname* { m a x } ( p _ { n } ^ { w } ) < \tau ) H ( q _ { n } ^ { w } , r _ { n } ^ { s } ) . +$$ + +The loss given above minimized by experiments is simply $\mathcal { L } _ { s u p } + \mathcal { L } _ { p } + \mathcal { L } _ { s e p } + \mathcal { L } _ { a b 2 }$ in (i), (ii) and $\mathcal { L } _ { s u p } + \mathcal { L } _ { p } + \mathcal { L } _ { a b 2 }$ in (iii). + +All models are trained on CIFAR-10 using four labels per class, and we show the results of all experiments in Figure 7. In the figure, the Hard-Hard indicates setting of (i), the Rand-Soft indicates setting of (ii) and the Rev-Norm indicates setting of (iii). (i), (ii) and (iii) achieve accuracy of $9 0 . 5 6 \%$ , $9 1 . 5 3 \%$ and $9 1 . 0 3 \%$ respectively. The default MutexMatch achieved an accuracy of $9 3 . 4 9 \%$ which outperforms other settings. Furthermore, experiments show that the accuracy and stability of the complementary pseudo-labels which are outputted by TNC of default MutexMatch are dominant. TNC uses hard pseudo-label for separate training to ensure the accuracy and stability of complementary pseudo-label, and uses soft pseudo-label to participate in mutex-based consistency regularization to exclude more potential error classes. + +![](images/8680b672c399d43c5b55e4c775ab7be58a50520aa8e62e96b9d77626f9b522b5.jpg) +Figure 7: The learning curve of ablation study on CIFAR-10 with 40 labels. The $\mathbf { X }$ -axis represents the training epoch and the y-axis represents the test accuracy in (a), the pseudo-label accuracy in (b), and the complementary pseudo-label accuracy in (c). + +# 6 CONCLUSION + +In this paper, we propose MutexMatch, a novel SSL algorithm using a mutex-based consistency regularization derived by two distinct classifiers, one is to predict “what it is” and the other is to predict “what it is not”. MutexMatch can achieve superior performance on various SSL benchmarks, especially under label-scarce conditions. Last but not least, we validate that low-confidence samples could still be well utilized in training from a novel way. We believe this usage of low-confidence samples could be borrowed to other semi-supervised tasks, e.g., segmentation and detection. + +# REFERENCES + +Phil Bachman, Ouais Alsharif, and Doina Precup. Learning with pseudo-ensembles. In Advances in Neural Information Processing Systems 27, volume 27, pp. 3365–3373, 2014. + +David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, and Colin A. Raffel. Mixmatch: A holistic approach to semi-supervised learning. In Advances in Neural Information Processing Systems, volume 32, pp. 5049–5059, 2019. + +David Berthelot, Nicholas Carlini, Ekin D. Cubuk, Alex Kurakin, Kihyuk Sohn, Han Zhang, and Colin Raffel. Remixmatch: Semi-supervised learning with distribution matching and augmentation anchoring. In Eighth International Conference on Learning Representations, 2020. + +Olivier Chapelle, Bernhard Scholkopf, and Alexander Zien. Semi-supervised learning (chapelle, o. et al., eds.; 2006)[book reviews]. IEEE Transactions on Neural Networks, 20(3):542–542, 2009. + +Ekin Dogus Cubuk, Barret Zoph, Jon Shlens, and Quoc Le. Randaugment: Practical automated data augmentation with a reduced search space. In Advances in Neural Information Processing Systems, volume 33, pp. 18613–18624, 2020. + +Takashi Ishida, Gang Niu, Weihua Hu, and Masashi Sugiyama. Learning from complementary labels. In Advances in Neural Information Processing Systems, volume 30, pp. 5639–5649, 2017. + +Takashi Ishida, Gang Niu, Aditya Krishna Menon, and Masashi Sugiyama. Complementary-label learning for arbitrary losses and models. In International Conference on Machine Learning, pp. 2971–2980, 2018. + +Zhanghan Ke, Daoye Wang, Qiong Yan, Jimmy Ren, and Rynson Lau. Dual student: Breaking the limits of the teacher in semi-supervised learning. In IEEE International Conference on Computer Vision, pp. 6728–6736, 2019. + +Youngdong Kim, Junho Yim, Juseung Yun, and Junmo Kim. Nlnl: Negative learning for noisy labels. In 2019 IEEE International Conference on Computer Vision, pp. 101–110, 2019. + +Samuli Laine and Timo Aila. Temporal ensembling for semi-supervised learning. arXiv preprint arXiv:1610.02242, 2016. + +Dong-Hyun Lee et al. Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks. In Workshop on challenges in representation learning, ICML, volume 3, pp. 896, 2013. + +Junnan Li, Caiming Xiong, and Steven C. H. Hoi. Comatch: Semi-supervised learning with contrastive graph regularization. arXiv preprint arXiv:2011.11183, 2020. + +Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts. In Fifth International Conference on Learning Representations, 2017. + +Giorgio Patrini, Alessandro Rozza, Aditya Krishna Menon, Richard Nock, and Lizhen Qu. Making deep neural networks robust to label noise: A loss correction approach. In IEEE Conference on Computer Vision and Pattern Recognition, pp. 2233–2241, 2017. + +Mamshad Nayeem Rizve, Kevin Duarte, Yogesh S Rawat, and Mubarak Shah. In defense of pseudolabeling: An uncertainty-aware pseudo-label selection framework for semi-supervised learning. In The Ninth International Conference on Learning Representations, 2021. + +Weiwei Shi, Yihong Gong, Chris Ding, Zhiheng MaXiaoyu Tao, and Nanning Zheng. Transductive semi-supervised deep learning using min-max features. In Proceedings of the European Conference on Computer Vision, pp. 299–315, 2018. + +Kihyuk Sohn, David Berthelot, Chun-Liang Li, Zizhao Zhang, Nicholas Carlini, Ekin D. Cubuk, Alex Kurakin, Han Zhang, and Colin Raffel. Fixmatch: Simplifying semi-supervised learning with consistency and confidence. arXiv preprint arXiv:2001.07685, 2020. + +Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. In Advances in neural information processing systems, 2017. + +Vikas Verma, Alex Lamb, Juho Kannala, Yoshua Bengio, and David Lopez-Paz. Interpolation consistency training for semi-supervised learning. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, pp. 3635–3641, 2019. + +Qizhe Xie, Zihang Dai, Eduard Hovy, Thang Luong, and Quoc Le. Unsupervised data augmentation for consistency training. Advances in Neural Information Processing Systems, 33, 2020. + +Xiyu Yu, Tongliang Liu, Mingming Gong, and Dacheng Tao. Learning with biased complementary labels. In Proceedings of the European Conference on Computer Vision, pp. 69–85, 2018. + +Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In British Machine Vision Conference, 2016. + +Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. mixup: Beyond empirical risk minimization. arXiv preprint arXiv:1710.09412, 2017. + +Xiaojin Zhu. Semi-supervised learning. Encyclopedia of Machine Learning and Data Mining, pp. 1142–1147, 2017. + +# A ALGORITHM + +# Algorithm 1: MutexMatch algorithm + +Input: batch of labeledfeature extractor ata , T $\mathcal { X } = \{ ( x _ { b } ^ { l b } , y _ { b } ^ { l b } ) \} _ { b = 1 } ^ { B }$ , batch of unlabeled data $\mathcal { U } = \{ x _ { b } ^ { u l b } \} _ { b = 1 } ^ { \mu B }$ +$\theta$ $\mathcal { P }$ $\mathcal { N }$ +1 for iteration $t$ do +2 $\begin{array} { r l } { \mathcal { L } _ { s u p } = \frac { 1 } { B } \sum _ { n = 1 } ^ { B } H ( y _ { n } ^ { l b } , \mathcal { P } ( x _ { n } ^ { l b } ) ) } & { { } / / \operatorname { \it S u p e r v i s e d l o s s f o r } x ^ { l b } } \end{array}$ +3 for iteration $b = 1$ to $\mu B$ do +4 $p _ { b } ^ { w } = \mathcal { P } ( \theta ( \alpha _ { w } ( x _ { b } ^ { u l b } ) ) )$ // Compute TPC’s prediction for weakly-augmented $x ^ { u l b }$ +5 $p _ { b } ^ { s } = \mathcal { P } ( \theta ( \alpha _ { s } ( x _ { b } ^ { u l b } ) ) )$ // Compute TPC’s prediction for strongly-augmented $x ^ { u l b }$ +6 $r _ { b } ^ { w } = \mathcal { N } ( \theta ( \alpha _ { w } ( x _ { b } ^ { u l b } ) ) )$ // Compute TNC’s prediction for weakly-augmented $x ^ { u l b }$ +7 $r _ { b } ^ { s } = \mathcal { N } ( \theta ( \alpha _ { s } ( x _ { b } ^ { u l b } ) ) )$ // Compute TNC’s prediction for strongly-augmented $x ^ { u l b }$ +8 $\hat { p } _ { b } ^ { w } = \arg \operatorname* { m a x } ( p _ { b } ^ { w } )$ // Select pseudo-labels for $x ^ { u l b }$ +9 $\hat { q } _ { b } ^ { w } = \arg \operatorname* { m i n } ( p _ { b } ^ { w } )$ // Select complementary pseudo-labels for $x ^ { u l b }$ +10 end +11 $\begin{array} { r l } { \mathcal { L } _ { s e p } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } H ( \hat { q } _ { n } ^ { w } , \mathcal { N } ( \hat { \theta } ( x _ { n } ^ { w } ) ) ) } \end{array}$ // Stop back-propagating gradients on θ +12 = 1µB PµBn=1 1(max(pwn ) ≥ τ )H(ˆpwn , psn) // Positive consistency loss for xulb +13 $\begin{array} { r } { \mathcal { L } _ { n } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } \mathbb { 1 } ( \operatorname* { m a x } ( p _ { n } ^ { w } ) < \tau ) H ( r _ { n } ^ { w } , r _ { n } ^ { s } ) } \end{array}$ // Negative consistency loss for $x ^ { u l b }$ +14 update $\theta , \mathcal { P } , \mathcal { N }$ by SGD to optimise $\mathcal { L } _ { s u p } + \lambda _ { s e p } \mathcal { L } _ { s e p } + \lambda _ { p } \mathcal { L } _ { p } + \lambda _ { n } \mathcal { L } _ { n }$ + +# B BARELY SUPERVISED LEARNING + +The experimental protocol of barely supervised learning (BSL) described in Sohn et al. (2020) assume a limited availability (e.g., 1 or 5) of labeled data from categories of interest. In order to test the performance of our method in extreme cases, we conduct experiments on CIFAR-10 with only one label per class, and consider developing a simple method to use our TNC in the test phase. As shown in Table 4, we use five different random seeds to extract one label of each class from CIFAR-10, and use MutexMatch to achieve test accuracy reaching between $6 5 . 3 0 \%$ and $9 3 . 0 7 \%$ with a mean of $7 8 . 7 3 \%$ . Compared with FixMatch (Sohn et al., 2020) reaching between $4 8 . 5 8 \%$ and $8 5 . 3 2 \%$ , the performance of MutexMatch is more superior. Then we consider using TNC to complete the test phase under this setting to obtain the test accuracy. We assume that in the ideal case, according to Equation (3), for test data $x$ , the prediction of TNC $r _ { x } = \mathcal { N } ( x )$ and the prediction of TPC $p _ { x } = \mathcal { P } ( x )$ should satisfy arg $\operatorname* { m a x } ( r _ { x } ) = \arg \operatorname* { m i n } ( p _ { x } )$ . + +According to negative learning proposed in Kim et al. (2019), we hypothesis TNC is trained to classify what input image does not belong to its complementary label, so that we can use $\hat { r } _ { x } = \arg \operatorname* { m i n } ( r _ { x } )$ to classify an input image $x$ . Compared with TPC, TNC may learn less error information when the label is extremely scarce, so as to obtain better test performance. In order to verify this idea, we used TNC to participate in the test phase showed in Figure 8. For test sample $x$ , we set a confidence threshold $T$ , if $p _ { x } > T$ we uses TPC to predict, if $p _ { x } < T$ uses TNC instead, that is, the leftmost point $T = 0$ ) in the figure represents only TNC for test, and the rightmost point $T = 1$ ) represents only TNC for test. Taking 20 labels as the dividing line, we can see that using TNC for prediction has more advantages in the case of fewer labels. + +![](images/30814809dbc0e0079295d450d91755d7597c6c1a4adaa51bca39e87484fccf02.jpg) +Figure 8: Test accuracy on CIFAR-10 in single run with various amount of labels using TNC to participate test phase. The $\mathbf { X }$ -axis represents confidence threshold $T$ and y-axis represents test accuracy. + +Table 4: Accuracy of MutexMatch on a single 1-label split of CIFAR-10 with different random seeds. Results are ordered by accuracy. + +
Fold12345
Accuracy65.3071.1277.8386.3393.07
+ +# C SEMI-SUPERVISED LEARNING WITH NOISY LABELS + +To evaluate the robustness of MutexMatch, we conduct our experiments following settings of semi-supervised learning with noisy labels on CIFAR-10. Semi-supervised learning and noise labels are challenging problems, and semi-supervised learning with noise labels is much more because the ability of the model to resist noise labels will be greatly weakened when there is only a small amount of labeled data. + +Setting. Following Kim et al. (2019); Patrini et al. (2017), we applied three different types of noise in experiments: + +(1) Symmetric-inc noise is created by randomly selecting the label from all classes. +(2) Symmetric-exc noise is created by randomly selecting the label from all classes without ground truth label. +(3) Asymmetric noise is generated by mapping TRUCK AUTOMOBILE, BIRD $\mathrm { \Sigma } \mathrm { P L A }$ NE, DEER $ \mathrm { H O R S E }$ , and $\mathbf { C A T } \mathbf { D O G }$ for CIFAR-10. +We evaluate MutexMatch and baselines with noisy labels mentioned above using the same settings as in Section 4.1. All experiments use 40 labeled data for training, varying radio of noisy labels in labeled data $( 2 5 \% \& 5 0 \% )$ . + +Results. Table 5 shows the accuracy comparison between MutexMatch and baselines. All the results are reported by averaging on 5 different folds. Experiments show the robustness of MutexMatch under this setting. For example, with 2 labels and 2 noisy labels (Symmetric-inc) per class, MutexMatch achieves $8 8 . 7 2 { \scriptstyle \pm 3 . 5 1 \% }$ accuracy, while training of FixMatch collapse reaching a lower $7 7 . 8 0 { \pm } 1 7 . 5 7 \%$ accuracy. MutexMatch contains the idea of negative learning. Learning from the perspective of complementary pseudo-label can prevents model from overfitting to noisy data (Kim et al., 2019) so that MutexMatch achieves superior performance in SSL with noisy labels. + +Table 5: Accuracy on CIFAR-10 with noisy labels averaged on 5 different folds. All experiments were based on 40 labeled data with varying radio of noisy labels. + +
MethodSymmetric-incSymmetric-excAsymmetric
25%noisy50%noisy25%noisy50%noisy25%noisy50%noisy
FixMatch77.80±17.5781.54±18.4780.05±5.8075.11±14.6684.58±5.9072.91±19.30
MutexMatch88.72±8.5177.18±10.5589.37±6.1081.85±8.0089.51±5.1478.28±15.44
+ +# D ADDITIONAL EXPERIMENTAL RESULTS + +# D.1 ABLATION STUDY ON LEARNING RATE AND LEARNING RATE SCHEDULE + +We note that learning rate and learning rate schedule are very important for MutexMatch. In this section, we use the experimental setting in Section 4.1 to conduct additional ablation experiments for both. Following Loshchilov & Hutter (2017), recent work (Sohn et al., 2020; Li et al., 2020) use a cosine learning rate decay and achieve best performance. However, as shown in Table 6, we found that MutexMatch achieves better results without learning rate decay on CIFAR-10, outperforming cosine learning rate decay by $0 . 3 2 \%$ . When there are many labels, the pseudo-labels outputted by TPC are more likely to have high-confidence and remain stable. It is necessary for MutexMatch to use cosine learning rate decay to jump out of the local optimum. + +Table 6: Ablation study on learning rate and learning rate schedule. Results are reported on CIFAR10 varying number of labels. + +
Decay ScheduleLearning RateLabelsBackboneAccuracy
No Decay0.0340WRN-28-293.54
No Decay0.0740WRN-28-293.02
No Decay0.1040WRN-28-292.89
Cosine Decay0.0340WRN-28-293.22
Cosine Decay0.0740WRN-28-293.20
Cosine Decay0.1040WRN-28-292.59
No Decay0.0380WRN-28-293.95
Cosine Decay0.0380WRN-28-294.53
No Decay0.031000CNN-1391.57
Cosine Decay0.031000CNN-1393.46
No Decay0.034000CNN-1392.75
Cosine Decay0.034000CNN-1394.41
+ +# D.2 HYPERPARAMETERS + +For MutexMatch, the choice of $\tau$ needs to be very cautious, because different $\tau$ will lead to the division of high and low-confidence portions, which will affect the impact of the mutex-based consistency regularization on the model. We use the identical setting of experiments in Section 4.1 for MutexMatch and vary $\tau$ to verify the sensitivity of MutexMatch to this hyperparameter. As shown in Table 7, MutexMatch needs to select appropriate $\tau$ to divide confidence portions. We note that when there are many labels, $\tau$ has a greater impact on performance. The more labels are available, the less confirmation bias will be when using TPC directly for classification, so the portion of TPC in mutex-based consistency regularization can be used directly for learning. Therefore, we guess that in general, we should choose a smaller $\tau$ to make more pseudo-labels participate in the training of TPC when the number of labels increases. + +Table 7: Ablation study on confidence threshold $\tau$ . Results are reported on CIFAR-10 varying number of labels. + +
TLabelsBackbone Accuracy
0.540WRN-28-293.52
0.7540WRN-28-293.44
0.8540WRN-28-293.28
0.9540WRN-28-293.54
0.9940WRN-28-292.17
0.580WRN-28-294.53
0.9580WRN-28-293.64
0.51000CNN-1393.46
0.951000CNN-1392.07
0.54000CNN-1394.41
0.954000CNN-1392.94
+ +At the same time, showed in Figure 9, we vary the weight $\lambda _ { s e p }$ of the separate training loss for TNC $\mathcal { L } _ { s e p }$ and $\lambda _ { n }$ of the negative consistency loss ${ \mathcal { L } } _ { n }$ . Choosing the appropriate weight of loss is very important for MutexMatch. Larger $\lambda _ { s e p }$ ensures the accuracy of complementary pseudo-labels, which helps TNC better participate in training. Appropriate $\lambda _ { n }$ weighs the contribution of TNC and TPC in mutex-based consistency regularization, so that the model can achieve better performance. + +Additionaly, we provide more ablation studies on various $\lambda _ { s e p }$ , $\lambda _ { n }$ and $\lambda _ { p }$ shown in Table 8. We find that increasing $\lambda _ { s e p }$ and $\lambda _ { n }$ at the same time will cause severe performance degradation, which shows that we must carefully control the importance of TNC in the learning process, because TPC has always maintained the most important position in completing our classification tasks. Meanwhile, appropriate $\lambda _ { p }$ ensures model can benefit from learning of TNC. More results of experiments about situation where $\lambda _ { p } = 0$ or $\lambda _ { n } = 0$ can be found in Section D.3. + +Table 8: Accuracy on CIFAR-10 with 40 labels and various $\lambda _ { s e p } , \lambda _ { n } , \lambda _ { p }$ . + +
Xsep入nXpAccuracy
11193.49
111091.27
112085.05
101188.94
201185.44
2010118.03
20101015.96
+ +![](images/0a082e49f276a6d9fd0da8cd59d788738b4a8a1da723759d2dd943276b71db30.jpg) +Figure 9: Accuracy on CIFAR-10 with $4 0 1 \mathrm { a }$ - bels and various $\lambda _ { s e p }$ , $\lambda _ { n }$ . + +# D.3 ABLATION STUDY ON TPC AND TNC + +We explain why we enforce consistency regularization on TNC as follows. Given two augmented variants derived from the same unlabeled instance, we claim that the class probability distributions (i.e., soft-labels) of their complementary predictions (i.e., the TNC’s outputs) should be consistent. Under the help of an independent training process of TNC, the model can be more confident on “what it is not”. As a result, such prediction consistency on TNC can effectively decrease the FalseNegative probability on TPC’s predictions. As shown in Figure 10, given a instance of class 1, the independent training of TNC can generate an accurate complementary prediction with extremely low probability of class 1. Then encouraging a similar prediction on its strongly augmented variant can help the model to learn more discriminative features. It can in turn affect the TPC’s prediction, such that the False-Negative probabilities (to be predicted as a class of 2, 3, 4, 5) can be effectively decreased. Consequently, the True-Positive probability of TPC’s predictions is enlarged. + +![](images/8ee3047fcae56fe3247e0e4b3879c64ca88844cbf1b6ab96f28751aeee398af0.jpg) +Figure 10: The correct component in prediction vector is class 1. + +We construct an experiment on CIFAR-10 with 40 labels to verify our findings. We denote MutexMatch without consistency reguarlization on TNC as M wo. c (i.e., MutexMatch degenerates to FixMatch). Although the confidence of the correct predictions of MutexMatch and M wo. $c$ is very high (are very close to $1 0 0 \%$ ), we check their wrong predictions as an example for comparison. In fact, for the correct part of the pseudo-labels, MutexMatch obtains more correct pseudo-labels than M wo. $c$ (FixMatch) thanks to the use of consistency regularization on TNC (i.e., MutexMatch’s accuracy of pseduo-labels is higher than FixMatch, which is shown in Figure 5). + +As shown in Figure 11(a), given the unlabeled instances belonging to “automobile”, the MutexMatch’s average probability of “automobile” component in prediction vector is higher than that of M wo. c. We can also obtain similar findings on other different classes, as shown in Figure 11(b). Such observations demonstrate enforcing prediction consistency on TNC can successfully help the model lower the False-Negative probability, which in turn improve the True-Positive probability in the TPC’s prediction vector. + +![](images/7493c3663455406ffd0325a930915cde0a0e05e37845af4add418830bf7a3b5a.jpg) +Figure 11: (a) Average probability of each component in prediction vector of automobile images. (b) Average probability of correct component in prediction vector of all classes in CIFAR-10. + +Moreover, this design is based on the consideration of it’s unreasonable to directly involve complementary label based negative learning in semi-supervised. In the early training stage of semisupervised learning, the accuracy of pseudo-labels is often not very high. In this process, the introduction of complementary labels directly into the learning process will not only be helpful, but even harmful, and lead to training collapse finally. Therefore, we use consistency regularization to “decouple” the part where complementary label are directly involved in training. We believe that it is reasonable to use consistency regularization on TNC at the sample level, because we only need TNC to show the various results for each class at the dataset level, which is shown in Section 4.3. This demonstrates TNC has learned discriminative information from the aspect of complementary label. Implementing consistency regularization on TNC at the sample level is to help TNC learn from the perspective of complementary labels, so that feature extraction can learn better data representation of unlabeled data with low confidence. In addition, we use soft labels for consistency regularization on TNC, which also ensures TNC can learn multi-class information, which is described in Section 3.3. For further discussion on the effectiveness of TPC and TNC, we consider removing these two components respectively. Given mentioned above, we designed the following experiments: + +(i) What happens if consistency regularization on TNC is abandoned (i.e., $\lambda _ { n } = 0$ )? Taking into account that in the default MutexMatch, training of TNC with complementary pseudolabels stops the gradient, so if we set $\lambda _ { n } = 0$ , then TNC is equivalent to not participating in the training of the model at all, which means that MutexMatch degenerates into FixMatch. So a more reasonable setting is to restore the backpropagation on feature extractor $\theta$ in Equation (3). Then we explore the effect of not using consistency regularization on TNC: + +$$ +\mathcal { L } _ { s e p } = \frac { 1 } { \mu B } \sum _ { n = 1 } ^ { \mu B } H ( \arg \operatorname* { m i n } ( \mathcal { P } ( \theta ( x _ { n } ^ { w } ) ) , \mathcal { N } ( \theta ( x _ { n } ^ { w } ) ) ) , +$$ + +where we restore the back propagation on $\theta$ and set $\lambda _ { n } = 0$ in Equation (1). + +(ii) What happens when we train TNC with complementary labels generated by TPC without stopping the gradient on TNC? We keep $\lambda _ { n } = 1$ in Equation (1), and restore the backpropagation on $\theta$ like Equation (12). +(iii) At the same time, in order to explore the role of TPC component, we set $\lambda _ { p } = 0$ in Equation (1) for ablation study. +(iv) We simply set $( \lambda _ { p } , \lambda _ { n } , \lambda _ { s e p } )$ to $( 0 , 0 , 0 ) , ( 0 , 1 , 0 )$ and $( 1 , 1 , 0 )$ . + +As shown in Figure 12, the default MutexMatch achieves dominant performance compared with other settings. In figures, wo. TNC represents setting of (i), wo. $_ { S G }$ represents setting of (ii), wo. $T P C$ represents setting of (iii) and $w . \ 0 0 0 , w . \ 0 l 0 , w . \ l l 0$ represent setting of (iv). Obviously, TNC participates in model training directly will cause training to collapse, which means that the model is seriously affected by the learning of TNC, and there is no way to learn effective information of “what it is”. (iii) shows that TPC in this case does not get adequate training and it can’t complete the classification task. i.e., the training of TPC is equivalent to using only labeled data. Meanwhile, The training of TNC is closely related to TPC. In this case, TNC has not been well trained, too. And in fact, we don’t use TNC to participate in the testing phase. (i) and (ii) illustrate the superiority of using consistency regularization for learning on TNC. This “decoupling” ensures that TNC can make the model learn a better data representation without affecting the learning of TPC. Finally, the combination of (iv) and other settings proves the necessity of each component in MutexMatch. + +![](images/df8412667028cb8aecb1f42e55870b95531041612f83866e9add7013c6a04086.jpg) +Figure 12: The learning curve of ablation study on SVHN. The $\mathbf { X }$ -axis represents the training epoch and $\mathbf { y }$ -axis represents the test accuracy in (a), (c) and the pseudo-label accuracy in (b), (d). \ No newline at end of file diff --git a/parse/dev/r5hq-Ooh_Ba/r5hq-Ooh_Ba_middle.json b/parse/dev/r5hq-Ooh_Ba/r5hq-Ooh_Ba_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..527e26536afa388d1ee424a0d1748857a56b45f6 --- /dev/null +++ b/parse/dev/r5hq-Ooh_Ba/r5hq-Ooh_Ba_middle.json @@ -0,0 +1,46363 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 504, + 136 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 226, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 226, + 96 + ], + "score": 1.0, + "content": "MUTEXMATCH:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 98, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 505, + 117 + ], + "score": 1.0, + "content": "SEMI-SUPERVISED LEARNING WITH MUTEX-BASED", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 118, + 354, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 354, + 137 + ], + "score": 1.0, + "content": "CONSISTENCY REGULARIZATION", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 112, + 155, + 244, + 177 + ], + "lines": [ + { + "bbox": [ + 113, + 155, + 201, + 167 + ], + "spans": [ + { + "bbox": [ + 113, + 155, + 201, + 167 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 111, + 166, + 245, + 178 + ], + "spans": [ + { + "bbox": [ + 111, + 166, + 245, + 178 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 278, + 206, + 333, + 218 + ], + "lines": [ + { + "bbox": [ + 277, + 206, + 335, + 219 + ], + "spans": [ + { + "bbox": [ + 277, + 206, + 335, + 219 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 143, + 231, + 468, + 418 + ], + "lines": [ + { + "bbox": [ + 142, + 232, + 469, + 244 + ], + "spans": [ + { + "bbox": [ + 142, + 232, + 469, + 244 + ], + "score": 1.0, + "content": "The core issue in semi-supervised learning (SSL) lies in how to effectively lever-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 243, + 470, + 255 + ], + "spans": [ + { + "bbox": [ + 141, + 243, + 470, + 255 + ], + "score": 1.0, + "content": "age unlabeled data, whereas most existing methods usually concentrate on the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 254, + 470, + 267 + ], + "spans": [ + { + "bbox": [ + 141, + 254, + 470, + 267 + ], + "score": 1.0, + "content": "utilization of high-confidence samples yet seldom fully explore the usage of low-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 264, + 470, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 264, + 470, + 277 + ], + "score": 1.0, + "content": "confidence samples. Early SSL methods mostly require low-confidence samples", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 275, + 470, + 289 + ], + "spans": [ + { + "bbox": [ + 141, + 275, + 470, + 289 + ], + "score": 1.0, + "content": "to optimize the same loss function as high-confidence samples, but this setting", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 142, + 287, + 469, + 299 + ], + "spans": [ + { + "bbox": [ + 142, + 287, + 469, + 299 + ], + "score": 1.0, + "content": "might largely challenge the low-confidence samples especially at the early train-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 297, + 470, + 310 + ], + "spans": [ + { + "bbox": [ + 141, + 297, + 470, + 310 + ], + "score": 1.0, + "content": "ing stage. In this paper, we aim to utilize low-confidence samples in a novel", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 308, + 469, + 322 + ], + "spans": [ + { + "bbox": [ + 141, + 308, + 469, + 322 + ], + "score": 1.0, + "content": "way, which is realized by our proposed mutex-based consistency regularization,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 319, + 470, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 319, + 470, + 332 + ], + "score": 1.0, + "content": "namely MutexMatch. To be specific, the high-confidence samples are required", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "score": 1.0, + "content": "to exactly predict “What it is” by conventional True-Positive Classifier, while the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 341, + 470, + 354 + ], + "spans": [ + { + "bbox": [ + 141, + 341, + 470, + 354 + ], + "score": 1.0, + "content": "low-confidence samples, for a much simpler goal, are employed to predict “What", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 352, + 470, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 352, + 470, + 365 + ], + "score": 1.0, + "content": "it is not” by True-Negative Classifier with ease. In this way, we not only mitigate", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 363, + 470, + 375 + ], + "spans": [ + { + "bbox": [ + 141, + 363, + 470, + 375 + ], + "score": 1.0, + "content": "the pseudo-labeling errors but also make full use of the low-confidence unlabeled", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 374, + 470, + 387 + ], + "spans": [ + { + "bbox": [ + 141, + 374, + 470, + 387 + ], + "score": 1.0, + "content": "data in the training stage. The proposed MutexMatch achieves superior perfor-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 385, + 470, + 397 + ], + "spans": [ + { + "bbox": [ + 141, + 385, + 470, + 397 + ], + "score": 1.0, + "content": "mance on multiple benchmark datasets, i.e., CIFAR-10, CIFAR-100, SVHN, and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 396, + 469, + 408 + ], + "spans": [ + { + "bbox": [ + 141, + 396, + 469, + 408 + ], + "score": 1.0, + "content": "STL-10. Particularly, our method shows further superiority under few quantities", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 407, + 459, + 419 + ], + "spans": [ + { + "bbox": [ + 141, + 407, + 225, + 419 + ], + "score": 1.0, + "content": "of labeled data, e.g.,", + "type": "text" + }, + { + "bbox": [ + 225, + 407, + 257, + 418 + ], + "score": 0.84, + "content": "9 1 . 7 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 407, + 459, + 419 + ], + "score": 1.0, + "content": "accuracy with only 20 labeled data on CIFAR-10.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 441, + 205, + 453 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 208, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 208, + 456 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 466, + 505, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 465, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 480 + ], + "score": 1.0, + "content": "Aiming to escape from time-consuming and laborious labeling tasks, semi-supervised learning", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 476, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 506, + 491 + ], + "score": 1.0, + "content": "(SSL) (Chapelle et al., 2009; Zhu, 2017) has been a longstanding yet important direction to leverage", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 104, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "a large quantity of unlabeled data along with few labeled data during training. Recent SSL models", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 500, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 505, + 512 + ], + "score": 1.0, + "content": "could be categorized into consistency regularization based or entropy minimization based methods", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "where the utilization of unlabeled data is crucial in both. In particular, consistency regularization", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 521, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 533 + ], + "score": 1.0, + "content": "based methods like (Laine & Aila, 2016; Tarvainen & Valpola, 2017) intend to utilize all unlabeled", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "data together with the supervision on labeled data, which is at the risk of strong confirmation bias.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 542, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 556 + ], + "score": 1.0, + "content": "Although recent holistic methods such as Sohn et al. (2020) realize consistency regularization by", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "combining entropy minimization via pseudo labeling, a fatal limitation is that they set a confidence", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 564, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 506, + 578 + ], + "score": 1.0, + "content": "threshold to control whether unlabeled data should participate in training, preventing low-confidence", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "score": 1.0, + "content": "unlabeled data from being effectively involved. Different from consistency based methods, recent", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "score": 1.0, + "content": "entropy minimization based methods Rizve et al. (2021) employ pseudo labeling to iteratively in-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 599, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 505, + 611 + ], + "score": 1.0, + "content": "corporate a part of low-confidence samples into training process. However, in this way, in addition", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 609, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 621 + ], + "score": 1.0, + "content": "to possible error accumulation, some low-confidence unlabeled samples are still neglected. In a", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "score": 1.0, + "content": "nutshell, the waste of unlabeled samples with low confidence causes the model difficult to learn the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 631, + 459, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 459, + 644 + ], + "score": 1.0, + "content": "potential pattern from all unlabeled data, which might deteriorate the final performance.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 505, + 736 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "Being aware of the aforementioned limitations, we try to answer — if we could treat these low-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 659, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 505, + 670 + ], + "score": 1.0, + "content": "confidence unlabeled samples in a novel way? As shown in Figure 1, imaging a low-confidence", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 669, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 505, + 682 + ], + "score": 1.0, + "content": "sample with its actual label of a “horse”, it might be hard for a trained model to propose an accurate", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "score": 1.0, + "content": "prediction, i.e., “it is a horse”. On the contrary, it can be much easier for the model to “guess what", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 691, + 506, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 506, + 704 + ], + "score": 1.0, + "content": "it is not, e.g., it is not a cat”. This drives us to consider a straightforward yet feasible direction", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "score": 1.0, + "content": "— for low-confidence samples, can we design a paradigm to exclude “what it is not” to benefit", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 713, + 506, + 727 + ], + "spans": [ + { + "bbox": [ + 105, + 713, + 506, + 727 + ], + "score": 1.0, + "content": "the learning of “what it is”. Intuitively, by introducing this paradigm, the space of searching the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 724, + 505, + 737 + ], + "spans": [ + { + "bbox": [ + 105, + 724, + 505, + 737 + ], + "score": 1.0, + "content": "optimal classifier could be largely reduced since the most impossible classes are initially excluded.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 79, + 504, + 136 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 226, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 226, + 96 + ], + "score": 1.0, + "content": "MUTEXMATCH:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 98, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 505, + 117 + ], + "score": 1.0, + "content": "SEMI-SUPERVISED LEARNING WITH MUTEX-BASED", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 118, + 354, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 354, + 137 + ], + "score": 1.0, + "content": "CONSISTENCY REGULARIZATION", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 112, + 155, + 244, + 177 + ], + "lines": [ + { + "bbox": [ + 113, + 155, + 201, + 167 + ], + "spans": [ + { + "bbox": [ + 113, + 155, + 201, + 167 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 111, + 166, + 245, + 178 + ], + "spans": [ + { + "bbox": [ + 111, + 166, + 245, + 178 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 111, + 155, + 245, + 178 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 206, + 333, + 218 + ], + "lines": [ + { + "bbox": [ + 277, + 206, + 335, + 219 + ], + "spans": [ + { + "bbox": [ + 277, + 206, + 335, + 219 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 143, + 231, + 468, + 418 + ], + "lines": [ + { + "bbox": [ + 142, + 232, + 469, + 244 + ], + "spans": [ + { + "bbox": [ + 142, + 232, + 469, + 244 + ], + "score": 1.0, + "content": "The core issue in semi-supervised learning (SSL) lies in how to effectively lever-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 243, + 470, + 255 + ], + "spans": [ + { + "bbox": [ + 141, + 243, + 470, + 255 + ], + "score": 1.0, + "content": "age unlabeled data, whereas most existing methods usually concentrate on the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 254, + 470, + 267 + ], + "spans": [ + { + "bbox": [ + 141, + 254, + 470, + 267 + ], + "score": 1.0, + "content": "utilization of high-confidence samples yet seldom fully explore the usage of low-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 264, + 470, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 264, + 470, + 277 + ], + "score": 1.0, + "content": "confidence samples. Early SSL methods mostly require low-confidence samples", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 275, + 470, + 289 + ], + "spans": [ + { + "bbox": [ + 141, + 275, + 470, + 289 + ], + "score": 1.0, + "content": "to optimize the same loss function as high-confidence samples, but this setting", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 142, + 287, + 469, + 299 + ], + "spans": [ + { + "bbox": [ + 142, + 287, + 469, + 299 + ], + "score": 1.0, + "content": "might largely challenge the low-confidence samples especially at the early train-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 297, + 470, + 310 + ], + "spans": [ + { + "bbox": [ + 141, + 297, + 470, + 310 + ], + "score": 1.0, + "content": "ing stage. In this paper, we aim to utilize low-confidence samples in a novel", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 308, + 469, + 322 + ], + "spans": [ + { + "bbox": [ + 141, + 308, + 469, + 322 + ], + "score": 1.0, + "content": "way, which is realized by our proposed mutex-based consistency regularization,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 319, + 470, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 319, + 470, + 332 + ], + "score": 1.0, + "content": "namely MutexMatch. To be specific, the high-confidence samples are required", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 331, + 469, + 343 + ], + "score": 1.0, + "content": "to exactly predict “What it is” by conventional True-Positive Classifier, while the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 341, + 470, + 354 + ], + "spans": [ + { + "bbox": [ + 141, + 341, + 470, + 354 + ], + "score": 1.0, + "content": "low-confidence samples, for a much simpler goal, are employed to predict “What", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 352, + 470, + 365 + ], + "spans": [ + { + "bbox": [ + 141, + 352, + 470, + 365 + ], + "score": 1.0, + "content": "it is not” by True-Negative Classifier with ease. In this way, we not only mitigate", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 363, + 470, + 375 + ], + "spans": [ + { + "bbox": [ + 141, + 363, + 470, + 375 + ], + "score": 1.0, + "content": "the pseudo-labeling errors but also make full use of the low-confidence unlabeled", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 374, + 470, + 387 + ], + "spans": [ + { + "bbox": [ + 141, + 374, + 470, + 387 + ], + "score": 1.0, + "content": "data in the training stage. The proposed MutexMatch achieves superior perfor-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 385, + 470, + 397 + ], + "spans": [ + { + "bbox": [ + 141, + 385, + 470, + 397 + ], + "score": 1.0, + "content": "mance on multiple benchmark datasets, i.e., CIFAR-10, CIFAR-100, SVHN, and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 396, + 469, + 408 + ], + "spans": [ + { + "bbox": [ + 141, + 396, + 469, + 408 + ], + "score": 1.0, + "content": "STL-10. Particularly, our method shows further superiority under few quantities", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 407, + 459, + 419 + ], + "spans": [ + { + "bbox": [ + 141, + 407, + 225, + 419 + ], + "score": 1.0, + "content": "of labeled data, e.g.,", + "type": "text" + }, + { + "bbox": [ + 225, + 407, + 257, + 418 + ], + "score": 0.84, + "content": "9 1 . 7 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 407, + 459, + 419 + ], + "score": 1.0, + "content": "accuracy with only 20 labeled data on CIFAR-10.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 14, + "bbox_fs": [ + 141, + 232, + 470, + 419 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 441, + 205, + 453 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 208, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 208, + 456 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 466, + 505, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 465, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 480 + ], + "score": 1.0, + "content": "Aiming to escape from time-consuming and laborious labeling tasks, semi-supervised learning", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 476, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 506, + 491 + ], + "score": 1.0, + "content": "(SSL) (Chapelle et al., 2009; Zhu, 2017) has been a longstanding yet important direction to leverage", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 104, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "a large quantity of unlabeled data along with few labeled data during training. Recent SSL models", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 500, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 505, + 512 + ], + "score": 1.0, + "content": "could be categorized into consistency regularization based or entropy minimization based methods", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "where the utilization of unlabeled data is crucial in both. In particular, consistency regularization", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 521, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 533 + ], + "score": 1.0, + "content": "based methods like (Laine & Aila, 2016; Tarvainen & Valpola, 2017) intend to utilize all unlabeled", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "data together with the supervision on labeled data, which is at the risk of strong confirmation bias.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 542, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 556 + ], + "score": 1.0, + "content": "Although recent holistic methods such as Sohn et al. (2020) realize consistency regularization by", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "combining entropy minimization via pseudo labeling, a fatal limitation is that they set a confidence", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 564, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 506, + 578 + ], + "score": 1.0, + "content": "threshold to control whether unlabeled data should participate in training, preventing low-confidence", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "score": 1.0, + "content": "unlabeled data from being effectively involved. Different from consistency based methods, recent", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "score": 1.0, + "content": "entropy minimization based methods Rizve et al. (2021) employ pseudo labeling to iteratively in-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 599, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 505, + 611 + ], + "score": 1.0, + "content": "corporate a part of low-confidence samples into training process. However, in this way, in addition", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 609, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 621 + ], + "score": 1.0, + "content": "to possible error accumulation, some low-confidence unlabeled samples are still neglected. In a", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "score": 1.0, + "content": "nutshell, the waste of unlabeled samples with low confidence causes the model difficult to learn the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 631, + 459, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 459, + 644 + ], + "score": 1.0, + "content": "potential pattern from all unlabeled data, which might deteriorate the final performance.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 31.5, + "bbox_fs": [ + 104, + 465, + 506, + 644 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 505, + 736 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "Being aware of the aforementioned limitations, we try to answer — if we could treat these low-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 659, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 505, + 670 + ], + "score": 1.0, + "content": "confidence unlabeled samples in a novel way? As shown in Figure 1, imaging a low-confidence", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 669, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 505, + 682 + ], + "score": 1.0, + "content": "sample with its actual label of a “horse”, it might be hard for a trained model to propose an accurate", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "score": 1.0, + "content": "prediction, i.e., “it is a horse”. On the contrary, it can be much easier for the model to “guess what", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 691, + 506, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 506, + 704 + ], + "score": 1.0, + "content": "it is not, e.g., it is not a cat”. This drives us to consider a straightforward yet feasible direction", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "score": 1.0, + "content": "— for low-confidence samples, can we design a paradigm to exclude “what it is not” to benefit", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 713, + 506, + 727 + ], + "spans": [ + { + "bbox": [ + 105, + 713, + 506, + 727 + ], + "score": 1.0, + "content": "the learning of “what it is”. Intuitively, by introducing this paradigm, the space of searching the", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 724, + 505, + 737 + ], + "spans": [ + { + "bbox": [ + 105, + 724, + 505, + 737 + ], + "score": 1.0, + "content": "optimal classifier could be largely reduced since the most impossible classes are initially excluded.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 648, + 506, + 737 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 310, + 137 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 311, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 311, + 95 + ], + "score": 1.0, + "content": "That is to say, for a low-confidence image, it is", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 310, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 310, + 105 + ], + "score": 1.0, + "content": "unnecessary to get a certain class, and its comple-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 311, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 311, + 117 + ], + "score": 1.0, + "content": "mentary pseudo-label is easier to obtain. Thus, we", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 311, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 311, + 127 + ], + "score": 1.0, + "content": "could learn less error information when using un-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 243, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 243, + 138 + ], + "score": 1.0, + "content": "labeled data with low confidence.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 144, + 309, + 242 + ], + "lines": [ + { + "bbox": [ + 106, + 143, + 311, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 311, + 155 + ], + "score": 1.0, + "content": "As aforementioned, to leverage all unlabeled data", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 311, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 311, + 166 + ], + "score": 1.0, + "content": "in a novel way, we propose MutexMatch, a new", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 165, + 311, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 311, + 177 + ], + "score": 1.0, + "content": "framework of SSL using mutex-based consistency", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 311, + 187 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 311, + 187 + ], + "score": 1.0, + "content": "regularization. We utilize True-Positive Classi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 186, + 311, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 311, + 199 + ], + "score": 1.0, + "content": "fier (TPC): to predict “What it is”, and True-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 311, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 311, + 210 + ], + "score": 1.0, + "content": "Negative Classifier (TNC): to predict “What it is", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 208, + 311, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 311, + 222 + ], + "score": 1.0, + "content": "not”, which is designed to learn feature representa-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 219, + 311, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 311, + 232 + ], + "score": 1.0, + "content": "tion of unlabeled data from a mutex perspective.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 231, + 311, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 311, + 243 + ], + "score": 1.0, + "content": "An improvement of MutexMatch compared with", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9 + }, + { + "type": "image", + "bbox": [ + 321, + 83, + 500, + 168 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 321, + 83, + 500, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 321, + 83, + 500, + 168 + ], + "spans": [ + { + "bbox": [ + 321, + 83, + 500, + 168 + ], + "score": 0.97, + "type": "image", + "image_path": "207647779515a6cfe4c418f405c73582ec86d916211dba5834302b336e5fbcd8.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 321, + 83, + 500, + 97.16666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 321, + 97.16666666666667, + 500, + 111.33333333333334 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 321, + 111.33333333333334, + 500, + 125.50000000000001 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 321, + 125.50000000000001, + 500, + 139.66666666666669 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 321, + 139.66666666666669, + 500, + 153.83333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 321, + 153.83333333333334, + 500, + 168.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 318, + 179, + 505, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 317, + 179, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 317, + 179, + 505, + 191 + ], + "score": 1.0, + "content": "Figure 1: Graphical explanations of the “Ex-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 318, + 190, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 318, + 190, + 505, + 201 + ], + "score": 1.0, + "content": "clusion Method” for classification task. For", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 317, + 201, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 317, + 201, + 505, + 213 + ], + "score": 1.0, + "content": "an image, the proposed method excludes some", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 317, + 212, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 317, + 212, + 505, + 223 + ], + "score": 1.0, + "content": "wrong classes via different images of the same", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 317, + 223, + 495, + 235 + ], + "spans": [ + { + "bbox": [ + 317, + 223, + 495, + 235 + ], + "score": 1.0, + "content": "class, so as to eliminate the wrong answers.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22 + } + ], + "index": 19.25 + }, + { + "type": "text", + "bbox": [ + 107, + 242, + 505, + 319 + ], + "lines": [ + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "score": 1.0, + "content": "the existing methods is that it allows low-confidence unlabeled samples to participate in training", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 253, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 505, + 266 + ], + "score": 1.0, + "content": "by optimizing a much simpler objective compared with that of high-confidence unlabeled samples.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 264, + 504, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 504, + 276 + ], + "score": 1.0, + "content": "Inspired by FixMatch (Sohn et al., 2020), the weakly-augmented unlabeled samples are used to gen-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "erate pseudo-labels, and we use RandAugument (Cubuk et al., 2020) for strong augmentation. We", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 285, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 299 + ], + "score": 1.0, + "content": "set a threshold to control the high-confidence portion and low-confidence portion of pseudo-labels.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 296, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 296, + 505, + 309 + ], + "score": 1.0, + "content": "In high-confidence portion and low-confidence portion, we enforce the consistency regularization", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 307, + 483, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 483, + 321 + ], + "score": 1.0, + "content": "on the output of TPC and TNC, respectively. A diagram of MutexMatch is shown in Figure 2.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 325, + 505, + 402 + ], + "lines": [ + { + "bbox": [ + 105, + 324, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 505, + 338 + ], + "score": 1.0, + "content": "In this work, the key contributions include three aspects: (1) We propose mutex-based consistency", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 336, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 505, + 348 + ], + "score": 1.0, + "content": "regularization for SSL, which can make full use of unlabeled data in a more effective way; (2) We", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 346, + 504, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 504, + 359 + ], + "score": 1.0, + "content": "use two classifiers (TPC and TNC) to construct MutexMatch, a novel framework using pseudo-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 357, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 505, + 371 + ], + "score": 1.0, + "content": "label and complementary label to learn an informative representation of unlabeled samples; (3) By", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "exploiting all unlabeled data, we can obtain better classification results in a label-scarce setting than", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 380, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 391 + ], + "score": 1.0, + "content": "recently-proposed SSL algorithm. For example, on the most commonly-studied SSL benchmark", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 390, + 477, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 416, + 403 + ], + "score": 1.0, + "content": "CIFAR-10, the accuracy of MutexMatch using only 20 labels can reach up to", + "type": "text" + }, + { + "bbox": [ + 416, + 390, + 473, + 401 + ], + "score": 0.88, + "content": "9 1 . 7 7 { \\pm } 2 . 6 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 390, + 477, + 403 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 108, + 429, + 210, + 442 + ], + "lines": [ + { + "bbox": [ + 104, + 428, + 213, + 445 + ], + "spans": [ + { + "bbox": [ + 104, + 428, + 213, + 445 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 462, + 505, + 637 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "score": 1.0, + "content": "Consistency regularization is a significant branch of recent state-of-the-art (SOTA) SSL methods,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 473, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 506, + 486 + ], + "score": 1.0, + "content": "which is proposed in Bachman et al. (2014). Such methods encourage the classifier to output same", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "class probability distribution after different versions of augmentation for the same unlabeled data.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 496, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 504, + 508 + ], + "score": 1.0, + "content": "Generally, the consistency regularization based models are trained with unlabeled data using the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 505, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 104, + 505, + 165, + 520 + ], + "score": 1.0, + "content": "loss function:", + "type": "text" + }, + { + "bbox": [ + 165, + 506, + 274, + 519 + ], + "score": 0.93, + "content": "| | p ( y | \\alpha ( x ) \\dot { - } p \\bar { ( y | \\alpha ( x ) ) ) } | | _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 505, + 305, + 520 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 306, + 508, + 313, + 516 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 505, + 406, + 520 + ], + "score": 1.0, + "content": "is the input image and", + "type": "text" + }, + { + "bbox": [ + 407, + 506, + 425, + 518 + ], + "score": 0.91, + "content": "\\alpha ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 505, + 506, + 520 + ], + "score": 1.0, + "content": "is an kind of trans-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 517, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 351, + 530 + ], + "score": 1.0, + "content": "formation that does not change the image label. Particularly,", + "type": "text" + }, + { + "bbox": [ + 351, + 517, + 369, + 529 + ], + "score": 0.9, + "content": "\\alpha ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 517, + 505, + 530 + ], + "score": 1.0, + "content": "can adopt different augmentation", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "methods, e.g., Mixup (Zhang et al., 2017) in Berthelot et al. (2019), RandAugment (Sohn et al.,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "2020) in Sohn et al. (2020) and CTAugment in Berthelot et al. (2020). Laine & Aila (2016) enforces", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "score": 1.0, + "content": "a loss of consistency on the predictions of two augmented variants of unlabeled data. In Tarvainen &", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "score": 1.0, + "content": "Valpola (2017), a teacher model is maintained to generate more stable targets for unlabeled data, and", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "the mean squared error is used to encourage same predictions of the student and teacher models. Xie", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "et al. (2020) adopts automatic augmentation for data perturbation and enforces a loss of consistency", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "by the KL divergence. Recently, some holistic methods (Sohn et al., 2020; Berthelot et al., 2020)", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 104, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 104, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "have been proposed to combine consistency regularization with pseudo-labeling for better SSL per-", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "formance. Differently, in MutexMatch, in addition to enforcing prediction consistency on TPC, we", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 627, + 455, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 455, + 639 + ], + "score": 1.0, + "content": "propose a novel mutex-based consistency to effectively leverage all unlabeled samples.", + "type": "text" + } + ], + "index": 55 + } + ], + "index": 47.5 + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "score": 1.0, + "content": "Pseudo-labeling is widely leveraged for entropy minimization by constructing one-hot labels from", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "predictions of unlabeled data with high-confidence and makes use of them based on cross-entropy", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "loss (Lee et al., 2013; Shi et al., 2018; Sohn et al., 2020; Xie et al., 2020). However, these methods", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "have a significant limitation, i.e., using confidence thresholds to select pseudo-labels results in that", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "all unlabeled data is not sufficiently exploited. Recent pseudo-labeling based method (Rizve et al.,", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "2021) proposes an uncertainty-aware pseudo-label selection framework to use both high and low-", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "confidence samples. However, it introduces two thresholds to control the pseudo-label generation,", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 105, + 720, + 307, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 307, + 732 + ], + "score": 1.0, + "content": "thus some unlabeled samples are still not utilized.", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 59.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 310, + 137 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 311, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 311, + 95 + ], + "score": 1.0, + "content": "That is to say, for a low-confidence image, it is", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 310, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 310, + 105 + ], + "score": 1.0, + "content": "unnecessary to get a certain class, and its comple-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 311, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 311, + 117 + ], + "score": 1.0, + "content": "mentary pseudo-label is easier to obtain. Thus, we", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 311, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 311, + 127 + ], + "score": 1.0, + "content": "could learn less error information when using un-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 243, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 243, + 138 + ], + "score": 1.0, + "content": "labeled data with low confidence.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 82, + 311, + 138 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 144, + 309, + 242 + ], + "lines": [ + { + "bbox": [ + 106, + 143, + 311, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 311, + 155 + ], + "score": 1.0, + "content": "As aforementioned, to leverage all unlabeled data", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 311, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 311, + 166 + ], + "score": 1.0, + "content": "in a novel way, we propose MutexMatch, a new", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 165, + 311, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 311, + 177 + ], + "score": 1.0, + "content": "framework of SSL using mutex-based consistency", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 311, + 187 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 311, + 187 + ], + "score": 1.0, + "content": "regularization. We utilize True-Positive Classi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 186, + 311, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 311, + 199 + ], + "score": 1.0, + "content": "fier (TPC): to predict “What it is”, and True-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 311, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 311, + 210 + ], + "score": 1.0, + "content": "Negative Classifier (TNC): to predict “What it is", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 208, + 311, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 311, + 222 + ], + "score": 1.0, + "content": "not”, which is designed to learn feature representa-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 219, + 311, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 311, + 232 + ], + "score": 1.0, + "content": "tion of unlabeled data from a mutex perspective.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 231, + 311, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 311, + 243 + ], + "score": 1.0, + "content": "An improvement of MutexMatch compared with", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "score": 1.0, + "content": "the existing methods is that it allows low-confidence unlabeled samples to participate in training", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 253, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 505, + 266 + ], + "score": 1.0, + "content": "by optimizing a much simpler objective compared with that of high-confidence unlabeled samples.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 264, + 504, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 504, + 276 + ], + "score": 1.0, + "content": "Inspired by FixMatch (Sohn et al., 2020), the weakly-augmented unlabeled samples are used to gen-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "erate pseudo-labels, and we use RandAugument (Cubuk et al., 2020) for strong augmentation. We", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 285, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 299 + ], + "score": 1.0, + "content": "set a threshold to control the high-confidence portion and low-confidence portion of pseudo-labels.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 296, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 296, + 505, + 309 + ], + "score": 1.0, + "content": "In high-confidence portion and low-confidence portion, we enforce the consistency regularization", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 307, + 483, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 483, + 321 + ], + "score": 1.0, + "content": "on the output of TPC and TNC, respectively. A diagram of MutexMatch is shown in Figure 2.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 143, + 311, + 243 + ] + }, + { + "type": "image", + "bbox": [ + 321, + 83, + 500, + 168 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 321, + 83, + 500, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 321, + 83, + 500, + 168 + ], + "spans": [ + { + "bbox": [ + 321, + 83, + 500, + 168 + ], + "score": 0.97, + "type": "image", + "image_path": "207647779515a6cfe4c418f405c73582ec86d916211dba5834302b336e5fbcd8.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 321, + 83, + 500, + 97.16666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 321, + 97.16666666666667, + 500, + 111.33333333333334 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 321, + 111.33333333333334, + 500, + 125.50000000000001 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 321, + 125.50000000000001, + 500, + 139.66666666666669 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 321, + 139.66666666666669, + 500, + 153.83333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 321, + 153.83333333333334, + 500, + 168.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 318, + 179, + 505, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 317, + 179, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 317, + 179, + 505, + 191 + ], + "score": 1.0, + "content": "Figure 1: Graphical explanations of the “Ex-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 318, + 190, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 318, + 190, + 505, + 201 + ], + "score": 1.0, + "content": "clusion Method” for classification task. For", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 317, + 201, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 317, + 201, + 505, + 213 + ], + "score": 1.0, + "content": "an image, the proposed method excludes some", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 317, + 212, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 317, + 212, + 505, + 223 + ], + "score": 1.0, + "content": "wrong classes via different images of the same", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 317, + 223, + 495, + 235 + ], + "spans": [ + { + "bbox": [ + 317, + 223, + 495, + 235 + ], + "score": 1.0, + "content": "class, so as to eliminate the wrong answers.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22 + } + ], + "index": 19.25 + }, + { + "type": "text", + "bbox": [ + 107, + 242, + 505, + 319 + ], + "lines": [], + "index": 28, + "bbox_fs": [ + 105, + 241, + 505, + 321 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 325, + 505, + 402 + ], + "lines": [ + { + "bbox": [ + 105, + 324, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 505, + 338 + ], + "score": 1.0, + "content": "In this work, the key contributions include three aspects: (1) We propose mutex-based consistency", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 336, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 505, + 348 + ], + "score": 1.0, + "content": "regularization for SSL, which can make full use of unlabeled data in a more effective way; (2) We", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 346, + 504, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 504, + 359 + ], + "score": 1.0, + "content": "use two classifiers (TPC and TNC) to construct MutexMatch, a novel framework using pseudo-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 357, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 505, + 371 + ], + "score": 1.0, + "content": "label and complementary label to learn an informative representation of unlabeled samples; (3) By", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "exploiting all unlabeled data, we can obtain better classification results in a label-scarce setting than", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 380, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 391 + ], + "score": 1.0, + "content": "recently-proposed SSL algorithm. For example, on the most commonly-studied SSL benchmark", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 390, + 477, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 416, + 403 + ], + "score": 1.0, + "content": "CIFAR-10, the accuracy of MutexMatch using only 20 labels can reach up to", + "type": "text" + }, + { + "bbox": [ + 416, + 390, + 473, + 401 + ], + "score": 0.88, + "content": "9 1 . 7 7 { \\pm } 2 . 6 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 390, + 477, + 403 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 324, + 505, + 403 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 429, + 210, + 442 + ], + "lines": [ + { + "bbox": [ + 104, + 428, + 213, + 445 + ], + "spans": [ + { + "bbox": [ + 104, + 428, + 213, + 445 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 462, + 505, + 637 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "score": 1.0, + "content": "Consistency regularization is a significant branch of recent state-of-the-art (SOTA) SSL methods,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 473, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 506, + 486 + ], + "score": 1.0, + "content": "which is proposed in Bachman et al. (2014). Such methods encourage the classifier to output same", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "class probability distribution after different versions of augmentation for the same unlabeled data.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 496, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 504, + 508 + ], + "score": 1.0, + "content": "Generally, the consistency regularization based models are trained with unlabeled data using the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 505, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 104, + 505, + 165, + 520 + ], + "score": 1.0, + "content": "loss function:", + "type": "text" + }, + { + "bbox": [ + 165, + 506, + 274, + 519 + ], + "score": 0.93, + "content": "| | p ( y | \\alpha ( x ) \\dot { - } p \\bar { ( y | \\alpha ( x ) ) ) } | | _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 505, + 305, + 520 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 306, + 508, + 313, + 516 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 505, + 406, + 520 + ], + "score": 1.0, + "content": "is the input image and", + "type": "text" + }, + { + "bbox": [ + 407, + 506, + 425, + 518 + ], + "score": 0.91, + "content": "\\alpha ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 505, + 506, + 520 + ], + "score": 1.0, + "content": "is an kind of trans-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 517, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 351, + 530 + ], + "score": 1.0, + "content": "formation that does not change the image label. Particularly,", + "type": "text" + }, + { + "bbox": [ + 351, + 517, + 369, + 529 + ], + "score": 0.9, + "content": "\\alpha ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 517, + 505, + 530 + ], + "score": 1.0, + "content": "can adopt different augmentation", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "methods, e.g., Mixup (Zhang et al., 2017) in Berthelot et al. (2019), RandAugment (Sohn et al.,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "2020) in Sohn et al. (2020) and CTAugment in Berthelot et al. (2020). Laine & Aila (2016) enforces", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 563 + ], + "score": 1.0, + "content": "a loss of consistency on the predictions of two augmented variants of unlabeled data. In Tarvainen &", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "score": 1.0, + "content": "Valpola (2017), a teacher model is maintained to generate more stable targets for unlabeled data, and", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "the mean squared error is used to encourage same predictions of the student and teacher models. Xie", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "et al. (2020) adopts automatic augmentation for data perturbation and enforces a loss of consistency", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "by the KL divergence. Recently, some holistic methods (Sohn et al., 2020; Berthelot et al., 2020)", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 104, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 104, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "have been proposed to combine consistency regularization with pseudo-labeling for better SSL per-", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "formance. Differently, in MutexMatch, in addition to enforcing prediction consistency on TPC, we", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 627, + 455, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 455, + 639 + ], + "score": 1.0, + "content": "propose a novel mutex-based consistency to effectively leverage all unlabeled samples.", + "type": "text" + } + ], + "index": 55 + } + ], + "index": 47.5, + "bbox_fs": [ + 104, + 462, + 506, + 639 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "score": 1.0, + "content": "Pseudo-labeling is widely leveraged for entropy minimization by constructing one-hot labels from", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "predictions of unlabeled data with high-confidence and makes use of them based on cross-entropy", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "loss (Lee et al., 2013; Shi et al., 2018; Sohn et al., 2020; Xie et al., 2020). However, these methods", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "have a significant limitation, i.e., using confidence thresholds to select pseudo-labels results in that", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "all unlabeled data is not sufficiently exploited. Recent pseudo-labeling based method (Rizve et al.,", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "2021) proposes an uncertainty-aware pseudo-label selection framework to use both high and low-", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "confidence samples. However, it introduces two thresholds to control the pseudo-label generation,", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 105, + 720, + 307, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 307, + 732 + ], + "score": 1.0, + "content": "thus some unlabeled samples are still not utilized.", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 59.5, + "bbox_fs": [ + 104, + 643, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 79, + 504, + 239 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 79, + 504, + 239 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 79, + 504, + 239 + ], + "spans": [ + { + "bbox": [ + 109, + 79, + 504, + 239 + ], + "score": 0.972, + "type": "image", + "image_path": "ea87b65a4a75f58c4f5b1c9dcf561cf90691b4f26b6f6e39703b4922c331498a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 79, + 504, + 132.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 132.33333333333334, + 504, + 185.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 185.66666666666669, + 504, + 239.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 257, + 505, + 335 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 258, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 474, + 270 + ], + "score": 1.0, + "content": "Figure 2: Diagram of the proposed MutexMatch. Given a batch of unlabeled samples, TPC", + "type": "text" + }, + { + "bbox": [ + 475, + 258, + 484, + 268 + ], + "score": 0.78, + "content": "\\mathcal { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 258, + 505, + 270 + ], + "score": 1.0, + "content": "uses", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 268, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 281 + ], + "score": 1.0, + "content": "their weakly-augmented variants to generate pseudo-labels. Then we adopt the classes with the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 355, + 292 + ], + "score": 1.0, + "content": "lowest confidence as the complementary labels to train TNC", + "type": "text" + }, + { + "bbox": [ + 356, + 280, + 367, + 290 + ], + "score": 0.72, + "content": "\\mathcal { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 280, + 506, + 292 + ], + "score": 1.0, + "content": "separately. Meanwhile, TPC and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "score": 1.0, + "content": "TNC are used for mutex-based consistency regularization in the high and low-confidence portion of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 300, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 236, + 315 + ], + "score": 1.0, + "content": "TPC’s predictions respectively.", + "type": "text" + }, + { + "bbox": [ + 236, + 302, + 244, + 313 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 300, + 359, + 315 + ], + "score": 1.0, + "content": "denotes output features and", + "type": "text" + }, + { + "bbox": [ + 360, + 303, + 377, + 313 + ], + "score": 0.26, + "content": "p , r", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 300, + 506, + 315 + ], + "score": 1.0, + "content": "denote predictions of TPC and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 183, + 326 + ], + "score": 1.0, + "content": "TNC. Superscripts", + "type": "text" + }, + { + "bbox": [ + 184, + 315, + 192, + 322 + ], + "score": 0.77, + "content": "w", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 311, + 211, + 326 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 211, + 315, + 217, + 322 + ], + "score": 0.67, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 311, + 506, + 326 + ], + "score": 1.0, + "content": "represent corresponding outputs for the weakly-augmented variant and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 325, + 272, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 272, + 336 + ], + "score": 1.0, + "content": "strongly-augmented variant, respectively.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 345, + 505, + 422 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 506, + 359 + ], + "score": 1.0, + "content": "Complementary label is used to help the model learn which class the input image does not belong", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 356, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 354, + 369 + ], + "score": 1.0, + "content": "to (Ishida et al., 2017; 2018; Yu et al., 2018). Considering a", + "type": "text" + }, + { + "bbox": [ + 354, + 358, + 360, + 366 + ], + "score": 0.73, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 356, + 506, + 369 + ], + "score": 1.0, + "content": "-class classification task, we denote", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 367, + 504, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 134, + 377 + ], + "score": 0.9, + "content": "x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 367, + 223, + 379 + ], + "score": 1.0, + "content": "as an input image and", + "type": "text" + }, + { + "bbox": [ + 224, + 367, + 299, + 379 + ], + "score": 0.92, + "content": "y \\in \\mathcal { Y } = \\{ 1 , . . . , c \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 367, + 435, + 379 + ], + "score": 1.0, + "content": "as its label. Complementary label", + "type": "text" + }, + { + "bbox": [ + 435, + 368, + 442, + 378 + ], + "score": 0.82, + "content": "\\overline { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 367, + 504, + 379 + ], + "score": 1.0, + "content": "is generated by", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 153, + 390 + ], + "score": 1.0, + "content": "select from", + "type": "text" + }, + { + "bbox": [ + 153, + 378, + 181, + 390 + ], + "score": 0.92, + "content": "\\mathcal { V } \\backslash \\{ \\boldsymbol { y } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 378, + 404, + 390 + ], + "score": 1.0, + "content": "at random. Inspired by Rizve et al. (2021) and Kim et al.", + "type": "text" + }, + { + "bbox": [ + 405, + 378, + 435, + 389 + ], + "score": 0.48, + "content": "( 2 0 1 9 ) ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 378, + 505, + 390 + ], + "score": 1.0, + "content": ", in MutexMatch,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "we design a novel way (detailed in Section 3.2) to propose complementary labels, so as to ensure", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "score": 1.0, + "content": "their effectiveness in semi supervised learning. Experiments about using standard complementary", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 411, + 285, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 285, + 423 + ], + "score": 1.0, + "content": "label selections are discussed in Section 5.2.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13 + }, + { + "type": "title", + "bbox": [ + 108, + 438, + 204, + 451 + ], + "lines": [ + { + "bbox": [ + 104, + 436, + 206, + 453 + ], + "spans": [ + { + "bbox": [ + 104, + 436, + 206, + 453 + ], + "score": 1.0, + "content": "3 MUTEXMATCH", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 107, + 462, + 178, + 474 + ], + "lines": [ + { + "bbox": [ + 105, + 461, + 180, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 180, + 476 + ], + "score": 1.0, + "content": "3.1 OVERVIEW", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 483, + 505, + 572 + ], + "lines": [ + { + "bbox": [ + 105, + 482, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 402, + 497 + ], + "score": 1.0, + "content": "Different from existing SSL approaches, in addition to a feature extractor", + "type": "text" + }, + { + "bbox": [ + 402, + 483, + 419, + 496 + ], + "score": 0.9, + "content": "\\theta ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 482, + 505, + 497 + ], + "score": 1.0, + "content": ", MutexMatch jointly", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 358, + 507 + ], + "score": 1.0, + "content": "trains two distinct classifiers, a True-Positive Classifier (TPC)", + "type": "text" + }, + { + "bbox": [ + 359, + 495, + 378, + 507 + ], + "score": 0.91, + "content": "\\mathcal { P } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "and a True-Negative Classifier", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 136, + 518 + ], + "score": 1.0, + "content": "(TNC)", + "type": "text" + }, + { + "bbox": [ + 136, + 505, + 157, + 517 + ], + "score": 0.88, + "content": "\\mathcal { N } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 505, + 505, + 518 + ], + "score": 1.0, + "content": ". To be specific, TPC is used to predict which class the instance belongs to (i.e., true", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 104, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "positive), while TNC is employed to indicate which class the instance is not (i.e., true negative).", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 410, + 540 + ], + "score": 1.0, + "content": "To mitigate pseudo-labeling errors, a pre-defined high-confidence threshold", + "type": "text" + }, + { + "bbox": [ + 410, + 529, + 417, + 537 + ], + "score": 0.79, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 527, + 505, + 540 + ], + "score": 1.0, + "content": "is utilized to split the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "unlabeled data into high-confidence and low-confidence portions. Besides training TPC on the high-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "score": 1.0, + "content": "confidence portion, we explore complementary labels on low-confidence samples to train TNC. In", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 561, + 354, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 354, + 573 + ], + "score": 1.0, + "content": "this way, all the unlabeled data could be effectively exploited.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 505, + 646 + ], + "lines": [ + { + "bbox": [ + 103, + 574, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 103, + 574, + 218, + 592 + ], + "score": 1.0, + "content": "In a mini-batch, we have", + "type": "text" + }, + { + "bbox": [ + 218, + 578, + 227, + 587 + ], + "score": 0.81, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 574, + 285, + 592 + ], + "score": 1.0, + "content": "labeled data", + "type": "text" + }, + { + "bbox": [ + 285, + 576, + 376, + 590 + ], + "score": 0.93, + "content": "\\mathcal { X } ~ = ~ \\{ ( x _ { b } ^ { l b } , y _ { b } ^ { l b } ) \\} _ { b = 1 } ^ { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 574, + 398, + 592 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 398, + 578, + 414, + 589 + ], + "score": 0.88, + "content": "\\mu B", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 574, + 480, + 592 + ], + "score": 1.0, + "content": "unlabeled data", + "type": "text" + }, + { + "bbox": [ + 480, + 578, + 505, + 589 + ], + "score": 0.86, + "content": "u \\ =", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 107, + 586, + 507, + 608 + ], + "spans": [ + { + "bbox": [ + 107, + 588, + 177, + 603 + ], + "score": 0.93, + "content": "\\{ ( x _ { b } ^ { u l b } , y _ { b } ^ { u l b } ) \\} _ { b = 1 } ^ { \\mu B }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 586, + 208, + 608 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 209, + 593, + 217, + 603 + ], + "score": 0.81, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 586, + 338, + 608 + ], + "score": 1.0, + "content": "represents the relative size of", + "type": "text" + }, + { + "bbox": [ + 339, + 591, + 348, + 601 + ], + "score": 0.83, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 586, + 366, + 608 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 367, + 591, + 376, + 601 + ], + "score": 0.78, + "content": "\\mathcal { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 586, + 507, + 608 + ], + "score": 1.0, + "content": ". Following (Sohn et al., 2020),", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 422, + 614 + ], + "score": 1.0, + "content": "we perform weak and strong augmentations for data perturbations, denoted by", + "type": "text" + }, + { + "bbox": [ + 422, + 602, + 446, + 613 + ], + "score": 0.91, + "content": "\\alpha _ { w } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 601, + 465, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 465, + 602, + 487, + 613 + ], + "score": 0.91, + "content": "\\alpha _ { s } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 601, + 505, + 614 + ], + "score": 1.0, + "content": ", re-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 613, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 289, + 625 + ], + "score": 1.0, + "content": "spectively. Given weakly-augmented instance", + "type": "text" + }, + { + "bbox": [ + 290, + 613, + 303, + 623 + ], + "score": 0.88, + "content": "x ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 613, + 437, + 625 + ], + "score": 1.0, + "content": "and strongly-augmented instance", + "type": "text" + }, + { + "bbox": [ + 437, + 614, + 447, + 623 + ], + "score": 0.85, + "content": "x ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 613, + 505, + 625 + ], + "score": 1.0, + "content": ", MutexMatch", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 622, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 104, + 622, + 335, + 637 + ], + "score": 1.0, + "content": "simultaneously optimizes four losses: the supervised loss", + "type": "text" + }, + { + "bbox": [ + 335, + 624, + 356, + 636 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s u p }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 622, + 466, + 637 + ], + "score": 1.0, + "content": ", the separated negative loss", + "type": "text" + }, + { + "bbox": [ + 467, + 624, + 487, + 636 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 622, + 506, + 637 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 634, + 488, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 207, + 648 + ], + "score": 1.0, + "content": "positive consistency loss", + "type": "text" + }, + { + "bbox": [ + 207, + 635, + 219, + 647 + ], + "score": 0.9, + "content": "{ \\mathcal { L } } _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 634, + 355, + 648 + ], + "score": 1.0, + "content": "and the negative consistency loss", + "type": "text" + }, + { + "bbox": [ + 355, + 635, + 368, + 646 + ], + "score": 0.9, + "content": "{ \\mathcal { L } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 634, + 488, + 648 + ], + "score": 1.0, + "content": ". In summary, the total loss is", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5 + }, + { + "type": "interline_equation", + "bbox": [ + 223, + 651, + 387, + 664 + ], + "lines": [ + { + "bbox": [ + 223, + 651, + 387, + 664 + ], + "spans": [ + { + "bbox": [ + 223, + 651, + 387, + 664 + ], + "score": 0.91, + "content": "\\mathcal { L } = \\mathcal { L } _ { s u p } + \\lambda _ { s e p } \\mathcal { L } _ { s e p } + \\lambda _ { p } \\mathcal { L } _ { p } + \\lambda _ { n } \\mathcal { L } _ { n } ,", + "type": "interline_equation", + "image_path": "17ceca4159129e15e6c22e3c347a7dca6a8f38ccfab8d67e0a1d93a92c01d37b.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 223, + 651, + 387, + 664 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 668, + 502, + 693 + ], + "lines": [ + { + "bbox": [ + 107, + 668, + 504, + 682 + ], + "spans": [ + { + "bbox": [ + 107, + 668, + 133, + 682 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 669, + 152, + 681 + ], + "score": 0.75, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 668, + 156, + 682 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 156, + 669, + 168, + 681 + ], + "score": 0.65, + "content": "\\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 668, + 186, + 682 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 186, + 669, + 199, + 680 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 668, + 504, + 682 + ], + "score": 1.0, + "content": "are scalar hyper-parameters to adjust the relative importance of correspond-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 680, + 504, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 239, + 694 + ], + "score": 1.0, + "content": "ing losses. The supervised loss", + "type": "text" + }, + { + "bbox": [ + 239, + 681, + 260, + 694 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s u p }", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 680, + 457, + 694 + ], + "score": 1.0, + "content": "is simply defined as the cross-entropy between", + "type": "text" + }, + { + "bbox": [ + 457, + 680, + 470, + 693 + ], + "score": 0.9, + "content": "y ^ { l b }", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 680, + 504, + 694 + ], + "score": 1.0, + "content": "and the", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 701, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 118, + 699, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 118, + 699, + 505, + 714 + ], + "score": 1.0, + "content": "1In Kim et al. (2019), complementary label based negative learning shows great potential for noisy labels.", + "type": "text" + } + ] + }, + { + "bbox": [ + 106, + 711, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 506, + 723 + ], + "score": 1.0, + "content": "Through our method using complementary label, we also achieve amazing performance under the more difficult", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 721, + 464, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 464, + 733 + ], + "score": 1.0, + "content": "setting of semi-supervised with noisy labels and the results can be found in Section C in Appendix.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 79, + 504, + 239 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 79, + 504, + 239 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 79, + 504, + 239 + ], + "spans": [ + { + "bbox": [ + 109, + 79, + 504, + 239 + ], + "score": 0.972, + "type": "image", + "image_path": "ea87b65a4a75f58c4f5b1c9dcf561cf90691b4f26b6f6e39703b4922c331498a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 79, + 504, + 132.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 132.33333333333334, + 504, + 185.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 185.66666666666669, + 504, + 239.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 257, + 505, + 335 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 258, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 474, + 270 + ], + "score": 1.0, + "content": "Figure 2: Diagram of the proposed MutexMatch. Given a batch of unlabeled samples, TPC", + "type": "text" + }, + { + "bbox": [ + 475, + 258, + 484, + 268 + ], + "score": 0.78, + "content": "\\mathcal { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 258, + 505, + 270 + ], + "score": 1.0, + "content": "uses", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 268, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 281 + ], + "score": 1.0, + "content": "their weakly-augmented variants to generate pseudo-labels. Then we adopt the classes with the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 355, + 292 + ], + "score": 1.0, + "content": "lowest confidence as the complementary labels to train TNC", + "type": "text" + }, + { + "bbox": [ + 356, + 280, + 367, + 290 + ], + "score": 0.72, + "content": "\\mathcal { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 280, + 506, + 292 + ], + "score": 1.0, + "content": "separately. Meanwhile, TPC and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 304 + ], + "score": 1.0, + "content": "TNC are used for mutex-based consistency regularization in the high and low-confidence portion of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 300, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 236, + 315 + ], + "score": 1.0, + "content": "TPC’s predictions respectively.", + "type": "text" + }, + { + "bbox": [ + 236, + 302, + 244, + 313 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 300, + 359, + 315 + ], + "score": 1.0, + "content": "denotes output features and", + "type": "text" + }, + { + "bbox": [ + 360, + 303, + 377, + 313 + ], + "score": 0.26, + "content": "p , r", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 300, + 506, + 315 + ], + "score": 1.0, + "content": "denote predictions of TPC and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 183, + 326 + ], + "score": 1.0, + "content": "TNC. Superscripts", + "type": "text" + }, + { + "bbox": [ + 184, + 315, + 192, + 322 + ], + "score": 0.77, + "content": "w", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 311, + 211, + 326 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 211, + 315, + 217, + 322 + ], + "score": 0.67, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 311, + 506, + 326 + ], + "score": 1.0, + "content": "represent corresponding outputs for the weakly-augmented variant and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 325, + 272, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 272, + 336 + ], + "score": 1.0, + "content": "strongly-augmented variant, respectively.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 345, + 505, + 422 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 506, + 359 + ], + "score": 1.0, + "content": "Complementary label is used to help the model learn which class the input image does not belong", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 356, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 354, + 369 + ], + "score": 1.0, + "content": "to (Ishida et al., 2017; 2018; Yu et al., 2018). Considering a", + "type": "text" + }, + { + "bbox": [ + 354, + 358, + 360, + 366 + ], + "score": 0.73, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 356, + 506, + 369 + ], + "score": 1.0, + "content": "-class classification task, we denote", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 367, + 504, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 134, + 377 + ], + "score": 0.9, + "content": "x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 367, + 223, + 379 + ], + "score": 1.0, + "content": "as an input image and", + "type": "text" + }, + { + "bbox": [ + 224, + 367, + 299, + 379 + ], + "score": 0.92, + "content": "y \\in \\mathcal { Y } = \\{ 1 , . . . , c \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 367, + 435, + 379 + ], + "score": 1.0, + "content": "as its label. Complementary label", + "type": "text" + }, + { + "bbox": [ + 435, + 368, + 442, + 378 + ], + "score": 0.82, + "content": "\\overline { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 367, + 504, + 379 + ], + "score": 1.0, + "content": "is generated by", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 153, + 390 + ], + "score": 1.0, + "content": "select from", + "type": "text" + }, + { + "bbox": [ + 153, + 378, + 181, + 390 + ], + "score": 0.92, + "content": "\\mathcal { V } \\backslash \\{ \\boldsymbol { y } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 378, + 404, + 390 + ], + "score": 1.0, + "content": "at random. Inspired by Rizve et al. (2021) and Kim et al.", + "type": "text" + }, + { + "bbox": [ + 405, + 378, + 435, + 389 + ], + "score": 0.48, + "content": "( 2 0 1 9 ) ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 378, + 505, + 390 + ], + "score": 1.0, + "content": ", in MutexMatch,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "we design a novel way (detailed in Section 3.2) to propose complementary labels, so as to ensure", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 414 + ], + "score": 1.0, + "content": "their effectiveness in semi supervised learning. Experiments about using standard complementary", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 411, + 285, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 285, + 423 + ], + "score": 1.0, + "content": "label selections are discussed in Section 5.2.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 344, + 506, + 423 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 438, + 204, + 451 + ], + "lines": [ + { + "bbox": [ + 104, + 436, + 206, + 453 + ], + "spans": [ + { + "bbox": [ + 104, + 436, + 206, + 453 + ], + "score": 1.0, + "content": "3 MUTEXMATCH", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 107, + 462, + 178, + 474 + ], + "lines": [ + { + "bbox": [ + 105, + 461, + 180, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 180, + 476 + ], + "score": 1.0, + "content": "3.1 OVERVIEW", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 483, + 505, + 572 + ], + "lines": [ + { + "bbox": [ + 105, + 482, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 402, + 497 + ], + "score": 1.0, + "content": "Different from existing SSL approaches, in addition to a feature extractor", + "type": "text" + }, + { + "bbox": [ + 402, + 483, + 419, + 496 + ], + "score": 0.9, + "content": "\\theta ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 482, + 505, + 497 + ], + "score": 1.0, + "content": ", MutexMatch jointly", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 358, + 507 + ], + "score": 1.0, + "content": "trains two distinct classifiers, a True-Positive Classifier (TPC)", + "type": "text" + }, + { + "bbox": [ + 359, + 495, + 378, + 507 + ], + "score": 0.91, + "content": "\\mathcal { P } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "and a True-Negative Classifier", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 136, + 518 + ], + "score": 1.0, + "content": "(TNC)", + "type": "text" + }, + { + "bbox": [ + 136, + 505, + 157, + 517 + ], + "score": 0.88, + "content": "\\mathcal { N } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 505, + 505, + 518 + ], + "score": 1.0, + "content": ". To be specific, TPC is used to predict which class the instance belongs to (i.e., true", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 104, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "positive), while TNC is employed to indicate which class the instance is not (i.e., true negative).", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 410, + 540 + ], + "score": 1.0, + "content": "To mitigate pseudo-labeling errors, a pre-defined high-confidence threshold", + "type": "text" + }, + { + "bbox": [ + 410, + 529, + 417, + 537 + ], + "score": 0.79, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 527, + 505, + 540 + ], + "score": 1.0, + "content": "is utilized to split the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "unlabeled data into high-confidence and low-confidence portions. Besides training TPC on the high-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 562 + ], + "score": 1.0, + "content": "confidence portion, we explore complementary labels on low-confidence samples to train TNC. In", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 561, + 354, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 354, + 573 + ], + "score": 1.0, + "content": "this way, all the unlabeled data could be effectively exploited.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22.5, + "bbox_fs": [ + 104, + 482, + 505, + 573 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 505, + 646 + ], + "lines": [ + { + "bbox": [ + 103, + 574, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 103, + 574, + 218, + 592 + ], + "score": 1.0, + "content": "In a mini-batch, we have", + "type": "text" + }, + { + "bbox": [ + 218, + 578, + 227, + 587 + ], + "score": 0.81, + "content": "B", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 574, + 285, + 592 + ], + "score": 1.0, + "content": "labeled data", + "type": "text" + }, + { + "bbox": [ + 285, + 576, + 376, + 590 + ], + "score": 0.93, + "content": "\\mathcal { X } ~ = ~ \\{ ( x _ { b } ^ { l b } , y _ { b } ^ { l b } ) \\} _ { b = 1 } ^ { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 574, + 398, + 592 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 398, + 578, + 414, + 589 + ], + "score": 0.88, + "content": "\\mu B", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 574, + 480, + 592 + ], + "score": 1.0, + "content": "unlabeled data", + "type": "text" + }, + { + "bbox": [ + 480, + 578, + 505, + 589 + ], + "score": 0.86, + "content": "u \\ =", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 107, + 586, + 507, + 608 + ], + "spans": [ + { + "bbox": [ + 107, + 588, + 177, + 603 + ], + "score": 0.93, + "content": "\\{ ( x _ { b } ^ { u l b } , y _ { b } ^ { u l b } ) \\} _ { b = 1 } ^ { \\mu B }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 586, + 208, + 608 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 209, + 593, + 217, + 603 + ], + "score": 0.81, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 586, + 338, + 608 + ], + "score": 1.0, + "content": "represents the relative size of", + "type": "text" + }, + { + "bbox": [ + 339, + 591, + 348, + 601 + ], + "score": 0.83, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 586, + 366, + 608 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 367, + 591, + 376, + 601 + ], + "score": 0.78, + "content": "\\mathcal { U }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 586, + 507, + 608 + ], + "score": 1.0, + "content": ". Following (Sohn et al., 2020),", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 422, + 614 + ], + "score": 1.0, + "content": "we perform weak and strong augmentations for data perturbations, denoted by", + "type": "text" + }, + { + "bbox": [ + 422, + 602, + 446, + 613 + ], + "score": 0.91, + "content": "\\alpha _ { w } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 601, + 465, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 465, + 602, + 487, + 613 + ], + "score": 0.91, + "content": "\\alpha _ { s } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 601, + 505, + 614 + ], + "score": 1.0, + "content": ", re-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 613, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 289, + 625 + ], + "score": 1.0, + "content": "spectively. Given weakly-augmented instance", + "type": "text" + }, + { + "bbox": [ + 290, + 613, + 303, + 623 + ], + "score": 0.88, + "content": "x ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 613, + 437, + 625 + ], + "score": 1.0, + "content": "and strongly-augmented instance", + "type": "text" + }, + { + "bbox": [ + 437, + 614, + 447, + 623 + ], + "score": 0.85, + "content": "x ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 613, + 505, + 625 + ], + "score": 1.0, + "content": ", MutexMatch", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 622, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 104, + 622, + 335, + 637 + ], + "score": 1.0, + "content": "simultaneously optimizes four losses: the supervised loss", + "type": "text" + }, + { + "bbox": [ + 335, + 624, + 356, + 636 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s u p }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 622, + 466, + 637 + ], + "score": 1.0, + "content": ", the separated negative loss", + "type": "text" + }, + { + "bbox": [ + 467, + 624, + 487, + 636 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 622, + 506, + 637 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 634, + 488, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 207, + 648 + ], + "score": 1.0, + "content": "positive consistency loss", + "type": "text" + }, + { + "bbox": [ + 207, + 635, + 219, + 647 + ], + "score": 0.9, + "content": "{ \\mathcal { L } } _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 634, + 355, + 648 + ], + "score": 1.0, + "content": "and the negative consistency loss", + "type": "text" + }, + { + "bbox": [ + 355, + 635, + 368, + 646 + ], + "score": 0.9, + "content": "{ \\mathcal { L } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 634, + 488, + 648 + ], + "score": 1.0, + "content": ". In summary, the total loss is", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5, + "bbox_fs": [ + 103, + 574, + 507, + 648 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 223, + 651, + 387, + 664 + ], + "lines": [ + { + "bbox": [ + 223, + 651, + 387, + 664 + ], + "spans": [ + { + "bbox": [ + 223, + 651, + 387, + 664 + ], + "score": 0.91, + "content": "\\mathcal { L } = \\mathcal { L } _ { s u p } + \\lambda _ { s e p } \\mathcal { L } _ { s e p } + \\lambda _ { p } \\mathcal { L } _ { p } + \\lambda _ { n } \\mathcal { L } _ { n } ,", + "type": "interline_equation", + "image_path": "17ceca4159129e15e6c22e3c347a7dca6a8f38ccfab8d67e0a1d93a92c01d37b.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 223, + 651, + 387, + 664 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 668, + 502, + 693 + ], + "lines": [ + { + "bbox": [ + 107, + 668, + 504, + 682 + ], + "spans": [ + { + "bbox": [ + 107, + 668, + 133, + 682 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 669, + 152, + 681 + ], + "score": 0.75, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 668, + 156, + 682 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 156, + 669, + 168, + 681 + ], + "score": 0.65, + "content": "\\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 668, + 186, + 682 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 186, + 669, + 199, + 680 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 668, + 504, + 682 + ], + "score": 1.0, + "content": "are scalar hyper-parameters to adjust the relative importance of correspond-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 680, + 504, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 239, + 694 + ], + "score": 1.0, + "content": "ing losses. The supervised loss", + "type": "text" + }, + { + "bbox": [ + 239, + 681, + 260, + 694 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s u p }", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 680, + 457, + 694 + ], + "score": 1.0, + "content": "is simply defined as the cross-entropy between", + "type": "text" + }, + { + "bbox": [ + 457, + 680, + 470, + 693 + ], + "score": 0.9, + "content": "y ^ { l b }", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 680, + 504, + 694 + ], + "score": 1.0, + "content": "and the", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 668, + 504, + 694 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 353, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 81, + 354, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 248, + 96 + ], + "score": 1.0, + "content": "predictions of TPC on labeled data", + "type": "text" + }, + { + "bbox": [ + 248, + 82, + 261, + 92 + ], + "score": 0.89, + "content": "x ^ { l b }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 81, + 354, + 96 + ], + "score": 1.0, + "content": ", calculated as follows:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 97, + 385, + 131 + ], + "lines": [ + { + "bbox": [ + 225, + 97, + 385, + 131 + ], + "spans": [ + { + "bbox": [ + 225, + 97, + 385, + 131 + ], + "score": 0.95, + "content": "\\mathcal { L } _ { s u p } = \\frac { 1 } { B } \\sum _ { n = 1 } ^ { B } H ( y _ { n } ^ { l b } , \\mathcal { P } ( \\theta ( \\alpha _ { w } ( x _ { n } ^ { l b } ) ) ) ) ,", + "type": "interline_equation", + "image_path": "94d5af6b19d4ea4eb5e473283cd7b9e924e3d1a33f3223ff5ad6886418087623.jpg" + } + ] + } + ], + "index": 1.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 97, + 385, + 114.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 225, + 114.0, + 385, + 131.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 133, + 442, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 132, + 439, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 133, + 148 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 133, + 165, + 146 + ], + "score": 0.93, + "content": "H ( p , q )", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 132, + 409, + 148 + ], + "score": 1.0, + "content": "denotes the standard cross-entropy loss between distribution", + "type": "text" + }, + { + "bbox": [ + 410, + 136, + 415, + 145 + ], + "score": 0.81, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 132, + 433, + 148 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 433, + 136, + 439, + 145 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 108, + 158, + 257, + 169 + ], + "lines": [ + { + "bbox": [ + 105, + 157, + 258, + 170 + ], + "spans": [ + { + "bbox": [ + 105, + 157, + 258, + 170 + ], + "score": 1.0, + "content": "3.2 TRUE-NEGATIVE CLASSIFIER", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 178, + 297, + 288 + ], + "lines": [ + { + "bbox": [ + 105, + 178, + 297, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 297, + 190 + ], + "score": 1.0, + "content": "In multi-class classification tasks, for a specific", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 189, + 297, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 297, + 201 + ], + "score": 1.0, + "content": "instance, it is easier to predict which class it", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 200, + 297, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 297, + 212 + ], + "score": 1.0, + "content": "does not belong to than to know which class it", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 211, + 297, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 297, + 223 + ], + "score": 1.0, + "content": "exactly is. For example, given an image of air-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 222, + 297, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 297, + 234 + ], + "score": 1.0, + "content": "plane in CIFAR-10, we can predict which class", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 233, + 297, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 265, + 245 + ], + "score": 1.0, + "content": "it does not belong to with a probability", + "type": "text" + }, + { + "bbox": [ + 266, + 234, + 286, + 244 + ], + "score": 0.85, + "content": "90 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 233, + 297, + 245 + ], + "score": 1.0, + "content": "at", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 244, + 296, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 296, + 257 + ], + "score": 1.0, + "content": "random, whereas we only have the probability", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 255, + 297, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 118, + 267 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 255, + 137, + 266 + ], + "score": 0.87, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 255, + 297, + 267 + ], + "score": 1.0, + "content": "to correctly predict it is an airplane. To", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 266, + 297, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 297, + 278 + ], + "score": 1.0, + "content": "this end, we design a True-Negative Classifier", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 277, + 297, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 297, + 289 + ], + "score": 1.0, + "content": "to predict which class it is not. Compared to", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 9.5 + }, + { + "type": "image", + "bbox": [ + 312, + 180, + 494, + 253 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 312, + 180, + 494, + 253 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 312, + 180, + 494, + 253 + ], + "spans": [ + { + "bbox": [ + 312, + 180, + 494, + 253 + ], + "score": 0.966, + "type": "image", + "image_path": "6de18e33fa1a4e3e4c06c6e0e1140d78d074d9ef3c09a6db8bcbe63da0b747b8.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 312, + 180, + 494, + 194.6 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 312, + 194.6, + 494, + 209.2 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 312, + 209.2, + 494, + 223.79999999999998 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 312, + 223.79999999999998, + 494, + 238.39999999999998 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 312, + 238.39999999999998, + 494, + 252.99999999999997 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 350, + 264, + 459, + 276 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 349, + 263, + 459, + 277 + ], + "spans": [ + { + "bbox": [ + 349, + 263, + 459, + 277 + ], + "score": 1.0, + "content": "Figure 3: Training of TNC", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 288, + 505, + 355 + ], + "lines": [ + { + "bbox": [ + 106, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 505, + 300 + ], + "score": 1.0, + "content": "TPC, it is much easier to obtain correct labels for TNC. Thus we exploit TNC to provide more", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 299, + 506, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 506, + 312 + ], + "score": 1.0, + "content": "guidance information on unlabeled data. We then propose a mutex-based prediction consistency on", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 310, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 505, + 321 + ], + "score": 1.0, + "content": "TPC and TNC to make full use of unlabeled data, which is described in Section 3.3. The high-level", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "score": 1.0, + "content": "training process of TNC is shown in Figure 3. Unlike the standard complementary label genera-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 331, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 506, + 344 + ], + "score": 1.0, + "content": "tions (Ishida et al., 2017; Yu et al., 2018), we use the class with the lowest confidence in TPC’s", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 343, + 494, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 390, + 356 + ], + "score": 1.0, + "content": "predictions as the complementary label to train TNC. The training loss", + "type": "text" + }, + { + "bbox": [ + 390, + 343, + 411, + 356 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 343, + 494, + 356 + ], + "score": 1.0, + "content": "can be calculated as", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23.5 + }, + { + "type": "interline_equation", + "bbox": [ + 199, + 358, + 411, + 392 + ], + "lines": [ + { + "bbox": [ + 199, + 358, + 411, + 392 + ], + "spans": [ + { + "bbox": [ + 199, + 358, + 411, + 392 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { s e p } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } H ( \\arg \\operatorname* { m i n } ( \\mathcal { P } ( \\theta ( x _ { n } ^ { w } ) ) , \\mathcal { N } ( \\hat { \\theta } ( x _ { n } ^ { w } ) ) ) ,", + "type": "interline_equation", + "image_path": "d2d9bd984b9451ef77afad1222366f11067fd5814b5cb4fbbd9aa85685eaf7e1.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 199, + 358, + 411, + 375.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 199, + 375.0, + 411, + 392.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 396, + 505, + 442 + ], + "lines": [ + { + "bbox": [ + 106, + 395, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 134, + 409 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 395, + 141, + 407 + ], + "score": 0.82, + "content": "\\hat { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 396, + 207, + 409 + ], + "score": 1.0, + "content": "represents that", + "type": "text" + }, + { + "bbox": [ + 207, + 397, + 213, + 407 + ], + "score": 0.81, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 396, + 505, + 409 + ], + "score": 1.0, + "content": "is considered constant for the generation of this loss, i.e., stop back-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 408, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 506, + 421 + ], + "score": 1.0, + "content": "propagating gradients. Since our downstream task is to accurately classify images, we adopt such", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 419, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 505, + 432 + ], + "score": 1.0, + "content": "gradient-blocking operation to ensure that the feature extractor will not be affected by the training", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 430, + 412, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 412, + 442 + ], + "score": 1.0, + "content": "of TNC. We extensively investigate the effectiveness of TNC in Section 4.3.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30.5 + }, + { + "type": "title", + "bbox": [ + 107, + 454, + 341, + 466 + ], + "lines": [ + { + "bbox": [ + 106, + 454, + 342, + 467 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 342, + 467 + ], + "score": 1.0, + "content": "3.3 MUTEX-BASED CONSISTENCY REGULARIZATION", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 474, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "In recent consistency-regularization based SSL methods, only samples with high-confidence pre-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 486, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 505, + 497 + ], + "score": 1.0, + "content": "dictions are leveraged to train models. However, it could lead to inefficient utilization of unlabeled", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 104, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "data, especially at the early stage of the training process. Differently, MutexMatch can also effec-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 507, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 505, + 521 + ], + "score": 1.0, + "content": "tively exploit low-confidence unlabeled samples via introducing a novel mutex-based consistency", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 283, + 532 + ], + "score": 1.0, + "content": "regularization. A high-confidence threshold", + "type": "text" + }, + { + "bbox": [ + 283, + 521, + 290, + 529 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 519, + 506, + 532 + ], + "score": 1.0, + "content": "on TPC’s predictions is defined to split the unlabeled", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 462, + 542 + ], + "score": 1.0, + "content": "samples into two portions with mutex confidence intervals, i.e., the high-confidence one", + "type": "text" + }, + { + "bbox": [ + 462, + 531, + 487, + 541 + ], + "score": 0.88, + "content": "( > \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 529, + 506, + 542 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 540, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 203, + 554 + ], + "score": 1.0, + "content": "the low-confidence one", + "type": "text" + }, + { + "bbox": [ + 203, + 541, + 227, + 552 + ], + "score": 0.87, + "content": "( \\leq \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 540, + 506, + 554 + ], + "score": 1.0, + "content": ". In the high-confidence portion, we use TPC to learn what the unla-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 551, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 506, + 565 + ], + "score": 1.0, + "content": "beled data is, while in the low-confidence portion, we employ TNC to learn what it is not, because", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 562, + 324, + 575 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 324, + 575 + ], + "score": 1.0, + "content": "it is difficult for us to obtain its real class information.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 579, + 505, + 624 + ], + "lines": [ + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 325, + 592 + ], + "score": 1.0, + "content": "On the one hand, we use weakly-augmented example", + "type": "text" + }, + { + "bbox": [ + 325, + 581, + 338, + 590 + ], + "score": 0.87, + "content": "x ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 579, + 506, + 592 + ], + "score": 1.0, + "content": "to generate pseduo-labels from TPC and", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 590, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 437, + 603 + ], + "score": 1.0, + "content": "enforce positive consistency against its corresponding strongly-augmented variant", + "type": "text" + }, + { + "bbox": [ + 437, + 591, + 448, + 601 + ], + "score": 0.86, + "content": "x ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 590, + 506, + 603 + ], + "score": 1.0, + "content": ". We can then", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 208, + 614 + ], + "score": 1.0, + "content": "obtain their predictions,", + "type": "text" + }, + { + "bbox": [ + 209, + 601, + 280, + 614 + ], + "score": 0.92, + "content": "p ^ { \\bar { w } } \\ \\stackrel { \\cdot } { = } \\ \\mathcal { P } ( \\theta ( x ^ { w } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 601, + 301, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 302, + 601, + 368, + 614 + ], + "score": 0.92, + "content": "p ^ { s } \\doteq \\mathcal { P } ( \\bar { \\theta ( } \\bar { x ^ { s } } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 601, + 393, + 614 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 394, + 602, + 479, + 614 + ], + "score": 0.91, + "content": "\\hat { p } ^ { w } = \\arg \\operatorname* { m a x } ( p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 601, + 505, + 614 + ], + "score": 1.0, + "content": ", such", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 612, + 332, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 315, + 627 + ], + "score": 1.0, + "content": "consistency can be achieved by minimizing the loss", + "type": "text" + }, + { + "bbox": [ + 315, + 614, + 327, + 625 + ], + "score": 0.88, + "content": "{ \\mathcal { L } } _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 612, + 332, + 627 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44.5 + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 628, + 394, + 662 + ], + "lines": [ + { + "bbox": [ + 216, + 628, + 394, + 662 + ], + "spans": [ + { + "bbox": [ + 216, + 628, + 394, + 662 + ], + "score": 0.95, + "content": "\\mathcal { L } _ { p } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) \\geq \\tau ) H ( \\hat { p } _ { n } ^ { w } , p _ { n } ^ { s } ) ,", + "type": "interline_equation", + "image_path": "b1552c990b4554c03dfd33150c9c16ccc4c327413d2ac430d228bec9cabea7d2.jpg" + } + ] + } + ], + "index": 47.5, + "virtual_lines": [ + { + "bbox": [ + 216, + 628, + 394, + 645.0 + ], + "spans": [], + "index": 47 + }, + { + "bbox": [ + 216, + 645.0, + 394, + 662.0 + ], + "spans": [], + "index": 48 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 664, + 505, + 699 + ], + "lines": [ + { + "bbox": [ + 106, + 664, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 135, + 678 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 665, + 209, + 677 + ], + "score": 0.9, + "content": "\\mathbb { 1 } ( \\operatorname* { m a x } ( p ^ { w } ) > \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 664, + 494, + 678 + ], + "score": 1.0, + "content": "retains the predictions whose maximum probabilities are larger than", + "type": "text" + }, + { + "bbox": [ + 495, + 668, + 501, + 675 + ], + "score": 0.73, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 664, + 505, + 678 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 675, + 505, + 688 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 505, + 688 + ], + "score": 1.0, + "content": "On the other hand, for these low-confidence samples, we enforce consistency regularization against", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 686, + 275, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 686, + 258, + 700 + ], + "score": 1.0, + "content": "TNC’s predictions by minimizing the", + "type": "text" + }, + { + "bbox": [ + 258, + 689, + 271, + 698 + ], + "score": 0.85, + "content": "{ \\mathcal { L } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 686, + 275, + 700 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 50 + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 701, + 394, + 735 + ], + "lines": [ + { + "bbox": [ + 216, + 701, + 394, + 735 + ], + "spans": [ + { + "bbox": [ + 216, + 701, + 394, + 735 + ], + "score": 0.95, + "content": "\\mathcal { L } _ { n } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( r _ { n } ^ { w } , r _ { n } ^ { s } ) ,", + "type": "interline_equation", + "image_path": "24e1cba47acce0e4aa3d9cb663ccce625aef87cb836ea2bc72c84e8a6414d468.jpg" + } + ] + } + ], + "index": 52.5, + "virtual_lines": [ + { + "bbox": [ + 216, + 701, + 394, + 718.0 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 216, + 718.0, + 394, + 735.0 + ], + "spans": [], + "index": 53 + } + ] + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 309, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 353, + 94 + ], + "lines": [ + { + "bbox": [ + 104, + 81, + 354, + 96 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 248, + 96 + ], + "score": 1.0, + "content": "predictions of TPC on labeled data", + "type": "text" + }, + { + "bbox": [ + 248, + 82, + 261, + 92 + ], + "score": 0.89, + "content": "x ^ { l b }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 81, + 354, + 96 + ], + "score": 1.0, + "content": ", calculated as follows:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 104, + 81, + 354, + 96 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 97, + 385, + 131 + ], + "lines": [ + { + "bbox": [ + 225, + 97, + 385, + 131 + ], + "spans": [ + { + "bbox": [ + 225, + 97, + 385, + 131 + ], + "score": 0.95, + "content": "\\mathcal { L } _ { s u p } = \\frac { 1 } { B } \\sum _ { n = 1 } ^ { B } H ( y _ { n } ^ { l b } , \\mathcal { P } ( \\theta ( \\alpha _ { w } ( x _ { n } ^ { l b } ) ) ) ) ,", + "type": "interline_equation", + "image_path": "94d5af6b19d4ea4eb5e473283cd7b9e924e3d1a33f3223ff5ad6886418087623.jpg" + } + ] + } + ], + "index": 1.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 97, + 385, + 114.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 225, + 114.0, + 385, + 131.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 133, + 442, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 132, + 439, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 133, + 148 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 133, + 165, + 146 + ], + "score": 0.93, + "content": "H ( p , q )", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 132, + 409, + 148 + ], + "score": 1.0, + "content": "denotes the standard cross-entropy loss between distribution", + "type": "text" + }, + { + "bbox": [ + 410, + 136, + 415, + 145 + ], + "score": 0.81, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 132, + 433, + 148 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 433, + 136, + 439, + 145 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 132, + 439, + 148 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 158, + 257, + 169 + ], + "lines": [ + { + "bbox": [ + 105, + 157, + 258, + 170 + ], + "spans": [ + { + "bbox": [ + 105, + 157, + 258, + 170 + ], + "score": 1.0, + "content": "3.2 TRUE-NEGATIVE CLASSIFIER", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 178, + 297, + 288 + ], + "lines": [ + { + "bbox": [ + 105, + 178, + 297, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 297, + 190 + ], + "score": 1.0, + "content": "In multi-class classification tasks, for a specific", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 189, + 297, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 297, + 201 + ], + "score": 1.0, + "content": "instance, it is easier to predict which class it", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 200, + 297, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 297, + 212 + ], + "score": 1.0, + "content": "does not belong to than to know which class it", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 211, + 297, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 297, + 223 + ], + "score": 1.0, + "content": "exactly is. For example, given an image of air-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 222, + 297, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 297, + 234 + ], + "score": 1.0, + "content": "plane in CIFAR-10, we can predict which class", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 233, + 297, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 265, + 245 + ], + "score": 1.0, + "content": "it does not belong to with a probability", + "type": "text" + }, + { + "bbox": [ + 266, + 234, + 286, + 244 + ], + "score": 0.85, + "content": "90 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 233, + 297, + 245 + ], + "score": 1.0, + "content": "at", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 244, + 296, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 296, + 257 + ], + "score": 1.0, + "content": "random, whereas we only have the probability", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 255, + 297, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 118, + 267 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 255, + 137, + 266 + ], + "score": 0.87, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 255, + 297, + 267 + ], + "score": 1.0, + "content": "to correctly predict it is an airplane. To", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 266, + 297, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 297, + 278 + ], + "score": 1.0, + "content": "this end, we design a True-Negative Classifier", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 277, + 297, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 297, + 289 + ], + "score": 1.0, + "content": "to predict which class it is not. Compared to", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 178, + 297, + 289 + ] + }, + { + "type": "image", + "bbox": [ + 312, + 180, + 494, + 253 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 312, + 180, + 494, + 253 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 312, + 180, + 494, + 253 + ], + "spans": [ + { + "bbox": [ + 312, + 180, + 494, + 253 + ], + "score": 0.966, + "type": "image", + "image_path": "6de18e33fa1a4e3e4c06c6e0e1140d78d074d9ef3c09a6db8bcbe63da0b747b8.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 312, + 180, + 494, + 194.6 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 312, + 194.6, + 494, + 209.2 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 312, + 209.2, + 494, + 223.79999999999998 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 312, + 223.79999999999998, + 494, + 238.39999999999998 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 312, + 238.39999999999998, + 494, + 252.99999999999997 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 350, + 264, + 459, + 276 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 349, + 263, + 459, + 277 + ], + "spans": [ + { + "bbox": [ + 349, + 263, + 459, + 277 + ], + "score": 1.0, + "content": "Figure 3: Training of TNC", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 288, + 505, + 355 + ], + "lines": [ + { + "bbox": [ + 106, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 505, + 300 + ], + "score": 1.0, + "content": "TPC, it is much easier to obtain correct labels for TNC. Thus we exploit TNC to provide more", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 299, + 506, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 506, + 312 + ], + "score": 1.0, + "content": "guidance information on unlabeled data. We then propose a mutex-based prediction consistency on", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 310, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 505, + 321 + ], + "score": 1.0, + "content": "TPC and TNC to make full use of unlabeled data, which is described in Section 3.3. The high-level", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "score": 1.0, + "content": "training process of TNC is shown in Figure 3. Unlike the standard complementary label genera-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 331, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 506, + 344 + ], + "score": 1.0, + "content": "tions (Ishida et al., 2017; Yu et al., 2018), we use the class with the lowest confidence in TPC’s", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 343, + 494, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 390, + 356 + ], + "score": 1.0, + "content": "predictions as the complementary label to train TNC. The training loss", + "type": "text" + }, + { + "bbox": [ + 390, + 343, + 411, + 356 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 343, + 494, + 356 + ], + "score": 1.0, + "content": "can be calculated as", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 287, + 506, + 356 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 199, + 358, + 411, + 392 + ], + "lines": [ + { + "bbox": [ + 199, + 358, + 411, + 392 + ], + "spans": [ + { + "bbox": [ + 199, + 358, + 411, + 392 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { s e p } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } H ( \\arg \\operatorname* { m i n } ( \\mathcal { P } ( \\theta ( x _ { n } ^ { w } ) ) , \\mathcal { N } ( \\hat { \\theta } ( x _ { n } ^ { w } ) ) ) ,", + "type": "interline_equation", + "image_path": "d2d9bd984b9451ef77afad1222366f11067fd5814b5cb4fbbd9aa85685eaf7e1.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 199, + 358, + 411, + 375.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 199, + 375.0, + 411, + 392.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 396, + 505, + 442 + ], + "lines": [ + { + "bbox": [ + 106, + 395, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 134, + 409 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 395, + 141, + 407 + ], + "score": 0.82, + "content": "\\hat { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 396, + 207, + 409 + ], + "score": 1.0, + "content": "represents that", + "type": "text" + }, + { + "bbox": [ + 207, + 397, + 213, + 407 + ], + "score": 0.81, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 396, + 505, + 409 + ], + "score": 1.0, + "content": "is considered constant for the generation of this loss, i.e., stop back-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 408, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 506, + 421 + ], + "score": 1.0, + "content": "propagating gradients. Since our downstream task is to accurately classify images, we adopt such", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 419, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 505, + 432 + ], + "score": 1.0, + "content": "gradient-blocking operation to ensure that the feature extractor will not be affected by the training", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 430, + 412, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 412, + 442 + ], + "score": 1.0, + "content": "of TNC. We extensively investigate the effectiveness of TNC in Section 4.3.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 395, + 506, + 442 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 454, + 341, + 466 + ], + "lines": [ + { + "bbox": [ + 106, + 454, + 342, + 467 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 342, + 467 + ], + "score": 1.0, + "content": "3.3 MUTEX-BASED CONSISTENCY REGULARIZATION", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 474, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "In recent consistency-regularization based SSL methods, only samples with high-confidence pre-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 486, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 505, + 497 + ], + "score": 1.0, + "content": "dictions are leveraged to train models. However, it could lead to inefficient utilization of unlabeled", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 104, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "data, especially at the early stage of the training process. Differently, MutexMatch can also effec-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 507, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 505, + 521 + ], + "score": 1.0, + "content": "tively exploit low-confidence unlabeled samples via introducing a novel mutex-based consistency", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 283, + 532 + ], + "score": 1.0, + "content": "regularization. A high-confidence threshold", + "type": "text" + }, + { + "bbox": [ + 283, + 521, + 290, + 529 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 519, + 506, + 532 + ], + "score": 1.0, + "content": "on TPC’s predictions is defined to split the unlabeled", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 462, + 542 + ], + "score": 1.0, + "content": "samples into two portions with mutex confidence intervals, i.e., the high-confidence one", + "type": "text" + }, + { + "bbox": [ + 462, + 531, + 487, + 541 + ], + "score": 0.88, + "content": "( > \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 529, + 506, + 542 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 540, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 203, + 554 + ], + "score": 1.0, + "content": "the low-confidence one", + "type": "text" + }, + { + "bbox": [ + 203, + 541, + 227, + 552 + ], + "score": 0.87, + "content": "( \\leq \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 540, + 506, + 554 + ], + "score": 1.0, + "content": ". In the high-confidence portion, we use TPC to learn what the unla-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 551, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 506, + 565 + ], + "score": 1.0, + "content": "beled data is, while in the low-confidence portion, we employ TNC to learn what it is not, because", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 562, + 324, + 575 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 324, + 575 + ], + "score": 1.0, + "content": "it is difficult for us to obtain its real class information.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 38, + "bbox_fs": [ + 104, + 476, + 506, + 575 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 579, + 505, + 624 + ], + "lines": [ + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 325, + 592 + ], + "score": 1.0, + "content": "On the one hand, we use weakly-augmented example", + "type": "text" + }, + { + "bbox": [ + 325, + 581, + 338, + 590 + ], + "score": 0.87, + "content": "x ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 579, + 506, + 592 + ], + "score": 1.0, + "content": "to generate pseduo-labels from TPC and", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 590, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 437, + 603 + ], + "score": 1.0, + "content": "enforce positive consistency against its corresponding strongly-augmented variant", + "type": "text" + }, + { + "bbox": [ + 437, + 591, + 448, + 601 + ], + "score": 0.86, + "content": "x ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 590, + 506, + 603 + ], + "score": 1.0, + "content": ". We can then", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 208, + 614 + ], + "score": 1.0, + "content": "obtain their predictions,", + "type": "text" + }, + { + "bbox": [ + 209, + 601, + 280, + 614 + ], + "score": 0.92, + "content": "p ^ { \\bar { w } } \\ \\stackrel { \\cdot } { = } \\ \\mathcal { P } ( \\theta ( x ^ { w } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 601, + 301, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 302, + 601, + 368, + 614 + ], + "score": 0.92, + "content": "p ^ { s } \\doteq \\mathcal { P } ( \\bar { \\theta ( } \\bar { x ^ { s } } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 601, + 393, + 614 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 394, + 602, + 479, + 614 + ], + "score": 0.91, + "content": "\\hat { p } ^ { w } = \\arg \\operatorname* { m a x } ( p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 601, + 505, + 614 + ], + "score": 1.0, + "content": ", such", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 612, + 332, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 315, + 627 + ], + "score": 1.0, + "content": "consistency can be achieved by minimizing the loss", + "type": "text" + }, + { + "bbox": [ + 315, + 614, + 327, + 625 + ], + "score": 0.88, + "content": "{ \\mathcal { L } } _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 612, + 332, + 627 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 579, + 506, + 627 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 628, + 394, + 662 + ], + "lines": [ + { + "bbox": [ + 216, + 628, + 394, + 662 + ], + "spans": [ + { + "bbox": [ + 216, + 628, + 394, + 662 + ], + "score": 0.95, + "content": "\\mathcal { L } _ { p } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) \\geq \\tau ) H ( \\hat { p } _ { n } ^ { w } , p _ { n } ^ { s } ) ,", + "type": "interline_equation", + "image_path": "b1552c990b4554c03dfd33150c9c16ccc4c327413d2ac430d228bec9cabea7d2.jpg" + } + ] + } + ], + "index": 47.5, + "virtual_lines": [ + { + "bbox": [ + 216, + 628, + 394, + 645.0 + ], + "spans": [], + "index": 47 + }, + { + "bbox": [ + 216, + 645.0, + 394, + 662.0 + ], + "spans": [], + "index": 48 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 664, + 505, + 699 + ], + "lines": [ + { + "bbox": [ + 106, + 664, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 135, + 678 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 665, + 209, + 677 + ], + "score": 0.9, + "content": "\\mathbb { 1 } ( \\operatorname* { m a x } ( p ^ { w } ) > \\tau )", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 664, + 494, + 678 + ], + "score": 1.0, + "content": "retains the predictions whose maximum probabilities are larger than", + "type": "text" + }, + { + "bbox": [ + 495, + 668, + 501, + 675 + ], + "score": 0.73, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 664, + 505, + 678 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 675, + 505, + 688 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 505, + 688 + ], + "score": 1.0, + "content": "On the other hand, for these low-confidence samples, we enforce consistency regularization against", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 686, + 275, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 686, + 258, + 700 + ], + "score": 1.0, + "content": "TNC’s predictions by minimizing the", + "type": "text" + }, + { + "bbox": [ + 258, + 689, + 271, + 698 + ], + "score": 0.85, + "content": "{ \\mathcal { L } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 686, + 275, + 700 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 50, + "bbox_fs": [ + 105, + 664, + 505, + 700 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 216, + 701, + 394, + 735 + ], + "lines": [ + { + "bbox": [ + 216, + 701, + 394, + 735 + ], + "spans": [ + { + "bbox": [ + 216, + 701, + 394, + 735 + ], + "score": 0.95, + "content": "\\mathcal { L } _ { n } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( r _ { n } ^ { w } , r _ { n } ^ { s } ) ,", + "type": "interline_equation", + "image_path": "24e1cba47acce0e4aa3d9cb663ccce625aef87cb836ea2bc72c84e8a6414d468.jpg" + } + ] + } + ], + "index": 52.5, + "virtual_lines": [ + { + "bbox": [ + 216, + 701, + 394, + 718.0 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 216, + 718.0, + 394, + 735.0 + ], + "spans": [], + "index": 53 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 107, + 142, + 503, + 217 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 504, + 134 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 90, + 505, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 90, + 505, + 101 + ], + "score": 1.0, + "content": "Table 1: Accuracy for CIFAR-10, CIFAR100 and SVHN averaged on 5 different folds. Results with", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 99, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 104, + 99, + 506, + 113 + ], + "score": 1.0, + "content": "∗ were reported in CoMatch (Li et al., 2020), while results with † are using our own reimplementa-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 112, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 123 + ], + "score": 1.0, + "content": "tion. Other results were reported in FixMatch (Sohn et al., 2020). Results with DA are achieved by", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 122, + 393, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 393, + 135 + ], + "score": 1.0, + "content": "combining the distribution alignment technique (Berthelot et al., 2020).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 107, + 142, + 503, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 142, + 503, + 217 + ], + "spans": [ + { + "bbox": [ + 107, + 142, + 503, + 217 + ], + "score": 0.98, + "html": "
MethodCIFAR-10CIFAR-100SVHN
10 labels20 labels40 labels80 labels200 labels400 labels2500 labels40 labels250 labels
UDA=70.95±5.93==40.72±0.8866.87±0.2247.37±20.5194.31±2.76
MixMatch27.84±10.63*52.46±11.5080.79±1.28*33.39±1.3260.06±0.3757.45±14.5396.02±0.23
ReMixMatch w. DA80.90±9.64=55.72±2.0672.57±0.3196.66±0.2097.08±0.48
FixMatch64.08±20.33†82.32±9.77*88.61±3.3592.06±0.88*38.87±2.50†51.15±1.7571.71±0.1196.04±2.1797.52±0.38
FixMatch w. DA*83.81±9.3586.98±3.4092.29±0.86=
CoMatch*69.87±11.82†87.67±8.4793.09±1.3993.97±0.6296.47 ± 1.29†97.75± 0.19†
MutexMatch78.73±11.2191.77±2.6093.49±0.2294.34±0.8140.38±2.3656.14±1.4671.80±0.2397.19±0.2697.73±0.18
", + "type": "table", + "image_path": "3d753cfb39766d9ad318e11e1f652721a22d3f3a3fd6bc47c43cbd28dd4835c5.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 107, + 142, + 503, + 167.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 107, + 167.0, + 503, + 192.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 107, + 192.0, + 503, + 217.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 236, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 135, + 252 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 237, + 208, + 250 + ], + "score": 0.92, + "content": "r ^ { w } \\ = \\ \\mathcal { N } ( \\theta ( x ^ { w } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 236, + 229, + 252 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 229, + 237, + 297, + 250 + ], + "score": 0.93, + "content": "r ^ { s } = \\mathcal { N } ( \\theta ( x ^ { s } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 236, + 455, + 252 + ], + "score": 1.0, + "content": "are the label predictions of TNC for", + "type": "text" + }, + { + "bbox": [ + 456, + 238, + 469, + 248 + ], + "score": 0.87, + "content": "x ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 236, + 489, + 252 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 490, + 238, + 501, + 248 + ], + "score": 0.85, + "content": "x ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 236, + 506, + 252 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "respectively. For the purpose of entropy minimization (Lee et al., 2013), we adopt hard pseudo-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 258, + 504, + 272 + ], + "spans": [ + { + "bbox": [ + 104, + 258, + 128, + 272 + ], + "score": 1.0, + "content": "label", + "type": "text" + }, + { + "bbox": [ + 128, + 260, + 141, + 271 + ], + "score": 0.89, + "content": "\\hat { p } ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 258, + 491, + 272 + ], + "score": 1.0, + "content": "to enforce the consistency regularization on TPC. Differently, we use soft pseudo-label", + "type": "text" + }, + { + "bbox": [ + 491, + 260, + 504, + 269 + ], + "score": 0.84, + "content": "r ^ { w }", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "score": 1.0, + "content": "for consistency regularization of TNC, so as MutexMatch can know more information of impossible", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "score": 1.0, + "content": "class for classification. We further discuss this soft-label setting in Section 5.2. The whole algorithm", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 292, + 261, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 261, + 305 + ], + "score": 1.0, + "content": "is presented in Section 1 of Appendix.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9.5 + }, + { + "type": "title", + "bbox": [ + 108, + 320, + 200, + 333 + ], + "lines": [ + { + "bbox": [ + 105, + 319, + 201, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 201, + 335 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 345, + 505, + 389 + ], + "lines": [ + { + "bbox": [ + 105, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 505, + 358 + ], + "score": 1.0, + "content": "Following Tarvainen & Valpola (2017); Sohn et al. (2020), we perform evaluation on four benchmark", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "datasets, including STL-10, CIFAR-10/100 and SVHN. We also conduct ablation studies in Section", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "5 to investigate the efficacy of MutexMatch. Other experiments, e.g., the impact of learning rate and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 379, + 345, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 345, + 391 + ], + "score": 1.0, + "content": "hyper-parameters, are shown in Section D in the Appendix.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 109, + 403, + 287, + 415 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 288, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 288, + 416 + ], + "score": 1.0, + "content": "4.1 CIFAR-10, CIFAR-100 AND SVHN", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 505, + 457 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 505, + 438 + ], + "score": 1.0, + "content": "We evaluate our method and baselines on three widely used SSL datasets: (1) CIFAR-10, consisting", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "score": 1.0, + "content": "of 50,000 images from 10 classes, (2) CIFAR-100, consisting of 50,000 images from 100 classes,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 446, + 499, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 499, + 460 + ], + "score": 1.0, + "content": "and (3) SVHN, consisting of more than 70,000 street view house number images from 10 classes.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 468, + 505, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 481 + ], + "score": 1.0, + "content": "Baselines. We introduce recent state-of-the-art SSL methods, i.e., CoMatch (Li et al., 2020),", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "FixMatch (Sohn et al., 2020) and FixMatch with distribution alignment (Berthelot et al., 2020)", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 491, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 506, + 502 + ], + "score": 1.0, + "content": "to compare with MutexMatch. Moreover, we compare our method with SSL methods such as", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 502, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 505, + 513 + ], + "score": 1.0, + "content": "UDA (Xie et al., 2020), MixMatch (Berthelot et al., 2019) and ReMixMatch (Berthelot et al., 2020).", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 523, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 535 + ], + "score": 1.0, + "content": "Settings. For all experiments, in MutexMatch, we adopt Wide ReseNet (Zagoruyko &", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 535, + 504, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 504, + 546 + ], + "score": 1.0, + "content": "Komodakis, 2016) as the backbone (WRN-28-2 for CIFAR-10, SVHN and WRN-28-8 for CIFAR-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 545, + 504, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 504, + 557 + ], + "score": 1.0, + "content": "100) following Sohn et al. (2020). In our implementation, TNC is the same two-layer MLP as TPC.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 554, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 554, + 506, + 570 + ], + "score": 1.0, + "content": "For fair comparison, We follow these baseline methods (Sohn et al., 2020; Li et al., 2020) using", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "SGD with a momentum of 0.9 and a weight decay of 0.0005 during training. Also, we train the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "model for 1024 epochs, using a learning rate of 0.03 without the decay schedule for CIFAR-10,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "and with cosine decay schedule for CIFAR-100 and SVHN. For hyper-parameters in MutexMatch,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 599, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 135, + 613 + ], + "score": 1.0, + "content": "we set", + "type": "text" + }, + { + "bbox": [ + 135, + 600, + 245, + 612 + ], + "score": 0.87, + "content": "\\tau = 0 . 9 5 , \\mu = 7 , B = 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 599, + 410, + 613 + ], + "score": 1.0, + "content": "for all experiments. Particularly, we set", + "type": "text" + }, + { + "bbox": [ + 410, + 600, + 446, + 611 + ], + "score": 0.9, + "content": "\\tau = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 599, + 505, + 613 + ], + "score": 1.0, + "content": "on CIFAR-10", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "with 80 labels and train the model with cosine decay schedule for learning rate. In our method,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 408, + 635 + ], + "score": 1.0, + "content": "RandAugment (Cubuk et al., 2020) is used for strong augmentation. Also,", + "type": "text" + }, + { + "bbox": [ + 409, + 622, + 428, + 634 + ], + "score": 0.82, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 622, + 432, + 635 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 432, + 622, + 444, + 634 + ], + "score": 0.69, + "content": "\\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 622, + 462, + 635 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 462, + 622, + 475, + 633 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 622, + 506, + 635 + ], + "score": 1.0, + "content": "are set", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "to 1 for simplicity. To reduce the influence from random data partition, we report the mean and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 644, + 385, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 385, + 656 + ], + "score": 1.0, + "content": "variance of accuracy on five different folds of labeled/unlabeled data.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 666, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "Results. Table 1 shows the comparison between MutexMatch and baselines. With only 4", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 380, + 689 + ], + "score": 1.0, + "content": "labeled data per class, MutexMatch achieves an accuracy of", + "type": "text" + }, + { + "bbox": [ + 380, + 677, + 438, + 688 + ], + "score": 0.93, + "content": "9 3 . 4 9 { \\pm } 0 . 2 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "on CIFAR-10,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 107, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 107, + 688, + 164, + 699 + ], + "score": 0.89, + "content": "5 6 . 1 4 { \\pm } 1 . 4 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 687, + 246, + 700 + ], + "score": 1.0, + "content": "on CIFAR-100 and", + "type": "text" + }, + { + "bbox": [ + 246, + 688, + 303, + 699 + ], + "score": 0.9, + "content": "9 7 . 1 9 { \\pm } 0 . 2 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "on SVHN, yielding improvement over prior SSL", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "results. Especially, we demonstrate the superiority of MutexMatch under the extremely label-scarce", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 504, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 297, + 721 + ], + "score": 1.0, + "content": "setting. e.g., achieving an average accuracy of", + "type": "text" + }, + { + "bbox": [ + 297, + 710, + 329, + 720 + ], + "score": 0.88, + "content": "9 1 . 7 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 710, + 471, + 721 + ], + "score": 1.0, + "content": "on CIFAR-10 with only 20 labels,", + "type": "text" + }, + { + "bbox": [ + 472, + 709, + 504, + 720 + ], + "score": 0.89, + "content": "4 0 . 3 8 \\%", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "on CIFAR-100 with 200 labels. In addition, details on barely supervised learning can be found in", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 107, + 142, + 503, + 217 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 504, + 134 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 90, + 505, + 101 + ], + "spans": [ + { + "bbox": [ + 106, + 90, + 505, + 101 + ], + "score": 1.0, + "content": "Table 1: Accuracy for CIFAR-10, CIFAR100 and SVHN averaged on 5 different folds. Results with", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 99, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 104, + 99, + 506, + 113 + ], + "score": 1.0, + "content": "∗ were reported in CoMatch (Li et al., 2020), while results with † are using our own reimplementa-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 112, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 123 + ], + "score": 1.0, + "content": "tion. Other results were reported in FixMatch (Sohn et al., 2020). Results with DA are achieved by", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 122, + 393, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 393, + 135 + ], + "score": 1.0, + "content": "combining the distribution alignment technique (Berthelot et al., 2020).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 107, + 142, + 503, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 142, + 503, + 217 + ], + "spans": [ + { + "bbox": [ + 107, + 142, + 503, + 217 + ], + "score": 0.98, + "html": "
MethodCIFAR-10CIFAR-100SVHN
10 labels20 labels40 labels80 labels200 labels400 labels2500 labels40 labels250 labels
UDA=70.95±5.93==40.72±0.8866.87±0.2247.37±20.5194.31±2.76
MixMatch27.84±10.63*52.46±11.5080.79±1.28*33.39±1.3260.06±0.3757.45±14.5396.02±0.23
ReMixMatch w. DA80.90±9.64=55.72±2.0672.57±0.3196.66±0.2097.08±0.48
FixMatch64.08±20.33†82.32±9.77*88.61±3.3592.06±0.88*38.87±2.50†51.15±1.7571.71±0.1196.04±2.1797.52±0.38
FixMatch w. DA*83.81±9.3586.98±3.4092.29±0.86=
CoMatch*69.87±11.82†87.67±8.4793.09±1.3993.97±0.6296.47 ± 1.29†97.75± 0.19†
MutexMatch78.73±11.2191.77±2.6093.49±0.2294.34±0.8140.38±2.3656.14±1.4671.80±0.2397.19±0.2697.73±0.18
", + "type": "table", + "image_path": "3d753cfb39766d9ad318e11e1f652721a22d3f3a3fd6bc47c43cbd28dd4835c5.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 107, + 142, + 503, + 167.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 107, + 167.0, + 503, + 192.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 107, + 192.0, + 503, + 217.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 236, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 135, + 252 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 237, + 208, + 250 + ], + "score": 0.92, + "content": "r ^ { w } \\ = \\ \\mathcal { N } ( \\theta ( x ^ { w } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 236, + 229, + 252 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 229, + 237, + 297, + 250 + ], + "score": 0.93, + "content": "r ^ { s } = \\mathcal { N } ( \\theta ( x ^ { s } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 236, + 455, + 252 + ], + "score": 1.0, + "content": "are the label predictions of TNC for", + "type": "text" + }, + { + "bbox": [ + 456, + 238, + 469, + 248 + ], + "score": 0.87, + "content": "x ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 236, + 489, + 252 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 490, + 238, + 501, + 248 + ], + "score": 0.85, + "content": "x ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 236, + 506, + 252 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "respectively. For the purpose of entropy minimization (Lee et al., 2013), we adopt hard pseudo-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 258, + 504, + 272 + ], + "spans": [ + { + "bbox": [ + 104, + 258, + 128, + 272 + ], + "score": 1.0, + "content": "label", + "type": "text" + }, + { + "bbox": [ + 128, + 260, + 141, + 271 + ], + "score": 0.89, + "content": "\\hat { p } ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 258, + 491, + 272 + ], + "score": 1.0, + "content": "to enforce the consistency regularization on TPC. Differently, we use soft pseudo-label", + "type": "text" + }, + { + "bbox": [ + 491, + 260, + 504, + 269 + ], + "score": 0.84, + "content": "r ^ { w }", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "score": 1.0, + "content": "for consistency regularization of TNC, so as MutexMatch can know more information of impossible", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 294 + ], + "score": 1.0, + "content": "class for classification. We further discuss this soft-label setting in Section 5.2. The whole algorithm", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 292, + 261, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 261, + 305 + ], + "score": 1.0, + "content": "is presented in Section 1 of Appendix.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9.5, + "bbox_fs": [ + 104, + 236, + 506, + 305 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 320, + 200, + 333 + ], + "lines": [ + { + "bbox": [ + 105, + 319, + 201, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 201, + 335 + ], + "score": 1.0, + "content": "4 EXPERIMENTS", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 345, + 505, + 389 + ], + "lines": [ + { + "bbox": [ + 105, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 505, + 358 + ], + "score": 1.0, + "content": "Following Tarvainen & Valpola (2017); Sohn et al. (2020), we perform evaluation on four benchmark", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "datasets, including STL-10, CIFAR-10/100 and SVHN. We also conduct ablation studies in Section", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "5 to investigate the efficacy of MutexMatch. Other experiments, e.g., the impact of learning rate and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 379, + 345, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 345, + 391 + ], + "score": 1.0, + "content": "hyper-parameters, are shown in Section D in the Appendix.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 346, + 505, + 391 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 403, + 287, + 415 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 288, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 288, + 416 + ], + "score": 1.0, + "content": "4.1 CIFAR-10, CIFAR-100 AND SVHN", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 505, + 457 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 505, + 438 + ], + "score": 1.0, + "content": "We evaluate our method and baselines on three widely used SSL datasets: (1) CIFAR-10, consisting", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "score": 1.0, + "content": "of 50,000 images from 10 classes, (2) CIFAR-100, consisting of 50,000 images from 100 classes,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 446, + 499, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 499, + 460 + ], + "score": 1.0, + "content": "and (3) SVHN, consisting of more than 70,000 street view house number images from 10 classes.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 423, + 506, + 460 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 468, + 505, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 481 + ], + "score": 1.0, + "content": "Baselines. We introduce recent state-of-the-art SSL methods, i.e., CoMatch (Li et al., 2020),", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "FixMatch (Sohn et al., 2020) and FixMatch with distribution alignment (Berthelot et al., 2020)", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 491, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 506, + 502 + ], + "score": 1.0, + "content": "to compare with MutexMatch. Moreover, we compare our method with SSL methods such as", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 502, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 505, + 513 + ], + "score": 1.0, + "content": "UDA (Xie et al., 2020), MixMatch (Berthelot et al., 2019) and ReMixMatch (Berthelot et al., 2020).", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 467, + 506, + 513 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 523, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 105, + 523, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 535 + ], + "score": 1.0, + "content": "Settings. For all experiments, in MutexMatch, we adopt Wide ReseNet (Zagoruyko &", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 535, + 504, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 504, + 546 + ], + "score": 1.0, + "content": "Komodakis, 2016) as the backbone (WRN-28-2 for CIFAR-10, SVHN and WRN-28-8 for CIFAR-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 545, + 504, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 504, + 557 + ], + "score": 1.0, + "content": "100) following Sohn et al. (2020). In our implementation, TNC is the same two-layer MLP as TPC.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 554, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 554, + 506, + 570 + ], + "score": 1.0, + "content": "For fair comparison, We follow these baseline methods (Sohn et al., 2020; Li et al., 2020) using", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "SGD with a momentum of 0.9 and a weight decay of 0.0005 during training. Also, we train the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "model for 1024 epochs, using a learning rate of 0.03 without the decay schedule for CIFAR-10,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "and with cosine decay schedule for CIFAR-100 and SVHN. For hyper-parameters in MutexMatch,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 599, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 135, + 613 + ], + "score": 1.0, + "content": "we set", + "type": "text" + }, + { + "bbox": [ + 135, + 600, + 245, + 612 + ], + "score": 0.87, + "content": "\\tau = 0 . 9 5 , \\mu = 7 , B = 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 599, + 410, + 613 + ], + "score": 1.0, + "content": "for all experiments. Particularly, we set", + "type": "text" + }, + { + "bbox": [ + 410, + 600, + 446, + 611 + ], + "score": 0.9, + "content": "\\tau = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 599, + 505, + 613 + ], + "score": 1.0, + "content": "on CIFAR-10", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "with 80 labels and train the model with cosine decay schedule for learning rate. In our method,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 408, + 635 + ], + "score": 1.0, + "content": "RandAugment (Cubuk et al., 2020) is used for strong augmentation. Also,", + "type": "text" + }, + { + "bbox": [ + 409, + 622, + 428, + 634 + ], + "score": 0.82, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 622, + 432, + 635 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 432, + 622, + 444, + 634 + ], + "score": 0.69, + "content": "\\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 622, + 462, + 635 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 462, + 622, + 475, + 633 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 622, + 506, + 635 + ], + "score": 1.0, + "content": "are set", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "to 1 for simplicity. To reduce the influence from random data partition, we report the mean and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 644, + 385, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 385, + 656 + ], + "score": 1.0, + "content": "variance of accuracy on five different folds of labeled/unlabeled data.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 31.5, + "bbox_fs": [ + 104, + 523, + 506, + 656 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 666, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "Results. Table 1 shows the comparison between MutexMatch and baselines. With only 4", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 380, + 689 + ], + "score": 1.0, + "content": "labeled data per class, MutexMatch achieves an accuracy of", + "type": "text" + }, + { + "bbox": [ + 380, + 677, + 438, + 688 + ], + "score": 0.93, + "content": "9 3 . 4 9 { \\pm } 0 . 2 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "on CIFAR-10,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 107, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 107, + 688, + 164, + 699 + ], + "score": 0.89, + "content": "5 6 . 1 4 { \\pm } 1 . 4 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 687, + 246, + 700 + ], + "score": 1.0, + "content": "on CIFAR-100 and", + "type": "text" + }, + { + "bbox": [ + 246, + 688, + 303, + 699 + ], + "score": 0.9, + "content": "9 7 . 1 9 { \\pm } 0 . 2 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "on SVHN, yielding improvement over prior SSL", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "results. Especially, we demonstrate the superiority of MutexMatch under the extremely label-scarce", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 504, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 297, + 721 + ], + "score": 1.0, + "content": "setting. e.g., achieving an average accuracy of", + "type": "text" + }, + { + "bbox": [ + 297, + 710, + 329, + 720 + ], + "score": 0.88, + "content": "9 1 . 7 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 710, + 471, + 721 + ], + "score": 1.0, + "content": "on CIFAR-10 with only 20 labels,", + "type": "text" + }, + { + "bbox": [ + 472, + 709, + 504, + 720 + ], + "score": 0.89, + "content": "4 0 . 3 8 \\%", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "on CIFAR-100 with 200 labels. In addition, details on barely supervised learning can be found in", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 665, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Section B of Appendix. The fewer labels will lead to the accumulation of more noise pseudo-labels", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 502, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 502, + 106 + ], + "score": 1.0, + "content": "in training, whereas MutexMatch uses all unlabeled data while introducing little error information.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 110, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "score": 1.0, + "content": "Moreover, we report additional results on CIFAR-10 with different backbone CNN-13 and more", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 120, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 120, + 506, + 135 + ], + "score": 1.0, + "content": "available labels. We compare MutexMatch with MT (Tarvainen & Valpola, 2017), ICT (Verma", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "score": 1.0, + "content": "et al., 2019), DualStudent (Ke et al., 2019) and UPS Rizve et al. (2021). We conduct experiments", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 144, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 505, + 155 + ], + "score": 1.0, + "content": "using the same setting as CIFAR-10 with 80 labels. In Table 2, we find that MutexMatch is not", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "score": 1.0, + "content": "backbone dependent, and achieves performance improvement when more labels are given, outper-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 505, + 177 + ], + "score": 1.0, + "content": "forming all baseline methods. More discussion of experimental results can be found in Section 4.3.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5 + }, + { + "type": "table", + "bbox": [ + 126, + 238, + 286, + 331 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 118, + 185, + 299, + 229 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 183, + 300, + 198 + ], + "spans": [ + { + "bbox": [ + 117, + 183, + 300, + 198 + ], + "score": 1.0, + "content": "Table 2: Accuracy on CIFAR-10 with larger", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 118, + 196, + 299, + 208 + ], + "spans": [ + { + "bbox": [ + 118, + 196, + 299, + 208 + ], + "score": 1.0, + "content": "amounts of labels and CNN-13 backbone.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 118, + 207, + 299, + 219 + ], + "spans": [ + { + "bbox": [ + 118, + 207, + 299, + 219 + ], + "score": 1.0, + "content": "Results of baseline methods are reported in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 118, + 217, + 221, + 230 + ], + "spans": [ + { + "bbox": [ + 118, + 217, + 221, + 230 + ], + "score": 1.0, + "content": "UPS (Rizve et al., 2021).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "table_body", + "bbox": [ + 126, + 238, + 286, + 331 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 238, + 286, + 331 + ], + "spans": [ + { + "bbox": [ + 126, + 238, + 286, + 331 + ], + "score": 0.977, + "html": "
CIFAR-10Method1000 labels 4000 labels
1000 labels4000 labels
MT80.96±0:5188.59±0.25
ICT84.52±0.7892.71±0.02
DualStudent85.83±0.3891.11±0.09
UPS91.82±0.1593.61±0.02
MutexMatch 93.01±0.32 94.10±0.24
", + "type": "table", + "image_path": "8feeb76c419d9c0cb9a702b8182db41c180114b5eacc238655d0d11c5a7bafe2.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 126, + 238, + 286, + 251.28571428571428 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 126, + 251.28571428571428, + 286, + 264.57142857142856 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 126, + 264.57142857142856, + 286, + 277.85714285714283 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 126, + 277.85714285714283, + 286, + 291.1428571428571 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 126, + 291.1428571428571, + 286, + 304.4285714285714 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 126, + 304.4285714285714, + 286, + 317.71428571428567 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 126, + 317.71428571428567, + 286, + 330.99999999999994 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 14.25 + }, + { + "type": "table", + "bbox": [ + 340, + 237, + 461, + 330 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 312, + 185, + 493, + 229 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 312, + 184, + 493, + 197 + ], + "spans": [ + { + "bbox": [ + 312, + 184, + 493, + 197 + ], + "score": 1.0, + "content": "Table 3: Accuracy on STL-10 averaged on 5", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 311, + 195, + 494, + 208 + ], + "spans": [ + { + "bbox": [ + 311, + 195, + 494, + 208 + ], + "score": 1.0, + "content": "pre-defined folds with ResNet-18 backbone.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 311, + 206, + 494, + 219 + ], + "spans": [ + { + "bbox": [ + 311, + 206, + 494, + 219 + ], + "score": 1.0, + "content": "Results of baseline methods are reported in", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 312, + 216, + 420, + 230 + ], + "spans": [ + { + "bbox": [ + 312, + 216, + 420, + 230 + ], + "score": 1.0, + "content": "CoMatch (Li et al., 2020).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "table_body", + "bbox": [ + 340, + 237, + 461, + 330 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 340, + 237, + 461, + 330 + ], + "spans": [ + { + "bbox": [ + 340, + 237, + 461, + 330 + ], + "score": 0.967, + "html": "
MethodSTL-10 ResNet-18
MixMatch38.02±8.29
FixMatch65.38±0.42
FixMatch w.DA66.53±0.39
CoMatch79.80±0.38
MutexMatch83.36±0.22
", + "type": "table", + "image_path": "1b98011c8c58a0a0a4355dca13eca0a8d2da1c11479ed2554baa334ff8777cc3.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 340, + 237, + 461, + 283.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 340, + 283.5, + 461, + 330.0 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 18.5 + }, + { + "type": "title", + "bbox": [ + 107, + 345, + 166, + 357 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 167, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 167, + 358 + ], + "score": 1.0, + "content": "4.2 STL-10", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 367, + 505, + 400 + ], + "lines": [ + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "score": 1.0, + "content": "STL-10 contains 10 classes of 5,000 labeled and 100,000 unlabeled images extracted from a similar", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "score": 1.0, + "content": "but broader distribution. The challenge of STL-10 lies in other unlabeled images contains out of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 389, + 490, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 490, + 401 + ], + "score": 1.0, + "content": "distribution images and this distribution shift enables us to test the robustness of SSL algorithm.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 411, + 505, + 477 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "Settings. For STL-10, we evaluate MutexMatch on the 5 pre-defined folds. Each fold con-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 421, + 505, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 434 + ], + "score": 1.0, + "content": "tains 1,000 labeled data and 100,000 unlabeled data. Therefore, we trained five models and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 432, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 445 + ], + "score": 1.0, + "content": "averaged their performance as the final result. Following Li et al. (2020), we use ResNet-18 as", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "score": 1.0, + "content": "the backbone because it consumes less computing resources than WRN-28-8 used in Sohn et al.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "score": 1.0, + "content": "(2020). We use the same hyperparameters and learning rate as CIFAR-100 in Section 4.1, and train", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 466, + 422, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 422, + 478 + ], + "score": 1.0, + "content": "the models using SGD with a momentum of 0.9 and a weight decay of 0.0005.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 483, + 504, + 537 + ], + "lines": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "Results. Table 3 shows the results, averaged on all 5 runs. In this setting, MutexMatch achieves", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 492, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 222, + 507 + ], + "score": 1.0, + "content": "accuracy improvement from", + "type": "text" + }, + { + "bbox": [ + 222, + 494, + 280, + 504 + ], + "score": 0.9, + "content": "7 9 . 8 0 { \\pm } 0 . 3 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 492, + 291, + 507 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 292, + 494, + 349, + 504 + ], + "score": 0.9, + "content": "8 3 . 3 6 { \\pm } 0 . 2 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 492, + 505, + 507 + ], + "score": 1.0, + "content": "compared with CoMatch. The perfor-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 504, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 506, + 518 + ], + "score": 1.0, + "content": "mance of MutexMatch on STL-10 is much better than that of the existing methods, showing TNC is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 515, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 506, + 528 + ], + "score": 1.0, + "content": "less sensitive to data distribution shift between labeled and unlabeled data, so that MutexMatch can", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 527, + 313, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 313, + 538 + ], + "score": 1.0, + "content": "maintain robust performance like on other datasets.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37 + }, + { + "type": "title", + "bbox": [ + 109, + 554, + 284, + 565 + ], + "lines": [ + { + "bbox": [ + 106, + 554, + 285, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 285, + 567 + ], + "score": 1.0, + "content": "4.3 EFFECTIVENESS ANALYSIS OF TNC", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 505, + 708 + ], + "lines": [ + { + "bbox": [ + 106, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "As shown in Figure 1, the TNC of MutexMatch uses exclusion method to help the model deal with", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 586, + 504, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 504, + 599 + ], + "score": 1.0, + "content": "unlabeled data with low confidence. Ideally, we think that for one class, the distribution of comple-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 598, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 610 + ], + "score": 1.0, + "content": "mentary pseudo-labels from TNC should be evenly dispersed or diverse unlike pseudo-label from", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "score": 1.0, + "content": "TPC, so MutexMatch can exclude more error classes as much as possible. We conduct experiments", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 620, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 620, + 505, + 632 + ], + "score": 1.0, + "content": "on CIFAR-10 with 40 labels using the same setting as in Section 4.1. We observe that the class", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 631, + 505, + 643 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 505, + 643 + ], + "score": 1.0, + "content": "prediction from TNC is indeed generally consistent with our hypothesis. As shown in Figure 4,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 642, + 504, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 642, + 504, + 653 + ], + "score": 1.0, + "content": "during training, for each class of CIFAR-10, the prediction of TNC is gradually dispersed to several", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 652, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 652, + 505, + 666 + ], + "score": 1.0, + "content": "classes (i.e., far away from the main diagonal of heat map), instead of gathering at a single class,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "score": 1.0, + "content": "indicating that TNC could play the role of exclusion method. On the contrary, the prediction out-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 675, + 506, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 506, + 687 + ], + "score": 1.0, + "content": "putted by TPC is gradually concentrated to the correct class (i.e., gathered to the main diagonal of", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 686, + 505, + 698 + ], + "spans": [ + { + "bbox": [ + 106, + 686, + 505, + 698 + ], + "score": 1.0, + "content": "the heat map). Note that, TNC uses soft labels for consistency regularization, which can explore", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 697, + 366, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 366, + 709 + ], + "score": 1.0, + "content": "more complementary information to help TPC classify correctly.", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 46.5 + }, + { + "type": "text", + "bbox": [ + 107, + 713, + 504, + 735 + ], + "lines": [ + { + "bbox": [ + 106, + 713, + 505, + 726 + ], + "spans": [ + { + "bbox": [ + 106, + 713, + 505, + 726 + ], + "score": 1.0, + "content": "Compared with other baseline methods shown in Table 1, MutexMatch performs better on CIFAR-", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 723, + 505, + 737 + ], + "spans": [ + { + "bbox": [ + 106, + 723, + 505, + 737 + ], + "score": 1.0, + "content": "10 with extremely scarce labels. We believe that confirmation bias (Yu et al., 2018) leads to the poor", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 53.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Section B of Appendix. The fewer labels will lead to the accumulation of more noise pseudo-labels", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 502, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 502, + 106 + ], + "score": 1.0, + "content": "in training, whereas MutexMatch uses all unlabeled data while introducing little error information.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 505, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 110, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "score": 1.0, + "content": "Moreover, we report additional results on CIFAR-10 with different backbone CNN-13 and more", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 120, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 120, + 506, + 135 + ], + "score": 1.0, + "content": "available labels. We compare MutexMatch with MT (Tarvainen & Valpola, 2017), ICT (Verma", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 131, + 506, + 145 + ], + "score": 1.0, + "content": "et al., 2019), DualStudent (Ke et al., 2019) and UPS Rizve et al. (2021). We conduct experiments", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 144, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 505, + 155 + ], + "score": 1.0, + "content": "using the same setting as CIFAR-10 with 80 labels. In Table 2, we find that MutexMatch is not", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "score": 1.0, + "content": "backbone dependent, and achieves performance improvement when more labels are given, outper-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 505, + 177 + ], + "score": 1.0, + "content": "forming all baseline methods. More discussion of experimental results can be found in Section 4.3.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 110, + 506, + 177 + ] + }, + { + "type": "table", + "bbox": [ + 126, + 238, + 286, + 331 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 118, + 185, + 299, + 229 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 183, + 300, + 198 + ], + "spans": [ + { + "bbox": [ + 117, + 183, + 300, + 198 + ], + "score": 1.0, + "content": "Table 2: Accuracy on CIFAR-10 with larger", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 118, + 196, + 299, + 208 + ], + "spans": [ + { + "bbox": [ + 118, + 196, + 299, + 208 + ], + "score": 1.0, + "content": "amounts of labels and CNN-13 backbone.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 118, + 207, + 299, + 219 + ], + "spans": [ + { + "bbox": [ + 118, + 207, + 299, + 219 + ], + "score": 1.0, + "content": "Results of baseline methods are reported in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 118, + 217, + 221, + 230 + ], + "spans": [ + { + "bbox": [ + 118, + 217, + 221, + 230 + ], + "score": 1.0, + "content": "UPS (Rizve et al., 2021).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "table_body", + "bbox": [ + 126, + 238, + 286, + 331 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 238, + 286, + 331 + ], + "spans": [ + { + "bbox": [ + 126, + 238, + 286, + 331 + ], + "score": 0.977, + "html": "
CIFAR-10Method1000 labels 4000 labels
1000 labels4000 labels
MT80.96±0:5188.59±0.25
ICT84.52±0.7892.71±0.02
DualStudent85.83±0.3891.11±0.09
UPS91.82±0.1593.61±0.02
MutexMatch 93.01±0.32 94.10±0.24
", + "type": "table", + "image_path": "8feeb76c419d9c0cb9a702b8182db41c180114b5eacc238655d0d11c5a7bafe2.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 126, + 238, + 286, + 251.28571428571428 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 126, + 251.28571428571428, + 286, + 264.57142857142856 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 126, + 264.57142857142856, + 286, + 277.85714285714283 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 126, + 277.85714285714283, + 286, + 291.1428571428571 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 126, + 291.1428571428571, + 286, + 304.4285714285714 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 126, + 304.4285714285714, + 286, + 317.71428571428567 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 126, + 317.71428571428567, + 286, + 330.99999999999994 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 14.25 + }, + { + "type": "table", + "bbox": [ + 340, + 237, + 461, + 330 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 312, + 185, + 493, + 229 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 312, + 184, + 493, + 197 + ], + "spans": [ + { + "bbox": [ + 312, + 184, + 493, + 197 + ], + "score": 1.0, + "content": "Table 3: Accuracy on STL-10 averaged on 5", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 311, + 195, + 494, + 208 + ], + "spans": [ + { + "bbox": [ + 311, + 195, + 494, + 208 + ], + "score": 1.0, + "content": "pre-defined folds with ResNet-18 backbone.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 311, + 206, + 494, + 219 + ], + "spans": [ + { + "bbox": [ + 311, + 206, + 494, + 219 + ], + "score": 1.0, + "content": "Results of baseline methods are reported in", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 312, + 216, + 420, + 230 + ], + "spans": [ + { + "bbox": [ + 312, + 216, + 420, + 230 + ], + "score": 1.0, + "content": "CoMatch (Li et al., 2020).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "table_body", + "bbox": [ + 340, + 237, + 461, + 330 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 340, + 237, + 461, + 330 + ], + "spans": [ + { + "bbox": [ + 340, + 237, + 461, + 330 + ], + "score": 0.967, + "html": "
MethodSTL-10 ResNet-18
MixMatch38.02±8.29
FixMatch65.38±0.42
FixMatch w.DA66.53±0.39
CoMatch79.80±0.38
MutexMatch83.36±0.22
", + "type": "table", + "image_path": "1b98011c8c58a0a0a4355dca13eca0a8d2da1c11479ed2554baa334ff8777cc3.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 340, + 237, + 461, + 283.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 340, + 283.5, + 461, + 330.0 + ], + "spans": [], + "index": 24 + } + ] + } + ], + "index": 18.5 + }, + { + "type": "title", + "bbox": [ + 107, + 345, + 166, + 357 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 167, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 167, + 358 + ], + "score": 1.0, + "content": "4.2 STL-10", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 367, + 505, + 400 + ], + "lines": [ + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "score": 1.0, + "content": "STL-10 contains 10 classes of 5,000 labeled and 100,000 unlabeled images extracted from a similar", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "score": 1.0, + "content": "but broader distribution. The challenge of STL-10 lies in other unlabeled images contains out of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 389, + 490, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 490, + 401 + ], + "score": 1.0, + "content": "distribution images and this distribution shift enables us to test the robustness of SSL algorithm.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 366, + 506, + 401 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 411, + 505, + 477 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "Settings. For STL-10, we evaluate MutexMatch on the 5 pre-defined folds. Each fold con-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 421, + 505, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 434 + ], + "score": 1.0, + "content": "tains 1,000 labeled data and 100,000 unlabeled data. Therefore, we trained five models and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 432, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 445 + ], + "score": 1.0, + "content": "averaged their performance as the final result. Following Li et al. (2020), we use ResNet-18 as", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "score": 1.0, + "content": "the backbone because it consumes less computing resources than WRN-28-8 used in Sohn et al.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "score": 1.0, + "content": "(2020). We use the same hyperparameters and learning rate as CIFAR-100 in Section 4.1, and train", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 466, + 422, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 422, + 478 + ], + "score": 1.0, + "content": "the models using SGD with a momentum of 0.9 and a weight decay of 0.0005.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 410, + 506, + 478 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 483, + 504, + 537 + ], + "lines": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "Results. Table 3 shows the results, averaged on all 5 runs. In this setting, MutexMatch achieves", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 492, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 222, + 507 + ], + "score": 1.0, + "content": "accuracy improvement from", + "type": "text" + }, + { + "bbox": [ + 222, + 494, + 280, + 504 + ], + "score": 0.9, + "content": "7 9 . 8 0 { \\pm } 0 . 3 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 492, + 291, + 507 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 292, + 494, + 349, + 504 + ], + "score": 0.9, + "content": "8 3 . 3 6 { \\pm } 0 . 2 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 492, + 505, + 507 + ], + "score": 1.0, + "content": "compared with CoMatch. The perfor-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 504, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 506, + 518 + ], + "score": 1.0, + "content": "mance of MutexMatch on STL-10 is much better than that of the existing methods, showing TNC is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 515, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 506, + 528 + ], + "score": 1.0, + "content": "less sensitive to data distribution shift between labeled and unlabeled data, so that MutexMatch can", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 527, + 313, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 313, + 538 + ], + "score": 1.0, + "content": "maintain robust performance like on other datasets.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 482, + 506, + 538 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 554, + 284, + 565 + ], + "lines": [ + { + "bbox": [ + 106, + 554, + 285, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 285, + 567 + ], + "score": 1.0, + "content": "4.3 EFFECTIVENESS ANALYSIS OF TNC", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 505, + 708 + ], + "lines": [ + { + "bbox": [ + 106, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "As shown in Figure 1, the TNC of MutexMatch uses exclusion method to help the model deal with", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 586, + 504, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 504, + 599 + ], + "score": 1.0, + "content": "unlabeled data with low confidence. Ideally, we think that for one class, the distribution of comple-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 598, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 610 + ], + "score": 1.0, + "content": "mentary pseudo-labels from TNC should be evenly dispersed or diverse unlike pseudo-label from", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "score": 1.0, + "content": "TPC, so MutexMatch can exclude more error classes as much as possible. We conduct experiments", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 620, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 620, + 505, + 632 + ], + "score": 1.0, + "content": "on CIFAR-10 with 40 labels using the same setting as in Section 4.1. We observe that the class", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 631, + 505, + 643 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 505, + 643 + ], + "score": 1.0, + "content": "prediction from TNC is indeed generally consistent with our hypothesis. As shown in Figure 4,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 642, + 504, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 642, + 504, + 653 + ], + "score": 1.0, + "content": "during training, for each class of CIFAR-10, the prediction of TNC is gradually dispersed to several", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 652, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 652, + 505, + 666 + ], + "score": 1.0, + "content": "classes (i.e., far away from the main diagonal of heat map), instead of gathering at a single class,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 505, + 676 + ], + "score": 1.0, + "content": "indicating that TNC could play the role of exclusion method. On the contrary, the prediction out-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 675, + 506, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 506, + 687 + ], + "score": 1.0, + "content": "putted by TPC is gradually concentrated to the correct class (i.e., gathered to the main diagonal of", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 686, + 505, + 698 + ], + "spans": [ + { + "bbox": [ + 106, + 686, + 505, + 698 + ], + "score": 1.0, + "content": "the heat map). Note that, TNC uses soft labels for consistency regularization, which can explore", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 697, + 366, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 366, + 709 + ], + "score": 1.0, + "content": "more complementary information to help TPC classify correctly.", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 46.5, + "bbox_fs": [ + 105, + 576, + 506, + 709 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 713, + 504, + 735 + ], + "lines": [ + { + "bbox": [ + 106, + 713, + 505, + 726 + ], + "spans": [ + { + "bbox": [ + 106, + 713, + 505, + 726 + ], + "score": 1.0, + "content": "Compared with other baseline methods shown in Table 1, MutexMatch performs better on CIFAR-", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 723, + 505, + 737 + ], + "spans": [ + { + "bbox": [ + 106, + 723, + 505, + 737 + ], + "score": 1.0, + "content": "10 with extremely scarce labels. We believe that confirmation bias (Yu et al., 2018) leads to the poor", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 337, + 311, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 311, + 348 + ], + "score": 1.0, + "content": "performance of other methods. Fewer labels will", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 348, + 311, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 311, + 360 + ], + "score": 1.0, + "content": "introduce more noisy pseudo-labeled examples to", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 360, + 311, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 311, + 370 + ], + "score": 1.0, + "content": "participate in the learning process. Nevertheless,", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 369, + 311, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 311, + 381 + ], + "score": 1.0, + "content": "MutexMatch utilizes the unlabeled samples with", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 381, + 311, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 311, + 392 + ], + "score": 1.0, + "content": "low confidence, in an exclusive manner by TNC,", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 391, + 311, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 311, + 403 + ], + "score": 1.0, + "content": "introducing few noisy pseudo-labels. As shown in", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 403, + 311, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 311, + 415 + ], + "score": 1.0, + "content": "Figure 5, our experiments on CIFAR-10 show Mu-", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 414, + 311, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 311, + 426 + ], + "score": 1.0, + "content": "texMatch produces more accurate pseudo-labels", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 424, + 311, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 311, + 437 + ], + "score": 1.0, + "content": "than FixMatch (Sohn et al., 2020), especially when", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 435, + 311, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 311, + 448 + ], + "score": 1.0, + "content": "there are very few labeled samples. In this figure,", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 446, + 311, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 119, + 457 + ], + "score": 0.68, + "content": "M", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 119, + 446, + 286, + 458 + ], + "score": 1.0, + "content": "indicates the results of MutexMatch and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 286, + 447, + 295, + 457 + ], + "score": 0.78, + "content": "F", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 296, + 446, + 311, + 458 + ], + "score": 1.0, + "content": "in-", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 458, + 234, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 458, + 234, + 469 + ], + "score": 1.0, + "content": "dicates the results of FixMatch.", + "type": "text", + "cross_page": true + } + ], + "index": 17 + } + ], + "index": 53.5, + "bbox_fs": [ + 106, + 713, + 505, + 737 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 87, + 503, + 279 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 87, + 503, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 87, + 503, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 87, + 503, + 279 + ], + "score": 0.974, + "type": "image", + "image_path": "c0040f1531bae28167b29b35bea5073b61fa823340a41d44c23ce2c3ec2a7714.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 87, + 503, + 151.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 151.0, + 503, + 215.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 215.0, + 503, + 279.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 291, + 505, + 325 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 290, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 183, + 305 + ], + "score": 1.0, + "content": "Figure 4: The rate", + "type": "text" + }, + { + "bbox": [ + 183, + 292, + 199, + 303 + ], + "score": 0.79, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 290, + 505, + 305 + ], + "score": 1.0, + "content": "of each class (column in heat map) in the pseudo-label and complementary", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "score": 1.0, + "content": "pseudo-label outputted by TPC and TNC respectively corresponding to each class (row in heat map)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 313, + 489, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 489, + 326 + ], + "score": 1.0, + "content": "in CIFAR-10. The darker, the higher. Results are reported in a run on CIFAR-10 with 40 labels.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 337, + 310, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 337, + 311, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 311, + 348 + ], + "score": 1.0, + "content": "performance of other methods. Fewer labels will", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 348, + 311, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 311, + 360 + ], + "score": 1.0, + "content": "introduce more noisy pseudo-labeled examples to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 360, + 311, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 311, + 370 + ], + "score": 1.0, + "content": "participate in the learning process. Nevertheless,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 369, + 311, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 311, + 381 + ], + "score": 1.0, + "content": "MutexMatch utilizes the unlabeled samples with", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 381, + 311, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 311, + 392 + ], + "score": 1.0, + "content": "low confidence, in an exclusive manner by TNC,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 391, + 311, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 311, + 403 + ], + "score": 1.0, + "content": "introducing few noisy pseudo-labels. As shown in", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 403, + 311, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 311, + 415 + ], + "score": 1.0, + "content": "Figure 5, our experiments on CIFAR-10 show Mu-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 414, + 311, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 311, + 426 + ], + "score": 1.0, + "content": "texMatch produces more accurate pseudo-labels", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 424, + 311, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 311, + 437 + ], + "score": 1.0, + "content": "than FixMatch (Sohn et al., 2020), especially when", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 435, + 311, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 311, + 448 + ], + "score": 1.0, + "content": "there are very few labeled samples. In this figure,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 446, + 311, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 119, + 457 + ], + "score": 0.68, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 446, + 286, + 458 + ], + "score": 1.0, + "content": "indicates the results of MutexMatch and", + "type": "text" + }, + { + "bbox": [ + 286, + 447, + 295, + 457 + ], + "score": 0.78, + "content": "F", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 446, + 311, + 458 + ], + "score": 1.0, + "content": "in-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 458, + 234, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 458, + 234, + 469 + ], + "score": 1.0, + "content": "dicates the results of FixMatch.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 108, + 475, + 309, + 541 + ], + "lines": [ + { + "bbox": [ + 106, + 474, + 311, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 311, + 487 + ], + "score": 1.0, + "content": "The accuracy of complementary pseudo-label is", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 486, + 311, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 311, + 497 + ], + "score": 1.0, + "content": "very crucial. An important premise for Mutex-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 496, + 310, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 310, + 508 + ], + "score": 1.0, + "content": "Match to work is that the complementary pseudo-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 507, + 311, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 311, + 519 + ], + "score": 1.0, + "content": "label outputted by TNC is easy to predict, so it will", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 518, + 310, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 310, + 530 + ], + "score": 1.0, + "content": "introduce less error information into the model.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 529, + 311, + 542 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 311, + 542 + ], + "score": 1.0, + "content": "Figure 5 shows that the complementary pseudo-", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20.5 + }, + { + "type": "image", + "bbox": [ + 323, + 339, + 499, + 481 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 323, + 339, + 499, + 481 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 323, + 339, + 499, + 481 + ], + "spans": [ + { + "bbox": [ + 323, + 339, + 499, + 481 + ], + "score": 0.966, + "type": "image", + "image_path": "bbe2a02403308f6a14471776d3983c2f8c4c669db531d3edc246903e5eaf57fe.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 323, + 339, + 499, + 351.90909090909093 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 323, + 351.90909090909093, + 499, + 364.81818181818187 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 323, + 364.81818181818187, + 499, + 377.7272727272728 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 323, + 377.7272727272728, + 499, + 390.63636363636374 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 323, + 390.63636363636374, + 499, + 403.5454545454547 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 323, + 403.5454545454547, + 499, + 416.4545454545456 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 323, + 416.4545454545456, + 499, + 429.36363636363654 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 323, + 429.36363636363654, + 499, + 442.2727272727275 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 323, + 442.2727272727275, + 499, + 455.1818181818184 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 323, + 455.1818181818184, + 499, + 468.09090909090935 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 323, + 468.09090909090935, + 499, + 481.0000000000003 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 318, + 495, + 504, + 528 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 318, + 494, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 318, + 494, + 505, + 506 + ], + "score": 1.0, + "content": "Figure 5: Accuracy of pseudo-label and com-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 317, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 317, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "plementary label on CIFAR-10 with different", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 318, + 517, + 414, + 528 + ], + "spans": [ + { + "bbox": [ + 318, + 517, + 414, + 528 + ], + "score": 1.0, + "content": "amount of labeled data.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 108, + 541, + 504, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 541, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 552 + ], + "score": 1.0, + "content": "labels outputted by TNC achieves high accuracy. Compared with the pseudo-label outputted by", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 551, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 505, + 564 + ], + "score": 1.0, + "content": "TPC, complementary label is more insensitive to the change of the number of labels. Even with", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 561, + 334, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 334, + 576 + ], + "score": 1.0, + "content": "only one label per class, it can maintain a high accuracy.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "title", + "bbox": [ + 108, + 591, + 218, + 604 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 221, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 221, + 606 + ], + "score": 1.0, + "content": "5 ABLATION STUDY", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 618, + 505, + 662 + ], + "lines": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "We conduct an extensive ablation study to verify the effectiveness of MutexMatch. The experiments", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "are mainly conducted on CIFAR-10 and SVHN using four labels per class, where MutexMatch", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 640, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 144, + 653 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + }, + { + "bbox": [ + 144, + 640, + 201, + 650 + ], + "score": 0.9, + "content": "9 3 . 4 9 { \\pm } 0 . 2 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 640, + 219, + 653 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 220, + 640, + 277, + 651 + ], + "score": 0.9, + "content": "9 7 . 1 9 { \\pm } 0 . 2 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 640, + 506, + 653 + ], + "score": 1.0, + "content": "accuracy using default setting. In the following experi-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 651, + 495, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 495, + 663 + ], + "score": 1.0, + "content": "ments, we keep the supervised loss as Equation (2) and positive consistency loss as Equation (4).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43.5 + }, + { + "type": "title", + "bbox": [ + 109, + 677, + 324, + 688 + ], + "lines": [ + { + "bbox": [ + 107, + 677, + 326, + 690 + ], + "spans": [ + { + "bbox": [ + 107, + 677, + 326, + 690 + ], + "score": 1.0, + "content": "5.1 UTILIZATION OF LOW-CONFIDENCE SAMPLES", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "In order to fairly verify the effectiveness of TNC, we use the same settings as Section 4.1. We believe", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "that the reason why the performance of MutexMatch is better than other earlier SSL algorithms is", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "that the existence of TNC enables the model to learn from all unlabeled data. For example, in", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 87, + 503, + 279 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 87, + 503, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 87, + 503, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 87, + 503, + 279 + ], + "score": 0.974, + "type": "image", + "image_path": "c0040f1531bae28167b29b35bea5073b61fa823340a41d44c23ce2c3ec2a7714.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 87, + 503, + 151.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 151.0, + 503, + 215.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 215.0, + 503, + 279.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 291, + 505, + 325 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 290, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 183, + 305 + ], + "score": 1.0, + "content": "Figure 4: The rate", + "type": "text" + }, + { + "bbox": [ + 183, + 292, + 199, + 303 + ], + "score": 0.79, + "content": "( \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 290, + 505, + 305 + ], + "score": 1.0, + "content": "of each class (column in heat map) in the pseudo-label and complementary", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 316 + ], + "score": 1.0, + "content": "pseudo-label outputted by TPC and TNC respectively corresponding to each class (row in heat map)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 313, + 489, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 489, + 326 + ], + "score": 1.0, + "content": "in CIFAR-10. The darker, the higher. Results are reported in a run on CIFAR-10 with 40 labels.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 337, + 310, + 469 + ], + "lines": [], + "index": 11.5, + "bbox_fs": [ + 105, + 337, + 311, + 469 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 108, + 475, + 309, + 541 + ], + "lines": [ + { + "bbox": [ + 106, + 474, + 311, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 311, + 487 + ], + "score": 1.0, + "content": "The accuracy of complementary pseudo-label is", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 486, + 311, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 311, + 497 + ], + "score": 1.0, + "content": "very crucial. An important premise for Mutex-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 496, + 310, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 310, + 508 + ], + "score": 1.0, + "content": "Match to work is that the complementary pseudo-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 507, + 311, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 311, + 519 + ], + "score": 1.0, + "content": "label outputted by TNC is easy to predict, so it will", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 518, + 310, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 310, + 530 + ], + "score": 1.0, + "content": "introduce less error information into the model.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 529, + 311, + 542 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 311, + 542 + ], + "score": 1.0, + "content": "Figure 5 shows that the complementary pseudo-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 541, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 552 + ], + "score": 1.0, + "content": "labels outputted by TNC achieves high accuracy. Compared with the pseudo-label outputted by", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 551, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 505, + 564 + ], + "score": 1.0, + "content": "TPC, complementary label is more insensitive to the change of the number of labels. Even with", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 561, + 334, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 334, + 576 + ], + "score": 1.0, + "content": "only one label per class, it can maintain a high accuracy.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 474, + 311, + 542 + ] + }, + { + "type": "image", + "bbox": [ + 323, + 339, + 499, + 481 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 323, + 339, + 499, + 481 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 323, + 339, + 499, + 481 + ], + "spans": [ + { + "bbox": [ + 323, + 339, + 499, + 481 + ], + "score": 0.966, + "type": "image", + "image_path": "bbe2a02403308f6a14471776d3983c2f8c4c669db531d3edc246903e5eaf57fe.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 323, + 339, + 499, + 351.90909090909093 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 323, + 351.90909090909093, + 499, + 364.81818181818187 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 323, + 364.81818181818187, + 499, + 377.7272727272728 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 323, + 377.7272727272728, + 499, + 390.63636363636374 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 323, + 390.63636363636374, + 499, + 403.5454545454547 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 323, + 403.5454545454547, + 499, + 416.4545454545456 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 323, + 416.4545454545456, + 499, + 429.36363636363654 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 323, + 429.36363636363654, + 499, + 442.2727272727275 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 323, + 442.2727272727275, + 499, + 455.1818181818184 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 323, + 455.1818181818184, + 499, + 468.09090909090935 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 323, + 468.09090909090935, + 499, + 481.0000000000003 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 318, + 495, + 504, + 528 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 318, + 494, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 318, + 494, + 505, + 506 + ], + "score": 1.0, + "content": "Figure 5: Accuracy of pseudo-label and com-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 317, + 505, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 317, + 505, + 506, + 518 + ], + "score": 1.0, + "content": "plementary label on CIFAR-10 with different", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 318, + 517, + 414, + 528 + ], + "spans": [ + { + "bbox": [ + 318, + 517, + 414, + 528 + ], + "score": 1.0, + "content": "amount of labeled data.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 108, + 541, + 504, + 574 + ], + "lines": [], + "index": 39, + "bbox_fs": [ + 105, + 541, + 505, + 576 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 591, + 218, + 604 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 221, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 221, + 606 + ], + "score": 1.0, + "content": "5 ABLATION STUDY", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 618, + 505, + 662 + ], + "lines": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "We conduct an extensive ablation study to verify the effectiveness of MutexMatch. The experiments", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "are mainly conducted on CIFAR-10 and SVHN using four labels per class, where MutexMatch", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 640, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 144, + 653 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + }, + { + "bbox": [ + 144, + 640, + 201, + 650 + ], + "score": 0.9, + "content": "9 3 . 4 9 { \\pm } 0 . 2 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 640, + 219, + 653 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 220, + 640, + 277, + 651 + ], + "score": 0.9, + "content": "9 7 . 1 9 { \\pm } 0 . 2 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 640, + 506, + 653 + ], + "score": 1.0, + "content": "accuracy using default setting. In the following experi-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 651, + 495, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 495, + 663 + ], + "score": 1.0, + "content": "ments, we keep the supervised loss as Equation (2) and positive consistency loss as Equation (4).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 618, + 506, + 663 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 677, + 324, + 688 + ], + "lines": [ + { + "bbox": [ + 107, + 677, + 326, + 690 + ], + "spans": [ + { + "bbox": [ + 107, + 677, + 326, + 690 + ], + "score": 1.0, + "content": "5.1 UTILIZATION OF LOW-CONFIDENCE SAMPLES", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "In order to fairly verify the effectiveness of TNC, we use the same settings as Section 4.1. We believe", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "that the reason why the performance of MutexMatch is better than other earlier SSL algorithms is", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "that the existence of TNC enables the model to learn from all unlabeled data. For example, in", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48, + "bbox_fs": [ + 106, + 699, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "FixMatch, with a predefined confidence threshold, the unlabeled samples whose confidence is less", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "than this threshold will not participate in the training. Therefore, we use the three most intuitive", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 405, + 117 + ], + "score": 1.0, + "content": "ways to use all the unlabeled data. We first use TPC to compute prediction", + "type": "text" + }, + { + "bbox": [ + 405, + 104, + 458, + 117 + ], + "score": 0.93, + "content": "p ^ { w } = \\mathcal { P } ( x ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "of weakly-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 268, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 213, + 127 + ], + "score": 1.0, + "content": "augmented unlabeled data", + "type": "text" + }, + { + "bbox": [ + 213, + 116, + 226, + 125 + ], + "score": 0.87, + "content": "x ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 115, + 268, + 127 + ], + "score": 1.0, + "content": "and then:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 128, + 136, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 126, + 136, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 126, + 136, + 176, + 150 + ], + "score": 1.0, + "content": "(i) We use", + "type": "text" + }, + { + "bbox": [ + 176, + 137, + 260, + 149 + ], + "score": 0.9, + "content": "\\hat { p } ^ { w } = \\arg \\operatorname* { m a x } ( p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 136, + 506, + 150 + ], + "score": 1.0, + "content": "as a hard pseudo-label, and enforce the cross-entropy loss", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 141, + 148, + 484, + 161 + ], + "spans": [ + { + "bbox": [ + 141, + 148, + 264, + 161 + ], + "score": 1.0, + "content": "against the model’s prediction", + "type": "text" + }, + { + "bbox": [ + 265, + 149, + 314, + 160 + ], + "score": 0.93, + "content": "p ^ { s } = \\mathcal { P } ( x ^ { s } )", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 148, + 468, + 161 + ], + "score": 1.0, + "content": "of strongly-augmented unlabeled data", + "type": "text" + }, + { + "bbox": [ + 469, + 149, + 479, + 158 + ], + "score": 0.87, + "content": "x ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 148, + 484, + 161 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "interline_equation", + "bbox": [ + 230, + 167, + 416, + 202 + ], + "lines": [ + { + "bbox": [ + 230, + 167, + 416, + 202 + ], + "spans": [ + { + "bbox": [ + 230, + 167, + 416, + 202 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { a b 1 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( \\hat { p } _ { n } ^ { w } , p _ { n } ^ { s } ) .", + "type": "interline_equation", + "image_path": "345ca509680ed0e7800e419cbdc53d2be960eebb39aab7bf9b30d2d997bfd455.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 230, + 167, + 416, + 184.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 230, + 184.5, + 416, + 202.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 124, + 211, + 460, + 223 + ], + "lines": [ + { + "bbox": [ + 124, + 210, + 461, + 226 + ], + "spans": [ + { + "bbox": [ + 124, + 210, + 173, + 226 + ], + "score": 1.0, + "content": "(ii) We use", + "type": "text" + }, + { + "bbox": [ + 174, + 212, + 186, + 223 + ], + "score": 0.89, + "content": "p ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 210, + 445, + 226 + ], + "score": 1.0, + "content": "as a soft pseudo-label and enforce the cross-entropy loss against", + "type": "text" + }, + { + "bbox": [ + 446, + 212, + 456, + 223 + ], + "score": 0.88, + "content": "p ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 210, + 461, + 226 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 230, + 230, + 416, + 265 + ], + "lines": [ + { + "bbox": [ + 230, + 230, + 416, + 265 + ], + "spans": [ + { + "bbox": [ + 230, + 230, + 416, + 265 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { a b 1 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( p _ { n } ^ { w } , p _ { n } ^ { s } ) .", + "type": "interline_equation", + "image_path": "27058594fad0bcbc0333840ef3f783984dd487cdae2c0151cc8c45fd8fc8b869.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 230, + 230, + 416, + 247.5 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 230, + 247.5, + 416, + 265.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 124, + 274, + 506, + 298 + ], + "lines": [ + { + "bbox": [ + 122, + 272, + 504, + 288 + ], + "spans": [ + { + "bbox": [ + 122, + 272, + 222, + 288 + ], + "score": 1.0, + "content": "(iii) We use the features", + "type": "text" + }, + { + "bbox": [ + 223, + 275, + 273, + 287 + ], + "score": 0.93, + "content": "f ^ { w } = \\theta ( x ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 272, + 456, + 288 + ], + "score": 1.0, + "content": "of weakly-augmented image and the features", + "type": "text" + }, + { + "bbox": [ + 457, + 275, + 504, + 287 + ], + "score": 0.93, + "content": "f ^ { s } = \\theta ( x ^ { s } )", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 286, + 399, + 298 + ], + "spans": [ + { + "bbox": [ + 142, + 286, + 399, + 298 + ], + "score": 1.0, + "content": "of strongly-augmented image extracted by the feature extractor:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 304, + 416, + 339 + ], + "lines": [ + { + "bbox": [ + 231, + 304, + 416, + 339 + ], + "spans": [ + { + "bbox": [ + 231, + 304, + 416, + 339 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { a b 1 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) E ( f _ { n } ^ { w } , f _ { n } ^ { s } ) ,", + "type": "interline_equation", + "image_path": "b2781ad6641286dcc02eaf9b0fe5530c1fb28ac3c546308fbb3968a0803a04a4.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 231, + 304, + 416, + 321.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 231, + 321.5, + 416, + 339.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 346, + 463, + 358 + ], + "lines": [ + { + "bbox": [ + 141, + 344, + 460, + 361 + ], + "spans": [ + { + "bbox": [ + 141, + 344, + 169, + 361 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 169, + 346, + 200, + 358 + ], + "score": 0.93, + "content": "E ( p , q )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 344, + 429, + 361 + ], + "score": 1.0, + "content": "denotes the mean squared loss between two distributions", + "type": "text" + }, + { + "bbox": [ + 430, + 349, + 437, + 358 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 344, + 454, + 361 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 454, + 349, + 460, + 358 + ], + "score": 0.77, + "content": "q", + "type": "inline_equation" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 352, + 381 + ], + "score": 1.0, + "content": "The loss given above minimized by experiments is simply", + "type": "text" + }, + { + "bbox": [ + 353, + 369, + 432, + 380 + ], + "score": 0.93, + "content": "\\mathcal { L } _ { s u p } + \\mathcal { L } _ { p } + \\mathcal { L } _ { a b 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 368, + 506, + 381 + ], + "score": 1.0, + "content": ". All models are", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 378, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 505, + 391 + ], + "score": 1.0, + "content": "trained on SVHN using four labels per class and we show the results of all experiments in Figure 6.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 390, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 334, + 402 + ], + "score": 1.0, + "content": "In this figure, the FULL indicates the setting of (i), the", + "type": "text" + }, + { + "bbox": [ + 334, + 390, + 360, + 401 + ], + "score": 0.26, + "content": "S O F T", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 390, + 505, + 402 + ], + "score": 1.0, + "content": "indicates the setting of (ii) and the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 400, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 505, + 414 + ], + "score": 1.0, + "content": "MSE indicates the setting of (iii). On this dataset, the default MutexMatch achieves an accuracy", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 412, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 118, + 424 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 412, + 175, + 423 + ], + "score": 0.92, + "content": "9 7 . 1 9 { \\pm } 0 . 2 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 412, + 505, + 424 + ], + "score": 1.0, + "content": ", outperforming all other experiments. Other ways using low-confidence samples", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 423, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 505, + 435 + ], + "score": 1.0, + "content": "will introduce more noisy pseudo-labels, resulting in the decline and instability of the accuracy of", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 434, + 376, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 376, + 446 + ], + "score": 1.0, + "content": "pseudo-label, which is not conducive to consistency regularization.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19 + }, + { + "type": "image", + "bbox": [ + 141, + 462, + 469, + 600 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 141, + 462, + 469, + 600 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 141, + 462, + 469, + 600 + ], + "spans": [ + { + "bbox": [ + 141, + 462, + 469, + 600 + ], + "score": 0.973, + "type": "image", + "image_path": "81b9511dbb909e0cbda8589a39505652619bd86c557ca4be519f147d7cd842e8.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 141, + 462, + 469, + 508.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 141, + 508.0, + 469, + 554.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 141, + 554.0, + 469, + 600.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 613, + 503, + 636 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 359, + 627 + ], + "score": 1.0, + "content": "Figure 6: The learning curve of ablation study on SVHN. The", + "type": "text" + }, + { + "bbox": [ + 359, + 615, + 366, + 624 + ], + "score": 0.47, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "-axis represents the training epoch", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 624, + 435, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 435, + 637 + ], + "score": 1.0, + "content": "and y-axis represents the test accuracy in (a) and the pseudo-label accuracy in (b).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + } + ], + "index": 25.25 + }, + { + "type": "title", + "bbox": [ + 107, + 655, + 327, + 666 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 328, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 328, + 668 + ], + "score": 1.0, + "content": "5.2 EVALUATION ON LEARNING SCHEME OF TNC", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 504, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 504, + 689 + ], + "score": 1.0, + "content": "The learning of TNC in MutexMatch is very important. In default MutexMatch, we use hard com-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 208, + 700 + ], + "score": 1.0, + "content": "plementary pseudo-label", + "type": "text" + }, + { + "bbox": [ + 208, + 687, + 286, + 699 + ], + "score": 0.92, + "content": "\\hat { q } ^ { w } = \\arg \\operatorname* { m i n } ( p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "to train TNC separately when stopping gradient back", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "propagation on the feature extractor, and enforce consistency regularization against soft pseudo-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 128, + 722 + ], + "score": 1.0, + "content": "label", + "type": "text" + }, + { + "bbox": [ + 128, + 709, + 183, + 722 + ], + "score": 0.93, + "content": "\\bar { r ^ { w } } = \\mathcal { N } ( x ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 708, + 313, + 722 + ], + "score": 1.0, + "content": "in the low-confidence portion of", + "type": "text" + }, + { + "bbox": [ + 314, + 710, + 345, + 721 + ], + "score": 0.9, + "content": "p ^ { w } \\leq \\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 708, + 506, + 722 + ], + "score": 1.0, + "content": ". In order to validate the effectiveness of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 720, + 501, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 501, + 733 + ], + "score": 1.0, + "content": "learning scheme of TNC in MutexMatch, we use three changed learning schemes for experiments:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "FixMatch, with a predefined confidence threshold, the unlabeled samples whose confidence is less", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "than this threshold will not participate in the training. Therefore, we use the three most intuitive", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 405, + 117 + ], + "score": 1.0, + "content": "ways to use all the unlabeled data. We first use TPC to compute prediction", + "type": "text" + }, + { + "bbox": [ + 405, + 104, + 458, + 117 + ], + "score": 0.93, + "content": "p ^ { w } = \\mathcal { P } ( x ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "of weakly-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 268, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 213, + 127 + ], + "score": 1.0, + "content": "augmented unlabeled data", + "type": "text" + }, + { + "bbox": [ + 213, + 116, + 226, + 125 + ], + "score": 0.87, + "content": "x ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 115, + 268, + 127 + ], + "score": 1.0, + "content": "and then:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 82, + 505, + 127 + ] + }, + { + "type": "text", + "bbox": [ + 128, + 136, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 126, + 136, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 126, + 136, + 176, + 150 + ], + "score": 1.0, + "content": "(i) We use", + "type": "text" + }, + { + "bbox": [ + 176, + 137, + 260, + 149 + ], + "score": 0.9, + "content": "\\hat { p } ^ { w } = \\arg \\operatorname* { m a x } ( p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 136, + 506, + 150 + ], + "score": 1.0, + "content": "as a hard pseudo-label, and enforce the cross-entropy loss", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 141, + 148, + 484, + 161 + ], + "spans": [ + { + "bbox": [ + 141, + 148, + 264, + 161 + ], + "score": 1.0, + "content": "against the model’s prediction", + "type": "text" + }, + { + "bbox": [ + 265, + 149, + 314, + 160 + ], + "score": 0.93, + "content": "p ^ { s } = \\mathcal { P } ( x ^ { s } )", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 148, + 468, + 161 + ], + "score": 1.0, + "content": "of strongly-augmented unlabeled data", + "type": "text" + }, + { + "bbox": [ + 469, + 149, + 479, + 158 + ], + "score": 0.87, + "content": "x ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 148, + 484, + 161 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 126, + 136, + 506, + 161 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 230, + 167, + 416, + 202 + ], + "lines": [ + { + "bbox": [ + 230, + 167, + 416, + 202 + ], + "spans": [ + { + "bbox": [ + 230, + 167, + 416, + 202 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { a b 1 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( \\hat { p } _ { n } ^ { w } , p _ { n } ^ { s } ) .", + "type": "interline_equation", + "image_path": "345ca509680ed0e7800e419cbdc53d2be960eebb39aab7bf9b30d2d997bfd455.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 230, + 167, + 416, + 184.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 230, + 184.5, + 416, + 202.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 124, + 211, + 460, + 223 + ], + "lines": [ + { + "bbox": [ + 124, + 210, + 461, + 226 + ], + "spans": [ + { + "bbox": [ + 124, + 210, + 173, + 226 + ], + "score": 1.0, + "content": "(ii) We use", + "type": "text" + }, + { + "bbox": [ + 174, + 212, + 186, + 223 + ], + "score": 0.89, + "content": "p ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 210, + 445, + 226 + ], + "score": 1.0, + "content": "as a soft pseudo-label and enforce the cross-entropy loss against", + "type": "text" + }, + { + "bbox": [ + 446, + 212, + 456, + 223 + ], + "score": 0.88, + "content": "p ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 210, + 461, + 226 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 124, + 210, + 461, + 226 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 230, + 230, + 416, + 265 + ], + "lines": [ + { + "bbox": [ + 230, + 230, + 416, + 265 + ], + "spans": [ + { + "bbox": [ + 230, + 230, + 416, + 265 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { a b 1 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( p _ { n } ^ { w } , p _ { n } ^ { s } ) .", + "type": "interline_equation", + "image_path": "27058594fad0bcbc0333840ef3f783984dd487cdae2c0151cc8c45fd8fc8b869.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 230, + 230, + 416, + 247.5 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 230, + 247.5, + 416, + 265.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 124, + 274, + 506, + 298 + ], + "lines": [ + { + "bbox": [ + 122, + 272, + 504, + 288 + ], + "spans": [ + { + "bbox": [ + 122, + 272, + 222, + 288 + ], + "score": 1.0, + "content": "(iii) We use the features", + "type": "text" + }, + { + "bbox": [ + 223, + 275, + 273, + 287 + ], + "score": 0.93, + "content": "f ^ { w } = \\theta ( x ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 272, + 456, + 288 + ], + "score": 1.0, + "content": "of weakly-augmented image and the features", + "type": "text" + }, + { + "bbox": [ + 457, + 275, + 504, + 287 + ], + "score": 0.93, + "content": "f ^ { s } = \\theta ( x ^ { s } )", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 286, + 399, + 298 + ], + "spans": [ + { + "bbox": [ + 142, + 286, + 399, + 298 + ], + "score": 1.0, + "content": "of strongly-augmented image extracted by the feature extractor:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 122, + 272, + 504, + 298 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 304, + 416, + 339 + ], + "lines": [ + { + "bbox": [ + 231, + 304, + 416, + 339 + ], + "spans": [ + { + "bbox": [ + 231, + 304, + 416, + 339 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { a b 1 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) E ( f _ { n } ^ { w } , f _ { n } ^ { s } ) ,", + "type": "interline_equation", + "image_path": "b2781ad6641286dcc02eaf9b0fe5530c1fb28ac3c546308fbb3968a0803a04a4.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 231, + 304, + 416, + 321.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 231, + 321.5, + 416, + 339.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 346, + 463, + 358 + ], + "lines": [ + { + "bbox": [ + 141, + 344, + 460, + 361 + ], + "spans": [ + { + "bbox": [ + 141, + 344, + 169, + 361 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 169, + 346, + 200, + 358 + ], + "score": 0.93, + "content": "E ( p , q )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 344, + 429, + 361 + ], + "score": 1.0, + "content": "denotes the mean squared loss between two distributions", + "type": "text" + }, + { + "bbox": [ + 430, + 349, + 437, + 358 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 344, + 454, + 361 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 454, + 349, + 460, + 358 + ], + "score": 0.77, + "content": "q", + "type": "inline_equation" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 141, + 344, + 460, + 361 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 352, + 381 + ], + "score": 1.0, + "content": "The loss given above minimized by experiments is simply", + "type": "text" + }, + { + "bbox": [ + 353, + 369, + 432, + 380 + ], + "score": 0.93, + "content": "\\mathcal { L } _ { s u p } + \\mathcal { L } _ { p } + \\mathcal { L } _ { a b 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 368, + 506, + 381 + ], + "score": 1.0, + "content": ". All models are", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 378, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 505, + 391 + ], + "score": 1.0, + "content": "trained on SVHN using four labels per class and we show the results of all experiments in Figure 6.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 390, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 334, + 402 + ], + "score": 1.0, + "content": "In this figure, the FULL indicates the setting of (i), the", + "type": "text" + }, + { + "bbox": [ + 334, + 390, + 360, + 401 + ], + "score": 0.26, + "content": "S O F T", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 390, + 505, + 402 + ], + "score": 1.0, + "content": "indicates the setting of (ii) and the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 400, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 505, + 414 + ], + "score": 1.0, + "content": "MSE indicates the setting of (iii). On this dataset, the default MutexMatch achieves an accuracy", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 412, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 118, + 424 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 412, + 175, + 423 + ], + "score": 0.92, + "content": "9 7 . 1 9 { \\pm } 0 . 2 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 412, + 505, + 424 + ], + "score": 1.0, + "content": ", outperforming all other experiments. Other ways using low-confidence samples", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 423, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 505, + 435 + ], + "score": 1.0, + "content": "will introduce more noisy pseudo-labels, resulting in the decline and instability of the accuracy of", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 434, + 376, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 376, + 446 + ], + "score": 1.0, + "content": "pseudo-label, which is not conducive to consistency regularization.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 368, + 506, + 446 + ] + }, + { + "type": "image", + "bbox": [ + 141, + 462, + 469, + 600 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 141, + 462, + 469, + 600 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 141, + 462, + 469, + 600 + ], + "spans": [ + { + "bbox": [ + 141, + 462, + 469, + 600 + ], + "score": 0.973, + "type": "image", + "image_path": "81b9511dbb909e0cbda8589a39505652619bd86c557ca4be519f147d7cd842e8.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 141, + 462, + 469, + 508.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 141, + 508.0, + 469, + 554.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 141, + 554.0, + 469, + 600.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 613, + 503, + 636 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 359, + 627 + ], + "score": 1.0, + "content": "Figure 6: The learning curve of ablation study on SVHN. The", + "type": "text" + }, + { + "bbox": [ + 359, + 615, + 366, + 624 + ], + "score": 0.47, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "-axis represents the training epoch", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 624, + 435, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 435, + 637 + ], + "score": 1.0, + "content": "and y-axis represents the test accuracy in (a) and the pseudo-label accuracy in (b).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + } + ], + "index": 25.25 + }, + { + "type": "title", + "bbox": [ + 107, + 655, + 327, + 666 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 328, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 328, + 668 + ], + "score": 1.0, + "content": "5.2 EVALUATION ON LEARNING SCHEME OF TNC", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 504, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 504, + 689 + ], + "score": 1.0, + "content": "The learning of TNC in MutexMatch is very important. In default MutexMatch, we use hard com-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 208, + 700 + ], + "score": 1.0, + "content": "plementary pseudo-label", + "type": "text" + }, + { + "bbox": [ + 208, + 687, + 286, + 699 + ], + "score": 0.92, + "content": "\\hat { q } ^ { w } = \\arg \\operatorname* { m i n } ( p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "to train TNC separately when stopping gradient back", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "propagation on the feature extractor, and enforce consistency regularization against soft pseudo-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 128, + 722 + ], + "score": 1.0, + "content": "label", + "type": "text" + }, + { + "bbox": [ + 128, + 709, + 183, + 722 + ], + "score": 0.93, + "content": "\\bar { r ^ { w } } = \\mathcal { N } ( x ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 708, + 313, + 722 + ], + "score": 1.0, + "content": "in the low-confidence portion of", + "type": "text" + }, + { + "bbox": [ + 314, + 710, + 345, + 721 + ], + "score": 0.9, + "content": "p ^ { w } \\leq \\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 708, + 506, + 722 + ], + "score": 1.0, + "content": ". In order to validate the effectiveness of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 720, + 501, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 501, + 733 + ], + "score": 1.0, + "content": "learning scheme of TNC in MutexMatch, we use three changed learning schemes for experiments:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 677, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 127, + 82, + 505, + 115 + ], + "lines": [ + { + "bbox": [ + 127, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 127, + 82, + 248, + 96 + ], + "score": 1.0, + "content": "(i) We use hard pseudo-label", + "type": "text" + }, + { + "bbox": [ + 248, + 83, + 326, + 95 + ], + "score": 0.91, + "content": "\\hat { q } ^ { w } = \\arg \\operatorname* { m i n } ( p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "to train TNC separately while stopping gra-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 142, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 142, + 94, + 250, + 106 + ], + "score": 1.0, + "content": "dient back propagation on", + "type": "text" + }, + { + "bbox": [ + 250, + 95, + 256, + 104 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 94, + 505, + 106 + ], + "score": 1.0, + "content": ", and enforce consistency regularization against hard comple-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 141, + 105, + 234, + 117 + ], + "spans": [ + { + "bbox": [ + 141, + 105, + 234, + 117 + ], + "score": 1.0, + "content": "mentary pseudo-label:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 174, + 116, + 471, + 151 + ], + "lines": [ + { + "bbox": [ + 174, + 116, + 471, + 151 + ], + "spans": [ + { + "bbox": [ + 174, + 116, + 471, + 151 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { s e p } = \\frac { 1 } { B } \\sum _ { n = 1 } ^ { B } H ( \\hat { q } _ { n } ^ { w } , r _ { n } ^ { w } ) , \\mathcal { L } _ { a b 2 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( \\hat { r } _ { n } ^ { w } , r _ { n } ^ { s } ) ,", + "type": "interline_equation", + "image_path": "fee88ee03e22a0ce7fd891453bbabd919a5221be570940ac2cb61e8433f2df7c.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 174, + 116, + 471, + 127.66666666666667 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 174, + 127.66666666666667, + 471, + 139.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 174, + 139.33333333333334, + 471, + 151.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 151, + 320, + 163 + ], + "lines": [ + { + "bbox": [ + 142, + 150, + 320, + 165 + ], + "spans": [ + { + "bbox": [ + 142, + 150, + 169, + 165 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 169, + 152, + 248, + 164 + ], + "score": 0.89, + "content": "\\hat { r } ^ { w } = \\arg \\operatorname* { m a x } ( r ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 150, + 266, + 165 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 267, + 152, + 317, + 164 + ], + "score": 0.9, + "content": "r ^ { s } = \\mathcal { N } ( { x } ^ { s } )", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 150, + 320, + 165 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 126, + 165, + 505, + 210 + ], + "lines": [ + { + "bbox": [ + 123, + 164, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 123, + 164, + 314, + 180 + ], + "score": 1.0, + "content": "(ii) We use hard complementary pseudo-label", + "type": "text" + }, + { + "bbox": [ + 314, + 166, + 327, + 177 + ], + "score": 0.89, + "content": "\\hat { \\gamma } ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 164, + 506, + 180 + ], + "score": 1.0, + "content": ", which is generated via randomly selecting", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 176, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 141, + 176, + 333, + 190 + ], + "score": 1.0, + "content": "the class without the highest confidence from", + "type": "text" + }, + { + "bbox": [ + 333, + 177, + 345, + 188 + ], + "score": 0.88, + "content": "p ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 176, + 505, + 190 + ], + "score": 1.0, + "content": "(just like the standard complementary", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 141, + 187, + 495, + 200 + ], + "score": 1.0, + "content": "label selection) to train TNC separately, while stopping gradient back propagation on", + "type": "text" + }, + { + "bbox": [ + 495, + 188, + 501, + 197 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 187, + 506, + 200 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 198, + 467, + 211 + ], + "spans": [ + { + "bbox": [ + 141, + 198, + 467, + 211 + ], + "score": 1.0, + "content": "and enforce consistency regularization against soft complementary pseudo-label:", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "interline_equation", + "bbox": [ + 166, + 210, + 464, + 245 + ], + "lines": [ + { + "bbox": [ + 166, + 210, + 464, + 245 + ], + "spans": [ + { + "bbox": [ + 166, + 210, + 464, + 245 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { s e p } = \\frac { 1 } { B } \\sum _ { n = 1 } ^ { B } H ( \\hat { \\gamma } _ { n } ^ { w } , r _ { n } ^ { w } ) , L _ { a b 2 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( r _ { n } ^ { w } , r _ { n } ^ { s } ) .", + "type": "interline_equation", + "image_path": "63551d9ac7ab886517fdf3cbdf8a21a9dfc8ad56091f8cc7d131cae3489313e3.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 166, + 210, + 464, + 221.66666666666666 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 166, + 221.66666666666666, + 464, + 233.33333333333331 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 166, + 233.33333333333331, + 464, + 244.99999999999997 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 123, + 246, + 504, + 290 + ], + "lines": [ + { + "bbox": [ + 122, + 246, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 122, + 246, + 505, + 259 + ], + "score": 1.0, + "content": "(iii) We remove the separately training part of TNC. The complementary pseudo-label for TNC", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 140, + 256, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 140, + 256, + 239, + 270 + ], + "score": 1.0, + "content": "is obtained directly by", + "type": "text" + }, + { + "bbox": [ + 240, + 258, + 333, + 269 + ], + "score": 0.89, + "content": "q ^ { w } \\ = \\ \\mathtt { N o r m } ( 1 \\ - \\ p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 256, + 363, + 270 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 364, + 257, + 399, + 269 + ], + "score": 0.83, + "content": "\\operatorname { N o r m } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 256, + 505, + 270 + ], + "score": 1.0, + "content": "is operation normalizing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 266, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 142, + 268, + 155, + 280 + ], + "score": 0.87, + "content": "q ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 266, + 208, + 282 + ], + "score": 1.0, + "content": "into interval", + "type": "text" + }, + { + "bbox": [ + 209, + 268, + 229, + 280 + ], + "score": 0.79, + "content": "[ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 266, + 505, + 282 + ], + "score": 1.0, + "content": ". We enforce consistency regularization against soft complementary", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 280, + 199, + 290 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 199, + 290 + ], + "score": 1.0, + "content": "pseudo-label:", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 289, + 416, + 324 + ], + "lines": [ + { + "bbox": [ + 231, + 289, + 416, + 324 + ], + "spans": [ + { + "bbox": [ + 231, + 289, + 416, + 324 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { a b 2 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( q _ { n } ^ { w } , r _ { n } ^ { s } ) .", + "type": "interline_equation", + "image_path": "f527d437da681eca39a369f28ad459e8def08e6a5a21f55db8ffb3a443b793e8.jpg" + } + ] + } + ], + "index": 18.5, + "virtual_lines": [ + { + "bbox": [ + 231, + 289, + 416, + 306.5 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 231, + 306.5, + 416, + 324.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 325, + 504, + 349 + ], + "lines": [ + { + "bbox": [ + 105, + 324, + 506, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 340, + 340 + ], + "score": 1.0, + "content": "The loss given above minimized by experiments is simply", + "type": "text" + }, + { + "bbox": [ + 340, + 326, + 447, + 338 + ], + "score": 0.93, + "content": "\\mathcal { L } _ { s u p } + \\mathcal { L } _ { p } + \\mathcal { L } _ { s e p } + \\mathcal { L } _ { a b 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 324, + 506, + 340 + ], + "score": 1.0, + "content": "in (i), (ii) and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 336, + 214, + 350 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 182, + 349 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { s u p } + \\mathcal { L } _ { p } + \\mathcal { L } _ { a b 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 336, + 214, + 350 + ], + "score": 1.0, + "content": "in (iii).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 106, + 353, + 505, + 453 + ], + "lines": [ + { + "bbox": [ + 106, + 353, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 505, + 367 + ], + "score": 1.0, + "content": "All models are trained on CIFAR-10 using four labels per class, and we show the results of all", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 365, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 505, + 377 + ], + "score": 1.0, + "content": "experiments in Figure 7. In the figure, the Hard-Hard indicates setting of (i), the Rand-Soft indi-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 376, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 389 + ], + "score": 1.0, + "content": "cates setting of (ii) and the Rev-Norm indicates setting of (iii). (i), (ii) and (iii) achieve accuracy", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 387, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 118, + 399 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 387, + 150, + 397 + ], + "score": 0.84, + "content": "9 0 . 5 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 387, + 155, + 399 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 155, + 387, + 187, + 397 + ], + "score": 0.84, + "content": "9 1 . 5 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 387, + 208, + 399 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 208, + 387, + 240, + 397 + ], + "score": 0.87, + "content": "9 1 . 0 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 387, + 506, + 399 + ], + "score": 1.0, + "content": "respectively. The default MutexMatch achieved an accuracy of", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 139, + 408 + ], + "score": 0.88, + "content": "9 3 . 4 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 397, + 506, + 411 + ], + "score": 1.0, + "content": "which outperforms other settings. Furthermore, experiments show that the accuracy and", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "score": 1.0, + "content": "stability of the complementary pseudo-labels which are outputted by TNC of default MutexMatch", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 419, + 504, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 504, + 433 + ], + "score": 1.0, + "content": "are dominant. TNC uses hard pseudo-label for separate training to ensure the accuracy and stability", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 430, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 443 + ], + "score": 1.0, + "content": "of complementary pseudo-label, and uses soft pseudo-label to participate in mutex-based consis-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 442, + 346, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 346, + 455 + ], + "score": 1.0, + "content": "tency regularization to exclude more potential error classes.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 26 + }, + { + "type": "image", + "bbox": [ + 113, + 466, + 497, + 581 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 466, + 497, + 581 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 466, + 497, + 581 + ], + "spans": [ + { + "bbox": [ + 113, + 466, + 497, + 581 + ], + "score": 0.971, + "type": "image", + "image_path": "8680b672c399d43c5b55e4c775ab7be58a50520aa8e62e96b9d77626f9b522b5.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 113, + 466, + 497, + 504.3333333333333 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 113, + 504.3333333333333, + 497, + 542.6666666666666 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 113, + 542.6666666666666, + 497, + 581.0 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 592, + 505, + 627 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 592, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 434, + 606 + ], + "score": 1.0, + "content": "Figure 7: The learning curve of ablation study on CIFAR-10 with 40 labels. The", + "type": "text" + }, + { + "bbox": [ + 435, + 595, + 441, + 603 + ], + "score": 0.26, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 592, + 505, + 606 + ], + "score": 1.0, + "content": "-axis represents", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "score": 1.0, + "content": "the training epoch and the y-axis represents the test accuracy in (a), the pseudo-label accuracy in (b),", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 615, + 320, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 320, + 627 + ], + "score": 1.0, + "content": "and the complementary pseudo-label accuracy in (c).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 107, + 641, + 195, + 654 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 198, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 198, + 657 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "In this paper, we propose MutexMatch, a novel SSL algorithm using a mutex-based consistency", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "regularization derived by two distinct classifiers, one is to predict “what it is” and the other is to", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 104, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "predict “what it is not”. MutexMatch can achieve superior performance on various SSL benchmarks,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "especially under label-scarce conditions. Last but not least, we validate that low-confidence samples", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "could still be well utilized in training from a novel way. We believe this usage of low-confidence", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 721, + 475, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 475, + 732 + ], + "score": 1.0, + "content": "samples could be borrowed to other semi-supervised tasks, e.g., segmentation and detection.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 127, + 82, + 505, + 115 + ], + "lines": [ + { + "bbox": [ + 127, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 127, + 82, + 248, + 96 + ], + "score": 1.0, + "content": "(i) We use hard pseudo-label", + "type": "text" + }, + { + "bbox": [ + 248, + 83, + 326, + 95 + ], + "score": 0.91, + "content": "\\hat { q } ^ { w } = \\arg \\operatorname* { m i n } ( p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "to train TNC separately while stopping gra-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 142, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 142, + 94, + 250, + 106 + ], + "score": 1.0, + "content": "dient back propagation on", + "type": "text" + }, + { + "bbox": [ + 250, + 95, + 256, + 104 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 94, + 505, + 106 + ], + "score": 1.0, + "content": ", and enforce consistency regularization against hard comple-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 141, + 105, + 234, + 117 + ], + "spans": [ + { + "bbox": [ + 141, + 105, + 234, + 117 + ], + "score": 1.0, + "content": "mentary pseudo-label:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 127, + 82, + 505, + 117 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 174, + 116, + 471, + 151 + ], + "lines": [ + { + "bbox": [ + 174, + 116, + 471, + 151 + ], + "spans": [ + { + "bbox": [ + 174, + 116, + 471, + 151 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { s e p } = \\frac { 1 } { B } \\sum _ { n = 1 } ^ { B } H ( \\hat { q } _ { n } ^ { w } , r _ { n } ^ { w } ) , \\mathcal { L } _ { a b 2 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( \\hat { r } _ { n } ^ { w } , r _ { n } ^ { s } ) ,", + "type": "interline_equation", + "image_path": "fee88ee03e22a0ce7fd891453bbabd919a5221be570940ac2cb61e8433f2df7c.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 174, + 116, + 471, + 127.66666666666667 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 174, + 127.66666666666667, + 471, + 139.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 174, + 139.33333333333334, + 471, + 151.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 151, + 320, + 163 + ], + "lines": [ + { + "bbox": [ + 142, + 150, + 320, + 165 + ], + "spans": [ + { + "bbox": [ + 142, + 150, + 169, + 165 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 169, + 152, + 248, + 164 + ], + "score": 0.89, + "content": "\\hat { r } ^ { w } = \\arg \\operatorname* { m a x } ( r ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 150, + 266, + 165 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 267, + 152, + 317, + 164 + ], + "score": 0.9, + "content": "r ^ { s } = \\mathcal { N } ( { x } ^ { s } )", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 150, + 320, + 165 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 142, + 150, + 320, + 165 + ] + }, + { + "type": "text", + "bbox": [ + 126, + 165, + 505, + 210 + ], + "lines": [ + { + "bbox": [ + 123, + 164, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 123, + 164, + 314, + 180 + ], + "score": 1.0, + "content": "(ii) We use hard complementary pseudo-label", + "type": "text" + }, + { + "bbox": [ + 314, + 166, + 327, + 177 + ], + "score": 0.89, + "content": "\\hat { \\gamma } ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 164, + 506, + 180 + ], + "score": 1.0, + "content": ", which is generated via randomly selecting", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 176, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 141, + 176, + 333, + 190 + ], + "score": 1.0, + "content": "the class without the highest confidence from", + "type": "text" + }, + { + "bbox": [ + 333, + 177, + 345, + 188 + ], + "score": 0.88, + "content": "p ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 176, + 505, + 190 + ], + "score": 1.0, + "content": "(just like the standard complementary", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 141, + 187, + 495, + 200 + ], + "score": 1.0, + "content": "label selection) to train TNC separately, while stopping gradient back propagation on", + "type": "text" + }, + { + "bbox": [ + 495, + 188, + 501, + 197 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 187, + 506, + 200 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 198, + 467, + 211 + ], + "spans": [ + { + "bbox": [ + 141, + 198, + 467, + 211 + ], + "score": 1.0, + "content": "and enforce consistency regularization against soft complementary pseudo-label:", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5, + "bbox_fs": [ + 123, + 164, + 506, + 211 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 166, + 210, + 464, + 245 + ], + "lines": [ + { + "bbox": [ + 166, + 210, + 464, + 245 + ], + "spans": [ + { + "bbox": [ + 166, + 210, + 464, + 245 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { s e p } = \\frac { 1 } { B } \\sum _ { n = 1 } ^ { B } H ( \\hat { \\gamma } _ { n } ^ { w } , r _ { n } ^ { w } ) , L _ { a b 2 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( r _ { n } ^ { w } , r _ { n } ^ { s } ) .", + "type": "interline_equation", + "image_path": "63551d9ac7ab886517fdf3cbdf8a21a9dfc8ad56091f8cc7d131cae3489313e3.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 166, + 210, + 464, + 221.66666666666666 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 166, + 221.66666666666666, + 464, + 233.33333333333331 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 166, + 233.33333333333331, + 464, + 244.99999999999997 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 123, + 246, + 504, + 290 + ], + "lines": [ + { + "bbox": [ + 122, + 246, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 122, + 246, + 505, + 259 + ], + "score": 1.0, + "content": "(iii) We remove the separately training part of TNC. The complementary pseudo-label for TNC", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 140, + 256, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 140, + 256, + 239, + 270 + ], + "score": 1.0, + "content": "is obtained directly by", + "type": "text" + }, + { + "bbox": [ + 240, + 258, + 333, + 269 + ], + "score": 0.89, + "content": "q ^ { w } \\ = \\ \\mathtt { N o r m } ( 1 \\ - \\ p ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 256, + 363, + 270 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 364, + 257, + 399, + 269 + ], + "score": 0.83, + "content": "\\operatorname { N o r m } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 256, + 505, + 270 + ], + "score": 1.0, + "content": "is operation normalizing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 266, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 142, + 268, + 155, + 280 + ], + "score": 0.87, + "content": "q ^ { w }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 266, + 208, + 282 + ], + "score": 1.0, + "content": "into interval", + "type": "text" + }, + { + "bbox": [ + 209, + 268, + 229, + 280 + ], + "score": 0.79, + "content": "[ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 266, + 505, + 282 + ], + "score": 1.0, + "content": ". We enforce consistency regularization against soft complementary", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 280, + 199, + 290 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 199, + 290 + ], + "score": 1.0, + "content": "pseudo-label:", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 122, + 246, + 505, + 290 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 289, + 416, + 324 + ], + "lines": [ + { + "bbox": [ + 231, + 289, + 416, + 324 + ], + "spans": [ + { + "bbox": [ + 231, + 289, + 416, + 324 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { a b 2 } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( q _ { n } ^ { w } , r _ { n } ^ { s } ) .", + "type": "interline_equation", + "image_path": "f527d437da681eca39a369f28ad459e8def08e6a5a21f55db8ffb3a443b793e8.jpg" + } + ] + } + ], + "index": 18.5, + "virtual_lines": [ + { + "bbox": [ + 231, + 289, + 416, + 306.5 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 231, + 306.5, + 416, + 324.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 325, + 504, + 349 + ], + "lines": [ + { + "bbox": [ + 105, + 324, + 506, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 340, + 340 + ], + "score": 1.0, + "content": "The loss given above minimized by experiments is simply", + "type": "text" + }, + { + "bbox": [ + 340, + 326, + 447, + 338 + ], + "score": 0.93, + "content": "\\mathcal { L } _ { s u p } + \\mathcal { L } _ { p } + \\mathcal { L } _ { s e p } + \\mathcal { L } _ { a b 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 324, + 506, + 340 + ], + "score": 1.0, + "content": "in (i), (ii) and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 336, + 214, + 350 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 182, + 349 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { s u p } + \\mathcal { L } _ { p } + \\mathcal { L } _ { a b 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 336, + 214, + 350 + ], + "score": 1.0, + "content": "in (iii).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 324, + 506, + 350 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 353, + 505, + 453 + ], + "lines": [ + { + "bbox": [ + 106, + 353, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 505, + 367 + ], + "score": 1.0, + "content": "All models are trained on CIFAR-10 using four labels per class, and we show the results of all", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 365, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 505, + 377 + ], + "score": 1.0, + "content": "experiments in Figure 7. In the figure, the Hard-Hard indicates setting of (i), the Rand-Soft indi-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 376, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 389 + ], + "score": 1.0, + "content": "cates setting of (ii) and the Rev-Norm indicates setting of (iii). (i), (ii) and (iii) achieve accuracy", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 387, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 118, + 399 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 387, + 150, + 397 + ], + "score": 0.84, + "content": "9 0 . 5 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 387, + 155, + 399 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 155, + 387, + 187, + 397 + ], + "score": 0.84, + "content": "9 1 . 5 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 387, + 208, + 399 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 208, + 387, + 240, + 397 + ], + "score": 0.87, + "content": "9 1 . 0 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 387, + 506, + 399 + ], + "score": 1.0, + "content": "respectively. The default MutexMatch achieved an accuracy of", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 139, + 408 + ], + "score": 0.88, + "content": "9 3 . 4 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 397, + 506, + 411 + ], + "score": 1.0, + "content": "which outperforms other settings. Furthermore, experiments show that the accuracy and", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "score": 1.0, + "content": "stability of the complementary pseudo-labels which are outputted by TNC of default MutexMatch", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 419, + 504, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 504, + 433 + ], + "score": 1.0, + "content": "are dominant. TNC uses hard pseudo-label for separate training to ensure the accuracy and stability", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 430, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 443 + ], + "score": 1.0, + "content": "of complementary pseudo-label, and uses soft pseudo-label to participate in mutex-based consis-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 442, + 346, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 346, + 455 + ], + "score": 1.0, + "content": "tency regularization to exclude more potential error classes.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 353, + 506, + 455 + ] + }, + { + "type": "image", + "bbox": [ + 113, + 466, + 497, + 581 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 466, + 497, + 581 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 466, + 497, + 581 + ], + "spans": [ + { + "bbox": [ + 113, + 466, + 497, + 581 + ], + "score": 0.971, + "type": "image", + "image_path": "8680b672c399d43c5b55e4c775ab7be58a50520aa8e62e96b9d77626f9b522b5.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 113, + 466, + 497, + 504.3333333333333 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 113, + 504.3333333333333, + 497, + 542.6666666666666 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 113, + 542.6666666666666, + 497, + 581.0 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 592, + 505, + 627 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 592, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 434, + 606 + ], + "score": 1.0, + "content": "Figure 7: The learning curve of ablation study on CIFAR-10 with 40 labels. The", + "type": "text" + }, + { + "bbox": [ + 435, + 595, + 441, + 603 + ], + "score": 0.26, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 592, + 505, + 606 + ], + "score": 1.0, + "content": "-axis represents", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 617 + ], + "score": 1.0, + "content": "the training epoch and the y-axis represents the test accuracy in (a), the pseudo-label accuracy in (b),", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 615, + 320, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 320, + 627 + ], + "score": 1.0, + "content": "and the complementary pseudo-label accuracy in (c).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 107, + 641, + 195, + 654 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 198, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 198, + 657 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "In this paper, we propose MutexMatch, a novel SSL algorithm using a mutex-based consistency", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "regularization derived by two distinct classifiers, one is to predict “what it is” and the other is to", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 104, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "predict “what it is not”. MutexMatch can achieve superior performance on various SSL benchmarks,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "especially under label-scarce conditions. Last but not least, we validate that low-confidence samples", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "could still be well utilized in training from a novel way. We believe this usage of low-confidence", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 721, + 475, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 475, + 732 + ], + "score": 1.0, + "content": "samples could be borrowed to other semi-supervised tasks, e.g., segmentation and detection.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40.5, + "bbox_fs": [ + 104, + 665, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 176, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 100, + 504, + 124 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 505, + 113 + ], + "score": 1.0, + "content": "Phil Bachman, Ouais Alsharif, and Doina Precup. Learning with pseudo-ensembles. In Advances", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 111, + 439, + 124 + ], + "spans": [ + { + "bbox": [ + 115, + 111, + 439, + 124 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems 27, volume 27, pp. 3365–3373, 2014.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 108, + 131, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 131, + 505, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 131, + 505, + 144 + ], + "score": 1.0, + "content": "David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, and Colin A.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 143, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 116, + 143, + 505, + 155 + ], + "score": 1.0, + "content": "Raffel. Mixmatch: A holistic approach to semi-supervised learning. In Advances in Neural", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 153, + 386, + 166 + ], + "spans": [ + { + "bbox": [ + 116, + 153, + 386, + 166 + ], + "score": 1.0, + "content": "Information Processing Systems, volume 32, pp. 5049–5059, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 108, + 173, + 503, + 208 + ], + "lines": [ + { + "bbox": [ + 106, + 174, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 505, + 186 + ], + "score": 1.0, + "content": "David Berthelot, Nicholas Carlini, Ekin D. Cubuk, Alex Kurakin, Kihyuk Sohn, Han Zhang, and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 116, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "Colin Raffel. Remixmatch: Semi-supervised learning with distribution matching and augmenta-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 196, + 468, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 196, + 468, + 208 + ], + "score": 1.0, + "content": "tion anchoring. In Eighth International Conference on Learning Representations, 2020.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 215, + 504, + 239 + ], + "lines": [ + { + "bbox": [ + 105, + 215, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 230 + ], + "score": 1.0, + "content": "Olivier Chapelle, Bernhard Scholkopf, and Alexander Zien. Semi-supervised learning (chapelle, o.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 227, + 503, + 240 + ], + "spans": [ + { + "bbox": [ + 115, + 227, + 503, + 240 + ], + "score": 1.0, + "content": "et al., eds.; 2006)[book reviews]. IEEE Transactions on Neural Networks, 20(3):542–542, 2009.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 246, + 505, + 281 + ], + "lines": [ + { + "bbox": [ + 105, + 246, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 505, + 260 + ], + "score": 1.0, + "content": "Ekin Dogus Cubuk, Barret Zoph, Jon Shlens, and Quoc Le. Randaugment: Practical automated", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 258, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 115, + 258, + 505, + 272 + ], + "score": 1.0, + "content": "data augmentation with a reduced search space. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 269, + 299, + 281 + ], + "spans": [ + { + "bbox": [ + 115, + 269, + 299, + 281 + ], + "score": 1.0, + "content": "Systems, volume 33, pp. 18613–18624, 2020.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 104, + 289, + 504, + 313 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 303 + ], + "score": 1.0, + "content": "Takashi Ishida, Gang Niu, Weihua Hu, and Masashi Sugiyama. Learning from complementary", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 116, + 300, + 505, + 313 + ], + "score": 1.0, + "content": "labels. In Advances in Neural Information Processing Systems, volume 30, pp. 5639–5649, 2017.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 107, + 320, + 504, + 354 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "score": 1.0, + "content": "Takashi Ishida, Gang Niu, Aditya Krishna Menon, and Masashi Sugiyama. Complementary-label", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 330, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 115, + 330, + 506, + 345 + ], + "score": 1.0, + "content": "learning for arbitrary losses and models. In International Conference on Machine Learning, pp.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 117, + 343, + 191, + 354 + ], + "spans": [ + { + "bbox": [ + 117, + 343, + 191, + 354 + ], + "score": 1.0, + "content": "2971–2980, 2018.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 362, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 376 + ], + "score": 1.0, + "content": "Zhanghan Ke, Daoye Wang, Qiong Yan, Jimmy Ren, and Rynson Lau. Dual student: Breaking the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 373, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 115, + 373, + 505, + 387 + ], + "score": 1.0, + "content": "limits of the teacher in semi-supervised learning. In IEEE International Conference on Computer", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 117, + 385, + 236, + 397 + ], + "spans": [ + { + "bbox": [ + 117, + 385, + 236, + 397 + ], + "score": 1.0, + "content": "Vision, pp. 6728–6736, 2019.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 404, + 503, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 419 + ], + "score": 1.0, + "content": "Youngdong Kim, Junho Yim, Juseung Yun, and Junmo Kim. Nlnl: Negative learning for noisy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 415, + 470, + 428 + ], + "spans": [ + { + "bbox": [ + 115, + 415, + 470, + 428 + ], + "score": 1.0, + "content": "labels. In 2019 IEEE International Conference on Computer Vision, pp. 101–110, 2019.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 504, + 459 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 450 + ], + "score": 1.0, + "content": "Samuli Laine and Timo Aila. Temporal ensembling for semi-supervised learning. arXiv preprint", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 447, + 219, + 458 + ], + "spans": [ + { + "bbox": [ + 116, + 447, + 219, + 458 + ], + "score": 1.0, + "content": "arXiv:1610.02242, 2016.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 106, + 467, + 504, + 501 + ], + "lines": [ + { + "bbox": [ + 106, + 468, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 505, + 480 + ], + "score": 1.0, + "content": "Dong-Hyun Lee et al. Pseudo-label: The simple and efficient semi-supervised learning method for", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 115, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "deep neural networks. In Workshop on challenges in representation learning, ICML, volume 3,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 114, + 489, + 177, + 501 + ], + "spans": [ + { + "bbox": [ + 114, + 489, + 177, + 501 + ], + "score": 1.0, + "content": "pp. 896, 2013.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 104, + 509, + 504, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "score": 1.0, + "content": "Junnan Li, Caiming Xiong, and Steven C. H. Hoi. Comatch: Semi-supervised learning with con-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 520, + 397, + 533 + ], + "spans": [ + { + "bbox": [ + 116, + 520, + 397, + 533 + ], + "score": 1.0, + "content": "trastive graph regularization. arXiv preprint arXiv:2011.11183, 2020.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 540, + 504, + 564 + ], + "lines": [ + { + "bbox": [ + 106, + 540, + 504, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 504, + 553 + ], + "score": 1.0, + "content": "Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts. In Fifth", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 551, + 365, + 565 + ], + "spans": [ + { + "bbox": [ + 116, + 551, + 365, + 565 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 571, + 504, + 606 + ], + "lines": [ + { + "bbox": [ + 106, + 570, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 506, + 585 + ], + "score": 1.0, + "content": "Giorgio Patrini, Alessandro Rozza, Aditya Krishna Menon, Richard Nock, and Lizhen Qu. Making", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 583, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 116, + 583, + 505, + 596 + ], + "score": 1.0, + "content": "deep neural networks robust to label noise: A loss correction approach. In IEEE Conference on", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 117, + 594, + 379, + 608 + ], + "spans": [ + { + "bbox": [ + 117, + 594, + 379, + 608 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition, pp. 2233–2241, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 613, + 503, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "score": 1.0, + "content": "Mamshad Nayeem Rizve, Kevin Duarte, Yogesh S Rawat, and Mubarak Shah. In defense of pseudo-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 624, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 115, + 624, + 505, + 639 + ], + "score": 1.0, + "content": "labeling: An uncertainty-aware pseudo-label selection framework for semi-supervised learning.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 635, + 418, + 649 + ], + "spans": [ + { + "bbox": [ + 115, + 635, + 418, + 649 + ], + "score": 1.0, + "content": "In The Ninth International Conference on Learning Representations, 2021.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 504, + 690 + ], + "lines": [ + { + "bbox": [ + 106, + 656, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 505, + 669 + ], + "score": 1.0, + "content": "Weiwei Shi, Yihong Gong, Chris Ding, Zhiheng MaXiaoyu Tao, and Nanning Zheng. Transductive", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 668, + 504, + 680 + ], + "spans": [ + { + "bbox": [ + 116, + 668, + 504, + 680 + ], + "score": 1.0, + "content": "semi-supervised deep learning using min-max features. In Proceedings of the European Confer-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 678, + 301, + 690 + ], + "spans": [ + { + "bbox": [ + 116, + 678, + 301, + 690 + ], + "score": 1.0, + "content": "ence on Computer Vision, pp. 299–315, 2018.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Kihyuk Sohn, David Berthelot, Chun-Liang Li, Zizhao Zhang, Nicholas Carlini, Ekin D. Cubuk,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 115, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "Alex Kurakin, Han Zhang, and Colin Raffel. Fixmatch: Simplifying semi-supervised learning", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 720, + 414, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 414, + 733 + ], + "score": 1.0, + "content": "with consistency and confidence. arXiv preprint arXiv:2001.07685, 2020.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 176, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 105, + 100, + 504, + 124 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 505, + 113 + ], + "score": 1.0, + "content": "Phil Bachman, Ouais Alsharif, and Doina Precup. Learning with pseudo-ensembles. In Advances", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 111, + 439, + 124 + ], + "spans": [ + { + "bbox": [ + 115, + 111, + 439, + 124 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems 27, volume 27, pp. 3365–3373, 2014.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 100, + 505, + 124 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 131, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 131, + 505, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 131, + 505, + 144 + ], + "score": 1.0, + "content": "David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, and Colin A.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 143, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 116, + 143, + 505, + 155 + ], + "score": 1.0, + "content": "Raffel. Mixmatch: A holistic approach to semi-supervised learning. In Advances in Neural", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 153, + 386, + 166 + ], + "spans": [ + { + "bbox": [ + 116, + 153, + 386, + 166 + ], + "score": 1.0, + "content": "Information Processing Systems, volume 32, pp. 5049–5059, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 131, + 505, + 166 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 173, + 503, + 208 + ], + "lines": [ + { + "bbox": [ + 106, + 174, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 505, + 186 + ], + "score": 1.0, + "content": "David Berthelot, Nicholas Carlini, Ekin D. Cubuk, Alex Kurakin, Kihyuk Sohn, Han Zhang, and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 116, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "Colin Raffel. Remixmatch: Semi-supervised learning with distribution matching and augmenta-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 196, + 468, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 196, + 468, + 208 + ], + "score": 1.0, + "content": "tion anchoring. In Eighth International Conference on Learning Representations, 2020.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 174, + 505, + 208 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 215, + 504, + 239 + ], + "lines": [ + { + "bbox": [ + 105, + 215, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 230 + ], + "score": 1.0, + "content": "Olivier Chapelle, Bernhard Scholkopf, and Alexander Zien. Semi-supervised learning (chapelle, o.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 227, + 503, + 240 + ], + "spans": [ + { + "bbox": [ + 115, + 227, + 503, + 240 + ], + "score": 1.0, + "content": "et al., eds.; 2006)[book reviews]. IEEE Transactions on Neural Networks, 20(3):542–542, 2009.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 215, + 505, + 240 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 246, + 505, + 281 + ], + "lines": [ + { + "bbox": [ + 105, + 246, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 505, + 260 + ], + "score": 1.0, + "content": "Ekin Dogus Cubuk, Barret Zoph, Jon Shlens, and Quoc Le. Randaugment: Practical automated", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 258, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 115, + 258, + 505, + 272 + ], + "score": 1.0, + "content": "data augmentation with a reduced search space. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 269, + 299, + 281 + ], + "spans": [ + { + "bbox": [ + 115, + 269, + 299, + 281 + ], + "score": 1.0, + "content": "Systems, volume 33, pp. 18613–18624, 2020.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 246, + 505, + 281 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 289, + 504, + 313 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 303 + ], + "score": 1.0, + "content": "Takashi Ishida, Gang Niu, Weihua Hu, and Masashi Sugiyama. Learning from complementary", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 116, + 300, + 505, + 313 + ], + "score": 1.0, + "content": "labels. In Advances in Neural Information Processing Systems, volume 30, pp. 5639–5649, 2017.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 288, + 505, + 313 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 320, + 504, + 354 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "score": 1.0, + "content": "Takashi Ishida, Gang Niu, Aditya Krishna Menon, and Masashi Sugiyama. Complementary-label", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 330, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 115, + 330, + 506, + 345 + ], + "score": 1.0, + "content": "learning for arbitrary losses and models. In International Conference on Machine Learning, pp.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 117, + 343, + 191, + 354 + ], + "spans": [ + { + "bbox": [ + 117, + 343, + 191, + 354 + ], + "score": 1.0, + "content": "2971–2980, 2018.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 320, + 506, + 354 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 362, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 376 + ], + "score": 1.0, + "content": "Zhanghan Ke, Daoye Wang, Qiong Yan, Jimmy Ren, and Rynson Lau. Dual student: Breaking the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 373, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 115, + 373, + 505, + 387 + ], + "score": 1.0, + "content": "limits of the teacher in semi-supervised learning. In IEEE International Conference on Computer", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 117, + 385, + 236, + 397 + ], + "spans": [ + { + "bbox": [ + 117, + 385, + 236, + 397 + ], + "score": 1.0, + "content": "Vision, pp. 6728–6736, 2019.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 106, + 362, + 505, + 397 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 404, + 503, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 419 + ], + "score": 1.0, + "content": "Youngdong Kim, Junho Yim, Juseung Yun, and Junmo Kim. Nlnl: Negative learning for noisy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 415, + 470, + 428 + ], + "spans": [ + { + "bbox": [ + 115, + 415, + 470, + 428 + ], + "score": 1.0, + "content": "labels. In 2019 IEEE International Conference on Computer Vision, pp. 101–110, 2019.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 106, + 403, + 505, + 428 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 504, + 459 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 450 + ], + "score": 1.0, + "content": "Samuli Laine and Timo Aila. Temporal ensembling for semi-supervised learning. arXiv preprint", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 447, + 219, + 458 + ], + "spans": [ + { + "bbox": [ + 116, + 447, + 219, + 458 + ], + "score": 1.0, + "content": "arXiv:1610.02242, 2016.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 434, + 506, + 458 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 467, + 504, + 501 + ], + "lines": [ + { + "bbox": [ + 106, + 468, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 505, + 480 + ], + "score": 1.0, + "content": "Dong-Hyun Lee et al. Pseudo-label: The simple and efficient semi-supervised learning method for", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 115, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "deep neural networks. In Workshop on challenges in representation learning, ICML, volume 3,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 114, + 489, + 177, + 501 + ], + "spans": [ + { + "bbox": [ + 114, + 489, + 177, + 501 + ], + "score": 1.0, + "content": "pp. 896, 2013.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 468, + 506, + 501 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 509, + 504, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 523 + ], + "score": 1.0, + "content": "Junnan Li, Caiming Xiong, and Steven C. H. Hoi. Comatch: Semi-supervised learning with con-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 520, + 397, + 533 + ], + "spans": [ + { + "bbox": [ + 116, + 520, + 397, + 533 + ], + "score": 1.0, + "content": "trastive graph regularization. arXiv preprint arXiv:2011.11183, 2020.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 509, + 505, + 533 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 540, + 504, + 564 + ], + "lines": [ + { + "bbox": [ + 106, + 540, + 504, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 504, + 553 + ], + "score": 1.0, + "content": "Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts. In Fifth", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 551, + 365, + 565 + ], + "spans": [ + { + "bbox": [ + 116, + 551, + 365, + 565 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 106, + 540, + 504, + 565 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 571, + 504, + 606 + ], + "lines": [ + { + "bbox": [ + 106, + 570, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 506, + 585 + ], + "score": 1.0, + "content": "Giorgio Patrini, Alessandro Rozza, Aditya Krishna Menon, Richard Nock, and Lizhen Qu. Making", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 583, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 116, + 583, + 505, + 596 + ], + "score": 1.0, + "content": "deep neural networks robust to label noise: A loss correction approach. In IEEE Conference on", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 117, + 594, + 379, + 608 + ], + "spans": [ + { + "bbox": [ + 117, + 594, + 379, + 608 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition, pp. 2233–2241, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 106, + 570, + 506, + 608 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 613, + 503, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "score": 1.0, + "content": "Mamshad Nayeem Rizve, Kevin Duarte, Yogesh S Rawat, and Mubarak Shah. In defense of pseudo-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 624, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 115, + 624, + 505, + 639 + ], + "score": 1.0, + "content": "labeling: An uncertainty-aware pseudo-label selection framework for semi-supervised learning.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 635, + 418, + 649 + ], + "spans": [ + { + "bbox": [ + 115, + 635, + 418, + 649 + ], + "score": 1.0, + "content": "In The Ninth International Conference on Learning Representations, 2021.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37, + "bbox_fs": [ + 106, + 614, + 505, + 649 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 504, + 690 + ], + "lines": [ + { + "bbox": [ + 106, + 656, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 505, + 669 + ], + "score": 1.0, + "content": "Weiwei Shi, Yihong Gong, Chris Ding, Zhiheng MaXiaoyu Tao, and Nanning Zheng. Transductive", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 668, + 504, + 680 + ], + "spans": [ + { + "bbox": [ + 116, + 668, + 504, + 680 + ], + "score": 1.0, + "content": "semi-supervised deep learning using min-max features. In Proceedings of the European Confer-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 678, + 301, + 690 + ], + "spans": [ + { + "bbox": [ + 116, + 678, + 301, + 690 + ], + "score": 1.0, + "content": "ence on Computer Vision, pp. 299–315, 2018.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40, + "bbox_fs": [ + 106, + 656, + 505, + 690 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Kihyuk Sohn, David Berthelot, Chun-Liang Li, Zizhao Zhang, Nicholas Carlini, Ekin D. Cubuk,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 115, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "Alex Kurakin, Han Zhang, and Colin Raffel. Fixmatch: Simplifying semi-supervised learning", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 720, + 414, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 414, + 733 + ], + "score": 1.0, + "content": "with consistency and confidence. arXiv preprint arXiv:2001.07685, 2020.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 106, + 698, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight-averaged consis-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "tency targets improve semi-supervised deep learning results. In Advances in neural information", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 114, + 105, + 222, + 116 + ], + "spans": [ + { + "bbox": [ + 114, + 105, + 222, + 116 + ], + "score": 1.0, + "content": "processing systems, 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 109, + 123, + 503, + 158 + ], + "lines": [ + { + "bbox": [ + 107, + 122, + 504, + 137 + ], + "spans": [ + { + "bbox": [ + 107, + 122, + 504, + 137 + ], + "score": 1.0, + "content": "Vikas Verma, Alex Lamb, Juho Kannala, Yoshua Bengio, and David Lopez-Paz. Interpolation con-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 135, + 505, + 147 + ], + "spans": [ + { + "bbox": [ + 116, + 135, + 505, + 147 + ], + "score": 1.0, + "content": "sistency training for semi-supervised learning. In Proceedings of the Twenty-Eighth International", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 146, + 378, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 146, + 378, + 158 + ], + "score": 1.0, + "content": "Joint Conference on Artificial Intelligence, pp. 3635–3641, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 108, + 164, + 504, + 188 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "score": 1.0, + "content": "Qizhe Xie, Zihang Dai, Eduard Hovy, Thang Luong, and Quoc Le. Unsupervised data augmentation", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 470, + 189 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 470, + 189 + ], + "score": 1.0, + "content": "for consistency training. Advances in Neural Information Processing Systems, 33, 2020.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 195, + 504, + 219 + ], + "lines": [ + { + "bbox": [ + 105, + 194, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 505, + 210 + ], + "score": 1.0, + "content": "Xiyu Yu, Tongliang Liu, Mingming Gong, and Dacheng Tao. Learning with biased complementary", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 207, + 477, + 220 + ], + "spans": [ + { + "bbox": [ + 116, + 207, + 477, + 220 + ], + "score": 1.0, + "content": "labels. In Proceedings of the European Conference on Computer Vision, pp. 69–85, 2018.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 504, + 249 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "score": 1.0, + "content": "Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In British Machine Vision", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 237, + 192, + 249 + ], + "spans": [ + { + "bbox": [ + 115, + 237, + 192, + 249 + ], + "score": 1.0, + "content": "Conference, 2016.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 504, + 280 + ], + "lines": [ + { + "bbox": [ + 105, + 256, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 505, + 270 + ], + "score": 1.0, + "content": "Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. mixup: Beyond empirical", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 267, + 355, + 280 + ], + "spans": [ + { + "bbox": [ + 115, + 267, + 355, + 280 + ], + "score": 1.0, + "content": "risk minimization. arXiv preprint arXiv:1710.09412, 2017.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 286, + 504, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 285, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 302 + ], + "score": 1.0, + "content": "Xiaojin Zhu. Semi-supervised learning. Encyclopedia of Machine Learning and Data Mining, pp.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 117, + 298, + 191, + 310 + ], + "spans": [ + { + "bbox": [ + 117, + 298, + 191, + 310 + ], + "score": 1.0, + "content": "1142–1147, 2017.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 108, + 332, + 194, + 344 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 195, + 347 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 195, + 347 + ], + "score": 1.0, + "content": "A ALGORITHM", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 108, + 365, + 258, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 259, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 259, + 379 + ], + "score": 1.0, + "content": "Algorithm 1: MutexMatch algorithm", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 97, + 379, + 473, + 589 + ], + "lines": [ + { + "bbox": [ + 100, + 371, + 466, + 405 + ], + "spans": [ + { + "bbox": [ + 100, + 371, + 204, + 405 + ], + "score": 1.0, + "content": "Input: batch of labeledfeature extractor", + "type": "text" + }, + { + "bbox": [ + 211, + 371, + 222, + 405 + ], + "score": 1.0, + "content": "ata , T", + "type": "text" + }, + { + "bbox": [ + 222, + 381, + 305, + 394 + ], + "score": 0.9, + "content": "\\mathcal { X } = \\{ ( x _ { b } ^ { l b } , y _ { b } ^ { l b } ) \\} _ { b = 1 } ^ { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 371, + 404, + 405 + ], + "score": 1.0, + "content": ", batch of unlabeled data", + "type": "text" + }, + { + "bbox": [ + 404, + 381, + 466, + 394 + ], + "score": 0.89, + "content": "\\mathcal { U } = \\{ x _ { b } ^ { u l b } \\} _ { b = 1 } ^ { \\mu B }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 205, + 393, + 283, + 403 + ], + "spans": [ + { + "bbox": [ + 205, + 393, + 211, + 402 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 393, + 245, + 403 + ], + "score": 0.7, + "content": "\\mathcal { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 394, + 283, + 402 + ], + "score": 0.53, + "content": "\\mathcal { N }", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 97, + 402, + 178, + 415 + ], + "spans": [ + { + "bbox": [ + 97, + 402, + 158, + 415 + ], + "score": 1.0, + "content": "1 for iteration", + "type": "text" + }, + { + "bbox": [ + 158, + 405, + 163, + 413 + ], + "score": 0.46, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 402, + 178, + 415 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 98, + 411, + 369, + 433 + ], + "spans": [ + { + "bbox": [ + 98, + 419, + 105, + 426 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 116, + 411, + 120, + 433 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 121, + 414, + 369, + 429 + ], + "score": 0.71, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { s u p } = \\frac { 1 } { B } \\sum _ { n = 1 } ^ { B } H ( y _ { n } ^ { l b } , \\mathcal { P } ( x _ { n } ^ { l b } ) ) } & { { } / / \\operatorname { \\it S u p e r v i s e d l o s s f o r } x ^ { l b } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 98, + 428, + 239, + 440 + ], + "spans": [ + { + "bbox": [ + 98, + 431, + 105, + 439 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 121, + 428, + 173, + 440 + ], + "score": 1.0, + "content": "for iteration", + "type": "text" + }, + { + "bbox": [ + 174, + 429, + 197, + 439 + ], + "score": 0.85, + "content": "b = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 428, + 209, + 440 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 209, + 429, + 224, + 440 + ], + "score": 0.54, + "content": "\\mu B", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 428, + 239, + 440 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 99, + 437, + 469, + 454 + ], + "spans": [ + { + "bbox": [ + 99, + 443, + 106, + 451 + ], + "score": 1.0, + "content": "4", + "type": "text" + }, + { + "bbox": [ + 136, + 439, + 228, + 452 + ], + "score": 0.7, + "content": "p _ { b } ^ { w } = \\mathcal { P } ( \\theta ( \\alpha _ { w } ( x _ { b } ^ { u l b } ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 437, + 451, + 454 + ], + "score": 1.0, + "content": "// Compute TPC’s prediction for weakly-augmented", + "type": "text" + }, + { + "bbox": [ + 451, + 441, + 469, + 451 + ], + "score": 0.73, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 99, + 449, + 469, + 467 + ], + "spans": [ + { + "bbox": [ + 99, + 456, + 106, + 464 + ], + "score": 1.0, + "content": "5", + "type": "text" + }, + { + "bbox": [ + 136, + 453, + 225, + 465 + ], + "score": 0.74, + "content": "p _ { b } ^ { s } = \\mathcal { P } ( \\theta ( \\alpha _ { s } ( x _ { b } ^ { u l b } ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 449, + 451, + 467 + ], + "score": 1.0, + "content": "// Compute TPC’s prediction for strongly-augmented", + "type": "text" + }, + { + "bbox": [ + 451, + 454, + 469, + 463 + ], + "score": 0.83, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 99, + 460, + 471, + 483 + ], + "spans": [ + { + "bbox": [ + 99, + 469, + 106, + 477 + ], + "score": 1.0, + "content": "6", + "type": "text" + }, + { + "bbox": [ + 136, + 466, + 229, + 478 + ], + "score": 0.85, + "content": "r _ { b } ^ { w } = \\mathcal { N } ( \\theta ( \\alpha _ { w } ( x _ { b } ^ { u l b } ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 460, + 453, + 483 + ], + "score": 1.0, + "content": "// Compute TNC’s prediction for weakly-augmented", + "type": "text" + }, + { + "bbox": [ + 453, + 466, + 471, + 476 + ], + "score": 0.84, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 99, + 475, + 471, + 493 + ], + "spans": [ + { + "bbox": [ + 99, + 481, + 105, + 489 + ], + "score": 1.0, + "content": "7", + "type": "text" + }, + { + "bbox": [ + 137, + 478, + 226, + 491 + ], + "score": 0.84, + "content": "r _ { b } ^ { s } = \\mathcal { N } ( \\theta ( \\alpha _ { s } ( x _ { b } ^ { u l b } ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 475, + 453, + 493 + ], + "score": 1.0, + "content": "// Compute TNC’s prediction for strongly-augmented", + "type": "text" + }, + { + "bbox": [ + 453, + 479, + 471, + 488 + ], + "score": 0.86, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 99, + 487, + 354, + 506 + ], + "spans": [ + { + "bbox": [ + 99, + 494, + 105, + 502 + ], + "score": 1.0, + "content": "8", + "type": "text" + }, + { + "bbox": [ + 137, + 492, + 217, + 504 + ], + "score": 0.64, + "content": "\\hat { p } _ { b } ^ { w } = \\arg \\operatorname* { m a x } ( p _ { b } ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 487, + 336, + 506 + ], + "score": 1.0, + "content": "// Select pseudo-labels for", + "type": "text" + }, + { + "bbox": [ + 336, + 492, + 354, + 501 + ], + "score": 0.69, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 98, + 498, + 416, + 521 + ], + "spans": [ + { + "bbox": [ + 98, + 506, + 105, + 515 + ], + "score": 1.0, + "content": "9", + "type": "text" + }, + { + "bbox": [ + 137, + 505, + 214, + 516 + ], + "score": 0.57, + "content": "\\hat { q } _ { b } ^ { w } = \\arg \\operatorname* { m i n } ( p _ { b } ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 498, + 398, + 521 + ], + "score": 1.0, + "content": "// Select complementary pseudo-labels for", + "type": "text" + }, + { + "bbox": [ + 398, + 505, + 416, + 514 + ], + "score": 0.65, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 95, + 515, + 142, + 529 + ], + "spans": [ + { + "bbox": [ + 95, + 518, + 106, + 529 + ], + "score": 1.0, + "content": "10", + "type": "text" + }, + { + "bbox": [ + 120, + 515, + 142, + 528 + ], + "score": 1.0, + "content": "end", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 95, + 524, + 452, + 546 + ], + "spans": [ + { + "bbox": [ + 95, + 531, + 105, + 543 + ], + "score": 1.0, + "content": "11", + "type": "text" + }, + { + "bbox": [ + 122, + 528, + 274, + 543 + ], + "score": 0.46, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { s e p } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } H ( \\hat { q } _ { n } ^ { w } , \\mathcal { N } ( \\hat { \\theta } ( x _ { n } ^ { w } ) ) ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 524, + 452, + 546 + ], + "score": 1.0, + "content": "// Stop back-propagating gradients on θ", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 95, + 540, + 462, + 562 + ], + "spans": [ + { + "bbox": [ + 95, + 547, + 106, + 558 + ], + "score": 1.0, + "content": "12", + "type": "text" + }, + { + "bbox": [ + 141, + 540, + 462, + 562 + ], + "score": 1.0, + "content": "= 1µB PµBn=1 1(max(pwn ) ≥ τ )H(ˆpwn , psn) // Positive consistency loss for xulb", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 95, + 558, + 461, + 577 + ], + "spans": [ + { + "bbox": [ + 95, + 562, + 106, + 573 + ], + "score": 1.0, + "content": "13", + "type": "text" + }, + { + "bbox": [ + 122, + 559, + 304, + 574 + ], + "score": 0.8, + "content": "\\begin{array} { r } { \\mathcal { L } _ { n } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( r _ { n } ^ { w } , r _ { n } ^ { s } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 558, + 443, + 577 + ], + "score": 1.0, + "content": "// Negative consistency loss for", + "type": "text" + }, + { + "bbox": [ + 444, + 562, + 461, + 571 + ], + "score": 0.38, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 32 + }, + { + "bbox": [ + 95, + 573, + 406, + 588 + ], + "spans": [ + { + "bbox": [ + 95, + 576, + 106, + 587 + ], + "score": 1.0, + "content": "14", + "type": "text" + }, + { + "bbox": [ + 120, + 573, + 151, + 588 + ], + "score": 1.0, + "content": "update", + "type": "text" + }, + { + "bbox": [ + 151, + 575, + 183, + 586 + ], + "score": 0.87, + "content": "\\theta , \\mathcal { P } , \\mathcal { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 573, + 267, + 588 + ], + "score": 1.0, + "content": "by SGD to optimise", + "type": "text" + }, + { + "bbox": [ + 267, + 575, + 406, + 587 + ], + "score": 0.87, + "content": "\\mathcal { L } _ { s u p } + \\lambda _ { s e p } \\mathcal { L } _ { s e p } + \\lambda _ { p } \\mathcal { L } _ { p } + \\lambda _ { n } \\mathcal { L } _ { n }", + "type": "inline_equation" + } + ], + "index": 33 + } + ], + "index": 25.5 + }, + { + "type": "title", + "bbox": [ + 107, + 629, + 296, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 297, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 297, + 644 + ], + "score": 1.0, + "content": "B BARELY SUPERVISED LEARNING", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "The experimental protocol of barely supervised learning (BSL) described in Sohn et al. (2020) as-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 665, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 104, + 665, + 506, + 679 + ], + "score": 1.0, + "content": "sume a limited availability (e.g., 1 or 5) of labeled data from categories of interest. In order to test", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "the performance of our method in extreme cases, we conduct experiments on CIFAR-10 with only", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "one label per class, and consider developing a simple method to use our TNC in the test phase.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "As shown in Table 4, we use five different random seeds to extract one label of each class from", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 419, + 722 + ], + "score": 1.0, + "content": "CIFAR-10, and use MutexMatch to achieve test accuracy reaching between", + "type": "text" + }, + { + "bbox": [ + 419, + 710, + 452, + 720 + ], + "score": 0.88, + "content": "6 5 . 3 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 709, + 471, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 472, + 710, + 504, + 720 + ], + "score": 0.86, + "content": "9 3 . 0 7 \\%", + "type": "inline_equation" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 721, + 504, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 171, + 732 + ], + "score": 1.0, + "content": "with a mean of", + "type": "text" + }, + { + "bbox": [ + 172, + 721, + 203, + 731 + ], + "score": 0.88, + "content": "7 8 . 7 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 721, + 471, + 732 + ], + "score": 1.0, + "content": ". Compared with FixMatch (Sohn et al., 2020) reaching between", + "type": "text" + }, + { + "bbox": [ + 471, + 721, + 504, + 731 + ], + "score": 0.85, + "content": "4 8 . 5 8 \\%", + "type": "inline_equation" + } + ], + "index": 41 + } + ], + "index": 38 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight-averaged consis-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "tency targets improve semi-supervised deep learning results. In Advances in neural information", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 114, + 105, + 222, + 116 + ], + "spans": [ + { + "bbox": [ + 114, + 105, + 222, + 116 + ], + "score": 1.0, + "content": "processing systems, 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 82, + 505, + 116 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 123, + 503, + 158 + ], + "lines": [ + { + "bbox": [ + 107, + 122, + 504, + 137 + ], + "spans": [ + { + "bbox": [ + 107, + 122, + 504, + 137 + ], + "score": 1.0, + "content": "Vikas Verma, Alex Lamb, Juho Kannala, Yoshua Bengio, and David Lopez-Paz. Interpolation con-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 135, + 505, + 147 + ], + "spans": [ + { + "bbox": [ + 116, + 135, + 505, + 147 + ], + "score": 1.0, + "content": "sistency training for semi-supervised learning. In Proceedings of the Twenty-Eighth International", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 146, + 378, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 146, + 378, + 158 + ], + "score": 1.0, + "content": "Joint Conference on Artificial Intelligence, pp. 3635–3641, 2019.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 107, + 122, + 505, + 158 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 164, + 504, + 188 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "score": 1.0, + "content": "Qizhe Xie, Zihang Dai, Eduard Hovy, Thang Luong, and Quoc Le. Unsupervised data augmentation", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 470, + 189 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 470, + 189 + ], + "score": 1.0, + "content": "for consistency training. Advances in Neural Information Processing Systems, 33, 2020.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 164, + 505, + 189 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 195, + 504, + 219 + ], + "lines": [ + { + "bbox": [ + 105, + 194, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 505, + 210 + ], + "score": 1.0, + "content": "Xiyu Yu, Tongliang Liu, Mingming Gong, and Dacheng Tao. Learning with biased complementary", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 207, + 477, + 220 + ], + "spans": [ + { + "bbox": [ + 116, + 207, + 477, + 220 + ], + "score": 1.0, + "content": "labels. In Proceedings of the European Conference on Computer Vision, pp. 69–85, 2018.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 194, + 505, + 220 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 504, + 249 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "score": 1.0, + "content": "Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In British Machine Vision", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 237, + 192, + 249 + ], + "spans": [ + { + "bbox": [ + 115, + 237, + 192, + 249 + ], + "score": 1.0, + "content": "Conference, 2016.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 225, + 506, + 249 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 504, + 280 + ], + "lines": [ + { + "bbox": [ + 105, + 256, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 505, + 270 + ], + "score": 1.0, + "content": "Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. mixup: Beyond empirical", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 267, + 355, + 280 + ], + "spans": [ + { + "bbox": [ + 115, + 267, + 355, + 280 + ], + "score": 1.0, + "content": "risk minimization. arXiv preprint arXiv:1710.09412, 2017.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 256, + 505, + 280 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 286, + 504, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 285, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 302 + ], + "score": 1.0, + "content": "Xiaojin Zhu. Semi-supervised learning. Encyclopedia of Machine Learning and Data Mining, pp.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 117, + 298, + 191, + 310 + ], + "spans": [ + { + "bbox": [ + 117, + 298, + 191, + 310 + ], + "score": 1.0, + "content": "1142–1147, 2017.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 285, + 505, + 310 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 332, + 194, + 344 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 195, + 347 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 195, + 347 + ], + "score": 1.0, + "content": "A ALGORITHM", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 108, + 365, + 258, + 376 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 259, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 259, + 379 + ], + "score": 1.0, + "content": "Algorithm 1: MutexMatch algorithm", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "index", + "bbox": [ + 97, + 379, + 473, + 589 + ], + "lines": [ + { + "bbox": [ + 100, + 371, + 466, + 405 + ], + "spans": [ + { + "bbox": [ + 100, + 371, + 204, + 405 + ], + "score": 1.0, + "content": "Input: batch of labeledfeature extractor", + "type": "text" + }, + { + "bbox": [ + 211, + 371, + 222, + 405 + ], + "score": 1.0, + "content": "ata , T", + "type": "text" + }, + { + "bbox": [ + 222, + 381, + 305, + 394 + ], + "score": 0.9, + "content": "\\mathcal { X } = \\{ ( x _ { b } ^ { l b } , y _ { b } ^ { l b } ) \\} _ { b = 1 } ^ { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 371, + 404, + 405 + ], + "score": 1.0, + "content": ", batch of unlabeled data", + "type": "text" + }, + { + "bbox": [ + 404, + 381, + 466, + 394 + ], + "score": 0.89, + "content": "\\mathcal { U } = \\{ x _ { b } ^ { u l b } \\} _ { b = 1 } ^ { \\mu B }", + "type": "inline_equation" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 205, + 393, + 283, + 403 + ], + "spans": [ + { + "bbox": [ + 205, + 393, + 211, + 402 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 393, + 245, + 403 + ], + "score": 0.7, + "content": "\\mathcal { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 394, + 283, + 402 + ], + "score": 0.53, + "content": "\\mathcal { N }", + "type": "inline_equation" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 97, + 402, + 178, + 415 + ], + "spans": [ + { + "bbox": [ + 97, + 402, + 158, + 415 + ], + "score": 1.0, + "content": "1 for iteration", + "type": "text" + }, + { + "bbox": [ + 158, + 405, + 163, + 413 + ], + "score": 0.46, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 402, + 178, + 415 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 98, + 411, + 369, + 433 + ], + "spans": [ + { + "bbox": [ + 98, + 419, + 105, + 426 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 116, + 411, + 120, + 433 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 121, + 414, + 369, + 429 + ], + "score": 0.71, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { s u p } = \\frac { 1 } { B } \\sum _ { n = 1 } ^ { B } H ( y _ { n } ^ { l b } , \\mathcal { P } ( x _ { n } ^ { l b } ) ) } & { { } / / \\operatorname { \\it S u p e r v i s e d l o s s f o r } x ^ { l b } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 98, + 428, + 239, + 440 + ], + "spans": [ + { + "bbox": [ + 98, + 431, + 105, + 439 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 121, + 428, + 173, + 440 + ], + "score": 1.0, + "content": "for iteration", + "type": "text" + }, + { + "bbox": [ + 174, + 429, + 197, + 439 + ], + "score": 0.85, + "content": "b = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 428, + 209, + 440 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 209, + 429, + 224, + 440 + ], + "score": 0.54, + "content": "\\mu B", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 428, + 239, + 440 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 99, + 437, + 469, + 454 + ], + "spans": [ + { + "bbox": [ + 99, + 443, + 106, + 451 + ], + "score": 1.0, + "content": "4", + "type": "text" + }, + { + "bbox": [ + 136, + 439, + 228, + 452 + ], + "score": 0.7, + "content": "p _ { b } ^ { w } = \\mathcal { P } ( \\theta ( \\alpha _ { w } ( x _ { b } ^ { u l b } ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 437, + 451, + 454 + ], + "score": 1.0, + "content": "// Compute TPC’s prediction for weakly-augmented", + "type": "text" + }, + { + "bbox": [ + 451, + 441, + 469, + 451 + ], + "score": 0.73, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 99, + 449, + 469, + 467 + ], + "spans": [ + { + "bbox": [ + 99, + 456, + 106, + 464 + ], + "score": 1.0, + "content": "5", + "type": "text" + }, + { + "bbox": [ + 136, + 453, + 225, + 465 + ], + "score": 0.74, + "content": "p _ { b } ^ { s } = \\mathcal { P } ( \\theta ( \\alpha _ { s } ( x _ { b } ^ { u l b } ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 449, + 451, + 467 + ], + "score": 1.0, + "content": "// Compute TPC’s prediction for strongly-augmented", + "type": "text" + }, + { + "bbox": [ + 451, + 454, + 469, + 463 + ], + "score": 0.83, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 99, + 460, + 471, + 483 + ], + "spans": [ + { + "bbox": [ + 99, + 469, + 106, + 477 + ], + "score": 1.0, + "content": "6", + "type": "text" + }, + { + "bbox": [ + 136, + 466, + 229, + 478 + ], + "score": 0.85, + "content": "r _ { b } ^ { w } = \\mathcal { N } ( \\theta ( \\alpha _ { w } ( x _ { b } ^ { u l b } ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 460, + 453, + 483 + ], + "score": 1.0, + "content": "// Compute TNC’s prediction for weakly-augmented", + "type": "text" + }, + { + "bbox": [ + 453, + 466, + 471, + 476 + ], + "score": 0.84, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 99, + 475, + 471, + 493 + ], + "spans": [ + { + "bbox": [ + 99, + 481, + 105, + 489 + ], + "score": 1.0, + "content": "7", + "type": "text" + }, + { + "bbox": [ + 137, + 478, + 226, + 491 + ], + "score": 0.84, + "content": "r _ { b } ^ { s } = \\mathcal { N } ( \\theta ( \\alpha _ { s } ( x _ { b } ^ { u l b } ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 475, + 453, + 493 + ], + "score": 1.0, + "content": "// Compute TNC’s prediction for strongly-augmented", + "type": "text" + }, + { + "bbox": [ + 453, + 479, + 471, + 488 + ], + "score": 0.86, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 99, + 487, + 354, + 506 + ], + "spans": [ + { + "bbox": [ + 99, + 494, + 105, + 502 + ], + "score": 1.0, + "content": "8", + "type": "text" + }, + { + "bbox": [ + 137, + 492, + 217, + 504 + ], + "score": 0.64, + "content": "\\hat { p } _ { b } ^ { w } = \\arg \\operatorname* { m a x } ( p _ { b } ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 487, + 336, + 506 + ], + "score": 1.0, + "content": "// Select pseudo-labels for", + "type": "text" + }, + { + "bbox": [ + 336, + 492, + 354, + 501 + ], + "score": 0.69, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 98, + 498, + 416, + 521 + ], + "spans": [ + { + "bbox": [ + 98, + 506, + 105, + 515 + ], + "score": 1.0, + "content": "9", + "type": "text" + }, + { + "bbox": [ + 137, + 505, + 214, + 516 + ], + "score": 0.57, + "content": "\\hat { q } _ { b } ^ { w } = \\arg \\operatorname* { m i n } ( p _ { b } ^ { w } )", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 498, + 398, + 521 + ], + "score": 1.0, + "content": "// Select complementary pseudo-labels for", + "type": "text" + }, + { + "bbox": [ + 398, + 505, + 416, + 514 + ], + "score": 0.65, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 95, + 515, + 142, + 529 + ], + "spans": [ + { + "bbox": [ + 95, + 518, + 106, + 529 + ], + "score": 1.0, + "content": "10", + "type": "text" + }, + { + "bbox": [ + 120, + 515, + 142, + 528 + ], + "score": 1.0, + "content": "end", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 95, + 524, + 452, + 546 + ], + "spans": [ + { + "bbox": [ + 95, + 531, + 105, + 543 + ], + "score": 1.0, + "content": "11", + "type": "text" + }, + { + "bbox": [ + 122, + 528, + 274, + 543 + ], + "score": 0.46, + "content": "\\begin{array} { r l } { \\mathcal { L } _ { s e p } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } H ( \\hat { q } _ { n } ^ { w } , \\mathcal { N } ( \\hat { \\theta } ( x _ { n } ^ { w } ) ) ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 524, + 452, + 546 + ], + "score": 1.0, + "content": "// Stop back-propagating gradients on θ", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 95, + 540, + 462, + 562 + ], + "spans": [ + { + "bbox": [ + 95, + 547, + 106, + 558 + ], + "score": 1.0, + "content": "12", + "type": "text" + }, + { + "bbox": [ + 141, + 540, + 462, + 562 + ], + "score": 1.0, + "content": "= 1µB PµBn=1 1(max(pwn ) ≥ τ )H(ˆpwn , psn) // Positive consistency loss for xulb", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 95, + 558, + 461, + 577 + ], + "spans": [ + { + "bbox": [ + 95, + 562, + 106, + 573 + ], + "score": 1.0, + "content": "13", + "type": "text" + }, + { + "bbox": [ + 122, + 559, + 304, + 574 + ], + "score": 0.8, + "content": "\\begin{array} { r } { \\mathcal { L } _ { n } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } \\mathbb { 1 } ( \\operatorname* { m a x } ( p _ { n } ^ { w } ) < \\tau ) H ( r _ { n } ^ { w } , r _ { n } ^ { s } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 558, + 443, + 577 + ], + "score": 1.0, + "content": "// Negative consistency loss for", + "type": "text" + }, + { + "bbox": [ + 444, + 562, + 461, + 571 + ], + "score": 0.38, + "content": "x ^ { u l b }", + "type": "inline_equation" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 95, + 573, + 406, + 588 + ], + "spans": [ + { + "bbox": [ + 95, + 576, + 106, + 587 + ], + "score": 1.0, + "content": "14", + "type": "text" + }, + { + "bbox": [ + 120, + 573, + 151, + 588 + ], + "score": 1.0, + "content": "update", + "type": "text" + }, + { + "bbox": [ + 151, + 575, + 183, + 586 + ], + "score": 0.87, + "content": "\\theta , \\mathcal { P } , \\mathcal { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 573, + 267, + 588 + ], + "score": 1.0, + "content": "by SGD to optimise", + "type": "text" + }, + { + "bbox": [ + 267, + 575, + 406, + 587 + ], + "score": 0.87, + "content": "\\mathcal { L } _ { s u p } + \\lambda _ { s e p } \\mathcal { L } _ { s e p } + \\lambda _ { p } \\mathcal { L } _ { p } + \\lambda _ { n } \\mathcal { L } _ { n }", + "type": "inline_equation" + } + ], + "index": 33, + "is_list_start_line": true + } + ], + "index": 25.5, + "bbox_fs": [ + 95, + 371, + 471, + 588 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 629, + 296, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 297, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 297, + 644 + ], + "score": 1.0, + "content": "B BARELY SUPERVISED LEARNING", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "The experimental protocol of barely supervised learning (BSL) described in Sohn et al. (2020) as-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 665, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 104, + 665, + 506, + 679 + ], + "score": 1.0, + "content": "sume a limited availability (e.g., 1 or 5) of labeled data from categories of interest. In order to test", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "the performance of our method in extreme cases, we conduct experiments on CIFAR-10 with only", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "one label per class, and consider developing a simple method to use our TNC in the test phase.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "As shown in Table 4, we use five different random seeds to extract one label of each class from", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 419, + 722 + ], + "score": 1.0, + "content": "CIFAR-10, and use MutexMatch to achieve test accuracy reaching between", + "type": "text" + }, + { + "bbox": [ + 419, + 710, + 452, + 720 + ], + "score": 0.88, + "content": "6 5 . 3 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 709, + 471, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 472, + 710, + 504, + 720 + ], + "score": 0.86, + "content": "9 3 . 0 7 \\%", + "type": "inline_equation" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 721, + 504, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 171, + 732 + ], + "score": 1.0, + "content": "with a mean of", + "type": "text" + }, + { + "bbox": [ + 172, + 721, + 203, + 731 + ], + "score": 0.88, + "content": "7 8 . 7 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 721, + 471, + 732 + ], + "score": 1.0, + "content": ". Compared with FixMatch (Sohn et al., 2020) reaching between", + "type": "text" + }, + { + "bbox": [ + 471, + 721, + 504, + 731 + ], + "score": 0.85, + "content": "4 8 . 5 8 \\%", + "type": "inline_equation" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 124, + 95 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 124, + 83, + 156, + 93 + ], + "score": 0.88, + "content": "8 5 . 3 2 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 156, + 82, + 506, + 95 + ], + "score": 1.0, + "content": ", the performance of MutexMatch is more superior. Then we consider using TNC to", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "complete the test phase under this setting to obtain the test accuracy. We assume that in the ideal", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 283, + 117 + ], + "score": 1.0, + "content": "case, according to Equation (3), for test data", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 283, + 106, + 290, + 114 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 290, + 104, + 383, + 117 + ], + "score": 1.0, + "content": ", the prediction of TNC", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 383, + 104, + 430, + 116 + ], + "score": 0.92, + "content": "r _ { x } = \\mathcal { N } ( x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 430, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "and the prediction", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 363, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 137, + 129 + ], + "score": 1.0, + "content": "of TPC", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 138, + 115, + 184, + 127 + ], + "score": 0.91, + "content": "p _ { x } = \\mathcal { P } ( x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 184, + 114, + 258, + 129 + ], + "score": 1.0, + "content": "should satisfy arg", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 258, + 115, + 359, + 128 + ], + "score": 0.88, + "content": "\\operatorname* { m a x } ( r _ { x } ) = \\arg \\operatorname* { m i n } ( p _ { x } )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 359, + 114, + 363, + 129 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 3 + } + ], + "index": 38, + "bbox_fs": [ + 104, + 655, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 124, + 95 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 83, + 156, + 93 + ], + "score": 0.88, + "content": "8 5 . 3 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 82, + 506, + 95 + ], + "score": 1.0, + "content": ", the performance of MutexMatch is more superior. Then we consider using TNC to", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "complete the test phase under this setting to obtain the test accuracy. We assume that in the ideal", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 283, + 117 + ], + "score": 1.0, + "content": "case, according to Equation (3), for test data", + "type": "text" + }, + { + "bbox": [ + 283, + 106, + 290, + 114 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 104, + 383, + 117 + ], + "score": 1.0, + "content": ", the prediction of TNC", + "type": "text" + }, + { + "bbox": [ + 383, + 104, + 430, + 116 + ], + "score": 0.92, + "content": "r _ { x } = \\mathcal { N } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "and the prediction", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 363, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 137, + 129 + ], + "score": 1.0, + "content": "of TPC", + "type": "text" + }, + { + "bbox": [ + 138, + 115, + 184, + 127 + ], + "score": 0.91, + "content": "p _ { x } = \\mathcal { P } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 114, + 258, + 129 + ], + "score": 1.0, + "content": "should satisfy arg", + "type": "text" + }, + { + "bbox": [ + 258, + 115, + 359, + 128 + ], + "score": 0.88, + "content": "\\operatorname* { m a x } ( r _ { x } ) = \\arg \\operatorname* { m i n } ( p _ { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 114, + 363, + 129 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 505, + 231 + ], + "lines": [ + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "score": 1.0, + "content": "According to negative learning proposed in Kim et al. (2019), we hypothesis TNC is trained", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 506, + 156 + ], + "score": 1.0, + "content": "to classify what input image does not belong to its complementary label, so that we can use", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 183, + 166 + ], + "score": 0.92, + "content": "\\hat { r } _ { x } = \\arg \\operatorname* { m i n } ( r _ { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 154, + 294, + 167 + ], + "score": 1.0, + "content": "to classify an input image", + "type": "text" + }, + { + "bbox": [ + 294, + 156, + 301, + 164 + ], + "score": 0.72, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 154, + 506, + 167 + ], + "score": 1.0, + "content": ". Compared with TPC, TNC may learn less error", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 505, + 178 + ], + "score": 1.0, + "content": "information when the label is extremely scarce, so as to obtain better test performance. In order to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 175, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 494, + 190 + ], + "score": 1.0, + "content": "verify this idea, we used TNC to participate in the test phase showed in Figure 8. For test sample", + "type": "text" + }, + { + "bbox": [ + 494, + 178, + 501, + 186 + ], + "score": 0.67, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 175, + 505, + 190 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 186, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 186, + 228, + 200 + ], + "score": 1.0, + "content": "we set a confidence threshold", + "type": "text" + }, + { + "bbox": [ + 228, + 188, + 237, + 198 + ], + "score": 0.74, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 186, + 250, + 200 + ], + "score": 1.0, + "content": ", if", + "type": "text" + }, + { + "bbox": [ + 250, + 187, + 284, + 199 + ], + "score": 0.92, + "content": "p _ { x } > T", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 186, + 393, + 200 + ], + "score": 1.0, + "content": "we uses TPC to predict, if", + "type": "text" + }, + { + "bbox": [ + 394, + 187, + 427, + 199 + ], + "score": 0.92, + "content": "p _ { x } < T", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 186, + 506, + 200 + ], + "score": 1.0, + "content": "uses TNC instead,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 210, + 211 + ], + "score": 1.0, + "content": "that is, the leftmost point", + "type": "text" + }, + { + "bbox": [ + 210, + 199, + 238, + 209 + ], + "score": 0.87, + "content": "T = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 198, + 505, + 211 + ], + "score": 1.0, + "content": ") in the figure represents only TNC for test, and the rightmost point", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 109, + 207, + 506, + 223 + ], + "spans": [ + { + "bbox": [ + 109, + 209, + 138, + 219 + ], + "score": 0.84, + "content": "T = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 207, + 506, + 223 + ], + "score": 1.0, + "content": ") represents only TNC for test. Taking 20 labels as the dividing line, we can see that using", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 219, + 380, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 380, + 233 + ], + "score": 1.0, + "content": "TNC for prediction has more advantages in the case of fewer labels.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8 + }, + { + "type": "image", + "bbox": [ + 215, + 243, + 396, + 378 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 215, + 243, + 396, + 378 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 215, + 243, + 396, + 378 + ], + "spans": [ + { + "bbox": [ + 215, + 243, + 396, + 378 + ], + "score": 0.969, + "type": "image", + "image_path": "30814809dbc0e0079295d450d91755d7597c6c1a4adaa51bca39e87484fccf02.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 215, + 243, + 396, + 256.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 215, + 256.5, + 396, + 270.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 215, + 270.0, + 396, + 283.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 215, + 283.5, + 396, + 297.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 215, + 297.0, + 396, + 310.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 215, + 310.5, + 396, + 324.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 215, + 324.0, + 396, + 337.5 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 215, + 337.5, + 396, + 351.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 215, + 351.0, + 396, + 364.5 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 215, + 364.5, + 396, + 378.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 391, + 505, + 414 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "score": 1.0, + "content": "Figure 8: Test accuracy on CIFAR-10 in single run with various amount of labels using TNC to par-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 401, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 200, + 416 + ], + "score": 1.0, + "content": "ticipate test phase. The", + "type": "text" + }, + { + "bbox": [ + 200, + 404, + 206, + 412 + ], + "score": 0.46, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 401, + 354, + 416 + ], + "score": 1.0, + "content": "-axis represents confidence threshold", + "type": "text" + }, + { + "bbox": [ + 354, + 403, + 363, + 412 + ], + "score": 0.8, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 401, + 505, + 416 + ], + "score": 1.0, + "content": "and y-axis represents test accuracy.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + } + ], + "index": 20.5 + }, + { + "type": "table", + "bbox": [ + 198, + 467, + 411, + 502 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 434, + 503, + 458 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "Table 4: Accuracy of MutexMatch on a single 1-label split of CIFAR-10 with different random", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 445, + 265, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 265, + 460 + ], + "score": 1.0, + "content": "seeds. Results are ordered by accuracy.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "table_body", + "bbox": [ + 198, + 467, + 411, + 502 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 467, + 411, + 502 + ], + "spans": [ + { + "bbox": [ + 198, + 467, + 411, + 502 + ], + "score": 0.965, + "html": "
Fold12345
Accuracy65.3071.1277.8386.3393.07
", + "type": "table", + "image_path": "eb8cedba45fdaa2435ea10c6989cdbe988a102ef8bf42b5d616d28a553b0c5d5.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 198, + 467, + 411, + 484.5 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 198, + 484.5, + 411, + 502.0 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "index": 26.5 + }, + { + "type": "title", + "bbox": [ + 107, + 520, + 393, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 395, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 395, + 534 + ], + "score": 1.0, + "content": "C SEMI-SUPERVISED LEARNING WITH NOISY LABELS", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 505, + 600 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "To evaluate the robustness of MutexMatch, we conduct our experiments following settings of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 557, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 505, + 568 + ], + "score": 1.0, + "content": "semi-supervised learning with noisy labels on CIFAR-10. Semi-supervised learning and noise", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "labels are challenging problems, and semi-supervised learning with noise labels is much more", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "because the ability of the model to resist noise labels will be greatly weakened when there is only a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 589, + 227, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 227, + 601 + ], + "score": 1.0, + "content": "small amount of labeled data.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 610, + 505, + 633 + ], + "lines": [ + { + "bbox": [ + 105, + 609, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 624 + ], + "score": 1.0, + "content": "Setting. Following Kim et al. (2019); Patrini et al. (2017), we applied three different types", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 621, + 205, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 205, + 635 + ], + "score": 1.0, + "content": "of noise in experiments:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 627, + 506, + 721 + ], + "lines": [ + { + "bbox": [ + 105, + 632, + 439, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 439, + 646 + ], + "score": 1.0, + "content": "(1) Symmetric-inc noise is created by randomly selecting the label from all classes.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "(2) Symmetric-exc noise is created by randomly selecting the label from all classes without ground", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 655, + 152, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 152, + 666 + ], + "score": 1.0, + "content": "truth label.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 104, + 665, + 340, + 678 + ], + "score": 1.0, + "content": "(3) Asymmetric noise is generated by mapping TRUCK", + "type": "text" + }, + { + "bbox": [ + 340, + 667, + 354, + 676 + ], + "score": 0.8, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 665, + 450, + 678 + ], + "score": 1.0, + "content": "AUTOMOBILE, BIRD", + "type": "text" + }, + { + "bbox": [ + 451, + 666, + 486, + 676 + ], + "score": 0.25, + "content": "\\mathrm { \\Sigma } \\mathrm { P L A }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "NE,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 677, + 321, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 134, + 689 + ], + "score": 1.0, + "content": "DEER", + "type": "text" + }, + { + "bbox": [ + 134, + 677, + 181, + 687 + ], + "score": 0.6, + "content": " \\mathrm { H O R S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 677, + 202, + 689 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 202, + 677, + 259, + 687 + ], + "score": 0.76, + "content": "\\mathbf { C A T } \\mathbf { D O G }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 677, + 321, + 689 + ], + "score": 1.0, + "content": "for CIFAR-10.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "We evaluate MutexMatch and baselines with noisy labels mentioned above using the same settings", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "as in Section 4.1. All experiments use 40 labeled data for training, varying radio of noisy labels in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 213, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 157, + 721 + ], + "score": 1.0, + "content": "labeled data", + "type": "text" + }, + { + "bbox": [ + 158, + 710, + 207, + 721 + ], + "score": 0.6, + "content": "( 2 5 \\% \\& 5 0 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 710, + 213, + 721 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 40.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [], + "index": 1.5, + "bbox_fs": [ + 105, + 82, + 506, + 129 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 505, + 231 + ], + "lines": [ + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "score": 1.0, + "content": "According to negative learning proposed in Kim et al. (2019), we hypothesis TNC is trained", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 506, + 156 + ], + "score": 1.0, + "content": "to classify what input image does not belong to its complementary label, so that we can use", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 183, + 166 + ], + "score": 0.92, + "content": "\\hat { r } _ { x } = \\arg \\operatorname* { m i n } ( r _ { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 154, + 294, + 167 + ], + "score": 1.0, + "content": "to classify an input image", + "type": "text" + }, + { + "bbox": [ + 294, + 156, + 301, + 164 + ], + "score": 0.72, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 154, + 506, + 167 + ], + "score": 1.0, + "content": ". Compared with TPC, TNC may learn less error", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 505, + 178 + ], + "score": 1.0, + "content": "information when the label is extremely scarce, so as to obtain better test performance. In order to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 175, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 494, + 190 + ], + "score": 1.0, + "content": "verify this idea, we used TNC to participate in the test phase showed in Figure 8. For test sample", + "type": "text" + }, + { + "bbox": [ + 494, + 178, + 501, + 186 + ], + "score": 0.67, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 175, + 505, + 190 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 186, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 186, + 228, + 200 + ], + "score": 1.0, + "content": "we set a confidence threshold", + "type": "text" + }, + { + "bbox": [ + 228, + 188, + 237, + 198 + ], + "score": 0.74, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 186, + 250, + 200 + ], + "score": 1.0, + "content": ", if", + "type": "text" + }, + { + "bbox": [ + 250, + 187, + 284, + 199 + ], + "score": 0.92, + "content": "p _ { x } > T", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 186, + 393, + 200 + ], + "score": 1.0, + "content": "we uses TPC to predict, if", + "type": "text" + }, + { + "bbox": [ + 394, + 187, + 427, + 199 + ], + "score": 0.92, + "content": "p _ { x } < T", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 186, + 506, + 200 + ], + "score": 1.0, + "content": "uses TNC instead,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 210, + 211 + ], + "score": 1.0, + "content": "that is, the leftmost point", + "type": "text" + }, + { + "bbox": [ + 210, + 199, + 238, + 209 + ], + "score": 0.87, + "content": "T = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 198, + 505, + 211 + ], + "score": 1.0, + "content": ") in the figure represents only TNC for test, and the rightmost point", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 109, + 207, + 506, + 223 + ], + "spans": [ + { + "bbox": [ + 109, + 209, + 138, + 219 + ], + "score": 0.84, + "content": "T = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 207, + 506, + 223 + ], + "score": 1.0, + "content": ") represents only TNC for test. Taking 20 labels as the dividing line, we can see that using", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 219, + 380, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 380, + 233 + ], + "score": 1.0, + "content": "TNC for prediction has more advantages in the case of fewer labels.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8, + "bbox_fs": [ + 104, + 132, + 506, + 233 + ] + }, + { + "type": "image", + "bbox": [ + 215, + 243, + 396, + 378 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 215, + 243, + 396, + 378 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 215, + 243, + 396, + 378 + ], + "spans": [ + { + "bbox": [ + 215, + 243, + 396, + 378 + ], + "score": 0.969, + "type": "image", + "image_path": "30814809dbc0e0079295d450d91755d7597c6c1a4adaa51bca39e87484fccf02.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 215, + 243, + 396, + 256.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 215, + 256.5, + 396, + 270.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 215, + 270.0, + 396, + 283.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 215, + 283.5, + 396, + 297.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 215, + 297.0, + 396, + 310.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 215, + 310.5, + 396, + 324.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 215, + 324.0, + 396, + 337.5 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 215, + 337.5, + 396, + 351.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 215, + 351.0, + 396, + 364.5 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 215, + 364.5, + 396, + 378.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 391, + 505, + 414 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "score": 1.0, + "content": "Figure 8: Test accuracy on CIFAR-10 in single run with various amount of labels using TNC to par-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 401, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 200, + 416 + ], + "score": 1.0, + "content": "ticipate test phase. The", + "type": "text" + }, + { + "bbox": [ + 200, + 404, + 206, + 412 + ], + "score": 0.46, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 401, + 354, + 416 + ], + "score": 1.0, + "content": "-axis represents confidence threshold", + "type": "text" + }, + { + "bbox": [ + 354, + 403, + 363, + 412 + ], + "score": 0.8, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 401, + 505, + 416 + ], + "score": 1.0, + "content": "and y-axis represents test accuracy.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + } + ], + "index": 20.5 + }, + { + "type": "table", + "bbox": [ + 198, + 467, + 411, + 502 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 434, + 503, + 458 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "Table 4: Accuracy of MutexMatch on a single 1-label split of CIFAR-10 with different random", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 445, + 265, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 265, + 460 + ], + "score": 1.0, + "content": "seeds. Results are ordered by accuracy.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "table_body", + "bbox": [ + 198, + 467, + 411, + 502 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 467, + 411, + 502 + ], + "spans": [ + { + "bbox": [ + 198, + 467, + 411, + 502 + ], + "score": 0.965, + "html": "
Fold12345
Accuracy65.3071.1277.8386.3393.07
", + "type": "table", + "image_path": "eb8cedba45fdaa2435ea10c6989cdbe988a102ef8bf42b5d616d28a553b0c5d5.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 198, + 467, + 411, + 484.5 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 198, + 484.5, + 411, + 502.0 + ], + "spans": [], + "index": 28 + } + ] + } + ], + "index": 26.5 + }, + { + "type": "title", + "bbox": [ + 107, + 520, + 393, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 395, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 395, + 534 + ], + "score": 1.0, + "content": "C SEMI-SUPERVISED LEARNING WITH NOISY LABELS", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 505, + 600 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "To evaluate the robustness of MutexMatch, we conduct our experiments following settings of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 557, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 505, + 568 + ], + "score": 1.0, + "content": "semi-supervised learning with noisy labels on CIFAR-10. Semi-supervised learning and noise", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 567, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 506, + 580 + ], + "score": 1.0, + "content": "labels are challenging problems, and semi-supervised learning with noise labels is much more", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "because the ability of the model to resist noise labels will be greatly weakened when there is only a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 589, + 227, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 227, + 601 + ], + "score": 1.0, + "content": "small amount of labeled data.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 545, + 506, + 601 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 610, + 505, + 633 + ], + "lines": [ + { + "bbox": [ + 105, + 609, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 624 + ], + "score": 1.0, + "content": "Setting. Following Kim et al. (2019); Patrini et al. (2017), we applied three different types", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 621, + 205, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 205, + 635 + ], + "score": 1.0, + "content": "of noise in experiments:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 609, + 505, + 635 + ] + }, + { + "type": "list", + "bbox": [ + 106, + 627, + 506, + 721 + ], + "lines": [ + { + "bbox": [ + 105, + 632, + 439, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 439, + 646 + ], + "score": 1.0, + "content": "(1) Symmetric-inc noise is created by randomly selecting the label from all classes.", + "type": "text" + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "(2) Symmetric-exc noise is created by randomly selecting the label from all classes without ground", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 655, + 152, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 152, + 666 + ], + "score": 1.0, + "content": "truth label.", + "type": "text" + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 104, + 665, + 340, + 678 + ], + "score": 1.0, + "content": "(3) Asymmetric noise is generated by mapping TRUCK", + "type": "text" + }, + { + "bbox": [ + 340, + 667, + 354, + 676 + ], + "score": 0.8, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 665, + 450, + 678 + ], + "score": 1.0, + "content": "AUTOMOBILE, BIRD", + "type": "text" + }, + { + "bbox": [ + 451, + 666, + 486, + 676 + ], + "score": 0.25, + "content": "\\mathrm { \\Sigma } \\mathrm { P L A }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "NE,", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 677, + 321, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 134, + 689 + ], + "score": 1.0, + "content": "DEER", + "type": "text" + }, + { + "bbox": [ + 134, + 677, + 181, + 687 + ], + "score": 0.6, + "content": " \\mathrm { H O R S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 677, + 202, + 689 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 202, + 677, + 259, + 687 + ], + "score": 0.76, + "content": "\\mathbf { C A T } \\mathbf { D O G }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 677, + 321, + 689 + ], + "score": 1.0, + "content": "for CIFAR-10.", + "type": "text" + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "We evaluate MutexMatch and baselines with noisy labels mentioned above using the same settings", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "as in Section 4.1. All experiments use 40 labeled data for training, varying radio of noisy labels in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 213, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 157, + 721 + ], + "score": 1.0, + "content": "labeled data", + "type": "text" + }, + { + "bbox": [ + 158, + 710, + 207, + 721 + ], + "score": 0.6, + "content": "( 2 5 \\% \\& 5 0 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 710, + 213, + 721 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44, + "is_list_end_line": true + } + ], + "index": 40.5, + "bbox_fs": [ + 104, + 632, + 506, + 721 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "Results. Table 5 shows the accuracy comparison between MutexMatch and baselines. All", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "the results are reported by averaging on 5 different folds. Experiments show the robustness of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "score": 1.0, + "content": "MutexMatch under this setting. For example, with 2 labels and 2 noisy labels (Symmetric-inc) per", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 113, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 222, + 129 + ], + "score": 1.0, + "content": "class, MutexMatch achieves", + "type": "text" + }, + { + "bbox": [ + 222, + 115, + 280, + 126 + ], + "score": 0.9, + "content": "8 8 . 7 2 { \\scriptstyle \\pm 3 . 5 1 \\% }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 113, + 506, + 129 + ], + "score": 1.0, + "content": "accuracy, while training of FixMatch collapse reaching", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 125, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 140, + 140 + ], + "score": 1.0, + "content": "a lower", + "type": "text" + }, + { + "bbox": [ + 140, + 126, + 203, + 137 + ], + "score": 0.9, + "content": "7 7 . 8 0 { \\pm } 1 7 . 5 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 125, + 505, + 140 + ], + "score": 1.0, + "content": "accuracy. MutexMatch contains the idea of negative learning. Learning", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "score": 1.0, + "content": "from the perspective of complementary pseudo-label can prevents model from overfitting to noisy", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "score": 1.0, + "content": "data (Kim et al., 2019) so that MutexMatch achieves superior performance in SSL with noisy labels.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "table", + "bbox": [ + 108, + 209, + 504, + 271 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 177, + 502, + 201 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 178, + 504, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 178, + 504, + 191 + ], + "score": 1.0, + "content": "Table 5: Accuracy on CIFAR-10 with noisy labels averaged on 5 different folds. All experiments", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 189, + 367, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 367, + 201 + ], + "score": 1.0, + "content": "were based on 40 labeled data with varying radio of noisy labels.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 209, + 504, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 209, + 504, + 271 + ], + "spans": [ + { + "bbox": [ + 108, + 209, + 504, + 271 + ], + "score": 0.98, + "html": "
MethodSymmetric-incSymmetric-excAsymmetric
25%noisy50%noisy25%noisy50%noisy25%noisy50%noisy
FixMatch77.80±17.5781.54±18.4780.05±5.8075.11±14.6684.58±5.9072.91±19.30
MutexMatch88.72±8.5177.18±10.5589.37±6.1081.85±8.0089.51±5.1478.28±15.44
", + "type": "table", + "image_path": "08fde405e88d6c290c97436dca1cb7f0cb2da75763009666c4a30e107b560a71.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 108, + 209, + 504, + 229.66666666666666 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 108, + 229.66666666666666, + 504, + 250.33333333333331 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 108, + 250.33333333333331, + 504, + 271.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 8.75 + }, + { + "type": "title", + "bbox": [ + 107, + 291, + 329, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 290, + 330, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 330, + 306 + ], + "score": 1.0, + "content": "D ADDITIONAL EXPERIMENTAL RESULTS", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 108, + 316, + 445, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 315, + 447, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 447, + 329 + ], + "score": 1.0, + "content": "D.1 ABLATION STUDY ON LEARNING RATE AND LEARNING RATE SCHEDULE", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 336, + 505, + 425 + ], + "lines": [ + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "We note that learning rate and learning rate schedule are very important for MutexMatch. In this", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 348, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 506, + 360 + ], + "score": 1.0, + "content": "section, we use the experimental setting in Section 4.1 to conduct additional ablation experiments", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "score": 1.0, + "content": "for both. Following Loshchilov & Hutter (2017), recent work (Sohn et al., 2020; Li et al., 2020) use", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 370, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 382 + ], + "score": 1.0, + "content": "a cosine learning rate decay and achieve best performance. However, as shown in Table 6, we found", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 380, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 394 + ], + "score": 1.0, + "content": "that MutexMatch achieves better results without learning rate decay on CIFAR-10, outperforming", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 391, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 104, + 391, + 227, + 405 + ], + "score": 1.0, + "content": "cosine learning rate decay by", + "type": "text" + }, + { + "bbox": [ + 227, + 392, + 254, + 402 + ], + "score": 0.88, + "content": "0 . 3 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 391, + 505, + 405 + ], + "score": 1.0, + "content": ". When there are many labels, the pseudo-labels outputted by", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 403, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 506, + 415 + ], + "score": 1.0, + "content": "TPC are more likely to have high-confidence and remain stable. It is necessary for MutexMatch to", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 414, + 365, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 365, + 426 + ], + "score": 1.0, + "content": "use cosine learning rate decay to jump out of the local optimum.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17.5 + }, + { + "type": "table", + "bbox": [ + 163, + 476, + 445, + 651 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 444, + 503, + 467 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "Table 6: Ablation study on learning rate and learning rate schedule. Results are reported on CIFAR-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 455, + 224, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 224, + 468 + ], + "score": 1.0, + "content": "10 varying number of labels.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "table_body", + "bbox": [ + 163, + 476, + 445, + 651 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 163, + 476, + 445, + 651 + ], + "spans": [ + { + "bbox": [ + 163, + 476, + 445, + 651 + ], + "score": 0.983, + "html": "
Decay ScheduleLearning RateLabelsBackboneAccuracy
No Decay0.0340WRN-28-293.54
No Decay0.0740WRN-28-293.02
No Decay0.1040WRN-28-292.89
Cosine Decay0.0340WRN-28-293.22
Cosine Decay0.0740WRN-28-293.20
Cosine Decay0.1040WRN-28-292.59
No Decay0.0380WRN-28-293.95
Cosine Decay0.0380WRN-28-294.53
No Decay0.031000CNN-1391.57
Cosine Decay0.031000CNN-1393.46
No Decay0.034000CNN-1392.75
Cosine Decay0.034000CNN-1394.41
", + "type": "table", + "image_path": "1797bd8421b54b54cde22d65e3fc8c69928690a8e8813685eb48b2c628103607.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 163, + 476, + 445, + 534.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 163, + 534.3333333333334, + 445, + 592.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 163, + 592.6666666666667, + 445, + 651.0000000000001 + ], + "spans": [], + "index": 26 + } + ] + } + ], + "index": 23.75 + }, + { + "type": "title", + "bbox": [ + 107, + 667, + 221, + 678 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 222, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 222, + 679 + ], + "score": 1.0, + "content": "D.2 HYPERPARAMETERS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 238, + 700 + ], + "score": 1.0, + "content": "For MutexMatch, the choice of", + "type": "text" + }, + { + "bbox": [ + 238, + 690, + 245, + 698 + ], + "score": 0.72, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 687, + 430, + 700 + ], + "score": 1.0, + "content": "needs to be very cautious, because different", + "type": "text" + }, + { + "bbox": [ + 431, + 690, + 438, + 698 + ], + "score": 0.73, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "will lead to the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "division of high and low-confidence portions, which will affect the impact of the mutex-based con-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "sistency regularization on the model. We use the identical setting of experiments in Section 4.1 for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 198, + 732 + ], + "score": 1.0, + "content": "MutexMatch and vary", + "type": "text" + }, + { + "bbox": [ + 198, + 722, + 205, + 730 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "to verify the sensitivity of MutexMatch to this hyperparameter. As shown", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "Results. Table 5 shows the accuracy comparison between MutexMatch and baselines. All", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "the results are reported by averaging on 5 different folds. Experiments show the robustness of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "score": 1.0, + "content": "MutexMatch under this setting. For example, with 2 labels and 2 noisy labels (Symmetric-inc) per", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 113, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 113, + 222, + 129 + ], + "score": 1.0, + "content": "class, MutexMatch achieves", + "type": "text" + }, + { + "bbox": [ + 222, + 115, + 280, + 126 + ], + "score": 0.9, + "content": "8 8 . 7 2 { \\scriptstyle \\pm 3 . 5 1 \\% }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 113, + 506, + 129 + ], + "score": 1.0, + "content": "accuracy, while training of FixMatch collapse reaching", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 125, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 140, + 140 + ], + "score": 1.0, + "content": "a lower", + "type": "text" + }, + { + "bbox": [ + 140, + 126, + 203, + 137 + ], + "score": 0.9, + "content": "7 7 . 8 0 { \\pm } 1 7 . 5 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 125, + 505, + 140 + ], + "score": 1.0, + "content": "accuracy. MutexMatch contains the idea of negative learning. Learning", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "score": 1.0, + "content": "from the perspective of complementary pseudo-label can prevents model from overfitting to noisy", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "score": 1.0, + "content": "data (Kim et al., 2019) so that MutexMatch achieves superior performance in SSL with noisy labels.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 82, + 506, + 161 + ] + }, + { + "type": "table", + "bbox": [ + 108, + 209, + 504, + 271 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 177, + 502, + 201 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 178, + 504, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 178, + 504, + 191 + ], + "score": 1.0, + "content": "Table 5: Accuracy on CIFAR-10 with noisy labels averaged on 5 different folds. All experiments", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 189, + 367, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 367, + 201 + ], + "score": 1.0, + "content": "were based on 40 labeled data with varying radio of noisy labels.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 209, + 504, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 209, + 504, + 271 + ], + "spans": [ + { + "bbox": [ + 108, + 209, + 504, + 271 + ], + "score": 0.98, + "html": "
MethodSymmetric-incSymmetric-excAsymmetric
25%noisy50%noisy25%noisy50%noisy25%noisy50%noisy
FixMatch77.80±17.5781.54±18.4780.05±5.8075.11±14.6684.58±5.9072.91±19.30
MutexMatch88.72±8.5177.18±10.5589.37±6.1081.85±8.0089.51±5.1478.28±15.44
", + "type": "table", + "image_path": "08fde405e88d6c290c97436dca1cb7f0cb2da75763009666c4a30e107b560a71.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 108, + 209, + 504, + 229.66666666666666 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 108, + 229.66666666666666, + 504, + 250.33333333333331 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 108, + 250.33333333333331, + 504, + 271.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 8.75 + }, + { + "type": "title", + "bbox": [ + 107, + 291, + 329, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 290, + 330, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 330, + 306 + ], + "score": 1.0, + "content": "D ADDITIONAL EXPERIMENTAL RESULTS", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 108, + 316, + 445, + 328 + ], + "lines": [ + { + "bbox": [ + 105, + 315, + 447, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 447, + 329 + ], + "score": 1.0, + "content": "D.1 ABLATION STUDY ON LEARNING RATE AND LEARNING RATE SCHEDULE", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 336, + 505, + 425 + ], + "lines": [ + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "We note that learning rate and learning rate schedule are very important for MutexMatch. In this", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 348, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 506, + 360 + ], + "score": 1.0, + "content": "section, we use the experimental setting in Section 4.1 to conduct additional ablation experiments", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 359, + 505, + 371 + ], + "score": 1.0, + "content": "for both. Following Loshchilov & Hutter (2017), recent work (Sohn et al., 2020; Li et al., 2020) use", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 370, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 382 + ], + "score": 1.0, + "content": "a cosine learning rate decay and achieve best performance. However, as shown in Table 6, we found", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 380, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 394 + ], + "score": 1.0, + "content": "that MutexMatch achieves better results without learning rate decay on CIFAR-10, outperforming", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 391, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 104, + 391, + 227, + 405 + ], + "score": 1.0, + "content": "cosine learning rate decay by", + "type": "text" + }, + { + "bbox": [ + 227, + 392, + 254, + 402 + ], + "score": 0.88, + "content": "0 . 3 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 391, + 505, + 405 + ], + "score": 1.0, + "content": ". When there are many labels, the pseudo-labels outputted by", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 403, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 506, + 415 + ], + "score": 1.0, + "content": "TPC are more likely to have high-confidence and remain stable. It is necessary for MutexMatch to", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 414, + 365, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 365, + 426 + ], + "score": 1.0, + "content": "use cosine learning rate decay to jump out of the local optimum.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17.5, + "bbox_fs": [ + 104, + 337, + 506, + 426 + ] + }, + { + "type": "table", + "bbox": [ + 163, + 476, + 445, + 651 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 108, + 444, + 503, + 467 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "Table 6: Ablation study on learning rate and learning rate schedule. Results are reported on CIFAR-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 455, + 224, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 224, + 468 + ], + "score": 1.0, + "content": "10 varying number of labels.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "table_body", + "bbox": [ + 163, + 476, + 445, + 651 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 163, + 476, + 445, + 651 + ], + "spans": [ + { + "bbox": [ + 163, + 476, + 445, + 651 + ], + "score": 0.983, + "html": "
Decay ScheduleLearning RateLabelsBackboneAccuracy
No Decay0.0340WRN-28-293.54
No Decay0.0740WRN-28-293.02
No Decay0.1040WRN-28-292.89
Cosine Decay0.0340WRN-28-293.22
Cosine Decay0.0740WRN-28-293.20
Cosine Decay0.1040WRN-28-292.59
No Decay0.0380WRN-28-293.95
Cosine Decay0.0380WRN-28-294.53
No Decay0.031000CNN-1391.57
Cosine Decay0.031000CNN-1393.46
No Decay0.034000CNN-1392.75
Cosine Decay0.034000CNN-1394.41
", + "type": "table", + "image_path": "1797bd8421b54b54cde22d65e3fc8c69928690a8e8813685eb48b2c628103607.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 163, + 476, + 445, + 534.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 163, + 534.3333333333334, + 445, + 592.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 163, + 592.6666666666667, + 445, + 651.0000000000001 + ], + "spans": [], + "index": 26 + } + ] + } + ], + "index": 23.75 + }, + { + "type": "title", + "bbox": [ + 107, + 667, + 221, + 678 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 222, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 222, + 679 + ], + "score": 1.0, + "content": "D.2 HYPERPARAMETERS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 238, + 700 + ], + "score": 1.0, + "content": "For MutexMatch, the choice of", + "type": "text" + }, + { + "bbox": [ + 238, + 690, + 245, + 698 + ], + "score": 0.72, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 687, + 430, + 700 + ], + "score": 1.0, + "content": "needs to be very cautious, because different", + "type": "text" + }, + { + "bbox": [ + 431, + 690, + 438, + 698 + ], + "score": 0.73, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "will lead to the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "division of high and low-confidence portions, which will affect the impact of the mutex-based con-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "sistency regularization on the model. We use the identical setting of experiments in Section 4.1 for", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 198, + 732 + ], + "score": 1.0, + "content": "MutexMatch and vary", + "type": "text" + }, + { + "bbox": [ + 198, + 722, + 205, + 730 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "to verify the sensitivity of MutexMatch to this hyperparameter. As shown", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 302, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 317, + 313 + ], + "score": 1.0, + "content": "in Table 7, MutexMatch needs to select appropriate", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 318, + 303, + 325, + 312 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 325, + 302, + 505, + 313 + ], + "score": 1.0, + "content": "to divide confidence portions. We note that", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 221, + 325 + ], + "score": 1.0, + "content": "when there are many labels,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 221, + 315, + 228, + 322 + ], + "score": 0.74, + "content": "\\tau", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 229, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "has a greater impact on performance. The more labels are available,", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 324, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 505, + 335 + ], + "score": 1.0, + "content": "the less confirmation bias will be when using TPC directly for classification, so the portion of TPC", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 334, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 505, + 348 + ], + "score": 1.0, + "content": "in mutex-based consistency regularization can be used directly for learning. Therefore, we guess", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 343, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 280, + 360 + ], + "score": 1.0, + "content": "that in general, we should choose a smaller", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 280, + 347, + 287, + 355 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 287, + 343, + 506, + 360 + ], + "score": 1.0, + "content": "to make more pseudo-labels participate in the training", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 356, + 288, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 288, + 369 + ], + "score": 1.0, + "content": "of TPC when the number of labels increases.", + "type": "text", + "cross_page": true + } + ], + "index": 19 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 687, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 222, + 118, + 388, + 278 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 503, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 86, + 505, + 104 + ], + "spans": [ + { + "bbox": [ + 104, + 86, + 311, + 104 + ], + "score": 1.0, + "content": "Table 7: Ablation study on confidence threshold", + "type": "text" + }, + { + "bbox": [ + 311, + 91, + 318, + 99 + ], + "score": 0.69, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 86, + 505, + 104 + ], + "score": 1.0, + "content": ". Results are reported on CIFAR-10 varying", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 100, + 178, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 178, + 111 + ], + "score": 1.0, + "content": "number of labels.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 222, + 118, + 388, + 278 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 222, + 118, + 388, + 278 + ], + "spans": [ + { + "bbox": [ + 222, + 118, + 388, + 278 + ], + "score": 0.971, + "html": "
TLabelsBackbone Accuracy
0.540WRN-28-293.52
0.7540WRN-28-293.44
0.8540WRN-28-293.28
0.9540WRN-28-293.54
0.9940WRN-28-292.17
0.580WRN-28-294.53
0.9580WRN-28-293.64
0.51000CNN-1393.46
0.951000CNN-1392.07
0.54000CNN-1394.41
0.954000CNN-1392.94
", + "type": "table", + "image_path": "141e1f3c6d38a859e21bf4a56ba0c61e587dcf1926658decdb9d26ae686edd8d.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 222, + 118, + 388, + 131.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 222, + 131.33333333333334, + 388, + 144.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 222, + 144.66666666666669, + 388, + 158.00000000000003 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 222, + 158.00000000000003, + 388, + 171.33333333333337 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 222, + 171.33333333333337, + 388, + 184.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 222, + 184.6666666666667, + 388, + 198.00000000000006 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 222, + 198.00000000000006, + 388, + 211.3333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 222, + 211.3333333333334, + 388, + 224.66666666666674 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 222, + 224.66666666666674, + 388, + 238.00000000000009 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 222, + 238.00000000000009, + 388, + 251.33333333333343 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 222, + 251.33333333333343, + 388, + 264.66666666666674 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 222, + 264.66666666666674, + 388, + 278.00000000000006 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "text", + "bbox": [ + 106, + 301, + 505, + 368 + ], + "lines": [ + { + "bbox": [ + 106, + 302, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 317, + 313 + ], + "score": 1.0, + "content": "in Table 7, MutexMatch needs to select appropriate", + "type": "text" + }, + { + "bbox": [ + 318, + 303, + 325, + 312 + ], + "score": 0.77, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 302, + 505, + 313 + ], + "score": 1.0, + "content": "to divide confidence portions. We note that", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 221, + 325 + ], + "score": 1.0, + "content": "when there are many labels,", + "type": "text" + }, + { + "bbox": [ + 221, + 315, + 228, + 322 + ], + "score": 0.74, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "has a greater impact on performance. The more labels are available,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 324, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 505, + 335 + ], + "score": 1.0, + "content": "the less confirmation bias will be when using TPC directly for classification, so the portion of TPC", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 334, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 505, + 348 + ], + "score": 1.0, + "content": "in mutex-based consistency regularization can be used directly for learning. Therefore, we guess", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 343, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 280, + 360 + ], + "score": 1.0, + "content": "that in general, we should choose a smaller", + "type": "text" + }, + { + "bbox": [ + 280, + 347, + 287, + 355 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 343, + 506, + 360 + ], + "score": 1.0, + "content": "to make more pseudo-labels participate in the training", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 356, + 288, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 288, + 369 + ], + "score": 1.0, + "content": "of TPC when the number of labels increases.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 107, + 373, + 505, + 428 + ], + "lines": [ + { + "bbox": [ + 105, + 372, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 352, + 387 + ], + "score": 1.0, + "content": "At the same time, showed in Figure 9, we vary the weight", + "type": "text" + }, + { + "bbox": [ + 352, + 374, + 371, + 385 + ], + "score": 0.91, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 372, + 506, + 387 + ], + "score": 1.0, + "content": "of the separate training loss for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 383, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 129, + 398 + ], + "score": 1.0, + "content": "TNC", + "type": "text" + }, + { + "bbox": [ + 129, + 384, + 150, + 397 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 383, + 168, + 398 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 168, + 385, + 181, + 395 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 383, + 312, + 398 + ], + "score": 1.0, + "content": "of the negative consistency loss", + "type": "text" + }, + { + "bbox": [ + 313, + 385, + 326, + 395 + ], + "score": 0.89, + "content": "{ \\mathcal { L } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 383, + 506, + 398 + ], + "score": 1.0, + "content": ". Choosing the appropriate weight of loss is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 266, + 408 + ], + "score": 1.0, + "content": "very important for MutexMatch. Larger", + "type": "text" + }, + { + "bbox": [ + 266, + 396, + 286, + 407 + ], + "score": 0.9, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "ensures the accuracy of complementary pseudo-labels,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 345, + 419 + ], + "score": 1.0, + "content": "which helps TNC better participate in training. Appropriate", + "type": "text" + }, + { + "bbox": [ + 345, + 407, + 358, + 417 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "weighs the contribution of TNC and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 417, + 500, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 500, + 430 + ], + "score": 1.0, + "content": "TPC in mutex-based consistency regularization, so that the model can achieve better performance.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 433, + 505, + 500 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 342, + 447 + ], + "score": 1.0, + "content": "Additionaly, we provide more ablation studies on various", + "type": "text" + }, + { + "bbox": [ + 343, + 434, + 361, + 446 + ], + "score": 0.9, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 434, + 366, + 447 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 366, + 434, + 379, + 446 + ], + "score": 0.83, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 434, + 398, + 447 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 398, + 434, + 410, + 447 + ], + "score": 0.89, + "content": "\\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "shown in Table 8. We", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 185, + 457 + ], + "score": 1.0, + "content": "find that increasing", + "type": "text" + }, + { + "bbox": [ + 185, + 445, + 204, + 457 + ], + "score": 0.91, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 445, + 222, + 457 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 222, + 445, + 235, + 456 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "at the same time will cause severe performance degradation, which", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "shows that we must carefully control the importance of TNC in the learning process, because TPC", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "has always maintained the most important position in completing our classification tasks. Mean-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 181, + 491 + ], + "score": 1.0, + "content": "while, appropriate", + "type": "text" + }, + { + "bbox": [ + 181, + 478, + 193, + 490 + ], + "score": 0.89, + "content": "\\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "ensures model can benefit from learning of TNC. More results of experiments", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 488, + 385, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 194, + 501 + ], + "score": 1.0, + "content": "about situation where", + "type": "text" + }, + { + "bbox": [ + 195, + 489, + 225, + 501 + ], + "score": 0.92, + "content": "\\lambda _ { p } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 488, + 236, + 501 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 237, + 489, + 268, + 500 + ], + "score": 0.92, + "content": "\\lambda _ { n } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 488, + 385, + 501 + ], + "score": 1.0, + "content": "can be found in Section D.3.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5 + }, + { + "type": "table", + "bbox": [ + 148, + 557, + 265, + 659 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 118, + 525, + 297, + 549 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 118, + 524, + 299, + 537 + ], + "spans": [ + { + "bbox": [ + 118, + 524, + 299, + 537 + ], + "score": 1.0, + "content": "Table 8: Accuracy on CIFAR-10 with 40 la-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 118, + 535, + 239, + 550 + ], + "spans": [ + { + "bbox": [ + 118, + 535, + 185, + 550 + ], + "score": 1.0, + "content": "bels and various", + "type": "text" + }, + { + "bbox": [ + 185, + 537, + 236, + 549 + ], + "score": 0.6, + "content": "\\lambda _ { s e p } , \\lambda _ { n } , \\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 535, + 239, + 550 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "table_body", + "bbox": [ + 148, + 557, + 265, + 659 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 148, + 557, + 265, + 659 + ], + "spans": [ + { + "bbox": [ + 148, + 557, + 265, + 659 + ], + "score": 0.969, + "html": "
Xsep入nXpAccuracy
11193.49
111091.27
112085.05
101188.94
201185.44
2010118.03
20101015.96
", + "type": "table", + "image_path": "e133e31142fb6bc8624942263fd6797c2f86519ad98243bb722c8bee4aed80a1.jpg" + } + ] + } + ], + "index": 37.0, + "virtual_lines": [ + { + "bbox": [ + 148, + 557, + 265, + 608.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 148, + 608.0, + 265, + 659.0 + ], + "spans": [], + "index": 41 + } + ] + } + ], + "index": 34.25 + }, + { + "type": "image", + "bbox": [ + 323, + 525, + 482, + 637 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 323, + 525, + 482, + 637 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 323, + 525, + 482, + 637 + ], + "spans": [ + { + "bbox": [ + 323, + 525, + 482, + 637 + ], + "score": 0.968, + "type": "image", + "image_path": "0a082e49f276a6d9fd0da8cd59d788738b4a8a1da723759d2dd943276b71db30.jpg" + } + ] + } + ], + "index": 37.5, + "virtual_lines": [ + { + "bbox": [ + 323, + 525, + 482, + 539.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 323, + 539.0, + 482, + 553.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 323, + 553.0, + 482, + 567.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 323, + 567.0, + 482, + 581.0 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 323, + 581.0, + 482, + 595.0 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 323, + 595.0, + 482, + 609.0 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 323, + 609.0, + 482, + 623.0 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 323, + 623.0, + 482, + 637.0 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 312, + 641, + 492, + 664 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 312, + 641, + 493, + 653 + ], + "spans": [ + { + "bbox": [ + 312, + 641, + 468, + 653 + ], + "score": 1.0, + "content": "Figure 9: Accuracy on CIFAR-10 with", + "type": "text" + }, + { + "bbox": [ + 468, + 641, + 489, + 652 + ], + "score": 0.36, + "content": "4 0 1 \\mathrm { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 641, + 493, + 653 + ], + "score": 1.0, + "content": "-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 312, + 650, + 419, + 666 + ], + "spans": [ + { + "bbox": [ + 312, + 650, + 379, + 666 + ], + "score": 1.0, + "content": "bels and various", + "type": "text" + }, + { + "bbox": [ + 380, + 653, + 398, + 664 + ], + "score": 0.8, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 650, + 402, + 666 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 402, + 653, + 415, + 663 + ], + "score": 0.64, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 650, + 419, + 666 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + } + ], + "index": 40.5 + }, + { + "type": "title", + "bbox": [ + 108, + 687, + 294, + 699 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 294, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 294, + 700 + ], + "score": 1.0, + "content": "D.3 ABLATION STUDY ON TPC AND TNC", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "We explain why we enforce consistency regularization on TNC as follows. Given two augmented", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "variants derived from the same unlabeled instance, we claim that the class probability distributions", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46.5 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 222, + 118, + 388, + 278 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 89, + 503, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 86, + 505, + 104 + ], + "spans": [ + { + "bbox": [ + 104, + 86, + 311, + 104 + ], + "score": 1.0, + "content": "Table 7: Ablation study on confidence threshold", + "type": "text" + }, + { + "bbox": [ + 311, + 91, + 318, + 99 + ], + "score": 0.69, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 86, + 505, + 104 + ], + "score": 1.0, + "content": ". Results are reported on CIFAR-10 varying", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 100, + 178, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 178, + 111 + ], + "score": 1.0, + "content": "number of labels.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 222, + 118, + 388, + 278 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 222, + 118, + 388, + 278 + ], + "spans": [ + { + "bbox": [ + 222, + 118, + 388, + 278 + ], + "score": 0.971, + "html": "
TLabelsBackbone Accuracy
0.540WRN-28-293.52
0.7540WRN-28-293.44
0.8540WRN-28-293.28
0.9540WRN-28-293.54
0.9940WRN-28-292.17
0.580WRN-28-294.53
0.9580WRN-28-293.64
0.51000CNN-1393.46
0.951000CNN-1392.07
0.54000CNN-1394.41
0.954000CNN-1392.94
", + "type": "table", + "image_path": "141e1f3c6d38a859e21bf4a56ba0c61e587dcf1926658decdb9d26ae686edd8d.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 222, + 118, + 388, + 131.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 222, + 131.33333333333334, + 388, + 144.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 222, + 144.66666666666669, + 388, + 158.00000000000003 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 222, + 158.00000000000003, + 388, + 171.33333333333337 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 222, + 171.33333333333337, + 388, + 184.6666666666667 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 222, + 184.6666666666667, + 388, + 198.00000000000006 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 222, + 198.00000000000006, + 388, + 211.3333333333334 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 222, + 211.3333333333334, + 388, + 224.66666666666674 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 222, + 224.66666666666674, + 388, + 238.00000000000009 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 222, + 238.00000000000009, + 388, + 251.33333333333343 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 222, + 251.33333333333343, + 388, + 264.66666666666674 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 222, + 264.66666666666674, + 388, + 278.00000000000006 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "text", + "bbox": [ + 106, + 301, + 505, + 368 + ], + "lines": [], + "index": 16.5, + "bbox_fs": [ + 104, + 302, + 506, + 369 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 373, + 505, + 428 + ], + "lines": [ + { + "bbox": [ + 105, + 372, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 352, + 387 + ], + "score": 1.0, + "content": "At the same time, showed in Figure 9, we vary the weight", + "type": "text" + }, + { + "bbox": [ + 352, + 374, + 371, + 385 + ], + "score": 0.91, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 372, + 506, + 387 + ], + "score": 1.0, + "content": "of the separate training loss for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 383, + 506, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 129, + 398 + ], + "score": 1.0, + "content": "TNC", + "type": "text" + }, + { + "bbox": [ + 129, + 384, + 150, + 397 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 383, + 168, + 398 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 168, + 385, + 181, + 395 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 383, + 312, + 398 + ], + "score": 1.0, + "content": "of the negative consistency loss", + "type": "text" + }, + { + "bbox": [ + 313, + 385, + 326, + 395 + ], + "score": 0.89, + "content": "{ \\mathcal { L } } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 383, + 506, + 398 + ], + "score": 1.0, + "content": ". Choosing the appropriate weight of loss is", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 266, + 408 + ], + "score": 1.0, + "content": "very important for MutexMatch. Larger", + "type": "text" + }, + { + "bbox": [ + 266, + 396, + 286, + 407 + ], + "score": 0.9, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "ensures the accuracy of complementary pseudo-labels,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 345, + 419 + ], + "score": 1.0, + "content": "which helps TNC better participate in training. Appropriate", + "type": "text" + }, + { + "bbox": [ + 345, + 407, + 358, + 417 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "weighs the contribution of TNC and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 417, + 500, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 500, + 430 + ], + "score": 1.0, + "content": "TPC in mutex-based consistency regularization, so that the model can achieve better performance.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 372, + 506, + 430 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 433, + 505, + 500 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 342, + 447 + ], + "score": 1.0, + "content": "Additionaly, we provide more ablation studies on various", + "type": "text" + }, + { + "bbox": [ + 343, + 434, + 361, + 446 + ], + "score": 0.9, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 434, + 366, + 447 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 366, + 434, + 379, + 446 + ], + "score": 0.83, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 434, + 398, + 447 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 398, + 434, + 410, + 447 + ], + "score": 0.89, + "content": "\\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "shown in Table 8. We", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 185, + 457 + ], + "score": 1.0, + "content": "find that increasing", + "type": "text" + }, + { + "bbox": [ + 185, + 445, + 204, + 457 + ], + "score": 0.91, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 445, + 222, + 457 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 222, + 445, + 235, + 456 + ], + "score": 0.89, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "at the same time will cause severe performance degradation, which", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "shows that we must carefully control the importance of TNC in the learning process, because TPC", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "has always maintained the most important position in completing our classification tasks. Mean-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 181, + 491 + ], + "score": 1.0, + "content": "while, appropriate", + "type": "text" + }, + { + "bbox": [ + 181, + 478, + 193, + 490 + ], + "score": 0.89, + "content": "\\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "ensures model can benefit from learning of TNC. More results of experiments", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 488, + 385, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 194, + 501 + ], + "score": 1.0, + "content": "about situation where", + "type": "text" + }, + { + "bbox": [ + 195, + 489, + 225, + 501 + ], + "score": 0.92, + "content": "\\lambda _ { p } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 488, + 236, + 501 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 237, + 489, + 268, + 500 + ], + "score": 0.92, + "content": "\\lambda _ { n } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 488, + 385, + 501 + ], + "score": 1.0, + "content": "can be found in Section D.3.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 434, + 506, + 501 + ] + }, + { + "type": "table", + "bbox": [ + 148, + 557, + 265, + 659 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 118, + 525, + 297, + 549 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 118, + 524, + 299, + 537 + ], + "spans": [ + { + "bbox": [ + 118, + 524, + 299, + 537 + ], + "score": 1.0, + "content": "Table 8: Accuracy on CIFAR-10 with 40 la-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 118, + 535, + 239, + 550 + ], + "spans": [ + { + "bbox": [ + 118, + 535, + 185, + 550 + ], + "score": 1.0, + "content": "bels and various", + "type": "text" + }, + { + "bbox": [ + 185, + 537, + 236, + 549 + ], + "score": 0.6, + "content": "\\lambda _ { s e p } , \\lambda _ { n } , \\lambda _ { p }", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 535, + 239, + 550 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "table_body", + "bbox": [ + 148, + 557, + 265, + 659 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 148, + 557, + 265, + 659 + ], + "spans": [ + { + "bbox": [ + 148, + 557, + 265, + 659 + ], + "score": 0.969, + "html": "
Xsep入nXpAccuracy
11193.49
111091.27
112085.05
101188.94
201185.44
2010118.03
20101015.96
", + "type": "table", + "image_path": "e133e31142fb6bc8624942263fd6797c2f86519ad98243bb722c8bee4aed80a1.jpg" + } + ] + } + ], + "index": 37.0, + "virtual_lines": [ + { + "bbox": [ + 148, + 557, + 265, + 608.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 148, + 608.0, + 265, + 659.0 + ], + "spans": [], + "index": 41 + } + ] + } + ], + "index": 34.25 + }, + { + "type": "image", + "bbox": [ + 323, + 525, + 482, + 637 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 323, + 525, + 482, + 637 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 323, + 525, + 482, + 637 + ], + "spans": [ + { + "bbox": [ + 323, + 525, + 482, + 637 + ], + "score": 0.968, + "type": "image", + "image_path": "0a082e49f276a6d9fd0da8cd59d788738b4a8a1da723759d2dd943276b71db30.jpg" + } + ] + } + ], + "index": 37.5, + "virtual_lines": [ + { + "bbox": [ + 323, + 525, + 482, + 539.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 323, + 539.0, + 482, + 553.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 323, + 553.0, + 482, + 567.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 323, + 567.0, + 482, + 581.0 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 323, + 581.0, + 482, + 595.0 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 323, + 595.0, + 482, + 609.0 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 323, + 609.0, + 482, + 623.0 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 323, + 623.0, + 482, + 637.0 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 312, + 641, + 492, + 664 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 312, + 641, + 493, + 653 + ], + "spans": [ + { + "bbox": [ + 312, + 641, + 468, + 653 + ], + "score": 1.0, + "content": "Figure 9: Accuracy on CIFAR-10 with", + "type": "text" + }, + { + "bbox": [ + 468, + 641, + 489, + 652 + ], + "score": 0.36, + "content": "4 0 1 \\mathrm { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 641, + 493, + 653 + ], + "score": 1.0, + "content": "-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 312, + 650, + 419, + 666 + ], + "spans": [ + { + "bbox": [ + 312, + 650, + 379, + 666 + ], + "score": 1.0, + "content": "bels and various", + "type": "text" + }, + { + "bbox": [ + 380, + 653, + 398, + 664 + ], + "score": 0.8, + "content": "\\lambda _ { s e p }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 650, + 402, + 666 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 402, + 653, + 415, + 663 + ], + "score": 0.64, + "content": "\\lambda _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 650, + 419, + 666 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + } + ], + "index": 40.5 + }, + { + "type": "title", + "bbox": [ + 108, + 687, + 294, + 699 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 294, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 294, + 700 + ], + "score": 1.0, + "content": "D.3 ABLATION STUDY ON TPC AND TNC", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "We explain why we enforce consistency regularization on TNC as follows. Given two augmented", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "variants derived from the same unlabeled instance, we claim that the class probability distributions", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 263, + 504, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 504, + 275 + ], + "score": 1.0, + "content": "(i.e., soft-labels) of their complementary predictions (i.e., the TNC’s outputs) should be consistent.", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 273, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 506, + 287 + ], + "score": 1.0, + "content": "Under the help of an independent training process of TNC, the model can be more confident on", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 284, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 298 + ], + "score": 1.0, + "content": "“what it is not”. As a result, such prediction consistency on TNC can effectively decrease the False-", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "Negative probability on TPC’s predictions. As shown in Figure 10, given a instance of class 1, the", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 306, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 321 + ], + "score": 1.0, + "content": "independent training of TNC can generate an accurate complementary prediction with extremely", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 317, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 506, + 331 + ], + "score": 1.0, + "content": "low probability of class 1. Then encouraging a similar prediction on its strongly augmented variant", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "can help the model to learn more discriminative features. It can in turn affect the TPC’s prediction,", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 339, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 353 + ], + "score": 1.0, + "content": "such that the False-Negative probabilities (to be predicted as a class of 2, 3, 4, 5) can be effectively", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 351, + 460, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 460, + 364 + ], + "score": 1.0, + "content": "decreased. Consequently, the True-Positive probability of TPC’s predictions is enlarged.", + "type": "text", + "cross_page": true + } + ], + "index": 19 + } + ], + "index": 46.5, + "bbox_fs": [ + 106, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 181, + 79, + 425, + 216 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 181, + 79, + 425, + 216 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 181, + 79, + 425, + 216 + ], + "spans": [ + { + "bbox": [ + 181, + 79, + 425, + 216 + ], + "score": 0.969, + "type": "image", + "image_path": "8ee3047fcae56fe3247e0e4b3879c64ca88844cbf1b6ab96f28751aeee398af0.jpg" + } + ] + } + ], + "index": 4.5, + "virtual_lines": [ + { + "bbox": [ + 181, + 79, + 425, + 92.7 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 181, + 92.7, + 425, + 106.4 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 181, + 106.4, + 425, + 120.10000000000001 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 181, + 120.10000000000001, + 425, + 133.8 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 181, + 133.8, + 425, + 147.5 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 181, + 147.5, + 425, + 161.2 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 181, + 161.2, + 425, + 174.89999999999998 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 181, + 174.89999999999998, + 425, + 188.59999999999997 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 181, + 188.59999999999997, + 425, + 202.29999999999995 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 181, + 202.29999999999995, + 425, + 215.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 174, + 227, + 434, + 239 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 174, + 226, + 436, + 240 + ], + "spans": [ + { + "bbox": [ + 174, + 226, + 436, + 240 + ], + "score": 1.0, + "content": "Figure 10: The correct component in prediction vector is class 1.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "index": 7.25 + }, + { + "type": "text", + "bbox": [ + 106, + 262, + 505, + 362 + ], + "lines": [ + { + "bbox": [ + 106, + 263, + 504, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 504, + 275 + ], + "score": 1.0, + "content": "(i.e., soft-labels) of their complementary predictions (i.e., the TNC’s outputs) should be consistent.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 273, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 506, + 287 + ], + "score": 1.0, + "content": "Under the help of an independent training process of TNC, the model can be more confident on", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 284, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 298 + ], + "score": 1.0, + "content": "“what it is not”. As a result, such prediction consistency on TNC can effectively decrease the False-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "Negative probability on TPC’s predictions. As shown in Figure 10, given a instance of class 1, the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 306, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 321 + ], + "score": 1.0, + "content": "independent training of TNC can generate an accurate complementary prediction with extremely", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 317, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 506, + 331 + ], + "score": 1.0, + "content": "low probability of class 1. Then encouraging a similar prediction on its strongly augmented variant", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "can help the model to learn more discriminative features. It can in turn affect the TPC’s prediction,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 339, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 505, + 353 + ], + "score": 1.0, + "content": "such that the False-Negative probabilities (to be predicted as a class of 2, 3, 4, 5) can be effectively", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 351, + 460, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 460, + 364 + ], + "score": 1.0, + "content": "decreased. Consequently, the True-Positive probability of TPC’s predictions is enlarged.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 505, + 445 + ], + "lines": [ + { + "bbox": [ + 106, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "We construct an experiment on CIFAR-10 with 40 labels to verify our findings. We denote Mu-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 379, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 505, + 391 + ], + "score": 1.0, + "content": "texMatch without consistency reguarlization on TNC as M wo. c (i.e., MutexMatch degenerates to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 388, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 468, + 403 + ], + "score": 1.0, + "content": "FixMatch). Although the confidence of the correct predictions of MutexMatch and M wo.", + "type": "text" + }, + { + "bbox": [ + 469, + 392, + 475, + 400 + ], + "score": 0.39, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 388, + 505, + 403 + ], + "score": 1.0, + "content": "is very", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 400, + 505, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 199, + 413 + ], + "score": 1.0, + "content": "high (are very close to", + "type": "text" + }, + { + "bbox": [ + 199, + 400, + 224, + 411 + ], + "score": 0.87, + "content": "1 0 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 400, + 505, + 413 + ], + "score": 1.0, + "content": "), we check their wrong predictions as an example for comparison. In", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 411, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 505, + 424 + ], + "score": 1.0, + "content": "fact, for the correct part of the pseudo-labels, MutexMatch obtains more correct pseudo-labels than", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 423, + 505, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 137, + 434 + ], + "score": 1.0, + "content": "M wo.", + "type": "text" + }, + { + "bbox": [ + 137, + 424, + 144, + 433 + ], + "score": 0.34, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 423, + 505, + 434 + ], + "score": 1.0, + "content": "(FixMatch) thanks to the use of consistency regularization on TNC (i.e., MutexMatch’s", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 434, + 429, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 429, + 446 + ], + "score": 1.0, + "content": "accuracy of pseduo-labels is higher than FixMatch, which is shown in Figure 5).", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 450, + 505, + 516 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 463 + ], + "score": 1.0, + "content": "As shown in Figure 11(a), given the unlabeled instances belonging to “automobile”, the Mutex-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "score": 1.0, + "content": "Match’s average probability of “automobile” component in prediction vector is higher than that of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "M wo. c. We can also obtain similar findings on other different classes, as shown in Figure 11(b).", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "Such observations demonstrate enforcing prediction consistency on TNC can successfully help the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 494, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 507 + ], + "score": 1.0, + "content": "model lower the False-Negative probability, which in turn improve the True-Positive probability in", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 506, + 220, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 220, + 517 + ], + "score": 1.0, + "content": "the TPC’s prediction vector.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5 + }, + { + "type": "image", + "bbox": [ + 140, + 534, + 472, + 694 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 140, + 534, + 472, + 694 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 140, + 534, + 472, + 694 + ], + "spans": [ + { + "bbox": [ + 140, + 534, + 472, + 694 + ], + "score": 0.97, + "type": "image", + "image_path": "7493c3663455406ffd0325a930915cde0a0e05e37845af4add418830bf7a3b5a.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 140, + 534, + 472, + 587.3333333333334 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 140, + 587.3333333333334, + 472, + 640.6666666666667 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 140, + 640.6666666666667, + 472, + 694.0000000000001 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 103, + 707, + 504, + 730 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 707, + 505, + 720 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 505, + 720 + ], + "score": 1.0, + "content": "Figure 11: (a) Average probability of each component in prediction vector of automobile images.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 719, + 480, + 730 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 480, + 730 + ], + "score": 1.0, + "content": "(b) Average probability of correct component in prediction vector of all classes in CIFAR-10.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + } + ], + "index": 35.25 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 181, + 79, + 425, + 216 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 181, + 79, + 425, + 216 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 181, + 79, + 425, + 216 + ], + "spans": [ + { + "bbox": [ + 181, + 79, + 425, + 216 + ], + "score": 0.969, + "type": "image", + "image_path": "8ee3047fcae56fe3247e0e4b3879c64ca88844cbf1b6ab96f28751aeee398af0.jpg" + } + ] + } + ], + "index": 4.5, + "virtual_lines": [ + { + "bbox": [ + 181, + 79, + 425, + 92.7 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 181, + 92.7, + 425, + 106.4 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 181, + 106.4, + 425, + 120.10000000000001 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 181, + 120.10000000000001, + 425, + 133.8 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 181, + 133.8, + 425, + 147.5 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 181, + 147.5, + 425, + 161.2 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 181, + 161.2, + 425, + 174.89999999999998 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 181, + 174.89999999999998, + 425, + 188.59999999999997 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 181, + 188.59999999999997, + 425, + 202.29999999999995 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 181, + 202.29999999999995, + 425, + 215.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 174, + 227, + 434, + 239 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 174, + 226, + 436, + 240 + ], + "spans": [ + { + "bbox": [ + 174, + 226, + 436, + 240 + ], + "score": 1.0, + "content": "Figure 10: The correct component in prediction vector is class 1.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "index": 7.25 + }, + { + "type": "text", + "bbox": [ + 106, + 262, + 505, + 362 + ], + "lines": [], + "index": 15, + "bbox_fs": [ + 105, + 263, + 506, + 364 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 505, + 445 + ], + "lines": [ + { + "bbox": [ + 106, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "We construct an experiment on CIFAR-10 with 40 labels to verify our findings. We denote Mu-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 379, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 505, + 391 + ], + "score": 1.0, + "content": "texMatch without consistency reguarlization on TNC as M wo. c (i.e., MutexMatch degenerates to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 388, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 468, + 403 + ], + "score": 1.0, + "content": "FixMatch). Although the confidence of the correct predictions of MutexMatch and M wo.", + "type": "text" + }, + { + "bbox": [ + 469, + 392, + 475, + 400 + ], + "score": 0.39, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 388, + 505, + 403 + ], + "score": 1.0, + "content": "is very", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 400, + 505, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 199, + 413 + ], + "score": 1.0, + "content": "high (are very close to", + "type": "text" + }, + { + "bbox": [ + 199, + 400, + 224, + 411 + ], + "score": 0.87, + "content": "1 0 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 400, + 505, + 413 + ], + "score": 1.0, + "content": "), we check their wrong predictions as an example for comparison. In", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 411, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 505, + 424 + ], + "score": 1.0, + "content": "fact, for the correct part of the pseudo-labels, MutexMatch obtains more correct pseudo-labels than", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 423, + 505, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 137, + 434 + ], + "score": 1.0, + "content": "M wo.", + "type": "text" + }, + { + "bbox": [ + 137, + 424, + 144, + 433 + ], + "score": 0.34, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 423, + 505, + 434 + ], + "score": 1.0, + "content": "(FixMatch) thanks to the use of consistency regularization on TNC (i.e., MutexMatch’s", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 434, + 429, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 429, + 446 + ], + "score": 1.0, + "content": "accuracy of pseduo-labels is higher than FixMatch, which is shown in Figure 5).", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 367, + 505, + 446 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 450, + 505, + 516 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 463 + ], + "score": 1.0, + "content": "As shown in Figure 11(a), given the unlabeled instances belonging to “automobile”, the Mutex-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 506, + 474 + ], + "score": 1.0, + "content": "Match’s average probability of “automobile” component in prediction vector is higher than that of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "M wo. c. We can also obtain similar findings on other different classes, as shown in Figure 11(b).", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "Such observations demonstrate enforcing prediction consistency on TNC can successfully help the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 494, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 507 + ], + "score": 1.0, + "content": "model lower the False-Negative probability, which in turn improve the True-Positive probability in", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 506, + 220, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 220, + 517 + ], + "score": 1.0, + "content": "the TPC’s prediction vector.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 450, + 506, + 517 + ] + }, + { + "type": "image", + "bbox": [ + 140, + 534, + 472, + 694 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 140, + 534, + 472, + 694 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 140, + 534, + 472, + 694 + ], + "spans": [ + { + "bbox": [ + 140, + 534, + 472, + 694 + ], + "score": 0.97, + "type": "image", + "image_path": "7493c3663455406ffd0325a930915cde0a0e05e37845af4add418830bf7a3b5a.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 140, + 534, + 472, + 587.3333333333334 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 140, + 587.3333333333334, + 472, + 640.6666666666667 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 140, + 640.6666666666667, + 472, + 694.0000000000001 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 103, + 707, + 504, + 730 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 707, + 505, + 720 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 505, + 720 + ], + "score": 1.0, + "content": "Figure 11: (a) Average probability of each component in prediction vector of automobile images.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 719, + 480, + 730 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 480, + 730 + ], + "score": 1.0, + "content": "(b) Average probability of correct component in prediction vector of all classes in CIFAR-10.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + } + ], + "index": 35.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 248 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Moreover, this design is based on the consideration of it’s unreasonable to directly involve com-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "plementary label based negative learning in semi-supervised. In the early training stage of semi-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "score": 1.0, + "content": "supervised learning, the accuracy of pseudo-labels is often not very high. In this process, the intro-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "duction of complementary labels directly into the learning process will not only be helpful, but even", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "harmful, and lead to training collapse finally. Therefore, we use consistency regularization to “de-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "couple” the part where complementary label are directly involved in training. We believe that it is", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 506, + 160 + ], + "score": 1.0, + "content": "reasonable to use consistency regularization on TNC at the sample level, because we only need TNC", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 160, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 505, + 171 + ], + "score": 1.0, + "content": "to show the various results for each class at the dataset level, which is shown in Section 4.3. This", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 169, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 505, + 182 + ], + "score": 1.0, + "content": "demonstrates TNC has learned discriminative information from the aspect of complementary label.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 181, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "Implementing consistency regularization on TNC at the sample level is to help TNC learn from the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 192, + 505, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 505, + 204 + ], + "score": 1.0, + "content": "perspective of complementary labels, so that feature extraction can learn better data representation", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "of unlabeled data with low confidence. In addition, we use soft labels for consistency regularization", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "on TNC, which also ensures TNC can learn multi-class information, which is described in Section", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 224, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 506, + 238 + ], + "score": 1.0, + "content": "3.3. For further discussion on the effectiveness of TPC and TNC, we consider removing these two", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 236, + 471, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 471, + 249 + ], + "score": 1.0, + "content": "components respectively. Given mentioned above, we designed the following experiments:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 128, + 256, + 504, + 323 + ], + "lines": [ + { + "bbox": [ + 127, + 256, + 506, + 270 + ], + "spans": [ + { + "bbox": [ + 127, + 256, + 431, + 270 + ], + "score": 1.0, + "content": "(i) What happens if consistency regularization on TNC is abandoned (i.e.,", + "type": "text" + }, + { + "bbox": [ + 432, + 257, + 465, + 268 + ], + "score": 0.83, + "content": "\\lambda _ { n } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 256, + 506, + 270 + ], + "score": 1.0, + "content": ")? Taking", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "into account that in the default MutexMatch, training of TNC with complementary pseudo-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 277, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 289, + 291 + ], + "score": 1.0, + "content": "labels stops the gradient, so if we set", + "type": "text" + }, + { + "bbox": [ + 289, + 278, + 320, + 289 + ], + "score": 0.91, + "content": "\\lambda _ { n } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 277, + 506, + 291 + ], + "score": 1.0, + "content": ", then TNC is equivalent to not participating in", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "the training of the model at all, which means that MutexMatch degenerates into FixMatch.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 300, + 486, + 312 + ], + "score": 1.0, + "content": "So a more reasonable setting is to restore the backpropagation on feature extractor", + "type": "text" + }, + { + "bbox": [ + 486, + 300, + 493, + 310 + ], + "score": 0.82, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 300, + 505, + 312 + ], + "score": 1.0, + "content": "in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 312, + 501, + 323 + ], + "spans": [ + { + "bbox": [ + 142, + 312, + 501, + 323 + ], + "score": 1.0, + "content": "Equation (3). Then we explore the effect of not using consistency regularization on TNC:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5 + }, + { + "type": "interline_equation", + "bbox": [ + 217, + 326, + 429, + 361 + ], + "lines": [ + { + "bbox": [ + 217, + 326, + 429, + 361 + ], + "spans": [ + { + "bbox": [ + 217, + 326, + 429, + 361 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { s e p } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } H ( \\arg \\operatorname* { m i n } ( \\mathcal { P } ( \\theta ( x _ { n } ^ { w } ) ) , \\mathcal { N } ( \\theta ( x _ { n } ^ { w } ) ) ) ,", + "type": "interline_equation", + "image_path": "e50c330df996bb6d8fd8fa02cac1cff320d2099b326a2dec7ecd6f2ace5eb284.jpg" + } + ] + } + ], + "index": 21.5, + "virtual_lines": [ + { + "bbox": [ + 217, + 326, + 429, + 343.5 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 217, + 343.5, + 429, + 361.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 364, + 446, + 376 + ], + "lines": [ + { + "bbox": [ + 141, + 362, + 446, + 378 + ], + "spans": [ + { + "bbox": [ + 141, + 362, + 311, + 378 + ], + "score": 1.0, + "content": "where we restore the back propagation on", + "type": "text" + }, + { + "bbox": [ + 311, + 365, + 318, + 374 + ], + "score": 0.81, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 362, + 349, + 378 + ], + "score": 1.0, + "content": "and set", + "type": "text" + }, + { + "bbox": [ + 349, + 364, + 380, + 375 + ], + "score": 0.91, + "content": "\\lambda _ { n } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 362, + 446, + 378 + ], + "score": 1.0, + "content": "in Equation (1).", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 123, + 378, + 505, + 454 + ], + "lines": [ + { + "bbox": [ + 124, + 379, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 124, + 379, + 506, + 392 + ], + "score": 1.0, + "content": "(ii) What happens when we train TNC with complementary labels generated by TPC without", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 389, + 504, + 403 + ], + "spans": [ + { + "bbox": [ + 141, + 389, + 304, + 403 + ], + "score": 1.0, + "content": "stopping the gradient on TNC? We keep", + "type": "text" + }, + { + "bbox": [ + 304, + 390, + 335, + 401 + ], + "score": 0.91, + "content": "\\lambda _ { n } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 389, + 504, + 403 + ], + "score": 1.0, + "content": "in Equation (1), and restore the backprop-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 142, + 401, + 270, + 414 + ], + "spans": [ + { + "bbox": [ + 142, + 401, + 186, + 414 + ], + "score": 1.0, + "content": "agation on", + "type": "text" + }, + { + "bbox": [ + 187, + 401, + 193, + 411 + ], + "score": 0.75, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 401, + 270, + 414 + ], + "score": 1.0, + "content": "like Equation (12).", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 123, + 414, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 123, + 414, + 426, + 429 + ], + "score": 1.0, + "content": "(iii) At the same time, in order to explore the role of TPC component, we set", + "type": "text" + }, + { + "bbox": [ + 426, + 416, + 456, + 428 + ], + "score": 0.92, + "content": "\\lambda _ { p } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 414, + 505, + 429 + ], + "score": 1.0, + "content": "in Equation", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 426, + 231, + 439 + ], + "spans": [ + { + "bbox": [ + 141, + 426, + 231, + 439 + ], + "score": 1.0, + "content": "(1) for ablation study.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 122, + 439, + 394, + 455 + ], + "spans": [ + { + "bbox": [ + 122, + 439, + 201, + 455 + ], + "score": 1.0, + "content": "(iv) We simply set", + "type": "text" + }, + { + "bbox": [ + 201, + 440, + 258, + 453 + ], + "score": 0.92, + "content": "( \\lambda _ { p } , \\lambda _ { n } , \\lambda _ { s e p } )", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 439, + 270, + 455 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 271, + 440, + 339, + 453 + ], + "score": 0.31, + "content": "( 0 , 0 , 0 ) , ( 0 , 1 , 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 439, + 358, + 455 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 358, + 441, + 390, + 453 + ], + "score": 0.86, + "content": "( 1 , 1 , 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 439, + 394, + 455 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 461, + 505, + 593 + ], + "lines": [ + { + "bbox": [ + 106, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "As shown in Figure 12, the default MutexMatch achieves dominant performance compared with", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 473, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 370, + 485 + ], + "score": 1.0, + "content": "other settings. In figures, wo. TNC represents setting of (i), wo.", + "type": "text" + }, + { + "bbox": [ + 371, + 473, + 385, + 482 + ], + "score": 0.47, + "content": "_ { S G }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 473, + 505, + 485 + ], + "score": 1.0, + "content": "represents setting of (ii), wo.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 107, + 483, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 107, + 483, + 127, + 494 + ], + "score": 0.35, + "content": "T P C", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 483, + 244, + 496 + ], + "score": 1.0, + "content": "represents setting of (iii) and", + "type": "text" + }, + { + "bbox": [ + 244, + 483, + 337, + 494 + ], + "score": 0.47, + "content": "w . \\ 0 0 0 , w . \\ 0 l 0 , w . \\ l l 0", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 483, + 506, + 496 + ], + "score": 1.0, + "content": "represent setting of (iv). Obviously, TNC", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 494, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 104, + 494, + 506, + 506 + ], + "score": 1.0, + "content": "participates in model training directly will cause training to collapse, which means that the model", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 505, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 505, + 517 + ], + "score": 1.0, + "content": "is seriously affected by the learning of TNC, and there is no way to learn effective information of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 515, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 506, + 529 + ], + "score": 1.0, + "content": "“what it is”. (iii) shows that TPC in this case does not get adequate training and it can’t complete", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 525, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 540 + ], + "score": 1.0, + "content": "the classification task. i.e., the training of TPC is equivalent to using only labeled data. Meanwhile,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 538, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 505, + 549 + ], + "score": 1.0, + "content": "The training of TNC is closely related to TPC. In this case, TNC has not been well trained, too. And", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 547, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 104, + 547, + 505, + 562 + ], + "score": 1.0, + "content": "in fact, we don’t use TNC to participate in the testing phase. (i) and (ii) illustrate the superiority", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "of using consistency regularization for learning on TNC. This “decoupling” ensures that TNC can", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 570, + 506, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 506, + 584 + ], + "score": 1.0, + "content": "make the model learn a better data representation without affecting the learning of TPC. Finally, the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 582, + 486, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 486, + 594 + ], + "score": 1.0, + "content": "combination of (iv) and other settings proves the necessity of each component in MutexMatch.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 35.5 + }, + { + "type": "image", + "bbox": [ + 109, + 607, + 501, + 696 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 607, + 501, + 696 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 607, + 501, + 696 + ], + "spans": [ + { + "bbox": [ + 109, + 607, + 501, + 696 + ], + "score": 0.969, + "type": "image", + "image_path": "df8412667028cb8aecb1f42e55870b95531041612f83866e9add7013c6a04086.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 109, + 607, + 501, + 636.6666666666666 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 109, + 636.6666666666666, + 501, + 666.3333333333333 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 109, + 666.3333333333333, + 501, + 695.9999999999999 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 708, + 504, + 731 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 360, + 721 + ], + "score": 1.0, + "content": "Figure 12: The learning curve of ablation study on SVHN. The", + "type": "text" + }, + { + "bbox": [ + 361, + 710, + 367, + 718 + ], + "score": 0.31, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 708, + 505, + 721 + ], + "score": 1.0, + "content": "-axis represents the training epoch", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 719, + 468, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 123, + 732 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 721, + 130, + 731 + ], + "score": 0.27, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 719, + 468, + 732 + ], + "score": 1.0, + "content": "-axis represents the test accuracy in (a), (c) and the pseudo-label accuracy in (b), (d).", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "index": 44.25 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 248 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Moreover, this design is based on the consideration of it’s unreasonable to directly involve com-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "plementary label based negative learning in semi-supervised. In the early training stage of semi-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 118 + ], + "score": 1.0, + "content": "supervised learning, the accuracy of pseudo-labels is often not very high. In this process, the intro-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "duction of complementary labels directly into the learning process will not only be helpful, but even", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "harmful, and lead to training collapse finally. Therefore, we use consistency regularization to “de-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "couple” the part where complementary label are directly involved in training. We believe that it is", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 506, + 160 + ], + "score": 1.0, + "content": "reasonable to use consistency regularization on TNC at the sample level, because we only need TNC", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 160, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 505, + 171 + ], + "score": 1.0, + "content": "to show the various results for each class at the dataset level, which is shown in Section 4.3. This", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 169, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 505, + 182 + ], + "score": 1.0, + "content": "demonstrates TNC has learned discriminative information from the aspect of complementary label.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 181, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "Implementing consistency regularization on TNC at the sample level is to help TNC learn from the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 192, + 505, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 505, + 204 + ], + "score": 1.0, + "content": "perspective of complementary labels, so that feature extraction can learn better data representation", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "of unlabeled data with low confidence. In addition, we use soft labels for consistency regularization", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "on TNC, which also ensures TNC can learn multi-class information, which is described in Section", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 224, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 506, + 238 + ], + "score": 1.0, + "content": "3.3. For further discussion on the effectiveness of TPC and TNC, we consider removing these two", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 236, + 471, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 471, + 249 + ], + "score": 1.0, + "content": "components respectively. Given mentioned above, we designed the following experiments:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 82, + 506, + 249 + ] + }, + { + "type": "text", + "bbox": [ + 128, + 256, + 504, + 323 + ], + "lines": [ + { + "bbox": [ + 127, + 256, + 506, + 270 + ], + "spans": [ + { + "bbox": [ + 127, + 256, + 431, + 270 + ], + "score": 1.0, + "content": "(i) What happens if consistency regularization on TNC is abandoned (i.e.,", + "type": "text" + }, + { + "bbox": [ + 432, + 257, + 465, + 268 + ], + "score": 0.83, + "content": "\\lambda _ { n } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 256, + 506, + 270 + ], + "score": 1.0, + "content": ")? Taking", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "into account that in the default MutexMatch, training of TNC with complementary pseudo-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 277, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 289, + 291 + ], + "score": 1.0, + "content": "labels stops the gradient, so if we set", + "type": "text" + }, + { + "bbox": [ + 289, + 278, + 320, + 289 + ], + "score": 0.91, + "content": "\\lambda _ { n } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 277, + 506, + 291 + ], + "score": 1.0, + "content": ", then TNC is equivalent to not participating in", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "the training of the model at all, which means that MutexMatch degenerates into FixMatch.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 300, + 486, + 312 + ], + "score": 1.0, + "content": "So a more reasonable setting is to restore the backpropagation on feature extractor", + "type": "text" + }, + { + "bbox": [ + 486, + 300, + 493, + 310 + ], + "score": 0.82, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 300, + 505, + 312 + ], + "score": 1.0, + "content": "in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 312, + 501, + 323 + ], + "spans": [ + { + "bbox": [ + 142, + 312, + 501, + 323 + ], + "score": 1.0, + "content": "Equation (3). Then we explore the effect of not using consistency regularization on TNC:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5, + "bbox_fs": [ + 127, + 256, + 506, + 323 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 217, + 326, + 429, + 361 + ], + "lines": [ + { + "bbox": [ + 217, + 326, + 429, + 361 + ], + "spans": [ + { + "bbox": [ + 217, + 326, + 429, + 361 + ], + "score": 0.94, + "content": "\\mathcal { L } _ { s e p } = \\frac { 1 } { \\mu B } \\sum _ { n = 1 } ^ { \\mu B } H ( \\arg \\operatorname* { m i n } ( \\mathcal { P } ( \\theta ( x _ { n } ^ { w } ) ) , \\mathcal { N } ( \\theta ( x _ { n } ^ { w } ) ) ) ,", + "type": "interline_equation", + "image_path": "e50c330df996bb6d8fd8fa02cac1cff320d2099b326a2dec7ecd6f2ace5eb284.jpg" + } + ] + } + ], + "index": 21.5, + "virtual_lines": [ + { + "bbox": [ + 217, + 326, + 429, + 343.5 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 217, + 343.5, + 429, + 361.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 364, + 446, + 376 + ], + "lines": [ + { + "bbox": [ + 141, + 362, + 446, + 378 + ], + "spans": [ + { + "bbox": [ + 141, + 362, + 311, + 378 + ], + "score": 1.0, + "content": "where we restore the back propagation on", + "type": "text" + }, + { + "bbox": [ + 311, + 365, + 318, + 374 + ], + "score": 0.81, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 362, + 349, + 378 + ], + "score": 1.0, + "content": "and set", + "type": "text" + }, + { + "bbox": [ + 349, + 364, + 380, + 375 + ], + "score": 0.91, + "content": "\\lambda _ { n } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 362, + 446, + 378 + ], + "score": 1.0, + "content": "in Equation (1).", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 141, + 362, + 446, + 378 + ] + }, + { + "type": "list", + "bbox": [ + 123, + 378, + 505, + 454 + ], + "lines": [ + { + "bbox": [ + 124, + 379, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 124, + 379, + 506, + 392 + ], + "score": 1.0, + "content": "(ii) What happens when we train TNC with complementary labels generated by TPC without", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 389, + 504, + 403 + ], + "spans": [ + { + "bbox": [ + 141, + 389, + 304, + 403 + ], + "score": 1.0, + "content": "stopping the gradient on TNC? We keep", + "type": "text" + }, + { + "bbox": [ + 304, + 390, + 335, + 401 + ], + "score": 0.91, + "content": "\\lambda _ { n } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 389, + 504, + 403 + ], + "score": 1.0, + "content": "in Equation (1), and restore the backprop-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 142, + 401, + 270, + 414 + ], + "spans": [ + { + "bbox": [ + 142, + 401, + 186, + 414 + ], + "score": 1.0, + "content": "agation on", + "type": "text" + }, + { + "bbox": [ + 187, + 401, + 193, + 411 + ], + "score": 0.75, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 401, + 270, + 414 + ], + "score": 1.0, + "content": "like Equation (12).", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 123, + 414, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 123, + 414, + 426, + 429 + ], + "score": 1.0, + "content": "(iii) At the same time, in order to explore the role of TPC component, we set", + "type": "text" + }, + { + "bbox": [ + 426, + 416, + 456, + 428 + ], + "score": 0.92, + "content": "\\lambda _ { p } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 414, + 505, + 429 + ], + "score": 1.0, + "content": "in Equation", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 426, + 231, + 439 + ], + "spans": [ + { + "bbox": [ + 141, + 426, + 231, + 439 + ], + "score": 1.0, + "content": "(1) for ablation study.", + "type": "text" + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 122, + 439, + 394, + 455 + ], + "spans": [ + { + "bbox": [ + 122, + 439, + 201, + 455 + ], + "score": 1.0, + "content": "(iv) We simply set", + "type": "text" + }, + { + "bbox": [ + 201, + 440, + 258, + 453 + ], + "score": 0.92, + "content": "( \\lambda _ { p } , \\lambda _ { n } , \\lambda _ { s e p } )", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 439, + 270, + 455 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 271, + 440, + 339, + 453 + ], + "score": 0.31, + "content": "( 0 , 0 , 0 ) , ( 0 , 1 , 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 439, + 358, + 455 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 358, + 441, + 390, + 453 + ], + "score": 0.86, + "content": "( 1 , 1 , 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 439, + 394, + 455 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 26.5, + "bbox_fs": [ + 122, + 379, + 506, + 455 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 461, + 505, + 593 + ], + "lines": [ + { + "bbox": [ + 106, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "As shown in Figure 12, the default MutexMatch achieves dominant performance compared with", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 473, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 370, + 485 + ], + "score": 1.0, + "content": "other settings. In figures, wo. TNC represents setting of (i), wo.", + "type": "text" + }, + { + "bbox": [ + 371, + 473, + 385, + 482 + ], + "score": 0.47, + "content": "_ { S G }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 473, + 505, + 485 + ], + "score": 1.0, + "content": "represents setting of (ii), wo.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 107, + 483, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 107, + 483, + 127, + 494 + ], + "score": 0.35, + "content": "T P C", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 483, + 244, + 496 + ], + "score": 1.0, + "content": "represents setting of (iii) and", + "type": "text" + }, + { + "bbox": [ + 244, + 483, + 337, + 494 + ], + "score": 0.47, + "content": "w . \\ 0 0 0 , w . \\ 0 l 0 , w . \\ l l 0", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 483, + 506, + 496 + ], + "score": 1.0, + "content": "represent setting of (iv). Obviously, TNC", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 494, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 104, + 494, + 506, + 506 + ], + "score": 1.0, + "content": "participates in model training directly will cause training to collapse, which means that the model", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 505, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 505, + 517 + ], + "score": 1.0, + "content": "is seriously affected by the learning of TNC, and there is no way to learn effective information of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 515, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 506, + 529 + ], + "score": 1.0, + "content": "“what it is”. (iii) shows that TPC in this case does not get adequate training and it can’t complete", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 525, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 540 + ], + "score": 1.0, + "content": "the classification task. i.e., the training of TPC is equivalent to using only labeled data. Meanwhile,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 538, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 505, + 549 + ], + "score": 1.0, + "content": "The training of TNC is closely related to TPC. In this case, TNC has not been well trained, too. And", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 547, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 104, + 547, + 505, + 562 + ], + "score": 1.0, + "content": "in fact, we don’t use TNC to participate in the testing phase. (i) and (ii) illustrate the superiority", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "of using consistency regularization for learning on TNC. This “decoupling” ensures that TNC can", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 570, + 506, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 506, + 584 + ], + "score": 1.0, + "content": "make the model learn a better data representation without affecting the learning of TPC. Finally, the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 582, + 486, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 486, + 594 + ], + "score": 1.0, + "content": "combination of (iv) and other settings proves the necessity of each component in MutexMatch.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 35.5, + "bbox_fs": [ + 104, + 461, + 506, + 594 + ] + }, + { + "type": "image", + "bbox": [ + 109, + 607, + 501, + 696 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 607, + 501, + 696 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 607, + 501, + 696 + ], + "spans": [ + { + "bbox": [ + 109, + 607, + 501, + 696 + ], + "score": 0.969, + "type": "image", + "image_path": "df8412667028cb8aecb1f42e55870b95531041612f83866e9add7013c6a04086.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 109, + 607, + 501, + 636.6666666666666 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 109, + 636.6666666666666, + 501, + 666.3333333333333 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 109, + 666.3333333333333, + 501, + 695.9999999999999 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 708, + 504, + 731 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 360, + 721 + ], + "score": 1.0, + "content": "Figure 12: The learning curve of ablation study on SVHN. The", + "type": "text" + }, + { + "bbox": [ + 361, + 710, + 367, + 718 + ], + "score": 0.31, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 708, + 505, + 721 + ], + "score": 1.0, + "content": "-axis represents the training epoch", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 719, + 468, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 123, + 732 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 721, + 130, + 731 + ], + "score": 0.27, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 719, + 468, + 732 + ], + "score": 1.0, + "content": "-axis represents the test accuracy in (a), (c) and the pseudo-label accuracy in (b), (d).", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "index": 44.25 + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/dev/tHK5ntjp-5K/tHK5ntjp-5K_content_list.json b/parse/dev/tHK5ntjp-5K/tHK5ntjp-5K_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..0187a9f1c47c87e03d2c9632c53f8d00f0f32fda --- /dev/null +++ b/parse/dev/tHK5ntjp-5K/tHK5ntjp-5K_content_list.json @@ -0,0 +1,1725 @@ +[ + { + "type": "text", + "text": "LION: Latent Point Diffusion Models for 3D Shape Generation ", + "text_level": 1, + "bbox": [ + 271, + 122, + 725, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Xiaohui Zeng1,2,3,∗ Arash Vahdat1 Francis Williams1 ", + "bbox": [ + 281, + 224, + 714, + 241 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Zan Gojcic1 Or Litany1 Sanja Fidler1,2,3 Karsten Kreis1 ", + "bbox": [ + 251, + 260, + 743, + 276 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1NVIDIA 2University of Toronto 3Vector Institute {xzeng,avahdat,fwilliams,zgojcic,olitany,sfidler,kkreis}@nvidia.com ", + "bbox": [ + 294, + 289, + 702, + 319 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 354, + 535, + 371 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Denoising diffusion models (DDMs) have shown promising results in 3D point cloud synthesis. To advance 3D DDMs and make them useful for digital artists, we require (i) high generation quality, (ii) flexibility for manipulation and applications such as conditional synthesis and shape interpolation, and (iii) the ability to output smooth surfaces or meshes. To this end, we introduce the hierarchical Latent Point Diffusion Model (LION) for 3D shape generation. LION is set up as a variational autoencoder (VAE) with a hierarchical latent space that combines a global shape latent representation with a point-structured latent space. For generation, we train two hierarchical DDMs in these latent spaces. The hierarchical VAE approach boosts performance compared to DDMs that operate on point clouds directly, while the point-structured latents are still ideally suited for DDM-based modeling. Experimentally, LION achieves state-of-the-art generation performance on multiple ShapeNet benchmarks. Furthermore, our VAE framework allows us to easily use LION for different relevant tasks: LION excels at multimodal shape denoising and voxel-conditioned synthesis, and it can be adapted for text- and image-driven 3D generation. We also demonstrate shape autoencoding and latent shape interpolation, and we augment LION with modern surface reconstruction techniques to generate smooth 3D meshes. We hope that LION provides a powerful tool for artists working with 3D shapes due to its high-quality generation, flexibility, and surface reconstruction. Project page and code: https://nv-tlabs.github.io/LION. ", + "bbox": [ + 233, + 387, + 766, + 664 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 Introduction ", + "text_level": 1, + "bbox": [ + 174, + 688, + 310, + 704 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Generative modeling of 3D shapes has extensive applications in 3D content creation and has become an active area of research [1–52]. However, to be useful as a tool for digital artists, generative models of 3D shapes have to fulfill several criteria: (i) Generated shapes need to be realistic and of highquality without artifacts. (ii) The model should enable flexible and interactive use and refinement: For example, a user may want to refine a generated shape and synthesize versions with varying details. Or an artist may provide a coarse or noisy input shape, thereby guiding the model to produce multiple realistic high-quality outputs. Similarly, a user may want to interpolate different shapes. (iii) The model should output smooth meshes, which are the standard representation in most graphics software. ", + "bbox": [ + 174, + 713, + 825, + 824 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Existing 3D generative models build on various frameworks, including generative adversarial networks (GANs) [1–23], variational autoencoders (VAEs) [24–30], normalizing flows [31–34], autoregressive models [35–38], and more [39–44]. Most recently, denoising diffusion models (DDMs) ", + "bbox": [ + 178, + 830, + 823, + 872 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/a5ecbde7333db061775c6917bb95434b89bc4661fe3aa6fbf5e867c83ff8aba7.jpg", + "image_caption": [ + "Figure 1: LION is set up as a hierarchical point cloud VAE with denoising diffusion models over the shape latent and latent point distributions. PointVoxel CNNs (PVCNN) with adaptive Group Normalization (Ada. GN) are used as neural networks. The latent points can be interpreted as a smoothed version of the input point cloud. Shape As Points (SAP) is optionally used for mesh reconstruction. " + ], + "image_footnote": [], + "bbox": [ + 173, + 59, + 648, + 285 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "have emerged as powerful generative models, achieving outstanding results not only on image synthesis [53–64] but also for point cloud-based 3D shape generation [45–47]. In DDMs, the data is gradually perturbed by a diffusion process, while a deep neural network is trained to denoise. This network can then be used to synthesize novel data in an iterative fashion when initialized from random noise [53, 65–67]. However, existing DDMs for 3D shape synthesis struggle with simultaneously satisfying all criteria discussed above for practically useful 3D generative models. ", + "bbox": [ + 173, + 289, + 825, + 372 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Here, we aim to develop a DDM-based generative model of 3D shapes overcoming these limitations. We introduce the Latent Point Diffusion Model (LION) for 3D shape generation (see Fig. 1). Similar to previous 3D DDMs, LION operates on point clouds, but it is constructed as a VAE with DDMs in latent space. LION comprises a hierarchical latent space with a vector-valued global shape latent and another point-structured latent space. The latent representations are predicted with point cloud processing encoders, and two latent DDMs are trained in these latent spaces. Synthesis in LION proceeds by drawing novel latent samples from the hierarchical latent DDMs and decoding back to the original point cloud space. Importantly, we also demonstrate how to augment LION with modern surface reconstruction methods [68] to synthesize smooth shapes as desired by artists. LION has multiple advantages: ", + "bbox": [ + 173, + 378, + 825, + 503 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Expressivity: By mapping point clouds into regularized latent spaces, the DDMs in latent space are effectively tasked with learning a smoothed distribution. This is easier than training on potentially complex point clouds directly [58], thereby improving expressivity. However, point clouds are, in principle, an ideal representation for DDMs. Because of that, we use latent points, this is, we keep a point cloud structure for our main latent representation. Augmenting the model with an additional global shape latent variable in a hierarchical manner further boosts expressivity. We validate LION on several popular ShapeNet benchmarks and achieve state-of-the-art synthesis performance. ", + "bbox": [ + 174, + 510, + 825, + 607 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Varying Output Types: Extending LION with Shape As Points (SAP) [68] geometry reconstruction allows us to also output smooth meshes. Fine-tuning SAP on data generated by LION’s autoencoder reduces synthesis noise and enables us to generate high-quality geometry. LION combines (latent) point cloud-based modeling, ideal for DDMs, with surface reconstruction, desired by artists. ", + "bbox": [ + 174, + 613, + 825, + 669 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Flexibility: Since LION is set up as a VAE, it can be easily adapted for different tasks without retraining the latent DDMs: We can efficiently fine-tune LION’s encoders on voxelized or noisy inputs, which a user can provide for guidance. This enables multimodal voxel-guided synthesis and shape denoising. We also leverage LION’s latent spaces for shape interpolation and autoencoding. Optionally training the DDMs conditioned on CLIP embeddings enables image- and text-driven 3D generation. ", + "bbox": [ + 174, + 674, + 825, + 744 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In summary, we make the following contributions: (i) We introduce LION, a novel generative model for 3D shape synthesis, which operates on point clouds and is built on a hierarchical VAE framework with two latent DDMs. (ii) We validate LION’s high synthesis quality by reaching state-of-the-art performance on widely used ShapeNet benchmarks. (iii) We achieve high-quality and diverse 3D shape synthesis with LION even when trained jointly over many classes without conditioning. (iv) We propose to combine LION with SAP-based surface reconstruction. (v) We demonstrate the flexibility of our framework by adapting it to relevant tasks such as multimodal voxel-guided synthesis. ", + "bbox": [ + 174, + 751, + 825, + 848 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 Background ", + "text_level": 1, + "bbox": [ + 174, + 858, + 308, + 875 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Traditionally, DDMs were introduced in a discrete-step fashion: Given samples $\\mathbf { x } _ { 0 } \\sim q ( \\mathbf { x } _ { 0 } )$ from a data distribution, DDMs use a Markovian fixed forward diffusion process defined as [65, 53] ", + "bbox": [ + 176, + 883, + 823, + 911 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/50a69b4fae41081a6f71ae4fa30d4db630414574be115671a763f0682d5ca458.jpg", + "image_caption": [ + "Figure 2: Generated meshes with LION. Right: Synthesizing different details by diffuse-denoise (see Sec. 3.1) in latent space, while preserving overall shapes. " + ], + "image_footnote": [], + "bbox": [ + 178, + 56, + 689, + 200 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/aab33c64cd41651771e73025dc2397f469be2bf99e574c43a343cd4228a63b61.jpg", + "text": "$$\nq ( \\mathbf { x } _ { 1 : T } | \\mathbf { x } _ { 0 } ) : = \\prod _ { t = 1 } ^ { T } q ( \\mathbf { x } _ { t } | \\mathbf { x } _ { t - 1 } ) , \\qquad q ( \\mathbf { x } _ { t } | \\mathbf { x } _ { t - 1 } ) : = \\mathcal { N } ( \\mathbf { x } _ { t } ; \\sqrt { 1 - \\beta _ { t } } \\mathbf { x } _ { t - 1 } , \\beta _ { t } I ) ,\n$$", + "text_format": "latex", + "bbox": [ + 238, + 198, + 758, + 241 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $T$ denotes the number of steps and $q \\big ( \\mathbf { x } _ { t } | \\mathbf { x } _ { t - 1 } \\big )$ is a Gaussian transition kernel, which gradually adds noise to the input with a variance schedule $\\beta _ { 1 } , . . . , \\beta _ { T }$ . The $\\beta _ { t }$ are chosen such that the chain approximately converges to a standard Gaussian distribution after $T$ steps, $q ( \\mathbf { x } _ { T } ) { \\approx } { \\mathcal { N } } ( \\mathbf { x } _ { T } ; \\mathbf { 0 } , I )$ . DDMs learn a parametrized reverse process (model parameters $\\pmb \\theta$ ) that inverts the forward diffusion: ", + "bbox": [ + 173, + 241, + 825, + 296 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/271c06d7336ca1d1b28e2438cf13cf561e03f0e84c8c3a63ce18a956aef8d0c8.jpg", + "text": "$$\np _ { \\theta } ( \\mathbf { x } _ { 0 : T } ) : = p ( \\mathbf { x } _ { T } ) \\prod _ { t = 1 } ^ { T } p _ { \\theta } ( \\mathbf { x } _ { t - 1 } | \\mathbf { x } _ { t } ) , \\qquad p _ { \\theta } ( \\mathbf { x } _ { t - 1 } | \\mathbf { x } _ { t } ) : = \\mathcal { N } ( \\mathbf { x } _ { t - 1 } ; \\mu _ { \\theta } ( \\mathbf { x } _ { t } , t ) , \\rho _ { t } ^ { 2 } I ) .\n$$", + "text_format": "latex", + "bbox": [ + 222, + 297, + 776, + 340 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "This generative reverse process is also Markovian with Gaussian transition kernels, which use fixed variances $\\rho _ { t } ^ { 2 }$ . DDMs can be interpreted as latent variable models, where $\\mathbf { x } _ { 1 } , . . . , \\mathbf { x } _ { T }$ are latents, and the forward process $q \\big ( \\mathbf { x } _ { 1 : T } | \\mathbf { x } _ { 0 } \\big )$ acts as a fixed approximate posterior, to which the generative $p _ { \\pmb { \\theta } } ( \\mathbf { x } _ { 0 : T } )$ is fit. DDMs are trained by minimizing the variational upper bound on the negative log-likelihood of the data $\\mathbf { x } _ { \\mathrm { 0 } }$ under $p _ { \\theta } ( \\mathbf { x } _ { 0 : T } )$ . Up to irrelevant constant terms, this objective can be expressed as [53] ", + "bbox": [ + 173, + 340, + 825, + 409 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/8758abfcbe929c2a1e7f5ef5591b7cf0e54b47e582ee409ed2060181a887379e.jpg", + "text": "$$\n\\operatorname* { m i n } _ { \\theta } \\mathbb { E } _ { t \\sim U \\left\\{ 1 , T \\right\\} , \\mathbf { x } _ { 0 } \\sim p \\left( \\mathbf { x } _ { 0 } \\right) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } \\left[ w ( t ) | | \\epsilon - \\epsilon _ { \\theta } ( \\alpha _ { t } \\mathbf { x } _ { 0 } + \\sigma _ { t } \\epsilon , t ) | | _ { 2 } ^ { 2 } \\right] , w ( t ) = \\frac { \\beta _ { t } ^ { 2 } } { 2 \\rho _ { t } ^ { 2 } ( 1 - \\beta _ { t } ) ( 1 - \\alpha _ { t } ^ { 2 } ) } ,\n$$", + "text_format": "latex", + "bbox": [ + 183, + 409, + 797, + 441 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\alpha _ { t } = \\sqrt { \\prod _ { s = 1 } ^ { t } ( 1 - \\beta _ { s } ) }$ and $\\sigma _ { t } ~ = ~ \\sqrt { 1 - \\alpha _ { t } ^ { 2 } }$ are the parameters of the tractable diffused distribution after $t$ steps $q ( \\mathbf { x } _ { t } | \\mathbf { x } _ { 0 } ) = \\mathcal { N } ( \\mathbf { x } _ { t } ; \\alpha _ { t } \\mathbf { x } _ { 0 } , \\sigma _ { t } ^ { 2 } I )$ . Furthermore, Eq. (3) employs the widely used parametrization µθ(xt, t) := √ 11−βt $\\begin{array} { r } { \\mu _ { \\theta } ( \\mathbf { x } _ { t } , t ) : = \\frac { 1 } { \\sqrt { 1 - \\beta _ { t } } } \\left( \\mathbf { x } _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - \\alpha _ { t } ^ { 2 } } } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) \\right) } \\end{array}$ . It is common practice to set $w ( t ) = 1$ , instead of the one in Eq. (3), which often promotes perceptual quality of the generated output. In the objective of Eq. (3), the model $\\epsilon _ { \\theta }$ is, for all possible steps $t$ along the diffusion process, effectively trained to predict the noise vector $\\epsilon$ that is necessary to denoise an observed diffused sample $\\mathbf { x } _ { t }$ . After training, the DDM can be sampled with ancestral sampling in an iterative fashion: ", + "bbox": [ + 173, + 444, + 825, + 568 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/6550edfaeeba37f2a7851c20144c611d252e6347668116264063a4c7f440725d.jpg", + "text": "$$\n\\begin{array} { r } { \\mathbf { x } _ { t - 1 } = \\frac { 1 } { \\sqrt { 1 - \\beta _ { t } } } ( \\mathbf { x } _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - \\alpha _ { t } ^ { 2 } } } \\pmb { \\epsilon } _ { \\theta } ( \\mathbf { x } _ { t } , t ) ) + \\rho _ { t } \\pmb { \\eta } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 346, + 569, + 650, + 595 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\eta \\sim \\mathcal { N } ( \\eta ; 0 , I )$ . This sampling chain is initialized from a random sample $\\mathbf { x } _ { T } \\sim \\mathcal { N } ( \\mathbf { x } _ { T } ; \\mathbf { 0 } , I )$ . \nFurthermore, the noise injection in Eq. 4 is usually omitted in the last sampling step. ", + "bbox": [ + 174, + 597, + 825, + 626 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "DDMs can also be expressed with a continuous-time framework [67, 69]. In this formulation, the diffusion and reverse generative processes are described by differential equations. This approach allows for deterministic sampling and encoding schemes based on ordinary differential equations (ODEs). We make use of this framework in Sec. 3.1 and we review this approach in more detail in App. B. ", + "bbox": [ + 174, + 631, + 826, + 688 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 Hierarchical Latent Point Diffusion Models ", + "text_level": 1, + "bbox": [ + 174, + 696, + 568, + 715 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We first formally introduce LION, then discuss various applications and extensions in Sec. 3.1, and finally recapitulate its unique advantages in Sec. 3.2. See Fig. 1 for a visualization of LION. ", + "bbox": [ + 174, + 720, + 823, + 750 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We are modeling point clouds $\\mathbf { x } \\in \\mathbb { R } ^ { 3 \\times N }$ , consisting of $N$ points with xyz-coordinates in $\\mathbb { R } ^ { 3 }$ . LION is set up as a hierarchical VAE with DDMs in latent space. It uses a vector-valued global shape latent $\\mathbf { z } _ { 0 } \\in \\mathbb { R } ^ { D _ { \\mathbf { z } } }$ and a point cloud-structured latent $\\mathbf { h } _ { 0 } \\in \\mathbf { \\bar { \\mathbb { R } } } ^ { ( 3 + D _ { \\mathbf { h } } ) \\times N }$ . Specifically, $\\mathbf { h } _ { 0 }$ is a latent point cloud consisting of $N$ points with $x y z$ -coordinates in $\\mathbb { R } ^ { 3 }$ . In addition, each latent point can carry additional $D _ { \\mathbf { h } }$ latent features. Training of LION is then performed in two stages—first, we train it as a regular VAE with standard Gaussian priors; then, we train the latent DDMs on the latent encodings. ", + "bbox": [ + 173, + 753, + 825, + 840 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "First Stage Training. Initially, LION is trained by maximizing a modified variational lower bound on the data log-likelihood (ELBO) with respect to the encoder and decoder parameters $\\phi$ and $\\boldsymbol { \\xi }$ [70, 71]: ", + "bbox": [ + 171, + 845, + 825, + 875 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/1fc516f62bded3377c6552c66b41784c29310ee897a56536681150999b6a39e6.jpg", + "text": "$$\n\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { E L B O } } ( \\boldsymbol { \\phi } , \\boldsymbol { \\xi } ) = \\mathbb { E } _ { p ( \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } ) } [ \\log p _ { \\xi } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } ) } \\\\ & { \\qquad - \\lambda _ { \\mathbf { z } } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) \\vert p ( \\mathbf { z } _ { 0 } ) ) - \\lambda _ { \\mathbf { h } } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } ) \\vert p ( \\mathbf { h } _ { 0 } ) ) ] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 251, + 876, + 745, + 917 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/794b4b6104af7191905670a7a7a67c74af466190e3278dba537a59617fa285aa.jpg", + "image_caption": [ + "Figure 3: Generated shapes (top: point clouds, bottom: corresponding meshes) from LION trained jointly over 13 classes of ShapeNet-vol without conditioning (Sec. 5.2). " + ], + "image_footnote": [], + "bbox": [ + 179, + 64, + 638, + 162 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Here, the global shape latent $\\mathbf { z } _ { 0 }$ is sampled from the posterior distribution $q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } )$ , which is parametrized by factorial Gaussians, whose means and variances are predicted via an encoder network. The point cloud latent $\\mathbf { h } _ { 0 }$ is sampled from a similarly parametrized posterior $q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } )$ , while also conditioning on $\\mathbf { z } _ { 0 }$ ( $\\cdot \\phi$ denotes the parameters of both encoders). Furthermore, $p _ { \\pmb { \\xi } } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } )$ denotes the decoder, parametrized as a factorial Laplace distribution with predicted means and fixed unit scale parameter (corresponding to an $L _ { 1 }$ reconstruction loss). $\\lambda _ { \\mathbf { z } }$ and $\\lambda _ { \\mathbf { h } }$ are hyperparameters balancing reconstruction accuracy and Kullback-Leibler regularization (note that only for $\\lambda _ { \\mathbf { z } } = \\lambda _ { \\mathbf { h } } = 1$ we are optimizing a rigorous ELBO). The priors $p ( \\mathbf { z } _ { 0 } )$ and $p ( \\mathbf { h } _ { 0 } )$ are $\\mathcal { N } ( \\mathbf { 0 } , \\pmb { I } )$ . Also see Fig. 1 again. ", + "bbox": [ + 173, + 175, + 825, + 287 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Second Stage Training. In principle, we could use the VAE’s priors to sample encodings and generate new shapes. However, the simple Gaussian priors will not accurately match the encoding distribution from the training data and therefore produce poor samples (prior hole problem [58, 72–79]). This motivates training highly expressive latent DDMs. In particular, in the second stage we freeze the VAE’s encoder and decoder networks and train two latent DDMs on the encodings $\\mathbf { z } _ { 0 }$ and $\\mathbf { h } _ { 0 }$ sampled from $q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } )$ and $q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } )$ , minimizing score matching (SM) objectives similar to Eq. (2): ", + "bbox": [ + 173, + 292, + 825, + 377 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/ebdf63e10a31b4b95e5bbc33639e727966fe0bf1c6ad4195a6e3d40902bc0426.jpg", + "text": "$$\n\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S M } ^ { \\mathbf { z } } } ( \\pmb { \\theta } ) = \\mathbb { E } _ { t \\sim U \\{ 1 , T \\} , p ( \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } | | \\epsilon - \\epsilon _ { \\theta } ( \\mathbf { z } _ { t } , t ) | | _ { 2 } ^ { 2 } , } \\\\ & { \\mathcal { L } _ { \\mathrm { S M } ^ { \\mathbf { h } } } ( \\pmb { \\psi } ) = \\mathbb { E } _ { t \\sim U \\{ 1 , T \\} , p ( \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } ) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } | | \\epsilon - \\epsilon _ { \\psi } ( \\mathbf { h } _ { t } , \\mathbf { z } _ { 0 } , t ) | | _ { 2 } ^ { 2 } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 241, + 382, + 756, + 424 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where ${ \\bf z } _ { t } = \\alpha _ { t } { \\bf z } _ { 0 } + \\sigma _ { t } \\epsilon$ and $\\mathbf { h } _ { t } = \\alpha _ { t } \\mathbf { h } _ { 0 } + \\sigma _ { t } \\mathbf { \\epsilon } \\epsilon$ are the diffused latent encodings. Furthermore, $\\pmb \\theta$ denotes the parameters of the global shape latent DDM $\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\mathbf { z } _ { t } , t )$ , and $\\psi$ refers to the parameters of the conditional DDM $\\mathbf { \\epsilon } _ { \\psi } ( \\mathbf { h } _ { t } , \\bar { \\mathbf { z } _ { 0 } } , t )$ trained over the latent point cloud (note the conditioning on $\\mathbf { z } _ { 0 }$ ). ", + "bbox": [ + 176, + 428, + 825, + 469 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Generation. With the latent DDMs, we can formally define a hierarchical generative model $\\begin{array} { r } { p _ { \\xi , \\psi , \\theta } ( \\mathbf { x } , \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } ) = p _ { \\xi } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } ) p _ { \\psi } ( \\mathbf { h } _ { 0 } | \\mathbf { z } _ { 0 } ) p _ { \\theta } ( \\mathbf { z } _ { 0 } ) } \\end{array}$ , where $p _ { \\pmb { \\theta } } ( \\mathbf { z } _ { 0 } )$ denotes the distribution of the global shape latent DDM, $p _ { \\psi } ( \\mathbf { h } _ { 0 } | \\mathbf { z } _ { 0 } )$ refers to the DDM modeling the point cloud-structured latents, and $p _ { \\pmb { \\xi } } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } )$ is LION’s decoder. We can hierarchically sample the latent DDMs following Eq. (4) and then translate the latent points back to the original point cloud space with the decoder. ", + "bbox": [ + 173, + 476, + 825, + 546 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Network Architectures and DDM Parametrization. Let us briefly summarize key implementation choices. The encoder networks, as well as the decoder and the latent point DDM, operating on point clouds $\\mathbf { x }$ , are all implemented based on Point-Voxel CNNs (PVCNNs) [80], following Zhou et al. [46]. PVCNNs efficiently combine the point-based processing of PointNets [81, 82] with the strong spatial inductive bias of convolutions. The DDM modeling the global shape latent uses a ResNet [83] structure with fully-connected layers (implemented as $1 \\times 1$ -convolutions). All conditionings on the global shape latent are implemented via adaptive Group Normalization [84] in the PVCNN layers. Furthermore, following Vahdat et al. [58] we use a mixed score parametrization in both latent DDMs. This means that the score models are parametrized to predict a residual correction to an analytic standard Gaussian score. This is beneficial since the latent encodings are regularized towards a standard Gaussian distribution during the first training stage (see App. D for all details). ", + "bbox": [ + 173, + 551, + 825, + 704 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.1 Applications and Extensions ", + "text_level": 1, + "bbox": [ + 174, + 719, + 410, + 734 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Here, we discuss how LION can be used and extended for different relevant applications. ", + "bbox": [ + 174, + 744, + 754, + 760 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Multimodal Generation. We can synthesize different variations of a given shape, enabling multimodal generation in a controlled manner: Given a shape, i.e., its point cloud $\\mathbf { x }$ , we encode it into latent space. Then, we diffuse its encodings $\\mathbf { z } _ { 0 }$ and $\\mathbf { h } _ { 0 }$ for a small number of steps $\\tau < T$ towards intermediate ${ \\bf z } _ { \\tau }$ and ${ \\bf h } _ { \\tau }$ along the diffusion process such that only local details are destroyed. Running the reverse generation process from this intermediate $\\tau$ , starting at ${ \\bf z } _ { \\tau }$ and ${ \\bf h } _ { \\tau }$ , leads to variations of the original shape with different details (see, for instance, Fig. 2). We refer to this procedure as diffuse-denoise (details in App. C.1). Similar techniques have been used for image editing [85]. ", + "bbox": [ + 174, + 765, + 825, + 863 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Encoder Fine-tuning for Voxel-Conditioned Synthesis and Denoising. In practice, an artist using a 3D generative model may have a rough idea of the desired shape. For instance, they may be able to quickly construct a coarse voxelized shape, to which the generative model then adds realistic details. ", + "bbox": [ + 176, + 869, + 825, + 911 + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/85429d590b634ce75ac7892abdbde7713e22da5e9b4c3ea48ef5c5e6e54822dc.jpg", + "image_caption": [ + "Figure 4: Voxelguided synthesis with LION. We run diffuse-denoise in latent space (see Sec. 3.1) to generate diverse plausible clean shapes. " + ], + "image_footnote": [], + "bbox": [ + 209, + 59, + 668, + 179 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In LION, we can support such applications: using a similar ELBO as in Eq. (5), but with a frozen decoder, we can fine-tune LION’s encoder networks to take voxelized shapes as input (we simply place points at the voxelized shape’s surface) and map them to the corresponding latent encodings $\\mathbf { z } _ { 0 }$ and $\\mathbf { h } _ { 0 }$ that reconstruct the original non-voxelized point cloud. Now, a user can utilize the fine-tuned encoders to encode voxelized shapes and generate plausible detailed shapes. Importantly, this can be naturally combined with the diffuse-denoise procedure to clean up imperfect encodings and to generate different possible detailed shapes (see Fig. 4). ", + "bbox": [ + 173, + 181, + 825, + 277 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Furthermore, this approach is general. Instead of voxel-conditioned synthesis, we can also fine-tune the encoder networks on noisy shapes to perform multimodal shape denoising, also potentially combined with diffuse-denoise. LION supports these applications easily without re-training the latent DDMs due to its VAE framework with additional encoders and decoders, in contrast to previous works that train DDMs on point clouds directly [46, 47]. See App. C.2 for technical details. ", + "bbox": [ + 174, + 284, + 825, + 353 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Shape Interpolation. LION also enables shape interpolation: We can encode different point clouds into LION’s hierarchical latent space and use the probability flow ODE (see App. B) to further encode into the latent DDMs’ Gaussian priors, where we can safely perform spherical interpolation and expect valid shapes along the interpolation path. We can use the intermediate encodings to generate the interpolated shapes (see Fig. 7; details in App. C.3). ", + "bbox": [ + 173, + 359, + 823, + 429 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Surface Reconstruction. While point clouds are an ideal 3D representation for DDMs, artists may prefer meshed outputs. Hence, we propose to combine LION with modern geometry reconstruction methods (see Figs. 2, 4 and 5). We use Shape As Points (SAP) [68], which is based on differentiable Poisson surface reconstruction and can be trained to extract smooth meshes from noisy point clouds. Moreover, we fine-tune SAP on training data generated by LION’s autoencoder to better adjust SAP to the noise distribution in point clouds generated by LION. Specifically, we take clean shapes, encode them into latent space, run a few steps of diffuse-denoise that only slightly modify some details, and decode back. The diffuse-denoise in latent space results in noise in the generated point ated points. clouds similar to what is observed during unconditional synthesis (details in App. C.4). ", + "bbox": [ + 174, + 435, + 741, + 587 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/e8246239ce1cb83b69f6bfa0190e0a475971dd591c1a606190b19903eaf14fce.jpg", + "image_caption": [ + "Figure 5: Reconstructing a mesh from LION’s gener" + ], + "image_footnote": [], + "bbox": [ + 732, + 417, + 802, + 513 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.2 LION’s Advantages ", + "text_level": 1, + "bbox": [ + 174, + 597, + 351, + 611 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We now recapitulate LION’s unique advantages. LION’s structure as a hierarchical VAE with latent DDMs is inspired by latent DDMs on images [57, 58, 77]. This framework has key benefits: ", + "bbox": [ + 173, + 614, + 823, + 643 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "(i) Expressivity: First training a VAE that regularizes the latent encodings to approximately fall under standard Gaussian distributions, which are also the DDMs’ equilibrium distributions towards which the diffusion processes converge, results in an easier modeling task for the DDMs: They have to model only the remaining mismatch between the actual encoding distributions and their own Gaussian priors [58]. This translates into improved expressivity, which is further enhanced by the additional decoder network. However, point clouds are, in principle, an ideal representation for the DDM framework, because they can be diffused and denoised easily and powerful point cloud processing architectures exist. Therefore, LION uses point cloud latents that combine the advantages of both latent DDMs and 3D point clouds. Our point cloud latents can be interpreted as smoothed versions of the original point clouds that are easier to model (see Fig. 1). Moreover, the hierarchical VAE setup with an additional global shape latent increases LION’s expressivity even further and results in natural disentanglement between overall shape and local details captured by the shape latents and latent points (Sec. 5.2). ", + "bbox": [ + 173, + 648, + 825, + 815 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "(ii) Flexibility: Another advantage of LION’s VAE framework is that its encoders can be fine-tuned for various relevant tasks, as discussed previously, and it also enables easy shape interpolation. Other 3D point cloud DDMs operating on point clouds directly [47, 46] do not offer simultaneously as much flexibility and expressivity out-of-the-box (see quantitative comparisons in Secs. 5.1 and 5.4). ", + "bbox": [ + 174, + 820, + 825, + 877 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "(iii) Mesh Reconstruction: As discussed, while point clouds are ideal for DDMs, artists likely prefer meshed outputs. As explained above, we propose to use LION together with modern surface reconstruction techniques [68], again combining the best of both worlds—a point cloud-based VAE backbone ideal for DDMs, and smooth geometry reconstruction methods operating on the synthesized point clouds to generate practically useful smooth surfaces, which can be easily transformed into meshes. ", + "bbox": [ + 174, + 882, + 823, + 911 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/3a7faca45d0a42a04bfe777486d20f82a5448e7fd0bfdf16d03a15e9ff8b2a17.jpg", + "image_caption": [ + "Figure 6: Unconditional shape generation with 2,048 points for airplane, car and chair classes (class-specific models trained on PointFlow’s ShapeNet data with global normalization). " + ], + "image_footnote": [], + "bbox": [ + 178, + 60, + 696, + 218 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 234, + 825, + 277 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4 Related Work ", + "text_level": 1, + "bbox": [ + 174, + 300, + 321, + 318 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We are building on DDMs [53, 65–67], which have been used most prominently for image [53–63] and speech synthesis [86–91]. We train DDMs in latent space, an idea that has been explored for image [57, 58, 77] and music [92] generation, too. However, these works did not train separate conditional DDMs. Hierarchical DDM training has been used for generative image upsampling [54], text-to-image generation [63, 64], and semantic image modeling [60]. Most relevant among these works is Preechakul et al. [60], which extracts a high-level semantic representation of an image with an auxiliary encoder and then trains a DDM that adds details directly in image space. We are the first to explore related concepts for 3D shape synthesis and we also train both DDMs in latent space. Furthermore, DDMs and VAEs have also been combined in such a way that the DDM improves the output of the VAE [93]. ", + "bbox": [ + 174, + 334, + 826, + 459 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Most related to LION are “Point-Voxel Diffusion” (PVD) [46] and “Diffusion Probabilistic Models for 3D Point Cloud Generation” (DPM) [47]. PVD trains a DDM directly on point clouds, and our decision to use PVCNNs is inspired by this work. DPM, like LION, uses a shape latent variable, but models its distribution with Normalizing Flows [94, 95], and then trains a weaker point-wise conditional DDM directly on the point cloud data (this allows DPM to learn useful representations in its latent variable, but sacrifices generation quality). As we show below, neither PVD nor DPM easily enables applications such as multimodal voxel-conditioned synthesis and denoising. Furthermore, LION achieves significantly stronger generation performance. Finally, neither PVD nor DPM reconstructs meshes from the generated point clouds. Point cloud and 3D shape generation have also been explored with other generative models: PointFlow [31], DPF-Net [33] and SoftFlow [32] rely on Normalizing Flows [94–97]. SetVAE [29] treats point cloud synthesis as set generation and uses VAEs. ShapeGF [45] learns distributions over gradient fields that model shape surfaces. Both IM-GAN [7], which models shapes as neural fields, and l-GAN [2] train GANs over latent variables that encode the shapes, similar to other works [3], while r-GAN [2] generates point clouds directly. PDGN [52] proposes progressive deconvolutional networks within a point cloud GAN. SP-GAN [19] uses a spherical point cloud prior. Other progressive [22, 37] and graph-based architectures [4, 6] have been used, too. Also generative cellular automata (GCAs) can be employed for voxel-based 3D shape generation [43]. In orthogonal work, point cloud DDMs have been used for generative shape completion [46, 98]. ", + "bbox": [ + 173, + 464, + 826, + 714 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Recently, image-driven [8–16, 44] training of 3D generative models as well as text-driven 3D generation [34, 49–51] have received much attention. These are complementary directions to ours; in fact, augmenting LION with additional image-based training or including text-guidance are promising future directions. Finally, we are relying on SAP [68] for mesh generation. Strong alternative approaches for reconstructing smooth surfaces from point clouds exist [99–103]. ", + "bbox": [ + 174, + 720, + 823, + 790 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/4415796a634b4f9ff0b2056e1bed79ad61f4bef0f5dc3c549987eef0c051b295.jpg", + "image_caption": [ + "Figure 7: Interpolating different shapes by interpolating their encodings in the standard Gaussian priors of LION’s latent DDMs (details in App. C.3). " + ], + "image_footnote": [], + "bbox": [ + 179, + 801, + 684, + 906 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/ebb39d1c01aa3875199305b3330015b1cff8fdde2ef129455e9fa1343d0c3270.jpg", + "table_caption": [ + "Table 1: Generation metrics (1-NNA↓) on airplane, chair, car categories from ShapeNet dataset from PointFlow [31]. Training and test data normalized globally into [-1, 1]. " + ], + "table_footnote": [], + "table_body": "
AirplaneChairCar
CD EMDCDEMDCDEMD
r-GAN[2]98.40 96.7983.6999.7094.46599.01
1-GAN (CD) [2]87.30 )93.9568.5883.8466.4988.78
1-GAN (EMD)[2]89.49 76.9171.9064.6571.1666.19
PointFlow [31]75.68 370.7462.8460.5758.1056.25
SoftFlow [32]76.05 65.8059.2160.0564.7760.09
SetVAE [29]76.54 67.6558.8460.5759.9459.94
DPF-Net [33]75.18 65.5562.0058.5362.3554.48
DPM[47]76.42 86.9160.0574.7768.8979.97
PVD [46]73.82 64.8156.2653.3254.5553.83
LION (ours)67.41 61.2353.7052.3453.41 51.14
", + "bbox": [ + 178, + 140, + 532, + 308 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/793bb3e9675457e4e80978eea616561af9ac7c1712012a40fe45e51606bd100c.jpg", + "table_caption": [ + "Table 2: Generation results (1-NNA↓) on ShapeNet dataset from PointFlow [31]. All data normalized individually into [-1, 1]. " + ], + "table_footnote": [], + "table_body": "
AirplaneChairCar
CDEMDCD EMDCD EMD
TreeGAN[ [6]97.53 99.88 88.3796.37 89.7794.89
ShapeGF[ [45]81.23 80.8658.0161.25 61.7957.24
SP-GAN[19]94.69 93.9572.5883.6987.36 85.94
PDGN [52]94.94 91.7371.8379.00 89.35 87.22
GCA [43]88.15 85.9364.2764.50 70.4564.20
LION(ours)76.30 67.04 56.50 53.85 59.52 49.29
", + "bbox": [ + 544, + 126, + 816, + 229 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/19fe95ccaec57a967b905abdcd29634f4271ede433ec69399086a8087bd349e1.jpg", + "table_caption": [ + "Table 3: Results (1-NNA↓) on ShapeNet-vol. " + ], + "table_footnote": [], + "table_body": "
AirplaneChairCar
CD EMDCD EMDCD EMD
IM-GAN[7]79.70 77.85 57.0958.20 88.9284.58
DPM[47] PVD [46]83.04 96.04 61.96 66.46 56.06 61.8974.96 77.30 57.90 64.4987.12 55.74
LION (ours) 53.47 53.84 52.0748.6754.81 50.53
", + "bbox": [ + 544, + 251, + 818, + 340 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5 Experiments ", + "text_level": 1, + "bbox": [ + 173, + 325, + 312, + 343 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We provide an overview of our most interesting experimental results in the main paper. All experiment details and extensive additional experiments can be found in App. E and App. F, respectively. ", + "bbox": [ + 174, + 351, + 821, + 378 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5.1 Single-Class 3D Shape Generation ", + "text_level": 1, + "bbox": [ + 174, + 385, + 452, + 400 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Datasets. To compare LION against existing methods, we use ShapeNet [104], the most widely used dataset to benchmark 3D shape generative models. Following previous works [31, 46, 47], we train on three categories: airplane, chair, car. Also like previous methods, we primarily rely on PointFlow’s [31] dataset splits and preprocssing. It normalizes the data globally across the whole dataset. However, some baselines require per-shape normalization [19, 43, 45, 52]; hence, we also train on such data. Furthermore, training SAP requires signed distance fields (SDFs) for volumetric supervision, which the PointFlow data does not offer. Hence, for simplicity we follow Peng et al. [68, 101] and also use their data splits and preprocessing, which includes SDFs.We train LION, DPM, PVD, and IM-GAN (which synthesizes shapes as SDFs) also on as ShapeNet-vol here). This data is also per-shape normalized. ", + "bbox": [ + 174, + 404, + 627, + 595 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Evaluation. Model evaluation follows previous works [31, 46]. Various metrics to evaluate point cloud generative models exist, with different advantages and disadvantages, discussed in detail by Yang et al. [31]. Following recent works [31, 46], we use 1-NNA (with both Chamfer distance (CD) and earth mover distance (EMD)) as our main metric. It quantifies the distributional similarity between generated shapes and validation set and measures both quality and diversity [31]. For fair comparisons, all metrics are computed on point clouds, not meshed outputs (App. E.2 discusses different metrics; further results on coverage (COV) and minimum matching distance (MMD) in App. F.2). ", + "bbox": [ + 174, + 603, + 627, + 742 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/5b55e07dfb5406b0da8472c26a7620d4b2d2f093be889f901eda2c6ce9df0988.jpg", + "image_caption": [ + "Figure 8: Samples from our unconditional 13-class model: In each column, we use the same global shape latent $\\mathbf { z } _ { 0 }$ . " + ], + "image_footnote": [], + "bbox": [ + 643, + 387, + 831, + 506 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "this dataset version (denoted Dataset details in App. E.1. ", + "bbox": [ + 633, + 569, + 825, + 597 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/e83d289bc06c0f5fc610ca5bc6ef67d763a882fba325d2247a81695db22c08c2.jpg", + "table_caption": [ + "Table 4: Generation results (1- NNA↓) of LION trained jointly on 13 classes of ShapeNet-vol. " + ], + "table_footnote": [], + "table_body": "
ModelCDEMD
TreeGAN [6]96.8096.60
PointFlow [31]63.2566.05
ShapeGF[45]55.6559.00
SetVAE [29]79.2595.25
PDGN [52]71.0586.00
DPF-Net [33]67.1064.75
DPM[47]62.3086.50
PVD [46]58.6557.85
LION (ours)51.8548.95
", + "bbox": [ + 642, + 656, + 826, + 784 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Results. Samples from LION are shown in Fig. 6 and quantitative results in Tabs. 1-3 (see Sec. 4 for details about baselines—to reduce the number of baselines to train, we are focusing on the most recent and competitive ones). LION outperforms all baselines and achieves state-of-the-art performance on all classes and dataset versions. Importantly, we outperform both PVD and DPM, which also leverage DDMs, by large margins. Our samples are diverse and appear visually pleasing. ", + "bbox": [ + 173, + 747, + 625, + 790 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 790, + 826, + 832 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/9b9618423aeeac129aa5e6e358feb49f5e1ebb18eaf986c614bfbfb4c5e35683.jpg", + "image_caption": [ + "Figure 9: Generated point clouds from LION trained jointly over 55 classes of ShapeNet-vol (no conditioning). " + ], + "image_footnote": [], + "bbox": [ + 181, + 839, + 794, + 885 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/7e583fae1e920eca77cfbcfa30b8e62210c56760f2035b8269f5be44b2ae0c30.jpg", + "image_caption": [ + "Figure 10: Samples from LION trained on ShapeNet’s Mug and Bottle classes, and on Turbosquid animals. " + ], + "image_footnote": [], + "bbox": [ + 178, + 71, + 826, + 198 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Mesh Reconstruction. As explained in Sec. 3.1, we combine LION with mesh reconstruction, to directly synthesize practically useful meshes. We show generated meshes in Fig. 2, which look smooth and of high quality. In Fig. 2, we also visually demonstrate how we can vary the local details of synthesized shapes while preserving the overall shape with our diffuse-denoise technique (Sec. 3.1). Details about the number of diffusion steps for all diffuse-denoise experiments are in App. E. ", + "bbox": [ + 174, + 233, + 825, + 303 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Shape Interpolation. As discussed in Sec. 3.1, LION also enables shape interpolation, potentially useful for shape editing applications. We show this in Fig. 7, combined with mesh reconstruction. The generated shapes are clean and semantically plausible along the entire interpolation path. In App. F.12.1, we also show interpolations from PVD [46] and DPM [47] for comparison. ", + "bbox": [ + 174, + 309, + 825, + 364 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5.2 Many-class Unconditional 3D Shape Generation ", + "text_level": 1, + "bbox": [ + 174, + 383, + 547, + 398 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "13-Class LION Model. We train a LION model jointly without any class conditioning on 13 different categories (airplane, chair, car, lamp, table, sofa, cabinet, bench, telephone, loudspeaker, display, watercraft, rifle) from ShapeNet (ShapeNet-vol version). Training a single model without conditioning over such diverse shapes is challenging, as the data distribution is highly complex and multimodal. We show LION’s generated samples in Fig. 3, including meshes: LION synthesizes high-quality and diverse plausible shapes even when trained on such complex data. We report the model’s quantitative generation performance in Tab. 4, and we also trained various strong baseline methods under the same setting for comparison. We find that LION significantly outperforms all baselines by a large margin. We further observe that the hierarchical VAE architecture of LION becomes crucial: The shape latent variable $\\mathbf { z } _ { 0 }$ captures global shape, while the latent points $\\mathbf { h } _ { 0 }$ model details. This can be seen in Fig. 8: we show samples when fixing the global shape latent $\\mathbf { z } _ { 0 }$ and only sample $\\mathbf { h } _ { 0 }$ (details in App. F.3). ", + "bbox": [ + 173, + 406, + 826, + 558 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "55-Class LION Model. Encouraged by these results, we also trained a LION model again jointly without any class conditioning on all 55 different categories from ShapeNet. Note that we did on purpose not use class-conditioning in these experiments to create a difficult 3D generation task and thereby explore LION’s scalability to highly complex and multimodal datasets. We show generated point cloud samples in Fig. 9 (we did not train an SAP model on the 55 classes data): LION synthesizes high-quality and diverse shapes. It can even generate samples from the cap class, which contributes with only 39 training data samples, indicating that LION has an excellent mode coverage that even includes the very rare classes. To the best of our knowledge no previous 3D shape generative models have demonstrated satisfactory generation performance for such diverse and multimodal 3D data without relying on conditioning information (details in App. F.4). In conclusion, we observe that LION out-of-the-box easily scales to highly complex multi-category shape generation. ", + "bbox": [ + 173, + 564, + 825, + 717 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5.3 Training LION on Small Datasets ", + "text_level": 1, + "bbox": [ + 176, + 728, + 446, + 742 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Next, we explore whether LION can also be trained successfully on very small datasets. To this end, we train models on the Mug and Bottle ShapeNet classes. The number of training samples is 149 and 340, respectively, which is much smaller than the common classes like chair, car and airplane. Furthermore, we also train LION on 553 animal assets from the TurboSquid data repository. Generated shapes from the three models are shown in Fig. 10. LION is able to generate correct mugs and bottles as well as diverse and high-quality animal shapes. We conclude that LION also performs well even when training in the challenging low-data setting (details in Apps. F.5 and F.6). ", + "bbox": [ + 174, + 751, + 823, + 779 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/2ade93952b2c1687a4aed247344a7e707a70e0829b0b8547f8b519ee299a389b.jpg", + "image_caption": [ + "Figure 11: Voxel-guided synthesis. We show different methods with 0 and 50 steps of diffuse-denoise. Voxelizations of generated points are also shown: Yellow boxes indicate generated points correctly fill input voxels, green boxes indicate voxels should be filled but are left empty, red boxes indicate extra voxels. " + ], + "image_footnote": [], + "bbox": [ + 187, + 789, + 575, + 915 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/a5ec7783f89ce350c646589554680fe14dc18d0cd5cd5692ac372dc06bdb76d5.jpg", + "image_caption": [ + "Figure 13: Voxel-guided generation. Quality metrics for output points (lower is better) and voxel IOU with respect to input (higher is better). $x -$ - axes denote diffuse-denoise steps. " + ], + "image_footnote": [], + "bbox": [ + 607, + 79, + 812, + 270 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/78f765294bdff9804bb3ae06cde9832dbc910c39d07cd4c4c2e5c738c6154ef6.jpg", + "image_caption": [ + "Figure 12: Reconstruction metrics with respect to clean inputs for airplane category (lower is better) when guiding synthesis with voxelized or noisy inputs (using uniform, outlier, and normal noise, see App. F.7). $_ x$ -axes denote number of diffuse-denoise steps. " + ], + "image_footnote": [], + "bbox": [ + 179, + 77, + 581, + 282 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 356, + 826, + 426 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "5.4 Voxel-guided Shape Synthesis and Denoising with Fine-tuned Encoders ", + "text_level": 1, + "bbox": [ + 174, + 438, + 707, + 453 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Next, we test our strategy for multimodal voxel-guided shape synthesis (see Sec. 3.1) using the airplane class LION model (experiment details in App. E, more experiments in App. F.7). We first voxelize our training set and fine-tune our encoder networks to produce the correct encodings to decode back the original shapes. When processing voxelized shapes with our point-cloud networks, we sample points on the surface of the voxels. As discussed, we can use different numbers of diffusedenoise steps in latent space to generate various plausible shapes and correct for poor encodings. Instead of voxelizations, we can also consider different noisy inputs (we use normal, uniform, and outlier noise, see App. F.7) and achieve multimodal denoising with the same approach. The same tasks can be attempted with the important DDM-based baselines PVD and DPM, by directly—not in a latent space—diffusing and denoising voxelized (converted to point clouds) or noisy point clouds. ", + "bbox": [ + 173, + 460, + 826, + 601 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Fig. 12 shows the reconstruction performance of LION, DPM and PVD for different numbers of diffuse-denoise steps (we voxelized or noised the validation set to measure this). We see that for almost all inputs—voxelized or different noises—LION performs best. PVD and DPM perform acceptably for normal and uniform noise, which is similar to the noise injected during training of their DDMs, but perform very poorly for outlier noise or voxel inputs, which is the most relevant case to us, because voxels can be easily placed by users. It is LION’s unique framework with additional fine-tuned encoders in its VAE and only latent DDMs that makes this possible. Performing more diffuse-denoise steps means that more independent, novel shapes are generated. These will be cleaner and of higher quality, but also correspond less to the noisy or voxel inputs used for guidance. In Fig. 13, we show this trade-off for the voxel-guidance experiment (other experiments in App. F.7), where (top) we measured the outputs’ synthesis quality by calculating 1-NNA with respect to the validation set, and (bottom) the average intersection over union (IOU) between the input voxels and the voxelized outputs. We generally see a trade-off: More diffuse-denoise steps result in lower 1-NNA (better quality), but also lower IOU. LION strikes the best balance by a large gap: Its additional encoder network directly generates plausible latent encodings from the perturbed inputs that are both high quality and also correspond well to the input. This trade-off is visualized in Fig. 11 for LION, DPM, and PVD, where we show generated point clouds and voxelizations (note that performing no diffuse-denoise at all for PVD and DPM corresponds to simply keeping the input, as these models’ DDMs operate directly on point clouds). We see that running 50 diffuse-denoise steps to generate diverse outputs for DPM and especially PVD results in a significant violation of the input voxelization. In contrast, LION generates realistic outputs that also obey the driving voxels. Overall, LION wins out both in this task and also in unconditional generation with large gaps over these previous DDM-based point cloud generative models. We conclude that LION does not only offer state-of-the-art 3D shape generation quality, but is also very versatile. Note that guided synthesis can also be combined with mesh reconstruction, as shown in Fig. 4. ", + "bbox": [ + 173, + 606, + 826, + 772 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/e439f9c4b4cd464eb59193841e41c04c3f2eabc99776306355fe2b44efd72e31.jpg", + "image_caption": [ + "Figure 14: We apply Text2Mesh [49] on meshes generated by LION. In Text2Mesh, textures are generated and meshes refined such that rendered images of the 3D objects are aligned with user-provided text prompts [105]. " + ], + "image_footnote": [], + "bbox": [ + 183, + 785, + 813, + 882 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/81667f186ae68d9e39acfe123525b7508d90e9671430dcd91425046a6551bf91.jpg", + "image_caption": [ + "Figure 16: Text-driven shape generation of chairs and cars with LION. Bottom row is the text prompt used as input. " + ], + "image_footnote": [], + "bbox": [ + 194, + 70, + 563, + 142 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 150, + 826, + 316 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "5.5 Sampling Time ", + "text_level": 1, + "bbox": [ + 174, + 325, + 320, + 340 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "While our main experiments use 1,000-step DDPM-based synthesis, which takes $\\approx 2 7 . 1 2$ seconds, we can significantly accelerate generation without significant loss in quality. Using DDIM-based sampling [106], we can generate high quality shapes in under one second (Fig. 15), which would enable real-time interactive applications. More analyses in App. F.9. ", + "bbox": [ + 174, + 344, + 660, + 414 + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/625795a9b501d90ded76d171008f1ad0b2a315ea890e5a177dc72ef8863de109.jpg", + "image_caption": [ + "Figure 15: 25-step DDIM [106] samples (0.89 seconds per shape). " + ], + "image_footnote": [], + "bbox": [ + 676, + 318, + 818, + 417 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "5.6 Overview of Additional Experiments in Appendix ", + "text_level": 1, + "bbox": [ + 173, + 422, + 560, + 438 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "(i) In App. F.1, we perform various ablation studies. The experiments quantitatively validate LION’s architecture choices and the advantage of our hierarchical VAE setup with conditional latent DDMs. (ii) In App. F.8, we measure LION’s autoencoding performance. (iii) To demonstrate the value of directly outputting meshes, in App. F.10 we use Text2Mesh [49] to generate textures based on text prompts for synthesized LION samples (Fig. 14). This would not be possible, if we only generated point clouds. (iv) To qualitatively show that LION can be adapted easily to other relevant tasks, in App. F.11 we condition LION on CLIP embeddings of the shapes’ rendered images, following CLIP-Forge [34] (Fig. 16). This enables text-driven 3D shape generation and single view 3D reconstruction (Fig. 17). (v) We also show many more samples (Apps. F.2-F.6) and shape interpolations (App. F.12) from our models, more examples of voxel-guided and noise-guided synthesis (App. F.7), and we further analyze our 13-class LION model (App. F.3.2). ", + "bbox": [ + 176, + 441, + 658, + 468 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 469, + 825, + 593 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "6 Conclusions ", + "text_level": 1, + "bbox": [ + 174, + 602, + 307, + 618 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We introduced LION, a novel generative model of 3D shapes. LION uses a VAE framework with hierarchical DDMs in latent space and can be combined with SAP for mesh generation. LION achieves state-of-the-art shape generation performance and enables applications such as voxel-conditioned synthesis, multimodal shape denoising, and shape interpolation. LION is currently trained on 3D point clouds only and can not directly generate textured shapes. A promising extension would be to include image-based training by incorporating neural or differentiable rendering [17, 107–111] and to also synthesize textures [16, 112–114]. Furthermore, LION currently focuses on single object generation only. It would be interesting to extend it to full 3D scene synthesis. Moreover, synthesis could be further accelerated by building on works on accelerated sampling from DDMs [61, 62, 67, 106, 115–121]. ", + "bbox": [ + 173, + 626, + 826, + 751 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Broader Impact. We believe that LION can potentially improve 3D content creation and assist the workflow of digital artists. We designed LION with such applications in mind and hope that it can grow into a practical tool enhancing artists’ creativity. Although we do not see any immediate negative use-cases for LION, it is important that practitioners apply an abundance of caution to mitigate impacts given generative modeling more generally can also be used for malicious purposes, discussed for instance in Vaccari and Chadwick [122], Nguyen et al. [123], Mirsky and Lee [124]. ", + "bbox": [ + 173, + 756, + 826, + 840 + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/2a8dd3f6899aade67630f04ae5ec6f0108def53fe8be061056ed6a9d207f21b7.jpg", + "image_caption": [ + "Figure 17: Single view 3D reconstructions of a car from an RGB image. LION can generate multiple plausible outputs using our diffuse-denoise technique. " + ], + "image_footnote": [], + "bbox": [ + 179, + 864, + 503, + 905 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "References \n[1] Jiajun Wu, Chengkai Zhang, Tianfan Xue, Bill Freeman, and Josh Tenenbaum. Learning a probabilistic latent space of object shapes via 3d generative-adversarial modeling. In Advances in Neural Information Processing Systems, 2016. [2] Panos Achlioptas, Olga Diamanti, Ioannis Mitliagkas, and Leonidas Guibas. Learning representations and generative models for 3D point clouds. In ICML, 2018. \n[3] Chun-Liang Li, Manzil Zaheer, Yang Zhang, Barnabas Poczos, and Ruslan Salakhutdinov. Point cloud gan. arXiv preprint arXiv:1810.05795, 2018. \n[4] Diego Valsesia, Giulia Fracastoro, and Enrico Magli. Learning localized generative models for 3d point clouds via graph convolution. In International Conference on Learning Representations (ICLR) 2019, 2019. [5] Wenlong Huang, Brian Lai, Weijian Xu, and Zhuowen Tu. 3d volumetric modeling with introspective neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33(01), pages 8481–8488, 2019. [6] Dong Wook Shu, Sung Woo Park, and Junseok Kwon. 3d point cloud generative adversarial network based on tree structured graph convolutions. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3859–3868, 2019. \n[7] Zhiqin Chen and Hao Zhang. Learning implicit fields for generative shape modeling. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019. [8] Michael Niemeyer and Andreas Geiger. Giraffe: Representing scenes as compositional generative neural feature fields. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2021. [9] Katja Schwarz, Yiyi Liao, Michael Niemeyer, and Andreas Geiger. Graf: Generative radiance fields for 3d-aware image synthesis. In Advances in Neural Information Processing Systems (NeurIPS), 2020. \n[10] Yiyi Liao, Katja Schwarz, Lars Mescheder, and Andreas Geiger. Towards unsupervised learning of generative models for 3d controllable image synthesis. In Proceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2020. \n[11] Eric Chan, Marco Monteiro, Petr Kellnhofer, Jiajun Wu, and Gordon Wetzstein. pi-gan: Periodic implicit generative adversarial networks for 3d-aware image synthesis. In Proc. CVPR, 2021. \n[12] Eric R. Chan, Connor Z. Lin, Matthew A. Chan, Koki Nagano, Boxiao Pan, Shalini De Mello, Orazio Gallo, Leonidas Guibas, Jonathan Tremblay, Sameh Khamis, Tero Karras, and Gordon Wetzstein. Efficient geometry-aware 3D generative adversarial networks. In arXiv, 2021. \n[13] Roy Or-El, Xuan Luo, Mengyi Shan, Eli Shechtman, Jeong Joon Park, and Ira Kemelmacher-Shlizerman. Stylesdf: High-resolution 3d-consistent image and geometry generation. arXiv preprint arXiv:2112.11427, 2021. \n[14] Jiatao Gu, Lingjie Liu, Peng Wang, and Christian Theobalt. Stylenerf: A style-based 3d aware generator for high-resolution image synthesis. In International Conference on Learning Representations, 2022. \n[15] Peng Zhou, Lingxi Xie, Bingbing Ni, and Qi Tian. CIPS-3D: A 3D-Aware Generator of GANs Based on Conditionally-Independent Pixel Synthesis. arXiv preprint arXiv:2110.09788, 2021. \n[16] Dario Pavllo, Jonas Kohler, Thomas Hofmann, and Aurelien Lucchi. Learning generative models of textured 3d meshes from real-world images. In IEEE/CVF International Conference on Computer Vision (ICCV), 2021. \n[17] Yuxuan Zhang, Wenzheng Chen, Huan Ling, Jun Gao, Yinan Zhang, Antonio Torralba, and Sanja Fidler. Image {gan}s meet differentiable rendering for inverse graphics and interpretable 3d neural rendering. In International Conference on Learning Representations, 2021. \n[18] Moritz Ibing, Isaak Lim, and Leif P. Kobbelt. 3d shape generation with grid-based implicit functions. In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021. \n[19] Ruihui Li, Xianzhi Li, Ke-Hei Hui, and Chi-Wing Fu. SP-GAN:sphere-guided 3d shape generation and manipulation. ACM Transactions on Graphics (Proc. SIGGRAPH), 40(4), 2021. \n[20] A. Luo, T. Li, W. Zhang, and T. Lee. Surfgen: Adversarial 3d shape synthesis with explicit surface discriminators. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021. \n[21] Zhiqin Chen, Vladimir G. Kim, Matthew Fisher, Noam Aigerman, Hao Zhang, and Siddhartha Chaudhuri. Decor-gan: 3d shape detailization by conditional refinement. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021. \n[22] Cheng Wen, Baosheng Yu, and Dacheng Tao. Learning progressive point embeddings for 3d point cloud generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10266–10275, 2021. \n[23] Zekun Hao, Arun Mallya, Serge Belongie, and Ming-Yu Liu. GANcraft: Unsupervised 3D Neural Rendering of Minecraft Worlds. In ICCV, 2021. \n[24] Or Litany, Alex Bronstein, Michael Bronstein, and Ameesh Makadia. Deformable shape completion with graph convolutional autoencoders. CVPR, 2018. \n[25] Qingyang Tan, Lin Gao, Yu-Kun Lai, and Shihong Xia. Variational autoencoders for deforming 3d mesh models. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5841–5850, 2018. \n[26] Kaichun Mo, Paul Guerrero, Li Yi, Hao Su, Peter Wonka, Niloy Mitra, and Leonidas J Guibas. Structurenet: Hierarchical graph networks for 3d shape generation. arXiv preprint arXiv:1908.00575, 2019. \n[27] Lin Gao, Jie Yang, Tong Wu, Yu-Jie Yuan, Hongbo Fu, Yu-Kun Lai, and Hao(Richard) Zhang. SDM-NET: Deep generative network for structured deformable mesh. ACM Transactions on Graphics (Proceedings of ACM SIGGRAPH Asia 2019), 38(6):243:1–243:15, 2019. \n[28] Lin Gao, Tong Wu, Yu-Jie Yuan, Ming-Xian Lin, Yu-Kun Lai, and Hao Zhang. Tm-net: Deep generative networks for textured meshes. ACM Transactions on Graphics (TOG), 40(6):263:1–263:15, 2021. \n[29] Jinwoo Kim, Jaehoon Yoo, Juho Lee, and Seunghoon Hong. Setvae: Learning hierarchical composition for generative modeling of set-structured data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15059–15068, June 2021. \n[30] Paritosh Mittal, Yen-Chi Cheng, Maneesh Singh, and Shubham Tulsiani. AutoSDF: Shape priors for 3d completion, reconstruction and generation. In CVPR, 2022. \n[31] Guandao Yang, Xun Huang, Zekun Hao, Ming-Yu Liu, Serge Belongie, and Bharath Hariharan. PointFlow: 3D point cloud generation with continuous normalizing flows. In ICCV, 2019. \n[32] Hyeongju Kim, Hyeonseung Lee, Woo Hyun Kang, Joun Yeop Lee, and Nam Soo Kim. SoftFlow: Probabilistic framework for normalizing flow on manifolds. In NeurIPS, 2020. \n[33] Roman Klokov, Edmond Boyer, and Jakob Verbeek. Discrete point flow networks for efficient point cloud generation. In ECCV, 2020. \n[34] Aditya Sanghi, Hang Chu, Joseph G Lambourne, Ye Wang, Chin-Yi Cheng, and Marco Fumero. Clipforge: Towards zero-shot text-to-shape generation. arXiv preprint arXiv:2110.02624, 2021. \n[35] Yongbin Sun, Yue Wang, Ziwei Liu, Joshua E Siegel, and Sanjay E Sarma. Pointgrow: Autoregressively learned point cloud generation with self-attention. In Winter Conference on Applications of Computer Vision, 2020. \n[36] Charlie Nash, Yaroslav Ganin, S. M. Ali Eslami, and Peter W. Battaglia. Polygen: An autoregressive generative model of 3d meshes. ICML, 2020. \n[37] Wei-Jan Ko, Hui-Yu Huang, Yu-Liang Kuo, Chen-Yi Chiu, Li-Heng Wang, and Wei-Chen Chiu. Rpg: Learning recursive point cloud generation. arXiv preprint arXiv:2105.14322, 2021. \n[38] Moritz Ibing, Gregor Kobsik, and Leif Kobbelt. Octree transformer: Autoregressive 3d shape generation on hierarchically structured sequences. arXiv preprint arXiv:2111.12480, 2021. \n[39] Jianwen Xie, Zilong Zheng, Ruiqi Gao, Wenguan Wang, Zhu Song-Chun, and Ying Nian Wu. Learning descriptor networks for 3d shape synthesis and analysis. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018. \n[40] Jianwen Xie, Yifei Xu, Zilong Zheng, Ruiqi Gao, Wenguan Wang, Zhu Song-Chun, and Ying Nian Wu. Generative pointnet: Deep energy-based learning on unordered point sets for 3d generation, reconstruction and classification. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021. \n[41] Tianchang Shen, Jun Gao, Kangxue Yin, Ming-Yu Liu, and Sanja Fidler. Deep marching tetrahedra: a hybrid representation for high-resolution 3d shape synthesis. In Advances in Neural Information Processing Systems (NeurIPS), 2021. \n[42] Kangxue Yin, Jun Gao, Maria Shugrina, Sameh Khamis, and Sanja Fidler. 3dstylenet: Creating 3d shapes with geometric and texture style variations. In Proceedings of International Conference on Computer Vision (ICCV), 2021. \n[43] Dongsu Zhang, Changwoon Choi, Jeonghwan Kim, and Young Min Kim. Learning to generate 3d shapes with generative cellular automata. In International Conference on Learning Representations, 2021. \n[44] Chen Chao, Zhizhong Han, Yu-Shen Liu, and Matthias Zwicker. Unsupervised learning of fine structure generation for 3d point clouds by 2d projection matching. arXiv preprint arXiv:2108.03746, 2021. \n[45] Ruojin Cai, Guandao Yang, Hadar Averbuch-Elor, Zekun Hao, Serge Belongie, Noah Snavely, and Bharath Hariharan. Learning gradient fields for shape generation. In Proceedings of the European Conference on Computer Vision (ECCV), 2020. \n[46] Linqi Zhou, Yilun Du, and Jiajun Wu. 3d shape generation and completion through point-voxel diffusion. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021. \n[47] Shitong Luo and Wei Hu. Diffusion probabilistic models for 3d point cloud generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021. \n[48] Yu Deng, Jiaolong Yang, and Xin Tong. Deformed implicit field: Modeling 3d shapes with learned dense correspondence. In IEEE Computer Vision and Pattern Recognition, 2021. \n[49] Oscar Michel, Roi Bar-On, Richard Liu, Sagie Benaim, and Rana Hanocka. Text2mesh: Text-driven neural stylization for meshes. arXiv preprint arXiv:2112.03221, 2021. \n[50] Ajay Jain, Ben Mildenhall, Jonathan T. Barron, Pieter Abbeel, and Ben Poole. Zero-shot text-guided object generation with dream fields. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. \n[51] Nasir Khalid, Tianhao Xie, Eugene Belilovsky, and Tiberiu Popa. Text to mesh without 3d supervision using limit subdivision. arXiv preprint arXiv:2203.13333, 2022. \n[52] Le Hui, Rui Xu, Jin Xie, Jianjun Qian, and Jian Yang. Progressive point cloud deconvolution generation network. In ECCV, 2020. \n[53] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, 2020. \n[54] Jonathan Ho, Chitwan Saharia, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans. Cascaded diffusion models for high fidelity image generation. arXiv preprint arXiv:2106.15282, 2021. \n[55] Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. In International Conference on Machine Learning, 2021. \n[56] Prafulla Dhariwal and Alexander Quinn Nichol. Diffusion models beat GANs on image synthesis. In Advances in Neural Information Processing Systems, 2021. \n[57] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. arXiv preprint arXiv:2112.10752, 2021. \n[58] Arash Vahdat, Karsten Kreis, and Jan Kautz. Score-based generative modeling in latent space. In Advances in Neural Information Processing Systems, 2021. \n[59] Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. Glide: Towards photorealistic image generation and editing with text-guided diffusion models. arXiv preprint arXiv:2112.10741, 2021. \n[60] Konpat Preechakul, Nattanat Chatthee, Suttisak Wizadwongsa, and Supasorn Suwajanakorn. Diffusion autoencoders: Toward a meaningful and decodable representation. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2022. \n[61] Tim Dockhorn, Arash Vahdat, and Karsten Kreis. Score-based generative modeling with critically-damped langevin diffusion. In International Conference on Learning Representations (ICLR), 2022. \n[62] Zhisheng Xiao, Karsten Kreis, and Arash Vahdat. Tackling the generative learning trilemma with denoising diffusion GANs. In International Conference on Learning Representations (ICLR), 2022. \n[63] Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv preprint arXiv:2204.06125, 2022. \n[64] Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed Kamyar Seyed Ghasemipour, Burcu Karagol Ayan, S. Sara Mahdavi, Rapha Gontijo Lopes, Tim Salimans, Jonathan Ho, David J Fleet, and Mohammad Norouzi. Photorealistic text-to-image diffusion models with deep language understanding. arXiv preprint arXiv:2205.11487, 2022. \n[65] Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International Conference on Machine Learning, 2015. \n[66] Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. In Proceedings of the 33rd Annual Conference on Neural Information Processing Systems, 2019. \n[67] Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021. \n[68] Songyou Peng, Chiyu \"Max\" Jiang, Yiyi Liao, Michael Niemeyer, Marc Pollefeys, and Andreas Geiger. Shape as points: A differentiable poisson solver. In Advances in Neural Information Processing Systems (NeurIPS), 2021. \n[69] Diederik P Kingma, Tim Salimans, Ben Poole, and Jonathan Ho. Variational diffusion models. In Advances in Neural Information Processing Systems, 2021. \n[70] Diederik P Kingma and Max Welling. Auto-encoding variational bayes. In The International Conference on Learning Representations, 2014. \n[71] Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and approximate inference in deep generative models. In International Conference on Machine Learning, 2014. \n[72] Jakub Tomczak and Max Welling. Vae with a vampprior. In International Conference on Artificial Intelligence and Statistics, pages 1214–1223, 2018. \n[73] Hiroshi Takahashi, Tomoharu Iwata, Yuki Yamanaka, Masanori Yamada, and Satoshi Yagi. Variational autoencoder with implicit optimal priors. Proceedings of the AAAI Conference on Artificial Intelligence, 33(01):5066–5073, Jul. 2019. \n[74] Matthias Bauer and Andriy Mnih. Resampled priors for variational autoencoders. In Kamalika Chaudhuri and Masashi Sugiyama, editors, Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, volume 89 of Proceedings of Machine Learning Research, pages 66–75. PMLR, 16–18 Apr 2019. \n[75] Arash Vahdat and Jan Kautz. NVAE: A deep hierarchical variational autoencoder. In Advances in Neural Information Processing Systems, 2020. \n[76] Jyoti Aneja, Alexander Schwing, Jan Kautz, and Arash Vahdat. NCP-VAE: Variational autoencoders with noise contrastive priors. In Advances in Neural Information Processing Systems, 2021. \n[77] Abhishek Sinha, Jiaming Song, Chenlin Meng, and Stefano Ermon. D2c: Diffusion-denoising models for few-shot conditional generation. In Advances in Neural Information Processing Systems, 2021. \n[78] Mihaela Rosca, Balaji Lakshminarayanan, and Shakir Mohamed. Distribution matching in variational inference. arXiv preprint arXiv:1802.06847, 2018. \n[79] Matthew D Hoffman and Matthew J Johnson. Elbo surgery: yet another way to carve up the variational evidence lower bound. In Workshop in Advances in Approximate Bayesian Inference, NeurIPS, 2016. \n[80] Zhijian Liu, Haotian Tang, Yujun Lin, and Song Han. Point-voxel cnn for efficient 3d deep learning. In Advances in Neural Information Processing Systems, 2019. \n[81] Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017. \n[82] Charles R Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. In Advances in Neural Information Processing Systems, 2017. \n[83] Kaiming He, X. Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. \n[84] Yuxin Wu and Kaiming He. Group normalization. In Proceedings of the European Conference on Computer Vision (ECCV), 2018. \n[85] Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon. SDEdit: Guided image synthesis and editing with stochastic differential equations. In International Conference on Learning Representations, 2022. \n[86] Nanxin Chen, Yu Zhang, Heiga Zen, Ron J Weiss, Mohammad Norouzi, and William Chan. Wavegrad: Estimating gradients for waveform generation. In International Conference on Learning Representations (ICLR), 2021. \n[87] Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao, and Bryan Catanzaro. DiffWave: A Versatile Diffusion Model for Audio Synthesis. In International Conference on Learning Representations, 2021. \n[88] Myeonghun Jeong, Hyeongju Kim, Sung Jun Cheon, Byoung Jin Choi, and Nam Soo Kim. Diff-tts: A denoising diffusion model for text-to-speech. arXiv preprint arXiv:2104.01409, 2021. \n[89] Nanxin Chen, Yu Zhang, Heiga Zen, Ron J. Weiss, Mohammad Norouzi, Najim Dehak, and William Chan. Wavegrad 2: Iterative refinement for text-to-speech synthesis. arXiv preprint arXiv:2106.09660, 2021. \n[90] Vadim Popov, Ivan Vovk, Vladimir Gogoryan, Tasnima Sadekova, and Mikhail Kudinov. Grad-tts: A diffusion probabilistic model for text-to-speech. In International Conference on Machine Learning, 2021. \n[91] Songxiang Liu, Dan Su, and Dong Yu. Diffgan-tts: High-fidelity and efficient text-to-speech with denoising diffusion gans. arXiv preprint arXiv:2201.11972, 2022. \n[92] Gautam Mittal, Jesse Engel, Curtis Hawthorne, and Ian Simon. Symbolic music generation with diffusion models. In Proceedings of the 22nd International Society for Music Information Retrieval Conference, 2021. \n[93] Kushagra Pandey, Avideep Mukherjee, Piyush Rai, and Abhishek Kumar. Diffusevae: Efficient, controllable and high-fidelity generation from low-dimensional latents. arXiv preprint arXiv:2201.00308, 2022. \n[94] Danilo Jimenez Rezende and Shakir Mohamed. Variational inference with normalizing flows. In International Conference on Machine Learning, 2015. \n[95] Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real NVP. In International Conference on Learning Representations ICLR, 2017. \n[96] Ricky T. Q. Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations. Advances in Neural Information Processing Systems, 2018. \n[97] Will Grathwohl, Ricky T. Q. Chen, Jesse Bettencourt, Ilya Sutskever, and David Duvenaud. FFJORD: Free-form continuous dynamics for scalable reversible generative models. In International Conference on Learning Representations, 2019. \n[98] Zhaoyang Lyu, Zhifeng Kong, Xudong XU, Liang Pan, and Dahua Lin. A conditional point diffusionrefinement paradigm for 3d point cloud completion. In International Conference on Learning Representations (ICLR), 2022. \n[99] Thibault Groueix, Matthew Fisher, Vladimir G. Kim, Bryan Russell, and Mathieu Aubry. AtlasNet: A Papier-Mâché Approach to Learning 3D Surface Generation. In Proceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2018. \n[100] Lars Mescheder, Michael Oechsle, Michael Niemeyer, Sebastian Nowozin, and Andreas Geiger. Occupancy networks: Learning 3d reconstruction in function space. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4460–4470, 2019. \n[101] Songyou Peng, Michael Niemeyer, Lars Mescheder, and Andreas Geiger Marc Pollefeys. Convolutional occupancy networks. In European Conference on Computer Vision (ECCV), 2020. ", + "bbox": [ + 174, + 82, + 828, + 920 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 178, + 53, + 828, + 920 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "", + "bbox": [ + 176, + 87, + 828, + 917 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "", + "bbox": [ + 178, + 50, + 828, + 919 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 65, + 828, + 917 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "[102] Francis Williams, Matthew Trager, Joan Bruna, and Denis Zorin. Neural splines: Fitting 3d surfaces with infinitely-wide neural networks. In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021. ", + "bbox": [ + 173, + 92, + 823, + 131 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[103] Francis Williams, Zan Gojcic, Sameh Khamis, Denis Zorin, Joan Bruna, Sanja Fidler, and Or Litany. Neural fields as learnable kernels for 3d reconstruction. arXiv preprint arXiv:2111.13674, 2021. ", + "bbox": [ + 169, + 140, + 823, + 167 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[104] Angel X. Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. ShapeNet: An Information-Rich 3D Model Repository. Technical Report arXiv:1512.03012 [cs.GR], Stanford University — Princeton University — Toyota Technological Institute at Chicago, 2015. ", + "bbox": [ + 174, + 176, + 825, + 229 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[105] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. In International Conference on Machine Learning, ICML, 2021. ", + "bbox": [ + 173, + 238, + 826, + 290 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[106] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In International Conference on Learning Representations, 2021. ", + "bbox": [ + 171, + 299, + 823, + 327 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[107] Wenzheng Chen, Huan Ling, Jun Gao, Edward Smith, Jaakko Lehtinen, Alec Jacobson, and Sanja Fidler. Learning to predict 3d objects with an interpolation-based differentiable renderer. In Advances in Neural Information Processing Systems, 2019. ", + "bbox": [ + 173, + 337, + 825, + 376 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[108] Samuli Laine, Janne Hellsten, Tero Karras, Yeongho Seol, Jaakko Lehtinen, and Timo Aila. Modular primitives for high-performance differentiable rendering. ACM Transactions on Graphics, 2020. ", + "bbox": [ + 173, + 386, + 823, + 412 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[109] Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. In ECCV, 2020. ", + "bbox": [ + 171, + 421, + 823, + 449 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[110] Wenzheng Chen, Joey Litalien, Jun Gao, Zian Wang, Clement Fuji Tsang, Sameh Khamis, Or Litany, and Sanja Fidler. DIB- $\\mathbf { \\cdot R } { + + }$ : Learning to predict lighting and material with a hybrid differentiable renderer. In Advances in Neural Information Processing Systems (NeurIPS), 2021. ", + "bbox": [ + 173, + 458, + 823, + 497 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[111] Ayush Tewari, Justus Thies, Ben Mildenhall, Pratul Srinivasan, Edgar Tretschk, Yifan Wang, Christoph Lassner, Vincent Sitzmann, Ricardo Martin-Brualla, Stephen Lombardi, Tomas Simon, Christian Theobalt, Matthias Niessner, Jonathan T. Barron, Gordon Wetzstein, Michael Zollhoefer, and Vladislav Golyanik. Advances in neural rendering. arXiv preprint arXiv:2111.05849, 2021. ", + "bbox": [ + 173, + 507, + 825, + 559 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[112] Michael Oechsle, Lars Mescheder, Michael Niemeyer, Thilo Strauss, and Andreas Geiger. Texture fields: Learning texture representations in function space. In Proceedings IEEE International Conf. on Computer Vision (ICCV), 2019. ", + "bbox": [ + 171, + 568, + 823, + 608 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[113] Zhiqin Chen, Kangxue Yin, and Sanja Fidler. Auv-net: Learning aligned uv maps for texture transfer and synthesis. In The Conference on Computer Vision and Pattern Recognition (CVPR), 2022. ", + "bbox": [ + 171, + 617, + 823, + 645 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[114] Yawar Siddiqui, Justus Thies, Fangchang Ma, Qi Shan, Matthias Nießner, and Angela Dai. Texturify: Generating textures on 3d shape surfaces. arXiv preprint arXiv:2204.02411, 2022. ", + "bbox": [ + 171, + 654, + 823, + 681 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[115] Daniel Watson, Jonathan Ho, Mohammad Norouzi, and William Chan. Learning to efficiently sample from diffusion probabilistic models. arXiv preprint arXiv:2106.03802, 2021. ", + "bbox": [ + 173, + 690, + 823, + 717 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[116] Zhifeng Kong and Wei Ping. On fast sampling of diffusion probabilistic models. arXiv preprint arXiv:2106.00132, 2021. ", + "bbox": [ + 173, + 727, + 823, + 753 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[117] Alexia Jolicoeur-Martineau, Ke Li, Rémi Piché-Taillefer, Tal Kachman, and Ioannis Mitliagkas. Gotta Go Fast When Generating Data with Score-Based Models. arXiv preprint arXiv:2105.14080, 2021. ", + "bbox": [ + 173, + 763, + 823, + 790 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[118] Daniel Watson, William Chan, Jonathan Ho, and Mohammad Norouzi. Learning Fast Samplers for Diffusion Models by Differentiating Through Sample Quality. In International Conference on Learning Representations, 2022. ", + "bbox": [ + 173, + 799, + 823, + 838 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[119] Luping Liu, Yi Ren, Zhijie Lin, and Zhou Zhao. Pseudo numerical methods for diffusion models on manifolds. In International Conference on Learning Representations (ICLR), 2022. ", + "bbox": [ + 171, + 848, + 821, + 876 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[120] Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. In International Conference on Learning Representations (ICLR), 2022. ", + "bbox": [ + 171, + 885, + 825, + 911 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "[121] Fan Bao, Chongxuan Li, Jun Zhu, and Bo Zhang. Analytic-DPM: an Analytic Estimate of the Optimal Reverse Variance in Diffusion Probabilistic Models. In International Conference on Learning Representations, 2022. \n[122] Cristian Vaccari and Andrew Chadwick. Deepfakes and disinformation: Exploring the impact of synthetic political video on deception, uncertainty, and trust in news. Social Media $^ +$ Society, 6(1): 2056305120903408, 2020. \n[123] Thanh Thi Nguyen, Quoc Viet Hung Nguyen, Cuong M. Nguyen, Dung Nguyen, Duc Thanh Nguyen, and Saeid Nahavandi. Deep learning for deepfakes creation and detection: A survey. arXiv preprint arXiv:1909.11573, 2021. \n[124] Yisroel Mirsky and Wenke Lee. The creation and detection of deepfakes: A survey. ACM Comput. Surv., 54(1), 2021. \n[125] Merlin Nimier-David, Delio Vicini, Tizian Zeltner, and Wenzel Jakob. Mitsuba 2: a retargetable forward and inverse renderer. ACM Transactions on Graphics, 38:1–17, 11 2019. doi: 10.1145/3355089.3356498. \n[126] Jonathan Krause, Michael Stark, Jia Deng, and Li Fei-Fei. 3d object representations for fine-grained categorization. In 4th International IEEE Workshop on 3D Representation and Recognition (3dRR-13), Sydney, Australia, 2013. \n[127] Sungjoon Choi, Qian-Yi Zhou, Stephen Miller, and Vladlen Koltun. A large dataset of object scans. arXiv:1602.02481, 2016. \n[128] Xingyuan Sun, Jiajun Wu, Xiuming Zhang, Zhoutong Zhang, Chengkai Zhang, Tianfan Xue, Joshua B Tenenbaum, and William T Freeman. Pix3d: Dataset and methods for single-image 3d shape modeling. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018. \n[129] Brian DO Anderson. Reverse-time diffusion equation models. Stochastic Processes and their Applications, 12(3):313–326, 1982. \n[130] Ulrich G Haussmann and Etienne Pardoux. Time reversal of diffusions. The Annals of Probability, pages 1188–1205, 1986. \n[131] Pascal Vincent. A connection between score matching and denoising autoencoders. Neural Computation, 23(7):1661–1674, 2011. \n[132] Radford M Neal and Geoffrey E Hinton. A view of the em algorithm that justifies incremental, sparse, and other variants. In Learning in graphical models, pages 355–368. Springer, 1998. \n[133] Diederik P. Kingma and Max Welling. An introduction to variational autoencoders. Foundations and Trends in Machine Learning, 12(4):307–392, 2019. \n[134] J. R. Dormand and P. J. Prince. A family of embedded runge–kutta formulae. Journal of Computational and Applied Mathematics, 6(1):19–26, 1980. \n[135] Michael Kazhdan, Matthew Bolitho, and Hugues Hoppe. Poisson surface reconstruction. In Proceedings of the fourth Eurographics symposium on Geometry processing, volume 7, 2006. \n[136] William E. Lorensen and Harvey E. Cline. Marching cubes: A high resolution 3d surface construction algorithm. In Proceedings of the 14th Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH ’87, page 163–169, New York, NY, USA, 1987. Association for Computing Machinery. \n[137] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7132–7141, 2018. \n[138] Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative adversarial networks. In International Conference on Learning Representations, 2018. \n[139] Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research, 9(11), 2008. \n[140] Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E. Sarma, Michael M. Bronstein, and Justin M. Solomon. Dynamic graph cnn for learning on point clouds. ACM Transactions on Graphics (TOG), 2019. \n[141] Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 16259–16268, 2021. ", + "bbox": [ + 171, + 78, + 828, + 917 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Checklist ", + "text_level": 1, + "bbox": [ + 174, + 89, + 254, + 106 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "1. For all authors... ", + "bbox": [ + 214, + 121, + 339, + 135 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? [Yes] \n(b) Did you describe the limitations of your work? [Yes] Please see Sec. 6. \n(c) Did you discuss any potential negative societal impacts of your work? [Yes] Please see Sec. 6. \n(d) Have you read the ethics review guidelines and ensured that your paper conforms to them? [Yes] ", + "bbox": [ + 238, + 138, + 825, + 244 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "2. If you are including theoretical results... ", + "bbox": [ + 214, + 248, + 493, + 262 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "(a) Did you state the full set of assumptions of all theoretical results? [N/A] We did not derive novel theoretical results. We rather propose a novel generative model of 3D shapes. \n(b) Did you include complete proofs of all theoretical results? [N/A] ", + "bbox": [ + 238, + 267, + 825, + 325 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "3. If you ran experiments... ", + "bbox": [ + 214, + 330, + 393, + 344 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "(a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [No] We will release code and instructions to reproduce all experiments upon acceptance of the manuscript. The internal guidelines of our institution prevent us from releasing code at this stage. \n(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] We provide all model training and evaluation details in the App. D, including all hyperparameters. \n(c) Did you report error bars (e.g., with respect to the random seed after running experiments multiple times)? [No] Following standard conventions in the related 3D generative modeling literature, we do not report error bars. Furthermore, we avoid running similar setups repeatedly to save computational resources (our main models are quite large and require substantial GPU resources for training, see App. E.9). \n(d) Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] Please see App. E.9. ", + "bbox": [ + 238, + 348, + 826, + 549 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets... ", + "bbox": [ + 222, + 553, + 823, + 569 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "(a) If your work uses existing assets, did you cite the creators? [Yes] For baseline comparisons we run publicly available code from previous, publicly available papers, which we cite. We also use various datasets. Here, we present a summary. To compare to baselines, we use the following codes: • r-GAN, l-GAN [2]: https://github.com/optas/latent_3d_points (MIT License) • PointFlow [31]: https://github.com/stevenygd/PointFlow (MIT License) • SoftFlow [32]: https://github.com/ANLGBOY/SoftFlow • Set-VAE [29]: https://github.com/jw9730/setvae (MIT License) • DPF-NET [33]: https://github.com/Regenerator/dpf-nets • DPM [47]: https://github.com/luost26/diffusion-point-cloud (MIT License) • PVD [46]: https://github.com/alexzhou907/PVD (MIT License) • ShapeGF [45]: https://github.com/RuojinCai/ShapeGF (MIT License) • SP-GAN [19]: https://github.com/liruihui/sp-gan (MIT License) • PDGN [52]: https://github.com/fpthink/PDGN (MIT License) • IM-GAN [7]: https://github.com/czq142857/implicit-decoder (MIT license) and https://github.com/czq142857/IM-NET-pytorch (MIT license) • GCA [43]: https://github.com/96lives/gca (MIT license) ", + "bbox": [ + 236, + 571, + 826, + 852 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "We use further codebases in other places: ", + "bbox": [ + 266, + 854, + 535, + 867 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "• We use the MitSuba renderer for visualizations [125]: https: //github.com/mitsuba-renderer/mitsuba2 (License: https: //github.com/mitsuba-renderer/mitsuba2/blob/master/LICENSE), ", + "bbox": [ + 276, + 869, + 826, + 911 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "and the code to generate the scene discription files for MitSuba [31]: https://github.com/zekunhao1995/PointFlowRenderer. \n• We rely on SAP [68] for mesh generation with the code at https://github.com/ autonomousvision/shape_as_points (MIT License). \n• For calculating the evaluation metrics, we use the implementation for CD at https: //github.com/ThibaultGROUEIX/ChamferDistancePytorch (MIT License) and for EMD at https://github.com/daerduoCarey/PyTorchEMD. \n• We use Text2Mesh [49] for per-sample text-driven texture synthesis: https:// github.com/threedle/text2mesh (MIT License) ", + "bbox": [ + 276, + 92, + 826, + 219 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "We also rely on the following datasets: ", + "bbox": [ + 266, + 222, + 519, + 236 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "• ShapeNet [104]. Its terms of use can be found at https://shapenet.org/ terms. \n• The Cars dataset [126] from http://ai.stanford.edu/\\~jkrause/cars/car_ dataset.html with ImageNet License: https://image-net.org/download. php. \n• The TurboSquid data repository, https://www.turbosquid.com. We obtained a custom license from TurboSquid. \n• Redwood 3DScan Dataset [127]: https://github.com/isl-org/ redwood-3dscan (Public Domain) \n• Pix3D [128]: https://github.com/xingyuansun/pix3d. (Creative Commons Attribution 4.0 International License). \n(b) Did you mention the license of the assets? [Yes] In App. E.8, we mention the licenses of the codes and other assets we are using. \n(c) Did you include any new assets either in the supplemental material or as a URL? [No] \n(d) Did you discuss whether and how consent was obtained from people whose data you’re using/curating? [N/A] \n(e) Did you discuss whether the data you are using/curating contains personally identifiable information or offensive content? [No] We are primarily using the publicly available ShapeNet [104] dataset, which has been widely used in the generative modeling literature as standard benchmark. It only consists of simple 3D models of shapes such as airplanes, chairs, cars, etc. ", + "bbox": [ + 274, + 238, + 826, + 393 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "", + "bbox": [ + 238, + 397, + 825, + 542 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "5. If you used crowdsourcing or conducted research with human subjects... ", + "bbox": [ + 214, + 547, + 704, + 563 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "(a) Did you include the full text of instructions given to participants and screenshots, if applicable? [N/A] \n(b) Did you describe any potential participant risks, with links to Institutional Review Board (IRB) approvals, if applicable? [N/A] \n(c) Did you include the estimated hourly wage paid to participants and the total amount spent on participant compensation? [N/A] ", + "bbox": [ + 238, + 565, + 825, + 655 + ], + "page_idx": 18 + } +] \ No newline at end of file diff --git a/parse/dev/tHK5ntjp-5K/tHK5ntjp-5K_middle.json b/parse/dev/tHK5ntjp-5K/tHK5ntjp-5K_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..53666a678ba80e32b9dd79892416233495837c72 --- /dev/null +++ b/parse/dev/tHK5ntjp-5K/tHK5ntjp-5K_middle.json @@ -0,0 +1,54862 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 166, + 97, + 444, + 137 + ], + "lines": [ + { + "bbox": [ + 165, + 97, + 446, + 116 + ], + "spans": [ + { + "bbox": [ + 165, + 97, + 446, + 116 + ], + "score": 1.0, + "content": "LION: Latent Point Diffusion Models", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 210, + 117, + 400, + 137 + ], + "spans": [ + { + "bbox": [ + 210, + 117, + 400, + 137 + ], + "score": 1.0, + "content": "for 3D Shape Generation", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 172, + 178, + 437, + 191 + ], + "lines": [ + { + "bbox": [ + 171, + 177, + 438, + 192 + ], + "spans": [ + { + "bbox": [ + 171, + 177, + 258, + 192 + ], + "score": 1.0, + "content": "Xiaohui Zeng1,2,3,∗", + "type": "text" + }, + { + "bbox": [ + 271, + 177, + 339, + 191 + ], + "score": 1.0, + "content": "Arash Vahdat1", + "type": "text" + }, + { + "bbox": [ + 357, + 178, + 438, + 192 + ], + "score": 1.0, + "content": "Francis Williams1", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 154, + 206, + 455, + 219 + ], + "lines": [ + { + "bbox": [ + 152, + 204, + 457, + 222 + ], + "spans": [ + { + "bbox": [ + 152, + 205, + 209, + 222 + ], + "score": 1.0, + "content": "Zan Gojcic1", + "type": "text" + }, + { + "bbox": [ + 227, + 205, + 278, + 222 + ], + "score": 1.0, + "content": "Or Litany1", + "type": "text" + }, + { + "bbox": [ + 296, + 204, + 372, + 222 + ], + "score": 1.0, + "content": "Sanja Fidler1,2,3", + "type": "text" + }, + { + "bbox": [ + 389, + 206, + 457, + 221 + ], + "score": 1.0, + "content": "Karsten Kreis1", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 180, + 229, + 430, + 253 + ], + "lines": [ + { + "bbox": [ + 204, + 226, + 406, + 243 + ], + "spans": [ + { + "bbox": [ + 204, + 226, + 406, + 243 + ], + "score": 1.0, + "content": "1NVIDIA 2University of Toronto 3Vector Institute", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 180, + 243, + 432, + 254 + ], + "spans": [ + { + "bbox": [ + 180, + 243, + 432, + 254 + ], + "score": 1.0, + "content": "{xzeng,avahdat,fwilliams,zgojcic,olitany,sfidler,kkreis}@nvidia.com", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "title", + "bbox": [ + 283, + 281, + 328, + 294 + ], + "lines": [ + { + "bbox": [ + 281, + 280, + 331, + 296 + ], + "spans": [ + { + "bbox": [ + 281, + 280, + 331, + 296 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 143, + 307, + 469, + 526 + ], + "lines": [ + { + "bbox": [ + 141, + 307, + 469, + 320 + ], + "spans": [ + { + "bbox": [ + 141, + 307, + 469, + 320 + ], + "score": 1.0, + "content": "Denoising diffusion models (DDMs) have shown promising results in 3D point", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 318, + 469, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 318, + 469, + 331 + ], + "score": 1.0, + "content": "cloud synthesis. To advance 3D DDMs and make them useful for digital artists, we", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 329, + 469, + 342 + ], + "spans": [ + { + "bbox": [ + 141, + 329, + 469, + 342 + ], + "score": 1.0, + "content": "require (i) high generation quality, (ii) flexibility for manipulation and applications", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 341, + 470, + 352 + ], + "spans": [ + { + "bbox": [ + 141, + 341, + 470, + 352 + ], + "score": 1.0, + "content": "such as conditional synthesis and shape interpolation, and (iii) the ability to output", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 351, + 470, + 363 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 470, + 363 + ], + "score": 1.0, + "content": "smooth surfaces or meshes. To this end, we introduce the hierarchical Latent Point", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 362, + 469, + 374 + ], + "spans": [ + { + "bbox": [ + 141, + 362, + 469, + 374 + ], + "score": 1.0, + "content": "Diffusion Model (LION) for 3D shape generation. LION is set up as a variational", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 373, + 469, + 385 + ], + "spans": [ + { + "bbox": [ + 141, + 373, + 469, + 385 + ], + "score": 1.0, + "content": "autoencoder (VAE) with a hierarchical latent space that combines a global shape", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 384, + 469, + 396 + ], + "spans": [ + { + "bbox": [ + 141, + 384, + 469, + 396 + ], + "score": 1.0, + "content": "latent representation with a point-structured latent space. For generation, we train", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 395, + 469, + 407 + ], + "spans": [ + { + "bbox": [ + 141, + 395, + 469, + 407 + ], + "score": 1.0, + "content": "two hierarchical DDMs in these latent spaces. The hierarchical VAE approach", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 405, + 470, + 419 + ], + "spans": [ + { + "bbox": [ + 141, + 405, + 470, + 419 + ], + "score": 1.0, + "content": "boosts performance compared to DDMs that operate on point clouds directly,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 416, + 471, + 429 + ], + "spans": [ + { + "bbox": [ + 141, + 416, + 471, + 429 + ], + "score": 1.0, + "content": "while the point-structured latents are still ideally suited for DDM-based modeling.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 427, + 470, + 440 + ], + "spans": [ + { + "bbox": [ + 141, + 427, + 470, + 440 + ], + "score": 1.0, + "content": "Experimentally, LION achieves state-of-the-art generation performance on multiple", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "spans": [ + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "score": 1.0, + "content": "ShapeNet benchmarks. Furthermore, our VAE framework allows us to easily use", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 448, + 470, + 462 + ], + "spans": [ + { + "bbox": [ + 141, + 448, + 470, + 462 + ], + "score": 1.0, + "content": "LION for different relevant tasks: LION excels at multimodal shape denoising and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 460, + 470, + 473 + ], + "spans": [ + { + "bbox": [ + 141, + 460, + 470, + 473 + ], + "score": 1.0, + "content": "voxel-conditioned synthesis, and it can be adapted for text- and image-driven 3D", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 472, + 470, + 483 + ], + "spans": [ + { + "bbox": [ + 142, + 472, + 470, + 483 + ], + "score": 1.0, + "content": "generation. We also demonstrate shape autoencoding and latent shape interpolation,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 482, + 470, + 495 + ], + "spans": [ + { + "bbox": [ + 141, + 482, + 470, + 495 + ], + "score": 1.0, + "content": "and we augment LION with modern surface reconstruction techniques to generate", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 493, + 469, + 505 + ], + "spans": [ + { + "bbox": [ + 141, + 493, + 469, + 505 + ], + "score": 1.0, + "content": "smooth 3D meshes. We hope that LION provides a powerful tool for artists", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 504, + 469, + 516 + ], + "spans": [ + { + "bbox": [ + 142, + 504, + 469, + 516 + ], + "score": 1.0, + "content": "working with 3D shapes due to its high-quality generation, flexibility, and surface", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 515, + 465, + 527 + ], + "spans": [ + { + "bbox": [ + 141, + 515, + 465, + 527 + ], + "score": 1.0, + "content": "reconstruction. Project page and code: https://nv-tlabs.github.io/LION.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 107, + 545, + 190, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 192, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 192, + 560 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 565, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 106, + 565, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 505, + 577 + ], + "score": 1.0, + "content": "Generative modeling of 3D shapes has extensive applications in 3D content creation and has become", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "an active area of research [1–52]. However, to be useful as a tool for digital artists, generative models", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 587, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 506, + 599 + ], + "score": 1.0, + "content": "of 3D shapes have to fulfill several criteria: (i) Generated shapes need to be realistic and of high-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 597, + 506, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 610 + ], + "score": 1.0, + "content": "quality without artifacts. (ii) The model should enable flexible and interactive use and refinement: For", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 609, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 621 + ], + "score": 1.0, + "content": "example, a user may want to refine a generated shape and synthesize versions with varying details.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 619, + 506, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 506, + 633 + ], + "score": 1.0, + "content": "Or an artist may provide a coarse or noisy input shape, thereby guiding the model to produce multiple", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 631, + 505, + 643 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 643 + ], + "score": 1.0, + "content": "realistic high-quality outputs. Similarly, a user may want to interpolate different shapes. (iii) The", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 641, + 506, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 506, + 654 + ], + "score": 1.0, + "content": "model should output smooth meshes, which are the standard representation in most graphics software.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 109, + 658, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 670 + ], + "score": 1.0, + "content": "Existing 3D generative models build on various frameworks, including generative adversarial net-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 669, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 107, + 669, + 506, + 681 + ], + "score": 1.0, + "content": "works (GANs) [1–23], variational autoencoders (VAEs) [24–30], normalizing flows [31–34], autore-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 680, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 506, + 692 + ], + "score": 1.0, + "content": "gressive models [35–38], and more [39–44]. Most recently, denoising diffusion models (DDMs)", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 119, + 701, + 272, + 711 + ], + "lines": [ + { + "bbox": [ + 118, + 699, + 273, + 713 + ], + "spans": [ + { + "bbox": [ + 118, + 699, + 273, + 713 + ], + "score": 1.0, + "content": "∗Work done during internship at NVIDIA.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 731, + 385, + 742 + ], + "lines": [ + { + "bbox": [ + 106, + 730, + 386, + 743 + ], + "spans": [ + { + "bbox": [ + 106, + 730, + 386, + 743 + ], + "score": 1.0, + "content": "36th Conference on Neural Information Processing Systems (NeurIPS 2022).", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 166, + 97, + 444, + 137 + ], + "lines": [ + { + "bbox": [ + 165, + 97, + 446, + 116 + ], + "spans": [ + { + "bbox": [ + 165, + 97, + 446, + 116 + ], + "score": 1.0, + "content": "LION: Latent Point Diffusion Models", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 210, + 117, + 400, + 137 + ], + "spans": [ + { + "bbox": [ + 210, + 117, + 400, + 137 + ], + "score": 1.0, + "content": "for 3D Shape Generation", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 172, + 178, + 437, + 191 + ], + "lines": [ + { + "bbox": [ + 171, + 177, + 438, + 192 + ], + "spans": [ + { + "bbox": [ + 171, + 177, + 258, + 192 + ], + "score": 1.0, + "content": "Xiaohui Zeng1,2,3,∗", + "type": "text" + }, + { + "bbox": [ + 271, + 177, + 339, + 191 + ], + "score": 1.0, + "content": "Arash Vahdat1", + "type": "text" + }, + { + "bbox": [ + 357, + 178, + 438, + 192 + ], + "score": 1.0, + "content": "Francis Williams1", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 171, + 177, + 438, + 192 + ] + }, + { + "type": "text", + "bbox": [ + 154, + 206, + 455, + 219 + ], + "lines": [ + { + "bbox": [ + 152, + 204, + 457, + 222 + ], + "spans": [ + { + "bbox": [ + 152, + 205, + 209, + 222 + ], + "score": 1.0, + "content": "Zan Gojcic1", + "type": "text" + }, + { + "bbox": [ + 227, + 205, + 278, + 222 + ], + "score": 1.0, + "content": "Or Litany1", + "type": "text" + }, + { + "bbox": [ + 296, + 204, + 372, + 222 + ], + "score": 1.0, + "content": "Sanja Fidler1,2,3", + "type": "text" + }, + { + "bbox": [ + 389, + 206, + 457, + 221 + ], + "score": 1.0, + "content": "Karsten Kreis1", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 152, + 204, + 457, + 222 + ] + }, + { + "type": "text", + "bbox": [ + 180, + 229, + 430, + 253 + ], + "lines": [ + { + "bbox": [ + 204, + 226, + 406, + 243 + ], + "spans": [ + { + "bbox": [ + 204, + 226, + 406, + 243 + ], + "score": 1.0, + "content": "1NVIDIA 2University of Toronto 3Vector Institute", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 180, + 243, + 432, + 254 + ], + "spans": [ + { + "bbox": [ + 180, + 243, + 432, + 254 + ], + "score": 1.0, + "content": "{xzeng,avahdat,fwilliams,zgojcic,olitany,sfidler,kkreis}@nvidia.com", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 180, + 226, + 432, + 254 + ] + }, + { + "type": "title", + "bbox": [ + 283, + 281, + 328, + 294 + ], + "lines": [ + { + "bbox": [ + 281, + 280, + 331, + 296 + ], + "spans": [ + { + "bbox": [ + 281, + 280, + 331, + 296 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 143, + 307, + 469, + 526 + ], + "lines": [ + { + "bbox": [ + 141, + 307, + 469, + 320 + ], + "spans": [ + { + "bbox": [ + 141, + 307, + 469, + 320 + ], + "score": 1.0, + "content": "Denoising diffusion models (DDMs) have shown promising results in 3D point", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 318, + 469, + 331 + ], + "spans": [ + { + "bbox": [ + 141, + 318, + 469, + 331 + ], + "score": 1.0, + "content": "cloud synthesis. To advance 3D DDMs and make them useful for digital artists, we", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 329, + 469, + 342 + ], + "spans": [ + { + "bbox": [ + 141, + 329, + 469, + 342 + ], + "score": 1.0, + "content": "require (i) high generation quality, (ii) flexibility for manipulation and applications", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 341, + 470, + 352 + ], + "spans": [ + { + "bbox": [ + 141, + 341, + 470, + 352 + ], + "score": 1.0, + "content": "such as conditional synthesis and shape interpolation, and (iii) the ability to output", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 351, + 470, + 363 + ], + "spans": [ + { + "bbox": [ + 141, + 351, + 470, + 363 + ], + "score": 1.0, + "content": "smooth surfaces or meshes. To this end, we introduce the hierarchical Latent Point", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 362, + 469, + 374 + ], + "spans": [ + { + "bbox": [ + 141, + 362, + 469, + 374 + ], + "score": 1.0, + "content": "Diffusion Model (LION) for 3D shape generation. LION is set up as a variational", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 373, + 469, + 385 + ], + "spans": [ + { + "bbox": [ + 141, + 373, + 469, + 385 + ], + "score": 1.0, + "content": "autoencoder (VAE) with a hierarchical latent space that combines a global shape", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 384, + 469, + 396 + ], + "spans": [ + { + "bbox": [ + 141, + 384, + 469, + 396 + ], + "score": 1.0, + "content": "latent representation with a point-structured latent space. For generation, we train", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 395, + 469, + 407 + ], + "spans": [ + { + "bbox": [ + 141, + 395, + 469, + 407 + ], + "score": 1.0, + "content": "two hierarchical DDMs in these latent spaces. The hierarchical VAE approach", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 405, + 470, + 419 + ], + "spans": [ + { + "bbox": [ + 141, + 405, + 470, + 419 + ], + "score": 1.0, + "content": "boosts performance compared to DDMs that operate on point clouds directly,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 416, + 471, + 429 + ], + "spans": [ + { + "bbox": [ + 141, + 416, + 471, + 429 + ], + "score": 1.0, + "content": "while the point-structured latents are still ideally suited for DDM-based modeling.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 427, + 470, + 440 + ], + "spans": [ + { + "bbox": [ + 141, + 427, + 470, + 440 + ], + "score": 1.0, + "content": "Experimentally, LION achieves state-of-the-art generation performance on multiple", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "spans": [ + { + "bbox": [ + 141, + 438, + 469, + 450 + ], + "score": 1.0, + "content": "ShapeNet benchmarks. Furthermore, our VAE framework allows us to easily use", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 448, + 470, + 462 + ], + "spans": [ + { + "bbox": [ + 141, + 448, + 470, + 462 + ], + "score": 1.0, + "content": "LION for different relevant tasks: LION excels at multimodal shape denoising and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 460, + 470, + 473 + ], + "spans": [ + { + "bbox": [ + 141, + 460, + 470, + 473 + ], + "score": 1.0, + "content": "voxel-conditioned synthesis, and it can be adapted for text- and image-driven 3D", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 472, + 470, + 483 + ], + "spans": [ + { + "bbox": [ + 142, + 472, + 470, + 483 + ], + "score": 1.0, + "content": "generation. We also demonstrate shape autoencoding and latent shape interpolation,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 482, + 470, + 495 + ], + "spans": [ + { + "bbox": [ + 141, + 482, + 470, + 495 + ], + "score": 1.0, + "content": "and we augment LION with modern surface reconstruction techniques to generate", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 493, + 469, + 505 + ], + "spans": [ + { + "bbox": [ + 141, + 493, + 469, + 505 + ], + "score": 1.0, + "content": "smooth 3D meshes. We hope that LION provides a powerful tool for artists", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 504, + 469, + 516 + ], + "spans": [ + { + "bbox": [ + 142, + 504, + 469, + 516 + ], + "score": 1.0, + "content": "working with 3D shapes due to its high-quality generation, flexibility, and surface", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 515, + 465, + 527 + ], + "spans": [ + { + "bbox": [ + 141, + 515, + 465, + 527 + ], + "score": 1.0, + "content": "reconstruction. Project page and code: https://nv-tlabs.github.io/LION.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 16.5, + "bbox_fs": [ + 141, + 307, + 471, + 527 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 545, + 190, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 192, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 192, + 560 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 565, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 106, + 565, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 505, + 577 + ], + "score": 1.0, + "content": "Generative modeling of 3D shapes has extensive applications in 3D content creation and has become", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "an active area of research [1–52]. However, to be useful as a tool for digital artists, generative models", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 587, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 506, + 599 + ], + "score": 1.0, + "content": "of 3D shapes have to fulfill several criteria: (i) Generated shapes need to be realistic and of high-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 597, + 506, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 610 + ], + "score": 1.0, + "content": "quality without artifacts. (ii) The model should enable flexible and interactive use and refinement: For", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 609, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 621 + ], + "score": 1.0, + "content": "example, a user may want to refine a generated shape and synthesize versions with varying details.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 619, + 506, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 506, + 633 + ], + "score": 1.0, + "content": "Or an artist may provide a coarse or noisy input shape, thereby guiding the model to produce multiple", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 631, + 505, + 643 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 643 + ], + "score": 1.0, + "content": "realistic high-quality outputs. Similarly, a user may want to interpolate different shapes. (iii) The", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 641, + 506, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 506, + 654 + ], + "score": 1.0, + "content": "model should output smooth meshes, which are the standard representation in most graphics software.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 565, + 506, + 654 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 658, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 670 + ], + "score": 1.0, + "content": "Existing 3D generative models build on various frameworks, including generative adversarial net-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 669, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 107, + 669, + 506, + 681 + ], + "score": 1.0, + "content": "works (GANs) [1–23], variational autoencoders (VAEs) [24–30], normalizing flows [31–34], autore-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 680, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 506, + 692 + ], + "score": 1.0, + "content": "gressive models [35–38], and more [39–44]. Most recently, denoising diffusion models (DDMs)", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37, + "bbox_fs": [ + 106, + 657, + 506, + 692 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 47, + 397, + 226 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 47, + 397, + 226 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 47, + 397, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 47, + 397, + 226 + ], + "score": 0.968, + "type": "image", + "image_path": "a5ecbde7333db061775c6917bb95434b89bc4661fe3aa6fbf5e867c83ff8aba7.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 106, + 47, + 397, + 106.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 106.66666666666666, + 397, + 166.33333333333331 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 106, + 166.33333333333331, + 397, + 225.99999999999997 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 409, + 52, + 503, + 214 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 408, + 52, + 504, + 65 + ], + "spans": [ + { + "bbox": [ + 408, + 52, + 504, + 65 + ], + "score": 1.0, + "content": "Figure 1: LION is set", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 407, + 63, + 503, + 75 + ], + "spans": [ + { + "bbox": [ + 407, + 63, + 503, + 75 + ], + "score": 1.0, + "content": "up as a hierarchical point", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 408, + 73, + 504, + 84 + ], + "spans": [ + { + "bbox": [ + 408, + 73, + 504, + 84 + ], + "score": 1.0, + "content": "cloud VAE with denois-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 408, + 83, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 408, + 83, + 504, + 95 + ], + "score": 1.0, + "content": "ing diffusion models over", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 408, + 93, + 503, + 104 + ], + "spans": [ + { + "bbox": [ + 408, + 93, + 503, + 104 + ], + "score": 1.0, + "content": "the shape latent and latent", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 408, + 103, + 504, + 113 + ], + "spans": [ + { + "bbox": [ + 408, + 103, + 504, + 113 + ], + "score": 1.0, + "content": "point distributions. Point-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 408, + 112, + 504, + 124 + ], + "spans": [ + { + "bbox": [ + 408, + 112, + 504, + 124 + ], + "score": 1.0, + "content": "Voxel CNNs (PVCNN)", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 408, + 123, + 504, + 135 + ], + "spans": [ + { + "bbox": [ + 408, + 123, + 504, + 135 + ], + "score": 1.0, + "content": "with adaptive Group Nor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 408, + 133, + 504, + 144 + ], + "spans": [ + { + "bbox": [ + 408, + 133, + 504, + 144 + ], + "score": 1.0, + "content": "malization (Ada. GN) are", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 408, + 143, + 504, + 154 + ], + "spans": [ + { + "bbox": [ + 408, + 143, + 504, + 154 + ], + "score": 1.0, + "content": "used as neural networks.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 407, + 152, + 503, + 165 + ], + "spans": [ + { + "bbox": [ + 407, + 152, + 503, + 165 + ], + "score": 1.0, + "content": "The latent points can be", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 408, + 163, + 502, + 173 + ], + "spans": [ + { + "bbox": [ + 408, + 163, + 502, + 173 + ], + "score": 1.0, + "content": "interpreted as a smoothed", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 408, + 173, + 503, + 184 + ], + "spans": [ + { + "bbox": [ + 408, + 173, + 503, + 184 + ], + "score": 1.0, + "content": "version of the input point", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 408, + 182, + 503, + 194 + ], + "spans": [ + { + "bbox": [ + 408, + 182, + 503, + 194 + ], + "score": 1.0, + "content": "cloud. Shape As Points", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 408, + 193, + 503, + 204 + ], + "spans": [ + { + "bbox": [ + 408, + 193, + 503, + 204 + ], + "score": 1.0, + "content": "(SAP) is optionally used", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 408, + 202, + 499, + 214 + ], + "spans": [ + { + "bbox": [ + 408, + 202, + 499, + 214 + ], + "score": 1.0, + "content": "for mesh reconstruction.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 9.0 + } + ], + "index": 9.0 + }, + { + "type": "text", + "bbox": [ + 106, + 229, + 505, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "score": 1.0, + "content": "have emerged as powerful generative models, achieving outstanding results not only on image syn-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 240, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 506, + 253 + ], + "score": 1.0, + "content": "thesis [53–64] but also for point cloud-based 3D shape generation [45–47]. In DDMs, the data is", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 251, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 506, + 264 + ], + "score": 1.0, + "content": "gradually perturbed by a diffusion process, while a deep neural network is trained to denoise. This", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 263, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 505, + 273 + ], + "score": 1.0, + "content": "network can then be used to synthesize novel data in an iterative fashion when initialized from random", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 273, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 505, + 286 + ], + "score": 1.0, + "content": "noise [53, 65–67]. However, existing DDMs for 3D shape synthesis struggle with simultaneously", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 284, + 435, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 435, + 297 + ], + "score": 1.0, + "content": "satisfying all criteria discussed above for practically useful 3D generative models.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 300, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 106, + 300, + 506, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 506, + 312 + ], + "score": 1.0, + "content": "Here, we aim to develop a DDM-based generative model of 3D shapes overcoming these limitations.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "score": 1.0, + "content": "We introduce the Latent Point Diffusion Model (LION) for 3D shape generation (see Fig. 1). Similar to", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "previous 3D DDMs, LION operates on point clouds, but it is constructed as a VAE with DDMs in latent", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 332, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 506, + 346 + ], + "score": 1.0, + "content": "space. LION comprises a hierarchical latent space with a vector-valued global shape latent and another", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 343, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 506, + 358 + ], + "score": 1.0, + "content": "point-structured latent space. The latent representations are predicted with point cloud processing", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 353, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 369 + ], + "score": 1.0, + "content": "encoders, and two latent DDMs are trained in these latent spaces. Synthesis in LION proceeds by draw-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 365, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 379 + ], + "score": 1.0, + "content": "ing novel latent samples from the hierarchical latent DDMs and decoding back to the original point", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "score": 1.0, + "content": "cloud space. Importantly, we also demonstrate how to augment LION with modern surface reconstruc-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "tion methods [68] to synthesize smooth shapes as desired by artists. LION has multiple advantages:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 505, + 481 + ], + "lines": [ + { + "bbox": [ + 105, + 403, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 506, + 417 + ], + "score": 1.0, + "content": "Expressivity: By mapping point clouds into regularized latent spaces, the DDMs in latent space are", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "effectively tasked with learning a smoothed distribution. This is easier than training on potentially", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "complex point clouds directly [58], thereby improving expressivity. However, point clouds are, in", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "principle, an ideal representation for DDMs. Because of that, we use latent points, this is, we keep a", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "point cloud structure for our main latent representation. Augmenting the model with an additional", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "global shape latent variable in a hierarchical manner further boosts expressivity. We validate LION", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 469, + 479, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 479, + 483 + ], + "score": 1.0, + "content": "on several popular ShapeNet benchmarks and achieve state-of-the-art synthesis performance.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 505, + 530 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 505, + 499 + ], + "score": 1.0, + "content": "Varying Output Types: Extending LION with Shape As Points (SAP) [68] geometry reconstruction", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "allows us to also output smooth meshes. Fine-tuning SAP on data generated by LION’s autoencoder", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 508, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 505, + 521 + ], + "score": 1.0, + "content": "reduces synthesis noise and enables us to generate high-quality geometry. LION combines (latent)", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 519, + 477, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 477, + 531 + ], + "score": 1.0, + "content": "point cloud-based modeling, ideal for DDMs, with surface reconstruction, desired by artists.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5 + }, + { + "type": "text", + "bbox": [ + 107, + 534, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 534, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 506, + 547 + ], + "score": 1.0, + "content": "Flexibility: Since LION is set up as a VAE, it can be easily adapted for different tasks without re-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 545, + 507, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 507, + 560 + ], + "score": 1.0, + "content": "training the latent DDMs: We can efficiently fine-tune LION’s encoders on voxelized or noisy inputs,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 558, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 506, + 569 + ], + "score": 1.0, + "content": "which a user can provide for guidance. This enables multimodal voxel-guided synthesis and shape de-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 567, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 582 + ], + "score": 1.0, + "content": "noising. We also leverage LION’s latent spaces for shape interpolation and autoencoding. Optionally", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 578, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 506, + 592 + ], + "score": 1.0, + "content": "training the DDMs conditioned on CLIP embeddings enables image- and text-driven 3D generation.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 47 + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "score": 1.0, + "content": "In summary, we make the following contributions: (i) We introduce LION, a novel generative model", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "for 3D shape synthesis, which operates on point clouds and is built on a hierarchical VAE framework", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 617, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 630 + ], + "score": 1.0, + "content": "with two latent DDMs. (ii) We validate LION’s high synthesis quality by reaching state-of-the-art", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 628, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 505, + 640 + ], + "score": 1.0, + "content": "performance on widely used ShapeNet benchmarks. (iii) We achieve high-quality and diverse 3D", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 639, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 651 + ], + "score": 1.0, + "content": "shape synthesis with LION even when trained jointly over many classes without conditioning. (iv) We", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 649, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 506, + 663 + ], + "score": 1.0, + "content": "propose to combine LION with SAP-based surface reconstruction. (v) We demonstrate the flexibility", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 660, + 479, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 479, + 673 + ], + "score": 1.0, + "content": "of our framework by adapting it to relevant tasks such as multimodal voxel-guided synthesis.", + "type": "text" + } + ], + "index": 56 + } + ], + "index": 53 + }, + { + "type": "title", + "bbox": [ + 107, + 680, + 189, + 693 + ], + "lines": [ + { + "bbox": [ + 104, + 678, + 190, + 696 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 190, + 696 + ], + "score": 1.0, + "content": "2 Background", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 57 + }, + { + "type": "text", + "bbox": [ + 108, + 700, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 426, + 712 + ], + "score": 1.0, + "content": "Traditionally, DDMs were introduced in a discrete-step fashion: Given samples", + "type": "text" + }, + { + "bbox": [ + 426, + 700, + 474, + 712 + ], + "score": 0.94, + "content": "\\mathbf { x } _ { 0 } \\sim q ( \\mathbf { x } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "from a", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 711, + 478, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 478, + 723 + ], + "score": 1.0, + "content": "data distribution, DDMs use a Markovian fixed forward diffusion process defined as [65, 53]", + "type": "text" + } + ], + "index": 59 + } + ], + "index": 58.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 741, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 741, + 310, + 753 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 47, + 397, + 226 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 47, + 397, + 226 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 47, + 397, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 47, + 397, + 226 + ], + "score": 0.968, + "type": "image", + "image_path": "a5ecbde7333db061775c6917bb95434b89bc4661fe3aa6fbf5e867c83ff8aba7.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 106, + 47, + 397, + 106.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 106.66666666666666, + 397, + 166.33333333333331 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 106, + 166.33333333333331, + 397, + 225.99999999999997 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 409, + 52, + 503, + 214 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 408, + 52, + 504, + 65 + ], + "spans": [ + { + "bbox": [ + 408, + 52, + 504, + 65 + ], + "score": 1.0, + "content": "Figure 1: LION is set", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 407, + 63, + 503, + 75 + ], + "spans": [ + { + "bbox": [ + 407, + 63, + 503, + 75 + ], + "score": 1.0, + "content": "up as a hierarchical point", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 408, + 73, + 504, + 84 + ], + "spans": [ + { + "bbox": [ + 408, + 73, + 504, + 84 + ], + "score": 1.0, + "content": "cloud VAE with denois-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 408, + 83, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 408, + 83, + 504, + 95 + ], + "score": 1.0, + "content": "ing diffusion models over", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 408, + 93, + 503, + 104 + ], + "spans": [ + { + "bbox": [ + 408, + 93, + 503, + 104 + ], + "score": 1.0, + "content": "the shape latent and latent", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 408, + 103, + 504, + 113 + ], + "spans": [ + { + "bbox": [ + 408, + 103, + 504, + 113 + ], + "score": 1.0, + "content": "point distributions. Point-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 408, + 112, + 504, + 124 + ], + "spans": [ + { + "bbox": [ + 408, + 112, + 504, + 124 + ], + "score": 1.0, + "content": "Voxel CNNs (PVCNN)", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 408, + 123, + 504, + 135 + ], + "spans": [ + { + "bbox": [ + 408, + 123, + 504, + 135 + ], + "score": 1.0, + "content": "with adaptive Group Nor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 408, + 133, + 504, + 144 + ], + "spans": [ + { + "bbox": [ + 408, + 133, + 504, + 144 + ], + "score": 1.0, + "content": "malization (Ada. GN) are", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 408, + 143, + 504, + 154 + ], + "spans": [ + { + "bbox": [ + 408, + 143, + 504, + 154 + ], + "score": 1.0, + "content": "used as neural networks.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 407, + 152, + 503, + 165 + ], + "spans": [ + { + "bbox": [ + 407, + 152, + 503, + 165 + ], + "score": 1.0, + "content": "The latent points can be", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 408, + 163, + 502, + 173 + ], + "spans": [ + { + "bbox": [ + 408, + 163, + 502, + 173 + ], + "score": 1.0, + "content": "interpreted as a smoothed", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 408, + 173, + 503, + 184 + ], + "spans": [ + { + "bbox": [ + 408, + 173, + 503, + 184 + ], + "score": 1.0, + "content": "version of the input point", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 408, + 182, + 503, + 194 + ], + "spans": [ + { + "bbox": [ + 408, + 182, + 503, + 194 + ], + "score": 1.0, + "content": "cloud. Shape As Points", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 408, + 193, + 503, + 204 + ], + "spans": [ + { + "bbox": [ + 408, + 193, + 503, + 204 + ], + "score": 1.0, + "content": "(SAP) is optionally used", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 408, + 202, + 499, + 214 + ], + "spans": [ + { + "bbox": [ + 408, + 202, + 499, + 214 + ], + "score": 1.0, + "content": "for mesh reconstruction.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 9.0 + } + ], + "index": 9.0 + }, + { + "type": "text", + "bbox": [ + 106, + 229, + 505, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "score": 1.0, + "content": "have emerged as powerful generative models, achieving outstanding results not only on image syn-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 240, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 506, + 253 + ], + "score": 1.0, + "content": "thesis [53–64] but also for point cloud-based 3D shape generation [45–47]. In DDMs, the data is", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 251, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 506, + 264 + ], + "score": 1.0, + "content": "gradually perturbed by a diffusion process, while a deep neural network is trained to denoise. This", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 263, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 505, + 273 + ], + "score": 1.0, + "content": "network can then be used to synthesize novel data in an iterative fashion when initialized from random", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 273, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 505, + 286 + ], + "score": 1.0, + "content": "noise [53, 65–67]. However, existing DDMs for 3D shape synthesis struggle with simultaneously", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 284, + 435, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 435, + 297 + ], + "score": 1.0, + "content": "satisfying all criteria discussed above for practically useful 3D generative models.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 228, + 506, + 297 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 300, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 106, + 300, + 506, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 506, + 312 + ], + "score": 1.0, + "content": "Here, we aim to develop a DDM-based generative model of 3D shapes overcoming these limitations.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "score": 1.0, + "content": "We introduce the Latent Point Diffusion Model (LION) for 3D shape generation (see Fig. 1). Similar to", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "previous 3D DDMs, LION operates on point clouds, but it is constructed as a VAE with DDMs in latent", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 332, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 506, + 346 + ], + "score": 1.0, + "content": "space. LION comprises a hierarchical latent space with a vector-valued global shape latent and another", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 343, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 506, + 358 + ], + "score": 1.0, + "content": "point-structured latent space. The latent representations are predicted with point cloud processing", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 353, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 369 + ], + "score": 1.0, + "content": "encoders, and two latent DDMs are trained in these latent spaces. Synthesis in LION proceeds by draw-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 365, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 506, + 379 + ], + "score": 1.0, + "content": "ing novel latent samples from the hierarchical latent DDMs and decoding back to the original point", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "score": 1.0, + "content": "cloud space. Importantly, we also demonstrate how to augment LION with modern surface reconstruc-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "tion methods [68] to synthesize smooth shapes as desired by artists. LION has multiple advantages:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29, + "bbox_fs": [ + 104, + 300, + 506, + 401 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 505, + 481 + ], + "lines": [ + { + "bbox": [ + 105, + 403, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 506, + 417 + ], + "score": 1.0, + "content": "Expressivity: By mapping point clouds into regularized latent spaces, the DDMs in latent space are", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "effectively tasked with learning a smoothed distribution. This is easier than training on potentially", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "complex point clouds directly [58], thereby improving expressivity. However, point clouds are, in", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "principle, an ideal representation for DDMs. Because of that, we use latent points, this is, we keep a", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "point cloud structure for our main latent representation. Augmenting the model with an additional", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "global shape latent variable in a hierarchical manner further boosts expressivity. We validate LION", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 469, + 479, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 479, + 483 + ], + "score": 1.0, + "content": "on several popular ShapeNet benchmarks and achieve state-of-the-art synthesis performance.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 403, + 506, + 483 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 505, + 530 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 505, + 499 + ], + "score": 1.0, + "content": "Varying Output Types: Extending LION with Shape As Points (SAP) [68] geometry reconstruction", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "allows us to also output smooth meshes. Fine-tuning SAP on data generated by LION’s autoencoder", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 508, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 505, + 521 + ], + "score": 1.0, + "content": "reduces synthesis noise and enables us to generate high-quality geometry. LION combines (latent)", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 519, + 477, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 477, + 531 + ], + "score": 1.0, + "content": "point cloud-based modeling, ideal for DDMs, with surface reconstruction, desired by artists.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 485, + 506, + 531 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 534, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 534, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 506, + 547 + ], + "score": 1.0, + "content": "Flexibility: Since LION is set up as a VAE, it can be easily adapted for different tasks without re-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 545, + 507, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 507, + 560 + ], + "score": 1.0, + "content": "training the latent DDMs: We can efficiently fine-tune LION’s encoders on voxelized or noisy inputs,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 558, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 506, + 569 + ], + "score": 1.0, + "content": "which a user can provide for guidance. This enables multimodal voxel-guided synthesis and shape de-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 567, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 506, + 582 + ], + "score": 1.0, + "content": "noising. We also leverage LION’s latent spaces for shape interpolation and autoencoding. Optionally", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 578, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 506, + 592 + ], + "score": 1.0, + "content": "training the DDMs conditioned on CLIP embeddings enables image- and text-driven 3D generation.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 47, + "bbox_fs": [ + 105, + 534, + 507, + 592 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "score": 1.0, + "content": "In summary, we make the following contributions: (i) We introduce LION, a novel generative model", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "for 3D shape synthesis, which operates on point clouds and is built on a hierarchical VAE framework", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 617, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 630 + ], + "score": 1.0, + "content": "with two latent DDMs. (ii) We validate LION’s high synthesis quality by reaching state-of-the-art", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 628, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 505, + 640 + ], + "score": 1.0, + "content": "performance on widely used ShapeNet benchmarks. (iii) We achieve high-quality and diverse 3D", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 639, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 651 + ], + "score": 1.0, + "content": "shape synthesis with LION even when trained jointly over many classes without conditioning. (iv) We", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 649, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 506, + 663 + ], + "score": 1.0, + "content": "propose to combine LION with SAP-based surface reconstruction. (v) We demonstrate the flexibility", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 660, + 479, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 479, + 673 + ], + "score": 1.0, + "content": "of our framework by adapting it to relevant tasks such as multimodal voxel-guided synthesis.", + "type": "text" + } + ], + "index": 56 + } + ], + "index": 53, + "bbox_fs": [ + 105, + 595, + 506, + 673 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 680, + 189, + 693 + ], + "lines": [ + { + "bbox": [ + 104, + 678, + 190, + 696 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 190, + 696 + ], + "score": 1.0, + "content": "2 Background", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 57 + }, + { + "type": "text", + "bbox": [ + 108, + 700, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 426, + 712 + ], + "score": 1.0, + "content": "Traditionally, DDMs were introduced in a discrete-step fashion: Given samples", + "type": "text" + }, + { + "bbox": [ + 426, + 700, + 474, + 712 + ], + "score": 0.94, + "content": "\\mathbf { x } _ { 0 } \\sim q ( \\mathbf { x } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "from a", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 106, + 711, + 478, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 478, + 723 + ], + "score": 1.0, + "content": "data distribution, DDMs use a Markovian fixed forward diffusion process defined as [65, 53]", + "type": "text" + } + ], + "index": 59 + } + ], + "index": 58.5, + "bbox_fs": [ + 106, + 699, + 505, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 45, + 422, + 159 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 45, + 422, + 159 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 45, + 422, + 159 + ], + "spans": [ + { + "bbox": [ + 109, + 45, + 422, + 159 + ], + "score": 0.827, + "type": "image", + "image_path": "50a69b4fae41081a6f71ae4fa30d4db630414574be115671a763f0682d5ca458.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 109, + 45, + 422, + 83.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 83.0, + 422, + 121.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 121.0, + 422, + 159.0 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 431, + 49, + 506, + 151 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 429, + 48, + 508, + 63 + ], + "spans": [ + { + "bbox": [ + 429, + 48, + 508, + 63 + ], + "score": 1.0, + "content": "Figure 2: Gen-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 430, + 59, + 506, + 71 + ], + "spans": [ + { + "bbox": [ + 430, + 60, + 457, + 70 + ], + "score": 1.0, + "content": "erated", + "type": "text" + }, + { + "bbox": [ + 474, + 59, + 506, + 71 + ], + "score": 1.0, + "content": "meshes", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 429, + 69, + 507, + 82 + ], + "spans": [ + { + "bbox": [ + 429, + 69, + 507, + 82 + ], + "score": 1.0, + "content": "with LION. Right:", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 430, + 78, + 507, + 92 + ], + "spans": [ + { + "bbox": [ + 430, + 78, + 483, + 92 + ], + "score": 1.0, + "content": "Synthesizing", + "type": "text" + }, + { + "bbox": [ + 488, + 79, + 507, + 91 + ], + "score": 1.0, + "content": "dif-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 430, + 88, + 504, + 101 + ], + "spans": [ + { + "bbox": [ + 430, + 88, + 493, + 101 + ], + "score": 1.0, + "content": "ferent details", + "type": "text" + }, + { + "bbox": [ + 492, + 90, + 504, + 101 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 430, + 99, + 490, + 111 + ], + "spans": [ + { + "bbox": [ + 430, + 99, + 490, + 111 + ], + "score": 1.0, + "content": "diffuse-denoise", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 430, + 109, + 506, + 122 + ], + "spans": [ + { + "bbox": [ + 430, + 109, + 506, + 122 + ], + "score": 1.0, + "content": "(see Sec. 3.1) in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 430, + 119, + 506, + 131 + ], + "spans": [ + { + "bbox": [ + 430, + 119, + 506, + 131 + ], + "score": 1.0, + "content": "latent space, while", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 429, + 129, + 506, + 141 + ], + "spans": [ + { + "bbox": [ + 429, + 129, + 506, + 141 + ], + "score": 1.0, + "content": "preserving overall", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 429, + 139, + 461, + 153 + ], + "spans": [ + { + "bbox": [ + 429, + 139, + 461, + 153 + ], + "score": 1.0, + "content": "shapes.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 6.5 + } + ], + "index": 4.75 + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 157, + 464, + 191 + ], + "lines": [ + { + "bbox": [ + 146, + 157, + 464, + 191 + ], + "spans": [ + { + "bbox": [ + 146, + 157, + 464, + 191 + ], + "score": 0.9, + "content": "q ( \\mathbf { x } _ { 1 : T } | \\mathbf { x } _ { 0 } ) : = \\prod _ { t = 1 } ^ { T } q ( \\mathbf { x } _ { t } | \\mathbf { x } _ { t - 1 } ) , \\qquad q ( \\mathbf { x } _ { t } | \\mathbf { x } _ { t - 1 } ) : = \\mathcal { N } ( \\mathbf { x } _ { t } ; \\sqrt { 1 - \\beta _ { t } } \\mathbf { x } _ { t - 1 } , \\beta _ { t } I ) ,", + "type": "interline_equation", + "image_path": "aab33c64cd41651771e73025dc2397f469be2bf99e574c43a343cd4228a63b61.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 146, + 157, + 464, + 168.33333333333334 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 146, + 168.33333333333334, + 464, + 179.66666666666669 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 146, + 179.66666666666669, + 464, + 191.00000000000003 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 191, + 505, + 235 + ], + "lines": [ + { + "bbox": [ + 105, + 190, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 132, + 204 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 191, + 141, + 201 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 190, + 270, + 204 + ], + "score": 1.0, + "content": "denotes the number of steps and", + "type": "text" + }, + { + "bbox": [ + 271, + 191, + 316, + 203 + ], + "score": 0.9, + "content": "q \\big ( \\mathbf { x } _ { t } | \\mathbf { x } _ { t - 1 } \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 190, + 506, + 204 + ], + "score": 1.0, + "content": "is a Gaussian transition kernel, which gradually", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 202, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 304, + 214 + ], + "score": 1.0, + "content": "adds noise to the input with a variance schedule", + "type": "text" + }, + { + "bbox": [ + 304, + 203, + 345, + 213 + ], + "score": 0.93, + "content": "\\beta _ { 1 } , . . . , \\beta _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 202, + 368, + 214 + ], + "score": 1.0, + "content": ". The", + "type": "text" + }, + { + "bbox": [ + 368, + 203, + 379, + 213 + ], + "score": 0.88, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 202, + 506, + 214 + ], + "score": 1.0, + "content": "are chosen such that the chain", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 213, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 381, + 225 + ], + "score": 1.0, + "content": "approximately converges to a standard Gaussian distribution after", + "type": "text" + }, + { + "bbox": [ + 382, + 213, + 391, + 223 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 213, + 419, + 225 + ], + "score": 1.0, + "content": "steps,", + "type": "text" + }, + { + "bbox": [ + 419, + 213, + 503, + 225 + ], + "score": 0.92, + "content": "q ( \\mathbf { x } _ { T } ) { \\approx } { \\mathcal { N } } ( \\mathbf { x } _ { T } ; \\mathbf { 0 } , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 213, + 506, + 225 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 223, + 503, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 356, + 236 + ], + "score": 1.0, + "content": "DDMs learn a parametrized reverse process (model parameters", + "type": "text" + }, + { + "bbox": [ + 356, + 225, + 363, + 234 + ], + "score": 0.75, + "content": "\\pmb \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 223, + 503, + 236 + ], + "score": 1.0, + "content": ") that inverts the forward diffusion:", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "interline_equation", + "bbox": [ + 136, + 236, + 475, + 270 + ], + "lines": [ + { + "bbox": [ + 136, + 236, + 475, + 270 + ], + "spans": [ + { + "bbox": [ + 136, + 236, + 475, + 270 + ], + "score": 0.94, + "content": "p _ { \\theta } ( \\mathbf { x } _ { 0 : T } ) : = p ( \\mathbf { x } _ { T } ) \\prod _ { t = 1 } ^ { T } p _ { \\theta } ( \\mathbf { x } _ { t - 1 } | \\mathbf { x } _ { t } ) , \\qquad p _ { \\theta } ( \\mathbf { x } _ { t - 1 } | \\mathbf { x } _ { t } ) : = \\mathcal { N } ( \\mathbf { x } _ { t - 1 } ; \\mu _ { \\theta } ( \\mathbf { x } _ { t } , t ) , \\rho _ { t } ^ { 2 } I ) .", + "type": "interline_equation", + "image_path": "271c06d7336ca1d1b28e2438cf13cf561e03f0e84c8c3a63ce18a956aef8d0c8.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 136, + 236, + 475, + 247.33333333333334 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 136, + 247.33333333333334, + 475, + 258.6666666666667 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 136, + 258.6666666666667, + 475, + 270.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 270, + 505, + 324 + ], + "lines": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "score": 1.0, + "content": "This generative reverse process is also Markovian with Gaussian transition kernels, which use fixed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 145, + 295 + ], + "score": 1.0, + "content": "variances", + "type": "text" + }, + { + "bbox": [ + 145, + 281, + 156, + 294 + ], + "score": 0.89, + "content": "\\rho _ { t } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 281, + 387, + 295 + ], + "score": 1.0, + "content": ". DDMs can be interpreted as latent variable models, where", + "type": "text" + }, + { + "bbox": [ + 388, + 283, + 429, + 293 + ], + "score": 0.9, + "content": "\\mathbf { x } _ { 1 } , . . . , \\mathbf { x } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 281, + 506, + 295 + ], + "score": 1.0, + "content": "are latents, and the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 290, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 173, + 306 + ], + "score": 1.0, + "content": "forward process", + "type": "text" + }, + { + "bbox": [ + 174, + 292, + 218, + 304 + ], + "score": 0.92, + "content": "q \\big ( \\mathbf { x } _ { 1 : T } | \\mathbf { x } _ { 0 } \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 290, + 467, + 306 + ], + "score": 1.0, + "content": "acts as a fixed approximate posterior, to which the generative", + "type": "text" + }, + { + "bbox": [ + 468, + 292, + 505, + 304 + ], + "score": 0.91, + "content": "p _ { \\pmb { \\theta } } ( \\mathbf { x } _ { 0 : T } )", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 301, + 506, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 506, + 314 + ], + "score": 1.0, + "content": "is fit. DDMs are trained by minimizing the variational upper bound on the negative log-likelihood of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 310, + 504, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 140, + 325 + ], + "score": 1.0, + "content": "the data", + "type": "text" + }, + { + "bbox": [ + 140, + 313, + 152, + 323 + ], + "score": 0.87, + "content": "\\mathbf { x } _ { \\mathrm { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 310, + 178, + 325 + ], + "score": 1.0, + "content": "under", + "type": "text" + }, + { + "bbox": [ + 178, + 312, + 216, + 324 + ], + "score": 0.92, + "content": "p _ { \\theta } ( \\mathbf { x } _ { 0 : T } )", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 310, + 504, + 325 + ], + "score": 1.0, + "content": ". Up to irrelevant constant terms, this objective can be expressed as [53]", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 324, + 488, + 350 + ], + "lines": [ + { + "bbox": [ + 112, + 324, + 488, + 350 + ], + "spans": [ + { + "bbox": [ + 112, + 324, + 488, + 350 + ], + "score": 0.89, + "content": "\\operatorname* { m i n } _ { \\theta } \\mathbb { E } _ { t \\sim U \\left\\{ 1 , T \\right\\} , \\mathbf { x } _ { 0 } \\sim p \\left( \\mathbf { x } _ { 0 } \\right) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } \\left[ w ( t ) | | \\epsilon - \\epsilon _ { \\theta } ( \\alpha _ { t } \\mathbf { x } _ { 0 } + \\sigma _ { t } \\epsilon , t ) | | _ { 2 } ^ { 2 } \\right] , w ( t ) = \\frac { \\beta _ { t } ^ { 2 } } { 2 \\rho _ { t } ^ { 2 } ( 1 - \\beta _ { t } ) ( 1 - \\alpha _ { t } ^ { 2 } ) } ,", + "type": "interline_equation", + "image_path": "8758abfcbe929c2a1e7f5ef5591b7cf0e54b47e582ee409ed2060181a887379e.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 112, + 324, + 488, + 350 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 352, + 505, + 450 + ], + "lines": [ + { + "bbox": [ + 104, + 350, + 507, + 372 + ], + "spans": [ + { + "bbox": [ + 104, + 350, + 135, + 372 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 351, + 232, + 371 + ], + "score": 0.95, + "content": "\\alpha _ { t } = \\sqrt { \\prod _ { s = 1 } ^ { t } ( 1 - \\beta _ { s } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 350, + 253, + 372 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 253, + 354, + 320, + 368 + ], + "score": 0.93, + "content": "\\sigma _ { t } ~ = ~ \\sqrt { 1 - \\alpha _ { t } ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 350, + 507, + 372 + ], + "score": 1.0, + "content": "are the parameters of the tractable diffused", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 369, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 177, + 384 + ], + "score": 1.0, + "content": "distribution after", + "type": "text" + }, + { + "bbox": [ + 177, + 372, + 182, + 380 + ], + "score": 0.78, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 369, + 207, + 384 + ], + "score": 1.0, + "content": "steps", + "type": "text" + }, + { + "bbox": [ + 207, + 370, + 330, + 382 + ], + "score": 0.9, + "content": "q ( \\mathbf { x } _ { t } | \\mathbf { x } _ { 0 } ) = \\mathcal { N } ( \\mathbf { x } _ { t } ; \\alpha _ { t } \\mathbf { x } _ { 0 } , \\sigma _ { t } ^ { 2 } I )", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 369, + 506, + 384 + ], + "score": 1.0, + "content": ". Furthermore, Eq. (3) employs the widely", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 382, + 508, + 407 + ], + "spans": [ + { + "bbox": [ + 104, + 384, + 281, + 405 + ], + "score": 1.0, + "content": "used parametrization µθ(xt, t) := √ 11−βt", + "type": "text" + }, + { + "bbox": [ + 195, + 382, + 380, + 407 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mu _ { \\theta } ( \\mathbf { x } _ { t } , t ) : = \\frac { 1 } { \\sqrt { 1 - \\beta _ { t } } } \\left( \\mathbf { x } _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - \\alpha _ { t } ^ { 2 } } } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 383, + 508, + 406 + ], + "score": 1.0, + "content": ". It is common practice to set", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 107, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 107, + 406, + 145, + 418 + ], + "score": 0.91, + "content": "w ( t ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 405, + 505, + 418 + ], + "score": 1.0, + "content": ", instead of the one in Eq. (3), which often promotes perceptual quality of the generated", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 416, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 283, + 430 + ], + "score": 1.0, + "content": "output. In the objective of Eq. (3), the model", + "type": "text" + }, + { + "bbox": [ + 284, + 418, + 295, + 428 + ], + "score": 0.85, + "content": "\\epsilon _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 416, + 389, + 430 + ], + "score": 1.0, + "content": "is, for all possible steps", + "type": "text" + }, + { + "bbox": [ + 390, + 418, + 395, + 427 + ], + "score": 0.76, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 416, + 506, + 430 + ], + "score": 1.0, + "content": "along the diffusion process,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 294, + 441 + ], + "score": 1.0, + "content": "effectively trained to predict the noise vector", + "type": "text" + }, + { + "bbox": [ + 294, + 430, + 300, + 438 + ], + "score": 0.66, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 428, + 505, + 441 + ], + "score": 1.0, + "content": "that is necessary to denoise an observed diffused", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 439, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 137, + 451 + ], + "score": 1.0, + "content": "sample", + "type": "text" + }, + { + "bbox": [ + 137, + 440, + 148, + 450 + ], + "score": 0.86, + "content": "\\mathbf { x } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 439, + 504, + 451 + ], + "score": 1.0, + "content": ". After training, the DDM can be sampled with ancestral sampling in an iterative fashion:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32 + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 451, + 398, + 472 + ], + "lines": [ + { + "bbox": [ + 212, + 451, + 398, + 472 + ], + "spans": [ + { + "bbox": [ + 212, + 451, + 398, + 472 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathbf { x } _ { t - 1 } = \\frac { 1 } { \\sqrt { 1 - \\beta _ { t } } } ( \\mathbf { x } _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - \\alpha _ { t } ^ { 2 } } } \\pmb { \\epsilon } _ { \\theta } ( \\mathbf { x } _ { t } , t ) ) + \\rho _ { t } \\pmb { \\eta } , } \\end{array}", + "type": "interline_equation", + "image_path": "6550edfaeeba37f2a7851c20144c611d252e6347668116264063a4c7f440725d.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 212, + 451, + 398, + 472 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 473, + 505, + 496 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 133, + 487 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 473, + 198, + 485 + ], + "score": 0.94, + "content": "\\eta \\sim \\mathcal { N } ( \\eta ; 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 471, + 426, + 487 + ], + "score": 1.0, + "content": ". This sampling chain is initialized from a random sample", + "type": "text" + }, + { + "bbox": [ + 426, + 473, + 502, + 485 + ], + "score": 0.91, + "content": "\\mathbf { x } _ { T } \\sim \\mathcal { N } ( \\mathbf { x } _ { T } ; \\mathbf { 0 } , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 471, + 506, + 487 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 482, + 445, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 445, + 497 + ], + "score": 1.0, + "content": "Furthermore, the noise injection in Eq. 4 is usually omitted in the last sampling step.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 506, + 545 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "DDMs can also be expressed with a continuous-time framework [67, 69]. In this formulation, the dif-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 510, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 524 + ], + "score": 1.0, + "content": "fusion and reverse generative processes are described by differential equations. This approach allows", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 522, + 507, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 507, + 534 + ], + "score": 1.0, + "content": "for deterministic sampling and encoding schemes based on ordinary differential equations (ODEs).", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 532, + 497, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 497, + 546 + ], + "score": 1.0, + "content": "We make use of this framework in Sec. 3.1 and we review this approach in more detail in App. B.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5 + }, + { + "type": "title", + "bbox": [ + 107, + 552, + 348, + 567 + ], + "lines": [ + { + "bbox": [ + 104, + 552, + 348, + 568 + ], + "spans": [ + { + "bbox": [ + 104, + 552, + 348, + 568 + ], + "score": 1.0, + "content": "3 Hierarchical Latent Point Diffusion Models", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 107, + 571, + 504, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 571, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 505, + 584 + ], + "score": 1.0, + "content": "We first formally introduce LION, then discuss various applications and extensions in Sec. 3.1, and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 582, + 477, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 477, + 595 + ], + "score": 1.0, + "content": "finally recapitulate its unique advantages in Sec. 3.2. See Fig. 1 for a visualization of LION.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + }, + { + "type": "text", + "bbox": [ + 106, + 597, + 505, + 666 + ], + "lines": [ + { + "bbox": [ + 104, + 595, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 104, + 595, + 225, + 613 + ], + "score": 1.0, + "content": "We are modeling point clouds", + "type": "text" + }, + { + "bbox": [ + 226, + 597, + 270, + 609 + ], + "score": 0.91, + "content": "\\mathbf { x } \\in \\mathbb { R } ^ { 3 \\times N }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 595, + 327, + 613 + ], + "score": 1.0, + "content": ", consisting of", + "type": "text" + }, + { + "bbox": [ + 327, + 599, + 337, + 609 + ], + "score": 0.81, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 595, + 461, + 613 + ], + "score": 1.0, + "content": "points with xyz-coordinates in", + "type": "text" + }, + { + "bbox": [ + 462, + 598, + 474, + 609 + ], + "score": 0.86, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 595, + 506, + 613 + ], + "score": 1.0, + "content": ". LION", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 610, + 504, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 504, + 621 + ], + "score": 1.0, + "content": "is set up as a hierarchical VAE with DDMs in latent space. It uses a vector-valued global shape latent", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 618, + 507, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 148, + 633 + ], + "score": 0.92, + "content": "\\mathbf { z } _ { 0 } \\in \\mathbb { R } ^ { D _ { \\mathbf { z } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 618, + 293, + 636 + ], + "score": 1.0, + "content": "and a point cloud-structured latent", + "type": "text" + }, + { + "bbox": [ + 294, + 621, + 367, + 633 + ], + "score": 0.93, + "content": "\\mathbf { h } _ { 0 } \\in \\mathbf { \\bar { \\mathbb { R } } } ^ { ( 3 + D _ { \\mathbf { h } } ) \\times N }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 618, + 425, + 636 + ], + "score": 1.0, + "content": ". Specifically,", + "type": "text" + }, + { + "bbox": [ + 425, + 622, + 437, + 633 + ], + "score": 0.86, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 618, + 507, + 636 + ], + "score": 1.0, + "content": "is a latent point", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 104, + 631, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 104, + 631, + 187, + 646 + ], + "score": 1.0, + "content": "cloud consisting of", + "type": "text" + }, + { + "bbox": [ + 187, + 633, + 198, + 642 + ], + "score": 0.79, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 631, + 247, + 646 + ], + "score": 1.0, + "content": "points with", + "type": "text" + }, + { + "bbox": [ + 248, + 634, + 264, + 644 + ], + "score": 0.57, + "content": "x y z", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 631, + 327, + 646 + ], + "score": 1.0, + "content": "-coordinates in", + "type": "text" + }, + { + "bbox": [ + 327, + 633, + 339, + 642 + ], + "score": 0.83, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 631, + 506, + 646 + ], + "score": 1.0, + "content": ". In addition, each latent point can carry", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 644, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 148, + 655 + ], + "score": 1.0, + "content": "additional", + "type": "text" + }, + { + "bbox": [ + 149, + 644, + 163, + 655 + ], + "score": 0.89, + "content": "D _ { \\mathbf { h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 644, + 505, + 655 + ], + "score": 1.0, + "content": "latent features. Training of LION is then performed in two stages—first, we train it as", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "score": 1.0, + "content": "a regular VAE with standard Gaussian priors; then, we train the latent DDMs on the latent encodings.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 48.5 + }, + { + "type": "text", + "bbox": [ + 105, + 670, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 670, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 683 + ], + "score": 1.0, + "content": "First Stage Training. Initially, LION is trained by maximizing a modified variational lower bound on", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 682, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 436, + 695 + ], + "score": 1.0, + "content": "the data log-likelihood (ELBO) with respect to the encoder and decoder parameters", + "type": "text" + }, + { + "bbox": [ + 436, + 682, + 444, + 694 + ], + "score": 0.87, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 682, + 461, + 695 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 462, + 682, + 469, + 694 + ], + "score": 0.73, + "content": "\\boldsymbol { \\xi }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 682, + 506, + 695 + ], + "score": 1.0, + "content": "[70, 71]:", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 52.5 + }, + { + "type": "interline_equation", + "bbox": [ + 154, + 694, + 456, + 727 + ], + "lines": [ + { + "bbox": [ + 154, + 694, + 456, + 727 + ], + "spans": [ + { + "bbox": [ + 154, + 694, + 456, + 727 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { E L B O } } ( \\boldsymbol { \\phi } , \\boldsymbol { \\xi } ) = \\mathbb { E } _ { p ( \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } ) } [ \\log p _ { \\xi } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } ) } \\\\ & { \\qquad - \\lambda _ { \\mathbf { z } } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) \\vert p ( \\mathbf { z } _ { 0 } ) ) - \\lambda _ { \\mathbf { h } } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } ) \\vert p ( \\mathbf { h } _ { 0 } ) ) ] . } \\end{array}", + "type": "interline_equation", + "image_path": "1fc516f62bded3377c6552c66b41784c29310ee897a56536681150999b6a39e6.jpg" + } + ] + } + ], + "index": 55, + "virtual_lines": [ + { + "bbox": [ + 154, + 694, + 456, + 705.0 + ], + "spans": [], + "index": 54 + }, + { + "bbox": [ + 154, + 705.0, + 456, + 716.0 + ], + "spans": [], + "index": 55 + }, + { + "bbox": [ + 154, + 716.0, + 456, + 727.0 + ], + "spans": [], + "index": 56 + } + ] + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 45, + 422, + 159 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 45, + 422, + 159 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 45, + 422, + 159 + ], + "spans": [ + { + "bbox": [ + 109, + 45, + 422, + 159 + ], + "score": 0.827, + "type": "image", + "image_path": "50a69b4fae41081a6f71ae4fa30d4db630414574be115671a763f0682d5ca458.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 109, + 45, + 422, + 83.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 83.0, + 422, + 121.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 121.0, + 422, + 159.0 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 431, + 49, + 506, + 151 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 429, + 48, + 508, + 63 + ], + "spans": [ + { + "bbox": [ + 429, + 48, + 508, + 63 + ], + "score": 1.0, + "content": "Figure 2: Gen-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 430, + 59, + 506, + 71 + ], + "spans": [ + { + "bbox": [ + 430, + 60, + 457, + 70 + ], + "score": 1.0, + "content": "erated", + "type": "text" + }, + { + "bbox": [ + 474, + 59, + 506, + 71 + ], + "score": 1.0, + "content": "meshes", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 429, + 69, + 507, + 82 + ], + "spans": [ + { + "bbox": [ + 429, + 69, + 507, + 82 + ], + "score": 1.0, + "content": "with LION. Right:", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 430, + 78, + 507, + 92 + ], + "spans": [ + { + "bbox": [ + 430, + 78, + 483, + 92 + ], + "score": 1.0, + "content": "Synthesizing", + "type": "text" + }, + { + "bbox": [ + 488, + 79, + 507, + 91 + ], + "score": 1.0, + "content": "dif-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 430, + 88, + 504, + 101 + ], + "spans": [ + { + "bbox": [ + 430, + 88, + 493, + 101 + ], + "score": 1.0, + "content": "ferent details", + "type": "text" + }, + { + "bbox": [ + 492, + 90, + 504, + 101 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 430, + 99, + 490, + 111 + ], + "spans": [ + { + "bbox": [ + 430, + 99, + 490, + 111 + ], + "score": 1.0, + "content": "diffuse-denoise", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 430, + 109, + 506, + 122 + ], + "spans": [ + { + "bbox": [ + 430, + 109, + 506, + 122 + ], + "score": 1.0, + "content": "(see Sec. 3.1) in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 430, + 119, + 506, + 131 + ], + "spans": [ + { + "bbox": [ + 430, + 119, + 506, + 131 + ], + "score": 1.0, + "content": "latent space, while", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 429, + 129, + 506, + 141 + ], + "spans": [ + { + "bbox": [ + 429, + 129, + 506, + 141 + ], + "score": 1.0, + "content": "preserving overall", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 429, + 139, + 461, + 153 + ], + "spans": [ + { + "bbox": [ + 429, + 139, + 461, + 153 + ], + "score": 1.0, + "content": "shapes.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 6.5 + } + ], + "index": 4.75 + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 157, + 464, + 191 + ], + "lines": [ + { + "bbox": [ + 146, + 157, + 464, + 191 + ], + "spans": [ + { + "bbox": [ + 146, + 157, + 464, + 191 + ], + "score": 0.9, + "content": "q ( \\mathbf { x } _ { 1 : T } | \\mathbf { x } _ { 0 } ) : = \\prod _ { t = 1 } ^ { T } q ( \\mathbf { x } _ { t } | \\mathbf { x } _ { t - 1 } ) , \\qquad q ( \\mathbf { x } _ { t } | \\mathbf { x } _ { t - 1 } ) : = \\mathcal { N } ( \\mathbf { x } _ { t } ; \\sqrt { 1 - \\beta _ { t } } \\mathbf { x } _ { t - 1 } , \\beta _ { t } I ) ,", + "type": "interline_equation", + "image_path": "aab33c64cd41651771e73025dc2397f469be2bf99e574c43a343cd4228a63b61.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 146, + 157, + 464, + 168.33333333333334 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 146, + 168.33333333333334, + 464, + 179.66666666666669 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 146, + 179.66666666666669, + 464, + 191.00000000000003 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 191, + 505, + 235 + ], + "lines": [ + { + "bbox": [ + 105, + 190, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 132, + 204 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 191, + 141, + 201 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 190, + 270, + 204 + ], + "score": 1.0, + "content": "denotes the number of steps and", + "type": "text" + }, + { + "bbox": [ + 271, + 191, + 316, + 203 + ], + "score": 0.9, + "content": "q \\big ( \\mathbf { x } _ { t } | \\mathbf { x } _ { t - 1 } \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 190, + 506, + 204 + ], + "score": 1.0, + "content": "is a Gaussian transition kernel, which gradually", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 202, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 304, + 214 + ], + "score": 1.0, + "content": "adds noise to the input with a variance schedule", + "type": "text" + }, + { + "bbox": [ + 304, + 203, + 345, + 213 + ], + "score": 0.93, + "content": "\\beta _ { 1 } , . . . , \\beta _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 202, + 368, + 214 + ], + "score": 1.0, + "content": ". The", + "type": "text" + }, + { + "bbox": [ + 368, + 203, + 379, + 213 + ], + "score": 0.88, + "content": "\\beta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 202, + 506, + 214 + ], + "score": 1.0, + "content": "are chosen such that the chain", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 213, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 381, + 225 + ], + "score": 1.0, + "content": "approximately converges to a standard Gaussian distribution after", + "type": "text" + }, + { + "bbox": [ + 382, + 213, + 391, + 223 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 213, + 419, + 225 + ], + "score": 1.0, + "content": "steps,", + "type": "text" + }, + { + "bbox": [ + 419, + 213, + 503, + 225 + ], + "score": 0.92, + "content": "q ( \\mathbf { x } _ { T } ) { \\approx } { \\mathcal { N } } ( \\mathbf { x } _ { T } ; \\mathbf { 0 } , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 213, + 506, + 225 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 223, + 503, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 356, + 236 + ], + "score": 1.0, + "content": "DDMs learn a parametrized reverse process (model parameters", + "type": "text" + }, + { + "bbox": [ + 356, + 225, + 363, + 234 + ], + "score": 0.75, + "content": "\\pmb \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 223, + 503, + 236 + ], + "score": 1.0, + "content": ") that inverts the forward diffusion:", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 190, + 506, + 236 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 136, + 236, + 475, + 270 + ], + "lines": [ + { + "bbox": [ + 136, + 236, + 475, + 270 + ], + "spans": [ + { + "bbox": [ + 136, + 236, + 475, + 270 + ], + "score": 0.94, + "content": "p _ { \\theta } ( \\mathbf { x } _ { 0 : T } ) : = p ( \\mathbf { x } _ { T } ) \\prod _ { t = 1 } ^ { T } p _ { \\theta } ( \\mathbf { x } _ { t - 1 } | \\mathbf { x } _ { t } ) , \\qquad p _ { \\theta } ( \\mathbf { x } _ { t - 1 } | \\mathbf { x } _ { t } ) : = \\mathcal { N } ( \\mathbf { x } _ { t - 1 } ; \\mu _ { \\theta } ( \\mathbf { x } _ { t } , t ) , \\rho _ { t } ^ { 2 } I ) .", + "type": "interline_equation", + "image_path": "271c06d7336ca1d1b28e2438cf13cf561e03f0e84c8c3a63ce18a956aef8d0c8.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 136, + 236, + 475, + 247.33333333333334 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 136, + 247.33333333333334, + 475, + 258.6666666666667 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 136, + 258.6666666666667, + 475, + 270.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 270, + 505, + 324 + ], + "lines": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 505, + 282 + ], + "score": 1.0, + "content": "This generative reverse process is also Markovian with Gaussian transition kernels, which use fixed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 145, + 295 + ], + "score": 1.0, + "content": "variances", + "type": "text" + }, + { + "bbox": [ + 145, + 281, + 156, + 294 + ], + "score": 0.89, + "content": "\\rho _ { t } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 281, + 387, + 295 + ], + "score": 1.0, + "content": ". DDMs can be interpreted as latent variable models, where", + "type": "text" + }, + { + "bbox": [ + 388, + 283, + 429, + 293 + ], + "score": 0.9, + "content": "\\mathbf { x } _ { 1 } , . . . , \\mathbf { x } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 281, + 506, + 295 + ], + "score": 1.0, + "content": "are latents, and the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 290, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 173, + 306 + ], + "score": 1.0, + "content": "forward process", + "type": "text" + }, + { + "bbox": [ + 174, + 292, + 218, + 304 + ], + "score": 0.92, + "content": "q \\big ( \\mathbf { x } _ { 1 : T } | \\mathbf { x } _ { 0 } \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 290, + 467, + 306 + ], + "score": 1.0, + "content": "acts as a fixed approximate posterior, to which the generative", + "type": "text" + }, + { + "bbox": [ + 468, + 292, + 505, + 304 + ], + "score": 0.91, + "content": "p _ { \\pmb { \\theta } } ( \\mathbf { x } _ { 0 : T } )", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 301, + 506, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 506, + 314 + ], + "score": 1.0, + "content": "is fit. DDMs are trained by minimizing the variational upper bound on the negative log-likelihood of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 310, + 504, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 140, + 325 + ], + "score": 1.0, + "content": "the data", + "type": "text" + }, + { + "bbox": [ + 140, + 313, + 152, + 323 + ], + "score": 0.87, + "content": "\\mathbf { x } _ { \\mathrm { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 310, + 178, + 325 + ], + "score": 1.0, + "content": "under", + "type": "text" + }, + { + "bbox": [ + 178, + 312, + 216, + 324 + ], + "score": 0.92, + "content": "p _ { \\theta } ( \\mathbf { x } _ { 0 : T } )", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 310, + 504, + 325 + ], + "score": 1.0, + "content": ". Up to irrelevant constant terms, this objective can be expressed as [53]", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 270, + 506, + 325 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 324, + 488, + 350 + ], + "lines": [ + { + "bbox": [ + 112, + 324, + 488, + 350 + ], + "spans": [ + { + "bbox": [ + 112, + 324, + 488, + 350 + ], + "score": 0.89, + "content": "\\operatorname* { m i n } _ { \\theta } \\mathbb { E } _ { t \\sim U \\left\\{ 1 , T \\right\\} , \\mathbf { x } _ { 0 } \\sim p \\left( \\mathbf { x } _ { 0 } \\right) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } \\left[ w ( t ) | | \\epsilon - \\epsilon _ { \\theta } ( \\alpha _ { t } \\mathbf { x } _ { 0 } + \\sigma _ { t } \\epsilon , t ) | | _ { 2 } ^ { 2 } \\right] , w ( t ) = \\frac { \\beta _ { t } ^ { 2 } } { 2 \\rho _ { t } ^ { 2 } ( 1 - \\beta _ { t } ) ( 1 - \\alpha _ { t } ^ { 2 } ) } ,", + "type": "interline_equation", + "image_path": "8758abfcbe929c2a1e7f5ef5591b7cf0e54b47e582ee409ed2060181a887379e.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 112, + 324, + 488, + 350 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 352, + 505, + 450 + ], + "lines": [ + { + "bbox": [ + 104, + 350, + 507, + 372 + ], + "spans": [ + { + "bbox": [ + 104, + 350, + 135, + 372 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 351, + 232, + 371 + ], + "score": 0.95, + "content": "\\alpha _ { t } = \\sqrt { \\prod _ { s = 1 } ^ { t } ( 1 - \\beta _ { s } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 350, + 253, + 372 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 253, + 354, + 320, + 368 + ], + "score": 0.93, + "content": "\\sigma _ { t } ~ = ~ \\sqrt { 1 - \\alpha _ { t } ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 350, + 507, + 372 + ], + "score": 1.0, + "content": "are the parameters of the tractable diffused", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 369, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 177, + 384 + ], + "score": 1.0, + "content": "distribution after", + "type": "text" + }, + { + "bbox": [ + 177, + 372, + 182, + 380 + ], + "score": 0.78, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 369, + 207, + 384 + ], + "score": 1.0, + "content": "steps", + "type": "text" + }, + { + "bbox": [ + 207, + 370, + 330, + 382 + ], + "score": 0.9, + "content": "q ( \\mathbf { x } _ { t } | \\mathbf { x } _ { 0 } ) = \\mathcal { N } ( \\mathbf { x } _ { t } ; \\alpha _ { t } \\mathbf { x } _ { 0 } , \\sigma _ { t } ^ { 2 } I )", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 369, + 506, + 384 + ], + "score": 1.0, + "content": ". Furthermore, Eq. (3) employs the widely", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 382, + 508, + 407 + ], + "spans": [ + { + "bbox": [ + 104, + 384, + 281, + 405 + ], + "score": 1.0, + "content": "used parametrization µθ(xt, t) := √ 11−βt", + "type": "text" + }, + { + "bbox": [ + 195, + 382, + 380, + 407 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mu _ { \\theta } ( \\mathbf { x } _ { t } , t ) : = \\frac { 1 } { \\sqrt { 1 - \\beta _ { t } } } \\left( \\mathbf { x } _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - \\alpha _ { t } ^ { 2 } } } \\epsilon _ { \\theta } ( \\mathbf { x } _ { t } , t ) \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 383, + 508, + 406 + ], + "score": 1.0, + "content": ". It is common practice to set", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 107, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 107, + 406, + 145, + 418 + ], + "score": 0.91, + "content": "w ( t ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 405, + 505, + 418 + ], + "score": 1.0, + "content": ", instead of the one in Eq. (3), which often promotes perceptual quality of the generated", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 416, + 506, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 283, + 430 + ], + "score": 1.0, + "content": "output. In the objective of Eq. (3), the model", + "type": "text" + }, + { + "bbox": [ + 284, + 418, + 295, + 428 + ], + "score": 0.85, + "content": "\\epsilon _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 416, + 389, + 430 + ], + "score": 1.0, + "content": "is, for all possible steps", + "type": "text" + }, + { + "bbox": [ + 390, + 418, + 395, + 427 + ], + "score": 0.76, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 416, + 506, + 430 + ], + "score": 1.0, + "content": "along the diffusion process,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 294, + 441 + ], + "score": 1.0, + "content": "effectively trained to predict the noise vector", + "type": "text" + }, + { + "bbox": [ + 294, + 430, + 300, + 438 + ], + "score": 0.66, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 428, + 505, + 441 + ], + "score": 1.0, + "content": "that is necessary to denoise an observed diffused", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 439, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 137, + 451 + ], + "score": 1.0, + "content": "sample", + "type": "text" + }, + { + "bbox": [ + 137, + 440, + 148, + 450 + ], + "score": 0.86, + "content": "\\mathbf { x } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 439, + 504, + 451 + ], + "score": 1.0, + "content": ". After training, the DDM can be sampled with ancestral sampling in an iterative fashion:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32, + "bbox_fs": [ + 104, + 350, + 508, + 451 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 451, + 398, + 472 + ], + "lines": [ + { + "bbox": [ + 212, + 451, + 398, + 472 + ], + "spans": [ + { + "bbox": [ + 212, + 451, + 398, + 472 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathbf { x } _ { t - 1 } = \\frac { 1 } { \\sqrt { 1 - \\beta _ { t } } } ( \\mathbf { x } _ { t } - \\frac { \\beta _ { t } } { \\sqrt { 1 - \\alpha _ { t } ^ { 2 } } } \\pmb { \\epsilon } _ { \\theta } ( \\mathbf { x } _ { t } , t ) ) + \\rho _ { t } \\pmb { \\eta } , } \\end{array}", + "type": "interline_equation", + "image_path": "6550edfaeeba37f2a7851c20144c611d252e6347668116264063a4c7f440725d.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 212, + 451, + 398, + 472 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "list", + "bbox": [ + 107, + 473, + 505, + 496 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 133, + 487 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 473, + 198, + 485 + ], + "score": 0.94, + "content": "\\eta \\sim \\mathcal { N } ( \\eta ; 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 471, + 426, + 487 + ], + "score": 1.0, + "content": ". This sampling chain is initialized from a random sample", + "type": "text" + }, + { + "bbox": [ + 426, + 473, + 502, + 485 + ], + "score": 0.91, + "content": "\\mathbf { x } _ { T } \\sim \\mathcal { N } ( \\mathbf { x } _ { T } ; \\mathbf { 0 } , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 471, + 506, + 487 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 482, + 445, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 445, + 497 + ], + "score": 1.0, + "content": "Furthermore, the noise injection in Eq. 4 is usually omitted in the last sampling step.", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 471, + 506, + 497 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 506, + 545 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "DDMs can also be expressed with a continuous-time framework [67, 69]. In this formulation, the dif-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 510, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 524 + ], + "score": 1.0, + "content": "fusion and reverse generative processes are described by differential equations. This approach allows", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 522, + 507, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 507, + 534 + ], + "score": 1.0, + "content": "for deterministic sampling and encoding schemes based on ordinary differential equations (ODEs).", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 532, + 497, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 497, + 546 + ], + "score": 1.0, + "content": "We make use of this framework in Sec. 3.1 and we review this approach in more detail in App. B.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 500, + 507, + 546 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 552, + 348, + 567 + ], + "lines": [ + { + "bbox": [ + 104, + 552, + 348, + 568 + ], + "spans": [ + { + "bbox": [ + 104, + 552, + 348, + 568 + ], + "score": 1.0, + "content": "3 Hierarchical Latent Point Diffusion Models", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 107, + 571, + 504, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 571, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 505, + 584 + ], + "score": 1.0, + "content": "We first formally introduce LION, then discuss various applications and extensions in Sec. 3.1, and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 582, + 477, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 477, + 595 + ], + "score": 1.0, + "content": "finally recapitulate its unique advantages in Sec. 3.2. See Fig. 1 for a visualization of LION.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5, + "bbox_fs": [ + 106, + 571, + 505, + 595 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 597, + 505, + 666 + ], + "lines": [ + { + "bbox": [ + 104, + 595, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 104, + 595, + 225, + 613 + ], + "score": 1.0, + "content": "We are modeling point clouds", + "type": "text" + }, + { + "bbox": [ + 226, + 597, + 270, + 609 + ], + "score": 0.91, + "content": "\\mathbf { x } \\in \\mathbb { R } ^ { 3 \\times N }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 595, + 327, + 613 + ], + "score": 1.0, + "content": ", consisting of", + "type": "text" + }, + { + "bbox": [ + 327, + 599, + 337, + 609 + ], + "score": 0.81, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 595, + 461, + 613 + ], + "score": 1.0, + "content": "points with xyz-coordinates in", + "type": "text" + }, + { + "bbox": [ + 462, + 598, + 474, + 609 + ], + "score": 0.86, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 595, + 506, + 613 + ], + "score": 1.0, + "content": ". LION", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 610, + 504, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 504, + 621 + ], + "score": 1.0, + "content": "is set up as a hierarchical VAE with DDMs in latent space. It uses a vector-valued global shape latent", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 618, + 507, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 148, + 633 + ], + "score": 0.92, + "content": "\\mathbf { z } _ { 0 } \\in \\mathbb { R } ^ { D _ { \\mathbf { z } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 618, + 293, + 636 + ], + "score": 1.0, + "content": "and a point cloud-structured latent", + "type": "text" + }, + { + "bbox": [ + 294, + 621, + 367, + 633 + ], + "score": 0.93, + "content": "\\mathbf { h } _ { 0 } \\in \\mathbf { \\bar { \\mathbb { R } } } ^ { ( 3 + D _ { \\mathbf { h } } ) \\times N }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 618, + 425, + 636 + ], + "score": 1.0, + "content": ". Specifically,", + "type": "text" + }, + { + "bbox": [ + 425, + 622, + 437, + 633 + ], + "score": 0.86, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 618, + 507, + 636 + ], + "score": 1.0, + "content": "is a latent point", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 104, + 631, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 104, + 631, + 187, + 646 + ], + "score": 1.0, + "content": "cloud consisting of", + "type": "text" + }, + { + "bbox": [ + 187, + 633, + 198, + 642 + ], + "score": 0.79, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 631, + 247, + 646 + ], + "score": 1.0, + "content": "points with", + "type": "text" + }, + { + "bbox": [ + 248, + 634, + 264, + 644 + ], + "score": 0.57, + "content": "x y z", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 631, + 327, + 646 + ], + "score": 1.0, + "content": "-coordinates in", + "type": "text" + }, + { + "bbox": [ + 327, + 633, + 339, + 642 + ], + "score": 0.83, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 631, + 506, + 646 + ], + "score": 1.0, + "content": ". In addition, each latent point can carry", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 644, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 148, + 655 + ], + "score": 1.0, + "content": "additional", + "type": "text" + }, + { + "bbox": [ + 149, + 644, + 163, + 655 + ], + "score": 0.89, + "content": "D _ { \\mathbf { h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 644, + 505, + 655 + ], + "score": 1.0, + "content": "latent features. Training of LION is then performed in two stages—first, we train it as", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "score": 1.0, + "content": "a regular VAE with standard Gaussian priors; then, we train the latent DDMs on the latent encodings.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 48.5, + "bbox_fs": [ + 104, + 595, + 507, + 667 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 670, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 670, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 683 + ], + "score": 1.0, + "content": "First Stage Training. Initially, LION is trained by maximizing a modified variational lower bound on", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 682, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 436, + 695 + ], + "score": 1.0, + "content": "the data log-likelihood (ELBO) with respect to the encoder and decoder parameters", + "type": "text" + }, + { + "bbox": [ + 436, + 682, + 444, + 694 + ], + "score": 0.87, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 682, + 461, + 695 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 462, + 682, + 469, + 694 + ], + "score": 0.73, + "content": "\\boldsymbol { \\xi }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 682, + 506, + 695 + ], + "score": 1.0, + "content": "[70, 71]:", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 52.5, + "bbox_fs": [ + 106, + 670, + 506, + 695 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 154, + 694, + 456, + 727 + ], + "lines": [ + { + "bbox": [ + 154, + 694, + 456, + 727 + ], + "spans": [ + { + "bbox": [ + 154, + 694, + 456, + 727 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { E L B O } } ( \\boldsymbol { \\phi } , \\boldsymbol { \\xi } ) = \\mathbb { E } _ { p ( \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } ) } [ \\log p _ { \\xi } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } ) } \\\\ & { \\qquad - \\lambda _ { \\mathbf { z } } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) \\vert p ( \\mathbf { z } _ { 0 } ) ) - \\lambda _ { \\mathbf { h } } D _ { \\mathrm { K L } } ( q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } ) \\vert p ( \\mathbf { h } _ { 0 } ) ) ] . } \\end{array}", + "type": "interline_equation", + "image_path": "1fc516f62bded3377c6552c66b41784c29310ee897a56536681150999b6a39e6.jpg" + } + ] + } + ], + "index": 55, + "virtual_lines": [ + { + "bbox": [ + 154, + 694, + 456, + 705.0 + ], + "spans": [], + "index": 54 + }, + { + "bbox": [ + 154, + 705.0, + 456, + 716.0 + ], + "spans": [], + "index": 55 + }, + { + "bbox": [ + 154, + 716.0, + 456, + 727.0 + ], + "spans": [], + "index": 56 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 51, + 391, + 129 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 51, + 391, + 129 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 51, + 391, + 129 + ], + "spans": [ + { + "bbox": [ + 110, + 51, + 391, + 129 + ], + "score": 0.945, + "type": "image", + "image_path": "794b4b6104af7191905670a7a7a67c74af466190e3278dba537a59617fa285aa.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 51, + 391, + 77.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 77.0, + 391, + 103.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 103.0, + 391, + 129.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 412, + 47, + 502, + 128 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 412, + 47, + 503, + 59 + ], + "spans": [ + { + "bbox": [ + 412, + 47, + 503, + 59 + ], + "score": 1.0, + "content": "Figure 3: Generated", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 412, + 57, + 503, + 70 + ], + "spans": [ + { + "bbox": [ + 412, + 57, + 503, + 70 + ], + "score": 1.0, + "content": "shapes (top: point", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 412, + 68, + 504, + 79 + ], + "spans": [ + { + "bbox": [ + 412, + 68, + 479, + 79 + ], + "score": 1.0, + "content": "clouds, bottom:", + "type": "text" + }, + { + "bbox": [ + 479, + 69, + 504, + 79 + ], + "score": 1.0, + "content": "cor-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 411, + 77, + 504, + 90 + ], + "spans": [ + { + "bbox": [ + 411, + 77, + 457, + 90 + ], + "score": 1.0, + "content": "responding", + "type": "text" + }, + { + "bbox": [ + 469, + 78, + 504, + 89 + ], + "score": 1.0, + "content": "meshes)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 412, + 87, + 503, + 98 + ], + "spans": [ + { + "bbox": [ + 412, + 87, + 503, + 98 + ], + "score": 1.0, + "content": "from LION trained", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 411, + 98, + 503, + 110 + ], + "spans": [ + { + "bbox": [ + 411, + 98, + 503, + 110 + ], + "score": 1.0, + "content": "jointly over 13 classes", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 411, + 108, + 503, + 119 + ], + "spans": [ + { + "bbox": [ + 411, + 108, + 503, + 119 + ], + "score": 1.0, + "content": "of ShapeNet-vol without", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 412, + 118, + 500, + 129 + ], + "spans": [ + { + "bbox": [ + 412, + 118, + 500, + 129 + ], + "score": 1.0, + "content": "conditioning (Sec. 5.2).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6.5 + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 106, + 139, + 505, + 228 + ], + "lines": [ + { + "bbox": [ + 105, + 139, + 506, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 231, + 152 + ], + "score": 1.0, + "content": "Here, the global shape latent", + "type": "text" + }, + { + "bbox": [ + 231, + 142, + 243, + 151 + ], + "score": 0.86, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 139, + 423, + 152 + ], + "score": 1.0, + "content": "is sampled from the posterior distribution", + "type": "text" + }, + { + "bbox": [ + 424, + 140, + 461, + 152 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 139, + 506, + 152 + ], + "score": 1.0, + "content": ", which is", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 151, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 162 + ], + "score": 1.0, + "content": "parametrized by factorial Gaussians, whose means and variances are predicted via an encoder network.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 161, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 105, + 161, + 194, + 174 + ], + "score": 1.0, + "content": "The point cloud latent", + "type": "text" + }, + { + "bbox": [ + 194, + 162, + 206, + 172 + ], + "score": 0.87, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 161, + 406, + 174 + ], + "score": 1.0, + "content": "is sampled from a similarly parametrized posterior", + "type": "text" + }, + { + "bbox": [ + 406, + 162, + 459, + 173 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 161, + 506, + 174 + ], + "score": 1.0, + "content": ", while also", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 173, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 171, + 185 + ], + "score": 1.0, + "content": "conditioning on", + "type": "text" + }, + { + "bbox": [ + 171, + 173, + 182, + 184 + ], + "score": 0.84, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 173, + 187, + 185 + ], + "score": 1.0, + "content": "(", + "type": "text" + }, + { + "bbox": [ + 187, + 173, + 195, + 184 + ], + "score": 0.83, + "content": "\\cdot \\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 173, + 419, + 185 + ], + "score": 1.0, + "content": "denotes the parameters of both encoders). Furthermore,", + "type": "text" + }, + { + "bbox": [ + 419, + 173, + 471, + 185 + ], + "score": 0.93, + "content": "p _ { \\pmb { \\xi } } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 173, + 505, + 185 + ], + "score": 1.0, + "content": "denotes", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 184, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 184, + 505, + 195 + ], + "score": 1.0, + "content": "the decoder, parametrized as a factorial Laplace distribution with predicted means and fixed unit scale", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 194, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 235, + 208 + ], + "score": 1.0, + "content": "parameter (corresponding to an", + "type": "text" + }, + { + "bbox": [ + 236, + 195, + 248, + 205 + ], + "score": 0.88, + "content": "L _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 194, + 334, + 208 + ], + "score": 1.0, + "content": "reconstruction loss).", + "type": "text" + }, + { + "bbox": [ + 334, + 194, + 346, + 205 + ], + "score": 0.87, + "content": "\\lambda _ { \\mathbf { z } }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 194, + 364, + 208 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 365, + 195, + 377, + 205 + ], + "score": 0.89, + "content": "\\lambda _ { \\mathbf { h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 194, + 505, + 208 + ], + "score": 1.0, + "content": "are hyperparameters balancing", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 204, + 506, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 420, + 218 + ], + "score": 1.0, + "content": "reconstruction accuracy and Kullback-Leibler regularization (note that only for", + "type": "text" + }, + { + "bbox": [ + 421, + 205, + 475, + 216 + ], + "score": 0.91, + "content": "\\lambda _ { \\mathbf { z } } = \\lambda _ { \\mathbf { h } } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 204, + 506, + 218 + ], + "score": 1.0, + "content": "we are", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 216, + 483, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 272, + 229 + ], + "score": 1.0, + "content": "optimizing a rigorous ELBO). The priors", + "type": "text" + }, + { + "bbox": [ + 273, + 216, + 296, + 228 + ], + "score": 0.92, + "content": "p ( \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 216, + 314, + 229 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 315, + 216, + 339, + 228 + ], + "score": 0.92, + "content": "p ( \\mathbf { h } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 216, + 355, + 229 + ], + "score": 1.0, + "content": "are", + "type": "text" + }, + { + "bbox": [ + 356, + 216, + 389, + 228 + ], + "score": 0.93, + "content": "\\mathcal { N } ( \\mathbf { 0 } , \\pmb { I } )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 216, + 483, + 229 + ], + "score": 1.0, + "content": ". Also see Fig. 1 again.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 505, + 299 + ], + "lines": [ + { + "bbox": [ + 105, + 231, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 245 + ], + "score": 1.0, + "content": "Second Stage Training. In principle, we could use the VAE’s priors to sample encodings and generate", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 244, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 505, + 255 + ], + "score": 1.0, + "content": "new shapes. However, the simple Gaussian priors will not accurately match the encoding distribution", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "score": 1.0, + "content": "from the training data and therefore produce poor samples (prior hole problem [58, 72–79]). This", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 266, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 505, + 277 + ], + "score": 1.0, + "content": "motivates training highly expressive latent DDMs. In particular, in the second stage we freeze the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 428, + 289 + ], + "score": 1.0, + "content": "VAE’s encoder and decoder networks and train two latent DDMs on the encodings", + "type": "text" + }, + { + "bbox": [ + 428, + 277, + 439, + 287 + ], + "score": 0.86, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 276, + 456, + 289 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 457, + 276, + 469, + 287 + ], + "score": 0.89, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 276, + 505, + 289 + ], + "score": 1.0, + "content": "sampled", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 286, + 492, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 128, + 300 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 129, + 287, + 166, + 299 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 286, + 184, + 300 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 185, + 287, + 237, + 299 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 286, + 492, + 300 + ], + "score": 1.0, + "content": ", minimizing score matching (SM) objectives similar to Eq. (2):", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "interline_equation", + "bbox": [ + 148, + 303, + 463, + 336 + ], + "lines": [ + { + "bbox": [ + 148, + 303, + 463, + 336 + ], + "spans": [ + { + "bbox": [ + 148, + 303, + 463, + 336 + ], + "score": 0.85, + "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S M } ^ { \\mathbf { z } } } ( \\pmb { \\theta } ) = \\mathbb { E } _ { t \\sim U \\{ 1 , T \\} , p ( \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } | | \\epsilon - \\epsilon _ { \\theta } ( \\mathbf { z } _ { t } , t ) | | _ { 2 } ^ { 2 } , } \\\\ & { \\mathcal { L } _ { \\mathrm { S M } ^ { \\mathbf { h } } } ( \\pmb { \\psi } ) = \\mathbb { E } _ { t \\sim U \\{ 1 , T \\} , p ( \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } ) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } | | \\epsilon - \\epsilon _ { \\psi } ( \\mathbf { h } _ { t } , \\mathbf { z } _ { 0 } , t ) | | _ { 2 } ^ { 2 } , } \\end{array}", + "type": "interline_equation", + "image_path": "ebdf63e10a31b4b95e5bbc33639e727966fe0bf1c6ad4195a6e3d40902bc0426.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 148, + 303, + 463, + 314.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 148, + 314.0, + 463, + 325.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 148, + 325.0, + 463, + 336.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 339, + 505, + 372 + ], + "lines": [ + { + "bbox": [ + 106, + 338, + 504, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 338, + 133, + 352 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 340, + 205, + 351 + ], + "score": 0.91, + "content": "{ \\bf z } _ { t } = \\alpha _ { t } { \\bf z } _ { 0 } + \\sigma _ { t } \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 338, + 225, + 352 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 225, + 340, + 298, + 350 + ], + "score": 0.91, + "content": "\\mathbf { h } _ { t } = \\alpha _ { t } \\mathbf { h } _ { 0 } + \\sigma _ { t } \\mathbf { \\epsilon } \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 338, + 496, + 352 + ], + "score": 1.0, + "content": "are the diffused latent encodings. Furthermore,", + "type": "text" + }, + { + "bbox": [ + 497, + 340, + 504, + 349 + ], + "score": 0.67, + "content": "\\pmb \\theta", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 350, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 330, + 362 + ], + "score": 1.0, + "content": "denotes the parameters of the global shape latent DDM", + "type": "text" + }, + { + "bbox": [ + 331, + 350, + 366, + 362 + ], + "score": 0.92, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\mathbf { z } _ { t } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 350, + 387, + 362 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 387, + 351, + 396, + 361 + ], + "score": 0.87, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 350, + 506, + 362 + ], + "score": 1.0, + "content": "refers to the parameters of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 361, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 194, + 374 + ], + "score": 1.0, + "content": "the conditional DDM", + "type": "text" + }, + { + "bbox": [ + 194, + 361, + 246, + 373 + ], + "score": 0.92, + "content": "\\mathbf { \\epsilon } _ { \\psi } ( \\mathbf { h } _ { t } , \\bar { \\mathbf { z } _ { 0 } } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 361, + 487, + 374 + ], + "score": 1.0, + "content": "trained over the latent point cloud (note the conditioning on", + "type": "text" + }, + { + "bbox": [ + 488, + 362, + 498, + 372 + ], + "score": 0.8, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 499, + 361, + 505, + 374 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 377, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "score": 1.0, + "content": "Generation. With the latent DDMs, we can formally define a hierarchical generative model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 314, + 400 + ], + "score": 0.87, + "content": "\\begin{array} { r } { p _ { \\xi , \\psi , \\theta } ( \\mathbf { x } , \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } ) = p _ { \\xi } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } ) p _ { \\psi } ( \\mathbf { h } _ { 0 } | \\mathbf { z } _ { 0 } ) p _ { \\theta } ( \\mathbf { z } _ { 0 } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 388, + 347, + 401 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 347, + 388, + 376, + 400 + ], + "score": 0.92, + "content": "p _ { \\pmb { \\theta } } ( \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "denotes the distribution of the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 398, + 507, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 211, + 412 + ], + "score": 1.0, + "content": "global shape latent DDM,", + "type": "text" + }, + { + "bbox": [ + 211, + 400, + 255, + 411 + ], + "score": 0.9, + "content": "p _ { \\psi } ( \\mathbf { h } _ { 0 } | \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 398, + 507, + 412 + ], + "score": 1.0, + "content": "refers to the DDM modeling the point cloud-structured latents,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 409, + 507, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 123, + 423 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 410, + 175, + 422 + ], + "score": 0.92, + "content": "p _ { \\pmb { \\xi } } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 409, + 507, + 423 + ], + "score": 1.0, + "content": "is LION’s decoder. We can hierarchically sample the latent DDMs following Eq. (4)", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 420, + 468, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 468, + 434 + ], + "score": 1.0, + "content": "and then translate the latent points back to the original point cloud space with the decoder.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 437, + 505, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "score": 1.0, + "content": "Network Architectures and DDM Parametrization. Let us briefly summarize key implementation", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "score": 1.0, + "content": "choices. The encoder networks, as well as the decoder and the latent point DDM, operating on point", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 135, + 471 + ], + "score": 1.0, + "content": "clouds", + "type": "text" + }, + { + "bbox": [ + 135, + 461, + 143, + 469 + ], + "score": 0.59, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 459, + 506, + 471 + ], + "score": 1.0, + "content": ", are all implemented based on Point-Voxel CNNs (PVCNNs) [80], following Zhou et al.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 468, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 506, + 483 + ], + "score": 1.0, + "content": "[46]. PVCNNs efficiently combine the point-based processing of PointNets [81, 82] with the strong", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 480, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 494 + ], + "score": 1.0, + "content": "spatial inductive bias of convolutions. The DDM modeling the global shape latent uses a ResNet [83]", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 492, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 324, + 503 + ], + "score": 1.0, + "content": "structure with fully-connected layers (implemented as", + "type": "text" + }, + { + "bbox": [ + 325, + 492, + 344, + 502 + ], + "score": 0.87, + "content": "1 \\times 1", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 492, + 505, + 503 + ], + "score": 1.0, + "content": "-convolutions). All conditionings on the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "global shape latent are implemented via adaptive Group Normalization [84] in the PVCNN layers.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 514, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 506, + 526 + ], + "score": 1.0, + "content": "Furthermore, following Vahdat et al. [58] we use a mixed score parametrization in both latent DDMs.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "This means that the score models are parametrized to predict a residual correction to an analytic", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 535, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 506, + 547 + ], + "score": 1.0, + "content": "standard Gaussian score. This is beneficial since the latent encodings are regularized towards a", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 546, + 459, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 459, + 559 + ], + "score": 1.0, + "content": "standard Gaussian distribution during the first training stage (see App. D for all details).", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 41 + }, + { + "type": "title", + "bbox": [ + 107, + 570, + 251, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 252, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 252, + 584 + ], + "score": 1.0, + "content": "3.1 Applications and Extensions", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47 + }, + { + "type": "text", + "bbox": [ + 107, + 590, + 462, + 602 + ], + "lines": [ + { + "bbox": [ + 104, + 588, + 465, + 605 + ], + "spans": [ + { + "bbox": [ + 104, + 588, + 465, + 605 + ], + "score": 1.0, + "content": "Here, we discuss how LION can be used and extended for different relevant applications.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 48 + }, + { + "type": "text", + "bbox": [ + 107, + 606, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 506, + 619 + ], + "score": 1.0, + "content": "Multimodal Generation. We can synthesize different variations of a given shape, enabling multi-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 619, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 619, + 420, + 630 + ], + "score": 1.0, + "content": "modal generation in a controlled manner: Given a shape, i.e., its point cloud", + "type": "text" + }, + { + "bbox": [ + 421, + 620, + 428, + 628 + ], + "score": 0.3, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 619, + 505, + 630 + ], + "score": 1.0, + "content": ", we encode it into", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 286, + 641 + ], + "score": 1.0, + "content": "latent space. Then, we diffuse its encodings", + "type": "text" + }, + { + "bbox": [ + 286, + 630, + 297, + 640 + ], + "score": 0.86, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 629, + 315, + 641 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 315, + 629, + 327, + 640 + ], + "score": 0.87, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 629, + 442, + 641 + ], + "score": 1.0, + "content": "for a small number of steps", + "type": "text" + }, + { + "bbox": [ + 442, + 629, + 470, + 639 + ], + "score": 0.9, + "content": "\\tau < T", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "towards", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 638, + 505, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 157, + 654 + ], + "score": 1.0, + "content": "intermediate", + "type": "text" + }, + { + "bbox": [ + 157, + 641, + 169, + 651 + ], + "score": 0.87, + "content": "{ \\bf z } _ { \\tau }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 638, + 185, + 654 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 186, + 640, + 198, + 651 + ], + "score": 0.88, + "content": "{ \\bf h } _ { \\tau }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 638, + 505, + 654 + ], + "score": 1.0, + "content": "along the diffusion process such that only local details are destroyed. Running", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 326, + 663 + ], + "score": 1.0, + "content": "the reverse generation process from this intermediate", + "type": "text" + }, + { + "bbox": [ + 326, + 653, + 333, + 660 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 651, + 380, + 663 + ], + "score": 1.0, + "content": ", starting at", + "type": "text" + }, + { + "bbox": [ + 380, + 652, + 392, + 662 + ], + "score": 0.86, + "content": "{ \\bf z } _ { \\tau }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 651, + 411, + 663 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 411, + 651, + 424, + 662 + ], + "score": 0.87, + "content": "{ \\bf h } _ { \\tau }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 651, + 505, + 663 + ], + "score": 1.0, + "content": ", leads to variations", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 661, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 505, + 674 + ], + "score": 1.0, + "content": "of the original shape with different details (see, for instance, Fig. 2). We refer to this procedure as", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 672, + 489, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 672, + 489, + 685 + ], + "score": 1.0, + "content": "diffuse-denoise (details in App. C.1). Similar techniques have been used for image editing [85].", + "type": "text" + } + ], + "index": 55 + } + ], + "index": 52 + }, + { + "type": "text", + "bbox": [ + 108, + 689, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 703 + ], + "score": 1.0, + "content": "Encoder Fine-tuning for Voxel-Conditioned Synthesis and Denoising. In practice, an artist using", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 700, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 712 + ], + "score": 1.0, + "content": "a 3D generative model may have a rough idea of the desired shape. For instance, they may be able to", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 105, + 711, + 507, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 507, + 723 + ], + "score": 1.0, + "content": "quickly construct a coarse voxelized shape, to which the generative model then adds realistic details.", + "type": "text" + } + ], + "index": 58 + } + ], + "index": 57 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 11, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 51, + 391, + 129 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 51, + 391, + 129 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 51, + 391, + 129 + ], + "spans": [ + { + "bbox": [ + 110, + 51, + 391, + 129 + ], + "score": 0.945, + "type": "image", + "image_path": "794b4b6104af7191905670a7a7a67c74af466190e3278dba537a59617fa285aa.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 51, + 391, + 77.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 77.0, + 391, + 103.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 103.0, + 391, + 129.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 412, + 47, + 502, + 128 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 412, + 47, + 503, + 59 + ], + "spans": [ + { + "bbox": [ + 412, + 47, + 503, + 59 + ], + "score": 1.0, + "content": "Figure 3: Generated", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 412, + 57, + 503, + 70 + ], + "spans": [ + { + "bbox": [ + 412, + 57, + 503, + 70 + ], + "score": 1.0, + "content": "shapes (top: point", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 412, + 68, + 504, + 79 + ], + "spans": [ + { + "bbox": [ + 412, + 68, + 479, + 79 + ], + "score": 1.0, + "content": "clouds, bottom:", + "type": "text" + }, + { + "bbox": [ + 479, + 69, + 504, + 79 + ], + "score": 1.0, + "content": "cor-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 411, + 77, + 504, + 90 + ], + "spans": [ + { + "bbox": [ + 411, + 77, + 457, + 90 + ], + "score": 1.0, + "content": "responding", + "type": "text" + }, + { + "bbox": [ + 469, + 78, + 504, + 89 + ], + "score": 1.0, + "content": "meshes)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 412, + 87, + 503, + 98 + ], + "spans": [ + { + "bbox": [ + 412, + 87, + 503, + 98 + ], + "score": 1.0, + "content": "from LION trained", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 411, + 98, + 503, + 110 + ], + "spans": [ + { + "bbox": [ + 411, + 98, + 503, + 110 + ], + "score": 1.0, + "content": "jointly over 13 classes", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 411, + 108, + 503, + 119 + ], + "spans": [ + { + "bbox": [ + 411, + 108, + 503, + 119 + ], + "score": 1.0, + "content": "of ShapeNet-vol without", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 412, + 118, + 500, + 129 + ], + "spans": [ + { + "bbox": [ + 412, + 118, + 500, + 129 + ], + "score": 1.0, + "content": "conditioning (Sec. 5.2).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6.5 + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 106, + 139, + 505, + 228 + ], + "lines": [ + { + "bbox": [ + 105, + 139, + 506, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 231, + 152 + ], + "score": 1.0, + "content": "Here, the global shape latent", + "type": "text" + }, + { + "bbox": [ + 231, + 142, + 243, + 151 + ], + "score": 0.86, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 139, + 423, + 152 + ], + "score": 1.0, + "content": "is sampled from the posterior distribution", + "type": "text" + }, + { + "bbox": [ + 424, + 140, + 461, + 152 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 139, + 506, + 152 + ], + "score": 1.0, + "content": ", which is", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 151, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 162 + ], + "score": 1.0, + "content": "parametrized by factorial Gaussians, whose means and variances are predicted via an encoder network.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 161, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 105, + 161, + 194, + 174 + ], + "score": 1.0, + "content": "The point cloud latent", + "type": "text" + }, + { + "bbox": [ + 194, + 162, + 206, + 172 + ], + "score": 0.87, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 161, + 406, + 174 + ], + "score": 1.0, + "content": "is sampled from a similarly parametrized posterior", + "type": "text" + }, + { + "bbox": [ + 406, + 162, + 459, + 173 + ], + "score": 0.91, + "content": "q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 161, + 506, + 174 + ], + "score": 1.0, + "content": ", while also", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 173, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 171, + 185 + ], + "score": 1.0, + "content": "conditioning on", + "type": "text" + }, + { + "bbox": [ + 171, + 173, + 182, + 184 + ], + "score": 0.84, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 173, + 187, + 185 + ], + "score": 1.0, + "content": "(", + "type": "text" + }, + { + "bbox": [ + 187, + 173, + 195, + 184 + ], + "score": 0.83, + "content": "\\cdot \\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 173, + 419, + 185 + ], + "score": 1.0, + "content": "denotes the parameters of both encoders). Furthermore,", + "type": "text" + }, + { + "bbox": [ + 419, + 173, + 471, + 185 + ], + "score": 0.93, + "content": "p _ { \\pmb { \\xi } } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 173, + 505, + 185 + ], + "score": 1.0, + "content": "denotes", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 184, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 184, + 505, + 195 + ], + "score": 1.0, + "content": "the decoder, parametrized as a factorial Laplace distribution with predicted means and fixed unit scale", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 194, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 235, + 208 + ], + "score": 1.0, + "content": "parameter (corresponding to an", + "type": "text" + }, + { + "bbox": [ + 236, + 195, + 248, + 205 + ], + "score": 0.88, + "content": "L _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 194, + 334, + 208 + ], + "score": 1.0, + "content": "reconstruction loss).", + "type": "text" + }, + { + "bbox": [ + 334, + 194, + 346, + 205 + ], + "score": 0.87, + "content": "\\lambda _ { \\mathbf { z } }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 194, + 364, + 208 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 365, + 195, + 377, + 205 + ], + "score": 0.89, + "content": "\\lambda _ { \\mathbf { h } }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 194, + 505, + 208 + ], + "score": 1.0, + "content": "are hyperparameters balancing", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 204, + 506, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 420, + 218 + ], + "score": 1.0, + "content": "reconstruction accuracy and Kullback-Leibler regularization (note that only for", + "type": "text" + }, + { + "bbox": [ + 421, + 205, + 475, + 216 + ], + "score": 0.91, + "content": "\\lambda _ { \\mathbf { z } } = \\lambda _ { \\mathbf { h } } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 204, + 506, + 218 + ], + "score": 1.0, + "content": "we are", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 216, + 483, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 272, + 229 + ], + "score": 1.0, + "content": "optimizing a rigorous ELBO). The priors", + "type": "text" + }, + { + "bbox": [ + 273, + 216, + 296, + 228 + ], + "score": 0.92, + "content": "p ( \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 216, + 314, + 229 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 315, + 216, + 339, + 228 + ], + "score": 0.92, + "content": "p ( \\mathbf { h } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 216, + 355, + 229 + ], + "score": 1.0, + "content": "are", + "type": "text" + }, + { + "bbox": [ + 356, + 216, + 389, + 228 + ], + "score": 0.93, + "content": "\\mathcal { N } ( \\mathbf { 0 } , \\pmb { I } )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 216, + 483, + 229 + ], + "score": 1.0, + "content": ". Also see Fig. 1 again.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 139, + 506, + 229 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 505, + 299 + ], + "lines": [ + { + "bbox": [ + 105, + 231, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 245 + ], + "score": 1.0, + "content": "Second Stage Training. In principle, we could use the VAE’s priors to sample encodings and generate", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 244, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 505, + 255 + ], + "score": 1.0, + "content": "new shapes. However, the simple Gaussian priors will not accurately match the encoding distribution", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "score": 1.0, + "content": "from the training data and therefore produce poor samples (prior hole problem [58, 72–79]). This", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 266, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 505, + 277 + ], + "score": 1.0, + "content": "motivates training highly expressive latent DDMs. In particular, in the second stage we freeze the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 428, + 289 + ], + "score": 1.0, + "content": "VAE’s encoder and decoder networks and train two latent DDMs on the encodings", + "type": "text" + }, + { + "bbox": [ + 428, + 277, + 439, + 287 + ], + "score": 0.86, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 276, + 456, + 289 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 457, + 276, + 469, + 287 + ], + "score": 0.89, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 276, + 505, + 289 + ], + "score": 1.0, + "content": "sampled", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 286, + 492, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 128, + 300 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 129, + 287, + 166, + 299 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 286, + 184, + 300 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 185, + 287, + 237, + 299 + ], + "score": 0.93, + "content": "q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 286, + 492, + 300 + ], + "score": 1.0, + "content": ", minimizing score matching (SM) objectives similar to Eq. (2):", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 231, + 505, + 300 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 148, + 303, + 463, + 336 + ], + "lines": [ + { + "bbox": [ + 148, + 303, + 463, + 336 + ], + "spans": [ + { + "bbox": [ + 148, + 303, + 463, + 336 + ], + "score": 0.85, + "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { S M } ^ { \\mathbf { z } } } ( \\pmb { \\theta } ) = \\mathbb { E } _ { t \\sim U \\{ 1 , T \\} , p ( \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } | | \\epsilon - \\epsilon _ { \\theta } ( \\mathbf { z } _ { t } , t ) | | _ { 2 } ^ { 2 } , } \\\\ & { \\mathcal { L } _ { \\mathrm { S M } ^ { \\mathbf { h } } } ( \\pmb { \\psi } ) = \\mathbb { E } _ { t \\sim U \\{ 1 , T \\} , p ( \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { z } _ { 0 } | \\mathbf { x } ) , q _ { \\phi } ( \\mathbf { h } _ { 0 } | \\mathbf { x } , \\mathbf { z } _ { 0 } ) , \\epsilon \\sim \\mathcal { N } ( \\mathbf { 0 } , I ) } | | \\epsilon - \\epsilon _ { \\psi } ( \\mathbf { h } _ { t } , \\mathbf { z } _ { 0 } , t ) | | _ { 2 } ^ { 2 } , } \\end{array}", + "type": "interline_equation", + "image_path": "ebdf63e10a31b4b95e5bbc33639e727966fe0bf1c6ad4195a6e3d40902bc0426.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 148, + 303, + 463, + 314.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 148, + 314.0, + 463, + 325.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 148, + 325.0, + 463, + 336.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 339, + 505, + 372 + ], + "lines": [ + { + "bbox": [ + 106, + 338, + 504, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 338, + 133, + 352 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 340, + 205, + 351 + ], + "score": 0.91, + "content": "{ \\bf z } _ { t } = \\alpha _ { t } { \\bf z } _ { 0 } + \\sigma _ { t } \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 338, + 225, + 352 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 225, + 340, + 298, + 350 + ], + "score": 0.91, + "content": "\\mathbf { h } _ { t } = \\alpha _ { t } \\mathbf { h } _ { 0 } + \\sigma _ { t } \\mathbf { \\epsilon } \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 338, + 496, + 352 + ], + "score": 1.0, + "content": "are the diffused latent encodings. Furthermore,", + "type": "text" + }, + { + "bbox": [ + 497, + 340, + 504, + 349 + ], + "score": 0.67, + "content": "\\pmb \\theta", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 350, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 330, + 362 + ], + "score": 1.0, + "content": "denotes the parameters of the global shape latent DDM", + "type": "text" + }, + { + "bbox": [ + 331, + 350, + 366, + 362 + ], + "score": 0.92, + "content": "\\boldsymbol { \\epsilon } _ { \\boldsymbol { \\theta } } ( \\mathbf { z } _ { t } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 350, + 387, + 362 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 387, + 351, + 396, + 361 + ], + "score": 0.87, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 350, + 506, + 362 + ], + "score": 1.0, + "content": "refers to the parameters of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 361, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 194, + 374 + ], + "score": 1.0, + "content": "the conditional DDM", + "type": "text" + }, + { + "bbox": [ + 194, + 361, + 246, + 373 + ], + "score": 0.92, + "content": "\\mathbf { \\epsilon } _ { \\psi } ( \\mathbf { h } _ { t } , \\bar { \\mathbf { z } _ { 0 } } , t )", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 361, + 487, + 374 + ], + "score": 1.0, + "content": "trained over the latent point cloud (note the conditioning on", + "type": "text" + }, + { + "bbox": [ + 488, + 362, + 498, + 372 + ], + "score": 0.8, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 499, + 361, + 505, + 374 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 106, + 338, + 506, + 374 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 377, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "score": 1.0, + "content": "Generation. With the latent DDMs, we can formally define a hierarchical generative model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 388, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 314, + 400 + ], + "score": 0.87, + "content": "\\begin{array} { r } { p _ { \\xi , \\psi , \\theta } ( \\mathbf { x } , \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } ) = p _ { \\xi } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } ) p _ { \\psi } ( \\mathbf { h } _ { 0 } | \\mathbf { z } _ { 0 } ) p _ { \\theta } ( \\mathbf { z } _ { 0 } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 388, + 347, + 401 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 347, + 388, + 376, + 400 + ], + "score": 0.92, + "content": "p _ { \\pmb { \\theta } } ( \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 388, + 506, + 401 + ], + "score": 1.0, + "content": "denotes the distribution of the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 398, + 507, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 211, + 412 + ], + "score": 1.0, + "content": "global shape latent DDM,", + "type": "text" + }, + { + "bbox": [ + 211, + 400, + 255, + 411 + ], + "score": 0.9, + "content": "p _ { \\psi } ( \\mathbf { h } _ { 0 } | \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 398, + 507, + 412 + ], + "score": 1.0, + "content": "refers to the DDM modeling the point cloud-structured latents,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 409, + 507, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 123, + 423 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 410, + 175, + 422 + ], + "score": 0.92, + "content": "p _ { \\pmb { \\xi } } ( \\mathbf { x } | \\mathbf { h } _ { 0 } , \\mathbf { z } _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 409, + 507, + 423 + ], + "score": 1.0, + "content": "is LION’s decoder. We can hierarchically sample the latent DDMs following Eq. (4)", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 420, + 468, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 468, + 434 + ], + "score": 1.0, + "content": "and then translate the latent points back to the original point cloud space with the decoder.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 377, + 507, + 434 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 437, + 505, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "score": 1.0, + "content": "Network Architectures and DDM Parametrization. Let us briefly summarize key implementation", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "score": 1.0, + "content": "choices. The encoder networks, as well as the decoder and the latent point DDM, operating on point", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 135, + 471 + ], + "score": 1.0, + "content": "clouds", + "type": "text" + }, + { + "bbox": [ + 135, + 461, + 143, + 469 + ], + "score": 0.59, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 459, + 506, + 471 + ], + "score": 1.0, + "content": ", are all implemented based on Point-Voxel CNNs (PVCNNs) [80], following Zhou et al.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 468, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 506, + 483 + ], + "score": 1.0, + "content": "[46]. PVCNNs efficiently combine the point-based processing of PointNets [81, 82] with the strong", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 480, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 494 + ], + "score": 1.0, + "content": "spatial inductive bias of convolutions. The DDM modeling the global shape latent uses a ResNet [83]", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 492, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 324, + 503 + ], + "score": 1.0, + "content": "structure with fully-connected layers (implemented as", + "type": "text" + }, + { + "bbox": [ + 325, + 492, + 344, + 502 + ], + "score": 0.87, + "content": "1 \\times 1", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 492, + 505, + 503 + ], + "score": 1.0, + "content": "-convolutions). All conditionings on the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "global shape latent are implemented via adaptive Group Normalization [84] in the PVCNN layers.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 514, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 506, + 526 + ], + "score": 1.0, + "content": "Furthermore, following Vahdat et al. [58] we use a mixed score parametrization in both latent DDMs.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "This means that the score models are parametrized to predict a residual correction to an analytic", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 535, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 506, + 547 + ], + "score": 1.0, + "content": "standard Gaussian score. This is beneficial since the latent encodings are regularized towards a", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 546, + 459, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 459, + 559 + ], + "score": 1.0, + "content": "standard Gaussian distribution during the first training stage (see App. D for all details).", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 436, + 506, + 559 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 570, + 251, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 252, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 252, + 584 + ], + "score": 1.0, + "content": "3.1 Applications and Extensions", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47 + }, + { + "type": "text", + "bbox": [ + 107, + 590, + 462, + 602 + ], + "lines": [ + { + "bbox": [ + 104, + 588, + 465, + 605 + ], + "spans": [ + { + "bbox": [ + 104, + 588, + 465, + 605 + ], + "score": 1.0, + "content": "Here, we discuss how LION can be used and extended for different relevant applications.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 48, + "bbox_fs": [ + 104, + 588, + 465, + 605 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 606, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 106, + 608, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 506, + 619 + ], + "score": 1.0, + "content": "Multimodal Generation. We can synthesize different variations of a given shape, enabling multi-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 619, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 619, + 420, + 630 + ], + "score": 1.0, + "content": "modal generation in a controlled manner: Given a shape, i.e., its point cloud", + "type": "text" + }, + { + "bbox": [ + 421, + 620, + 428, + 628 + ], + "score": 0.3, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 619, + 505, + 630 + ], + "score": 1.0, + "content": ", we encode it into", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 286, + 641 + ], + "score": 1.0, + "content": "latent space. Then, we diffuse its encodings", + "type": "text" + }, + { + "bbox": [ + 286, + 630, + 297, + 640 + ], + "score": 0.86, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 629, + 315, + 641 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 315, + 629, + 327, + 640 + ], + "score": 0.87, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 629, + 442, + 641 + ], + "score": 1.0, + "content": "for a small number of steps", + "type": "text" + }, + { + "bbox": [ + 442, + 629, + 470, + 639 + ], + "score": 0.9, + "content": "\\tau < T", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "towards", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 638, + 505, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 157, + 654 + ], + "score": 1.0, + "content": "intermediate", + "type": "text" + }, + { + "bbox": [ + 157, + 641, + 169, + 651 + ], + "score": 0.87, + "content": "{ \\bf z } _ { \\tau }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 638, + 185, + 654 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 186, + 640, + 198, + 651 + ], + "score": 0.88, + "content": "{ \\bf h } _ { \\tau }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 638, + 505, + 654 + ], + "score": 1.0, + "content": "along the diffusion process such that only local details are destroyed. Running", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 326, + 663 + ], + "score": 1.0, + "content": "the reverse generation process from this intermediate", + "type": "text" + }, + { + "bbox": [ + 326, + 653, + 333, + 660 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 651, + 380, + 663 + ], + "score": 1.0, + "content": ", starting at", + "type": "text" + }, + { + "bbox": [ + 380, + 652, + 392, + 662 + ], + "score": 0.86, + "content": "{ \\bf z } _ { \\tau }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 651, + 411, + 663 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 411, + 651, + 424, + 662 + ], + "score": 0.87, + "content": "{ \\bf h } _ { \\tau }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 651, + 505, + 663 + ], + "score": 1.0, + "content": ", leads to variations", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 661, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 505, + 674 + ], + "score": 1.0, + "content": "of the original shape with different details (see, for instance, Fig. 2). We refer to this procedure as", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 672, + 489, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 672, + 489, + 685 + ], + "score": 1.0, + "content": "diffuse-denoise (details in App. C.1). Similar techniques have been used for image editing [85].", + "type": "text" + } + ], + "index": 55 + } + ], + "index": 52, + "bbox_fs": [ + 105, + 608, + 506, + 685 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 689, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 703 + ], + "score": 1.0, + "content": "Encoder Fine-tuning for Voxel-Conditioned Synthesis and Denoising. In practice, an artist using", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 700, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 712 + ], + "score": 1.0, + "content": "a 3D generative model may have a rough idea of the desired shape. For instance, they may be able to", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 105, + 711, + 507, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 507, + 723 + ], + "score": 1.0, + "content": "quickly construct a coarse voxelized shape, to which the generative model then adds realistic details.", + "type": "text" + } + ], + "index": 58 + } + ], + "index": 57, + "bbox_fs": [ + 105, + 687, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 128, + 47, + 409, + 142 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 128, + 47, + 409, + 142 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 128, + 47, + 409, + 142 + ], + "spans": [ + { + "bbox": [ + 128, + 47, + 409, + 142 + ], + "score": 0.965, + "type": "image", + "image_path": "85429d590b634ce75ac7892abdbde7713e22da5e9b4c3ea48ef5c5e6e54822dc.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 128, + 47, + 409, + 78.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 128, + 78.66666666666667, + 409, + 110.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 128, + 110.33333333333334, + 409, + 142.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 427, + 50, + 505, + 131 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 426, + 49, + 507, + 63 + ], + "spans": [ + { + "bbox": [ + 426, + 49, + 507, + 63 + ], + "score": 1.0, + "content": "Figure 4: Voxel-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 425, + 60, + 505, + 72 + ], + "spans": [ + { + "bbox": [ + 425, + 61, + 457, + 72 + ], + "score": 1.0, + "content": "guided", + "type": "text" + }, + { + "bbox": [ + 466, + 60, + 505, + 72 + ], + "score": 1.0, + "content": "synthesis", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 426, + 71, + 505, + 81 + ], + "spans": [ + { + "bbox": [ + 426, + 71, + 505, + 81 + ], + "score": 1.0, + "content": "with LION. We run", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 427, + 80, + 505, + 91 + ], + "spans": [ + { + "bbox": [ + 427, + 80, + 486, + 91 + ], + "score": 1.0, + "content": "diffuse-denoise", + "type": "text" + }, + { + "bbox": [ + 490, + 81, + 505, + 91 + ], + "score": 1.0, + "content": "in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 425, + 90, + 505, + 103 + ], + "spans": [ + { + "bbox": [ + 425, + 90, + 505, + 103 + ], + "score": 1.0, + "content": "latent space (see", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 426, + 100, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 426, + 100, + 506, + 113 + ], + "score": 1.0, + "content": "Sec. 3.1) to generate", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 426, + 110, + 506, + 122 + ], + "spans": [ + { + "bbox": [ + 426, + 110, + 457, + 122 + ], + "score": 1.0, + "content": "diverse", + "type": "text" + }, + { + "bbox": [ + 467, + 110, + 506, + 122 + ], + "score": 1.0, + "content": "plausible", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 426, + 120, + 478, + 132 + ], + "spans": [ + { + "bbox": [ + 426, + 120, + 478, + 132 + ], + "score": 1.0, + "content": "clean shapes.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6.5 + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 106, + 144, + 505, + 220 + ], + "lines": [ + { + "bbox": [ + 105, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "In LION, we can support such applications: using a similar ELBO as in Eq. (5), but with a frozen", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "score": 1.0, + "content": "decoder, we can fine-tune LION’s encoder networks to take voxelized shapes as input (we simply", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 164, + 504, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 493, + 179 + ], + "score": 1.0, + "content": "place points at the voxelized shape’s surface) and map them to the corresponding latent encodings", + "type": "text" + }, + { + "bbox": [ + 493, + 167, + 504, + 176 + ], + "score": 0.8, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 176, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 123, + 189 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 176, + 135, + 187 + ], + "score": 0.85, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 176, + 505, + 189 + ], + "score": 1.0, + "content": "that reconstruct the original non-voxelized point cloud. Now, a user can utilize the fine-tuned", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 199 + ], + "score": 1.0, + "content": "encoders to encode voxelized shapes and generate plausible detailed shapes. Importantly, this can", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "score": 1.0, + "content": "be naturally combined with the diffuse-denoise procedure to clean up imperfect encodings and to", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 208, + 327, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 327, + 222 + ], + "score": 1.0, + "content": "generate different possible detailed shapes (see Fig. 4).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 225, + 505, + 280 + ], + "lines": [ + { + "bbox": [ + 106, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "Furthermore, this approach is general. Instead of voxel-conditioned synthesis, we can also fine-tune", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 235, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 249 + ], + "score": 1.0, + "content": "the encoder networks on noisy shapes to perform multimodal shape denoising, also potentially", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 246, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 506, + 260 + ], + "score": 1.0, + "content": "combined with diffuse-denoise. LION supports these applications easily without re-training the latent", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 257, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 505, + 270 + ], + "score": 1.0, + "content": "DDMs due to its VAE framework with additional encoders and decoders, in contrast to previous", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 268, + 474, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 474, + 282 + ], + "score": 1.0, + "content": "works that train DDMs on point clouds directly [46, 47]. See App. C.2 for technical details.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 285, + 504, + 340 + ], + "lines": [ + { + "bbox": [ + 106, + 284, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 505, + 298 + ], + "score": 1.0, + "content": "Shape Interpolation. LION also enables shape interpolation: We can encode different point clouds", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 295, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 309 + ], + "score": 1.0, + "content": "into LION’s hierarchical latent space and use the probability flow ODE (see App. B) to further encode", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "into the latent DDMs’ Gaussian priors, where we can safely perform spherical interpolation and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 317, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 331 + ], + "score": 1.0, + "content": "expect valid shapes along the interpolation path. We can use the intermediate encodings to generate", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 329, + 330, + 341 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 330, + 341 + ], + "score": 1.0, + "content": "the interpolated shapes (see Fig. 7; details in App. C.3).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 345, + 454, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 345, + 424, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 424, + 357 + ], + "score": 1.0, + "content": "Surface Reconstruction. While point clouds are an ideal 3D representation for", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 355, + 424, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 424, + 369 + ], + "score": 1.0, + "content": "DDMs, artists may prefer meshed outputs. Hence, we propose to combine LION", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 367, + 424, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 424, + 380 + ], + "score": 1.0, + "content": "with modern geometry reconstruction methods (see Figs. 2, 4 and 5). We use", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 378, + 424, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 424, + 389 + ], + "score": 1.0, + "content": "Shape As Points (SAP) [68], which is based on differentiable Poisson surface", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 388, + 424, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 424, + 401 + ], + "score": 1.0, + "content": "reconstruction and can be trained to extract smooth meshes from noisy point", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 399, + 425, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 425, + 412 + ], + "score": 1.0, + "content": "clouds. Moreover, we fine-tune SAP on training data generated by LION’s au-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 410, + 425, + 423 + ], + "spans": [ + { + "bbox": [ + 104, + 410, + 425, + 423 + ], + "score": 1.0, + "content": "toencoder to better adjust SAP to the noise distribution in point clouds generated", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 421, + 424, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 424, + 434 + ], + "score": 1.0, + "content": "by LION. Specifically, we take clean shapes, encode them into latent space, run", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 431, + 425, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 431, + 425, + 446 + ], + "score": 1.0, + "content": "a few steps of diffuse-denoise that only slightly modify some details, and decode", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 439, + 475, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 424, + 456 + ], + "score": 1.0, + "content": "back. The diffuse-denoise in latent space results in noise in the generated point", + "type": "text" + }, + { + "bbox": [ + 430, + 439, + 475, + 451 + ], + "score": 1.0, + "content": "ated points.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 453, + 456, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 456, + 467 + ], + "score": 1.0, + "content": "clouds similar to what is observed during unconditional synthesis (details in App. C.4).", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33 + }, + { + "type": "image", + "bbox": [ + 448, + 331, + 491, + 407 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 448, + 331, + 491, + 407 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 448, + 331, + 491, + 407 + ], + "spans": [ + { + "bbox": [ + 448, + 331, + 491, + 407 + ], + "score": 0.841, + "type": "image", + "image_path": "e8246239ce1cb83b69f6bfa0190e0a475971dd591c1a606190b19903eaf14fce.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 448, + 331, + 491, + 407 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 431, + 409, + 506, + 450 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 430, + 408, + 507, + 421 + ], + "spans": [ + { + "bbox": [ + 430, + 408, + 507, + 421 + ], + "score": 1.0, + "content": "Figure 5: Recon-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 431, + 419, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 431, + 419, + 505, + 430 + ], + "score": 1.0, + "content": "structing a mesh", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 430, + 428, + 507, + 441 + ], + "spans": [ + { + "bbox": [ + 430, + 428, + 507, + 441 + ], + "score": 1.0, + "content": "from LION’s gener-", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + } + ], + "index": 40.0 + }, + { + "type": "title", + "bbox": [ + 107, + 473, + 215, + 484 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 217, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 217, + 487 + ], + "score": 1.0, + "content": "3.2 LION’s Advantages", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 106, + 487, + 504, + 510 + ], + "lines": [ + { + "bbox": [ + 106, + 487, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 499 + ], + "score": 1.0, + "content": "We now recapitulate LION’s unique advantages. LION’s structure as a hierarchical VAE with latent", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 497, + 477, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 477, + 510 + ], + "score": 1.0, + "content": "DDMs is inspired by latent DDMs on images [57, 58, 77]. This framework has key benefits:", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 505, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 514, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 506, + 527 + ], + "score": 1.0, + "content": "(i) Expressivity: First training a VAE that regularizes the latent encodings to approximately fall under", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 525, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 537 + ], + "score": 1.0, + "content": "standard Gaussian distributions, which are also the DDMs’ equilibrium distributions towards which", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 536, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 505, + 549 + ], + "score": 1.0, + "content": "the diffusion processes converge, results in an easier modeling task for the DDMs: They have to", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "score": 1.0, + "content": "model only the remaining mismatch between the actual encoding distributions and their own Gaussian", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 558, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 506, + 570 + ], + "score": 1.0, + "content": "priors [58]. This translates into improved expressivity, which is further enhanced by the additional", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 569, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 569, + 506, + 581 + ], + "score": 1.0, + "content": "decoder network. However, point clouds are, in principle, an ideal representation for the DDM frame-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 580, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 506, + 592 + ], + "score": 1.0, + "content": "work, because they can be diffused and denoised easily and powerful point cloud processing architec-", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 591, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 505, + 603 + ], + "score": 1.0, + "content": "tures exist. Therefore, LION uses point cloud latents that combine the advantages of both latent DDMs", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "and 3D point clouds. Our point cloud latents can be interpreted as smoothed versions of the original", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 612, + 507, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 507, + 625 + ], + "score": 1.0, + "content": "point clouds that are easier to model (see Fig. 1). Moreover, the hierarchical VAE setup with an addi-", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 106, + 623, + 506, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 506, + 636 + ], + "score": 1.0, + "content": "tional global shape latent increases LION’s expressivity even further and results in natural disentangle-", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 106, + 635, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 506, + 647 + ], + "score": 1.0, + "content": "ment between overall shape and local details captured by the shape latents and latent points (Sec. 5.2).", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 51.5 + }, + { + "type": "text", + "bbox": [ + 107, + 650, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 651, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 505, + 662 + ], + "score": 1.0, + "content": "(ii) Flexibility: Another advantage of LION’s VAE framework is that its encoders can be fine-tuned", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 662, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 674 + ], + "score": 1.0, + "content": "for various relevant tasks, as discussed previously, and it also enables easy shape interpolation. Other", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "score": 1.0, + "content": "3D point cloud DDMs operating on point clouds directly [47, 46] do not offer simultaneously as", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 683, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 105, + 683, + 506, + 696 + ], + "score": 1.0, + "content": "much flexibility and expressivity out-of-the-box (see quantitative comparisons in Secs. 5.1 and 5.4).", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 59.5 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "(iii) Mesh Reconstruction: As discussed, while point clouds are ideal for DDMs, artists likely prefer", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "meshed outputs. As explained above, we propose to use LION together with modern surface recon-", + "type": "text" + } + ], + "index": 63 + } + ], + "index": 62.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 128, + 47, + 409, + 142 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 128, + 47, + 409, + 142 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 128, + 47, + 409, + 142 + ], + "spans": [ + { + "bbox": [ + 128, + 47, + 409, + 142 + ], + "score": 0.965, + "type": "image", + "image_path": "85429d590b634ce75ac7892abdbde7713e22da5e9b4c3ea48ef5c5e6e54822dc.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 128, + 47, + 409, + 78.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 128, + 78.66666666666667, + 409, + 110.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 128, + 110.33333333333334, + 409, + 142.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 427, + 50, + 505, + 131 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 426, + 49, + 507, + 63 + ], + "spans": [ + { + "bbox": [ + 426, + 49, + 507, + 63 + ], + "score": 1.0, + "content": "Figure 4: Voxel-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 425, + 60, + 505, + 72 + ], + "spans": [ + { + "bbox": [ + 425, + 61, + 457, + 72 + ], + "score": 1.0, + "content": "guided", + "type": "text" + }, + { + "bbox": [ + 466, + 60, + 505, + 72 + ], + "score": 1.0, + "content": "synthesis", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 426, + 71, + 505, + 81 + ], + "spans": [ + { + "bbox": [ + 426, + 71, + 505, + 81 + ], + "score": 1.0, + "content": "with LION. We run", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 427, + 80, + 505, + 91 + ], + "spans": [ + { + "bbox": [ + 427, + 80, + 486, + 91 + ], + "score": 1.0, + "content": "diffuse-denoise", + "type": "text" + }, + { + "bbox": [ + 490, + 81, + 505, + 91 + ], + "score": 1.0, + "content": "in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 425, + 90, + 505, + 103 + ], + "spans": [ + { + "bbox": [ + 425, + 90, + 505, + 103 + ], + "score": 1.0, + "content": "latent space (see", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 426, + 100, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 426, + 100, + 506, + 113 + ], + "score": 1.0, + "content": "Sec. 3.1) to generate", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 426, + 110, + 506, + 122 + ], + "spans": [ + { + "bbox": [ + 426, + 110, + 457, + 122 + ], + "score": 1.0, + "content": "diverse", + "type": "text" + }, + { + "bbox": [ + 467, + 110, + 506, + 122 + ], + "score": 1.0, + "content": "plausible", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 426, + 120, + 478, + 132 + ], + "spans": [ + { + "bbox": [ + 426, + 120, + 478, + 132 + ], + "score": 1.0, + "content": "clean shapes.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6.5 + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 106, + 144, + 505, + 220 + ], + "lines": [ + { + "bbox": [ + 105, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "In LION, we can support such applications: using a similar ELBO as in Eq. (5), but with a frozen", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 505, + 167 + ], + "score": 1.0, + "content": "decoder, we can fine-tune LION’s encoder networks to take voxelized shapes as input (we simply", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 164, + 504, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 493, + 179 + ], + "score": 1.0, + "content": "place points at the voxelized shape’s surface) and map them to the corresponding latent encodings", + "type": "text" + }, + { + "bbox": [ + 493, + 167, + 504, + 176 + ], + "score": 0.8, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 176, + 505, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 123, + 189 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 176, + 135, + 187 + ], + "score": 0.85, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 176, + 505, + 189 + ], + "score": 1.0, + "content": "that reconstruct the original non-voxelized point cloud. Now, a user can utilize the fine-tuned", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 199 + ], + "score": 1.0, + "content": "encoders to encode voxelized shapes and generate plausible detailed shapes. Importantly, this can", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "score": 1.0, + "content": "be naturally combined with the diffuse-denoise procedure to clean up imperfect encodings and to", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 208, + 327, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 327, + 222 + ], + "score": 1.0, + "content": "generate different possible detailed shapes (see Fig. 4).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 142, + 506, + 222 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 225, + 505, + 280 + ], + "lines": [ + { + "bbox": [ + 106, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "Furthermore, this approach is general. Instead of voxel-conditioned synthesis, we can also fine-tune", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 235, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 249 + ], + "score": 1.0, + "content": "the encoder networks on noisy shapes to perform multimodal shape denoising, also potentially", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 246, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 506, + 260 + ], + "score": 1.0, + "content": "combined with diffuse-denoise. LION supports these applications easily without re-training the latent", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 257, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 505, + 270 + ], + "score": 1.0, + "content": "DDMs due to its VAE framework with additional encoders and decoders, in contrast to previous", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 268, + 474, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 474, + 282 + ], + "score": 1.0, + "content": "works that train DDMs on point clouds directly [46, 47]. See App. C.2 for technical details.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 225, + 506, + 282 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 285, + 504, + 340 + ], + "lines": [ + { + "bbox": [ + 106, + 284, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 505, + 298 + ], + "score": 1.0, + "content": "Shape Interpolation. LION also enables shape interpolation: We can encode different point clouds", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 295, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 309 + ], + "score": 1.0, + "content": "into LION’s hierarchical latent space and use the probability flow ODE (see App. B) to further encode", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "into the latent DDMs’ Gaussian priors, where we can safely perform spherical interpolation and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 317, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 331 + ], + "score": 1.0, + "content": "expect valid shapes along the interpolation path. We can use the intermediate encodings to generate", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 329, + 330, + 341 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 330, + 341 + ], + "score": 1.0, + "content": "the interpolated shapes (see Fig. 7; details in App. C.3).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 284, + 505, + 341 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 345, + 454, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 345, + 424, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 424, + 357 + ], + "score": 1.0, + "content": "Surface Reconstruction. While point clouds are an ideal 3D representation for", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 355, + 424, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 424, + 369 + ], + "score": 1.0, + "content": "DDMs, artists may prefer meshed outputs. Hence, we propose to combine LION", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 367, + 424, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 424, + 380 + ], + "score": 1.0, + "content": "with modern geometry reconstruction methods (see Figs. 2, 4 and 5). We use", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 378, + 424, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 424, + 389 + ], + "score": 1.0, + "content": "Shape As Points (SAP) [68], which is based on differentiable Poisson surface", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 388, + 424, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 424, + 401 + ], + "score": 1.0, + "content": "reconstruction and can be trained to extract smooth meshes from noisy point", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 399, + 425, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 425, + 412 + ], + "score": 1.0, + "content": "clouds. Moreover, we fine-tune SAP on training data generated by LION’s au-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 410, + 425, + 423 + ], + "spans": [ + { + "bbox": [ + 104, + 410, + 425, + 423 + ], + "score": 1.0, + "content": "toencoder to better adjust SAP to the noise distribution in point clouds generated", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 421, + 424, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 424, + 434 + ], + "score": 1.0, + "content": "by LION. Specifically, we take clean shapes, encode them into latent space, run", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 431, + 425, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 431, + 425, + 446 + ], + "score": 1.0, + "content": "a few steps of diffuse-denoise that only slightly modify some details, and decode", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 439, + 475, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 424, + 456 + ], + "score": 1.0, + "content": "back. The diffuse-denoise in latent space results in noise in the generated point", + "type": "text" + }, + { + "bbox": [ + 430, + 439, + 475, + 451 + ], + "score": 1.0, + "content": "ated points.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 453, + 456, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 456, + 467 + ], + "score": 1.0, + "content": "clouds similar to what is observed during unconditional synthesis (details in App. C.4).", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33, + "bbox_fs": [ + 104, + 345, + 475, + 467 + ] + }, + { + "type": "image", + "bbox": [ + 448, + 331, + 491, + 407 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 448, + 331, + 491, + 407 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 448, + 331, + 491, + 407 + ], + "spans": [ + { + "bbox": [ + 448, + 331, + 491, + 407 + ], + "score": 0.841, + "type": "image", + "image_path": "e8246239ce1cb83b69f6bfa0190e0a475971dd591c1a606190b19903eaf14fce.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 448, + 331, + 491, + 407 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 431, + 409, + 506, + 450 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 430, + 408, + 507, + 421 + ], + "spans": [ + { + "bbox": [ + 430, + 408, + 507, + 421 + ], + "score": 1.0, + "content": "Figure 5: Recon-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 431, + 419, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 431, + 419, + 505, + 430 + ], + "score": 1.0, + "content": "structing a mesh", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 430, + 428, + 507, + 441 + ], + "spans": [ + { + "bbox": [ + 430, + 428, + 507, + 441 + ], + "score": 1.0, + "content": "from LION’s gener-", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + } + ], + "index": 40.0 + }, + { + "type": "title", + "bbox": [ + 107, + 473, + 215, + 484 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 217, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 217, + 487 + ], + "score": 1.0, + "content": "3.2 LION’s Advantages", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 106, + 487, + 504, + 510 + ], + "lines": [ + { + "bbox": [ + 106, + 487, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 499 + ], + "score": 1.0, + "content": "We now recapitulate LION’s unique advantages. LION’s structure as a hierarchical VAE with latent", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 497, + 477, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 477, + 510 + ], + "score": 1.0, + "content": "DDMs is inspired by latent DDMs on images [57, 58, 77]. This framework has key benefits:", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5, + "bbox_fs": [ + 106, + 487, + 505, + 510 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 505, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 514, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 506, + 527 + ], + "score": 1.0, + "content": "(i) Expressivity: First training a VAE that regularizes the latent encodings to approximately fall under", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 525, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 505, + 537 + ], + "score": 1.0, + "content": "standard Gaussian distributions, which are also the DDMs’ equilibrium distributions towards which", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 536, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 505, + 549 + ], + "score": 1.0, + "content": "the diffusion processes converge, results in an easier modeling task for the DDMs: They have to", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "score": 1.0, + "content": "model only the remaining mismatch between the actual encoding distributions and their own Gaussian", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 558, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 506, + 570 + ], + "score": 1.0, + "content": "priors [58]. This translates into improved expressivity, which is further enhanced by the additional", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 569, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 569, + 506, + 581 + ], + "score": 1.0, + "content": "decoder network. However, point clouds are, in principle, an ideal representation for the DDM frame-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 580, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 506, + 592 + ], + "score": 1.0, + "content": "work, because they can be diffused and denoised easily and powerful point cloud processing architec-", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 591, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 505, + 603 + ], + "score": 1.0, + "content": "tures exist. Therefore, LION uses point cloud latents that combine the advantages of both latent DDMs", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "and 3D point clouds. Our point cloud latents can be interpreted as smoothed versions of the original", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 612, + 507, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 507, + 625 + ], + "score": 1.0, + "content": "point clouds that are easier to model (see Fig. 1). Moreover, the hierarchical VAE setup with an addi-", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 106, + 623, + 506, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 506, + 636 + ], + "score": 1.0, + "content": "tional global shape latent increases LION’s expressivity even further and results in natural disentangle-", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 106, + 635, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 506, + 647 + ], + "score": 1.0, + "content": "ment between overall shape and local details captured by the shape latents and latent points (Sec. 5.2).", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 51.5, + "bbox_fs": [ + 105, + 514, + 507, + 647 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 650, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 651, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 505, + 662 + ], + "score": 1.0, + "content": "(ii) Flexibility: Another advantage of LION’s VAE framework is that its encoders can be fine-tuned", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 662, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 674 + ], + "score": 1.0, + "content": "for various relevant tasks, as discussed previously, and it also enables easy shape interpolation. Other", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "score": 1.0, + "content": "3D point cloud DDMs operating on point clouds directly [47, 46] do not offer simultaneously as", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 683, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 105, + 683, + 506, + 696 + ], + "score": 1.0, + "content": "much flexibility and expressivity out-of-the-box (see quantitative comparisons in Secs. 5.1 and 5.4).", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 59.5, + "bbox_fs": [ + 105, + 651, + 506, + 696 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "(iii) Mesh Reconstruction: As discussed, while point clouds are ideal for DDMs, artists likely prefer", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "meshed outputs. As explained above, we propose to use LION together with modern surface recon-", + "type": "text" + } + ], + "index": 63 + }, + { + "bbox": [ + 105, + 186, + 506, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 506, + 199 + ], + "score": 1.0, + "content": "struction techniques [68], again combining the best of both worlds—a point cloud-based VAE back-", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "score": 1.0, + "content": "bone ideal for DDMs, and smooth geometry reconstruction methods operating on the synthesized point", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 208, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 106, + 208, + 506, + 221 + ], + "score": 1.0, + "content": "clouds to generate practically useful smooth surfaces, which can be easily transformed into meshes.", + "type": "text", + "cross_page": true + } + ], + "index": 17 + } + ], + "index": 62.5, + "bbox_fs": [ + 106, + 700, + 505, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 48, + 426, + 173 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 48, + 426, + 173 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 48, + 426, + 173 + ], + "spans": [ + { + "bbox": [ + 109, + 48, + 426, + 173 + ], + "score": 0.953, + "type": "image", + "image_path": "3a7faca45d0a42a04bfe777486d20f82a5448e7fd0bfdf16d03a15e9ff8b2a17.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 109, + 48, + 426, + 89.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 89.66666666666666, + 426, + 131.33333333333331 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 109, + 131.33333333333331, + 426, + 172.99999999999997 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 435, + 47, + 505, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 434, + 47, + 507, + 59 + ], + "spans": [ + { + "bbox": [ + 434, + 47, + 507, + 59 + ], + "score": 1.0, + "content": "Figure 6: Un-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 434, + 58, + 506, + 70 + ], + "spans": [ + { + "bbox": [ + 434, + 58, + 506, + 70 + ], + "score": 1.0, + "content": "conditional shape", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 433, + 68, + 506, + 79 + ], + "spans": [ + { + "bbox": [ + 433, + 68, + 506, + 79 + ], + "score": 1.0, + "content": "generation with", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 433, + 77, + 506, + 90 + ], + "spans": [ + { + "bbox": [ + 433, + 77, + 506, + 90 + ], + "score": 1.0, + "content": "2,048 points for", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 433, + 87, + 506, + 100 + ], + "spans": [ + { + "bbox": [ + 433, + 87, + 472, + 100 + ], + "score": 1.0, + "content": "airplane,", + "type": "text" + }, + { + "bbox": [ + 489, + 89, + 506, + 99 + ], + "score": 1.0, + "content": "car", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 434, + 98, + 506, + 109 + ], + "spans": [ + { + "bbox": [ + 434, + 98, + 506, + 109 + ], + "score": 1.0, + "content": "and chair classes", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 433, + 108, + 490, + 120 + ], + "spans": [ + { + "bbox": [ + 433, + 108, + 490, + 120 + ], + "score": 1.0, + "content": "(class-specific", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 434, + 117, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 434, + 118, + 466, + 128 + ], + "score": 1.0, + "content": "models", + "type": "text" + }, + { + "bbox": [ + 475, + 117, + 506, + 129 + ], + "score": 1.0, + "content": "trained", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 434, + 127, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 434, + 129, + 447, + 138 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 456, + 127, + 506, + 138 + ], + "score": 1.0, + "content": "PointFlow’s", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 434, + 137, + 507, + 150 + ], + "spans": [ + { + "bbox": [ + 434, + 137, + 475, + 150 + ], + "score": 1.0, + "content": "ShapeNet", + "type": "text" + }, + { + "bbox": [ + 486, + 137, + 507, + 149 + ], + "score": 1.0, + "content": "data", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 433, + 147, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 433, + 147, + 455, + 159 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 477, + 147, + 506, + 160 + ], + "score": 1.0, + "content": "global", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 433, + 157, + 494, + 169 + ], + "spans": [ + { + "bbox": [ + 433, + 157, + 494, + 169 + ], + "score": 1.0, + "content": "normalization).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 7.5 + } + ], + "index": 6.75 + }, + { + "type": "text", + "bbox": [ + 107, + 186, + 505, + 220 + ], + "lines": [ + { + "bbox": [ + 105, + 186, + 506, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 506, + 199 + ], + "score": 1.0, + "content": "struction techniques [68], again combining the best of both worlds—a point cloud-based VAE back-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "score": 1.0, + "content": "bone ideal for DDMs, and smooth geometry reconstruction methods operating on the synthesized point", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 208, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 106, + 208, + 506, + 221 + ], + "score": 1.0, + "content": "clouds to generate practically useful smooth surfaces, which can be easily transformed into meshes.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 107, + 238, + 197, + 252 + ], + "lines": [ + { + "bbox": [ + 105, + 238, + 198, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 198, + 254 + ], + "score": 1.0, + "content": "4 Related Work", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 265, + 506, + 364 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 278 + ], + "score": 1.0, + "content": "We are building on DDMs [53, 65–67], which have been used most prominently for image [53–63] and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 275, + 507, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 507, + 289 + ], + "score": 1.0, + "content": "speech synthesis [86–91]. We train DDMs in latent space, an idea that has been explored for image [57,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 288, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 506, + 299 + ], + "score": 1.0, + "content": "58, 77] and music [92] generation, too. However, these works did not train separate conditional DDMs.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 298, + 507, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 507, + 311 + ], + "score": 1.0, + "content": "Hierarchical DDM training has been used for generative image upsampling [54], text-to-image gen-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 309, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 506, + 322 + ], + "score": 1.0, + "content": "eration [63, 64], and semantic image modeling [60]. Most relevant among these works is Preechakul", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 320, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 506, + 333 + ], + "score": 1.0, + "content": "et al. [60], which extracts a high-level semantic representation of an image with an auxiliary encoder", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 330, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 505, + 343 + ], + "score": 1.0, + "content": "and then trains a DDM that adds details directly in image space. We are the first to explore related", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 342, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 354 + ], + "score": 1.0, + "content": "concepts for 3D shape synthesis and we also train both DDMs in latent space. Furthermore, DDMs", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "and VAEs have also been combined in such a way that the DDM improves the output of the VAE [93].", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 368, + 506, + 566 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 506, + 381 + ], + "score": 1.0, + "content": "Most related to LION are “Point-Voxel Diffusion” (PVD) [46] and “Diffusion Probabilistic Models", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 380, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 392 + ], + "score": 1.0, + "content": "for 3D Point Cloud Generation” (DPM) [47]. PVD trains a DDM directly on point clouds, and our", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "decision to use PVCNNs is inspired by this work. DPM, like LION, uses a shape latent variable,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "score": 1.0, + "content": "but models its distribution with Normalizing Flows [94, 95], and then trains a weaker point-wise", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 414, + 505, + 425 + ], + "score": 1.0, + "content": "conditional DDM directly on the point cloud data (this allows DPM to learn useful representations in", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 424, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 436 + ], + "score": 1.0, + "content": "its latent variable, but sacrifices generation quality). As we show below, neither PVD nor DPM easily", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 433, + 507, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 507, + 448 + ], + "score": 1.0, + "content": "enables applications such as multimodal voxel-conditioned synthesis and denoising. Furthermore,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 445, + 507, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 507, + 459 + ], + "score": 1.0, + "content": "LION achieves significantly stronger generation performance. Finally, neither PVD nor DPM recon-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "structs meshes from the generated point clouds. Point cloud and 3D shape generation have also been", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "explored with other generative models: PointFlow [31], DPF-Net [33] and SoftFlow [32] rely on Nor-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 478, + 507, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 507, + 491 + ], + "score": 1.0, + "content": "malizing Flows [94–97]. SetVAE [29] treats point cloud synthesis as set generation and uses VAEs.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 488, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 503 + ], + "score": 1.0, + "content": "ShapeGF [45] learns distributions over gradient fields that model shape surfaces. Both IM-GAN [7],", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "which models shapes as neural fields, and l-GAN [2] train GANs over latent variables that encode the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 511, + 507, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 507, + 524 + ], + "score": 1.0, + "content": "shapes, similar to other works [3], while r-GAN [2] generates point clouds directly. PDGN [52] pro-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 522, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 104, + 522, + 506, + 535 + ], + "score": 1.0, + "content": "poses progressive deconvolutional networks within a point cloud GAN. SP-GAN [19] uses a spherical", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 104, + 532, + 507, + 546 + ], + "spans": [ + { + "bbox": [ + 104, + 532, + 507, + 546 + ], + "score": 1.0, + "content": "point cloud prior. Other progressive [22, 37] and graph-based architectures [4, 6] have been used, too.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 544, + 507, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 507, + 556 + ], + "score": 1.0, + "content": "Also generative cellular automata (GCAs) can be employed for voxel-based 3D shape generation [43].", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 554, + 495, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 495, + 567 + ], + "score": 1.0, + "content": "In orthogonal work, point cloud DDMs have been used for generative shape completion [46, 98].", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 107, + 571, + 504, + 626 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "score": 1.0, + "content": "Recently, image-driven [8–16, 44] training of 3D generative models as well as text-driven 3D", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "score": 1.0, + "content": "generation [34, 49–51] have received much attention. These are complementary directions to ours;", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 593, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 505, + 605 + ], + "score": 1.0, + "content": "in fact, augmenting LION with additional image-based training or including text-guidance are", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 603, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 506, + 618 + ], + "score": 1.0, + "content": "promising future directions. Finally, we are relying on SAP [68] for mesh generation. Strong", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 615, + 473, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 473, + 627 + ], + "score": 1.0, + "content": "alternative approaches for reconstructing smooth surfaces from point clouds exist [99–103].", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48 + }, + { + "type": "image", + "bbox": [ + 110, + 635, + 419, + 718 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 635, + 419, + 718 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 635, + 419, + 718 + ], + "spans": [ + { + "bbox": [ + 110, + 635, + 419, + 718 + ], + "score": 0.932, + "type": "image", + "image_path": "4415796a634b4f9ff0b2056e1bed79ad61f4bef0f5dc3c549987eef0c051b295.jpg" + } + ] + } + ], + "index": 52, + "virtual_lines": [ + { + "bbox": [ + 110, + 635, + 419, + 662.6666666666666 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 110, + 662.6666666666666, + 419, + 690.3333333333333 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 110, + 690.3333333333333, + 419, + 717.9999999999999 + ], + "spans": [], + "index": 53 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 427, + 633, + 505, + 714 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 426, + 632, + 507, + 645 + ], + "spans": [ + { + "bbox": [ + 426, + 632, + 507, + 645 + ], + "score": 1.0, + "content": "Figure 7: Interpolat-", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 426, + 643, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 426, + 643, + 506, + 655 + ], + "score": 1.0, + "content": "ing different shapes", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 425, + 653, + 506, + 665 + ], + "spans": [ + { + "bbox": [ + 425, + 653, + 506, + 665 + ], + "score": 1.0, + "content": "by interpolating their", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 426, + 663, + 507, + 675 + ], + "spans": [ + { + "bbox": [ + 426, + 663, + 507, + 675 + ], + "score": 1.0, + "content": "encodings in the stan-", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 426, + 672, + 507, + 685 + ], + "spans": [ + { + "bbox": [ + 426, + 672, + 507, + 685 + ], + "score": 1.0, + "content": "dard Gaussian pri-", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 427, + 683, + 507, + 694 + ], + "spans": [ + { + "bbox": [ + 427, + 683, + 507, + 694 + ], + "score": 1.0, + "content": "ors of LION’s la-", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 426, + 693, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 426, + 693, + 505, + 704 + ], + "score": 1.0, + "content": "tent DDMs (details in", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 426, + 703, + 468, + 715 + ], + "spans": [ + { + "bbox": [ + 426, + 703, + 468, + 715 + ], + "score": 1.0, + "content": "App. C.3).", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 57.5 + } + ], + "index": 54.75 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 48, + 426, + 173 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 48, + 426, + 173 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 48, + 426, + 173 + ], + "spans": [ + { + "bbox": [ + 109, + 48, + 426, + 173 + ], + "score": 0.953, + "type": "image", + "image_path": "3a7faca45d0a42a04bfe777486d20f82a5448e7fd0bfdf16d03a15e9ff8b2a17.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 109, + 48, + 426, + 89.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 89.66666666666666, + 426, + 131.33333333333331 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 109, + 131.33333333333331, + 426, + 172.99999999999997 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 435, + 47, + 505, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 434, + 47, + 507, + 59 + ], + "spans": [ + { + "bbox": [ + 434, + 47, + 507, + 59 + ], + "score": 1.0, + "content": "Figure 6: Un-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 434, + 58, + 506, + 70 + ], + "spans": [ + { + "bbox": [ + 434, + 58, + 506, + 70 + ], + "score": 1.0, + "content": "conditional shape", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 433, + 68, + 506, + 79 + ], + "spans": [ + { + "bbox": [ + 433, + 68, + 506, + 79 + ], + "score": 1.0, + "content": "generation with", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 433, + 77, + 506, + 90 + ], + "spans": [ + { + "bbox": [ + 433, + 77, + 506, + 90 + ], + "score": 1.0, + "content": "2,048 points for", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 433, + 87, + 506, + 100 + ], + "spans": [ + { + "bbox": [ + 433, + 87, + 472, + 100 + ], + "score": 1.0, + "content": "airplane,", + "type": "text" + }, + { + "bbox": [ + 489, + 89, + 506, + 99 + ], + "score": 1.0, + "content": "car", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 434, + 98, + 506, + 109 + ], + "spans": [ + { + "bbox": [ + 434, + 98, + 506, + 109 + ], + "score": 1.0, + "content": "and chair classes", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 433, + 108, + 490, + 120 + ], + "spans": [ + { + "bbox": [ + 433, + 108, + 490, + 120 + ], + "score": 1.0, + "content": "(class-specific", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 434, + 117, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 434, + 118, + 466, + 128 + ], + "score": 1.0, + "content": "models", + "type": "text" + }, + { + "bbox": [ + 475, + 117, + 506, + 129 + ], + "score": 1.0, + "content": "trained", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 434, + 127, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 434, + 129, + 447, + 138 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 456, + 127, + 506, + 138 + ], + "score": 1.0, + "content": "PointFlow’s", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 434, + 137, + 507, + 150 + ], + "spans": [ + { + "bbox": [ + 434, + 137, + 475, + 150 + ], + "score": 1.0, + "content": "ShapeNet", + "type": "text" + }, + { + "bbox": [ + 486, + 137, + 507, + 149 + ], + "score": 1.0, + "content": "data", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 433, + 147, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 433, + 147, + 455, + 159 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 477, + 147, + 506, + 160 + ], + "score": 1.0, + "content": "global", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 433, + 157, + 494, + 169 + ], + "spans": [ + { + "bbox": [ + 433, + 157, + 494, + 169 + ], + "score": 1.0, + "content": "normalization).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 7.5 + } + ], + "index": 6.75 + }, + { + "type": "text", + "bbox": [ + 107, + 186, + 505, + 220 + ], + "lines": [], + "index": 16, + "bbox_fs": [ + 105, + 186, + 506, + 221 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 238, + 197, + 252 + ], + "lines": [ + { + "bbox": [ + 105, + 238, + 198, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 198, + 254 + ], + "score": 1.0, + "content": "4 Related Work", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 265, + 506, + 364 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 506, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 278 + ], + "score": 1.0, + "content": "We are building on DDMs [53, 65–67], which have been used most prominently for image [53–63] and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 275, + 507, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 507, + 289 + ], + "score": 1.0, + "content": "speech synthesis [86–91]. We train DDMs in latent space, an idea that has been explored for image [57,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 288, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 506, + 299 + ], + "score": 1.0, + "content": "58, 77] and music [92] generation, too. However, these works did not train separate conditional DDMs.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 298, + 507, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 507, + 311 + ], + "score": 1.0, + "content": "Hierarchical DDM training has been used for generative image upsampling [54], text-to-image gen-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 309, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 506, + 322 + ], + "score": 1.0, + "content": "eration [63, 64], and semantic image modeling [60]. Most relevant among these works is Preechakul", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 320, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 506, + 333 + ], + "score": 1.0, + "content": "et al. [60], which extracts a high-level semantic representation of an image with an auxiliary encoder", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 330, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 505, + 343 + ], + "score": 1.0, + "content": "and then trains a DDM that adds details directly in image space. We are the first to explore related", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 342, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 354 + ], + "score": 1.0, + "content": "concepts for 3D shape synthesis and we also train both DDMs in latent space. Furthermore, DDMs", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "and VAEs have also been combined in such a way that the DDM improves the output of the VAE [93].", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 264, + 507, + 365 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 368, + 506, + 566 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 506, + 381 + ], + "score": 1.0, + "content": "Most related to LION are “Point-Voxel Diffusion” (PVD) [46] and “Diffusion Probabilistic Models", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 380, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 392 + ], + "score": 1.0, + "content": "for 3D Point Cloud Generation” (DPM) [47]. PVD trains a DDM directly on point clouds, and our", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "decision to use PVCNNs is inspired by this work. DPM, like LION, uses a shape latent variable,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 415 + ], + "score": 1.0, + "content": "but models its distribution with Normalizing Flows [94, 95], and then trains a weaker point-wise", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 414, + 505, + 425 + ], + "score": 1.0, + "content": "conditional DDM directly on the point cloud data (this allows DPM to learn useful representations in", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 424, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 436 + ], + "score": 1.0, + "content": "its latent variable, but sacrifices generation quality). As we show below, neither PVD nor DPM easily", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 433, + 507, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 507, + 448 + ], + "score": 1.0, + "content": "enables applications such as multimodal voxel-conditioned synthesis and denoising. Furthermore,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 445, + 507, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 507, + 459 + ], + "score": 1.0, + "content": "LION achieves significantly stronger generation performance. Finally, neither PVD nor DPM recon-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "structs meshes from the generated point clouds. Point cloud and 3D shape generation have also been", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "explored with other generative models: PointFlow [31], DPF-Net [33] and SoftFlow [32] rely on Nor-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 478, + 507, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 507, + 491 + ], + "score": 1.0, + "content": "malizing Flows [94–97]. SetVAE [29] treats point cloud synthesis as set generation and uses VAEs.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 488, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 503 + ], + "score": 1.0, + "content": "ShapeGF [45] learns distributions over gradient fields that model shape surfaces. Both IM-GAN [7],", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "which models shapes as neural fields, and l-GAN [2] train GANs over latent variables that encode the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 511, + 507, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 507, + 524 + ], + "score": 1.0, + "content": "shapes, similar to other works [3], while r-GAN [2] generates point clouds directly. PDGN [52] pro-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 522, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 104, + 522, + 506, + 535 + ], + "score": 1.0, + "content": "poses progressive deconvolutional networks within a point cloud GAN. SP-GAN [19] uses a spherical", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 104, + 532, + 507, + 546 + ], + "spans": [ + { + "bbox": [ + 104, + 532, + 507, + 546 + ], + "score": 1.0, + "content": "point cloud prior. Other progressive [22, 37] and graph-based architectures [4, 6] have been used, too.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 544, + 507, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 507, + 556 + ], + "score": 1.0, + "content": "Also generative cellular automata (GCAs) can be employed for voxel-based 3D shape generation [43].", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 554, + 495, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 495, + 567 + ], + "score": 1.0, + "content": "In orthogonal work, point cloud DDMs have been used for generative shape completion [46, 98].", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 36.5, + "bbox_fs": [ + 104, + 368, + 507, + 567 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 571, + 504, + 626 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "score": 1.0, + "content": "Recently, image-driven [8–16, 44] training of 3D generative models as well as text-driven 3D", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "score": 1.0, + "content": "generation [34, 49–51] have received much attention. These are complementary directions to ours;", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 593, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 505, + 605 + ], + "score": 1.0, + "content": "in fact, augmenting LION with additional image-based training or including text-guidance are", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 603, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 506, + 618 + ], + "score": 1.0, + "content": "promising future directions. Finally, we are relying on SAP [68] for mesh generation. Strong", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 615, + 473, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 473, + 627 + ], + "score": 1.0, + "content": "alternative approaches for reconstructing smooth surfaces from point clouds exist [99–103].", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48, + "bbox_fs": [ + 105, + 570, + 506, + 627 + ] + }, + { + "type": "image", + "bbox": [ + 110, + 635, + 419, + 718 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 635, + 419, + 718 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 110, + 635, + 419, + 718 + ], + "spans": [ + { + "bbox": [ + 110, + 635, + 419, + 718 + ], + "score": 0.932, + "type": "image", + "image_path": "4415796a634b4f9ff0b2056e1bed79ad61f4bef0f5dc3c549987eef0c051b295.jpg" + } + ] + } + ], + "index": 52, + "virtual_lines": [ + { + "bbox": [ + 110, + 635, + 419, + 662.6666666666666 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 110, + 662.6666666666666, + 419, + 690.3333333333333 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 110, + 690.3333333333333, + 419, + 717.9999999999999 + ], + "spans": [], + "index": 53 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 427, + 633, + 505, + 714 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 426, + 632, + 507, + 645 + ], + "spans": [ + { + "bbox": [ + 426, + 632, + 507, + 645 + ], + "score": 1.0, + "content": "Figure 7: Interpolat-", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 426, + 643, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 426, + 643, + 506, + 655 + ], + "score": 1.0, + "content": "ing different shapes", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 425, + 653, + 506, + 665 + ], + "spans": [ + { + "bbox": [ + 425, + 653, + 506, + 665 + ], + "score": 1.0, + "content": "by interpolating their", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 426, + 663, + 507, + 675 + ], + "spans": [ + { + "bbox": [ + 426, + 663, + 507, + 675 + ], + "score": 1.0, + "content": "encodings in the stan-", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 426, + 672, + 507, + 685 + ], + "spans": [ + { + "bbox": [ + 426, + 672, + 507, + 685 + ], + "score": 1.0, + "content": "dard Gaussian pri-", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 427, + 683, + 507, + 694 + ], + "spans": [ + { + "bbox": [ + 427, + 683, + 507, + 694 + ], + "score": 1.0, + "content": "ors of LION’s la-", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 426, + 693, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 426, + 693, + 505, + 704 + ], + "score": 1.0, + "content": "tent DDMs (details in", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 426, + 703, + 468, + 715 + ], + "spans": [ + { + "bbox": [ + 426, + 703, + 468, + 715 + ], + "score": 1.0, + "content": "App. C.3).", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 57.5 + } + ], + "index": 54.75 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 109, + 111, + 326, + 244 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 81, + 320, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 81, + 320, + 93 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 320, + 93 + ], + "score": 1.0, + "content": "Table 1: Generation metrics (1-NNA↓) on airplane, chair,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 91, + 321, + 103 + ], + "spans": [ + { + "bbox": [ + 106, + 91, + 321, + 103 + ], + "score": 1.0, + "content": "car categories from ShapeNet dataset from PointFlow [31].", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 100, + 303, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 303, + 113 + ], + "score": 1.0, + "content": "Training and test data normalized globally into [-1, 1].", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 109, + 111, + 326, + 244 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 111, + 326, + 244 + ], + "spans": [ + { + "bbox": [ + 109, + 111, + 326, + 244 + ], + "score": 0.98, + "html": "
AirplaneChairCar
CD EMDCDEMDCDEMD
r-GAN[2]98.40 96.7983.6999.7094.46599.01
1-GAN (CD) [2]87.30 )93.9568.5883.8466.4988.78
1-GAN (EMD)[2]89.49 76.9171.9064.6571.1666.19
PointFlow [31]75.68 370.7462.8460.5758.1056.25
SoftFlow [32]76.05 65.8059.2160.0564.7760.09
SetVAE [29]76.54 67.6558.8460.5759.9459.94
DPF-Net [33]75.18 65.5562.0058.5362.3554.48
DPM[47]76.42 86.9160.0574.7768.8979.97
PVD [46]73.82 64.8156.2653.3254.5553.83
LION (ours)67.41 61.2353.7052.3453.41 51.14
", + "type": "table", + "image_path": "ebb39d1c01aa3875199305b3330015b1cff8fdde2ef129455e9fa1343d0c3270.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 109, + 111, + 326, + 123.0909090909091 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 123.0909090909091, + 326, + 135.1818181818182 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 109, + 135.1818181818182, + 326, + 147.27272727272728 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 109, + 147.27272727272728, + 326, + 159.36363636363637 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 109, + 159.36363636363637, + 326, + 171.45454545454547 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 109, + 171.45454545454547, + 326, + 183.54545454545456 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 109, + 183.54545454545456, + 326, + 195.63636363636365 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 109, + 195.63636363636365, + 326, + 207.72727272727275 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 109, + 207.72727272727275, + 326, + 219.81818181818184 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 109, + 219.81818181818184, + 326, + 231.90909090909093 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 109, + 231.90909090909093, + 326, + 244.00000000000003 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 4.5 + }, + { + "type": "table", + "bbox": [ + 333, + 100, + 500, + 182 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 329, + 69, + 502, + 99 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 329, + 69, + 502, + 81 + ], + "spans": [ + { + "bbox": [ + 329, + 69, + 502, + 81 + ], + "score": 1.0, + "content": "Table 2: Generation results (1-NNA↓) on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 330, + 79, + 503, + 90 + ], + "spans": [ + { + "bbox": [ + 330, + 79, + 503, + 90 + ], + "score": 1.0, + "content": "ShapeNet dataset from PointFlow [31]. All data", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 329, + 88, + 461, + 101 + ], + "spans": [ + { + "bbox": [ + 329, + 88, + 461, + 101 + ], + "score": 1.0, + "content": "normalized individually into [-1, 1].", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "table_body", + "bbox": [ + 333, + 100, + 500, + 182 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 333, + 100, + 500, + 182 + ], + "spans": [ + { + "bbox": [ + 333, + 100, + 500, + 182 + ], + "score": 0.957, + "html": "
AirplaneChairCar
CDEMDCD EMDCD EMD
TreeGAN[ [6]97.53 99.88 88.3796.37 89.7794.89
ShapeGF[ [45]81.23 80.8658.0161.25 61.7957.24
SP-GAN[19]94.69 93.9572.5883.6987.36 85.94
PDGN [52]94.94 91.7371.8379.00 89.35 87.22
GCA [43]88.15 85.9364.2764.50 70.4564.20
LION(ours)76.30 67.04 56.50 53.85 59.52 49.29
", + "type": "table", + "image_path": "793bb3e9675457e4e80978eea616561af9ac7c1712012a40fe45e51606bd100c.jpg" + } + ] + } + ], + "index": 19.5, + "virtual_lines": [ + { + "bbox": [ + 333, + 100, + 500, + 113.66666666666667 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 333, + 113.66666666666667, + 500, + 127.33333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 333, + 127.33333333333334, + 500, + 141.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 333, + 141.0, + 500, + 154.66666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 333, + 154.66666666666666, + 500, + 168.33333333333331 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 333, + 168.33333333333331, + 500, + 181.99999999999997 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 17.25 + }, + { + "type": "table", + "bbox": [ + 333, + 199, + 501, + 270 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 334, + 188, + 499, + 199 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 332, + 187, + 500, + 200 + ], + "spans": [ + { + "bbox": [ + 332, + 187, + 500, + 200 + ], + "score": 1.0, + "content": "Table 3: Results (1-NNA↓) on ShapeNet-vol.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "table_body", + "bbox": [ + 333, + 199, + 501, + 270 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 333, + 199, + 501, + 270 + ], + "spans": [ + { + "bbox": [ + 333, + 199, + 501, + 270 + ], + "score": 0.929, + "html": "
AirplaneChairCar
CD EMDCD EMDCD EMD
IM-GAN[7]79.70 77.85 57.0958.20 88.9284.58
DPM[47] PVD [46]83.04 96.04 61.96 66.46 56.06 61.8974.96 77.30 57.90 64.4987.12 55.74
LION (ours) 53.47 53.84 52.0748.6754.81 50.53
", + "type": "table", + "image_path": "19fe95ccaec57a967b905abdcd29634f4271ede433ec69399086a8087bd349e1.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 333, + 199, + 501, + 213.2 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 333, + 213.2, + 501, + 227.39999999999998 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 333, + 227.39999999999998, + 501, + 241.59999999999997 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 333, + 241.59999999999997, + 501, + 255.79999999999995 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 333, + 255.79999999999995, + 501, + 269.99999999999994 + ], + "spans": [], + "index": 29 + } + ] + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 106, + 258, + 191, + 272 + ], + "lines": [ + { + "bbox": [ + 103, + 256, + 193, + 276 + ], + "spans": [ + { + "bbox": [ + 103, + 256, + 193, + 276 + ], + "score": 1.0, + "content": "5 Experiments", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 278, + 503, + 300 + ], + "lines": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "score": 1.0, + "content": "We provide an overview of our most interesting experimental results in the main paper. All experiment", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 288, + 481, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 481, + 301 + ], + "score": 1.0, + "content": "details and extensive additional experiments can be found in App. E and App. F, respectively.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "title", + "bbox": [ + 107, + 305, + 277, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 278, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 278, + 320 + ], + "score": 1.0, + "content": "5.1 Single-Class 3D Shape Generation", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 320, + 384, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 320, + 384, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 384, + 332 + ], + "score": 1.0, + "content": "Datasets. To compare LION against existing methods, we use", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 330, + 385, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 385, + 343 + ], + "score": 1.0, + "content": "ShapeNet [104], the most widely used dataset to benchmark 3D", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 342, + 385, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 385, + 354 + ], + "score": 1.0, + "content": "shape generative models. Following previous works [31, 46, 47],", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 353, + 385, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 385, + 364 + ], + "score": 1.0, + "content": "we train on three categories: airplane, chair, car. Also like previ-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 364, + 385, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 385, + 375 + ], + "score": 1.0, + "content": "ous methods, we primarily rely on PointFlow’s [31] dataset splits", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 375, + 385, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 385, + 386 + ], + "score": 1.0, + "content": "and preprocssing. It normalizes the data globally across the whole", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 384, + 386, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 386, + 398 + ], + "score": 1.0, + "content": "dataset. However, some baselines require per-shape normaliza-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 396, + 385, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 385, + 408 + ], + "score": 1.0, + "content": "tion [19, 43, 45, 52]; hence, we also train on such data. Furthermore,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 407, + 385, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 385, + 420 + ], + "score": 1.0, + "content": "training SAP requires signed distance fields (SDFs) for volumetric", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 418, + 385, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 385, + 430 + ], + "score": 1.0, + "content": "supervision, which the PointFlow data does not offer. Hence, for", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 429, + 384, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 384, + 441 + ], + "score": 1.0, + "content": "simplicity we follow Peng et al. [68, 101] and also use their data", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 440, + 385, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 385, + 452 + ], + "score": 1.0, + "content": "splits and preprocessing, which includes SDFs.We train LION, DPM,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 450, + 382, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 382, + 463 + ], + "score": 1.0, + "content": "PVD, and IM-GAN (which synthesizes shapes as SDFs) also on", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 461, + 380, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 380, + 474 + ], + "score": 1.0, + "content": "as ShapeNet-vol here). This data is also per-shape normalized.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 107, + 478, + 384, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 386, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 386, + 490 + ], + "score": 1.0, + "content": "Evaluation. Model evaluation follows previous works [31, 46]. Var-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 489, + 386, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 386, + 500 + ], + "score": 1.0, + "content": "ious metrics to evaluate point cloud generative models exist, with dif-", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 499, + 386, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 386, + 512 + ], + "score": 1.0, + "content": "ferent advantages and disadvantages, discussed in detail by Yang et al.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 510, + 386, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 386, + 523 + ], + "score": 1.0, + "content": "[31]. Following recent works [31, 46], we use 1-NNA (with both Cham-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 522, + 385, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 385, + 533 + ], + "score": 1.0, + "content": "fer distance (CD) and earth mover distance (EMD)) as our main metric.", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 533, + 385, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 385, + 545 + ], + "score": 1.0, + "content": "It quantifies the distributional similarity between generated shapes and", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 543, + 385, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 385, + 555 + ], + "score": 1.0, + "content": "validation set and measures both quality and diversity [31]. For fair", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 555, + 385, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 385, + 566 + ], + "score": 1.0, + "content": "comparisons, all metrics are computed on point clouds, not meshed out-", + "type": "text" + } + ], + "index": 65 + }, + { + "bbox": [ + 105, + 565, + 385, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 385, + 579 + ], + "score": 1.0, + "content": "puts (App. E.2 discusses different metrics; further results on coverage", + "type": "text" + } + ], + "index": 66 + }, + { + "bbox": [ + 105, + 576, + 356, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 356, + 589 + ], + "score": 1.0, + "content": "(COV) and minimum matching distance (MMD) in App. F.2).", + "type": "text" + } + ], + "index": 67 + } + ], + "index": 51.5 + }, + { + "type": "image", + "bbox": [ + 394, + 307, + 509, + 401 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 394, + 307, + 509, + 401 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 394, + 307, + 509, + 401 + ], + "spans": [ + { + "bbox": [ + 394, + 307, + 509, + 401 + ], + "score": 0.958, + "type": "image", + "image_path": "5b55e07dfb5406b0da8472c26a7620d4b2d2f093be889f901eda2c6ce9df0988.jpg" + } + ] + } + ], + "index": 54.5, + "virtual_lines": [ + { + "bbox": [ + 394, + 307, + 509, + 354.0 + ], + "spans": [], + "index": 54 + }, + { + "bbox": [ + 394, + 354.0, + 509, + 401.0 + ], + "spans": [], + "index": 55 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 391, + 406, + 505, + 446 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 391, + 405, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 391, + 405, + 506, + 417 + ], + "score": 1.0, + "content": "Figure 8: Samples from our", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 391, + 416, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 391, + 416, + 506, + 426 + ], + "score": 1.0, + "content": "unconditional 13-class model:", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 391, + 426, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 391, + 426, + 505, + 436 + ], + "score": 1.0, + "content": "In each column, we use the", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 390, + 435, + 496, + 448 + ], + "spans": [ + { + "bbox": [ + 390, + 435, + 481, + 448 + ], + "score": 1.0, + "content": "same global shape latent", + "type": "text" + }, + { + "bbox": [ + 482, + 437, + 492, + 446 + ], + "score": 0.83, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 435, + 496, + 448 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 59 + } + ], + "index": 57.5 + } + ], + "index": 56.0 + }, + { + "type": "text", + "bbox": [ + 388, + 451, + 505, + 473 + ], + "lines": [ + { + "bbox": [ + 385, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 385, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "this dataset version (denoted", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 386, + 462, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 386, + 462, + 506, + 474 + ], + "score": 1.0, + "content": "Dataset details in App. E.1.", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 60.5 + }, + { + "type": "table", + "bbox": [ + 393, + 520, + 506, + 621 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 392, + 483, + 505, + 514 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 390, + 482, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 390, + 482, + 506, + 494 + ], + "score": 1.0, + "content": "Table 4: Generation results (1-", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 390, + 492, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 390, + 492, + 506, + 505 + ], + "score": 1.0, + "content": "NNA↓) of LION trained jointly", + "type": "text" + } + ], + "index": 63 + }, + { + "bbox": [ + 391, + 504, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 391, + 504, + 505, + 515 + ], + "score": 1.0, + "content": "on 13 classes of ShapeNet-vol.", + "type": "text" + } + ], + "index": 64 + } + ], + "index": 63 + }, + { + "type": "table_body", + "bbox": [ + 393, + 520, + 506, + 621 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 393, + 520, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 393, + 520, + 506, + 621 + ], + "score": 0.967, + "html": "
ModelCDEMD
TreeGAN [6]96.8096.60
PointFlow [31]63.2566.05
ShapeGF[45]55.6559.00
SetVAE [29]79.2595.25
PDGN [52]71.0586.00
DPF-Net [33]67.1064.75
DPM[47]62.3086.50
PVD [46]58.6557.85
LION (ours)51.8548.95
", + "type": "table", + "image_path": "e83d289bc06c0f5fc610ca5bc6ef67d763a882fba325d2247a81695db22c08c2.jpg" + } + ] + } + ], + "index": 71.5, + "virtual_lines": [ + { + "bbox": [ + 393, + 520, + 506, + 570.5 + ], + "spans": [], + "index": 71 + }, + { + "bbox": [ + 393, + 570.5, + 506, + 621.0 + ], + "spans": [], + "index": 72 + } + ] + } + ], + "index": 67.25 + }, + { + "type": "text", + "bbox": [ + 106, + 592, + 383, + 626 + ], + "lines": [ + { + "bbox": [ + 105, + 592, + 384, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 384, + 605 + ], + "score": 1.0, + "content": "Results. Samples from LION are shown in Fig. 6 and quantitative", + "type": "text" + } + ], + "index": 68 + }, + { + "bbox": [ + 105, + 603, + 384, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 384, + 616 + ], + "score": 1.0, + "content": "results in Tabs. 1-3 (see Sec. 4 for details about baselines—to reduce", + "type": "text" + } + ], + "index": 69 + }, + { + "bbox": [ + 105, + 614, + 385, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 385, + 627 + ], + "score": 1.0, + "content": "the number of baselines to train, we are focusing on the most recent", + "type": "text" + } + ], + "index": 70 + } + ], + "index": 69 + }, + { + "type": "text", + "bbox": [ + 107, + 626, + 506, + 659 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 506, + 639 + ], + "score": 1.0, + "content": "and competitive ones). LION outperforms all baselines and achieves state-of-the-art performance", + "type": "text" + } + ], + "index": 73 + }, + { + "bbox": [ + 106, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "on all classes and dataset versions. Importantly, we outperform both PVD and DPM, which also", + "type": "text" + } + ], + "index": 74 + }, + { + "bbox": [ + 105, + 646, + 463, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 463, + 661 + ], + "score": 1.0, + "content": "leverage DDMs, by large margins. Our samples are diverse and appear visually pleasing.", + "type": "text" + } + ], + "index": 75 + } + ], + "index": 74 + }, + { + "type": "image", + "bbox": [ + 111, + 665, + 486, + 701 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 665, + 486, + 701 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 111, + 665, + 486, + 701 + ], + "spans": [ + { + "bbox": [ + 111, + 665, + 486, + 701 + ], + "score": 0.943, + "type": "image", + "image_path": "9b9618423aeeac129aa5e6e358feb49f5e1ebb18eaf986c614bfbfb4c5e35683.jpg" + } + ] + } + ], + "index": 77, + "virtual_lines": [ + { + "bbox": [ + 111, + 665, + 486, + 677.0 + ], + "spans": [], + "index": 76 + }, + { + "bbox": [ + 111, + 677.0, + 486, + 689.0 + ], + "spans": [], + "index": 77 + }, + { + "bbox": [ + 111, + 689.0, + 486, + 701.0 + ], + "spans": [], + "index": 78 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 104, + 708, + 503, + 720 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 706, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 706, + 505, + 722 + ], + "score": 1.0, + "content": "Figure 9: Generated point clouds from LION trained jointly over 55 classes of ShapeNet-vol (no conditioning).", + "type": "text" + } + ], + "index": 79 + } + ], + "index": 79 + } + ], + "index": 78.0 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 109, + 111, + 326, + 244 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 81, + 320, + 111 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 81, + 320, + 93 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 320, + 93 + ], + "score": 1.0, + "content": "Table 1: Generation metrics (1-NNA↓) on airplane, chair,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 91, + 321, + 103 + ], + "spans": [ + { + "bbox": [ + 106, + 91, + 321, + 103 + ], + "score": 1.0, + "content": "car categories from ShapeNet dataset from PointFlow [31].", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 100, + 303, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 303, + 113 + ], + "score": 1.0, + "content": "Training and test data normalized globally into [-1, 1].", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 109, + 111, + 326, + 244 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 111, + 326, + 244 + ], + "spans": [ + { + "bbox": [ + 109, + 111, + 326, + 244 + ], + "score": 0.98, + "html": "
AirplaneChairCar
CD EMDCDEMDCDEMD
r-GAN[2]98.40 96.7983.6999.7094.46599.01
1-GAN (CD) [2]87.30 )93.9568.5883.8466.4988.78
1-GAN (EMD)[2]89.49 76.9171.9064.6571.1666.19
PointFlow [31]75.68 370.7462.8460.5758.1056.25
SoftFlow [32]76.05 65.8059.2160.0564.7760.09
SetVAE [29]76.54 67.6558.8460.5759.9459.94
DPF-Net [33]75.18 65.5562.0058.5362.3554.48
DPM[47]76.42 86.9160.0574.7768.8979.97
PVD [46]73.82 64.8156.2653.3254.5553.83
LION (ours)67.41 61.2353.7052.3453.41 51.14
", + "type": "table", + "image_path": "ebb39d1c01aa3875199305b3330015b1cff8fdde2ef129455e9fa1343d0c3270.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 109, + 111, + 326, + 123.0909090909091 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 109, + 123.0909090909091, + 326, + 135.1818181818182 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 109, + 135.1818181818182, + 326, + 147.27272727272728 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 109, + 147.27272727272728, + 326, + 159.36363636363637 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 109, + 159.36363636363637, + 326, + 171.45454545454547 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 109, + 171.45454545454547, + 326, + 183.54545454545456 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 109, + 183.54545454545456, + 326, + 195.63636363636365 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 109, + 195.63636363636365, + 326, + 207.72727272727275 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 109, + 207.72727272727275, + 326, + 219.81818181818184 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 109, + 219.81818181818184, + 326, + 231.90909090909093 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 109, + 231.90909090909093, + 326, + 244.00000000000003 + ], + "spans": [], + "index": 13 + } + ] + } + ], + "index": 4.5 + }, + { + "type": "table", + "bbox": [ + 333, + 100, + 500, + 182 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 329, + 69, + 502, + 99 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 329, + 69, + 502, + 81 + ], + "spans": [ + { + "bbox": [ + 329, + 69, + 502, + 81 + ], + "score": 1.0, + "content": "Table 2: Generation results (1-NNA↓) on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 330, + 79, + 503, + 90 + ], + "spans": [ + { + "bbox": [ + 330, + 79, + 503, + 90 + ], + "score": 1.0, + "content": "ShapeNet dataset from PointFlow [31]. All data", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 329, + 88, + 461, + 101 + ], + "spans": [ + { + "bbox": [ + 329, + 88, + 461, + 101 + ], + "score": 1.0, + "content": "normalized individually into [-1, 1].", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "table_body", + "bbox": [ + 333, + 100, + 500, + 182 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 333, + 100, + 500, + 182 + ], + "spans": [ + { + "bbox": [ + 333, + 100, + 500, + 182 + ], + "score": 0.957, + "html": "
AirplaneChairCar
CDEMDCD EMDCD EMD
TreeGAN[ [6]97.53 99.88 88.3796.37 89.7794.89
ShapeGF[ [45]81.23 80.8658.0161.25 61.7957.24
SP-GAN[19]94.69 93.9572.5883.6987.36 85.94
PDGN [52]94.94 91.7371.8379.00 89.35 87.22
GCA [43]88.15 85.9364.2764.50 70.4564.20
LION(ours)76.30 67.04 56.50 53.85 59.52 49.29
", + "type": "table", + "image_path": "793bb3e9675457e4e80978eea616561af9ac7c1712012a40fe45e51606bd100c.jpg" + } + ] + } + ], + "index": 19.5, + "virtual_lines": [ + { + "bbox": [ + 333, + 100, + 500, + 113.66666666666667 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 333, + 113.66666666666667, + 500, + 127.33333333333334 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 333, + 127.33333333333334, + 500, + 141.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 333, + 141.0, + 500, + 154.66666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 333, + 154.66666666666666, + 500, + 168.33333333333331 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 333, + 168.33333333333331, + 500, + 181.99999999999997 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 17.25 + }, + { + "type": "table", + "bbox": [ + 333, + 199, + 501, + 270 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 334, + 188, + 499, + 199 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 332, + 187, + 500, + 200 + ], + "spans": [ + { + "bbox": [ + 332, + 187, + 500, + 200 + ], + "score": 1.0, + "content": "Table 3: Results (1-NNA↓) on ShapeNet-vol.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "table_body", + "bbox": [ + 333, + 199, + 501, + 270 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 333, + 199, + 501, + 270 + ], + "spans": [ + { + "bbox": [ + 333, + 199, + 501, + 270 + ], + "score": 0.929, + "html": "
AirplaneChairCar
CD EMDCD EMDCD EMD
IM-GAN[7]79.70 77.85 57.0958.20 88.9284.58
DPM[47] PVD [46]83.04 96.04 61.96 66.46 56.06 61.8974.96 77.30 57.90 64.4987.12 55.74
LION (ours) 53.47 53.84 52.0748.6754.81 50.53
", + "type": "table", + "image_path": "19fe95ccaec57a967b905abdcd29634f4271ede433ec69399086a8087bd349e1.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 333, + 199, + 501, + 213.2 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 333, + 213.2, + 501, + 227.39999999999998 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 333, + 227.39999999999998, + 501, + 241.59999999999997 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 333, + 241.59999999999997, + 501, + 255.79999999999995 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 333, + 255.79999999999995, + 501, + 269.99999999999994 + ], + "spans": [], + "index": 29 + } + ] + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 106, + 258, + 191, + 272 + ], + "lines": [ + { + "bbox": [ + 103, + 256, + 193, + 276 + ], + "spans": [ + { + "bbox": [ + 103, + 256, + 193, + 276 + ], + "score": 1.0, + "content": "5 Experiments", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 278, + 503, + 300 + ], + "lines": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 290 + ], + "score": 1.0, + "content": "We provide an overview of our most interesting experimental results in the main paper. All experiment", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 288, + 481, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 481, + 301 + ], + "score": 1.0, + "content": "details and extensive additional experiments can be found in App. E and App. F, respectively.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 277, + 505, + 301 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 305, + 277, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 278, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 278, + 320 + ], + "score": 1.0, + "content": "5.1 Single-Class 3D Shape Generation", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 320, + 384, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 320, + 384, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 384, + 332 + ], + "score": 1.0, + "content": "Datasets. To compare LION against existing methods, we use", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 330, + 385, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 385, + 343 + ], + "score": 1.0, + "content": "ShapeNet [104], the most widely used dataset to benchmark 3D", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 342, + 385, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 385, + 354 + ], + "score": 1.0, + "content": "shape generative models. Following previous works [31, 46, 47],", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 353, + 385, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 385, + 364 + ], + "score": 1.0, + "content": "we train on three categories: airplane, chair, car. Also like previ-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 364, + 385, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 385, + 375 + ], + "score": 1.0, + "content": "ous methods, we primarily rely on PointFlow’s [31] dataset splits", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 375, + 385, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 385, + 386 + ], + "score": 1.0, + "content": "and preprocssing. It normalizes the data globally across the whole", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 384, + 386, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 386, + 398 + ], + "score": 1.0, + "content": "dataset. However, some baselines require per-shape normaliza-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 396, + 385, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 385, + 408 + ], + "score": 1.0, + "content": "tion [19, 43, 45, 52]; hence, we also train on such data. Furthermore,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 407, + 385, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 385, + 420 + ], + "score": 1.0, + "content": "training SAP requires signed distance fields (SDFs) for volumetric", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 418, + 385, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 385, + 430 + ], + "score": 1.0, + "content": "supervision, which the PointFlow data does not offer. Hence, for", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 429, + 384, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 384, + 441 + ], + "score": 1.0, + "content": "simplicity we follow Peng et al. [68, 101] and also use their data", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 440, + 385, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 385, + 452 + ], + "score": 1.0, + "content": "splits and preprocessing, which includes SDFs.We train LION, DPM,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 450, + 382, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 382, + 463 + ], + "score": 1.0, + "content": "PVD, and IM-GAN (which synthesizes shapes as SDFs) also on", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 461, + 380, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 380, + 474 + ], + "score": 1.0, + "content": "as ShapeNet-vol here). This data is also per-shape normalized.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 320, + 386, + 474 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 478, + 384, + 588 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 386, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 386, + 490 + ], + "score": 1.0, + "content": "Evaluation. Model evaluation follows previous works [31, 46]. Var-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 489, + 386, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 386, + 500 + ], + "score": 1.0, + "content": "ious metrics to evaluate point cloud generative models exist, with dif-", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 499, + 386, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 386, + 512 + ], + "score": 1.0, + "content": "ferent advantages and disadvantages, discussed in detail by Yang et al.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 510, + 386, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 386, + 523 + ], + "score": 1.0, + "content": "[31]. Following recent works [31, 46], we use 1-NNA (with both Cham-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 522, + 385, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 385, + 533 + ], + "score": 1.0, + "content": "fer distance (CD) and earth mover distance (EMD)) as our main metric.", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 533, + 385, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 385, + 545 + ], + "score": 1.0, + "content": "It quantifies the distributional similarity between generated shapes and", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 543, + 385, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 385, + 555 + ], + "score": 1.0, + "content": "validation set and measures both quality and diversity [31]. For fair", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 555, + 385, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 385, + 566 + ], + "score": 1.0, + "content": "comparisons, all metrics are computed on point clouds, not meshed out-", + "type": "text" + } + ], + "index": 65 + }, + { + "bbox": [ + 105, + 565, + 385, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 385, + 579 + ], + "score": 1.0, + "content": "puts (App. E.2 discusses different metrics; further results on coverage", + "type": "text" + } + ], + "index": 66 + }, + { + "bbox": [ + 105, + 576, + 356, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 356, + 589 + ], + "score": 1.0, + "content": "(COV) and minimum matching distance (MMD) in App. F.2).", + "type": "text" + } + ], + "index": 67 + } + ], + "index": 51.5, + "bbox_fs": [ + 105, + 478, + 386, + 589 + ] + }, + { + "type": "image", + "bbox": [ + 394, + 307, + 509, + 401 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 394, + 307, + 509, + 401 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 394, + 307, + 509, + 401 + ], + "spans": [ + { + "bbox": [ + 394, + 307, + 509, + 401 + ], + "score": 0.958, + "type": "image", + "image_path": "5b55e07dfb5406b0da8472c26a7620d4b2d2f093be889f901eda2c6ce9df0988.jpg" + } + ] + } + ], + "index": 54.5, + "virtual_lines": [ + { + "bbox": [ + 394, + 307, + 509, + 354.0 + ], + "spans": [], + "index": 54 + }, + { + "bbox": [ + 394, + 354.0, + 509, + 401.0 + ], + "spans": [], + "index": 55 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 391, + 406, + 505, + 446 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 391, + 405, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 391, + 405, + 506, + 417 + ], + "score": 1.0, + "content": "Figure 8: Samples from our", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 391, + 416, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 391, + 416, + 506, + 426 + ], + "score": 1.0, + "content": "unconditional 13-class model:", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 391, + 426, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 391, + 426, + 505, + 436 + ], + "score": 1.0, + "content": "In each column, we use the", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 390, + 435, + 496, + 448 + ], + "spans": [ + { + "bbox": [ + 390, + 435, + 481, + 448 + ], + "score": 1.0, + "content": "same global shape latent", + "type": "text" + }, + { + "bbox": [ + 482, + 437, + 492, + 446 + ], + "score": 0.83, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 435, + 496, + 448 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 59 + } + ], + "index": 57.5 + } + ], + "index": 56.0 + }, + { + "type": "text", + "bbox": [ + 388, + 451, + 505, + 473 + ], + "lines": [ + { + "bbox": [ + 385, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 385, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "this dataset version (denoted", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 386, + 462, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 386, + 462, + 506, + 474 + ], + "score": 1.0, + "content": "Dataset details in App. E.1.", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 60.5, + "bbox_fs": [ + 385, + 450, + 506, + 474 + ] + }, + { + "type": "table", + "bbox": [ + 393, + 520, + 506, + 621 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 392, + 483, + 505, + 514 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 390, + 482, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 390, + 482, + 506, + 494 + ], + "score": 1.0, + "content": "Table 4: Generation results (1-", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 390, + 492, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 390, + 492, + 506, + 505 + ], + "score": 1.0, + "content": "NNA↓) of LION trained jointly", + "type": "text" + } + ], + "index": 63 + }, + { + "bbox": [ + 391, + 504, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 391, + 504, + 505, + 515 + ], + "score": 1.0, + "content": "on 13 classes of ShapeNet-vol.", + "type": "text" + } + ], + "index": 64 + } + ], + "index": 63 + }, + { + "type": "table_body", + "bbox": [ + 393, + 520, + 506, + 621 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 393, + 520, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 393, + 520, + 506, + 621 + ], + "score": 0.967, + "html": "
ModelCDEMD
TreeGAN [6]96.8096.60
PointFlow [31]63.2566.05
ShapeGF[45]55.6559.00
SetVAE [29]79.2595.25
PDGN [52]71.0586.00
DPF-Net [33]67.1064.75
DPM[47]62.3086.50
PVD [46]58.6557.85
LION (ours)51.8548.95
", + "type": "table", + "image_path": "e83d289bc06c0f5fc610ca5bc6ef67d763a882fba325d2247a81695db22c08c2.jpg" + } + ] + } + ], + "index": 71.5, + "virtual_lines": [ + { + "bbox": [ + 393, + 520, + 506, + 570.5 + ], + "spans": [], + "index": 71 + }, + { + "bbox": [ + 393, + 570.5, + 506, + 621.0 + ], + "spans": [], + "index": 72 + } + ] + } + ], + "index": 67.25 + }, + { + "type": "text", + "bbox": [ + 106, + 592, + 383, + 626 + ], + "lines": [ + { + "bbox": [ + 105, + 592, + 384, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 384, + 605 + ], + "score": 1.0, + "content": "Results. Samples from LION are shown in Fig. 6 and quantitative", + "type": "text" + } + ], + "index": 68 + }, + { + "bbox": [ + 105, + 603, + 384, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 384, + 616 + ], + "score": 1.0, + "content": "results in Tabs. 1-3 (see Sec. 4 for details about baselines—to reduce", + "type": "text" + } + ], + "index": 69 + }, + { + "bbox": [ + 105, + 614, + 385, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 385, + 627 + ], + "score": 1.0, + "content": "the number of baselines to train, we are focusing on the most recent", + "type": "text" + } + ], + "index": 70 + }, + { + "bbox": [ + 105, + 624, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 506, + 639 + ], + "score": 1.0, + "content": "and competitive ones). LION outperforms all baselines and achieves state-of-the-art performance", + "type": "text" + } + ], + "index": 73 + }, + { + "bbox": [ + 106, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "on all classes and dataset versions. Importantly, we outperform both PVD and DPM, which also", + "type": "text" + } + ], + "index": 74 + }, + { + "bbox": [ + 105, + 646, + 463, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 463, + 661 + ], + "score": 1.0, + "content": "leverage DDMs, by large margins. Our samples are diverse and appear visually pleasing.", + "type": "text" + } + ], + "index": 75 + } + ], + "index": 69, + "bbox_fs": [ + 105, + 592, + 385, + 627 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 626, + 506, + 659 + ], + "lines": [], + "index": 74, + "bbox_fs": [ + 105, + 624, + 506, + 661 + ], + "lines_deleted": true + }, + { + "type": "image", + "bbox": [ + 111, + 665, + 486, + 701 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 665, + 486, + 701 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 111, + 665, + 486, + 701 + ], + "spans": [ + { + "bbox": [ + 111, + 665, + 486, + 701 + ], + "score": 0.943, + "type": "image", + "image_path": "9b9618423aeeac129aa5e6e358feb49f5e1ebb18eaf986c614bfbfb4c5e35683.jpg" + } + ] + } + ], + "index": 77, + "virtual_lines": [ + { + "bbox": [ + 111, + 665, + 486, + 677.0 + ], + "spans": [], + "index": 76 + }, + { + "bbox": [ + 111, + 677.0, + 486, + 689.0 + ], + "spans": [], + "index": 77 + }, + { + "bbox": [ + 111, + 689.0, + 486, + 701.0 + ], + "spans": [], + "index": 78 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 104, + 708, + 503, + 720 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 706, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 706, + 505, + 722 + ], + "score": 1.0, + "content": "Figure 9: Generated point clouds from LION trained jointly over 55 classes of ShapeNet-vol (no conditioning).", + "type": "text" + } + ], + "index": 79 + } + ], + "index": 79 + } + ], + "index": 78.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 57, + 506, + 157 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 57, + 506, + 157 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 57, + 506, + 157 + ], + "spans": [ + { + "bbox": [ + 109, + 57, + 506, + 157 + ], + "score": 0.967, + "type": "image", + "image_path": "7e583fae1e920eca77cfbcfa30b8e62210c56760f2035b8269f5be44b2ae0c30.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 57, + 506, + 90.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 90.33333333333334, + 506, + 123.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 123.66666666666669, + 506, + 157.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 163, + 501, + 174 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 161, + 502, + 175 + ], + "spans": [ + { + "bbox": [ + 109, + 161, + 502, + 175 + ], + "score": 1.0, + "content": "Figure 10: Samples from LION trained on ShapeNet’s Mug and Bottle classes, and on Turbosquid animals.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 107, + 185, + 505, + 240 + ], + "lines": [ + { + "bbox": [ + 106, + 185, + 506, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 506, + 198 + ], + "score": 1.0, + "content": "Mesh Reconstruction. As explained in Sec. 3.1, we combine LION with mesh reconstruction, to", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 196, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 106, + 196, + 505, + 208 + ], + "score": 1.0, + "content": "directly synthesize practically useful meshes. We show generated meshes in Fig. 2, which look", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "score": 1.0, + "content": "smooth and of high quality. In Fig. 2, we also visually demonstrate how we can vary the local details", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 218, + 507, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 507, + 231 + ], + "score": 1.0, + "content": "of synthesized shapes while preserving the overall shape with our diffuse-denoise technique (Sec. 3.1).", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 228, + 479, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 479, + 242 + ], + "score": 1.0, + "content": "Details about the number of diffusion steps for all diffuse-denoise experiments are in App. E.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 245, + 505, + 289 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "score": 1.0, + "content": "Shape Interpolation. As discussed in Sec. 3.1, LION also enables shape interpolation, potentially", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 256, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 506, + 268 + ], + "score": 1.0, + "content": "useful for shape editing applications. We show this in Fig. 7, combined with mesh reconstruction.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 267, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 506, + 280 + ], + "score": 1.0, + "content": "The generated shapes are clean and semantically plausible along the entire interpolation path. In", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 278, + 460, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 460, + 291 + ], + "score": 1.0, + "content": "App. F.12.1, we also show interpolations from PVD [46] and DPM [47] for comparison.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "title", + "bbox": [ + 107, + 304, + 335, + 316 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 336, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 336, + 320 + ], + "score": 1.0, + "content": "5.2 Many-class Unconditional 3D Shape Generation", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 322, + 506, + 442 + ], + "lines": [ + { + "bbox": [ + 105, + 322, + 506, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 334 + ], + "score": 1.0, + "content": "13-Class LION Model. We train a LION model jointly without any class conditioning on 13 different", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 333, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 506, + 345 + ], + "score": 1.0, + "content": "categories (airplane, chair, car, lamp, table, sofa, cabinet, bench, telephone, loudspeaker, display, wa-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 342, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 358 + ], + "score": 1.0, + "content": "tercraft, rifle) from ShapeNet (ShapeNet-vol version). Training a single model without conditioning", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 355, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 506, + 367 + ], + "score": 1.0, + "content": "over such diverse shapes is challenging, as the data distribution is highly complex and multimodal.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 364, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 506, + 379 + ], + "score": 1.0, + "content": "We show LION’s generated samples in Fig. 3, including meshes: LION synthesizes high-quality and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "score": 1.0, + "content": "diverse plausible shapes even when trained on such complex data. We report the model’s quantitative", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 387, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 506, + 400 + ], + "score": 1.0, + "content": "generation performance in Tab. 4, and we also trained various strong baseline methods under the same", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 398, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 506, + 411 + ], + "score": 1.0, + "content": "setting for comparison. We find that LION significantly outperforms all baselines by a large margin.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "score": 1.0, + "content": "We further observe that the hierarchical VAE architecture of LION becomes crucial: The shape latent", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 420, + 507, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 139, + 433 + ], + "score": 1.0, + "content": "variable", + "type": "text" + }, + { + "bbox": [ + 140, + 421, + 151, + 431 + ], + "score": 0.85, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 420, + 329, + 433 + ], + "score": 1.0, + "content": "captures global shape, while the latent points", + "type": "text" + }, + { + "bbox": [ + 330, + 420, + 342, + 431 + ], + "score": 0.86, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 420, + 507, + 433 + ], + "score": 1.0, + "content": "model details. This can be seen in Fig. 8:", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 431, + 497, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 319, + 444 + ], + "score": 1.0, + "content": "we show samples when fixing the global shape latent", + "type": "text" + }, + { + "bbox": [ + 320, + 433, + 330, + 442 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 432, + 399, + 444 + ], + "score": 1.0, + "content": "and only sample", + "type": "text" + }, + { + "bbox": [ + 399, + 431, + 412, + 442 + ], + "score": 0.86, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 432, + 497, + 444 + ], + "score": 1.0, + "content": "(details in App. F.3).", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 505, + 568 + ], + "lines": [ + { + "bbox": [ + 106, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "55-Class LION Model. Encouraged by these results, we also trained a LION model again jointly", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 458, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 458, + 505, + 471 + ], + "score": 1.0, + "content": "without any class conditioning on all 55 different categories from ShapeNet. Note that we did on", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "score": 1.0, + "content": "purpose not use class-conditioning in these experiments to create a difficult 3D generation task and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 480, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 506, + 492 + ], + "score": 1.0, + "content": "thereby explore LION’s scalability to highly complex and multimodal datasets. We show generated", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 491, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 506, + 503 + ], + "score": 1.0, + "content": "point cloud samples in Fig. 9 (we did not train an SAP model on the 55 classes data): LION", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 502, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 514 + ], + "score": 1.0, + "content": "synthesizes high-quality and diverse shapes. It can even generate samples from the cap class, which", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "score": 1.0, + "content": "contributes with only 39 training data samples, indicating that LION has an excellent mode coverage", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "score": 1.0, + "content": "that even includes the very rare classes. To the best of our knowledge no previous 3D shape generative", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "models have demonstrated satisfactory generation performance for such diverse and multimodal 3D", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 546, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 506, + 558 + ], + "score": 1.0, + "content": "data without relying on conditioning information (details in App. F.4). In conclusion, we observe that", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 556, + 453, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 453, + 569 + ], + "score": 1.0, + "content": "LION out-of-the-box easily scales to highly complex multi-category shape generation.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30 + }, + { + "type": "title", + "bbox": [ + 108, + 577, + 273, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 275, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 275, + 591 + ], + "score": 1.0, + "content": "5.3 Training LION on Small Datasets", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 504, + 617 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "score": 1.0, + "content": "Next, we explore whether LION can also be trained successfully on very small datasets. To this", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "end, we train models on the Mug and Bottle ShapeNet classes. The number of training samples", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "image", + "bbox": [ + 115, + 625, + 352, + 725 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 625, + 352, + 725 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 115, + 625, + 352, + 725 + ], + "spans": [ + { + "bbox": [ + 115, + 625, + 352, + 725 + ], + "score": 0.964, + "type": "image", + "image_path": "2ade93952b2c1687a4aed247344a7e707a70e0829b0b8547f8b519ee299a389b.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 115, + 625, + 352, + 639.2857142857143 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 115, + 639.2857142857143, + 352, + 653.5714285714287 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 115, + 653.5714285714287, + 352, + 667.857142857143 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 115, + 667.857142857143, + 352, + 682.1428571428573 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 115, + 682.1428571428573, + 352, + 696.4285714285717 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 115, + 696.4285714285717, + 352, + 710.714285714286 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 115, + 710.714285714286, + 352, + 725.0000000000003 + ], + "spans": [], + "index": 45 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 357, + 630, + 502, + 711 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 357, + 630, + 502, + 641 + ], + "spans": [ + { + "bbox": [ + 357, + 630, + 502, + 641 + ], + "score": 1.0, + "content": "Figure 11: Voxel-guided synthesis. We", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 357, + 641, + 502, + 651 + ], + "spans": [ + { + "bbox": [ + 357, + 641, + 502, + 651 + ], + "score": 1.0, + "content": "show different methods with 0 and 50", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 357, + 651, + 503, + 661 + ], + "spans": [ + { + "bbox": [ + 357, + 651, + 503, + 661 + ], + "score": 1.0, + "content": "steps of diffuse-denoise. Voxelizations", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 357, + 660, + 504, + 672 + ], + "spans": [ + { + "bbox": [ + 357, + 660, + 504, + 672 + ], + "score": 1.0, + "content": "of generated points are also shown: Yel-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 357, + 671, + 504, + 681 + ], + "spans": [ + { + "bbox": [ + 357, + 671, + 504, + 681 + ], + "score": 1.0, + "content": "low boxes indicate generated points cor-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 357, + 680, + 504, + 691 + ], + "spans": [ + { + "bbox": [ + 357, + 680, + 504, + 691 + ], + "score": 1.0, + "content": "rectly fill input voxels, green boxes indi-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 357, + 690, + 502, + 700 + ], + "spans": [ + { + "bbox": [ + 357, + 690, + 502, + 700 + ], + "score": 1.0, + "content": "cate voxels should be filled but are left", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 357, + 701, + 498, + 711 + ], + "spans": [ + { + "bbox": [ + 357, + 701, + 498, + 711 + ], + "score": 1.0, + "content": "empty, red boxes indicate extra voxels.", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 49.5 + } + ], + "index": 45.75 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 57, + 506, + 157 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 57, + 506, + 157 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 57, + 506, + 157 + ], + "spans": [ + { + "bbox": [ + 109, + 57, + 506, + 157 + ], + "score": 0.967, + "type": "image", + "image_path": "7e583fae1e920eca77cfbcfa30b8e62210c56760f2035b8269f5be44b2ae0c30.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 57, + 506, + 90.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 90.33333333333334, + 506, + 123.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 123.66666666666669, + 506, + 157.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 163, + 501, + 174 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 161, + 502, + 175 + ], + "spans": [ + { + "bbox": [ + 109, + 161, + 502, + 175 + ], + "score": 1.0, + "content": "Figure 10: Samples from LION trained on ShapeNet’s Mug and Bottle classes, and on Turbosquid animals.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 107, + 185, + 505, + 240 + ], + "lines": [ + { + "bbox": [ + 106, + 185, + 506, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 506, + 198 + ], + "score": 1.0, + "content": "Mesh Reconstruction. As explained in Sec. 3.1, we combine LION with mesh reconstruction, to", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 196, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 106, + 196, + 505, + 208 + ], + "score": 1.0, + "content": "directly synthesize practically useful meshes. We show generated meshes in Fig. 2, which look", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "score": 1.0, + "content": "smooth and of high quality. In Fig. 2, we also visually demonstrate how we can vary the local details", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 218, + 507, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 507, + 231 + ], + "score": 1.0, + "content": "of synthesized shapes while preserving the overall shape with our diffuse-denoise technique (Sec. 3.1).", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 228, + 479, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 479, + 242 + ], + "score": 1.0, + "content": "Details about the number of diffusion steps for all diffuse-denoise experiments are in App. E.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 185, + 507, + 242 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 245, + 505, + 289 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "score": 1.0, + "content": "Shape Interpolation. As discussed in Sec. 3.1, LION also enables shape interpolation, potentially", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 256, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 506, + 268 + ], + "score": 1.0, + "content": "useful for shape editing applications. We show this in Fig. 7, combined with mesh reconstruction.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 267, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 506, + 280 + ], + "score": 1.0, + "content": "The generated shapes are clean and semantically plausible along the entire interpolation path. In", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 278, + 460, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 460, + 291 + ], + "score": 1.0, + "content": "App. F.12.1, we also show interpolations from PVD [46] and DPM [47] for comparison.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 245, + 506, + 291 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 304, + 335, + 316 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 336, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 336, + 320 + ], + "score": 1.0, + "content": "5.2 Many-class Unconditional 3D Shape Generation", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 322, + 506, + 442 + ], + "lines": [ + { + "bbox": [ + 105, + 322, + 506, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 334 + ], + "score": 1.0, + "content": "13-Class LION Model. We train a LION model jointly without any class conditioning on 13 different", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 333, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 506, + 345 + ], + "score": 1.0, + "content": "categories (airplane, chair, car, lamp, table, sofa, cabinet, bench, telephone, loudspeaker, display, wa-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 342, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 358 + ], + "score": 1.0, + "content": "tercraft, rifle) from ShapeNet (ShapeNet-vol version). Training a single model without conditioning", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 355, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 506, + 367 + ], + "score": 1.0, + "content": "over such diverse shapes is challenging, as the data distribution is highly complex and multimodal.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 364, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 506, + 379 + ], + "score": 1.0, + "content": "We show LION’s generated samples in Fig. 3, including meshes: LION synthesizes high-quality and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 506, + 390 + ], + "score": 1.0, + "content": "diverse plausible shapes even when trained on such complex data. We report the model’s quantitative", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 387, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 506, + 400 + ], + "score": 1.0, + "content": "generation performance in Tab. 4, and we also trained various strong baseline methods under the same", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 398, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 506, + 411 + ], + "score": 1.0, + "content": "setting for comparison. We find that LION significantly outperforms all baselines by a large margin.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 506, + 422 + ], + "score": 1.0, + "content": "We further observe that the hierarchical VAE architecture of LION becomes crucial: The shape latent", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 420, + 507, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 139, + 433 + ], + "score": 1.0, + "content": "variable", + "type": "text" + }, + { + "bbox": [ + 140, + 421, + 151, + 431 + ], + "score": 0.85, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 420, + 329, + 433 + ], + "score": 1.0, + "content": "captures global shape, while the latent points", + "type": "text" + }, + { + "bbox": [ + 330, + 420, + 342, + 431 + ], + "score": 0.86, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 420, + 507, + 433 + ], + "score": 1.0, + "content": "model details. This can be seen in Fig. 8:", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 431, + 497, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 319, + 444 + ], + "score": 1.0, + "content": "we show samples when fixing the global shape latent", + "type": "text" + }, + { + "bbox": [ + 320, + 433, + 330, + 442 + ], + "score": 0.87, + "content": "\\mathbf { z } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 432, + 399, + 444 + ], + "score": 1.0, + "content": "and only sample", + "type": "text" + }, + { + "bbox": [ + 399, + 431, + 412, + 442 + ], + "score": 0.86, + "content": "\\mathbf { h } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 432, + 497, + 444 + ], + "score": 1.0, + "content": "(details in App. F.3).", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 322, + 507, + 444 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 505, + 568 + ], + "lines": [ + { + "bbox": [ + 106, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "55-Class LION Model. Encouraged by these results, we also trained a LION model again jointly", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 458, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 458, + 505, + 471 + ], + "score": 1.0, + "content": "without any class conditioning on all 55 different categories from ShapeNet. Note that we did on", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "score": 1.0, + "content": "purpose not use class-conditioning in these experiments to create a difficult 3D generation task and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 480, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 506, + 492 + ], + "score": 1.0, + "content": "thereby explore LION’s scalability to highly complex and multimodal datasets. We show generated", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 491, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 506, + 503 + ], + "score": 1.0, + "content": "point cloud samples in Fig. 9 (we did not train an SAP model on the 55 classes data): LION", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 502, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 514 + ], + "score": 1.0, + "content": "synthesizes high-quality and diverse shapes. It can even generate samples from the cap class, which", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 526 + ], + "score": 1.0, + "content": "contributes with only 39 training data samples, indicating that LION has an excellent mode coverage", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "score": 1.0, + "content": "that even includes the very rare classes. To the best of our knowledge no previous 3D shape generative", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "models have demonstrated satisfactory generation performance for such diverse and multimodal 3D", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 546, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 506, + 558 + ], + "score": 1.0, + "content": "data without relying on conditioning information (details in App. F.4). In conclusion, we observe that", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 556, + 453, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 453, + 569 + ], + "score": 1.0, + "content": "LION out-of-the-box easily scales to highly complex multi-category shape generation.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 448, + 506, + 569 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 577, + 273, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 275, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 275, + 591 + ], + "score": 1.0, + "content": "5.3 Training LION on Small Datasets", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 504, + 617 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "score": 1.0, + "content": "Next, we explore whether LION can also be trained successfully on very small datasets. To this", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "end, we train models on the Mug and Bottle ShapeNet classes. The number of training samples", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "is 149 and 340, respectively, which is much smaller than the common classes like chair, car and", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 292, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 506, + 307 + ], + "score": 1.0, + "content": "airplane. Furthermore, we also train LION on 553 animal assets from the TurboSquid data repository.", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 302, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 317 + ], + "score": 1.0, + "content": "Generated shapes from the three models are shown in Fig. 10. LION is able to generate correct mugs", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "and bottles as well as diverse and high-quality animal shapes. We conclude that LION also performs", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 325, + 464, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 464, + 339 + ], + "score": 1.0, + "content": "well even when training in the challenging low-data setting (details in Apps. F.5 and F.6).", + "type": "text", + "cross_page": true + } + ], + "index": 17 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 594, + 505, + 618 + ] + }, + { + "type": "image", + "bbox": [ + 115, + 625, + 352, + 725 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 625, + 352, + 725 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 115, + 625, + 352, + 725 + ], + "spans": [ + { + "bbox": [ + 115, + 625, + 352, + 725 + ], + "score": 0.964, + "type": "image", + "image_path": "2ade93952b2c1687a4aed247344a7e707a70e0829b0b8547f8b519ee299a389b.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 115, + 625, + 352, + 639.2857142857143 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 115, + 639.2857142857143, + 352, + 653.5714285714287 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 115, + 653.5714285714287, + 352, + 667.857142857143 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 115, + 667.857142857143, + 352, + 682.1428571428573 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 115, + 682.1428571428573, + 352, + 696.4285714285717 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 115, + 696.4285714285717, + 352, + 710.714285714286 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 115, + 710.714285714286, + 352, + 725.0000000000003 + ], + "spans": [], + "index": 45 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 357, + 630, + 502, + 711 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 357, + 630, + 502, + 641 + ], + "spans": [ + { + "bbox": [ + 357, + 630, + 502, + 641 + ], + "score": 1.0, + "content": "Figure 11: Voxel-guided synthesis. We", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 357, + 641, + 502, + 651 + ], + "spans": [ + { + "bbox": [ + 357, + 641, + 502, + 651 + ], + "score": 1.0, + "content": "show different methods with 0 and 50", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 357, + 651, + 503, + 661 + ], + "spans": [ + { + "bbox": [ + 357, + 651, + 503, + 661 + ], + "score": 1.0, + "content": "steps of diffuse-denoise. Voxelizations", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 357, + 660, + 504, + 672 + ], + "spans": [ + { + "bbox": [ + 357, + 660, + 504, + 672 + ], + "score": 1.0, + "content": "of generated points are also shown: Yel-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 357, + 671, + 504, + 681 + ], + "spans": [ + { + "bbox": [ + 357, + 671, + 504, + 681 + ], + "score": 1.0, + "content": "low boxes indicate generated points cor-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 357, + 680, + 504, + 691 + ], + "spans": [ + { + "bbox": [ + 357, + 680, + 504, + 691 + ], + "score": 1.0, + "content": "rectly fill input voxels, green boxes indi-", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 357, + 690, + 502, + 700 + ], + "spans": [ + { + "bbox": [ + 357, + 690, + 502, + 700 + ], + "score": 1.0, + "content": "cate voxels should be filled but are left", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 357, + 701, + 498, + 711 + ], + "spans": [ + { + "bbox": [ + 357, + 701, + 498, + 711 + ], + "score": 1.0, + "content": "empty, red boxes indicate extra voxels.", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 49.5 + } + ], + "index": 45.75 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 372, + 63, + 497, + 214 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 372, + 63, + 497, + 214 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 372, + 63, + 497, + 214 + ], + "spans": [ + { + "bbox": [ + 372, + 63, + 497, + 214 + ], + "score": 0.959, + "type": "image", + "image_path": "a5ec7783f89ce350c646589554680fe14dc18d0cd5cd5692ac372dc06bdb76d5.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 372, + 63, + 497, + 214 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 368, + 216, + 503, + 267 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 367, + 214, + 504, + 228 + ], + "spans": [ + { + "bbox": [ + 367, + 214, + 504, + 228 + ], + "score": 1.0, + "content": "Figure 13: Voxel-guided genera-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 368, + 226, + 502, + 236 + ], + "spans": [ + { + "bbox": [ + 368, + 226, + 502, + 236 + ], + "score": 1.0, + "content": "tion. Quality metrics for output points", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 367, + 236, + 502, + 246 + ], + "spans": [ + { + "bbox": [ + 367, + 236, + 502, + 246 + ], + "score": 1.0, + "content": "(lower is better) and voxel IOU with", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 367, + 246, + 504, + 257 + ], + "spans": [ + { + "bbox": [ + 367, + 246, + 493, + 257 + ], + "score": 1.0, + "content": "respect to input (higher is better).", + "type": "text" + }, + { + "bbox": [ + 493, + 248, + 500, + 255 + ], + "score": 0.54, + "content": "x -", + "type": "inline_equation" + }, + { + "bbox": [ + 500, + 246, + 504, + 257 + ], + "score": 1.0, + "content": "-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 367, + 256, + 491, + 267 + ], + "spans": [ + { + "bbox": [ + 367, + 256, + 491, + 267 + ], + "score": 1.0, + "content": "axes denote diffuse-denoise steps.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8 + } + ], + "index": 4.5 + }, + { + "type": "image", + "bbox": [ + 110, + 61, + 356, + 224 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 61, + 356, + 224 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 61, + 356, + 224 + ], + "spans": [ + { + "bbox": [ + 110, + 61, + 356, + 224 + ], + "score": 0.968, + "type": "image", + "image_path": "78f765294bdff9804bb3ae06cde9832dbc910c39d07cd4c4c2e5c738c6154ef6.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 110, + 61, + 356, + 115.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 115.33333333333334, + 356, + 169.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 110, + 169.66666666666669, + 356, + 224.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 226, + 361, + 266 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 225, + 362, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 362, + 237 + ], + "score": 1.0, + "content": "Figure 12: Reconstruction metrics with respect to clean inputs for air-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 236, + 361, + 246 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 361, + 246 + ], + "score": 1.0, + "content": "plane category (lower is better) when guiding synthesis with voxelized", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 245, + 362, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 362, + 257 + ], + "score": 1.0, + "content": "or noisy inputs (using uniform, outlier, and normal noise, see App. F.7).", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 255, + 277, + 268 + ], + "spans": [ + { + "bbox": [ + 107, + 258, + 113, + 265 + ], + "score": 0.78, + "content": "_ x", + "type": "inline_equation" + }, + { + "bbox": [ + 113, + 255, + 277, + 268 + ], + "score": 1.0, + "content": "-axes denote number of diffuse-denoise steps.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8.0 + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 107, + 282, + 506, + 338 + ], + "lines": [ + { + "bbox": [ + 106, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "is 149 and 340, respectively, which is much smaller than the common classes like chair, car and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 292, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 506, + 307 + ], + "score": 1.0, + "content": "airplane. Furthermore, we also train LION on 553 animal assets from the TurboSquid data repository.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 302, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 317 + ], + "score": 1.0, + "content": "Generated shapes from the three models are shown in Fig. 10. LION is able to generate correct mugs", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "and bottles as well as diverse and high-quality animal shapes. We conclude that LION also performs", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 325, + 464, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 464, + 339 + ], + "score": 1.0, + "content": "well even when training in the challenging low-data setting (details in Apps. F.5 and F.6).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 347, + 433, + 359 + ], + "lines": [ + { + "bbox": [ + 105, + 346, + 434, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 434, + 361 + ], + "score": 1.0, + "content": "5.4 Voxel-guided Shape Synthesis and Denoising with Fine-tuned Encoders", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 365, + 506, + 476 + ], + "lines": [ + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "Next, we test our strategy for multimodal voxel-guided shape synthesis (see Sec. 3.1) using the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 376, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 506, + 389 + ], + "score": 1.0, + "content": "airplane class LION model (experiment details in App. E, more experiments in App. F.7). We first", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "voxelize our training set and fine-tune our encoder networks to produce the correct encodings to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 398, + 507, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 507, + 411 + ], + "score": 1.0, + "content": "decode back the original shapes. When processing voxelized shapes with our point-cloud networks,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 410, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 506, + 421 + ], + "score": 1.0, + "content": "we sample points on the surface of the voxels. As discussed, we can use different numbers of diffuse-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "denoise steps in latent space to generate various plausible shapes and correct for poor encodings.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "score": 1.0, + "content": "Instead of voxelizations, we can also consider different noisy inputs (we use normal, uniform, and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 442, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 506, + 455 + ], + "score": 1.0, + "content": "outlier noise, see App. F.7) and achieve multimodal denoising with the same approach. The same", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 454, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 465 + ], + "score": 1.0, + "content": "tasks can be attempted with the important DDM-based baselines PVD and DPM, by directly—not in", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "a latent space—diffusing and denoising voxelized (converted to point clouds) or noisy point clouds.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 480, + 506, + 612 + ], + "lines": [ + { + "bbox": [ + 105, + 480, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 506, + 492 + ], + "score": 1.0, + "content": "Fig. 12 shows the reconstruction performance of LION, DPM and PVD for different numbers of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 506, + 504 + ], + "score": 1.0, + "content": "diffuse-denoise steps (we voxelized or noised the validation set to measure this). We see that for almost", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "score": 1.0, + "content": "all inputs—voxelized or different noises—LION performs best. PVD and DPM perform acceptably", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 505, + 525 + ], + "score": 1.0, + "content": "for normal and uniform noise, which is similar to the noise injected during training of their DDMs, but", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "score": 1.0, + "content": "perform very poorly for outlier noise or voxel inputs, which is the most relevant case to us, because", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 535, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 506, + 547 + ], + "score": 1.0, + "content": "voxels can be easily placed by users. It is LION’s unique framework with additional fine-tuned", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "encoders in its VAE and only latent DDMs that makes this possible. Performing more diffuse-denoise", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 557, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 557, + 506, + 570 + ], + "score": 1.0, + "content": "steps means that more independent, novel shapes are generated. These will be cleaner and of higher", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "score": 1.0, + "content": "quality, but also correspond less to the noisy or voxel inputs used for guidance. In Fig. 13, we show this", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 578, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 506, + 592 + ], + "score": 1.0, + "content": "trade-off for the voxel-guidance experiment (other experiments in App. F.7), where (top) we measured", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 588, + 505, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 505, + 602 + ], + "score": 1.0, + "content": "the outputs’ synthesis quality by calculating 1-NNA with respect to the validation set, and (bottom) the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 601, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 506, + 613 + ], + "score": 1.0, + "content": "average intersection over union (IOU) between the input voxels and the voxelized outputs. We gener-", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 34.5 + }, + { + "type": "image", + "bbox": [ + 112, + 622, + 498, + 699 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 622, + 498, + 699 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 112, + 622, + 498, + 699 + ], + "spans": [ + { + "bbox": [ + 112, + 622, + 498, + 699 + ], + "score": 0.845, + "type": "image", + "image_path": "e439f9c4b4cd464eb59193841e41c04c3f2eabc99776306355fe2b44efd72e31.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 112, + 622, + 498, + 647.6666666666666 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 112, + 647.6666666666666, + 498, + 673.3333333333333 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 112, + 673.3333333333333, + 498, + 698.9999999999999 + ], + "spans": [], + "index": 43 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 704, + 505, + 725 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 703, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 106, + 703, + 505, + 716 + ], + "score": 1.0, + "content": "Figure 14: We apply Text2Mesh [49] on meshes generated by LION. In Text2Mesh, textures are generated and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 714, + 504, + 725 + ], + "spans": [ + { + "bbox": [ + 106, + 714, + 504, + 725 + ], + "score": 1.0, + "content": "meshes refined such that rendered images of the 3D objects are aligned with user-provided text prompts [105].", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "index": 43.25 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 372, + 63, + 497, + 214 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 372, + 63, + 497, + 214 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 372, + 63, + 497, + 214 + ], + "spans": [ + { + "bbox": [ + 372, + 63, + 497, + 214 + ], + "score": 0.959, + "type": "image", + "image_path": "a5ec7783f89ce350c646589554680fe14dc18d0cd5cd5692ac372dc06bdb76d5.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 372, + 63, + 497, + 214 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 368, + 216, + 503, + 267 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 367, + 214, + 504, + 228 + ], + "spans": [ + { + "bbox": [ + 367, + 214, + 504, + 228 + ], + "score": 1.0, + "content": "Figure 13: Voxel-guided genera-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 368, + 226, + 502, + 236 + ], + "spans": [ + { + "bbox": [ + 368, + 226, + 502, + 236 + ], + "score": 1.0, + "content": "tion. Quality metrics for output points", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 367, + 236, + 502, + 246 + ], + "spans": [ + { + "bbox": [ + 367, + 236, + 502, + 246 + ], + "score": 1.0, + "content": "(lower is better) and voxel IOU with", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 367, + 246, + 504, + 257 + ], + "spans": [ + { + "bbox": [ + 367, + 246, + 493, + 257 + ], + "score": 1.0, + "content": "respect to input (higher is better).", + "type": "text" + }, + { + "bbox": [ + 493, + 248, + 500, + 255 + ], + "score": 0.54, + "content": "x -", + "type": "inline_equation" + }, + { + "bbox": [ + 500, + 246, + 504, + 257 + ], + "score": 1.0, + "content": "-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 367, + 256, + 491, + 267 + ], + "spans": [ + { + "bbox": [ + 367, + 256, + 491, + 267 + ], + "score": 1.0, + "content": "axes denote diffuse-denoise steps.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8 + } + ], + "index": 4.5 + }, + { + "type": "image", + "bbox": [ + 110, + 61, + 356, + 224 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 61, + 356, + 224 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 61, + 356, + 224 + ], + "spans": [ + { + "bbox": [ + 110, + 61, + 356, + 224 + ], + "score": 0.968, + "type": "image", + "image_path": "78f765294bdff9804bb3ae06cde9832dbc910c39d07cd4c4c2e5c738c6154ef6.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 110, + 61, + 356, + 115.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 115.33333333333334, + 356, + 169.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 110, + 169.66666666666669, + 356, + 224.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 226, + 361, + 266 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 225, + 362, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 362, + 237 + ], + "score": 1.0, + "content": "Figure 12: Reconstruction metrics with respect to clean inputs for air-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 236, + 361, + 246 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 361, + 246 + ], + "score": 1.0, + "content": "plane category (lower is better) when guiding synthesis with voxelized", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 245, + 362, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 362, + 257 + ], + "score": 1.0, + "content": "or noisy inputs (using uniform, outlier, and normal noise, see App. F.7).", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 255, + 277, + 268 + ], + "spans": [ + { + "bbox": [ + 107, + 258, + 113, + 265 + ], + "score": 0.78, + "content": "_ x", + "type": "inline_equation" + }, + { + "bbox": [ + 113, + 255, + 277, + 268 + ], + "score": 1.0, + "content": "-axes denote number of diffuse-denoise steps.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8.0 + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 107, + 282, + 506, + 338 + ], + "lines": [], + "index": 15, + "bbox_fs": [ + 105, + 282, + 506, + 339 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 347, + 433, + 359 + ], + "lines": [ + { + "bbox": [ + 105, + 346, + 434, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 434, + 361 + ], + "score": 1.0, + "content": "5.4 Voxel-guided Shape Synthesis and Denoising with Fine-tuned Encoders", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 365, + 506, + 476 + ], + "lines": [ + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "Next, we test our strategy for multimodal voxel-guided shape synthesis (see Sec. 3.1) using the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 376, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 506, + 389 + ], + "score": 1.0, + "content": "airplane class LION model (experiment details in App. E, more experiments in App. F.7). We first", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "voxelize our training set and fine-tune our encoder networks to produce the correct encodings to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 398, + 507, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 507, + 411 + ], + "score": 1.0, + "content": "decode back the original shapes. When processing voxelized shapes with our point-cloud networks,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 410, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 506, + 421 + ], + "score": 1.0, + "content": "we sample points on the surface of the voxels. As discussed, we can use different numbers of diffuse-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "denoise steps in latent space to generate various plausible shapes and correct for poor encodings.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "score": 1.0, + "content": "Instead of voxelizations, we can also consider different noisy inputs (we use normal, uniform, and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 442, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 506, + 455 + ], + "score": 1.0, + "content": "outlier noise, see App. F.7) and achieve multimodal denoising with the same approach. The same", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 454, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 465 + ], + "score": 1.0, + "content": "tasks can be attempted with the important DDM-based baselines PVD and DPM, by directly—not in", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "a latent space—diffusing and denoising voxelized (converted to point clouds) or noisy point clouds.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 366, + 507, + 477 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 480, + 506, + 612 + ], + "lines": [ + { + "bbox": [ + 105, + 480, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 506, + 492 + ], + "score": 1.0, + "content": "Fig. 12 shows the reconstruction performance of LION, DPM and PVD for different numbers of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 506, + 504 + ], + "score": 1.0, + "content": "diffuse-denoise steps (we voxelized or noised the validation set to measure this). We see that for almost", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "score": 1.0, + "content": "all inputs—voxelized or different noises—LION performs best. PVD and DPM perform acceptably", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 505, + 525 + ], + "score": 1.0, + "content": "for normal and uniform noise, which is similar to the noise injected during training of their DDMs, but", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 506, + 537 + ], + "score": 1.0, + "content": "perform very poorly for outlier noise or voxel inputs, which is the most relevant case to us, because", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 535, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 506, + 547 + ], + "score": 1.0, + "content": "voxels can be easily placed by users. It is LION’s unique framework with additional fine-tuned", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "encoders in its VAE and only latent DDMs that makes this possible. Performing more diffuse-denoise", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 557, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 557, + 506, + 570 + ], + "score": 1.0, + "content": "steps means that more independent, novel shapes are generated. These will be cleaner and of higher", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "score": 1.0, + "content": "quality, but also correspond less to the noisy or voxel inputs used for guidance. In Fig. 13, we show this", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 578, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 506, + 592 + ], + "score": 1.0, + "content": "trade-off for the voxel-guidance experiment (other experiments in App. F.7), where (top) we measured", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 588, + 505, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 505, + 602 + ], + "score": 1.0, + "content": "the outputs’ synthesis quality by calculating 1-NNA with respect to the validation set, and (bottom) the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 601, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 506, + 613 + ], + "score": 1.0, + "content": "average intersection over union (IOU) between the input voxels and the voxelized outputs. We gener-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 120, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 106, + 120, + 506, + 132 + ], + "score": 1.0, + "content": "ally see a trade-off: More diffuse-denoise steps result in lower 1-NNA (better quality), but also lower", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 130, + 506, + 143 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 506, + 143 + ], + "score": 1.0, + "content": "IOU. LION strikes the best balance by a large gap: Its additional encoder network directly generates", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 141, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 506, + 154 + ], + "score": 1.0, + "content": "plausible latent encodings from the perturbed inputs that are both high quality and also correspond well", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 153, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 506, + 165 + ], + "score": 1.0, + "content": "to the input. This trade-off is visualized in Fig. 11 for LION, DPM, and PVD, where we show gener-", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 163, + 506, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 506, + 176 + ], + "score": 1.0, + "content": "ated point clouds and voxelizations (note that performing no diffuse-denoise at all for PVD and DPM", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 174, + 506, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 186 + ], + "score": 1.0, + "content": "corresponds to simply keeping the input, as these models’ DDMs operate directly on point clouds).", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 185, + 506, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 506, + 198 + ], + "score": 1.0, + "content": "We see that running 50 diffuse-denoise steps to generate diverse outputs for DPM and especially PVD", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 196, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 506, + 208 + ], + "score": 1.0, + "content": "results in a significant violation of the input voxelization. In contrast, LION generates realistic outputs", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "that also obey the driving voxels. Overall, LION wins out both in this task and also in unconditional", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 218, + 507, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 507, + 230 + ], + "score": 1.0, + "content": "generation with large gaps over these previous DDM-based point cloud generative models. We con-", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 228, + 507, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 507, + 241 + ], + "score": 1.0, + "content": "clude that LION does not only offer state-of-the-art 3D shape generation quality, but is also very ver-", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 240, + 507, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 507, + 252 + ], + "score": 1.0, + "content": "satile. Note that guided synthesis can also be combined with mesh reconstruction, as shown in Fig. 4.", + "type": "text", + "cross_page": true + } + ], + "index": 18 + } + ], + "index": 34.5, + "bbox_fs": [ + 104, + 480, + 506, + 613 + ] + }, + { + "type": "image", + "bbox": [ + 112, + 622, + 498, + 699 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 622, + 498, + 699 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 112, + 622, + 498, + 699 + ], + "spans": [ + { + "bbox": [ + 112, + 622, + 498, + 699 + ], + "score": 0.845, + "type": "image", + "image_path": "e439f9c4b4cd464eb59193841e41c04c3f2eabc99776306355fe2b44efd72e31.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 112, + 622, + 498, + 647.6666666666666 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 112, + 647.6666666666666, + 498, + 673.3333333333333 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 112, + 673.3333333333333, + 498, + 698.9999999999999 + ], + "spans": [], + "index": 43 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 704, + 505, + 725 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 703, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 106, + 703, + 505, + 716 + ], + "score": 1.0, + "content": "Figure 14: We apply Text2Mesh [49] on meshes generated by LION. In Text2Mesh, textures are generated and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 714, + 504, + 725 + ], + "spans": [ + { + "bbox": [ + 106, + 714, + 504, + 725 + ], + "score": 1.0, + "content": "meshes refined such that rendered images of the 3D objects are aligned with user-provided text prompts [105].", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "index": 43.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 56, + 345, + 113 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 56, + 345, + 113 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 56, + 345, + 113 + ], + "spans": [ + { + "bbox": [ + 119, + 56, + 345, + 113 + ], + "score": 0.73, + "type": "image", + "image_path": "81667f186ae68d9e39acfe123525b7508d90e9671430dcd91425046a6551bf91.jpg" + } + ] + } + ], + "index": 3.0, + "virtual_lines": [ + { + "bbox": [ + 119, + 56, + 345, + 70.25 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 70.25, + 345, + 84.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 119, + 84.5, + 345, + 98.75 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 119, + 98.75, + 345, + 113.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 357, + 65, + 503, + 96 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 357, + 64, + 505, + 77 + ], + "spans": [ + { + "bbox": [ + 357, + 64, + 505, + 77 + ], + "score": 1.0, + "content": "Figure 16: Text-driven shape genera-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 357, + 75, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 357, + 75, + 505, + 86 + ], + "score": 1.0, + "content": "tion of chairs and cars with LION. Bot-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 357, + 86, + 503, + 97 + ], + "spans": [ + { + "bbox": [ + 357, + 86, + 503, + 97 + ], + "score": 1.0, + "content": "tom row is the text prompt used as input.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 106, + 119, + 506, + 251 + ], + "lines": [ + { + "bbox": [ + 106, + 120, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 106, + 120, + 506, + 132 + ], + "score": 1.0, + "content": "ally see a trade-off: More diffuse-denoise steps result in lower 1-NNA (better quality), but also lower", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 130, + 506, + 143 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 506, + 143 + ], + "score": 1.0, + "content": "IOU. LION strikes the best balance by a large gap: Its additional encoder network directly generates", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 141, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 506, + 154 + ], + "score": 1.0, + "content": "plausible latent encodings from the perturbed inputs that are both high quality and also correspond well", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 153, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 506, + 165 + ], + "score": 1.0, + "content": "to the input. This trade-off is visualized in Fig. 11 for LION, DPM, and PVD, where we show gener-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 163, + 506, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 506, + 176 + ], + "score": 1.0, + "content": "ated point clouds and voxelizations (note that performing no diffuse-denoise at all for PVD and DPM", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 174, + 506, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 186 + ], + "score": 1.0, + "content": "corresponds to simply keeping the input, as these models’ DDMs operate directly on point clouds).", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 185, + 506, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 506, + 198 + ], + "score": 1.0, + "content": "We see that running 50 diffuse-denoise steps to generate diverse outputs for DPM and especially PVD", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 196, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 506, + 208 + ], + "score": 1.0, + "content": "results in a significant violation of the input voxelization. In contrast, LION generates realistic outputs", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "that also obey the driving voxels. Overall, LION wins out both in this task and also in unconditional", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 218, + 507, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 507, + 230 + ], + "score": 1.0, + "content": "generation with large gaps over these previous DDM-based point cloud generative models. We con-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 228, + 507, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 507, + 241 + ], + "score": 1.0, + "content": "clude that LION does not only offer state-of-the-art 3D shape generation quality, but is also very ver-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 240, + 507, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 507, + 252 + ], + "score": 1.0, + "content": "satile. Note that guided synthesis can also be combined with mesh reconstruction, as shown in Fig. 4.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 12.5 + }, + { + "type": "title", + "bbox": [ + 107, + 258, + 196, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 197, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 197, + 273 + ], + "score": 1.0, + "content": "5.5 Sampling Time", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 273, + 404, + 328 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 404, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 404, + 285 + ], + "score": 1.0, + "content": "While our main experiments use 1,000-step DDPM-based synthesis, which", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 283, + 404, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 128, + 296 + ], + "score": 1.0, + "content": "takes", + "type": "text" + }, + { + "bbox": [ + 129, + 284, + 164, + 294 + ], + "score": 0.78, + "content": "\\approx 2 7 . 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 283, + 404, + 296 + ], + "score": 1.0, + "content": "seconds, we can significantly accelerate generation without", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 294, + 404, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 404, + 307 + ], + "score": 1.0, + "content": "significant loss in quality. Using DDIM-based sampling [106], we can", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 306, + 404, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 404, + 317 + ], + "score": 1.0, + "content": "generate high quality shapes in under one second (Fig. 15), which would", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 317, + 381, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 381, + 329 + ], + "score": 1.0, + "content": "enable real-time interactive applications. More analyses in App. F.9.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "image", + "bbox": [ + 414, + 252, + 501, + 331 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 414, + 252, + 501, + 331 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 414, + 252, + 501, + 331 + ], + "spans": [ + { + "bbox": [ + 414, + 252, + 501, + 331 + ], + "score": 0.958, + "type": "image", + "image_path": "625795a9b501d90ded76d171008f1ad0b2a315ea890e5a177dc72ef8863de109.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 414, + 252, + 501, + 291.5 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 414, + 291.5, + 501, + 331.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 411, + 336, + 504, + 367 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 410, + 334, + 507, + 349 + ], + "spans": [ + { + "bbox": [ + 410, + 334, + 465, + 348 + ], + "score": 1.0, + "content": "Figure 15:", + "type": "text" + }, + { + "bbox": [ + 469, + 334, + 507, + 349 + ], + "score": 1.0, + "content": "25-step", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 410, + 344, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 410, + 344, + 506, + 358 + ], + "score": 1.0, + "content": "DDIM [106] samples", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 410, + 356, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 410, + 356, + 505, + 368 + ], + "score": 1.0, + "content": "(0.89 seconds per shape).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + } + ], + "index": 26.75 + }, + { + "type": "title", + "bbox": [ + 106, + 335, + 343, + 347 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 343, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 343, + 349 + ], + "score": 1.0, + "content": "5.6 Overview of Additional Experiments in Appendix", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 108, + 350, + 403, + 371 + ], + "lines": [ + { + "bbox": [ + 105, + 349, + 405, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 405, + 363 + ], + "score": 1.0, + "content": "(i) In App. F.1, we perform various ablation studies. The experiments quan-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 360, + 405, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 405, + 373 + ], + "score": 1.0, + "content": "titatively validate LION’s architecture choices and the advantage of our", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.0 + }, + { + "type": "text", + "bbox": [ + 106, + 372, + 505, + 470 + ], + "lines": [ + { + "bbox": [ + 106, + 371, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 506, + 383 + ], + "score": 1.0, + "content": "hierarchical VAE setup with conditional latent DDMs. (ii) In App. F.8, we measure LION’s autoen-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 381, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 506, + 396 + ], + "score": 1.0, + "content": "coding performance. (iii) To demonstrate the value of directly outputting meshes, in App. F.10 we use", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 392, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 104, + 392, + 506, + 407 + ], + "score": 1.0, + "content": "Text2Mesh [49] to generate textures based on text prompts for synthesized LION samples (Fig. 14).", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 402, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 104, + 402, + 506, + 417 + ], + "score": 1.0, + "content": "This would not be possible, if we only generated point clouds. (iv) To qualitatively show that LION", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "can be adapted easily to other relevant tasks, in App. F.11 we condition LION on CLIP embeddings", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 425, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 104, + 425, + 506, + 439 + ], + "score": 1.0, + "content": "of the shapes’ rendered images, following CLIP-Forge [34] (Fig. 16). This enables text-driven 3D", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "shape generation and single view 3D reconstruction (Fig. 17). (v) We also show many more samples", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 447, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 460 + ], + "score": 1.0, + "content": "(Apps. F.2-F.6) and shape interpolations (App. F.12) from our models, more examples of voxel-guided", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "score": 1.0, + "content": "and noise-guided synthesis (App. F.7), and we further analyze our 13-class LION model (App. F.3.2).", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37 + }, + { + "type": "title", + "bbox": [ + 107, + 477, + 188, + 490 + ], + "lines": [ + { + "bbox": [ + 104, + 474, + 190, + 493 + ], + "spans": [ + { + "bbox": [ + 104, + 474, + 190, + 493 + ], + "score": 1.0, + "content": "6 Conclusions", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 506, + 595 + ], + "lines": [ + { + "bbox": [ + 106, + 496, + 507, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 507, + 509 + ], + "score": 1.0, + "content": "We introduced LION, a novel generative model of 3D shapes. LION uses a VAE framework with hi-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "erarchical DDMs in latent space and can be combined with SAP for mesh generation. LION achieves", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 518, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 506, + 530 + ], + "score": 1.0, + "content": "state-of-the-art shape generation performance and enables applications such as voxel-conditioned syn-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "score": 1.0, + "content": "thesis, multimodal shape denoising, and shape interpolation. LION is currently trained on 3D point", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 539, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 506, + 553 + ], + "score": 1.0, + "content": "clouds only and can not directly generate textured shapes. A promising extension would be to include", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 550, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 564 + ], + "score": 1.0, + "content": "image-based training by incorporating neural or differentiable rendering [17, 107–111] and to also syn-", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "thesize textures [16, 112–114]. Furthermore, LION currently focuses on single object generation only.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "It would be interesting to extend it to full 3D scene synthesis. Moreover, synthesis could be further", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 583, + 503, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 503, + 595 + ], + "score": 1.0, + "content": "accelerated by building on works on accelerated sampling from DDMs [61, 62, 67, 106, 115–121].", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 47 + }, + { + "type": "text", + "bbox": [ + 106, + 599, + 506, + 666 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 506, + 611 + ], + "score": 1.0, + "content": "Broader Impact. We believe that LION can potentially improve 3D content creation and assist", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "the workflow of digital artists. We designed LION with such applications in mind and hope that it", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "score": 1.0, + "content": "can grow into a practical tool enhancing artists’ creativity. Although we do not see any immediate", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "score": 1.0, + "content": "negative use-cases for LION, it is important that practitioners apply an abundance of caution to", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 642, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 657 + ], + "score": 1.0, + "content": "mitigate impacts given generative modeling more generally can also be used for malicious purposes,", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 653, + 501, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 501, + 667 + ], + "score": 1.0, + "content": "discussed for instance in Vaccari and Chadwick [122], Nguyen et al. [123], Mirsky and Lee [124].", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 54.5 + }, + { + "type": "image", + "bbox": [ + 110, + 685, + 308, + 717 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 685, + 308, + 717 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 110, + 685, + 308, + 717 + ], + "spans": [ + { + "bbox": [ + 110, + 685, + 308, + 717 + ], + "score": 0.807, + "type": "image", + "image_path": "2a8dd3f6899aade67630f04ae5ec6f0108def53fe8be061056ed6a9d207f21b7.jpg" + } + ] + } + ], + "index": 59.5, + "virtual_lines": [ + { + "bbox": [ + 110, + 685, + 308, + 701.0 + ], + "spans": [], + "index": 58 + }, + { + "bbox": [ + 110, + 701.0, + 308, + 717.0 + ], + "spans": [], + "index": 61 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 313, + 683, + 506, + 714 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 314, + 682, + 507, + 694 + ], + "spans": [ + { + "bbox": [ + 314, + 682, + 507, + 694 + ], + "score": 1.0, + "content": "Figure 17: Single view 3D reconstructions of a car", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 314, + 693, + 506, + 704 + ], + "spans": [ + { + "bbox": [ + 314, + 693, + 506, + 704 + ], + "score": 1.0, + "content": "from an RGB image. LION can generate multiple", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 313, + 703, + 507, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 703, + 507, + 715 + ], + "score": 1.0, + "content": "plausible outputs using our diffuse-denoise technique.", + "type": "text" + } + ], + "index": 62 + } + ], + "index": 60 + } + ], + "index": 59.75 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 56, + 345, + 113 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 56, + 345, + 113 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 56, + 345, + 113 + ], + "spans": [ + { + "bbox": [ + 119, + 56, + 345, + 113 + ], + "score": 0.73, + "type": "image", + "image_path": "81667f186ae68d9e39acfe123525b7508d90e9671430dcd91425046a6551bf91.jpg" + } + ] + } + ], + "index": 3.0, + "virtual_lines": [ + { + "bbox": [ + 119, + 56, + 345, + 70.25 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 70.25, + 345, + 84.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 119, + 84.5, + 345, + 98.75 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 119, + 98.75, + 345, + 113.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 357, + 65, + 503, + 96 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 357, + 64, + 505, + 77 + ], + "spans": [ + { + "bbox": [ + 357, + 64, + 505, + 77 + ], + "score": 1.0, + "content": "Figure 16: Text-driven shape genera-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 357, + 75, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 357, + 75, + 505, + 86 + ], + "score": 1.0, + "content": "tion of chairs and cars with LION. Bot-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 357, + 86, + 503, + 97 + ], + "spans": [ + { + "bbox": [ + 357, + 86, + 503, + 97 + ], + "score": 1.0, + "content": "tom row is the text prompt used as input.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 106, + 119, + 506, + 251 + ], + "lines": [], + "index": 12.5, + "bbox_fs": [ + 105, + 120, + 507, + 252 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 258, + 196, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 197, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 197, + 273 + ], + "score": 1.0, + "content": "5.5 Sampling Time", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 273, + 404, + 328 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 404, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 404, + 285 + ], + "score": 1.0, + "content": "While our main experiments use 1,000-step DDPM-based synthesis, which", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 283, + 404, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 128, + 296 + ], + "score": 1.0, + "content": "takes", + "type": "text" + }, + { + "bbox": [ + 129, + 284, + 164, + 294 + ], + "score": 0.78, + "content": "\\approx 2 7 . 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 283, + 404, + 296 + ], + "score": 1.0, + "content": "seconds, we can significantly accelerate generation without", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 294, + 404, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 404, + 307 + ], + "score": 1.0, + "content": "significant loss in quality. Using DDIM-based sampling [106], we can", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 306, + 404, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 404, + 317 + ], + "score": 1.0, + "content": "generate high quality shapes in under one second (Fig. 15), which would", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 317, + 381, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 381, + 329 + ], + "score": 1.0, + "content": "enable real-time interactive applications. More analyses in App. F.9.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 273, + 404, + 329 + ] + }, + { + "type": "image", + "bbox": [ + 414, + 252, + 501, + 331 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 414, + 252, + 501, + 331 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 414, + 252, + 501, + 331 + ], + "spans": [ + { + "bbox": [ + 414, + 252, + 501, + 331 + ], + "score": 0.958, + "type": "image", + "image_path": "625795a9b501d90ded76d171008f1ad0b2a315ea890e5a177dc72ef8863de109.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 414, + 252, + 501, + 291.5 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 414, + 291.5, + 501, + 331.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 411, + 336, + 504, + 367 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 410, + 334, + 507, + 349 + ], + "spans": [ + { + "bbox": [ + 410, + 334, + 465, + 348 + ], + "score": 1.0, + "content": "Figure 15:", + "type": "text" + }, + { + "bbox": [ + 469, + 334, + 507, + 349 + ], + "score": 1.0, + "content": "25-step", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 410, + 344, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 410, + 344, + 506, + 358 + ], + "score": 1.0, + "content": "DDIM [106] samples", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 410, + 356, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 410, + 356, + 505, + 368 + ], + "score": 1.0, + "content": "(0.89 seconds per shape).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + } + ], + "index": 26.75 + }, + { + "type": "title", + "bbox": [ + 106, + 335, + 343, + 347 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 343, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 343, + 349 + ], + "score": 1.0, + "content": "5.6 Overview of Additional Experiments in Appendix", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 108, + 350, + 403, + 371 + ], + "lines": [ + { + "bbox": [ + 105, + 349, + 405, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 405, + 363 + ], + "score": 1.0, + "content": "(i) In App. F.1, we perform various ablation studies. The experiments quan-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 360, + 405, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 405, + 373 + ], + "score": 1.0, + "content": "titatively validate LION’s architecture choices and the advantage of our", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 371, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 506, + 383 + ], + "score": 1.0, + "content": "hierarchical VAE setup with conditional latent DDMs. (ii) In App. F.8, we measure LION’s autoen-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 381, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 506, + 396 + ], + "score": 1.0, + "content": "coding performance. (iii) To demonstrate the value of directly outputting meshes, in App. F.10 we use", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 392, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 104, + 392, + 506, + 407 + ], + "score": 1.0, + "content": "Text2Mesh [49] to generate textures based on text prompts for synthesized LION samples (Fig. 14).", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 402, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 104, + 402, + 506, + 417 + ], + "score": 1.0, + "content": "This would not be possible, if we only generated point clouds. (iv) To qualitatively show that LION", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "can be adapted easily to other relevant tasks, in App. F.11 we condition LION on CLIP embeddings", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 425, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 104, + 425, + 506, + 439 + ], + "score": 1.0, + "content": "of the shapes’ rendered images, following CLIP-Forge [34] (Fig. 16). This enables text-driven 3D", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "shape generation and single view 3D reconstruction (Fig. 17). (v) We also show many more samples", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 447, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 460 + ], + "score": 1.0, + "content": "(Apps. F.2-F.6) and shape interpolations (App. F.12) from our models, more examples of voxel-guided", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "score": 1.0, + "content": "and noise-guided synthesis (App. F.7), and we further analyze our 13-class LION model (App. F.3.2).", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 30.0, + "bbox_fs": [ + 105, + 349, + 405, + 373 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 372, + 505, + 470 + ], + "lines": [], + "index": 37, + "bbox_fs": [ + 104, + 371, + 506, + 471 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 477, + 188, + 490 + ], + "lines": [ + { + "bbox": [ + 104, + 474, + 190, + 493 + ], + "spans": [ + { + "bbox": [ + 104, + 474, + 190, + 493 + ], + "score": 1.0, + "content": "6 Conclusions", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 506, + 595 + ], + "lines": [ + { + "bbox": [ + 106, + 496, + 507, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 507, + 509 + ], + "score": 1.0, + "content": "We introduced LION, a novel generative model of 3D shapes. LION uses a VAE framework with hi-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "erarchical DDMs in latent space and can be combined with SAP for mesh generation. LION achieves", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 518, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 506, + 530 + ], + "score": 1.0, + "content": "state-of-the-art shape generation performance and enables applications such as voxel-conditioned syn-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "score": 1.0, + "content": "thesis, multimodal shape denoising, and shape interpolation. LION is currently trained on 3D point", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 539, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 506, + 553 + ], + "score": 1.0, + "content": "clouds only and can not directly generate textured shapes. A promising extension would be to include", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 550, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 564 + ], + "score": 1.0, + "content": "image-based training by incorporating neural or differentiable rendering [17, 107–111] and to also syn-", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "thesize textures [16, 112–114]. Furthermore, LION currently focuses on single object generation only.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "It would be interesting to extend it to full 3D scene synthesis. Moreover, synthesis could be further", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 583, + 503, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 503, + 595 + ], + "score": 1.0, + "content": "accelerated by building on works on accelerated sampling from DDMs [61, 62, 67, 106, 115–121].", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 47, + "bbox_fs": [ + 105, + 496, + 507, + 595 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 599, + 506, + 666 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 506, + 611 + ], + "score": 1.0, + "content": "Broader Impact. We believe that LION can potentially improve 3D content creation and assist", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "the workflow of digital artists. We designed LION with such applications in mind and hope that it", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "score": 1.0, + "content": "can grow into a practical tool enhancing artists’ creativity. Although we do not see any immediate", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "score": 1.0, + "content": "negative use-cases for LION, it is important that practitioners apply an abundance of caution to", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 642, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 657 + ], + "score": 1.0, + "content": "mitigate impacts given generative modeling more generally can also be used for malicious purposes,", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 105, + 653, + 501, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 501, + 667 + ], + "score": 1.0, + "content": "discussed for instance in Vaccari and Chadwick [122], Nguyen et al. [123], Mirsky and Lee [124].", + "type": "text" + } + ], + "index": 57 + } + ], + "index": 54.5, + "bbox_fs": [ + 105, + 600, + 506, + 667 + ] + }, + { + "type": "image", + "bbox": [ + 110, + 685, + 308, + 717 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 685, + 308, + 717 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 110, + 685, + 308, + 717 + ], + "spans": [ + { + "bbox": [ + 110, + 685, + 308, + 717 + ], + "score": 0.807, + "type": "image", + "image_path": "2a8dd3f6899aade67630f04ae5ec6f0108def53fe8be061056ed6a9d207f21b7.jpg" + } + ] + } + ], + "index": 59.5, + "virtual_lines": [ + { + "bbox": [ + 110, + 685, + 308, + 701.0 + ], + "spans": [], + "index": 58 + }, + { + "bbox": [ + 110, + 701.0, + 308, + 717.0 + ], + "spans": [], + "index": 61 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 313, + 683, + 506, + 714 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 314, + 682, + 507, + 694 + ], + "spans": [ + { + "bbox": [ + 314, + 682, + 507, + 694 + ], + "score": 1.0, + "content": "Figure 17: Single view 3D reconstructions of a car", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 314, + 693, + 506, + 704 + ], + "spans": [ + { + "bbox": [ + 314, + 693, + 506, + 704 + ], + "score": 1.0, + "content": "from an RGB image. LION can generate multiple", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 313, + 703, + 507, + 715 + ], + "spans": [ + { + "bbox": [ + 313, + 703, + 507, + 715 + ], + "score": 1.0, + "content": "plausible outputs using our diffuse-denoise technique.", + "type": "text" + } + ], + "index": 62 + } + ], + "index": 60 + } + ], + "index": 59.75 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 65, + 507, + 729 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 113, + 88, + 505, + 102 + ], + "spans": [ + { + "bbox": [ + 113, + 88, + 505, + 102 + ], + "score": 1.0, + "content": "[1] Jiajun Wu, Chengkai Zhang, Tianfan Xue, Bill Freeman, and Josh Tenenbaum. Learning a probabilistic", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 131, + 100, + 505, + 111 + ], + "spans": [ + { + "bbox": [ + 131, + 100, + 505, + 111 + ], + "score": 1.0, + "content": "latent space of object shapes via 3d generative-adversarial modeling. In Advances in Neural Information", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 129, + 108, + 229, + 122 + ], + "spans": [ + { + "bbox": [ + 129, + 108, + 229, + 122 + ], + "score": 1.0, + "content": "Processing Systems, 2016.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 114, + 127, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 114, + 127, + 506, + 140 + ], + "score": 1.0, + "content": "[2] Panos Achlioptas, Olga Diamanti, Ioannis Mitliagkas, and Leonidas Guibas. Learning representations", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 128, + 136, + 348, + 149 + ], + "spans": [ + { + "bbox": [ + 128, + 136, + 348, + 149 + ], + "score": 1.0, + "content": "and generative models for 3D point clouds. In ICML, 2018.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 113, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 113, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "[3] Chun-Liang Li, Manzil Zaheer, Yang Zhang, Barnabas Poczos, and Ruslan Salakhutdinov. Point cloud", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 128, + 164, + 296, + 176 + ], + "spans": [ + { + "bbox": [ + 128, + 164, + 296, + 176 + ], + "score": 1.0, + "content": "gan. arXiv preprint arXiv:1810.05795, 2018.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 113, + 181, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 113, + 181, + 506, + 194 + ], + "score": 1.0, + "content": "[4] Diego Valsesia, Giulia Fracastoro, and Enrico Magli. Learning localized generative models for 3d point", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 129, + 191, + 507, + 205 + ], + "spans": [ + { + "bbox": [ + 129, + 191, + 507, + 205 + ], + "score": 1.0, + "content": "clouds via graph convolution. In International Conference on Learning Representations (ICLR) 2019,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 130, + 202, + 154, + 214 + ], + "spans": [ + { + "bbox": [ + 130, + 202, + 154, + 214 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 114, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 114, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "[5] Wenlong Huang, Brian Lai, Weijian Xu, and Zhuowen Tu. 3d volumetric modeling with introspective", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 128, + 228, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 128, + 228, + 505, + 243 + ], + "score": 1.0, + "content": "neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33(01), pages", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 130, + 240, + 200, + 251 + ], + "spans": [ + { + "bbox": [ + 130, + 240, + 200, + 251 + ], + "score": 1.0, + "content": "8481–8488, 2019.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 257, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 115, + 257, + 505, + 269 + ], + "score": 1.0, + "content": "[6] Dong Wook Shu, Sung Woo Park, and Junseok Kwon. 3d point cloud generative adversarial network", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 129, + 266, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 129, + 266, + 506, + 280 + ], + "score": 1.0, + "content": "based on tree structured graph convolutions. In Proceedings of the IEEE/CVF International Conference", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 130, + 277, + 298, + 289 + ], + "spans": [ + { + "bbox": [ + 130, + 277, + 298, + 289 + ], + "score": 1.0, + "content": "on Computer Vision, pages 3859–3868, 2019.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 113, + 294, + 507, + 307 + ], + "spans": [ + { + "bbox": [ + 113, + 294, + 507, + 307 + ], + "score": 1.0, + "content": "[7] Zhiqin Chen and Hao Zhang. Learning implicit fields for generative shape modeling. In Proceedings of", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 130, + 304, + 446, + 316 + ], + "spans": [ + { + "bbox": [ + 130, + 304, + 446, + 316 + ], + "score": 1.0, + "content": "the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 114, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 114, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "[8] Michael Niemeyer and Andreas Geiger. Giraffe: Representing scenes as compositional generative neural", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 131, + 332, + 477, + 344 + ], + "spans": [ + { + "bbox": [ + 131, + 332, + 477, + 344 + ], + "score": 1.0, + "content": "feature fields. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 114, + 350, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 114, + 350, + 505, + 361 + ], + "score": 1.0, + "content": "[9] Katja Schwarz, Yiyi Liao, Michael Niemeyer, and Andreas Geiger. Graf: Generative radiance fields for", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 129, + 358, + 493, + 372 + ], + "spans": [ + { + "bbox": [ + 129, + 358, + 493, + 372 + ], + "score": 1.0, + "content": "3d-aware image synthesis. In Advances in Neural Information Processing Systems (NeurIPS), 2020.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 109, + 376, + 507, + 390 + ], + "spans": [ + { + "bbox": [ + 109, + 376, + 507, + 390 + ], + "score": 1.0, + "content": "[10] Yiyi Liao, Katja Schwarz, Lars Mescheder, and Andreas Geiger. Towards unsupervised learning of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 128, + 385, + 507, + 400 + ], + "spans": [ + { + "bbox": [ + 128, + 385, + 507, + 400 + ], + "score": 1.0, + "content": "generative models for 3d controllable image synthesis. In Proceedings IEEE Conf. on Computer Vision", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 131, + 398, + 276, + 407 + ], + "spans": [ + { + "bbox": [ + 131, + 398, + 276, + 407 + ], + "score": 1.0, + "content": "and Pattern Recognition (CVPR), 2020.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 109, + 413, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 109, + 413, + 506, + 427 + ], + "score": 1.0, + "content": "[11] Eric Chan, Marco Monteiro, Petr Kellnhofer, Jiajun Wu, and Gordon Wetzstein. pi-gan: Periodic implicit", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 128, + 424, + 439, + 437 + ], + "spans": [ + { + "bbox": [ + 128, + 424, + 439, + 437 + ], + "score": 1.0, + "content": "generative adversarial networks for 3d-aware image synthesis. In Proc. CVPR, 2021.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 110, + 442, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 110, + 442, + 505, + 453 + ], + "score": 1.0, + "content": "[12] Eric R. Chan, Connor Z. Lin, Matthew A. Chan, Koki Nagano, Boxiao Pan, Shalini De Mello, Orazio", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 131, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 131, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "Gallo, Leonidas Guibas, Jonathan Tremblay, Sameh Khamis, Tero Karras, and Gordon Wetzstein. Efficient", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 129, + 463, + 380, + 473 + ], + "spans": [ + { + "bbox": [ + 129, + 463, + 380, + 473 + ], + "score": 1.0, + "content": "geometry-aware 3D generative adversarial networks. In arXiv, 2021.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 110, + 479, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 110, + 479, + 506, + 491 + ], + "score": 1.0, + "content": "[13] Roy Or-El, Xuan Luo, Mengyi Shan, Eli Shechtman, Jeong Joon Park, and Ira Kemelmacher-Shlizerman.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 129, + 488, + 507, + 502 + ], + "spans": [ + { + "bbox": [ + 129, + 488, + 507, + 502 + ], + "score": 1.0, + "content": "Stylesdf: High-resolution 3d-consistent image and geometry generation. arXiv preprint arXiv:2112.11427,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 129, + 497, + 155, + 512 + ], + "spans": [ + { + "bbox": [ + 129, + 497, + 155, + 512 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 109, + 516, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 109, + 516, + 506, + 530 + ], + "score": 1.0, + "content": "[14] Jiatao Gu, Lingjie Liu, Peng Wang, and Christian Theobalt. Stylenerf: A style-based 3d aware generator", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 131, + 527, + 496, + 539 + ], + "spans": [ + { + "bbox": [ + 131, + 527, + 496, + 539 + ], + "score": 1.0, + "content": "for high-resolution image synthesis. In International Conference on Learning Representations, 2022.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 110, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 110, + 545, + 505, + 556 + ], + "score": 1.0, + "content": "[15] Peng Zhou, Lingxi Xie, Bingbing Ni, and Qi Tian. CIPS-3D: A 3D-Aware Generator of GANs Based on", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 129, + 554, + 438, + 568 + ], + "spans": [ + { + "bbox": [ + 129, + 554, + 438, + 568 + ], + "score": 1.0, + "content": "Conditionally-Independent Pixel Synthesis. arXiv preprint arXiv:2110.09788, 2021.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 109, + 571, + 507, + 585 + ], + "spans": [ + { + "bbox": [ + 109, + 571, + 507, + 585 + ], + "score": 1.0, + "content": "[16] Dario Pavllo, Jonas Kohler, Thomas Hofmann, and Aurelien Lucchi. Learning generative models of", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 128, + 580, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 128, + 580, + 506, + 595 + ], + "score": 1.0, + "content": "textured 3d meshes from real-world images. In IEEE/CVF International Conference on Computer Vision", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 129, + 592, + 184, + 604 + ], + "spans": [ + { + "bbox": [ + 129, + 592, + 184, + 604 + ], + "score": 1.0, + "content": "(ICCV), 2021.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 110, + 609, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 110, + 609, + 506, + 621 + ], + "score": 1.0, + "content": "[17] Yuxuan Zhang, Wenzheng Chen, Huan Ling, Jun Gao, Yinan Zhang, Antonio Torralba, and Sanja Fidler.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 128, + 617, + 507, + 633 + ], + "spans": [ + { + "bbox": [ + 128, + 617, + 507, + 633 + ], + "score": 1.0, + "content": "Image {gan}s meet differentiable rendering for inverse graphics and interpretable 3d neural rendering. In", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 130, + 629, + 354, + 641 + ], + "spans": [ + { + "bbox": [ + 130, + 629, + 354, + 641 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2021.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 109, + 646, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 109, + 646, + 506, + 660 + ], + "score": 1.0, + "content": "[18] Moritz Ibing, Isaak Lim, and Leif P. Kobbelt. 3d shape generation with grid-based implicit functions. In", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 130, + 657, + 454, + 668 + ], + "spans": [ + { + "bbox": [ + 130, + 657, + 454, + 668 + ], + "score": 1.0, + "content": "2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 109, + 671, + 507, + 688 + ], + "spans": [ + { + "bbox": [ + 109, + 671, + 507, + 688 + ], + "score": 1.0, + "content": "[19] Ruihui Li, Xianzhi Li, Ke-Hei Hui, and Chi-Wing Fu. SP-GAN:sphere-guided 3d shape generation and", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 131, + 685, + 421, + 696 + ], + "spans": [ + { + "bbox": [ + 131, + 685, + 421, + 696 + ], + "score": 1.0, + "content": "manipulation. ACM Transactions on Graphics (Proc. SIGGRAPH), 40(4), 2021.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 109, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 109, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "[20] A. Luo, T. Li, W. Zhang, and T. Lee. Surfgen: Adversarial 3d shape synthesis with explicit surface", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 130, + 712, + 478, + 723 + ], + "spans": [ + { + "bbox": [ + 130, + 712, + 478, + 723 + ], + "score": 1.0, + "content": "discriminators. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 24.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 310, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 107, + 65, + 507, + 729 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 113, + 88, + 505, + 102 + ], + "spans": [ + { + "bbox": [ + 113, + 88, + 505, + 102 + ], + "score": 1.0, + "content": "[1] Jiajun Wu, Chengkai Zhang, Tianfan Xue, Bill Freeman, and Josh Tenenbaum. Learning a probabilistic", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 131, + 100, + 505, + 111 + ], + "spans": [ + { + "bbox": [ + 131, + 100, + 505, + 111 + ], + "score": 1.0, + "content": "latent space of object shapes via 3d generative-adversarial modeling. In Advances in Neural Information", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 129, + 108, + 229, + 122 + ], + "spans": [ + { + "bbox": [ + 129, + 108, + 229, + 122 + ], + "score": 1.0, + "content": "Processing Systems, 2016.", + "type": "text" + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 114, + 127, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 114, + 127, + 506, + 140 + ], + "score": 1.0, + "content": "[2] Panos Achlioptas, Olga Diamanti, Ioannis Mitliagkas, and Leonidas Guibas. Learning representations", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 128, + 136, + 348, + 149 + ], + "spans": [ + { + "bbox": [ + 128, + 136, + 348, + 149 + ], + "score": 1.0, + "content": "and generative models for 3D point clouds. In ICML, 2018.", + "type": "text" + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 113, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 113, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "[3] Chun-Liang Li, Manzil Zaheer, Yang Zhang, Barnabas Poczos, and Ruslan Salakhutdinov. Point cloud", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 164, + 296, + 176 + ], + "spans": [ + { + "bbox": [ + 128, + 164, + 296, + 176 + ], + "score": 1.0, + "content": "gan. arXiv preprint arXiv:1810.05795, 2018.", + "type": "text" + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 113, + 181, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 113, + 181, + 506, + 194 + ], + "score": 1.0, + "content": "[4] Diego Valsesia, Giulia Fracastoro, and Enrico Magli. Learning localized generative models for 3d point", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 191, + 507, + 205 + ], + "spans": [ + { + "bbox": [ + 129, + 191, + 507, + 205 + ], + "score": 1.0, + "content": "clouds via graph convolution. In International Conference on Learning Representations (ICLR) 2019,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 130, + 202, + 154, + 214 + ], + "spans": [ + { + "bbox": [ + 130, + 202, + 154, + 214 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 114, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 114, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "[5] Wenlong Huang, Brian Lai, Weijian Xu, and Zhuowen Tu. 3d volumetric modeling with introspective", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 128, + 228, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 128, + 228, + 505, + 243 + ], + "score": 1.0, + "content": "neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33(01), pages", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 130, + 240, + 200, + 251 + ], + "spans": [ + { + "bbox": [ + 130, + 240, + 200, + 251 + ], + "score": 1.0, + "content": "8481–8488, 2019.", + "type": "text" + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 115, + 257, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 115, + 257, + 505, + 269 + ], + "score": 1.0, + "content": "[6] Dong Wook Shu, Sung Woo Park, and Junseok Kwon. 3d point cloud generative adversarial network", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 129, + 266, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 129, + 266, + 506, + 280 + ], + "score": 1.0, + "content": "based on tree structured graph convolutions. In Proceedings of the IEEE/CVF International Conference", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 130, + 277, + 298, + 289 + ], + "spans": [ + { + "bbox": [ + 130, + 277, + 298, + 289 + ], + "score": 1.0, + "content": "on Computer Vision, pages 3859–3868, 2019.", + "type": "text" + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 113, + 294, + 507, + 307 + ], + "spans": [ + { + "bbox": [ + 113, + 294, + 507, + 307 + ], + "score": 1.0, + "content": "[7] Zhiqin Chen and Hao Zhang. Learning implicit fields for generative shape modeling. In Proceedings of", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 304, + 446, + 316 + ], + "spans": [ + { + "bbox": [ + 130, + 304, + 446, + 316 + ], + "score": 1.0, + "content": "the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019.", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 114, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 114, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "[8] Michael Niemeyer and Andreas Geiger. Giraffe: Representing scenes as compositional generative neural", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 131, + 332, + 477, + 344 + ], + "spans": [ + { + "bbox": [ + 131, + 332, + 477, + 344 + ], + "score": 1.0, + "content": "feature fields. In Proc. IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 114, + 350, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 114, + 350, + 505, + 361 + ], + "score": 1.0, + "content": "[9] Katja Schwarz, Yiyi Liao, Michael Niemeyer, and Andreas Geiger. Graf: Generative radiance fields for", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 129, + 358, + 493, + 372 + ], + "spans": [ + { + "bbox": [ + 129, + 358, + 493, + 372 + ], + "score": 1.0, + "content": "3d-aware image synthesis. In Advances in Neural Information Processing Systems (NeurIPS), 2020.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 109, + 376, + 507, + 390 + ], + "spans": [ + { + "bbox": [ + 109, + 376, + 507, + 390 + ], + "score": 1.0, + "content": "[10] Yiyi Liao, Katja Schwarz, Lars Mescheder, and Andreas Geiger. Towards unsupervised learning of", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 385, + 507, + 400 + ], + "spans": [ + { + "bbox": [ + 128, + 385, + 507, + 400 + ], + "score": 1.0, + "content": "generative models for 3d controllable image synthesis. In Proceedings IEEE Conf. on Computer Vision", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 131, + 398, + 276, + 407 + ], + "spans": [ + { + "bbox": [ + 131, + 398, + 276, + 407 + ], + "score": 1.0, + "content": "and Pattern Recognition (CVPR), 2020.", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 413, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 109, + 413, + 506, + 427 + ], + "score": 1.0, + "content": "[11] Eric Chan, Marco Monteiro, Petr Kellnhofer, Jiajun Wu, and Gordon Wetzstein. pi-gan: Periodic implicit", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 424, + 439, + 437 + ], + "spans": [ + { + "bbox": [ + 128, + 424, + 439, + 437 + ], + "score": 1.0, + "content": "generative adversarial networks for 3d-aware image synthesis. In Proc. CVPR, 2021.", + "type": "text" + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 442, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 110, + 442, + 505, + 453 + ], + "score": 1.0, + "content": "[12] Eric R. Chan, Connor Z. Lin, Matthew A. Chan, Koki Nagano, Boxiao Pan, Shalini De Mello, Orazio", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 131, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 131, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "Gallo, Leonidas Guibas, Jonathan Tremblay, Sameh Khamis, Tero Karras, and Gordon Wetzstein. Efficient", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 129, + 463, + 380, + 473 + ], + "spans": [ + { + "bbox": [ + 129, + 463, + 380, + 473 + ], + "score": 1.0, + "content": "geometry-aware 3D generative adversarial networks. In arXiv, 2021.", + "type": "text" + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 479, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 110, + 479, + 506, + 491 + ], + "score": 1.0, + "content": "[13] Roy Or-El, Xuan Luo, Mengyi Shan, Eli Shechtman, Jeong Joon Park, and Ira Kemelmacher-Shlizerman.", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 488, + 507, + 502 + ], + "spans": [ + { + "bbox": [ + 129, + 488, + 507, + 502 + ], + "score": 1.0, + "content": "Stylesdf: High-resolution 3d-consistent image and geometry generation. arXiv preprint arXiv:2112.11427,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 129, + 497, + 155, + 512 + ], + "spans": [ + { + "bbox": [ + 129, + 497, + 155, + 512 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 516, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 109, + 516, + 506, + 530 + ], + "score": 1.0, + "content": "[14] Jiatao Gu, Lingjie Liu, Peng Wang, and Christian Theobalt. Stylenerf: A style-based 3d aware generator", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 131, + 527, + 496, + 539 + ], + "spans": [ + { + "bbox": [ + 131, + 527, + 496, + 539 + ], + "score": 1.0, + "content": "for high-resolution image synthesis. In International Conference on Learning Representations, 2022.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 110, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 110, + 545, + 505, + 556 + ], + "score": 1.0, + "content": "[15] Peng Zhou, Lingxi Xie, Bingbing Ni, and Qi Tian. CIPS-3D: A 3D-Aware Generator of GANs Based on", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 554, + 438, + 568 + ], + "spans": [ + { + "bbox": [ + 129, + 554, + 438, + 568 + ], + "score": 1.0, + "content": "Conditionally-Independent Pixel Synthesis. arXiv preprint arXiv:2110.09788, 2021.", + "type": "text" + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 571, + 507, + 585 + ], + "spans": [ + { + "bbox": [ + 109, + 571, + 507, + 585 + ], + "score": 1.0, + "content": "[16] Dario Pavllo, Jonas Kohler, Thomas Hofmann, and Aurelien Lucchi. Learning generative models of", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 580, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 128, + 580, + 506, + 595 + ], + "score": 1.0, + "content": "textured 3d meshes from real-world images. In IEEE/CVF International Conference on Computer Vision", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 129, + 592, + 184, + 604 + ], + "spans": [ + { + "bbox": [ + 129, + 592, + 184, + 604 + ], + "score": 1.0, + "content": "(ICCV), 2021.", + "type": "text" + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 609, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 110, + 609, + 506, + 621 + ], + "score": 1.0, + "content": "[17] Yuxuan Zhang, Wenzheng Chen, Huan Ling, Jun Gao, Yinan Zhang, Antonio Torralba, and Sanja Fidler.", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 617, + 507, + 633 + ], + "spans": [ + { + "bbox": [ + 128, + 617, + 507, + 633 + ], + "score": 1.0, + "content": "Image {gan}s meet differentiable rendering for inverse graphics and interpretable 3d neural rendering. In", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 130, + 629, + 354, + 641 + ], + "spans": [ + { + "bbox": [ + 130, + 629, + 354, + 641 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2021.", + "type": "text" + } + ], + "index": 43, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 646, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 109, + 646, + 506, + 660 + ], + "score": 1.0, + "content": "[18] Moritz Ibing, Isaak Lim, and Leif P. Kobbelt. 3d shape generation with grid-based implicit functions. In", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 657, + 454, + 668 + ], + "spans": [ + { + "bbox": [ + 130, + 657, + 454, + 668 + ], + "score": 1.0, + "content": "2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 671, + 507, + 688 + ], + "spans": [ + { + "bbox": [ + 109, + 671, + 507, + 688 + ], + "score": 1.0, + "content": "[19] Ruihui Li, Xianzhi Li, Ke-Hei Hui, and Chi-Wing Fu. SP-GAN:sphere-guided 3d shape generation and", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 131, + 685, + 421, + 696 + ], + "spans": [ + { + "bbox": [ + 131, + 685, + 421, + 696 + ], + "score": 1.0, + "content": "manipulation. ACM Transactions on Graphics (Proc. SIGGRAPH), 40(4), 2021.", + "type": "text" + } + ], + "index": 47, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 109, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "[20] A. Luo, T. Li, W. Zhang, and T. Lee. Surfgen: Adversarial 3d shape synthesis with explicit surface", + "type": "text" + } + ], + "index": 48, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 712, + 478, + 723 + ], + "spans": [ + { + "bbox": [ + 130, + 712, + 478, + 723 + ], + "score": 1.0, + "content": "discriminators. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021.", + "type": "text" + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 73, + 507, + 85 + ], + "spans": [ + { + "bbox": [ + 110, + 73, + 507, + 85 + ], + "score": 1.0, + "content": "[21] Zhiqin Chen, Vladimir G. Kim, Matthew Fisher, Noam Aigerman, Hao Zhang, and Siddhartha Chaudhuri.", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 82, + 507, + 97 + ], + "spans": [ + { + "bbox": [ + 129, + 82, + 507, + 97 + ], + "score": 1.0, + "content": "Decor-gan: 3d shape detailization by conditional refinement. In Proceedings of IEEE Conference on", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 129, + 93, + 338, + 106 + ], + "spans": [ + { + "bbox": [ + 129, + 93, + 338, + 106 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 112, + 507, + 126 + ], + "spans": [ + { + "bbox": [ + 110, + 112, + 507, + 126 + ], + "score": 1.0, + "content": "[22] Cheng Wen, Baosheng Yu, and Dacheng Tao. Learning progressive point embeddings for 3d point cloud", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 121, + 507, + 135 + ], + "spans": [ + { + "bbox": [ + 129, + 121, + 507, + 135 + ], + "score": 1.0, + "content": "generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 128, + 132, + 232, + 144 + ], + "spans": [ + { + "bbox": [ + 128, + 132, + 232, + 144 + ], + "score": 1.0, + "content": "pages 10266–10275, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 110, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "[23] Zekun Hao, Arun Mallya, Serge Belongie, and Ming-Yu Liu. GANcraft: Unsupervised 3D Neural", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 160, + 307, + 173 + ], + "spans": [ + { + "bbox": [ + 130, + 160, + 307, + 173 + ], + "score": 1.0, + "content": "Rendering of Minecraft Worlds. In ICCV, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 180, + 504, + 192 + ], + "spans": [ + { + "bbox": [ + 110, + 180, + 504, + 192 + ], + "score": 1.0, + "content": "[24] Or Litany, Alex Bronstein, Michael Bronstein, and Ameesh Makadia. Deformable shape completion with", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 189, + 308, + 203 + ], + "spans": [ + { + "bbox": [ + 129, + 189, + 308, + 203 + ], + "score": 1.0, + "content": "graph convolutional autoencoders. CVPR, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 208, + 507, + 222 + ], + "spans": [ + { + "bbox": [ + 109, + 208, + 507, + 222 + ], + "score": 1.0, + "content": "[25] Qingyang Tan, Lin Gao, Yu-Kun Lai, and Shihong Xia. Variational autoencoders for deforming 3d", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 217, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 128, + 217, + 506, + 233 + ], + "score": 1.0, + "content": "mesh models. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 129, + 229, + 199, + 241 + ], + "spans": [ + { + "bbox": [ + 129, + 229, + 199, + 241 + ], + "score": 1.0, + "content": "5841–5850, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 248, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 110, + 248, + 506, + 260 + ], + "score": 1.0, + "content": "[26] Kaichun Mo, Paul Guerrero, Li Yi, Hao Su, Peter Wonka, Niloy Mitra, and Leonidas J Guibas. Struc-", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 256, + 501, + 271 + ], + "spans": [ + { + "bbox": [ + 129, + 256, + 501, + 271 + ], + "score": 1.0, + "content": "turenet: Hierarchical graph networks for 3d shape generation. arXiv preprint arXiv:1908.00575, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 110, + 276, + 507, + 289 + ], + "spans": [ + { + "bbox": [ + 110, + 276, + 507, + 289 + ], + "score": 1.0, + "content": "[27] Lin Gao, Jie Yang, Tong Wu, Yu-Jie Yuan, Hongbo Fu, Yu-Kun Lai, and Hao(Richard) Zhang. SDM-NET:", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 129, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "Deep generative network for structured deformable mesh. ACM Transactions on Graphics (Proceedings", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 129, + 295, + 350, + 309 + ], + "spans": [ + { + "bbox": [ + 129, + 295, + 350, + 309 + ], + "score": 1.0, + "content": "of ACM SIGGRAPH Asia 2019), 38(6):243:1–243:15, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 316, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 110, + 316, + 505, + 328 + ], + "score": 1.0, + "content": "[28] Lin Gao, Tong Wu, Yu-Jie Yuan, Ming-Xian Lin, Yu-Kun Lai, and Hao Zhang. Tm-net: Deep generative", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 325, + 486, + 338 + ], + "spans": [ + { + "bbox": [ + 130, + 325, + 486, + 338 + ], + "score": 1.0, + "content": "networks for textured meshes. ACM Transactions on Graphics (TOG), 40(6):263:1–263:15, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 344, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 110, + 344, + 506, + 356 + ], + "score": 1.0, + "content": "[29] Jinwoo Kim, Jaehoon Yoo, Juho Lee, and Seunghoon Hong. Setvae: Learning hierarchical composition", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 353, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 129, + 353, + 505, + 367 + ], + "score": 1.0, + "content": "for generative modeling of set-structured data. In Proceedings of the IEEE/CVF Conference on Computer", + "type": "text", + "cross_page": true + } + ], + "index": 21 + }, + { + "bbox": [ + 130, + 365, + 396, + 376 + ], + "spans": [ + { + "bbox": [ + 130, + 365, + 396, + 376 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition (CVPR), pages 15059–15068, June 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 110, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "[30] Paritosh Mittal, Yen-Chi Cheng, Maneesh Singh, and Shubham Tulsiani. AutoSDF: Shape priors for 3d", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 393, + 346, + 405 + ], + "spans": [ + { + "bbox": [ + 130, + 393, + 346, + 405 + ], + "score": 1.0, + "content": "completion, reconstruction and generation. In CVPR, 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 412, + 507, + 424 + ], + "spans": [ + { + "bbox": [ + 110, + 412, + 507, + 424 + ], + "score": 1.0, + "content": "[31] Guandao Yang, Xun Huang, Zekun Hao, Ming-Yu Liu, Serge Belongie, and Bharath Hariharan. PointFlow:", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 422, + 414, + 434 + ], + "spans": [ + { + "bbox": [ + 130, + 422, + 414, + 434 + ], + "score": 1.0, + "content": "3D point cloud generation with continuous normalizing flows. In ICCV, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 440, + 508, + 453 + ], + "spans": [ + { + "bbox": [ + 109, + 440, + 508, + 453 + ], + "score": 1.0, + "content": "[32] Hyeongju Kim, Hyeonseung Lee, Woo Hyun Kang, Joun Yeop Lee, and Nam Soo Kim. SoftFlow:", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 451, + 417, + 463 + ], + "spans": [ + { + "bbox": [ + 129, + 451, + 417, + 463 + ], + "score": 1.0, + "content": "Probabilistic framework for normalizing flow on manifolds. In NeurIPS, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 468, + 507, + 483 + ], + "spans": [ + { + "bbox": [ + 110, + 468, + 507, + 483 + ], + "score": 1.0, + "content": "[33] Roman Klokov, Edmond Boyer, and Jakob Verbeek. Discrete point flow networks for efficient point cloud", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 480, + 235, + 492 + ], + "spans": [ + { + "bbox": [ + 128, + 480, + 235, + 492 + ], + "score": 1.0, + "content": "generation. In ECCV, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 496, + 507, + 513 + ], + "spans": [ + { + "bbox": [ + 108, + 496, + 507, + 513 + ], + "score": 1.0, + "content": "[34] Aditya Sanghi, Hang Chu, Joseph G Lambourne, Ye Wang, Chin-Yi Cheng, and Marco Fumero. Clip-", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 507, + 464, + 521 + ], + "spans": [ + { + "bbox": [ + 128, + 507, + 464, + 521 + ], + "score": 1.0, + "content": "forge: Towards zero-shot text-to-shape generation. arXiv preprint arXiv:2110.02624, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 526, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 110, + 526, + 506, + 541 + ], + "score": 1.0, + "content": "[35] Yongbin Sun, Yue Wang, Ziwei Liu, Joshua E Siegel, and Sanjay E Sarma. Pointgrow: Autoregressively", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 535, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 129, + 535, + 506, + 551 + ], + "score": 1.0, + "content": "learned point cloud generation with self-attention. In Winter Conference on Applications of Computer", + "type": "text", + "cross_page": true + } + ], + "index": 34 + }, + { + "bbox": [ + 130, + 547, + 181, + 559 + ], + "spans": [ + { + "bbox": [ + 130, + 547, + 181, + 559 + ], + "score": 1.0, + "content": "Vision, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 564, + 507, + 579 + ], + "spans": [ + { + "bbox": [ + 108, + 564, + 507, + 579 + ], + "score": 1.0, + "content": "[36] Charlie Nash, Yaroslav Ganin, S. M. Ali Eslami, and Peter W. Battaglia. Polygen: An autoregressive", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 575, + 297, + 589 + ], + "spans": [ + { + "bbox": [ + 128, + 575, + 297, + 589 + ], + "score": 1.0, + "content": "generative model of 3d meshes. ICML, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 594, + 507, + 608 + ], + "spans": [ + { + "bbox": [ + 110, + 594, + 507, + 608 + ], + "score": 1.0, + "content": "[37] Wei-Jan Ko, Hui-Yu Huang, Yu-Liang Kuo, Chen-Yi Chiu, Li-Heng Wang, and Wei-Chen Chiu. Rpg:", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 605, + 433, + 617 + ], + "spans": [ + { + "bbox": [ + 129, + 605, + 433, + 617 + ], + "score": 1.0, + "content": "Learning recursive point cloud generation. arXiv preprint arXiv:2105.14322, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 110, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "[38] Moritz Ibing, Gregor Kobsik, and Leif Kobbelt. Octree transformer: Autoregressive 3d shape generation", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 634, + 420, + 646 + ], + "spans": [ + { + "bbox": [ + 130, + 634, + 420, + 646 + ], + "score": 1.0, + "content": "on hierarchically structured sequences. arXiv preprint arXiv:2111.12480, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 651, + 507, + 666 + ], + "spans": [ + { + "bbox": [ + 109, + 651, + 507, + 666 + ], + "score": 1.0, + "content": "[39] Jianwen Xie, Zilong Zheng, Ruiqi Gao, Wenguan Wang, Zhu Song-Chun, and Ying Nian Wu. Learning", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 663, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 130, + 663, + 506, + 675 + ], + "score": 1.0, + "content": "descriptor networks for 3d shape synthesis and analysis. In The IEEE Conference on Computer Vision", + "type": "text", + "cross_page": true + } + ], + "index": 43 + }, + { + "bbox": [ + 129, + 672, + 276, + 684 + ], + "spans": [ + { + "bbox": [ + 129, + 672, + 276, + 684 + ], + "score": 1.0, + "content": "and Pattern Recognition (CVPR), 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 690, + 507, + 705 + ], + "spans": [ + { + "bbox": [ + 110, + 690, + 507, + 705 + ], + "score": 1.0, + "content": "[40] Jianwen Xie, Yifei Xu, Zilong Zheng, Ruiqi Gao, Wenguan Wang, Zhu Song-Chun, and Ying Nian Wu.", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 702, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 130, + 702, + 506, + 714 + ], + "score": 1.0, + "content": "Generative pointnet: Deep energy-based learning on unordered point sets for 3d generation, reconstruction", + "type": "text", + "cross_page": true + } + ], + "index": 46 + }, + { + "bbox": [ + 129, + 712, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 129, + 712, + 506, + 723 + ], + "score": 1.0, + "content": "and classification. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 47 + }, + { + "bbox": [ + 111, + 73, + 505, + 84 + ], + "spans": [ + { + "bbox": [ + 111, + 73, + 505, + 84 + ], + "score": 1.0, + "content": "[41] Tianchang Shen, Jun Gao, Kangxue Yin, Ming-Yu Liu, and Sanja Fidler. Deep marching tetrahedra:", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 128, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "a hybrid representation for high-resolution 3d shape synthesis. In Advances in Neural Information", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 128, + 93, + 268, + 106 + ], + "spans": [ + { + "bbox": [ + 128, + 93, + 268, + 106 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 111, + 112, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 111, + 112, + 505, + 124 + ], + "score": 1.0, + "content": "[42] Kangxue Yin, Jun Gao, Maria Shugrina, Sameh Khamis, and Sanja Fidler. 3dstylenet: Creating 3d shapes", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 121, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 128, + 121, + 505, + 135 + ], + "score": 1.0, + "content": "with geometric and texture style variations. In Proceedings of International Conference on Computer", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 131, + 132, + 209, + 144 + ], + "spans": [ + { + "bbox": [ + 131, + 132, + 209, + 144 + ], + "score": 1.0, + "content": "Vision (ICCV), 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 110, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "[43] Dongsu Zhang, Changwoon Choi, Jeonghwan Kim, and Young Min Kim. Learning to generate 3d shapes", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 160, + 488, + 172 + ], + "spans": [ + { + "bbox": [ + 130, + 160, + 488, + 172 + ], + "score": 1.0, + "content": "with generative cellular automata. In International Conference on Learning Representations, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 178, + 506, + 191 + ], + "spans": [ + { + "bbox": [ + 109, + 178, + 506, + 191 + ], + "score": 1.0, + "content": "[44] Chen Chao, Zhizhong Han, Yu-Shen Liu, and Matthias Zwicker. Unsupervised learning of fine structure", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 187, + 489, + 202 + ], + "spans": [ + { + "bbox": [ + 128, + 187, + 489, + 202 + ], + "score": 1.0, + "content": "generation for 3d point clouds by 2d projection matching. arXiv preprint arXiv:2108.03746, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 208, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 110, + 208, + 505, + 219 + ], + "score": 1.0, + "content": "[45] Ruojin Cai, Guandao Yang, Hadar Averbuch-Elor, Zekun Hao, Serge Belongie, Noah Snavely, and", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 214, + 506, + 232 + ], + "spans": [ + { + "bbox": [ + 128, + 214, + 506, + 232 + ], + "score": 1.0, + "content": "Bharath Hariharan. Learning gradient fields for shape generation. In Proceedings of the European", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 130, + 227, + 305, + 239 + ], + "spans": [ + { + "bbox": [ + 130, + 227, + 305, + 239 + ], + "score": 1.0, + "content": "Conference on Computer Vision (ECCV), 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 245, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 110, + 245, + 506, + 257 + ], + "score": 1.0, + "content": "[46] Linqi Zhou, Yilun Du, and Jiajun Wu. 3d shape generation and completion through point-voxel diffusion.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 255, + 470, + 267 + ], + "spans": [ + { + "bbox": [ + 130, + 255, + 470, + 267 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 272, + 507, + 287 + ], + "spans": [ + { + "bbox": [ + 109, + 272, + 507, + 287 + ], + "score": 1.0, + "content": "[47] Shitong Luo and Wei Hu. Diffusion probabilistic models for 3d point cloud generation. In Proceedings of", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 284, + 448, + 296 + ], + "spans": [ + { + "bbox": [ + 130, + 284, + 448, + 296 + ], + "score": 1.0, + "content": "the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 111, + 303, + 505, + 314 + ], + "spans": [ + { + "bbox": [ + 111, + 303, + 505, + 314 + ], + "score": 1.0, + "content": "[48] Yu Deng, Jiaolong Yang, and Xin Tong. Deformed implicit field: Modeling 3d shapes with learned dense", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 312, + 402, + 324 + ], + "spans": [ + { + "bbox": [ + 128, + 312, + 402, + 324 + ], + "score": 1.0, + "content": "correspondence. In IEEE Computer Vision and Pattern Recognition, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 331, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 110, + 331, + 505, + 342 + ], + "score": 1.0, + "content": "[49] Oscar Michel, Roi Bar-On, Richard Liu, Sagie Benaim, and Rana Hanocka. Text2mesh: Text-driven", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 340, + 387, + 353 + ], + "spans": [ + { + "bbox": [ + 128, + 340, + 387, + 353 + ], + "score": 1.0, + "content": "neural stylization for meshes. arXiv preprint arXiv:2112.03221, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 358, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 109, + 358, + 506, + 372 + ], + "score": 1.0, + "content": "[50] Ajay Jain, Ben Mildenhall, Jonathan T. Barron, Pieter Abbeel, and Ben Poole. Zero-shot text-guided", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 128, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "object generation with dream fields. In IEEE Conference on Computer Vision and Pattern Recognition", + "type": "text", + "cross_page": true + } + ], + "index": 22 + }, + { + "bbox": [ + 129, + 379, + 186, + 391 + ], + "spans": [ + { + "bbox": [ + 129, + 379, + 186, + 391 + ], + "score": 1.0, + "content": "(CVPR), 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 111, + 398, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 111, + 398, + 505, + 410 + ], + "score": 1.0, + "content": "[51] Nasir Khalid, Tianhao Xie, Eugene Belilovsky, and Tiberiu Popa. Text to mesh without 3d supervision", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 407, + 365, + 420 + ], + "spans": [ + { + "bbox": [ + 128, + 407, + 365, + 420 + ], + "score": 1.0, + "content": "using limit subdivision. arXiv preprint arXiv:2203.13333, 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 425, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 109, + 425, + 506, + 439 + ], + "score": 1.0, + "content": "[52] Le Hui, Rui Xu, Jin Xie, Jianjun Qian, and Jian Yang. Progressive point cloud deconvolution generation", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 131, + 437, + 226, + 447 + ], + "spans": [ + { + "bbox": [ + 131, + 437, + 226, + 447 + ], + "score": 1.0, + "content": "network. In ECCV, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 454, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 110, + 454, + 506, + 468 + ], + "score": 1.0, + "content": "[53] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 465, + 301, + 476 + ], + "spans": [ + { + "bbox": [ + 130, + 465, + 301, + 476 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 482, + 507, + 496 + ], + "spans": [ + { + "bbox": [ + 109, + 482, + 507, + 496 + ], + "score": 1.0, + "content": "[54] Jonathan Ho, Chitwan Saharia, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 492, + 503, + 506 + ], + "spans": [ + { + "bbox": [ + 129, + 492, + 503, + 506 + ], + "score": 1.0, + "content": "Cascaded diffusion models for high fidelity image generation. arXiv preprint arXiv:2106.15282, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 31 + }, + { + "bbox": [ + 109, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 109, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "[55] Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. In", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 522, + 327, + 533 + ], + "spans": [ + { + "bbox": [ + 130, + 522, + 327, + 533 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 111, + 540, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 111, + 540, + 505, + 551 + ], + "score": 1.0, + "content": "[56] Prafulla Dhariwal and Alexander Quinn Nichol. Diffusion models beat GANs on image synthesis. In", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 550, + 346, + 561 + ], + "spans": [ + { + "bbox": [ + 129, + 550, + 346, + 561 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 568, + 505, + 580 + ], + "spans": [ + { + "bbox": [ + 110, + 568, + 505, + 580 + ], + "score": 1.0, + "content": "[57] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 578, + 444, + 591 + ], + "spans": [ + { + "bbox": [ + 128, + 578, + 444, + 591 + ], + "score": 1.0, + "content": "image synthesis with latent diffusion models. arXiv preprint arXiv:2112.10752, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 596, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 110, + 596, + 506, + 609 + ], + "score": 1.0, + "content": "[58] Arash Vahdat, Karsten Kreis, and Jan Kautz. Score-based generative modeling in latent space. In", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 606, + 346, + 619 + ], + "spans": [ + { + "bbox": [ + 128, + 606, + 346, + 619 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 624, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 110, + 624, + 506, + 637 + ], + "score": 1.0, + "content": "[59] Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 632, + 507, + 649 + ], + "spans": [ + { + "bbox": [ + 128, + 632, + 507, + 649 + ], + "score": 1.0, + "content": "Sutskever, and Mark Chen. Glide: Towards photorealistic image generation and editing with text-guided", + "type": "text", + "cross_page": true + } + ], + "index": 41 + }, + { + "bbox": [ + 131, + 645, + 344, + 657 + ], + "spans": [ + { + "bbox": [ + 131, + 645, + 344, + 657 + ], + "score": 1.0, + "content": "diffusion models. arXiv preprint arXiv:2112.10741, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_end_line": true + }, + { + "bbox": [ + 111, + 664, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 111, + 664, + 505, + 675 + ], + "score": 1.0, + "content": "[60] Konpat Preechakul, Nattanat Chatthee, Suttisak Wizadwongsa, and Supasorn Suwajanakorn. Diffusion", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 672, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 128, + 672, + 505, + 686 + ], + "score": 1.0, + "content": "autoencoders: Toward a meaningful and decodable representation. In IEEE Conference on Computer", + "type": "text", + "cross_page": true + } + ], + "index": 44 + }, + { + "bbox": [ + 130, + 682, + 300, + 696 + ], + "spans": [ + { + "bbox": [ + 130, + 682, + 300, + 696 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition (CVPR), 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 109, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "[61] Tim Dockhorn, Arash Vahdat, and Karsten Kreis. Score-based generative modeling with critically-damped", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 712, + 462, + 724 + ], + "spans": [ + { + "bbox": [ + 130, + 712, + 462, + 724 + ], + "score": 1.0, + "content": "langevin diffusion. In International Conference on Learning Representations (ICLR), 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 47, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 73, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 109, + 73, + 507, + 87 + ], + "score": 1.0, + "content": "[62] Zhisheng Xiao, Karsten Kreis, and Arash Vahdat. Tackling the generative learning trilemma with", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 82, + 492, + 97 + ], + "spans": [ + { + "bbox": [ + 128, + 82, + 492, + 97 + ], + "score": 1.0, + "content": "denoising diffusion GANs. In International Conference on Learning Representations (ICLR), 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 101, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 108, + 101, + 506, + 114 + ], + "score": 1.0, + "content": "[63] Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 111, + 405, + 125 + ], + "spans": [ + { + "bbox": [ + 128, + 111, + 405, + 125 + ], + "score": 1.0, + "content": "image generation with clip latents. arXiv preprint arXiv:2204.06125, 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 129, + 507, + 143 + ], + "spans": [ + { + "bbox": [ + 109, + 129, + 507, + 143 + ], + "score": 1.0, + "content": "[64] Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed Kam-", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 131, + 141, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 131, + 141, + 506, + 153 + ], + "score": 1.0, + "content": "yar Seyed Ghasemipour, Burcu Karagol Ayan, S. Sara Mahdavi, Rapha Gontijo Lopes, Tim Salimans,", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 128, + 148, + 507, + 163 + ], + "spans": [ + { + "bbox": [ + 128, + 148, + 507, + 163 + ], + "score": 1.0, + "content": "Jonathan Ho, David J Fleet, and Mohammad Norouzi. Photorealistic text-to-image diffusion models with", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 130, + 160, + 388, + 172 + ], + "spans": [ + { + "bbox": [ + 130, + 160, + 388, + 172 + ], + "score": 1.0, + "content": "deep language understanding. arXiv preprint arXiv:2205.11487, 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 178, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 110, + 178, + 505, + 190 + ], + "score": 1.0, + "content": "[65] Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 189, + 507, + 201 + ], + "spans": [ + { + "bbox": [ + 130, + 189, + 507, + 201 + ], + "score": 1.0, + "content": "learning using nonequilibrium thermodynamics. In International Conference on Machine Learning, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 110, + 207, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 110, + 207, + 506, + 219 + ], + "score": 1.0, + "content": "[66] Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. In", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 217, + 470, + 230 + ], + "spans": [ + { + "bbox": [ + 130, + 217, + 470, + 230 + ], + "score": 1.0, + "content": "Proceedings of the 33rd Annual Conference on Neural Information Processing Systems, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 234, + 507, + 248 + ], + "spans": [ + { + "bbox": [ + 109, + 234, + 507, + 248 + ], + "score": 1.0, + "content": "[67] Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 245, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 129, + 245, + 505, + 257 + ], + "score": 1.0, + "content": "Score-based generative modeling through stochastic differential equations. In International Conference", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 131, + 257, + 262, + 266 + ], + "spans": [ + { + "bbox": [ + 131, + 257, + 262, + 266 + ], + "score": 1.0, + "content": "on Learning Representations, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 273, + 507, + 285 + ], + "spans": [ + { + "bbox": [ + 110, + 273, + 507, + 285 + ], + "score": 1.0, + "content": "[68] Songyou Peng, Chiyu \"Max\" Jiang, Yiyi Liao, Michael Niemeyer, Marc Pollefeys, and Andreas Geiger.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 284, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 130, + 284, + 505, + 296 + ], + "score": 1.0, + "content": "Shape as points: A differentiable poisson solver. In Advances in Neural Information Processing Systems", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 130, + 294, + 196, + 304 + ], + "spans": [ + { + "bbox": [ + 130, + 294, + 196, + 304 + ], + "score": 1.0, + "content": "(NeurIPS), 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 312, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 110, + 312, + 506, + 324 + ], + "score": 1.0, + "content": "[69] Diederik P Kingma, Tim Salimans, Ben Poole, and Jonathan Ho. Variational diffusion models. In", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 321, + 345, + 334 + ], + "spans": [ + { + "bbox": [ + 129, + 321, + 345, + 334 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 339, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 109, + 339, + 506, + 353 + ], + "score": 1.0, + "content": "[70] Diederik P Kingma and Max Welling. Auto-encoding variational bayes. In The International Conference", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 131, + 352, + 263, + 361 + ], + "spans": [ + { + "bbox": [ + 131, + 352, + 263, + 361 + ], + "score": 1.0, + "content": "on Learning Representations, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 367, + 507, + 382 + ], + "spans": [ + { + "bbox": [ + 109, + 367, + 507, + 382 + ], + "score": 1.0, + "content": "[71] Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and ap-", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 379, + 506, + 391 + ], + "spans": [ + { + "bbox": [ + 130, + 379, + 506, + 391 + ], + "score": 1.0, + "content": "proximate inference in deep generative models. In International Conference on Machine Learning,", + "type": "text", + "cross_page": true + } + ], + "index": 23 + }, + { + "bbox": [ + 130, + 388, + 155, + 401 + ], + "spans": [ + { + "bbox": [ + 130, + 388, + 155, + 401 + ], + "score": 1.0, + "content": "2014.", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 407, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 110, + 407, + 506, + 419 + ], + "score": 1.0, + "content": "[72] Jakub Tomczak and Max Welling. Vae with a vampprior. In International Conference on Artificial", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 417, + 319, + 429 + ], + "spans": [ + { + "bbox": [ + 130, + 417, + 319, + 429 + ], + "score": 1.0, + "content": "Intelligence and Statistics, pages 1214–1223, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 108, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "[73] Hiroshi Takahashi, Tomoharu Iwata, Yuki Yamanaka, Masanori Yamada, and Satoshi Yagi. Variational", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 443, + 507, + 459 + ], + "spans": [ + { + "bbox": [ + 129, + 443, + 507, + 459 + ], + "score": 1.0, + "content": "autoencoder with implicit optimal priors. Proceedings of the AAAI Conference on Artificial Intelligence,", + "type": "text", + "cross_page": true + } + ], + "index": 28 + }, + { + "bbox": [ + 130, + 455, + 240, + 467 + ], + "spans": [ + { + "bbox": [ + 130, + 455, + 240, + 467 + ], + "score": 1.0, + "content": "33(01):5066–5073, Jul. 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 472, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 109, + 472, + 506, + 486 + ], + "score": 1.0, + "content": "[74] Matthias Bauer and Andriy Mnih. Resampled priors for variational autoencoders. In Kamalika Chaudhuri", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 129, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "and Masashi Sugiyama, editors, Proceedings of the Twenty-Second International Conference on Artificial", + "type": "text", + "cross_page": true + } + ], + "index": 31 + }, + { + "bbox": [ + 129, + 492, + 507, + 507 + ], + "spans": [ + { + "bbox": [ + 129, + 492, + 507, + 507 + ], + "score": 1.0, + "content": "Intelligence and Statistics, volume 89 of Proceedings of Machine Learning Research, pages 66–75.", + "type": "text", + "cross_page": true + } + ], + "index": 32 + }, + { + "bbox": [ + 130, + 503, + 223, + 515 + ], + "spans": [ + { + "bbox": [ + 130, + 503, + 223, + 515 + ], + "score": 1.0, + "content": "PMLR, 16–18 Apr 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 520, + 507, + 535 + ], + "spans": [ + { + "bbox": [ + 109, + 520, + 507, + 535 + ], + "score": 1.0, + "content": "[75] Arash Vahdat and Jan Kautz. NVAE: A deep hierarchical variational autoencoder. In Advances in Neural", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 532, + 274, + 544 + ], + "spans": [ + { + "bbox": [ + 130, + 532, + 274, + 544 + ], + "score": 1.0, + "content": "Information Processing Systems, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 550, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 110, + 550, + 506, + 562 + ], + "score": 1.0, + "content": "[76] Jyoti Aneja, Alexander Schwing, Jan Kautz, and Arash Vahdat. NCP-VAE: Variational autoencoders with", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 560, + 445, + 572 + ], + "spans": [ + { + "bbox": [ + 130, + 560, + 445, + 572 + ], + "score": 1.0, + "content": "noise contrastive priors. In Advances in Neural Information Processing Systems, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 110, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "[77] Abhishek Sinha, Jiaming Song, Chenlin Meng, and Stefano Ermon. D2c: Diffusion-denoising models for", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 588, + 475, + 600 + ], + "spans": [ + { + "bbox": [ + 130, + 588, + 475, + 600 + ], + "score": 1.0, + "content": "few-shot conditional generation. In Advances in Neural Information Processing Systems, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 606, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 109, + 606, + 506, + 620 + ], + "score": 1.0, + "content": "[78] Mihaela Rosca, Balaji Lakshminarayanan, and Shakir Mohamed. Distribution matching in variational", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 617, + 316, + 629 + ], + "spans": [ + { + "bbox": [ + 130, + 617, + 316, + 629 + ], + "score": 1.0, + "content": "inference. arXiv preprint arXiv:1802.06847, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 635, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 110, + 635, + 505, + 647 + ], + "score": 1.0, + "content": "[79] Matthew D Hoffman and Matthew J Johnson. Elbo surgery: yet another way to carve up the variational", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 644, + 498, + 658 + ], + "spans": [ + { + "bbox": [ + 129, + 644, + 498, + 658 + ], + "score": 1.0, + "content": "evidence lower bound. In Workshop in Advances in Approximate Bayesian Inference, NeurIPS, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 43 + }, + { + "bbox": [ + 109, + 662, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 109, + 662, + 506, + 676 + ], + "score": 1.0, + "content": "[80] Zhijian Liu, Haotian Tang, Yujun Lin, and Song Han. Point-voxel cnn for efficient 3d deep learning. In", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 673, + 345, + 686 + ], + "spans": [ + { + "bbox": [ + 129, + 673, + 345, + 686 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 691, + 506, + 705 + ], + "spans": [ + { + "bbox": [ + 109, + 691, + 506, + 705 + ], + "score": 1.0, + "content": "[81] Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 701, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 130, + 701, + 506, + 713 + ], + "score": 1.0, + "content": "3d classification and segmentation. In IEEE Conference on Computer Vision and Pattern Recognition", + "type": "text", + "cross_page": true + } + ], + "index": 47 + }, + { + "bbox": [ + 129, + 711, + 187, + 723 + ], + "spans": [ + { + "bbox": [ + 129, + 711, + 187, + 723 + ], + "score": 1.0, + "content": "(CVPR), 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 72, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 108, + 72, + 507, + 87 + ], + "score": 1.0, + "content": "[82] Charles R Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 83, + 459, + 96 + ], + "spans": [ + { + "bbox": [ + 128, + 83, + 459, + 96 + ], + "score": 1.0, + "content": "point sets in a metric space. In Advances in Neural Information Processing Systems, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 101, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 109, + 101, + 506, + 115 + ], + "score": 1.0, + "content": "[83] Kaiming He, X. Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 112, + 435, + 124 + ], + "spans": [ + { + "bbox": [ + 130, + 112, + 435, + 124 + ], + "score": 1.0, + "content": "2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 131, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 110, + 131, + 505, + 142 + ], + "score": 1.0, + "content": "[84] Yuxin Wu and Kaiming He. Group normalization. In Proceedings of the European Conference on", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 131, + 141, + 249, + 151 + ], + "spans": [ + { + "bbox": [ + 131, + 141, + 249, + 151 + ], + "score": 1.0, + "content": "Computer Vision (ECCV), 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 158, + 507, + 172 + ], + "spans": [ + { + "bbox": [ + 109, + 158, + 507, + 172 + ], + "score": 1.0, + "content": "[85] Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon.", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 167, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 129, + 167, + 506, + 182 + ], + "score": 1.0, + "content": "SDEdit: Guided image synthesis and editing with stochastic differential equations. In International", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 178, + 305, + 190 + ], + "spans": [ + { + "bbox": [ + 130, + 178, + 305, + 190 + ], + "score": 1.0, + "content": "Conference on Learning Representations, 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 196, + 507, + 209 + ], + "spans": [ + { + "bbox": [ + 109, + 196, + 507, + 209 + ], + "score": 1.0, + "content": "[86] Nanxin Chen, Yu Zhang, Heiga Zen, Ron J Weiss, Mohammad Norouzi, and William Chan. Wavegrad:", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 130, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "Estimating gradients for waveform generation. In International Conference on Learning Representations", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 129, + 217, + 185, + 229 + ], + "spans": [ + { + "bbox": [ + 129, + 217, + 185, + 229 + ], + "score": 1.0, + "content": "(ICLR), 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 235, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 110, + 235, + 505, + 247 + ], + "score": 1.0, + "content": "[87] Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao, and Bryan Catanzaro. DiffWave: A Versatile Diffusion", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 245, + 467, + 257 + ], + "spans": [ + { + "bbox": [ + 129, + 245, + 467, + 257 + ], + "score": 1.0, + "content": "Model for Audio Synthesis. In International Conference on Learning Representations, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 262, + 507, + 276 + ], + "spans": [ + { + "bbox": [ + 108, + 262, + 507, + 276 + ], + "score": 1.0, + "content": "[88] Myeonghun Jeong, Hyeongju Kim, Sung Jun Cheon, Byoung Jin Choi, and Nam Soo Kim. Diff-tts: A", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 273, + 442, + 285 + ], + "spans": [ + { + "bbox": [ + 130, + 273, + 442, + 285 + ], + "score": 1.0, + "content": "denoising diffusion model for text-to-speech. arXiv preprint arXiv:2104.01409, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 292, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 110, + 292, + 505, + 304 + ], + "score": 1.0, + "content": "[89] Nanxin Chen, Yu Zhang, Heiga Zen, Ron J. Weiss, Mohammad Norouzi, Najim Dehak, and William", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 300, + 507, + 315 + ], + "spans": [ + { + "bbox": [ + 128, + 300, + 507, + 315 + ], + "score": 1.0, + "content": "Chan. Wavegrad 2: Iterative refinement for text-to-speech synthesis. arXiv preprint arXiv:2106.09660,", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 130, + 312, + 155, + 322 + ], + "spans": [ + { + "bbox": [ + 130, + 312, + 155, + 322 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 330, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 110, + 330, + 506, + 342 + ], + "score": 1.0, + "content": "[90] Vadim Popov, Ivan Vovk, Vladimir Gogoryan, Tasnima Sadekova, and Mikhail Kudinov. Grad-tts: A", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 340, + 507, + 352 + ], + "spans": [ + { + "bbox": [ + 130, + 340, + 507, + 352 + ], + "score": 1.0, + "content": "diffusion probabilistic model for text-to-speech. In International Conference on Machine Learning, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 20 + }, + { + "bbox": [ + 110, + 359, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 110, + 359, + 505, + 370 + ], + "score": 1.0, + "content": "[91] Songxiang Liu, Dan Su, and Dong Yu. Diffgan-tts: High-fidelity and efficient text-to-speech with", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 369, + 371, + 381 + ], + "spans": [ + { + "bbox": [ + 130, + 369, + 371, + 381 + ], + "score": 1.0, + "content": "denoising diffusion gans. arXiv preprint arXiv:2201.11972, 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 387, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 110, + 387, + 506, + 399 + ], + "score": 1.0, + "content": "[92] Gautam Mittal, Jesse Engel, Curtis Hawthorne, and Ian Simon. Symbolic music generation with diffusion", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 396, + 507, + 411 + ], + "spans": [ + { + "bbox": [ + 128, + 396, + 507, + 411 + ], + "score": 1.0, + "content": "models. In Proceedings of the 22nd International Society for Music Information Retrieval Conference,", + "type": "text", + "cross_page": true + } + ], + "index": 24 + }, + { + "bbox": [ + 129, + 406, + 156, + 419 + ], + "spans": [ + { + "bbox": [ + 129, + 406, + 156, + 419 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 425, + 507, + 437 + ], + "spans": [ + { + "bbox": [ + 110, + 425, + 507, + 437 + ], + "score": 1.0, + "content": "[93] Kushagra Pandey, Avideep Mukherjee, Piyush Rai, and Abhishek Kumar. Diffusevae: Efficient, con-", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 434, + 507, + 448 + ], + "spans": [ + { + "bbox": [ + 129, + 434, + 507, + 448 + ], + "score": 1.0, + "content": "trollable and high-fidelity generation from low-dimensional latents. arXiv preprint arXiv:2201.00308,", + "type": "text", + "cross_page": true + } + ], + "index": 27 + }, + { + "bbox": [ + 129, + 444, + 155, + 458 + ], + "spans": [ + { + "bbox": [ + 129, + 444, + 155, + 458 + ], + "score": 1.0, + "content": "2022.", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 462, + 507, + 476 + ], + "spans": [ + { + "bbox": [ + 108, + 462, + 507, + 476 + ], + "score": 1.0, + "content": "[94] Danilo Jimenez Rezende and Shakir Mohamed. Variational inference with normalizing flows. In", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 474, + 329, + 486 + ], + "spans": [ + { + "bbox": [ + 130, + 474, + 329, + 486 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 491, + 507, + 505 + ], + "spans": [ + { + "bbox": [ + 109, + 491, + 507, + 505 + ], + "score": 1.0, + "content": "[95] Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real NVP. In", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 502, + 377, + 514 + ], + "spans": [ + { + "bbox": [ + 129, + 502, + 377, + 514 + ], + "score": 1.0, + "content": "International Conference on Learning Representations ICLR, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 110, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "[96] Ricky T. Q. Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 530, + 387, + 542 + ], + "spans": [ + { + "bbox": [ + 129, + 530, + 387, + 542 + ], + "score": 1.0, + "content": "equations. Advances in Neural Information Processing Systems, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 547, + 507, + 561 + ], + "spans": [ + { + "bbox": [ + 108, + 547, + 507, + 561 + ], + "score": 1.0, + "content": "[97] Will Grathwohl, Ricky T. Q. Chen, Jesse Bettencourt, Ilya Sutskever, and David Duvenaud. FFJORD:", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 559, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 129, + 559, + 505, + 570 + ], + "score": 1.0, + "content": "Free-form continuous dynamics for scalable reversible generative models. In International Conference", + "type": "text", + "cross_page": true + } + ], + "index": 36 + }, + { + "bbox": [ + 131, + 570, + 261, + 579 + ], + "spans": [ + { + "bbox": [ + 131, + 570, + 261, + 579 + ], + "score": 1.0, + "content": "on Learning Representations, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 587, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 110, + 587, + 506, + 599 + ], + "score": 1.0, + "content": "[98] Zhaoyang Lyu, Zhifeng Kong, Xudong XU, Liang Pan, and Dahua Lin. A conditional point diffusion-", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 130, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "refinement paradigm for 3d point cloud completion. In International Conference on Learning Representa-", + "type": "text", + "cross_page": true + } + ], + "index": 39 + }, + { + "bbox": [ + 129, + 607, + 204, + 619 + ], + "spans": [ + { + "bbox": [ + 129, + 607, + 204, + 619 + ], + "score": 1.0, + "content": "tions (ICLR), 2022.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 626, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 110, + 626, + 505, + 637 + ], + "score": 1.0, + "content": "[99] Thibault Groueix, Matthew Fisher, Vladimir G. Kim, Bryan Russell, and Mathieu Aubry. AtlasNet: A", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 633, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 128, + 633, + 506, + 648 + ], + "score": 1.0, + "content": "Papier-Mâché Approach to Learning 3D Surface Generation. In Proceedings IEEE Conf. on Computer", + "type": "text", + "cross_page": true + } + ], + "index": 42 + }, + { + "bbox": [ + 130, + 645, + 301, + 657 + ], + "spans": [ + { + "bbox": [ + 130, + 645, + 301, + 657 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition (CVPR), 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 663, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 506, + 675 + ], + "score": 1.0, + "content": "[100] Lars Mescheder, Michael Oechsle, Michael Niemeyer, Sebastian Nowozin, and Andreas Geiger. Oc-", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 673, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 130, + 673, + 505, + 685 + ], + "score": 1.0, + "content": "cupancy networks: Learning 3d reconstruction in function space. In Proceedings of the IEEE/CVF", + "type": "text", + "cross_page": true + } + ], + "index": 45 + }, + { + "bbox": [ + 129, + 682, + 431, + 696 + ], + "spans": [ + { + "bbox": [ + 129, + 682, + 431, + 696 + ], + "score": 1.0, + "content": "Conference on Computer Vision and Pattern Recognition, pages 4460–4470, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 702, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 505, + 713 + ], + "score": 1.0, + "content": "[101] Songyou Peng, Michael Niemeyer, Lars Mescheder, and Andreas Geiger Marc Pollefeys. Convolutional", + "type": "text", + "cross_page": true + } + ], + "index": 47, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 711, + 431, + 724 + ], + "spans": [ + { + "bbox": [ + 129, + 711, + 431, + 724 + ], + "score": 1.0, + "content": "occupancy networks. In European Conference on Computer Vision (ECCV), 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_end_line": true + } + ], + "index": 24.5, + "bbox_fs": [ + 106, + 70, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 109, + 42, + 507, + 729 + ], + "lines": [ + { + "bbox": [ + 110, + 73, + 507, + 85 + ], + "spans": [ + { + "bbox": [ + 110, + 73, + 507, + 85 + ], + "score": 1.0, + "content": "[21] Zhiqin Chen, Vladimir G. Kim, Matthew Fisher, Noam Aigerman, Hao Zhang, and Siddhartha Chaudhuri.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 129, + 82, + 507, + 97 + ], + "spans": [ + { + "bbox": [ + 129, + 82, + 507, + 97 + ], + "score": 1.0, + "content": "Decor-gan: 3d shape detailization by conditional refinement. In Proceedings of IEEE Conference on", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 129, + 93, + 338, + 106 + ], + "spans": [ + { + "bbox": [ + 129, + 93, + 338, + 106 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 110, + 112, + 507, + 126 + ], + "spans": [ + { + "bbox": [ + 110, + 112, + 507, + 126 + ], + "score": 1.0, + "content": "[22] Cheng Wen, Baosheng Yu, and Dacheng Tao. Learning progressive point embeddings for 3d point cloud", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 129, + 121, + 507, + 135 + ], + "spans": [ + { + "bbox": [ + 129, + 121, + 507, + 135 + ], + "score": 1.0, + "content": "generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 128, + 132, + 232, + 144 + ], + "spans": [ + { + "bbox": [ + 128, + 132, + 232, + 144 + ], + "score": 1.0, + "content": "pages 10266–10275, 2021.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 110, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 110, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "[23] Zekun Hao, Arun Mallya, Serge Belongie, and Ming-Yu Liu. GANcraft: Unsupervised 3D Neural", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 130, + 160, + 307, + 173 + ], + "spans": [ + { + "bbox": [ + 130, + 160, + 307, + 173 + ], + "score": 1.0, + "content": "Rendering of Minecraft Worlds. In ICCV, 2021.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 110, + 180, + 504, + 192 + ], + "spans": [ + { + "bbox": [ + 110, + 180, + 504, + 192 + ], + "score": 1.0, + "content": "[24] Or Litany, Alex Bronstein, Michael Bronstein, and Ameesh Makadia. Deformable shape completion with", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 129, + 189, + 308, + 203 + ], + "spans": [ + { + "bbox": [ + 129, + 189, + 308, + 203 + ], + "score": 1.0, + "content": "graph convolutional autoencoders. CVPR, 2018.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 109, + 208, + 507, + 222 + ], + "spans": [ + { + "bbox": [ + 109, + 208, + 507, + 222 + ], + "score": 1.0, + "content": "[25] Qingyang Tan, Lin Gao, Yu-Kun Lai, and Shihong Xia. Variational autoencoders for deforming 3d", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 128, + 217, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 128, + 217, + 506, + 233 + ], + "score": 1.0, + "content": "mesh models. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 129, + 229, + 199, + 241 + ], + "spans": [ + { + "bbox": [ + 129, + 229, + 199, + 241 + ], + "score": 1.0, + "content": "5841–5850, 2018.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 110, + 248, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 110, + 248, + 506, + 260 + ], + "score": 1.0, + "content": "[26] Kaichun Mo, Paul Guerrero, Li Yi, Hao Su, Peter Wonka, Niloy Mitra, and Leonidas J Guibas. Struc-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 129, + 256, + 501, + 271 + ], + "spans": [ + { + "bbox": [ + 129, + 256, + 501, + 271 + ], + "score": 1.0, + "content": "turenet: Hierarchical graph networks for 3d shape generation. arXiv preprint arXiv:1908.00575, 2019.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 110, + 276, + 507, + 289 + ], + "spans": [ + { + "bbox": [ + 110, + 276, + 507, + 289 + ], + "score": 1.0, + "content": "[27] Lin Gao, Jie Yang, Tong Wu, Yu-Jie Yuan, Hongbo Fu, Yu-Kun Lai, and Hao(Richard) Zhang. SDM-NET:", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 129, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 129, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "Deep generative network for structured deformable mesh. ACM Transactions on Graphics (Proceedings", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 129, + 295, + 350, + 309 + ], + "spans": [ + { + "bbox": [ + 129, + 295, + 350, + 309 + ], + "score": 1.0, + "content": "of ACM SIGGRAPH Asia 2019), 38(6):243:1–243:15, 2019.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 110, + 316, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 110, + 316, + 505, + 328 + ], + "score": 1.0, + "content": "[28] Lin Gao, Tong Wu, Yu-Jie Yuan, Ming-Xian Lin, Yu-Kun Lai, and Hao Zhang. Tm-net: Deep generative", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 130, + 325, + 486, + 338 + ], + "spans": [ + { + "bbox": [ + 130, + 325, + 486, + 338 + ], + "score": 1.0, + "content": "networks for textured meshes. ACM Transactions on Graphics (TOG), 40(6):263:1–263:15, 2021.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 110, + 344, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 110, + 344, + 506, + 356 + ], + "score": 1.0, + "content": "[29] Jinwoo Kim, Jaehoon Yoo, Juho Lee, and Seunghoon Hong. Setvae: Learning hierarchical composition", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 129, + 353, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 129, + 353, + 505, + 367 + ], + "score": 1.0, + "content": "for generative modeling of set-structured data. In Proceedings of the IEEE/CVF Conference on Computer", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 130, + 365, + 396, + 376 + ], + "spans": [ + { + "bbox": [ + 130, + 365, + 396, + 376 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition (CVPR), pages 15059–15068, June 2021.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 110, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 110, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "[30] Paritosh Mittal, Yen-Chi Cheng, Maneesh Singh, and Shubham Tulsiani. AutoSDF: Shape priors for 3d", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 130, + 393, + 346, + 405 + ], + "spans": [ + { + "bbox": [ + 130, + 393, + 346, + 405 + ], + "score": 1.0, + "content": "completion, reconstruction and generation. In CVPR, 2022.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 110, + 412, + 507, + 424 + ], + "spans": [ + { + "bbox": [ + 110, + 412, + 507, + 424 + ], + "score": 1.0, + "content": "[31] Guandao Yang, Xun Huang, Zekun Hao, Ming-Yu Liu, Serge Belongie, and Bharath Hariharan. PointFlow:", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 130, + 422, + 414, + 434 + ], + "spans": [ + { + "bbox": [ + 130, + 422, + 414, + 434 + ], + "score": 1.0, + "content": "3D point cloud generation with continuous normalizing flows. In ICCV, 2019.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 109, + 440, + 508, + 453 + ], + "spans": [ + { + "bbox": [ + 109, + 440, + 508, + 453 + ], + "score": 1.0, + "content": "[32] Hyeongju Kim, Hyeonseung Lee, Woo Hyun Kang, Joun Yeop Lee, and Nam Soo Kim. SoftFlow:", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 129, + 451, + 417, + 463 + ], + "spans": [ + { + "bbox": [ + 129, + 451, + 417, + 463 + ], + "score": 1.0, + "content": "Probabilistic framework for normalizing flow on manifolds. In NeurIPS, 2020.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 110, + 468, + 507, + 483 + ], + "spans": [ + { + "bbox": [ + 110, + 468, + 507, + 483 + ], + "score": 1.0, + "content": "[33] Roman Klokov, Edmond Boyer, and Jakob Verbeek. Discrete point flow networks for efficient point cloud", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 128, + 480, + 235, + 492 + ], + "spans": [ + { + "bbox": [ + 128, + 480, + 235, + 492 + ], + "score": 1.0, + "content": "generation. In ECCV, 2020.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 108, + 496, + 507, + 513 + ], + "spans": [ + { + "bbox": [ + 108, + 496, + 507, + 513 + ], + "score": 1.0, + "content": "[34] Aditya Sanghi, Hang Chu, Joseph G Lambourne, Ye Wang, Chin-Yi Cheng, and Marco Fumero. Clip-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 128, + 507, + 464, + 521 + ], + "spans": [ + { + "bbox": [ + 128, + 507, + 464, + 521 + ], + "score": 1.0, + "content": "forge: Towards zero-shot text-to-shape generation. arXiv preprint arXiv:2110.02624, 2021.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 110, + 526, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 110, + 526, + 506, + 541 + ], + "score": 1.0, + "content": "[35] Yongbin Sun, Yue Wang, Ziwei Liu, Joshua E Siegel, and Sanjay E Sarma. Pointgrow: Autoregressively", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 129, + 535, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 129, + 535, + 506, + 551 + ], + "score": 1.0, + "content": "learned point cloud generation with self-attention. In Winter Conference on Applications of Computer", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 130, + 547, + 181, + 559 + ], + "spans": [ + { + "bbox": [ + 130, + 547, + 181, + 559 + ], + "score": 1.0, + "content": "Vision, 2020.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 108, + 564, + 507, + 579 + ], + "spans": [ + { + "bbox": [ + 108, + 564, + 507, + 579 + ], + "score": 1.0, + "content": "[36] Charlie Nash, Yaroslav Ganin, S. M. Ali Eslami, and Peter W. Battaglia. Polygen: An autoregressive", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 128, + 575, + 297, + 589 + ], + "spans": [ + { + "bbox": [ + 128, + 575, + 297, + 589 + ], + "score": 1.0, + "content": "generative model of 3d meshes. ICML, 2020.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 110, + 594, + 507, + 608 + ], + "spans": [ + { + "bbox": [ + 110, + 594, + 507, + 608 + ], + "score": 1.0, + "content": "[37] Wei-Jan Ko, Hui-Yu Huang, Yu-Liang Kuo, Chen-Yi Chiu, Li-Heng Wang, and Wei-Chen Chiu. Rpg:", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 129, + 605, + 433, + 617 + ], + "spans": [ + { + "bbox": [ + 129, + 605, + 433, + 617 + ], + "score": 1.0, + "content": "Learning recursive point cloud generation. arXiv preprint arXiv:2105.14322, 2021.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 110, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 110, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "[38] Moritz Ibing, Gregor Kobsik, and Leif Kobbelt. Octree transformer: Autoregressive 3d shape generation", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 130, + 634, + 420, + 646 + ], + "spans": [ + { + "bbox": [ + 130, + 634, + 420, + 646 + ], + "score": 1.0, + "content": "on hierarchically structured sequences. arXiv preprint arXiv:2111.12480, 2021.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 109, + 651, + 507, + 666 + ], + "spans": [ + { + "bbox": [ + 109, + 651, + 507, + 666 + ], + "score": 1.0, + "content": "[39] Jianwen Xie, Zilong Zheng, Ruiqi Gao, Wenguan Wang, Zhu Song-Chun, and Ying Nian Wu. Learning", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 130, + 663, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 130, + 663, + 506, + 675 + ], + "score": 1.0, + "content": "descriptor networks for 3d shape synthesis and analysis. In The IEEE Conference on Computer Vision", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 129, + 672, + 276, + 684 + ], + "spans": [ + { + "bbox": [ + 129, + 672, + 276, + 684 + ], + "score": 1.0, + "content": "and Pattern Recognition (CVPR), 2018.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 110, + 690, + 507, + 705 + ], + "spans": [ + { + "bbox": [ + 110, + 690, + 507, + 705 + ], + "score": 1.0, + "content": "[40] Jianwen Xie, Yifei Xu, Zilong Zheng, Ruiqi Gao, Wenguan Wang, Zhu Song-Chun, and Ying Nian Wu.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 130, + 702, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 130, + 702, + 506, + 714 + ], + "score": 1.0, + "content": "Generative pointnet: Deep energy-based learning on unordered point sets for 3d generation, reconstruction", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 129, + 712, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 129, + 712, + 506, + 723 + ], + "score": 1.0, + "content": "and classification. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 23.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 109, + 42, + 507, + 729 + ], + "lines": [], + "index": 23.5, + "bbox_fs": [ + 108, + 73, + 508, + 723 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 69, + 507, + 727 + ], + "lines": [ + { + "bbox": [ + 111, + 73, + 505, + 84 + ], + "spans": [ + { + "bbox": [ + 111, + 73, + 505, + 84 + ], + "score": 1.0, + "content": "[41] Tianchang Shen, Jun Gao, Kangxue Yin, Ming-Yu Liu, and Sanja Fidler. Deep marching tetrahedra:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 128, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 128, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "a hybrid representation for high-resolution 3d shape synthesis. In Advances in Neural Information", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 128, + 93, + 268, + 106 + ], + "spans": [ + { + "bbox": [ + 128, + 93, + 268, + 106 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), 2021.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 112, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 111, + 112, + 505, + 124 + ], + "score": 1.0, + "content": "[42] Kangxue Yin, Jun Gao, Maria Shugrina, Sameh Khamis, and Sanja Fidler. 3dstylenet: Creating 3d shapes", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 128, + 121, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 128, + 121, + 505, + 135 + ], + "score": 1.0, + "content": "with geometric and texture style variations. In Proceedings of International Conference on Computer", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 131, + 132, + 209, + 144 + ], + "spans": [ + { + "bbox": [ + 131, + 132, + 209, + 144 + ], + "score": 1.0, + "content": "Vision (ICCV), 2021.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 110, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 110, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "[43] Dongsu Zhang, Changwoon Choi, Jeonghwan Kim, and Young Min Kim. Learning to generate 3d shapes", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 130, + 160, + 488, + 172 + ], + "spans": [ + { + "bbox": [ + 130, + 160, + 488, + 172 + ], + "score": 1.0, + "content": "with generative cellular automata. In International Conference on Learning Representations, 2021.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 109, + 178, + 506, + 191 + ], + "spans": [ + { + "bbox": [ + 109, + 178, + 506, + 191 + ], + "score": 1.0, + "content": "[44] Chen Chao, Zhizhong Han, Yu-Shen Liu, and Matthias Zwicker. Unsupervised learning of fine structure", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 128, + 187, + 489, + 202 + ], + "spans": [ + { + "bbox": [ + 128, + 187, + 489, + 202 + ], + "score": 1.0, + "content": "generation for 3d point clouds by 2d projection matching. arXiv preprint arXiv:2108.03746, 2021.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 110, + 208, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 110, + 208, + 505, + 219 + ], + "score": 1.0, + "content": "[45] Ruojin Cai, Guandao Yang, Hadar Averbuch-Elor, Zekun Hao, Serge Belongie, Noah Snavely, and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 128, + 214, + 506, + 232 + ], + "spans": [ + { + "bbox": [ + 128, + 214, + 506, + 232 + ], + "score": 1.0, + "content": "Bharath Hariharan. Learning gradient fields for shape generation. In Proceedings of the European", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 130, + 227, + 305, + 239 + ], + "spans": [ + { + "bbox": [ + 130, + 227, + 305, + 239 + ], + "score": 1.0, + "content": "Conference on Computer Vision (ECCV), 2020.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 110, + 245, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 110, + 245, + 506, + 257 + ], + "score": 1.0, + "content": "[46] Linqi Zhou, Yilun Du, and Jiajun Wu. 3d shape generation and completion through point-voxel diffusion.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 130, + 255, + 470, + 267 + ], + "spans": [ + { + "bbox": [ + 130, + 255, + 470, + 267 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 109, + 272, + 507, + 287 + ], + "spans": [ + { + "bbox": [ + 109, + 272, + 507, + 287 + ], + "score": 1.0, + "content": "[47] Shitong Luo and Wei Hu. Diffusion probabilistic models for 3d point cloud generation. In Proceedings of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 130, + 284, + 448, + 296 + ], + "spans": [ + { + "bbox": [ + 130, + 284, + 448, + 296 + ], + "score": 1.0, + "content": "the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 111, + 303, + 505, + 314 + ], + "spans": [ + { + "bbox": [ + 111, + 303, + 505, + 314 + ], + "score": 1.0, + "content": "[48] Yu Deng, Jiaolong Yang, and Xin Tong. Deformed implicit field: Modeling 3d shapes with learned dense", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 128, + 312, + 402, + 324 + ], + "spans": [ + { + "bbox": [ + 128, + 312, + 402, + 324 + ], + "score": 1.0, + "content": "correspondence. In IEEE Computer Vision and Pattern Recognition, 2021.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 110, + 331, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 110, + 331, + 505, + 342 + ], + "score": 1.0, + "content": "[49] Oscar Michel, Roi Bar-On, Richard Liu, Sagie Benaim, and Rana Hanocka. Text2mesh: Text-driven", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 128, + 340, + 387, + 353 + ], + "spans": [ + { + "bbox": [ + 128, + 340, + 387, + 353 + ], + "score": 1.0, + "content": "neural stylization for meshes. arXiv preprint arXiv:2112.03221, 2021.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 109, + 358, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 109, + 358, + 506, + 372 + ], + "score": 1.0, + "content": "[50] Ajay Jain, Ben Mildenhall, Jonathan T. Barron, Pieter Abbeel, and Ben Poole. Zero-shot text-guided", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 128, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 128, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "object generation with dream fields. In IEEE Conference on Computer Vision and Pattern Recognition", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 129, + 379, + 186, + 391 + ], + "spans": [ + { + "bbox": [ + 129, + 379, + 186, + 391 + ], + "score": 1.0, + "content": "(CVPR), 2022.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 111, + 398, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 111, + 398, + 505, + 410 + ], + "score": 1.0, + "content": "[51] Nasir Khalid, Tianhao Xie, Eugene Belilovsky, and Tiberiu Popa. Text to mesh without 3d supervision", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 128, + 407, + 365, + 420 + ], + "spans": [ + { + "bbox": [ + 128, + 407, + 365, + 420 + ], + "score": 1.0, + "content": "using limit subdivision. arXiv preprint arXiv:2203.13333, 2022.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 109, + 425, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 109, + 425, + 506, + 439 + ], + "score": 1.0, + "content": "[52] Le Hui, Rui Xu, Jin Xie, Jianjun Qian, and Jian Yang. Progressive point cloud deconvolution generation", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 131, + 437, + 226, + 447 + ], + "spans": [ + { + "bbox": [ + 131, + 437, + 226, + 447 + ], + "score": 1.0, + "content": "network. In ECCV, 2020.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 110, + 454, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 110, + 454, + 506, + 468 + ], + "score": 1.0, + "content": "[53] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 130, + 465, + 301, + 476 + ], + "spans": [ + { + "bbox": [ + 130, + 465, + 301, + 476 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, 2020.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 109, + 482, + 507, + 496 + ], + "spans": [ + { + "bbox": [ + 109, + 482, + 507, + 496 + ], + "score": 1.0, + "content": "[54] Jonathan Ho, Chitwan Saharia, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 129, + 492, + 503, + 506 + ], + "spans": [ + { + "bbox": [ + 129, + 492, + 503, + 506 + ], + "score": 1.0, + "content": "Cascaded diffusion models for high fidelity image generation. arXiv preprint arXiv:2106.15282, 2021.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 109, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 109, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "[55] Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. In", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 130, + 522, + 327, + 533 + ], + "spans": [ + { + "bbox": [ + 130, + 522, + 327, + 533 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, 2021.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 111, + 540, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 111, + 540, + 505, + 551 + ], + "score": 1.0, + "content": "[56] Prafulla Dhariwal and Alexander Quinn Nichol. Diffusion models beat GANs on image synthesis. In", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 129, + 550, + 346, + 561 + ], + "spans": [ + { + "bbox": [ + 129, + 550, + 346, + 561 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2021.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 110, + 568, + 505, + 580 + ], + "spans": [ + { + "bbox": [ + 110, + 568, + 505, + 580 + ], + "score": 1.0, + "content": "[57] Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 128, + 578, + 444, + 591 + ], + "spans": [ + { + "bbox": [ + 128, + 578, + 444, + 591 + ], + "score": 1.0, + "content": "image synthesis with latent diffusion models. arXiv preprint arXiv:2112.10752, 2021.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 110, + 596, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 110, + 596, + 506, + 609 + ], + "score": 1.0, + "content": "[58] Arash Vahdat, Karsten Kreis, and Jan Kautz. Score-based generative modeling in latent space. In", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 128, + 606, + 346, + 619 + ], + "spans": [ + { + "bbox": [ + 128, + 606, + 346, + 619 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2021.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 110, + 624, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 110, + 624, + 506, + 637 + ], + "score": 1.0, + "content": "[59] Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 128, + 632, + 507, + 649 + ], + "spans": [ + { + "bbox": [ + 128, + 632, + 507, + 649 + ], + "score": 1.0, + "content": "Sutskever, and Mark Chen. Glide: Towards photorealistic image generation and editing with text-guided", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 131, + 645, + 344, + 657 + ], + "spans": [ + { + "bbox": [ + 131, + 645, + 344, + 657 + ], + "score": 1.0, + "content": "diffusion models. arXiv preprint arXiv:2112.10741, 2021.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 111, + 664, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 111, + 664, + 505, + 675 + ], + "score": 1.0, + "content": "[60] Konpat Preechakul, Nattanat Chatthee, Suttisak Wizadwongsa, and Supasorn Suwajanakorn. Diffusion", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 128, + 672, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 128, + 672, + 505, + 686 + ], + "score": 1.0, + "content": "autoencoders: Toward a meaningful and decodable representation. In IEEE Conference on Computer", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 130, + 682, + 300, + 696 + ], + "spans": [ + { + "bbox": [ + 130, + 682, + 300, + 696 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition (CVPR), 2022.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 109, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 109, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "[61] Tim Dockhorn, Arash Vahdat, and Karsten Kreis. Score-based generative modeling with critically-damped", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 130, + 712, + 462, + 724 + ], + "spans": [ + { + "bbox": [ + 130, + 712, + 462, + 724 + ], + "score": 1.0, + "content": "langevin diffusion. In International Conference on Learning Representations (ICLR), 2022.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 23.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 108, + 69, + 507, + 727 + ], + "lines": [], + "index": 23.5, + "bbox_fs": [ + 109, + 73, + 507, + 724 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 109, + 40, + 507, + 728 + ], + "lines": [ + { + "bbox": [ + 109, + 73, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 109, + 73, + 507, + 87 + ], + "score": 1.0, + "content": "[62] Zhisheng Xiao, Karsten Kreis, and Arash Vahdat. Tackling the generative learning trilemma with", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 128, + 82, + 492, + 97 + ], + "spans": [ + { + "bbox": [ + 128, + 82, + 492, + 97 + ], + "score": 1.0, + "content": "denoising diffusion GANs. In International Conference on Learning Representations (ICLR), 2022.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 108, + 101, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 108, + 101, + 506, + 114 + ], + "score": 1.0, + "content": "[63] Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 128, + 111, + 405, + 125 + ], + "spans": [ + { + "bbox": [ + 128, + 111, + 405, + 125 + ], + "score": 1.0, + "content": "image generation with clip latents. arXiv preprint arXiv:2204.06125, 2022.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 109, + 129, + 507, + 143 + ], + "spans": [ + { + "bbox": [ + 109, + 129, + 507, + 143 + ], + "score": 1.0, + "content": "[64] Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed Kam-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 131, + 141, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 131, + 141, + 506, + 153 + ], + "score": 1.0, + "content": "yar Seyed Ghasemipour, Burcu Karagol Ayan, S. Sara Mahdavi, Rapha Gontijo Lopes, Tim Salimans,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 128, + 148, + 507, + 163 + ], + "spans": [ + { + "bbox": [ + 128, + 148, + 507, + 163 + ], + "score": 1.0, + "content": "Jonathan Ho, David J Fleet, and Mohammad Norouzi. Photorealistic text-to-image diffusion models with", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 130, + 160, + 388, + 172 + ], + "spans": [ + { + "bbox": [ + 130, + 160, + 388, + 172 + ], + "score": 1.0, + "content": "deep language understanding. arXiv preprint arXiv:2205.11487, 2022.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 110, + 178, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 110, + 178, + 505, + 190 + ], + "score": 1.0, + "content": "[65] Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 130, + 189, + 507, + 201 + ], + "spans": [ + { + "bbox": [ + 130, + 189, + 507, + 201 + ], + "score": 1.0, + "content": "learning using nonequilibrium thermodynamics. In International Conference on Machine Learning, 2015.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 110, + 207, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 110, + 207, + 506, + 219 + ], + "score": 1.0, + "content": "[66] Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. In", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 130, + 217, + 470, + 230 + ], + "spans": [ + { + "bbox": [ + 130, + 217, + 470, + 230 + ], + "score": 1.0, + "content": "Proceedings of the 33rd Annual Conference on Neural Information Processing Systems, 2019.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 109, + 234, + 507, + 248 + ], + "spans": [ + { + "bbox": [ + 109, + 234, + 507, + 248 + ], + "score": 1.0, + "content": "[67] Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 129, + 245, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 129, + 245, + 505, + 257 + ], + "score": 1.0, + "content": "Score-based generative modeling through stochastic differential equations. In International Conference", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 131, + 257, + 262, + 266 + ], + "spans": [ + { + "bbox": [ + 131, + 257, + 262, + 266 + ], + "score": 1.0, + "content": "on Learning Representations, 2021.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 110, + 273, + 507, + 285 + ], + "spans": [ + { + "bbox": [ + 110, + 273, + 507, + 285 + ], + "score": 1.0, + "content": "[68] Songyou Peng, Chiyu \"Max\" Jiang, Yiyi Liao, Michael Niemeyer, Marc Pollefeys, and Andreas Geiger.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 130, + 284, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 130, + 284, + 505, + 296 + ], + "score": 1.0, + "content": "Shape as points: A differentiable poisson solver. In Advances in Neural Information Processing Systems", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 130, + 294, + 196, + 304 + ], + "spans": [ + { + "bbox": [ + 130, + 294, + 196, + 304 + ], + "score": 1.0, + "content": "(NeurIPS), 2021.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 110, + 312, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 110, + 312, + 506, + 324 + ], + "score": 1.0, + "content": "[69] Diederik P Kingma, Tim Salimans, Ben Poole, and Jonathan Ho. Variational diffusion models. In", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 129, + 321, + 345, + 334 + ], + "spans": [ + { + "bbox": [ + 129, + 321, + 345, + 334 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2021.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 109, + 339, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 109, + 339, + 506, + 353 + ], + "score": 1.0, + "content": "[70] Diederik P Kingma and Max Welling. Auto-encoding variational bayes. In The International Conference", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 131, + 352, + 263, + 361 + ], + "spans": [ + { + "bbox": [ + 131, + 352, + 263, + 361 + ], + "score": 1.0, + "content": "on Learning Representations, 2014.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 109, + 367, + 507, + 382 + ], + "spans": [ + { + "bbox": [ + 109, + 367, + 507, + 382 + ], + "score": 1.0, + "content": "[71] Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and ap-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 130, + 379, + 506, + 391 + ], + "spans": [ + { + "bbox": [ + 130, + 379, + 506, + 391 + ], + "score": 1.0, + "content": "proximate inference in deep generative models. In International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 130, + 388, + 155, + 401 + ], + "spans": [ + { + "bbox": [ + 130, + 388, + 155, + 401 + ], + "score": 1.0, + "content": "2014.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 110, + 407, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 110, + 407, + 506, + 419 + ], + "score": 1.0, + "content": "[72] Jakub Tomczak and Max Welling. Vae with a vampprior. In International Conference on Artificial", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 130, + 417, + 319, + 429 + ], + "spans": [ + { + "bbox": [ + 130, + 417, + 319, + 429 + ], + "score": 1.0, + "content": "Intelligence and Statistics, pages 1214–1223, 2018.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 108, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 108, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "[73] Hiroshi Takahashi, Tomoharu Iwata, Yuki Yamanaka, Masanori Yamada, and Satoshi Yagi. Variational", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 129, + 443, + 507, + 459 + ], + "spans": [ + { + "bbox": [ + 129, + 443, + 507, + 459 + ], + "score": 1.0, + "content": "autoencoder with implicit optimal priors. Proceedings of the AAAI Conference on Artificial Intelligence,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 130, + 455, + 240, + 467 + ], + "spans": [ + { + "bbox": [ + 130, + 455, + 240, + 467 + ], + "score": 1.0, + "content": "33(01):5066–5073, Jul. 2019.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 109, + 472, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 109, + 472, + 506, + 486 + ], + "score": 1.0, + "content": "[74] Matthias Bauer and Andriy Mnih. Resampled priors for variational autoencoders. In Kamalika Chaudhuri", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 129, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 129, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "and Masashi Sugiyama, editors, Proceedings of the Twenty-Second International Conference on Artificial", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 129, + 492, + 507, + 507 + ], + "spans": [ + { + "bbox": [ + 129, + 492, + 507, + 507 + ], + "score": 1.0, + "content": "Intelligence and Statistics, volume 89 of Proceedings of Machine Learning Research, pages 66–75.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 130, + 503, + 223, + 515 + ], + "spans": [ + { + "bbox": [ + 130, + 503, + 223, + 515 + ], + "score": 1.0, + "content": "PMLR, 16–18 Apr 2019.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 109, + 520, + 507, + 535 + ], + "spans": [ + { + "bbox": [ + 109, + 520, + 507, + 535 + ], + "score": 1.0, + "content": "[75] Arash Vahdat and Jan Kautz. NVAE: A deep hierarchical variational autoencoder. In Advances in Neural", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 130, + 532, + 274, + 544 + ], + "spans": [ + { + "bbox": [ + 130, + 532, + 274, + 544 + ], + "score": 1.0, + "content": "Information Processing Systems, 2020.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 110, + 550, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 110, + 550, + 506, + 562 + ], + "score": 1.0, + "content": "[76] Jyoti Aneja, Alexander Schwing, Jan Kautz, and Arash Vahdat. NCP-VAE: Variational autoencoders with", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 130, + 560, + 445, + 572 + ], + "spans": [ + { + "bbox": [ + 130, + 560, + 445, + 572 + ], + "score": 1.0, + "content": "noise contrastive priors. In Advances in Neural Information Processing Systems, 2021.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 110, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 110, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "[77] Abhishek Sinha, Jiaming Song, Chenlin Meng, and Stefano Ermon. D2c: Diffusion-denoising models for", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 130, + 588, + 475, + 600 + ], + "spans": [ + { + "bbox": [ + 130, + 588, + 475, + 600 + ], + "score": 1.0, + "content": "few-shot conditional generation. In Advances in Neural Information Processing Systems, 2021.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 109, + 606, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 109, + 606, + 506, + 620 + ], + "score": 1.0, + "content": "[78] Mihaela Rosca, Balaji Lakshminarayanan, and Shakir Mohamed. Distribution matching in variational", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 130, + 617, + 316, + 629 + ], + "spans": [ + { + "bbox": [ + 130, + 617, + 316, + 629 + ], + "score": 1.0, + "content": "inference. arXiv preprint arXiv:1802.06847, 2018.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 110, + 635, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 110, + 635, + 505, + 647 + ], + "score": 1.0, + "content": "[79] Matthew D Hoffman and Matthew J Johnson. Elbo surgery: yet another way to carve up the variational", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 129, + 644, + 498, + 658 + ], + "spans": [ + { + "bbox": [ + 129, + 644, + 498, + 658 + ], + "score": 1.0, + "content": "evidence lower bound. In Workshop in Advances in Approximate Bayesian Inference, NeurIPS, 2016.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 109, + 662, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 109, + 662, + 506, + 676 + ], + "score": 1.0, + "content": "[80] Zhijian Liu, Haotian Tang, Yujun Lin, and Song Han. Point-voxel cnn for efficient 3d deep learning. In", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 129, + 673, + 345, + 686 + ], + "spans": [ + { + "bbox": [ + 129, + 673, + 345, + 686 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, 2019.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 109, + 691, + 506, + 705 + ], + "spans": [ + { + "bbox": [ + 109, + 691, + 506, + 705 + ], + "score": 1.0, + "content": "[81] Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 130, + 701, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 130, + 701, + 506, + 713 + ], + "score": 1.0, + "content": "3d classification and segmentation. In IEEE Conference on Computer Vision and Pattern Recognition", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 129, + 711, + 187, + 723 + ], + "spans": [ + { + "bbox": [ + 129, + 711, + 187, + 723 + ], + "score": 1.0, + "content": "(CVPR), 2017.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 24 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 109, + 40, + 507, + 728 + ], + "lines": [], + "index": 24, + "bbox_fs": [ + 108, + 73, + 507, + 723 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 52, + 507, + 727 + ], + "lines": [ + { + "bbox": [ + 108, + 72, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 108, + 72, + 507, + 87 + ], + "score": 1.0, + "content": "[82] Charles R Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointnet++: Deep hierarchical feature learning on", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 128, + 83, + 459, + 96 + ], + "spans": [ + { + "bbox": [ + 128, + 83, + 459, + 96 + ], + "score": 1.0, + "content": "point sets in a metric space. In Advances in Neural Information Processing Systems, 2017.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 109, + 101, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 109, + 101, + 506, + 115 + ], + "score": 1.0, + "content": "[83] Kaiming He, X. Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 130, + 112, + 435, + 124 + ], + "spans": [ + { + "bbox": [ + 130, + 112, + 435, + 124 + ], + "score": 1.0, + "content": "2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 110, + 131, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 110, + 131, + 505, + 142 + ], + "score": 1.0, + "content": "[84] Yuxin Wu and Kaiming He. Group normalization. In Proceedings of the European Conference on", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 131, + 141, + 249, + 151 + ], + "spans": [ + { + "bbox": [ + 131, + 141, + 249, + 151 + ], + "score": 1.0, + "content": "Computer Vision (ECCV), 2018.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 109, + 158, + 507, + 172 + ], + "spans": [ + { + "bbox": [ + 109, + 158, + 507, + 172 + ], + "score": 1.0, + "content": "[85] Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 129, + 167, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 129, + 167, + 506, + 182 + ], + "score": 1.0, + "content": "SDEdit: Guided image synthesis and editing with stochastic differential equations. In International", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 178, + 305, + 190 + ], + "spans": [ + { + "bbox": [ + 130, + 178, + 305, + 190 + ], + "score": 1.0, + "content": "Conference on Learning Representations, 2022.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 109, + 196, + 507, + 209 + ], + "spans": [ + { + "bbox": [ + 109, + 196, + 507, + 209 + ], + "score": 1.0, + "content": "[86] Nanxin Chen, Yu Zhang, Heiga Zen, Ron J Weiss, Mohammad Norouzi, and William Chan. Wavegrad:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 130, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 130, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "Estimating gradients for waveform generation. In International Conference on Learning Representations", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 129, + 217, + 185, + 229 + ], + "spans": [ + { + "bbox": [ + 129, + 217, + 185, + 229 + ], + "score": 1.0, + "content": "(ICLR), 2021.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 110, + 235, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 110, + 235, + 505, + 247 + ], + "score": 1.0, + "content": "[87] Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao, and Bryan Catanzaro. DiffWave: A Versatile Diffusion", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 129, + 245, + 467, + 257 + ], + "spans": [ + { + "bbox": [ + 129, + 245, + 467, + 257 + ], + "score": 1.0, + "content": "Model for Audio Synthesis. In International Conference on Learning Representations, 2021.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 108, + 262, + 507, + 276 + ], + "spans": [ + { + "bbox": [ + 108, + 262, + 507, + 276 + ], + "score": 1.0, + "content": "[88] Myeonghun Jeong, Hyeongju Kim, Sung Jun Cheon, Byoung Jin Choi, and Nam Soo Kim. Diff-tts: A", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 130, + 273, + 442, + 285 + ], + "spans": [ + { + "bbox": [ + 130, + 273, + 442, + 285 + ], + "score": 1.0, + "content": "denoising diffusion model for text-to-speech. arXiv preprint arXiv:2104.01409, 2021.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 110, + 292, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 110, + 292, + 505, + 304 + ], + "score": 1.0, + "content": "[89] Nanxin Chen, Yu Zhang, Heiga Zen, Ron J. Weiss, Mohammad Norouzi, Najim Dehak, and William", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 128, + 300, + 507, + 315 + ], + "spans": [ + { + "bbox": [ + 128, + 300, + 507, + 315 + ], + "score": 1.0, + "content": "Chan. Wavegrad 2: Iterative refinement for text-to-speech synthesis. arXiv preprint arXiv:2106.09660,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 130, + 312, + 155, + 322 + ], + "spans": [ + { + "bbox": [ + 130, + 312, + 155, + 322 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 110, + 330, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 110, + 330, + 506, + 342 + ], + "score": 1.0, + "content": "[90] Vadim Popov, Ivan Vovk, Vladimir Gogoryan, Tasnima Sadekova, and Mikhail Kudinov. Grad-tts: A", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 130, + 340, + 507, + 352 + ], + "spans": [ + { + "bbox": [ + 130, + 340, + 507, + 352 + ], + "score": 1.0, + "content": "diffusion probabilistic model for text-to-speech. In International Conference on Machine Learning, 2021.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 110, + 359, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 110, + 359, + 505, + 370 + ], + "score": 1.0, + "content": "[91] Songxiang Liu, Dan Su, and Dong Yu. Diffgan-tts: High-fidelity and efficient text-to-speech with", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 130, + 369, + 371, + 381 + ], + "spans": [ + { + "bbox": [ + 130, + 369, + 371, + 381 + ], + "score": 1.0, + "content": "denoising diffusion gans. arXiv preprint arXiv:2201.11972, 2022.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 110, + 387, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 110, + 387, + 506, + 399 + ], + "score": 1.0, + "content": "[92] Gautam Mittal, Jesse Engel, Curtis Hawthorne, and Ian Simon. Symbolic music generation with diffusion", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 128, + 396, + 507, + 411 + ], + "spans": [ + { + "bbox": [ + 128, + 396, + 507, + 411 + ], + "score": 1.0, + "content": "models. In Proceedings of the 22nd International Society for Music Information Retrieval Conference,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 129, + 406, + 156, + 419 + ], + "spans": [ + { + "bbox": [ + 129, + 406, + 156, + 419 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 110, + 425, + 507, + 437 + ], + "spans": [ + { + "bbox": [ + 110, + 425, + 507, + 437 + ], + "score": 1.0, + "content": "[93] Kushagra Pandey, Avideep Mukherjee, Piyush Rai, and Abhishek Kumar. Diffusevae: Efficient, con-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 129, + 434, + 507, + 448 + ], + "spans": [ + { + "bbox": [ + 129, + 434, + 507, + 448 + ], + "score": 1.0, + "content": "trollable and high-fidelity generation from low-dimensional latents. arXiv preprint arXiv:2201.00308,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 129, + 444, + 155, + 458 + ], + "spans": [ + { + "bbox": [ + 129, + 444, + 155, + 458 + ], + "score": 1.0, + "content": "2022.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 108, + 462, + 507, + 476 + ], + "spans": [ + { + "bbox": [ + 108, + 462, + 507, + 476 + ], + "score": 1.0, + "content": "[94] Danilo Jimenez Rezende and Shakir Mohamed. Variational inference with normalizing flows. In", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 130, + 474, + 329, + 486 + ], + "spans": [ + { + "bbox": [ + 130, + 474, + 329, + 486 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, 2015.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 109, + 491, + 507, + 505 + ], + "spans": [ + { + "bbox": [ + 109, + 491, + 507, + 505 + ], + "score": 1.0, + "content": "[95] Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real NVP. In", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 129, + 502, + 377, + 514 + ], + "spans": [ + { + "bbox": [ + 129, + 502, + 377, + 514 + ], + "score": 1.0, + "content": "International Conference on Learning Representations ICLR, 2017.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 110, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 110, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "[96] Ricky T. Q. Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 129, + 530, + 387, + 542 + ], + "spans": [ + { + "bbox": [ + 129, + 530, + 387, + 542 + ], + "score": 1.0, + "content": "equations. Advances in Neural Information Processing Systems, 2018.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 108, + 547, + 507, + 561 + ], + "spans": [ + { + "bbox": [ + 108, + 547, + 507, + 561 + ], + "score": 1.0, + "content": "[97] Will Grathwohl, Ricky T. Q. Chen, Jesse Bettencourt, Ilya Sutskever, and David Duvenaud. FFJORD:", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 129, + 559, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 129, + 559, + 505, + 570 + ], + "score": 1.0, + "content": "Free-form continuous dynamics for scalable reversible generative models. In International Conference", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 131, + 570, + 261, + 579 + ], + "spans": [ + { + "bbox": [ + 131, + 570, + 261, + 579 + ], + "score": 1.0, + "content": "on Learning Representations, 2019.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 110, + 587, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 110, + 587, + 506, + 599 + ], + "score": 1.0, + "content": "[98] Zhaoyang Lyu, Zhifeng Kong, Xudong XU, Liang Pan, and Dahua Lin. A conditional point diffusion-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 130, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 130, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "refinement paradigm for 3d point cloud completion. In International Conference on Learning Representa-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 129, + 607, + 204, + 619 + ], + "spans": [ + { + "bbox": [ + 129, + 607, + 204, + 619 + ], + "score": 1.0, + "content": "tions (ICLR), 2022.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 110, + 626, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 110, + 626, + 505, + 637 + ], + "score": 1.0, + "content": "[99] Thibault Groueix, Matthew Fisher, Vladimir G. Kim, Bryan Russell, and Mathieu Aubry. AtlasNet: A", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 128, + 633, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 128, + 633, + 506, + 648 + ], + "score": 1.0, + "content": "Papier-Mâché Approach to Learning 3D Surface Generation. In Proceedings IEEE Conf. on Computer", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 130, + 645, + 301, + 657 + ], + "spans": [ + { + "bbox": [ + 130, + 645, + 301, + 657 + ], + "score": 1.0, + "content": "Vision and Pattern Recognition (CVPR), 2018.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 663, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 506, + 675 + ], + "score": 1.0, + "content": "[100] Lars Mescheder, Michael Oechsle, Michael Niemeyer, Sebastian Nowozin, and Andreas Geiger. Oc-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 130, + 673, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 130, + 673, + 505, + 685 + ], + "score": 1.0, + "content": "cupancy networks: Learning 3d reconstruction in function space. In Proceedings of the IEEE/CVF", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 129, + 682, + 431, + 696 + ], + "spans": [ + { + "bbox": [ + 129, + 682, + 431, + 696 + ], + "score": 1.0, + "content": "Conference on Computer Vision and Pattern Recognition, pages 4460–4470, 2019.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 702, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 505, + 713 + ], + "score": 1.0, + "content": "[101] Songyou Peng, Michael Niemeyer, Lars Mescheder, and Andreas Geiger Marc Pollefeys. Convolutional", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 129, + 711, + 431, + 724 + ], + "spans": [ + { + "bbox": [ + 129, + 711, + 431, + 724 + ], + "score": 1.0, + "content": "occupancy networks. In European Conference on Computer Vision (ECCV), 2020.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 24 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 52, + 507, + 727 + ], + "lines": [], + "index": 24, + "bbox_fs": [ + 105, + 72, + 507, + 724 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 73, + 504, + 104 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "[102] Francis Williams, Matthew Trager, Joan Bruna, and Denis Zorin. Neural splines: Fitting 3d surfaces", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 130, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 130, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "with infinitely-wide neural networks. In 2021 IEEE/CVF Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 130, + 93, + 232, + 105 + ], + "spans": [ + { + "bbox": [ + 130, + 93, + 232, + 105 + ], + "score": 1.0, + "content": "Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 104, + 111, + 504, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 507, + 125 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 507, + 125 + ], + "score": 1.0, + "content": "[103] Francis Williams, Zan Gojcic, Sameh Khamis, Denis Zorin, Joan Bruna, Sanja Fidler, and Or Litany.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 129, + 122, + 479, + 134 + ], + "spans": [ + { + "bbox": [ + 129, + 122, + 479, + 134 + ], + "score": 1.0, + "content": "Neural fields as learnable kernels for 3d reconstruction. arXiv preprint arXiv:2111.13674, 2021.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 140, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 140, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 140, + 505, + 152 + ], + "score": 1.0, + "content": "[104] Angel X. Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 131, + 151, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 131, + 151, + 505, + 162 + ], + "score": 1.0, + "content": "Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. ShapeNet: An", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 129, + 159, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 129, + 159, + 505, + 173 + ], + "score": 1.0, + "content": "Information-Rich 3D Model Repository. Technical Report arXiv:1512.03012 [cs.GR], Stanford University", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 170, + 405, + 182 + ], + "spans": [ + { + "bbox": [ + 130, + 170, + 405, + 182 + ], + "score": 1.0, + "content": "— Princeton University — Toyota Technological Institute at Chicago, 2015.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 189, + 506, + 230 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 201 + ], + "score": 1.0, + "content": "[105] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 129, + 198, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 129, + 198, + 506, + 213 + ], + "score": 1.0, + "content": "Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 130, + 209, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 130, + 209, + 505, + 221 + ], + "score": 1.0, + "content": "transferable visual models from natural language supervision. In International Conference on Machine", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 130, + 219, + 217, + 231 + ], + "spans": [ + { + "bbox": [ + 130, + 219, + 217, + 231 + ], + "score": 1.0, + "content": "Learning, ICML, 2021.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 105, + 237, + 504, + 259 + ], + "lines": [ + { + "bbox": [ + 107, + 238, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 107, + 238, + 505, + 249 + ], + "score": 1.0, + "content": "[106] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In International", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 130, + 249, + 305, + 259 + ], + "spans": [ + { + "bbox": [ + 130, + 249, + 305, + 259 + ], + "score": 1.0, + "content": "Conference on Learning Representations, 2021.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 267, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 266, + 507, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 507, + 279 + ], + "score": 1.0, + "content": "[107] Wenzheng Chen, Huan Ling, Jun Gao, Edward Smith, Jaakko Lehtinen, Alec Jacobson, and Sanja Fidler.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 130, + 277, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 130, + 277, + 506, + 288 + ], + "score": 1.0, + "content": "Learning to predict 3d objects with an interpolation-based differentiable renderer. In Advances in Neural", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 130, + 287, + 273, + 298 + ], + "spans": [ + { + "bbox": [ + 130, + 287, + 273, + 298 + ], + "score": 1.0, + "content": "Information Processing Systems, 2019.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 306, + 504, + 327 + ], + "lines": [ + { + "bbox": [ + 107, + 306, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 107, + 306, + 505, + 317 + ], + "score": 1.0, + "content": "[108] Samuli Laine, Janne Hellsten, Tero Karras, Yeongho Seol, Jaakko Lehtinen, and Timo Aila. Modular", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 129, + 316, + 479, + 327 + ], + "spans": [ + { + "bbox": [ + 129, + 316, + 479, + 327 + ], + "score": 1.0, + "content": "primitives for high-performance differentiable rendering. ACM Transactions on Graphics, 2020.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 105, + 334, + 504, + 356 + ], + "lines": [ + { + "bbox": [ + 106, + 335, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 505, + 346 + ], + "score": 1.0, + "content": "[109] Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 129, + 344, + 463, + 356 + ], + "spans": [ + { + "bbox": [ + 129, + 344, + 463, + 356 + ], + "score": 1.0, + "content": "Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. In ECCV, 2020.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 106, + 363, + 504, + 394 + ], + "lines": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "[110] Wenzheng Chen, Joey Litalien, Jun Gao, Zian Wang, Clement Fuji Tsang, Sameh Khamis, Or Litany, and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 130, + 373, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 130, + 373, + 196, + 385 + ], + "score": 1.0, + "content": "Sanja Fidler. DIB-", + "type": "text" + }, + { + "bbox": [ + 197, + 374, + 214, + 383 + ], + "score": 0.74, + "content": "\\mathbf { \\cdot R } { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 373, + 506, + 385 + ], + "score": 1.0, + "content": ": Learning to predict lighting and material with a hybrid differentiable renderer. In", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 129, + 383, + 385, + 395 + ], + "spans": [ + { + "bbox": [ + 129, + 383, + 385, + 395 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems (NeurIPS), 2021.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 402, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 402, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 505, + 414 + ], + "score": 1.0, + "content": "[111] Ayush Tewari, Justus Thies, Ben Mildenhall, Pratul Srinivasan, Edgar Tretschk, Yifan Wang, Christoph", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 130, + 413, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 130, + 413, + 506, + 424 + ], + "score": 1.0, + "content": "Lassner, Vincent Sitzmann, Ricardo Martin-Brualla, Stephen Lombardi, Tomas Simon, Christian Theobalt,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 130, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 130, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "Matthias Niessner, Jonathan T. Barron, Gordon Wetzstein, Michael Zollhoefer, and Vladislav Golyanik.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 130, + 432, + 387, + 444 + ], + "spans": [ + { + "bbox": [ + 130, + 432, + 387, + 444 + ], + "score": 1.0, + "content": "Advances in neural rendering. arXiv preprint arXiv:2111.05849, 2021.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 105, + 450, + 504, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 451, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 462 + ], + "score": 1.0, + "content": "[112] Michael Oechsle, Lars Mescheder, Michael Niemeyer, Thilo Strauss, and Andreas Geiger. Texture fields:", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 130, + 462, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 130, + 462, + 505, + 472 + ], + "score": 1.0, + "content": "Learning texture representations in function space. In Proceedings IEEE International Conf. on Computer", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 130, + 471, + 208, + 482 + ], + "spans": [ + { + "bbox": [ + 130, + 471, + 208, + 482 + ], + "score": 1.0, + "content": "Vision (ICCV), 2019.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 105, + 489, + 504, + 511 + ], + "lines": [ + { + "bbox": [ + 106, + 490, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 505, + 501 + ], + "score": 1.0, + "content": "[113] Zhiqin Chen, Kangxue Yin, and Sanja Fidler. Auv-net: Learning aligned uv maps for texture transfer and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 130, + 500, + 456, + 511 + ], + "spans": [ + { + "bbox": [ + 130, + 500, + 456, + 511 + ], + "score": 1.0, + "content": "synthesis. In The Conference on Computer Vision and Pattern Recognition (CVPR), 2022.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 105, + 518, + 504, + 540 + ], + "lines": [ + { + "bbox": [ + 106, + 518, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 506, + 531 + ], + "score": 1.0, + "content": "[114] Yawar Siddiqui, Justus Thies, Fangchang Ma, Qi Shan, Matthias Nießner, and Angela Dai. Texturify:", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 130, + 528, + 429, + 539 + ], + "spans": [ + { + "bbox": [ + 130, + 528, + 429, + 539 + ], + "score": 1.0, + "content": "Generating textures on 3d shape surfaces. arXiv preprint arXiv:2204.02411, 2022.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 106, + 547, + 504, + 568 + ], + "lines": [ + { + "bbox": [ + 106, + 548, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 548, + 505, + 559 + ], + "score": 1.0, + "content": "[115] Daniel Watson, Jonathan Ho, Mohammad Norouzi, and William Chan. Learning to efficiently sample", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 130, + 558, + 409, + 569 + ], + "spans": [ + { + "bbox": [ + 130, + 558, + 409, + 569 + ], + "score": 1.0, + "content": "from diffusion probabilistic models. arXiv preprint arXiv:2106.03802, 2021.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 504, + 597 + ], + "lines": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "score": 1.0, + "content": "[116] Zhifeng Kong and Wei Ping. On fast sampling of diffusion probabilistic models. arXiv preprint", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 129, + 586, + 223, + 596 + ], + "spans": [ + { + "bbox": [ + 129, + 586, + 223, + 596 + ], + "score": 1.0, + "content": "arXiv:2106.00132, 2021.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 106, + 605, + 504, + 626 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "[117] Alexia Jolicoeur-Martineau, Ke Li, Rémi Piché-Taillefer, Tal Kachman, and Ioannis Mitliagkas. Gotta", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 131, + 615, + 490, + 626 + ], + "spans": [ + { + "bbox": [ + 131, + 615, + 490, + 626 + ], + "score": 1.0, + "content": "Go Fast When Generating Data with Score-Based Models. arXiv preprint arXiv:2105.14080, 2021.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 106, + 633, + 504, + 664 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 506, + 646 + ], + "score": 1.0, + "content": "[118] Daniel Watson, William Chan, Jonathan Ho, and Mohammad Norouzi. Learning Fast Samplers for", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 129, + 642, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 129, + 642, + 506, + 657 + ], + "score": 1.0, + "content": "Diffusion Models by Differentiating Through Sample Quality. In International Conference on Learning", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 130, + 654, + 214, + 665 + ], + "spans": [ + { + "bbox": [ + 130, + 654, + 214, + 665 + ], + "score": 1.0, + "content": "Representations, 2022.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 105, + 672, + 503, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 672, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 505, + 684 + ], + "score": 1.0, + "content": "[119] Luping Liu, Yi Ren, Zhijie Lin, and Zhou Zhao. Pseudo numerical methods for diffusion models on", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 129, + 682, + 433, + 694 + ], + "spans": [ + { + "bbox": [ + 129, + 682, + 433, + 694 + ], + "score": 1.0, + "content": "manifolds. In International Conference on Learning Representations (ICLR), 2022.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + }, + { + "type": "text", + "bbox": [ + 105, + 701, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 701, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 701, + 505, + 712 + ], + "score": 1.0, + "content": "[120] Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. In", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 130, + 712, + 382, + 723 + ], + "spans": [ + { + "bbox": [ + 130, + 712, + 382, + 723 + ], + "score": 1.0, + "content": "International Conference on Learning Representations (ICLR), 2022.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 73, + 504, + 104 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "[102] Francis Williams, Matthew Trager, Joan Bruna, and Denis Zorin. Neural splines: Fitting 3d surfaces", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 130, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 130, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "with infinitely-wide neural networks. In 2021 IEEE/CVF Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 130, + 93, + 232, + 105 + ], + "spans": [ + { + "bbox": [ + 130, + 93, + 232, + 105 + ], + "score": 1.0, + "content": "Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 72, + 505, + 105 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 111, + 504, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 507, + 125 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 507, + 125 + ], + "score": 1.0, + "content": "[103] Francis Williams, Zan Gojcic, Sameh Khamis, Denis Zorin, Joan Bruna, Sanja Fidler, and Or Litany.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 129, + 122, + 479, + 134 + ], + "spans": [ + { + "bbox": [ + 129, + 122, + 479, + 134 + ], + "score": 1.0, + "content": "Neural fields as learnable kernels for 3d reconstruction. arXiv preprint arXiv:2111.13674, 2021.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 110, + 507, + 134 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 140, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 140, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 140, + 505, + 152 + ], + "score": 1.0, + "content": "[104] Angel X. Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 131, + 151, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 131, + 151, + 505, + 162 + ], + "score": 1.0, + "content": "Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. ShapeNet: An", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 129, + 159, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 129, + 159, + 505, + 173 + ], + "score": 1.0, + "content": "Information-Rich 3D Model Repository. Technical Report arXiv:1512.03012 [cs.GR], Stanford University", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 130, + 170, + 405, + 182 + ], + "spans": [ + { + "bbox": [ + 130, + 170, + 405, + 182 + ], + "score": 1.0, + "content": "— Princeton University — Toyota Technological Institute at Chicago, 2015.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 140, + 505, + 182 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 189, + 506, + 230 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 201 + ], + "score": 1.0, + "content": "[105] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 129, + 198, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 129, + 198, + 506, + 213 + ], + "score": 1.0, + "content": "Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 130, + 209, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 130, + 209, + 505, + 221 + ], + "score": 1.0, + "content": "transferable visual models from natural language supervision. In International Conference on Machine", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 130, + 219, + 217, + 231 + ], + "spans": [ + { + "bbox": [ + 130, + 219, + 217, + 231 + ], + "score": 1.0, + "content": "Learning, ICML, 2021.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 106, + 190, + 506, + 231 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 237, + 504, + 259 + ], + "lines": [ + { + "bbox": [ + 107, + 238, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 107, + 238, + 505, + 249 + ], + "score": 1.0, + "content": "[106] Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In International", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 130, + 249, + 305, + 259 + ], + "spans": [ + { + "bbox": [ + 130, + 249, + 305, + 259 + ], + "score": 1.0, + "content": "Conference on Learning Representations, 2021.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 107, + 238, + 505, + 259 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 267, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 266, + 507, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 507, + 279 + ], + "score": 1.0, + "content": "[107] Wenzheng Chen, Huan Ling, Jun Gao, Edward Smith, Jaakko Lehtinen, Alec Jacobson, and Sanja Fidler.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 130, + 277, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 130, + 277, + 506, + 288 + ], + "score": 1.0, + "content": "Learning to predict 3d objects with an interpolation-based differentiable renderer. In Advances in Neural", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 130, + 287, + 273, + 298 + ], + "spans": [ + { + "bbox": [ + 130, + 287, + 273, + 298 + ], + "score": 1.0, + "content": "Information Processing Systems, 2019.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 266, + 507, + 298 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 306, + 504, + 327 + ], + "lines": [ + { + "bbox": [ + 107, + 306, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 107, + 306, + 505, + 317 + ], + "score": 1.0, + "content": "[108] Samuli Laine, Janne Hellsten, Tero Karras, Yeongho Seol, Jaakko Lehtinen, and Timo Aila. Modular", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 129, + 316, + 479, + 327 + ], + "spans": [ + { + "bbox": [ + 129, + 316, + 479, + 327 + ], + "score": 1.0, + "content": "primitives for high-performance differentiable rendering. ACM Transactions on Graphics, 2020.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 107, + 306, + 505, + 327 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 334, + 504, + 356 + ], + "lines": [ + { + "bbox": [ + 106, + 335, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 505, + 346 + ], + "score": 1.0, + "content": "[109] Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 129, + 344, + 463, + 356 + ], + "spans": [ + { + "bbox": [ + 129, + 344, + 463, + 356 + ], + "score": 1.0, + "content": "Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. In ECCV, 2020.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 106, + 335, + 505, + 356 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 363, + 504, + 394 + ], + "lines": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "[110] Wenzheng Chen, Joey Litalien, Jun Gao, Zian Wang, Clement Fuji Tsang, Sameh Khamis, Or Litany, and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 130, + 373, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 130, + 373, + 196, + 385 + ], + "score": 1.0, + "content": "Sanja Fidler. DIB-", + "type": "text" + }, + { + "bbox": [ + 197, + 374, + 214, + 383 + ], + "score": 0.74, + "content": "\\mathbf { \\cdot R } { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 373, + 506, + 385 + ], + "score": 1.0, + "content": ": Learning to predict lighting and material with a hybrid differentiable renderer. In", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 129, + 383, + 385, + 395 + ], + "spans": [ + { + "bbox": [ + 129, + 383, + 385, + 395 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems (NeurIPS), 2021.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 363, + 506, + 395 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 402, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 402, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 505, + 414 + ], + "score": 1.0, + "content": "[111] Ayush Tewari, Justus Thies, Ben Mildenhall, Pratul Srinivasan, Edgar Tretschk, Yifan Wang, Christoph", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 130, + 413, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 130, + 413, + 506, + 424 + ], + "score": 1.0, + "content": "Lassner, Vincent Sitzmann, Ricardo Martin-Brualla, Stephen Lombardi, Tomas Simon, Christian Theobalt,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 130, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 130, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "Matthias Niessner, Jonathan T. Barron, Gordon Wetzstein, Michael Zollhoefer, and Vladislav Golyanik.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 130, + 432, + 387, + 444 + ], + "spans": [ + { + "bbox": [ + 130, + 432, + 387, + 444 + ], + "score": 1.0, + "content": "Advances in neural rendering. arXiv preprint arXiv:2111.05849, 2021.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 106, + 402, + 506, + 444 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 450, + 504, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 451, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 506, + 462 + ], + "score": 1.0, + "content": "[112] Michael Oechsle, Lars Mescheder, Michael Niemeyer, Thilo Strauss, and Andreas Geiger. Texture fields:", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 130, + 462, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 130, + 462, + 505, + 472 + ], + "score": 1.0, + "content": "Learning texture representations in function space. In Proceedings IEEE International Conf. on Computer", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 130, + 471, + 208, + 482 + ], + "spans": [ + { + "bbox": [ + 130, + 471, + 208, + 482 + ], + "score": 1.0, + "content": "Vision (ICCV), 2019.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 451, + 506, + 482 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 489, + 504, + 511 + ], + "lines": [ + { + "bbox": [ + 106, + 490, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 505, + 501 + ], + "score": 1.0, + "content": "[113] Zhiqin Chen, Kangxue Yin, and Sanja Fidler. Auv-net: Learning aligned uv maps for texture transfer and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 130, + 500, + 456, + 511 + ], + "spans": [ + { + "bbox": [ + 130, + 500, + 456, + 511 + ], + "score": 1.0, + "content": "synthesis. In The Conference on Computer Vision and Pattern Recognition (CVPR), 2022.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 106, + 490, + 505, + 511 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 518, + 504, + 540 + ], + "lines": [ + { + "bbox": [ + 106, + 518, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 506, + 531 + ], + "score": 1.0, + "content": "[114] Yawar Siddiqui, Justus Thies, Fangchang Ma, Qi Shan, Matthias Nießner, and Angela Dai. Texturify:", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 130, + 528, + 429, + 539 + ], + "spans": [ + { + "bbox": [ + 130, + 528, + 429, + 539 + ], + "score": 1.0, + "content": "Generating textures on 3d shape surfaces. arXiv preprint arXiv:2204.02411, 2022.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 106, + 518, + 506, + 539 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 547, + 504, + 568 + ], + "lines": [ + { + "bbox": [ + 106, + 548, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 548, + 505, + 559 + ], + "score": 1.0, + "content": "[115] Daniel Watson, Jonathan Ho, Mohammad Norouzi, and William Chan. Learning to efficiently sample", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 130, + 558, + 409, + 569 + ], + "spans": [ + { + "bbox": [ + 130, + 558, + 409, + 569 + ], + "score": 1.0, + "content": "from diffusion probabilistic models. arXiv preprint arXiv:2106.03802, 2021.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 548, + 505, + 569 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 576, + 504, + 597 + ], + "lines": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 506, + 589 + ], + "score": 1.0, + "content": "[116] Zhifeng Kong and Wei Ping. On fast sampling of diffusion probabilistic models. arXiv preprint", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 129, + 586, + 223, + 596 + ], + "spans": [ + { + "bbox": [ + 129, + 586, + 223, + 596 + ], + "score": 1.0, + "content": "arXiv:2106.00132, 2021.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 575, + 506, + 596 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 605, + 504, + 626 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "[117] Alexia Jolicoeur-Martineau, Ke Li, Rémi Piché-Taillefer, Tal Kachman, and Ioannis Mitliagkas. Gotta", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 131, + 615, + 490, + 626 + ], + "spans": [ + { + "bbox": [ + 131, + 615, + 490, + 626 + ], + "score": 1.0, + "content": "Go Fast When Generating Data with Score-Based Models. arXiv preprint arXiv:2105.14080, 2021.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 106, + 605, + 505, + 626 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 633, + 504, + 664 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 506, + 646 + ], + "score": 1.0, + "content": "[118] Daniel Watson, William Chan, Jonathan Ho, and Mohammad Norouzi. Learning Fast Samplers for", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 129, + 642, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 129, + 642, + 506, + 657 + ], + "score": 1.0, + "content": "Diffusion Models by Differentiating Through Sample Quality. In International Conference on Learning", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 130, + 654, + 214, + 665 + ], + "spans": [ + { + "bbox": [ + 130, + 654, + 214, + 665 + ], + "score": 1.0, + "content": "Representations, 2022.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 633, + 506, + 665 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 672, + 503, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 672, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 505, + 684 + ], + "score": 1.0, + "content": "[119] Luping Liu, Yi Ren, Zhijie Lin, and Zhou Zhao. Pseudo numerical methods for diffusion models on", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 129, + 682, + 433, + 694 + ], + "spans": [ + { + "bbox": [ + 129, + 682, + 433, + 694 + ], + "score": 1.0, + "content": "manifolds. In International Conference on Learning Representations (ICLR), 2022.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 672, + 505, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 701, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 701, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 701, + 505, + 712 + ], + "score": 1.0, + "content": "[120] Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. In", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 130, + 712, + 382, + 723 + ], + "spans": [ + { + "bbox": [ + 130, + 712, + 382, + 723 + ], + "score": 1.0, + "content": "International Conference on Learning Representations (ICLR), 2022.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5, + "bbox_fs": [ + 106, + 701, + 505, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 62, + 507, + 727 + ], + "lines": [ + { + "bbox": [ + 106, + 74, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 74, + 505, + 86 + ], + "score": 1.0, + "content": "[121] Fan Bao, Chongxuan Li, Jun Zhu, and Bo Zhang. Analytic-DPM: an Analytic Estimate of the Op-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 128, + 81, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 128, + 81, + 506, + 97 + ], + "score": 1.0, + "content": "timal Reverse Variance in Diffusion Probabilistic Models. In International Conference on Learning", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 129, + 93, + 215, + 105 + ], + "spans": [ + { + "bbox": [ + 129, + 93, + 215, + 105 + ], + "score": 1.0, + "content": "Representations, 2022.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 112, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 506, + 123 + ], + "score": 1.0, + "content": "[122] Cristian Vaccari and Andrew Chadwick. Deepfakes and disinformation: Exploring the impact of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 130, + 122, + 506, + 134 + ], + "spans": [ + { + "bbox": [ + 130, + 122, + 445, + 134 + ], + "score": 1.0, + "content": "synthetic political video on deception, uncertainty, and trust in news. Social Media", + "type": "text" + }, + { + "bbox": [ + 446, + 123, + 452, + 131 + ], + "score": 0.52, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 122, + 506, + 134 + ], + "score": 1.0, + "content": "Society, 6(1):", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 131, + 132, + 230, + 142 + ], + "spans": [ + { + "bbox": [ + 131, + 132, + 230, + 142 + ], + "score": 1.0, + "content": "2056305120903408, 2020.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 148, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 104, + 148, + 506, + 162 + ], + "score": 1.0, + "content": "[123] Thanh Thi Nguyen, Quoc Viet Hung Nguyen, Cuong M. Nguyen, Dung Nguyen, Duc Thanh Nguyen,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 128, + 158, + 506, + 172 + ], + "spans": [ + { + "bbox": [ + 128, + 158, + 506, + 172 + ], + "score": 1.0, + "content": "and Saeid Nahavandi. Deep learning for deepfakes creation and detection: A survey. arXiv preprint", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 129, + 169, + 225, + 180 + ], + "spans": [ + { + "bbox": [ + 129, + 169, + 225, + 180 + ], + "score": 1.0, + "content": "arXiv:1909.11573, 2021.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 186, + 507, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 186, + 507, + 200 + ], + "score": 1.0, + "content": "[124] Yisroel Mirsky and Wenke Lee. The creation and detection of deepfakes: A survey. ACM Comput. Surv.,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 129, + 196, + 178, + 210 + ], + "spans": [ + { + "bbox": [ + 129, + 196, + 178, + 210 + ], + "score": 1.0, + "content": "54(1), 2021.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "[125] Merlin Nimier-David, Delio Vicini, Tizian Zeltner, and Wenzel Jakob. Mitsuba 2: a retargetable forward", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 129, + 225, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 129, + 225, + 506, + 237 + ], + "score": 1.0, + "content": "and inverse renderer. ACM Transactions on Graphics, 38:1–17, 11 2019. doi: 10.1145/3355089.3356498.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 243, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 506, + 255 + ], + "score": 1.0, + "content": "[126] Jonathan Krause, Michael Stark, Jia Deng, and Li Fei-Fei. 3d object representations for fine-grained", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 129, + 252, + 507, + 266 + ], + "spans": [ + { + "bbox": [ + 129, + 252, + 507, + 266 + ], + "score": 1.0, + "content": "categorization. In 4th International IEEE Workshop on 3D Representation and Recognition (3dRR-13),", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 129, + 262, + 221, + 274 + ], + "spans": [ + { + "bbox": [ + 129, + 262, + 221, + 274 + ], + "score": 1.0, + "content": "Sydney, Australia, 2013.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 280, + 507, + 294 + ], + "spans": [ + { + "bbox": [ + 104, + 280, + 507, + 294 + ], + "score": 1.0, + "content": "[127] Sungjoon Choi, Qian-Yi Zhou, Stephen Miller, and Vladlen Koltun. A large dataset of object scans.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 129, + 290, + 225, + 303 + ], + "spans": [ + { + "bbox": [ + 129, + 290, + 225, + 303 + ], + "score": 1.0, + "content": "arXiv:1602.02481, 2016.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 309, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 505, + 321 + ], + "score": 1.0, + "content": "[128] Xingyuan Sun, Jiajun Wu, Xiuming Zhang, Zhoutong Zhang, Chengkai Zhang, Tianfan Xue, Joshua B", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 129, + 317, + 506, + 332 + ], + "spans": [ + { + "bbox": [ + 129, + 317, + 506, + 332 + ], + "score": 1.0, + "content": "Tenenbaum, and William T Freeman. Pix3d: Dataset and methods for single-image 3d shape modeling.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 129, + 328, + 425, + 342 + ], + "spans": [ + { + "bbox": [ + 129, + 328, + 425, + 342 + ], + "score": 1.0, + "content": "In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 345, + 507, + 360 + ], + "spans": [ + { + "bbox": [ + 104, + 345, + 507, + 360 + ], + "score": 1.0, + "content": "[129] Brian DO Anderson. Reverse-time diffusion equation models. Stochastic Processes and their Applications,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 129, + 356, + 210, + 368 + ], + "spans": [ + { + "bbox": [ + 129, + 356, + 210, + 368 + ], + "score": 1.0, + "content": "12(3):313–326, 1982.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 372, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 104, + 372, + 506, + 388 + ], + "score": 1.0, + "content": "[130] Ulrich G Haussmann and Etienne Pardoux. Time reversal of diffusions. The Annals of Probability, pages", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 129, + 384, + 198, + 396 + ], + "spans": [ + { + "bbox": [ + 129, + 384, + 198, + 396 + ], + "score": 1.0, + "content": "1188–1205, 1986.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 401, + 507, + 416 + ], + "spans": [ + { + "bbox": [ + 104, + 401, + 507, + 416 + ], + "score": 1.0, + "content": "[131] Pascal Vincent. A connection between score matching and denoising autoencoders. Neural Computation,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 129, + 413, + 219, + 424 + ], + "spans": [ + { + "bbox": [ + 129, + 413, + 219, + 424 + ], + "score": 1.0, + "content": "23(7):1661–1674, 2011.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 429, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 104, + 429, + 506, + 443 + ], + "score": 1.0, + "content": "[132] Radford M Neal and Geoffrey E Hinton. A view of the em algorithm that justifies incremental, sparse,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 130, + 440, + 437, + 451 + ], + "spans": [ + { + "bbox": [ + 130, + 440, + 437, + 451 + ], + "score": 1.0, + "content": "and other variants. In Learning in graphical models, pages 355–368. Springer, 1998.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 457, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 104, + 457, + 506, + 471 + ], + "score": 1.0, + "content": "[133] Diederik P. Kingma and Max Welling. An introduction to variational autoencoders. Foundations and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 129, + 468, + 318, + 481 + ], + "spans": [ + { + "bbox": [ + 129, + 468, + 318, + 481 + ], + "score": 1.0, + "content": "Trends in Machine Learning, 12(4):307–392, 2019.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 484, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 104, + 484, + 506, + 499 + ], + "score": 1.0, + "content": "[134] J. R. Dormand and P. J. Prince. A family of embedded runge–kutta formulae. Journal of Computational", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 130, + 496, + 294, + 508 + ], + "spans": [ + { + "bbox": [ + 130, + 496, + 294, + 508 + ], + "score": 1.0, + "content": "and Applied Mathematics, 6(1):19–26, 1980.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 513, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 104, + 513, + 506, + 527 + ], + "score": 1.0, + "content": "[135] Michael Kazhdan, Matthew Bolitho, and Hugues Hoppe. Poisson surface reconstruction. In Proceedings", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 129, + 524, + 424, + 538 + ], + "spans": [ + { + "bbox": [ + 129, + 524, + 424, + 538 + ], + "score": 1.0, + "content": "of the fourth Eurographics symposium on Geometry processing, volume 7, 2006.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 542, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 506, + 554 + ], + "score": 1.0, + "content": "[136] William E. Lorensen and Harvey E. Cline. Marching cubes: A high resolution 3d surface construction", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 129, + 551, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 129, + 551, + 505, + 563 + ], + "score": 1.0, + "content": "algorithm. In Proceedings of the 14th Annual Conference on Computer Graphics and Interactive", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 128, + 560, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 128, + 560, + 506, + 576 + ], + "score": 1.0, + "content": "Techniques, SIGGRAPH ’87, page 163–169, New York, NY, USA, 1987. Association for Computing", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 130, + 571, + 175, + 583 + ], + "spans": [ + { + "bbox": [ + 130, + 571, + 175, + 583 + ], + "score": 1.0, + "content": "Machinery.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 590, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 506, + 601 + ], + "score": 1.0, + "content": "[137] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE conference", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 129, + 599, + 382, + 612 + ], + "spans": [ + { + "bbox": [ + 129, + 599, + 382, + 612 + ], + "score": 1.0, + "content": "on computer vision and pattern recognition, pages 7132–7141, 2018.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "[138] Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 129, + 627, + 484, + 640 + ], + "spans": [ + { + "bbox": [ + 129, + 627, + 484, + 640 + ], + "score": 1.0, + "content": "generative adversarial networks. In International Conference on Learning Representations, 2018.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 104, + 643, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 643, + 506, + 659 + ], + "score": 1.0, + "content": "[139] Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 130, + 656, + 214, + 668 + ], + "spans": [ + { + "bbox": [ + 130, + 656, + 214, + 668 + ], + "score": 1.0, + "content": "research, 9(11), 2008.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 673, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 506, + 685 + ], + "score": 1.0, + "content": "[140] Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E. Sarma, Michael M. Bronstein, and Justin M. Solomon.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 129, + 682, + 476, + 695 + ], + "spans": [ + { + "bbox": [ + 129, + 682, + 476, + 695 + ], + "score": 1.0, + "content": "Dynamic graph cnn for learning on point clouds. ACM Transactions on Graphics (TOG), 2019.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 701, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 701, + 506, + 713 + ], + "score": 1.0, + "content": "[141] Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. In", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 129, + 711, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 129, + 711, + 506, + 725 + ], + "score": 1.0, + "content": "Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 16259–16268, 2021.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 24 + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "score": 1.0, + "content": "17", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 62, + 507, + 727 + ], + "lines": [ + { + "bbox": [ + 106, + 74, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 74, + 505, + 86 + ], + "score": 1.0, + "content": "[121] Fan Bao, Chongxuan Li, Jun Zhu, and Bo Zhang. Analytic-DPM: an Analytic Estimate of the Op-", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 81, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 128, + 81, + 506, + 97 + ], + "score": 1.0, + "content": "timal Reverse Variance in Diffusion Probabilistic Models. In International Conference on Learning", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 129, + 93, + 215, + 105 + ], + "spans": [ + { + "bbox": [ + 129, + 93, + 215, + 105 + ], + "score": 1.0, + "content": "Representations, 2022.", + "type": "text" + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 112, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 506, + 123 + ], + "score": 1.0, + "content": "[122] Cristian Vaccari and Andrew Chadwick. Deepfakes and disinformation: Exploring the impact of", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 122, + 506, + 134 + ], + "spans": [ + { + "bbox": [ + 130, + 122, + 445, + 134 + ], + "score": 1.0, + "content": "synthetic political video on deception, uncertainty, and trust in news. Social Media", + "type": "text" + }, + { + "bbox": [ + 446, + 123, + 452, + 131 + ], + "score": 0.52, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 122, + 506, + 134 + ], + "score": 1.0, + "content": "Society, 6(1):", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 131, + 132, + 230, + 142 + ], + "spans": [ + { + "bbox": [ + 131, + 132, + 230, + 142 + ], + "score": 1.0, + "content": "2056305120903408, 2020.", + "type": "text" + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 148, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 104, + 148, + 506, + 162 + ], + "score": 1.0, + "content": "[123] Thanh Thi Nguyen, Quoc Viet Hung Nguyen, Cuong M. Nguyen, Dung Nguyen, Duc Thanh Nguyen,", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 158, + 506, + 172 + ], + "spans": [ + { + "bbox": [ + 128, + 158, + 506, + 172 + ], + "score": 1.0, + "content": "and Saeid Nahavandi. Deep learning for deepfakes creation and detection: A survey. arXiv preprint", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 129, + 169, + 225, + 180 + ], + "spans": [ + { + "bbox": [ + 129, + 169, + 225, + 180 + ], + "score": 1.0, + "content": "arXiv:1909.11573, 2021.", + "type": "text" + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 186, + 507, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 186, + 507, + 200 + ], + "score": 1.0, + "content": "[124] Yisroel Mirsky and Wenke Lee. The creation and detection of deepfakes: A survey. ACM Comput. Surv.,", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 196, + 178, + 210 + ], + "spans": [ + { + "bbox": [ + 129, + 196, + 178, + 210 + ], + "score": 1.0, + "content": "54(1), 2021.", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 228 + ], + "score": 1.0, + "content": "[125] Merlin Nimier-David, Delio Vicini, Tizian Zeltner, and Wenzel Jakob. Mitsuba 2: a retargetable forward", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 225, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 129, + 225, + 506, + 237 + ], + "score": 1.0, + "content": "and inverse renderer. ACM Transactions on Graphics, 38:1–17, 11 2019. doi: 10.1145/3355089.3356498.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 243, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 506, + 255 + ], + "score": 1.0, + "content": "[126] Jonathan Krause, Michael Stark, Jia Deng, and Li Fei-Fei. 3d object representations for fine-grained", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 252, + 507, + 266 + ], + "spans": [ + { + "bbox": [ + 129, + 252, + 507, + 266 + ], + "score": 1.0, + "content": "categorization. In 4th International IEEE Workshop on 3D Representation and Recognition (3dRR-13),", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 129, + 262, + 221, + 274 + ], + "spans": [ + { + "bbox": [ + 129, + 262, + 221, + 274 + ], + "score": 1.0, + "content": "Sydney, Australia, 2013.", + "type": "text" + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 280, + 507, + 294 + ], + "spans": [ + { + "bbox": [ + 104, + 280, + 507, + 294 + ], + "score": 1.0, + "content": "[127] Sungjoon Choi, Qian-Yi Zhou, Stephen Miller, and Vladlen Koltun. A large dataset of object scans.", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 290, + 225, + 303 + ], + "spans": [ + { + "bbox": [ + 129, + 290, + 225, + 303 + ], + "score": 1.0, + "content": "arXiv:1602.02481, 2016.", + "type": "text" + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 309, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 505, + 321 + ], + "score": 1.0, + "content": "[128] Xingyuan Sun, Jiajun Wu, Xiuming Zhang, Zhoutong Zhang, Chengkai Zhang, Tianfan Xue, Joshua B", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 317, + 506, + 332 + ], + "spans": [ + { + "bbox": [ + 129, + 317, + 506, + 332 + ], + "score": 1.0, + "content": "Tenenbaum, and William T Freeman. Pix3d: Dataset and methods for single-image 3d shape modeling.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 129, + 328, + 425, + 342 + ], + "spans": [ + { + "bbox": [ + 129, + 328, + 425, + 342 + ], + "score": 1.0, + "content": "In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 345, + 507, + 360 + ], + "spans": [ + { + "bbox": [ + 104, + 345, + 507, + 360 + ], + "score": 1.0, + "content": "[129] Brian DO Anderson. Reverse-time diffusion equation models. Stochastic Processes and their Applications,", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 356, + 210, + 368 + ], + "spans": [ + { + "bbox": [ + 129, + 356, + 210, + 368 + ], + "score": 1.0, + "content": "12(3):313–326, 1982.", + "type": "text" + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 372, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 104, + 372, + 506, + 388 + ], + "score": 1.0, + "content": "[130] Ulrich G Haussmann and Etienne Pardoux. Time reversal of diffusions. The Annals of Probability, pages", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 384, + 198, + 396 + ], + "spans": [ + { + "bbox": [ + 129, + 384, + 198, + 396 + ], + "score": 1.0, + "content": "1188–1205, 1986.", + "type": "text" + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 401, + 507, + 416 + ], + "spans": [ + { + "bbox": [ + 104, + 401, + 507, + 416 + ], + "score": 1.0, + "content": "[131] Pascal Vincent. A connection between score matching and denoising autoencoders. Neural Computation,", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 413, + 219, + 424 + ], + "spans": [ + { + "bbox": [ + 129, + 413, + 219, + 424 + ], + "score": 1.0, + "content": "23(7):1661–1674, 2011.", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 429, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 104, + 429, + 506, + 443 + ], + "score": 1.0, + "content": "[132] Radford M Neal and Geoffrey E Hinton. A view of the em algorithm that justifies incremental, sparse,", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 440, + 437, + 451 + ], + "spans": [ + { + "bbox": [ + 130, + 440, + 437, + 451 + ], + "score": 1.0, + "content": "and other variants. In Learning in graphical models, pages 355–368. Springer, 1998.", + "type": "text" + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 457, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 104, + 457, + 506, + 471 + ], + "score": 1.0, + "content": "[133] Diederik P. Kingma and Max Welling. An introduction to variational autoencoders. Foundations and", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 468, + 318, + 481 + ], + "spans": [ + { + "bbox": [ + 129, + 468, + 318, + 481 + ], + "score": 1.0, + "content": "Trends in Machine Learning, 12(4):307–392, 2019.", + "type": "text" + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 484, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 104, + 484, + 506, + 499 + ], + "score": 1.0, + "content": "[134] J. R. Dormand and P. J. Prince. A family of embedded runge–kutta formulae. Journal of Computational", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 496, + 294, + 508 + ], + "spans": [ + { + "bbox": [ + 130, + 496, + 294, + 508 + ], + "score": 1.0, + "content": "and Applied Mathematics, 6(1):19–26, 1980.", + "type": "text" + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 513, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 104, + 513, + 506, + 527 + ], + "score": 1.0, + "content": "[135] Michael Kazhdan, Matthew Bolitho, and Hugues Hoppe. Poisson surface reconstruction. In Proceedings", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 524, + 424, + 538 + ], + "spans": [ + { + "bbox": [ + 129, + 524, + 424, + 538 + ], + "score": 1.0, + "content": "of the fourth Eurographics symposium on Geometry processing, volume 7, 2006.", + "type": "text" + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 542, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 506, + 554 + ], + "score": 1.0, + "content": "[136] William E. Lorensen and Harvey E. Cline. Marching cubes: A high resolution 3d surface construction", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 551, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 129, + 551, + 505, + 563 + ], + "score": 1.0, + "content": "algorithm. In Proceedings of the 14th Annual Conference on Computer Graphics and Interactive", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 128, + 560, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 128, + 560, + 506, + 576 + ], + "score": 1.0, + "content": "Techniques, SIGGRAPH ’87, page 163–169, New York, NY, USA, 1987. Association for Computing", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 130, + 571, + 175, + 583 + ], + "spans": [ + { + "bbox": [ + 130, + 571, + 175, + 583 + ], + "score": 1.0, + "content": "Machinery.", + "type": "text" + } + ], + "index": 38, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 590, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 506, + 601 + ], + "score": 1.0, + "content": "[137] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In Proceedings of the IEEE conference", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 599, + 382, + 612 + ], + "spans": [ + { + "bbox": [ + 129, + 599, + 382, + 612 + ], + "score": 1.0, + "content": "on computer vision and pattern recognition, pages 7132–7141, 2018.", + "type": "text" + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "[138] Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 627, + 484, + 640 + ], + "spans": [ + { + "bbox": [ + 129, + 627, + 484, + 640 + ], + "score": 1.0, + "content": "generative adversarial networks. In International Conference on Learning Representations, 2018.", + "type": "text" + } + ], + "index": 42, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 643, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 643, + 506, + 659 + ], + "score": 1.0, + "content": "[139] Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 130, + 656, + 214, + 668 + ], + "spans": [ + { + "bbox": [ + 130, + 656, + 214, + 668 + ], + "score": 1.0, + "content": "research, 9(11), 2008.", + "type": "text" + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 673, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 506, + 685 + ], + "score": 1.0, + "content": "[140] Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E. Sarma, Michael M. Bronstein, and Justin M. Solomon.", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 682, + 476, + 695 + ], + "spans": [ + { + "bbox": [ + 129, + 682, + 476, + 695 + ], + "score": 1.0, + "content": "Dynamic graph cnn for learning on point clouds. ACM Transactions on Graphics (TOG), 2019.", + "type": "text" + } + ], + "index": 46, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 701, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 701, + 506, + 713 + ], + "score": 1.0, + "content": "[141] Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip HS Torr, and Vladlen Koltun. Point transformer. In", + "type": "text" + } + ], + "index": 47, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 711, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 129, + 711, + 506, + 725 + ], + "score": 1.0, + "content": "Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 16259–16268, 2021.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 24, + "bbox_fs": [ + 104, + 74, + 507, + 725 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 71, + 156, + 84 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 158, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 158, + 85 + ], + "score": 1.0, + "content": "Checklist", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 131, + 96, + 208, + 107 + ], + "lines": [ + { + "bbox": [ + 129, + 94, + 210, + 109 + ], + "spans": [ + { + "bbox": [ + 129, + 94, + 210, + 109 + ], + "score": 1.0, + "content": "1. For all authors...", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 146, + 110, + 505, + 194 + ], + "lines": [ + { + "bbox": [ + 145, + 109, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 145, + 109, + 505, + 123 + ], + "score": 1.0, + "content": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 162, + 121, + 288, + 133 + ], + "spans": [ + { + "bbox": [ + 162, + 121, + 288, + 133 + ], + "score": 1.0, + "content": "contributions and scope? [Yes]", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 144, + 133, + 448, + 147 + ], + "spans": [ + { + "bbox": [ + 144, + 133, + 448, + 147 + ], + "score": 1.0, + "content": "(b) Did you describe the limitations of your work? [Yes] Please see Sec. 6.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 146, + 146, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 146, + 146, + 505, + 160 + ], + "score": 1.0, + "content": "(c) Did you discuss any potential negative societal impacts of your work? [Yes] Please see", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 161, + 158, + 192, + 170 + ], + "spans": [ + { + "bbox": [ + 161, + 158, + 192, + 170 + ], + "score": 1.0, + "content": "Sec. 6.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 146, + 170, + 505, + 184 + ], + "spans": [ + { + "bbox": [ + 146, + 170, + 505, + 184 + ], + "score": 1.0, + "content": "(d) Have you read the ethics review guidelines and ensured that your paper conforms to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 162, + 181, + 214, + 195 + ], + "spans": [ + { + "bbox": [ + 162, + 181, + 214, + 195 + ], + "score": 1.0, + "content": "them? [Yes]", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 131, + 197, + 302, + 208 + ], + "lines": [ + { + "bbox": [ + 129, + 196, + 304, + 210 + ], + "spans": [ + { + "bbox": [ + 129, + 196, + 304, + 210 + ], + "score": 1.0, + "content": "2. If you are including theoretical results...", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 146, + 212, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 145, + 211, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 145, + 211, + 506, + 225 + ], + "score": 1.0, + "content": "(a) Did you state the full set of assumptions of all theoretical results? [N/A] We did not", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 161, + 222, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 161, + 222, + 505, + 236 + ], + "score": 1.0, + "content": "derive novel theoretical results. We rather propose a novel generative model of 3D", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 161, + 232, + 195, + 248 + ], + "spans": [ + { + "bbox": [ + 161, + 232, + 195, + 248 + ], + "score": 1.0, + "content": "shapes.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 144, + 245, + 424, + 261 + ], + "spans": [ + { + "bbox": [ + 144, + 245, + 424, + 261 + ], + "score": 1.0, + "content": "(b) Did you include complete proofs of all theoretical results? [N/A]", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 131, + 262, + 241, + 273 + ], + "lines": [ + { + "bbox": [ + 129, + 259, + 242, + 275 + ], + "spans": [ + { + "bbox": [ + 129, + 259, + 242, + 275 + ], + "score": 1.0, + "content": "3. If you ran experiments...", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 146, + 276, + 506, + 435 + ], + "lines": [ + { + "bbox": [ + 146, + 276, + 507, + 289 + ], + "spans": [ + { + "bbox": [ + 146, + 276, + 507, + 289 + ], + "score": 1.0, + "content": "(a) Did you include the code, data, and instructions needed to reproduce the main experi-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 161, + 287, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 161, + 287, + 506, + 299 + ], + "score": 1.0, + "content": "mental results (either in the supplemental material or as a URL)? [No] We will release", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 161, + 298, + 506, + 310 + ], + "spans": [ + { + "bbox": [ + 161, + 298, + 506, + 310 + ], + "score": 1.0, + "content": "code and instructions to reproduce all experiments upon acceptance of the manuscript.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 161, + 308, + 504, + 322 + ], + "spans": [ + { + "bbox": [ + 161, + 308, + 504, + 322 + ], + "score": 1.0, + "content": "The internal guidelines of our institution prevent us from releasing code at this stage.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 146, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 146, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 161, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 161, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "were chosen)? [Yes] We provide all model training and evaluation details in the App. D,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 161, + 343, + 286, + 356 + ], + "spans": [ + { + "bbox": [ + 161, + 343, + 286, + 356 + ], + "score": 1.0, + "content": "including all hyperparameters.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 147, + 356, + 507, + 369 + ], + "spans": [ + { + "bbox": [ + 147, + 356, + 507, + 369 + ], + "score": 1.0, + "content": "(c) Did you report error bars (e.g., with respect to the random seed after running ex-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 161, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 161, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "periments multiple times)? [No] Following standard conventions in the related 3D", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 161, + 378, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 161, + 378, + 506, + 390 + ], + "score": 1.0, + "content": "generative modeling literature, we do not report error bars. Furthermore, we avoid", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 161, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 161, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "running similar setups repeatedly to save computational resources (our main models", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 162, + 400, + 486, + 412 + ], + "spans": [ + { + "bbox": [ + 162, + 400, + 486, + 412 + ], + "score": 1.0, + "content": "are quite large and require substantial GPU resources for training, see App. E.9).", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 146, + 411, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 146, + 411, + 505, + 426 + ], + "score": 1.0, + "content": "(d) Did you include the total amount of compute and the type of resources used (e.g., type", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 162, + 423, + 453, + 436 + ], + "spans": [ + { + "bbox": [ + 162, + 423, + 453, + 436 + ], + "score": 1.0, + "content": "of GPUs, internal cluster, or cloud provider)? [Yes] Please see App. E.9.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 136, + 438, + 504, + 451 + ], + "lines": [ + { + "bbox": [ + 133, + 437, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 133, + 437, + 506, + 453 + ], + "score": 1.0, + "content": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets...", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 145, + 453, + 506, + 675 + ], + "lines": [ + { + "bbox": [ + 145, + 453, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 145, + 453, + 506, + 465 + ], + "score": 1.0, + "content": "(a) If your work uses existing assets, did you cite the creators? [Yes] For baseline compar-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 161, + 464, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 161, + 464, + 506, + 476 + ], + "score": 1.0, + "content": "isons we run publicly available code from previous, publicly available papers, which", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 161, + 474, + 433, + 488 + ], + "spans": [ + { + "bbox": [ + 161, + 474, + 433, + 488 + ], + "score": 1.0, + "content": "we cite. We also use various datasets. Here, we present a summary.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 162, + 487, + 376, + 498 + ], + "spans": [ + { + "bbox": [ + 162, + 487, + 376, + 498 + ], + "score": 1.0, + "content": "To compare to baselines, we use the following codes:", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 167, + 499, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 167, + 499, + 506, + 512 + ], + "score": 1.0, + "content": "• r-GAN, l-GAN [2]: https://github.com/optas/latent_3d_points (MIT", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 175, + 510, + 214, + 522 + ], + "spans": [ + { + "bbox": [ + 175, + 510, + 214, + 522 + ], + "score": 1.0, + "content": "License)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 167, + 520, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 167, + 520, + 506, + 535 + ], + "score": 1.0, + "content": "• PointFlow [31]: https://github.com/stevenygd/PointFlow (MIT License)", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 168, + 532, + 425, + 547 + ], + "spans": [ + { + "bbox": [ + 168, + 532, + 425, + 547 + ], + "score": 1.0, + "content": "• SoftFlow [32]: https://github.com/ANLGBOY/SoftFlow", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 168, + 545, + 468, + 558 + ], + "spans": [ + { + "bbox": [ + 168, + 545, + 468, + 558 + ], + "score": 1.0, + "content": "• Set-VAE [29]: https://github.com/jw9730/setvae (MIT License)", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 167, + 557, + 450, + 570 + ], + "spans": [ + { + "bbox": [ + 167, + 557, + 450, + 570 + ], + "score": 1.0, + "content": "• DPF-NET [33]: https://github.com/Regenerator/dpf-nets", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 167, + 569, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 167, + 569, + 506, + 582 + ], + "score": 1.0, + "content": "• DPM [47]: https://github.com/luost26/diffusion-point-cloud (MIT", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 175, + 580, + 214, + 593 + ], + "spans": [ + { + "bbox": [ + 175, + 580, + 214, + 593 + ], + "score": 1.0, + "content": "License)", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 167, + 591, + 464, + 606 + ], + "spans": [ + { + "bbox": [ + 167, + 591, + 464, + 606 + ], + "score": 1.0, + "content": "• PVD [46]: https://github.com/alexzhou907/PVD (MIT License)", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 167, + 603, + 492, + 618 + ], + "spans": [ + { + "bbox": [ + 167, + 603, + 492, + 618 + ], + "score": 1.0, + "content": "• ShapeGF [45]: https://github.com/RuojinCai/ShapeGF (MIT License)", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 167, + 615, + 481, + 630 + ], + "spans": [ + { + "bbox": [ + 167, + 615, + 481, + 630 + ], + "score": 1.0, + "content": "• SP-GAN [19]: https://github.com/liruihui/sp-gan (MIT License)", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 167, + 627, + 455, + 641 + ], + "spans": [ + { + "bbox": [ + 167, + 627, + 455, + 641 + ], + "score": 1.0, + "content": "• PDGN [52]: https://github.com/fpthink/PDGN (MIT License)", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 167, + 640, + 507, + 652 + ], + "spans": [ + { + "bbox": [ + 167, + 640, + 507, + 652 + ], + "score": 1.0, + "content": "• IM-GAN [7]: https://github.com/czq142857/implicit-decoder (MIT li-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 176, + 652, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 176, + 652, + 505, + 663 + ], + "score": 1.0, + "content": "cense) and https://github.com/czq142857/IM-NET-pytorch (MIT license)", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 168, + 662, + 440, + 676 + ], + "spans": [ + { + "bbox": [ + 168, + 662, + 440, + 676 + ], + "score": 1.0, + "content": "• GCA [43]: https://github.com/96lives/gca (MIT license)", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 163, + 677, + 328, + 687 + ], + "lines": [ + { + "bbox": [ + 162, + 675, + 329, + 689 + ], + "spans": [ + { + "bbox": [ + 162, + 675, + 329, + 689 + ], + "score": 1.0, + "content": "We use further codebases in other places:", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 49 + }, + { + "type": "text", + "bbox": [ + 169, + 689, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 167, + 687, + 508, + 702 + ], + "spans": [ + { + "bbox": [ + 167, + 687, + 458, + 701 + ], + "score": 1.0, + "content": "• We use the MitSuba renderer for visualizations [125]:", + "type": "text" + }, + { + "bbox": [ + 472, + 687, + 508, + 702 + ], + "score": 1.0, + "content": "https:", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 177, + 699, + 507, + 713 + ], + "spans": [ + { + "bbox": [ + 177, + 700, + 379, + 712 + ], + "score": 1.0, + "content": "//github.com/mitsuba-renderer/mitsuba2", + "type": "text" + }, + { + "bbox": [ + 396, + 700, + 437, + 711 + ], + "score": 1.0, + "content": "(License:", + "type": "text" + }, + { + "bbox": [ + 472, + 699, + 507, + 713 + ], + "score": 1.0, + "content": "https:", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 177, + 710, + 489, + 722 + ], + "spans": [ + { + "bbox": [ + 177, + 710, + 489, + 722 + ], + "score": 1.0, + "content": "//github.com/mitsuba-renderer/mitsuba2/blob/master/LICENSE),", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 51 + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 742, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 71, + 156, + 84 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 158, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 158, + 85 + ], + "score": 1.0, + "content": "Checklist", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 131, + 96, + 208, + 107 + ], + "lines": [ + { + "bbox": [ + 129, + 94, + 210, + 109 + ], + "spans": [ + { + "bbox": [ + 129, + 94, + 210, + 109 + ], + "score": 1.0, + "content": "1. For all authors...", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 129, + 94, + 210, + 109 + ] + }, + { + "type": "list", + "bbox": [ + 146, + 110, + 505, + 194 + ], + "lines": [ + { + "bbox": [ + 145, + 109, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 145, + 109, + 505, + 123 + ], + "score": 1.0, + "content": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 121, + 288, + 133 + ], + "spans": [ + { + "bbox": [ + 162, + 121, + 288, + 133 + ], + "score": 1.0, + "content": "contributions and scope? [Yes]", + "type": "text" + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 144, + 133, + 448, + 147 + ], + "spans": [ + { + "bbox": [ + 144, + 133, + 448, + 147 + ], + "score": 1.0, + "content": "(b) Did you describe the limitations of your work? [Yes] Please see Sec. 6.", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 146, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 146, + 146, + 505, + 160 + ], + "score": 1.0, + "content": "(c) Did you discuss any potential negative societal impacts of your work? [Yes] Please see", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 158, + 192, + 170 + ], + "spans": [ + { + "bbox": [ + 161, + 158, + 192, + 170 + ], + "score": 1.0, + "content": "Sec. 6.", + "type": "text" + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 170, + 505, + 184 + ], + "spans": [ + { + "bbox": [ + 146, + 170, + 505, + 184 + ], + "score": 1.0, + "content": "(d) Have you read the ethics review guidelines and ensured that your paper conforms to", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 181, + 214, + 195 + ], + "spans": [ + { + "bbox": [ + 162, + 181, + 214, + 195 + ], + "score": 1.0, + "content": "them? [Yes]", + "type": "text" + } + ], + "index": 8, + "is_list_end_line": true + } + ], + "index": 5, + "bbox_fs": [ + 144, + 109, + 505, + 195 + ] + }, + { + "type": "text", + "bbox": [ + 131, + 197, + 302, + 208 + ], + "lines": [ + { + "bbox": [ + 129, + 196, + 304, + 210 + ], + "spans": [ + { + "bbox": [ + 129, + 196, + 304, + 210 + ], + "score": 1.0, + "content": "2. If you are including theoretical results...", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 129, + 196, + 304, + 210 + ] + }, + { + "type": "list", + "bbox": [ + 146, + 212, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 145, + 211, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 145, + 211, + 506, + 225 + ], + "score": 1.0, + "content": "(a) Did you state the full set of assumptions of all theoretical results? [N/A] We did not", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 222, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 161, + 222, + 505, + 236 + ], + "score": 1.0, + "content": "derive novel theoretical results. We rather propose a novel generative model of 3D", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 161, + 232, + 195, + 248 + ], + "spans": [ + { + "bbox": [ + 161, + 232, + 195, + 248 + ], + "score": 1.0, + "content": "shapes.", + "type": "text" + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 144, + 245, + 424, + 261 + ], + "spans": [ + { + "bbox": [ + 144, + 245, + 424, + 261 + ], + "score": 1.0, + "content": "(b) Did you include complete proofs of all theoretical results? [N/A]", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 11.5, + "bbox_fs": [ + 144, + 211, + 506, + 261 + ] + }, + { + "type": "text", + "bbox": [ + 131, + 262, + 241, + 273 + ], + "lines": [ + { + "bbox": [ + 129, + 259, + 242, + 275 + ], + "spans": [ + { + "bbox": [ + 129, + 259, + 242, + 275 + ], + "score": 1.0, + "content": "3. If you ran experiments...", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 129, + 259, + 242, + 275 + ] + }, + { + "type": "list", + "bbox": [ + 146, + 276, + 506, + 435 + ], + "lines": [ + { + "bbox": [ + 146, + 276, + 507, + 289 + ], + "spans": [ + { + "bbox": [ + 146, + 276, + 507, + 289 + ], + "score": 1.0, + "content": "(a) Did you include the code, data, and instructions needed to reproduce the main experi-", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 287, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 161, + 287, + 506, + 299 + ], + "score": 1.0, + "content": "mental results (either in the supplemental material or as a URL)? [No] We will release", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 161, + 298, + 506, + 310 + ], + "spans": [ + { + "bbox": [ + 161, + 298, + 506, + 310 + ], + "score": 1.0, + "content": "code and instructions to reproduce all experiments upon acceptance of the manuscript.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 161, + 308, + 504, + 322 + ], + "spans": [ + { + "bbox": [ + 161, + 308, + 504, + 322 + ], + "score": 1.0, + "content": "The internal guidelines of our institution prevent us from releasing code at this stage.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 146, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 146, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 161, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "were chosen)? [Yes] We provide all model training and evaluation details in the App. D,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 161, + 343, + 286, + 356 + ], + "spans": [ + { + "bbox": [ + 161, + 343, + 286, + 356 + ], + "score": 1.0, + "content": "including all hyperparameters.", + "type": "text" + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 147, + 356, + 507, + 369 + ], + "spans": [ + { + "bbox": [ + 147, + 356, + 507, + 369 + ], + "score": 1.0, + "content": "(c) Did you report error bars (e.g., with respect to the random seed after running ex-", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 161, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "periments multiple times)? [No] Following standard conventions in the related 3D", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 161, + 378, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 161, + 378, + 506, + 390 + ], + "score": 1.0, + "content": "generative modeling literature, we do not report error bars. Furthermore, we avoid", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 161, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 161, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "running similar setups repeatedly to save computational resources (our main models", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 162, + 400, + 486, + 412 + ], + "spans": [ + { + "bbox": [ + 162, + 400, + 486, + 412 + ], + "score": 1.0, + "content": "are quite large and require substantial GPU resources for training, see App. E.9).", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 411, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 146, + 411, + 505, + 426 + ], + "score": 1.0, + "content": "(d) Did you include the total amount of compute and the type of resources used (e.g., type", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 423, + 453, + 436 + ], + "spans": [ + { + "bbox": [ + 162, + 423, + 453, + 436 + ], + "score": 1.0, + "content": "of GPUs, internal cluster, or cloud provider)? [Yes] Please see App. E.9.", + "type": "text" + } + ], + "index": 28, + "is_list_end_line": true + } + ], + "index": 21.5, + "bbox_fs": [ + 146, + 276, + 507, + 436 + ] + }, + { + "type": "text", + "bbox": [ + 136, + 438, + 504, + 451 + ], + "lines": [ + { + "bbox": [ + 133, + 437, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 133, + 437, + 506, + 453 + ], + "score": 1.0, + "content": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets...", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 133, + 437, + 506, + 453 + ] + }, + { + "type": "text", + "bbox": [ + 145, + 453, + 506, + 675 + ], + "lines": [ + { + "bbox": [ + 145, + 453, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 145, + 453, + 506, + 465 + ], + "score": 1.0, + "content": "(a) If your work uses existing assets, did you cite the creators? [Yes] For baseline compar-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 161, + 464, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 161, + 464, + 506, + 476 + ], + "score": 1.0, + "content": "isons we run publicly available code from previous, publicly available papers, which", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 161, + 474, + 433, + 488 + ], + "spans": [ + { + "bbox": [ + 161, + 474, + 433, + 488 + ], + "score": 1.0, + "content": "we cite. We also use various datasets. Here, we present a summary.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 162, + 487, + 376, + 498 + ], + "spans": [ + { + "bbox": [ + 162, + 487, + 376, + 498 + ], + "score": 1.0, + "content": "To compare to baselines, we use the following codes:", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 167, + 499, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 167, + 499, + 506, + 512 + ], + "score": 1.0, + "content": "• r-GAN, l-GAN [2]: https://github.com/optas/latent_3d_points (MIT", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 175, + 510, + 214, + 522 + ], + "spans": [ + { + "bbox": [ + 175, + 510, + 214, + 522 + ], + "score": 1.0, + "content": "License)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 167, + 520, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 167, + 520, + 506, + 535 + ], + "score": 1.0, + "content": "• PointFlow [31]: https://github.com/stevenygd/PointFlow (MIT License)", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 168, + 532, + 425, + 547 + ], + "spans": [ + { + "bbox": [ + 168, + 532, + 425, + 547 + ], + "score": 1.0, + "content": "• SoftFlow [32]: https://github.com/ANLGBOY/SoftFlow", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 168, + 545, + 468, + 558 + ], + "spans": [ + { + "bbox": [ + 168, + 545, + 468, + 558 + ], + "score": 1.0, + "content": "• Set-VAE [29]: https://github.com/jw9730/setvae (MIT License)", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 167, + 557, + 450, + 570 + ], + "spans": [ + { + "bbox": [ + 167, + 557, + 450, + 570 + ], + "score": 1.0, + "content": "• DPF-NET [33]: https://github.com/Regenerator/dpf-nets", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 167, + 569, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 167, + 569, + 506, + 582 + ], + "score": 1.0, + "content": "• DPM [47]: https://github.com/luost26/diffusion-point-cloud (MIT", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 175, + 580, + 214, + 593 + ], + "spans": [ + { + "bbox": [ + 175, + 580, + 214, + 593 + ], + "score": 1.0, + "content": "License)", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 167, + 591, + 464, + 606 + ], + "spans": [ + { + "bbox": [ + 167, + 591, + 464, + 606 + ], + "score": 1.0, + "content": "• PVD [46]: https://github.com/alexzhou907/PVD (MIT License)", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 167, + 603, + 492, + 618 + ], + "spans": [ + { + "bbox": [ + 167, + 603, + 492, + 618 + ], + "score": 1.0, + "content": "• ShapeGF [45]: https://github.com/RuojinCai/ShapeGF (MIT License)", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 167, + 615, + 481, + 630 + ], + "spans": [ + { + "bbox": [ + 167, + 615, + 481, + 630 + ], + "score": 1.0, + "content": "• SP-GAN [19]: https://github.com/liruihui/sp-gan (MIT License)", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 167, + 627, + 455, + 641 + ], + "spans": [ + { + "bbox": [ + 167, + 627, + 455, + 641 + ], + "score": 1.0, + "content": "• PDGN [52]: https://github.com/fpthink/PDGN (MIT License)", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 167, + 640, + 507, + 652 + ], + "spans": [ + { + "bbox": [ + 167, + 640, + 507, + 652 + ], + "score": 1.0, + "content": "• IM-GAN [7]: https://github.com/czq142857/implicit-decoder (MIT li-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 176, + 652, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 176, + 652, + 505, + 663 + ], + "score": 1.0, + "content": "cense) and https://github.com/czq142857/IM-NET-pytorch (MIT license)", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 168, + 662, + 440, + 676 + ], + "spans": [ + { + "bbox": [ + 168, + 662, + 440, + 676 + ], + "score": 1.0, + "content": "• GCA [43]: https://github.com/96lives/gca (MIT license)", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 39, + "bbox_fs": [ + 145, + 453, + 507, + 676 + ] + }, + { + "type": "text", + "bbox": [ + 163, + 677, + 328, + 687 + ], + "lines": [ + { + "bbox": [ + 162, + 675, + 329, + 689 + ], + "spans": [ + { + "bbox": [ + 162, + 675, + 329, + 689 + ], + "score": 1.0, + "content": "We use further codebases in other places:", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 49, + "bbox_fs": [ + 162, + 675, + 329, + 689 + ] + }, + { + "type": "text", + "bbox": [ + 169, + 689, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 167, + 687, + 508, + 702 + ], + "spans": [ + { + "bbox": [ + 167, + 687, + 458, + 701 + ], + "score": 1.0, + "content": "• We use the MitSuba renderer for visualizations [125]:", + "type": "text" + }, + { + "bbox": [ + 472, + 687, + 508, + 702 + ], + "score": 1.0, + "content": "https:", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 177, + 699, + 507, + 713 + ], + "spans": [ + { + "bbox": [ + 177, + 700, + 379, + 712 + ], + "score": 1.0, + "content": "//github.com/mitsuba-renderer/mitsuba2", + "type": "text" + }, + { + "bbox": [ + 396, + 700, + 437, + 711 + ], + "score": 1.0, + "content": "(License:", + "type": "text" + }, + { + "bbox": [ + 472, + 699, + 507, + 713 + ], + "score": 1.0, + "content": "https:", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 177, + 710, + 489, + 722 + ], + "spans": [ + { + "bbox": [ + 177, + 710, + 489, + 722 + ], + "score": 1.0, + "content": "//github.com/mitsuba-renderer/mitsuba2/blob/master/LICENSE),", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 51, + "bbox_fs": [ + 167, + 687, + 508, + 722 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 169, + 73, + 506, + 174 + ], + "lines": [ + { + "bbox": [ + 176, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 176, + 72, + 506, + 85 + ], + "score": 1.0, + "content": "and the code to generate the scene discription files for MitSuba [31]:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 176, + 83, + 439, + 96 + ], + "spans": [ + { + "bbox": [ + 176, + 83, + 439, + 96 + ], + "score": 1.0, + "content": "https://github.com/zekunhao1995/PointFlowRenderer.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 168, + 94, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 168, + 94, + 505, + 108 + ], + "score": 1.0, + "content": "• We rely on SAP [68] for mesh generation with the code at https://github.com/", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 176, + 106, + 410, + 118 + ], + "spans": [ + { + "bbox": [ + 176, + 106, + 410, + 118 + ], + "score": 1.0, + "content": "autonomousvision/shape_as_points (MIT License).", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 168, + 117, + 507, + 130 + ], + "spans": [ + { + "bbox": [ + 168, + 117, + 507, + 130 + ], + "score": 1.0, + "content": "• For calculating the evaluation metrics, we use the implementation for CD at https:", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 176, + 128, + 506, + 142 + ], + "spans": [ + { + "bbox": [ + 176, + 128, + 506, + 142 + ], + "score": 1.0, + "content": "//github.com/ThibaultGROUEIX/ChamferDistancePytorch (MIT License)", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 176, + 140, + 465, + 152 + ], + "spans": [ + { + "bbox": [ + 176, + 140, + 465, + 152 + ], + "score": 1.0, + "content": "and for EMD at https://github.com/daerduoCarey/PyTorchEMD.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 168, + 150, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 168, + 150, + 506, + 165 + ], + "score": 1.0, + "content": "• We use Text2Mesh [49] for per-sample text-driven texture synthesis: https://", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 176, + 163, + 392, + 175 + ], + "spans": [ + { + "bbox": [ + 176, + 163, + 392, + 175 + ], + "score": 1.0, + "content": "github.com/threedle/text2mesh (MIT License)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 163, + 176, + 318, + 187 + ], + "lines": [ + { + "bbox": [ + 161, + 174, + 319, + 189 + ], + "spans": [ + { + "bbox": [ + 161, + 174, + 319, + 189 + ], + "score": 1.0, + "content": "We also rely on the following datasets:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 168, + 189, + 506, + 312 + ], + "lines": [ + { + "bbox": [ + 167, + 186, + 504, + 202 + ], + "spans": [ + { + "bbox": [ + 167, + 186, + 504, + 202 + ], + "score": 1.0, + "content": "• ShapeNet [104]. Its terms of use can be found at https://shapenet.org/", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 176, + 200, + 208, + 211 + ], + "spans": [ + { + "bbox": [ + 176, + 200, + 208, + 211 + ], + "score": 1.0, + "content": "terms.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 167, + 210, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 167, + 210, + 505, + 224 + ], + "score": 1.0, + "content": "• The Cars dataset [126] from http://ai.stanford.edu/~jkrause/cars/car_", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 176, + 222, + 507, + 235 + ], + "spans": [ + { + "bbox": [ + 176, + 222, + 507, + 235 + ], + "score": 1.0, + "content": "dataset.html with ImageNet License: https://image-net.org/download.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 175, + 234, + 198, + 246 + ], + "spans": [ + { + "bbox": [ + 175, + 234, + 198, + 246 + ], + "score": 1.0, + "content": "php.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 168, + 244, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 168, + 244, + 506, + 258 + ], + "score": 1.0, + "content": "• The TurboSquid data repository, https://www.turbosquid.com. We obtained", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 175, + 256, + 319, + 267 + ], + "spans": [ + { + "bbox": [ + 175, + 256, + 319, + 267 + ], + "score": 1.0, + "content": "a custom license from TurboSquid.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 168, + 267, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 168, + 267, + 343, + 281 + ], + "score": 1.0, + "content": "• Redwood 3DScan Dataset [127]:", + "type": "text" + }, + { + "bbox": [ + 360, + 267, + 505, + 281 + ], + "score": 1.0, + "content": "https://github.com/isl-org/", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 176, + 279, + 322, + 291 + ], + "spans": [ + { + "bbox": [ + 176, + 279, + 322, + 291 + ], + "score": 1.0, + "content": "redwood-3dscan (Public Domain)", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 167, + 289, + 507, + 303 + ], + "spans": [ + { + "bbox": [ + 167, + 289, + 507, + 303 + ], + "score": 1.0, + "content": "• Pix3D [128]: https://github.com/xingyuansun/pix3d. (Creative Com-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 177, + 302, + 355, + 313 + ], + "spans": [ + { + "bbox": [ + 177, + 302, + 355, + 313 + ], + "score": 1.0, + "content": "mons Attribution 4.0 International License).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 146, + 315, + 505, + 430 + ], + "lines": [ + { + "bbox": [ + 146, + 315, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 146, + 315, + 505, + 327 + ], + "score": 1.0, + "content": "(b) Did you mention the license of the assets? [Yes] In App. E.8, we mention the licenses", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 162, + 324, + 333, + 339 + ], + "spans": [ + { + "bbox": [ + 162, + 324, + 333, + 339 + ], + "score": 1.0, + "content": "of the codes and other assets we are using.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 146, + 337, + 503, + 352 + ], + "spans": [ + { + "bbox": [ + 146, + 337, + 503, + 352 + ], + "score": 1.0, + "content": "(c) Did you include any new assets either in the supplemental material or as a URL? [No]", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 145, + 352, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 145, + 352, + 505, + 364 + ], + "score": 1.0, + "content": "(d) Did you discuss whether and how consent was obtained from people whose data you’re", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 162, + 362, + 253, + 375 + ], + "spans": [ + { + "bbox": [ + 162, + 362, + 253, + 375 + ], + "score": 1.0, + "content": "using/curating? [N/A]", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 146, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 146, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "(e) Did you discuss whether the data you are using/curating contains personally identifiable", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 162, + 386, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 162, + 386, + 505, + 398 + ], + "score": 1.0, + "content": "information or offensive content? [No] We are primarily using the publicly available", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 162, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 162, + 396, + 505, + 410 + ], + "score": 1.0, + "content": "ShapeNet [104] dataset, which has been widely used in the generative modeling", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 162, + 408, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 162, + 408, + 505, + 420 + ], + "score": 1.0, + "content": "literature as standard benchmark. It only consists of simple 3D models of shapes such", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 162, + 419, + 281, + 431 + ], + "spans": [ + { + "bbox": [ + 162, + 419, + 281, + 431 + ], + "score": 1.0, + "content": "as airplanes, chairs, cars, etc.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 131, + 434, + 431, + 446 + ], + "lines": [ + { + "bbox": [ + 128, + 432, + 433, + 448 + ], + "spans": [ + { + "bbox": [ + 128, + 432, + 433, + 448 + ], + "score": 1.0, + "content": "5. If you used crowdsourcing or conducted research with human subjects...", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 146, + 448, + 505, + 519 + ], + "lines": [ + { + "bbox": [ + 145, + 448, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 145, + 448, + 506, + 461 + ], + "score": 1.0, + "content": "(a) Did you include the full text of instructions given to participants and screenshots, if", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 161, + 459, + 237, + 472 + ], + "spans": [ + { + "bbox": [ + 161, + 459, + 237, + 472 + ], + "score": 1.0, + "content": "applicable? [N/A]", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 145, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 145, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "(b) Did you describe any potential participant risks, with links to Institutional Review", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 161, + 483, + 342, + 496 + ], + "spans": [ + { + "bbox": [ + 161, + 483, + 342, + 496 + ], + "score": 1.0, + "content": "Board (IRB) approvals, if applicable? [N/A]", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 146, + 496, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 146, + 496, + 505, + 509 + ], + "score": 1.0, + "content": "(c) Did you include the estimated hourly wage paid to participants and the total amount", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 161, + 508, + 333, + 520 + ], + "spans": [ + { + "bbox": [ + 161, + 508, + 333, + 520 + ], + "score": 1.0, + "content": "spent on participant compensation? [N/A]", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34.5 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "19", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 169, + 73, + 506, + 174 + ], + "lines": [ + { + "bbox": [ + 176, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 176, + 72, + 506, + 85 + ], + "score": 1.0, + "content": "and the code to generate the scene discription files for MitSuba [31]:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 176, + 83, + 439, + 96 + ], + "spans": [ + { + "bbox": [ + 176, + 83, + 439, + 96 + ], + "score": 1.0, + "content": "https://github.com/zekunhao1995/PointFlowRenderer.", + "type": "text" + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 168, + 94, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 168, + 94, + 505, + 108 + ], + "score": 1.0, + "content": "• We rely on SAP [68] for mesh generation with the code at https://github.com/", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 176, + 106, + 410, + 118 + ], + "spans": [ + { + "bbox": [ + 176, + 106, + 410, + 118 + ], + "score": 1.0, + "content": "autonomousvision/shape_as_points (MIT License).", + "type": "text" + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 168, + 117, + 507, + 130 + ], + "spans": [ + { + "bbox": [ + 168, + 117, + 507, + 130 + ], + "score": 1.0, + "content": "• For calculating the evaluation metrics, we use the implementation for CD at https:", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 176, + 128, + 506, + 142 + ], + "spans": [ + { + "bbox": [ + 176, + 128, + 506, + 142 + ], + "score": 1.0, + "content": "//github.com/ThibaultGROUEIX/ChamferDistancePytorch (MIT License)", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 176, + 140, + 465, + 152 + ], + "spans": [ + { + "bbox": [ + 176, + 140, + 465, + 152 + ], + "score": 1.0, + "content": "and for EMD at https://github.com/daerduoCarey/PyTorchEMD.", + "type": "text" + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 168, + 150, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 168, + 150, + 506, + 165 + ], + "score": 1.0, + "content": "• We use Text2Mesh [49] for per-sample text-driven texture synthesis: https://", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 176, + 163, + 392, + 175 + ], + "spans": [ + { + "bbox": [ + 176, + 163, + 392, + 175 + ], + "score": 1.0, + "content": "github.com/threedle/text2mesh (MIT License)", + "type": "text" + } + ], + "index": 8, + "is_list_end_line": true + } + ], + "index": 4, + "bbox_fs": [ + 168, + 72, + 507, + 175 + ] + }, + { + "type": "text", + "bbox": [ + 163, + 176, + 318, + 187 + ], + "lines": [ + { + "bbox": [ + 161, + 174, + 319, + 189 + ], + "spans": [ + { + "bbox": [ + 161, + 174, + 319, + 189 + ], + "score": 1.0, + "content": "We also rely on the following datasets:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 161, + 174, + 319, + 189 + ] + }, + { + "type": "list", + "bbox": [ + 168, + 189, + 506, + 312 + ], + "lines": [ + { + "bbox": [ + 167, + 186, + 504, + 202 + ], + "spans": [ + { + "bbox": [ + 167, + 186, + 504, + 202 + ], + "score": 1.0, + "content": "• ShapeNet [104]. Its terms of use can be found at https://shapenet.org/", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 176, + 200, + 208, + 211 + ], + "spans": [ + { + "bbox": [ + 176, + 200, + 208, + 211 + ], + "score": 1.0, + "content": "terms.", + "type": "text" + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 167, + 210, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 167, + 210, + 505, + 224 + ], + "score": 1.0, + "content": "• The Cars dataset [126] from http://ai.stanford.edu/~jkrause/cars/car_", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 176, + 222, + 507, + 235 + ], + "spans": [ + { + "bbox": [ + 176, + 222, + 507, + 235 + ], + "score": 1.0, + "content": "dataset.html with ImageNet License: https://image-net.org/download.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 175, + 234, + 198, + 246 + ], + "spans": [ + { + "bbox": [ + 175, + 234, + 198, + 246 + ], + "score": 1.0, + "content": "php.", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 168, + 244, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 168, + 244, + 506, + 258 + ], + "score": 1.0, + "content": "• The TurboSquid data repository, https://www.turbosquid.com. We obtained", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 175, + 256, + 319, + 267 + ], + "spans": [ + { + "bbox": [ + 175, + 256, + 319, + 267 + ], + "score": 1.0, + "content": "a custom license from TurboSquid.", + "type": "text" + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 168, + 267, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 168, + 267, + 343, + 281 + ], + "score": 1.0, + "content": "• Redwood 3DScan Dataset [127]:", + "type": "text" + }, + { + "bbox": [ + 360, + 267, + 505, + 281 + ], + "score": 1.0, + "content": "https://github.com/isl-org/", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 176, + 279, + 322, + 291 + ], + "spans": [ + { + "bbox": [ + 176, + 279, + 322, + 291 + ], + "score": 1.0, + "content": "redwood-3dscan (Public Domain)", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 167, + 289, + 507, + 303 + ], + "spans": [ + { + "bbox": [ + 167, + 289, + 507, + 303 + ], + "score": 1.0, + "content": "• Pix3D [128]: https://github.com/xingyuansun/pix3d. (Creative Com-", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 177, + 302, + 355, + 313 + ], + "spans": [ + { + "bbox": [ + 177, + 302, + 355, + 313 + ], + "score": 1.0, + "content": "mons Attribution 4.0 International License).", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 315, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 146, + 315, + 505, + 327 + ], + "score": 1.0, + "content": "(b) Did you mention the license of the assets? [Yes] In App. E.8, we mention the licenses", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 324, + 333, + 339 + ], + "spans": [ + { + "bbox": [ + 162, + 324, + 333, + 339 + ], + "score": 1.0, + "content": "of the codes and other assets we are using.", + "type": "text" + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 337, + 503, + 352 + ], + "spans": [ + { + "bbox": [ + 146, + 337, + 503, + 352 + ], + "score": 1.0, + "content": "(c) Did you include any new assets either in the supplemental material or as a URL? [No]", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 145, + 352, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 145, + 352, + 505, + 364 + ], + "score": 1.0, + "content": "(d) Did you discuss whether and how consent was obtained from people whose data you’re", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 362, + 253, + 375 + ], + "spans": [ + { + "bbox": [ + 162, + 362, + 253, + 375 + ], + "score": 1.0, + "content": "using/curating? [N/A]", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 146, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "(e) Did you discuss whether the data you are using/curating contains personally identifiable", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 386, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 162, + 386, + 505, + 398 + ], + "score": 1.0, + "content": "information or offensive content? [No] We are primarily using the publicly available", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 162, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 162, + 396, + 505, + 410 + ], + "score": 1.0, + "content": "ShapeNet [104] dataset, which has been widely used in the generative modeling", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 162, + 408, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 162, + 408, + 505, + 420 + ], + "score": 1.0, + "content": "literature as standard benchmark. It only consists of simple 3D models of shapes such", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 162, + 419, + 281, + 431 + ], + "spans": [ + { + "bbox": [ + 162, + 419, + 281, + 431 + ], + "score": 1.0, + "content": "as airplanes, chairs, cars, etc.", + "type": "text" + } + ], + "index": 30, + "is_list_end_line": true + } + ], + "index": 15, + "bbox_fs": [ + 167, + 186, + 507, + 313 + ] + }, + { + "type": "list", + "bbox": [ + 146, + 315, + 505, + 430 + ], + "lines": [], + "index": 25.5, + "bbox_fs": [ + 145, + 315, + 505, + 431 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 131, + 434, + 431, + 446 + ], + "lines": [ + { + "bbox": [ + 128, + 432, + 433, + 448 + ], + "spans": [ + { + "bbox": [ + 128, + 432, + 433, + 448 + ], + "score": 1.0, + "content": "5. If you used crowdsourcing or conducted research with human subjects...", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31, + "bbox_fs": [ + 128, + 432, + 433, + 448 + ] + }, + { + "type": "list", + "bbox": [ + 146, + 448, + 505, + 519 + ], + "lines": [ + { + "bbox": [ + 145, + 448, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 145, + 448, + 506, + 461 + ], + "score": 1.0, + "content": "(a) Did you include the full text of instructions given to participants and screenshots, if", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 459, + 237, + 472 + ], + "spans": [ + { + "bbox": [ + 161, + 459, + 237, + 472 + ], + "score": 1.0, + "content": "applicable? [N/A]", + "type": "text" + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 145, + 472, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 145, + 472, + 505, + 485 + ], + "score": 1.0, + "content": "(b) Did you describe any potential participant risks, with links to Institutional Review", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 483, + 342, + 496 + ], + "spans": [ + { + "bbox": [ + 161, + 483, + 342, + 496 + ], + "score": 1.0, + "content": "Board (IRB) approvals, if applicable? [N/A]", + "type": "text" + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 496, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 146, + 496, + 505, + 509 + ], + "score": 1.0, + "content": "(c) Did you include the estimated hourly wage paid to participants and the total amount", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 508, + 333, + 520 + ], + "spans": [ + { + "bbox": [ + 161, + 508, + 333, + 520 + ], + "score": 1.0, + "content": "spent on participant compensation? [N/A]", + "type": "text" + } + ], + "index": 37, + "is_list_end_line": true + } + ], + "index": 34.5, + "bbox_fs": [ + 145, + 448, + 506, + 520 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/dev/v54eUIayFh/images/0164e828eb13380270f067c363b0186d1bbcb1960acfff09d450029d44bfae3b.jpg b/parse/dev/v54eUIayFh/images/0164e828eb13380270f067c363b0186d1bbcb1960acfff09d450029d44bfae3b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..213189e427b2e2f94191568a02d7026a0e09cf68 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/0164e828eb13380270f067c363b0186d1bbcb1960acfff09d450029d44bfae3b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25e4bfdbe899fe7a43b59a5b41117ba0db007b75fd2a084fa67f54b5bece16f2 +size 18851 diff --git a/parse/dev/v54eUIayFh/images/05a4ebc5894b0db0cea1eaa2b852f1eb3fba973b98e17810371f725e3b5e3c51.jpg b/parse/dev/v54eUIayFh/images/05a4ebc5894b0db0cea1eaa2b852f1eb3fba973b98e17810371f725e3b5e3c51.jpg new file mode 100644 index 0000000000000000000000000000000000000000..80a59de479416ae14034d3b709847edbbef0e509 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/05a4ebc5894b0db0cea1eaa2b852f1eb3fba973b98e17810371f725e3b5e3c51.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c3d3fb659aeeab1eb998499c202e2d44337e5e2d26c5ca668fac64dd79119c6 +size 45030 diff --git a/parse/dev/v54eUIayFh/images/08c58339e14946f66088b5521c45f1e8932b94d016071608b8865306d9348e5f.jpg b/parse/dev/v54eUIayFh/images/08c58339e14946f66088b5521c45f1e8932b94d016071608b8865306d9348e5f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..758b1ffd41455014813b7170f2ea954710d147ef --- /dev/null +++ b/parse/dev/v54eUIayFh/images/08c58339e14946f66088b5521c45f1e8932b94d016071608b8865306d9348e5f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1ed5611082d1b266c48829c701a382198906f9cd0c0a86ac0dc261ef0510061 +size 40635 diff --git a/parse/dev/v54eUIayFh/images/0f48ffb5077bb8a7f0ec71d24de98c629e02d50df7b51f1664154ac9a5c945f5.jpg b/parse/dev/v54eUIayFh/images/0f48ffb5077bb8a7f0ec71d24de98c629e02d50df7b51f1664154ac9a5c945f5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..73bc7ca64db59dde22ee00ead7e3bc866340e99d --- /dev/null +++ b/parse/dev/v54eUIayFh/images/0f48ffb5077bb8a7f0ec71d24de98c629e02d50df7b51f1664154ac9a5c945f5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0dadf8b8c10f8580f68f98a35ab522d2552a1a388067fed3a38bcdc8a104c685 +size 121656 diff --git a/parse/dev/v54eUIayFh/images/0ff793313409af7320cb2ee71b0336a463c8f08a5e0364bf684fad7d31969317.jpg b/parse/dev/v54eUIayFh/images/0ff793313409af7320cb2ee71b0336a463c8f08a5e0364bf684fad7d31969317.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cb8715121ab5dfea100ec38f809188410cd97442 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/0ff793313409af7320cb2ee71b0336a463c8f08a5e0364bf684fad7d31969317.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a6efe274e0d02c88cb8008ce3c86e820563ff4ac66c06d4afc39b123e91982b +size 14030 diff --git a/parse/dev/v54eUIayFh/images/16698d49a4b5517d756c624b1810c4b1688ca335a2f7854252080f45c1e5e283.jpg b/parse/dev/v54eUIayFh/images/16698d49a4b5517d756c624b1810c4b1688ca335a2f7854252080f45c1e5e283.jpg new file mode 100644 index 0000000000000000000000000000000000000000..33b8fbc343b01501fa8d1f562e5fecd6008520c2 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/16698d49a4b5517d756c624b1810c4b1688ca335a2f7854252080f45c1e5e283.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0dd6cf63ce0c13316f85e938527d3c3f8da5ecfe1e3e0cac25bb67c6e3610d27 +size 42283 diff --git a/parse/dev/v54eUIayFh/images/169a4907042a8a0f146ea427fe8b9df037cad21d932aaa9d2b192489c82562b1.jpg b/parse/dev/v54eUIayFh/images/169a4907042a8a0f146ea427fe8b9df037cad21d932aaa9d2b192489c82562b1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4c0204d27e632bae0a52d7ba972d073c3be23697 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/169a4907042a8a0f146ea427fe8b9df037cad21d932aaa9d2b192489c82562b1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c7950299d553f3d075272a4ebfc2bd0fa6954fb73afa13e97bd77da0b868b4d +size 27085 diff --git a/parse/dev/v54eUIayFh/images/1fe38f8b80c193320b1b25b181e91c2293b6f9e87787b2cf78ff1bc057b56e01.jpg b/parse/dev/v54eUIayFh/images/1fe38f8b80c193320b1b25b181e91c2293b6f9e87787b2cf78ff1bc057b56e01.jpg new file mode 100644 index 0000000000000000000000000000000000000000..87970f665864c19026a683a975c6f33df8191ca4 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/1fe38f8b80c193320b1b25b181e91c2293b6f9e87787b2cf78ff1bc057b56e01.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:236692dbaa05bcf184eeb8e53f6ab1fdfca3d0946f99a9b6711238a35b4141a0 +size 21269 diff --git a/parse/dev/v54eUIayFh/images/2894b41f5846561381beb85496e0b282142f7b08075deb7087b257189fa6b884.jpg b/parse/dev/v54eUIayFh/images/2894b41f5846561381beb85496e0b282142f7b08075deb7087b257189fa6b884.jpg new file mode 100644 index 0000000000000000000000000000000000000000..43828bd8817c32ecd5bb25e5b1c07fc98708d2d0 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/2894b41f5846561381beb85496e0b282142f7b08075deb7087b257189fa6b884.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:72fbbed92f05c8fd98081b43c11b76f1835cbb3a987c48036199e3a246363f46 +size 7963 diff --git a/parse/dev/v54eUIayFh/images/2d355454522171fa2e1b083602cdc59ebab9a153e42804ce92e584efa68ebab9.jpg b/parse/dev/v54eUIayFh/images/2d355454522171fa2e1b083602cdc59ebab9a153e42804ce92e584efa68ebab9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..86a9b9181cef6199bbd75cf7f2e43a81e319ae16 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/2d355454522171fa2e1b083602cdc59ebab9a153e42804ce92e584efa68ebab9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e11cce0eab581c8e388f99bfc685aec4fd7eaf82033e54c640cdb9640521811 +size 144707 diff --git a/parse/dev/v54eUIayFh/images/406de263650c6bfb5fe8a6742a82c576f8bdf7771416a4ff2faecafc56ca8408.jpg b/parse/dev/v54eUIayFh/images/406de263650c6bfb5fe8a6742a82c576f8bdf7771416a4ff2faecafc56ca8408.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ed02e304720356daec1f98aa74c680fe0f28e2ff --- /dev/null +++ b/parse/dev/v54eUIayFh/images/406de263650c6bfb5fe8a6742a82c576f8bdf7771416a4ff2faecafc56ca8408.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6162ef906979fd0bd3514aeed5b050c991fb9feee3e15b82c6a0c995be0995a +size 34921 diff --git a/parse/dev/v54eUIayFh/images/47cd8d08dfba650e8d11972cc7c3f075e2af3fdcaee7764d368b5a4950bb88b7.jpg b/parse/dev/v54eUIayFh/images/47cd8d08dfba650e8d11972cc7c3f075e2af3fdcaee7764d368b5a4950bb88b7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..be477733d352ef49c5b23dc808f5a0148cb4e133 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/47cd8d08dfba650e8d11972cc7c3f075e2af3fdcaee7764d368b5a4950bb88b7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1428416c3d718f8d8c5f5dbeeeb408d1bc8c9d8a12db25f480d5df320236c42 +size 27027 diff --git a/parse/dev/v54eUIayFh/images/533f5d7e0c6cc374e12d4d3c734e3d41fc6ad14014462c57fa1def5de3219eb5.jpg b/parse/dev/v54eUIayFh/images/533f5d7e0c6cc374e12d4d3c734e3d41fc6ad14014462c57fa1def5de3219eb5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..50182e85878bb1253e122c35386dad0b5b9ae926 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/533f5d7e0c6cc374e12d4d3c734e3d41fc6ad14014462c57fa1def5de3219eb5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6b48f8d75e3aefd1d1095df353340cd240bef0ea1773eb8dde63cf2a61457c6 +size 16805 diff --git a/parse/dev/v54eUIayFh/images/6c0ea7ba2a2a2c99b0745134a484112af74d27e5d6570f3bad24e796d908e009.jpg b/parse/dev/v54eUIayFh/images/6c0ea7ba2a2a2c99b0745134a484112af74d27e5d6570f3bad24e796d908e009.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eb48aff8b57cd469d507e4ec561b18694c9a60d0 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/6c0ea7ba2a2a2c99b0745134a484112af74d27e5d6570f3bad24e796d908e009.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e653249a71fe317bca8fc612696e703564c274685eb30a312f9262590d81a456 +size 130227 diff --git a/parse/dev/v54eUIayFh/images/79a82e9968f139c9101cd625411b9d9d91949e9bac0b603c4ff327082f276f1f.jpg b/parse/dev/v54eUIayFh/images/79a82e9968f139c9101cd625411b9d9d91949e9bac0b603c4ff327082f276f1f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..53e2cd209a53c22a2f7f2792a68b891198ae7c6d --- /dev/null +++ b/parse/dev/v54eUIayFh/images/79a82e9968f139c9101cd625411b9d9d91949e9bac0b603c4ff327082f276f1f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e31e50ba1403826d639f75a04f1d41af41e07e06ba0c6de2603b90d7dcfa2b5c +size 25346 diff --git a/parse/dev/v54eUIayFh/images/864ebf88f6ec3183b5f05edf0269ae982cf62ec77abde0f6f63e4d7710948da3.jpg b/parse/dev/v54eUIayFh/images/864ebf88f6ec3183b5f05edf0269ae982cf62ec77abde0f6f63e4d7710948da3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..190b1a52a9875ccfbc4072da3fc80e711cd0909f --- /dev/null +++ b/parse/dev/v54eUIayFh/images/864ebf88f6ec3183b5f05edf0269ae982cf62ec77abde0f6f63e4d7710948da3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:272df17d4a10a768d7468558bcd3892ee251f6f0e1979604f18adadc7f2b3e49 +size 2756 diff --git a/parse/dev/v54eUIayFh/images/8b48892dcdf5997bd253d939ad73025ccc80eec99a968be71ac95455d87347cc.jpg b/parse/dev/v54eUIayFh/images/8b48892dcdf5997bd253d939ad73025ccc80eec99a968be71ac95455d87347cc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f5a61edc2283d96c6a7a7e9d994f2b4e83b4b3de --- /dev/null +++ b/parse/dev/v54eUIayFh/images/8b48892dcdf5997bd253d939ad73025ccc80eec99a968be71ac95455d87347cc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52a86babb60cb917f5e68425e685559cf8c00525a77e7bd42aa75685042553de +size 9910 diff --git a/parse/dev/v54eUIayFh/images/8e4c74d2cca91527f6427f7df6ce14f59247ff3cf6638d066c185853ca69c6df.jpg b/parse/dev/v54eUIayFh/images/8e4c74d2cca91527f6427f7df6ce14f59247ff3cf6638d066c185853ca69c6df.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ebb5d7ee077b54d1ba80214803c585f53dbdffc4 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/8e4c74d2cca91527f6427f7df6ce14f59247ff3cf6638d066c185853ca69c6df.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:debe6b7fd2904c69c6e00a33c65a52eb284651dca0b4b48a4082e6de50a24582 +size 245486 diff --git a/parse/dev/v54eUIayFh/images/9c7333bbce3405b4086a56ad84a32c1c3e9b3432694e06f759f5790b04f6ad72.jpg b/parse/dev/v54eUIayFh/images/9c7333bbce3405b4086a56ad84a32c1c3e9b3432694e06f759f5790b04f6ad72.jpg new file mode 100644 index 0000000000000000000000000000000000000000..da55e94d951ef7e4e63d6047a67ed22c348c42c3 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/9c7333bbce3405b4086a56ad84a32c1c3e9b3432694e06f759f5790b04f6ad72.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4db693278d9548a488441b5700426450550b8d6dda1b03d02f4976bfcaf3541 +size 38521 diff --git a/parse/dev/v54eUIayFh/images/9ecf91e675f8d8d800075a4ae88a33aef4033b041c54d7252883d8413fccc53c.jpg b/parse/dev/v54eUIayFh/images/9ecf91e675f8d8d800075a4ae88a33aef4033b041c54d7252883d8413fccc53c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3389be9d11e630acebabff7c3901750f294e5619 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/9ecf91e675f8d8d800075a4ae88a33aef4033b041c54d7252883d8413fccc53c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d77a8ad67311534f8bc8a73ab7554151848826c683ae578e5cc2b1c2da46e70d +size 175083 diff --git a/parse/dev/v54eUIayFh/images/a3270977f741084bdf043f1e398e79f8a4eda204077b26f6b699967ba6bf420a.jpg b/parse/dev/v54eUIayFh/images/a3270977f741084bdf043f1e398e79f8a4eda204077b26f6b699967ba6bf420a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..203c4adcc53b1bcf5d0744c2e5642e61daac494a --- /dev/null +++ b/parse/dev/v54eUIayFh/images/a3270977f741084bdf043f1e398e79f8a4eda204077b26f6b699967ba6bf420a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42616f3e838d5faec155d6eaed678d15370297dd1f055aa5a2b23fdab4f8ef35 +size 17994 diff --git a/parse/dev/v54eUIayFh/images/b9bc5fe441a84ca5ed0902d7991aa92463777b428224fad431090fe7457e7254.jpg b/parse/dev/v54eUIayFh/images/b9bc5fe441a84ca5ed0902d7991aa92463777b428224fad431090fe7457e7254.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7e5214616a5542474a2429938ea261dcf4e8ecaf --- /dev/null +++ b/parse/dev/v54eUIayFh/images/b9bc5fe441a84ca5ed0902d7991aa92463777b428224fad431090fe7457e7254.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8df7626cf6f8e9d5666b42b58c32f537abe68ee52e914b19e9330d34db3d39be +size 12959 diff --git a/parse/dev/v54eUIayFh/images/c113ff5911f48b8ab6744e8638c79653ea69d2209d8794f4319072c9f2846759.jpg b/parse/dev/v54eUIayFh/images/c113ff5911f48b8ab6744e8638c79653ea69d2209d8794f4319072c9f2846759.jpg new file mode 100644 index 0000000000000000000000000000000000000000..401c76e862bd637632177e901bfb6d33bd976134 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/c113ff5911f48b8ab6744e8638c79653ea69d2209d8794f4319072c9f2846759.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:309ee86f0200a0091cc555f790bfb71c8c307e9302ee16eae4ca0d7b3d1e75ed +size 29527 diff --git a/parse/dev/v54eUIayFh/images/c799b6b76ab35eaa3b515c74d2b485fd146a5a355e015d55c85a4189e876416e.jpg b/parse/dev/v54eUIayFh/images/c799b6b76ab35eaa3b515c74d2b485fd146a5a355e015d55c85a4189e876416e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b43dd3219fae780c1327b65873fa0f2db49e46fc --- /dev/null +++ b/parse/dev/v54eUIayFh/images/c799b6b76ab35eaa3b515c74d2b485fd146a5a355e015d55c85a4189e876416e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:97e83f7063e95538d0babb65eeb608f9e50e34f2b8cb3a45090718931e4cfc0c +size 9886 diff --git a/parse/dev/v54eUIayFh/images/ce300439a093037af0e22018b9d26d71a58799df23ce9342b111dfd41d49307c.jpg b/parse/dev/v54eUIayFh/images/ce300439a093037af0e22018b9d26d71a58799df23ce9342b111dfd41d49307c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..517a64408350b4e3683746252b59011836a1b4b6 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/ce300439a093037af0e22018b9d26d71a58799df23ce9342b111dfd41d49307c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ea4e66366389ac270ecb32f7086ee40acd97b36eb0054d622b61a69595eb4df +size 81749 diff --git a/parse/dev/v54eUIayFh/images/dc7804f648f95a2a22cf80a3473e45e1d7e39a8f68d1387d94119c16ed8279e3.jpg b/parse/dev/v54eUIayFh/images/dc7804f648f95a2a22cf80a3473e45e1d7e39a8f68d1387d94119c16ed8279e3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8f625766637a424fe468ad05d773fd4d3cea5382 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/dc7804f648f95a2a22cf80a3473e45e1d7e39a8f68d1387d94119c16ed8279e3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:229728f0fcb2d4221d0a2daa4d08401761720ec899a7529c330356a91c3ac8dc +size 154575 diff --git a/parse/dev/v54eUIayFh/images/ddfdf7bb249088e3f366e8b8ba221b8f472a1b46e62e984081e2fc92dd49c629.jpg b/parse/dev/v54eUIayFh/images/ddfdf7bb249088e3f366e8b8ba221b8f472a1b46e62e984081e2fc92dd49c629.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5b8f1da6148efbe25de536da3253a63dcfc30f0f --- /dev/null +++ b/parse/dev/v54eUIayFh/images/ddfdf7bb249088e3f366e8b8ba221b8f472a1b46e62e984081e2fc92dd49c629.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8c106cc57f0f924ff4514019e4bf5f08a52c857c7bd7c8bd7df9ad52b49d032 +size 38992 diff --git a/parse/dev/v54eUIayFh/images/e2c5a982dc82ee252be359b3af2b07dde3d18070a621e58b6df7d3118d6b8b1a.jpg b/parse/dev/v54eUIayFh/images/e2c5a982dc82ee252be359b3af2b07dde3d18070a621e58b6df7d3118d6b8b1a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fcd1a1596534ff029f3e4dcb1fbcc8ecd21feb7c --- /dev/null +++ b/parse/dev/v54eUIayFh/images/e2c5a982dc82ee252be359b3af2b07dde3d18070a621e58b6df7d3118d6b8b1a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6fdde8a8a35dcb4bef95630193c460e2afd99694add1261b90a734016761e4da +size 6702 diff --git a/parse/dev/v54eUIayFh/images/e2f7a1a82640708ece2e219ec1c356d11a92c8831241ba8bcd8afc8f321db7de.jpg b/parse/dev/v54eUIayFh/images/e2f7a1a82640708ece2e219ec1c356d11a92c8831241ba8bcd8afc8f321db7de.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2fd94e7b780c5669b343d646b71d09a6cbcc3a08 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/e2f7a1a82640708ece2e219ec1c356d11a92c8831241ba8bcd8afc8f321db7de.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26c8a3f48f6442f288c0f392e13b4cfcd3f93db03bb24b125320e750dd50bf67 +size 301361 diff --git a/parse/dev/v54eUIayFh/images/e97c1a92c260daa28e8cf10d26b2723974c19da904ccbd13e4ab533559dc605b.jpg b/parse/dev/v54eUIayFh/images/e97c1a92c260daa28e8cf10d26b2723974c19da904ccbd13e4ab533559dc605b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..28bda83c498fd2a4450bd5fe89b198991005e513 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/e97c1a92c260daa28e8cf10d26b2723974c19da904ccbd13e4ab533559dc605b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3909d4df99f9cd3ba0a7949a4f9232e28a25df015fb0f4f552aac1485ec470ea +size 29839 diff --git a/parse/dev/v54eUIayFh/images/f135750c3bf538a1ff4c9800056787e401a4ba3c39c05f42b7e5365257228845.jpg b/parse/dev/v54eUIayFh/images/f135750c3bf538a1ff4c9800056787e401a4ba3c39c05f42b7e5365257228845.jpg new file mode 100644 index 0000000000000000000000000000000000000000..200867d8ec63f22366cb37fe7a2158819334a296 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/f135750c3bf538a1ff4c9800056787e401a4ba3c39c05f42b7e5365257228845.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ad3cbd10c2e814d82f40f66883c46a897baebf702ed21e7f244820715093b23 +size 48697 diff --git a/parse/dev/v54eUIayFh/images/fa35fbc7943b8c85b6b5ea26f058940226239c3b9410569a34218ac3e9c58baa.jpg b/parse/dev/v54eUIayFh/images/fa35fbc7943b8c85b6b5ea26f058940226239c3b9410569a34218ac3e9c58baa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0b0b81c6ccc2ce522c574d05a804bd52edef5373 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/fa35fbc7943b8c85b6b5ea26f058940226239c3b9410569a34218ac3e9c58baa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a83e4358d2f5237f7df24d3f2b0b2cf229ea37c63494bc9abfd99ed74193d26 +size 5124 diff --git a/parse/dev/v54eUIayFh/images/fcb8b0ba4b724590545b882ca6e410285660539ed63b095b991438633dd6c62a.jpg b/parse/dev/v54eUIayFh/images/fcb8b0ba4b724590545b882ca6e410285660539ed63b095b991438633dd6c62a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2a381a886fa3f78c61d173dd21f79fc99199dab6 --- /dev/null +++ b/parse/dev/v54eUIayFh/images/fcb8b0ba4b724590545b882ca6e410285660539ed63b095b991438633dd6c62a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16c5cbc71aef70951543dbd1d9aa2dfb03237c22f5d161589ff80c4de67192d2 +size 8951 diff --git a/parse/dev/wqD6TfbYkrn/images/002d4532dd55eabda2053e09d84cc1480c88db3fec7d0555278e1e2def56a2c3.jpg b/parse/dev/wqD6TfbYkrn/images/002d4532dd55eabda2053e09d84cc1480c88db3fec7d0555278e1e2def56a2c3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e991ea1f6c039246a2ec4cac17ecc7c9d675beb9 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/002d4532dd55eabda2053e09d84cc1480c88db3fec7d0555278e1e2def56a2c3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0e13cf82dd349097042a278657aaa77109a26b3002d9a9a72c25445556d0248 +size 85995 diff --git a/parse/dev/wqD6TfbYkrn/images/114d3abf121ebcb2b7a9910244b0c519c052310577bb81572c584b7fcb146fde.jpg b/parse/dev/wqD6TfbYkrn/images/114d3abf121ebcb2b7a9910244b0c519c052310577bb81572c584b7fcb146fde.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c81aa1a29e7dc5a517089d1e28bfd2e452d35001 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/114d3abf121ebcb2b7a9910244b0c519c052310577bb81572c584b7fcb146fde.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:20e534b38bff342e24600604a0d1b8445e4fdef5092f97ee8512d3e0b1cd0058 +size 81525 diff --git a/parse/dev/wqD6TfbYkrn/images/221115eeaa8a2f9e735ea5ee873fb0d82a1b812139a29c395a3eed7875c82db3.jpg b/parse/dev/wqD6TfbYkrn/images/221115eeaa8a2f9e735ea5ee873fb0d82a1b812139a29c395a3eed7875c82db3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4b30270c1a57c821b5e84a464885757cc46b111f --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/221115eeaa8a2f9e735ea5ee873fb0d82a1b812139a29c395a3eed7875c82db3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc3ae820607084ae30c8689bfac9cc47390f6bfbf5745e5b6be7591770d433af +size 95332 diff --git a/parse/dev/wqD6TfbYkrn/images/46ceec4e08d6a670f515c1ce3d4080b88d8ee4ef95ffe1930aaa07353237179a.jpg b/parse/dev/wqD6TfbYkrn/images/46ceec4e08d6a670f515c1ce3d4080b88d8ee4ef95ffe1930aaa07353237179a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1c1b5e5229505798d503cb9b54a7c5296781913a --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/46ceec4e08d6a670f515c1ce3d4080b88d8ee4ef95ffe1930aaa07353237179a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23a3609b4abf0f6bd34b16ed811bb92ce4344b4ec513e23297cc7b58301c737e +size 12498 diff --git a/parse/dev/wqD6TfbYkrn/images/494f64743fa9478f801158b279f22090bcfdcbb1e21204c18fbeed949ea3fbfc.jpg b/parse/dev/wqD6TfbYkrn/images/494f64743fa9478f801158b279f22090bcfdcbb1e21204c18fbeed949ea3fbfc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5b4907f036192d51d69084c06e880e83ae6c6fe2 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/494f64743fa9478f801158b279f22090bcfdcbb1e21204c18fbeed949ea3fbfc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb8f493cd62f54a7301880cf21259591da9ef2c342f57d17c0902800f8cd5f62 +size 11041 diff --git a/parse/dev/wqD6TfbYkrn/images/68421485387e4e90d59122d210e17a18be03567b2ff02c307faf4374de85d4db.jpg b/parse/dev/wqD6TfbYkrn/images/68421485387e4e90d59122d210e17a18be03567b2ff02c307faf4374de85d4db.jpg new file mode 100644 index 0000000000000000000000000000000000000000..716fc9dc90c5a8737a55edc624d86b16e69589c5 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/68421485387e4e90d59122d210e17a18be03567b2ff02c307faf4374de85d4db.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb559266f5ca4465f02da8424d3421a5fda8b0e949642741a3608c40eb8100b0 +size 36780 diff --git a/parse/dev/wqD6TfbYkrn/images/6f875b691a8ebf42c6f671ceeb602cfee8378328a3c2ee47303aac8f3d15bc6e.jpg b/parse/dev/wqD6TfbYkrn/images/6f875b691a8ebf42c6f671ceeb602cfee8378328a3c2ee47303aac8f3d15bc6e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f8331aa682141d97c8c3077e8f3d247a2afa3522 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/6f875b691a8ebf42c6f671ceeb602cfee8378328a3c2ee47303aac8f3d15bc6e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b1f01708c11d16c593991426a35ee70143a8d7b303eb4fece88de9f03f6cdb2 +size 52602 diff --git a/parse/dev/wqD6TfbYkrn/images/782116d643a7bf605695d0838f2d6f75df2c4680a8c888b2324704751bc50922.jpg b/parse/dev/wqD6TfbYkrn/images/782116d643a7bf605695d0838f2d6f75df2c4680a8c888b2324704751bc50922.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89111e1f0926e667b92e98b46bfbdaae03f6d49b --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/782116d643a7bf605695d0838f2d6f75df2c4680a8c888b2324704751bc50922.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:51c523c54df8a79e976aeaafe27542fa14138469405dad37aabe8536e527c62a +size 57627 diff --git a/parse/dev/wqD6TfbYkrn/images/7dc2b01b9e116f38475b01ec60a451c327b6f33938df01845266cbf47addfbfa.jpg b/parse/dev/wqD6TfbYkrn/images/7dc2b01b9e116f38475b01ec60a451c327b6f33938df01845266cbf47addfbfa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..932a8d95baf47e39ace649797d451bb5d6a4ac87 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/7dc2b01b9e116f38475b01ec60a451c327b6f33938df01845266cbf47addfbfa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5fa97559612ef57b43ad2788864f8ada467f1a0115fcbb83d3b2777e1df4ef1 +size 121519 diff --git a/parse/dev/wqD6TfbYkrn/images/a530ad05a001f7a54142747c96873955b3ce7fc911e5b5dcd7a9143d14f554b3.jpg b/parse/dev/wqD6TfbYkrn/images/a530ad05a001f7a54142747c96873955b3ce7fc911e5b5dcd7a9143d14f554b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bc5ae1c9aaafa6fc07c1019e01b011a999e29b1d --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/a530ad05a001f7a54142747c96873955b3ce7fc911e5b5dcd7a9143d14f554b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:29a8534af716281bbae1f7768f09a884d9bd1f8d09e109f85be3c16b9499b96c +size 6668 diff --git a/parse/dev/wqD6TfbYkrn/images/ac957d9ba25772fa06f7a4d95381a3b77e98cbe8cd93574a843b690de33f697e.jpg b/parse/dev/wqD6TfbYkrn/images/ac957d9ba25772fa06f7a4d95381a3b77e98cbe8cd93574a843b690de33f697e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a63f00a665e84cc0ff2c97924c1cf40ecb4411e --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/ac957d9ba25772fa06f7a4d95381a3b77e98cbe8cd93574a843b690de33f697e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c3d8aebc5aef515ca6658250424ee6fe8baa5c1baab0f258743b65c3badec858 +size 72358 diff --git a/parse/dev/wqD6TfbYkrn/images/ba5f5f51d1d0dd9ed8564cb0736d88ed93de4815984e45ec3cbc72656cdf5f4f.jpg b/parse/dev/wqD6TfbYkrn/images/ba5f5f51d1d0dd9ed8564cb0736d88ed93de4815984e45ec3cbc72656cdf5f4f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..84898222305afe12de2a06d5cdf5ed2b71b619f1 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/ba5f5f51d1d0dd9ed8564cb0736d88ed93de4815984e45ec3cbc72656cdf5f4f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01db5815f3d29498a21b638f2597c853689e3bb2ce7dde4c6c600218cb819a13 +size 99840 diff --git a/parse/dev/wqD6TfbYkrn/images/dc6935be2f895ec27cc2573c566824ec638be193709947a8271c1e403183f97e.jpg b/parse/dev/wqD6TfbYkrn/images/dc6935be2f895ec27cc2573c566824ec638be193709947a8271c1e403183f97e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c004e448a2be9bbad04c22b9dc3f391d905a9d68 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/dc6935be2f895ec27cc2573c566824ec638be193709947a8271c1e403183f97e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d93ed7f99c9f4f9e777db44d75ab11f5c8cd9a5efc0516f47c4526911318c36 +size 108135 diff --git a/parse/dev/wqD6TfbYkrn/images/fb18d39234c7bacfc53ce62fcc2930ffcd7811f5615dfba7f0fe0befe8f1a2a4.jpg b/parse/dev/wqD6TfbYkrn/images/fb18d39234c7bacfc53ce62fcc2930ffcd7811f5615dfba7f0fe0befe8f1a2a4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..053a310780730726d7abde5aadbe757b70de98c0 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/fb18d39234c7bacfc53ce62fcc2930ffcd7811f5615dfba7f0fe0befe8f1a2a4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f5c7723d309b0584609303f840c2c3e4dbd42ef8ea4148b9518fc88bf163619 +size 152270 diff --git a/parse/dev/wqD6TfbYkrn/images/fc50d311e0c8990988ccf2d9da282e1d6c44a4af34e86f92756b3ec0a771ba84.jpg b/parse/dev/wqD6TfbYkrn/images/fc50d311e0c8990988ccf2d9da282e1d6c44a4af34e86f92756b3ec0a771ba84.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d6f39d51812e18fee3cfa0b2ddaeb9afe6a0e13 --- /dev/null +++ b/parse/dev/wqD6TfbYkrn/images/fc50d311e0c8990988ccf2d9da282e1d6c44a4af34e86f92756b3ec0a771ba84.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53e55804cb3bb0f9614de2b66db1993ce69eb780b5f73d7457689ae0101b8a06 +size 9700 diff --git a/parse/dev/z7p2V6KROOV/z7p2V6KROOV_middle.json b/parse/dev/z7p2V6KROOV/z7p2V6KROOV_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..7d6ba03f82aa7154e720a86de05681bbf593caeb --- /dev/null +++ b/parse/dev/z7p2V6KROOV/z7p2V6KROOV_middle.json @@ -0,0 +1,119606 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 78, + 429, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 431, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 431, + 97 + ], + "score": 1.0, + "content": "EXTENDING THE WILDS BENCHMARK FOR", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 98, + 324, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 98, + 324, + 117 + ], + "score": 1.0, + "content": "UNSUPERVISED ADAPTATION", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 113, + 134, + 504, + 204 + ], + "lines": [ + { + "bbox": [ + 111, + 134, + 371, + 147 + ], + "spans": [ + { + "bbox": [ + 111, + 134, + 230, + 147 + ], + "score": 1.0, + "content": "Shiori Sagawa1∗, Pang Wei", + "type": "text" + }, + { + "bbox": [ + 230, + 135, + 257, + 146 + ], + "score": 0.81, + "content": "\\mathbf { K o h } ^ { \\mathrm { 1 * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 134, + 285, + 147 + ], + "score": 1.0, + "content": ", Tony", + "type": "text" + }, + { + "bbox": [ + 285, + 135, + 310, + 146 + ], + "score": 0.59, + "content": "\\mathbf { L e e ^ { 1 * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 134, + 339, + 147 + ], + "score": 1.0, + "content": ", Irena", + "type": "text" + }, + { + "bbox": [ + 340, + 135, + 367, + 146 + ], + "score": 0.7, + "content": "\\mathbf { G a o ^ { 1 * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 134, + 371, + 147 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 145, + 500, + 160 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 172, + 160 + ], + "score": 1.0, + "content": "Sang Michael", + "type": "text" + }, + { + "bbox": [ + 173, + 146, + 192, + 157 + ], + "score": 0.4, + "content": "\\mathbf { X i e } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 145, + 305, + 160 + ], + "score": 1.0, + "content": ", Kendrick Shen1, Ananya", + "type": "text" + }, + { + "bbox": [ + 306, + 146, + 342, + 157 + ], + "score": 0.29, + "content": "\\mathbf { K u m a r } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 145, + 380, + 160 + ], + "score": 1.0, + "content": ", Weihua", + "type": "text" + }, + { + "bbox": [ + 381, + 146, + 399, + 157 + ], + "score": 0.86, + "content": "\\mathbf { H } \\mathbf { u } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 145, + 500, + 160 + ], + "score": 1.0, + "content": ", Michihiro Yasunaga1,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 111, + 156, + 504, + 171 + ], + "spans": [ + { + "bbox": [ + 111, + 156, + 406, + 171 + ], + "score": 1.0, + "content": "Henrik Marklund1, Sara Beery2, Etienne David3, Ian Stavness4, Wei", + "type": "text" + }, + { + "bbox": [ + 407, + 157, + 430, + 169 + ], + "score": 0.56, + "content": "\\mathbf { G u o ^ { 5 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 156, + 504, + 171 + ], + "score": 1.0, + "content": ", Jure Leskovec1,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 111, + 167, + 467, + 183 + ], + "spans": [ + { + "bbox": [ + 111, + 167, + 380, + 183 + ], + "score": 1.0, + "content": "Kate Saenko6, Tatsunori Hashimoto1, Sergey Levine7, Chelsea", + "type": "text" + }, + { + "bbox": [ + 380, + 169, + 406, + 180 + ], + "score": 0.62, + "content": "\\mathbf { F i n n ^ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 167, + 467, + 183 + ], + "score": 1.0, + "content": ", Percy Liang1", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 111, + 180, + 472, + 193 + ], + "spans": [ + { + "bbox": [ + 111, + 180, + 472, + 193 + ], + "score": 1.0, + "content": "∗Equal contribution 1Stanford University 2Caltech 3INRAE 4University of Saskatchewan", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 111, + 191, + 421, + 205 + ], + "spans": [ + { + "bbox": [ + 111, + 191, + 199, + 205 + ], + "score": 1.0, + "content": "5University of Tokyo", + "type": "text" + }, + { + "bbox": [ + 199, + 192, + 211, + 203 + ], + "score": 0.4, + "content": "{ } ^ { 6 } \\mathrm { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 191, + 421, + 205 + ], + "score": 1.0, + "content": "oston University 7University of California, Berkeley", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5 + }, + { + "type": "title", + "bbox": [ + 278, + 232, + 333, + 244 + ], + "lines": [ + { + "bbox": [ + 276, + 232, + 335, + 245 + ], + "spans": [ + { + "bbox": [ + 276, + 232, + 335, + 245 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 143, + 258, + 468, + 477 + ], + "lines": [ + { + "bbox": [ + 141, + 258, + 469, + 270 + ], + "spans": [ + { + "bbox": [ + 141, + 258, + 469, + 270 + ], + "score": 1.0, + "content": "Machine learning systems deployed in the wild are often trained on a source dis-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 268, + 469, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 268, + 469, + 280 + ], + "score": 1.0, + "content": "tribution but deployed on a different target distribution. Unlabeled data can be", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 280, + 469, + 292 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 469, + 292 + ], + "score": 1.0, + "content": "a powerful point of leverage for mitigating these distribution shifts, as it is fre-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 291, + 469, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 291, + 469, + 302 + ], + "score": 1.0, + "content": "quently much more available than labeled data and can often be obtained from", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 142, + 302, + 469, + 313 + ], + "spans": [ + { + "bbox": [ + 142, + 302, + 469, + 313 + ], + "score": 1.0, + "content": "distributions beyond the source distribution as well. However, existing distribu-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 313, + 469, + 324 + ], + "spans": [ + { + "bbox": [ + 142, + 313, + 469, + 324 + ], + "score": 1.0, + "content": "tion shift benchmarks with unlabeled data do not reflect the breadth of scenarios", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 324, + 469, + 335 + ], + "spans": [ + { + "bbox": [ + 141, + 324, + 469, + 335 + ], + "score": 1.0, + "content": "that arise in real-world applications. In this work, we present the WILDS 2.0", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 335, + 469, + 346 + ], + "spans": [ + { + "bbox": [ + 142, + 335, + 469, + 346 + ], + "score": 1.0, + "content": "update, which extends 8 of the 10 datasets in the WILDS benchmark of distribu-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 345, + 469, + 357 + ], + "spans": [ + { + "bbox": [ + 141, + 345, + 469, + 357 + ], + "score": 1.0, + "content": "tion shifts to include curated unlabeled data that would be realistically obtainable", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 356, + 470, + 369 + ], + "spans": [ + { + "bbox": [ + 141, + 356, + 470, + 369 + ], + "score": 1.0, + "content": "in deployment. These datasets span a wide range of applications (from histol-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 367, + 470, + 379 + ], + "spans": [ + { + "bbox": [ + 141, + 367, + 470, + 379 + ], + "score": 1.0, + "content": "ogy to wildlife conservation), tasks (classification, regression, and detection), and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 378, + 469, + 391 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 469, + 391 + ], + "score": 1.0, + "content": "modalities (photos, satellite images, microscope slides, text, molecular graphs).", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 388, + 470, + 402 + ], + "spans": [ + { + "bbox": [ + 141, + 388, + 470, + 402 + ], + "score": 1.0, + "content": "The update maintains consistency with the original WILDS benchmark by using", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 400, + 470, + 412 + ], + "spans": [ + { + "bbox": [ + 141, + 400, + 470, + 412 + ], + "score": 1.0, + "content": "identical labeled training, validation, and test sets, as well as identical evaluation", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 411, + 470, + 423 + ], + "spans": [ + { + "bbox": [ + 141, + 411, + 470, + 423 + ], + "score": 1.0, + "content": "metrics. We systematically benchmark state-of-the-art methods that use unlabeled", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 422, + 470, + 434 + ], + "spans": [ + { + "bbox": [ + 141, + 422, + 470, + 434 + ], + "score": 1.0, + "content": "data, including domain-invariant, self-training, and self-supervised methods, and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 433, + 469, + 445 + ], + "spans": [ + { + "bbox": [ + 141, + 433, + 469, + 445 + ], + "score": 1.0, + "content": "show that their success on WILDS is limited. To facilitate method development,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 142, + 444, + 469, + 456 + ], + "spans": [ + { + "bbox": [ + 142, + 444, + 469, + 456 + ], + "score": 1.0, + "content": "we provide an open-source package that automates data loading and contains the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 455, + 470, + 467 + ], + "spans": [ + { + "bbox": [ + 141, + 455, + 470, + 467 + ], + "score": 1.0, + "content": "model architectures and methods used in this paper. Code and leaderboards are", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 466, + 350, + 477 + ], + "spans": [ + { + "bbox": [ + 141, + 466, + 350, + 477 + ], + "score": 1.0, + "content": "available at https://wilds.stanford.edu.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 18.5 + }, + { + "type": "title", + "bbox": [ + 108, + 503, + 205, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 502, + 208, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 208, + 519 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 528, + 505, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 505, + 540 + ], + "score": 1.0, + "content": "Distribution shifts—when models are trained on a source distribution but deployed on a different", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 539, + 504, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 504, + 551 + ], + "score": 1.0, + "content": "target distribution—are frequent problems for machine learning systems in the wild (Quinonero- ˜", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "Candela et al., 2009; Geirhos et al., 2020; Koh et al., 2021). In this paper, we focus on the use of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "score": 1.0, + "content": "unlabeled data to mitigate these shifts. Unlabeled data is a powerful point of leverage as it is more", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "readily available than labeled data and can often be obtained from distributions beyond the source", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "score": 1.0, + "content": "distribution. For example, in the crop detection task in Figure 1, we wish to learn a model that can", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "extrapolate to a set of target domains (farms) (David et al., 2020), and while we only have labeled", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "training examples from some source domains, we have many more unlabeled examples from the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 617, + 427, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 427, + 628 + ], + "score": 1.0, + "content": "source domains, from extra domains, and even directly from the target domains.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 633, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "Many methods for leveraging unlabeled data have been highly successful on some types of distri-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "score": 1.0, + "content": "bution shifts (Berthelot et al., 2021; Zhang et al., 2021). However, the datasets typically used for", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 655, + 504, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 504, + 666 + ], + "score": 1.0, + "content": "evaluating these methods do not reflect many of the realistic shifts that might occur in the wild.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 664, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 505, + 678 + ], + "score": 1.0, + "content": "These evaluations tend instead to focus on shifts between photos and stylized versions like sketches", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "(Li et al., 2017; Venkateswara et al., 2017; Peng et al., 2019) or synthetic renderings (Peng et al.,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "2018), or between variants of digits datasets like MNIST (LeCun et al., 1998) and SVHN (Netzer", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "et al., 2011). Unfortunately, prior work has shown that methods that work well on one type of shift", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "need not generalize to others (Taori et al., 2020; Djolonga et al., 2020; Xie et al., 2021a; Miller et al.,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 720, + 496, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 496, + 732 + ], + "score": 1.0, + "content": "2021), which raises the question of how well they would work on a wider array of realistic shifts.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 78, + 429, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 431, + 97 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 431, + 97 + ], + "score": 1.0, + "content": "EXTENDING THE WILDS BENCHMARK FOR", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 98, + 324, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 98, + 324, + 117 + ], + "score": 1.0, + "content": "UNSUPERVISED ADAPTATION", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 113, + 134, + 504, + 204 + ], + "lines": [ + { + "bbox": [ + 111, + 134, + 371, + 147 + ], + "spans": [ + { + "bbox": [ + 111, + 134, + 230, + 147 + ], + "score": 1.0, + "content": "Shiori Sagawa1∗, Pang Wei", + "type": "text" + }, + { + "bbox": [ + 230, + 135, + 257, + 146 + ], + "score": 0.81, + "content": "\\mathbf { K o h } ^ { \\mathrm { 1 * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 134, + 285, + 147 + ], + "score": 1.0, + "content": ", Tony", + "type": "text" + }, + { + "bbox": [ + 285, + 135, + 310, + 146 + ], + "score": 0.59, + "content": "\\mathbf { L e e ^ { 1 * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 134, + 339, + 147 + ], + "score": 1.0, + "content": ", Irena", + "type": "text" + }, + { + "bbox": [ + 340, + 135, + 367, + 146 + ], + "score": 0.7, + "content": "\\mathbf { G a o ^ { 1 * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 134, + 371, + 147 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 145, + 500, + 160 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 172, + 160 + ], + "score": 1.0, + "content": "Sang Michael", + "type": "text" + }, + { + "bbox": [ + 173, + 146, + 192, + 157 + ], + "score": 0.4, + "content": "\\mathbf { X i e } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 145, + 305, + 160 + ], + "score": 1.0, + "content": ", Kendrick Shen1, Ananya", + "type": "text" + }, + { + "bbox": [ + 306, + 146, + 342, + 157 + ], + "score": 0.29, + "content": "\\mathbf { K u m a r } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 145, + 380, + 160 + ], + "score": 1.0, + "content": ", Weihua", + "type": "text" + }, + { + "bbox": [ + 381, + 146, + 399, + 157 + ], + "score": 0.86, + "content": "\\mathbf { H } \\mathbf { u } ^ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 145, + 500, + 160 + ], + "score": 1.0, + "content": ", Michihiro Yasunaga1,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 111, + 156, + 504, + 171 + ], + "spans": [ + { + "bbox": [ + 111, + 156, + 406, + 171 + ], + "score": 1.0, + "content": "Henrik Marklund1, Sara Beery2, Etienne David3, Ian Stavness4, Wei", + "type": "text" + }, + { + "bbox": [ + 407, + 157, + 430, + 169 + ], + "score": 0.56, + "content": "\\mathbf { G u o ^ { 5 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 156, + 504, + 171 + ], + "score": 1.0, + "content": ", Jure Leskovec1,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 111, + 167, + 467, + 183 + ], + "spans": [ + { + "bbox": [ + 111, + 167, + 380, + 183 + ], + "score": 1.0, + "content": "Kate Saenko6, Tatsunori Hashimoto1, Sergey Levine7, Chelsea", + "type": "text" + }, + { + "bbox": [ + 380, + 169, + 406, + 180 + ], + "score": 0.62, + "content": "\\mathbf { F i n n ^ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 167, + 467, + 183 + ], + "score": 1.0, + "content": ", Percy Liang1", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 111, + 180, + 472, + 193 + ], + "spans": [ + { + "bbox": [ + 111, + 180, + 472, + 193 + ], + "score": 1.0, + "content": "∗Equal contribution 1Stanford University 2Caltech 3INRAE 4University of Saskatchewan", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 111, + 191, + 421, + 205 + ], + "spans": [ + { + "bbox": [ + 111, + 191, + 199, + 205 + ], + "score": 1.0, + "content": "5University of Tokyo", + "type": "text" + }, + { + "bbox": [ + 199, + 192, + 211, + 203 + ], + "score": 0.4, + "content": "{ } ^ { 6 } \\mathrm { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 191, + 421, + 205 + ], + "score": 1.0, + "content": "oston University 7University of California, Berkeley", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5, + "bbox_fs": [ + 111, + 134, + 504, + 205 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 232, + 333, + 244 + ], + "lines": [ + { + "bbox": [ + 276, + 232, + 335, + 245 + ], + "spans": [ + { + "bbox": [ + 276, + 232, + 335, + 245 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 143, + 258, + 468, + 477 + ], + "lines": [ + { + "bbox": [ + 141, + 258, + 469, + 270 + ], + "spans": [ + { + "bbox": [ + 141, + 258, + 469, + 270 + ], + "score": 1.0, + "content": "Machine learning systems deployed in the wild are often trained on a source dis-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 268, + 469, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 268, + 469, + 280 + ], + "score": 1.0, + "content": "tribution but deployed on a different target distribution. Unlabeled data can be", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 280, + 469, + 292 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 469, + 292 + ], + "score": 1.0, + "content": "a powerful point of leverage for mitigating these distribution shifts, as it is fre-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 291, + 469, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 291, + 469, + 302 + ], + "score": 1.0, + "content": "quently much more available than labeled data and can often be obtained from", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 142, + 302, + 469, + 313 + ], + "spans": [ + { + "bbox": [ + 142, + 302, + 469, + 313 + ], + "score": 1.0, + "content": "distributions beyond the source distribution as well. However, existing distribu-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 313, + 469, + 324 + ], + "spans": [ + { + "bbox": [ + 142, + 313, + 469, + 324 + ], + "score": 1.0, + "content": "tion shift benchmarks with unlabeled data do not reflect the breadth of scenarios", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 324, + 469, + 335 + ], + "spans": [ + { + "bbox": [ + 141, + 324, + 469, + 335 + ], + "score": 1.0, + "content": "that arise in real-world applications. In this work, we present the WILDS 2.0", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 335, + 469, + 346 + ], + "spans": [ + { + "bbox": [ + 142, + 335, + 469, + 346 + ], + "score": 1.0, + "content": "update, which extends 8 of the 10 datasets in the WILDS benchmark of distribu-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 345, + 469, + 357 + ], + "spans": [ + { + "bbox": [ + 141, + 345, + 469, + 357 + ], + "score": 1.0, + "content": "tion shifts to include curated unlabeled data that would be realistically obtainable", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 356, + 470, + 369 + ], + "spans": [ + { + "bbox": [ + 141, + 356, + 470, + 369 + ], + "score": 1.0, + "content": "in deployment. These datasets span a wide range of applications (from histol-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 367, + 470, + 379 + ], + "spans": [ + { + "bbox": [ + 141, + 367, + 470, + 379 + ], + "score": 1.0, + "content": "ogy to wildlife conservation), tasks (classification, regression, and detection), and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 378, + 469, + 391 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 469, + 391 + ], + "score": 1.0, + "content": "modalities (photos, satellite images, microscope slides, text, molecular graphs).", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 388, + 470, + 402 + ], + "spans": [ + { + "bbox": [ + 141, + 388, + 470, + 402 + ], + "score": 1.0, + "content": "The update maintains consistency with the original WILDS benchmark by using", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 400, + 470, + 412 + ], + "spans": [ + { + "bbox": [ + 141, + 400, + 470, + 412 + ], + "score": 1.0, + "content": "identical labeled training, validation, and test sets, as well as identical evaluation", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 411, + 470, + 423 + ], + "spans": [ + { + "bbox": [ + 141, + 411, + 470, + 423 + ], + "score": 1.0, + "content": "metrics. We systematically benchmark state-of-the-art methods that use unlabeled", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 422, + 470, + 434 + ], + "spans": [ + { + "bbox": [ + 141, + 422, + 470, + 434 + ], + "score": 1.0, + "content": "data, including domain-invariant, self-training, and self-supervised methods, and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 433, + 469, + 445 + ], + "spans": [ + { + "bbox": [ + 141, + 433, + 469, + 445 + ], + "score": 1.0, + "content": "show that their success on WILDS is limited. To facilitate method development,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 142, + 444, + 469, + 456 + ], + "spans": [ + { + "bbox": [ + 142, + 444, + 469, + 456 + ], + "score": 1.0, + "content": "we provide an open-source package that automates data loading and contains the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 455, + 470, + 467 + ], + "spans": [ + { + "bbox": [ + 141, + 455, + 470, + 467 + ], + "score": 1.0, + "content": "model architectures and methods used in this paper. Code and leaderboards are", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 466, + 350, + 477 + ], + "spans": [ + { + "bbox": [ + 141, + 466, + 350, + 477 + ], + "score": 1.0, + "content": "available at https://wilds.stanford.edu.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 18.5, + "bbox_fs": [ + 141, + 258, + 470, + 477 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 503, + 205, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 502, + 208, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 208, + 519 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 528, + 505, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 505, + 540 + ], + "score": 1.0, + "content": "Distribution shifts—when models are trained on a source distribution but deployed on a different", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 539, + 504, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 504, + 551 + ], + "score": 1.0, + "content": "target distribution—are frequent problems for machine learning systems in the wild (Quinonero- ˜", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 106, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "Candela et al., 2009; Geirhos et al., 2020; Koh et al., 2021). In this paper, we focus on the use of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "score": 1.0, + "content": "unlabeled data to mitigate these shifts. Unlabeled data is a powerful point of leverage as it is more", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "readily available than labeled data and can often be obtained from distributions beyond the source", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "score": 1.0, + "content": "distribution. For example, in the crop detection task in Figure 1, we wish to learn a model that can", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "extrapolate to a set of target domains (farms) (David et al., 2020), and while we only have labeled", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "training examples from some source domains, we have many more unlabeled examples from the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 617, + 427, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 427, + 628 + ], + "score": 1.0, + "content": "source domains, from extra domains, and even directly from the target domains.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 528, + 506, + 628 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 633, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "Many methods for leveraging unlabeled data have been highly successful on some types of distri-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "score": 1.0, + "content": "bution shifts (Berthelot et al., 2021; Zhang et al., 2021). However, the datasets typically used for", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 655, + 504, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 504, + 666 + ], + "score": 1.0, + "content": "evaluating these methods do not reflect many of the realistic shifts that might occur in the wild.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 664, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 505, + 678 + ], + "score": 1.0, + "content": "These evaluations tend instead to focus on shifts between photos and stylized versions like sketches", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 690 + ], + "score": 1.0, + "content": "(Li et al., 2017; Venkateswara et al., 2017; Peng et al., 2019) or synthetic renderings (Peng et al.,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "2018), or between variants of digits datasets like MNIST (LeCun et al., 1998) and SVHN (Netzer", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "et al., 2011). Unfortunately, prior work has shown that methods that work well on one type of shift", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "need not generalize to others (Taori et al., 2020; Djolonga et al., 2020; Xie et al., 2021a; Miller et al.,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 720, + 496, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 496, + 732 + ], + "score": 1.0, + "content": "2021), which raises the question of how well they would work on a wider array of realistic shifts.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 632, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 51, + 502, + 215 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 51, + 502, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 51, + 502, + 215 + ], + "spans": [ + { + "bbox": [ + 109, + 51, + 502, + 215 + ], + "score": 0.972, + "type": "image", + "image_path": "3fd45fe4623769a5a90206e38ceef7c83a5a9207abe2235df7671e48419696eb.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 51, + 502, + 105.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 105.66666666666666, + 502, + 160.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 160.33333333333331, + 502, + 214.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 225, + 505, + 301 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 224, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 505, + 237 + ], + "score": 1.0, + "content": "Figure 1: Each WILDS dataset (Koh et al., 2021) contains labeled data from the source domains", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 236, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 505, + 248 + ], + "score": 1.0, + "content": "(for training), validation domains (for hyperparameter selection), and target domains (for held-out", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 245, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 104, + 245, + 505, + 259 + ], + "score": 1.0, + "content": "evaluation). In the WILDS 2.0 update, we extend these datasets with unlabeled data from a com-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 257, + 506, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 506, + 270 + ], + "score": 1.0, + "content": "bination of source, validation, or target domains, as well as extra domains from which there is no", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "labeled data. The labeled data is exactly the same as in WILDS 1.0. In this figure, we illustrate", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 279, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 505, + 292 + ], + "score": 1.0, + "content": "the setting with the GLOBALWHEAT-WILDS dataset, where domains correspond to images acquired", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 290, + 293, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 293, + 302 + ], + "score": 1.0, + "content": "from different locations and at different times.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 322, + 505, + 455 + ], + "lines": [ + { + "bbox": [ + 105, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "In this paper, we make two contributions. First, we present WILDS 2.0 (Figure 2), an updated ver-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 335, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 505, + 346 + ], + "score": 1.0, + "content": "sion of the recent WILDS benchmark of in-the-wild distribution shifts (Koh et al., 2021). WILDS", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 345, + 504, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 504, + 357 + ], + "score": 1.0, + "content": "datasets span a wide range of tasks and modalities, and each dataset reflects a domain generalization", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "or subpopulation shift setting with a substantial gap between in-distribution and out-of-distribution", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 367, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 505, + 379 + ], + "score": 1.0, + "content": "performance. However, WILDS 1.0 only contained labeled data, which limits the leverage for learn-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "ing robust models. In WILDS 2.0, we extend 8 of the 10 WILDS datasets1 with curated unlabeled", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 389, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 506, + 401 + ], + "score": 1.0, + "content": "data acquired from the same source and target domains as the labeled data, as well as from extra", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 399, + 506, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 506, + 413 + ], + "score": 1.0, + "content": "domains of the same type: e.g., in the GLOBALWHEAT-WILDS dataset pictured in Figure 1, we", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 411, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 506, + 423 + ], + "score": 1.0, + "content": "acquired unlabeled photos of wheat fields from the source and target farms as well as extra farms", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 421, + 505, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 434 + ], + "score": 1.0, + "content": "that were not in the original labeled dataset. In total, WILDS 2.0 adds 14.5 million unlabeled ex-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 433, + 505, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 358, + 445 + ], + "score": 1.0, + "content": "amples, expanding the number of examples for each dataset by", + "type": "text" + }, + { + "bbox": [ + 358, + 433, + 388, + 443 + ], + "score": 0.89, + "content": "3 { - } 1 3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 433, + 505, + 445 + ], + "score": 1.0, + "content": "and allowing us to combine", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 443, + 410, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 410, + 456 + ], + "score": 1.0, + "content": "the real-world relevance of WILDS with the leverage of unlabeled data.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 461, + 505, + 538 + ], + "lines": [ + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "score": 1.0, + "content": "Second, we developed a standardized and consistent protocol for evaluating methods that lever-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "age the unlabeled data in WILDS 2.0. We assessed representatives from three popular categories:", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "methods for learning domain-invariant representations (Sun & Saenko, 2016; Ganin et al., 2016),", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 494, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 505, + 505 + ], + "score": 1.0, + "content": "self-training methods (Lee, 2013; Sohn et al., 2020; Xie et al., 2020), and pre-training methods that", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "score": 1.0, + "content": "rely on self-supervision (Devlin et al., 2019; Caron et al., 2020). These methods have been success-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 516, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 505, + 527 + ], + "score": 1.0, + "content": "ful on some types of shifts, such as going from photos to sketches, or from handwritten digits to", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 527, + 326, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 326, + 538 + ], + "score": 1.0, + "content": "street signs (Berthelot et al., 2021; Zhang et al., 2021).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 543, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "Our results on WILDS are mixed: many methods did not outperform standard supervised", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 554, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 566 + ], + "score": 1.0, + "content": "training despite using additional unlabeled data, and the only clear successes were on two im-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "score": 1.0, + "content": "age classification datasets (CAMELYON17-WILDS and FMOW-WILDS). Successful methods relied", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "heavily on data augmentation (Xie et al., 2020; Caron et al., 2020), which limited their applicability", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "score": 1.0, + "content": "to modalities where augmentations are not as well developed, such as text and molecular graphs.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 598, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 610 + ], + "score": 1.0, + "content": "The same methods were unsuccessful on image regression and detection tasks, which have been", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "score": 1.0, + "content": "relatively understudied: e.g., pseudolabel-based methods do not straightforwardly apply to regres-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 619, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 506, + 632 + ], + "score": 1.0, + "content": "sion. For the text datasets, continued language model pre-training did not help, unlike in prior work", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 630, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 505, + 645 + ], + "score": 1.0, + "content": "(Gururangan et al., 2020). Our results suggest fruitful avenues for future work, such as developing", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 642, + 501, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 501, + 655 + ], + "score": 1.0, + "content": "data augmentations for non-image modalities and more effective hyperparameter tuning protocols.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 107, + 659, + 504, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "Overall, our results underscore the importance of developing and evaluating methods for unlabeled", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 669, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 506, + 682 + ], + "score": 1.0, + "content": "data on a wider variety of real-world shifts than is typically studied. To this end, we have updated", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "the open-source Python WILDS package to include unlabeled data loaders, compatible implementa-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 505, + 705 + ], + "score": 1.0, + "content": "tions of all the methods we benchmarked, and scripts to replicate all experiments in this paper (Ap-", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 712, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 118, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 118, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "1 We omitted PY150-WILDS, as code completion data is always labeled by nature of the task, and RXRX1-", + "type": "text" + } + ] + }, + { + "bbox": [ + 106, + 721, + 411, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 411, + 732 + ], + "score": 1.0, + "content": "WILDS, as unlabeled data for that genetic perturbation task is not typically available.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 51, + 502, + 215 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 51, + 502, + 215 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 51, + 502, + 215 + ], + "spans": [ + { + "bbox": [ + 109, + 51, + 502, + 215 + ], + "score": 0.972, + "type": "image", + "image_path": "3fd45fe4623769a5a90206e38ceef7c83a5a9207abe2235df7671e48419696eb.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 51, + 502, + 105.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 105.66666666666666, + 502, + 160.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 160.33333333333331, + 502, + 214.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 225, + 505, + 301 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 224, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 505, + 237 + ], + "score": 1.0, + "content": "Figure 1: Each WILDS dataset (Koh et al., 2021) contains labeled data from the source domains", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 236, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 505, + 248 + ], + "score": 1.0, + "content": "(for training), validation domains (for hyperparameter selection), and target domains (for held-out", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 245, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 104, + 245, + 505, + 259 + ], + "score": 1.0, + "content": "evaluation). In the WILDS 2.0 update, we extend these datasets with unlabeled data from a com-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 257, + 506, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 506, + 270 + ], + "score": 1.0, + "content": "bination of source, validation, or target domains, as well as extra domains from which there is no", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "labeled data. The labeled data is exactly the same as in WILDS 1.0. In this figure, we illustrate", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 279, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 505, + 292 + ], + "score": 1.0, + "content": "the setting with the GLOBALWHEAT-WILDS dataset, where domains correspond to images acquired", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 290, + 293, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 293, + 302 + ], + "score": 1.0, + "content": "from different locations and at different times.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 322, + 505, + 455 + ], + "lines": [ + { + "bbox": [ + 105, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "In this paper, we make two contributions. First, we present WILDS 2.0 (Figure 2), an updated ver-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 335, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 505, + 346 + ], + "score": 1.0, + "content": "sion of the recent WILDS benchmark of in-the-wild distribution shifts (Koh et al., 2021). WILDS", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 345, + 504, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 504, + 357 + ], + "score": 1.0, + "content": "datasets span a wide range of tasks and modalities, and each dataset reflects a domain generalization", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "or subpopulation shift setting with a substantial gap between in-distribution and out-of-distribution", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 367, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 505, + 379 + ], + "score": 1.0, + "content": "performance. However, WILDS 1.0 only contained labeled data, which limits the leverage for learn-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 378, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 505, + 390 + ], + "score": 1.0, + "content": "ing robust models. In WILDS 2.0, we extend 8 of the 10 WILDS datasets1 with curated unlabeled", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 389, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 506, + 401 + ], + "score": 1.0, + "content": "data acquired from the same source and target domains as the labeled data, as well as from extra", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 399, + 506, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 506, + 413 + ], + "score": 1.0, + "content": "domains of the same type: e.g., in the GLOBALWHEAT-WILDS dataset pictured in Figure 1, we", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 411, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 506, + 423 + ], + "score": 1.0, + "content": "acquired unlabeled photos of wheat fields from the source and target farms as well as extra farms", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 421, + 505, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 434 + ], + "score": 1.0, + "content": "that were not in the original labeled dataset. In total, WILDS 2.0 adds 14.5 million unlabeled ex-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 433, + 505, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 358, + 445 + ], + "score": 1.0, + "content": "amples, expanding the number of examples for each dataset by", + "type": "text" + }, + { + "bbox": [ + 358, + 433, + 388, + 443 + ], + "score": 0.89, + "content": "3 { - } 1 3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 433, + 505, + 445 + ], + "score": 1.0, + "content": "and allowing us to combine", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 443, + 410, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 410, + 456 + ], + "score": 1.0, + "content": "the real-world relevance of WILDS with the leverage of unlabeled data.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 323, + 506, + 456 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 461, + 505, + 538 + ], + "lines": [ + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "score": 1.0, + "content": "Second, we developed a standardized and consistent protocol for evaluating methods that lever-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "age the unlabeled data in WILDS 2.0. We assessed representatives from three popular categories:", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "methods for learning domain-invariant representations (Sun & Saenko, 2016; Ganin et al., 2016),", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 494, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 494, + 505, + 505 + ], + "score": 1.0, + "content": "self-training methods (Lee, 2013; Sohn et al., 2020; Xie et al., 2020), and pre-training methods that", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "score": 1.0, + "content": "rely on self-supervision (Devlin et al., 2019; Caron et al., 2020). These methods have been success-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 516, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 505, + 527 + ], + "score": 1.0, + "content": "ful on some types of shifts, such as going from photos to sketches, or from handwritten digits to", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 527, + 326, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 326, + 538 + ], + "score": 1.0, + "content": "street signs (Berthelot et al., 2021; Zhang et al., 2021).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 461, + 505, + 538 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 543, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "Our results on WILDS are mixed: many methods did not outperform standard supervised", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 554, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 566 + ], + "score": 1.0, + "content": "training despite using additional unlabeled data, and the only clear successes were on two im-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "score": 1.0, + "content": "age classification datasets (CAMELYON17-WILDS and FMOW-WILDS). Successful methods relied", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "heavily on data augmentation (Xie et al., 2020; Caron et al., 2020), which limited their applicability", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "score": 1.0, + "content": "to modalities where augmentations are not as well developed, such as text and molecular graphs.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 598, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 610 + ], + "score": 1.0, + "content": "The same methods were unsuccessful on image regression and detection tasks, which have been", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 505, + 622 + ], + "score": 1.0, + "content": "relatively understudied: e.g., pseudolabel-based methods do not straightforwardly apply to regres-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 619, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 506, + 632 + ], + "score": 1.0, + "content": "sion. For the text datasets, continued language model pre-training did not help, unlike in prior work", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 630, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 505, + 645 + ], + "score": 1.0, + "content": "(Gururangan et al., 2020). Our results suggest fruitful avenues for future work, such as developing", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 642, + 501, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 501, + 655 + ], + "score": 1.0, + "content": "data augmentations for non-image modalities and more effective hyperparameter tuning protocols.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 543, + 506, + 655 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 659, + 504, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "Overall, our results underscore the importance of developing and evaluating methods for unlabeled", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 669, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 506, + 682 + ], + "score": 1.0, + "content": "data on a wider variety of real-world shifts than is typically studied. To this end, we have updated", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "the open-source Python WILDS package to include unlabeled data loaders, compatible implementa-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 505, + 705 + ], + "score": 1.0, + "content": "tions of all the methods we benchmarked, and scripts to replicate all experiments in this paper (Ap-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "pendix G). Code and public leaderboards are available at https://wilds.stanford.edu. By", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "score": 1.0, + "content": "allowing developers to easily test algorithms across the variety of datasets in WILDS 2.0, we hope", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 399, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 505, + 410 + ], + "score": 1.0, + "content": "to accelerate the development of methods that can leverage unlabeled data to improve robustness to", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 410, + 224, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 224, + 420 + ], + "score": 1.0, + "content": "real-world distribution shifts.", + "type": "text", + "cross_page": true + } + ], + "index": 14 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 658, + 506, + 705 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 50, + 504, + 259 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 50, + 504, + 259 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 50, + 504, + 259 + ], + "spans": [ + { + "bbox": [ + 110, + 50, + 504, + 259 + ], + "score": 0.937, + "type": "image", + "image_path": "63768da805fbf7337aad7f451ce99e743999480ded617c21d5b9be4ee6f97cb2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 50, + 504, + 119.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 119.66666666666667, + 504, + 189.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 189.33333333333334, + 504, + 259.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 268, + 505, + 356 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 267, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 506, + 280 + ], + "score": 1.0, + "content": "Figure 2: The WILDS 2.0 update adds unlabeled data to 8 WILDS datasets. For each dataset, we", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 279, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 370, + 291 + ], + "score": 1.0, + "content": "kept the labeled data from WILDS and expanded the datasets by", + "type": "text" + }, + { + "bbox": [ + 370, + 279, + 399, + 290 + ], + "score": 0.89, + "content": "3 { - } 1 3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 279, + 505, + 291 + ], + "score": 1.0, + "content": "with unlabeled data from", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 289, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 303 + ], + "score": 1.0, + "content": "the same underlying dataset. The type of unlabeled data (i.e., whether it comes from source, extra,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 302, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 505, + 312 + ], + "score": 1.0, + "content": "validation, or target domains) depends on what is realistic and available for the application. Beyond", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 312, + 504, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 504, + 323 + ], + "score": 1.0, + "content": "these 8 datasets, WILDS also contains 2 datasets without unlabeled data: the PY150-WILDS code", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "completion dataset and the RXRX1-WILDS genetic perturbation dataset. For all datasets, the labeled", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 333, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 345 + ], + "score": 1.0, + "content": "data and evaluation metrics are exactly the same as in WILDS 1.0. Figure adapted with permission", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 343, + 201, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 201, + 357 + ], + "score": 1.0, + "content": "from Koh et al. (2021).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6.5 + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 106, + 376, + 505, + 420 + ], + "lines": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "pendix G). Code and public leaderboards are available at https://wilds.stanford.edu. By", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "score": 1.0, + "content": "allowing developers to easily test algorithms across the variety of datasets in WILDS 2.0, we hope", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 399, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 505, + 410 + ], + "score": 1.0, + "content": "to accelerate the development of methods that can leverage unlabeled data to improve robustness to", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 410, + 224, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 224, + 420 + ], + "score": 1.0, + "content": "real-world distribution shifts.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 426, + 505, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "Finally, we note that WILDS 2.0 not a separate benchmark from WILDS 1.0: the labeled data and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 438, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 449 + ], + "score": 1.0, + "content": "evaluation metrics are exactly the same in WILDS 1.0 and WILDS 2.0, and future results should be", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 448, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 506, + 460 + ], + "score": 1.0, + "content": "reported on the overall WILDS benchmark, with a note describing what kind of unlabeled data (if", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "score": 1.0, + "content": "any) was used. In this paper, we discuss the addition of unlabeled data and analyze the performance", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "of methods that use the unlabeled data. For a more detailed description of the datasets, evaluation", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 481, + 451, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 451, + 494 + ], + "score": 1.0, + "content": "metrics, and models used, please refer to the original WILDS paper (Koh et al., 2021).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 108, + 509, + 502, + 521 + ], + "lines": [ + { + "bbox": [ + 105, + 507, + 504, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 504, + 523 + ], + "score": 1.0, + "content": "2 COMPARISON WITH EXISTING UNSUPERVISED ADAPTATION BENCHMARKS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 505, + 566 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "WILDS 2.0 offers a diverse range of applications and modalities while also providing an extensive", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "amount of unlabeled data that can be used as leverage for training robust models. In this section, we", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 555, + 432, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 432, + 568 + ], + "score": 1.0, + "content": "briefly compare with other existing ML benchmarks for unsupervised adaptation.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "Images. Evaluations of unsupervised adaptation methods for image classification have focused on", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "generalizing from natural photos to a range of stylized images, such as sketches and cartoons (PACS", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "(Li et al., 2017), Office-Home (Venkateswara et al., 2017), and DomainNet (Peng et al., 2019)),", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "product images (Office-31 (Saenko et al., 2010)), and synthetic renderings (VisDA (Peng et al.,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 617, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 505, + 627 + ], + "score": 1.0, + "content": "2018)), though location-based shifts have also been recently explored (Dubey et al., 2021). It is", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "also popular to evaluate on shifts between digits datasets, such as MNIST (LeCun et al., 1998),", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "SVHN (Netzer et al., 2011), and USPS (Hull, 1994). In image detection and segmentation, existing", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "adaptation benchmarks tend to focus on generalizing from synthetic to natural scenes (Ros et al.,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "2016; Richter et al., 2016; Cordts et al., 2016; Hoffman et al., 2018), which can be an important", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "score": 1.0, + "content": "tool for realistic problems but is not the focus of this work. In contrast, WILDS considers real-world", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 681, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 694 + ], + "score": 1.0, + "content": "distribution shifts, and it spans diverse modalities (satellite, microscope, agriculture, and camera", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 693, + 349, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 349, + 705 + ], + "score": 1.0, + "content": "trap images) and tasks (classification, regression, detection).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Text. Methods for unsupervised adaptation in NLP are typically evaluated on domain shifts between", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "different textual sources, such as news articles, different categories of product reviews, Wikipedia,", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 50, + 504, + 259 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 50, + 504, + 259 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 50, + 504, + 259 + ], + "spans": [ + { + "bbox": [ + 110, + 50, + 504, + 259 + ], + "score": 0.937, + "type": "image", + "image_path": "63768da805fbf7337aad7f451ce99e743999480ded617c21d5b9be4ee6f97cb2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 50, + 504, + 119.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 119.66666666666667, + 504, + 189.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 189.33333333333334, + 504, + 259.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 268, + 505, + 356 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 267, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 506, + 280 + ], + "score": 1.0, + "content": "Figure 2: The WILDS 2.0 update adds unlabeled data to 8 WILDS datasets. For each dataset, we", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 279, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 370, + 291 + ], + "score": 1.0, + "content": "kept the labeled data from WILDS and expanded the datasets by", + "type": "text" + }, + { + "bbox": [ + 370, + 279, + 399, + 290 + ], + "score": 0.89, + "content": "3 { - } 1 3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 279, + 505, + 291 + ], + "score": 1.0, + "content": "with unlabeled data from", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 289, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 303 + ], + "score": 1.0, + "content": "the same underlying dataset. The type of unlabeled data (i.e., whether it comes from source, extra,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 302, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 505, + 312 + ], + "score": 1.0, + "content": "validation, or target domains) depends on what is realistic and available for the application. Beyond", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 312, + 504, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 504, + 323 + ], + "score": 1.0, + "content": "these 8 datasets, WILDS also contains 2 datasets without unlabeled data: the PY150-WILDS code", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "completion dataset and the RXRX1-WILDS genetic perturbation dataset. For all datasets, the labeled", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 333, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 345 + ], + "score": 1.0, + "content": "data and evaluation metrics are exactly the same as in WILDS 1.0. Figure adapted with permission", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 343, + 201, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 201, + 357 + ], + "score": 1.0, + "content": "from Koh et al. (2021).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6.5 + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 106, + 376, + 505, + 420 + ], + "lines": [], + "index": 12.5, + "bbox_fs": [ + 105, + 376, + 505, + 420 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 426, + 505, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "Finally, we note that WILDS 2.0 not a separate benchmark from WILDS 1.0: the labeled data and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 438, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 449 + ], + "score": 1.0, + "content": "evaluation metrics are exactly the same in WILDS 1.0 and WILDS 2.0, and future results should be", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 448, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 506, + 460 + ], + "score": 1.0, + "content": "reported on the overall WILDS benchmark, with a note describing what kind of unlabeled data (if", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "score": 1.0, + "content": "any) was used. In this paper, we discuss the addition of unlabeled data and analyze the performance", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "of methods that use the unlabeled data. For a more detailed description of the datasets, evaluation", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 481, + 451, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 451, + 494 + ], + "score": 1.0, + "content": "metrics, and models used, please refer to the original WILDS paper (Koh et al., 2021).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 426, + 506, + 494 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 509, + 502, + 521 + ], + "lines": [ + { + "bbox": [ + 105, + 507, + 504, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 504, + 523 + ], + "score": 1.0, + "content": "2 COMPARISON WITH EXISTING UNSUPERVISED ADAPTATION BENCHMARKS", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 505, + 566 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "WILDS 2.0 offers a diverse range of applications and modalities while also providing an extensive", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "amount of unlabeled data that can be used as leverage for training robust models. In this section, we", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 555, + 432, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 432, + 568 + ], + "score": 1.0, + "content": "briefly compare with other existing ML benchmarks for unsupervised adaptation.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 532, + 505, + 568 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "Images. Evaluations of unsupervised adaptation methods for image classification have focused on", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "generalizing from natural photos to a range of stylized images, such as sketches and cartoons (PACS", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "(Li et al., 2017), Office-Home (Venkateswara et al., 2017), and DomainNet (Peng et al., 2019)),", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "product images (Office-31 (Saenko et al., 2010)), and synthetic renderings (VisDA (Peng et al.,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 617, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 505, + 627 + ], + "score": 1.0, + "content": "2018)), though location-based shifts have also been recently explored (Dubey et al., 2021). It is", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "also popular to evaluate on shifts between digits datasets, such as MNIST (LeCun et al., 1998),", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "SVHN (Netzer et al., 2011), and USPS (Hull, 1994). In image detection and segmentation, existing", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "adaptation benchmarks tend to focus on generalizing from synthetic to natural scenes (Ros et al.,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "2016; Richter et al., 2016; Cordts et al., 2016; Hoffman et al., 2018), which can be an important", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "score": 1.0, + "content": "tool for realistic problems but is not the focus of this work. In contrast, WILDS considers real-world", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 681, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 694 + ], + "score": 1.0, + "content": "distribution shifts, and it spans diverse modalities (satellite, microscope, agriculture, and camera", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 693, + 349, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 349, + 705 + ], + "score": 1.0, + "content": "trap images) and tasks (classification, regression, detection).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 572, + 506, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Text. Methods for unsupervised adaptation in NLP are typically evaluated on domain shifts between", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "different textual sources, such as news articles, different categories of product reviews, Wikipedia,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 82, + 504, + 94 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 504, + 94 + ], + "score": 1.0, + "content": "or social media platforms (Blitzer et al., 2007; Mansour et al., 2009; Oren et al., 2019; Miller et al.,", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 92, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 92, + 505, + 105 + ], + "score": 1.0, + "content": "2020; Kamath et al., 2020; Hendrycks et al., 2020), or even more specialized sources such as legal", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 504, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 504, + 116 + ], + "score": 1.0, + "content": "documents (Chalkidis et al., 2020) or biomedical papers (Lee et al., 2020b; Gu et al., 2020). Multi-", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "lingual tasks can also be a setting for unsupervised adaptation (Conneau et al., 2018; Conneau &", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "Lample, 2019; Hu et al., 2020a; Clark et al., 2020), especially when generalizing to low-resource", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 149 + ], + "score": 1.0, + "content": "languages (Nekoto et al., 2020). The WILDS text datasets differ in that they focus on subpopulation", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 148, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 148, + 505, + 160 + ], + "score": 1.0, + "content": "performance, either to particular demographics in CIVILCOMMENTS-WILDS or to tail populations", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 417, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 417, + 171 + ], + "score": 1.0, + "content": "in AMAZON-WILDS, rather than on adapting to a completely distinct domain.", + "type": "text", + "cross_page": true + } + ], + "index": 7 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 171 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 504, + 94 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 504, + 94 + ], + "score": 1.0, + "content": "or social media platforms (Blitzer et al., 2007; Mansour et al., 2009; Oren et al., 2019; Miller et al.,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 92, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 92, + 505, + 105 + ], + "score": 1.0, + "content": "2020; Kamath et al., 2020; Hendrycks et al., 2020), or even more specialized sources such as legal", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 504, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 504, + 116 + ], + "score": 1.0, + "content": "documents (Chalkidis et al., 2020) or biomedical papers (Lee et al., 2020b; Gu et al., 2020). Multi-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "lingual tasks can also be a setting for unsupervised adaptation (Conneau et al., 2018; Conneau &", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "Lample, 2019; Hu et al., 2020a; Clark et al., 2020), especially when generalizing to low-resource", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 149 + ], + "score": 1.0, + "content": "languages (Nekoto et al., 2020). The WILDS text datasets differ in that they focus on subpopulation", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 148, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 148, + 505, + 160 + ], + "score": 1.0, + "content": "performance, either to particular demographics in CIVILCOMMENTS-WILDS or to tail populations", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 417, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 417, + 171 + ], + "score": 1.0, + "content": "in AMAZON-WILDS, rather than on adapting to a completely distinct domain.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 176, + 504, + 199 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 190 + ], + "score": 1.0, + "content": "Molecules. While unlabeled molecules have been used for pre-training (Hu et al., 2020c; Rong", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 461, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 461, + 200 + ], + "score": 1.0, + "content": "et al., 2020), no standardized unsupervised adaptation benchmarks have been developed.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 108, + 214, + 223, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 225, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 225, + 229 + ], + "score": 1.0, + "content": "3 PROBLEM SETTING", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 239, + 505, + 284 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 505, + 251 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 474, + 251 + ], + "score": 1.0, + "content": "As in WILDS 1.0, we study the domain shift setting where the data is drawn from domains", + "type": "text" + }, + { + "bbox": [ + 474, + 240, + 501, + 250 + ], + "score": 0.9, + "content": "d \\in \\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 239, + 505, + 251 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 163, + 262 + ], + "score": 1.0, + "content": "Each domain", + "type": "text" + }, + { + "bbox": [ + 163, + 251, + 170, + 260 + ], + "score": 0.72, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 250, + 310, + 262 + ], + "score": 1.0, + "content": "corresponds to a data distribution", + "type": "text" + }, + { + "bbox": [ + 310, + 251, + 323, + 261 + ], + "score": 0.89, + "content": "P _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 250, + 345, + 262 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 346, + 250, + 379, + 262 + ], + "score": 0.93, + "content": "( x , y , d )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 250, + 411, + 262 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 411, + 252, + 419, + 260 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 250, + 471, + 262 + ], + "score": 1.0, + "content": "is the input,", + "type": "text" + }, + { + "bbox": [ + 472, + 252, + 479, + 262 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 262, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 236, + 273 + ], + "score": 1.0, + "content": "prediction, and all points from", + "type": "text" + }, + { + "bbox": [ + 236, + 262, + 248, + 272 + ], + "score": 0.88, + "content": "P _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 262, + 306, + 273 + ], + "score": 1.0, + "content": "have domain", + "type": "text" + }, + { + "bbox": [ + 306, + 262, + 312, + 271 + ], + "score": 0.74, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 262, + 506, + 273 + ], + "score": 1.0, + "content": ". See Koh et al. (2021) for more details. The", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 271, + 223, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 223, + 285 + ], + "score": 1.0, + "content": "domains come in four types:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "table", + "bbox": [ + 110, + 294, + 501, + 361 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 294, + 501, + 361 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 294, + 501, + 361 + ], + "spans": [ + { + "bbox": [ + 110, + 294, + 501, + 361 + ], + "score": 0.98, + "html": "
Type of domainLabeled dataUnlabeled data
Source domainsExtra domainsValidation domainsTarget domainsUsed for trainingNoneUsed for hyperparameter tuningUsed for held-out evaluationCan be used for training, if available
", + "type": "table", + "image_path": "51c28112b45f4b13dd639c79d0ca870e0370d80d76315171c5476bf9844d8639.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 110, + 294, + 501, + 316.3333333333333 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 110, + 316.3333333333333, + 501, + 338.66666666666663 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 110, + 338.66666666666663, + 501, + 360.99999999999994 + ], + "spans": [], + "index": 17 + } + ] + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 369, + 504, + 392 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 505, + 382 + ], + "score": 1.0, + "content": "Table 1: All datasets have labeled source, validation, and target data, as well as unlabeled data from", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 380, + 429, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 429, + 393 + ], + "score": 1.0, + "content": "one or more types of domains, depending on what is realistic for the application.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 506, + 417 + ], + "score": 1.0, + "content": "We consider several variants of the domain shift setting. In some applications, all four types of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "score": 1.0, + "content": "domains are disjoint (e.g., if we are training on labeled data from some hospitals but seeking to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "generalize to new hospitals); in others, the target domains are a subset of the source domains (e.g., if", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "score": 1.0, + "content": "we are training on a heterogeneous dataset but seeking to measure model performance on particular", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 449, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 505, + 460 + ], + "score": 1.0, + "content": "demographic subpopulations). Models are trained on labeled data from the source domains, as well", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 459, + 504, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 504, + 471 + ], + "score": 1.0, + "content": "as unlabeled data of one or more types of domains, depending on what is realistic for the application.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5 + }, + { + "type": "title", + "bbox": [ + 107, + 486, + 179, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 181, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 181, + 501 + ], + "score": 1.0, + "content": "4 DATASETS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 511, + 505, + 599 + ], + "lines": [ + { + "bbox": [ + 107, + 512, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 107, + 512, + 505, + 523 + ], + "score": 1.0, + "content": "WILDS 2.0 augments 8 WILDS datasets with curated unlabeled data. For consistency, the labeled", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 523, + 504, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 504, + 534 + ], + "score": 1.0, + "content": "datasets and evaluation metrics are exactly the same as in WILDS 1.0, which allows direct evalua-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "tions of the utility of unlabeled training data. The labeled and unlabeled data are disjoint, e.g., the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 544, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 557 + ], + "score": 1.0, + "content": "unlabeled target data is different from the labeled target data used for evaluation. Here, we briefly", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 555, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 568 + ], + "score": 1.0, + "content": "describe each dataset, why unlabeled data can be realistically obtained for the corresponding task,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 566, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 578 + ], + "score": 1.0, + "content": "and how it might help. In Appendix A, we provide more information on each dataset, including data", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "score": 1.0, + "content": "provenance and details on data processing. In general, all of the unlabeled datasets in WILDS 2.0", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 588, + 466, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 466, + 600 + ], + "score": 1.0, + "content": "were processed in a similar way as their corresponding labeled datasets from WILDS 1.0.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 107, + 604, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "IWILDCAM2020-WILDS: Species classification across different camera traps. The task is to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 617, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 505, + 627 + ], + "score": 1.0, + "content": "classify the animal species in a camera trap image (Beery et al., 2020). We aim to generalize to new", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "camera trap locations despite variations in illumination, background, and label frequencies (Beery", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "et al., 2018). While hundreds of thousands of camera traps are active worldwide, only a small subset", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "of these traps have had images labeled, and the unlabeled data from the other camera traps capture", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "diverse operating conditions that can be used to learn robust models. In this work, we add unlabeled", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "images from 3,215 extra camera traps also in the WCS Camera Traps dataset (Beery et al., 2020).", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 681, + 452, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 286, + 694 + ], + "score": 1.0, + "content": "This expands the number of camera traps by", + "type": "text" + }, + { + "bbox": [ + 286, + 682, + 305, + 692 + ], + "score": 0.86, + "content": "1 1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 681, + 433, + 694 + ], + "score": 1.0, + "content": "and the number of examples by", + "type": "text" + }, + { + "bbox": [ + 433, + 682, + 447, + 693 + ], + "score": 0.86, + "content": "5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 681, + 452, + 694 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "CAMELYON17-WILDS: Tumor identification across different hospitals. The task is to clas-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "sify image patches from lymph node sections as tumor or normal tissue. We seek to generalize to", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "new hospitals, which can differ in their patient demographics and data acquisition protocols (Veta", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 171 + ], + "lines": [], + "index": 3.5, + "bbox_fs": [ + 105, + 82, + 505, + 171 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 176, + 504, + 199 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 190 + ], + "score": 1.0, + "content": "Molecules. While unlabeled molecules have been used for pre-training (Hu et al., 2020c; Rong", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 461, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 461, + 200 + ], + "score": 1.0, + "content": "et al., 2020), no standardized unsupervised adaptation benchmarks have been developed.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 174, + 506, + 200 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 214, + 223, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 225, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 225, + 229 + ], + "score": 1.0, + "content": "3 PROBLEM SETTING", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 239, + 505, + 284 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 505, + 251 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 474, + 251 + ], + "score": 1.0, + "content": "As in WILDS 1.0, we study the domain shift setting where the data is drawn from domains", + "type": "text" + }, + { + "bbox": [ + 474, + 240, + 501, + 250 + ], + "score": 0.9, + "content": "d \\in \\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 239, + 505, + 251 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 163, + 262 + ], + "score": 1.0, + "content": "Each domain", + "type": "text" + }, + { + "bbox": [ + 163, + 251, + 170, + 260 + ], + "score": 0.72, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 250, + 310, + 262 + ], + "score": 1.0, + "content": "corresponds to a data distribution", + "type": "text" + }, + { + "bbox": [ + 310, + 251, + 323, + 261 + ], + "score": 0.89, + "content": "P _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 250, + 345, + 262 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 346, + 250, + 379, + 262 + ], + "score": 0.93, + "content": "( x , y , d )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 250, + 411, + 262 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 411, + 252, + 419, + 260 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 250, + 471, + 262 + ], + "score": 1.0, + "content": "is the input,", + "type": "text" + }, + { + "bbox": [ + 472, + 252, + 479, + 262 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 262, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 236, + 273 + ], + "score": 1.0, + "content": "prediction, and all points from", + "type": "text" + }, + { + "bbox": [ + 236, + 262, + 248, + 272 + ], + "score": 0.88, + "content": "P _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 262, + 306, + 273 + ], + "score": 1.0, + "content": "have domain", + "type": "text" + }, + { + "bbox": [ + 306, + 262, + 312, + 271 + ], + "score": 0.74, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 262, + 506, + 273 + ], + "score": 1.0, + "content": ". See Koh et al. (2021) for more details. The", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 271, + 223, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 223, + 285 + ], + "score": 1.0, + "content": "domains come in four types:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 239, + 506, + 285 + ] + }, + { + "type": "table", + "bbox": [ + 110, + 294, + 501, + 361 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 294, + 501, + 361 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 294, + 501, + 361 + ], + "spans": [ + { + "bbox": [ + 110, + 294, + 501, + 361 + ], + "score": 0.98, + "html": "
Type of domainLabeled dataUnlabeled data
Source domainsExtra domainsValidation domainsTarget domainsUsed for trainingNoneUsed for hyperparameter tuningUsed for held-out evaluationCan be used for training, if available
", + "type": "table", + "image_path": "51c28112b45f4b13dd639c79d0ca870e0370d80d76315171c5476bf9844d8639.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 110, + 294, + 501, + 316.3333333333333 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 110, + 316.3333333333333, + 501, + 338.66666666666663 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 110, + 338.66666666666663, + 501, + 360.99999999999994 + ], + "spans": [], + "index": 17 + } + ] + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 369, + 504, + 392 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 505, + 382 + ], + "score": 1.0, + "content": "Table 1: All datasets have labeled source, validation, and target data, as well as unlabeled data from", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 380, + 429, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 429, + 393 + ], + "score": 1.0, + "content": "one or more types of domains, depending on what is realistic for the application.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 368, + 505, + 393 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 404, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 506, + 417 + ], + "score": 1.0, + "content": "We consider several variants of the domain shift setting. In some applications, all four types of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "score": 1.0, + "content": "domains are disjoint (e.g., if we are training on labeled data from some hospitals but seeking to", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "generalize to new hospitals); in others, the target domains are a subset of the source domains (e.g., if", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "score": 1.0, + "content": "we are training on a heterogeneous dataset but seeking to measure model performance on particular", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 449, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 505, + 460 + ], + "score": 1.0, + "content": "demographic subpopulations). Models are trained on labeled data from the source domains, as well", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 459, + 504, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 504, + 471 + ], + "score": 1.0, + "content": "as unlabeled data of one or more types of domains, depending on what is realistic for the application.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 403, + 506, + 471 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 486, + 179, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 485, + 181, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 181, + 501 + ], + "score": 1.0, + "content": "4 DATASETS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 511, + 505, + 599 + ], + "lines": [ + { + "bbox": [ + 107, + 512, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 107, + 512, + 505, + 523 + ], + "score": 1.0, + "content": "WILDS 2.0 augments 8 WILDS datasets with curated unlabeled data. For consistency, the labeled", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 523, + 504, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 504, + 534 + ], + "score": 1.0, + "content": "datasets and evaluation metrics are exactly the same as in WILDS 1.0, which allows direct evalua-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "tions of the utility of unlabeled training data. The labeled and unlabeled data are disjoint, e.g., the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 544, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 557 + ], + "score": 1.0, + "content": "unlabeled target data is different from the labeled target data used for evaluation. Here, we briefly", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 555, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 505, + 568 + ], + "score": 1.0, + "content": "describe each dataset, why unlabeled data can be realistically obtained for the corresponding task,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 566, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 578 + ], + "score": 1.0, + "content": "and how it might help. In Appendix A, we provide more information on each dataset, including data", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "score": 1.0, + "content": "provenance and details on data processing. In general, all of the unlabeled datasets in WILDS 2.0", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 588, + 466, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 466, + 600 + ], + "score": 1.0, + "content": "were processed in a similar way as their corresponding labeled datasets from WILDS 1.0.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 512, + 506, + 600 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 604, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "IWILDCAM2020-WILDS: Species classification across different camera traps. The task is to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 617, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 505, + 627 + ], + "score": 1.0, + "content": "classify the animal species in a camera trap image (Beery et al., 2020). We aim to generalize to new", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "camera trap locations despite variations in illumination, background, and label frequencies (Beery", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "et al., 2018). While hundreds of thousands of camera traps are active worldwide, only a small subset", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "of these traps have had images labeled, and the unlabeled data from the other camera traps capture", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "diverse operating conditions that can be used to learn robust models. In this work, we add unlabeled", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "images from 3,215 extra camera traps also in the WCS Camera Traps dataset (Beery et al., 2020).", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 681, + 452, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 286, + 694 + ], + "score": 1.0, + "content": "This expands the number of camera traps by", + "type": "text" + }, + { + "bbox": [ + 286, + 682, + 305, + 692 + ], + "score": 0.86, + "content": "1 1 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 681, + 433, + 694 + ], + "score": 1.0, + "content": "and the number of examples by", + "type": "text" + }, + { + "bbox": [ + 433, + 682, + 447, + 693 + ], + "score": 0.86, + "content": "5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 681, + 452, + 694 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 605, + 505, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "CAMELYON17-WILDS: Tumor identification across different hospitals. The task is to clas-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "sify image patches from lymph node sections as tumor or normal tissue. We seek to generalize to", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "new hospitals, which can differ in their patient demographics and data acquisition protocols (Veta", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 82, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 93 + ], + "score": 1.0, + "content": "et al., 2016; AlBadawy et al., 2018; Komura & Ishikawa, 2018; Tellez et al., 2019). While ob-", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "taining labeled data for histopathology applications requires pain-staking annotations from expert", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "pathologists, hospitals typically accumulate unlabeled slide images during normal operation. These", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "score": 1.0, + "content": "unlabeled images could be used to adapt to differences between hospitals (e.g., different staining", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "protocols might lead to different color distributions). We provide unlabeled patches from train and", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 351, + 149 + ], + "score": 1.0, + "content": "test hospitals, which expands the total number of patches by", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 351, + 137, + 373, + 148 + ], + "score": 0.87, + "content": "7 . 5 \\times", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 373, + 137, + 505, + 149 + ], + "score": 1.0, + "content": ". Both the labeled and unlabeled", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 376, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 376, + 160 + ], + "score": 1.0, + "content": "data are adapted from the Camelyon17 dataset (Bandi et al., 2018).", + "type": "text", + "cross_page": true + } + ], + "index": 6 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 699, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 159 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 93 + ], + "score": 1.0, + "content": "et al., 2016; AlBadawy et al., 2018; Komura & Ishikawa, 2018; Tellez et al., 2019). While ob-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "taining labeled data for histopathology applications requires pain-staking annotations from expert", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "pathologists, hospitals typically accumulate unlabeled slide images during normal operation. These", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "score": 1.0, + "content": "unlabeled images could be used to adapt to differences between hospitals (e.g., different staining", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "protocols might lead to different color distributions). We provide unlabeled patches from train and", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 351, + 149 + ], + "score": 1.0, + "content": "test hospitals, which expands the total number of patches by", + "type": "text" + }, + { + "bbox": [ + 351, + 137, + 373, + 148 + ], + "score": 0.87, + "content": "7 . 5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 137, + 505, + 149 + ], + "score": 1.0, + "content": ". Both the labeled and unlabeled", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 376, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 376, + 160 + ], + "score": 1.0, + "content": "data are adapted from the Camelyon17 dataset (Bandi et al., 2018).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 505, + 275 + ], + "lines": [ + { + "bbox": [ + 106, + 165, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 505, + 177 + ], + "score": 1.0, + "content": "FMOW-WILDS: Land use classification across different regions and years. The task is to clas-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "sify the type of building or land usage in a satellite image. Given training data from before 2013, we", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 188, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 505, + 199 + ], + "score": 1.0, + "content": "aim to generalize to satellite imagery taken after 2013, while maintaining high accuracy across all", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "score": 1.0, + "content": "geographic regions. While labeling land use requires combining map data and expert annotations,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 209, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 209, + 506, + 222 + ], + "score": 1.0, + "content": "unlabeled data is available in all locations in the world through constant streams of global satellite", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "imagery. Prior work has shown that unlabeled satellite data can improve OOD accuracy in land-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 506, + 243 + ], + "score": 1.0, + "content": "cover and cropland prediction (Xie et al., 2021a) as well as aerial object and scene classification", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 241, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 506, + 254 + ], + "score": 1.0, + "content": "(Reed et al., 2021). We provide unlabeled satellite imagery across all regions from the train and test", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 333, + 264 + ], + "score": 1.0, + "content": "timeframes defined in WILDS, expanding the dataset by", + "type": "text" + }, + { + "bbox": [ + 333, + 253, + 354, + 264 + ], + "score": 0.9, + "content": "3 . 5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 253, + 505, + 264 + ], + "score": 1.0, + "content": ". Both the labeled and unlabeled data", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 343, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 343, + 276 + ], + "score": 1.0, + "content": "are adapted from the FMoW dataset (Christie et al., 2018).", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 505, + 380 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 293 + ], + "score": 1.0, + "content": "POVERTYMAP-WILDS: Poverty mapping across different countries. The task is to predict a", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 290, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 305 + ], + "score": 1.0, + "content": "real-valued asset wealth index of the area in a satellite image. We consider generalizing across", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 302, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 315 + ], + "score": 1.0, + "content": "different countries. Like FMOW-WILDS, unlabeled satellite imagery is available globally, while", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 313, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 505, + 326 + ], + "score": 1.0, + "content": "labeled data is expensive to collect as it requires conducting nationally representative surveys in the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 324, + 504, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 504, + 337 + ], + "score": 1.0, + "content": "field. Prior work on poverty prediction has used unlabeled data for entropy minimization (Jean et al.,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "2018) and pre-training on auxiliary tasks such as nighttime light prediction (Xie et al., 2016; Jean", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 505, + 358 + ], + "score": 1.0, + "content": "et al., 2016), but these studies do not study generalization to new countries. We provide unlabeled", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 357, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 414, + 370 + ], + "score": 1.0, + "content": "satellite imagery from both train and test countries, expanding the dataset by", + "type": "text" + }, + { + "bbox": [ + 414, + 358, + 433, + 369 + ], + "score": 0.88, + "content": "1 4 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 357, + 505, + 370 + ], + "score": 1.0, + "content": ". Both the labeled", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 367, + 325, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 325, + 381 + ], + "score": 1.0, + "content": "and unlabeled data are adapted from Yeh et al. (2020).", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 505, + 495 + ], + "lines": [ + { + "bbox": [ + 106, + 385, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 505, + 397 + ], + "score": 1.0, + "content": "GLOBALWHEAT-WILDS: Wheat head detection across different regions. The task is to localize", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "score": 1.0, + "content": "wheat heads in overhead field images. We seek to generalize across image acquisition sessions,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 406, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 506, + 421 + ], + "score": 1.0, + "content": "each of which represents a particular location, time, and sensor; these can differ in wheat genotype,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "score": 1.0, + "content": "wheat head appearance, growing conditions, background appearance, illumination, and acquisition", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 428, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 506, + 443 + ], + "score": 1.0, + "content": "protocols. Wheat field images contain many densely packed and overlapping instances, making", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 440, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 506, + 453 + ], + "score": 1.0, + "content": "labeling wheat heads in images costly, tedious and sensitive to the individual annotator. However,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 451, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 462 + ], + "score": 1.0, + "content": "hundreds of agricultural research institutes around the world collect terabytes of unlabeled field", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 462, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 506, + 474 + ], + "score": 1.0, + "content": "images which could be used for training. We add unlabeled field images from train, test, and extra", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 473, + 504, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 289, + 484 + ], + "score": 1.0, + "content": "acquisition sessions, expanding the dataset by", + "type": "text" + }, + { + "bbox": [ + 289, + 473, + 308, + 484 + ], + "score": 0.87, + "content": "1 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 474, + 504, + 484 + ], + "score": 1.0, + "content": ". The labeled and unlabeled data are adapted from", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 484, + 488, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 488, + 496 + ], + "score": 1.0, + "content": "the Global Wheat Head Detection dataset and its underlying sources (David et al., 2020; 2021).", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 501, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "OGB-MOLPCBA: Molecular property prediction across different scaffolds. The task is to", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 512, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 524 + ], + "score": 1.0, + "content": "predict the biological activity of small molecules represented as molecular graphs (Wu et al., 2018;", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 522, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 505, + 535 + ], + "score": 1.0, + "content": "Hu et al., 2020b). We seek to generalize to molecules with new scaffold structures. Labels on", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "biological activity are only available for a small portion of molecules, as they require expensive lab", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "experiments to obtain. However, unlabeled molecule structures are readily available in large-scale", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 555, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 568 + ], + "score": 1.0, + "content": "chemical databases such as PubChem (Bolton et al., 2008), and have been previously used for pre-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "training (Hu et al., 2020c) and semi-supervised learning (Sun et al., 2020). We provide 5 million", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "score": 1.0, + "content": "unlabeled molecules from source and target scaffolds, which expands the number of molecules by", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 589, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 133, + 599 + ], + "score": 0.86, + "content": "1 2 . 5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 589, + 505, + 600 + ], + "score": 1.0, + "content": ". The original labeled data was curated by MoleculeNet (Wu et al., 2018) from PubChem, and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 599, + 421, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 421, + 612 + ], + "score": 1.0, + "content": "we similarly extracted the unlabeled data from PubChem (Bolton et al., 2008).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 107, + 616, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 628 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS: Toxicity classification across demographic identities. The task is to", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "classify whether a text comment is toxic or not. We consider the subpopulation shift setting, where", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 639, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 651 + ], + "score": 1.0, + "content": "the model must classify accurately across groups of comments mentioning different demographic", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "identities. While labels require large-scale crowdsourcing annotations on both comment toxicity,", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "unlabeled article comments are widely available on the internet. We provide unannotated comments", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 349, + 683 + ], + "score": 1.0, + "content": "as unlabeled data, which expands the size of the dataset by", + "type": "text" + }, + { + "bbox": [ + 349, + 672, + 370, + 682 + ], + "score": 0.86, + "content": "4 . 5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 671, + 505, + 683 + ], + "score": 1.0, + "content": ". Both the labeled and unlabeled", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 683, + 280, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 683, + 280, + 694 + ], + "score": 1.0, + "content": "data are adapted from Borkan et al. (2019).", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 49 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 502, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "AMAZON-WILDS: Sentiment classification across different users. The task is to classify the star", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 710, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 504, + 722 + ], + "score": 1.0, + "content": "ratings of Amazon reviews. We seek to perform consistently well across new reviewers. While the", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 53.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 763 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 159 + ], + "lines": [], + "index": 3, + "bbox_fs": [ + 105, + 82, + 506, + 160 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 505, + 275 + ], + "lines": [ + { + "bbox": [ + 106, + 165, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 505, + 177 + ], + "score": 1.0, + "content": "FMOW-WILDS: Land use classification across different regions and years. The task is to clas-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "sify the type of building or land usage in a satellite image. Given training data from before 2013, we", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 188, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 505, + 199 + ], + "score": 1.0, + "content": "aim to generalize to satellite imagery taken after 2013, while maintaining high accuracy across all", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "score": 1.0, + "content": "geographic regions. While labeling land use requires combining map data and expert annotations,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 209, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 209, + 506, + 222 + ], + "score": 1.0, + "content": "unlabeled data is available in all locations in the world through constant streams of global satellite", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "imagery. Prior work has shown that unlabeled satellite data can improve OOD accuracy in land-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 506, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 506, + 243 + ], + "score": 1.0, + "content": "cover and cropland prediction (Xie et al., 2021a) as well as aerial object and scene classification", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 241, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 506, + 254 + ], + "score": 1.0, + "content": "(Reed et al., 2021). We provide unlabeled satellite imagery across all regions from the train and test", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 333, + 264 + ], + "score": 1.0, + "content": "timeframes defined in WILDS, expanding the dataset by", + "type": "text" + }, + { + "bbox": [ + 333, + 253, + 354, + 264 + ], + "score": 0.9, + "content": "3 . 5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 253, + 505, + 264 + ], + "score": 1.0, + "content": ". Both the labeled and unlabeled data", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 343, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 343, + 276 + ], + "score": 1.0, + "content": "are adapted from the FMoW dataset (Christie et al., 2018).", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 165, + 506, + 276 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 505, + 380 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 293 + ], + "score": 1.0, + "content": "POVERTYMAP-WILDS: Poverty mapping across different countries. The task is to predict a", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 290, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 305 + ], + "score": 1.0, + "content": "real-valued asset wealth index of the area in a satellite image. We consider generalizing across", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 302, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 315 + ], + "score": 1.0, + "content": "different countries. Like FMOW-WILDS, unlabeled satellite imagery is available globally, while", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 313, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 505, + 326 + ], + "score": 1.0, + "content": "labeled data is expensive to collect as it requires conducting nationally representative surveys in the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 324, + 504, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 504, + 337 + ], + "score": 1.0, + "content": "field. Prior work on poverty prediction has used unlabeled data for entropy minimization (Jean et al.,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "2018) and pre-training on auxiliary tasks such as nighttime light prediction (Xie et al., 2016; Jean", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 346, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 505, + 358 + ], + "score": 1.0, + "content": "et al., 2016), but these studies do not study generalization to new countries. We provide unlabeled", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 357, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 414, + 370 + ], + "score": 1.0, + "content": "satellite imagery from both train and test countries, expanding the dataset by", + "type": "text" + }, + { + "bbox": [ + 414, + 358, + 433, + 369 + ], + "score": 0.88, + "content": "1 4 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 357, + 505, + 370 + ], + "score": 1.0, + "content": ". Both the labeled", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 367, + 325, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 325, + 381 + ], + "score": 1.0, + "content": "and unlabeled data are adapted from Yeh et al. (2020).", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 280, + 506, + 381 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 505, + 495 + ], + "lines": [ + { + "bbox": [ + 106, + 385, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 505, + 397 + ], + "score": 1.0, + "content": "GLOBALWHEAT-WILDS: Wheat head detection across different regions. The task is to localize", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "score": 1.0, + "content": "wheat heads in overhead field images. We seek to generalize across image acquisition sessions,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 406, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 506, + 421 + ], + "score": 1.0, + "content": "each of which represents a particular location, time, and sensor; these can differ in wheat genotype,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 505, + 431 + ], + "score": 1.0, + "content": "wheat head appearance, growing conditions, background appearance, illumination, and acquisition", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 428, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 506, + 443 + ], + "score": 1.0, + "content": "protocols. Wheat field images contain many densely packed and overlapping instances, making", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 440, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 506, + 453 + ], + "score": 1.0, + "content": "labeling wheat heads in images costly, tedious and sensitive to the individual annotator. However,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 451, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 462 + ], + "score": 1.0, + "content": "hundreds of agricultural research institutes around the world collect terabytes of unlabeled field", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 462, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 506, + 474 + ], + "score": 1.0, + "content": "images which could be used for training. We add unlabeled field images from train, test, and extra", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 473, + 504, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 289, + 484 + ], + "score": 1.0, + "content": "acquisition sessions, expanding the dataset by", + "type": "text" + }, + { + "bbox": [ + 289, + 473, + 308, + 484 + ], + "score": 0.87, + "content": "1 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 474, + 504, + 484 + ], + "score": 1.0, + "content": ". The labeled and unlabeled data are adapted from", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 484, + 488, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 488, + 496 + ], + "score": 1.0, + "content": "the Global Wheat Head Detection dataset and its underlying sources (David et al., 2020; 2021).", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 385, + 506, + 496 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 501, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "OGB-MOLPCBA: Molecular property prediction across different scaffolds. The task is to", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 512, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 505, + 524 + ], + "score": 1.0, + "content": "predict the biological activity of small molecules represented as molecular graphs (Wu et al., 2018;", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 522, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 505, + 535 + ], + "score": 1.0, + "content": "Hu et al., 2020b). We seek to generalize to molecules with new scaffold structures. Labels on", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "biological activity are only available for a small portion of molecules, as they require expensive lab", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "experiments to obtain. However, unlabeled molecule structures are readily available in large-scale", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 555, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 568 + ], + "score": 1.0, + "content": "chemical databases such as PubChem (Bolton et al., 2008), and have been previously used for pre-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "training (Hu et al., 2020c) and semi-supervised learning (Sun et al., 2020). We provide 5 million", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 505, + 590 + ], + "score": 1.0, + "content": "unlabeled molecules from source and target scaffolds, which expands the number of molecules by", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 589, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 133, + 599 + ], + "score": 0.86, + "content": "1 2 . 5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 589, + 505, + 600 + ], + "score": 1.0, + "content": ". The original labeled data was curated by MoleculeNet (Wu et al., 2018) from PubChem, and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 599, + 421, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 421, + 612 + ], + "score": 1.0, + "content": "we similarly extracted the unlabeled data from PubChem (Bolton et al., 2008).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 500, + 506, + 612 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 616, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 505, + 628 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS: Toxicity classification across demographic identities. The task is to", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "classify whether a text comment is toxic or not. We consider the subpopulation shift setting, where", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 639, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 651 + ], + "score": 1.0, + "content": "the model must classify accurately across groups of comments mentioning different demographic", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 649, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 662 + ], + "score": 1.0, + "content": "identities. While labels require large-scale crowdsourcing annotations on both comment toxicity,", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 660, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 505, + 673 + ], + "score": 1.0, + "content": "unlabeled article comments are widely available on the internet. We provide unannotated comments", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 349, + 683 + ], + "score": 1.0, + "content": "as unlabeled data, which expands the size of the dataset by", + "type": "text" + }, + { + "bbox": [ + 349, + 672, + 370, + 682 + ], + "score": 0.86, + "content": "4 . 5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 671, + 505, + 683 + ], + "score": 1.0, + "content": ". Both the labeled and unlabeled", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 106, + 683, + 280, + 694 + ], + "spans": [ + { + "bbox": [ + 106, + 683, + 280, + 694 + ], + "score": 1.0, + "content": "data are adapted from Borkan et al. (2019).", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 49, + "bbox_fs": [ + 105, + 615, + 505, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 502, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "AMAZON-WILDS: Sentiment classification across different users. The task is to classify the star", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 710, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 504, + 722 + ], + "score": 1.0, + "content": "ratings of Amazon reviews. We seek to perform consistently well across new reviewers. While the", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "labels (star ratings) are always available for Amazon reviews in practice, unlabeled data is a common", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "source of leverage for sentiment classification more generally, with prior work in domain adaptation", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 505, + 117 + ], + "score": 1.0, + "content": "(Blitzer & Pereira, 2007; Glorot et al., 2011) and semi-supervised learning (Dasgupta & Ng, 2009;", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "Li et al., 2011). We provide unlabeled reviews from test and extra reviewers, which expands the", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 215, + 138 + ], + "score": 1.0, + "content": "total number of reviews by", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 215, + 127, + 236, + 137 + ], + "score": 0.86, + "content": "7 . 5 \\times", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 237, + 127, + 505, + 138 + ], + "score": 1.0, + "content": ". Both the labeled and unlabeled data are adapted from the Amazon", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 243, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 243, + 149 + ], + "score": 1.0, + "content": "review dataset by Ni et al. (2019).", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 53.5, + "bbox_fs": [ + 105, + 698, + 505, + 722 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "labels (star ratings) are always available for Amazon reviews in practice, unlabeled data is a common", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "source of leverage for sentiment classification more generally, with prior work in domain adaptation", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 505, + 117 + ], + "score": 1.0, + "content": "(Blitzer & Pereira, 2007; Glorot et al., 2011) and semi-supervised learning (Dasgupta & Ng, 2009;", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "Li et al., 2011). We provide unlabeled reviews from test and extra reviewers, which expands the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 215, + 138 + ], + "score": 1.0, + "content": "total number of reviews by", + "type": "text" + }, + { + "bbox": [ + 215, + 127, + 236, + 137 + ], + "score": 0.86, + "content": "7 . 5 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 127, + 505, + 138 + ], + "score": 1.0, + "content": ". Both the labeled and unlabeled data are adapted from the Amazon", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 243, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 243, + 149 + ], + "score": 1.0, + "content": "review dataset by Ni et al. (2019).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 107, + 168, + 196, + 180 + ], + "lines": [ + { + "bbox": [ + 105, + 166, + 198, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 166, + 198, + 182 + ], + "score": 1.0, + "content": "5 ALGORITHMS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 194, + 505, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 194, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 505, + 206 + ], + "score": 1.0, + "content": "For our evaluation, we selected representative methods from the three categories described below.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 204, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 505, + 218 + ], + "score": 1.0, + "content": "These methods exemplify current approaches to using unlabeled data to improve robustness, and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 215, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 230 + ], + "score": 1.0, + "content": "they have been successful on popular domain adaptation benchmarks like DomainNet (Peng et al.,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 227, + 504, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 504, + 239 + ], + "score": 1.0, + "content": "2019) and semi-supervised settings like improving ImageNet accuracy by leveraging unlabeled im-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 237, + 484, + 251 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 484, + 251 + ], + "score": 1.0, + "content": "ages from the internet (Xie et al., 2020; Caron et al., 2020). For more details, see Appendix B.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 255, + 505, + 310 + ], + "lines": [ + { + "bbox": [ + 107, + 255, + 504, + 266 + ], + "spans": [ + { + "bbox": [ + 107, + 255, + 504, + 266 + ], + "score": 1.0, + "content": "Domain-invariant methods. Domain-invariant methods learn feature representations that are in-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 264, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 279 + ], + "score": 1.0, + "content": "variant across different domains by penalizing differences between learned source and target rep-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "resentations (Long et al., 2015; Ganin et al., 2016; Sun & Saenko, 2016; Long et al., 2017; 2018;", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 288, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 505, + 300 + ], + "score": 1.0, + "content": "Saito et al., 2018; Zhang et al., 2018; Xu et al., 2019; Zhang et al., 2019b). We discuss these methods", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 299, + 435, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 435, + 311 + ], + "score": 1.0, + "content": "further in Appendix B.2. For our experiments, we evaluate two classical methods:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 321, + 505, + 372 + ], + "lines": [ + { + "bbox": [ + 106, + 320, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 505, + 334 + ], + "score": 1.0, + "content": "• Domain-Adversarial Neural Networks (DANN) (Ganin et al., 2016) penalize representations on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 331, + 473, + 345 + ], + "spans": [ + { + "bbox": [ + 115, + 331, + 473, + 345 + ], + "score": 1.0, + "content": "which an auxiliary classifier can easily discriminate between source and target examples.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 348, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 107, + 348, + 505, + 362 + ], + "score": 1.0, + "content": "• Correlation Alignment (CORAL) (Sun et al., 2016; Sun & Saenko, 2016) penalizes differences", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 360, + 441, + 373 + ], + "spans": [ + { + "bbox": [ + 115, + 360, + 441, + 373 + ], + "score": 1.0, + "content": "between the means and covariances of the source and target feature distributions.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 107, + 383, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "score": 1.0, + "content": "Self-training. Self-training methods “pseudo-label” unlabeled examples with the model’s own pre-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 392, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 505, + 406 + ], + "score": 1.0, + "content": "dictions and then train on them as if they were labeled examples. These methods often also use con-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "score": 1.0, + "content": "sistency regularization, which encourages the model to make consistent predictions on augmented", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "views of unlabeled examples (Sohn et al., 2020; Xie et al., 2020; Berthelot et al., 2021). Self-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 426, + 504, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 504, + 438 + ], + "score": 1.0, + "content": "training methods have recently been successfully applied to unsupervised adaptation (Saito et al.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 436, + 477, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 477, + 450 + ], + "score": 1.0, + "content": "2017; Berthelot et al., 2021; Zhang et al., 2021). We include three representative algorithms:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 459, + 505, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 459, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 505, + 472 + ], + "score": 1.0, + "content": "• Pseudo-Label (Lee, 2013) dynamically generates pseudolabels and updates the model each batch.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 476, + 504, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 504, + 490 + ], + "score": 1.0, + "content": "• FixMatch (Sohn et al., 2020) adds consistency regularization on top of the Pseudo-Label algo-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 487, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 115, + 487, + 505, + 501 + ], + "score": 1.0, + "content": "rithm. Specifically, it generates pseudolabels on a weakly augmented view of the unlabeled data,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 500, + 456, + 511 + ], + "spans": [ + { + "bbox": [ + 116, + 500, + 456, + 511 + ], + "score": 1.0, + "content": "and then minimizes the loss of the model’s prediction on a strongly augmented view.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 107, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "• Noisy Student (Xie et al., 2020) leverages weak and strong augmentations like FixMatch, but", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 115, + 528, + 505, + 540 + ], + "score": 1.0, + "content": "instead of dynamically generating pseudolabels for each batch, it alternates between a few teacher", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 116, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "phases, where it generates pseudolabels, and student phases, where it trains to convergence on the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 550, + 204, + 561 + ], + "spans": [ + { + "bbox": [ + 116, + 550, + 204, + 561 + ], + "score": 1.0, + "content": "(pseudo)labeled data.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 571, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 506, + 583 + ], + "score": 1.0, + "content": "Self-supervision. Self-supervised methods learn useful representations by training on unlabeled", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 582, + 504, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 504, + 594 + ], + "score": 1.0, + "content": "data via auxiliary proxy tasks. Common approaches include reconstruction tasks (Vincent et al.,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 592, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 505, + 606 + ], + "score": 1.0, + "content": "2008; Erhan et al., 2010; Devlin et al., 2019; Gidaris et al., 2018; Lewis et al., 2020), and contrastive", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 603, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 616 + ], + "score": 1.0, + "content": "learning (He et al., 2020; Chen et al., 2020b; Caron et al., 2020; Radford et al., 2021b), and recent", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "score": 1.0, + "content": "work has shown that self-supervised methods can reduce dependence on spurious correlations and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "improve performance on domain adaptation tasks (Wang et al., 2021; Tsai et al., 2021; Mishra", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 636, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 505, + 650 + ], + "score": 1.0, + "content": "et al., 2021). We use these self-supervision methods for unsupervised adaptation by first pre-training", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 648, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 504, + 660 + ], + "score": 1.0, + "content": "models on the unlabeled data, and then finetuning them on the labeled source data (Shen et al., 2021).", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 658, + 388, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 388, + 673 + ], + "score": 1.0, + "content": "We evaluate popular self-supervised methods for vision and language:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 681, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 681, + 506, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 506, + 693 + ], + "score": 1.0, + "content": "• SwAV (Caron et al., 2020) is a contrastive learning algorithm that maps representations to a set of", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 116, + 692, + 381, + 704 + ], + "spans": [ + { + "bbox": [ + 116, + 692, + 381, + 704 + ], + "score": 1.0, + "content": "clusters and then enforces similarity between cluster assignments.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "• Masked language modeling (MLM) (Devlin et al., 2019) randomly masks some of the tokens from", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 721, + 359, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 359, + 732 + ], + "score": 1.0, + "content": "input text and trains the model to predict the missing tokens.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [], + "index": 2.5, + "bbox_fs": [ + 105, + 81, + 505, + 149 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 168, + 196, + 180 + ], + "lines": [ + { + "bbox": [ + 105, + 166, + 198, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 166, + 198, + 182 + ], + "score": 1.0, + "content": "5 ALGORITHMS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 194, + 505, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 194, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 505, + 206 + ], + "score": 1.0, + "content": "For our evaluation, we selected representative methods from the three categories described below.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 204, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 505, + 218 + ], + "score": 1.0, + "content": "These methods exemplify current approaches to using unlabeled data to improve robustness, and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 215, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 505, + 230 + ], + "score": 1.0, + "content": "they have been successful on popular domain adaptation benchmarks like DomainNet (Peng et al.,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 227, + 504, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 504, + 239 + ], + "score": 1.0, + "content": "2019) and semi-supervised settings like improving ImageNet accuracy by leveraging unlabeled im-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 237, + 484, + 251 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 484, + 251 + ], + "score": 1.0, + "content": "ages from the internet (Xie et al., 2020; Caron et al., 2020). For more details, see Appendix B.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 194, + 505, + 251 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 255, + 505, + 310 + ], + "lines": [ + { + "bbox": [ + 107, + 255, + 504, + 266 + ], + "spans": [ + { + "bbox": [ + 107, + 255, + 504, + 266 + ], + "score": 1.0, + "content": "Domain-invariant methods. Domain-invariant methods learn feature representations that are in-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 264, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 279 + ], + "score": 1.0, + "content": "variant across different domains by penalizing differences between learned source and target rep-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "resentations (Long et al., 2015; Ganin et al., 2016; Sun & Saenko, 2016; Long et al., 2017; 2018;", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 288, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 505, + 300 + ], + "score": 1.0, + "content": "Saito et al., 2018; Zhang et al., 2018; Xu et al., 2019; Zhang et al., 2019b). We discuss these methods", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 299, + 435, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 435, + 311 + ], + "score": 1.0, + "content": "further in Appendix B.2. For our experiments, we evaluate two classical methods:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 255, + 505, + 311 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 321, + 505, + 372 + ], + "lines": [ + { + "bbox": [ + 106, + 320, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 505, + 334 + ], + "score": 1.0, + "content": "• Domain-Adversarial Neural Networks (DANN) (Ganin et al., 2016) penalize representations on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 331, + 473, + 345 + ], + "spans": [ + { + "bbox": [ + 115, + 331, + 473, + 345 + ], + "score": 1.0, + "content": "which an auxiliary classifier can easily discriminate between source and target examples.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 348, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 107, + 348, + 505, + 362 + ], + "score": 1.0, + "content": "• Correlation Alignment (CORAL) (Sun et al., 2016; Sun & Saenko, 2016) penalizes differences", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 360, + 441, + 373 + ], + "spans": [ + { + "bbox": [ + 115, + 360, + 441, + 373 + ], + "score": 1.0, + "content": "between the means and covariances of the source and target feature distributions.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 106, + 320, + 505, + 373 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 383, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "score": 1.0, + "content": "Self-training. Self-training methods “pseudo-label” unlabeled examples with the model’s own pre-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 392, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 505, + 406 + ], + "score": 1.0, + "content": "dictions and then train on them as if they were labeled examples. These methods often also use con-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 506, + 417 + ], + "score": 1.0, + "content": "sistency regularization, which encourages the model to make consistent predictions on augmented", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "views of unlabeled examples (Sohn et al., 2020; Xie et al., 2020; Berthelot et al., 2021). Self-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 426, + 504, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 504, + 438 + ], + "score": 1.0, + "content": "training methods have recently been successfully applied to unsupervised adaptation (Saito et al.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 436, + 477, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 477, + 450 + ], + "score": 1.0, + "content": "2017; Berthelot et al., 2021; Zhang et al., 2021). We include three representative algorithms:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 381, + 506, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 459, + 505, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 459, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 505, + 472 + ], + "score": 1.0, + "content": "• Pseudo-Label (Lee, 2013) dynamically generates pseudolabels and updates the model each batch.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 476, + 504, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 504, + 490 + ], + "score": 1.0, + "content": "• FixMatch (Sohn et al., 2020) adds consistency regularization on top of the Pseudo-Label algo-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 487, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 115, + 487, + 505, + 501 + ], + "score": 1.0, + "content": "rithm. Specifically, it generates pseudolabels on a weakly augmented view of the unlabeled data,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 500, + 456, + 511 + ], + "spans": [ + { + "bbox": [ + 116, + 500, + 456, + 511 + ], + "score": 1.0, + "content": "and then minimizes the loss of the model’s prediction on a strongly augmented view.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 107, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "• Noisy Student (Xie et al., 2020) leverages weak and strong augmentations like FixMatch, but", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 115, + 528, + 505, + 540 + ], + "score": 1.0, + "content": "instead of dynamically generating pseudolabels for each batch, it alternates between a few teacher", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 116, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "phases, where it generates pseudolabels, and student phases, where it trains to convergence on the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 550, + 204, + 561 + ], + "spans": [ + { + "bbox": [ + 116, + 550, + 204, + 561 + ], + "score": 1.0, + "content": "(pseudo)labeled data.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 30.5, + "bbox_fs": [ + 106, + 459, + 505, + 561 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 571, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 506, + 583 + ], + "score": 1.0, + "content": "Self-supervision. Self-supervised methods learn useful representations by training on unlabeled", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 582, + 504, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 504, + 594 + ], + "score": 1.0, + "content": "data via auxiliary proxy tasks. Common approaches include reconstruction tasks (Vincent et al.,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 592, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 505, + 606 + ], + "score": 1.0, + "content": "2008; Erhan et al., 2010; Devlin et al., 2019; Gidaris et al., 2018; Lewis et al., 2020), and contrastive", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 603, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 505, + 616 + ], + "score": 1.0, + "content": "learning (He et al., 2020; Chen et al., 2020b; Caron et al., 2020; Radford et al., 2021b), and recent", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 504, + 626 + ], + "score": 1.0, + "content": "work has shown that self-supervised methods can reduce dependence on spurious correlations and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 639 + ], + "score": 1.0, + "content": "improve performance on domain adaptation tasks (Wang et al., 2021; Tsai et al., 2021; Mishra", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 636, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 505, + 650 + ], + "score": 1.0, + "content": "et al., 2021). We use these self-supervision methods for unsupervised adaptation by first pre-training", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 648, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 504, + 660 + ], + "score": 1.0, + "content": "models on the unlabeled data, and then finetuning them on the labeled source data (Shen et al., 2021).", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 658, + 388, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 388, + 673 + ], + "score": 1.0, + "content": "We evaluate popular self-supervised methods for vision and language:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 570, + 506, + 673 + ] + }, + { + "type": "list", + "bbox": [ + 106, + 681, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 681, + 506, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 506, + 693 + ], + "score": 1.0, + "content": "• SwAV (Caron et al., 2020) is a contrastive learning algorithm that maps representations to a set of", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 692, + 381, + 704 + ], + "spans": [ + { + "bbox": [ + 116, + 692, + 381, + 704 + ], + "score": 1.0, + "content": "clusters and then enforces similarity between cluster assignments.", + "type": "text" + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "• Masked language modeling (MLM) (Devlin et al., 2019) randomly masks some of the tokens from", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 116, + 721, + 359, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 359, + 732 + ], + "score": 1.0, + "content": "input text and trains the model to predict the missing tokens.", + "type": "text" + } + ], + "index": 47, + "is_list_end_line": true + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 681, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 200, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 201, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 201, + 96 + ], + "score": 1.0, + "content": "6 EXPERIMENTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 504, + 129 + ], + "lines": [ + { + "bbox": [ + 107, + 107, + 504, + 118 + ], + "spans": [ + { + "bbox": [ + 107, + 107, + 504, + 118 + ], + "score": 1.0, + "content": "To evaluate how well existing methods can leverage unlabeled data to be robust to in-the-wild dis-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 117, + 470, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 470, + 130 + ], + "score": 1.0, + "content": "tribution shifts, we benchmarked the methods above on all applicable WILDS 2.0 datasets.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "title", + "bbox": [ + 107, + 143, + 160, + 154 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 161, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 161, + 156 + ], + "score": 1.0, + "content": "6.1 SETUP", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 164, + 486, + 175 + ], + "lines": [ + { + "bbox": [ + 106, + 163, + 488, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 488, + 177 + ], + "score": 1.0, + "content": "We used the default models, labeled training and test sets, and evaluation metrics from WILDS.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 181, + 504, + 247 + ], + "lines": [ + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "score": 1.0, + "content": "Unlabeled data. WILDS 2.0 contains multiple types of unlabeled data (from source, extra, valida-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 192, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 506, + 205 + ], + "score": 1.0, + "content": "tion, and/or target domains). For simplicity, we ran experiments on a single type of unlabeled data", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 202, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 506, + 216 + ], + "score": 1.0, + "content": "for each dataset. Where possible, we used unlabeled target data to allow methods to directly adapt to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 215, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 505, + 226 + ], + "score": 1.0, + "content": "the target distribution; for IWILDCAM2020-WILDS and CIVILCOMMENTS-WILDS, which do not", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 226, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 506, + 237 + ], + "score": 1.0, + "content": "have unlabeled target data, we used the extra domains instead. All methods use exactly the same", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 236, + 479, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 479, + 248 + ], + "score": 1.0, + "content": "sets of labeled and unlabeled training data (except ERM, which does not use unlabeled data).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 253, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 252, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 252, + 505, + 265 + ], + "score": 1.0, + "content": "Hyperparameters. We tuned each method on each dataset separately using random hyperparameter", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "search. Following WILDS 1.0, we used the labeled out-of-distribution (OOD) validation set to", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "select hyperparameters and for early stopping (Koh et al., 2021). This validation set is drawn from", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 285, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 104, + 285, + 505, + 298 + ], + "score": 1.0, + "content": "a different distribution than both the training and the OOD test set, so tuning on it does not leak", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "score": 1.0, + "content": "information on the test distribution. We did not use the in-distribution (ID) validation set. For", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 308, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 505, + 319 + ], + "score": 1.0, + "content": "image classification and regression, we used both RandAugment (Cubuk et al., 2020) and Cutout", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "(DeVries & Taylor, 2017) as data augmentation for all methods. We did not use data augmentation", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "for the remaining datasets. For some datasets, we also had ground truth labels for the “unlabeled”", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 429, + 353 + ], + "score": 1.0, + "content": "data, which we used to run fully-labeled ERM experiments. Overall, we ran", + "type": "text" + }, + { + "bbox": [ + 429, + 341, + 452, + 351 + ], + "score": 0.86, + "content": "^ { 6 0 0 + }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "experiments", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "for 7,000 GPU hours on NVIDIA V100s. See Appendix B for a discussion of which methods were", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "applicable to which datasets; Appendix C for augmentation details; Appendix F for the fully-labeled", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 374, + 341, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 341, + 385 + ], + "score": 1.0, + "content": "experiments; Appendix D for further experimental details.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 107, + 399, + 170, + 410 + ], + "lines": [ + { + "bbox": [ + 106, + 398, + 171, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 171, + 412 + ], + "score": 1.0, + "content": "6.2 RESULTS", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 420, + 504, + 486 + ], + "lines": [ + { + "bbox": [ + 106, + 420, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 505, + 432 + ], + "score": 1.0, + "content": "Table 2 shows mixed results on WILDS: most methods do not improve over standard empirical risk", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 432, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 505, + 443 + ], + "score": 1.0, + "content": "minimization (ERM) despite access to unlabeled data and careful hyperparameter tuning. In con-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 442, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 506, + 455 + ], + "score": 1.0, + "content": "trast, these methods have been shown to perform well on prior unsupervised adaptation benchmarks;", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "score": 1.0, + "content": "in Appendix E, we verify our implementations by showing that these methods (with the exception", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 278, + 477 + ], + "score": 1.0, + "content": "of CORAL) outperform ERM on the real", + "type": "text" + }, + { + "bbox": [ + 279, + 465, + 292, + 474 + ], + "score": 0.76, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "sketch shift in DomainNet, a standard unsupervised", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 474, + 371, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 371, + 487 + ], + "score": 1.0, + "content": "adaptation benchmark for object classification (Peng et al., 2019).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 536 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 503 + ], + "score": 1.0, + "content": "Image classification (IWILDCAM2020-WILDS, CAMELYON17-WILDS, and FMOW-WILDS).", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 503, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 503, + 505, + 515 + ], + "score": 1.0, + "content": "Data augmentation improved OOD performance on all three image classification datasets. The gain", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 513, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 436, + 527 + ], + "score": 1.0, + "content": "was the most substantial on CAMELYON17-WILDS, where vanilla ERM achieved", + "type": "text" + }, + { + "bbox": [ + 437, + 514, + 464, + 524 + ], + "score": 0.84, + "content": "7 0 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 513, + 506, + 527 + ], + "score": 1.0, + "content": "accuracy,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 524, + 359, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 289, + 538 + ], + "score": 1.0, + "content": "while ERM with data augmentation achieved", + "type": "text" + }, + { + "bbox": [ + 289, + 525, + 316, + 536 + ], + "score": 0.85, + "content": "8 2 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 524, + 359, + 538 + ], + "score": 1.0, + "content": "accuracy.2", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 542, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "On CAMELYON17-WILDS and FMOW-WILDS, where we had access to unlabeled target data, Noisy", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "Student and SwAV pre-training consistently improved OOD performance and reduced variability", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "across replicates. However, the other methods—CORAL, DANN, Pseudo-Label, and FixMatch—", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "score": 1.0, + "content": "underperformed ERM. This was especially surprising for FixMatch, which performed very well on", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 585, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 598 + ], + "score": 1.0, + "content": "DomainNet (Appendix E). Both FixMatch and Noisy Student use pseudo-labeling and consistency", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 596, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 506, + 609 + ], + "score": 1.0, + "content": "regularization, but FixMatch dynamically computes pseudo-labels in each batch from the start of", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 607, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 506, + 620 + ], + "score": 1.0, + "content": "training, whereas Noisy Student first trains a teacher model to convergence on the labeled data and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 618, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 505, + 632 + ], + "score": 1.0, + "content": "updates pseudolabels at a much slower rate. As in Xie et al. (2020), this suggests that dynamically", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 630, + 307, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 307, + 642 + ], + "score": 1.0, + "content": "updating pseudo-labels might hurt generalization.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 646, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 331, + 658 + ], + "score": 1.0, + "content": "On IWILDCAM2020-WILDS, where we had access to", + "type": "text" + }, + { + "bbox": [ + 332, + 647, + 347, + 657 + ], + "score": 0.86, + "content": "4 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 646, + 505, + 658 + ], + "score": 1.0, + "content": "as many unlabeled images from extra", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 657, + 504, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 504, + 670 + ], + "score": 1.0, + "content": "domains (distinct camera traps) but not to any images from the target domains, none of the bench-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 668, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 681 + ], + "score": 1.0, + "content": "marked methods improved OOD performance compared to ERM. This was surprising, as many of", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "these methods were originally shown to work in semi-supervised settings. One difference could be", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "score": 1.0, + "content": "that the labeled and unlabeled examples in IWILDCAM2020-WILDS differ more significantly (as", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 712, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 117, + 709, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 117, + 709, + 506, + 724 + ], + "score": 1.0, + "content": "2The data augmentation involves color jitter, which simulates the difference in staining protocols between", + "type": "text" + } + ] + }, + { + "bbox": [ + 106, + 721, + 461, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 461, + 732 + ], + "score": 1.0, + "content": "the source and target distributions in CAMELYON17-WILDS (Koh et al., 2021; Robey et al., 2021).", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 200, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 201, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 201, + 96 + ], + "score": 1.0, + "content": "6 EXPERIMENTS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 504, + 129 + ], + "lines": [ + { + "bbox": [ + 107, + 107, + 504, + 118 + ], + "spans": [ + { + "bbox": [ + 107, + 107, + 504, + 118 + ], + "score": 1.0, + "content": "To evaluate how well existing methods can leverage unlabeled data to be robust to in-the-wild dis-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 117, + 470, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 470, + 130 + ], + "score": 1.0, + "content": "tribution shifts, we benchmarked the methods above on all applicable WILDS 2.0 datasets.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 107, + 504, + 130 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 143, + 160, + 154 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 161, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 161, + 156 + ], + "score": 1.0, + "content": "6.1 SETUP", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 164, + 486, + 175 + ], + "lines": [ + { + "bbox": [ + 106, + 163, + 488, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 488, + 177 + ], + "score": 1.0, + "content": "We used the default models, labeled training and test sets, and evaluation metrics from WILDS.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 163, + 488, + 177 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 181, + 504, + 247 + ], + "lines": [ + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "score": 1.0, + "content": "Unlabeled data. WILDS 2.0 contains multiple types of unlabeled data (from source, extra, valida-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 192, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 506, + 205 + ], + "score": 1.0, + "content": "tion, and/or target domains). For simplicity, we ran experiments on a single type of unlabeled data", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 202, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 506, + 216 + ], + "score": 1.0, + "content": "for each dataset. Where possible, we used unlabeled target data to allow methods to directly adapt to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 215, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 505, + 226 + ], + "score": 1.0, + "content": "the target distribution; for IWILDCAM2020-WILDS and CIVILCOMMENTS-WILDS, which do not", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 226, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 506, + 237 + ], + "score": 1.0, + "content": "have unlabeled target data, we used the extra domains instead. All methods use exactly the same", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 236, + 479, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 479, + 248 + ], + "score": 1.0, + "content": "sets of labeled and unlabeled training data (except ERM, which does not use unlabeled data).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 180, + 506, + 248 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 253, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 252, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 252, + 505, + 265 + ], + "score": 1.0, + "content": "Hyperparameters. We tuned each method on each dataset separately using random hyperparameter", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "search. Following WILDS 1.0, we used the labeled out-of-distribution (OOD) validation set to", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "select hyperparameters and for early stopping (Koh et al., 2021). This validation set is drawn from", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 285, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 104, + 285, + 505, + 298 + ], + "score": 1.0, + "content": "a different distribution than both the training and the OOD test set, so tuning on it does not leak", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "score": 1.0, + "content": "information on the test distribution. We did not use the in-distribution (ID) validation set. For", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 308, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 505, + 319 + ], + "score": 1.0, + "content": "image classification and regression, we used both RandAugment (Cubuk et al., 2020) and Cutout", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "(DeVries & Taylor, 2017) as data augmentation for all methods. We did not use data augmentation", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "for the remaining datasets. For some datasets, we also had ground truth labels for the “unlabeled”", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 429, + 353 + ], + "score": 1.0, + "content": "data, which we used to run fully-labeled ERM experiments. Overall, we ran", + "type": "text" + }, + { + "bbox": [ + 429, + 341, + 452, + 351 + ], + "score": 0.86, + "content": "^ { 6 0 0 + }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "experiments", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "for 7,000 GPU hours on NVIDIA V100s. See Appendix B for a discussion of which methods were", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "applicable to which datasets; Appendix C for augmentation details; Appendix F for the fully-labeled", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 374, + 341, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 341, + 385 + ], + "score": 1.0, + "content": "experiments; Appendix D for further experimental details.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 16.5, + "bbox_fs": [ + 104, + 252, + 505, + 385 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 399, + 170, + 410 + ], + "lines": [ + { + "bbox": [ + 106, + 398, + 171, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 171, + 412 + ], + "score": 1.0, + "content": "6.2 RESULTS", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 420, + 504, + 486 + ], + "lines": [ + { + "bbox": [ + 106, + 420, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 505, + 432 + ], + "score": 1.0, + "content": "Table 2 shows mixed results on WILDS: most methods do not improve over standard empirical risk", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 432, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 505, + 443 + ], + "score": 1.0, + "content": "minimization (ERM) despite access to unlabeled data and careful hyperparameter tuning. In con-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 442, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 506, + 455 + ], + "score": 1.0, + "content": "trast, these methods have been shown to perform well on prior unsupervised adaptation benchmarks;", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 466 + ], + "score": 1.0, + "content": "in Appendix E, we verify our implementations by showing that these methods (with the exception", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 278, + 477 + ], + "score": 1.0, + "content": "of CORAL) outperform ERM on the real", + "type": "text" + }, + { + "bbox": [ + 279, + 465, + 292, + 474 + ], + "score": 0.76, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "sketch shift in DomainNet, a standard unsupervised", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 474, + 371, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 371, + 487 + ], + "score": 1.0, + "content": "adaptation benchmark for object classification (Peng et al., 2019).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 420, + 506, + 487 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 536 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 503 + ], + "score": 1.0, + "content": "Image classification (IWILDCAM2020-WILDS, CAMELYON17-WILDS, and FMOW-WILDS).", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 503, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 503, + 505, + 515 + ], + "score": 1.0, + "content": "Data augmentation improved OOD performance on all three image classification datasets. The gain", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 513, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 436, + 527 + ], + "score": 1.0, + "content": "was the most substantial on CAMELYON17-WILDS, where vanilla ERM achieved", + "type": "text" + }, + { + "bbox": [ + 437, + 514, + 464, + 524 + ], + "score": 0.84, + "content": "7 0 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 513, + 506, + 527 + ], + "score": 1.0, + "content": "accuracy,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 524, + 359, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 289, + 538 + ], + "score": 1.0, + "content": "while ERM with data augmentation achieved", + "type": "text" + }, + { + "bbox": [ + 289, + 525, + 316, + 536 + ], + "score": 0.85, + "content": "8 2 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 524, + 359, + 538 + ], + "score": 1.0, + "content": "accuracy.2", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 492, + 506, + 538 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 542, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "On CAMELYON17-WILDS and FMOW-WILDS, where we had access to unlabeled target data, Noisy", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "Student and SwAV pre-training consistently improved OOD performance and reduced variability", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "across replicates. However, the other methods—CORAL, DANN, Pseudo-Label, and FixMatch—", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "score": 1.0, + "content": "underperformed ERM. This was especially surprising for FixMatch, which performed very well on", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 585, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 598 + ], + "score": 1.0, + "content": "DomainNet (Appendix E). Both FixMatch and Noisy Student use pseudo-labeling and consistency", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 596, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 506, + 609 + ], + "score": 1.0, + "content": "regularization, but FixMatch dynamically computes pseudo-labels in each batch from the start of", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 607, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 506, + 620 + ], + "score": 1.0, + "content": "training, whereas Noisy Student first trains a teacher model to convergence on the labeled data and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 618, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 505, + 632 + ], + "score": 1.0, + "content": "updates pseudolabels at a much slower rate. As in Xie et al. (2020), this suggests that dynamically", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 630, + 307, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 307, + 642 + ], + "score": 1.0, + "content": "updating pseudo-labels might hurt generalization.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 541, + 506, + 642 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 646, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 505, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 331, + 658 + ], + "score": 1.0, + "content": "On IWILDCAM2020-WILDS, where we had access to", + "type": "text" + }, + { + "bbox": [ + 332, + 647, + 347, + 657 + ], + "score": 0.86, + "content": "4 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 646, + 505, + 658 + ], + "score": 1.0, + "content": "as many unlabeled images from extra", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 657, + 504, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 504, + 670 + ], + "score": 1.0, + "content": "domains (distinct camera traps) but not to any images from the target domains, none of the bench-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 668, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 681 + ], + "score": 1.0, + "content": "marked methods improved OOD performance compared to ERM. This was surprising, as many of", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "these methods were originally shown to work in semi-supervised settings. One difference could be", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "score": 1.0, + "content": "that the labeled and unlabeled examples in IWILDCAM2020-WILDS differ more significantly (as", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 598, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 612 + ], + "score": 1.0, + "content": "they originate from different camera traps) than in the original FixMatch paper (Sohn et al., 2020),", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 609, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 622 + ], + "score": 1.0, + "content": "which used i.i.d. labeled and unlabeled data, or the Noisy Student paper (Xie et al., 2020), which", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "score": 1.0, + "content": "used ImageNet labeled data (Russakovsky et al., 2015) and JFT unlabeled data (Hinton et al., 2015).", + "type": "text", + "cross_page": true + } + ], + "index": 12 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 646, + 506, + 703 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 47, + 505, + 494 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 47, + 505, + 494 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 47, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 47, + 505, + 494 + ], + "score": 0.983, + "html": "
ERM (-data aug) ERMIWILDCAM2020-WILDSFMoW-WILDS
(Unlabeled extra, macro F1)(Unlabeled target, worst-region acc)
In-distributionOut-of-distributionIn-distributionOut-of-distribution
46.7 (0.6)30.6 (1.1)59.3 (0.7)33.7 (1.5)
47.0 (1.4)32.2 (1.2)60.6 (0.6)34.8 (1.5)
CORAL27.9 (0.4)58.9 (0.3)34.1 (0.6)
DANN40.5 (1.4) 48.5 (2.8) 31.9 (1.4)57.9 (0.8)34.6 (1.7)
Pseudo-Label47.3 (0.4)60.9 (0.5)33.7 (0.2)
FixMatch46.3 (0.5)30.3 (0.4) 31.0 (1.3) 58.6 (2.4)32.1 (2.0)
Noisy Student47.5 (0.9)32.1 (0.7)61.3 (0.4) 37.8 (0.6)
SwAV47.3 (1.4)29.0 (2.0) 61.8 (1.0)36.3 (1.0)
ERM (fully-labeled)54.6 (1.5)44.0 (2.3) 65.4 (0.4)58.7 (1.4)
ERM Noisy StudentCAMELYON17-WILDSPOVERTYMAP-WILDS
(Unlabeled target, avg acc)(Unlabeled target, worst U/R corr)
In-distributionOut-of-distributionIn-distributionOut-of-distribution
ERM (-data aug) 85.8 (1.9)70.8 (7.2)0.65 (0.03)0.50 (0.07)
90.6 (1.2)82.0 (7.4)0.66 (0.04)0.49 (0.06)
CORAL 90.4 (0.9)77.9 (6.6)0.54 (0.10)0.36 (0.08)
DANN 86.9 (2.2)68.4 (9.2)0.50 (0.07)0.33 (0.10)
Pseudo-Label67.7 (8.2)
FixMatch91.3 (1.3) 91.3 (1.1)71.0 (4.9) 0.54 (0.11)0.30 (0.11)
Noisy Student93.2 (0.5)86.7 (1.7) 0.61 (0.07)0.42 (0.11)
SwAV92.3 (0.4)91.4 (2.0) 0.60 (0.13)0.45 (0.05)
ERM CORAL DANN Pseudo-LabelGLOBALWHEAT-WILDSOGB-MOLPCBA
(Unlabeled target, avg domain acc)(Unlabeled target, avg AP)
In-distributionOut-of-distributionIn-distributionOut-of-distribution
77.8 (0.1)50.5 (1.7)28.3 (0.1)
126.6 (0.2)
一20.4 (0.8)
75.2 (1.2)42.7 (4.8)19.7 (0.1)
78.8 (0.5)49.3 (3.7)27.5 (0.1)
ERM CORAL DANN Pseudo-LabelCIVILCOMMENTS-WILDSAMAZON-WILDS
(Unlabeled extra, worst-group acc)(Unlabeled target,1Oth percentile acc)
In-distributionOut-of-distributionIn-distributionOut-of-distribution
89.8 (0.8)66.6 (1.6)72.0 (0.1)54.2 (0.8)
71.7 (0.1)53.3 (0.0)
一71.7 (0.1)53.3 (0.0)
71.6 (0.1)52.3 (1.1)
90.3 (0.5) 89.4 (1.2)66.9 (2.6) 65.7 (2.3)71.9 (0.4)53.9 (0.7)
Masked LM ERM (fully-labeled)89.9 (0.1)69.4 (0.6)73.6 (0.1)56.4 (0.8)
", + "type": "table", + "image_path": "430ae4267ba6354907f1e856bd415bc19ef072327dfd0b2577ec9220ccf52a7b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 47, + 505, + 196.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 196.0, + 505, + 345.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 345.0, + 505, + 494.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 498, + 505, + 577 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 498, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 511 + ], + "score": 1.0, + "content": "Table 2: The in-distribution (ID) and out-of-distribution (OOD) performance of each method on", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 510, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 505, + 522 + ], + "score": 1.0, + "content": "each applicable dataset. Following WILDS 1.0, we ran 3–10 replicates (random seeds) for each", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "cell, depending on the dataset. We report the standard deviation across replicates in parentheses; the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "standard error (of the mean) is lower by the square root of the number of replicates. Fully-labeled", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 555 + ], + "score": 1.0, + "content": "experiments use ground truth labels on the “unlabeled” data. We bold the highest non-fully-labeled", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "OOD performance numbers as well as others where the standard error is within range. Below each", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 565, + 377, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 377, + 578 + ], + "score": 1.0, + "content": "dataset name, we report the type of unlabeled data and metric used.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 105, + 598, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 612 + ], + "score": 1.0, + "content": "they originate from different camera traps) than in the original FixMatch paper (Sohn et al., 2020),", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 609, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 622 + ], + "score": 1.0, + "content": "which used i.i.d. labeled and unlabeled data, or the Noisy Student paper (Xie et al., 2020), which", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "score": 1.0, + "content": "used ImageNet labeled data (Russakovsky et al., 2015) and JFT unlabeled data (Hinton et al., 2015).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 108, + 638, + 503, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "Fully-labeled ERM models that used ground truth labels for the “unlabeled” data were available", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "for FMOW-WILDS and IWILDCAM2020-WILDS. They significantly outperformed other methods,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 660, + 402, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 402, + 673 + ], + "score": 1.0, + "content": "suggesting room for improvement in how we leverage the unlabeled data.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Image regression (POVERTYMAP-WILDS). Data augmentation had no effect on performance on", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "POVERTYMAP-WILDS, which differs from the above image datasets in that it is a regression task", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "and involves multi-spectral satellite images (with 7 channels); both of these aspects are relatively", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "unstudied compared to standard RGB image classification. All applicable methods underperformed", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 721, + 473, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 473, + 733 + ], + "score": 1.0, + "content": "standard ERM, despite having access to unlabeled data from the target domains (countries).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 47, + 505, + 494 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 47, + 505, + 494 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 47, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 47, + 505, + 494 + ], + "score": 0.983, + "html": "
ERM (-data aug) ERMIWILDCAM2020-WILDSFMoW-WILDS
(Unlabeled extra, macro F1)(Unlabeled target, worst-region acc)
In-distributionOut-of-distributionIn-distributionOut-of-distribution
46.7 (0.6)30.6 (1.1)59.3 (0.7)33.7 (1.5)
47.0 (1.4)32.2 (1.2)60.6 (0.6)34.8 (1.5)
CORAL27.9 (0.4)58.9 (0.3)34.1 (0.6)
DANN40.5 (1.4) 48.5 (2.8) 31.9 (1.4)57.9 (0.8)34.6 (1.7)
Pseudo-Label47.3 (0.4)60.9 (0.5)33.7 (0.2)
FixMatch46.3 (0.5)30.3 (0.4) 31.0 (1.3) 58.6 (2.4)32.1 (2.0)
Noisy Student47.5 (0.9)32.1 (0.7)61.3 (0.4) 37.8 (0.6)
SwAV47.3 (1.4)29.0 (2.0) 61.8 (1.0)36.3 (1.0)
ERM (fully-labeled)54.6 (1.5)44.0 (2.3) 65.4 (0.4)58.7 (1.4)
ERM Noisy StudentCAMELYON17-WILDSPOVERTYMAP-WILDS
(Unlabeled target, avg acc)(Unlabeled target, worst U/R corr)
In-distributionOut-of-distributionIn-distributionOut-of-distribution
ERM (-data aug) 85.8 (1.9)70.8 (7.2)0.65 (0.03)0.50 (0.07)
90.6 (1.2)82.0 (7.4)0.66 (0.04)0.49 (0.06)
CORAL 90.4 (0.9)77.9 (6.6)0.54 (0.10)0.36 (0.08)
DANN 86.9 (2.2)68.4 (9.2)0.50 (0.07)0.33 (0.10)
Pseudo-Label67.7 (8.2)
FixMatch91.3 (1.3) 91.3 (1.1)71.0 (4.9) 0.54 (0.11)0.30 (0.11)
Noisy Student93.2 (0.5)86.7 (1.7) 0.61 (0.07)0.42 (0.11)
SwAV92.3 (0.4)91.4 (2.0) 0.60 (0.13)0.45 (0.05)
ERM CORAL DANN Pseudo-LabelGLOBALWHEAT-WILDSOGB-MOLPCBA
(Unlabeled target, avg domain acc)(Unlabeled target, avg AP)
In-distributionOut-of-distributionIn-distributionOut-of-distribution
77.8 (0.1)50.5 (1.7)28.3 (0.1)
126.6 (0.2)
一20.4 (0.8)
75.2 (1.2)42.7 (4.8)19.7 (0.1)
78.8 (0.5)49.3 (3.7)27.5 (0.1)
ERM CORAL DANN Pseudo-LabelCIVILCOMMENTS-WILDSAMAZON-WILDS
(Unlabeled extra, worst-group acc)(Unlabeled target,1Oth percentile acc)
In-distributionOut-of-distributionIn-distributionOut-of-distribution
89.8 (0.8)66.6 (1.6)72.0 (0.1)54.2 (0.8)
71.7 (0.1)53.3 (0.0)
一71.7 (0.1)53.3 (0.0)
71.6 (0.1)52.3 (1.1)
90.3 (0.5) 89.4 (1.2)66.9 (2.6) 65.7 (2.3)71.9 (0.4)53.9 (0.7)
Masked LM ERM (fully-labeled)89.9 (0.1)69.4 (0.6)73.6 (0.1)56.4 (0.8)
", + "type": "table", + "image_path": "430ae4267ba6354907f1e856bd415bc19ef072327dfd0b2577ec9220ccf52a7b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 47, + 505, + 196.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 196.0, + 505, + 345.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 345.0, + 505, + 494.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 498, + 505, + 577 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 498, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 511 + ], + "score": 1.0, + "content": "Table 2: The in-distribution (ID) and out-of-distribution (OOD) performance of each method on", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 510, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 505, + 522 + ], + "score": 1.0, + "content": "each applicable dataset. Following WILDS 1.0, we ran 3–10 replicates (random seeds) for each", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "cell, depending on the dataset. We report the standard deviation across replicates in parentheses; the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 545 + ], + "score": 1.0, + "content": "standard error (of the mean) is lower by the square root of the number of replicates. Fully-labeled", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 555 + ], + "score": 1.0, + "content": "experiments use ground truth labels on the “unlabeled” data. We bold the highest non-fully-labeled", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "OOD performance numbers as well as others where the standard error is within range. Below each", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 565, + 377, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 377, + 578 + ], + "score": 1.0, + "content": "dataset name, we report the type of unlabeled data and metric used.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 505, + 632 + ], + "lines": [], + "index": 11, + "bbox_fs": [ + 105, + 598, + 506, + 633 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 108, + 638, + 503, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "Fully-labeled ERM models that used ground truth labels for the “unlabeled” data were available", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "for FMOW-WILDS and IWILDCAM2020-WILDS. They significantly outperformed other methods,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 660, + 402, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 402, + 673 + ], + "score": 1.0, + "content": "suggesting room for improvement in how we leverage the unlabeled data.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 637, + 505, + 673 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "Image regression (POVERTYMAP-WILDS). Data augmentation had no effect on performance on", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "POVERTYMAP-WILDS, which differs from the above image datasets in that it is a regression task", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "and involves multi-spectral satellite images (with 7 channels); both of these aspects are relatively", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "unstudied compared to standard RGB image classification. All applicable methods underperformed", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 721, + 473, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 473, + 733 + ], + "score": 1.0, + "content": "standard ERM, despite having access to unlabeled data from the target domains (countries).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 677, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "Image detection (GLOBALWHEAT-WILDS). We did not apply data augmentation here, as standard", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "augmentation changes the labels (e.g., cropping the image might remove bounding boxes) and would", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "violate the assumption that labels are invariant under augmentations, which contrastive and consis-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "tency regularization methods like SwAV, Noisy Student, and FixMatch rely on. Accordingly, we", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 139 + ], + "score": 1.0, + "content": "did not evaluate FixMatch and SwAV, and we modified Noisy Student to remove data augmentation", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 321, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 321, + 150 + ], + "score": 1.0, + "content": "noise. All applicable methods underperformed ERM.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 154, + 505, + 187 + ], + "lines": [ + { + "bbox": [ + 105, + 153, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 505, + 168 + ], + "score": 1.0, + "content": "Molecule classification (OGB-MOLPCBA). We did not apply data augmentation techniques to", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 504, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 504, + 178 + ], + "score": 1.0, + "content": "OGB-MOLPCBA as they are not well-developed for molecular graphs. All methods underper-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 175, + 446, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 446, + 190 + ], + "score": 1.0, + "content": "formed ERM. We did not report ID results as this dataset has no separate ID test set.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 193, + 505, + 347 + ], + "lines": [ + { + "bbox": [ + 105, + 191, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 206 + ], + "score": 1.0, + "content": "Text classification (CIVILCOMMENTS-WILDS, AMAZON-WILDS). Likewise, we did not apply", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 203, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 506, + 217 + ], + "score": 1.0, + "content": "data augmentation to the text datasets. On both datasets, other methods performed similarly to", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 213, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 228 + ], + "score": 1.0, + "content": "ERM (with class-balancing for CIVILCOMMENTS-WILDS). Continued masked LM pre-training on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 225, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 104, + 225, + 505, + 239 + ], + "score": 1.0, + "content": "the unlabeled data did not improve target performance, unlike in prior work (Gururangan et al.,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "2020); this might be because the BERT pre-training corpus (Devlin et al., 2019; Hendrycks et al.,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 247, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 505, + 260 + ], + "score": 1.0, + "content": "2020) is more similar to the online comments in CIVILCOMMENTS-WILDS and product reviews", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 257, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 104, + 257, + 506, + 272 + ], + "score": 1.0, + "content": "in AMAZON-WILDS than to the biomedical/CS text studied in Gururangan et al. (2020). Also,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 268, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 282 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS and AMAZON-WILDS measure subpopulation performance (on minority", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "score": 1.0, + "content": "demographics and on the tail subpopulation, respectively), whereas prior work adapted models to", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "new areas of the input space (e.g., from news to biomedical articles). Fully-labeled ERM models", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 303, + 504, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 504, + 315 + ], + "score": 1.0, + "content": "showed modest gains compared to FMOW-WILDS and IWILDCAM2020-WILDS. As our evalua-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 313, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 505, + 325 + ], + "score": 1.0, + "content": "tions on these text datasets focus on subpopulations performance, these results are consistent with", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "prior observations that ERM models can have poor subpopulation performance even with large la-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 335, + 491, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 491, + 348 + ], + "score": 1.0, + "content": "beled training sets (Sagawa et al., 2020), necessitating other approaches to subpopulation shifts.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 108, + 364, + 190, + 377 + ], + "lines": [ + { + "bbox": [ + 104, + 362, + 192, + 380 + ], + "spans": [ + { + "bbox": [ + 104, + 362, + 192, + 380 + ], + "score": 1.0, + "content": "7 DISCUSSION", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 389, + 456, + 401 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 458, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 458, + 403 + ], + "score": 1.0, + "content": "We conclude by discussing several takeaways and promising directions for future work.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 407, + 504, + 473 + ], + "lines": [ + { + "bbox": [ + 105, + 405, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 506, + 421 + ], + "score": 1.0, + "content": "The role of data augmentation. Many unsupervised adaptation methods rely strongly on data aug-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "score": 1.0, + "content": "mentation for consistency regularization or contrastive learning. This reliance on data augmenta-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "tion techniques—which are largely image-specific—restricts their generality, as they do not readily", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "generalize to other modalities (or even other types of images besides photos). Developing data aug-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 450, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 506, + 462 + ], + "score": 1.0, + "content": "mentation techniques that can work well in other applications and modalities could be crucial for", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 462, + 372, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 372, + 474 + ], + "score": 1.0, + "content": "expanding the applicability of these methods (Verma et al., 2021).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 478, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "Hyperparameter tuning. Unsupervised adaptation methods have even more hyperparameters than", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "standard supervised methods, and consistent with prior work, we found that these hyperparameters", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "can significantly affect OOD performance (Saito et al., 2021). Moreover, unlike in standard i.i.d. set-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "score": 1.0, + "content": "tings, we do not have labeled target data that we can use for hyperparameter selection. Improved", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 523, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 535 + ], + "score": 1.0, + "content": "methods for hyperparameter tuning could significantly improve OOD performance. Such methods", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 534, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 545 + ], + "score": 1.0, + "content": "might make use of the unlabeled target data, or even the combination of labeled and unlabeled OOD", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 545, + 375, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 375, + 556 + ], + "score": 1.0, + "content": "validation data, which is provided for most datasets in WILDS 2.0.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 561, + 504, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 506, + 573 + ], + "score": 1.0, + "content": "Pre-training on broader unlabeled data. Pre-training on huge amounts of unlabeled data improves", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 572, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 583 + ], + "score": 1.0, + "content": "robustness to distribution shifts in some settings (Bommasani et al., 2021). The unlabeled data need", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "not be related to the task: e.g., CLIP was pre-trained on text-image pairs from the internet but tested", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 594, + 506, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 506, + 607 + ], + "score": 1.0, + "content": "on tasks including histopathology and satellite image classification (Radford et al., 2021a). Existing", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "techniques for this type of broad pre-training appear insufficient for WILDS: many of our models", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "were initialized with ImageNet-pretrained weights or derivatives of BERT, but do not generalize", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "well OOD. While we focused on providing curated unlabeled data that is closely tailored to the task,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 638, + 364, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 364, + 649 + ], + "score": 1.0, + "content": "it could be fruitful to use both broad and curated unlabeled data.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "Leveraging domain annotations and task-specific structure. OOD robustness is ill-posed in gen-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "eral, as models cannot be robust to arbitrary distribution shifts. Beyond unlabeled data, WILDS", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 675, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 506, + 690 + ], + "score": 1.0, + "content": "also has domain annotations and other structured metadata for both labeled and unlabeled data (e.g.,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "in IWILDCAM2020-WILDS, we know which images were taken from which cameras). Exploit-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "ing this type of fine-grained domain structure for unsupervised adaptation—e.g., through multi-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "source/multi-target domain adaptation methods (Zhao et al., 2018; Peng et al., 2019)—could be a", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 721, + 469, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 469, + 732 + ], + "score": 1.0, + "content": "promising avenue for learning models that are more robust to the domain shifts in WILDS.", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 49 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "Image detection (GLOBALWHEAT-WILDS). We did not apply data augmentation here, as standard", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "augmentation changes the labels (e.g., cropping the image might remove bounding boxes) and would", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "violate the assumption that labels are invariant under augmentations, which contrastive and consis-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "tency regularization methods like SwAV, Noisy Student, and FixMatch rely on. Accordingly, we", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 139 + ], + "score": 1.0, + "content": "did not evaluate FixMatch and SwAV, and we modified Noisy Student to remove data augmentation", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 321, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 321, + 150 + ], + "score": 1.0, + "content": "noise. All applicable methods underperformed ERM.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 82, + 506, + 150 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 154, + 505, + 187 + ], + "lines": [ + { + "bbox": [ + 105, + 153, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 505, + 168 + ], + "score": 1.0, + "content": "Molecule classification (OGB-MOLPCBA). We did not apply data augmentation techniques to", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 504, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 504, + 178 + ], + "score": 1.0, + "content": "OGB-MOLPCBA as they are not well-developed for molecular graphs. All methods underper-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 175, + 446, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 446, + 190 + ], + "score": 1.0, + "content": "formed ERM. We did not report ID results as this dataset has no separate ID test set.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 153, + 505, + 190 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 193, + 505, + 347 + ], + "lines": [ + { + "bbox": [ + 105, + 191, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 206 + ], + "score": 1.0, + "content": "Text classification (CIVILCOMMENTS-WILDS, AMAZON-WILDS). Likewise, we did not apply", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 203, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 506, + 217 + ], + "score": 1.0, + "content": "data augmentation to the text datasets. On both datasets, other methods performed similarly to", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 213, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 228 + ], + "score": 1.0, + "content": "ERM (with class-balancing for CIVILCOMMENTS-WILDS). Continued masked LM pre-training on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 225, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 104, + 225, + 505, + 239 + ], + "score": 1.0, + "content": "the unlabeled data did not improve target performance, unlike in prior work (Gururangan et al.,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "2020); this might be because the BERT pre-training corpus (Devlin et al., 2019; Hendrycks et al.,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 247, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 505, + 260 + ], + "score": 1.0, + "content": "2020) is more similar to the online comments in CIVILCOMMENTS-WILDS and product reviews", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 257, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 104, + 257, + 506, + 272 + ], + "score": 1.0, + "content": "in AMAZON-WILDS than to the biomedical/CS text studied in Gururangan et al. (2020). Also,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 268, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 282 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS and AMAZON-WILDS measure subpopulation performance (on minority", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "score": 1.0, + "content": "demographics and on the tail subpopulation, respectively), whereas prior work adapted models to", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "new areas of the input space (e.g., from news to biomedical articles). Fully-labeled ERM models", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 303, + 504, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 504, + 315 + ], + "score": 1.0, + "content": "showed modest gains compared to FMOW-WILDS and IWILDCAM2020-WILDS. As our evalua-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 313, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 505, + 325 + ], + "score": 1.0, + "content": "tions on these text datasets focus on subpopulations performance, these results are consistent with", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "prior observations that ERM models can have poor subpopulation performance even with large la-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 335, + 491, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 491, + 348 + ], + "score": 1.0, + "content": "beled training sets (Sagawa et al., 2020), necessitating other approaches to subpopulation shifts.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 15.5, + "bbox_fs": [ + 104, + 191, + 506, + 348 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 364, + 190, + 377 + ], + "lines": [ + { + "bbox": [ + 104, + 362, + 192, + 380 + ], + "spans": [ + { + "bbox": [ + 104, + 362, + 192, + 380 + ], + "score": 1.0, + "content": "7 DISCUSSION", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 389, + 456, + 401 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 458, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 458, + 403 + ], + "score": 1.0, + "content": "We conclude by discussing several takeaways and promising directions for future work.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 106, + 389, + 458, + 403 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 407, + 504, + 473 + ], + "lines": [ + { + "bbox": [ + 105, + 405, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 506, + 421 + ], + "score": 1.0, + "content": "The role of data augmentation. Many unsupervised adaptation methods rely strongly on data aug-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "score": 1.0, + "content": "mentation for consistency regularization or contrastive learning. This reliance on data augmenta-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "tion techniques—which are largely image-specific—restricts their generality, as they do not readily", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "generalize to other modalities (or even other types of images besides photos). Developing data aug-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 450, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 506, + 462 + ], + "score": 1.0, + "content": "mentation techniques that can work well in other applications and modalities could be crucial for", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 462, + 372, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 372, + 474 + ], + "score": 1.0, + "content": "expanding the applicability of these methods (Verma et al., 2021).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 405, + 506, + 474 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 478, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "Hyperparameter tuning. Unsupervised adaptation methods have even more hyperparameters than", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "standard supervised methods, and consistent with prior work, we found that these hyperparameters", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "can significantly affect OOD performance (Saito et al., 2021). Moreover, unlike in standard i.i.d. set-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "score": 1.0, + "content": "tings, we do not have labeled target data that we can use for hyperparameter selection. Improved", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 523, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 535 + ], + "score": 1.0, + "content": "methods for hyperparameter tuning could significantly improve OOD performance. Such methods", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 534, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 545 + ], + "score": 1.0, + "content": "might make use of the unlabeled target data, or even the combination of labeled and unlabeled OOD", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 545, + 375, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 375, + 556 + ], + "score": 1.0, + "content": "validation data, which is provided for most datasets in WILDS 2.0.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 478, + 505, + 556 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 561, + 504, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 506, + 573 + ], + "score": 1.0, + "content": "Pre-training on broader unlabeled data. Pre-training on huge amounts of unlabeled data improves", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 572, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 583 + ], + "score": 1.0, + "content": "robustness to distribution shifts in some settings (Bommasani et al., 2021). The unlabeled data need", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "not be related to the task: e.g., CLIP was pre-trained on text-image pairs from the internet but tested", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 594, + 506, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 506, + 607 + ], + "score": 1.0, + "content": "on tasks including histopathology and satellite image classification (Radford et al., 2021a). Existing", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "techniques for this type of broad pre-training appear insufficient for WILDS: many of our models", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "were initialized with ImageNet-pretrained weights or derivatives of BERT, but do not generalize", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 639 + ], + "score": 1.0, + "content": "well OOD. While we focused on providing curated unlabeled data that is closely tailored to the task,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 638, + 364, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 364, + 649 + ], + "score": 1.0, + "content": "it could be fruitful to use both broad and curated unlabeled data.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 561, + 506, + 649 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "Leveraging domain annotations and task-specific structure. OOD robustness is ill-posed in gen-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "eral, as models cannot be robust to arbitrary distribution shifts. Beyond unlabeled data, WILDS", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 675, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 506, + 690 + ], + "score": 1.0, + "content": "also has domain annotations and other structured metadata for both labeled and unlabeled data (e.g.,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "in IWILDCAM2020-WILDS, we know which images were taken from which cameras). Exploit-", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "ing this type of fine-grained domain structure for unsupervised adaptation—e.g., through multi-", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "source/multi-target domain adaptation methods (Zhao et al., 2018; Peng et al., 2019)—could be a", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 721, + 469, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 469, + 732 + ], + "score": 1.0, + "content": "promising avenue for learning models that are more robust to the domain shifts in WILDS.", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 49, + "bbox_fs": [ + 105, + 655, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 209, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 210, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 210, + 94 + ], + "score": 1.0, + "content": "ETHICS STATEMENT", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 106, + 504, + 128 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "All WILDS datasets are curated and adapted from public data sources, with licenses that allow for", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 118, + 300, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 300, + 129 + ], + "score": 1.0, + "content": "public release. The datasets are all anonymized.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 106, + 134, + 505, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 135, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 505, + 145 + ], + "score": 1.0, + "content": "The distribution shifts in several of the WILDS datasets deal with issues of discrimination and bias", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 145, + 506, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 506, + 158 + ], + "score": 1.0, + "content": "that arise in real-world applications. For example, CIVILCOMMENTS-WILDS studies disparate", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 157, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 157, + 505, + 168 + ], + "score": 1.0, + "content": "model performance across online comments that mention different demographic groups, while", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 167, + 505, + 179 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 505, + 179 + ], + "score": 1.0, + "content": "FMOW-WILDS and POVERTYMAP-WILDS study countries and regions where labeled satellite data", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 178, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 506, + 190 + ], + "score": 1.0, + "content": "is less readily available. As our results suggest, standard models trained on these datasets will not", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 189, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 505, + 201 + ], + "score": 1.0, + "content": "perform well on those subpopulations, and their learned representations might also be biased in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 200, + 504, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 504, + 212 + ], + "score": 1.0, + "content": "undesirable ways (Bolukbasi et al., 2016; Caliskan et al., 2017; Garg et al., 2018; Tan & Celis,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 210, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 505, + 223 + ], + "score": 1.0, + "content": "2019; Steed & Caliskan, 2021). We also encourage caution in interpreting positive results on these", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 221, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 505, + 234 + ], + "score": 1.0, + "content": "datasets, as our evaluation metrics might not encompass all relevant facets of discrimination and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 233, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 505, + 245 + ], + "score": 1.0, + "content": "bias: e.g., the “ground truth” toxicity annotations in CIVILCOMMENTS-WILDS can themselves be", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "score": 1.0, + "content": "biased, and the particular choice of regions in FMOW-WILDS might obscure lower model perfor-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 254, + 197, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 197, + 268 + ], + "score": 1.0, + "content": "mance in sub-regions.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 272, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 271, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 285 + ], + "score": 1.0, + "content": "For FMOW-WILDS and POVERTYMAP-WILDS, surveillance and privacy issues also need to be", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 283, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 505, + 295 + ], + "score": 1.0, + "content": "considered. In FMOW-WILDS, the image resolution is lower than that of other public satellite data", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 294, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 506, + 307 + ], + "score": 1.0, + "content": "(e.g., from Google Maps), and in POVERTYMAP-WILDS, the location metadata is noised to protect", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 304, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 317 + ], + "score": 1.0, + "content": "privacy. For a deeper discussion of the ethics of remote sensing in the context of humanitarian aid", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 316, + 435, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 435, + 329 + ], + "score": 1.0, + "content": "and development, we refer readers to the UNICEF report by Berman et al. (2018).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 108, + 344, + 266, + 356 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 268, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 268, + 358 + ], + "score": 1.0, + "content": "REPRODUCIBILITY STATEMENT", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 109, + 369, + 504, + 402 + ], + "lines": [ + { + "bbox": [ + 106, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "All WILDS datasets are publicly available at https://wilds.stanford.edu, together with", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 380, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 392 + ], + "score": 1.0, + "content": "code and scripts to replicate all of the experiments in this paper. We also provide all trained model", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 391, + 390, + 403 + ], + "spans": [ + { + "bbox": [ + 107, + 391, + 390, + 403 + ], + "score": 1.0, + "content": "checkpoints and results, together with the exact hyperparameters used.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 407, + 411, + 419 + ], + "lines": [ + { + "bbox": [ + 105, + 407, + 413, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 413, + 422 + ], + "score": 1.0, + "content": "In our appendices, we provide more details on the datasets and experiments:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 132, + 428, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 132, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 132, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "• In Appendix A, we describe each of the updated datasets in WILDS 2.0 and their sources", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 440, + 411, + 452 + ], + "spans": [ + { + "bbox": [ + 141, + 440, + 411, + 452 + ], + "score": 1.0, + "content": "of unlabeled data as well as what data processing steps were taken.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 134, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 134, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "• In Appendix B, we describe the implementations of each of our benchmarked methods in", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 141, + 466, + 506, + 479 + ], + "score": 1.0, + "content": "detail. In particular, we discuss any changes we made to their original implementations,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 142, + 478, + 504, + 489 + ], + "spans": [ + { + "bbox": [ + 142, + 478, + 504, + 489 + ], + "score": 1.0, + "content": "either for consistency with other methods or with prior implementations of these methods.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 131, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 131, + 492, + 505, + 506 + ], + "score": 1.0, + "content": "• In Appendix C, we describe details of the data augmentations (if any) that we used across", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 503, + 195, + 515 + ], + "spans": [ + { + "bbox": [ + 141, + 503, + 195, + 515 + ], + "score": 1.0, + "content": "each dataset.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 131, + 518, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 131, + 518, + 506, + 532 + ], + "score": 1.0, + "content": "• In Appendix D, we describe our experimental protocol, including the hyperparameter se-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 141, + 530, + 460, + 543 + ], + "spans": [ + { + "bbox": [ + 141, + 530, + 460, + 543 + ], + "score": 1.0, + "content": "lection procedure and hyperparameter grids for all of the methods and datasets.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 132, + 545, + 439, + 558 + ], + "spans": [ + { + "bbox": [ + 132, + 545, + 439, + 558 + ], + "score": 1.0, + "content": "• In Appendix E, we describe the details of our experiments on DomainNet.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 131, + 560, + 456, + 574 + ], + "spans": [ + { + "bbox": [ + 131, + 560, + 456, + 574 + ], + "score": 1.0, + "content": "• In Appendix F, we describe the details of our fully-labeled ERM experiments.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 132, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 132, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "• Finally, in Appendix G, we include an illustrative code snippet of how to use the data", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 141, + 586, + 262, + 600 + ], + "spans": [ + { + "bbox": [ + 141, + 586, + 262, + 600 + ], + "score": 1.0, + "content": "loaders in the WILDS library.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 31 + }, + { + "type": "title", + "bbox": [ + 108, + 616, + 240, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 614, + 241, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 241, + 630 + ], + "score": 1.0, + "content": "AUTHOR CONTRIBUTIONS", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 640, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "score": 1.0, + "content": "The project was initiated by Shiori Sagawa, Pang Wei Koh, and Percy Liang. Shiori Sagawa and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "score": 1.0, + "content": "Pang Wei Koh led the project and coordinated the activities below. Tony Lee developed the exper-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 662, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 505, + 674 + ], + "score": 1.0, + "content": "imental infrastructure and ran the experiments. Tony Lee, Irena Gao, Sang Michael Xie, Kendrick", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 672, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 505, + 686 + ], + "score": 1.0, + "content": "Shen, Ananya Kumar, and Michihiro Yasunaga designed the evaluation framework and implemented", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 683, + 485, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 683, + 485, + 698 + ], + "score": 1.0, + "content": "the algorithms. The unlabeled data loaders and corresponding dataset writeups were added by:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 132, + 705, + 280, + 732 + ], + "lines": [ + { + "bbox": [ + 132, + 705, + 259, + 717 + ], + "spans": [ + { + "bbox": [ + 132, + 705, + 259, + 717 + ], + "score": 1.0, + "content": "• AMAZON-WILDS: Tony Lee", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 132, + 720, + 280, + 733 + ], + "spans": [ + { + "bbox": [ + 132, + 720, + 280, + 733 + ], + "score": 1.0, + "content": "• CAMELYON17-WILDS: Tony Lee", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 209, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 210, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 210, + 94 + ], + "score": 1.0, + "content": "ETHICS STATEMENT", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 106, + 504, + 128 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "All WILDS datasets are curated and adapted from public data sources, with licenses that allow for", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 118, + 300, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 300, + 129 + ], + "score": 1.0, + "content": "public release. The datasets are all anonymized.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 106, + 505, + 129 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 134, + 505, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 135, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 505, + 145 + ], + "score": 1.0, + "content": "The distribution shifts in several of the WILDS datasets deal with issues of discrimination and bias", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 145, + 506, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 506, + 158 + ], + "score": 1.0, + "content": "that arise in real-world applications. For example, CIVILCOMMENTS-WILDS studies disparate", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 157, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 157, + 505, + 168 + ], + "score": 1.0, + "content": "model performance across online comments that mention different demographic groups, while", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 167, + 505, + 179 + ], + "spans": [ + { + "bbox": [ + 106, + 167, + 505, + 179 + ], + "score": 1.0, + "content": "FMOW-WILDS and POVERTYMAP-WILDS study countries and regions where labeled satellite data", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 178, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 506, + 190 + ], + "score": 1.0, + "content": "is less readily available. As our results suggest, standard models trained on these datasets will not", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 189, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 505, + 201 + ], + "score": 1.0, + "content": "perform well on those subpopulations, and their learned representations might also be biased in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 200, + 504, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 200, + 504, + 212 + ], + "score": 1.0, + "content": "undesirable ways (Bolukbasi et al., 2016; Caliskan et al., 2017; Garg et al., 2018; Tan & Celis,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 210, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 505, + 223 + ], + "score": 1.0, + "content": "2019; Steed & Caliskan, 2021). We also encourage caution in interpreting positive results on these", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 221, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 505, + 234 + ], + "score": 1.0, + "content": "datasets, as our evaluation metrics might not encompass all relevant facets of discrimination and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 233, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 505, + 245 + ], + "score": 1.0, + "content": "bias: e.g., the “ground truth” toxicity annotations in CIVILCOMMENTS-WILDS can themselves be", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "score": 1.0, + "content": "biased, and the particular choice of regions in FMOW-WILDS might obscure lower model perfor-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 254, + 197, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 197, + 268 + ], + "score": 1.0, + "content": "mance in sub-regions.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 135, + 506, + 268 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 272, + 505, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 271, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 285 + ], + "score": 1.0, + "content": "For FMOW-WILDS and POVERTYMAP-WILDS, surveillance and privacy issues also need to be", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 283, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 505, + 295 + ], + "score": 1.0, + "content": "considered. In FMOW-WILDS, the image resolution is lower than that of other public satellite data", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 294, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 506, + 307 + ], + "score": 1.0, + "content": "(e.g., from Google Maps), and in POVERTYMAP-WILDS, the location metadata is noised to protect", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 304, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 317 + ], + "score": 1.0, + "content": "privacy. For a deeper discussion of the ethics of remote sensing in the context of humanitarian aid", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 316, + 435, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 435, + 329 + ], + "score": 1.0, + "content": "and development, we refer readers to the UNICEF report by Berman et al. (2018).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 271, + 506, + 329 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 344, + 266, + 356 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 268, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 268, + 358 + ], + "score": 1.0, + "content": "REPRODUCIBILITY STATEMENT", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 109, + 369, + 504, + 402 + ], + "lines": [ + { + "bbox": [ + 106, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "All WILDS datasets are publicly available at https://wilds.stanford.edu, together with", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 380, + 505, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 392 + ], + "score": 1.0, + "content": "code and scripts to replicate all of the experiments in this paper. We also provide all trained model", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 391, + 390, + 403 + ], + "spans": [ + { + "bbox": [ + 107, + 391, + 390, + 403 + ], + "score": 1.0, + "content": "checkpoints and results, together with the exact hyperparameters used.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 368, + 505, + 403 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 407, + 411, + 419 + ], + "lines": [ + { + "bbox": [ + 105, + 407, + 413, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 413, + 422 + ], + "score": 1.0, + "content": "In our appendices, we provide more details on the datasets and experiments:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 407, + 413, + 422 + ] + }, + { + "type": "list", + "bbox": [ + 132, + 428, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 132, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 132, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "• In Appendix A, we describe each of the updated datasets in WILDS 2.0 and their sources", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 440, + 411, + 452 + ], + "spans": [ + { + "bbox": [ + 141, + 440, + 411, + 452 + ], + "score": 1.0, + "content": "of unlabeled data as well as what data processing steps were taken.", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 134, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 134, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "• In Appendix B, we describe the implementations of each of our benchmarked methods in", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 141, + 466, + 506, + 479 + ], + "score": 1.0, + "content": "detail. In particular, we discuss any changes we made to their original implementations,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 142, + 478, + 504, + 489 + ], + "spans": [ + { + "bbox": [ + 142, + 478, + 504, + 489 + ], + "score": 1.0, + "content": "either for consistency with other methods or with prior implementations of these methods.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 131, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 131, + 492, + 505, + 506 + ], + "score": 1.0, + "content": "• In Appendix C, we describe details of the data augmentations (if any) that we used across", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 503, + 195, + 515 + ], + "spans": [ + { + "bbox": [ + 141, + 503, + 195, + 515 + ], + "score": 1.0, + "content": "each dataset.", + "type": "text" + } + ], + "index": 31, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 518, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 131, + 518, + 506, + 532 + ], + "score": 1.0, + "content": "• In Appendix D, we describe our experimental protocol, including the hyperparameter se-", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 530, + 460, + 543 + ], + "spans": [ + { + "bbox": [ + 141, + 530, + 460, + 543 + ], + "score": 1.0, + "content": "lection procedure and hyperparameter grids for all of the methods and datasets.", + "type": "text" + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 545, + 439, + 558 + ], + "spans": [ + { + "bbox": [ + 132, + 545, + 439, + 558 + ], + "score": 1.0, + "content": "• In Appendix E, we describe the details of our experiments on DomainNet.", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 560, + 456, + 574 + ], + "spans": [ + { + "bbox": [ + 131, + 560, + 456, + 574 + ], + "score": 1.0, + "content": "• In Appendix F, we describe the details of our fully-labeled ERM experiments.", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 132, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "• Finally, in Appendix G, we include an illustrative code snippet of how to use the data", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 586, + 262, + 600 + ], + "spans": [ + { + "bbox": [ + 141, + 586, + 262, + 600 + ], + "score": 1.0, + "content": "loaders in the WILDS library.", + "type": "text" + } + ], + "index": 37, + "is_list_end_line": true + } + ], + "index": 31, + "bbox_fs": [ + 131, + 429, + 506, + 600 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 616, + 240, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 614, + 241, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 241, + 630 + ], + "score": 1.0, + "content": "AUTHOR CONTRIBUTIONS", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 640, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 505, + 652 + ], + "score": 1.0, + "content": "The project was initiated by Shiori Sagawa, Pang Wei Koh, and Percy Liang. Shiori Sagawa and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "score": 1.0, + "content": "Pang Wei Koh led the project and coordinated the activities below. Tony Lee developed the exper-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 662, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 505, + 674 + ], + "score": 1.0, + "content": "imental infrastructure and ran the experiments. Tony Lee, Irena Gao, Sang Michael Xie, Kendrick", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 672, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 505, + 686 + ], + "score": 1.0, + "content": "Shen, Ananya Kumar, and Michihiro Yasunaga designed the evaluation framework and implemented", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 683, + 485, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 683, + 485, + 698 + ], + "score": 1.0, + "content": "the algorithms. The unlabeled data loaders and corresponding dataset writeups were added by:", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 640, + 505, + 698 + ] + }, + { + "type": "text", + "bbox": [ + 132, + 705, + 280, + 732 + ], + "lines": [ + { + "bbox": [ + 132, + 705, + 259, + 717 + ], + "spans": [ + { + "bbox": [ + 132, + 705, + 259, + 717 + ], + "score": 1.0, + "content": "• AMAZON-WILDS: Tony Lee", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 132, + 720, + 280, + 733 + ], + "spans": [ + { + "bbox": [ + 132, + 720, + 280, + 733 + ], + "score": 1.0, + "content": "• CAMELYON17-WILDS: Tony Lee", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5, + "bbox_fs": [ + 132, + 705, + 280, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 131, + 82, + 422, + 175 + ], + "lines": [ + { + "bbox": [ + 131, + 82, + 297, + 96 + ], + "spans": [ + { + "bbox": [ + 131, + 82, + 297, + 96 + ], + "score": 1.0, + "content": "• CIVILCOMMENTS-WILDS: Irena Gao", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 131, + 99, + 285, + 112 + ], + "spans": [ + { + "bbox": [ + 131, + 99, + 285, + 112 + ], + "score": 1.0, + "content": "• FMOW-WILDS: Sang Michael Xie", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 131, + 115, + 385, + 128 + ], + "spans": [ + { + "bbox": [ + 131, + 115, + 385, + 128 + ], + "score": 1.0, + "content": "• IWILDCAM2020-WILDS: Henrik Marklund and Sara Beery", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 131, + 132, + 270, + 143 + ], + "spans": [ + { + "bbox": [ + 131, + 132, + 270, + 143 + ], + "score": 1.0, + "content": "• OGB-MOLPCBA: Weihua Hu", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 131, + 146, + 315, + 160 + ], + "spans": [ + { + "bbox": [ + 131, + 146, + 315, + 160 + ], + "score": 1.0, + "content": "• POVERTYMAP-WILDS: Sang Michael Xie", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 131, + 163, + 422, + 176 + ], + "spans": [ + { + "bbox": [ + 131, + 163, + 422, + 176 + ], + "score": 1.0, + "content": "• GLOBALWHEAT-WILDS: Etienne David, Ian Stavness, and Wei Guo.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 185, + 503, + 229 + ], + "lines": [ + { + "bbox": [ + 105, + 184, + 505, + 197 + ], + "spans": [ + { + "bbox": [ + 105, + 184, + 505, + 197 + ], + "score": 1.0, + "content": "Tony Lee and Henrik Marklund set up the website and leaderboards. Jure Leskovec, Kate Saenko,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 195, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 505, + 208 + ], + "score": 1.0, + "content": "Tatsunori Hashimoto, Sergey Levine, Chelsea Finn and Percy Liang provided advice on the overall", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "project direction and experimental design and analysis throughout. Shiori Sagawa, Pang Wei Koh,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 217, + 457, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 457, + 231 + ], + "score": 1.0, + "content": "and Irena Gao drafted the paper; all authors contributed towards writing the final paper.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 108, + 247, + 225, + 259 + ], + "lines": [ + { + "bbox": [ + 106, + 246, + 226, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 226, + 262 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENTS", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 108, + 272, + 504, + 305 + ], + "lines": [ + { + "bbox": [ + 105, + 270, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 506, + 286 + ], + "score": 1.0, + "content": "We would like to thank Ashwin Ramaswami, Berton Earnshaw, Bowen Liu, Hongseok Namkoong,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 283, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 506, + 295 + ], + "score": 1.0, + "content": "Junguang Jiang, Ludwig Schmidt, Robbie Jones, Robin Jia, Ruijia Xu, and Yabin Zhang for their", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 293, + 168, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 168, + 307 + ], + "score": 1.0, + "content": "helpful advice.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 108, + 311, + 504, + 344 + ], + "lines": [ + { + "bbox": [ + 106, + 310, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 505, + 324 + ], + "score": 1.0, + "content": "The design of the WILDS benchmark was inspired by the Open Graph Benchmark (Hu et al., 2020b),", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "and we are grateful to the Open Graph Benchmark team for their advice and help in setting up our", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 333, + 155, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 155, + 344 + ], + "score": 1.0, + "content": "benchmark.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 350, + 505, + 470 + ], + "lines": [ + { + "bbox": [ + 105, + 349, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 506, + 363 + ], + "score": 1.0, + "content": "This project was funded by an Open Philanthropy Project Award and NSF Award Grant", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 360, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 374 + ], + "score": 1.0, + "content": "No. 1805310. Shiori Sagawa was supported by the Herbert Kunzel Stanford Graduate Fellowship", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 505, + 384 + ], + "score": 1.0, + "content": "and the Apple Scholars in AI/ML PhD fellowship. Sang Michael Xie was supported by a NDSEG", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "Graduate Fellowship. Ananya Kumar was supported by the Rambus Corporation Stanford Graduate", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 393, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 407 + ], + "score": 1.0, + "content": "Fellowship. Weihua Hu was supported by the Funai Overseas Scholarship and the Masason Foun-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 404, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 505, + 418 + ], + "score": 1.0, + "content": "dation Fellowship. Michihiro Yasunaga was supported by the Microsoft Research PhD Fellowship.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "Henrik Marklund was supported by the Dr. Tech. Marcus Wallenberg Foundation for Education in", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "International Industrial Entrepreneurship, CIFAR, and Google. Sara Beery was supported by an", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "NSF Graduate Research Fellowship and is a PIMCO Fellow in Data Science. Jure Leskovec is a", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "Chan Zuckerberg Biohub investigator. Chelsea Finn is a CIFAR Fellow in the Learning in Machines", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 459, + 191, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 191, + 473 + ], + "score": 1.0, + "content": "and Brains Program.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 476, + 505, + 543 + ], + "lines": [ + { + "bbox": [ + 106, + 475, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 505, + 490 + ], + "score": 1.0, + "content": "We also gratefully acknowledge the support of DARPA under Nos. N660011924033 (MCS); ARO", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "under Nos. W911NF-16-1-0342 (MURI), W911NF-16-1-0171 (DURIP); NSF under Nos. OAC-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 497, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 505, + 512 + ], + "score": 1.0, + "content": "1835598 (CINES), OAC-1934578 (HDR), CCF-1918940 (Expeditions), IIS-2030477 (RAPID);", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "Stanford Data Science Initiative, Wu Tsai Neurosciences Institute, Chan Zuckerberg Biohub, Ama-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "score": 1.0, + "content": "zon, JPMorgan Chase, Docomo, Hitachi, JD.com, KDDI, NVIDIA, Dell, Toshiba, and UnitedHealth", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 529, + 138, + 546 + ], + "spans": [ + { + "bbox": [ + 104, + 529, + 138, + 546 + ], + "score": 1.0, + "content": "Group.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30.5 + }, + { + "type": "title", + "bbox": [ + 108, + 561, + 175, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 176, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 176, + 574 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 105, + 580, + 504, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 577, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 595 + ], + "score": 1.0, + "content": "Abubakar Abid, Maheen Farooqi, and James Zou. Persistent anti-muslim bias in large language", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 591, + 314, + 603 + ], + "spans": [ + { + "bbox": [ + 115, + 591, + 314, + 603 + ], + "score": 1.0, + "content": "models. arXiv preprint arXiv:2101.05783, 2021.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 611, + 505, + 657 + ], + "lines": [ + { + "bbox": [ + 105, + 611, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 505, + 625 + ], + "score": 1.0, + "content": "Jorge A Ahumada, Eric Fegraus, Tanya Birch, Nicole Flores, Roland Kays, Timothy G OBrien,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 624, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 116, + 624, + 505, + 635 + ], + "score": 1.0, + "content": "Jonathan Palmer, Stephanie Schuttler, Jennifer Y Zhao, Walter Jetz, et al. Wildlife insights: A", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 635, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 115, + 635, + 505, + 646 + ], + "score": 1.0, + "content": "platform to maximize the potential of camera trap and other passive sensor wildlife data for the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 645, + 335, + 657 + ], + "spans": [ + { + "bbox": [ + 115, + 645, + 335, + 657 + ], + "score": 1.0, + "content": "planet. Environmental Conservation, 47(1):1–6, 2020.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 109, + 666, + 504, + 700 + ], + "lines": [ + { + "bbox": [ + 107, + 666, + 504, + 678 + ], + "spans": [ + { + "bbox": [ + 107, + 666, + 504, + 678 + ], + "score": 1.0, + "content": "Saad Ullah Akram, Talha Qaiser, Simon Graham, Juho Kannala, Janne Heikkila, and Nasir Rajpoot.¨", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 115, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "Leveraging unlabeled whole-slide-images for mitosis detection. Computational Pathology and", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 117, + 689, + 329, + 701 + ], + "spans": [ + { + "bbox": [ + 117, + 689, + 329, + 701 + ], + "score": 1.0, + "content": "Ophthalmic Medical Image Analysis, 1:69–77, 2018.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "EA AlBadawy, A Saha, and MA Mazurowski. Deep learning for segmentation of brain tumors:", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 116, + 720, + 402, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 402, + 733 + ], + "score": 1.0, + "content": "Impact of cross-institutional training and testing. Med Phys., 45, 2018.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 131, + 82, + 422, + 175 + ], + "lines": [ + { + "bbox": [ + 131, + 82, + 297, + 96 + ], + "spans": [ + { + "bbox": [ + 131, + 82, + 297, + 96 + ], + "score": 1.0, + "content": "• CIVILCOMMENTS-WILDS: Irena Gao", + "type": "text" + } + ], + "index": 0, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 99, + 285, + 112 + ], + "spans": [ + { + "bbox": [ + 131, + 99, + 285, + 112 + ], + "score": 1.0, + "content": "• FMOW-WILDS: Sang Michael Xie", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 115, + 385, + 128 + ], + "spans": [ + { + "bbox": [ + 131, + 115, + 385, + 128 + ], + "score": 1.0, + "content": "• IWILDCAM2020-WILDS: Henrik Marklund and Sara Beery", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 132, + 270, + 143 + ], + "spans": [ + { + "bbox": [ + 131, + 132, + 270, + 143 + ], + "score": 1.0, + "content": "• OGB-MOLPCBA: Weihua Hu", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 146, + 315, + 160 + ], + "spans": [ + { + "bbox": [ + 131, + 146, + 315, + 160 + ], + "score": 1.0, + "content": "• POVERTYMAP-WILDS: Sang Michael Xie", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 163, + 422, + 176 + ], + "spans": [ + { + "bbox": [ + 131, + 163, + 422, + 176 + ], + "score": 1.0, + "content": "• GLOBALWHEAT-WILDS: Etienne David, Ian Stavness, and Wei Guo.", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + } + ], + "index": 2.5, + "bbox_fs": [ + 131, + 82, + 422, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 185, + 503, + 229 + ], + "lines": [ + { + "bbox": [ + 105, + 184, + 505, + 197 + ], + "spans": [ + { + "bbox": [ + 105, + 184, + 505, + 197 + ], + "score": 1.0, + "content": "Tony Lee and Henrik Marklund set up the website and leaderboards. Jure Leskovec, Kate Saenko,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 195, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 505, + 208 + ], + "score": 1.0, + "content": "Tatsunori Hashimoto, Sergey Levine, Chelsea Finn and Percy Liang provided advice on the overall", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "project direction and experimental design and analysis throughout. Shiori Sagawa, Pang Wei Koh,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 217, + 457, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 457, + 231 + ], + "score": 1.0, + "content": "and Irena Gao drafted the paper; all authors contributed towards writing the final paper.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 184, + 505, + 231 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 247, + 225, + 259 + ], + "lines": [ + { + "bbox": [ + 106, + 246, + 226, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 226, + 262 + ], + "score": 1.0, + "content": "ACKNOWLEDGEMENTS", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 108, + 272, + 504, + 305 + ], + "lines": [ + { + "bbox": [ + 105, + 270, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 506, + 286 + ], + "score": 1.0, + "content": "We would like to thank Ashwin Ramaswami, Berton Earnshaw, Bowen Liu, Hongseok Namkoong,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 283, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 506, + 295 + ], + "score": 1.0, + "content": "Junguang Jiang, Ludwig Schmidt, Robbie Jones, Robin Jia, Ruijia Xu, and Yabin Zhang for their", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 293, + 168, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 168, + 307 + ], + "score": 1.0, + "content": "helpful advice.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 270, + 506, + 307 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 311, + 504, + 344 + ], + "lines": [ + { + "bbox": [ + 106, + 310, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 505, + 324 + ], + "score": 1.0, + "content": "The design of the WILDS benchmark was inspired by the Open Graph Benchmark (Hu et al., 2020b),", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "and we are grateful to the Open Graph Benchmark team for their advice and help in setting up our", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 333, + 155, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 155, + 344 + ], + "score": 1.0, + "content": "benchmark.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 106, + 310, + 505, + 344 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 350, + 505, + 470 + ], + "lines": [ + { + "bbox": [ + 105, + 349, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 506, + 363 + ], + "score": 1.0, + "content": "This project was funded by an Open Philanthropy Project Award and NSF Award Grant", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 360, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 374 + ], + "score": 1.0, + "content": "No. 1805310. Shiori Sagawa was supported by the Herbert Kunzel Stanford Graduate Fellowship", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 505, + 384 + ], + "score": 1.0, + "content": "and the Apple Scholars in AI/ML PhD fellowship. Sang Michael Xie was supported by a NDSEG", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "Graduate Fellowship. Ananya Kumar was supported by the Rambus Corporation Stanford Graduate", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 393, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 407 + ], + "score": 1.0, + "content": "Fellowship. Weihua Hu was supported by the Funai Overseas Scholarship and the Masason Foun-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 404, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 505, + 418 + ], + "score": 1.0, + "content": "dation Fellowship. Michihiro Yasunaga was supported by the Microsoft Research PhD Fellowship.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "Henrik Marklund was supported by the Dr. Tech. Marcus Wallenberg Foundation for Education in", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "International Industrial Entrepreneurship, CIFAR, and Google. Sara Beery was supported by an", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "NSF Graduate Research Fellowship and is a PIMCO Fellow in Data Science. Jure Leskovec is a", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "Chan Zuckerberg Biohub investigator. Chelsea Finn is a CIFAR Fellow in the Learning in Machines", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 459, + 191, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 191, + 473 + ], + "score": 1.0, + "content": "and Brains Program.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 349, + 506, + 473 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 476, + 505, + 543 + ], + "lines": [ + { + "bbox": [ + 106, + 475, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 505, + 490 + ], + "score": 1.0, + "content": "We also gratefully acknowledge the support of DARPA under Nos. N660011924033 (MCS); ARO", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "under Nos. W911NF-16-1-0342 (MURI), W911NF-16-1-0171 (DURIP); NSF under Nos. OAC-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 497, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 505, + 512 + ], + "score": 1.0, + "content": "1835598 (CINES), OAC-1934578 (HDR), CCF-1918940 (Expeditions), IIS-2030477 (RAPID);", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "Stanford Data Science Initiative, Wu Tsai Neurosciences Institute, Chan Zuckerberg Biohub, Ama-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "score": 1.0, + "content": "zon, JPMorgan Chase, Docomo, Hitachi, JD.com, KDDI, NVIDIA, Dell, Toshiba, and UnitedHealth", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 529, + 138, + 546 + ], + "spans": [ + { + "bbox": [ + 104, + 529, + 138, + 546 + ], + "score": 1.0, + "content": "Group.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30.5, + "bbox_fs": [ + 104, + 475, + 505, + 546 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 561, + 175, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 176, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 176, + 574 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 105, + 580, + 504, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 577, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 595 + ], + "score": 1.0, + "content": "Abubakar Abid, Maheen Farooqi, and James Zou. Persistent anti-muslim bias in large language", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 591, + 314, + 603 + ], + "spans": [ + { + "bbox": [ + 115, + 591, + 314, + 603 + ], + "score": 1.0, + "content": "models. arXiv preprint arXiv:2101.05783, 2021.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 577, + 506, + 603 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 611, + 505, + 657 + ], + "lines": [ + { + "bbox": [ + 105, + 611, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 505, + 625 + ], + "score": 1.0, + "content": "Jorge A Ahumada, Eric Fegraus, Tanya Birch, Nicole Flores, Roland Kays, Timothy G OBrien,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 624, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 116, + 624, + 505, + 635 + ], + "score": 1.0, + "content": "Jonathan Palmer, Stephanie Schuttler, Jennifer Y Zhao, Walter Jetz, et al. Wildlife insights: A", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 635, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 115, + 635, + 505, + 646 + ], + "score": 1.0, + "content": "platform to maximize the potential of camera trap and other passive sensor wildlife data for the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 645, + 335, + 657 + ], + "spans": [ + { + "bbox": [ + 115, + 645, + 335, + 657 + ], + "score": 1.0, + "content": "planet. Environmental Conservation, 47(1):1–6, 2020.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 611, + 505, + 657 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 666, + 504, + 700 + ], + "lines": [ + { + "bbox": [ + 107, + 666, + 504, + 678 + ], + "spans": [ + { + "bbox": [ + 107, + 666, + 504, + 678 + ], + "score": 1.0, + "content": "Saad Ullah Akram, Talha Qaiser, Simon Graham, Juho Kannala, Janne Heikkila, and Nasir Rajpoot.¨", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 115, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "Leveraging unlabeled whole-slide-images for mitosis detection. Computational Pathology and", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 117, + 689, + 329, + 701 + ], + "spans": [ + { + "bbox": [ + 117, + 689, + 329, + 701 + ], + "score": 1.0, + "content": "Ophthalmic Medical Image Analysis, 1:69–77, 2018.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 107, + 666, + 505, + 701 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "EA AlBadawy, A Saha, and MA Mazurowski. Deep learning for segmentation of brain tumors:", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 116, + 720, + 402, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 402, + 733 + ], + "score": 1.0, + "content": "Impact of cross-institutional training and testing. Med Phys., 45, 2018.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5, + "bbox_fs": [ + 106, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Shekoofeh Azizi, Basil Mustafa, Fiona Ryan, Zachary Beaver, Jan Freyberg, Jonathan Deaton,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "Aaron Loh, Alan Karthikesalingam, Simon Kornblith, Ting Chen, et al. Big self-supervised", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 104, + 464, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 104, + 464, + 117 + ], + "score": 1.0, + "content": "models advance medical image classification. arXiv preprint arXiv:2101.05224, 2021.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 505, + 167 + ], + "lines": [ + { + "bbox": [ + 106, + 123, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 505, + 135 + ], + "score": 1.0, + "content": "Peter Bandi, Oscar Geessink, Quirine Manson, Marcory Van Dijk, Maschenka Balkenhol, Meyke", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 133, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 133, + 505, + 146 + ], + "score": 1.0, + "content": "Hermsen, Babak Ehteshami Bejnordi, Byungjae Lee, Kyunghyun Paeng, Aoxiao Zhong, et al.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 505, + 157 + ], + "score": 1.0, + "content": "From detection of individual metastases to classification of lymph node status at the patient level:", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 156, + 495, + 168 + ], + "spans": [ + { + "bbox": [ + 115, + 156, + 495, + 168 + ], + "score": 1.0, + "content": "the CAMELYON17 challenge. IEEE Transactions on Medical Imaging, 38(2):550–560, 2018.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 108, + 174, + 502, + 198 + ], + "lines": [ + { + "bbox": [ + 106, + 173, + 504, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 504, + 188 + ], + "score": 1.0, + "content": "Sara Beery, Grant Van Horn, and Pietro Perona. Recognition in terra incognita. In European Con-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 114, + 186, + 347, + 198 + ], + "spans": [ + { + "bbox": [ + 114, + 186, + 347, + 198 + ], + "score": 1.0, + "content": "ference on Computer Vision (ECCV), pp. 456–473, 2018.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 108, + 204, + 503, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 218 + ], + "score": 1.0, + "content": "Sara Beery, Dan Morris, and Siyu Yang. Efficient pipeline for camera trap image review. arXiv", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 216, + 254, + 227 + ], + "spans": [ + { + "bbox": [ + 115, + 216, + 254, + 227 + ], + "score": 1.0, + "content": "preprint arXiv:1907.06772, 2019.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 233, + 503, + 257 + ], + "lines": [ + { + "bbox": [ + 105, + 232, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 247 + ], + "score": 1.0, + "content": "Sara Beery, Elijah Cole, and Arvi Gjoka. The iwildcam 2020 competition dataset. arXiv preprint", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 244, + 219, + 257 + ], + "spans": [ + { + "bbox": [ + 115, + 244, + 219, + 257 + ], + "score": 1.0, + "content": "arXiv:2004.10340, 2020.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 263, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 106, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wort-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 116, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "man Vaughan. A theory of learning from different domains. Machine Learning, 79(1):151–175,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 284, + 143, + 298 + ], + "spans": [ + { + "bbox": [ + 115, + 284, + 143, + 298 + ], + "score": 1.0, + "content": "2010.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 304, + 505, + 338 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 505, + 317 + ], + "score": 1.0, + "content": "Gabrielle Berman, Sara de la Rosa, and Tanya Accone. Ethical considerations when using geospatial", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 316, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 116, + 316, + 505, + 328 + ], + "score": 1.0, + "content": "technologies for evidence generation. Innocenti Discussion Paper, UNICEF Office of Research,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 325, + 143, + 339 + ], + "spans": [ + { + "bbox": [ + 115, + 325, + 143, + 339 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 345, + 505, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 504, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 504, + 357 + ], + "score": 1.0, + "content": "David Berthelot, Rebecca Roelofs, Kihyuk Sohn, Nicholas Carlini, and Alex Kurakin. Adamatch:", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 116, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "A unified approach to semi-supervised learning and domain adaptation. arXiv preprint", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 367, + 219, + 378 + ], + "spans": [ + { + "bbox": [ + 116, + 367, + 219, + 378 + ], + "score": 1.0, + "content": "arXiv:2106.04732, 2021.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 385, + 502, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 384, + 504, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 504, + 399 + ], + "score": 1.0, + "content": "Lukas Biewald. Experiment tracking with weights and biases, 2020. URL https://www.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 396, + 331, + 409 + ], + "spans": [ + { + "bbox": [ + 116, + 396, + 331, + 409 + ], + "score": 1.0, + "content": "wandb.com/. Software available from wandb.com.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 415, + 503, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 429 + ], + "score": 1.0, + "content": "John Blitzer and Fernando Pereira. Domain adaptation of natural language processing systems.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 117, + 426, + 253, + 438 + ], + "spans": [ + { + "bbox": [ + 117, + 426, + 253, + 438 + ], + "score": 1.0, + "content": "University of Pennsylvania, 2007.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 107, + 444, + 505, + 479 + ], + "lines": [ + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "John Blitzer, Mark Dredze, and Fernando Pereira. Biographies, bollywood, boom-boxes and", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 116, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "blenders: Domain adaptation for sentiment classification. In Proceedings of the 45th annual", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 467, + 420, + 479 + ], + "spans": [ + { + "bbox": [ + 116, + 467, + 420, + 479 + ], + "score": 1.0, + "content": "meeting of the association of computational linguistics, pp. 440–447, 2007.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 485, + 505, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 498 + ], + "score": 1.0, + "content": "Evan E Bolton, Yanli Wang, Paul A Thiessen, and Stephen H Bryant. Pubchem: integrated plat-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 496, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 115, + 496, + 505, + 510 + ], + "score": 1.0, + "content": "form of small molecules and biological activities. In Annual reports in computational chemistry,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 117, + 507, + 276, + 519 + ], + "spans": [ + { + "bbox": [ + 117, + 507, + 276, + 519 + ], + "score": 1.0, + "content": "volume 4, pp. 217–241. Elsevier, 2008.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 108, + 525, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 525, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 506, + 540 + ], + "score": 1.0, + "content": "Tolga Bolukbasi, Kai-Wei Chang, James Y Zou, Venkatesh Saligrama, and Adam T Kalai. Man is", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 538, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 538, + 505, + 550 + ], + "score": 1.0, + "content": "to computer programmer as woman is to homemaker? Debiasing word embeddings. In Advances", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 548, + 423, + 561 + ], + "spans": [ + { + "bbox": [ + 115, + 548, + 423, + 561 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems (NeurIPS), pp. 4349–4357, 2016.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 109, + 567, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 107, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 107, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "Rishi Bommasani, Drew A. Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 577, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 115, + 577, + 505, + 591 + ], + "score": 1.0, + "content": "Michael S. Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, Erik Brynjolfsson,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 116, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "Shyamal Buch, Dallas Card, Rodrigo Castellon, Niladri Chatterji, Annie Chen, Kathleen Creel,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 115, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "Jared Quincy Davis, Dorottya Demszky, Chris Donahue, Moussa Doumbouya, Esin Durmus, Ste-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 115, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "fano Ermon, John Etchemendy, Kawin Ethayarajh, Li Fei-Fei, Chelsea Finn, Trevor Gale, Lauren", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 115, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "Gillespie, Karan Goel, Noah Goodman, Shelby Grossman, Neel Guha, Tatsunori Hashimoto, Pe-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 633, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 115, + 633, + 505, + 646 + ], + "score": 1.0, + "content": "ter Henderson, John Hewitt, Daniel E. Ho, Jenny Hong, Kyle Hsu, Jing Huang, Thomas Icard,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 115, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "Saahil Jain, Dan Jurafsky, Pratyusha Kalluri, Siddharth Karamcheti, Geoff Keeling, Fereshte", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 115, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "Khani, Omar Khattab, Pang Wei Koh, Mark Krass, Ranjay Krishna, Rohith Kuditipudi, Ananya", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 115, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "Kumar, Faisal Ladhak, Mina Lee, Tony Lee, Jure Leskovec, Isabelle Levent, Xiang Lisa Li,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 115, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "Xuechen Li, Tengyu Ma, Ali Malik, Christopher D. Manning, Suvir Mirchandani, Eric Mitchell,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 115, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "Zanele Munyikwa, Suraj Nair, Avanika Narayan, Deepak Narayanan, Ben Newman, Allen Nie,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 115, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 115, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Juan Carlos Niebles, Hamed Nilforoshan, Julian Nyarko, Giray Ogut, Laurel Orr, Isabel Papadim-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 115, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 115, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "itriou, Joon Sung Park, Chris Piech, Eva Portelance, Christopher Potts, Aditi Raghunathan, Rob", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 115, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "Reich, Hongyu Ren, Frieda Rong, Yusuf Roohani, Camilo Ruiz, Jack Ryan, Christopher R, Dorsa", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 42 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Shekoofeh Azizi, Basil Mustafa, Fiona Ryan, Zachary Beaver, Jan Freyberg, Jonathan Deaton,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "Aaron Loh, Alan Karthikesalingam, Simon Kornblith, Ting Chen, et al. Big self-supervised", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 104, + 464, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 104, + 464, + 117 + ], + "score": 1.0, + "content": "models advance medical image classification. arXiv preprint arXiv:2101.05224, 2021.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 81, + 505, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 505, + 167 + ], + "lines": [ + { + "bbox": [ + 106, + 123, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 106, + 123, + 505, + 135 + ], + "score": 1.0, + "content": "Peter Bandi, Oscar Geessink, Quirine Manson, Marcory Van Dijk, Maschenka Balkenhol, Meyke", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 133, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 133, + 505, + 146 + ], + "score": 1.0, + "content": "Hermsen, Babak Ehteshami Bejnordi, Byungjae Lee, Kyunghyun Paeng, Aoxiao Zhong, et al.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 505, + 157 + ], + "score": 1.0, + "content": "From detection of individual metastases to classification of lymph node status at the patient level:", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 156, + 495, + 168 + ], + "spans": [ + { + "bbox": [ + 115, + 156, + 495, + 168 + ], + "score": 1.0, + "content": "the CAMELYON17 challenge. IEEE Transactions on Medical Imaging, 38(2):550–560, 2018.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 106, + 123, + 505, + 168 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 174, + 502, + 198 + ], + "lines": [ + { + "bbox": [ + 106, + 173, + 504, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 504, + 188 + ], + "score": 1.0, + "content": "Sara Beery, Grant Van Horn, and Pietro Perona. Recognition in terra incognita. In European Con-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 114, + 186, + 347, + 198 + ], + "spans": [ + { + "bbox": [ + 114, + 186, + 347, + 198 + ], + "score": 1.0, + "content": "ference on Computer Vision (ECCV), pp. 456–473, 2018.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 106, + 173, + 504, + 198 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 204, + 503, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 218 + ], + "score": 1.0, + "content": "Sara Beery, Dan Morris, and Siyu Yang. Efficient pipeline for camera trap image review. arXiv", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 216, + 254, + 227 + ], + "spans": [ + { + "bbox": [ + 115, + 216, + 254, + 227 + ], + "score": 1.0, + "content": "preprint arXiv:1907.06772, 2019.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 203, + 505, + 227 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 233, + 503, + 257 + ], + "lines": [ + { + "bbox": [ + 105, + 232, + 505, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 247 + ], + "score": 1.0, + "content": "Sara Beery, Elijah Cole, and Arvi Gjoka. The iwildcam 2020 competition dataset. arXiv preprint", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 244, + 219, + 257 + ], + "spans": [ + { + "bbox": [ + 115, + 244, + 219, + 257 + ], + "score": 1.0, + "content": "arXiv:2004.10340, 2020.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 232, + 505, + 257 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 263, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 106, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wort-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 116, + 275, + 505, + 287 + ], + "score": 1.0, + "content": "man Vaughan. A theory of learning from different domains. Machine Learning, 79(1):151–175,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 284, + 143, + 298 + ], + "spans": [ + { + "bbox": [ + 115, + 284, + 143, + 298 + ], + "score": 1.0, + "content": "2010.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 106, + 263, + 505, + 298 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 304, + 505, + 338 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 505, + 317 + ], + "score": 1.0, + "content": "Gabrielle Berman, Sara de la Rosa, and Tanya Accone. Ethical considerations when using geospatial", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 316, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 116, + 316, + 505, + 328 + ], + "score": 1.0, + "content": "technologies for evidence generation. Innocenti Discussion Paper, UNICEF Office of Research,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 325, + 143, + 339 + ], + "spans": [ + { + "bbox": [ + 115, + 325, + 143, + 339 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 303, + 505, + 339 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 345, + 505, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 344, + 504, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 504, + 357 + ], + "score": 1.0, + "content": "David Berthelot, Rebecca Roelofs, Kihyuk Sohn, Nicholas Carlini, and Alex Kurakin. Adamatch:", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 116, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "A unified approach to semi-supervised learning and domain adaptation. arXiv preprint", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 367, + 219, + 378 + ], + "spans": [ + { + "bbox": [ + 116, + 367, + 219, + 378 + ], + "score": 1.0, + "content": "arXiv:2106.04732, 2021.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 106, + 344, + 505, + 378 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 385, + 502, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 384, + 504, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 504, + 399 + ], + "score": 1.0, + "content": "Lukas Biewald. Experiment tracking with weights and biases, 2020. URL https://www.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 396, + 331, + 409 + ], + "spans": [ + { + "bbox": [ + 116, + 396, + 331, + 409 + ], + "score": 1.0, + "content": "wandb.com/. Software available from wandb.com.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 384, + 504, + 409 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 415, + 503, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 429 + ], + "score": 1.0, + "content": "John Blitzer and Fernando Pereira. Domain adaptation of natural language processing systems.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 117, + 426, + 253, + 438 + ], + "spans": [ + { + "bbox": [ + 117, + 426, + 253, + 438 + ], + "score": 1.0, + "content": "University of Pennsylvania, 2007.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 414, + 505, + 438 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 444, + 505, + 479 + ], + "lines": [ + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "John Blitzer, Mark Dredze, and Fernando Pereira. Biographies, bollywood, boom-boxes and", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 116, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "blenders: Domain adaptation for sentiment classification. In Proceedings of the 45th annual", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 467, + 420, + 479 + ], + "spans": [ + { + "bbox": [ + 116, + 467, + 420, + 479 + ], + "score": 1.0, + "content": "meeting of the association of computational linguistics, pp. 440–447, 2007.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 444, + 505, + 479 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 485, + 505, + 519 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 498 + ], + "score": 1.0, + "content": "Evan E Bolton, Yanli Wang, Paul A Thiessen, and Stephen H Bryant. Pubchem: integrated plat-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 496, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 115, + 496, + 505, + 510 + ], + "score": 1.0, + "content": "form of small molecules and biological activities. In Annual reports in computational chemistry,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 117, + 507, + 276, + 519 + ], + "spans": [ + { + "bbox": [ + 117, + 507, + 276, + 519 + ], + "score": 1.0, + "content": "volume 4, pp. 217–241. Elsevier, 2008.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 484, + 505, + 519 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 525, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 525, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 506, + 540 + ], + "score": 1.0, + "content": "Tolga Bolukbasi, Kai-Wei Chang, James Y Zou, Venkatesh Saligrama, and Adam T Kalai. Man is", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 538, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 538, + 505, + 550 + ], + "score": 1.0, + "content": "to computer programmer as woman is to homemaker? Debiasing word embeddings. In Advances", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 548, + 423, + 561 + ], + "spans": [ + { + "bbox": [ + 115, + 548, + 423, + 561 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems (NeurIPS), pp. 4349–4357, 2016.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 525, + 506, + 561 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 567, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 107, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 107, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "Rishi Bommasani, Drew A. Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 577, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 115, + 577, + 505, + 591 + ], + "score": 1.0, + "content": "Michael S. Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, Erik Brynjolfsson,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 116, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "Shyamal Buch, Dallas Card, Rodrigo Castellon, Niladri Chatterji, Annie Chen, Kathleen Creel,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 115, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "Jared Quincy Davis, Dorottya Demszky, Chris Donahue, Moussa Doumbouya, Esin Durmus, Ste-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 115, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "fano Ermon, John Etchemendy, Kawin Ethayarajh, Li Fei-Fei, Chelsea Finn, Trevor Gale, Lauren", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 115, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "Gillespie, Karan Goel, Noah Goodman, Shelby Grossman, Neel Guha, Tatsunori Hashimoto, Pe-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 633, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 115, + 633, + 505, + 646 + ], + "score": 1.0, + "content": "ter Henderson, John Hewitt, Daniel E. Ho, Jenny Hong, Kyle Hsu, Jing Huang, Thomas Icard,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 643, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 115, + 643, + 505, + 656 + ], + "score": 1.0, + "content": "Saahil Jain, Dan Jurafsky, Pratyusha Kalluri, Siddharth Karamcheti, Geoff Keeling, Fereshte", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 115, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "Khani, Omar Khattab, Pang Wei Koh, Mark Krass, Ranjay Krishna, Rohith Kuditipudi, Ananya", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 115, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "Kumar, Faisal Ladhak, Mina Lee, Tony Lee, Jure Leskovec, Isabelle Levent, Xiang Lisa Li,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 115, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "Xuechen Li, Tengyu Ma, Ali Malik, Christopher D. Manning, Suvir Mirchandani, Eric Mitchell,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 115, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "Zanele Munyikwa, Suraj Nair, Avanika Narayan, Deepak Narayanan, Ben Newman, Allen Nie,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 115, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 115, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Juan Carlos Niebles, Hamed Nilforoshan, Julian Nyarko, Giray Ogut, Laurel Orr, Isabel Papadim-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 115, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 115, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "itriou, Joon Sung Park, Chris Piech, Eva Portelance, Christopher Potts, Aditi Raghunathan, Rob", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 115, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "Reich, Hongyu Ren, Frieda Rong, Yusuf Roohani, Camilo Ruiz, Jack Ryan, Christopher R, Dorsa", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 42, + "bbox_fs": [ + 107, + 567, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 116, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 116, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 116, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "Sadigh, Shiori Sagawa, Keshav Santhanam, Andy Shih, Krishnan Srinivasan, Alex Tamkin, Ro-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "han Taori, Armin W. Thomas, Florian Tramr, Rose E. Wang, William Wang, Bohan Wu, Jiajun", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 103, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 103, + 505, + 117 + ], + "score": 1.0, + "content": "Wu, Yuhuai Wu, Sang Michael Xie, Michihiro Yasunaga, Jiaxuan You, Matei Zaharia, Michael", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 113, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 115, + 113, + 505, + 130 + ], + "score": 1.0, + "content": "Zhang, Tianyi Zhang, Xikun Zhang, Yuhui Zhang, Lucia Zheng, Kaitlyn Zhou, and Percy Liang.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 126, + 492, + 140 + ], + "spans": [ + { + "bbox": [ + 116, + 126, + 492, + 140 + ], + "score": 1.0, + "content": "On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258, 2021.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 108, + 146, + 505, + 180 + ], + "lines": [ + { + "bbox": [ + 106, + 146, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 505, + 158 + ], + "score": 1.0, + "content": "Daniel Borkan, Lucas Dixon, Jeffrey Sorensen, Nithum Thain, and Lucy Vasserman. Nuanced", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 157, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 115, + 157, + 505, + 169 + ], + "score": 1.0, + "content": "metrics for measuring unintended bias with real data for text classification. In World Wide Web", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 168, + 233, + 180 + ], + "spans": [ + { + "bbox": [ + 115, + 168, + 233, + 180 + ], + "score": 1.0, + "content": "(WWW), pp. 491–500, 2019.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 264 + ], + "lines": [ + { + "bbox": [ + 106, + 187, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 200 + ], + "score": 1.0, + "content": "Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhari-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 199, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 116, + 199, + 505, + 211 + ], + "score": 1.0, + "content": "wal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 209, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 115, + 209, + 505, + 222 + ], + "score": 1.0, + "content": "Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 219, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 115, + 219, + 505, + 234 + ], + "score": 1.0, + "content": "Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 114, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 114, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 114, + 242, + 506, + 256 + ], + "spans": [ + { + "bbox": [ + 114, + 242, + 506, + 256 + ], + "score": 1.0, + "content": "Ilya Sutskever, and Dario Amodei. Language models are few-shot learners. arXiv preprint", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 253, + 219, + 266 + ], + "spans": [ + { + "bbox": [ + 115, + 253, + 219, + 266 + ], + "score": 1.0, + "content": "arXiv:2005.14165, 2020.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 273, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 505, + 285 + ], + "score": 1.0, + "content": "Maxwell Burnette, Rob Kooper, J. D. Maloney, Gareth S. Rohde, Jeffrey A. Terstriep, Craig Willis,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 283, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 115, + 283, + 506, + 297 + ], + "score": 1.0, + "content": "Noah Fahlgren, Todd Mockler, Maria Newcomb, Vasit Sagan, Pedro Andrade-Sanchez, Nadia", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 294, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 115, + 294, + 506, + 308 + ], + "score": 1.0, + "content": "Shakoor, Paheding Sidike, Rick Ward, and David LeBauer. Terra-ref data processing infrastruc-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 306, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 116, + 306, + 506, + 319 + ], + "score": 1.0, + "content": "ture. In Proceedings of the Practice and Experience on Advanced Research Computing, PEARC", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 317, + 504, + 330 + ], + "spans": [ + { + "bbox": [ + 116, + 317, + 504, + 330 + ], + "score": 1.0, + "content": "’18, New York, NY, USA, 2018. Association for Computing Machinery. ISBN 9781450364461.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 336, + 503, + 359 + ], + "lines": [ + { + "bbox": [ + 106, + 336, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 505, + 349 + ], + "score": 1.0, + "content": "Aylin Caliskan, Joanna J Bryson, and Arvind Narayanan. Semantics derived automatically from", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 348, + 443, + 360 + ], + "spans": [ + { + "bbox": [ + 115, + 348, + 443, + 360 + ], + "score": 1.0, + "content": "language corpora contain human-like biases. Science, 356(6334):183–186, 2017.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 503, + 401 + ], + "lines": [ + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin. Un-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 379, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 505, + 391 + ], + "score": 1.0, + "content": "supervised learning of visual features by contrasting cluster assignments. In Advances in Neural", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 389, + 430, + 403 + ], + "spans": [ + { + "bbox": [ + 116, + 389, + 430, + 403 + ], + "score": 1.0, + "content": "Information Processing Systems (NeurIPS), volume 33, pp. 9912–9924, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 408, + 504, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 408, + 504, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 504, + 421 + ], + "score": 1.0, + "content": "Ilias Chalkidis, Manos Fergadiotis, Prodromos Malakasiotis, Nikolaos Aletras, and Ion Androut-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 116, + 420, + 506, + 433 + ], + "score": 1.0, + "content": "sopoulos. LEGAL-BERT:”preparing the muppets for court”. In Empirical Methods in Natural", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 431, + 339, + 443 + ], + "spans": [ + { + "bbox": [ + 116, + 431, + 339, + 443 + ], + "score": 1.0, + "content": "Language Processing (EMNLP), pp. 2898–2904, 2020.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 450, + 505, + 484 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 464 + ], + "score": 1.0, + "content": "Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 460, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 460, + 505, + 476 + ], + "score": 1.0, + "content": "contrastive learning of visual representations. In International Conference on Machine Learning", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 117, + 473, + 246, + 485 + ], + "spans": [ + { + "bbox": [ + 117, + 473, + 246, + 485 + ], + "score": 1.0, + "content": "(ICML), pp. 1597–1607, 2020a.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 526 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 505 + ], + "score": 1.0, + "content": "Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 502, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 115, + 502, + 506, + 517 + ], + "score": 1.0, + "content": "contrastive learning of visual representations. In International conference on machine learning,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 114, + 514, + 244, + 526 + ], + "spans": [ + { + "bbox": [ + 114, + 514, + 244, + 526 + ], + "score": 1.0, + "content": "pp. 1597–1607. PMLR, 2020b.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 534, + 504, + 557 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 547 + ], + "score": 1.0, + "content": "Gordon Christie, Neil Fendley, James Wilson, and Ryan Mukherjee. Functional map of the world.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 545, + 358, + 558 + ], + "spans": [ + { + "bbox": [ + 116, + 545, + 358, + 558 + ], + "score": 1.0, + "content": "In Computer Vision and Pattern Recognition (CVPR), 2018.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 504, + 587 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 578 + ], + "score": 1.0, + "content": "Ozan Ciga, Anne L Martel, and Tony Xu. Self supervised contrastive learning for digital histopathol-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 576, + 299, + 588 + ], + "spans": [ + { + "bbox": [ + 115, + 576, + 299, + 588 + ], + "score": 1.0, + "content": "ogy. arXiv preprint arXiv:2011.13971, 2020.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 505, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "score": 1.0, + "content": "Jonathan H Clark, Eunsol Choi, Michael Collins, Dan Garrette, Tom Kwiatkowski, Vitaly Nikolaev,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 606, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 115, + 606, + 505, + 619 + ], + "score": 1.0, + "content": "and Jennimaria Palomaki. Tydi qa: A benchmark for information-seeking question answering in", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 617, + 411, + 630 + ], + "spans": [ + { + "bbox": [ + 116, + 617, + 411, + 630 + ], + "score": 1.0, + "content": "typologically diverse languages. arXiv preprint arXiv:2003.05002, 2020.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 105, + 636, + 504, + 660 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "Elijah Cole, Xuan Yang, Kimberly Wilber, Oisin Mac Aodha, and Serge Belongie. When does", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 648, + 474, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 648, + 474, + 660 + ], + "score": 1.0, + "content": "contrastive visual representation learning work? arXiv preprint arXiv:2105.05837, 2021.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 104, + 667, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 505, + 681 + ], + "score": 1.0, + "content": "Alexis Conneau and Guillaume Lample. Cross-lingual language model pretraining. In Advances in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 678, + 412, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 678, + 412, + 691 + ], + "score": 1.0, + "content": "Neural Information Processing Systems (NeurIPS), pp. 7059–7069, 2019.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 697, + 505, + 712 + ], + "score": 1.0, + "content": "Alexis Conneau, Ruty Rinott, Guillaume Lample, Adina Williams, Samuel Bowman, Holger", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Schwenk, and Veselin Stoyanov. Xnli: Evaluating cross-lingual sentence representations. In", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 720, + 462, + 734 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 462, + 734 + ], + "score": 1.0, + "content": "Empirical Methods in Natural Language Processing (EMNLP), pp. 2475–2485, 2018.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 116, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 116, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 116, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "Sadigh, Shiori Sagawa, Keshav Santhanam, Andy Shih, Krishnan Srinivasan, Alex Tamkin, Ro-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "han Taori, Armin W. Thomas, Florian Tramr, Rose E. Wang, William Wang, Bohan Wu, Jiajun", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 103, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 103, + 505, + 117 + ], + "score": 1.0, + "content": "Wu, Yuhuai Wu, Sang Michael Xie, Michihiro Yasunaga, Jiaxuan You, Matei Zaharia, Michael", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 113, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 115, + 113, + 505, + 130 + ], + "score": 1.0, + "content": "Zhang, Tianyi Zhang, Xikun Zhang, Yuhui Zhang, Lucia Zheng, Kaitlyn Zhou, and Percy Liang.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 126, + 492, + 140 + ], + "spans": [ + { + "bbox": [ + 116, + 126, + 492, + 140 + ], + "score": 1.0, + "content": "On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258, 2021.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 115, + 83, + 505, + 140 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 146, + 505, + 180 + ], + "lines": [ + { + "bbox": [ + 106, + 146, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 505, + 158 + ], + "score": 1.0, + "content": "Daniel Borkan, Lucas Dixon, Jeffrey Sorensen, Nithum Thain, and Lucy Vasserman. Nuanced", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 157, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 115, + 157, + 505, + 169 + ], + "score": 1.0, + "content": "metrics for measuring unintended bias with real data for text classification. In World Wide Web", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 168, + 233, + 180 + ], + "spans": [ + { + "bbox": [ + 115, + 168, + 233, + 180 + ], + "score": 1.0, + "content": "(WWW), pp. 491–500, 2019.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 106, + 146, + 505, + 180 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 264 + ], + "lines": [ + { + "bbox": [ + 106, + 187, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 200 + ], + "score": 1.0, + "content": "Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhari-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 199, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 116, + 199, + 505, + 211 + ], + "score": 1.0, + "content": "wal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 209, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 115, + 209, + 505, + 222 + ], + "score": 1.0, + "content": "Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 219, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 115, + 219, + 505, + 234 + ], + "score": 1.0, + "content": "Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 114, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 114, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 114, + 242, + 506, + 256 + ], + "spans": [ + { + "bbox": [ + 114, + 242, + 506, + 256 + ], + "score": 1.0, + "content": "Ilya Sutskever, and Dario Amodei. Language models are few-shot learners. arXiv preprint", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 253, + 219, + 266 + ], + "spans": [ + { + "bbox": [ + 115, + 253, + 219, + 266 + ], + "score": 1.0, + "content": "arXiv:2005.14165, 2020.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 187, + 506, + 266 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 273, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 505, + 285 + ], + "score": 1.0, + "content": "Maxwell Burnette, Rob Kooper, J. D. Maloney, Gareth S. Rohde, Jeffrey A. Terstriep, Craig Willis,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 283, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 115, + 283, + 506, + 297 + ], + "score": 1.0, + "content": "Noah Fahlgren, Todd Mockler, Maria Newcomb, Vasit Sagan, Pedro Andrade-Sanchez, Nadia", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 294, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 115, + 294, + 506, + 308 + ], + "score": 1.0, + "content": "Shakoor, Paheding Sidike, Rick Ward, and David LeBauer. Terra-ref data processing infrastruc-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 306, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 116, + 306, + 506, + 319 + ], + "score": 1.0, + "content": "ture. In Proceedings of the Practice and Experience on Advanced Research Computing, PEARC", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 317, + 504, + 330 + ], + "spans": [ + { + "bbox": [ + 116, + 317, + 504, + 330 + ], + "score": 1.0, + "content": "’18, New York, NY, USA, 2018. Association for Computing Machinery. ISBN 9781450364461.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17, + "bbox_fs": [ + 106, + 273, + 506, + 330 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 336, + 503, + 359 + ], + "lines": [ + { + "bbox": [ + 106, + 336, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 505, + 349 + ], + "score": 1.0, + "content": "Aylin Caliskan, Joanna J Bryson, and Arvind Narayanan. Semantics derived automatically from", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 348, + 443, + 360 + ], + "spans": [ + { + "bbox": [ + 115, + 348, + 443, + 360 + ], + "score": 1.0, + "content": "language corpora contain human-like biases. Science, 356(6334):183–186, 2017.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 106, + 336, + 505, + 360 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 367, + 503, + 401 + ], + "lines": [ + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 505, + 380 + ], + "score": 1.0, + "content": "Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin. Un-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 379, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 505, + 391 + ], + "score": 1.0, + "content": "supervised learning of visual features by contrasting cluster assignments. In Advances in Neural", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 389, + 430, + 403 + ], + "spans": [ + { + "bbox": [ + 116, + 389, + 430, + 403 + ], + "score": 1.0, + "content": "Information Processing Systems (NeurIPS), volume 33, pp. 9912–9924, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 367, + 505, + 403 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 408, + 504, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 408, + 504, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 504, + 421 + ], + "score": 1.0, + "content": "Ilias Chalkidis, Manos Fergadiotis, Prodromos Malakasiotis, Nikolaos Aletras, and Ion Androut-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 116, + 420, + 506, + 433 + ], + "score": 1.0, + "content": "sopoulos. LEGAL-BERT:”preparing the muppets for court”. In Empirical Methods in Natural", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 431, + 339, + 443 + ], + "spans": [ + { + "bbox": [ + 116, + 431, + 339, + 443 + ], + "score": 1.0, + "content": "Language Processing (EMNLP), pp. 2898–2904, 2020.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 408, + 506, + 443 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 450, + 505, + 484 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 464 + ], + "score": 1.0, + "content": "Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 460, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 460, + 505, + 476 + ], + "score": 1.0, + "content": "contrastive learning of visual representations. In International Conference on Machine Learning", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 117, + 473, + 246, + 485 + ], + "spans": [ + { + "bbox": [ + 117, + 473, + 246, + 485 + ], + "score": 1.0, + "content": "(ICML), pp. 1597–1607, 2020a.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 106, + 450, + 505, + 485 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 492, + 505, + 526 + ], + "lines": [ + { + "bbox": [ + 106, + 492, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 505 + ], + "score": 1.0, + "content": "Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 502, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 115, + 502, + 506, + 517 + ], + "score": 1.0, + "content": "contrastive learning of visual representations. In International conference on machine learning,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 114, + 514, + 244, + 526 + ], + "spans": [ + { + "bbox": [ + 114, + 514, + 244, + 526 + ], + "score": 1.0, + "content": "pp. 1597–1607. PMLR, 2020b.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32, + "bbox_fs": [ + 106, + 492, + 506, + 526 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 534, + 504, + 557 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 547 + ], + "score": 1.0, + "content": "Gordon Christie, Neil Fendley, James Wilson, and Ryan Mukherjee. Functional map of the world.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 545, + 358, + 558 + ], + "spans": [ + { + "bbox": [ + 116, + 545, + 358, + 558 + ], + "score": 1.0, + "content": "In Computer Vision and Pattern Recognition (CVPR), 2018.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 106, + 533, + 505, + 558 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 504, + 587 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 578 + ], + "score": 1.0, + "content": "Ozan Ciga, Anne L Martel, and Tony Xu. Self supervised contrastive learning for digital histopathol-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 576, + 299, + 588 + ], + "spans": [ + { + "bbox": [ + 115, + 576, + 299, + 588 + ], + "score": 1.0, + "content": "ogy. arXiv preprint arXiv:2011.13971, 2020.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 564, + 505, + 588 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 505, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "score": 1.0, + "content": "Jonathan H Clark, Eunsol Choi, Michael Collins, Dan Garrette, Tom Kwiatkowski, Vitaly Nikolaev,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 606, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 115, + 606, + 505, + 619 + ], + "score": 1.0, + "content": "and Jennimaria Palomaki. Tydi qa: A benchmark for information-seeking question answering in", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 617, + 411, + 630 + ], + "spans": [ + { + "bbox": [ + 116, + 617, + 411, + 630 + ], + "score": 1.0, + "content": "typologically diverse languages. arXiv preprint arXiv:2003.05002, 2020.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 594, + 505, + 630 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 636, + 504, + 660 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "Elijah Cole, Xuan Yang, Kimberly Wilber, Oisin Mac Aodha, and Serge Belongie. When does", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 648, + 474, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 648, + 474, + 660 + ], + "score": 1.0, + "content": "contrastive visual representation learning work? arXiv preprint arXiv:2105.05837, 2021.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5, + "bbox_fs": [ + 106, + 637, + 505, + 660 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 667, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 505, + 681 + ], + "score": 1.0, + "content": "Alexis Conneau and Guillaume Lample. Cross-lingual language model pretraining. In Advances in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 678, + 412, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 678, + 412, + 691 + ], + "score": 1.0, + "content": "Neural Information Processing Systems (NeurIPS), pp. 7059–7069, 2019.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 106, + 667, + 505, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 697, + 505, + 712 + ], + "score": 1.0, + "content": "Alexis Conneau, Ruty Rinott, Guillaume Lample, Adina Williams, Samuel Bowman, Holger", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Schwenk, and Veselin Stoyanov. Xnli: Evaluating cross-lingual sentence representations. In", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 720, + 462, + 734 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 462, + 734 + ], + "score": 1.0, + "content": "Empirical Methods in Natural Language Processing (EMNLP), pp. 2475–2485, 2018.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46, + "bbox_fs": [ + 106, + 697, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "scene understanding. In Proceedings of the IEEE conference on computer vision and pattern", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 115, + 257, + 127 + ], + "spans": [ + { + "bbox": [ + 115, + 115, + 257, + 127 + ], + "score": 1.0, + "content": "recognition, pp. 3213–3223, 2016.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 106, + 133, + 504, + 167 + ], + "lines": [ + { + "bbox": [ + 106, + 134, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 106, + 134, + 505, + 146 + ], + "score": 1.0, + "content": "Ekin D Cubuk, Barret Zoph, Jonathon Shlens, and Quoc V Le. Randaugment: Practical automated", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 117, + 146, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 117, + 146, + 505, + 157 + ], + "score": 1.0, + "content": "data augmentation with a reduced search space. In Computer Vision and Pattern Recognition", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 156, + 232, + 168 + ], + "spans": [ + { + "bbox": [ + 116, + 156, + 232, + 168 + ], + "score": 1.0, + "content": "(CVPR), pp. 702–703, 2020.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 505, + 208 + ], + "lines": [ + { + "bbox": [ + 106, + 174, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 505, + 187 + ], + "score": 1.0, + "content": "Sajib Dasgupta and Vincent Ng. Mine the easy, classify the hard: a semi-supervised approach to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 185, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 115, + 185, + 505, + 199 + ], + "score": 1.0, + "content": "automatic sentiment classification. In Conference on Natural Language Processing (KONVENS),", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 197, + 196, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 197, + 196, + 208 + ], + "score": 1.0, + "content": "pp. 701–709, 2009.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 214, + 505, + 281 + ], + "lines": [ + { + "bbox": [ + 104, + 212, + 506, + 230 + ], + "spans": [ + { + "bbox": [ + 104, + 212, + 506, + 230 + ], + "score": 1.0, + "content": "Etienne David, Simon Madec, Pouria Sadeghi-Tehran, Helge Aasen, Bangyou Zheng, Shouyang", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 227, + 504, + 239 + ], + "spans": [ + { + "bbox": [ + 116, + 227, + 504, + 239 + ], + "score": 1.0, + "content": "Liu, Norbert Kirchgessner, Goro Ishikawa, Koichi Nagasawa, Minhajul A Badhon, Curtis Poz-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 238, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 116, + 238, + 505, + 249 + ], + "score": 1.0, + "content": "niak, Benoit de Solan, Andreas Hund, Scott C. Chapman, Frederic Baret, Ian Stavness, and Wei", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 115, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "Guo. Global wheat head detection (gwhd) dataset: a large and diverse dataset of high-resolution", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 258, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 114, + 258, + 506, + 273 + ], + "score": 1.0, + "content": "rgb-labelled images to develop and benchmark wheat head detection methods. Plant Phenomics,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 270, + 166, + 282 + ], + "spans": [ + { + "bbox": [ + 115, + 270, + 166, + 282 + ], + "score": 1.0, + "content": "2020, 2020.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 288, + 505, + 388 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "Etienne David, Mario Serouart, Daniel Smith, Simon Madec, Kaaviya Velumani, Shouyang Liu,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 299, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 116, + 299, + 505, + 313 + ], + "score": 1.0, + "content": "Xu Wang, Francisco Pinto, Shahameh Shafiee, Izzat S. A. Tahir, Hisashi Tsujimoto, Shuhei", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 311, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 116, + 311, + 505, + 324 + ], + "score": 1.0, + "content": "Nasuda, Bangyou Zheng, Norbert Kirchgessner, Helge Aasen, Andreas Hund, Pouria Sadhegi-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 321, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 115, + 321, + 505, + 334 + ], + "score": 1.0, + "content": "Tehran, Koichi Nagasawa, Goro Ishikawa, Sebastien Dandrifosse, Alexis Carlier, Benjamin Du- ´", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 333, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 116, + 333, + 505, + 345 + ], + "score": 1.0, + "content": "mont, Benoit Mercatoris, Byron Evers, Ken Kuroki, Haozhou Wang, Masanori Ishii, Minhajul A.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 343, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 115, + 343, + 505, + 357 + ], + "score": 1.0, + "content": "Badhon, Curtis Pozniak, David Shaner LeBauer, Morten Lillemo, Jesse Poland, Scott Chapman,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 355, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 116, + 355, + 505, + 366 + ], + "score": 1.0, + "content": "Benoit de Solan, Fred´ eric Baret, Ian Stavness, and Wei Guo. Global wheat head detection 2021: ´", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 365, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 115, + 365, + 506, + 378 + ], + "score": 1.0, + "content": "An improved dataset for benchmarking wheat head detection methods. Plant Phenomics, 2021,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 376, + 142, + 389 + ], + "spans": [ + { + "bbox": [ + 115, + 376, + 142, + 389 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 395, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 394, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 505, + 407 + ], + "score": 1.0, + "content": "Olivier Dehaene, Axel Camara, Olivier Moindrot, Axel de Lavergne, and Pierre Courtiol. Self-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 115, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "supervision closes the gap between weak and strong supervision in histology. arXiv preprint", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 417, + 219, + 428 + ], + "spans": [ + { + "bbox": [ + 115, + 417, + 219, + 428 + ], + "score": 1.0, + "content": "arXiv:2012.03583, 2020.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 505, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 450 + ], + "score": 1.0, + "content": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 116, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "bidirectional transformers for language understanding. In Association for Computational Linguis-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 458, + 252, + 470 + ], + "spans": [ + { + "bbox": [ + 115, + 458, + 252, + 470 + ], + "score": 1.0, + "content": "tics (ACL), pp. 4171–4186, 2019.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 476, + 504, + 499 + ], + "lines": [ + { + "bbox": [ + 106, + 476, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 489 + ], + "score": 1.0, + "content": "Terrance DeVries and Graham W Taylor. Improved regularization of convolutional neural networks", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 487, + 330, + 499 + ], + "spans": [ + { + "bbox": [ + 115, + 487, + 330, + 499 + ], + "score": 1.0, + "content": "with cutout. arXiv preprint arXiv:1708.04552, 2017.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "Josip Djolonga, Jessica Yung, Michael Tschannen, Rob Romijnders, Lucas Beyer, Alexander", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 517, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 116, + 517, + 505, + 529 + ], + "score": 1.0, + "content": "Kolesnikov, Joan Puigcerver, Matthias Minderer, Alexander D’Amour, Dan Moldovan, et al. On", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 527, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 115, + 527, + 506, + 541 + ], + "score": 1.0, + "content": "robustness and transferability of convolutional neural networks. arXiv preprint arXiv:2007.08558,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 538, + 142, + 552 + ], + "spans": [ + { + "bbox": [ + 115, + 538, + 142, + 552 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 105, + 557, + 504, + 581 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "Abhimanyu Dubey, Vignesh Ramanathan, Alex Pentland, and Dhruv Mahajan. Adaptive methods", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 569, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 116, + 569, + 505, + 582 + ], + "score": 1.0, + "content": "for real-world domain generalization. In Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 107, + 587, + 504, + 621 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 601 + ], + "score": 1.0, + "content": "Dumitru Erhan, Aaron Courville, Yoshua Bengio, and Pascal Vincent. Why does unsupervised pre-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 599, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 116, + 599, + 505, + 611 + ], + "score": 1.0, + "content": "training help deep learning? In Artificial Intelligence and Statistics (AISTATS), pp. 201–208,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 607, + 143, + 622 + ], + "spans": [ + { + "bbox": [ + 115, + 607, + 143, + 622 + ], + "score": 1.0, + "content": "2010.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 108, + 628, + 504, + 662 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, Francois", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 640, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 116, + 640, + 505, + 652 + ], + "score": 1.0, + "content": "Laviolette, Mario March, and Victor Lempitsky. Domain-adversarial training of neural networks.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 650, + 351, + 663 + ], + "spans": [ + { + "bbox": [ + 115, + 650, + 351, + 663 + ], + "score": 1.0, + "content": "Journal of Machine Learning Research (JMLR), 17, 2016.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 103, + 668, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 668, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 668, + 505, + 681 + ], + "score": 1.0, + "content": "Nikhil Garg, Londa Schiebinger, Dan Jurafsky, and James Zou. Word embeddings quantify 100", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 680, + 356, + 692 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 356, + 692 + ], + "score": 1.0, + "content": "years of gender and ethnic stereotypes. Science, 115, 2018.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Samuel Gehman, Suchin Gururangan, Maarten Sap, Yejin Choi, and Noah A Smith. Real-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "toxicityprompts: Evaluating neural toxic degeneration in language models. arXiv preprint", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 116, + 721, + 218, + 731 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 218, + 731 + ], + "score": 1.0, + "content": "arXiv:2009.11462, 2020.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Marius Cordts, Mohamed Omran, Sebastian Ramos, Timo Rehfeld, Markus Enzweiler, Rodrigo", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "Benenson, Uwe Franke, Stefan Roth, and Bernt Schiele. The cityscapes dataset for semantic urban", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "scene understanding. In Proceedings of the IEEE conference on computer vision and pattern", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 115, + 257, + 127 + ], + "spans": [ + { + "bbox": [ + 115, + 115, + 257, + 127 + ], + "score": 1.0, + "content": "recognition, pp. 3213–3223, 2016.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 82, + 505, + 127 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 133, + 504, + 167 + ], + "lines": [ + { + "bbox": [ + 106, + 134, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 106, + 134, + 505, + 146 + ], + "score": 1.0, + "content": "Ekin D Cubuk, Barret Zoph, Jonathon Shlens, and Quoc V Le. Randaugment: Practical automated", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 117, + 146, + 505, + 157 + ], + "spans": [ + { + "bbox": [ + 117, + 146, + 505, + 157 + ], + "score": 1.0, + "content": "data augmentation with a reduced search space. In Computer Vision and Pattern Recognition", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 156, + 232, + 168 + ], + "spans": [ + { + "bbox": [ + 116, + 156, + 232, + 168 + ], + "score": 1.0, + "content": "(CVPR), pp. 702–703, 2020.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 106, + 134, + 505, + 168 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 505, + 208 + ], + "lines": [ + { + "bbox": [ + 106, + 174, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 505, + 187 + ], + "score": 1.0, + "content": "Sajib Dasgupta and Vincent Ng. Mine the easy, classify the hard: a semi-supervised approach to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 185, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 115, + 185, + 505, + 199 + ], + "score": 1.0, + "content": "automatic sentiment classification. In Conference on Natural Language Processing (KONVENS),", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 197, + 196, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 197, + 196, + 208 + ], + "score": 1.0, + "content": "pp. 701–709, 2009.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 106, + 174, + 505, + 208 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 214, + 505, + 281 + ], + "lines": [ + { + "bbox": [ + 104, + 212, + 506, + 230 + ], + "spans": [ + { + "bbox": [ + 104, + 212, + 506, + 230 + ], + "score": 1.0, + "content": "Etienne David, Simon Madec, Pouria Sadeghi-Tehran, Helge Aasen, Bangyou Zheng, Shouyang", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 227, + 504, + 239 + ], + "spans": [ + { + "bbox": [ + 116, + 227, + 504, + 239 + ], + "score": 1.0, + "content": "Liu, Norbert Kirchgessner, Goro Ishikawa, Koichi Nagasawa, Minhajul A Badhon, Curtis Poz-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 238, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 116, + 238, + 505, + 249 + ], + "score": 1.0, + "content": "niak, Benoit de Solan, Andreas Hund, Scott C. Chapman, Frederic Baret, Ian Stavness, and Wei", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 115, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "Guo. Global wheat head detection (gwhd) dataset: a large and diverse dataset of high-resolution", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 258, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 114, + 258, + 506, + 273 + ], + "score": 1.0, + "content": "rgb-labelled images to develop and benchmark wheat head detection methods. Plant Phenomics,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 270, + 166, + 282 + ], + "spans": [ + { + "bbox": [ + 115, + 270, + 166, + 282 + ], + "score": 1.0, + "content": "2020, 2020.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5, + "bbox_fs": [ + 104, + 212, + 506, + 282 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 288, + 505, + 388 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "Etienne David, Mario Serouart, Daniel Smith, Simon Madec, Kaaviya Velumani, Shouyang Liu,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 299, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 116, + 299, + 505, + 313 + ], + "score": 1.0, + "content": "Xu Wang, Francisco Pinto, Shahameh Shafiee, Izzat S. A. Tahir, Hisashi Tsujimoto, Shuhei", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 311, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 116, + 311, + 505, + 324 + ], + "score": 1.0, + "content": "Nasuda, Bangyou Zheng, Norbert Kirchgessner, Helge Aasen, Andreas Hund, Pouria Sadhegi-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 321, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 115, + 321, + 505, + 334 + ], + "score": 1.0, + "content": "Tehran, Koichi Nagasawa, Goro Ishikawa, Sebastien Dandrifosse, Alexis Carlier, Benjamin Du- ´", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 333, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 116, + 333, + 505, + 345 + ], + "score": 1.0, + "content": "mont, Benoit Mercatoris, Byron Evers, Ken Kuroki, Haozhou Wang, Masanori Ishii, Minhajul A.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 343, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 115, + 343, + 505, + 357 + ], + "score": 1.0, + "content": "Badhon, Curtis Pozniak, David Shaner LeBauer, Morten Lillemo, Jesse Poland, Scott Chapman,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 355, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 116, + 355, + 505, + 366 + ], + "score": 1.0, + "content": "Benoit de Solan, Fred´ eric Baret, Ian Stavness, and Wei Guo. Global wheat head detection 2021: ´", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 365, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 115, + 365, + 506, + 378 + ], + "score": 1.0, + "content": "An improved dataset for benchmarking wheat head detection methods. Plant Phenomics, 2021,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 376, + 142, + 389 + ], + "spans": [ + { + "bbox": [ + 115, + 376, + 142, + 389 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 288, + 506, + 389 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 395, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 394, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 505, + 407 + ], + "score": 1.0, + "content": "Olivier Dehaene, Axel Camara, Olivier Moindrot, Axel de Lavergne, and Pierre Courtiol. Self-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 115, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "supervision closes the gap between weak and strong supervision in histology. arXiv preprint", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 417, + 219, + 428 + ], + "spans": [ + { + "bbox": [ + 115, + 417, + 219, + 428 + ], + "score": 1.0, + "content": "arXiv:2012.03583, 2020.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 394, + 505, + 428 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 505, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 450 + ], + "score": 1.0, + "content": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 116, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "bidirectional transformers for language understanding. In Association for Computational Linguis-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 458, + 252, + 470 + ], + "spans": [ + { + "bbox": [ + 115, + 458, + 252, + 470 + ], + "score": 1.0, + "content": "tics (ACL), pp. 4171–4186, 2019.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 434, + 505, + 470 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 476, + 504, + 499 + ], + "lines": [ + { + "bbox": [ + 106, + 476, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 489 + ], + "score": 1.0, + "content": "Terrance DeVries and Graham W Taylor. Improved regularization of convolutional neural networks", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 487, + 330, + 499 + ], + "spans": [ + { + "bbox": [ + 115, + 487, + 330, + 499 + ], + "score": 1.0, + "content": "with cutout. arXiv preprint arXiv:1708.04552, 2017.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 106, + 476, + 505, + 499 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "Josip Djolonga, Jessica Yung, Michael Tschannen, Rob Romijnders, Lucas Beyer, Alexander", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 517, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 116, + 517, + 505, + 529 + ], + "score": 1.0, + "content": "Kolesnikov, Joan Puigcerver, Matthias Minderer, Alexander D’Amour, Dan Moldovan, et al. On", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 527, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 115, + 527, + 506, + 541 + ], + "score": 1.0, + "content": "robustness and transferability of convolutional neural networks. arXiv preprint arXiv:2007.08558,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 538, + 142, + 552 + ], + "spans": [ + { + "bbox": [ + 115, + 538, + 142, + 552 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 506, + 506, + 552 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 557, + 504, + 581 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "Abhimanyu Dubey, Vignesh Ramanathan, Alex Pentland, and Dhruv Mahajan. Adaptive methods", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 569, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 116, + 569, + 505, + 582 + ], + "score": 1.0, + "content": "for real-world domain generalization. In Computer Vision and Pattern Recognition (CVPR), 2021.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 106, + 558, + 505, + 582 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 587, + 504, + 621 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 601 + ], + "score": 1.0, + "content": "Dumitru Erhan, Aaron Courville, Yoshua Bengio, and Pascal Vincent. Why does unsupervised pre-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 599, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 116, + 599, + 505, + 611 + ], + "score": 1.0, + "content": "training help deep learning? In Artificial Intelligence and Statistics (AISTATS), pp. 201–208,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 607, + 143, + 622 + ], + "spans": [ + { + "bbox": [ + 115, + 607, + 143, + 622 + ], + "score": 1.0, + "content": "2010.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 586, + 505, + 622 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 628, + 504, + 662 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, Francois", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 640, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 116, + 640, + 505, + 652 + ], + "score": 1.0, + "content": "Laviolette, Mario March, and Victor Lempitsky. Domain-adversarial training of neural networks.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 650, + 351, + 663 + ], + "spans": [ + { + "bbox": [ + 115, + 650, + 351, + 663 + ], + "score": 1.0, + "content": "Journal of Machine Learning Research (JMLR), 17, 2016.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 106, + 627, + 505, + 663 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 668, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 668, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 668, + 505, + 681 + ], + "score": 1.0, + "content": "Nikhil Garg, Londa Schiebinger, Dan Jurafsky, and James Zou. Word embeddings quantify 100", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 680, + 356, + 692 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 356, + 692 + ], + "score": 1.0, + "content": "years of gender and ethnic stereotypes. Science, 115, 2018.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 106, + 668, + 505, + 692 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Samuel Gehman, Suchin Gururangan, Maarten Sap, Yejin Choi, and Noah A Smith. Real-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "toxicityprompts: Evaluating neural toxic degeneration in language models. arXiv preprint", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 116, + 721, + 218, + 731 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 218, + 731 + ], + "score": 1.0, + "content": "arXiv:2009.11462, 2020.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48, + "bbox_fs": [ + 106, + 699, + 505, + 731 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Robert Geirhos, Jorn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, ¨", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "Matthias Bethge, and Felix A Wichmann. Shortcut learning in deep neural networks. Nature", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 104, + 231, + 117 + ], + "spans": [ + { + "bbox": [ + 116, + 104, + 231, + 117 + ], + "score": 1.0, + "content": "Machine Intelligence, 2020.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 504, + 157 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 505, + 138 + ], + "score": 1.0, + "content": "Spyros Gidaris, Praveer Singh, and Nikos Komodakis. Unsupervised representation learning by", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 115, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "predicting image rotations. In International Conference on Learning Representations (ICLR),", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 142, + 158 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 142, + 158 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 502, + 188 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "score": 1.0, + "content": "Xavier Glorot, Antoine Bordes, and Yoshua Bengio. Deep sparse rectifier neural networks. In", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 387, + 189 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 387, + 189 + ], + "score": 1.0, + "content": "Artificial Intelligence and Statistics (AISTATS), pp. 315–323, 2011.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 195, + 501, + 208 + ], + "lines": [ + { + "bbox": [ + 106, + 193, + 501, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 193, + 501, + 210 + ], + "score": 1.0, + "content": "Yves Grandvalet and Yoshua Bengio. Entropy regularization. In Semi-Supervised Learning, 2005.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 214, + 505, + 249 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 230 + ], + "score": 1.0, + "content": "Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 226, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 115, + 226, + 505, + 239 + ], + "score": 1.0, + "content": "Jianfeng Gao, and Hoifung Poon. Domain-specific language model pretraining for biomedical", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 237, + 396, + 250 + ], + "spans": [ + { + "bbox": [ + 115, + 237, + 396, + 250 + ], + "score": 1.0, + "content": "natural language processing. arXiv preprint arXiv:2007.15779, 2020.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 256, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 255, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 505, + 270 + ], + "score": 1.0, + "content": "Ishaan Gulrajani and David Lopez-Paz. In search of lost domain generalization. arXiv preprint", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 268, + 219, + 278 + ], + "spans": [ + { + "bbox": [ + 116, + 268, + 219, + 278 + ], + "score": 1.0, + "content": "arXiv:2007.01434, 2020.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 286, + 505, + 320 + ], + "lines": [ + { + "bbox": [ + 105, + 285, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 300 + ], + "score": 1.0, + "content": "Suchin Gururangan, Ana Marasovic, Swabha Swayamdipta, Kyle Lo, Iz Beltagy, Doug Downey, ´", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 297, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 115, + 297, + 506, + 311 + ], + "score": 1.0, + "content": "and Noah A Smith. Don’t stop pretraining: adapt language models to domains and tasks. arXiv", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 309, + 254, + 320 + ], + "spans": [ + { + "bbox": [ + 115, + 309, + 254, + 320 + ], + "score": 1.0, + "content": "preprint arXiv:2004.10964, 2020.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 105, + 327, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 327, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 341 + ], + "score": 1.0, + "content": "Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Ming-Wei Chang. Realm: Retrieval-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 340, + 339, + 352 + ], + "spans": [ + { + "bbox": [ + 115, + 340, + 339, + 352 + ], + "score": 1.0, + "content": "augmented language model pre-training. In icml, 2020.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 105, + 358, + 503, + 382 + ], + "lines": [ + { + "bbox": [ + 105, + 357, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 505, + 373 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 370, + 387, + 382 + ], + "spans": [ + { + "bbox": [ + 115, + 370, + 387, + 382 + ], + "score": 1.0, + "content": "nition. In Computer Vision and Pattern Recognition (CVPR), 2016.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 388, + 505, + 423 + ], + "lines": [ + { + "bbox": [ + 105, + 388, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 505, + 402 + ], + "score": 1.0, + "content": "Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for un-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 401, + 505, + 413 + ], + "spans": [ + { + "bbox": [ + 115, + 401, + 505, + 413 + ], + "score": 1.0, + "content": "supervised visual representation learning. In Computer Vision and Pattern Recognition (CVPR),", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 410, + 142, + 423 + ], + "spans": [ + { + "bbox": [ + 115, + 410, + 142, + 423 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 430, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 428, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 444 + ], + "score": 1.0, + "content": "Dan Hendrycks, Xiaoyuan Liu, Eric Wallace, Adam Dziedzic, Rishabh Krishnan, and Dawn Song.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 441, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 115, + 441, + 505, + 454 + ], + "score": 1.0, + "content": "Pretrained transformers improve out-of-distribution robustness. arXiv preprint arXiv:2004.06100,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 452, + 142, + 464 + ], + "spans": [ + { + "bbox": [ + 115, + 452, + 142, + 464 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 106, + 471, + 504, + 495 + ], + "lines": [ + { + "bbox": [ + 106, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "Geoffrey Hinton, Oriol Vinyals, and Jeffrey Dean. Distilling the knowledge in a neural network. In", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 483, + 391, + 496 + ], + "spans": [ + { + "bbox": [ + 116, + 483, + 391, + 496 + ], + "score": 1.0, + "content": "NIPS Deep Learning and Representation Learning Workshop, 2015.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 502, + 505, + 536 + ], + "lines": [ + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "score": 1.0, + "content": "Judy Hoffman, Eric Tzeng, Taesung Park, Jun-Yan Zhu, Phillip Isola, Kate Saenko, Alexei A. Efros,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 116, + 513, + 505, + 525 + ], + "score": 1.0, + "content": "and Trevor Darrell. Cycada: Cycle consistent adversarial domain adaptation. In International", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 117, + 524, + 313, + 537 + ], + "spans": [ + { + "bbox": [ + 117, + 524, + 313, + 537 + ], + "score": 1.0, + "content": "Conference on Machine Learning (ICML), 2018.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 543, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 557 + ], + "score": 1.0, + "content": "Junjie Hu, Sebastian Ruder, Aditya Siddhant, Graham Neubig, Orhan Firat, and Melvin Johnson.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 555, + 504, + 568 + ], + "spans": [ + { + "bbox": [ + 116, + 555, + 504, + 568 + ], + "score": 1.0, + "content": "Xtreme: A massively multilingual multi-task benchmark for evaluating cross-lingual generaliza-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 566, + 304, + 578 + ], + "spans": [ + { + "bbox": [ + 115, + 566, + 304, + 578 + ], + "score": 1.0, + "content": "tion. arXiv preprint arXiv:2003.11080, 2020a.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 108, + 585, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 584, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 505, + 599 + ], + "score": 1.0, + "content": "Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 115, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "and Jure Leskovec. Open Graph Benchmark: Datasets for machine learning on graphs. In Ad-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 608, + 392, + 619 + ], + "spans": [ + { + "bbox": [ + 116, + 608, + 392, + 619 + ], + "score": 1.0, + "content": "vances in Neural Information Processing Systems (NeurIPS), 2020b.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 109, + 626, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 107, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 107, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "Weihua Hu, Bowen Liu, Joseph Gomes, Marinka Zitnik, Percy Liang, Vijay Pande, and Jure", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 115, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "Leskovec. Strategies for pre-training graph neural networks. In International Conference on", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 117, + 649, + 285, + 661 + ], + "spans": [ + { + "bbox": [ + 117, + 649, + 285, + 661 + ], + "score": 1.0, + "content": "Learning Representations (ICLR), 2020c.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 108, + 668, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 681 + ], + "score": 1.0, + "content": "Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. Densely connected", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 678, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 115, + 678, + 506, + 692 + ], + "score": 1.0, + "content": "convolutional networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 690, + 259, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 259, + 703 + ], + "score": 1.0, + "content": "Recognition, pp. 4700–4708, 2017.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Jonathan J. Hull. A database for handwritten text recognition research. IEEE Transactions on", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 721, + 378, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 721, + 378, + 732 + ], + "score": 1.0, + "content": "pattern analysis and machine intelligence, 16(5):550–554, 1994.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Robert Geirhos, Jorn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, ¨", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "Matthias Bethge, and Felix A Wichmann. Shortcut learning in deep neural networks. Nature", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 104, + 231, + 117 + ], + "spans": [ + { + "bbox": [ + 116, + 104, + 231, + 117 + ], + "score": 1.0, + "content": "Machine Intelligence, 2020.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 505, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 504, + 157 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 505, + 138 + ], + "score": 1.0, + "content": "Spyros Gidaris, Praveer Singh, and Nikos Komodakis. Unsupervised representation learning by", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 115, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "predicting image rotations. In International Conference on Learning Representations (ICLR),", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 142, + 158 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 142, + 158 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 122, + 505, + 158 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 502, + 188 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 178 + ], + "score": 1.0, + "content": "Xavier Glorot, Antoine Bordes, and Yoshua Bengio. Deep sparse rectifier neural networks. In", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 387, + 189 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 387, + 189 + ], + "score": 1.0, + "content": "Artificial Intelligence and Statistics (AISTATS), pp. 315–323, 2011.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 164, + 505, + 189 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 195, + 501, + 208 + ], + "lines": [ + { + "bbox": [ + 106, + 193, + 501, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 193, + 501, + 210 + ], + "score": 1.0, + "content": "Yves Grandvalet and Yoshua Bengio. Entropy regularization. In Semi-Supervised Learning, 2005.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 106, + 193, + 501, + 210 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 214, + 505, + 249 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 230 + ], + "score": 1.0, + "content": "Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 226, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 115, + 226, + 505, + 239 + ], + "score": 1.0, + "content": "Jianfeng Gao, and Hoifung Poon. Domain-specific language model pretraining for biomedical", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 237, + 396, + 250 + ], + "spans": [ + { + "bbox": [ + 115, + 237, + 396, + 250 + ], + "score": 1.0, + "content": "natural language processing. arXiv preprint arXiv:2007.15779, 2020.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 213, + 505, + 250 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 256, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 255, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 505, + 270 + ], + "score": 1.0, + "content": "Ishaan Gulrajani and David Lopez-Paz. In search of lost domain generalization. arXiv preprint", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 268, + 219, + 278 + ], + "spans": [ + { + "bbox": [ + 116, + 268, + 219, + 278 + ], + "score": 1.0, + "content": "arXiv:2007.01434, 2020.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 255, + 505, + 278 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 286, + 505, + 320 + ], + "lines": [ + { + "bbox": [ + 105, + 285, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 300 + ], + "score": 1.0, + "content": "Suchin Gururangan, Ana Marasovic, Swabha Swayamdipta, Kyle Lo, Iz Beltagy, Doug Downey, ´", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 297, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 115, + 297, + 506, + 311 + ], + "score": 1.0, + "content": "and Noah A Smith. Don’t stop pretraining: adapt language models to domains and tasks. arXiv", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 309, + 254, + 320 + ], + "spans": [ + { + "bbox": [ + 115, + 309, + 254, + 320 + ], + "score": 1.0, + "content": "preprint arXiv:2004.10964, 2020.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 285, + 506, + 320 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 327, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 327, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 341 + ], + "score": 1.0, + "content": "Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Ming-Wei Chang. Realm: Retrieval-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 340, + 339, + 352 + ], + "spans": [ + { + "bbox": [ + 115, + 340, + 339, + 352 + ], + "score": 1.0, + "content": "augmented language model pre-training. In icml, 2020.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 327, + 505, + 352 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 358, + 503, + 382 + ], + "lines": [ + { + "bbox": [ + 105, + 357, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 505, + 373 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 370, + 387, + 382 + ], + "spans": [ + { + "bbox": [ + 115, + 370, + 387, + 382 + ], + "score": 1.0, + "content": "nition. In Computer Vision and Pattern Recognition (CVPR), 2016.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 357, + 505, + 382 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 388, + 505, + 423 + ], + "lines": [ + { + "bbox": [ + 105, + 388, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 505, + 402 + ], + "score": 1.0, + "content": "Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for un-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 401, + 505, + 413 + ], + "spans": [ + { + "bbox": [ + 115, + 401, + 505, + 413 + ], + "score": 1.0, + "content": "supervised visual representation learning. In Computer Vision and Pattern Recognition (CVPR),", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 410, + 142, + 423 + ], + "spans": [ + { + "bbox": [ + 115, + 410, + 142, + 423 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 388, + 505, + 423 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 430, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 428, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 444 + ], + "score": 1.0, + "content": "Dan Hendrycks, Xiaoyuan Liu, Eric Wallace, Adam Dziedzic, Rishabh Krishnan, and Dawn Song.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 441, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 115, + 441, + 505, + 454 + ], + "score": 1.0, + "content": "Pretrained transformers improve out-of-distribution robustness. arXiv preprint arXiv:2004.06100,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 452, + 142, + 464 + ], + "spans": [ + { + "bbox": [ + 115, + 452, + 142, + 464 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 428, + 505, + 464 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 471, + 504, + 495 + ], + "lines": [ + { + "bbox": [ + 106, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "Geoffrey Hinton, Oriol Vinyals, and Jeffrey Dean. Distilling the knowledge in a neural network. In", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 483, + 391, + 496 + ], + "spans": [ + { + "bbox": [ + 116, + 483, + 391, + 496 + ], + "score": 1.0, + "content": "NIPS Deep Learning and Representation Learning Workshop, 2015.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 471, + 505, + 496 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 502, + 505, + 536 + ], + "lines": [ + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 515 + ], + "score": 1.0, + "content": "Judy Hoffman, Eric Tzeng, Taesung Park, Jun-Yan Zhu, Phillip Isola, Kate Saenko, Alexei A. Efros,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 116, + 513, + 505, + 525 + ], + "score": 1.0, + "content": "and Trevor Darrell. Cycada: Cycle consistent adversarial domain adaptation. In International", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 117, + 524, + 313, + 537 + ], + "spans": [ + { + "bbox": [ + 117, + 524, + 313, + 537 + ], + "score": 1.0, + "content": "Conference on Machine Learning (ICML), 2018.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 502, + 505, + 537 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 543, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 557 + ], + "score": 1.0, + "content": "Junjie Hu, Sebastian Ruder, Aditya Siddhant, Graham Neubig, Orhan Firat, and Melvin Johnson.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 555, + 504, + 568 + ], + "spans": [ + { + "bbox": [ + 116, + 555, + 504, + 568 + ], + "score": 1.0, + "content": "Xtreme: A massively multilingual multi-task benchmark for evaluating cross-lingual generaliza-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 566, + 304, + 578 + ], + "spans": [ + { + "bbox": [ + 115, + 566, + 304, + 578 + ], + "score": 1.0, + "content": "tion. arXiv preprint arXiv:2003.11080, 2020a.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 543, + 505, + 578 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 585, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 584, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 505, + 599 + ], + "score": 1.0, + "content": "Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 115, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "and Jure Leskovec. Open Graph Benchmark: Datasets for machine learning on graphs. In Ad-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 608, + 392, + 619 + ], + "spans": [ + { + "bbox": [ + 116, + 608, + 392, + 619 + ], + "score": 1.0, + "content": "vances in Neural Information Processing Systems (NeurIPS), 2020b.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 584, + 505, + 619 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 626, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 107, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 107, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "Weihua Hu, Bowen Liu, Joseph Gomes, Marinka Zitnik, Percy Liang, Vijay Pande, and Jure", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 115, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "Leskovec. Strategies for pre-training graph neural networks. In International Conference on", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 117, + 649, + 285, + 661 + ], + "spans": [ + { + "bbox": [ + 117, + 649, + 285, + 661 + ], + "score": 1.0, + "content": "Learning Representations (ICLR), 2020c.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 107, + 626, + 505, + 661 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 668, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 681 + ], + "score": 1.0, + "content": "Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. Densely connected", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 678, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 115, + 678, + 506, + 692 + ], + "score": 1.0, + "content": "convolutional networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 690, + 259, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 259, + 703 + ], + "score": 1.0, + "content": "Recognition, pp. 4700–4708, 2017.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 667, + 506, + 703 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Jonathan J. Hull. A database for handwritten text recognition research. IEEE Transactions on", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 721, + 378, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 721, + 378, + 732 + ], + "score": 1.0, + "content": "pattern analysis and machine intelligence, 16(5):550–554, 1994.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5, + "bbox_fs": [ + 106, + 709, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Neal Jean, Marshall Burke, Michael Xie, W. Matthew Davis, David B. Lobell, and Stefano Ermon.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 477, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 477, + 106 + ], + "score": 1.0, + "content": "Combining satellite imagery and machine learning to predict poverty. Science, 353, 2016.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 108, + 112, + 503, + 146 + ], + "lines": [ + { + "bbox": [ + 105, + 111, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 505, + 126 + ], + "score": 1.0, + "content": "Neal Jean, Sang Michael Xie, and Stefano Ermon. Semi-supervised deep kernel learning: Regres-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 505, + 135 + ], + "score": 1.0, + "content": "sion with unlabeled data by minimizing predictive variance. In Advances in Neural Information", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 134, + 268, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 134, + 268, + 146 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), 2018.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 105, + 153, + 503, + 176 + ], + "lines": [ + { + "bbox": [ + 104, + 150, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 104, + 150, + 505, + 167 + ], + "score": 1.0, + "content": "Junguang Jiang, Baixu Chen, Bo Fu, and Mingsheng Long. Transfer learning library. https:", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 117, + 163, + 411, + 176 + ], + "spans": [ + { + "bbox": [ + 117, + 163, + 411, + 176 + ], + "score": 1.0, + "content": "//github.com/thuml/Transfer-Learning-Library, 2020.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 105, + 182, + 504, + 205 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "score": 1.0, + "content": "Amita Kamath, Robin Jia, and Percy Liang. Selective question answering under domain shift. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 194, + 342, + 206 + ], + "spans": [ + { + "bbox": [ + 115, + 194, + 342, + 206 + ], + "score": 1.0, + "content": "Association for Computational Linguistics (ACL), 2020.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 108, + 212, + 504, + 246 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 506, + 225 + ], + "score": 1.0, + "content": "Benjamin Kellenberger, Diego Marcos, Sylvain Lobry, and Devis Tuia. Half a percent of labels is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 224, + 506, + 236 + ], + "spans": [ + { + "bbox": [ + 115, + 224, + 506, + 236 + ], + "score": 1.0, + "content": "enough: Efficient animal detection in uav imagery using deep cnns and active learning. IEEE", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 234, + 421, + 247 + ], + "spans": [ + { + "bbox": [ + 115, + 234, + 421, + 247 + ], + "score": 1.0, + "content": "Transactions on Geoscience and Remote Sensing, 57(12):9524–9533, 2019.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 253, + 505, + 320 + ], + "lines": [ + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "score": 1.0, + "content": "Pang Wei Koh, Shiori Sagawa, Henrik Marklund, Sang Michael Xie, Marvin Zhang, Akshay Balsub-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 263, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 115, + 263, + 505, + 277 + ], + "score": 1.0, + "content": "ramani, Weihua Hu, Michihiro Yasunaga, Richard Lanas Phillips, Irena Gao, Tony Lee, Etienne", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 273, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 115, + 273, + 505, + 289 + ], + "score": 1.0, + "content": "David, Ian Stavness, Wei Guo, Berton A. Earnshaw, Imran S. Haque, Sara Beery, Jure Leskovec,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 286, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 116, + 286, + 505, + 299 + ], + "score": 1.0, + "content": "Anshul Kundaje, Emma Pierson, Sergey Levine, Chelsea Finn, and Percy Liang. WILDS: A", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 296, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 115, + 296, + 506, + 311 + ], + "score": 1.0, + "content": "benchmark of in-the-wild distribution shifts. In International Conference on Machine Learning", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 308, + 177, + 321 + ], + "spans": [ + { + "bbox": [ + 115, + 308, + 177, + 321 + ], + "score": 1.0, + "content": "(ICML), 2021.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 326, + 503, + 350 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 504, + 340 + ], + "score": 1.0, + "content": "Daisuke Komura and Shumpei Ishikawa. Machine learning methods for histopathological image", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 338, + 437, + 351 + ], + "spans": [ + { + "bbox": [ + 115, + 338, + 437, + 351 + ], + "score": 1.0, + "content": "analysis. Computational and structural biotechnology journal, 16:34–42, 2018.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 108, + 356, + 504, + 390 + ], + "lines": [ + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "score": 1.0, + "content": "Navid Alemi Koohbanani, Balagopal Unnikrishnan, Syed Ali Khurram, Pavitra Krishnaswamy, and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 368, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 116, + 368, + 505, + 380 + ], + "score": 1.0, + "content": "Nasir Rajpoot. Self-path: Self-supervision for classification of pathology images with limited", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 379, + 367, + 391 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 367, + 391 + ], + "score": 1.0, + "content": "annotations. IEEE Transactions on Medical Imaging, 1, 2021.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 397, + 368, + 409 + ], + "lines": [ + { + "bbox": [ + 106, + 396, + 368, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 368, + 410 + ], + "score": 1.0, + "content": "Greg Landrum et al. Rdkit: Open-source cheminformatics, 2006.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 108, + 416, + 504, + 439 + ], + "lines": [ + { + "bbox": [ + 107, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 107, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to ´", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 427, + 417, + 440 + ], + "spans": [ + { + "bbox": [ + 116, + 427, + 417, + 440 + ], + "score": 1.0, + "content": "document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 106, + 445, + 504, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 444, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 506, + 460 + ], + "score": 1.0, + "content": "Dong-Hyun Lee. Pseudo-label: The simple and efficient semi-supervised learning method for deep", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 457, + 464, + 469 + ], + "spans": [ + { + "bbox": [ + 115, + 457, + 464, + 469 + ], + "score": 1.0, + "content": "neural networks. In ICML Workshop on Challenges in Representation Learning, 2013.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 475, + 504, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 475, + 504, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 504, + 489 + ], + "score": 1.0, + "content": "Jason D Lee, Qi Lei, Nikunj Saunshi, and Jiacheng Zhuo. Predicting what you already know helps:", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 487, + 422, + 499 + ], + "spans": [ + { + "bbox": [ + 115, + 487, + 422, + 499 + ], + "score": 1.0, + "content": "Provable self-supervised learning. arXiv preprint arXiv:2008.01064, 2020a.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 504, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 505, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 506, + 519 + ], + "score": 1.0, + "content": "Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jae-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 516, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 116, + 516, + 506, + 530 + ], + "score": 1.0, + "content": "woo Kang. Biobert: a pre-trained biomedical language representation model for biomedical text", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 528, + 318, + 540 + ], + "spans": [ + { + "bbox": [ + 116, + 528, + 318, + 540 + ], + "score": 1.0, + "content": "mining. Bioinformatics, 36(4):1234–1240, 2020b.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 546, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 506, + 559 + ], + "score": 1.0, + "content": "Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 114, + 557, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 114, + 557, + 505, + 571 + ], + "score": 1.0, + "content": "Levy, Ves Stoyanov, and Luke Zettlemoyer. Bart: Denoising sequence-to-sequence pre-training", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 568, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 115, + 568, + 505, + 582 + ], + "score": 1.0, + "content": "for natural language generation, translation, and comprehension. In Association for Computa-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 579, + 243, + 592 + ], + "spans": [ + { + "bbox": [ + 115, + 579, + 243, + 592 + ], + "score": 1.0, + "content": "tional Linguistics (ACL), 2020.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 106, + 597, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 598, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 598, + 505, + 610 + ], + "score": 1.0, + "content": "Da Li, Yongxin Yang, Yi-Zhe Song, and Timothy M Hospedales. Deeper, broader and artier domain", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 608, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 115, + 608, + 506, + 624 + ], + "score": 1.0, + "content": "generalization. In Proceedings of the IEEE international conference on computer vision, pp.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 620, + 192, + 631 + ], + "spans": [ + { + "bbox": [ + 116, + 620, + 192, + 631 + ], + "score": 1.0, + "content": "5542–5550, 2017.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 639, + 504, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 653 + ], + "score": 1.0, + "content": "Shoushan Li, Zhongqing Wang, Guodong Zhou, and Sophia Yat Mei Lee. Semi-supervised learning", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 650, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 116, + 650, + 505, + 662 + ], + "score": 1.0, + "content": "for imbalanced sentiment classification. In International Joint Conference on Artificial Intelli-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 662, + 203, + 673 + ], + "spans": [ + { + "bbox": [ + 115, + 662, + 203, + 673 + ], + "score": 1.0, + "content": "gence (IJCAI), 2011.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 679, + 503, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 679, + 505, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 505, + 692 + ], + "score": 1.0, + "content": "Mingsheng Long, Yue Cao, Jianmin Wang, and Michael Jordan. Learning transferable features with", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 691, + 498, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 691, + 498, + 703 + ], + "score": 1.0, + "content": "deep adaptation networks. In International conference on machine learning, pp. 97–105, 2015.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Mingsheng Long, Han Zhu, Jianmin Wang, and Michael I Jordan. Deep transfer learning with joint", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 720, + 492, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 492, + 732 + ], + "score": 1.0, + "content": "adaptation networks. In International conference on machine learning, pp. 2208–2217, 2017.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Neal Jean, Marshall Burke, Michael Xie, W. Matthew Davis, David B. Lobell, and Stefano Ermon.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 477, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 477, + 106 + ], + "score": 1.0, + "content": "Combining satellite imagery and machine learning to predict poverty. Science, 353, 2016.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 505, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 112, + 503, + 146 + ], + "lines": [ + { + "bbox": [ + 105, + 111, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 505, + 126 + ], + "score": 1.0, + "content": "Neal Jean, Sang Michael Xie, and Stefano Ermon. Semi-supervised deep kernel learning: Regres-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 505, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 505, + 135 + ], + "score": 1.0, + "content": "sion with unlabeled data by minimizing predictive variance. In Advances in Neural Information", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 134, + 268, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 134, + 268, + 146 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), 2018.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 111, + 505, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 153, + 503, + 176 + ], + "lines": [ + { + "bbox": [ + 104, + 150, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 104, + 150, + 505, + 167 + ], + "score": 1.0, + "content": "Junguang Jiang, Baixu Chen, Bo Fu, and Mingsheng Long. Transfer learning library. https:", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 117, + 163, + 411, + 176 + ], + "spans": [ + { + "bbox": [ + 117, + 163, + 411, + 176 + ], + "score": 1.0, + "content": "//github.com/thuml/Transfer-Learning-Library, 2020.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 104, + 150, + 505, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 182, + 504, + 205 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "score": 1.0, + "content": "Amita Kamath, Robin Jia, and Percy Liang. Selective question answering under domain shift. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 194, + 342, + 206 + ], + "spans": [ + { + "bbox": [ + 115, + 194, + 342, + 206 + ], + "score": 1.0, + "content": "Association for Computational Linguistics (ACL), 2020.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 182, + 506, + 206 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 212, + 504, + 246 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 506, + 225 + ], + "score": 1.0, + "content": "Benjamin Kellenberger, Diego Marcos, Sylvain Lobry, and Devis Tuia. Half a percent of labels is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 224, + 506, + 236 + ], + "spans": [ + { + "bbox": [ + 115, + 224, + 506, + 236 + ], + "score": 1.0, + "content": "enough: Efficient animal detection in uav imagery using deep cnns and active learning. IEEE", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 234, + 421, + 247 + ], + "spans": [ + { + "bbox": [ + 115, + 234, + 421, + 247 + ], + "score": 1.0, + "content": "Transactions on Geoscience and Remote Sensing, 57(12):9524–9533, 2019.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 213, + 506, + 247 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 253, + 505, + 320 + ], + "lines": [ + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 266 + ], + "score": 1.0, + "content": "Pang Wei Koh, Shiori Sagawa, Henrik Marklund, Sang Michael Xie, Marvin Zhang, Akshay Balsub-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 263, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 115, + 263, + 505, + 277 + ], + "score": 1.0, + "content": "ramani, Weihua Hu, Michihiro Yasunaga, Richard Lanas Phillips, Irena Gao, Tony Lee, Etienne", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 273, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 115, + 273, + 505, + 289 + ], + "score": 1.0, + "content": "David, Ian Stavness, Wei Guo, Berton A. Earnshaw, Imran S. Haque, Sara Beery, Jure Leskovec,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 286, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 116, + 286, + 505, + 299 + ], + "score": 1.0, + "content": "Anshul Kundaje, Emma Pierson, Sergey Levine, Chelsea Finn, and Percy Liang. WILDS: A", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 296, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 115, + 296, + 506, + 311 + ], + "score": 1.0, + "content": "benchmark of in-the-wild distribution shifts. In International Conference on Machine Learning", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 308, + 177, + 321 + ], + "spans": [ + { + "bbox": [ + 115, + 308, + 177, + 321 + ], + "score": 1.0, + "content": "(ICML), 2021.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5, + "bbox_fs": [ + 106, + 253, + 506, + 321 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 326, + 503, + 350 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 504, + 340 + ], + "score": 1.0, + "content": "Daisuke Komura and Shumpei Ishikawa. Machine learning methods for histopathological image", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 338, + 437, + 351 + ], + "spans": [ + { + "bbox": [ + 115, + 338, + 437, + 351 + ], + "score": 1.0, + "content": "analysis. Computational and structural biotechnology journal, 16:34–42, 2018.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 106, + 326, + 504, + 351 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 356, + 504, + 390 + ], + "lines": [ + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 505, + 370 + ], + "score": 1.0, + "content": "Navid Alemi Koohbanani, Balagopal Unnikrishnan, Syed Ali Khurram, Pavitra Krishnaswamy, and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 368, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 116, + 368, + 505, + 380 + ], + "score": 1.0, + "content": "Nasir Rajpoot. Self-path: Self-supervision for classification of pathology images with limited", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 379, + 367, + 391 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 367, + 391 + ], + "score": 1.0, + "content": "annotations. IEEE Transactions on Medical Imaging, 1, 2021.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 356, + 505, + 391 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 397, + 368, + 409 + ], + "lines": [ + { + "bbox": [ + 106, + 396, + 368, + 410 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 368, + 410 + ], + "score": 1.0, + "content": "Greg Landrum et al. Rdkit: Open-source cheminformatics, 2006.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 396, + 368, + 410 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 416, + 504, + 439 + ], + "lines": [ + { + "bbox": [ + 107, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 107, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "Yann LeCun, Leon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to ´", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 427, + 417, + 440 + ], + "spans": [ + { + "bbox": [ + 116, + 427, + 417, + 440 + ], + "score": 1.0, + "content": "document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 107, + 416, + 505, + 440 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 445, + 504, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 444, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 506, + 460 + ], + "score": 1.0, + "content": "Dong-Hyun Lee. Pseudo-label: The simple and efficient semi-supervised learning method for deep", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 457, + 464, + 469 + ], + "spans": [ + { + "bbox": [ + 115, + 457, + 464, + 469 + ], + "score": 1.0, + "content": "neural networks. In ICML Workshop on Challenges in Representation Learning, 2013.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 444, + 506, + 469 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 475, + 504, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 475, + 504, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 504, + 489 + ], + "score": 1.0, + "content": "Jason D Lee, Qi Lei, Nikunj Saunshi, and Jiacheng Zhuo. Predicting what you already know helps:", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 487, + 422, + 499 + ], + "spans": [ + { + "bbox": [ + 115, + 487, + 422, + 499 + ], + "score": 1.0, + "content": "Provable self-supervised learning. arXiv preprint arXiv:2008.01064, 2020a.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 475, + 504, + 499 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 504, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 505, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 506, + 519 + ], + "score": 1.0, + "content": "Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jae-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 516, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 116, + 516, + 506, + 530 + ], + "score": 1.0, + "content": "woo Kang. Biobert: a pre-trained biomedical language representation model for biomedical text", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 528, + 318, + 540 + ], + "spans": [ + { + "bbox": [ + 116, + 528, + 318, + 540 + ], + "score": 1.0, + "content": "mining. Bioinformatics, 36(4):1234–1240, 2020b.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 505, + 506, + 540 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 546, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 506, + 559 + ], + "score": 1.0, + "content": "Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 114, + 557, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 114, + 557, + 505, + 571 + ], + "score": 1.0, + "content": "Levy, Ves Stoyanov, and Luke Zettlemoyer. Bart: Denoising sequence-to-sequence pre-training", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 568, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 115, + 568, + 505, + 582 + ], + "score": 1.0, + "content": "for natural language generation, translation, and comprehension. In Association for Computa-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 579, + 243, + 592 + ], + "spans": [ + { + "bbox": [ + 115, + 579, + 243, + 592 + ], + "score": 1.0, + "content": "tional Linguistics (ACL), 2020.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 546, + 506, + 592 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 597, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 106, + 598, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 598, + 505, + 610 + ], + "score": 1.0, + "content": "Da Li, Yongxin Yang, Yi-Zhe Song, and Timothy M Hospedales. Deeper, broader and artier domain", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 608, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 115, + 608, + 506, + 624 + ], + "score": 1.0, + "content": "generalization. In Proceedings of the IEEE international conference on computer vision, pp.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 620, + 192, + 631 + ], + "spans": [ + { + "bbox": [ + 116, + 620, + 192, + 631 + ], + "score": 1.0, + "content": "5542–5550, 2017.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38, + "bbox_fs": [ + 106, + 598, + 506, + 631 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 639, + 504, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 653 + ], + "score": 1.0, + "content": "Shoushan Li, Zhongqing Wang, Guodong Zhou, and Sophia Yat Mei Lee. Semi-supervised learning", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 650, + 505, + 662 + ], + "spans": [ + { + "bbox": [ + 116, + 650, + 505, + 662 + ], + "score": 1.0, + "content": "for imbalanced sentiment classification. In International Joint Conference on Artificial Intelli-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 662, + 203, + 673 + ], + "spans": [ + { + "bbox": [ + 115, + 662, + 203, + 673 + ], + "score": 1.0, + "content": "gence (IJCAI), 2011.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 637, + 506, + 673 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 679, + 503, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 679, + 505, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 505, + 692 + ], + "score": 1.0, + "content": "Mingsheng Long, Yue Cao, Jianmin Wang, and Michael Jordan. Learning transferable features with", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 691, + 498, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 691, + 498, + 703 + ], + "score": 1.0, + "content": "deep adaptation networks. In International conference on machine learning, pp. 97–105, 2015.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 679, + 505, + 703 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Mingsheng Long, Han Zhu, Jianmin Wang, and Michael I Jordan. Deep transfer learning with joint", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 720, + 492, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 492, + 732 + ], + "score": 1.0, + "content": "adaptation networks. In International conference on machine learning, pp. 2208–2217, 2017.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 106, + 709, + 504, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Mingsheng Long, Zhangjie Cao, Jianmin Wang, and Michael I Jordan. Conditional adversarial", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 489, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 489, + 106 + ], + "score": 1.0, + "content": "domain adaptation. In Advances in Neural Information Processing Systems (NeurIPS), 2018.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 111, + 504, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 111, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 505, + 125 + ], + "score": 1.0, + "content": "Ming Y Lu, Richard J Chen, Jingwen Wang, Debora Dillon, and Faisal Mahmood. Semi-supervised", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 122, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 115, + 122, + 505, + 137 + ], + "score": 1.0, + "content": "histology classification using deep multiple instance learning and contrastive predictive coding.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 135, + 279, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 135, + 279, + 146 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1910.10825, 2019.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 105, + 151, + 505, + 175 + ], + "lines": [ + { + "bbox": [ + 107, + 153, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 107, + 153, + 505, + 164 + ], + "score": 1.0, + "content": "Yishay Mansour, Mehryar Mohri, and Afshin Rostamizadeh. Domain adaptation with multiple", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 163, + 505, + 175 + ], + "spans": [ + { + "bbox": [ + 115, + 163, + 505, + 175 + ], + "score": 1.0, + "content": "sources. In Advances in Neural Information Processing Systems (NeurIPS), pp. 1041–1048, 2009.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 105, + 181, + 504, + 204 + ], + "lines": [ + { + "bbox": [ + 106, + 180, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 180, + 505, + 194 + ], + "score": 1.0, + "content": "John Miller, Karl Krauth, Benjamin Recht, and Ludwig Schmidt. The effect of natural distribution", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 192, + 425, + 205 + ], + "spans": [ + { + "bbox": [ + 116, + 192, + 425, + 205 + ], + "score": 1.0, + "content": "shift on question answering models. arXiv preprint arXiv:2004.14444, 2020.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 210, + 505, + 255 + ], + "lines": [ + { + "bbox": [ + 106, + 211, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 505, + 223 + ], + "score": 1.0, + "content": "John Miller, Rohan Taori, Aditi Raghunathan, Shiori Sagawa, Pang Wei Koh, Vaishaal Shankar,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 222, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 116, + 222, + 505, + 234 + ], + "score": 1.0, + "content": "Percy Liang, Yair Carmon, and Ludwig Schmidt. Accuracy on the line: on the strong correlation", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 232, + 506, + 246 + ], + "spans": [ + { + "bbox": [ + 115, + 232, + 506, + 246 + ], + "score": 1.0, + "content": "between out-of-distribution and in-distribution generalization. In International Conference on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 244, + 253, + 256 + ], + "spans": [ + { + "bbox": [ + 116, + 244, + 253, + 256 + ], + "score": 1.0, + "content": "Machine Learning (ICML), 2021.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 105, + 262, + 504, + 285 + ], + "lines": [ + { + "bbox": [ + 106, + 262, + 504, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 504, + 275 + ], + "score": 1.0, + "content": "Samarth Mishra, Kate Saenko, and Venkatesh Saligrama. Surprisingly simple semi-supervised do-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 273, + 480, + 285 + ], + "spans": [ + { + "bbox": [ + 115, + 273, + 480, + 285 + ], + "score": 1.0, + "content": "main adaptation with pretraining and consistency. arXiv preprint arXiv:2101.12727, 2021.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 105, + 291, + 504, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "Moin Nadeem, Anna Bethke, and Siva Reddy. Stereoset: Measuring stereotypical bias in pretrained", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 303, + 352, + 315 + ], + "spans": [ + { + "bbox": [ + 115, + 303, + 352, + 315 + ], + "score": 1.0, + "content": "language models. arXiv preprint arXiv:2004.09456, 2020.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 108, + 321, + 506, + 453 + ], + "lines": [ + { + "bbox": [ + 106, + 321, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 505, + 333 + ], + "score": 1.0, + "content": "Wilhelmina Nekoto, Vukosi Marivate, Tshinondiwa Matsila, Timi Fasubaa, Tajudeen Kola-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 332, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 116, + 332, + 505, + 344 + ], + "score": 1.0, + "content": "wole, Taiwo Fagbohungbe, Solomon Oluwole Akinola, Shamsuddee Hassan Muhammad, Sa-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 114, + 342, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 114, + 342, + 506, + 356 + ], + "score": 1.0, + "content": "lomon Kabongo, Salomey Osei, Sackey Freshia, Rubungo Andre Niyongabo, Ricky Macharm,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 114, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 114, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "Perez Ogayo, Orevaoghene Ahia, Musie Meressa, Mofe Adeyemi, Masabata Mokgesi-Selinga,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 114, + 364, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 114, + 364, + 506, + 378 + ], + "score": 1.0, + "content": "Lawrence Okegbemi, Laura Jane Martinus, Kolawole Tajudeen, Kevin Degila, Kelechi Ogueji,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 375, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 115, + 375, + 506, + 389 + ], + "score": 1.0, + "content": "Kathleen Siminyu, Julia Kreutzer, Jason Webster, Jamiil Toure Ali, Jade Abbott, Iroro Orife,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 386, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 115, + 386, + 505, + 399 + ], + "score": 1.0, + "content": "Ignatius Ezeani, Idris Abdulkabir Dangana, Herman Kamper, Hady Elsahar, Goodness Duru,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 397, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 115, + 397, + 505, + 410 + ], + "score": 1.0, + "content": "Ghollah Kioko, Espoir Murhabazi, Elan van Biljon, Daniel Whitenack, Christopher Onyefuluchi,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 408, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 115, + 408, + 505, + 421 + ], + "score": 1.0, + "content": "Chris Emezue, Bonaventure Dossou, Blessing Sibanda, Blessing Itoro Bassey, Ayodele Olabiyi,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 418, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 115, + 418, + 506, + 433 + ], + "score": 1.0, + "content": "Arshath Ramkilowan, Alp ktem, Adewale Akinfaderin, and Abdallah Bashir. Participatory re-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 430, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 115, + 430, + 506, + 443 + ], + "score": 1.0, + "content": "search for low-resourced machine translation: A case study in African languages. In Findings of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 442, + 444, + 454 + ], + "spans": [ + { + "bbox": [ + 116, + 442, + 444, + 454 + ], + "score": 1.0, + "content": "Empirical Methods in Natural Language Processing (Findings of EMNLP), 2020.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 108, + 460, + 503, + 494 + ], + "lines": [ + { + "bbox": [ + 106, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 505, + 473 + ], + "score": 1.0, + "content": "Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu, and Andrew Y Ng. Reading", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 470, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 115, + 470, + 505, + 484 + ], + "score": 1.0, + "content": "digits in natural images with unsupervised feature learning. In NIPS Workshop on Deep Learning", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 482, + 291, + 495 + ], + "spans": [ + { + "bbox": [ + 116, + 482, + 291, + 495 + ], + "score": 1.0, + "content": "and Unsupervised Feature Learning, 2011.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 505, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "score": 1.0, + "content": "Jianmo Ni, Jiacheng Li, and Julian McAuley. Justifying recommendations using distantly-labeled re-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 510, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 115, + 510, + 505, + 525 + ], + "score": 1.0, + "content": "views and fine-grained aspects. In Empirical Methods in Natural Language Processing (EMNLP),", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 523, + 196, + 533 + ], + "spans": [ + { + "bbox": [ + 115, + 523, + 196, + 533 + ], + "score": 1.0, + "content": "pp. 188–197, 2019.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 541, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 506, + 553 + ], + "score": 1.0, + "content": "Mohammad Sadegh Norouzzadeh, Dan Morris, Sara Beery, Neel Joshi, Nebojsa Jojic, and Jeff", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 551, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 116, + 551, + 505, + 565 + ], + "score": 1.0, + "content": "Clune. A deep active learning system for species identification and counting in camera trap", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 563, + 383, + 575 + ], + "spans": [ + { + "bbox": [ + 116, + 563, + 383, + 575 + ], + "score": 1.0, + "content": "images. Methods in Ecology and Evolution, 12(1):150–161, 2021.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 106, + 581, + 343, + 593 + ], + "lines": [ + { + "bbox": [ + 106, + 580, + 344, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 344, + 594 + ], + "score": 1.0, + "content": "Jill Nugent. inaturalist. Science Scope, 41(7):12–13, 2018.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 106, + 599, + 504, + 622 + ], + "lines": [ + { + "bbox": [ + 107, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 107, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "Yonatan Oren, Shiori Sagawa, Tatsunori Hashimoto, and Percy Liang. Distributionally robust lan-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 610, + 479, + 624 + ], + "spans": [ + { + "bbox": [ + 115, + 610, + 479, + 624 + ], + "score": 1.0, + "content": "guage modeling. In Empirical Methods in Natural Language Processing (EMNLP), 2019.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 105, + 628, + 503, + 651 + ], + "lines": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "Omiros Pantazis, Gabriel Brostow, Kate Jones, and Oisin Mac Aodha. Focus on the positives: Self-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 640, + 473, + 652 + ], + "spans": [ + { + "bbox": [ + 115, + 640, + 473, + 652 + ], + "score": 1.0, + "content": "supervised learning for biodiversity monitoring. arXiv preprint arXiv:2108.06435, 2021.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 106, + 658, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 504, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 504, + 669 + ], + "score": 1.0, + "content": "Mohammad Peikari, Sherine Salama, Sharon Nofech-Mozes, and Anne L Martel. A cluster-then-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 668, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 115, + 668, + 505, + 682 + ], + "score": 1.0, + "content": "label semi-supervised learning approach for pathology image classification. Scientific reports, 8", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 680, + 181, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 181, + 691 + ], + "score": 1.0, + "content": "(1):1–13, 2018.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 504, + 711 + ], + "score": 1.0, + "content": "Xingchao Peng, Ben Usman, Neela Kaushik, Dequan Wang, Judy Hoffman, and Kate Saenko.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Visda: A synthetic-to-real benchmark for visual domain adaptation. In Computer Vision and", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 116, + 720, + 325, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 325, + 732 + ], + "score": 1.0, + "content": "Pattern Recognition (CVPR), pp. 2021–2026, 2018.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47 + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "17", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Mingsheng Long, Zhangjie Cao, Jianmin Wang, and Michael I Jordan. Conditional adversarial", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 489, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 489, + 106 + ], + "score": 1.0, + "content": "domain adaptation. In Advances in Neural Information Processing Systems (NeurIPS), 2018.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 505, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 111, + 504, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 111, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 505, + 125 + ], + "score": 1.0, + "content": "Ming Y Lu, Richard J Chen, Jingwen Wang, Debora Dillon, and Faisal Mahmood. Semi-supervised", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 122, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 115, + 122, + 505, + 137 + ], + "score": 1.0, + "content": "histology classification using deep multiple instance learning and contrastive predictive coding.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 135, + 279, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 135, + 279, + 146 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1910.10825, 2019.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 111, + 505, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 151, + 505, + 175 + ], + "lines": [ + { + "bbox": [ + 107, + 153, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 107, + 153, + 505, + 164 + ], + "score": 1.0, + "content": "Yishay Mansour, Mehryar Mohri, and Afshin Rostamizadeh. Domain adaptation with multiple", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 163, + 505, + 175 + ], + "spans": [ + { + "bbox": [ + 115, + 163, + 505, + 175 + ], + "score": 1.0, + "content": "sources. In Advances in Neural Information Processing Systems (NeurIPS), pp. 1041–1048, 2009.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 107, + 153, + 505, + 175 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 181, + 504, + 204 + ], + "lines": [ + { + "bbox": [ + 106, + 180, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 180, + 505, + 194 + ], + "score": 1.0, + "content": "John Miller, Karl Krauth, Benjamin Recht, and Ludwig Schmidt. The effect of natural distribution", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 192, + 425, + 205 + ], + "spans": [ + { + "bbox": [ + 116, + 192, + 425, + 205 + ], + "score": 1.0, + "content": "shift on question answering models. arXiv preprint arXiv:2004.14444, 2020.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 106, + 180, + 505, + 205 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 210, + 505, + 255 + ], + "lines": [ + { + "bbox": [ + 106, + 211, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 505, + 223 + ], + "score": 1.0, + "content": "John Miller, Rohan Taori, Aditi Raghunathan, Shiori Sagawa, Pang Wei Koh, Vaishaal Shankar,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 222, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 116, + 222, + 505, + 234 + ], + "score": 1.0, + "content": "Percy Liang, Yair Carmon, and Ludwig Schmidt. Accuracy on the line: on the strong correlation", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 232, + 506, + 246 + ], + "spans": [ + { + "bbox": [ + 115, + 232, + 506, + 246 + ], + "score": 1.0, + "content": "between out-of-distribution and in-distribution generalization. In International Conference on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 244, + 253, + 256 + ], + "spans": [ + { + "bbox": [ + 116, + 244, + 253, + 256 + ], + "score": 1.0, + "content": "Machine Learning (ICML), 2021.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 106, + 211, + 506, + 256 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 262, + 504, + 285 + ], + "lines": [ + { + "bbox": [ + 106, + 262, + 504, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 504, + 275 + ], + "score": 1.0, + "content": "Samarth Mishra, Kate Saenko, and Venkatesh Saligrama. Surprisingly simple semi-supervised do-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 273, + 480, + 285 + ], + "spans": [ + { + "bbox": [ + 115, + 273, + 480, + 285 + ], + "score": 1.0, + "content": "main adaptation with pretraining and consistency. arXiv preprint arXiv:2101.12727, 2021.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 106, + 262, + 504, + 285 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 291, + 504, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "Moin Nadeem, Anna Bethke, and Siva Reddy. Stereoset: Measuring stereotypical bias in pretrained", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 303, + 352, + 315 + ], + "spans": [ + { + "bbox": [ + 115, + 303, + 352, + 315 + ], + "score": 1.0, + "content": "language models. arXiv preprint arXiv:2004.09456, 2020.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 291, + 505, + 315 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 321, + 506, + 453 + ], + "lines": [ + { + "bbox": [ + 106, + 321, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 505, + 333 + ], + "score": 1.0, + "content": "Wilhelmina Nekoto, Vukosi Marivate, Tshinondiwa Matsila, Timi Fasubaa, Tajudeen Kola-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 332, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 116, + 332, + 505, + 344 + ], + "score": 1.0, + "content": "wole, Taiwo Fagbohungbe, Solomon Oluwole Akinola, Shamsuddee Hassan Muhammad, Sa-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 114, + 342, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 114, + 342, + 506, + 356 + ], + "score": 1.0, + "content": "lomon Kabongo, Salomey Osei, Sackey Freshia, Rubungo Andre Niyongabo, Ricky Macharm,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 114, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 114, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "Perez Ogayo, Orevaoghene Ahia, Musie Meressa, Mofe Adeyemi, Masabata Mokgesi-Selinga,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 114, + 364, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 114, + 364, + 506, + 378 + ], + "score": 1.0, + "content": "Lawrence Okegbemi, Laura Jane Martinus, Kolawole Tajudeen, Kevin Degila, Kelechi Ogueji,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 375, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 115, + 375, + 506, + 389 + ], + "score": 1.0, + "content": "Kathleen Siminyu, Julia Kreutzer, Jason Webster, Jamiil Toure Ali, Jade Abbott, Iroro Orife,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 386, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 115, + 386, + 505, + 399 + ], + "score": 1.0, + "content": "Ignatius Ezeani, Idris Abdulkabir Dangana, Herman Kamper, Hady Elsahar, Goodness Duru,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 397, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 115, + 397, + 505, + 410 + ], + "score": 1.0, + "content": "Ghollah Kioko, Espoir Murhabazi, Elan van Biljon, Daniel Whitenack, Christopher Onyefuluchi,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 408, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 115, + 408, + 505, + 421 + ], + "score": 1.0, + "content": "Chris Emezue, Bonaventure Dossou, Blessing Sibanda, Blessing Itoro Bassey, Ayodele Olabiyi,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 418, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 115, + 418, + 506, + 433 + ], + "score": 1.0, + "content": "Arshath Ramkilowan, Alp ktem, Adewale Akinfaderin, and Abdallah Bashir. Participatory re-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 430, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 115, + 430, + 506, + 443 + ], + "score": 1.0, + "content": "search for low-resourced machine translation: A case study in African languages. In Findings of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 442, + 444, + 454 + ], + "spans": [ + { + "bbox": [ + 116, + 442, + 444, + 454 + ], + "score": 1.0, + "content": "Empirical Methods in Natural Language Processing (Findings of EMNLP), 2020.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 22.5, + "bbox_fs": [ + 106, + 321, + 506, + 454 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 460, + 503, + 494 + ], + "lines": [ + { + "bbox": [ + 106, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 505, + 473 + ], + "score": 1.0, + "content": "Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu, and Andrew Y Ng. Reading", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 470, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 115, + 470, + 505, + 484 + ], + "score": 1.0, + "content": "digits in natural images with unsupervised feature learning. In NIPS Workshop on Deep Learning", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 482, + 291, + 495 + ], + "spans": [ + { + "bbox": [ + 116, + 482, + 291, + 495 + ], + "score": 1.0, + "content": "and Unsupervised Feature Learning, 2011.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 459, + 505, + 495 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 500, + 505, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 505, + 513 + ], + "score": 1.0, + "content": "Jianmo Ni, Jiacheng Li, and Julian McAuley. Justifying recommendations using distantly-labeled re-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 510, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 115, + 510, + 505, + 525 + ], + "score": 1.0, + "content": "views and fine-grained aspects. In Empirical Methods in Natural Language Processing (EMNLP),", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 523, + 196, + 533 + ], + "spans": [ + { + "bbox": [ + 115, + 523, + 196, + 533 + ], + "score": 1.0, + "content": "pp. 188–197, 2019.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 499, + 505, + 533 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 541, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 506, + 553 + ], + "score": 1.0, + "content": "Mohammad Sadegh Norouzzadeh, Dan Morris, Sara Beery, Neel Joshi, Nebojsa Jojic, and Jeff", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 551, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 116, + 551, + 505, + 565 + ], + "score": 1.0, + "content": "Clune. A deep active learning system for species identification and counting in camera trap", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 563, + 383, + 575 + ], + "spans": [ + { + "bbox": [ + 116, + 563, + 383, + 575 + ], + "score": 1.0, + "content": "images. Methods in Ecology and Evolution, 12(1):150–161, 2021.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 106, + 541, + 506, + 575 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 581, + 343, + 593 + ], + "lines": [ + { + "bbox": [ + 106, + 580, + 344, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 344, + 594 + ], + "score": 1.0, + "content": "Jill Nugent. inaturalist. Science Scope, 41(7):12–13, 2018.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38, + "bbox_fs": [ + 106, + 580, + 344, + 594 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 599, + 504, + 622 + ], + "lines": [ + { + "bbox": [ + 107, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 107, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "Yonatan Oren, Shiori Sagawa, Tatsunori Hashimoto, and Percy Liang. Distributionally robust lan-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 610, + 479, + 624 + ], + "spans": [ + { + "bbox": [ + 115, + 610, + 479, + 624 + ], + "score": 1.0, + "content": "guage modeling. In Empirical Methods in Natural Language Processing (EMNLP), 2019.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 107, + 599, + 505, + 624 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 628, + 503, + 651 + ], + "lines": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "Omiros Pantazis, Gabriel Brostow, Kate Jones, and Oisin Mac Aodha. Focus on the positives: Self-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 640, + 473, + 652 + ], + "spans": [ + { + "bbox": [ + 115, + 640, + 473, + 652 + ], + "score": 1.0, + "content": "supervised learning for biodiversity monitoring. arXiv preprint arXiv:2108.06435, 2021.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5, + "bbox_fs": [ + 106, + 629, + 505, + 652 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 658, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 504, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 504, + 669 + ], + "score": 1.0, + "content": "Mohammad Peikari, Sherine Salama, Sharon Nofech-Mozes, and Anne L Martel. A cluster-then-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 668, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 115, + 668, + 505, + 682 + ], + "score": 1.0, + "content": "label semi-supervised learning approach for pathology image classification. Scientific reports, 8", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 680, + 181, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 181, + 691 + ], + "score": 1.0, + "content": "(1):1–13, 2018.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44, + "bbox_fs": [ + 106, + 658, + 505, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 504, + 711 + ], + "score": 1.0, + "content": "Xingchao Peng, Ben Usman, Neela Kaushik, Dequan Wang, Judy Hoffman, and Kate Saenko.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 116, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Visda: A synthetic-to-real benchmark for visual domain adaptation. In Computer Vision and", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 116, + 720, + 325, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 325, + 732 + ], + "score": 1.0, + "content": "Pattern Recognition (CVPR), pp. 2021–2026, 2018.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47, + "bbox_fs": [ + 106, + 699, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 503, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Xingchao Peng, Qinxun Bai, Xide Xia, Zijun Huang, Kate Saenko, and Bo Wang. Moment matching", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "for multi-source domain adaptation. In International Conference on Computer Vision (ICCV),", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 114, + 103, + 143, + 117 + ], + "spans": [ + { + "bbox": [ + 114, + 103, + 143, + 117 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 105, + 123, + 503, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 123, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 505, + 136 + ], + "score": 1.0, + "content": "Joaquin Quinonero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D. Lawrence. ˜", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 133, + 345, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 133, + 345, + 146 + ], + "score": 1.0, + "content": "Dataset shift in machine learning. The MIT Press, 2009.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 152, + 505, + 197 + ], + "lines": [ + { + "bbox": [ + 105, + 151, + 505, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 505, + 165 + ], + "score": 1.0, + "content": "Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agar-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 163, + 506, + 176 + ], + "spans": [ + { + "bbox": [ + 115, + 163, + 506, + 176 + ], + "score": 1.0, + "content": "wal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 174, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 115, + 174, + 505, + 187 + ], + "score": 1.0, + "content": "Sutskever. Learning transferable visual models from natural language supervision. In Interna-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 185, + 461, + 198 + ], + "spans": [ + { + "bbox": [ + 115, + 185, + 461, + 198 + ], + "score": 1.0, + "content": "tional Conference on Machine Learning (ICML), volume 139, pp. 8748–8763, 2021a.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 504, + 238 + ], + "lines": [ + { + "bbox": [ + 106, + 204, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 216 + ], + "score": 1.0, + "content": "Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 216, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 117, + 216, + 506, + 228 + ], + "score": 1.0, + "content": "Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 117, + 226, + 457, + 239 + ], + "spans": [ + { + "bbox": [ + 117, + 226, + 457, + 239 + ], + "score": 1.0, + "content": "models from natural language supervision. arXiv preprint arXiv:2103.00020, 2021b.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 244, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 106, + 243, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 258 + ], + "score": 1.0, + "content": "Colorado J. Reed, Xiangyu Yue, Ani Nrusimha, Sayna Ebrahimi, Vivek Vijaykumar, Richard Mao,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 255, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 115, + 255, + 505, + 268 + ], + "score": 1.0, + "content": "Bo Li, Shanghang Zhang, Devin Guillory, Sean Metzger, Kurt Keutzer, and Trevor Darrell. Self-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 267, + 412, + 279 + ], + "spans": [ + { + "bbox": [ + 116, + 267, + 412, + 279 + ], + "score": 1.0, + "content": "supervised pretraining improves self-supervised pretraining. arXiv, 2021.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 284, + 505, + 319 + ], + "lines": [ + { + "bbox": [ + 105, + 284, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 298 + ], + "score": 1.0, + "content": "Jian Ren, Ilker Hacihaliloglu, Eric A Singer, David J Foran, and Xin Qi. Adversarial domain adapta-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 117, + 297, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 117, + 297, + 505, + 309 + ], + "score": 1.0, + "content": "tion for classification of prostate histopathology whole-slide images. In International Conference", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 307, + 468, + 320 + ], + "spans": [ + { + "bbox": [ + 115, + 307, + 468, + 320 + ], + "score": 1.0, + "content": "on Medical Image Computing and Computer-Assisted Intervention, pp. 201–209, 2018.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 108, + 325, + 505, + 360 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 505, + 338 + ], + "score": 1.0, + "content": "Shaoqing Ren, Kaiming He, Ross B. Girshick, and Jian Sun. Faster R-CNN: Towards real-time", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 116, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "object detection with region proposal networks. IEEE Transactions on Pattern Analysis and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 348, + 325, + 360 + ], + "spans": [ + { + "bbox": [ + 116, + 348, + 325, + 360 + ], + "score": 1.0, + "content": "Machine Intelligence (PAMI), 39:1137–1149, 2015.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 366, + 504, + 389 + ], + "lines": [ + { + "bbox": [ + 107, + 367, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 107, + 367, + 505, + 378 + ], + "score": 1.0, + "content": "Stephan R Richter, Vibhav Vineet, Stefan Roth, and Vladlen Koltun. Playing for data: Ground truth", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 377, + 469, + 389 + ], + "spans": [ + { + "bbox": [ + 115, + 377, + 469, + 389 + ], + "score": 1.0, + "content": "from computer games. In European conference on computer vision, pp. 102–118, 2016.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 395, + 504, + 430 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 409 + ], + "score": 1.0, + "content": "Tim Robertson, Markus Doring, Robert Guralnick, David Bloom, John Wieczorek, Kyle Braak, ¨", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 114, + 406, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 114, + 406, + 506, + 420 + ], + "score": 1.0, + "content": "Javier Otegui, Laura Russell, and Peter Desmet. The gbif integrated publishing toolkit: facilitating", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 418, + 478, + 430 + ], + "spans": [ + { + "bbox": [ + 116, + 418, + 478, + 430 + ], + "score": 1.0, + "content": "the efficient publishing of biodiversity data on the internet. PloS one, 9(8):e102623, 2014.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 436, + 504, + 459 + ], + "lines": [ + { + "bbox": [ + 106, + 436, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 505, + 449 + ], + "score": 1.0, + "content": "Alexander Robey, George J Pappas, and Hamed Hassani. Model-based domain generalization. arXiv", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 114, + 448, + 254, + 459 + ], + "spans": [ + { + "bbox": [ + 114, + 448, + 254, + 459 + ], + "score": 1.0, + "content": "preprint arXiv:2102.11436, 2021.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 466, + 504, + 489 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 504, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 504, + 479 + ], + "score": 1.0, + "content": "David Rogers and Mathew Hahn. Extended-connectivity fingerprints. Journal of Chemical Infor-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 477, + 397, + 489 + ], + "spans": [ + { + "bbox": [ + 115, + 477, + 397, + 489 + ], + "score": 1.0, + "content": "mation and Modeling, 50(5):742–754, 2010. doi: 10.1021/ci100050t.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 504, + 529 + ], + "lines": [ + { + "bbox": [ + 106, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "Yu Rong, Yatao Bian, Tingyang Xu, Weiyang Xie, Ying Wei, Wenbing Huang, and Junzhou", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 506, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 116, + 506, + 505, + 520 + ], + "score": 1.0, + "content": "Huang. Self-supervised graph transformer on large-scale molecular data. arXiv preprint", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 517, + 219, + 529 + ], + "spans": [ + { + "bbox": [ + 115, + 517, + 219, + 529 + ], + "score": 1.0, + "content": "arXiv:2007.02835, 2020.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 536, + 505, + 581 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "score": 1.0, + "content": "German Ros, Laura Sellart, Joanna Materzynska, David Vazquez, and Antonio M Lopez. The", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 548, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 116, + 548, + 505, + 560 + ], + "score": 1.0, + "content": "synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 114, + 558, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 114, + 558, + 506, + 572 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 3234–", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 569, + 167, + 581 + ], + "spans": [ + { + "bbox": [ + 115, + 569, + 167, + 581 + ], + "score": 1.0, + "content": "3243, 2016.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 108, + 587, + 504, + 622 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 506, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 602 + ], + "score": 1.0, + "content": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 117, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 117, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. ImageNet large scale visual", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 610, + 474, + 623 + ], + "spans": [ + { + "bbox": [ + 116, + 610, + 474, + 623 + ], + "score": 1.0, + "content": "recognition challenge. International Journal of Computer Vision, 115(3):211–252, 2015.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 105, + 628, + 503, + 651 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "Kate Saenko, Brian Kulis, Mario Fritz, and Trevor Darrell. Adapting visual category models to new", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 639, + 415, + 651 + ], + "spans": [ + { + "bbox": [ + 115, + 639, + 415, + 651 + ], + "score": 1.0, + "content": "domains. In European conference on computer vision, pp. 213–226, 2010.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 503, + 692 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "Shiori Sagawa, Pang Wei Koh, Tatsunori B. Hashimoto, and Percy Liang. Distributionally robust", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 669, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 115, + 669, + 505, + 682 + ], + "score": 1.0, + "content": "neural networks for group shifts: On the importance of regularization for worst-case generaliza-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 680, + 428, + 694 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 428, + 694 + ], + "score": 1.0, + "content": "tion. In International Conference on Learning Representations (ICLR), 2020.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Kuniaki Saito, Yoshitaka Ushiku, and Tatsuya Harada. Asymmetric tri-training for unsupervised", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "domain adaptation. In International Conference on Machine Learning (ICML), pp. 2988–2997,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 115, + 719, + 143, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 719, + 143, + 732 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46 + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 503, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Xingchao Peng, Qinxun Bai, Xide Xia, Zijun Huang, Kate Saenko, and Bo Wang. Moment matching", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "for multi-source domain adaptation. In International Conference on Computer Vision (ICCV),", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 114, + 103, + 143, + 117 + ], + "spans": [ + { + "bbox": [ + 114, + 103, + 143, + 117 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 505, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 123, + 503, + 145 + ], + "lines": [ + { + "bbox": [ + 105, + 123, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 505, + 136 + ], + "score": 1.0, + "content": "Joaquin Quinonero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D. Lawrence. ˜", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 133, + 345, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 133, + 345, + 146 + ], + "score": 1.0, + "content": "Dataset shift in machine learning. The MIT Press, 2009.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 123, + 505, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 152, + 505, + 197 + ], + "lines": [ + { + "bbox": [ + 105, + 151, + 505, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 505, + 165 + ], + "score": 1.0, + "content": "Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agar-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 163, + 506, + 176 + ], + "spans": [ + { + "bbox": [ + 115, + 163, + 506, + 176 + ], + "score": 1.0, + "content": "wal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 174, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 115, + 174, + 505, + 187 + ], + "score": 1.0, + "content": "Sutskever. Learning transferable visual models from natural language supervision. In Interna-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 185, + 461, + 198 + ], + "spans": [ + { + "bbox": [ + 115, + 185, + 461, + 198 + ], + "score": 1.0, + "content": "tional Conference on Machine Learning (ICML), volume 139, pp. 8748–8763, 2021a.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 151, + 506, + 198 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 504, + 238 + ], + "lines": [ + { + "bbox": [ + 106, + 204, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 216 + ], + "score": 1.0, + "content": "Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 216, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 117, + 216, + 506, + 228 + ], + "score": 1.0, + "content": "Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 117, + 226, + 457, + 239 + ], + "spans": [ + { + "bbox": [ + 117, + 226, + 457, + 239 + ], + "score": 1.0, + "content": "models from natural language supervision. arXiv preprint arXiv:2103.00020, 2021b.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 204, + 506, + 239 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 244, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 106, + 243, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 258 + ], + "score": 1.0, + "content": "Colorado J. Reed, Xiangyu Yue, Ani Nrusimha, Sayna Ebrahimi, Vivek Vijaykumar, Richard Mao,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 255, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 115, + 255, + 505, + 268 + ], + "score": 1.0, + "content": "Bo Li, Shanghang Zhang, Devin Guillory, Sean Metzger, Kurt Keutzer, and Trevor Darrell. Self-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 267, + 412, + 279 + ], + "spans": [ + { + "bbox": [ + 116, + 267, + 412, + 279 + ], + "score": 1.0, + "content": "supervised pretraining improves self-supervised pretraining. arXiv, 2021.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 106, + 243, + 505, + 279 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 284, + 505, + 319 + ], + "lines": [ + { + "bbox": [ + 105, + 284, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 505, + 298 + ], + "score": 1.0, + "content": "Jian Ren, Ilker Hacihaliloglu, Eric A Singer, David J Foran, and Xin Qi. Adversarial domain adapta-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 117, + 297, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 117, + 297, + 505, + 309 + ], + "score": 1.0, + "content": "tion for classification of prostate histopathology whole-slide images. In International Conference", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 307, + 468, + 320 + ], + "spans": [ + { + "bbox": [ + 115, + 307, + 468, + 320 + ], + "score": 1.0, + "content": "on Medical Image Computing and Computer-Assisted Intervention, pp. 201–209, 2018.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 284, + 505, + 320 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 325, + 505, + 360 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 505, + 338 + ], + "score": 1.0, + "content": "Shaoqing Ren, Kaiming He, Ross B. Girshick, and Jian Sun. Faster R-CNN: Towards real-time", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 116, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "object detection with region proposal networks. IEEE Transactions on Pattern Analysis and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 348, + 325, + 360 + ], + "spans": [ + { + "bbox": [ + 116, + 348, + 325, + 360 + ], + "score": 1.0, + "content": "Machine Intelligence (PAMI), 39:1137–1149, 2015.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 106, + 325, + 505, + 360 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 366, + 504, + 389 + ], + "lines": [ + { + "bbox": [ + 107, + 367, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 107, + 367, + 505, + 378 + ], + "score": 1.0, + "content": "Stephan R Richter, Vibhav Vineet, Stefan Roth, and Vladlen Koltun. Playing for data: Ground truth", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 377, + 469, + 389 + ], + "spans": [ + { + "bbox": [ + 115, + 377, + 469, + 389 + ], + "score": 1.0, + "content": "from computer games. In European conference on computer vision, pp. 102–118, 2016.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 107, + 367, + 505, + 389 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 395, + 504, + 430 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 505, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 409 + ], + "score": 1.0, + "content": "Tim Robertson, Markus Doring, Robert Guralnick, David Bloom, John Wieczorek, Kyle Braak, ¨", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 114, + 406, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 114, + 406, + 506, + 420 + ], + "score": 1.0, + "content": "Javier Otegui, Laura Russell, and Peter Desmet. The gbif integrated publishing toolkit: facilitating", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 418, + 478, + 430 + ], + "spans": [ + { + "bbox": [ + 116, + 418, + 478, + 430 + ], + "score": 1.0, + "content": "the efficient publishing of biodiversity data on the internet. PloS one, 9(8):e102623, 2014.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 395, + 506, + 430 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 436, + 504, + 459 + ], + "lines": [ + { + "bbox": [ + 106, + 436, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 505, + 449 + ], + "score": 1.0, + "content": "Alexander Robey, George J Pappas, and Hamed Hassani. Model-based domain generalization. arXiv", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 114, + 448, + 254, + 459 + ], + "spans": [ + { + "bbox": [ + 114, + 448, + 254, + 459 + ], + "score": 1.0, + "content": "preprint arXiv:2102.11436, 2021.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 106, + 436, + 505, + 459 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 466, + 504, + 489 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 504, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 504, + 479 + ], + "score": 1.0, + "content": "David Rogers and Mathew Hahn. Extended-connectivity fingerprints. Journal of Chemical Infor-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 477, + 397, + 489 + ], + "spans": [ + { + "bbox": [ + 115, + 477, + 397, + 489 + ], + "score": 1.0, + "content": "mation and Modeling, 50(5):742–754, 2010. doi: 10.1021/ci100050t.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 106, + 466, + 504, + 489 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 504, + 529 + ], + "lines": [ + { + "bbox": [ + 106, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "Yu Rong, Yatao Bian, Tingyang Xu, Weiyang Xie, Ying Wei, Wenbing Huang, and Junzhou", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 506, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 116, + 506, + 505, + 520 + ], + "score": 1.0, + "content": "Huang. Self-supervised graph transformer on large-scale molecular data. arXiv preprint", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 517, + 219, + 529 + ], + "spans": [ + { + "bbox": [ + 115, + 517, + 219, + 529 + ], + "score": 1.0, + "content": "arXiv:2007.02835, 2020.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 106, + 495, + 505, + 529 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 536, + 505, + 581 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 505, + 549 + ], + "score": 1.0, + "content": "German Ros, Laura Sellart, Joanna Materzynska, David Vazquez, and Antonio M Lopez. The", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 548, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 116, + 548, + 505, + 560 + ], + "score": 1.0, + "content": "synthia dataset: A large collection of synthetic images for semantic segmentation of urban scenes.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 114, + 558, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 114, + 558, + 506, + 572 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 3234–", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 569, + 167, + 581 + ], + "spans": [ + { + "bbox": [ + 115, + 569, + 167, + 581 + ], + "score": 1.0, + "content": "3243, 2016.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 535, + 506, + 581 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 587, + 504, + 622 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 506, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 506, + 602 + ], + "score": 1.0, + "content": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 117, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 117, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. ImageNet large scale visual", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 610, + 474, + 623 + ], + "spans": [ + { + "bbox": [ + 116, + 610, + 474, + 623 + ], + "score": 1.0, + "content": "recognition challenge. International Journal of Computer Vision, 115(3):211–252, 2015.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 586, + 506, + 623 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 628, + 503, + 651 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "Kate Saenko, Brian Kulis, Mario Fritz, and Trevor Darrell. Adapting visual category models to new", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 639, + 415, + 651 + ], + "spans": [ + { + "bbox": [ + 115, + 639, + 415, + 651 + ], + "score": 1.0, + "content": "domains. In European conference on computer vision, pp. 213–226, 2010.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 106, + 627, + 505, + 651 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 503, + 692 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "Shiori Sagawa, Pang Wei Koh, Tatsunori B. Hashimoto, and Percy Liang. Distributionally robust", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 669, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 115, + 669, + 505, + 682 + ], + "score": 1.0, + "content": "neural networks for group shifts: On the importance of regularization for worst-case generaliza-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 680, + 428, + 694 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 428, + 694 + ], + "score": 1.0, + "content": "tion. In International Conference on Learning Representations (ICLR), 2020.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 106, + 657, + 506, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Kuniaki Saito, Yoshitaka Ushiku, and Tatsuya Harada. Asymmetric tri-training for unsupervised", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "domain adaptation. In International Conference on Machine Learning (ICML), pp. 2988–2997,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 115, + 719, + 143, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 719, + 143, + 732 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46, + "bbox_fs": [ + 106, + 699, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 504, + 95 + ], + "score": 1.0, + "content": "Kuniaki Saito, Kohei Watanabe, Yoshitaka Ushiku, and Tatsuya Harada. Maximum classifier dis-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 107 + ], + "score": 1.0, + "content": "crepancy for unsupervised domain adaptation. In Proceedings of the IEEE conference on com-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 105, + 356, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 105, + 356, + 117 + ], + "score": 1.0, + "content": "puter vision and pattern recognition, pp. 3723–3732, 2018.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 503, + 157 + ], + "lines": [ + { + "bbox": [ + 106, + 124, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 124, + 505, + 136 + ], + "score": 1.0, + "content": "Kuniaki Saito, Donghyun Kim, Piotr Teterwak, Stan Sclaroff, Trevor Darrell, and Kate Saenko. Tune", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 505, + 148 + ], + "score": 1.0, + "content": "it the right way: Unsupervised validation of domain adaptation via soft neighborhood density.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 146, + 279, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 146, + 279, + 158 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2108.10860, 2021.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 502, + 188 + ], + "lines": [ + { + "bbox": [ + 107, + 164, + 504, + 177 + ], + "spans": [ + { + "bbox": [ + 107, + 164, + 504, + 177 + ], + "score": 1.0, + "content": "Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. Distilbert, a distilled version of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 442, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 442, + 188 + ], + "score": 1.0, + "content": "bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108, 2019.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 194, + 506, + 228 + ], + "lines": [ + { + "bbox": [ + 106, + 195, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 505, + 207 + ], + "score": 1.0, + "content": "Shayne Shaw, Maciej Pajak, Aneta Lisowska, Sotirios A Tsaftaris, and Alison Q O’Neil. Teacher-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 206, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 115, + 206, + 505, + 219 + ], + "score": 1.0, + "content": "student chain for efficient semi-supervised histology image classification. arXiv preprint", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 217, + 219, + 228 + ], + "spans": [ + { + "bbox": [ + 115, + 217, + 219, + 228 + ], + "score": 1.0, + "content": "arXiv:2003.08797, 2020.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 236, + 505, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 235, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 250 + ], + "score": 1.0, + "content": "Kendrick Shen, Robbie Matthew Jones, Ananya Kumar, Sang Michael Xie, and Percy Liang. How", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 247, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 116, + 247, + 505, + 259 + ], + "score": 1.0, + "content": "does contrastive pre-training connect disparate domains? In NeurIPS Workshop on Distribution", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 258, + 169, + 271 + ], + "spans": [ + { + "bbox": [ + 115, + 258, + 169, + 271 + ], + "score": 1.0, + "content": "Shifts, 2021.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 277, + 505, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 277, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 505, + 290 + ], + "score": 1.0, + "content": "Kihyuk Sohn, David Berthelot, Chun-Liang Li, Zizhao Zhang, Nicholas Carlini, Ekin D. Cubuk,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 115, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "Alex Kurakin, Han Zhang, and Colin Raffel. Fixmatch: Simplifying semi-supervised learning", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 299, + 303, + 311 + ], + "spans": [ + { + "bbox": [ + 116, + 299, + 303, + 311 + ], + "score": 1.0, + "content": "with consistency and confidence. arXiv, 2020.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 318, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 505, + 333 + ], + "score": 1.0, + "content": "Ryan Steed and Aylin Caliskan. Image representations learned with unsupervised pre-training", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 329, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 115, + 329, + 505, + 344 + ], + "score": 1.0, + "content": "contain human-like biases. In ACM Conference on Fairness, Accountability, and Transparency", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 341, + 234, + 353 + ], + "spans": [ + { + "bbox": [ + 116, + 341, + 234, + 353 + ], + "score": 1.0, + "content": "(FAccT), pp. 701–713, 2021.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 103, + 360, + 504, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "score": 1.0, + "content": "Jong-Chyi Su and Subhransu Maji. The semi-supervised inaturalist-aves challenge at fgvc7 work-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 371, + 304, + 383 + ], + "spans": [ + { + "bbox": [ + 116, + 371, + 304, + 383 + ], + "score": 1.0, + "content": "shop. arXiv preprint arXiv:2103.06937, 2021.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 104, + 390, + 504, + 413 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "score": 1.0, + "content": "Baochen Sun and Kate Saenko. Deep coral: Correlation alignment for deep domain adaptation. In", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 401, + 349, + 414 + ], + "spans": [ + { + "bbox": [ + 116, + 401, + 349, + 414 + ], + "score": 1.0, + "content": "European Conference on Computer Vision (ECCV), 2016.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 105, + 420, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 421, + 505, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 505, + 434 + ], + "score": 1.0, + "content": "Baochen Sun, Jiashi Feng, and Kate Saenko. Return of frustratingly easy domain adaptation. In", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 431, + 406, + 444 + ], + "spans": [ + { + "bbox": [ + 115, + 431, + 406, + 444 + ], + "score": 1.0, + "content": "Association for the Advancement of Artificial Intelligence (AAAI), 2016.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 106, + 450, + 505, + 485 + ], + "lines": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "Fan-Yun Sun, Jordan Hoffmann, Vikas Verma, and Jian Tang. Infograph: Unsupervised and semi-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 462, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 116, + 462, + 505, + 475 + ], + "score": 1.0, + "content": "supervised graph-level representation learning via mutual information maximization. In Interna-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 473, + 366, + 486 + ], + "spans": [ + { + "bbox": [ + 116, + 473, + 366, + 486 + ], + "score": 1.0, + "content": "tional Conference on Learning Representations (ICLR), 2020.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 503, + 515 + ], + "lines": [ + { + "bbox": [ + 106, + 491, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 505, + 505 + ], + "score": 1.0, + "content": "Yi Chern Tan and L Elisa Celis. Assessing social and intersectional biases in contextualized word", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 504, + 346, + 515 + ], + "spans": [ + { + "bbox": [ + 115, + 504, + 346, + 515 + ], + "score": 1.0, + "content": "representations. arXiv preprint arXiv:1911.01485, 2019.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 505, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 520, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 536 + ], + "score": 1.0, + "content": "Rohan Taori, Achal Dave, Vaishaal Shankar, Nicholas Carlini, Benjamin Recht, and Ludwig", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 116, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "Schmidt. Measuring robustness to natural distribution shifts in image classification. arXiv preprint", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 545, + 219, + 556 + ], + "spans": [ + { + "bbox": [ + 115, + 545, + 219, + 556 + ], + "score": 1.0, + "content": "arXiv:2007.00644, 2020.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "David Tellez, Geert Litjens, Peter B ´ andi, Wouter Bulten, John-Melle Bokhorst, Francesco Ciompi, ´", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 116, + 575, + 505, + 587 + ], + "score": 1.0, + "content": "and Jeroen van der Laak. Quantifying the effects of data augmentation and stain color normaliza-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 585, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 115, + 585, + 505, + 598 + ], + "score": 1.0, + "content": "tion in convolutional neural networks for computational pathology. Medical Image Analysis, 58,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 596, + 142, + 608 + ], + "spans": [ + { + "bbox": [ + 114, + 596, + 142, + 608 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 107, + 615, + 504, + 650 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 504, + 629 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 504, + 629 + ], + "score": 1.0, + "content": "Yao-Hung Hubert Tsai, Martin Q Ma, Han Zhao, Kun Zhang, Louis-Philippe Morency, and Rus-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 627, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 116, + 627, + 504, + 639 + ], + "score": 1.0, + "content": "lan Salakhutdinov. Conditional contrastive learning: Removing undesirable information in self-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 639, + 390, + 650 + ], + "spans": [ + { + "bbox": [ + 116, + 639, + 390, + 650 + ], + "score": 1.0, + "content": "supervised representations. arXiv preprint arXiv:2106.02866, 2021.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 671 + ], + "score": 1.0, + "content": "Lifu Tu, Garima Lalwani, Spandana Gella, and He He. An empirical study on robustness to spurious", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 669, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 116, + 669, + 505, + 681 + ], + "score": 1.0, + "content": "correlations using pre-trained language models. Transactions of the Association for Computa-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 679, + 295, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 679, + 295, + 691 + ], + "score": 1.0, + "content": "tional Linguistics (TACL), 8:621–633, 2020.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "score": 1.0, + "content": "Grant Van Horn, Elijah Cole, Sara Beery, Kimberly Wilber, Serge Belongie, and Oisin Mac Aodha.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "Benchmarking representation learning for natural world image collections. In Proceedings of the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 720, + 494, + 734 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 494, + 734 + ], + "score": 1.0, + "content": "IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 12884–12893, 2021.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "19", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 504, + 95 + ], + "score": 1.0, + "content": "Kuniaki Saito, Kohei Watanabe, Yoshitaka Ushiku, and Tatsuya Harada. Maximum classifier dis-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 107 + ], + "score": 1.0, + "content": "crepancy for unsupervised domain adaptation. In Proceedings of the IEEE conference on com-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 105, + 356, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 105, + 356, + 117 + ], + "score": 1.0, + "content": "puter vision and pattern recognition, pp. 3723–3732, 2018.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 82, + 505, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 503, + 157 + ], + "lines": [ + { + "bbox": [ + 106, + 124, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 124, + 505, + 136 + ], + "score": 1.0, + "content": "Kuniaki Saito, Donghyun Kim, Piotr Teterwak, Stan Sclaroff, Trevor Darrell, and Kate Saenko. Tune", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 505, + 148 + ], + "score": 1.0, + "content": "it the right way: Unsupervised validation of domain adaptation via soft neighborhood density.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 146, + 279, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 146, + 279, + 158 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2108.10860, 2021.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 124, + 505, + 158 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 164, + 502, + 188 + ], + "lines": [ + { + "bbox": [ + 107, + 164, + 504, + 177 + ], + "spans": [ + { + "bbox": [ + 107, + 164, + 504, + 177 + ], + "score": 1.0, + "content": "Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. Distilbert, a distilled version of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 442, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 442, + 188 + ], + "score": 1.0, + "content": "bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108, 2019.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 107, + 164, + 504, + 188 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 194, + 506, + 228 + ], + "lines": [ + { + "bbox": [ + 106, + 195, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 505, + 207 + ], + "score": 1.0, + "content": "Shayne Shaw, Maciej Pajak, Aneta Lisowska, Sotirios A Tsaftaris, and Alison Q O’Neil. Teacher-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 206, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 115, + 206, + 505, + 219 + ], + "score": 1.0, + "content": "student chain for efficient semi-supervised histology image classification. arXiv preprint", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 217, + 219, + 228 + ], + "spans": [ + { + "bbox": [ + 115, + 217, + 219, + 228 + ], + "score": 1.0, + "content": "arXiv:2003.08797, 2020.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 106, + 195, + 505, + 228 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 236, + 505, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 235, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 505, + 250 + ], + "score": 1.0, + "content": "Kendrick Shen, Robbie Matthew Jones, Ananya Kumar, Sang Michael Xie, and Percy Liang. How", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 247, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 116, + 247, + 505, + 259 + ], + "score": 1.0, + "content": "does contrastive pre-training connect disparate domains? In NeurIPS Workshop on Distribution", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 258, + 169, + 271 + ], + "spans": [ + { + "bbox": [ + 115, + 258, + 169, + 271 + ], + "score": 1.0, + "content": "Shifts, 2021.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 235, + 505, + 271 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 277, + 505, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 277, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 505, + 290 + ], + "score": 1.0, + "content": "Kihyuk Sohn, David Berthelot, Chun-Liang Li, Zizhao Zhang, Nicholas Carlini, Ekin D. Cubuk,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 115, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "Alex Kurakin, Han Zhang, and Colin Raffel. Fixmatch: Simplifying semi-supervised learning", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 299, + 303, + 311 + ], + "spans": [ + { + "bbox": [ + 116, + 299, + 303, + 311 + ], + "score": 1.0, + "content": "with consistency and confidence. arXiv, 2020.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 106, + 277, + 505, + 311 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 318, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 505, + 333 + ], + "score": 1.0, + "content": "Ryan Steed and Aylin Caliskan. Image representations learned with unsupervised pre-training", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 329, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 115, + 329, + 505, + 344 + ], + "score": 1.0, + "content": "contain human-like biases. In ACM Conference on Fairness, Accountability, and Transparency", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 341, + 234, + 353 + ], + "spans": [ + { + "bbox": [ + 116, + 341, + 234, + 353 + ], + "score": 1.0, + "content": "(FAccT), pp. 701–713, 2021.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 318, + 505, + 353 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 360, + 504, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "score": 1.0, + "content": "Jong-Chyi Su and Subhransu Maji. The semi-supervised inaturalist-aves challenge at fgvc7 work-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 371, + 304, + 383 + ], + "spans": [ + { + "bbox": [ + 116, + 371, + 304, + 383 + ], + "score": 1.0, + "content": "shop. arXiv preprint arXiv:2103.06937, 2021.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 360, + 505, + 383 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 390, + 504, + 413 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 404 + ], + "score": 1.0, + "content": "Baochen Sun and Kate Saenko. Deep coral: Correlation alignment for deep domain adaptation. In", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 401, + 349, + 414 + ], + "spans": [ + { + "bbox": [ + 116, + 401, + 349, + 414 + ], + "score": 1.0, + "content": "European Conference on Computer Vision (ECCV), 2016.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 390, + 505, + 414 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 420, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 421, + 505, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 505, + 434 + ], + "score": 1.0, + "content": "Baochen Sun, Jiashi Feng, and Kate Saenko. Return of frustratingly easy domain adaptation. In", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 431, + 406, + 444 + ], + "spans": [ + { + "bbox": [ + 115, + 431, + 406, + 444 + ], + "score": 1.0, + "content": "Association for the Advancement of Artificial Intelligence (AAAI), 2016.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 106, + 421, + 505, + 444 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 450, + 505, + 485 + ], + "lines": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "Fan-Yun Sun, Jordan Hoffmann, Vikas Verma, and Jian Tang. Infograph: Unsupervised and semi-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 462, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 116, + 462, + 505, + 475 + ], + "score": 1.0, + "content": "supervised graph-level representation learning via mutual information maximization. In Interna-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 473, + 366, + 486 + ], + "spans": [ + { + "bbox": [ + 116, + 473, + 366, + 486 + ], + "score": 1.0, + "content": "tional Conference on Learning Representations (ICLR), 2020.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 451, + 505, + 486 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 503, + 515 + ], + "lines": [ + { + "bbox": [ + 106, + 491, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 505, + 505 + ], + "score": 1.0, + "content": "Yi Chern Tan and L Elisa Celis. Assessing social and intersectional biases in contextualized word", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 504, + 346, + 515 + ], + "spans": [ + { + "bbox": [ + 115, + 504, + 346, + 515 + ], + "score": 1.0, + "content": "representations. arXiv preprint arXiv:1911.01485, 2019.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 106, + 491, + 505, + 515 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 505, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 520, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 536 + ], + "score": 1.0, + "content": "Rohan Taori, Achal Dave, Vaishaal Shankar, Nicholas Carlini, Benjamin Recht, and Ludwig", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 116, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "Schmidt. Measuring robustness to natural distribution shifts in image classification. arXiv preprint", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 545, + 219, + 556 + ], + "spans": [ + { + "bbox": [ + 115, + 545, + 219, + 556 + ], + "score": 1.0, + "content": "arXiv:2007.00644, 2020.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 520, + 506, + 556 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "David Tellez, Geert Litjens, Peter B ´ andi, Wouter Bulten, John-Melle Bokhorst, Francesco Ciompi, ´", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 116, + 575, + 505, + 587 + ], + "score": 1.0, + "content": "and Jeroen van der Laak. Quantifying the effects of data augmentation and stain color normaliza-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 585, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 115, + 585, + 505, + 598 + ], + "score": 1.0, + "content": "tion in convolutional neural networks for computational pathology. Medical Image Analysis, 58,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 596, + 142, + 608 + ], + "spans": [ + { + "bbox": [ + 114, + 596, + 142, + 608 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 563, + 505, + 608 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 615, + 504, + 650 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 504, + 629 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 504, + 629 + ], + "score": 1.0, + "content": "Yao-Hung Hubert Tsai, Martin Q Ma, Han Zhao, Kun Zhang, Louis-Philippe Morency, and Rus-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 627, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 116, + 627, + 504, + 639 + ], + "score": 1.0, + "content": "lan Salakhutdinov. Conditional contrastive learning: Removing undesirable information in self-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 639, + 390, + 650 + ], + "spans": [ + { + "bbox": [ + 116, + 639, + 390, + 650 + ], + "score": 1.0, + "content": "supervised representations. arXiv preprint arXiv:2106.02866, 2021.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 106, + 615, + 504, + 650 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 671 + ], + "score": 1.0, + "content": "Lifu Tu, Garima Lalwani, Spandana Gella, and He He. An empirical study on robustness to spurious", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 669, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 116, + 669, + 505, + 681 + ], + "score": 1.0, + "content": "correlations using pre-trained language models. Transactions of the Association for Computa-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 679, + 295, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 679, + 295, + 691 + ], + "score": 1.0, + "content": "tional Linguistics (TACL), 8:621–633, 2020.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 106, + 657, + 505, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "score": 1.0, + "content": "Grant Van Horn, Elijah Cole, Sara Beery, Kimberly Wilber, Serge Belongie, and Oisin Mac Aodha.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 116, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "Benchmarking representation learning for natural world image collections. In Proceedings of the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 720, + 494, + 734 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 494, + 734 + ], + "score": 1.0, + "content": "IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 12884–12893, 2021.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 697, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethuraman Panchanathan. Deep", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "hashing network for unsupervised domain adaptation. In Computer Vision and Pattern Recogni-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 104, + 261, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 104, + 261, + 116 + ], + "score": 1.0, + "content": "tion (CVPR), pp. 5018–5027, 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 105, + 124, + 504, + 147 + ], + "lines": [ + { + "bbox": [ + 106, + 124, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 124, + 505, + 137 + ], + "score": 1.0, + "content": "Vikas Verma, Thang Luong, Kenji Kawaguchi, Hieu Pham, and Quoc Le. Towards domain-agnostic", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 136, + 465, + 148 + ], + "spans": [ + { + "bbox": [ + 115, + 136, + 465, + 148 + ], + "score": 1.0, + "content": "contrastive learning. In International Conference on Machine Learning (ICML), 2021.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 155, + 505, + 189 + ], + "lines": [ + { + "bbox": [ + 105, + 155, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 155, + 505, + 168 + ], + "score": 1.0, + "content": "Mitko Veta, Paul J Van Diest, Mehdi Jiwa, Shaimaa Al-Janabi, and Josien PW Pluim. Mitosis", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 166, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 116, + 166, + 505, + 178 + ], + "score": 1.0, + "content": "counting in breast cancer: Object-level interobserver agreement and comparison to an automatic", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 177, + 245, + 189 + ], + "spans": [ + { + "bbox": [ + 116, + 177, + 245, + 189 + ], + "score": 1.0, + "content": "method. PloS one, 11(8), 2016.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 197, + 505, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "score": 1.0, + "content": "Pascal Vincent, Hugo Larochelle, Yoshua Bengio, , and Pierre-Antoine Manzagol. Extracting and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 208, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 115, + 208, + 505, + 221 + ], + "score": 1.0, + "content": "composing robust features with denoising autoencoders. In International Conference on Machine", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 219, + 216, + 231 + ], + "spans": [ + { + "bbox": [ + 116, + 219, + 216, + 231 + ], + "score": 1.0, + "content": "Learning (ICML), 2008.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 104, + 239, + 504, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "score": 1.0, + "content": "Rui Wang, Zuxuan Wu, Zejia Weng, Jingjing Chen, Guo-Jun Qi, and Yu-Gang Jiang. Cross-domain", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 250, + 399, + 262 + ], + "spans": [ + { + "bbox": [ + 115, + 250, + 399, + 262 + ], + "score": 1.0, + "content": "contrastive learning for unsupervised domain adaptation. arXiv, 2021.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 105, + 270, + 503, + 293 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 504, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 504, + 282 + ], + "score": 1.0, + "content": "Colin Wei, Sang Michael Xie, and Tengyu Ma. Why do pretrained language models help in down-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 281, + 494, + 294 + ], + "spans": [ + { + "bbox": [ + 116, + 281, + 494, + 294 + ], + "score": 1.0, + "content": "stream tasks? an analysis of head and prompt tuning. arXiv preprint arXiv:2106.09226, 2021.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 107, + 301, + 504, + 334 + ], + "lines": [ + { + "bbox": [ + 106, + 301, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 505, + 313 + ], + "score": 1.0, + "content": "Ben G Weinstein, Sergio Marconi, Stephanie Bohlman, Alina Zare, and Ethan White. Individual", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 115, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "tree-crown detection in rgb imagery using semi-supervised deep learning neural networks. Remote", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 323, + 231, + 335 + ], + "spans": [ + { + "bbox": [ + 116, + 323, + 231, + 335 + ], + "score": 1.0, + "content": "Sensing, 11(11):1309, 2019.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 342, + 505, + 377 + ], + "lines": [ + { + "bbox": [ + 105, + 342, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 505, + 356 + ], + "score": 1.0, + "content": "Ben G Weinstein, Lindsey Gardner, Vienna Saccomanno, Ashley Steinkraus, Andrew Ortega, Kris-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 354, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 115, + 354, + 505, + 367 + ], + "score": 1.0, + "content": "ten Brush, Glenda Yenni, Ann E McKellar, Rowan Converse, Christopher Lippitt, et al. A general", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 365, + 479, + 379 + ], + "spans": [ + { + "bbox": [ + 116, + 365, + 479, + 379 + ], + "score": 1.0, + "content": "deep learning model for bird detection in high resolution airborne imagery. bioRxiv, 2021.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 384, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 104, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 104, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "John N Weinstein, Eric A Collisson, Gordon B Mills, Kenna R Mills Shaw, Brad A Ozenberger,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "Kyle Ellrott, Ilya Shmulevich, Chris Sander, Joshua M Stuart, Cancer Genome Atlas Research", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 114, + 406, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 114, + 406, + 506, + 420 + ], + "score": 1.0, + "content": "Network, et al. The cancer genome atlas pan-cancer analysis project. Nature genetics, 45(10),", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 417, + 142, + 429 + ], + "spans": [ + { + "bbox": [ + 115, + 417, + 142, + 429 + ], + "score": 1.0, + "content": "2013.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 108, + 437, + 503, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "score": 1.0, + "content": "Zhenqin Wu, Bharath Ramsundar, Evan N Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 115, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "Pappu, Karl Leswing, and Vijay Pande. Moleculenet: a benchmark for molecular machine learn-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 460, + 294, + 472 + ], + "spans": [ + { + "bbox": [ + 115, + 460, + 294, + 472 + ], + "score": 1.0, + "content": "ing. Chemical science, 9(2):513–530, 2018.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 479, + 505, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "score": 1.0, + "content": "Michael Xie, Neal Jean, Marshall Burke, David Lobell, and Stefano Ermon. Transfer learning from", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 115, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "deep features for remote sensing and poverty mapping. In Association for the Advancement of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 502, + 262, + 514 + ], + "spans": [ + { + "bbox": [ + 116, + 502, + 262, + 514 + ], + "score": 1.0, + "content": "Artificial Intelligence (AAAI), 2016.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 521, + 503, + 545 + ], + "lines": [ + { + "bbox": [ + 106, + 520, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 505, + 535 + ], + "score": 1.0, + "content": "Qizhe Xie, Minh-Thang Luong, Eduard Hovy, and Quoc V. Le. Self-training with noisy student", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 533, + 305, + 545 + ], + "spans": [ + { + "bbox": [ + 116, + 533, + 305, + 545 + ], + "score": 1.0, + "content": "improves imagenet classification. arXiv, 2020.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 552, + 505, + 586 + ], + "lines": [ + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "score": 1.0, + "content": "Sang Michael Xie, Ananya Kumar, Robbie Jones, Fereshte Khani, Tengyu Ma, and Percy Liang.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 563, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 115, + 563, + 505, + 577 + ], + "score": 1.0, + "content": "In-N-out: Pre-training and self-training using auxiliary information for out-of-distribution robust-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 575, + 434, + 587 + ], + "spans": [ + { + "bbox": [ + 116, + 575, + 434, + 587 + ], + "score": 1.0, + "content": "ness. In International Conference on Learning Representations (ICLR), 2021a.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 594, + 504, + 617 + ], + "lines": [ + { + "bbox": [ + 106, + 593, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 506, + 608 + ], + "score": 1.0, + "content": "Yaochen Xie, Zhao Xu, Jingtun Zhang, Zhengyang Wang, and Shuiwang Ji. Self-supervised learning", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 605, + 461, + 618 + ], + "spans": [ + { + "bbox": [ + 116, + 605, + 461, + 618 + ], + "score": 1.0, + "content": "of graph neural networks: A unified review. arXiv preprint arXiv:2102.10757, 2021b.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 105, + 625, + 504, + 649 + ], + "lines": [ + { + "bbox": [ + 106, + 624, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 638 + ], + "score": 1.0, + "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 636, + 452, + 649 + ], + "spans": [ + { + "bbox": [ + 116, + 636, + 452, + 649 + ], + "score": 1.0, + "content": "networks? In International Conference on Learning Representations (ICLR), 2018.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 504, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 670 + ], + "score": 1.0, + "content": "Ruijia Xu, Guanbin Li, Jihan Yang, and Liang Lin. Larger norm more transferable: An adap-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 668, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 116, + 668, + 505, + 680 + ], + "score": 1.0, + "content": "tive feature norm approach for unsupervised domain adaptation. In International Conference on", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 678, + 310, + 692 + ], + "spans": [ + { + "bbox": [ + 116, + 678, + 310, + 692 + ], + "score": 1.0, + "content": "Computer Vision (ICCV), pp. 1426–1435, 2019.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "Christopher Yeh, Anthony Perez, Anne Driscoll, George Azzari, Zhongyi Tang, David Lobell, Ste-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "fano Ermon, and Marshall Burke. Using publicly available satellite imagery and deep learning to", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 116, + 720, + 430, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 430, + 732 + ], + "score": 1.0, + "content": "understand economic well-being in africa. Nature Communications, 11, 2020.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethuraman Panchanathan. Deep", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "hashing network for unsupervised domain adaptation. In Computer Vision and Pattern Recogni-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 104, + 261, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 104, + 261, + 116 + ], + "score": 1.0, + "content": "tion (CVPR), pp. 5018–5027, 2017.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 505, + 116 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 124, + 504, + 147 + ], + "lines": [ + { + "bbox": [ + 106, + 124, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 106, + 124, + 505, + 137 + ], + "score": 1.0, + "content": "Vikas Verma, Thang Luong, Kenji Kawaguchi, Hieu Pham, and Quoc Le. Towards domain-agnostic", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 136, + 465, + 148 + ], + "spans": [ + { + "bbox": [ + 115, + 136, + 465, + 148 + ], + "score": 1.0, + "content": "contrastive learning. In International Conference on Machine Learning (ICML), 2021.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 106, + 124, + 505, + 148 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 155, + 505, + 189 + ], + "lines": [ + { + "bbox": [ + 105, + 155, + 505, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 155, + 505, + 168 + ], + "score": 1.0, + "content": "Mitko Veta, Paul J Van Diest, Mehdi Jiwa, Shaimaa Al-Janabi, and Josien PW Pluim. Mitosis", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 166, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 116, + 166, + 505, + 178 + ], + "score": 1.0, + "content": "counting in breast cancer: Object-level interobserver agreement and comparison to an automatic", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 177, + 245, + 189 + ], + "spans": [ + { + "bbox": [ + 116, + 177, + 245, + 189 + ], + "score": 1.0, + "content": "method. PloS one, 11(8), 2016.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 155, + 505, + 189 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 197, + 505, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "score": 1.0, + "content": "Pascal Vincent, Hugo Larochelle, Yoshua Bengio, , and Pierre-Antoine Manzagol. Extracting and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 208, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 115, + 208, + 505, + 221 + ], + "score": 1.0, + "content": "composing robust features with denoising autoencoders. In International Conference on Machine", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 219, + 216, + 231 + ], + "spans": [ + { + "bbox": [ + 116, + 219, + 216, + 231 + ], + "score": 1.0, + "content": "Learning (ICML), 2008.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 197, + 505, + 231 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 239, + 504, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 252 + ], + "score": 1.0, + "content": "Rui Wang, Zuxuan Wu, Zejia Weng, Jingjing Chen, Guo-Jun Qi, and Yu-Gang Jiang. Cross-domain", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 250, + 399, + 262 + ], + "spans": [ + { + "bbox": [ + 115, + 250, + 399, + 262 + ], + "score": 1.0, + "content": "contrastive learning for unsupervised domain adaptation. arXiv, 2021.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 239, + 505, + 262 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 270, + 503, + 293 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 504, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 504, + 282 + ], + "score": 1.0, + "content": "Colin Wei, Sang Michael Xie, and Tengyu Ma. Why do pretrained language models help in down-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 281, + 494, + 294 + ], + "spans": [ + { + "bbox": [ + 116, + 281, + 494, + 294 + ], + "score": 1.0, + "content": "stream tasks? an analysis of head and prompt tuning. arXiv preprint arXiv:2106.09226, 2021.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 106, + 271, + 504, + 294 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 301, + 504, + 334 + ], + "lines": [ + { + "bbox": [ + 106, + 301, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 505, + 313 + ], + "score": 1.0, + "content": "Ben G Weinstein, Sergio Marconi, Stephanie Bohlman, Alina Zare, and Ethan White. Individual", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 115, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "tree-crown detection in rgb imagery using semi-supervised deep learning neural networks. Remote", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 323, + 231, + 335 + ], + "spans": [ + { + "bbox": [ + 116, + 323, + 231, + 335 + ], + "score": 1.0, + "content": "Sensing, 11(11):1309, 2019.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 106, + 301, + 505, + 335 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 342, + 505, + 377 + ], + "lines": [ + { + "bbox": [ + 105, + 342, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 505, + 356 + ], + "score": 1.0, + "content": "Ben G Weinstein, Lindsey Gardner, Vienna Saccomanno, Ashley Steinkraus, Andrew Ortega, Kris-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 354, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 115, + 354, + 505, + 367 + ], + "score": 1.0, + "content": "ten Brush, Glenda Yenni, Ann E McKellar, Rowan Converse, Christopher Lippitt, et al. A general", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 365, + 479, + 379 + ], + "spans": [ + { + "bbox": [ + 116, + 365, + 479, + 379 + ], + "score": 1.0, + "content": "deep learning model for bird detection in high resolution airborne imagery. bioRxiv, 2021.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 342, + 505, + 379 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 384, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 104, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 104, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "John N Weinstein, Eric A Collisson, Gordon B Mills, Kenna R Mills Shaw, Brad A Ozenberger,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "Kyle Ellrott, Ilya Shmulevich, Chris Sander, Joshua M Stuart, Cancer Genome Atlas Research", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 114, + 406, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 114, + 406, + 506, + 420 + ], + "score": 1.0, + "content": "Network, et al. The cancer genome atlas pan-cancer analysis project. Nature genetics, 45(10),", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 417, + 142, + 429 + ], + "spans": [ + { + "bbox": [ + 115, + 417, + 142, + 429 + ], + "score": 1.0, + "content": "2013.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 104, + 384, + 506, + 429 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 437, + 503, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "score": 1.0, + "content": "Zhenqin Wu, Bharath Ramsundar, Evan N Feinberg, Joseph Gomes, Caleb Geniesse, Aneesh S", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 449, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 115, + 449, + 505, + 461 + ], + "score": 1.0, + "content": "Pappu, Karl Leswing, and Vijay Pande. Moleculenet: a benchmark for molecular machine learn-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 460, + 294, + 472 + ], + "spans": [ + { + "bbox": [ + 115, + 460, + 294, + 472 + ], + "score": 1.0, + "content": "ing. Chemical science, 9(2):513–530, 2018.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 438, + 505, + 472 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 479, + 505, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "score": 1.0, + "content": "Michael Xie, Neal Jean, Marshall Burke, David Lobell, and Stefano Ermon. Transfer learning from", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 115, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "deep features for remote sensing and poverty mapping. In Association for the Advancement of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 502, + 262, + 514 + ], + "spans": [ + { + "bbox": [ + 116, + 502, + 262, + 514 + ], + "score": 1.0, + "content": "Artificial Intelligence (AAAI), 2016.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 479, + 506, + 514 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 521, + 503, + 545 + ], + "lines": [ + { + "bbox": [ + 106, + 520, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 505, + 535 + ], + "score": 1.0, + "content": "Qizhe Xie, Minh-Thang Luong, Eduard Hovy, and Quoc V. Le. Self-training with noisy student", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 533, + 305, + 545 + ], + "spans": [ + { + "bbox": [ + 116, + 533, + 305, + 545 + ], + "score": 1.0, + "content": "improves imagenet classification. arXiv, 2020.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 106, + 520, + 505, + 545 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 552, + 505, + 586 + ], + "lines": [ + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "score": 1.0, + "content": "Sang Michael Xie, Ananya Kumar, Robbie Jones, Fereshte Khani, Tengyu Ma, and Percy Liang.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 563, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 115, + 563, + 505, + 577 + ], + "score": 1.0, + "content": "In-N-out: Pre-training and self-training using auxiliary information for out-of-distribution robust-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 575, + 434, + 587 + ], + "spans": [ + { + "bbox": [ + 116, + 575, + 434, + 587 + ], + "score": 1.0, + "content": "ness. In International Conference on Learning Representations (ICLR), 2021a.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 551, + 505, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 594, + 504, + 617 + ], + "lines": [ + { + "bbox": [ + 106, + 593, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 506, + 608 + ], + "score": 1.0, + "content": "Yaochen Xie, Zhao Xu, Jingtun Zhang, Zhengyang Wang, and Shuiwang Ji. Self-supervised learning", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 605, + 461, + 618 + ], + "spans": [ + { + "bbox": [ + 116, + 605, + 461, + 618 + ], + "score": 1.0, + "content": "of graph neural networks: A unified review. arXiv preprint arXiv:2102.10757, 2021b.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 593, + 506, + 618 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 625, + 504, + 649 + ], + "lines": [ + { + "bbox": [ + 106, + 624, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 505, + 638 + ], + "score": 1.0, + "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 636, + 452, + 649 + ], + "spans": [ + { + "bbox": [ + 116, + 636, + 452, + 649 + ], + "score": 1.0, + "content": "networks? In International Conference on Learning Representations (ICLR), 2018.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 106, + 624, + 505, + 649 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 504, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 670 + ], + "score": 1.0, + "content": "Ruijia Xu, Guanbin Li, Jihan Yang, and Liang Lin. Larger norm more transferable: An adap-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 668, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 116, + 668, + 505, + 680 + ], + "score": 1.0, + "content": "tive feature norm approach for unsupervised domain adaptation. In International Conference on", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 678, + 310, + 692 + ], + "spans": [ + { + "bbox": [ + 116, + 678, + 310, + 692 + ], + "score": 1.0, + "content": "Computer Vision (ICCV), pp. 1426–1435, 2019.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 655, + 505, + 692 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "Christopher Yeh, Anthony Perez, Anne Driscoll, George Azzari, Zhongyi Tang, David Lobell, Ste-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "fano Ermon, and Marshall Burke. Using publicly available satellite imagery and deep learning to", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 116, + 720, + 430, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 720, + 430, + 732 + ], + "score": 1.0, + "content": "understand economic well-being in africa. Nature Communications, 11, 2020.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44, + "bbox_fs": [ + 106, + 698, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 115 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Weichen Zhang, Wanli Ouyang, Wen Li, and Dong Xu. Collaborative and adversarial network", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 92, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 115, + 92, + 505, + 108 + ], + "score": 1.0, + "content": "for unsupervised domain adaptation. In Computer Vision and Pattern Recognition (CVPR), pp.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 105, + 191, + 116 + ], + "spans": [ + { + "bbox": [ + 116, + 105, + 191, + 116 + ], + "score": 1.0, + "content": "3801–3809, 2018.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 107, + 123, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 107, + 123, + 506, + 136 + ], + "score": 1.0, + "content": "Yabin Zhang, Haojian Zhang, Bin Deng, Shuai Li, Kui Jia, and Lei Zhang. Semi-supervised models", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 135, + 489, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 135, + 489, + 146 + ], + "score": 1.0, + "content": "are strong unsupervised domain adaptation learners. arXiv preprint arXiv:2106.00417, 2021.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 153, + 505, + 198 + ], + "lines": [ + { + "bbox": [ + 106, + 153, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 505, + 166 + ], + "score": 1.0, + "content": "Yifan Zhang, Hanbo Chen, Ying Wei, Peilin Zhao, Jiezhang Cao, Xinjuan Fan, Xiaoying Lou, Hail-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 163, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 115, + 163, + 505, + 178 + ], + "score": 1.0, + "content": "ing Liu, Jinlong Hou, Xiao Han, et al. From whole slide imaging to microscopy: Deep microscopy", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 175, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 175, + 506, + 188 + ], + "score": 1.0, + "content": "adaptation network for histopathology cancer image classification. In International Conference", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 186, + 471, + 199 + ], + "spans": [ + { + "bbox": [ + 115, + 186, + 471, + 199 + ], + "score": 1.0, + "content": "on Medical Image Computing and Computer-Assisted Intervention, pp. 360–368, 2019a.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 505, + 238 + ], + "lines": [ + { + "bbox": [ + 106, + 204, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 218 + ], + "score": 1.0, + "content": "Yuchen Zhang, Tianle Liu, Mingsheng Long, and Michael Jordan. Bridging theory and algorithm for", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 215, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 115, + 215, + 505, + 229 + ], + "score": 1.0, + "content": "domain adaptation. In International Conference on Machine Learning (ICML), pp. 7404–7413,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 226, + 147, + 239 + ], + "spans": [ + { + "bbox": [ + 115, + 226, + 147, + 239 + ], + "score": 1.0, + "content": "2019b.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 245, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 505, + 259 + ], + "score": 1.0, + "content": "Han Zhao, Shanghang Zhang, Guanhang Wu, Jose MF Moura, Joao P Costeira, and Geoffrey J Gor- ´", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 256, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 115, + 256, + 505, + 271 + ], + "score": 1.0, + "content": "don. Adversarial multiple source domain adaptation. Advances in neural information processing", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 268, + 240, + 279 + ], + "spans": [ + { + "bbox": [ + 115, + 268, + 240, + 279 + ], + "score": 1.0, + "content": "systems, 31:8559–8570, 2018.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 115 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Weichen Zhang, Wanli Ouyang, Wen Li, and Dong Xu. Collaborative and adversarial network", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 92, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 115, + 92, + 505, + 108 + ], + "score": 1.0, + "content": "for unsupervised domain adaptation. In Computer Vision and Pattern Recognition (CVPR), pp.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 105, + 191, + 116 + ], + "spans": [ + { + "bbox": [ + 116, + 105, + 191, + 116 + ], + "score": 1.0, + "content": "3801–3809, 2018.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 82, + 505, + 116 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 123, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 107, + 123, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 107, + 123, + 506, + 136 + ], + "score": 1.0, + "content": "Yabin Zhang, Haojian Zhang, Bin Deng, Shuai Li, Kui Jia, and Lei Zhang. Semi-supervised models", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 135, + 489, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 135, + 489, + 146 + ], + "score": 1.0, + "content": "are strong unsupervised domain adaptation learners. arXiv preprint arXiv:2106.00417, 2021.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 107, + 123, + 506, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 153, + 505, + 198 + ], + "lines": [ + { + "bbox": [ + 106, + 153, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 505, + 166 + ], + "score": 1.0, + "content": "Yifan Zhang, Hanbo Chen, Ying Wei, Peilin Zhao, Jiezhang Cao, Xinjuan Fan, Xiaoying Lou, Hail-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 163, + 505, + 178 + ], + "spans": [ + { + "bbox": [ + 115, + 163, + 505, + 178 + ], + "score": 1.0, + "content": "ing Liu, Jinlong Hou, Xiao Han, et al. From whole slide imaging to microscopy: Deep microscopy", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 175, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 175, + 506, + 188 + ], + "score": 1.0, + "content": "adaptation network for histopathology cancer image classification. In International Conference", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 186, + 471, + 199 + ], + "spans": [ + { + "bbox": [ + 115, + 186, + 471, + 199 + ], + "score": 1.0, + "content": "on Medical Image Computing and Computer-Assisted Intervention, pp. 360–368, 2019a.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 153, + 506, + 199 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 505, + 238 + ], + "lines": [ + { + "bbox": [ + 106, + 204, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 218 + ], + "score": 1.0, + "content": "Yuchen Zhang, Tianle Liu, Mingsheng Long, and Michael Jordan. Bridging theory and algorithm for", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 215, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 115, + 215, + 505, + 229 + ], + "score": 1.0, + "content": "domain adaptation. In International Conference on Machine Learning (ICML), pp. 7404–7413,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 226, + 147, + 239 + ], + "spans": [ + { + "bbox": [ + 115, + 226, + 147, + 239 + ], + "score": 1.0, + "content": "2019b.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 204, + 505, + 239 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 245, + 505, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 505, + 259 + ], + "score": 1.0, + "content": "Han Zhao, Shanghang Zhang, Guanhang Wu, Jose MF Moura, Joao P Costeira, and Geoffrey J Gor- ´", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 256, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 115, + 256, + 505, + 271 + ], + "score": 1.0, + "content": "don. Adversarial multiple source domain adaptation. Advances in neural information processing", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 268, + 240, + 279 + ], + "spans": [ + { + "bbox": [ + 115, + 268, + 240, + 279 + ], + "score": 1.0, + "content": "systems, 31:8559–8570, 2018.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 245, + 505, + 279 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 174, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 174, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 174, + 96 + ], + "score": 1.0, + "content": "APPENDICES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 109, + 505, + 121 + ], + "lines": [ + { + "bbox": [ + 108, + 108, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 108, + 109, + 232, + 120 + ], + "score": 1.0, + "content": "A Additional dataset details", + "type": "text" + }, + { + "bbox": [ + 491, + 108, + 506, + 123 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 120, + 126, + 506, + 257 + ], + "lines": [ + { + "bbox": [ + 121, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 121, + 126, + 251, + 138 + ], + "score": 1.0, + "content": "A.1 IWILDCAM2020-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 120, + 142, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 120, + 142, + 239, + 156 + ], + "score": 1.0, + "content": "A.2 CAMELYON17-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 142, + 506, + 156 + ], + "score": 1.0, + "content": "24", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 121, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 121, + 160, + 140, + 171 + ], + "score": 1.0, + "content": "A.3", + "type": "text" + }, + { + "bbox": [ + 142, + 159, + 210, + 172 + ], + "score": 1.0, + "content": "FMOW-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 159, + 506, + 173 + ], + "score": 1.0, + "content": "25", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 121, + 176, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 121, + 177, + 140, + 188 + ], + "score": 1.0, + "content": "A.4", + "type": "text" + }, + { + "bbox": [ + 142, + 176, + 240, + 190 + ], + "score": 1.0, + "content": "POVERTYMAP-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 176, + 506, + 190 + ], + "score": 1.0, + "content": "26", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 121, + 193, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 121, + 194, + 140, + 205 + ], + "score": 1.0, + "content": "A.5", + "type": "text" + }, + { + "bbox": [ + 144, + 194, + 246, + 206 + ], + "score": 1.0, + "content": "GLOBALWHEAT-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 193, + 506, + 208 + ], + "score": 1.0, + "content": "27", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 120, + 209, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 120, + 210, + 140, + 222 + ], + "score": 1.0, + "content": "A.6", + "type": "text" + }, + { + "bbox": [ + 144, + 210, + 222, + 222 + ], + "score": 1.0, + "content": "OGB-MOLPCBA", + "type": "text" + }, + { + "bbox": [ + 491, + 209, + 506, + 224 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 121, + 226, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 121, + 228, + 140, + 239 + ], + "score": 1.0, + "content": "A.7", + "type": "text" + }, + { + "bbox": [ + 143, + 228, + 253, + 240 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 226, + 506, + 241 + ], + "score": 1.0, + "content": "32", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 120, + 244, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 120, + 244, + 218, + 257 + ], + "score": 1.0, + "content": "A.8 AMAZON-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 244, + 506, + 257 + ], + "score": 1.0, + "content": "33", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 5.5 + }, + { + "type": "title", + "bbox": [ + 107, + 272, + 505, + 284 + ], + "lines": [ + { + "bbox": [ + 108, + 270, + 506, + 285 + ], + "spans": [ + { + "bbox": [ + 108, + 270, + 199, + 284 + ], + "score": 1.0, + "content": "B Algorithm details", + "type": "text" + }, + { + "bbox": [ + 491, + 270, + 506, + 285 + ], + "score": 1.0, + "content": "35", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 120, + 288, + 506, + 352 + ], + "lines": [ + { + "bbox": [ + 120, + 287, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 120, + 287, + 290, + 301 + ], + "score": 1.0, + "content": "B.1 Empirical risk minimization (ERM)", + "type": "text" + }, + { + "bbox": [ + 492, + 288, + 506, + 300 + ], + "score": 1.0, + "content": "35", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 121, + 304, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 121, + 305, + 254, + 317 + ], + "score": 1.0, + "content": "B.2 Domain-invariant methods", + "type": "text" + }, + { + "bbox": [ + 492, + 304, + 506, + 317 + ], + "score": 1.0, + "content": "35", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 120, + 321, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 120, + 321, + 236, + 335 + ], + "score": 1.0, + "content": "B.3 Self-training methods", + "type": "text" + }, + { + "bbox": [ + 491, + 321, + 506, + 335 + ], + "score": 1.0, + "content": "37", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 120, + 338, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 120, + 338, + 252, + 352 + ], + "score": 1.0, + "content": "B.4 Self-supervision methods", + "type": "text" + }, + { + "bbox": [ + 491, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "41", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "title", + "bbox": [ + 107, + 366, + 505, + 378 + ], + "lines": [ + { + "bbox": [ + 108, + 365, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 108, + 368, + 117, + 376 + ], + "score": 1.0, + "content": "C", + "type": "text" + }, + { + "bbox": [ + 119, + 366, + 205, + 379 + ], + "score": 1.0, + "content": "Data augmentation", + "type": "text" + }, + { + "bbox": [ + 491, + 365, + 506, + 379 + ], + "score": 1.0, + "content": "43", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 393, + 504, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 507, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 212, + 404 + ], + "score": 1.0, + "content": "D Experimental details", + "type": "text" + }, + { + "bbox": [ + 491, + 391, + 507, + 407 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 120, + 410, + 506, + 507 + ], + "lines": [ + { + "bbox": [ + 120, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 120, + 410, + 348, + 423 + ], + "score": 1.0, + "content": "D.1 In-distribution vs. out-of-distribution performance", + "type": "text" + }, + { + "bbox": [ + 491, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 120, + 426, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 120, + 426, + 228, + 438 + ], + "score": 1.0, + "content": "D.2 Model architectures", + "type": "text" + }, + { + "bbox": [ + 491, + 426, + 506, + 440 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 120, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 120, + 443, + 291, + 456 + ], + "score": 1.0, + "content": "D.3 Batch sizes and batch normalization", + "type": "text" + }, + { + "bbox": [ + 491, + 443, + 506, + 456 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 121, + 460, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 121, + 461, + 140, + 472 + ], + "score": 1.0, + "content": "D.4", + "type": "text" + }, + { + "bbox": [ + 141, + 460, + 241, + 474 + ], + "score": 1.0, + "content": "Hyperparameter tuning", + "type": "text" + }, + { + "bbox": [ + 491, + 460, + 506, + 474 + ], + "score": 1.0, + "content": "45", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 120, + 475, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 120, + 475, + 293, + 492 + ], + "score": 1.0, + "content": "D.5 Algorithm-specific hyperparameters", + "type": "text" + }, + { + "bbox": [ + 491, + 477, + 506, + 491 + ], + "score": 1.0, + "content": "46", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 120, + 495, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 120, + 495, + 240, + 507 + ], + "score": 1.0, + "content": "D.6 Compute infrastructure", + "type": "text" + }, + { + "bbox": [ + 491, + 495, + 506, + 507 + ], + "score": 1.0, + "content": "46", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 105, + 520, + 506, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 520, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 243, + 533 + ], + "score": 1.0, + "content": "E Experiments on DomainNet", + "type": "text" + }, + { + "bbox": [ + 491, + 520, + 506, + 536 + ], + "score": 1.0, + "content": "47", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 547, + 507, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 294, + 560 + ], + "score": 1.0, + "content": "F Fully-labeled ERM experimental details", + "type": "text" + }, + { + "bbox": [ + 491, + 547, + 507, + 562 + ], + "score": 1.0, + "content": "49", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 573, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 317, + 588 + ], + "score": 1.0, + "content": "G Using the WILDS library with unlabeled data", + "type": "text" + }, + { + "bbox": [ + 491, + 574, + 506, + 589 + ], + "score": 1.0, + "content": "51", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + } + ], + "page_idx": 21, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 174, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 174, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 174, + 96 + ], + "score": 1.0, + "content": "APPENDICES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 109, + 505, + 121 + ], + "lines": [ + { + "bbox": [ + 108, + 108, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 108, + 109, + 232, + 120 + ], + "score": 1.0, + "content": "A Additional dataset details", + "type": "text" + }, + { + "bbox": [ + 491, + 108, + 506, + 123 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "index", + "bbox": [ + 120, + 126, + 506, + 257 + ], + "lines": [ + { + "bbox": [ + 121, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 121, + 126, + 251, + 138 + ], + "score": 1.0, + "content": "A.1 IWILDCAM2020-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 142, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 120, + 142, + 239, + 156 + ], + "score": 1.0, + "content": "A.2 CAMELYON17-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 142, + 506, + 156 + ], + "score": 1.0, + "content": "24", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 121, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 121, + 160, + 140, + 171 + ], + "score": 1.0, + "content": "A.3", + "type": "text" + }, + { + "bbox": [ + 142, + 159, + 210, + 172 + ], + "score": 1.0, + "content": "FMOW-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 159, + 506, + 173 + ], + "score": 1.0, + "content": "25", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 121, + 176, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 121, + 177, + 140, + 188 + ], + "score": 1.0, + "content": "A.4", + "type": "text" + }, + { + "bbox": [ + 142, + 176, + 240, + 190 + ], + "score": 1.0, + "content": "POVERTYMAP-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 176, + 506, + 190 + ], + "score": 1.0, + "content": "26", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 121, + 193, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 121, + 194, + 140, + 205 + ], + "score": 1.0, + "content": "A.5", + "type": "text" + }, + { + "bbox": [ + 144, + 194, + 246, + 206 + ], + "score": 1.0, + "content": "GLOBALWHEAT-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 193, + 506, + 208 + ], + "score": 1.0, + "content": "27", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 209, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 120, + 210, + 140, + 222 + ], + "score": 1.0, + "content": "A.6", + "type": "text" + }, + { + "bbox": [ + 144, + 210, + 222, + 222 + ], + "score": 1.0, + "content": "OGB-MOLPCBA", + "type": "text" + }, + { + "bbox": [ + 491, + 209, + 506, + 224 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 121, + 226, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 121, + 228, + 140, + 239 + ], + "score": 1.0, + "content": "A.7", + "type": "text" + }, + { + "bbox": [ + 143, + 228, + 253, + 240 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 226, + 506, + 241 + ], + "score": 1.0, + "content": "32", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 244, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 120, + 244, + 218, + 257 + ], + "score": 1.0, + "content": "A.8 AMAZON-WILDS", + "type": "text" + }, + { + "bbox": [ + 491, + 244, + 506, + 257 + ], + "score": 1.0, + "content": "33", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + } + ], + "index": 5.5, + "bbox_fs": [ + 120, + 126, + 506, + 257 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 272, + 505, + 284 + ], + "lines": [ + { + "bbox": [ + 108, + 270, + 506, + 285 + ], + "spans": [ + { + "bbox": [ + 108, + 270, + 199, + 284 + ], + "score": 1.0, + "content": "B Algorithm details", + "type": "text" + }, + { + "bbox": [ + 491, + 270, + 506, + 285 + ], + "score": 1.0, + "content": "35", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "index", + "bbox": [ + 120, + 288, + 506, + 352 + ], + "lines": [ + { + "bbox": [ + 120, + 287, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 120, + 287, + 290, + 301 + ], + "score": 1.0, + "content": "B.1 Empirical risk minimization (ERM)", + "type": "text" + }, + { + "bbox": [ + 492, + 288, + 506, + 300 + ], + "score": 1.0, + "content": "35", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 121, + 304, + 506, + 317 + ], + "spans": [ + { + "bbox": [ + 121, + 305, + 254, + 317 + ], + "score": 1.0, + "content": "B.2 Domain-invariant methods", + "type": "text" + }, + { + "bbox": [ + 492, + 304, + 506, + 317 + ], + "score": 1.0, + "content": "35", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 321, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 120, + 321, + 236, + 335 + ], + "score": 1.0, + "content": "B.3 Self-training methods", + "type": "text" + }, + { + "bbox": [ + 491, + 321, + 506, + 335 + ], + "score": 1.0, + "content": "37", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 338, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 120, + 338, + 252, + 352 + ], + "score": 1.0, + "content": "B.4 Self-supervision methods", + "type": "text" + }, + { + "bbox": [ + 491, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "41", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + } + ], + "index": 12.5, + "bbox_fs": [ + 120, + 287, + 506, + 352 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 366, + 505, + 378 + ], + "lines": [ + { + "bbox": [ + 108, + 365, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 108, + 368, + 117, + 376 + ], + "score": 1.0, + "content": "C", + "type": "text" + }, + { + "bbox": [ + 119, + 366, + 205, + 379 + ], + "score": 1.0, + "content": "Data augmentation", + "type": "text" + }, + { + "bbox": [ + 491, + 365, + 506, + 379 + ], + "score": 1.0, + "content": "43", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 393, + 504, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 507, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 212, + 404 + ], + "score": 1.0, + "content": "D Experimental details", + "type": "text" + }, + { + "bbox": [ + 491, + 391, + 507, + 407 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "index", + "bbox": [ + 120, + 410, + 506, + 507 + ], + "lines": [ + { + "bbox": [ + 120, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 120, + 410, + 348, + 423 + ], + "score": 1.0, + "content": "D.1 In-distribution vs. out-of-distribution performance", + "type": "text" + }, + { + "bbox": [ + 491, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 426, + 506, + 440 + ], + "spans": [ + { + "bbox": [ + 120, + 426, + 228, + 438 + ], + "score": 1.0, + "content": "D.2 Model architectures", + "type": "text" + }, + { + "bbox": [ + 491, + 426, + 506, + 440 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 120, + 443, + 291, + 456 + ], + "score": 1.0, + "content": "D.3 Batch sizes and batch normalization", + "type": "text" + }, + { + "bbox": [ + 491, + 443, + 506, + 456 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 121, + 460, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 121, + 461, + 140, + 472 + ], + "score": 1.0, + "content": "D.4", + "type": "text" + }, + { + "bbox": [ + 141, + 460, + 241, + 474 + ], + "score": 1.0, + "content": "Hyperparameter tuning", + "type": "text" + }, + { + "bbox": [ + 491, + 460, + 506, + 474 + ], + "score": 1.0, + "content": "45", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 475, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 120, + 475, + 293, + 492 + ], + "score": 1.0, + "content": "D.5 Algorithm-specific hyperparameters", + "type": "text" + }, + { + "bbox": [ + 491, + 477, + 506, + 491 + ], + "score": 1.0, + "content": "46", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 120, + 495, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 120, + 495, + 240, + 507 + ], + "score": 1.0, + "content": "D.6 Compute infrastructure", + "type": "text" + }, + { + "bbox": [ + 491, + 495, + 506, + 507 + ], + "score": 1.0, + "content": "46", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 520, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 243, + 533 + ], + "score": 1.0, + "content": "E Experiments on DomainNet", + "type": "text" + }, + { + "bbox": [ + 491, + 520, + 506, + 536 + ], + "score": 1.0, + "content": "47", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 547, + 507, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 294, + 560 + ], + "score": 1.0, + "content": "F Fully-labeled ERM experimental details", + "type": "text" + }, + { + "bbox": [ + 491, + 547, + 507, + 562 + ], + "score": 1.0, + "content": "49", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 573, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 317, + 588 + ], + "score": 1.0, + "content": "G Using the WILDS library with unlabeled data", + "type": "text" + }, + { + "bbox": [ + 491, + 574, + 506, + 589 + ], + "score": 1.0, + "content": "51", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + } + ], + "index": 19.5, + "bbox_fs": [ + 120, + 410, + 506, + 507 + ] + }, + { + "type": "index", + "bbox": [ + 105, + 520, + 506, + 588 + ], + "lines": [], + "index": 24, + "bbox_fs": [ + 105, + 520, + 507, + 589 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 109, + 82, + 290, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 291, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 291, + 95 + ], + "score": 1.0, + "content": "A ADDITIONAL DATASET DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 504, + 140 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "In this appendix, we provide additional details on the unlabeled data in WILDS 2.0. For more", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 118, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 505, + 128 + ], + "score": 1.0, + "content": "context on the motivation behind each dataset, the choice of evaluation metric, and the labeled data,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 345, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 345, + 140 + ], + "score": 1.0, + "content": "please refer to the original WILDS paper (Koh et al., 2021).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 108, + 153, + 236, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 153, + 237, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 237, + 166 + ], + "score": 1.0, + "content": "A.1 IWILDCAM2020-WILDS", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 505, + 306 + ], + "lines": [ + { + "bbox": [ + 106, + 174, + 506, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 506, + 186 + ], + "score": 1.0, + "content": "The IWILDCAM2020-WILDS dataset was adapted from the iWildCam 2020 competition dataset", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "made up of data provided by the Wildlife Conservation Society (WCS) (Beery et al., 2020) 3. Cam-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 197, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 208 + ], + "score": 1.0, + "content": "era trap images are captured by motion-triggered static cameras placed in the wild to study wildlife", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "score": 1.0, + "content": "in a non-invasive manner. Images are captured at high volumes – a single camera trap can capture", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 218, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 505, + 231 + ], + "score": 1.0, + "content": "10K images in a month – and annotating these images requires species identification expertise and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "score": 1.0, + "content": "is time-intensive. However, there are tens of thousands of camera traps worldwide capturing images", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 241, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 252 + ], + "score": 1.0, + "content": "of wildlife that could be used as unlabeled training data. For example, Wildlife Insights (Ahumada", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 250, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 104, + 250, + 505, + 264 + ], + "score": 1.0, + "content": "et al., 2020) now contains almost 20M camera trap images collected across the globe, but a large", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 262, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 104, + 262, + 506, + 275 + ], + "score": 1.0, + "content": "proportion of that data is still unlabeled. Ideally we could capture value from those images despite", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 272, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 505, + 285 + ], + "score": 1.0, + "content": "the lack of available labels. We extend IWILDCAM2020-WILDS with unlabeled data from a set of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 284, + 506, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 506, + 296 + ], + "score": 1.0, + "content": "WCS camera traps entirely disjoint with the labeled dataset, representative of unlabeled data from a", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 295, + 238, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 238, + 307 + ], + "score": 1.0, + "content": "newly-deployed sensor network.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 107, + 318, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 504, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 496, + 332 + ], + "score": 1.0, + "content": "Problem setting. The task is to classify the species of animals in camera trap images. The input", + "type": "text" + }, + { + "bbox": [ + 497, + 321, + 504, + 329 + ], + "score": 0.67, + "content": "x", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 330, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 301, + 342 + ], + "score": 1.0, + "content": "is an image from a camera trap, and the domain", + "type": "text" + }, + { + "bbox": [ + 302, + 330, + 308, + 340 + ], + "score": 0.73, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 330, + 505, + 342 + ], + "score": 1.0, + "content": "corresponds to the camera trap that captured the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 179, + 354 + ], + "score": 1.0, + "content": "image. The target", + "type": "text" + }, + { + "bbox": [ + 179, + 343, + 186, + 352 + ], + "score": 0.69, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 341, + 505, + 354 + ], + "score": 1.0, + "content": ", provided only for the labeled training images, is one of 182 classes of animals.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "We seek to learn models that generalize well to new camera trap deployments, so the test data comes", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 362, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 376 + ], + "score": 1.0, + "content": "from domains unseen during training. Additionally, we evaluate the in-distribution performance on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 373, + 310, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 310, + 387 + ], + "score": 1.0, + "content": "held-out images from camera traps in the train set.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 108, + 398, + 503, + 431 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "score": 1.0, + "content": "Data. The data comes from multiple camera traps around the world, all provided by the Wildlife", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 408, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 505, + 420 + ], + "score": 1.0, + "content": "Conservation Society (WCS). The labeled data is the same as in Koh et al. (2021) and the unlabeled", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 420, + 483, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 483, + 432 + ], + "score": 1.0, + "content": "data comprise 819,120 images from 3215 WCS camera traps not included in iWildCam 2020:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 129, + 440, + 291, + 498 + ], + "lines": [ + { + "bbox": [ + 130, + 441, + 250, + 453 + ], + "spans": [ + { + "bbox": [ + 130, + 441, + 250, + 453 + ], + "score": 1.0, + "content": "1. Source: 243 camera traps.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 128, + 455, + 292, + 469 + ], + "spans": [ + { + "bbox": [ + 128, + 455, + 292, + 469 + ], + "score": 1.0, + "content": "2. Validation (OOD): 32 camera traps.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 128, + 470, + 276, + 484 + ], + "spans": [ + { + "bbox": [ + 128, + 470, + 276, + 484 + ], + "score": 1.0, + "content": "3. Target (OOD): 48 camera traps.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 129, + 486, + 250, + 500 + ], + "spans": [ + { + "bbox": [ + 129, + 486, + 250, + 500 + ], + "score": 1.0, + "content": "4. Extra: 3215 camera traps.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 108, + 507, + 505, + 541 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 505, + 520 + ], + "score": 1.0, + "content": "The four sets of camera traps are disjoint. The distributions of the labeled and unlabeled camera", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 519, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 531 + ], + "score": 1.0, + "content": "traps are very similar, except that the labeled data does not contain cameras with photos taken before", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 530, + 227, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 227, + 541 + ], + "score": 1.0, + "content": "LandSat 8 data was available.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "table", + "bbox": [ + 108, + 549, + 504, + 676 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 108, + 549, + 504, + 676 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 549, + 504, + 676 + ], + "spans": [ + { + "bbox": [ + 108, + 549, + 504, + 676 + ], + "score": 0.744, + "html": "
Split# Domains (camera traps)#Labeled examples# Unlabeled examples
Source243129,8090
Validation (ID)7,3140
Target (ID)8,1540
Validation (OOD)14,9610
Target (OOD)32 4842,7910
Extra (OOD)32150819,120
Total3538203,029819,120
", + "type": "table", + "image_path": "19062542cd179466f80bca4a7225806e57c13f6d6f9615b69bf0b4d21c7de9cd.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 108, + 549, + 504, + 591.3333333333334 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 108, + 591.3333333333334, + 504, + 633.6666666666667 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 108, + 633.6666666666667, + 504, + 676.0000000000001 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 110, + 683, + 500, + 695 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 681, + 500, + 697 + ], + "spans": [ + { + "bbox": [ + 110, + 681, + 500, + 697 + ], + "score": 1.0, + "content": "Table 3: Data for IWILDCAM2020-WILDS. Each domain corresponds to a different camera trap.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + } + ], + "index": 35.0 + } + ], + "page_idx": 22, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 711, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 117, + 707, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 117, + 707, + 505, + 724 + ], + "score": 1.0, + "content": "3The WCS Camera Traps Dataset can be found at http://lila.science/datasets/", + "type": "text" + } + ] + }, + { + "bbox": [ + 105, + 720, + 185, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 185, + 733 + ], + "score": 1.0, + "content": "wcscameratraps", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 109, + 82, + 290, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 291, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 291, + 95 + ], + "score": 1.0, + "content": "A ADDITIONAL DATASET DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 504, + 140 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "In this appendix, we provide additional details on the unlabeled data in WILDS 2.0. For more", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 118, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 505, + 128 + ], + "score": 1.0, + "content": "context on the motivation behind each dataset, the choice of evaluation metric, and the labeled data,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 128, + 345, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 345, + 140 + ], + "score": 1.0, + "content": "please refer to the original WILDS paper (Koh et al., 2021).", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 105, + 506, + 140 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 153, + 236, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 153, + 237, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 237, + 166 + ], + "score": 1.0, + "content": "A.1 IWILDCAM2020-WILDS", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 174, + 505, + 306 + ], + "lines": [ + { + "bbox": [ + 106, + 174, + 506, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 174, + 506, + 186 + ], + "score": 1.0, + "content": "The IWILDCAM2020-WILDS dataset was adapted from the iWildCam 2020 competition dataset", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 505, + 198 + ], + "score": 1.0, + "content": "made up of data provided by the Wildlife Conservation Society (WCS) (Beery et al., 2020) 3. Cam-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 197, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 208 + ], + "score": 1.0, + "content": "era trap images are captured by motion-triggered static cameras placed in the wild to study wildlife", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 506, + 220 + ], + "score": 1.0, + "content": "in a non-invasive manner. Images are captured at high volumes – a single camera trap can capture", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 218, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 106, + 218, + 505, + 231 + ], + "score": 1.0, + "content": "10K images in a month – and annotating these images requires species identification expertise and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 506, + 242 + ], + "score": 1.0, + "content": "is time-intensive. However, there are tens of thousands of camera traps worldwide capturing images", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 241, + 505, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 252 + ], + "score": 1.0, + "content": "of wildlife that could be used as unlabeled training data. For example, Wildlife Insights (Ahumada", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 250, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 104, + 250, + 505, + 264 + ], + "score": 1.0, + "content": "et al., 2020) now contains almost 20M camera trap images collected across the globe, but a large", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 262, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 104, + 262, + 506, + 275 + ], + "score": 1.0, + "content": "proportion of that data is still unlabeled. Ideally we could capture value from those images despite", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 272, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 505, + 285 + ], + "score": 1.0, + "content": "the lack of available labels. We extend IWILDCAM2020-WILDS with unlabeled data from a set of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 284, + 506, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 506, + 296 + ], + "score": 1.0, + "content": "WCS camera traps entirely disjoint with the labeled dataset, representative of unlabeled data from a", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 295, + 238, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 238, + 307 + ], + "score": 1.0, + "content": "newly-deployed sensor network.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 10.5, + "bbox_fs": [ + 104, + 174, + 506, + 307 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 318, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 504, + 332 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 496, + 332 + ], + "score": 1.0, + "content": "Problem setting. The task is to classify the species of animals in camera trap images. The input", + "type": "text" + }, + { + "bbox": [ + 497, + 321, + 504, + 329 + ], + "score": 0.67, + "content": "x", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 330, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 301, + 342 + ], + "score": 1.0, + "content": "is an image from a camera trap, and the domain", + "type": "text" + }, + { + "bbox": [ + 302, + 330, + 308, + 340 + ], + "score": 0.73, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 330, + 505, + 342 + ], + "score": 1.0, + "content": "corresponds to the camera trap that captured the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 341, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 179, + 354 + ], + "score": 1.0, + "content": "image. The target", + "type": "text" + }, + { + "bbox": [ + 179, + 343, + 186, + 352 + ], + "score": 0.69, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 341, + 505, + 354 + ], + "score": 1.0, + "content": ", provided only for the labeled training images, is one of 182 classes of animals.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "We seek to learn models that generalize well to new camera trap deployments, so the test data comes", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 362, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 376 + ], + "score": 1.0, + "content": "from domains unseen during training. Additionally, we evaluate the in-distribution performance on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 373, + 310, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 310, + 387 + ], + "score": 1.0, + "content": "held-out images from camera traps in the train set.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 318, + 505, + 387 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 398, + 503, + 431 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "score": 1.0, + "content": "Data. The data comes from multiple camera traps around the world, all provided by the Wildlife", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 408, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 505, + 420 + ], + "score": 1.0, + "content": "Conservation Society (WCS). The labeled data is the same as in Koh et al. (2021) and the unlabeled", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 420, + 483, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 483, + 432 + ], + "score": 1.0, + "content": "data comprise 819,120 images from 3215 WCS camera traps not included in iWildCam 2020:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 397, + 505, + 432 + ] + }, + { + "type": "index", + "bbox": [ + 129, + 440, + 291, + 498 + ], + "lines": [ + { + "bbox": [ + 130, + 441, + 250, + 453 + ], + "spans": [ + { + "bbox": [ + 130, + 441, + 250, + 453 + ], + "score": 1.0, + "content": "1. Source: 243 camera traps.", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 455, + 292, + 469 + ], + "spans": [ + { + "bbox": [ + 128, + 455, + 292, + 469 + ], + "score": 1.0, + "content": "2. Validation (OOD): 32 camera traps.", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 470, + 276, + 484 + ], + "spans": [ + { + "bbox": [ + 128, + 470, + 276, + 484 + ], + "score": 1.0, + "content": "3. Target (OOD): 48 camera traps.", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 486, + 250, + 500 + ], + "spans": [ + { + "bbox": [ + 129, + 486, + 250, + 500 + ], + "score": 1.0, + "content": "4. Extra: 3215 camera traps.", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + } + ], + "index": 27.5, + "bbox_fs": [ + 128, + 441, + 292, + 500 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 507, + 505, + 541 + ], + "lines": [ + { + "bbox": [ + 106, + 506, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 505, + 520 + ], + "score": 1.0, + "content": "The four sets of camera traps are disjoint. The distributions of the labeled and unlabeled camera", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 519, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 531 + ], + "score": 1.0, + "content": "traps are very similar, except that the labeled data does not contain cameras with photos taken before", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 530, + 227, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 227, + 541 + ], + "score": 1.0, + "content": "LandSat 8 data was available.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 506, + 505, + 541 + ] + }, + { + "type": "table", + "bbox": [ + 108, + 549, + 504, + 676 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 108, + 549, + 504, + 676 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 549, + 504, + 676 + ], + "spans": [ + { + "bbox": [ + 108, + 549, + 504, + 676 + ], + "score": 0.744, + "html": "
Split# Domains (camera traps)#Labeled examples# Unlabeled examples
Source243129,8090
Validation (ID)7,3140
Target (ID)8,1540
Validation (OOD)14,9610
Target (OOD)32 4842,7910
Extra (OOD)32150819,120
Total3538203,029819,120
", + "type": "table", + "image_path": "19062542cd179466f80bca4a7225806e57c13f6d6f9615b69bf0b4d21c7de9cd.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 108, + 549, + 504, + 591.3333333333334 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 108, + 591.3333333333334, + 504, + 633.6666666666667 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 108, + 633.6666666666667, + 504, + 676.0000000000001 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 110, + 683, + 500, + 695 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 681, + 500, + 697 + ], + "spans": [ + { + "bbox": [ + 110, + 681, + 500, + 697 + ], + "score": 1.0, + "content": "Table 3: Data for IWILDCAM2020-WILDS. Each domain corresponds to a different camera trap.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + } + ], + "index": 35.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 83, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 94 + ], + "score": 1.0, + "content": "Broader context. There are large volumes of unlabeled natural world data that have been collected", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "in growing repositories such as iNaturalist (Nugent, 2018), Wildlife Insights (Ahumada et al., 2020),", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "score": 1.0, + "content": "and GBIF (Robertson et al., 2014). This data includes images or video collected by remote sensors or", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "community scientists, GPS track data from an-animal devices, aerial data from drones or satellites,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "score": 1.0, + "content": "underwater sonar, bioacoustics, and eDNA. Methods that can harness the wealth of information in", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 148 + ], + "score": 1.0, + "content": "unlabeled ecological data are well-posed to make significant breakthroughs in how we think about", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 504, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 504, + 160 + ], + "score": 1.0, + "content": "ecological and conservation-focused research. Natural-world and ecological benchmarks that pro-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "score": 1.0, + "content": "vide unlabeled data include NEWT (Van Horn et al., 2021), investigating efficient task learning, and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 505, + 182 + ], + "score": 1.0, + "content": "Semi-Supervised iNat (Su & Maji, 2021), which provides labeled data for only a subset of the taxo-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "nomic tree. Recent work has begun to adapt weakly-supervised and self-supervised approaches for", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 192, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 506, + 204 + ], + "score": 1.0, + "content": "these natural world settings, including probing the generality and efficacy of self-supervision (Cole", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 215 + ], + "score": 1.0, + "content": "et al., 2021), incorporating domain-relevant context into self-supervision (Pantazis et al., 2021), or", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 214, + 504, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 504, + 226 + ], + "score": 1.0, + "content": "leveraging weak supervision from alternative data modalities (Weinstein et al., 2019) or pre-trained,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 224, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 505, + 237 + ], + "score": 1.0, + "content": "generic models (Weinstein et al., 2021; Beery et al., 2019). Active learning also plays a role here", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 236, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 505, + 248 + ], + "score": 1.0, + "content": "in seeking to adapt models efficiently to unlabeled data from novel regions with only a few targeted", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 246, + 347, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 347, + 259 + ], + "score": 1.0, + "content": "labels (Kellenberger et al., 2019; Norouzzadeh et al., 2021).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 108, + 272, + 226, + 283 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 227, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 227, + 285 + ], + "score": 1.0, + "content": "A.2 CAMELYON17-WILDS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 293, + 505, + 391 + ], + "lines": [ + { + "bbox": [ + 106, + 293, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 505, + 304 + ], + "score": 1.0, + "content": "The CAMELYON17-WILDS dataset (Koh et al., 2021) was adapted from the Camelyon17 dataset", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "(Bandi et al., 2018), which is a collection of whole-slide images (WSIs) of breast cancer metastases", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 312, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 104, + 312, + 505, + 329 + ], + "score": 1.0, + "content": "in lymph node sections from 5 hospitals in the Netherlands. The labels were obtained by asking", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 325, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 505, + 338 + ], + "score": 1.0, + "content": "expert pathologists to perform pixel-level annotations of each WSI, which is an expensive and pain-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "staking process. In practice, unlabeled WSIs (i.e., WSIs without pixel-level annotations) are much", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "score": 1.0, + "content": "easier to obtain. For example, only a fraction of the WSIs in the original Camelyon17 dataset (Bandi", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "score": 1.0, + "content": "et al., 2018) were labeled; the other WSIs, which are taken from the same 5 hospitals, were provided", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 369, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 506, + 382 + ], + "score": 1.0, + "content": "without labels. In this work, we augment the CAMELYON17-WILDS dataset with unlabeled data", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 380, + 178, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 178, + 392 + ], + "score": 1.0, + "content": "from these WSIs.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 404, + 505, + 492 + ], + "lines": [ + { + "bbox": [ + 105, + 404, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 505, + 417 + ], + "score": 1.0, + "content": "Problem setting. The task is to classify whether a histological image patch contains any tumor", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "tissue. We consider generalizing from a set of training hospitals to new hospitals at test time. The", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 129, + 439 + ], + "score": 1.0, + "content": "input", + "type": "text" + }, + { + "bbox": [ + 130, + 428, + 137, + 436 + ], + "score": 0.72, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 426, + 208, + 439 + ], + "score": 1.0, + "content": "corresponds to a", + "type": "text" + }, + { + "bbox": [ + 208, + 426, + 238, + 437 + ], + "score": 0.87, + "content": "9 6 \\times 9 6", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "image patch extracted from an WSI of a lymph node section, the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 128, + 450 + ], + "score": 1.0, + "content": "label", + "type": "text" + }, + { + "bbox": [ + 128, + 439, + 135, + 449 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 437, + 309, + 450 + ], + "score": 1.0, + "content": "is a binary indicator of whether the central", + "type": "text" + }, + { + "bbox": [ + 309, + 437, + 338, + 448 + ], + "score": 0.89, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "patch of the input contains any pixel that", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 447, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 330, + 461 + ], + "score": 1.0, + "content": "was annotated as a tumor in the WSI, and the domain", + "type": "text" + }, + { + "bbox": [ + 330, + 449, + 337, + 458 + ], + "score": 0.73, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 447, + 505, + 461 + ], + "score": 1.0, + "content": "identifies which hospital the patch came", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "from. Each patch also includes metadata on which WSI it was extracted from, though we do not", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "use this metadata for training or evaluation. Models are evaluated by their average accuracy on a", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 481, + 216, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 481, + 216, + 492 + ], + "score": 1.0, + "content": "class-balanced test dataset.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 549 + ], + "lines": [ + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "score": 1.0, + "content": "Data. All of the labeled and unlabeled data are taken from the Camelyon17 dataset (Bandi et al.,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "2018), which consists of WSIs from 5 hospitals (domains) in the Netherlands. We provide unlabeled", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 527, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 505, + 538 + ], + "score": 1.0, + "content": "data from same domains as the labeled CAMELYON17-WILDS dataset (no extra domains). The", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 538, + 223, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 223, + 549 + ], + "score": 1.0, + "content": "domains are split as follows:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 129, + 558, + 269, + 601 + ], + "lines": [ + { + "bbox": [ + 131, + 559, + 264, + 570 + ], + "spans": [ + { + "bbox": [ + 131, + 559, + 264, + 570 + ], + "score": 1.0, + "content": "1. Source: Hospitals 1, 2, and 3.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 129, + 573, + 270, + 586 + ], + "spans": [ + { + "bbox": [ + 129, + 573, + 270, + 586 + ], + "score": 1.0, + "content": "2. Validation (OOD): Hospital 4.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 129, + 589, + 254, + 601 + ], + "spans": [ + { + "bbox": [ + 129, + 589, + 254, + 601 + ], + "score": 1.0, + "content": "3. Target (OOD): Hospital 5.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 108, + 610, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 104, + 607, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 104, + 607, + 506, + 624 + ], + "score": 1.0, + "content": "CAMELYON17-WILDS also includes a Validation (ID) set which contains data from the training", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 619, + 149, + 635 + ], + "spans": [ + { + "bbox": [ + 104, + 619, + 149, + 635 + ], + "score": 1.0, + "content": "hospitals.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "The CAMELYON17-WILDS dataset has a total of 455,954 labeled patches across these splits, derived", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "from the 10 WSIs per hospital that have full pixel-level annotations. We augment the dataset with a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "total of 2,999,307 unlabeled patches, extracted from an additional 90 unlabeled WSIs per hospital.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "score": 1.0, + "content": "There is no overlap between the WSIs used for the labeled versus unlabeled data. To extract and", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 682, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 694 + ], + "score": 1.0, + "content": "process each patch, we followed the same data processing steps that were carried out for the labeled", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 692, + 209, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 209, + 705 + ], + "score": 1.0, + "content": "data in Koh et al. (2021).", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 45.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Unlike the labeled patches, which were sampled in a class-balanced manner (i.e., half of the patches", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "have positive labels), we sampled the unlabeled patches uniformly at random from the unlabeled", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 49.5 + } + ], + "page_idx": 23, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 311, + 760 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 83, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 94 + ], + "score": 1.0, + "content": "Broader context. There are large volumes of unlabeled natural world data that have been collected", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "in growing repositories such as iNaturalist (Nugent, 2018), Wildlife Insights (Ahumada et al., 2020),", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "score": 1.0, + "content": "and GBIF (Robertson et al., 2014). This data includes images or video collected by remote sensors or", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "community scientists, GPS track data from an-animal devices, aerial data from drones or satellites,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 505, + 138 + ], + "score": 1.0, + "content": "underwater sonar, bioacoustics, and eDNA. Methods that can harness the wealth of information in", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 148 + ], + "score": 1.0, + "content": "unlabeled ecological data are well-posed to make significant breakthroughs in how we think about", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 504, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 504, + 160 + ], + "score": 1.0, + "content": "ecological and conservation-focused research. Natural-world and ecological benchmarks that pro-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "score": 1.0, + "content": "vide unlabeled data include NEWT (Van Horn et al., 2021), investigating efficient task learning, and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 505, + 182 + ], + "score": 1.0, + "content": "Semi-Supervised iNat (Su & Maji, 2021), which provides labeled data for only a subset of the taxo-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "nomic tree. Recent work has begun to adapt weakly-supervised and self-supervised approaches for", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 192, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 506, + 204 + ], + "score": 1.0, + "content": "these natural world settings, including probing the generality and efficacy of self-supervision (Cole", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 215 + ], + "score": 1.0, + "content": "et al., 2021), incorporating domain-relevant context into self-supervision (Pantazis et al., 2021), or", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 214, + 504, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 504, + 226 + ], + "score": 1.0, + "content": "leveraging weak supervision from alternative data modalities (Weinstein et al., 2019) or pre-trained,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 224, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 505, + 237 + ], + "score": 1.0, + "content": "generic models (Weinstein et al., 2021; Beery et al., 2019). Active learning also plays a role here", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 236, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 505, + 248 + ], + "score": 1.0, + "content": "in seeking to adapt models efficiently to unlabeled data from novel regions with only a few targeted", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 246, + 347, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 347, + 259 + ], + "score": 1.0, + "content": "labels (Kellenberger et al., 2019; Norouzzadeh et al., 2021).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 83, + 506, + 259 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 272, + 226, + 283 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 227, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 227, + 285 + ], + "score": 1.0, + "content": "A.2 CAMELYON17-WILDS", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 293, + 505, + 391 + ], + "lines": [ + { + "bbox": [ + 106, + 293, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 505, + 304 + ], + "score": 1.0, + "content": "The CAMELYON17-WILDS dataset (Koh et al., 2021) was adapted from the Camelyon17 dataset", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "(Bandi et al., 2018), which is a collection of whole-slide images (WSIs) of breast cancer metastases", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 312, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 104, + 312, + 505, + 329 + ], + "score": 1.0, + "content": "in lymph node sections from 5 hospitals in the Netherlands. The labels were obtained by asking", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 325, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 505, + 338 + ], + "score": 1.0, + "content": "expert pathologists to perform pixel-level annotations of each WSI, which is an expensive and pain-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "staking process. In practice, unlabeled WSIs (i.e., WSIs without pixel-level annotations) are much", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "score": 1.0, + "content": "easier to obtain. For example, only a fraction of the WSIs in the original Camelyon17 dataset (Bandi", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "score": 1.0, + "content": "et al., 2018) were labeled; the other WSIs, which are taken from the same 5 hospitals, were provided", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 369, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 506, + 382 + ], + "score": 1.0, + "content": "without labels. In this work, we augment the CAMELYON17-WILDS dataset with unlabeled data", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 380, + 178, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 178, + 392 + ], + "score": 1.0, + "content": "from these WSIs.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21, + "bbox_fs": [ + 104, + 293, + 506, + 392 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 404, + 505, + 492 + ], + "lines": [ + { + "bbox": [ + 105, + 404, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 505, + 417 + ], + "score": 1.0, + "content": "Problem setting. The task is to classify whether a histological image patch contains any tumor", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "score": 1.0, + "content": "tissue. We consider generalizing from a set of training hospitals to new hospitals at test time. The", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 129, + 439 + ], + "score": 1.0, + "content": "input", + "type": "text" + }, + { + "bbox": [ + 130, + 428, + 137, + 436 + ], + "score": 0.72, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 426, + 208, + 439 + ], + "score": 1.0, + "content": "corresponds to a", + "type": "text" + }, + { + "bbox": [ + 208, + 426, + 238, + 437 + ], + "score": 0.87, + "content": "9 6 \\times 9 6", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "image patch extracted from an WSI of a lymph node section, the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 128, + 450 + ], + "score": 1.0, + "content": "label", + "type": "text" + }, + { + "bbox": [ + 128, + 439, + 135, + 449 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 437, + 309, + 450 + ], + "score": 1.0, + "content": "is a binary indicator of whether the central", + "type": "text" + }, + { + "bbox": [ + 309, + 437, + 338, + 448 + ], + "score": 0.89, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "patch of the input contains any pixel that", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 447, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 330, + 461 + ], + "score": 1.0, + "content": "was annotated as a tumor in the WSI, and the domain", + "type": "text" + }, + { + "bbox": [ + 330, + 449, + 337, + 458 + ], + "score": 0.73, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 447, + 505, + 461 + ], + "score": 1.0, + "content": "identifies which hospital the patch came", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "from. Each patch also includes metadata on which WSI it was extracted from, though we do not", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "use this metadata for training or evaluation. Models are evaluated by their average accuracy on a", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 481, + 216, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 481, + 216, + 492 + ], + "score": 1.0, + "content": "class-balanced test dataset.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 404, + 506, + 492 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 549 + ], + "lines": [ + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "score": 1.0, + "content": "Data. All of the labeled and unlabeled data are taken from the Camelyon17 dataset (Bandi et al.,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 528 + ], + "score": 1.0, + "content": "2018), which consists of WSIs from 5 hospitals (domains) in the Netherlands. We provide unlabeled", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 527, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 505, + 538 + ], + "score": 1.0, + "content": "data from same domains as the labeled CAMELYON17-WILDS dataset (no extra domains). The", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 538, + 223, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 223, + 549 + ], + "score": 1.0, + "content": "domains are split as follows:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 504, + 505, + 549 + ] + }, + { + "type": "index", + "bbox": [ + 129, + 558, + 269, + 601 + ], + "lines": [ + { + "bbox": [ + 131, + 559, + 264, + 570 + ], + "spans": [ + { + "bbox": [ + 131, + 559, + 264, + 570 + ], + "score": 1.0, + "content": "1. Source: Hospitals 1, 2, and 3.", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 573, + 270, + 586 + ], + "spans": [ + { + "bbox": [ + 129, + 573, + 270, + 586 + ], + "score": 1.0, + "content": "2. Validation (OOD): Hospital 4.", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 589, + 254, + 601 + ], + "spans": [ + { + "bbox": [ + 129, + 589, + 254, + 601 + ], + "score": 1.0, + "content": "3. Target (OOD): Hospital 5.", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + } + ], + "index": 39, + "bbox_fs": [ + 129, + 559, + 270, + 601 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 610, + 504, + 632 + ], + "lines": [ + { + "bbox": [ + 104, + 607, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 104, + 607, + 506, + 624 + ], + "score": 1.0, + "content": "CAMELYON17-WILDS also includes a Validation (ID) set which contains data from the training", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 619, + 149, + 635 + ], + "spans": [ + { + "bbox": [ + 104, + 619, + 149, + 635 + ], + "score": 1.0, + "content": "hospitals.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5, + "bbox_fs": [ + 104, + 607, + 506, + 635 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "The CAMELYON17-WILDS dataset has a total of 455,954 labeled patches across these splits, derived", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "from the 10 WSIs per hospital that have full pixel-level annotations. We augment the dataset with a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "total of 2,999,307 unlabeled patches, extracted from an additional 90 unlabeled WSIs per hospital.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 682 + ], + "score": 1.0, + "content": "There is no overlap between the WSIs used for the labeled versus unlabeled data. To extract and", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 682, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 694 + ], + "score": 1.0, + "content": "process each patch, we followed the same data processing steps that were carried out for the labeled", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 692, + 209, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 209, + 705 + ], + "score": 1.0, + "content": "data in Koh et al. (2021).", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 637, + 506, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Unlike the labeled patches, which were sampled in a class-balanced manner (i.e., half of the patches", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "have positive labels), we sampled the unlabeled patches uniformly at random from the unlabeled", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 49.5, + "bbox_fs": [ + 106, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 116, + 80, + 495, + 180 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 116, + 80, + 495, + 180 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 80, + 495, + 180 + ], + "spans": [ + { + "bbox": [ + 116, + 80, + 495, + 180 + ], + "score": 0.983, + "html": "
Split# Domains (hospitals) # Labeled examples# Unlabeled examples
Source Validation (ID)302,436 33,5601,799,247
Validation (OOD)34,9040 600,030
Target (OOD)1 185,054600,030
Total5455,9542,999,307
", + "type": "table", + "image_path": "52b0b0bdfb1607a6f2a621e0fc268d5bc5b7ec73c46c09aeb0d90e4b09b6a825.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 80, + 495, + 113.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 113.33333333333334, + 495, + 146.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 146.66666666666669, + 495, + 180.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 122, + 187, + 487, + 198 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 186, + 487, + 200 + ], + "spans": [ + { + "bbox": [ + 123, + 186, + 487, + 200 + ], + "score": 1.0, + "content": "Table 4: Data for CAMELYON17-WILDS. Each domain corresponds to a different hospital.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 505, + 276 + ], + "lines": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "score": 1.0, + "content": "WSIs. We sampled 6,667 patches per unlabeled WSI, with the single exception of one WSI which", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "score": 1.0, + "content": "had only 5,824 valid patches, resulting in a total of 3,000,150 unlabeled patches (Table 4). While the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 243, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 505, + 256 + ], + "score": 1.0, + "content": "labeled patches were sampled in a class-balanced manner, the underlying label distribution skews", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 253, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 237, + 267 + ], + "score": 1.0, + "content": "heavily negative (approximately", + "type": "text" + }, + { + "bbox": [ + 237, + 254, + 258, + 265 + ], + "score": 0.86, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 253, + 505, + 267 + ], + "score": 1.0, + "content": "of the patches in a WSI are negative), so we expect the unla-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 264, + 358, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 358, + 277 + ], + "score": 1.0, + "content": "beled patches to be similarly skewed in their label distribution.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 290, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "Broader context. We focused on providing unlabeled data from the same hospitals (domains) as", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "score": 1.0, + "content": "in the original labeled CAMELYON17-WILDS dataset. This unlabeled data from the training and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "score": 1.0, + "content": "test hospitals can be used to develop and evaluate methods for semi-supervised learning (Peikari", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 322, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 505, + 335 + ], + "score": 1.0, + "content": "et al., 2018; Akram et al., 2018; Lu et al., 2019; Shaw et al., 2020) and domain adaptation (Ren", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "score": 1.0, + "content": "et al., 2018; Zhang et al., 2019a; Koohbanani et al., 2021), respectively. In practice, there is also a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 345, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 505, + 357 + ], + "score": 1.0, + "content": "large amount of unlabeled data from different domains that is publicly available: for example, The", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "score": 1.0, + "content": "Cancer Genome Atlas (TCGA) hosts tens of thousands of publicly-available slide images across", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 365, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 104, + 365, + 505, + 381 + ], + "score": 1.0, + "content": "a variety of cancer types and from many different hospitals (Weinstein et al., 2013). These large", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 376, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 390 + ], + "score": 1.0, + "content": "and diverse datasets need not even be directly relevant to the task at hand, e.g., one could pre-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "train a model on images for different types of cancer even if the goal were to develop a model for", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 400, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 505, + 411 + ], + "score": 1.0, + "content": "breast cancer. Recent work has started to explore the use of these large and diverse datasets for", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "computational pathology applications (Ciga et al., 2020; Dehaene et al., 2020) and in other medical", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 422, + 272, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 272, + 433 + ], + "score": 1.0, + "content": "imaging applications (Azizi et al., 2021).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 108, + 447, + 197, + 459 + ], + "lines": [ + { + "bbox": [ + 105, + 446, + 198, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 198, + 461 + ], + "score": 1.0, + "content": "A.3 FMOW-WILDS", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 469, + 505, + 546 + ], + "lines": [ + { + "bbox": [ + 106, + 468, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 505, + 480 + ], + "score": 1.0, + "content": "The FMOW-WILDS dataset (Koh et al., 2021) was adapted from the FMoW dataset (Christie et al.,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "2018), which consists of global satellite images from 2002–2018, labeled with the functional pur-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 104, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "pose of the buildings or land in the image. The labels are collected by a process which combines map", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 501, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 514 + ], + "score": 1.0, + "content": "data with crowdsourced annotations (from a trusted crowd). In contrast, unlabeled satellite imagery", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "score": 1.0, + "content": "is readily available across the globe. In this work, we augment the FMOW-WILDS dataset with", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 523, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 536 + ], + "score": 1.0, + "content": "unused satellite images that were part of the original FMoW dataset but not in the FMOW-WILDS", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 533, + 140, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 140, + 547 + ], + "score": 1.0, + "content": "dataset.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 505, + 647 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "Problem setting. The task is to classify the building or land-use type of a satellite image. We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 582 + ], + "score": 1.0, + "content": "consider generalizing from images before 2013 to after 2013, as well as considering the performance", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 581, + 504, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 497, + 594 + ], + "score": 1.0, + "content": "on the worst-case geographic region (Africa, the Americas, Oceania, Asia, or Europe). The input", + "type": "text" + }, + { + "bbox": [ + 497, + 583, + 504, + 591 + ], + "score": 0.69, + "content": "x", + "type": "inline_equation" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 592, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 212, + 604 + ], + "score": 1.0, + "content": "is an RGB satellite image (", + "type": "text" + }, + { + "bbox": [ + 212, + 592, + 253, + 603 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 592, + 324, + 604 + ], + "score": 1.0, + "content": "pixels). The label", + "type": "text" + }, + { + "bbox": [ + 325, + 594, + 331, + 604 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 592, + 505, + 604 + ], + "score": 1.0, + "content": "is one of 62 building or land use categories.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 602, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 157, + 617 + ], + "score": 1.0, + "content": "The domain", + "type": "text" + }, + { + "bbox": [ + 158, + 604, + 164, + 614 + ], + "score": 0.76, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 602, + 506, + 617 + ], + "score": 1.0, + "content": "represents both the year and the geographical region of the image. Each image also", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "includes metadata on the location and time of the image, although we do not use these except for", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "score": 1.0, + "content": "splitting the domains. Models are evaluated by their average and worst-region accuracies in the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 635, + 177, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 177, + 648 + ], + "score": 1.0, + "content": "OOD timeframe.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 108, + 661, + 504, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 660, + 504, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 504, + 673 + ], + "score": 1.0, + "content": "Data. The labeled and unlabeled data are taken from the FMoW dataset (Christie et al., 2018).", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "We provide unlabeled data from same domains as the labeled FMOW-WILDS dataset (no extra", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 107, + 684, + 263, + 694 + ], + "spans": [ + { + "bbox": [ + 107, + 684, + 263, + 694 + ], + "score": 1.0, + "content": "domains). The domains are as follows:", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 130, + 704, + 326, + 732 + ], + "lines": [ + { + "bbox": [ + 129, + 705, + 281, + 717 + ], + "spans": [ + { + "bbox": [ + 129, + 705, + 281, + 717 + ], + "score": 1.0, + "content": "1. Source: Images from 2002–2013.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 129, + 720, + 327, + 733 + ], + "spans": [ + { + "bbox": [ + 129, + 720, + 327, + 733 + ], + "score": 1.0, + "content": "2. Validation (OOD): Images from 2013–2016.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + } + ], + "page_idx": 24, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 116, + 80, + 495, + 180 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 116, + 80, + 495, + 180 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 80, + 495, + 180 + ], + "spans": [ + { + "bbox": [ + 116, + 80, + 495, + 180 + ], + "score": 0.983, + "html": "
Split# Domains (hospitals) # Labeled examples# Unlabeled examples
Source Validation (ID)302,436 33,5601,799,247
Validation (OOD)34,9040 600,030
Target (OOD)1 185,054600,030
Total5455,9542,999,307
", + "type": "table", + "image_path": "52b0b0bdfb1607a6f2a621e0fc268d5bc5b7ec73c46c09aeb0d90e4b09b6a825.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 80, + 495, + 113.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 113.33333333333334, + 495, + 146.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 146.66666666666669, + 495, + 180.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 122, + 187, + 487, + 198 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 186, + 487, + 200 + ], + "spans": [ + { + "bbox": [ + 123, + 186, + 487, + 200 + ], + "score": 1.0, + "content": "Table 4: Data for CAMELYON17-WILDS. Each domain corresponds to a different hospital.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 221, + 505, + 276 + ], + "lines": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 505, + 234 + ], + "score": 1.0, + "content": "WSIs. We sampled 6,667 patches per unlabeled WSI, with the single exception of one WSI which", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "score": 1.0, + "content": "had only 5,824 valid patches, resulting in a total of 3,000,150 unlabeled patches (Table 4). While the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 243, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 505, + 256 + ], + "score": 1.0, + "content": "labeled patches were sampled in a class-balanced manner, the underlying label distribution skews", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 253, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 237, + 267 + ], + "score": 1.0, + "content": "heavily negative (approximately", + "type": "text" + }, + { + "bbox": [ + 237, + 254, + 258, + 265 + ], + "score": 0.86, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 253, + 505, + 267 + ], + "score": 1.0, + "content": "of the patches in a WSI are negative), so we expect the unla-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 264, + 358, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 358, + 277 + ], + "score": 1.0, + "content": "beled patches to be similarly skewed in their label distribution.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 221, + 505, + 277 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 290, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "Broader context. We focused on providing unlabeled data from the same hospitals (domains) as", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "score": 1.0, + "content": "in the original labeled CAMELYON17-WILDS dataset. This unlabeled data from the training and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 505, + 324 + ], + "score": 1.0, + "content": "test hospitals can be used to develop and evaluate methods for semi-supervised learning (Peikari", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 322, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 505, + 335 + ], + "score": 1.0, + "content": "et al., 2018; Akram et al., 2018; Lu et al., 2019; Shaw et al., 2020) and domain adaptation (Ren", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "score": 1.0, + "content": "et al., 2018; Zhang et al., 2019a; Koohbanani et al., 2021), respectively. In practice, there is also a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 345, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 505, + 357 + ], + "score": 1.0, + "content": "large amount of unlabeled data from different domains that is publicly available: for example, The", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 506, + 368 + ], + "score": 1.0, + "content": "Cancer Genome Atlas (TCGA) hosts tens of thousands of publicly-available slide images across", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 365, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 104, + 365, + 505, + 381 + ], + "score": 1.0, + "content": "a variety of cancer types and from many different hospitals (Weinstein et al., 2013). These large", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 376, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 390 + ], + "score": 1.0, + "content": "and diverse datasets need not even be directly relevant to the task at hand, e.g., one could pre-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 505, + 401 + ], + "score": 1.0, + "content": "train a model on images for different types of cancer even if the goal were to develop a model for", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 400, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 505, + 411 + ], + "score": 1.0, + "content": "breast cancer. Recent work has started to explore the use of these large and diverse datasets for", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "computational pathology applications (Ciga et al., 2020; Dehaene et al., 2020) and in other medical", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 422, + 272, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 272, + 433 + ], + "score": 1.0, + "content": "imaging applications (Azizi et al., 2021).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 15, + "bbox_fs": [ + 104, + 289, + 506, + 433 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 447, + 197, + 459 + ], + "lines": [ + { + "bbox": [ + 105, + 446, + 198, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 198, + 461 + ], + "score": 1.0, + "content": "A.3 FMOW-WILDS", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 469, + 505, + 546 + ], + "lines": [ + { + "bbox": [ + 106, + 468, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 505, + 480 + ], + "score": 1.0, + "content": "The FMOW-WILDS dataset (Koh et al., 2021) was adapted from the FMoW dataset (Christie et al.,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "2018), which consists of global satellite images from 2002–2018, labeled with the functional pur-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 490, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 104, + 490, + 506, + 504 + ], + "score": 1.0, + "content": "pose of the buildings or land in the image. The labels are collected by a process which combines map", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 501, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 514 + ], + "score": 1.0, + "content": "data with crowdsourced annotations (from a trusted crowd). In contrast, unlabeled satellite imagery", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "score": 1.0, + "content": "is readily available across the globe. In this work, we augment the FMOW-WILDS dataset with", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 523, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 536 + ], + "score": 1.0, + "content": "unused satellite images that were part of the original FMoW dataset but not in the FMOW-WILDS", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 533, + 140, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 140, + 547 + ], + "score": 1.0, + "content": "dataset.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26, + "bbox_fs": [ + 104, + 468, + 506, + 547 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 505, + 647 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "Problem setting. The task is to classify the building or land-use type of a satellite image. We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 582 + ], + "score": 1.0, + "content": "consider generalizing from images before 2013 to after 2013, as well as considering the performance", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 581, + 504, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 497, + 594 + ], + "score": 1.0, + "content": "on the worst-case geographic region (Africa, the Americas, Oceania, Asia, or Europe). The input", + "type": "text" + }, + { + "bbox": [ + 497, + 583, + 504, + 591 + ], + "score": 0.69, + "content": "x", + "type": "inline_equation" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 592, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 212, + 604 + ], + "score": 1.0, + "content": "is an RGB satellite image (", + "type": "text" + }, + { + "bbox": [ + 212, + 592, + 253, + 603 + ], + "score": 0.89, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 592, + 324, + 604 + ], + "score": 1.0, + "content": "pixels). The label", + "type": "text" + }, + { + "bbox": [ + 325, + 594, + 331, + 604 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 592, + 505, + 604 + ], + "score": 1.0, + "content": "is one of 62 building or land use categories.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 602, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 157, + 617 + ], + "score": 1.0, + "content": "The domain", + "type": "text" + }, + { + "bbox": [ + 158, + 604, + 164, + 614 + ], + "score": 0.76, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 602, + 506, + 617 + ], + "score": 1.0, + "content": "represents both the year and the geographical region of the image. Each image also", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "includes metadata on the location and time of the image, although we do not use these except for", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 626, + 505, + 637 + ], + "score": 1.0, + "content": "splitting the domains. Models are evaluated by their average and worst-region accuracies in the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 635, + 177, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 635, + 177, + 648 + ], + "score": 1.0, + "content": "OOD timeframe.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 560, + 506, + 648 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 661, + 504, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 660, + 504, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 504, + 673 + ], + "score": 1.0, + "content": "Data. The labeled and unlabeled data are taken from the FMoW dataset (Christie et al., 2018).", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "We provide unlabeled data from same domains as the labeled FMOW-WILDS dataset (no extra", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 107, + 684, + 263, + 694 + ], + "spans": [ + { + "bbox": [ + 107, + 684, + 263, + 694 + ], + "score": 1.0, + "content": "domains). The domains are as follows:", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 660, + 505, + 694 + ] + }, + { + "type": "index", + "bbox": [ + 130, + 704, + 326, + 732 + ], + "lines": [ + { + "bbox": [ + 129, + 705, + 281, + 717 + ], + "spans": [ + { + "bbox": [ + 129, + 705, + 281, + 717 + ], + "score": 1.0, + "content": "1. Source: Images from 2002–2013.", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 720, + 327, + 733 + ], + "spans": [ + { + "bbox": [ + 129, + 720, + 327, + 733 + ], + "score": 1.0, + "content": "2. Validation (OOD): Images from 2013–2016.", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + } + ], + "index": 41.5, + "bbox_fs": [ + 129, + 705, + 327, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 80, + 507, + 190 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 80, + 507, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 507, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 507, + 190 + ], + "score": 0.981, + "html": "
Split# Domains (years X region)# Labeled examples# Unlabeled examples
Source11 × 576,86311,948
Validation (ID)11,483 11,3270 0
Target (ID)
Validation (OOD)3×519,915155,313
Target (OOD)2×522,108173,208
Total16×5141,696340,469
", + "type": "table", + "image_path": "7920c1371fa61c88d9e4b4dc7e54757a3a5adfb20f1a0192a9116f6b599cee83.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 80, + 507, + 116.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 116.66666666666666, + 507, + 153.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 153.33333333333331, + 507, + 189.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 198, + 505, + 220 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 197, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 211 + ], + "score": 1.0, + "content": "Table 5: Data for FMOW-WILDS. Each domain corresponds to a different year and geographical", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 103, + 208, + 138, + 222 + ], + "spans": [ + { + "bbox": [ + 103, + 208, + 138, + 222 + ], + "score": 1.0, + "content": "region.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 132, + 243, + 309, + 255 + ], + "lines": [ + { + "bbox": [ + 129, + 243, + 311, + 258 + ], + "spans": [ + { + "bbox": [ + 129, + 243, + 311, + 258 + ], + "score": 1.0, + "content": "3. Target (OOD): Images from 2016–2018.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 108, + 265, + 504, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 265, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 505, + 278 + ], + "score": 1.0, + "content": "All of these domains have disjoint locations. FMOW-WILDS also includes Validation (ID) and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 107, + 277, + 415, + 288 + ], + "spans": [ + { + "bbox": [ + 107, + 277, + 415, + 288 + ], + "score": 1.0, + "content": "Target (ID) sets which contain data from the training domains of 2002–2013.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 108, + 293, + 503, + 316 + ], + "lines": [ + { + "bbox": [ + 106, + 292, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 505, + 306 + ], + "score": 1.0, + "content": "The FMOW-WILDS dataset has 141,696 labeled images across these splits. We augment the dataset", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 303, + 392, + 317 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 392, + 317 + ], + "score": 1.0, + "content": "with 340,469 unlabeled images. These images come from two sources:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 130, + 326, + 505, + 431 + ], + "lines": [ + { + "bbox": [ + 130, + 327, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 130, + 327, + 505, + 338 + ], + "score": 1.0, + "content": "1. We use a sequestered split of the dataset, which consists of new locations that are not in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 142, + 337, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 142, + 337, + 506, + 349 + ], + "score": 1.0, + "content": "the original labeled FMOW-WILDS dataset; these unlabeled data are drawn from the same", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 349, + 268, + 360 + ], + "spans": [ + { + "bbox": [ + 142, + 349, + 268, + 360 + ], + "score": 1.0, + "content": "distribution as the labeled data.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 129, + 365, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 129, + 365, + 505, + 376 + ], + "score": 1.0, + "content": "2. For the unlabeled target and validation splits, we also add unlabeled data in their respective", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 376, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 141, + 376, + 505, + 388 + ], + "score": 1.0, + "content": "timeframes from the training set locations. While the unlabeled data from the Validation", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 385, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 141, + 385, + 506, + 400 + ], + "score": 1.0, + "content": "(OOD) and Target (OOD) domains can come from the same locations as the labeled training", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 397, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 141, + 397, + 505, + 410 + ], + "score": 1.0, + "content": "data, we note that none of the locations in the labeled Validation (OOD) or Target (OOD)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 408, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 141, + 408, + 505, + 421 + ], + "score": 1.0, + "content": "data, which is used for evaluation, is shared with any of the unlabeled or labeled data used", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 419, + 192, + 433 + ], + "spans": [ + { + "bbox": [ + 141, + 419, + 192, + 433 + ], + "score": 1.0, + "content": "for training.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 450, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 461 + ], + "score": 1.0, + "content": "Broader context. We focus on providing unlabeled data from the years (domains) that were in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "score": 1.0, + "content": "the original FMOW-WILDS dataset. Prior works have used unlabeled satellite imagery for pre-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "training (Xie et al., 2016; Jean et al., 2016; Xie et al., 2021a; Reed et al., 2021), self-training (Xie", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "et al., 2021a), and semi-supervised learning (Reed et al., 2021). Leveraging unlabeled satellite", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "score": 1.0, + "content": "imagery is powerful since it is widely available and can reduce the frequency at which we need to", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 504, + 200, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 200, + 516 + ], + "score": 1.0, + "content": "re-collect labeled data.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 108, + 531, + 227, + 542 + ], + "lines": [ + { + "bbox": [ + 106, + 530, + 227, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 227, + 544 + ], + "score": 1.0, + "content": "A.4 POVERTYMAP-WILDS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 552, + 505, + 629 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "The POVERTYMAP-WILDS dataset (Koh et al., 2021) was adapted from Yeh et al. (2020). The", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "dataset consists of satellite images from 23 African countries, labeled with a village-level real-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 505, + 587 + ], + "score": 1.0, + "content": "valued asset wealth index (measure of wealth). The labels are collected by conducting a nationally", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 586, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 597 + ], + "score": 1.0, + "content": "representative survey, which requires sending workers into the field to ask each household a number", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "of questions and can be very expensive. In contrast, unlabeled satellite imagery is readily available", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "score": 1.0, + "content": "across the globe. In this work, we augment the POVERTYMAP-WILDS dataset with satellite images", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 618, + 237, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 237, + 630 + ], + "score": 1.0, + "content": "from the same LandSat satellite.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "score": 1.0, + "content": "Problem setting. The task is to predict a real-valued asset wealth index from a satellite image.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "We consider generalizing across country borders (the dataset contains 5 different cross validation", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 331, + 678 + ], + "score": 1.0, + "content": "folds, each splitting the countries differently). The input", + "type": "text" + }, + { + "bbox": [ + 331, + 668, + 338, + 676 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "is a multispectral LandSat satellite image", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 215, + 689 + ], + "score": 1.0, + "content": "with 8 channels (resized to", + "type": "text" + }, + { + "bbox": [ + 216, + 677, + 259, + 687 + ], + "score": 0.9, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 677, + 338, + 689 + ], + "score": 1.0, + "content": "pixels). The output", + "type": "text" + }, + { + "bbox": [ + 338, + 679, + 345, + 689 + ], + "score": 0.82, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "is a real-valued asset wealth index. The", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 688, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 139, + 700 + ], + "score": 1.0, + "content": "domain", + "type": "text" + }, + { + "bbox": [ + 139, + 688, + 146, + 698 + ], + "score": 0.77, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 688, + 506, + 700 + ], + "score": 1.0, + "content": "represents the country the image was taken in, as well as whether the image was taken at", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "an urban or rural area. Each image also includes metadata on the location and time, although we do", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "not make use of these except for defining the domains. Models are evaluated by the average Pearson", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 152, + 732 + ], + "score": 1.0, + "content": "correlation", + "type": "text" + }, + { + "bbox": [ + 153, + 721, + 165, + 732 + ], + "score": 0.68, + "content": "( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "across 5 folds, as well as the lower of the Pearson correlations on the urban or rural", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36.5 + } + ], + "page_idx": 25, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 80, + 507, + 190 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 80, + 507, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 507, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 507, + 190 + ], + "score": 0.981, + "html": "
Split# Domains (years X region)# Labeled examples# Unlabeled examples
Source11 × 576,86311,948
Validation (ID)11,483 11,3270 0
Target (ID)
Validation (OOD)3×519,915155,313
Target (OOD)2×522,108173,208
Total16×5141,696340,469
", + "type": "table", + "image_path": "7920c1371fa61c88d9e4b4dc7e54757a3a5adfb20f1a0192a9116f6b599cee83.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 80, + 507, + 116.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 116.66666666666666, + 507, + 153.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 153.33333333333331, + 507, + 189.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 198, + 505, + 220 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 197, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 211 + ], + "score": 1.0, + "content": "Table 5: Data for FMOW-WILDS. Each domain corresponds to a different year and geographical", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 103, + 208, + 138, + 222 + ], + "spans": [ + { + "bbox": [ + 103, + 208, + 138, + 222 + ], + "score": 1.0, + "content": "region.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 132, + 243, + 309, + 255 + ], + "lines": [ + { + "bbox": [ + 129, + 243, + 311, + 258 + ], + "spans": [ + { + "bbox": [ + 129, + 243, + 311, + 258 + ], + "score": 1.0, + "content": "3. Target (OOD): Images from 2016–2018.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 108, + 265, + 504, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 265, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 505, + 278 + ], + "score": 1.0, + "content": "All of these domains have disjoint locations. FMOW-WILDS also includes Validation (ID) and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 107, + 277, + 415, + 288 + ], + "spans": [ + { + "bbox": [ + 107, + 277, + 415, + 288 + ], + "score": 1.0, + "content": "Target (ID) sets which contain data from the training domains of 2002–2013.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 265, + 505, + 288 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 293, + 503, + 316 + ], + "lines": [ + { + "bbox": [ + 106, + 292, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 505, + 306 + ], + "score": 1.0, + "content": "The FMOW-WILDS dataset has 141,696 labeled images across these splits. We augment the dataset", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 303, + 392, + 317 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 392, + 317 + ], + "score": 1.0, + "content": "with 340,469 unlabeled images. These images come from two sources:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 106, + 292, + 505, + 317 + ] + }, + { + "type": "list", + "bbox": [ + 130, + 326, + 505, + 431 + ], + "lines": [ + { + "bbox": [ + 130, + 327, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 130, + 327, + 505, + 338 + ], + "score": 1.0, + "content": "1. We use a sequestered split of the dataset, which consists of new locations that are not in", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 337, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 142, + 337, + 506, + 349 + ], + "score": 1.0, + "content": "the original labeled FMOW-WILDS dataset; these unlabeled data are drawn from the same", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 349, + 268, + 360 + ], + "spans": [ + { + "bbox": [ + 142, + 349, + 268, + 360 + ], + "score": 1.0, + "content": "distribution as the labeled data.", + "type": "text" + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 129, + 365, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 129, + 365, + 505, + 376 + ], + "score": 1.0, + "content": "2. For the unlabeled target and validation splits, we also add unlabeled data in their respective", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 376, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 141, + 376, + 505, + 388 + ], + "score": 1.0, + "content": "timeframes from the training set locations. While the unlabeled data from the Validation", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 385, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 141, + 385, + 506, + 400 + ], + "score": 1.0, + "content": "(OOD) and Target (OOD) domains can come from the same locations as the labeled training", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 397, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 141, + 397, + 505, + 410 + ], + "score": 1.0, + "content": "data, we note that none of the locations in the labeled Validation (OOD) or Target (OOD)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 408, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 141, + 408, + 505, + 421 + ], + "score": 1.0, + "content": "data, which is used for evaluation, is shared with any of the unlabeled or labeled data used", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 419, + 192, + 433 + ], + "spans": [ + { + "bbox": [ + 141, + 419, + 192, + 433 + ], + "score": 1.0, + "content": "for training.", + "type": "text" + } + ], + "index": 18, + "is_list_end_line": true + } + ], + "index": 14, + "bbox_fs": [ + 129, + 327, + 506, + 433 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 450, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 106, + 450, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 461 + ], + "score": 1.0, + "content": "Broader context. We focus on providing unlabeled data from the years (domains) that were in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 473 + ], + "score": 1.0, + "content": "the original FMOW-WILDS dataset. Prior works have used unlabeled satellite imagery for pre-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "training (Xie et al., 2016; Jean et al., 2016; Xie et al., 2021a; Reed et al., 2021), self-training (Xie", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 495 + ], + "score": 1.0, + "content": "et al., 2021a), and semi-supervised learning (Reed et al., 2021). Leveraging unlabeled satellite", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "score": 1.0, + "content": "imagery is powerful since it is widely available and can reduce the frequency at which we need to", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 504, + 200, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 200, + 516 + ], + "score": 1.0, + "content": "re-collect labeled data.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 450, + 506, + 516 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 531, + 227, + 542 + ], + "lines": [ + { + "bbox": [ + 106, + 530, + 227, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 227, + 544 + ], + "score": 1.0, + "content": "A.4 POVERTYMAP-WILDS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 552, + 505, + 629 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "The POVERTYMAP-WILDS dataset (Koh et al., 2021) was adapted from Yeh et al. (2020). The", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "dataset consists of satellite images from 23 African countries, labeled with a village-level real-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 575, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 505, + 587 + ], + "score": 1.0, + "content": "valued asset wealth index (measure of wealth). The labels are collected by conducting a nationally", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 586, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 597 + ], + "score": 1.0, + "content": "representative survey, which requires sending workers into the field to ask each household a number", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "of questions and can be very expensive. In contrast, unlabeled satellite imagery is readily available", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "score": 1.0, + "content": "across the globe. In this work, we augment the POVERTYMAP-WILDS dataset with satellite images", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 618, + 237, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 237, + 630 + ], + "score": 1.0, + "content": "from the same LandSat satellite.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 552, + 505, + 630 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "score": 1.0, + "content": "Problem setting. The task is to predict a real-valued asset wealth index from a satellite image.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "We consider generalizing across country borders (the dataset contains 5 different cross validation", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 331, + 678 + ], + "score": 1.0, + "content": "folds, each splitting the countries differently). The input", + "type": "text" + }, + { + "bbox": [ + 331, + 668, + 338, + 676 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "is a multispectral LandSat satellite image", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 215, + 689 + ], + "score": 1.0, + "content": "with 8 channels (resized to", + "type": "text" + }, + { + "bbox": [ + 216, + 677, + 259, + 687 + ], + "score": 0.9, + "content": "2 2 4 \\times 2 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 677, + 338, + 689 + ], + "score": 1.0, + "content": "pixels). The output", + "type": "text" + }, + { + "bbox": [ + 338, + 679, + 345, + 689 + ], + "score": 0.82, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "is a real-valued asset wealth index. The", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 688, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 139, + 700 + ], + "score": 1.0, + "content": "domain", + "type": "text" + }, + { + "bbox": [ + 139, + 688, + 146, + 698 + ], + "score": 0.77, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 688, + 506, + 700 + ], + "score": 1.0, + "content": "represents the country the image was taken in, as well as whether the image was taken at", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "an urban or rural area. Each image also includes metadata on the location and time, although we do", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "not make use of these except for defining the domains. Models are evaluated by the average Pearson", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 152, + 732 + ], + "score": 1.0, + "content": "correlation", + "type": "text" + }, + { + "bbox": [ + 153, + 721, + 165, + 732 + ], + "score": 0.68, + "content": "( r )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "across 5 folds, as well as the lower of the Pearson correlations on the urban or rural", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 243, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 256 + ], + "score": 1.0, + "content": "subpopulations to test generalization to these subpopulations. In particular, generalization to rural", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 255, + 389, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 389, + 267 + ], + "score": 1.0, + "content": "subpopulations is important as poverty is more common in rural areas.", + "type": "text", + "cross_page": true + } + ], + "index": 6 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 642, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 112, + 80, + 499, + 190 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 112, + 80, + 499, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 80, + 499, + 190 + ], + "spans": [ + { + "bbox": [ + 112, + 80, + 499, + 190 + ], + "score": 0.982, + "html": "
Split# Domains (countries × rural-urban)# Labeled ex.# Unlabeled ex.
Source13 ×29,797181,948
Validation (ID)1,0000
Target (ID)1,0000
Validation (OOD)5×23,90924,173
Target (OOD)5×23,96355,275
Total23×219,669261,396
", + "type": "table", + "image_path": "ef0ef1c99d9087e1c97003b8502ca380e9f86f0fa57106704bade106544566b9.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 80, + 499, + 116.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 116.66666666666666, + 499, + 153.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 153.33333333333331, + 499, + 189.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 107, + 198, + 504, + 220 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 196, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 505, + 212 + ], + "score": 1.0, + "content": "Table 6: Data for POVERTYMAP-WILDS (Fold A). Each domain corresponds to a different country", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 209, + 324, + 221 + ], + "spans": [ + { + "bbox": [ + 106, + 209, + 324, + 221 + ], + "score": 1.0, + "content": "and whether the image was from a rural or urban area.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 244, + 503, + 266 + ], + "lines": [ + { + "bbox": [ + 106, + 243, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 256 + ], + "score": 1.0, + "content": "subpopulations to test generalization to these subpopulations. In particular, generalization to rural", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 255, + 389, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 389, + 267 + ], + "score": 1.0, + "content": "subpopulations is important as poverty is more common in rural areas.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 105, + 280, + 504, + 302 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "score": 1.0, + "content": "Data. We provide unlabeled data from same domains as the labeled POVERTYMAP-WILDS dataset", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 291, + 321, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 321, + 302 + ], + "score": 1.0, + "content": "(no extra domains). The domains are split as follows:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 127, + 312, + 404, + 358 + ], + "lines": [ + { + "bbox": [ + 129, + 312, + 349, + 325 + ], + "spans": [ + { + "bbox": [ + 129, + 312, + 349, + 325 + ], + "score": 1.0, + "content": "1. Source: Images from training countries in the fold.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 129, + 328, + 404, + 342 + ], + "spans": [ + { + "bbox": [ + 129, + 328, + 404, + 342 + ], + "score": 1.0, + "content": "2. Validation (OOD): Images from validation countries in the fold.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 129, + 345, + 362, + 358 + ], + "spans": [ + { + "bbox": [ + 129, + 345, + 362, + 358 + ], + "score": 1.0, + "content": "3. Target (OOD): Images from test countries in the fold.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 108, + 367, + 504, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 365, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 506, + 380 + ], + "score": 1.0, + "content": "All the countries in these splits are disjoint. Folds also contain a Validation (ID) and Target (ID) set", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 379, + 257, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 379, + 257, + 390 + ], + "score": 1.0, + "content": "with data from the training countries.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 395, + 505, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "The POVERTYMAP-WILDS dataset has 19,669 labeled images across these splits. We augment the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 406, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 505, + 418 + ], + "score": 1.0, + "content": "dataset with 261,396 unlabeled images from the same 23 countries. These images are collected", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 417, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 505, + 429 + ], + "score": 1.0, + "content": "using the same process as Yeh et al. (2020) from the same LandSat satellite. The image locations", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 428, + 500, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 500, + 440 + ], + "score": 1.0, + "content": "are chosen to be roughly near survey locations from the Demographic and Health Surveys (DHS).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 453, + 505, + 552 + ], + "lines": [ + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "score": 1.0, + "content": "Broader context. We focus on providing unlabeled data from the countries (domains) that were", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "score": 1.0, + "content": "in the original POVERTYMAP-WILDS dataset. Prior works on poverty prediction have used pre-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 474, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 505, + 488 + ], + "score": 1.0, + "content": "training on unlabeled data (to predict an auxiliary task such as nighttime light prediction) (Xie", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 485, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 506, + 499 + ], + "score": 1.0, + "content": "et al., 2016; Jean et al., 2016) and for semi-supervised learning via entropy minimization (Jean", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "et al., 2018). However, these works focus on generalization to new locations in the countries in", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 508, + 504, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 504, + 520 + ], + "score": 1.0, + "content": "the training set. Poverty prediction is different from usual tasks in that the output is real-valued.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 519, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 505, + 531 + ], + "score": 1.0, + "content": "Most methods for unlabeled data are made for classification tasks, and we hope that our dataset", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 529, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 543 + ], + "score": 1.0, + "content": "will encourage more work on methods for using unlabeled data for improving OOD performance in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 542, + 174, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 174, + 553 + ], + "score": 1.0, + "content": "regression tasks.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 108, + 568, + 234, + 579 + ], + "lines": [ + { + "bbox": [ + 106, + 567, + 235, + 580 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 235, + 580 + ], + "score": 1.0, + "content": "A.5 GLOBALWHEAT-WILDS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 589, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "The GLOBALWHEAT-WILDS dataset was extended from the Global Wheat Head Dataset developed", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "by David et al. (2020; 2021). The goal of the dataset is to localize wheat heads from field images to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 611, + 504, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 504, + 623 + ], + "score": 1.0, + "content": "assist plant scientists to assess the density, size, and health of wheat heads in a particular wheat field.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "This imagery is acquired during different periods to cover the development of the vegetation, from", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 632, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 646 + ], + "score": 1.0, + "content": "the emergence to organ appearance. Examples in GLOBALWHEAT-WILDS are labeled by bounding", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "score": 1.0, + "content": "box annotations of each wheat head in the image. Wheat heads are densely packed and overlap-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "ping, making object annotation highly tedious. Thus, the Global Wheat Head Dataset (GWHD) is", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "relatively small, while in reality more field images are available. We supplement GLOBALWHEAT-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "WILDS with unlabeled examples from the same set of field vehicles and sensors but taken in different", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "acquisition sessions, i.e., at different locations or the same location in a different year. The inclusion", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "of this unlabeled data allows: 1) a much higher spatial coverage of a field location when the data", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "comes from an acquisition session which is already included, 2) a much higher temporal resolution", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "when the data comes from a location which is already included, so we have a larger range of wheat", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 34 + } + ], + "page_idx": 26, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 112, + 80, + 499, + 190 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 112, + 80, + 499, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 80, + 499, + 190 + ], + "spans": [ + { + "bbox": [ + 112, + 80, + 499, + 190 + ], + "score": 0.982, + "html": "
Split# Domains (countries × rural-urban)# Labeled ex.# Unlabeled ex.
Source13 ×29,797181,948
Validation (ID)1,0000
Target (ID)1,0000
Validation (OOD)5×23,90924,173
Target (OOD)5×23,96355,275
Total23×219,669261,396
", + "type": "table", + "image_path": "ef0ef1c99d9087e1c97003b8502ca380e9f86f0fa57106704bade106544566b9.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 112, + 80, + 499, + 116.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 112, + 116.66666666666666, + 499, + 153.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 112, + 153.33333333333331, + 499, + 189.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 107, + 198, + 504, + 220 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 196, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 505, + 212 + ], + "score": 1.0, + "content": "Table 6: Data for POVERTYMAP-WILDS (Fold A). Each domain corresponds to a different country", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 209, + 324, + 221 + ], + "spans": [ + { + "bbox": [ + 106, + 209, + 324, + 221 + ], + "score": 1.0, + "content": "and whether the image was from a rural or urban area.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 107, + 244, + 503, + 266 + ], + "lines": [], + "index": 5.5, + "bbox_fs": [ + 106, + 243, + 505, + 267 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 105, + 280, + 504, + 302 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 292 + ], + "score": 1.0, + "content": "Data. We provide unlabeled data from same domains as the labeled POVERTYMAP-WILDS dataset", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 291, + 321, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 321, + 302 + ], + "score": 1.0, + "content": "(no extra domains). The domains are split as follows:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 280, + 506, + 302 + ] + }, + { + "type": "index", + "bbox": [ + 127, + 312, + 404, + 358 + ], + "lines": [ + { + "bbox": [ + 129, + 312, + 349, + 325 + ], + "spans": [ + { + "bbox": [ + 129, + 312, + 349, + 325 + ], + "score": 1.0, + "content": "1. Source: Images from training countries in the fold.", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 328, + 404, + 342 + ], + "spans": [ + { + "bbox": [ + 129, + 328, + 404, + 342 + ], + "score": 1.0, + "content": "2. Validation (OOD): Images from validation countries in the fold.", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 345, + 362, + 358 + ], + "spans": [ + { + "bbox": [ + 129, + 345, + 362, + 358 + ], + "score": 1.0, + "content": "3. Target (OOD): Images from test countries in the fold.", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + } + ], + "index": 10, + "bbox_fs": [ + 129, + 312, + 404, + 358 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 367, + 504, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 365, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 365, + 506, + 380 + ], + "score": 1.0, + "content": "All the countries in these splits are disjoint. Folds also contain a Validation (ID) and Target (ID) set", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 379, + 257, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 379, + 257, + 390 + ], + "score": 1.0, + "content": "with data from the training countries.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 106, + 365, + 506, + 390 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 395, + 505, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "The POVERTYMAP-WILDS dataset has 19,669 labeled images across these splits. We augment the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 406, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 505, + 418 + ], + "score": 1.0, + "content": "dataset with 261,396 unlabeled images from the same 23 countries. These images are collected", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 417, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 505, + 429 + ], + "score": 1.0, + "content": "using the same process as Yeh et al. (2020) from the same LandSat satellite. The image locations", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 428, + 500, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 500, + 440 + ], + "score": 1.0, + "content": "are chosen to be roughly near survey locations from the Demographic and Health Surveys (DHS).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 395, + 505, + 440 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 453, + 505, + 552 + ], + "lines": [ + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "score": 1.0, + "content": "Broader context. We focus on providing unlabeled data from the countries (domains) that were", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "score": 1.0, + "content": "in the original POVERTYMAP-WILDS dataset. Prior works on poverty prediction have used pre-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 474, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 505, + 488 + ], + "score": 1.0, + "content": "training on unlabeled data (to predict an auxiliary task such as nighttime light prediction) (Xie", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 485, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 506, + 499 + ], + "score": 1.0, + "content": "et al., 2016; Jean et al., 2016) and for semi-supervised learning via entropy minimization (Jean", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "et al., 2018). However, these works focus on generalization to new locations in the countries in", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 508, + 504, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 504, + 520 + ], + "score": 1.0, + "content": "the training set. Poverty prediction is different from usual tasks in that the output is real-valued.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 519, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 505, + 531 + ], + "score": 1.0, + "content": "Most methods for unlabeled data are made for classification tasks, and we hope that our dataset", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 529, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 543 + ], + "score": 1.0, + "content": "will encourage more work on methods for using unlabeled data for improving OOD performance in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 542, + 174, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 174, + 553 + ], + "score": 1.0, + "content": "regression tasks.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 453, + 506, + 553 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 568, + 234, + 579 + ], + "lines": [ + { + "bbox": [ + 106, + 567, + 235, + 580 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 235, + 580 + ], + "score": 1.0, + "content": "A.5 GLOBALWHEAT-WILDS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 589, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "The GLOBALWHEAT-WILDS dataset was extended from the Global Wheat Head Dataset developed", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "by David et al. (2020; 2021). The goal of the dataset is to localize wheat heads from field images to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 611, + 504, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 504, + 623 + ], + "score": 1.0, + "content": "assist plant scientists to assess the density, size, and health of wheat heads in a particular wheat field.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "This imagery is acquired during different periods to cover the development of the vegetation, from", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 632, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 646 + ], + "score": 1.0, + "content": "the emergence to organ appearance. Examples in GLOBALWHEAT-WILDS are labeled by bounding", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "score": 1.0, + "content": "box annotations of each wheat head in the image. Wheat heads are densely packed and overlap-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "ping, making object annotation highly tedious. Thus, the Global Wheat Head Dataset (GWHD) is", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "relatively small, while in reality more field images are available. We supplement GLOBALWHEAT-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "WILDS with unlabeled examples from the same set of field vehicles and sensors but taken in different", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "acquisition sessions, i.e., at different locations or the same location in a different year. The inclusion", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "of this unlabeled data allows: 1) a much higher spatial coverage of a field location when the data", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "comes from an acquisition session which is already included, 2) a much higher temporal resolution", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "when the data comes from a location which is already included, so we have a larger range of wheat", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "score": 1.0, + "content": "growth stages, and 3) slightly more diversity when the session comes from a different location, but", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 263, + 460, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 460, + 276 + ], + "score": 1.0, + "content": "with the same image acquisition protocol (i.e., the same field vehicle and image sensor).", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 589, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 80, + 524, + 206 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 80, + 524, + 206 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 524, + 206 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 524, + 206 + ], + "score": 0.982, + "html": "
Split# Domains (acquisition session)#Labeled examples# Unlabeled examples
Source182,9435,997
Validation (ID)3570
Target (ID)3570
Validation (OOD)1,4242.000
Target (OOD)8 211,4348,997
Extra53042,445
Total1006,51559,439
", + "type": "table", + "image_path": "7a6772c1e47b3149c5abf17c8c5660e2ab84d1d92113d9d1ad0c5a39e0ea610e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 80, + 524, + 122.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 122.0, + 524, + 164.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 164.0, + 524, + 206.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 219, + 214, + 393, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 218, + 214, + 393, + 226 + ], + "spans": [ + { + "bbox": [ + 218, + 214, + 393, + 226 + ], + "score": 1.0, + "content": "Table 7: Data for GLOBALWHEAT-WILDS.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 252, + 505, + 275 + ], + "lines": [ + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "score": 1.0, + "content": "growth stages, and 3) slightly more diversity when the session comes from a different location, but", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 263, + 460, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 460, + 276 + ], + "score": 1.0, + "content": "with the same image acquisition protocol (i.e., the same field vehicle and image sensor).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 292, + 505, + 391 + ], + "lines": [ + { + "bbox": [ + 105, + 291, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 305 + ], + "score": 1.0, + "content": "Problem setting. The task is to localize wheat heads in high resolution overhead field images", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "score": 1.0, + "content": "taken from above the crop canopy. We consider generalizing across acquisition sessions represent-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "ing a particular location, time and sensor with which the images were captured. Variation across ses-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "sions includes changes in wheat genotype, wheat head appearance, growing conditions, background", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 336, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 360, + 349 + ], + "score": 1.0, + "content": "appearance, illumination and acquisition protocol. The input", + "type": "text" + }, + { + "bbox": [ + 360, + 338, + 367, + 346 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 336, + 506, + 349 + ], + "score": 1.0, + "content": "is an overhead outdoor image of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 347, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 219, + 359 + ], + "score": 1.0, + "content": "wheat canopy, and the label", + "type": "text" + }, + { + "bbox": [ + 219, + 349, + 226, + 359 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 347, + 505, + 359 + ], + "score": 1.0, + "content": "is a set of box coordinates bounding the wheat heads (the spike at the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "score": 1.0, + "content": "top of the wheat plant holding grain), omitting any hair-like awns that may extend from the head.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 367, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 157, + 383 + ], + "score": 1.0, + "content": "The domain", + "type": "text" + }, + { + "bbox": [ + 157, + 370, + 164, + 379 + ], + "score": 0.72, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 367, + 506, + 383 + ], + "score": 1.0, + "content": "designates an acquisition session, which corresponds to a certain location, time, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 380, + 171, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 171, + 392 + ], + "score": 1.0, + "content": "imaging sensor.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 409, + 504, + 443 + ], + "lines": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "score": 1.0, + "content": "Data. We provide unlabeled data from same domains as the labeled GLOBALWHEAT-WILDS", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "score": 1.0, + "content": "dataset. Additionally, we provide unlabeled data from extra acquisition sessions not in the labeled", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 430, + 441, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 441, + 444 + ], + "score": 1.0, + "content": "GLOBALWHEAT-WILDS dataset (extra domains). The domains are split as follows:", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 129, + 454, + 505, + 559 + ], + "lines": [ + { + "bbox": [ + 129, + 453, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 129, + 453, + 347, + 469 + ], + "score": 1.0, + "content": "1. Source: 18 acquisition sessions in Europe (France", + "type": "text" + }, + { + "bbox": [ + 347, + 455, + 366, + 466 + ], + "score": 0.85, + "content": "\\times 1 3", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 453, + 405, + 469 + ], + "score": 1.0, + "content": ", Norway", + "type": "text" + }, + { + "bbox": [ + 405, + 455, + 419, + 466 + ], + "score": 0.83, + "content": "\\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 453, + 506, + 469 + ], + "score": 1.0, + "content": ", Switzerland, United", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 465, + 228, + 480 + ], + "spans": [ + { + "bbox": [ + 141, + 465, + 228, + 480 + ], + "score": 1.0, + "content": "Kingdom, Belgium).", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 128, + 484, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 128, + 484, + 394, + 499 + ], + "score": 1.0, + "content": "2. Validation (OOD): 8 acquisition sessions: 7 in Asia (Japan", + "type": "text" + }, + { + "bbox": [ + 394, + 486, + 412, + 497 + ], + "score": 0.76, + "content": "\\times ~ 4", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 484, + 444, + 499 + ], + "score": 1.0, + "content": ", China", + "type": "text" + }, + { + "bbox": [ + 444, + 486, + 463, + 497 + ], + "score": 0.78, + "content": "\\times \\ 3", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 484, + 505, + 499 + ], + "score": 1.0, + "content": ") and 1 in", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 496, + 207, + 510 + ], + "spans": [ + { + "bbox": [ + 141, + 496, + 207, + 510 + ], + "score": 1.0, + "content": "Africa (Sudan).", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 128, + 515, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 128, + 515, + 471, + 530 + ], + "score": 1.0, + "content": "3. Target (OOD): 21 acquisition sessions: 11 in Australia and 10 in North America", + "type": "text" + }, + { + "bbox": [ + 471, + 516, + 505, + 528 + ], + "score": 0.3, + "content": "( \\mathrm { U S A } \\times", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 528, + 244, + 540 + ], + "spans": [ + { + "bbox": [ + 141, + 528, + 185, + 540 + ], + "score": 1.0, + "content": "6, Mexico", + "type": "text" + }, + { + "bbox": [ + 185, + 528, + 201, + 538 + ], + "score": 0.76, + "content": "\\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 528, + 244, + 540 + ], + "score": 1.0, + "content": ", Canada).", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 128, + 547, + 414, + 561 + ], + "spans": [ + { + "bbox": [ + 128, + 547, + 414, + 561 + ], + "score": 1.0, + "content": "4. Extra (OOD): 53 acquisition sessions distributed across the world.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 570, + 505, + 615 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "score": 1.0, + "content": "The source, validation, and target sessions are split by continent, while the extra sessions are taken", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 582, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 505, + 594 + ], + "score": 1.0, + "content": "from across the world. For acquisition sessions with both labeled and unlabeled data, we randomly", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 593, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 505, + 605 + ], + "score": 1.0, + "content": "selected new patches of 1024x1024 pixels from the original underlying data. The images were", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 604, + 370, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 370, + 616 + ], + "score": 1.0, + "content": "preprocessed in the same way as described in David et al. (2021).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 632, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 631, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 505, + 646 + ], + "score": 1.0, + "content": "Broader context. Utilizing unlabeled data is relatively new in the context of plant phenotyping,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 643, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 506, + 657 + ], + "score": 1.0, + "content": "due to the lack of a large, unlabeled database of plant images. However, larger plant image datasets", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "are starting become available, such as from the Terraphenotying Reference Platform (TERRA-Ref,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "Burnette et al. (2018)). Increasing the sample size and variation within plant datasets is an important", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "goal, because plants from the same species are fairly self-similar within the same field and therefore", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "increasing the number of locations, times and image types included in a dataset can be beneficial", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "for making fine-grained visual classifications for plants. Further, for plant phenotyping to be used", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "in farming applications, such as for precisely spraying weeds in a field with herbicide, models must", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 721, + 324, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 324, + 732 + ], + "score": 1.0, + "content": "be highly robust to variations between different fields.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33 + } + ], + "page_idx": 27, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "28", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 80, + 524, + 206 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 80, + 524, + 206 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 524, + 206 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 524, + 206 + ], + "score": 0.982, + "html": "
Split# Domains (acquisition session)#Labeled examples# Unlabeled examples
Source182,9435,997
Validation (ID)3570
Target (ID)3570
Validation (OOD)1,4242.000
Target (OOD)8 211,4348,997
Extra53042,445
Total1006,51559,439
", + "type": "table", + "image_path": "7a6772c1e47b3149c5abf17c8c5660e2ab84d1d92113d9d1ad0c5a39e0ea610e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 80, + 524, + 122.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 122.0, + 524, + 164.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 164.0, + 524, + 206.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 219, + 214, + 393, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 218, + 214, + 393, + 226 + ], + "spans": [ + { + "bbox": [ + 218, + 214, + 393, + 226 + ], + "score": 1.0, + "content": "Table 7: Data for GLOBALWHEAT-WILDS.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 252, + 505, + 275 + ], + "lines": [], + "index": 4.5, + "bbox_fs": [ + 105, + 252, + 506, + 276 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 292, + 505, + 391 + ], + "lines": [ + { + "bbox": [ + 105, + 291, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 305 + ], + "score": 1.0, + "content": "Problem setting. The task is to localize wheat heads in high resolution overhead field images", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 316 + ], + "score": 1.0, + "content": "taken from above the crop canopy. We consider generalizing across acquisition sessions represent-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "ing a particular location, time and sensor with which the images were captured. Variation across ses-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "sions includes changes in wheat genotype, wheat head appearance, growing conditions, background", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 336, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 360, + 349 + ], + "score": 1.0, + "content": "appearance, illumination and acquisition protocol. The input", + "type": "text" + }, + { + "bbox": [ + 360, + 338, + 367, + 346 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 336, + 506, + 349 + ], + "score": 1.0, + "content": "is an overhead outdoor image of", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 347, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 219, + 359 + ], + "score": 1.0, + "content": "wheat canopy, and the label", + "type": "text" + }, + { + "bbox": [ + 219, + 349, + 226, + 359 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 347, + 505, + 359 + ], + "score": 1.0, + "content": "is a set of box coordinates bounding the wheat heads (the spike at the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 505, + 371 + ], + "score": 1.0, + "content": "top of the wheat plant holding grain), omitting any hair-like awns that may extend from the head.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 367, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 157, + 383 + ], + "score": 1.0, + "content": "The domain", + "type": "text" + }, + { + "bbox": [ + 157, + 370, + 164, + 379 + ], + "score": 0.72, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 367, + 506, + 383 + ], + "score": 1.0, + "content": "designates an acquisition session, which corresponds to a certain location, time, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 380, + 171, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 171, + 392 + ], + "score": 1.0, + "content": "imaging sensor.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 291, + 506, + 392 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 409, + 504, + 443 + ], + "lines": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "score": 1.0, + "content": "Data. We provide unlabeled data from same domains as the labeled GLOBALWHEAT-WILDS", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "score": 1.0, + "content": "dataset. Additionally, we provide unlabeled data from extra acquisition sessions not in the labeled", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 430, + 441, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 441, + 444 + ], + "score": 1.0, + "content": "GLOBALWHEAT-WILDS dataset (extra domains). The domains are split as follows:", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 409, + 506, + 444 + ] + }, + { + "type": "list", + "bbox": [ + 129, + 454, + 505, + 559 + ], + "lines": [ + { + "bbox": [ + 129, + 453, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 129, + 453, + 347, + 469 + ], + "score": 1.0, + "content": "1. Source: 18 acquisition sessions in Europe (France", + "type": "text" + }, + { + "bbox": [ + 347, + 455, + 366, + 466 + ], + "score": 0.85, + "content": "\\times 1 3", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 453, + 405, + 469 + ], + "score": 1.0, + "content": ", Norway", + "type": "text" + }, + { + "bbox": [ + 405, + 455, + 419, + 466 + ], + "score": 0.83, + "content": "\\times 2", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 453, + 506, + 469 + ], + "score": 1.0, + "content": ", Switzerland, United", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 465, + 228, + 480 + ], + "spans": [ + { + "bbox": [ + 141, + 465, + 228, + 480 + ], + "score": 1.0, + "content": "Kingdom, Belgium).", + "type": "text" + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 484, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 128, + 484, + 394, + 499 + ], + "score": 1.0, + "content": "2. Validation (OOD): 8 acquisition sessions: 7 in Asia (Japan", + "type": "text" + }, + { + "bbox": [ + 394, + 486, + 412, + 497 + ], + "score": 0.76, + "content": "\\times ~ 4", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 484, + 444, + 499 + ], + "score": 1.0, + "content": ", China", + "type": "text" + }, + { + "bbox": [ + 444, + 486, + 463, + 497 + ], + "score": 0.78, + "content": "\\times \\ 3", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 484, + 505, + 499 + ], + "score": 1.0, + "content": ") and 1 in", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 496, + 207, + 510 + ], + "spans": [ + { + "bbox": [ + 141, + 496, + 207, + 510 + ], + "score": 1.0, + "content": "Africa (Sudan).", + "type": "text" + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 515, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 128, + 515, + 471, + 530 + ], + "score": 1.0, + "content": "3. Target (OOD): 21 acquisition sessions: 11 in Australia and 10 in North America", + "type": "text" + }, + { + "bbox": [ + 471, + 516, + 505, + 528 + ], + "score": 0.3, + "content": "( \\mathrm { U S A } \\times", + "type": "inline_equation" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 528, + 244, + 540 + ], + "spans": [ + { + "bbox": [ + 141, + 528, + 185, + 540 + ], + "score": 1.0, + "content": "6, Mexico", + "type": "text" + }, + { + "bbox": [ + 185, + 528, + 201, + 538 + ], + "score": 0.76, + "content": "\\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 528, + 244, + 540 + ], + "score": 1.0, + "content": ", Canada).", + "type": "text" + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 547, + 414, + 561 + ], + "spans": [ + { + "bbox": [ + 128, + 547, + 414, + 561 + ], + "score": 1.0, + "content": "4. Extra (OOD): 53 acquisition sessions distributed across the world.", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 21, + "bbox_fs": [ + 128, + 453, + 506, + 561 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 570, + 505, + 615 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "score": 1.0, + "content": "The source, validation, and target sessions are split by continent, while the extra sessions are taken", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 582, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 582, + 505, + 594 + ], + "score": 1.0, + "content": "from across the world. For acquisition sessions with both labeled and unlabeled data, we randomly", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 593, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 505, + 605 + ], + "score": 1.0, + "content": "selected new patches of 1024x1024 pixels from the original underlying data. The images were", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 604, + 370, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 370, + 616 + ], + "score": 1.0, + "content": "preprocessed in the same way as described in David et al. (2021).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 570, + 505, + 616 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 632, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 631, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 631, + 505, + 646 + ], + "score": 1.0, + "content": "Broader context. Utilizing unlabeled data is relatively new in the context of plant phenotyping,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 643, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 506, + 657 + ], + "score": 1.0, + "content": "due to the lack of a large, unlabeled database of plant images. However, larger plant image datasets", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 668 + ], + "score": 1.0, + "content": "are starting become available, such as from the Terraphenotying Reference Platform (TERRA-Ref,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "Burnette et al. (2018)). Increasing the sample size and variation within plant datasets is an important", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "goal, because plants from the same species are fairly self-similar within the same field and therefore", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "increasing the number of locations, times and image types included in a dataset can be beneficial", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "for making fine-grained visual classifications for plants. Further, for plant phenotyping to be used", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "in farming applications, such as for precisely spraying weeds in a field with herbicide, models must", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 721, + 324, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 324, + 732 + ], + "score": 1.0, + "content": "be highly robust to variations between different fields.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 631, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 110, + 203, + 503, + 583 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 203, + 503, + 583 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 203, + 503, + 583 + ], + "spans": [ + { + "bbox": [ + 110, + 203, + 503, + 583 + ], + "score": 0.982, + "html": "
SplitNameCountrySiteDateSensorStage#Labeled#Heads#Unlabeled
SourceArvalis_1FranceGreoux6/2/2018HandheldPF6629350
SourceArvalis_2FranceGréoux6/16/2018HandheldF401210030
SourceArvalis_3FranceGreoux7/1/2018HandheldF-R588218930
SourceArvalis_4FranceGreoux5/27/2019HandheldF20442700
SourceArvalis_5FranceVLB *6/6/2019HandheldF44881800
SourceArvalis_6FranceVSC*6/26/2019HandheldF-R16086980
SourceArvalis_7FranceVLB*6/1/2019HandheldF-R2412470
SourceArvalis_8FranceVLB *6/1/2019HandheldF-R2010620
SourceArvalis_9FranceVLB*6/1/2020HandheldR3218940
SourceArvalis_10FranceMons6/10/2020HandheldF6015631000
SourceArvalis_11FranceVLB *6/18/2020HandheldF6028180
SourceArvalis_12FranceGreoux6/15/2020HandheldF2912771000
SourceETHZ_1SwitzerlandEschikon6/6/2018SpidercamF747496030
SourceINRAE_1FranceToulouse5/28/2019HandheldF-R17636341000
SourceNMBU_1NorwayNMBU7/24/2020CartF827345999
SourceNMBU_2NorwayNMBU8/7/2020CartR985211998
SourceRres_1UKRothamsted7/13/2015GantryF-R432192100
SourceULiege_1BelgiumGembloux7/28/2020CartR3018471000
ValidationARC_1SudanWadMedani3/1/2021HandheldF3011690
ValidationNAU_1ChinaBaiman/aHandheldPF2012400
ValidationNAU_2ChinaBaima5/2/2020CartPF10049181000
ValidationNAU_3ChinaBaima5/9/2020CartF10045961000
ValidationUkyoto_1JapanKyoto4/30/2020HandheldPF6026700
ValidationUtokyo_1JapanTsukuba5/22/2018CartR538141850
ValidationUtokyo_2JapanTsukuba5/22/2018CartR456130100
ValidationUtokyo_3JapanHokkaido6/16/2021Handheldmultiple12030850
TargetCIMMYT_1MexicoCiudadObregon3/24/2020CartPF6928431000
TargetCIMMYT_2MexicoCiudadObregon3/19/2020CartPF7727711000
TargetCIMMYT_3MexicoCiudadObregon3/23/2020CartPF6015611000
TargetKSU_1USManhattan,KS5/19/2016TractorPF10064351000
TargetKSU_2USManhattan,KS5/12/2017TractorPF10053021000
TargetKSUL3USManhattan,KS5/25/2017TractorF9552171000
TargetKSU_4USManhattan,KS5/25/2017TractorR6032851000
TargetTerraref_1USMaricopa4/2/2020GantryR1443360997
TargetTerraref_2USMaricopa3/20/2020GantryF10612741000
TargetUQ-1AustraliaGatton8/12/2015TractorPF226400
TargetUQ-2AustraliaGatton9/8/2015TractorPF16390
TargetUQ-3AustraliaGatton9/15/2015TractorF142970
TargetUQ-4AustraliaGatton10/1/2015TractorF3010390
TargetUQ.5AustraliaGatton10/9/2015TractorF-R303680 11470
Target TargetUQ-6 UQ-7Australia AustraliaGatton Gatton10/14/2015 10/6/2020Tractor HandheldF-R R30 17
", + "type": "table", + "image_path": "10d9d9f9c255494ff7a9baf1ea0abd60576fd2e2a842ac37717725fe954f9e51.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 203, + 503, + 329.6666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 329.6666666666667, + 503, + 456.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 456.33333333333337, + 503, + 583.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 156, + 591, + 454, + 602 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 155, + 590, + 456, + 604 + ], + "spans": [ + { + "bbox": [ + 155, + 590, + 456, + 604 + ], + "score": 1.0, + "content": "Table 8: Source, validation, and test domains for GLOBALWHEAT-WILDS.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 28, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 110, + 203, + 503, + 583 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 203, + 503, + 583 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 203, + 503, + 583 + ], + "spans": [ + { + "bbox": [ + 110, + 203, + 503, + 583 + ], + "score": 0.982, + "html": "
SplitNameCountrySiteDateSensorStage#Labeled#Heads#Unlabeled
SourceArvalis_1FranceGreoux6/2/2018HandheldPF6629350
SourceArvalis_2FranceGréoux6/16/2018HandheldF401210030
SourceArvalis_3FranceGreoux7/1/2018HandheldF-R588218930
SourceArvalis_4FranceGreoux5/27/2019HandheldF20442700
SourceArvalis_5FranceVLB *6/6/2019HandheldF44881800
SourceArvalis_6FranceVSC*6/26/2019HandheldF-R16086980
SourceArvalis_7FranceVLB*6/1/2019HandheldF-R2412470
SourceArvalis_8FranceVLB *6/1/2019HandheldF-R2010620
SourceArvalis_9FranceVLB*6/1/2020HandheldR3218940
SourceArvalis_10FranceMons6/10/2020HandheldF6015631000
SourceArvalis_11FranceVLB *6/18/2020HandheldF6028180
SourceArvalis_12FranceGreoux6/15/2020HandheldF2912771000
SourceETHZ_1SwitzerlandEschikon6/6/2018SpidercamF747496030
SourceINRAE_1FranceToulouse5/28/2019HandheldF-R17636341000
SourceNMBU_1NorwayNMBU7/24/2020CartF827345999
SourceNMBU_2NorwayNMBU8/7/2020CartR985211998
SourceRres_1UKRothamsted7/13/2015GantryF-R432192100
SourceULiege_1BelgiumGembloux7/28/2020CartR3018471000
ValidationARC_1SudanWadMedani3/1/2021HandheldF3011690
ValidationNAU_1ChinaBaiman/aHandheldPF2012400
ValidationNAU_2ChinaBaima5/2/2020CartPF10049181000
ValidationNAU_3ChinaBaima5/9/2020CartF10045961000
ValidationUkyoto_1JapanKyoto4/30/2020HandheldPF6026700
ValidationUtokyo_1JapanTsukuba5/22/2018CartR538141850
ValidationUtokyo_2JapanTsukuba5/22/2018CartR456130100
ValidationUtokyo_3JapanHokkaido6/16/2021Handheldmultiple12030850
TargetCIMMYT_1MexicoCiudadObregon3/24/2020CartPF6928431000
TargetCIMMYT_2MexicoCiudadObregon3/19/2020CartPF7727711000
TargetCIMMYT_3MexicoCiudadObregon3/23/2020CartPF6015611000
TargetKSU_1USManhattan,KS5/19/2016TractorPF10064351000
TargetKSU_2USManhattan,KS5/12/2017TractorPF10053021000
TargetKSUL3USManhattan,KS5/25/2017TractorF9552171000
TargetKSU_4USManhattan,KS5/25/2017TractorR6032851000
TargetTerraref_1USMaricopa4/2/2020GantryR1443360997
TargetTerraref_2USMaricopa3/20/2020GantryF10612741000
TargetUQ-1AustraliaGatton8/12/2015TractorPF226400
TargetUQ-2AustraliaGatton9/8/2015TractorPF16390
TargetUQ-3AustraliaGatton9/15/2015TractorF142970
TargetUQ-4AustraliaGatton10/1/2015TractorF3010390
TargetUQ.5AustraliaGatton10/9/2015TractorF-R303680 11470
Target TargetUQ-6 UQ-7Australia AustraliaGatton Gatton10/14/2015 10/6/2020Tractor HandheldF-R R30 17
", + "type": "table", + "image_path": "10d9d9f9c255494ff7a9baf1ea0abd60576fd2e2a842ac37717725fe954f9e51.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 203, + 503, + 329.6666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 329.6666666666667, + 503, + 456.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 456.33333333333337, + 503, + 583.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 156, + 591, + 454, + 602 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 155, + 590, + 456, + 604 + ], + "spans": [ + { + "bbox": [ + 155, + 590, + 456, + 604 + ], + "score": 1.0, + "content": "Table 8: Source, validation, and test domains for GLOBALWHEAT-WILDS.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 110, + 170, + 503, + 615 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 170, + 503, + 615 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 170, + 503, + 615 + ], + "spans": [ + { + "bbox": [ + 110, + 170, + 503, + 615 + ], + "score": 0.981, + "html": "
Split Name Country Site Date Sensor Stage #Labeled #Heads #Unlabeled
Extra Arvalis_13 France Mons 6/15/2018 Handheld F-R 0 0 995
ExtraArvalis_14 France Gréoux 5/25/2020 Handheld F 0 0 1000
ExtraArvalis_15 France VLB* 6/2/2020 Handheld F 0 0 1000
ExtraArvalis_16 France Gréoux 6/22/2020HandheldF-R 0 0 1000
ExtraArvalis_17 France Bignan 5/18/2021HandheldF-R 0 0 1000
ExtraArvalis_18 France VLB* 5/28/2021HandheldPF0 0 1000
ExtraArvalis_19 France Encrambade 6/2/2021Handheld F0 0 1000
ExtraArvalis_20 FranceOLM*6/2/2021HandheldF0 0 1000
ExtraArvalis_21 FranceEncrambade6/11/2021HandheldPF0 0 1000
ExtraArvalis_22 FranceVLB*6/14/2021HandheldF0 0 1000
ExtraArvalis_23 FranceOLM*6/17/2021HandheldF-R0 0 1000
ExtraCIMMYT_4 MexicoCiudadObregon3/11/2020CartF0 0 1000
ExtraCIMMYT_5 MexicoCiudadObregon3/12/2020CartF0 0 1000
ExtraCIMMYT_6 MexicoCiudadObregon3/13/2020CartF0 0 1000
ExtraCIMMYT_7 MexicoCiudadObregon3/13/2020CartF0 0 1000
ExtraCIMMYT_8 MexicoCiudadObregon3/13/2020CartF0 0 1000
ExtraCIMMYT_9 MexicoCiudadObregon3/19/2020CartF0 0 1000
ExtraCIMMYT_10 MexicoCiudadObregon4/15/2020CartE0 0 1000
ExtraCIMMYT_11 MexicoCiudadObregon4/22/2020CartE0 0 1000
ExtraCIMMYT_12 MexicoCiudadObregon4/22/2020CartE00 1000
ExtraCIMMYT_13MexicoCiudadObregon4/22/2020CartE001000
ExtraCIMMYT_14MexicoCiudadObregon4/22/2020CartPF001000
ExtraCIMMYT_15MexicoCiudadObregon4/28/2020CartPF001000
ExtraCIMMYT_16MexicoCiudadObregon5/3/2020CartF-R001000
ExtraETHZ_2SwitzerlandEschikon6/6/2018SpidercamF00750
ExtraINRAE_2FranceClermont-Ferrand5/29/2019HandheldF001000
ExtraKSU_5USManhattan,KS5/4/2016TractorF001000
ExtraKSU_6USManhattan,KS4/23/2017TractorP-F001000
ExtraRres_2UKRothamsted7/7/2015GantryR001000
ExtraRres_3UKRothamsted7/10/2015GantryF001000
ExtraRres_4UKRothamsted7/13/2015GantryF-R001000
ExtraRres.5UKRothamsted7/20/2015GantryF-R001000
ExtraULiege_2BelgiumGembloux6/11/2020CartPF001000
ExtraULiege_3BelgiumGembloux6/15/2020CartF001000
ExtraULiege_4BelgiumGembloux6/16/2020CartF001000
ExtraULiege_5BelgiumGembloux6/18/2020CartF001000
ExtraULiege_6BelgiumGembloux6/23/2020CartF001000
ExtraULiege_7BelgiumGembloux6/26/2020CartF001000
ExtraULiege_8BelgiumGembloux7/7/2020CartF-R001000
ExtraULiege_9BelgiumGembloux7/13/2020CartF-R001000
ExtraUsask_2CanadaSaskatchewan8/6/2019TractorF00800
ExtraUsask_3CanadaSaskatchewan8/12/2019TractorF-R00 800
ExtraUtokyo_4JapanHokkaido6/7/2021HandheldPF00 100
ExtraUtokyo.5JapanHokkaido6/9/2021HandheldF00 100
ExtraUtokyo_6JapanHokkaido6/16/2021HandheldPF00 100
ExtraUtokyo_7JapanHokkaido6/23/2021HandheldF00 100
ExtraUtokyo_8JapanHokkaido7/3/2021HandheldF00 100
ExtraUtokyo.9JapanHokkaido7/10/2021HandheldF00 100
ExtraUtokyo_10 JapanHokkaido7/10/2021HandheldF-R00 100
ExtraUtokyo_11 JapanHokkaido7/11/2021HandheldF-R00 100
ExtraUtokyo_12 JapanHokkaido7/20/2021HandheldR00 100
ExtraUtokyo_13 Japan Hokkaido 7/20/2021Handheld R00 100
ExtraUtokyo_14 Japan Hokkaido 7/28/2021 Handheld R 0 0 100
", + "type": "table", + "image_path": "a0d67484d9b05e23f10d20dd160ac1472e4e65c2230f8af44be846928dd1d04b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 170, + 503, + 318.33333333333337 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 318.33333333333337, + 503, + 466.66666666666674 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 466.66666666666674, + 503, + 615.0000000000001 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 199, + 623, + 412, + 634 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 622, + 413, + 635 + ], + "spans": [ + { + "bbox": [ + 198, + 622, + 413, + 635 + ], + "score": 1.0, + "content": "Table 9: Extra domains for GLOBALWHEAT-WILDS.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 29, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "30", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 110, + 170, + 503, + 615 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 110, + 170, + 503, + 615 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 170, + 503, + 615 + ], + "spans": [ + { + "bbox": [ + 110, + 170, + 503, + 615 + ], + "score": 0.981, + "html": "
Split Name Country Site Date Sensor Stage #Labeled #Heads #Unlabeled
Extra Arvalis_13 France Mons 6/15/2018 Handheld F-R 0 0 995
ExtraArvalis_14 France Gréoux 5/25/2020 Handheld F 0 0 1000
ExtraArvalis_15 France VLB* 6/2/2020 Handheld F 0 0 1000
ExtraArvalis_16 France Gréoux 6/22/2020HandheldF-R 0 0 1000
ExtraArvalis_17 France Bignan 5/18/2021HandheldF-R 0 0 1000
ExtraArvalis_18 France VLB* 5/28/2021HandheldPF0 0 1000
ExtraArvalis_19 France Encrambade 6/2/2021Handheld F0 0 1000
ExtraArvalis_20 FranceOLM*6/2/2021HandheldF0 0 1000
ExtraArvalis_21 FranceEncrambade6/11/2021HandheldPF0 0 1000
ExtraArvalis_22 FranceVLB*6/14/2021HandheldF0 0 1000
ExtraArvalis_23 FranceOLM*6/17/2021HandheldF-R0 0 1000
ExtraCIMMYT_4 MexicoCiudadObregon3/11/2020CartF0 0 1000
ExtraCIMMYT_5 MexicoCiudadObregon3/12/2020CartF0 0 1000
ExtraCIMMYT_6 MexicoCiudadObregon3/13/2020CartF0 0 1000
ExtraCIMMYT_7 MexicoCiudadObregon3/13/2020CartF0 0 1000
ExtraCIMMYT_8 MexicoCiudadObregon3/13/2020CartF0 0 1000
ExtraCIMMYT_9 MexicoCiudadObregon3/19/2020CartF0 0 1000
ExtraCIMMYT_10 MexicoCiudadObregon4/15/2020CartE0 0 1000
ExtraCIMMYT_11 MexicoCiudadObregon4/22/2020CartE0 0 1000
ExtraCIMMYT_12 MexicoCiudadObregon4/22/2020CartE00 1000
ExtraCIMMYT_13MexicoCiudadObregon4/22/2020CartE001000
ExtraCIMMYT_14MexicoCiudadObregon4/22/2020CartPF001000
ExtraCIMMYT_15MexicoCiudadObregon4/28/2020CartPF001000
ExtraCIMMYT_16MexicoCiudadObregon5/3/2020CartF-R001000
ExtraETHZ_2SwitzerlandEschikon6/6/2018SpidercamF00750
ExtraINRAE_2FranceClermont-Ferrand5/29/2019HandheldF001000
ExtraKSU_5USManhattan,KS5/4/2016TractorF001000
ExtraKSU_6USManhattan,KS4/23/2017TractorP-F001000
ExtraRres_2UKRothamsted7/7/2015GantryR001000
ExtraRres_3UKRothamsted7/10/2015GantryF001000
ExtraRres_4UKRothamsted7/13/2015GantryF-R001000
ExtraRres.5UKRothamsted7/20/2015GantryF-R001000
ExtraULiege_2BelgiumGembloux6/11/2020CartPF001000
ExtraULiege_3BelgiumGembloux6/15/2020CartF001000
ExtraULiege_4BelgiumGembloux6/16/2020CartF001000
ExtraULiege_5BelgiumGembloux6/18/2020CartF001000
ExtraULiege_6BelgiumGembloux6/23/2020CartF001000
ExtraULiege_7BelgiumGembloux6/26/2020CartF001000
ExtraULiege_8BelgiumGembloux7/7/2020CartF-R001000
ExtraULiege_9BelgiumGembloux7/13/2020CartF-R001000
ExtraUsask_2CanadaSaskatchewan8/6/2019TractorF00800
ExtraUsask_3CanadaSaskatchewan8/12/2019TractorF-R00 800
ExtraUtokyo_4JapanHokkaido6/7/2021HandheldPF00 100
ExtraUtokyo.5JapanHokkaido6/9/2021HandheldF00 100
ExtraUtokyo_6JapanHokkaido6/16/2021HandheldPF00 100
ExtraUtokyo_7JapanHokkaido6/23/2021HandheldF00 100
ExtraUtokyo_8JapanHokkaido7/3/2021HandheldF00 100
ExtraUtokyo.9JapanHokkaido7/10/2021HandheldF00 100
ExtraUtokyo_10 JapanHokkaido7/10/2021HandheldF-R00 100
ExtraUtokyo_11 JapanHokkaido7/11/2021HandheldF-R00 100
ExtraUtokyo_12 JapanHokkaido7/20/2021HandheldR00 100
ExtraUtokyo_13 Japan Hokkaido 7/20/2021Handheld R00 100
ExtraUtokyo_14 Japan Hokkaido 7/28/2021 Handheld R 0 0 100
", + "type": "table", + "image_path": "a0d67484d9b05e23f10d20dd160ac1472e4e65c2230f8af44be846928dd1d04b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 170, + 503, + 318.33333333333337 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 318.33333333333337, + 503, + 466.66666666666674 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 466.66666666666674, + 503, + 615.0000000000001 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 199, + 623, + 412, + 634 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 622, + 413, + 635 + ], + "spans": [ + { + "bbox": [ + 198, + 622, + 413, + 635 + ], + "score": 1.0, + "content": "Table 9: Extra domains for GLOBALWHEAT-WILDS.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 116, + 80, + 496, + 168 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 116, + 80, + 496, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 80, + 496, + 168 + ], + "spans": [ + { + "bbox": [ + 116, + 80, + 496, + 168 + ], + "score": 0.982, + "html": "
Split# Domains (scaffolds)# Labeled examples# Unlabeled examples
Source44,930350,3434,052,627
Validation (OOD)31,36143,793430,325
Target (OOD)43,79343,793517,048
Total120,084437,9295,000,000
", + "type": "table", + "image_path": "d6bc86c34f892e70d17c5599fd9faf15b9e071d58bdbe0259455de7c75ae12eb.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 80, + 496, + 109.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 109.33333333333333, + 496, + 138.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 138.66666666666666, + 496, + 168.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 108, + 176, + 505, + 198 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 505, + 187 + ], + "score": 1.0, + "content": "Table 10: Data for OGB-MOLPCBA. Each domain corresponds to a different molecule scaffold", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 188, + 146, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 146, + 199 + ], + "score": 1.0, + "content": "structure.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 108, + 223, + 209, + 234 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 210, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 210, + 237 + ], + "score": 1.0, + "content": "A.6 OGB-MOLPCBA", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 505, + 355 + ], + "lines": [ + { + "bbox": [ + 106, + 244, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 505, + 257 + ], + "score": 1.0, + "content": "The OGB-MOLPCBA dataset was adapted from the Open Graph Benchmark (Hu et al., 2020b) and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 254, + 506, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 506, + 269 + ], + "score": 1.0, + "content": "originally curated by the MoleculeNet (Wu et al., 2018) from the PubChem database (Bolton et al.,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 264, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 281 + ], + "score": 1.0, + "content": "2008). The dataset is a collection of molecules annotated with 128 kinds of binary labels indicating", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 278, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 278, + 506, + 291 + ], + "score": 1.0, + "content": "the outcome of different biological assays. Performing biological assays is expensive, and as a", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 288, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 506, + 301 + ], + "score": 1.0, + "content": "result, the assay labels are only sparsely available over a tiny portion of the molecules curated in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 505, + 312 + ], + "score": 1.0, + "content": "the large-scale PubChem database (Bolton et al., 2008). On the other hand, unlabeled molecule", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 309, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 506, + 325 + ], + "score": 1.0, + "content": "data is abundant and readily available from the database. Prior work in graph machine learning", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "has leveraged unlabeled molecules to perform pre-training (Hu et al., 2020c) and semi-supervised", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 332, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 506, + 344 + ], + "score": 1.0, + "content": "learning (Sun et al., 2020). In this work, we augment the OGB-MOLPCBA dataset with unlabeled", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 344, + 316, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 316, + 356 + ], + "score": 1.0, + "content": "molecules subsampled from the PubChem database.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 371, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 384 + ], + "score": 1.0, + "content": "Problem setting. The task is multi-task molecule classification, and we consider generalizing to", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 382, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 331, + 394 + ], + "score": 1.0, + "content": "new molecule scaffold structures at test time. The input", + "type": "text" + }, + { + "bbox": [ + 331, + 384, + 338, + 392 + ], + "score": 0.65, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 382, + 505, + 394 + ], + "score": 1.0, + "content": "corresponds to a molecular graph (where", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 392, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 350, + 406 + ], + "score": 1.0, + "content": "nodes are atoms and edges are chemical bounds), the label", + "type": "text" + }, + { + "bbox": [ + 350, + 394, + 357, + 405 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 392, + 505, + 406 + ], + "score": 1.0, + "content": "is a 128-dimensional binary vector,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 372, + 416 + ], + "score": 1.0, + "content": "representing the binary outcomes of the biological assay results.", + "type": "text" + }, + { + "bbox": [ + 372, + 406, + 379, + 415 + ], + "score": 0.63, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "could contain NaN values, in-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "dicating that the corresponding biological assays were not performed on the given molecule. The", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 138, + 438 + ], + "score": 1.0, + "content": "domain", + "type": "text" + }, + { + "bbox": [ + 138, + 426, + 145, + 435 + ], + "score": 0.76, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "indicates the scaffold group a molecule belongs to. As the binary labels are highly-skewed,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 437, + 428, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 428, + 449 + ], + "score": 1.0, + "content": "the model’s classification performance is evaluated using the Average Precision.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 463, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 463, + 504, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 504, + 475 + ], + "score": 1.0, + "content": "Data. All of the labeled and unlabeled data are taken from the PubChem database (Bolton et al.,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 475, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 505, + 486 + ], + "score": 1.0, + "content": "2008). We provide unlabeled data from same domains as the labeled OGB-MOLPCBA dataset", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 486, + 504, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 504, + 497 + ], + "score": 1.0, + "content": "(no extra domains). We curate the unlabeled data by randomly sampling 5 million molecules from", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 497, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 504, + 508 + ], + "score": 1.0, + "content": "the PubChem database. We then assign these unlabeled molecules to the existing labeled scaffold", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "groups that contain the most similar molecules. Specifically, we first compute the 1024-dimensional", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "score": 1.0, + "content": "Morgan fingerprints for all the molecules (Rogers & Hahn, 2010; Landrum et al., 2006). Then,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "score": 1.0, + "content": "for each unlabeled molecule, we compute its Jaccard similarity against all the labeled molecules in", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "OGB-MOLPCBA and obtain a labeled molecule with the highest Jaccard similarity. Finally, we", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 550, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 565 + ], + "score": 1.0, + "content": "assign the unlabeled molecule to the scaffold group that the most similar labeled molecule belongs", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 563, + 496, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 496, + 574 + ], + "score": 1.0, + "content": "to. This way, the molecules within the same scaffold groups are structurally similar to each other.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 578, + 352, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 578, + 354, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 354, + 591 + ], + "score": 1.0, + "content": "The domains in the OGB-MOLPCBA dataset are as follows:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 129, + 601, + 319, + 650 + ], + "lines": [ + { + "bbox": [ + 129, + 601, + 274, + 615 + ], + "spans": [ + { + "bbox": [ + 129, + 601, + 274, + 615 + ], + "score": 1.0, + "content": "1. Source: 44,930 scaffold groups.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 128, + 619, + 320, + 633 + ], + "spans": [ + { + "bbox": [ + 128, + 619, + 320, + 633 + ], + "score": 1.0, + "content": "2. Validation (OOD): 31,361 scaffold groups.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 129, + 637, + 303, + 652 + ], + "spans": [ + { + "bbox": [ + 129, + 637, + 303, + 652 + ], + "score": 1.0, + "content": "3. Target (OOD): 43,793 scaffold groups.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 660, + 504, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "score": 1.0, + "content": "The largest scaffolds are in the source split and the smallest scaffolds in the target split. We assign", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 671, + 474, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 474, + 683 + ], + "score": 1.0, + "content": "all of the unlabeled molecules to the existing domains, so there are no extra domains added.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "While the unlabeled data are similar to the labeled data in that they were all derived from PubChem", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "(Bolton et al., 2008), it is quite possible that there was some selection bias in which molecules", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "in PubChem were chosen to be labeled, which would lead to an undocumented distribution shift", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 721, + 283, + 731 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 283, + 731 + ], + "score": 1.0, + "content": "between the unlabeled and labeled datasets.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5 + } + ], + "page_idx": 30, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 116, + 80, + 496, + 168 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 116, + 80, + 496, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 80, + 496, + 168 + ], + "spans": [ + { + "bbox": [ + 116, + 80, + 496, + 168 + ], + "score": 0.982, + "html": "
Split# Domains (scaffolds)# Labeled examples# Unlabeled examples
Source44,930350,3434,052,627
Validation (OOD)31,36143,793430,325
Target (OOD)43,79343,793517,048
Total120,084437,9295,000,000
", + "type": "table", + "image_path": "d6bc86c34f892e70d17c5599fd9faf15b9e071d58bdbe0259455de7c75ae12eb.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 80, + 496, + 109.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 109.33333333333333, + 496, + 138.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 138.66666666666666, + 496, + 168.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 108, + 176, + 505, + 198 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 175, + 505, + 187 + ], + "spans": [ + { + "bbox": [ + 105, + 175, + 505, + 187 + ], + "score": 1.0, + "content": "Table 10: Data for OGB-MOLPCBA. Each domain corresponds to a different molecule scaffold", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 188, + 146, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 146, + 199 + ], + "score": 1.0, + "content": "structure.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 108, + 223, + 209, + 234 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 210, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 210, + 237 + ], + "score": 1.0, + "content": "A.6 OGB-MOLPCBA", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 505, + 355 + ], + "lines": [ + { + "bbox": [ + 106, + 244, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 505, + 257 + ], + "score": 1.0, + "content": "The OGB-MOLPCBA dataset was adapted from the Open Graph Benchmark (Hu et al., 2020b) and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 254, + 506, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 506, + 269 + ], + "score": 1.0, + "content": "originally curated by the MoleculeNet (Wu et al., 2018) from the PubChem database (Bolton et al.,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 264, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 281 + ], + "score": 1.0, + "content": "2008). The dataset is a collection of molecules annotated with 128 kinds of binary labels indicating", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 278, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 278, + 506, + 291 + ], + "score": 1.0, + "content": "the outcome of different biological assays. Performing biological assays is expensive, and as a", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 288, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 506, + 301 + ], + "score": 1.0, + "content": "result, the assay labels are only sparsely available over a tiny portion of the molecules curated in", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 505, + 312 + ], + "score": 1.0, + "content": "the large-scale PubChem database (Bolton et al., 2008). On the other hand, unlabeled molecule", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 309, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 506, + 325 + ], + "score": 1.0, + "content": "data is abundant and readily available from the database. Prior work in graph machine learning", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "has leveraged unlabeled molecules to perform pre-training (Hu et al., 2020c) and semi-supervised", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 332, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 506, + 344 + ], + "score": 1.0, + "content": "learning (Sun et al., 2020). In this work, we augment the OGB-MOLPCBA dataset with unlabeled", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 344, + 316, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 316, + 356 + ], + "score": 1.0, + "content": "molecules subsampled from the PubChem database.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 244, + 506, + 356 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 371, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 384 + ], + "score": 1.0, + "content": "Problem setting. The task is multi-task molecule classification, and we consider generalizing to", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 382, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 331, + 394 + ], + "score": 1.0, + "content": "new molecule scaffold structures at test time. The input", + "type": "text" + }, + { + "bbox": [ + 331, + 384, + 338, + 392 + ], + "score": 0.65, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 382, + 505, + 394 + ], + "score": 1.0, + "content": "corresponds to a molecular graph (where", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 392, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 350, + 406 + ], + "score": 1.0, + "content": "nodes are atoms and edges are chemical bounds), the label", + "type": "text" + }, + { + "bbox": [ + 350, + 394, + 357, + 405 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 392, + 505, + 406 + ], + "score": 1.0, + "content": "is a 128-dimensional binary vector,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 372, + 416 + ], + "score": 1.0, + "content": "representing the binary outcomes of the biological assay results.", + "type": "text" + }, + { + "bbox": [ + 372, + 406, + 379, + 415 + ], + "score": 0.63, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "could contain NaN values, in-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "dicating that the corresponding biological assays were not performed on the given molecule. The", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 138, + 438 + ], + "score": 1.0, + "content": "domain", + "type": "text" + }, + { + "bbox": [ + 138, + 426, + 145, + 435 + ], + "score": 0.76, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "indicates the scaffold group a molecule belongs to. As the binary labels are highly-skewed,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 437, + 428, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 428, + 449 + ], + "score": 1.0, + "content": "the model’s classification performance is evaluated using the Average Precision.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 370, + 505, + 449 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 463, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 463, + 504, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 504, + 475 + ], + "score": 1.0, + "content": "Data. All of the labeled and unlabeled data are taken from the PubChem database (Bolton et al.,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 475, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 505, + 486 + ], + "score": 1.0, + "content": "2008). We provide unlabeled data from same domains as the labeled OGB-MOLPCBA dataset", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 486, + 504, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 504, + 497 + ], + "score": 1.0, + "content": "(no extra domains). We curate the unlabeled data by randomly sampling 5 million molecules from", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 497, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 106, + 497, + 504, + 508 + ], + "score": 1.0, + "content": "the PubChem database. We then assign these unlabeled molecules to the existing labeled scaffold", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "groups that contain the most similar molecules. Specifically, we first compute the 1024-dimensional", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "score": 1.0, + "content": "Morgan fingerprints for all the molecules (Rogers & Hahn, 2010; Landrum et al., 2006). Then,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "score": 1.0, + "content": "for each unlabeled molecule, we compute its Jaccard similarity against all the labeled molecules in", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "OGB-MOLPCBA and obtain a labeled molecule with the highest Jaccard similarity. Finally, we", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 550, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 505, + 565 + ], + "score": 1.0, + "content": "assign the unlabeled molecule to the scaffold group that the most similar labeled molecule belongs", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 563, + 496, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 496, + 574 + ], + "score": 1.0, + "content": "to. This way, the molecules within the same scaffold groups are structurally similar to each other.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 463, + 505, + 574 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 578, + 352, + 590 + ], + "lines": [ + { + "bbox": [ + 106, + 578, + 354, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 354, + 591 + ], + "score": 1.0, + "content": "The domains in the OGB-MOLPCBA dataset are as follows:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 578, + 354, + 591 + ] + }, + { + "type": "index", + "bbox": [ + 129, + 601, + 319, + 650 + ], + "lines": [ + { + "bbox": [ + 129, + 601, + 274, + 615 + ], + "spans": [ + { + "bbox": [ + 129, + 601, + 274, + 615 + ], + "score": 1.0, + "content": "1. Source: 44,930 scaffold groups.", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 619, + 320, + 633 + ], + "spans": [ + { + "bbox": [ + 128, + 619, + 320, + 633 + ], + "score": 1.0, + "content": "2. Validation (OOD): 31,361 scaffold groups.", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 637, + 303, + 652 + ], + "spans": [ + { + "bbox": [ + 129, + 637, + 303, + 652 + ], + "score": 1.0, + "content": "3. Target (OOD): 43,793 scaffold groups.", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + } + ], + "index": 35, + "bbox_fs": [ + 128, + 601, + 320, + 652 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 660, + 504, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "score": 1.0, + "content": "The largest scaffolds are in the source split and the smallest scaffolds in the target split. We assign", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 671, + 474, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 474, + 683 + ], + "score": 1.0, + "content": "all of the unlabeled molecules to the existing domains, so there are no extra domains added.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 659, + 505, + 683 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "While the unlabeled data are similar to the labeled data in that they were all derived from PubChem", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "(Bolton et al., 2008), it is quite possible that there was some selection bias in which molecules", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "in PubChem were chosen to be labeled, which would lead to an undocumented distribution shift", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 721, + 283, + 731 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 283, + 731 + ], + "score": 1.0, + "content": "between the unlabeled and labeled datasets.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5, + "bbox_fs": [ + 106, + 687, + 505, + 731 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Broader context. We focused on providing unlabeled data for both training and OOD test do-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "mains. Unlabeled molecules can be used to develop and evaluate methods for domain adaptation,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "score": 1.0, + "content": "self-training, as well as pre-training (Hu et al., 2020c) and semi-supervised learning (Sun et al.,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "2020). In terms of future directions, we think it is fruitful to explore both graph-agnostic meth-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 139 + ], + "score": 1.0, + "content": "ods (e.g., pseudo-label training) and more graph-specific methods (e.g., self-supervised learning of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 280, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 280, + 150 + ], + "score": 1.0, + "content": "graph neural networks (Xie et al., 2021b)).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 163, + 241, + 174 + ], + "lines": [ + { + "bbox": [ + 106, + 162, + 242, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 162, + 242, + 176 + ], + "score": 1.0, + "content": "A.7 CIVILCOMMENTS-WILDS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 184, + 505, + 294 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "score": 1.0, + "content": "The CIVILCOMMENTS-WILDS dataset (Koh et al., 2021) was adapted from the CivilComments", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 194, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 505, + 207 + ], + "score": 1.0, + "content": "dataset (Borkan et al., 2019), which is a collection of text comments made on online articles. The", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 205, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 505, + 219 + ], + "score": 1.0, + "content": "data in CIVILCOMMENTS-WILDS underwent a significant labeling and annotation process: each ex-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "ample was labeled toxic or non-toxic and annotated for whether they mentioned certain demographic", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 227, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 505, + 240 + ], + "score": 1.0, + "content": "identities by at least 10 crowdworkers. Such a substantial labeling and identity annotation process is", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 239, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 250 + ], + "score": 1.0, + "content": "expensive and time-consuming. On the other hand, unlabeled, unannotated text comments are read-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "ily available. For example, CIVILCOMMENTS-WILDS only contains a subset of all data available in", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "the original CivilComments dataset (Borkan et al., 2019), most of which Koh et al. (2021) excluded", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "because these examples were not annotated for mentioning identities. In this work, we augment the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 282, + 430, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 430, + 295 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS dataset with these unlabeled, unannotated comments.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 107, + 307, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 480, + 320 + ], + "score": 1.0, + "content": "Problem setting. The task is to classify whether a text comment is toxic or not. The input", + "type": "text" + }, + { + "bbox": [ + 480, + 310, + 487, + 317 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 307, + 506, + 320 + ], + "score": 1.0, + "content": "is a", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 319, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 453, + 331 + ], + "score": 1.0, + "content": "text comment (at least one sentence long) originally made on an online article, the label", + "type": "text" + }, + { + "bbox": [ + 454, + 320, + 461, + 330 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 319, + 505, + 331 + ], + "score": 1.0, + "content": "is a binary", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 328, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 389, + 343 + ], + "score": 1.0, + "content": "indicator of whether the comment is rated toxic or not, and the domain", + "type": "text" + }, + { + "bbox": [ + 389, + 330, + 396, + 339 + ], + "score": 0.79, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 328, + 505, + 343 + ], + "score": 1.0, + "content": "is an 8-dimensional binary", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 340, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 352 + ], + "score": 1.0, + "content": "vector, where each dimension corresponds to whether the comment mentions each of 8 demographic", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "identities: male, female, LGBTQ, Christian, Muslim, other religions, Black, or White, respectively.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "Each comment also includes metadata on which article the comment was made on, although we do", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 374, + 297, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 297, + 385 + ], + "score": 1.0, + "content": "not use this metadata for training or evaluation.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 390, + 505, + 434 + ], + "lines": [ + { + "bbox": [ + 105, + 389, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 505, + 403 + ], + "score": 1.0, + "content": "We consider the subpopulation shift setting, where the model must perform well across all subpop-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 401, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 256, + 414 + ], + "score": 1.0, + "content": "ulations, which are defined based on", + "type": "text" + }, + { + "bbox": [ + 257, + 402, + 263, + 411 + ], + "score": 0.43, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 401, + 506, + 414 + ], + "score": 1.0, + "content": ". Koh et al. (2021) define 16 subpopulations (groups) based", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 412, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 119, + 425 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 119, + 413, + 126, + 422 + ], + "score": 0.47, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 412, + 506, + 425 + ], + "score": 1.0, + "content": ". Models are then evaluated by their worst-group accuracy, i.e., the lowest accuracy over the 16", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 423, + 374, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 374, + 436 + ], + "score": 1.0, + "content": "groups considered. In our work, we use the same evaluation setup.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 448, + 505, + 546 + ], + "lines": [ + { + "bbox": [ + 106, + 448, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 505, + 459 + ], + "score": 1.0, + "content": "Data. All of the labeled and unlabeled data are taken from the CivilComments dataset (Borkan", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 458, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 505, + 470 + ], + "score": 1.0, + "content": "et al., 2019). After preprocessing, Koh et al. (2021) created the CIVILCOMMENTS-WILDS dataset", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 399, + 482 + ], + "score": 1.0, + "content": "using the 448,000 examples that were fully annotated for both toxicity", + "type": "text" + }, + { + "bbox": [ + 400, + 472, + 407, + 482 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "and the mention of de-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 192, + 492 + ], + "score": 1.0, + "content": "mographic identities", + "type": "text" + }, + { + "bbox": [ + 193, + 481, + 200, + 491 + ], + "score": 0.52, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 480, + 505, + 492 + ], + "score": 1.0, + "content": ". In this work, we augment CIVILCOMMENTS-WILDS with an additional", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "1,551,515 examples collected by Borkan et al. (2019). We use these examples as unlabeled data.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 501, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 506, + 516 + ], + "score": 1.0, + "content": "We follow the same preprocessing steps as was done with the labeled data in Koh et al. (2021). The", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 344, + 525 + ], + "score": 1.0, + "content": "resulting unlabeled examples have no identity annotations", + "type": "text" + }, + { + "bbox": [ + 344, + 514, + 351, + 523 + ], + "score": 0.78, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 513, + 438, + 525 + ], + "score": 1.0, + "content": "and no toxicity label", + "type": "text" + }, + { + "bbox": [ + 438, + 515, + 445, + 525 + ], + "score": 0.77, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 513, + 505, + 525 + ], + "score": 1.0, + "content": ". We note that", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "Borkan et al. (2019) actually do provide toxicity labels for these examples in the original CivilCom-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 535, + 465, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 465, + 547 + ], + "score": 1.0, + "content": "ments dataset, but we ignore these labels and use them neither for training nor evaluation.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 552, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "Because our unlabeled examples have no identity annotations, we cannot group these examples as", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "score": 1.0, + "content": "Koh et al. (2021) group the labeled examples; thus we refer to this data as unlabeled data coming", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "score": 1.0, + "content": "from extra domains (Table 11). In practice, these comments may actually mention any number of", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 584, + 149, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 149, + 597 + ], + "score": 1.0, + "content": "identities.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 108, + 602, + 505, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 257, + 614 + ], + "score": 1.0, + "content": "A substantial amount (1,427,848 or", + "type": "text" + }, + { + "bbox": [ + 257, + 603, + 277, + 613 + ], + "score": 0.83, + "content": "92 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 601, + 505, + 614 + ], + "score": 1.0, + "content": ") of the unlabeled comments are drawn from the same", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 614, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 505, + 624 + ], + "score": 1.0, + "content": "articles as the labeled comments. In particular, 140,082 unlabeled comments are from the same", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 624, + 286, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 286, + 636 + ], + "score": 1.0, + "content": "articles as labeled comments in the test split.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 641, + 504, + 663 + ], + "lines": [ + { + "bbox": [ + 106, + 641, + 504, + 654 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 504, + 654 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS exhibits class imbalance. We account for this when benchmarking meth-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 651, + 471, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 471, + 665 + ], + "score": 1.0, + "content": "ods by sampling class-balanced batches of labeled data when applicable (see Appendix B).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + }, + { + "type": "text", + "bbox": [ + 108, + 677, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "Broader context. In this work, we focused on supplementing CIVILCOMMENTS-WILDS with", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "extra unannotated data from the original CivilComments dataset (Borkan et al., 2019). In practice,", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "unannotated text comments are widely available on the internet. Whether using such unlabeled", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "data, as we do in this work, can help with bias is still an open question. Previous work suggests", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "that training on large amounts of data alone is not sufficient to avoid unwanted biases, since many", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48 + } + ], + "page_idx": 31, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "32", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 94 + ], + "score": 1.0, + "content": "Broader context. We focused on providing unlabeled data for both training and OOD test do-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "mains. Unlabeled molecules can be used to develop and evaluate methods for domain adaptation,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 505, + 116 + ], + "score": 1.0, + "content": "self-training, as well as pre-training (Hu et al., 2020c) and semi-supervised learning (Sun et al.,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "2020). In terms of future directions, we think it is fruitful to explore both graph-agnostic meth-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 506, + 139 + ], + "score": 1.0, + "content": "ods (e.g., pseudo-label training) and more graph-specific methods (e.g., self-supervised learning of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 280, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 280, + 150 + ], + "score": 1.0, + "content": "graph neural networks (Xie et al., 2021b)).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 82, + 506, + 150 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 163, + 241, + 174 + ], + "lines": [ + { + "bbox": [ + 106, + 162, + 242, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 162, + 242, + 176 + ], + "score": 1.0, + "content": "A.7 CIVILCOMMENTS-WILDS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 184, + 505, + 294 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "score": 1.0, + "content": "The CIVILCOMMENTS-WILDS dataset (Koh et al., 2021) was adapted from the CivilComments", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 194, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 505, + 207 + ], + "score": 1.0, + "content": "dataset (Borkan et al., 2019), which is a collection of text comments made on online articles. The", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 205, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 505, + 219 + ], + "score": 1.0, + "content": "data in CIVILCOMMENTS-WILDS underwent a significant labeling and annotation process: each ex-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "ample was labeled toxic or non-toxic and annotated for whether they mentioned certain demographic", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 227, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 505, + 240 + ], + "score": 1.0, + "content": "identities by at least 10 crowdworkers. Such a substantial labeling and identity annotation process is", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 239, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 250 + ], + "score": 1.0, + "content": "expensive and time-consuming. On the other hand, unlabeled, unannotated text comments are read-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "ily available. For example, CIVILCOMMENTS-WILDS only contains a subset of all data available in", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 261, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 273 + ], + "score": 1.0, + "content": "the original CivilComments dataset (Borkan et al., 2019), most of which Koh et al. (2021) excluded", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "because these examples were not annotated for mentioning identities. In this work, we augment the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 282, + 430, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 430, + 295 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS dataset with these unlabeled, unannotated comments.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 183, + 505, + 295 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 307, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 480, + 320 + ], + "score": 1.0, + "content": "Problem setting. The task is to classify whether a text comment is toxic or not. The input", + "type": "text" + }, + { + "bbox": [ + 480, + 310, + 487, + 317 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 307, + 506, + 320 + ], + "score": 1.0, + "content": "is a", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 319, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 453, + 331 + ], + "score": 1.0, + "content": "text comment (at least one sentence long) originally made on an online article, the label", + "type": "text" + }, + { + "bbox": [ + 454, + 320, + 461, + 330 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 319, + 505, + 331 + ], + "score": 1.0, + "content": "is a binary", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 328, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 389, + 343 + ], + "score": 1.0, + "content": "indicator of whether the comment is rated toxic or not, and the domain", + "type": "text" + }, + { + "bbox": [ + 389, + 330, + 396, + 339 + ], + "score": 0.79, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 328, + 505, + 343 + ], + "score": 1.0, + "content": "is an 8-dimensional binary", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 340, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 352 + ], + "score": 1.0, + "content": "vector, where each dimension corresponds to whether the comment mentions each of 8 demographic", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "identities: male, female, LGBTQ, Christian, Muslim, other religions, Black, or White, respectively.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "Each comment also includes metadata on which article the comment was made on, although we do", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 374, + 297, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 297, + 385 + ], + "score": 1.0, + "content": "not use this metadata for training or evaluation.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 307, + 506, + 385 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 390, + 505, + 434 + ], + "lines": [ + { + "bbox": [ + 105, + 389, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 505, + 403 + ], + "score": 1.0, + "content": "We consider the subpopulation shift setting, where the model must perform well across all subpop-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 401, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 256, + 414 + ], + "score": 1.0, + "content": "ulations, which are defined based on", + "type": "text" + }, + { + "bbox": [ + 257, + 402, + 263, + 411 + ], + "score": 0.43, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 401, + 506, + 414 + ], + "score": 1.0, + "content": ". Koh et al. (2021) define 16 subpopulations (groups) based", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 412, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 119, + 425 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 119, + 413, + 126, + 422 + ], + "score": 0.47, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 412, + 506, + 425 + ], + "score": 1.0, + "content": ". Models are then evaluated by their worst-group accuracy, i.e., the lowest accuracy over the 16", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 423, + 374, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 374, + 436 + ], + "score": 1.0, + "content": "groups considered. In our work, we use the same evaluation setup.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 389, + 506, + 436 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 448, + 505, + 546 + ], + "lines": [ + { + "bbox": [ + 106, + 448, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 505, + 459 + ], + "score": 1.0, + "content": "Data. All of the labeled and unlabeled data are taken from the CivilComments dataset (Borkan", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 458, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 505, + 470 + ], + "score": 1.0, + "content": "et al., 2019). After preprocessing, Koh et al. (2021) created the CIVILCOMMENTS-WILDS dataset", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 399, + 482 + ], + "score": 1.0, + "content": "using the 448,000 examples that were fully annotated for both toxicity", + "type": "text" + }, + { + "bbox": [ + 400, + 472, + 407, + 482 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "and the mention of de-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 192, + 492 + ], + "score": 1.0, + "content": "mographic identities", + "type": "text" + }, + { + "bbox": [ + 193, + 481, + 200, + 491 + ], + "score": 0.52, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 480, + 505, + 492 + ], + "score": 1.0, + "content": ". In this work, we augment CIVILCOMMENTS-WILDS with an additional", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "1,551,515 examples collected by Borkan et al. (2019). We use these examples as unlabeled data.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 501, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 506, + 516 + ], + "score": 1.0, + "content": "We follow the same preprocessing steps as was done with the labeled data in Koh et al. (2021). The", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 513, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 344, + 525 + ], + "score": 1.0, + "content": "resulting unlabeled examples have no identity annotations", + "type": "text" + }, + { + "bbox": [ + 344, + 514, + 351, + 523 + ], + "score": 0.78, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 513, + 438, + 525 + ], + "score": 1.0, + "content": "and no toxicity label", + "type": "text" + }, + { + "bbox": [ + 438, + 515, + 445, + 525 + ], + "score": 0.77, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 513, + 505, + 525 + ], + "score": 1.0, + "content": ". We note that", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "Borkan et al. (2019) actually do provide toxicity labels for these examples in the original CivilCom-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 535, + 465, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 465, + 547 + ], + "score": 1.0, + "content": "ments dataset, but we ignore these labels and use them neither for training nor evaluation.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 448, + 506, + 547 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 552, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "Because our unlabeled examples have no identity annotations, we cannot group these examples as", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "score": 1.0, + "content": "Koh et al. (2021) group the labeled examples; thus we refer to this data as unlabeled data coming", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "score": 1.0, + "content": "from extra domains (Table 11). In practice, these comments may actually mention any number of", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 584, + 149, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 149, + 597 + ], + "score": 1.0, + "content": "identities.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 552, + 506, + 597 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 602, + 505, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 601, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 257, + 614 + ], + "score": 1.0, + "content": "A substantial amount (1,427,848 or", + "type": "text" + }, + { + "bbox": [ + 257, + 603, + 277, + 613 + ], + "score": 0.83, + "content": "92 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 601, + 505, + 614 + ], + "score": 1.0, + "content": ") of the unlabeled comments are drawn from the same", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 614, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 505, + 624 + ], + "score": 1.0, + "content": "articles as the labeled comments. In particular, 140,082 unlabeled comments are from the same", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 624, + 286, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 286, + 636 + ], + "score": 1.0, + "content": "articles as labeled comments in the test split.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 601, + 505, + 636 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 641, + 504, + 663 + ], + "lines": [ + { + "bbox": [ + 106, + 641, + 504, + 654 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 504, + 654 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS exhibits class imbalance. We account for this when benchmarking meth-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 651, + 471, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 471, + 665 + ], + "score": 1.0, + "content": "ods by sampling class-balanced batches of labeled data when applicable (see Appendix B).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 641, + 504, + 665 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 677, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "Broader context. In this work, we focused on supplementing CIVILCOMMENTS-WILDS with", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "extra unannotated data from the original CivilComments dataset (Borkan et al., 2019). In practice,", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "unannotated text comments are widely available on the internet. Whether using such unlabeled", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "data, as we do in this work, can help with bias is still an open question. Previous work suggests", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "that training on large amounts of data alone is not sufficient to avoid unwanted biases, since many", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 226, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 238 + ], + "score": 1.0, + "content": "papers have pointed out biases in large language models (Abid et al., 2021; Nadeem et al., 2020;", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "Gehman et al., 2020). However, recent work has also suggested that pre-trained models can be", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 249, + 504, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 504, + 261 + ], + "score": 1.0, + "content": "trained to be more robust against some types of spurious correlations (Hendrycks et al., 2020; Tu", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "score": 1.0, + "content": "et al., 2020) and that additional domain- and task-specific pre-training (Gururangan et al., 2020) can", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "score": 1.0, + "content": "also improve performance. We hope our contributions to the CIVILCOMMENTS-WILDS dataset can", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "encourage future study on whether unlabeled data can be leveraged to improve generalization across", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 292, + 193, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 193, + 304 + ], + "score": 1.0, + "content": "subpopulation shifts.", + "type": "text", + "cross_page": true + } + ], + "index": 10 + } + ], + "index": 48, + "bbox_fs": [ + 105, + 676, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 80, + 510, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 80, + 510, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 510, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 510, + 185 + ], + "score": 0.981, + "html": "
Split# Domains (label × identity groups) # Labeled examples # Unlabeled examples
Source16269,0380
Validation1645,1800
Target16133,7820
Extra101,551,515
Total16448,0001,551,515
", + "type": "table", + "image_path": "d6dad8ab0ef2406f024eda8fc2af0c48643105d62b23e0019490bd2bd35e525a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 80, + 510, + 115.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 115.0, + 510, + 150.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 150.0, + 510, + 185.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 123, + 192, + 484, + 204 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 191, + 485, + 205 + ], + "spans": [ + { + "bbox": [ + 124, + 191, + 485, + 205 + ], + "score": 1.0, + "content": "Table 11: Data for CIVILCOMMENTS-WILDS. All of the splits are identically distributed.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 505, + 303 + ], + "lines": [ + { + "bbox": [ + 105, + 226, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 238 + ], + "score": 1.0, + "content": "papers have pointed out biases in large language models (Abid et al., 2021; Nadeem et al., 2020;", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "Gehman et al., 2020). However, recent work has also suggested that pre-trained models can be", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 249, + 504, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 504, + 261 + ], + "score": 1.0, + "content": "trained to be more robust against some types of spurious correlations (Hendrycks et al., 2020; Tu", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "score": 1.0, + "content": "et al., 2020) and that additional domain- and task-specific pre-training (Gururangan et al., 2020) can", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "score": 1.0, + "content": "also improve performance. We hope our contributions to the CIVILCOMMENTS-WILDS dataset can", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "encourage future study on whether unlabeled data can be leveraged to improve generalization across", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 292, + 193, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 193, + 304 + ], + "score": 1.0, + "content": "subpopulation shifts.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 108, + 319, + 205, + 330 + ], + "lines": [ + { + "bbox": [ + 106, + 318, + 206, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 206, + 332 + ], + "score": 1.0, + "content": "A.8 AMAZON-WILDS", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 106, + 340, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 506, + 353 + ], + "score": 1.0, + "content": "The AMAZON-WILDS dataset (Koh et al., 2021) was adapted from the Amazon reviews dataset", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 351, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 506, + 363 + ], + "score": 1.0, + "content": "(Ni et al., 2019), which is a collection of product reviews written by reviewers. While Amazon", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "score": 1.0, + "content": "reviews are always labeled by the star ratings in practice, unlabeled data is a common source of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "leverage more generally for sentiment classification, with prior work in domain adaptation (Blitzer", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "& Pereira, 2007; Glorot et al., 2011) and semi-supervised learning (Dasgupta & Ng, 2009; Li et al.,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 394, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 506, + 408 + ], + "score": 1.0, + "content": "2011). In this work, we augment the AMAZON-WILDS dataset with unlabeled reviews, whose star", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 407, + 217, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 217, + 418 + ], + "score": 1.0, + "content": "ratings have been removed.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 432, + 505, + 498 + ], + "lines": [ + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "score": 1.0, + "content": "Problem setting. The task is sentiment classification, and we consider generalizing from a set of", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 443, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 309, + 455 + ], + "score": 1.0, + "content": "reviewers to new reviewers at test time. The input", + "type": "text" + }, + { + "bbox": [ + 309, + 445, + 317, + 453 + ], + "score": 0.7, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 443, + 473, + 455 + ], + "score": 1.0, + "content": "corresponds to a review text, the label", + "type": "text" + }, + { + "bbox": [ + 473, + 444, + 480, + 454 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 443, + 505, + 455 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 267, + 466 + ], + "score": 1.0, + "content": "star rating from 1 to 5, and the domain", + "type": "text" + }, + { + "bbox": [ + 268, + 454, + 274, + 464 + ], + "score": 0.73, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "identifies which user wrote the review. For each review,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "score": 1.0, + "content": "additional metadata (product ID, product category, review time, and summary) are also available.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 476, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 487 + ], + "score": 1.0, + "content": "Because the goal is to train a model that performs well across a wide range of reviewers, models are", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 487, + 489, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 489, + 498 + ], + "score": 1.0, + "content": "evaluated by their tail performance, concretely, their accuracy on the user at the 10th percentile.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 511, + 505, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "score": 1.0, + "content": "Data. All of the labeled and unlabeled data are taken from the Amazon reviews dataset (Ni et al.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "2019). We provide unlabeled data from same domains as the labeled AMAZON-WILDS dataset.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 534, + 504, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 504, + 545 + ], + "score": 1.0, + "content": "Additionally, we provide unlabeled data from extra reviewers not in the labeled AMAZON-WILDS", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 545, + 339, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 339, + 557 + ], + "score": 1.0, + "content": "dataset (extra domains). The domains are split as follows:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 129, + 566, + 291, + 628 + ], + "lines": [ + { + "bbox": [ + 130, + 567, + 245, + 578 + ], + "spans": [ + { + "bbox": [ + 130, + 567, + 245, + 578 + ], + "score": 1.0, + "content": "1. Source: 1,252 reviewers.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 129, + 582, + 292, + 596 + ], + "spans": [ + { + "bbox": [ + 129, + 582, + 292, + 596 + ], + "score": 1.0, + "content": "2. Validation (OOD): 1,334 reviewers.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 129, + 599, + 276, + 612 + ], + "spans": [ + { + "bbox": [ + 129, + 599, + 276, + 612 + ], + "score": 1.0, + "content": "3. Target (OOD): 1,334 reviewers.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 129, + 615, + 277, + 628 + ], + "spans": [ + { + "bbox": [ + 129, + 615, + 277, + 628 + ], + "score": 1.0, + "content": "4. Extra (OOD): 21,694 reviewers.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 107, + 638, + 504, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "The reviewers in each split are distinct, and all reviewers have at least 75 reviews. The distributions", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "of reviewers in each split are identical. AMAZON-WILDS also includes Validation (ID) and Target", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 659, + 330, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 330, + 672 + ], + "score": 1.0, + "content": "(ID) sets which contain data from the source reviewers.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "The AMAZON-WILDS dataset has a total of 539,502 labeled reviews across these splits, and we", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "augment the dataset with a total of 3,462,668 unlabeled reviews. For each split of the unlabeled", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 696, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 506, + 712 + ], + "score": 1.0, + "content": "data, we include all available reviews that are written by the reviewer. For the Extra (OOD) split,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "we include all reviewers with at least 75 reviews that are not in Source, Validation (OOD), or Target", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 721, + 163, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 163, + 733 + ], + "score": 1.0, + "content": "(OOD) splits.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38 + } + ], + "page_idx": 32, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "33", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 80, + 510, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 80, + 510, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 510, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 510, + 185 + ], + "score": 0.981, + "html": "
Split# Domains (label × identity groups) # Labeled examples # Unlabeled examples
Source16269,0380
Validation1645,1800
Target16133,7820
Extra101,551,515
Total16448,0001,551,515
", + "type": "table", + "image_path": "d6dad8ab0ef2406f024eda8fc2af0c48643105d62b23e0019490bd2bd35e525a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 80, + 510, + 115.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 115.0, + 510, + 150.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 150.0, + 510, + 185.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 123, + 192, + 484, + 204 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 191, + 485, + 205 + ], + "spans": [ + { + "bbox": [ + 124, + 191, + 485, + 205 + ], + "score": 1.0, + "content": "Table 11: Data for CIVILCOMMENTS-WILDS. All of the splits are identically distributed.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 505, + 303 + ], + "lines": [], + "index": 7, + "bbox_fs": [ + 105, + 226, + 505, + 304 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 319, + 205, + 330 + ], + "lines": [ + { + "bbox": [ + 106, + 318, + 206, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 206, + 332 + ], + "score": 1.0, + "content": "A.8 AMAZON-WILDS", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 340, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 106, + 340, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 506, + 353 + ], + "score": 1.0, + "content": "The AMAZON-WILDS dataset (Koh et al., 2021) was adapted from the Amazon reviews dataset", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 351, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 506, + 363 + ], + "score": 1.0, + "content": "(Ni et al., 2019), which is a collection of product reviews written by reviewers. While Amazon", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "score": 1.0, + "content": "reviews are always labeled by the star ratings in practice, unlabeled data is a common source of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "leverage more generally for sentiment classification, with prior work in domain adaptation (Blitzer", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "& Pereira, 2007; Glorot et al., 2011) and semi-supervised learning (Dasgupta & Ng, 2009; Li et al.,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 394, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 506, + 408 + ], + "score": 1.0, + "content": "2011). In this work, we augment the AMAZON-WILDS dataset with unlabeled reviews, whose star", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 407, + 217, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 217, + 418 + ], + "score": 1.0, + "content": "ratings have been removed.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 340, + 506, + 418 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 432, + 505, + 498 + ], + "lines": [ + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 444 + ], + "score": 1.0, + "content": "Problem setting. The task is sentiment classification, and we consider generalizing from a set of", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 443, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 309, + 455 + ], + "score": 1.0, + "content": "reviewers to new reviewers at test time. The input", + "type": "text" + }, + { + "bbox": [ + 309, + 445, + 317, + 453 + ], + "score": 0.7, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 443, + 473, + 455 + ], + "score": 1.0, + "content": "corresponds to a review text, the label", + "type": "text" + }, + { + "bbox": [ + 473, + 444, + 480, + 454 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 443, + 505, + 455 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 267, + 466 + ], + "score": 1.0, + "content": "star rating from 1 to 5, and the domain", + "type": "text" + }, + { + "bbox": [ + 268, + 454, + 274, + 464 + ], + "score": 0.73, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "identifies which user wrote the review. For each review,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "score": 1.0, + "content": "additional metadata (product ID, product category, review time, and summary) are also available.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 476, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 487 + ], + "score": 1.0, + "content": "Because the goal is to train a model that performs well across a wide range of reviewers, models are", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 487, + 489, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 489, + 498 + ], + "score": 1.0, + "content": "evaluated by their tail performance, concretely, their accuracy on the user at the 10th percentile.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 431, + 506, + 498 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 511, + 505, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 523 + ], + "score": 1.0, + "content": "Data. All of the labeled and unlabeled data are taken from the Amazon reviews dataset (Ni et al.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 534 + ], + "score": 1.0, + "content": "2019). We provide unlabeled data from same domains as the labeled AMAZON-WILDS dataset.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 534, + 504, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 504, + 545 + ], + "score": 1.0, + "content": "Additionally, we provide unlabeled data from extra reviewers not in the labeled AMAZON-WILDS", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 545, + 339, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 339, + 557 + ], + "score": 1.0, + "content": "dataset (extra domains). The domains are split as follows:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 511, + 505, + 557 + ] + }, + { + "type": "index", + "bbox": [ + 129, + 566, + 291, + 628 + ], + "lines": [ + { + "bbox": [ + 130, + 567, + 245, + 578 + ], + "spans": [ + { + "bbox": [ + 130, + 567, + 245, + 578 + ], + "score": 1.0, + "content": "1. Source: 1,252 reviewers.", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 582, + 292, + 596 + ], + "spans": [ + { + "bbox": [ + 129, + 582, + 292, + 596 + ], + "score": 1.0, + "content": "2. Validation (OOD): 1,334 reviewers.", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 599, + 276, + 612 + ], + "spans": [ + { + "bbox": [ + 129, + 599, + 276, + 612 + ], + "score": 1.0, + "content": "3. Target (OOD): 1,334 reviewers.", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 615, + 277, + 628 + ], + "spans": [ + { + "bbox": [ + 129, + 615, + 277, + 628 + ], + "score": 1.0, + "content": "4. Extra (OOD): 21,694 reviewers.", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + } + ], + "index": 30.5, + "bbox_fs": [ + 129, + 567, + 292, + 628 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 638, + 504, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "The reviewers in each split are distinct, and all reviewers have at least 75 reviews. The distributions", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 661 + ], + "score": 1.0, + "content": "of reviewers in each split are identical. AMAZON-WILDS also includes Validation (ID) and Target", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 659, + 330, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 330, + 672 + ], + "score": 1.0, + "content": "(ID) sets which contain data from the source reviewers.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 637, + 505, + 672 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 676, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "The AMAZON-WILDS dataset has a total of 539,502 labeled reviews across these splits, and we", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "augment the dataset with a total of 3,462,668 unlabeled reviews. For each split of the unlabeled", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 696, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 696, + 506, + 712 + ], + "score": 1.0, + "content": "data, we include all available reviews that are written by the reviewer. For the Extra (OOD) split,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "we include all reviewers with at least 75 reviews that are not in Source, Validation (OOD), or Target", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 721, + 163, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 163, + 733 + ], + "score": 1.0, + "content": "(OOD) splits.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 677, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 114, + 80, + 496, + 206 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 114, + 80, + 496, + 206 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 80, + 496, + 206 + ], + "spans": [ + { + "bbox": [ + 114, + 80, + 496, + 206 + ], + "score": 0.983, + "html": "
Split# Domains (reviewers)# Labeled examples# Unlabeled examples
Source1,252245,5020
Validation (ID)46,9500
Target (ID)46,9500
Validation (OOD)1,334100,050266,066
Target (OOD)1,334100,050268,761
Extra (OOD)21,69402,927,841
Total25,614539,5023,462,668
", + "type": "table", + "image_path": "e46afc614bd91d2cefd7cac4d6d2a80ba7095deefde1796782ae0ed75c863cdf.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 114, + 80, + 496, + 122.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 114, + 122.0, + 496, + 164.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 114, + 164.0, + 496, + 206.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 129, + 214, + 480, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 213, + 481, + 226 + ], + "spans": [ + { + "bbox": [ + 129, + 213, + 481, + 226 + ], + "score": 1.0, + "content": "Table 12: Data for AMAZON-WILDS. Each domain corresponds to a different reviewer.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 108, + 247, + 504, + 269 + ], + "lines": [ + { + "bbox": [ + 106, + 246, + 506, + 259 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 506, + 259 + ], + "score": 1.0, + "content": "To filter and process reviews, we followed the same data processing steps as for the labeled data in", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 257, + 255, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 255, + 270 + ], + "score": 1.0, + "content": "AMAZON-WILDS (Koh et al., 2021).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 505, + 381 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "score": 1.0, + "content": "Broader context. We focused on providing unlabeled data from OOD domains, including both", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "score": 1.0, + "content": "test and extra domains. Unlabeled data from the test reviewers can be used to develop and evaluate", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 304, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 505, + 315 + ], + "score": 1.0, + "content": "methods for domain adaptation (Ren et al., 2018; Zhang et al., 2019a; Koohbanani et al., 2021),", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 315, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 505, + 326 + ], + "score": 1.0, + "content": "which has been well-studied in the context of sentiment classification (Blitzer & Pereira, 2007;", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "Glorot et al., 2011). While there is limited prior work on leveraging unlabeled data from extra", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 336, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 506, + 349 + ], + "score": 1.0, + "content": "domains, some domain adaptation techniques can be readily adapted to leverage such unlabeled", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 347, + 504, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 504, + 359 + ], + "score": 1.0, + "content": "data (Ganin et al., 2016). Finally, we focus on unlabeled data specific to the task in this work,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 358, + 506, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 506, + 371 + ], + "score": 1.0, + "content": "varying only the domains, and this contrasts with the type of unlabeled data used for pre-training in", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 370, + 449, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 449, + 381 + ], + "score": 1.0, + "content": "NLP, which is much larger and more diverse (Devlin et al., 2019; Brown et al., 2020).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 10 + } + ], + "page_idx": 33, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 114, + 80, + 496, + 206 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 114, + 80, + 496, + 206 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 80, + 496, + 206 + ], + "spans": [ + { + "bbox": [ + 114, + 80, + 496, + 206 + ], + "score": 0.983, + "html": "
Split# Domains (reviewers)# Labeled examples# Unlabeled examples
Source1,252245,5020
Validation (ID)46,9500
Target (ID)46,9500
Validation (OOD)1,334100,050266,066
Target (OOD)1,334100,050268,761
Extra (OOD)21,69402,927,841
Total25,614539,5023,462,668
", + "type": "table", + "image_path": "e46afc614bd91d2cefd7cac4d6d2a80ba7095deefde1796782ae0ed75c863cdf.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 114, + 80, + 496, + 122.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 114, + 122.0, + 496, + 164.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 114, + 164.0, + 496, + 206.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 129, + 214, + 480, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 213, + 481, + 226 + ], + "spans": [ + { + "bbox": [ + 129, + 213, + 481, + 226 + ], + "score": 1.0, + "content": "Table 12: Data for AMAZON-WILDS. Each domain corresponds to a different reviewer.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 108, + 247, + 504, + 269 + ], + "lines": [ + { + "bbox": [ + 106, + 246, + 506, + 259 + ], + "spans": [ + { + "bbox": [ + 106, + 246, + 506, + 259 + ], + "score": 1.0, + "content": "To filter and process reviews, we followed the same data processing steps as for the labeled data in", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 257, + 255, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 255, + 270 + ], + "score": 1.0, + "content": "AMAZON-WILDS (Koh et al., 2021).", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 106, + 246, + 506, + 270 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 505, + 381 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 506, + 294 + ], + "score": 1.0, + "content": "Broader context. We focused on providing unlabeled data from OOD domains, including both", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "score": 1.0, + "content": "test and extra domains. Unlabeled data from the test reviewers can be used to develop and evaluate", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 304, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 505, + 315 + ], + "score": 1.0, + "content": "methods for domain adaptation (Ren et al., 2018; Zhang et al., 2019a; Koohbanani et al., 2021),", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 315, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 505, + 326 + ], + "score": 1.0, + "content": "which has been well-studied in the context of sentiment classification (Blitzer & Pereira, 2007;", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "Glorot et al., 2011). While there is limited prior work on leveraging unlabeled data from extra", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 336, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 336, + 506, + 349 + ], + "score": 1.0, + "content": "domains, some domain adaptation techniques can be readily adapted to leverage such unlabeled", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 347, + 504, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 504, + 359 + ], + "score": 1.0, + "content": "data (Ganin et al., 2016). Finally, we focus on unlabeled data specific to the task in this work,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 358, + 506, + 371 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 506, + 371 + ], + "score": 1.0, + "content": "varying only the domains, and this contrasts with the type of unlabeled data used for pre-training in", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 370, + 449, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 449, + 381 + ], + "score": 1.0, + "content": "NLP, which is much larger and more diverse (Devlin et al., 2019; Brown et al., 2020).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 280, + 506, + 381 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 239, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 240, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 240, + 96 + ], + "score": 1.0, + "content": "B ALGORITHM DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 106, + 106, + 304, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 303, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 303, + 119 + ], + "score": 1.0, + "content": "B.1 EMPIRICAL RISK MINIMIZATION (ERM)", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 126, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 126, + 504, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 504, + 138 + ], + "score": 1.0, + "content": "As a baseline, we consider Empirical Risk Minimization (ERM). ERM ignores unlabeled data and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 138, + 504, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 504, + 150 + ], + "score": 1.0, + "content": "minimizes the average labeled loss. We additionally evaluate ERM with strong data augmentation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "score": 1.0, + "content": "on applicable datasets, i.e., on IWILDCAM2020-WILDS, CAMELYON17-WILDS, POVERTYMAP-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "score": 1.0, + "content": "WILDS, and FMOW-WILDS (see Appendix C). ERM with strong data augmentation learns a model", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 169, + 274, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 114, + 181 + ], + "score": 0.76, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 169, + 274, + 184 + ], + "score": 1.0, + "content": "that minimizes the labeled training loss", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 220, + 188, + 390, + 220 + ], + "lines": [ + { + "bbox": [ + 220, + 188, + 390, + 220 + ], + "spans": [ + { + "bbox": [ + 220, + 188, + 390, + 220 + ], + "score": 0.94, + "content": "L _ { \\mathrm { L } } ( h ) = \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\bigl ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } \\bigr ) ,", + "type": "interline_equation", + "image_path": "58bf468bc898514ba5a4fba814f0c0ba890ba52c7438e0c4f7d51be14dd263a1.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 220, + 188, + 390, + 204.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 220, + 204.0, + 390, + 220.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 225, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 134, + 238 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 226, + 161, + 237 + ], + "score": 0.91, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 225, + 360, + 238 + ], + "score": 1.0, + "content": "is a stochastic data augmentation operation, and", + "type": "text" + }, + { + "bbox": [ + 360, + 226, + 366, + 235 + ], + "score": 0.73, + "content": "\\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 225, + 505, + 238 + ], + "score": 1.0, + "content": "measures the prediction loss. We", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 236, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 123, + 249 + ], + "score": 1.0, + "content": "use", + "type": "text" + }, + { + "bbox": [ + 123, + 237, + 136, + 247 + ], + "score": 0.88, + "content": "L _ { \\mathrm { L } }", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 236, + 505, + 249 + ], + "score": 1.0, + "content": "throughout this appendix to refer to the above labeled loss with strong augmentations (on", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 247, + 190, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 190, + 259 + ], + "score": 1.0, + "content": "applicable datasets).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 264, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 263, + 504, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 504, + 277 + ], + "score": 1.0, + "content": "For all dataset except CIVILCOMMENTS-WILDS, we sample labeled batches uniformly at random.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "In our experiments, we account for class imbalance in CIVILCOMMENTS-WILDS by explicitly sam-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 285, + 383, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 353, + 299 + ], + "score": 1.0, + "content": "pling class-balanced batches of labeled data when computing", + "type": "text" + }, + { + "bbox": [ + 353, + 286, + 379, + 298 + ], + "score": 0.93, + "content": "L _ { \\mathrm { L } } ( h )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 285, + 383, + 299 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "title", + "bbox": [ + 108, + 311, + 266, + 322 + ], + "lines": [ + { + "bbox": [ + 106, + 311, + 267, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 267, + 323 + ], + "score": 1.0, + "content": "B.2 DOMAIN-INVARIANT METHODS", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "score": 1.0, + "content": "Domain-invariant methods seek to learn feature representations that are invariant across domains.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 342, + 504, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 504, + 355 + ], + "score": 1.0, + "content": "These methods are motivated by earlier theoretical results showing that the gap between in- and out-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 353, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 366 + ], + "score": 1.0, + "content": "of-distribution performance depends on some measure of divergence between the source and target", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 364, + 504, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 504, + 377 + ], + "score": 1.0, + "content": "distributions (Ben-David et al., 2010). To minimize this divergence, the methods described below", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "penalize divergence between feature representations across domains, i.e., they encourage the model", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 387, + 373, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 373, + 399 + ], + "score": 1.0, + "content": "to produce feature representations that are similar across domains.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 107, + 403, + 505, + 437 + ], + "lines": [ + { + "bbox": [ + 105, + 402, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 181, + 417 + ], + "score": 1.0, + "content": "Consider a model", + "type": "text" + }, + { + "bbox": [ + 182, + 404, + 227, + 415 + ], + "score": 0.9, + "content": "h = g \\circ f", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 402, + 316, + 417 + ], + "score": 1.0, + "content": ", where the featurizer", + "type": "text" + }, + { + "bbox": [ + 316, + 404, + 370, + 415 + ], + "score": 0.91, + "content": "f : \\mathcal { X } \\mathcal { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 402, + 505, + 417 + ], + "score": 1.0, + "content": "maps the inputs to some feature", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 414, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 185, + 427 + ], + "score": 1.0, + "content": "space, and the head", + "type": "text" + }, + { + "bbox": [ + 186, + 415, + 231, + 426 + ], + "score": 0.89, + "content": "g : { \\mathcal { F } } { \\mathcal { V } }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 414, + 505, + 427 + ], + "score": 1.0, + "content": "maps feature representations to prediction targets. Domain-invariant", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 425, + 477, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 212, + 438 + ], + "score": 1.0, + "content": "methods seek to constrain", + "type": "text" + }, + { + "bbox": [ + 212, + 426, + 219, + 437 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 425, + 477, + 438 + ], + "score": 1.0, + "content": "to output similar representations for labeled and unlabeled data.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 108, + 442, + 503, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "In this work, we adapt all of our domain-invariant methods to use data augmentations on applicable", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 453, + 406, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 305, + 465 + ], + "score": 1.0, + "content": "datasets (see Appendix C), and thus the output of", + "type": "text" + }, + { + "bbox": [ + 306, + 454, + 312, + 465 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 453, + 406, + 465 + ], + "score": 1.0, + "content": "on the labeled batch is", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 470, + 392, + 487 + ], + "lines": [ + { + "bbox": [ + 219, + 470, + 392, + 487 + ], + "spans": [ + { + "bbox": [ + 219, + 470, + 392, + 487 + ], + "score": 0.93, + "content": "B _ { \\mathrm { L } } = \\{ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "interline_equation", + "image_path": "0e6c867b077507e98667d23152591687ffdd18d37caeb5b592ef7b4d46b47926.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 219, + 470, + 392, + 487 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 309, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 491, + 309, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 201, + 505 + ], + "score": 1.0, + "content": "Similarly, the output of", + "type": "text" + }, + { + "bbox": [ + 201, + 493, + 209, + 504 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 491, + 309, + 505 + ], + "score": 1.0, + "content": "on an unlabeled batch is", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 509, + 393, + 526 + ], + "lines": [ + { + "bbox": [ + 218, + 509, + 393, + 526 + ], + "spans": [ + { + "bbox": [ + 218, + 509, + 393, + 526 + ], + "score": 0.92, + "content": "B _ { \\mathrm { U } } = \\{ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "interline_equation", + "image_path": "e56ec1d2cca324a8a0772d04fe743593bc9fbbdfd644078c36ec6233ab8f87e9.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 218, + 509, + 393, + 526 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 536, + 505, + 570 + ], + "lines": [ + { + "bbox": [ + 105, + 536, + 506, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 356, + 549 + ], + "score": 1.0, + "content": "Domain-invariant methods seek to minimize some divergence", + "type": "text" + }, + { + "bbox": [ + 356, + 537, + 422, + 548 + ], + "score": 0.9, + "content": "\\xi : \\mathcal { F } \\times \\mathcal { F } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 536, + 506, + 549 + ], + "score": 1.0, + "content": "between the labeled", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 547, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 126, + 560 + ], + "score": 1.0, + "content": "data", + "type": "text" + }, + { + "bbox": [ + 126, + 548, + 139, + 559 + ], + "score": 0.89, + "content": "B _ { \\mathrm { L } }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 547, + 231, + 560 + ], + "score": 1.0, + "content": "and the unlabeled data", + "type": "text" + }, + { + "bbox": [ + 232, + 548, + 246, + 559 + ], + "score": 0.9, + "content": "B _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 547, + 505, + 560 + ], + "score": 1.0, + "content": ", where the choice of divergence depends on the specific method.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 558, + 294, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 294, + 572 + ], + "score": 1.0, + "content": "The divergence is expressed as a penalty term:", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 575, + 357, + 597 + ], + "lines": [ + { + "bbox": [ + 253, + 575, + 357, + 597 + ], + "spans": [ + { + "bbox": [ + 253, + 575, + 357, + 597 + ], + "score": 0.94, + "content": "L _ { \\mathrm { p e n a l t y } } ( f ) = \\xi \\Big ( B _ { \\mathrm { L } } , B _ { \\mathrm { U } } \\Big )", + "type": "interline_equation", + "image_path": "aa1ace8260c12a1b0e3df5b1b99ade859b497a39018e772130f5bd45823fb44d.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 253, + 575, + 357, + 597 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 601, + 504, + 624 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 614 + ], + "score": 1.0, + "content": "The final objective is a combination of the labeled loss and penalty loss. The balance between the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 613, + 369, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 279, + 626 + ], + "score": 1.0, + "content": "two losses is controlled by hyperparameter", + "type": "text" + }, + { + "bbox": [ + 280, + 613, + 286, + 623 + ], + "score": 0.79, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 613, + 369, + 626 + ], + "score": 1.0, + "content": ", the penalty weight.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "interline_equation", + "bbox": [ + 245, + 630, + 366, + 644 + ], + "lines": [ + { + "bbox": [ + 245, + 630, + 366, + 644 + ], + "spans": [ + { + "bbox": [ + 245, + 630, + 366, + 644 + ], + "score": 0.91, + "content": "L ( h ) = L _ { \\mathrm { L } } ( h ) + \\lambda L _ { \\mathrm { p e n a l t y } } ( f )", + "type": "interline_equation", + "image_path": "0fd4b2bc6b009fb50f1b1a103fdda864224f3c6caf05d68b09f3a2189b43049a.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 245, + 630, + 366, + 644 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "In our experiments, we study two classical domain-invariant methods, Correlation Alignment", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 666, + 504, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 504, + 677 + ], + "score": 1.0, + "content": "(CORAL) (Sun et al., 2016; Sun & Saenko, 2016) and Domain-Adversarial Neural Networks", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 677, + 504, + 688 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 504, + 688 + ], + "score": 1.0, + "content": "(DANN) (Ganin et al., 2016). These methods are well-known and established, but their perfor-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "mance can be lower than that of newer domain-invariant methods that employ different penalties to", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "encourage the source and target representations to be similar (Jiang et al., 2020; Zhang et al., 2021).", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "Examples of these newer methods are Joint Adaptation Networks (JAN) (Long et al., 2017), Con-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "ditional Domain Adversarial Networks (CDAN) (Long et al., 2018), Collaborative and Adversarial", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40 + } + ], + "page_idx": 34, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 239, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 240, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 240, + 96 + ], + "score": 1.0, + "content": "B ALGORITHM DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 106, + 106, + 304, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 105, + 303, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 303, + 119 + ], + "score": 1.0, + "content": "B.1 EMPIRICAL RISK MINIMIZATION (ERM)", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 126, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 126, + 504, + 138 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 504, + 138 + ], + "score": 1.0, + "content": "As a baseline, we consider Empirical Risk Minimization (ERM). ERM ignores unlabeled data and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 138, + 504, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 504, + 150 + ], + "score": 1.0, + "content": "minimizes the average labeled loss. We additionally evaluate ERM with strong data augmentation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 161 + ], + "score": 1.0, + "content": "on applicable datasets, i.e., on IWILDCAM2020-WILDS, CAMELYON17-WILDS, POVERTYMAP-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "score": 1.0, + "content": "WILDS, and FMOW-WILDS (see Appendix C). ERM with strong data augmentation learns a model", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 169, + 274, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 114, + 181 + ], + "score": 0.76, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 169, + 274, + 184 + ], + "score": 1.0, + "content": "that minimizes the labeled training loss", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 126, + 506, + 184 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 220, + 188, + 390, + 220 + ], + "lines": [ + { + "bbox": [ + 220, + 188, + 390, + 220 + ], + "spans": [ + { + "bbox": [ + 220, + 188, + 390, + 220 + ], + "score": 0.94, + "content": "L _ { \\mathrm { L } } ( h ) = \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\bigl ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } \\bigr ) ,", + "type": "interline_equation", + "image_path": "58bf468bc898514ba5a4fba814f0c0ba890ba52c7438e0c4f7d51be14dd263a1.jpg" + } + ] + } + ], + "index": 7.5, + "virtual_lines": [ + { + "bbox": [ + 220, + 188, + 390, + 204.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 220, + 204.0, + 390, + 220.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 225, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 134, + 238 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 226, + 161, + 237 + ], + "score": 0.91, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 225, + 360, + 238 + ], + "score": 1.0, + "content": "is a stochastic data augmentation operation, and", + "type": "text" + }, + { + "bbox": [ + 360, + 226, + 366, + 235 + ], + "score": 0.73, + "content": "\\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 225, + 505, + 238 + ], + "score": 1.0, + "content": "measures the prediction loss. We", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 236, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 123, + 249 + ], + "score": 1.0, + "content": "use", + "type": "text" + }, + { + "bbox": [ + 123, + 237, + 136, + 247 + ], + "score": 0.88, + "content": "L _ { \\mathrm { L } }", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 236, + 505, + 249 + ], + "score": 1.0, + "content": "throughout this appendix to refer to the above labeled loss with strong augmentations (on", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 247, + 190, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 190, + 259 + ], + "score": 1.0, + "content": "applicable datasets).", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 225, + 505, + 259 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 264, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 263, + 504, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 504, + 277 + ], + "score": 1.0, + "content": "For all dataset except CIVILCOMMENTS-WILDS, we sample labeled batches uniformly at random.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "In our experiments, we account for class imbalance in CIVILCOMMENTS-WILDS by explicitly sam-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 285, + 383, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 353, + 299 + ], + "score": 1.0, + "content": "pling class-balanced batches of labeled data when computing", + "type": "text" + }, + { + "bbox": [ + 353, + 286, + 379, + 298 + ], + "score": 0.93, + "content": "L _ { \\mathrm { L } } ( h )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 285, + 383, + 299 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 263, + 505, + 299 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 311, + 266, + 322 + ], + "lines": [ + { + "bbox": [ + 106, + 311, + 267, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 267, + 323 + ], + "score": 1.0, + "content": "B.2 DOMAIN-INVARIANT METHODS", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "score": 1.0, + "content": "Domain-invariant methods seek to learn feature representations that are invariant across domains.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 342, + 504, + 355 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 504, + 355 + ], + "score": 1.0, + "content": "These methods are motivated by earlier theoretical results showing that the gap between in- and out-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 353, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 366 + ], + "score": 1.0, + "content": "of-distribution performance depends on some measure of divergence between the source and target", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 364, + 504, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 504, + 377 + ], + "score": 1.0, + "content": "distributions (Ben-David et al., 2010). To minimize this divergence, the methods described below", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "penalize divergence between feature representations across domains, i.e., they encourage the model", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 387, + 373, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 373, + 399 + ], + "score": 1.0, + "content": "to produce feature representations that are similar across domains.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 332, + 506, + 399 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 403, + 505, + 437 + ], + "lines": [ + { + "bbox": [ + 105, + 402, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 181, + 417 + ], + "score": 1.0, + "content": "Consider a model", + "type": "text" + }, + { + "bbox": [ + 182, + 404, + 227, + 415 + ], + "score": 0.9, + "content": "h = g \\circ f", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 402, + 316, + 417 + ], + "score": 1.0, + "content": ", where the featurizer", + "type": "text" + }, + { + "bbox": [ + 316, + 404, + 370, + 415 + ], + "score": 0.91, + "content": "f : \\mathcal { X } \\mathcal { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 402, + 505, + 417 + ], + "score": 1.0, + "content": "maps the inputs to some feature", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 414, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 185, + 427 + ], + "score": 1.0, + "content": "space, and the head", + "type": "text" + }, + { + "bbox": [ + 186, + 415, + 231, + 426 + ], + "score": 0.89, + "content": "g : { \\mathcal { F } } { \\mathcal { V } }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 414, + 505, + 427 + ], + "score": 1.0, + "content": "maps feature representations to prediction targets. Domain-invariant", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 425, + 477, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 212, + 438 + ], + "score": 1.0, + "content": "methods seek to constrain", + "type": "text" + }, + { + "bbox": [ + 212, + 426, + 219, + 437 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 425, + 477, + 438 + ], + "score": 1.0, + "content": "to output similar representations for labeled and unlabeled data.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 402, + 505, + 438 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 442, + 503, + 465 + ], + "lines": [ + { + "bbox": [ + 106, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "In this work, we adapt all of our domain-invariant methods to use data augmentations on applicable", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 453, + 406, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 305, + 465 + ], + "score": 1.0, + "content": "datasets (see Appendix C), and thus the output of", + "type": "text" + }, + { + "bbox": [ + 306, + 454, + 312, + 465 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 453, + 406, + 465 + ], + "score": 1.0, + "content": "on the labeled batch is", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 442, + 505, + 465 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 470, + 392, + 487 + ], + "lines": [ + { + "bbox": [ + 219, + 470, + 392, + 487 + ], + "spans": [ + { + "bbox": [ + 219, + 470, + 392, + 487 + ], + "score": 0.93, + "content": "B _ { \\mathrm { L } } = \\{ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "interline_equation", + "image_path": "0e6c867b077507e98667d23152591687ffdd18d37caeb5b592ef7b4d46b47926.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 219, + 470, + 392, + 487 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 309, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 491, + 309, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 201, + 505 + ], + "score": 1.0, + "content": "Similarly, the output of", + "type": "text" + }, + { + "bbox": [ + 201, + 493, + 209, + 504 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 491, + 309, + 505 + ], + "score": 1.0, + "content": "on an unlabeled batch is", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 106, + 491, + 309, + 505 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 509, + 393, + 526 + ], + "lines": [ + { + "bbox": [ + 218, + 509, + 393, + 526 + ], + "spans": [ + { + "bbox": [ + 218, + 509, + 393, + 526 + ], + "score": 0.92, + "content": "B _ { \\mathrm { U } } = \\{ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "interline_equation", + "image_path": "e56ec1d2cca324a8a0772d04fe743593bc9fbbdfd644078c36ec6233ab8f87e9.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 218, + 509, + 393, + 526 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 536, + 505, + 570 + ], + "lines": [ + { + "bbox": [ + 105, + 536, + 506, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 356, + 549 + ], + "score": 1.0, + "content": "Domain-invariant methods seek to minimize some divergence", + "type": "text" + }, + { + "bbox": [ + 356, + 537, + 422, + 548 + ], + "score": 0.9, + "content": "\\xi : \\mathcal { F } \\times \\mathcal { F } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 536, + 506, + 549 + ], + "score": 1.0, + "content": "between the labeled", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 547, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 126, + 560 + ], + "score": 1.0, + "content": "data", + "type": "text" + }, + { + "bbox": [ + 126, + 548, + 139, + 559 + ], + "score": 0.89, + "content": "B _ { \\mathrm { L } }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 547, + 231, + 560 + ], + "score": 1.0, + "content": "and the unlabeled data", + "type": "text" + }, + { + "bbox": [ + 232, + 548, + 246, + 559 + ], + "score": 0.9, + "content": "B _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 547, + 505, + 560 + ], + "score": 1.0, + "content": ", where the choice of divergence depends on the specific method.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 558, + 294, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 294, + 572 + ], + "score": 1.0, + "content": "The divergence is expressed as a penalty term:", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 536, + 506, + 572 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 575, + 357, + 597 + ], + "lines": [ + { + "bbox": [ + 253, + 575, + 357, + 597 + ], + "spans": [ + { + "bbox": [ + 253, + 575, + 357, + 597 + ], + "score": 0.94, + "content": "L _ { \\mathrm { p e n a l t y } } ( f ) = \\xi \\Big ( B _ { \\mathrm { L } } , B _ { \\mathrm { U } } \\Big )", + "type": "interline_equation", + "image_path": "aa1ace8260c12a1b0e3df5b1b99ade859b497a39018e772130f5bd45823fb44d.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 253, + 575, + 357, + 597 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 601, + 504, + 624 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 614 + ], + "score": 1.0, + "content": "The final objective is a combination of the labeled loss and penalty loss. The balance between the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 613, + 369, + 626 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 279, + 626 + ], + "score": 1.0, + "content": "two losses is controlled by hyperparameter", + "type": "text" + }, + { + "bbox": [ + 280, + 613, + 286, + 623 + ], + "score": 0.79, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 613, + 369, + 626 + ], + "score": 1.0, + "content": ", the penalty weight.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 106, + 600, + 505, + 626 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 245, + 630, + 366, + 644 + ], + "lines": [ + { + "bbox": [ + 245, + 630, + 366, + 644 + ], + "spans": [ + { + "bbox": [ + 245, + 630, + 366, + 644 + ], + "score": 0.91, + "content": "L ( h ) = L _ { \\mathrm { L } } ( h ) + \\lambda L _ { \\mathrm { p e n a l t y } } ( f )", + "type": "interline_equation", + "image_path": "0fd4b2bc6b009fb50f1b1a103fdda864224f3c6caf05d68b09f3a2189b43049a.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 245, + 630, + 366, + 644 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "In our experiments, we study two classical domain-invariant methods, Correlation Alignment", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 666, + 504, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 504, + 677 + ], + "score": 1.0, + "content": "(CORAL) (Sun et al., 2016; Sun & Saenko, 2016) and Domain-Adversarial Neural Networks", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 677, + 504, + 688 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 504, + 688 + ], + "score": 1.0, + "content": "(DANN) (Ganin et al., 2016). These methods are well-known and established, but their perfor-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "mance can be lower than that of newer domain-invariant methods that employ different penalties to", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "encourage the source and target representations to be similar (Jiang et al., 2020; Zhang et al., 2021).", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "Examples of these newer methods are Joint Adaptation Networks (JAN) (Long et al., 2017), Con-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "ditional Domain Adversarial Networks (CDAN) (Long et al., 2018), Collaborative and Adversarial", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 654, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "score": 1.0, + "content": "Networks (CAN) (Zhang et al., 2018), and models with Adaptive Feature Norm (AFN) (Xu et al.,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "2019), as well as methods that minimize the Maximum Classifier Discrepancy (MCD) (Saito et al.,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 408, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 408, + 117 + ], + "score": 1.0, + "content": "2018) and the Margin Disparity Discrepancy (MDD) (Zhang et al., 2019b).", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 121, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "score": 1.0, + "content": "All of the above methods were developed for the single-source single-target setting, where the source", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "score": 1.0, + "content": "domain is treated as a single distribution, and likewise for the target domain. As each WILDS 2.0", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 144, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 505, + 155 + ], + "score": 1.0, + "content": "dataset comprises multiple source domains and multiple target domains, it is likely that methods", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 153, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 506, + 167 + ], + "score": 1.0, + "content": "that can leverage this additional structure could perform better. Examples of these methods include", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "Multi-source Domain Adversarial Networks (MDAN) (Zhao et al., 2018) and Moment Matching", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "score": 1.0, + "content": "for Multi-Source Domain Adaptation (M3SDA) (Peng et al., 2019). The DomainBed (Gulrajani &", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 186, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 505, + 200 + ], + "score": 1.0, + "content": "Lopez-Paz, 2020) and WILDS (Koh et al., 2021) benchmarks also extended single-source algorithms", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 197, + 506, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 506, + 210 + ], + "score": 1.0, + "content": "like CORAL and DANN to take advantage of multiple source domains in the domain generalization", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 208, + 442, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 442, + 223 + ], + "score": 1.0, + "content": "setting, and similar extensions in the domain adaptation setting could be promising.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 233, + 504, + 278 + ], + "lines": [ + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "score": 1.0, + "content": "Correlation Alignment (CORAL). Algorithm 1 describes CORAL, proposed by Sun et al.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 243, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 369, + 257 + ], + "score": 1.0, + "content": "(2016); Sun & Saenko (2016). CORAL measures the divergence", + "type": "text" + }, + { + "bbox": [ + 369, + 245, + 375, + 256 + ], + "score": 0.85, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 243, + 506, + 257 + ], + "score": 1.0, + "content": "between batches of feature rep-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 256, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 505, + 267 + ], + "score": 1.0, + "content": "resentations in terms of the deviation between their first and second order statistics. Given a labeled", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 264, + 488, + 280 + ], + "spans": [ + { + "bbox": [ + 104, + 264, + 257, + 280 + ], + "score": 1.0, + "content": "batch and unlabeled batch of features", + "type": "text" + }, + { + "bbox": [ + 258, + 266, + 312, + 278 + ], + "score": 0.58, + "content": "B _ { \\mathrm { L } } \\in \\mathbb { R } ^ { n _ { \\mathrm { L } } \\times m }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 264, + 315, + 280 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 316, + 266, + 372, + 278 + ], + "score": 0.81, + "content": "B _ { \\mathrm { U } } \\in \\mathbb { R } ^ { n _ { \\mathrm { U } } \\times m }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 264, + 488, + 280 + ], + "score": 1.0, + "content": ", define the feature means as", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "interline_equation", + "bbox": [ + 272, + 283, + 338, + 336 + ], + "lines": [ + { + "bbox": [ + 272, + 283, + 338, + 336 + ], + "spans": [ + { + "bbox": [ + 272, + 283, + 338, + 336 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\mu _ { \\mathrm { L } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } 1 ^ { T } B _ { \\mathrm { L } } } \\\\ & { \\mu _ { \\mathrm { U } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } } 1 ^ { T } B _ { \\mathrm { U } } } \\end{array}", + "type": "interline_equation", + "image_path": "f8e795785fc6306492006067277bc312197cacd8c0bd183c47489761ac6848f3.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 272, + 283, + 338, + 309.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 272, + 309.5, + 338, + 336.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 215, + 351 + ], + "lines": [ + { + "bbox": [ + 106, + 340, + 216, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 216, + 352 + ], + "score": 1.0, + "content": "and covariance matrices as", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 203, + 355, + 408, + 412 + ], + "lines": [ + { + "bbox": [ + 203, + 355, + 408, + 412 + ], + "spans": [ + { + "bbox": [ + 203, + 355, + 408, + 412 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { C _ { \\mathrm { L } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } - 1 } \\bigg ( B _ { \\mathrm { L } } ^ { T } B _ { \\mathrm { L } } - \\frac { 1 } { n _ { \\mathrm { L } } } \\big ( \\boldsymbol { 1 } ^ { T } B _ { \\mathrm { L } } \\big ) ^ { T } \\big ( \\boldsymbol { 1 } ^ { T } B _ { \\mathrm { L } } \\big ) \\bigg ) } \\\\ & { C _ { \\mathrm { U } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } - 1 } \\bigg ( B _ { \\mathrm { U } } ^ { T } B _ { \\mathrm { U } } - \\frac { 1 } { n _ { \\mathrm { U } } } \\big ( \\boldsymbol { 1 } ^ { T } B _ { \\mathrm { U } } \\big ) ^ { T } \\big ( \\boldsymbol { 1 } ^ { T } B _ { \\mathrm { U } } \\big ) \\bigg ) . } \\end{array}", + "type": "interline_equation", + "image_path": "11e5b3b1c84119f539bd9bde44ecc7d677988b6fc532090f7c5d1bc2800de208.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 203, + 355, + 408, + 369.25 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 203, + 369.25, + 408, + 383.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 203, + 383.5, + 408, + 397.75 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 203, + 397.75, + 408, + 412.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 416, + 271, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 415, + 272, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 272, + 430 + ], + "score": 1.0, + "content": "We then compute the CORAL penalty as", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 433, + 397, + 455 + ], + "lines": [ + { + "bbox": [ + 214, + 433, + 397, + 455 + ], + "spans": [ + { + "bbox": [ + 214, + 433, + 397, + 455 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\xi \\Big ( B _ { \\mathrm { L } } , B _ { \\mathrm { U } } \\Big ) = | | \\mu _ { \\mathrm { L } } - \\mu _ { \\mathrm { U } } | | ^ { 2 } + | | C _ { \\mathrm { L } } - C _ { \\mathrm { U } } | | _ { F } ^ { 2 } . } \\end{array}", + "type": "interline_equation", + "image_path": "64e8532e7e445056fd2af3152b103059949c7758341a6fd41811ae566dbe3872.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 214, + 433, + 397, + 455 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 466, + 505, + 544 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "We adapted our implementation from DomainBed (Gulrajani & Lopez-Paz, 2020), as done in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "WILDS 1.0. We note that these implementations compute the penalty as a sum of deviations in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 501 + ], + "score": 1.0, + "content": "means and covariances, whereas Sun et al. (2016); Sun & Saenko (2016) penalize deviations in", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "covariances only. (Sun et al. (2016) considers features that are normalized to zero mean.) On", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 509, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 104, + 509, + 474, + 526 + ], + "score": 1.0, + "content": "applicable datasets, we also strongly augmented all labeled and unlabeled examples using", + "type": "text" + }, + { + "bbox": [ + 475, + 511, + 501, + 523 + ], + "score": 0.87, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 509, + 506, + 526 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "whereas Sun et al. (2016); Sun & Saenko (2016) do not explicitly require data augmentations. We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 487, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 487, + 545 + ], + "score": 1.0, + "content": "add augmentations to allow for a fairer comparison to other methods which use augmentations.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 549, + 505, + 617 + ], + "lines": [ + { + "bbox": [ + 106, + 550, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 505, + 561 + ], + "score": 1.0, + "content": "Note that CORAL has also been adapted by Gulrajani & Lopez-Paz (2020); Koh et al. (2021) for", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 423, + 574 + ], + "score": 1.0, + "content": "domain generalization. In particular, where the original CORAL paper defines", + "type": "text" + }, + { + "bbox": [ + 423, + 561, + 452, + 573 + ], + "score": 0.85, + "content": "L _ { \\mathrm { p e n a l t y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 560, + 506, + 574 + ], + "score": 1.0, + "content": "as the diver-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 571, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 448, + 585 + ], + "score": 1.0, + "content": "gence between just two kinds of batches (labeled and unlabeled), these works define", + "type": "text" + }, + { + "bbox": [ + 449, + 572, + 478, + 584 + ], + "score": 0.88, + "content": "L _ { \\mathrm { p e n a l t y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 571, + 506, + 585 + ], + "score": 1.0, + "content": "as the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "score": 1.0, + "content": "divergence between many kinds of batches, where batches are grouped based on domain annotation", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 592, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 122, + 605 + ], + "score": 0.88, + "content": "\\boldsymbol { d } ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 592, + 506, + 608 + ], + "score": 1.0, + "content": ". For simplicity, we followed the original CORAL formulation and differentiate only between", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 478, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 409, + 617 + ], + "score": 1.0, + "content": "labeled and unlabeled batches. We leave leveraging the domain adaptations", + "type": "text" + }, + { + "bbox": [ + 409, + 606, + 416, + 615 + ], + "score": 0.79, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 605, + 478, + 617 + ], + "score": 1.0, + "content": "to future work.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 107, + 630, + 504, + 686 + ], + "lines": [ + { + "bbox": [ + 106, + 631, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 641 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We run CORAL on all datasets except GLOBALWHEAT-WILDS and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 641, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 505, + 653 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS. We do not evaluate domain invariant methods on CIVILCOMMENTS-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 651, + 506, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 506, + 665 + ], + "score": 1.0, + "content": "WILDS, since the labeled and unlabeled data are drawn from the same distribution. We do not", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 662, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 677 + ], + "score": 1.0, + "content": "evaluate CORAL on GLOBALWHEAT-WILDS because CORAL does not port straightforwardly to", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 672, + 181, + 688 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 181, + 688 + ], + "score": 1.0, + "content": "detection settings.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "DANN. Algorithm 2 describes DANN, proposed by Ganin et al. (2016). DANN measures the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 152, + 722 + ], + "score": 1.0, + "content": "divergence", + "type": "text" + }, + { + "bbox": [ + 153, + 710, + 159, + 721 + ], + "score": 0.83, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "between batches of feature representations using the performance of a discriminator", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 141, + 733 + ], + "score": 1.0, + "content": "network", + "type": "text" + }, + { + "bbox": [ + 141, + 721, + 153, + 732 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 720, + 458, + 733 + ], + "score": 1.0, + "content": "that aims to discriminate between domains. Given a batch of features (either", + "type": "text" + }, + { + "bbox": [ + 459, + 721, + 472, + 732 + ], + "score": 0.89, + "content": "B _ { \\mathrm { L } }", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 720, + 483, + 733 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 484, + 721, + 498, + 732 + ], + "score": 0.86, + "content": "B _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 499, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "),", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "page_idx": 35, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 13, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "score": 1.0, + "content": "Networks (CAN) (Zhang et al., 2018), and models with Adaptive Feature Norm (AFN) (Xu et al.,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "2019), as well as methods that minimize the Maximum Classifier Discrepancy (MCD) (Saito et al.,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 408, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 408, + 117 + ], + "score": 1.0, + "content": "2018) and the Margin Disparity Discrepancy (MDD) (Zhang et al., 2019b).", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 506, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 121, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "spans": [ + { + "bbox": [ + 106, + 121, + 505, + 133 + ], + "score": 1.0, + "content": "All of the above methods were developed for the single-source single-target setting, where the source", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 132, + 505, + 144 + ], + "score": 1.0, + "content": "domain is treated as a single distribution, and likewise for the target domain. As each WILDS 2.0", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 144, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 505, + 155 + ], + "score": 1.0, + "content": "dataset comprises multiple source domains and multiple target domains, it is likely that methods", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 153, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 153, + 506, + 167 + ], + "score": 1.0, + "content": "that can leverage this additional structure could perform better. Examples of these methods include", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "Multi-source Domain Adversarial Networks (MDAN) (Zhao et al., 2018) and Moment Matching", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 505, + 188 + ], + "score": 1.0, + "content": "for Multi-Source Domain Adaptation (M3SDA) (Peng et al., 2019). The DomainBed (Gulrajani &", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 186, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 505, + 200 + ], + "score": 1.0, + "content": "Lopez-Paz, 2020) and WILDS (Koh et al., 2021) benchmarks also extended single-source algorithms", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 197, + 506, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 506, + 210 + ], + "score": 1.0, + "content": "like CORAL and DANN to take advantage of multiple source domains in the domain generalization", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 208, + 442, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 442, + 223 + ], + "score": 1.0, + "content": "setting, and similar extensions in the domain adaptation setting could be promising.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 121, + 506, + 223 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 233, + 504, + 278 + ], + "lines": [ + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "score": 1.0, + "content": "Correlation Alignment (CORAL). Algorithm 1 describes CORAL, proposed by Sun et al.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 243, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 369, + 257 + ], + "score": 1.0, + "content": "(2016); Sun & Saenko (2016). CORAL measures the divergence", + "type": "text" + }, + { + "bbox": [ + 369, + 245, + 375, + 256 + ], + "score": 0.85, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 243, + 506, + 257 + ], + "score": 1.0, + "content": "between batches of feature rep-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 256, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 505, + 267 + ], + "score": 1.0, + "content": "resentations in terms of the deviation between their first and second order statistics. Given a labeled", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 264, + 488, + 280 + ], + "spans": [ + { + "bbox": [ + 104, + 264, + 257, + 280 + ], + "score": 1.0, + "content": "batch and unlabeled batch of features", + "type": "text" + }, + { + "bbox": [ + 258, + 266, + 312, + 278 + ], + "score": 0.58, + "content": "B _ { \\mathrm { L } } \\in \\mathbb { R } ^ { n _ { \\mathrm { L } } \\times m }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 264, + 315, + 280 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 316, + 266, + 372, + 278 + ], + "score": 0.81, + "content": "B _ { \\mathrm { U } } \\in \\mathbb { R } ^ { n _ { \\mathrm { U } } \\times m }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 264, + 488, + 280 + ], + "score": 1.0, + "content": ", define the feature means as", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5, + "bbox_fs": [ + 104, + 233, + 506, + 280 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 272, + 283, + 338, + 336 + ], + "lines": [ + { + "bbox": [ + 272, + 283, + 338, + 336 + ], + "spans": [ + { + "bbox": [ + 272, + 283, + 338, + 336 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\mu _ { \\mathrm { L } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } 1 ^ { T } B _ { \\mathrm { L } } } \\\\ & { \\mu _ { \\mathrm { U } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } } 1 ^ { T } B _ { \\mathrm { U } } } \\end{array}", + "type": "interline_equation", + "image_path": "f8e795785fc6306492006067277bc312197cacd8c0bd183c47489761ac6848f3.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 272, + 283, + 338, + 309.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 272, + 309.5, + 338, + 336.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 215, + 351 + ], + "lines": [ + { + "bbox": [ + 106, + 340, + 216, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 216, + 352 + ], + "score": 1.0, + "content": "and covariance matrices as", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 106, + 340, + 216, + 352 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 203, + 355, + 408, + 412 + ], + "lines": [ + { + "bbox": [ + 203, + 355, + 408, + 412 + ], + "spans": [ + { + "bbox": [ + 203, + 355, + 408, + 412 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { C _ { \\mathrm { L } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } - 1 } \\bigg ( B _ { \\mathrm { L } } ^ { T } B _ { \\mathrm { L } } - \\frac { 1 } { n _ { \\mathrm { L } } } \\big ( \\boldsymbol { 1 } ^ { T } B _ { \\mathrm { L } } \\big ) ^ { T } \\big ( \\boldsymbol { 1 } ^ { T } B _ { \\mathrm { L } } \\big ) \\bigg ) } \\\\ & { C _ { \\mathrm { U } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } - 1 } \\bigg ( B _ { \\mathrm { U } } ^ { T } B _ { \\mathrm { U } } - \\frac { 1 } { n _ { \\mathrm { U } } } \\big ( \\boldsymbol { 1 } ^ { T } B _ { \\mathrm { U } } \\big ) ^ { T } \\big ( \\boldsymbol { 1 } ^ { T } B _ { \\mathrm { U } } \\big ) \\bigg ) . } \\end{array}", + "type": "interline_equation", + "image_path": "11e5b3b1c84119f539bd9bde44ecc7d677988b6fc532090f7c5d1bc2800de208.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 203, + 355, + 408, + 369.25 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 203, + 369.25, + 408, + 383.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 203, + 383.5, + 408, + 397.75 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 203, + 397.75, + 408, + 412.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 416, + 271, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 415, + 272, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 272, + 430 + ], + "score": 1.0, + "content": "We then compute the CORAL penalty as", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 415, + 272, + 430 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 433, + 397, + 455 + ], + "lines": [ + { + "bbox": [ + 214, + 433, + 397, + 455 + ], + "spans": [ + { + "bbox": [ + 214, + 433, + 397, + 455 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\xi \\Big ( B _ { \\mathrm { L } } , B _ { \\mathrm { U } } \\Big ) = | | \\mu _ { \\mathrm { L } } - \\mu _ { \\mathrm { U } } | | ^ { 2 } + | | C _ { \\mathrm { L } } - C _ { \\mathrm { U } } | | _ { F } ^ { 2 } . } \\end{array}", + "type": "interline_equation", + "image_path": "64e8532e7e445056fd2af3152b103059949c7758341a6fd41811ae566dbe3872.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 214, + 433, + 397, + 455 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 466, + 505, + 544 + ], + "lines": [ + { + "bbox": [ + 106, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "We adapted our implementation from DomainBed (Gulrajani & Lopez-Paz, 2020), as done in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "WILDS 1.0. We note that these implementations compute the penalty as a sum of deviations in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 501 + ], + "score": 1.0, + "content": "means and covariances, whereas Sun et al. (2016); Sun & Saenko (2016) penalize deviations in", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "covariances only. (Sun et al. (2016) considers features that are normalized to zero mean.) On", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 509, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 104, + 509, + 474, + 526 + ], + "score": 1.0, + "content": "applicable datasets, we also strongly augmented all labeled and unlabeled examples using", + "type": "text" + }, + { + "bbox": [ + 475, + 511, + 501, + 523 + ], + "score": 0.87, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 509, + 506, + 526 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "whereas Sun et al. (2016); Sun & Saenko (2016) do not explicitly require data augmentations. We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 487, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 487, + 545 + ], + "score": 1.0, + "content": "add augmentations to allow for a fairer comparison to other methods which use augmentations.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28, + "bbox_fs": [ + 104, + 466, + 506, + 545 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 549, + 505, + 617 + ], + "lines": [ + { + "bbox": [ + 106, + 550, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 505, + 561 + ], + "score": 1.0, + "content": "Note that CORAL has also been adapted by Gulrajani & Lopez-Paz (2020); Koh et al. (2021) for", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 560, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 423, + 574 + ], + "score": 1.0, + "content": "domain generalization. In particular, where the original CORAL paper defines", + "type": "text" + }, + { + "bbox": [ + 423, + 561, + 452, + 573 + ], + "score": 0.85, + "content": "L _ { \\mathrm { p e n a l t y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 560, + 506, + 574 + ], + "score": 1.0, + "content": "as the diver-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 571, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 571, + 448, + 585 + ], + "score": 1.0, + "content": "gence between just two kinds of batches (labeled and unlabeled), these works define", + "type": "text" + }, + { + "bbox": [ + 449, + 572, + 478, + 584 + ], + "score": 0.88, + "content": "L _ { \\mathrm { p e n a l t y } }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 571, + 506, + 585 + ], + "score": 1.0, + "content": "as the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 595 + ], + "score": 1.0, + "content": "divergence between many kinds of batches, where batches are grouped based on domain annotation", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 592, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 593, + 122, + 605 + ], + "score": 0.88, + "content": "\\boldsymbol { d } ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 592, + 506, + 608 + ], + "score": 1.0, + "content": ". For simplicity, we followed the original CORAL formulation and differentiate only between", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 478, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 409, + 617 + ], + "score": 1.0, + "content": "labeled and unlabeled batches. We leave leveraging the domain adaptations", + "type": "text" + }, + { + "bbox": [ + 409, + 606, + 416, + 615 + ], + "score": 0.79, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 605, + 478, + 617 + ], + "score": 1.0, + "content": "to future work.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 550, + 506, + 617 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 630, + 504, + 686 + ], + "lines": [ + { + "bbox": [ + 106, + 631, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 641 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We run CORAL on all datasets except GLOBALWHEAT-WILDS and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 641, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 505, + 653 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS. We do not evaluate domain invariant methods on CIVILCOMMENTS-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 651, + 506, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 506, + 665 + ], + "score": 1.0, + "content": "WILDS, since the labeled and unlabeled data are drawn from the same distribution. We do not", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 662, + 506, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 677 + ], + "score": 1.0, + "content": "evaluate CORAL on GLOBALWHEAT-WILDS because CORAL does not port straightforwardly to", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 672, + 181, + 688 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 181, + 688 + ], + "score": 1.0, + "content": "detection settings.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 631, + 506, + 688 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "DANN. Algorithm 2 describes DANN, proposed by Ganin et al. (2016). DANN measures the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 152, + 722 + ], + "score": 1.0, + "content": "divergence", + "type": "text" + }, + { + "bbox": [ + 153, + 710, + 159, + 721 + ], + "score": 0.83, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "between batches of feature representations using the performance of a discriminator", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 141, + 733 + ], + "score": 1.0, + "content": "network", + "type": "text" + }, + { + "bbox": [ + 141, + 721, + 153, + 732 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 720, + 458, + 733 + ], + "score": 1.0, + "content": "that aims to discriminate between domains. Given a batch of features (either", + "type": "text" + }, + { + "bbox": [ + 459, + 721, + 472, + 732 + ], + "score": 0.89, + "content": "B _ { \\mathrm { L } }", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 720, + 483, + 733 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 484, + 721, + 498, + 732 + ], + "score": 0.86, + "content": "B _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 499, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "),", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 698, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 85, + 199, + 97 + ], + "lines": [ + { + "bbox": [ + 106, + 84, + 201, + 100 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 201, + 100 + ], + "score": 1.0, + "content": "Algorithm 1: CORAL", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 99, + 478, + 140 + ], + "lines": [ + { + "bbox": [ + 102, + 96, + 412, + 118 + ], + "spans": [ + { + "bbox": [ + 102, + 96, + 196, + 118 + ], + "score": 1.0, + "content": "Input: Labeled batch", + "type": "text" + }, + { + "bbox": [ + 196, + 100, + 339, + 115 + ], + "score": 0.91, + "content": "\\{ ( x _ { \\mathrm { L } } ^ { ( i ) } , y _ { \\mathrm { L } } ^ { ( i ) } , d _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 96, + 412, + 118 + ], + "score": 1.0, + "content": ", unlabeled batch", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 137, + 110, + 482, + 135 + ], + "spans": [ + { + "bbox": [ + 137, + 114, + 262, + 129 + ], + "score": 0.88, + "content": "\\{ ( x _ { \\mathrm { U } } ^ { ( i ) } , d _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( \\bar { 1 } , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 110, + 385, + 135 + ], + "score": 1.0, + "content": ", strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 386, + 117, + 412, + 129 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 110, + 482, + 135 + ], + "score": 1.0, + "content": ", penalty weight", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 137, + 127, + 325, + 140 + ], + "spans": [ + { + "bbox": [ + 137, + 129, + 163, + 138 + ], + "score": 0.83, + "content": "\\lambda \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 127, + 315, + 140 + ], + "score": 1.0, + "content": ", dimension of feature representations", + "type": "text" + }, + { + "bbox": [ + 315, + 130, + 325, + 138 + ], + "score": 0.67, + "content": "m", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 99, + 139, + 374, + 150 + ], + "lines": [ + { + "bbox": [ + 99, + 138, + 374, + 151 + ], + "spans": [ + { + "bbox": [ + 99, + 138, + 374, + 151 + ], + "score": 1.0, + "content": "1 Compute feature representations for labeled and unlabeled batches", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 220, + 156, + 396, + 192 + ], + "lines": [ + { + "bbox": [ + 220, + 156, + 396, + 192 + ], + "spans": [ + { + "bbox": [ + 220, + 156, + 396, + 192 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { B _ { \\mathrm { L } } = \\{ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot \\cdot , n _ { \\mathrm { L } } ) \\} } \\\\ & { B _ { \\mathrm { U } } = \\{ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\} } \\end{array}", + "type": "interline_equation", + "image_path": "28d8b9a06e3af1aa7c565afb7b13b017db27a66e5f161744f2fb06945419d18d.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 220, + 156, + 396, + 174.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 220, + 174.0, + 396, + 192.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 99, + 195, + 401, + 208 + ], + "lines": [ + { + "bbox": [ + 97, + 195, + 402, + 209 + ], + "spans": [ + { + "bbox": [ + 97, + 195, + 402, + 209 + ], + "score": 1.0, + "content": "2 Compute feature mean and covariances for labeled and unlabeled batches", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 208, + 213, + 408, + 321 + ], + "lines": [ + { + "bbox": [ + 208, + 213, + 408, + 321 + ], + "spans": [ + { + "bbox": [ + 208, + 213, + 408, + 321 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\mu _ { \\mathrm { L } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { L } } } \\\\ & { \\mu _ { \\mathrm { U } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } } \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { U } } } \\\\ & { C _ { \\mathrm { L } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } - 1 } \\bigg ( \\boldsymbol { B } _ { \\mathrm { L } } ^ { T } \\boldsymbol { B } _ { \\mathrm { L } } - \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } \\big ( \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { L } } \\big ) ^ { T } \\big ( \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { L } } \\big ) \\bigg ) } \\\\ & { C _ { \\mathrm { U } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } - 1 } \\bigg ( \\boldsymbol { B } _ { \\mathrm { U } } ^ { T } \\boldsymbol { B } _ { \\mathrm { U } } - \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } } \\big ( \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { U } } \\big ) ^ { T } \\big ( \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { U } } \\big ) \\bigg ) } \\end{array}", + "type": "interline_equation", + "image_path": "a4aab7451b43947e90847c847d8f51666a640a5c80f14c57008f98c071a01874.jpg" + } + ] + } + ], + "index": 11.5, + "virtual_lines": [ + { + "bbox": [ + 208, + 213, + 408, + 226.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 208, + 226.5, + 408, + 240.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 208, + 240.0, + 408, + 253.5 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 208, + 253.5, + 408, + 267.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 208, + 267.0, + 408, + 280.5 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 208, + 280.5, + 408, + 294.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 208, + 294.0, + 408, + 307.5 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 208, + 307.5, + 408, + 321.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 98, + 325, + 237, + 337 + ], + "lines": [ + { + "bbox": [ + 97, + 325, + 238, + 339 + ], + "spans": [ + { + "bbox": [ + 97, + 325, + 165, + 339 + ], + "score": 1.0, + "content": "3 Update model", + "type": "text" + }, + { + "bbox": [ + 165, + 326, + 206, + 337 + ], + "score": 0.92, + "content": "h = g \\circ f", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 325, + 238, + 339 + ], + "score": 1.0, + "content": "on loss", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 170, + 343, + 445, + 376 + ], + "lines": [ + { + "bbox": [ + 170, + 343, + 445, + 376 + ], + "spans": [ + { + "bbox": [ + 170, + 343, + 445, + 376 + ], + "score": 0.92, + "content": "\\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } \\big ) + \\lambda \\left( | | \\mu _ { \\mathrm { L } } - \\mu _ { \\mathrm { U } } | | ^ { 2 } + | | C _ { \\mathrm { L } } - C _ { \\mathrm { U } } | | _ { F } ^ { 2 } \\right)", + "type": "interline_equation", + "image_path": "0f95fb7704a3d68c48d59481837a3ef521f8cc338253a21db82c2fe23a38251f.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 170, + 343, + 445, + 354.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 170, + 354.0, + 445, + 365.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 170, + 365.0, + 445, + 376.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 406, + 503, + 429 + ], + "lines": [ + { + "bbox": [ + 106, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 181, + 419 + ], + "score": 1.0, + "content": "this deep network", + "type": "text" + }, + { + "bbox": [ + 181, + 407, + 193, + 418 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "must classify whether examples are from the labeled data or unlabeled data.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 417, + 302, + 430 + ], + "spans": [ + { + "bbox": [ + 107, + 418, + 118, + 429 + ], + "score": 0.88, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 417, + 302, + 430 + ], + "score": 1.0, + "content": "is optimized using a binary classification loss", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 431, + 468, + 464 + ], + "lines": [ + { + "bbox": [ + 142, + 431, + 468, + 464 + ], + "spans": [ + { + "bbox": [ + 142, + 431, + 468, + 464 + ], + "score": 0.94, + "content": "L ( h _ { d } ) = \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell ( h _ { d } \\circ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , 1 ) + \\frac { 1 } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\ell ( h _ { d } \\circ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) , 0 )", + "type": "interline_equation", + "image_path": "7f2312d7e10587604b0f7d813077989ca0770c07e6e23e3552d645f99ad08881.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 142, + 431, + 468, + 442.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 142, + 442.0, + 468, + 453.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 142, + 453.0, + 468, + 464.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 465, + 264, + 476 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 264, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 153, + 478 + ], + "score": 1.0, + "content": "The loss of", + "type": "text" + }, + { + "bbox": [ + 154, + 465, + 165, + 476 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 462, + 246, + 478 + ], + "score": 1.0, + "content": "is exactly related to", + "type": "text" + }, + { + "bbox": [ + 246, + 465, + 252, + 477 + ], + "score": 0.83, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 462, + 264, + 478 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "interline_equation", + "bbox": [ + 260, + 478, + 351, + 492 + ], + "lines": [ + { + "bbox": [ + 260, + 478, + 351, + 492 + ], + "spans": [ + { + "bbox": [ + 260, + 478, + 351, + 492 + ], + "score": 0.94, + "content": "\\xi ( B _ { \\mathrm { L } } , B _ { \\mathrm { U } } ) = - L ( h _ { d } )", + "type": "interline_equation", + "image_path": "ae361de6bf37d9c00f76352d72460355c46c8722c895acddbf5a459b1c7f9546.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 260, + 478, + 351, + 492 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 493, + 504, + 526 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 255, + 506 + ], + "score": 1.0, + "content": "In other words, at the same time that", + "type": "text" + }, + { + "bbox": [ + 256, + 494, + 267, + 504 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 492, + 401, + 506 + ], + "score": 1.0, + "content": "is optimized to minimize its loss", + "type": "text" + }, + { + "bbox": [ + 402, + 493, + 427, + 505 + ], + "score": 0.9, + "content": "L ( h _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 492, + 487, + 506 + ], + "score": 1.0, + "content": ", the featurizer", + "type": "text" + }, + { + "bbox": [ + 487, + 494, + 495, + 505 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 492, + 506, + 506 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 503, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 503, + 209, + 518 + ], + "score": 1.0, + "content": "incentivized to minimize", + "type": "text" + }, + { + "bbox": [ + 210, + 504, + 345, + 517 + ], + "score": 0.9, + "content": "L _ { \\mathrm { p e n a l t y } } = \\xi ( B _ { \\mathrm { L } } , B _ { \\mathrm { U } } ) = - L ( h _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 503, + 403, + 518 + ], + "score": 1.0, + "content": ", or maximize", + "type": "text" + }, + { + "bbox": [ + 403, + 505, + 429, + 516 + ], + "score": 0.92, + "content": "L ( h _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 503, + 506, + 518 + ], + "score": 1.0, + "content": ". See Algorithm 2", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 514, + 152, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 152, + 527 + ], + "score": 1.0, + "content": "for details.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 531, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "score": 1.0, + "content": "We adapted our implementation from the Transfer Learning Library (Jiang et al., 2020) and matched", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 543, + 504, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 504, + 554 + ], + "score": 1.0, + "content": "all details to the formulation given by Ganin et al. (2016), except for two changes. On applicable", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 387, + 567 + ], + "score": 1.0, + "content": "datasets, we have strongly all labeled and unlabeled examples using", + "type": "text" + }, + { + "bbox": [ + 388, + 555, + 414, + 566 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 554, + 505, + 567 + ], + "score": 1.0, + "content": ", whereas Ganin et al.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 565, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 506, + 577 + ], + "score": 1.0, + "content": "(2016) do not explicitly require data augmentations. We add augmentations to allow for a fairer", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "comparison to other methods which use augmentations. Second, where Ganin et al. (2016) opti-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 584, + 507, + 602 + ], + "spans": [ + { + "bbox": [ + 104, + 584, + 128, + 602 + ], + "score": 1.0, + "content": "mize", + "type": "text" + }, + { + "bbox": [ + 129, + 587, + 147, + 599 + ], + "score": 0.58, + "content": "f , g", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 584, + 168, + 602 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 168, + 587, + 180, + 598 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 584, + 298, + 602 + ], + "score": 1.0, + "content": "using the same learning rate", + "type": "text" + }, + { + "bbox": [ + 298, + 589, + 305, + 599 + ], + "score": 0.76, + "content": "\\eta", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 584, + 457, + 602 + ], + "score": 1.0, + "content": ", we use three different learning rates", + "type": "text" + }, + { + "bbox": [ + 457, + 589, + 501, + 599 + ], + "score": 0.9, + "content": "\\eta _ { f } , \\eta _ { g } , \\eta _ { h _ { d } }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 584, + 507, + 602 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 598, + 438, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 598, + 438, + 610 + ], + "score": 1.0, + "content": "following the implementation of the Transfer Learning Library (Jiang et al., 2020).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 621, + 504, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 622, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 632 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We run DANN on all datasets except GLOBALWHEAT-WILDS and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 630, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 506, + 645 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS. We do not evaluate domain invariant methods on CIVILCOMMENTS-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "WILDS, since the labeled and unlabeled data are drawn from the same distribution. We do not evalu-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 653, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 506, + 667 + ], + "score": 1.0, + "content": "ate DANN on GLOBALWHEAT-WILDS because DANN does not port straightforwardly to detection", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 663, + 142, + 679 + ], + "spans": [ + { + "bbox": [ + 104, + 663, + 142, + 679 + ], + "score": 1.0, + "content": "settings.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 39 + }, + { + "type": "title", + "bbox": [ + 108, + 689, + 245, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 689, + 247, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 247, + 701 + ], + "score": 1.0, + "content": "B.3 SELF-TRAINING METHODS", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Self-training methods leverage unlabeled data by “pseudo-labeling” unlabeled examples with the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "model’s own predictions and training on them as if they were labeled examples. In certain for-", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + } + ], + "page_idx": 36, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 85, + 199, + 97 + ], + "lines": [ + { + "bbox": [ + 106, + 84, + 201, + 100 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 201, + 100 + ], + "score": 1.0, + "content": "Algorithm 1: CORAL", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 99, + 478, + 140 + ], + "lines": [ + { + "bbox": [ + 102, + 96, + 412, + 118 + ], + "spans": [ + { + "bbox": [ + 102, + 96, + 196, + 118 + ], + "score": 1.0, + "content": "Input: Labeled batch", + "type": "text" + }, + { + "bbox": [ + 196, + 100, + 339, + 115 + ], + "score": 0.91, + "content": "\\{ ( x _ { \\mathrm { L } } ^ { ( i ) } , y _ { \\mathrm { L } } ^ { ( i ) } , d _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 96, + 412, + 118 + ], + "score": 1.0, + "content": ", unlabeled batch", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 137, + 110, + 482, + 135 + ], + "spans": [ + { + "bbox": [ + 137, + 114, + 262, + 129 + ], + "score": 0.88, + "content": "\\{ ( x _ { \\mathrm { U } } ^ { ( i ) } , d _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( \\bar { 1 } , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 110, + 385, + 135 + ], + "score": 1.0, + "content": ", strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 386, + 117, + 412, + 129 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 110, + 482, + 135 + ], + "score": 1.0, + "content": ", penalty weight", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 137, + 127, + 325, + 140 + ], + "spans": [ + { + "bbox": [ + 137, + 129, + 163, + 138 + ], + "score": 0.83, + "content": "\\lambda \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 127, + 315, + 140 + ], + "score": 1.0, + "content": ", dimension of feature representations", + "type": "text" + }, + { + "bbox": [ + 315, + 130, + 325, + 138 + ], + "score": 0.67, + "content": "m", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 102, + 96, + 482, + 140 + ] + }, + { + "type": "text", + "bbox": [ + 99, + 139, + 374, + 150 + ], + "lines": [ + { + "bbox": [ + 99, + 138, + 374, + 151 + ], + "spans": [ + { + "bbox": [ + 99, + 138, + 374, + 151 + ], + "score": 1.0, + "content": "1 Compute feature representations for labeled and unlabeled batches", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 99, + 138, + 374, + 151 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 220, + 156, + 396, + 192 + ], + "lines": [ + { + "bbox": [ + 220, + 156, + 396, + 192 + ], + "spans": [ + { + "bbox": [ + 220, + 156, + 396, + 192 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { B _ { \\mathrm { L } } = \\{ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot \\cdot , n _ { \\mathrm { L } } ) \\} } \\\\ & { B _ { \\mathrm { U } } = \\{ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\} } \\end{array}", + "type": "interline_equation", + "image_path": "28d8b9a06e3af1aa7c565afb7b13b017db27a66e5f161744f2fb06945419d18d.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 220, + 156, + 396, + 174.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 220, + 174.0, + 396, + 192.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 99, + 195, + 401, + 208 + ], + "lines": [ + { + "bbox": [ + 97, + 195, + 402, + 209 + ], + "spans": [ + { + "bbox": [ + 97, + 195, + 402, + 209 + ], + "score": 1.0, + "content": "2 Compute feature mean and covariances for labeled and unlabeled batches", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 97, + 195, + 402, + 209 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 208, + 213, + 408, + 321 + ], + "lines": [ + { + "bbox": [ + 208, + 213, + 408, + 321 + ], + "spans": [ + { + "bbox": [ + 208, + 213, + 408, + 321 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\mu _ { \\mathrm { L } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { L } } } \\\\ & { \\mu _ { \\mathrm { U } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } } \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { U } } } \\\\ & { C _ { \\mathrm { L } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } - 1 } \\bigg ( \\boldsymbol { B } _ { \\mathrm { L } } ^ { T } \\boldsymbol { B } _ { \\mathrm { L } } - \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } \\big ( \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { L } } \\big ) ^ { T } \\big ( \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { L } } \\big ) \\bigg ) } \\\\ & { C _ { \\mathrm { U } } = \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } - 1 } \\bigg ( \\boldsymbol { B } _ { \\mathrm { U } } ^ { T } \\boldsymbol { B } _ { \\mathrm { U } } - \\displaystyle \\frac { 1 } { n _ { \\mathrm { U } } } \\big ( \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { U } } \\big ) ^ { T } \\big ( \\boldsymbol { 1 } ^ { T } \\boldsymbol { B } _ { \\mathrm { U } } \\big ) \\bigg ) } \\end{array}", + "type": "interline_equation", + "image_path": "a4aab7451b43947e90847c847d8f51666a640a5c80f14c57008f98c071a01874.jpg" + } + ] + } + ], + "index": 11.5, + "virtual_lines": [ + { + "bbox": [ + 208, + 213, + 408, + 226.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 208, + 226.5, + 408, + 240.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 208, + 240.0, + 408, + 253.5 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 208, + 253.5, + 408, + 267.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 208, + 267.0, + 408, + 280.5 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 208, + 280.5, + 408, + 294.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 208, + 294.0, + 408, + 307.5 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 208, + 307.5, + 408, + 321.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 98, + 325, + 237, + 337 + ], + "lines": [ + { + "bbox": [ + 97, + 325, + 238, + 339 + ], + "spans": [ + { + "bbox": [ + 97, + 325, + 165, + 339 + ], + "score": 1.0, + "content": "3 Update model", + "type": "text" + }, + { + "bbox": [ + 165, + 326, + 206, + 337 + ], + "score": 0.92, + "content": "h = g \\circ f", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 325, + 238, + 339 + ], + "score": 1.0, + "content": "on loss", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 97, + 325, + 238, + 339 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 170, + 343, + 445, + 376 + ], + "lines": [ + { + "bbox": [ + 170, + 343, + 445, + 376 + ], + "spans": [ + { + "bbox": [ + 170, + 343, + 445, + 376 + ], + "score": 0.92, + "content": "\\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } \\big ) + \\lambda \\left( | | \\mu _ { \\mathrm { L } } - \\mu _ { \\mathrm { U } } | | ^ { 2 } + | | C _ { \\mathrm { L } } - C _ { \\mathrm { U } } | | _ { F } ^ { 2 } \\right)", + "type": "interline_equation", + "image_path": "0f95fb7704a3d68c48d59481837a3ef521f8cc338253a21db82c2fe23a38251f.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 170, + 343, + 445, + 354.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 170, + 354.0, + 445, + 365.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 170, + 365.0, + 445, + 376.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 406, + 503, + 429 + ], + "lines": [ + { + "bbox": [ + 106, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 181, + 419 + ], + "score": 1.0, + "content": "this deep network", + "type": "text" + }, + { + "bbox": [ + 181, + 407, + 193, + 418 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "must classify whether examples are from the labeled data or unlabeled data.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 417, + 302, + 430 + ], + "spans": [ + { + "bbox": [ + 107, + 418, + 118, + 429 + ], + "score": 0.88, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 118, + 417, + 302, + 430 + ], + "score": 1.0, + "content": "is optimized using a binary classification loss", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 106, + 406, + 505, + 430 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 431, + 468, + 464 + ], + "lines": [ + { + "bbox": [ + 142, + 431, + 468, + 464 + ], + "spans": [ + { + "bbox": [ + 142, + 431, + 468, + 464 + ], + "score": 0.94, + "content": "L ( h _ { d } ) = \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell ( h _ { d } \\circ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , 1 ) + \\frac { 1 } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\ell ( h _ { d } \\circ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) , 0 )", + "type": "interline_equation", + "image_path": "7f2312d7e10587604b0f7d813077989ca0770c07e6e23e3552d645f99ad08881.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 142, + 431, + 468, + 442.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 142, + 442.0, + 468, + 453.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 142, + 453.0, + 468, + 464.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 465, + 264, + 476 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 264, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 153, + 478 + ], + "score": 1.0, + "content": "The loss of", + "type": "text" + }, + { + "bbox": [ + 154, + 465, + 165, + 476 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 462, + 246, + 478 + ], + "score": 1.0, + "content": "is exactly related to", + "type": "text" + }, + { + "bbox": [ + 246, + 465, + 252, + 477 + ], + "score": 0.83, + "content": "\\xi", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 462, + 264, + 478 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 462, + 264, + 478 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 260, + 478, + 351, + 492 + ], + "lines": [ + { + "bbox": [ + 260, + 478, + 351, + 492 + ], + "spans": [ + { + "bbox": [ + 260, + 478, + 351, + 492 + ], + "score": 0.94, + "content": "\\xi ( B _ { \\mathrm { L } } , B _ { \\mathrm { U } } ) = - L ( h _ { d } )", + "type": "interline_equation", + "image_path": "ae361de6bf37d9c00f76352d72460355c46c8722c895acddbf5a459b1c7f9546.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 260, + 478, + 351, + 492 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 493, + 504, + 526 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 255, + 506 + ], + "score": 1.0, + "content": "In other words, at the same time that", + "type": "text" + }, + { + "bbox": [ + 256, + 494, + 267, + 504 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 492, + 401, + 506 + ], + "score": 1.0, + "content": "is optimized to minimize its loss", + "type": "text" + }, + { + "bbox": [ + 402, + 493, + 427, + 505 + ], + "score": 0.9, + "content": "L ( h _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 492, + 487, + 506 + ], + "score": 1.0, + "content": ", the featurizer", + "type": "text" + }, + { + "bbox": [ + 487, + 494, + 495, + 505 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 492, + 506, + 506 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 503, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 503, + 209, + 518 + ], + "score": 1.0, + "content": "incentivized to minimize", + "type": "text" + }, + { + "bbox": [ + 210, + 504, + 345, + 517 + ], + "score": 0.9, + "content": "L _ { \\mathrm { p e n a l t y } } = \\xi ( B _ { \\mathrm { L } } , B _ { \\mathrm { U } } ) = - L ( h _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 503, + 403, + 518 + ], + "score": 1.0, + "content": ", or maximize", + "type": "text" + }, + { + "bbox": [ + 403, + 505, + 429, + 516 + ], + "score": 0.92, + "content": "L ( h _ { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 503, + 506, + 518 + ], + "score": 1.0, + "content": ". See Algorithm 2", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 514, + 152, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 152, + 527 + ], + "score": 1.0, + "content": "for details.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 104, + 492, + 506, + 527 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 531, + 505, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "score": 1.0, + "content": "We adapted our implementation from the Transfer Learning Library (Jiang et al., 2020) and matched", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 543, + 504, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 504, + 554 + ], + "score": 1.0, + "content": "all details to the formulation given by Ganin et al. (2016), except for two changes. On applicable", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 387, + 567 + ], + "score": 1.0, + "content": "datasets, we have strongly all labeled and unlabeled examples using", + "type": "text" + }, + { + "bbox": [ + 388, + 555, + 414, + 566 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 554, + 505, + 567 + ], + "score": 1.0, + "content": ", whereas Ganin et al.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 565, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 106, + 565, + 506, + 577 + ], + "score": 1.0, + "content": "(2016) do not explicitly require data augmentations. We add augmentations to allow for a fairer", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "comparison to other methods which use augmentations. Second, where Ganin et al. (2016) opti-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 584, + 507, + 602 + ], + "spans": [ + { + "bbox": [ + 104, + 584, + 128, + 602 + ], + "score": 1.0, + "content": "mize", + "type": "text" + }, + { + "bbox": [ + 129, + 587, + 147, + 599 + ], + "score": 0.58, + "content": "f , g", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 584, + 168, + 602 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 168, + 587, + 180, + 598 + ], + "score": 0.89, + "content": "h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 584, + 298, + 602 + ], + "score": 1.0, + "content": "using the same learning rate", + "type": "text" + }, + { + "bbox": [ + 298, + 589, + 305, + 599 + ], + "score": 0.76, + "content": "\\eta", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 584, + 457, + 602 + ], + "score": 1.0, + "content": ", we use three different learning rates", + "type": "text" + }, + { + "bbox": [ + 457, + 589, + 501, + 599 + ], + "score": 0.9, + "content": "\\eta _ { f } , \\eta _ { g } , \\eta _ { h _ { d } }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 584, + 507, + 602 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 598, + 438, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 598, + 438, + 610 + ], + "score": 1.0, + "content": "following the implementation of the Transfer Learning Library (Jiang et al., 2020).", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33, + "bbox_fs": [ + 104, + 532, + 507, + 610 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 621, + 504, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 622, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 632 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We run DANN on all datasets except GLOBALWHEAT-WILDS and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 630, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 506, + 645 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS. We do not evaluate domain invariant methods on CIVILCOMMENTS-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "WILDS, since the labeled and unlabeled data are drawn from the same distribution. We do not evalu-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 653, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 506, + 667 + ], + "score": 1.0, + "content": "ate DANN on GLOBALWHEAT-WILDS because DANN does not port straightforwardly to detection", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 663, + 142, + 679 + ], + "spans": [ + { + "bbox": [ + 104, + 663, + 142, + 679 + ], + "score": 1.0, + "content": "settings.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 39, + "bbox_fs": [ + 104, + 622, + 506, + 679 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 689, + 245, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 689, + 247, + 701 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 247, + 701 + ], + "score": 1.0, + "content": "B.3 SELF-TRAINING METHODS", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Self-training methods leverage unlabeled data by “pseudo-labeling” unlabeled examples with the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "model’s own predictions and training on them as if they were labeled examples. In certain for-", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 106, + 709, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 85, + 194, + 97 + ], + "lines": [ + { + "bbox": [ + 105, + 84, + 195, + 99 + ], + "spans": [ + { + "bbox": [ + 105, + 84, + 195, + 99 + ], + "score": 1.0, + "content": "Algorithm 2: DANN", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 99, + 477, + 139 + ], + "lines": [ + { + "bbox": [ + 101, + 93, + 413, + 121 + ], + "spans": [ + { + "bbox": [ + 101, + 93, + 196, + 121 + ], + "score": 1.0, + "content": "Input: Labeled batch", + "type": "text" + }, + { + "bbox": [ + 196, + 100, + 339, + 115 + ], + "score": 0.91, + "content": "\\{ ( x _ { \\mathrm { L } } ^ { ( i ) } , y _ { \\mathrm { L } } ^ { ( i ) } , d _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 93, + 413, + 121 + ], + "score": 1.0, + "content": ", unlabeled batch", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 137, + 109, + 482, + 136 + ], + "spans": [ + { + "bbox": [ + 137, + 114, + 262, + 129 + ], + "score": 0.87, + "content": "\\{ ( x _ { \\mathrm { U } } ^ { ( i ) } , d _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( \\bar { 1 } , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 109, + 385, + 136 + ], + "score": 1.0, + "content": ", strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 386, + 117, + 412, + 129 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 109, + 482, + 136 + ], + "score": 1.0, + "content": ", penalty weight", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 137, + 126, + 267, + 141 + ], + "spans": [ + { + "bbox": [ + 137, + 129, + 163, + 138 + ], + "score": 0.83, + "content": "\\lambda \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 126, + 223, + 141 + ], + "score": 1.0, + "content": ", learning rates", + "type": "text" + }, + { + "bbox": [ + 224, + 130, + 267, + 140 + ], + "score": 0.8, + "content": "\\eta _ { f } , \\eta _ { g } , \\eta _ { h _ { d } }", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 101, + 139, + 278, + 150 + ], + "lines": [ + { + "bbox": [ + 99, + 137, + 277, + 152 + ], + "spans": [ + { + "bbox": [ + 99, + 137, + 266, + 152 + ], + "score": 1.0, + "content": "1 Compute loss for domain discriminator", + "type": "text" + }, + { + "bbox": [ + 266, + 140, + 277, + 150 + ], + "score": 0.86, + "content": "h _ { d }", + "type": "inline_equation" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 156, + 470, + 189 + ], + "lines": [ + { + "bbox": [ + 145, + 156, + 470, + 189 + ], + "spans": [ + { + "bbox": [ + 145, + 156, + 470, + 189 + ], + "score": 0.92, + "content": "L ( h _ { d } ) = \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell ( h _ { d } \\circ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , 1 ) + \\frac { 1 } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\ell ( h _ { d } \\circ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) , 0 )", + "type": "interline_equation", + "image_path": "a0a2cff673d28acb708c8f45fd8d53e3b79f9b1b2bddc2c507ffaf8465f8ee7d.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 145, + 156, + 470, + 167.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 145, + 167.0, + 470, + 178.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 145, + 178.0, + 470, + 189.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 99, + 195, + 247, + 207 + ], + "lines": [ + { + "bbox": [ + 97, + 194, + 246, + 208 + ], + "spans": [ + { + "bbox": [ + 97, + 194, + 205, + 208 + ], + "score": 1.0, + "content": "2 Compute loss for model", + "type": "text" + }, + { + "bbox": [ + 205, + 195, + 246, + 207 + ], + "score": 0.91, + "content": "h = g \\circ f", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 213, + 394, + 245 + ], + "lines": [ + { + "bbox": [ + 221, + 213, + 394, + 245 + ], + "spans": [ + { + "bbox": [ + 221, + 213, + 394, + 245 + ], + "score": 0.93, + "content": "\\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } \\big ) - \\lambda L ( h _ { d } )", + "type": "interline_equation", + "image_path": "686196a8fcfdf7abd6d0a8cb0b3d56d23019f72be056036211ddaa4937fc13d7.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 221, + 213, + 394, + 229.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 221, + 229.0, + 394, + 245.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 98, + 251, + 342, + 263 + ], + "lines": [ + { + "bbox": [ + 96, + 248, + 342, + 267 + ], + "spans": [ + { + "bbox": [ + 96, + 248, + 137, + 267 + ], + "score": 1.0, + "content": "3 Update", + "type": "text" + }, + { + "bbox": [ + 138, + 251, + 169, + 263 + ], + "score": 0.91, + "content": "f , g , h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 248, + 298, + 267 + ], + "score": 1.0, + "content": "using appropriate learning rates", + "type": "text" + }, + { + "bbox": [ + 298, + 253, + 342, + 263 + ], + "score": 0.86, + "content": "\\eta _ { f } , \\eta _ { g } , \\eta _ { h _ { d } }", + "type": "inline_equation" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 289, + 505, + 356 + ], + "lines": [ + { + "bbox": [ + 106, + 290, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 505, + 301 + ], + "score": 1.0, + "content": "mulations, this is equivalent to minimizing the model’s conditional entropy on the unlabeled data", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 301, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 107, + 301, + 505, + 312 + ], + "score": 1.0, + "content": "(Grandvalet & Bengio, 2005). Contemporary self-training methods also often make use of consis-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 311, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 324 + ], + "score": 1.0, + "content": "tency regularization, i.e., encouraging the model to make similar predictions on noisy/augmented", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 323, + 504, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 504, + 334 + ], + "score": 1.0, + "content": "versions of unlabeled examples. Self-training methods have recently been shown to be empirically", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 333, + 504, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 504, + 345 + ], + "score": 1.0, + "content": "successful at unsupervised domain adaptation (Saito et al., 2017; Berthelot et al., 2021; Zhang et al.,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 343, + 136, + 357 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 136, + 357 + ], + "score": 1.0, + "content": "2021).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 361, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 360, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 488, + 375 + ], + "score": 1.0, + "content": "The self-training methods we study follow this general structure: given an unlabeled example", + "type": "text" + }, + { + "bbox": [ + 489, + 364, + 501, + 372 + ], + "score": 0.8, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 360, + 505, + 375 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 371, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 104, + 371, + 265, + 385 + ], + "score": 1.0, + "content": "these algorithms generate a pseudolabel", + "type": "text" + }, + { + "bbox": [ + 266, + 372, + 316, + 384 + ], + "score": 0.93, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\psi ( \\bar { x } _ { \\mathrm { U } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 371, + 490, + 385 + ], + "score": 1.0, + "content": ", where the pseudolabel-generating function", + "type": "text" + }, + { + "bbox": [ + 490, + 373, + 498, + 384 + ], + "score": 0.72, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 371, + 505, + 385 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 141, + 394 + ], + "score": 0.91, + "content": "\\mathcal { X } \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "differs between algorithms. For classification problems, we study algorithms that produce", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 395, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 406 + ], + "score": 1.0, + "content": "hard pseudolabels, which are one-hot class predictions, rather than soft pseudolabels, which are", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 405, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 436, + 417 + ], + "score": 1.0, + "content": "continuous distributions over the classes. Next, algorithms define an unlabeled loss", + "type": "text" + }, + { + "bbox": [ + 436, + 405, + 463, + 417 + ], + "score": 0.94, + "content": "L _ { \\mathrm { U } } ( h )", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 405, + 506, + 417 + ], + "score": 1.0, + "content": "for model", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 416, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 114, + 426 + ], + "score": 0.71, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 416, + 298, + 429 + ], + "score": 1.0, + "content": "by minimizing the loss between pseudolabels", + "type": "text" + }, + { + "bbox": [ + 299, + 417, + 310, + 428 + ], + "score": 0.87, + "content": "\\tilde { y } _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 416, + 506, + 429 + ], + "score": 1.0, + "content": "and the model’s predictions. The algorithms we", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 427, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 205, + 440 + ], + "score": 1.0, + "content": "consider below augment", + "type": "text" + }, + { + "bbox": [ + 205, + 429, + 218, + 438 + ], + "score": 0.84, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 427, + 461, + 440 + ], + "score": 1.0, + "content": "during training; i.e., rather than minimizing the loss between", + "type": "text" + }, + { + "bbox": [ + 461, + 428, + 473, + 438 + ], + "score": 0.85, + "content": "\\tilde { y } _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 427, + 505, + 440 + ], + "score": 1.0, + "content": "and the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 197, + 450 + ], + "score": 1.0, + "content": "model’s prediction on", + "type": "text" + }, + { + "bbox": [ + 197, + 440, + 209, + 450 + ], + "score": 0.85, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 439, + 446, + 450 + ], + "score": 1.0, + "content": ", the algorithms below minimize the loss of predictions on", + "type": "text" + }, + { + "bbox": [ + 447, + 438, + 474, + 450 + ], + "score": 0.92, + "content": "A ( x _ { \\mathrm { U } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 439, + 505, + 450 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 448, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 107, + 449, + 115, + 459 + ], + "score": 0.77, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 448, + 373, + 462 + ], + "score": 1.0, + "content": "is a stochastic, label-preserving augmentation. Assuming model", + "type": "text" + }, + { + "bbox": [ + 374, + 450, + 380, + 459 + ], + "score": 0.82, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 448, + 505, + 462 + ], + "score": 1.0, + "content": "that outputs real-valued logits,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 461, + 228, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 228, + 472 + ], + "score": 1.0, + "content": "the complete unlabeled loss is", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 22.5 + }, + { + "type": "interline_equation", + "bbox": [ + 230, + 477, + 381, + 509 + ], + "lines": [ + { + "bbox": [ + 230, + 477, + 381, + 509 + ], + "spans": [ + { + "bbox": [ + 230, + 477, + 381, + 509 + ], + "score": 0.94, + "content": "L _ { \\mathrm { U } } ( h ) = \\frac { 1 } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\ell \\big ( h \\circ A ( x _ { \\mathrm { U } } ^ { ( i ) } ) , \\tilde { y } _ { \\mathrm { U } } ^ { ( i ) } \\big )", + "type": "interline_equation", + "image_path": "d1a9730f30eeeb6789c4a37c722914d98b7ab20513e77a495d91e8720b261fc8.jpg" + } + ] + } + ], + "index": 28.5, + "virtual_lines": [ + { + "bbox": [ + 230, + 477, + 381, + 493.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 230, + 493.0, + 381, + 509.0 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 513, + 506, + 536 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 505, + 526 + ], + "score": 1.0, + "content": "This unlabeled loss is jointly optimized with the standard ERM labeled loss. The balance between", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 524, + 479, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 294, + 537 + ], + "score": 1.0, + "content": "the two losses is controlled by hyperparameter", + "type": "text" + }, + { + "bbox": [ + 294, + 525, + 312, + 537 + ], + "score": 0.91, + "content": "\\lambda ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 524, + 470, + 537 + ], + "score": 1.0, + "content": ", which is a function of the current step", + "type": "text" + }, + { + "bbox": [ + 470, + 526, + 474, + 534 + ], + "score": 0.7, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 524, + 479, + 537 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "interline_equation", + "bbox": [ + 247, + 541, + 364, + 555 + ], + "lines": [ + { + "bbox": [ + 247, + 541, + 364, + 555 + ], + "spans": [ + { + "bbox": [ + 247, + 541, + 364, + 555 + ], + "score": 0.91, + "content": "L ( h ) = L _ { \\mathrm { L } } ( h ) + \\lambda ( t ) L _ { \\mathrm { U } } ( h )", + "type": "interline_equation", + "image_path": "c2723df428584cbe5aafb4dba0e1915d2b857ad0ad88b3b1b9fa2bf34da85e7c.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 247, + 541, + 364, + 555 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 565, + 505, + 599 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 505, + 579 + ], + "score": 1.0, + "content": "Pseudo-Label. Algorithm 3 describes Pseudo-Label, proposed by Lee (2013). In this algorithm,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 576, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 505, + 590 + ], + "score": 1.0, + "content": "the model dynamically generates pseudolabels and updates each batch. Formally, the pseudolabel-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 588, + 242, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 186, + 601 + ], + "score": 1.0, + "content": "generating function", + "type": "text" + }, + { + "bbox": [ + 187, + 588, + 195, + 600 + ], + "score": 0.86, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 588, + 242, + 601 + ], + "score": 1.0, + "content": "is given by", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 604, + 392, + 625 + ], + "lines": [ + { + "bbox": [ + 219, + 604, + 392, + 625 + ], + "spans": [ + { + "bbox": [ + 219, + 604, + 392, + 625 + ], + "score": 0.93, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\psi ( x _ { \\mathrm { U } } ) = \\arg \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "interline_equation", + "image_path": "4f96f3d98b0e2a21af0520bac17bcee4733cfb3c8b7c2f6f6cb0bd208b56d043.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 219, + 604, + 392, + 625 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 630, + 503, + 653 + ], + "lines": [ + { + "bbox": [ + 106, + 629, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 133, + 644 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 631, + 160, + 642 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 629, + 505, + 644 + ], + "score": 1.0, + "content": "is the strong augmentation function described in Appendix C. Pseudo-Label then com-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 641, + 447, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 447, + 653 + ], + "score": 1.0, + "content": "putes the loss between a strongly augmented example and its associated pseudolabel.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 657, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "In order to more fairly compare Pseudo-Label to FixMatch, we add on confidence thresholding to", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 669, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 505, + 681 + ], + "score": 1.0, + "content": "the Pseudo-Label algorithm, a feature also added in the implementation of Pseudo-Label by Sohn", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 680, + 505, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 692 + ], + "score": 1.0, + "content": "et al. (2020). When confidence thresholding, examples on which the model has low confidence have", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 467, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 306, + 704 + ], + "score": 1.0, + "content": "zero loss, i.e., for some threshold hyperparameter", + "type": "text" + }, + { + "bbox": [ + 306, + 693, + 312, + 701 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 690, + 397, + 704 + ], + "score": 1.0, + "content": ", the loss an example", + "type": "text" + }, + { + "bbox": [ + 397, + 693, + 410, + 702 + ], + "score": 0.84, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 690, + 467, + 704 + ], + "score": 1.0, + "content": "contributes is", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5 + }, + { + "type": "interline_equation", + "bbox": [ + 168, + 707, + 442, + 730 + ], + "lines": [ + { + "bbox": [ + 168, + 707, + 442, + 730 + ], + "spans": [ + { + "bbox": [ + 168, + 707, + 442, + 730 + ], + "score": 0.92, + "content": "\\mathbf { 1 } \\Big \\{ \\mathrm { S o f t m a x } \\Big ( \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ] \\Big ) \\geq \\tau \\Big \\} \\cdot \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) , y _ { \\mathrm { U } } \\big )", + "type": "interline_equation", + "image_path": "a5cb20ddaa31196454d73f3a34f904e8e7c658589557d6e3dda067a73b371c70.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 168, + 707, + 442, + 730 + ], + "spans": [], + "index": 43 + } + ] + } + ], + "page_idx": 37, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "38", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 85, + 194, + 97 + ], + "lines": [ + { + "bbox": [ + 105, + 84, + 195, + 99 + ], + "spans": [ + { + "bbox": [ + 105, + 84, + 195, + 99 + ], + "score": 1.0, + "content": "Algorithm 2: DANN", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 99, + 477, + 139 + ], + "lines": [ + { + "bbox": [ + 101, + 93, + 413, + 121 + ], + "spans": [ + { + "bbox": [ + 101, + 93, + 196, + 121 + ], + "score": 1.0, + "content": "Input: Labeled batch", + "type": "text" + }, + { + "bbox": [ + 196, + 100, + 339, + 115 + ], + "score": 0.91, + "content": "\\{ ( x _ { \\mathrm { L } } ^ { ( i ) } , y _ { \\mathrm { L } } ^ { ( i ) } , d _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 93, + 413, + 121 + ], + "score": 1.0, + "content": ", unlabeled batch", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 137, + 109, + 482, + 136 + ], + "spans": [ + { + "bbox": [ + 137, + 114, + 262, + 129 + ], + "score": 0.87, + "content": "\\{ ( x _ { \\mathrm { U } } ^ { ( i ) } , d _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( \\bar { 1 } , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 109, + 385, + 136 + ], + "score": 1.0, + "content": ", strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 386, + 117, + 412, + 129 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 109, + 482, + 136 + ], + "score": 1.0, + "content": ", penalty weight", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 137, + 126, + 267, + 141 + ], + "spans": [ + { + "bbox": [ + 137, + 129, + 163, + 138 + ], + "score": 0.83, + "content": "\\lambda \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 126, + 223, + 141 + ], + "score": 1.0, + "content": ", learning rates", + "type": "text" + }, + { + "bbox": [ + 224, + 130, + 267, + 140 + ], + "score": 0.8, + "content": "\\eta _ { f } , \\eta _ { g } , \\eta _ { h _ { d } }", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 101, + 93, + 482, + 141 + ] + }, + { + "type": "text", + "bbox": [ + 101, + 139, + 278, + 150 + ], + "lines": [ + { + "bbox": [ + 99, + 137, + 277, + 152 + ], + "spans": [ + { + "bbox": [ + 99, + 137, + 266, + 152 + ], + "score": 1.0, + "content": "1 Compute loss for domain discriminator", + "type": "text" + }, + { + "bbox": [ + 266, + 140, + 277, + 150 + ], + "score": 0.86, + "content": "h _ { d }", + "type": "inline_equation" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 99, + 137, + 277, + 152 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 156, + 470, + 189 + ], + "lines": [ + { + "bbox": [ + 145, + 156, + 470, + 189 + ], + "spans": [ + { + "bbox": [ + 145, + 156, + 470, + 189 + ], + "score": 0.92, + "content": "L ( h _ { d } ) = \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell ( h _ { d } \\circ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , 1 ) + \\frac { 1 } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\ell ( h _ { d } \\circ f \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) , 0 )", + "type": "interline_equation", + "image_path": "a0a2cff673d28acb708c8f45fd8d53e3b79f9b1b2bddc2c507ffaf8465f8ee7d.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 145, + 156, + 470, + 167.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 145, + 167.0, + 470, + 178.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 145, + 178.0, + 470, + 189.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 99, + 195, + 247, + 207 + ], + "lines": [ + { + "bbox": [ + 97, + 194, + 246, + 208 + ], + "spans": [ + { + "bbox": [ + 97, + 194, + 205, + 208 + ], + "score": 1.0, + "content": "2 Compute loss for model", + "type": "text" + }, + { + "bbox": [ + 205, + 195, + 246, + 207 + ], + "score": 0.91, + "content": "h = g \\circ f", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 97, + 194, + 246, + 208 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 213, + 394, + 245 + ], + "lines": [ + { + "bbox": [ + 221, + 213, + 394, + 245 + ], + "spans": [ + { + "bbox": [ + 221, + 213, + 394, + 245 + ], + "score": 0.93, + "content": "\\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } \\big ) - \\lambda L ( h _ { d } )", + "type": "interline_equation", + "image_path": "686196a8fcfdf7abd6d0a8cb0b3d56d23019f72be056036211ddaa4937fc13d7.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 221, + 213, + 394, + 229.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 221, + 229.0, + 394, + 245.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 98, + 251, + 342, + 263 + ], + "lines": [ + { + "bbox": [ + 96, + 248, + 342, + 267 + ], + "spans": [ + { + "bbox": [ + 96, + 248, + 137, + 267 + ], + "score": 1.0, + "content": "3 Update", + "type": "text" + }, + { + "bbox": [ + 138, + 251, + 169, + 263 + ], + "score": 0.91, + "content": "f , g , h _ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 248, + 298, + 267 + ], + "score": 1.0, + "content": "using appropriate learning rates", + "type": "text" + }, + { + "bbox": [ + 298, + 253, + 342, + 263 + ], + "score": 0.86, + "content": "\\eta _ { f } , \\eta _ { g } , \\eta _ { h _ { d } }", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 290, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 505, + 301 + ], + "score": 1.0, + "content": "mulations, this is equivalent to minimizing the model’s conditional entropy on the unlabeled data", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 301, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 107, + 301, + 505, + 312 + ], + "score": 1.0, + "content": "(Grandvalet & Bengio, 2005). Contemporary self-training methods also often make use of consis-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 311, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 324 + ], + "score": 1.0, + "content": "tency regularization, i.e., encouraging the model to make similar predictions on noisy/augmented", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 323, + 504, + 334 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 504, + 334 + ], + "score": 1.0, + "content": "versions of unlabeled examples. Self-training methods have recently been shown to be empirically", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 333, + 504, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 504, + 345 + ], + "score": 1.0, + "content": "successful at unsupervised domain adaptation (Saito et al., 2017; Berthelot et al., 2021; Zhang et al.,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 343, + 136, + 357 + ], + "spans": [ + { + "bbox": [ + 104, + 343, + 136, + 357 + ], + "score": 1.0, + "content": "2021).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 11, + "bbox_fs": [ + 96, + 248, + 342, + 267 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 289, + 505, + 356 + ], + "lines": [], + "index": 14.5, + "bbox_fs": [ + 104, + 290, + 506, + 357 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 361, + 505, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 360, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 488, + 375 + ], + "score": 1.0, + "content": "The self-training methods we study follow this general structure: given an unlabeled example", + "type": "text" + }, + { + "bbox": [ + 489, + 364, + 501, + 372 + ], + "score": 0.8, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 360, + 505, + 375 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 371, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 104, + 371, + 265, + 385 + ], + "score": 1.0, + "content": "these algorithms generate a pseudolabel", + "type": "text" + }, + { + "bbox": [ + 266, + 372, + 316, + 384 + ], + "score": 0.93, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\psi ( \\bar { x } _ { \\mathrm { U } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 371, + 490, + 385 + ], + "score": 1.0, + "content": ", where the pseudolabel-generating function", + "type": "text" + }, + { + "bbox": [ + 490, + 373, + 498, + 384 + ], + "score": 0.72, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 371, + 505, + 385 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 141, + 394 + ], + "score": 0.91, + "content": "\\mathcal { X } \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "differs between algorithms. For classification problems, we study algorithms that produce", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 395, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 106, + 395, + 505, + 406 + ], + "score": 1.0, + "content": "hard pseudolabels, which are one-hot class predictions, rather than soft pseudolabels, which are", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 405, + 506, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 436, + 417 + ], + "score": 1.0, + "content": "continuous distributions over the classes. Next, algorithms define an unlabeled loss", + "type": "text" + }, + { + "bbox": [ + 436, + 405, + 463, + 417 + ], + "score": 0.94, + "content": "L _ { \\mathrm { U } } ( h )", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 405, + 506, + 417 + ], + "score": 1.0, + "content": "for model", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 416, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 114, + 426 + ], + "score": 0.71, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 416, + 298, + 429 + ], + "score": 1.0, + "content": "by minimizing the loss between pseudolabels", + "type": "text" + }, + { + "bbox": [ + 299, + 417, + 310, + 428 + ], + "score": 0.87, + "content": "\\tilde { y } _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 416, + 506, + 429 + ], + "score": 1.0, + "content": "and the model’s predictions. The algorithms we", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 427, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 205, + 440 + ], + "score": 1.0, + "content": "consider below augment", + "type": "text" + }, + { + "bbox": [ + 205, + 429, + 218, + 438 + ], + "score": 0.84, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 427, + 461, + 440 + ], + "score": 1.0, + "content": "during training; i.e., rather than minimizing the loss between", + "type": "text" + }, + { + "bbox": [ + 461, + 428, + 473, + 438 + ], + "score": 0.85, + "content": "\\tilde { y } _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 427, + 505, + 440 + ], + "score": 1.0, + "content": "and the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 438, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 197, + 450 + ], + "score": 1.0, + "content": "model’s prediction on", + "type": "text" + }, + { + "bbox": [ + 197, + 440, + 209, + 450 + ], + "score": 0.85, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 439, + 446, + 450 + ], + "score": 1.0, + "content": ", the algorithms below minimize the loss of predictions on", + "type": "text" + }, + { + "bbox": [ + 447, + 438, + 474, + 450 + ], + "score": 0.92, + "content": "A ( x _ { \\mathrm { U } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 439, + 505, + 450 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 448, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 107, + 449, + 115, + 459 + ], + "score": 0.77, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 448, + 373, + 462 + ], + "score": 1.0, + "content": "is a stochastic, label-preserving augmentation. Assuming model", + "type": "text" + }, + { + "bbox": [ + 374, + 450, + 380, + 459 + ], + "score": 0.82, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 448, + 505, + 462 + ], + "score": 1.0, + "content": "that outputs real-valued logits,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 461, + 228, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 228, + 472 + ], + "score": 1.0, + "content": "the complete unlabeled loss is", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 22.5, + "bbox_fs": [ + 104, + 360, + 506, + 472 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 230, + 477, + 381, + 509 + ], + "lines": [ + { + "bbox": [ + 230, + 477, + 381, + 509 + ], + "spans": [ + { + "bbox": [ + 230, + 477, + 381, + 509 + ], + "score": 0.94, + "content": "L _ { \\mathrm { U } } ( h ) = \\frac { 1 } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\ell \\big ( h \\circ A ( x _ { \\mathrm { U } } ^ { ( i ) } ) , \\tilde { y } _ { \\mathrm { U } } ^ { ( i ) } \\big )", + "type": "interline_equation", + "image_path": "d1a9730f30eeeb6789c4a37c722914d98b7ab20513e77a495d91e8720b261fc8.jpg" + } + ] + } + ], + "index": 28.5, + "virtual_lines": [ + { + "bbox": [ + 230, + 477, + 381, + 493.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 230, + 493.0, + 381, + 509.0 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 513, + 506, + 536 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 505, + 526 + ], + "score": 1.0, + "content": "This unlabeled loss is jointly optimized with the standard ERM labeled loss. The balance between", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 524, + 479, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 294, + 537 + ], + "score": 1.0, + "content": "the two losses is controlled by hyperparameter", + "type": "text" + }, + { + "bbox": [ + 294, + 525, + 312, + 537 + ], + "score": 0.91, + "content": "\\lambda ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 524, + 470, + 537 + ], + "score": 1.0, + "content": ", which is a function of the current step", + "type": "text" + }, + { + "bbox": [ + 470, + 526, + 474, + 534 + ], + "score": 0.7, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 524, + 479, + 537 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 106, + 513, + 505, + 537 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 247, + 541, + 364, + 555 + ], + "lines": [ + { + "bbox": [ + 247, + 541, + 364, + 555 + ], + "spans": [ + { + "bbox": [ + 247, + 541, + 364, + 555 + ], + "score": 0.91, + "content": "L ( h ) = L _ { \\mathrm { L } } ( h ) + \\lambda ( t ) L _ { \\mathrm { U } } ( h )", + "type": "interline_equation", + "image_path": "c2723df428584cbe5aafb4dba0e1915d2b857ad0ad88b3b1b9fa2bf34da85e7c.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 247, + 541, + 364, + 555 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 565, + 505, + 599 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 505, + 579 + ], + "score": 1.0, + "content": "Pseudo-Label. Algorithm 3 describes Pseudo-Label, proposed by Lee (2013). In this algorithm,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 576, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 505, + 590 + ], + "score": 1.0, + "content": "the model dynamically generates pseudolabels and updates each batch. Formally, the pseudolabel-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 588, + 242, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 186, + 601 + ], + "score": 1.0, + "content": "generating function", + "type": "text" + }, + { + "bbox": [ + 187, + 588, + 195, + 600 + ], + "score": 0.86, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 588, + 242, + 601 + ], + "score": 1.0, + "content": "is given by", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 564, + 505, + 601 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 604, + 392, + 625 + ], + "lines": [ + { + "bbox": [ + 219, + 604, + 392, + 625 + ], + "spans": [ + { + "bbox": [ + 219, + 604, + 392, + 625 + ], + "score": 0.93, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\psi ( x _ { \\mathrm { U } } ) = \\arg \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "interline_equation", + "image_path": "4f96f3d98b0e2a21af0520bac17bcee4733cfb3c8b7c2f6f6cb0bd208b56d043.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 219, + 604, + 392, + 625 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 630, + 503, + 653 + ], + "lines": [ + { + "bbox": [ + 106, + 629, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 133, + 644 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 631, + 160, + 642 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 629, + 505, + 644 + ], + "score": 1.0, + "content": "is the strong augmentation function described in Appendix C. Pseudo-Label then com-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 641, + 447, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 641, + 447, + 653 + ], + "score": 1.0, + "content": "putes the loss between a strongly augmented example and its associated pseudolabel.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 629, + 505, + 653 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 657, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 671 + ], + "score": 1.0, + "content": "In order to more fairly compare Pseudo-Label to FixMatch, we add on confidence thresholding to", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 669, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 505, + 681 + ], + "score": 1.0, + "content": "the Pseudo-Label algorithm, a feature also added in the implementation of Pseudo-Label by Sohn", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 680, + 505, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 692 + ], + "score": 1.0, + "content": "et al. (2020). When confidence thresholding, examples on which the model has low confidence have", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 467, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 306, + 704 + ], + "score": 1.0, + "content": "zero loss, i.e., for some threshold hyperparameter", + "type": "text" + }, + { + "bbox": [ + 306, + 693, + 312, + 701 + ], + "score": 0.75, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 690, + 397, + 704 + ], + "score": 1.0, + "content": ", the loss an example", + "type": "text" + }, + { + "bbox": [ + 397, + 693, + 410, + 702 + ], + "score": 0.84, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 690, + 467, + 704 + ], + "score": 1.0, + "content": "contributes is", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 657, + 506, + 704 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 168, + 707, + 442, + 730 + ], + "lines": [ + { + "bbox": [ + 168, + 707, + 442, + 730 + ], + "spans": [ + { + "bbox": [ + 168, + 707, + 442, + 730 + ], + "score": 0.92, + "content": "\\mathbf { 1 } \\Big \\{ \\mathrm { S o f t m a x } \\Big ( \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ] \\Big ) \\geq \\tau \\Big \\} \\cdot \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) , y _ { \\mathrm { U } } \\big )", + "type": "interline_equation", + "image_path": "a5cb20ddaa31196454d73f3a34f904e8e7c658589557d6e3dda067a73b371c70.jpg" + } + ] + } + ], + "index": 43, + "virtual_lines": [ + { + "bbox": [ + 168, + 707, + 442, + 730 + ], + "spans": [], + "index": 43 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 286, + 95 + ], + "score": 1.0, + "content": "Finally, Pseudo-Label increases the balance", + "type": "text" + }, + { + "bbox": [ + 286, + 82, + 305, + 95 + ], + "score": 0.92, + "content": "\\lambda ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "between labeled and unlabeled losses over time,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 227, + 106 + ], + "score": 1.0, + "content": "initially placing 0 weight on", + "type": "text" + }, + { + "bbox": [ + 227, + 93, + 254, + 106 + ], + "score": 0.92, + "content": "L _ { \\mathrm { U } } ( h )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "and then linearly stepping the unlabeled loss weight until it", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 504, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 274, + 117 + ], + "score": 1.0, + "content": "reaches the full value of hyperparameter", + "type": "text" + }, + { + "bbox": [ + 274, + 105, + 281, + 114 + ], + "score": 0.83, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 105, + 485, + 117 + ], + "score": 1.0, + "content": "at some threshold step. We fix the step at which", + "type": "text" + }, + { + "bbox": [ + 486, + 104, + 504, + 117 + ], + "score": 0.88, + "content": "\\lambda ( t )", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 220, + 128 + ], + "score": 1.0, + "content": "reaches its maximum value", + "type": "text" + }, + { + "bbox": [ + 221, + 116, + 234, + 126 + ], + "score": 0.78, + "content": "( \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 114, + 259, + 128 + ], + "score": 1.0, + "content": "to be", + "type": "text" + }, + { + "bbox": [ + 259, + 115, + 279, + 126 + ], + "score": 0.89, + "content": "4 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "of the total number of training steps, matching the im-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "plementation of Sohn et al. (2020). This scheduling allows the algorithm to initially prioritize the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "labeled loss, as generated pseudolabels are mostly incorrect while the model has low accuracy. For-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 149, + 308, + 161 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 162, + 161 + ], + "score": 1.0, + "content": "mally, at step", + "type": "text" + }, + { + "bbox": [ + 162, + 150, + 167, + 158 + ], + "score": 0.65, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 149, + 295, + 161 + ], + "score": 1.0, + "content": "and given total number of steps", + "type": "text" + }, + { + "bbox": [ + 295, + 149, + 304, + 158 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 149, + 308, + 161 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 164, + 358, + 189 + ], + "lines": [ + { + "bbox": [ + 253, + 164, + 358, + 189 + ], + "spans": [ + { + "bbox": [ + 253, + 164, + 358, + 189 + ], + "score": 0.94, + "content": "\\lambda ( t ) = \\operatorname* { m i n } \\{ \\frac { t } { 0 . 4 T } , 1 \\} \\cdot \\lambda", + "type": "interline_equation", + "image_path": "89ea13c5e265f19ec57baed34440f78a55c502383ed5d080e551352df9d7970f.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 253, + 164, + 358, + 189 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 198, + 505, + 243 + ], + "lines": [ + { + "bbox": [ + 106, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "We add augmentations to Pseudo-Label in order to allow for a fairer comparison to other meth-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 209, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 506, + 222 + ], + "score": 1.0, + "content": "ods that use augmentations. On applicable datasets, we have strongly augmented all labeled and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 220, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 213, + 234 + ], + "score": 1.0, + "content": "unlabeled examples using", + "type": "text" + }, + { + "bbox": [ + 214, + 221, + 240, + 232 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 220, + 506, + 234 + ], + "score": 1.0, + "content": ", whereas Lee (2013) do not use any data augmentations, i.e., all", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 232, + 342, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 156, + 245 + ], + "score": 1.0, + "content": "instances of", + "type": "text" + }, + { + "bbox": [ + 157, + 232, + 183, + 244 + ], + "score": 0.91, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 232, + 342, + 245 + ], + "score": 1.0, + "content": "are replaced with the identity function.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "title", + "bbox": [ + 107, + 259, + 220, + 270 + ], + "lines": [ + { + "bbox": [ + 106, + 258, + 221, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 221, + 272 + ], + "score": 1.0, + "content": "Algorithm 3: Pseudo-Label", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 273, + 505, + 313 + ], + "lines": [ + { + "bbox": [ + 104, + 269, + 411, + 293 + ], + "spans": [ + { + "bbox": [ + 104, + 269, + 196, + 293 + ], + "score": 1.0, + "content": "Input: Labeled batch", + "type": "text" + }, + { + "bbox": [ + 196, + 273, + 339, + 289 + ], + "score": 0.91, + "content": "\\{ ( x _ { \\mathrm { L } } ^ { ( i ) } , y _ { \\mathrm { L } } ^ { ( i ) } , d _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 269, + 411, + 293 + ], + "score": 1.0, + "content": ", unlabeled batch", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 137, + 283, + 509, + 308 + ], + "spans": [ + { + "bbox": [ + 137, + 288, + 261, + 303 + ], + "score": 0.89, + "content": "\\{ ( x _ { \\mathrm { U } } ^ { ( i ) } , d _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( \\bar { 1 } , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 283, + 385, + 308 + ], + "score": 1.0, + "content": ", strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 386, + 291, + 412, + 303 + ], + "score": 0.88, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 283, + 509, + 308 + ], + "score": 1.0, + "content": ", unlabeled loss weight", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 135, + 299, + 368, + 315 + ], + "spans": [ + { + "bbox": [ + 135, + 299, + 200, + 315 + ], + "score": 1.0, + "content": "for current step", + "type": "text" + }, + { + "bbox": [ + 200, + 302, + 238, + 313 + ], + "score": 0.85, + "content": "\\lambda ( t ) \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 299, + 326, + 315 + ], + "score": 1.0, + "content": ", confidence threshold", + "type": "text" + }, + { + "bbox": [ + 327, + 302, + 366, + 313 + ], + "score": 0.92, + "content": "\\tau \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 299, + 368, + 315 + ], + "score": 0.0, + "content": "", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 100, + 313, + 425, + 325 + ], + "lines": [ + { + "bbox": [ + 97, + 311, + 426, + 326 + ], + "spans": [ + { + "bbox": [ + 97, + 311, + 199, + 326 + ], + "score": 1.0, + "content": "1 Generate pseudolabels", + "type": "text" + }, + { + "bbox": [ + 199, + 312, + 334, + 326 + ], + "score": 0.85, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\arg \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 311, + 426, + 326 + ], + "score": 1.0, + "content": "for the unlabeled data", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 99, + 325, + 203, + 336 + ], + "lines": [ + { + "bbox": [ + 97, + 324, + 204, + 338 + ], + "spans": [ + { + "bbox": [ + 97, + 324, + 165, + 338 + ], + "score": 1.0, + "content": "2 Update model", + "type": "text" + }, + { + "bbox": [ + 165, + 325, + 172, + 335 + ], + "score": 0.81, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 324, + 204, + 338 + ], + "score": 1.0, + "content": "on loss", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 342, + 467, + 410 + ], + "lines": [ + { + "bbox": [ + 147, + 342, + 467, + 410 + ], + "spans": [ + { + "bbox": [ + 147, + 342, + 467, + 410 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } ) \\big ) } \\\\ & { \\displaystyle + \\frac { \\lambda ( t ) } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\mathbf { 1 } \\Big \\{ \\operatorname { S o f t m a x } \\Big ( \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ] \\Big ) \\geq \\tau \\Big \\} \\cdot \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) , \\tilde { y } _ { \\mathrm { U } } \\big ) } \\end{array}", + "type": "interline_equation", + "image_path": "9ce3fd9294a4f0f406a129d8baf0ccc8a5ab7c4e8d1ec9b3fdc42b8eb73f3ddc.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 147, + 342, + 467, + 364.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 147, + 364.6666666666667, + 467, + 387.33333333333337 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 147, + 387.33333333333337, + 467, + 410.00000000000006 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 438, + 504, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 438, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 504, + 451 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate Pseudo-Label on all datasets except POVERTYMAP-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "WILDS, as POVERTYMAP-WILDS is a regression dataset, and hard pseudolabeling does not port", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 461, + 257, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 257, + 473 + ], + "score": 1.0, + "content": "straightforwardly to regression tasks.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 484, + 505, + 562 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "FixMatch. Algorithm 4 describes FixMatch, proposed by Sohn et al. (2020). Like Pseudo-Label,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 495, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 504, + 508 + ], + "score": 1.0, + "content": "this algorithm dynamically generates pseudolabels and updates each batch. FixMatch additionally", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "employs consistency regularization on the unlabeled data. While pseudolabels are generated on a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 505, + 530 + ], + "score": 1.0, + "content": "weakly augmented view of the unlabeled examples, the loss is computed with respect to predictions", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 541 + ], + "score": 1.0, + "content": "on a strongly augmented view. This encourages models to make predictions on a strongly augmented", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "example consistent with its prediction on the same example when weakly augmented. For details", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 550, + 386, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 386, + 563 + ], + "score": 1.0, + "content": "about the strong versus weak augmentations we use, see Appendix C.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 567, + 348, + 579 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 348, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 293, + 582 + ], + "score": 1.0, + "content": "Formally, the pseudolabel-generating function", + "type": "text" + }, + { + "bbox": [ + 293, + 568, + 301, + 579 + ], + "score": 0.85, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 564, + 348, + 582 + ], + "score": 1.0, + "content": "is given by", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 583, + 389, + 605 + ], + "lines": [ + { + "bbox": [ + 221, + 583, + 389, + 605 + ], + "spans": [ + { + "bbox": [ + 221, + 583, + 389, + 605 + ], + "score": 0.93, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\psi ( x _ { \\mathrm { U } } ) = \\arg \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { w e a k } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "interline_equation", + "image_path": "c31488c5f6757187bc88df13313b8e5a2516aeea9537d0e509b588bf88580bbe.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 221, + 583, + 389, + 605 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 615, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 615, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 628 + ], + "score": 1.0, + "content": "Like Pseudo-Label, FixMatch uses confidence thresholding, and unlabeled examples on which the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "score": 1.0, + "content": "model has low confidence have zero loss. Following Sohn et al. (2020), we keep the balance between", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 636, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 275, + 652 + ], + "score": 1.0, + "content": "labeled and unlabeled losses constant at", + "type": "text" + }, + { + "bbox": [ + 275, + 638, + 317, + 650 + ], + "score": 0.92, + "content": "\\lambda ( t ) = \\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 636, + 505, + 652 + ], + "score": 1.0, + "content": ". FixMatch’s original authors justify keeping", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 125, + 661 + ], + "score": 0.91, + "content": "\\lambda ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 649, + 344, + 660 + ], + "score": 1.0, + "content": "at a fixed magnitude (as opposed to slowly increasing", + "type": "text" + }, + { + "bbox": [ + 344, + 649, + 362, + 661 + ], + "score": 0.92, + "content": "\\lambda ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "as in Pseudo-Label) by noting that", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "most predictions made by FixMatch are initially low confidence, so for sufficiently high confidence", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 147, + 684 + ], + "score": 1.0, + "content": "threshold", + "type": "text" + }, + { + "bbox": [ + 147, + 673, + 154, + 681 + ], + "score": 0.73, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 671, + 442, + 684 + ], + "score": 1.0, + "content": ", most unlabeled examples have loss zero, keeping the magnitude of", + "type": "text" + }, + { + "bbox": [ + 442, + 671, + 469, + 682 + ], + "score": 0.91, + "content": "L _ { \\mathrm { U } } ( h )", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "initially", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 682, + 470, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 470, + 694 + ], + "score": 1.0, + "content": "small. This magnitude grows over time, providing a natural curriculum (Sohn et al., 2020).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "We endeavored to match our implementation of FixMatch to the formulation of Sohn et al. (2020),", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "except in the use of augmentations for labeled data. Where we have strongly augmented all labeled", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 171, + 733 + ], + "score": 1.0, + "content": "examples using", + "type": "text" + }, + { + "bbox": [ + 172, + 721, + 198, + 733 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 720, + 506, + 733 + ], + "score": 1.0, + "content": "in Algorithm 4, Sohn et al. (2020) explicitly choose to use weak instead of", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + } + ], + "page_idx": 38, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "39", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 286, + 95 + ], + "score": 1.0, + "content": "Finally, Pseudo-Label increases the balance", + "type": "text" + }, + { + "bbox": [ + 286, + 82, + 305, + 95 + ], + "score": 0.92, + "content": "\\lambda ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "between labeled and unlabeled losses over time,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 227, + 106 + ], + "score": 1.0, + "content": "initially placing 0 weight on", + "type": "text" + }, + { + "bbox": [ + 227, + 93, + 254, + 106 + ], + "score": 0.92, + "content": "L _ { \\mathrm { U } } ( h )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "and then linearly stepping the unlabeled loss weight until it", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 504, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 274, + 117 + ], + "score": 1.0, + "content": "reaches the full value of hyperparameter", + "type": "text" + }, + { + "bbox": [ + 274, + 105, + 281, + 114 + ], + "score": 0.83, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 105, + 485, + 117 + ], + "score": 1.0, + "content": "at some threshold step. We fix the step at which", + "type": "text" + }, + { + "bbox": [ + 486, + 104, + 504, + 117 + ], + "score": 0.88, + "content": "\\lambda ( t )", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 220, + 128 + ], + "score": 1.0, + "content": "reaches its maximum value", + "type": "text" + }, + { + "bbox": [ + 221, + 116, + 234, + 126 + ], + "score": 0.78, + "content": "( \\lambda )", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 114, + 259, + 128 + ], + "score": 1.0, + "content": "to be", + "type": "text" + }, + { + "bbox": [ + 259, + 115, + 279, + 126 + ], + "score": 0.89, + "content": "4 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "of the total number of training steps, matching the im-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 138 + ], + "score": 1.0, + "content": "plementation of Sohn et al. (2020). This scheduling allows the algorithm to initially prioritize the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "labeled loss, as generated pseudolabels are mostly incorrect while the model has low accuracy. For-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 149, + 308, + 161 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 162, + 161 + ], + "score": 1.0, + "content": "mally, at step", + "type": "text" + }, + { + "bbox": [ + 162, + 150, + 167, + 158 + ], + "score": 0.65, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 149, + 295, + 161 + ], + "score": 1.0, + "content": "and given total number of steps", + "type": "text" + }, + { + "bbox": [ + 295, + 149, + 304, + 158 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 149, + 308, + 161 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 82, + 506, + 161 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 164, + 358, + 189 + ], + "lines": [ + { + "bbox": [ + 253, + 164, + 358, + 189 + ], + "spans": [ + { + "bbox": [ + 253, + 164, + 358, + 189 + ], + "score": 0.94, + "content": "\\lambda ( t ) = \\operatorname* { m i n } \\{ \\frac { t } { 0 . 4 T } , 1 \\} \\cdot \\lambda", + "type": "interline_equation", + "image_path": "89ea13c5e265f19ec57baed34440f78a55c502383ed5d080e551352df9d7970f.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 253, + 164, + 358, + 189 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 198, + 505, + 243 + ], + "lines": [ + { + "bbox": [ + 106, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "We add augmentations to Pseudo-Label in order to allow for a fairer comparison to other meth-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 209, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 506, + 222 + ], + "score": 1.0, + "content": "ods that use augmentations. On applicable datasets, we have strongly augmented all labeled and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 220, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 213, + 234 + ], + "score": 1.0, + "content": "unlabeled examples using", + "type": "text" + }, + { + "bbox": [ + 214, + 221, + 240, + 232 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 220, + 506, + 234 + ], + "score": 1.0, + "content": ", whereas Lee (2013) do not use any data augmentations, i.e., all", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 232, + 342, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 156, + 245 + ], + "score": 1.0, + "content": "instances of", + "type": "text" + }, + { + "bbox": [ + 157, + 232, + 183, + 244 + ], + "score": 0.91, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 232, + 342, + 245 + ], + "score": 1.0, + "content": "are replaced with the identity function.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 198, + 506, + 245 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 259, + 220, + 270 + ], + "lines": [ + { + "bbox": [ + 106, + 258, + 221, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 221, + 272 + ], + "score": 1.0, + "content": "Algorithm 3: Pseudo-Label", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 273, + 505, + 313 + ], + "lines": [ + { + "bbox": [ + 104, + 269, + 411, + 293 + ], + "spans": [ + { + "bbox": [ + 104, + 269, + 196, + 293 + ], + "score": 1.0, + "content": "Input: Labeled batch", + "type": "text" + }, + { + "bbox": [ + 196, + 273, + 339, + 289 + ], + "score": 0.91, + "content": "\\{ ( x _ { \\mathrm { L } } ^ { ( i ) } , y _ { \\mathrm { L } } ^ { ( i ) } , d _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 269, + 411, + 293 + ], + "score": 1.0, + "content": ", unlabeled batch", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 137, + 283, + 509, + 308 + ], + "spans": [ + { + "bbox": [ + 137, + 288, + 261, + 303 + ], + "score": 0.89, + "content": "\\{ ( x _ { \\mathrm { U } } ^ { ( i ) } , d _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( \\bar { 1 } , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 283, + 385, + 308 + ], + "score": 1.0, + "content": ", strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 386, + 291, + 412, + 303 + ], + "score": 0.88, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 283, + 509, + 308 + ], + "score": 1.0, + "content": ", unlabeled loss weight", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 135, + 299, + 368, + 315 + ], + "spans": [ + { + "bbox": [ + 135, + 299, + 200, + 315 + ], + "score": 1.0, + "content": "for current step", + "type": "text" + }, + { + "bbox": [ + 200, + 302, + 238, + 313 + ], + "score": 0.85, + "content": "\\lambda ( t ) \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 299, + 326, + 315 + ], + "score": 1.0, + "content": ", confidence threshold", + "type": "text" + }, + { + "bbox": [ + 327, + 302, + 366, + 313 + ], + "score": 0.92, + "content": "\\tau \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 299, + 368, + 315 + ], + "score": 0.0, + "content": "", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 269, + 509, + 315 + ] + }, + { + "type": "text", + "bbox": [ + 100, + 313, + 425, + 325 + ], + "lines": [ + { + "bbox": [ + 97, + 311, + 426, + 326 + ], + "spans": [ + { + "bbox": [ + 97, + 311, + 199, + 326 + ], + "score": 1.0, + "content": "1 Generate pseudolabels", + "type": "text" + }, + { + "bbox": [ + 199, + 312, + 334, + 326 + ], + "score": 0.85, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\arg \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 311, + 426, + 326 + ], + "score": 1.0, + "content": "for the unlabeled data", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 97, + 311, + 426, + 326 + ] + }, + { + "type": "text", + "bbox": [ + 99, + 325, + 203, + 336 + ], + "lines": [ + { + "bbox": [ + 97, + 324, + 204, + 338 + ], + "spans": [ + { + "bbox": [ + 97, + 324, + 165, + 338 + ], + "score": 1.0, + "content": "2 Update model", + "type": "text" + }, + { + "bbox": [ + 165, + 325, + 172, + 335 + ], + "score": 0.81, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 324, + 204, + 338 + ], + "score": 1.0, + "content": "on loss", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 97, + 324, + 204, + 338 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 342, + 467, + 410 + ], + "lines": [ + { + "bbox": [ + 147, + 342, + 467, + 410 + ], + "spans": [ + { + "bbox": [ + 147, + 342, + 467, + 410 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } ) \\big ) } \\\\ & { \\displaystyle + \\frac { \\lambda ( t ) } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\mathbf { 1 } \\Big \\{ \\operatorname { S o f t m a x } \\Big ( \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ] \\Big ) \\geq \\tau \\Big \\} \\cdot \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) , \\tilde { y } _ { \\mathrm { U } } \\big ) } \\end{array}", + "type": "interline_equation", + "image_path": "9ce3fd9294a4f0f406a129d8baf0ccc8a5ab7c4e8d1ec9b3fdc42b8eb73f3ddc.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 147, + 342, + 467, + 364.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 147, + 364.6666666666667, + 467, + 387.33333333333337 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 147, + 387.33333333333337, + 467, + 410.00000000000006 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 438, + 504, + 472 + ], + "lines": [ + { + "bbox": [ + 106, + 438, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 504, + 451 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate Pseudo-Label on all datasets except POVERTYMAP-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "WILDS, as POVERTYMAP-WILDS is a regression dataset, and hard pseudolabeling does not port", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 461, + 257, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 257, + 473 + ], + "score": 1.0, + "content": "straightforwardly to regression tasks.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 106, + 438, + 505, + 473 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 484, + 505, + 562 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "FixMatch. Algorithm 4 describes FixMatch, proposed by Sohn et al. (2020). Like Pseudo-Label,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 495, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 504, + 508 + ], + "score": 1.0, + "content": "this algorithm dynamically generates pseudolabels and updates each batch. FixMatch additionally", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "employs consistency regularization on the unlabeled data. While pseudolabels are generated on a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 505, + 530 + ], + "score": 1.0, + "content": "weakly augmented view of the unlabeled examples, the loss is computed with respect to predictions", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 541 + ], + "score": 1.0, + "content": "on a strongly augmented view. This encourages models to make predictions on a strongly augmented", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "example consistent with its prediction on the same example when weakly augmented. For details", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 550, + 386, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 386, + 563 + ], + "score": 1.0, + "content": "about the strong versus weak augmentations we use, see Appendix C.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 484, + 506, + 563 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 567, + 348, + 579 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 348, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 293, + 582 + ], + "score": 1.0, + "content": "Formally, the pseudolabel-generating function", + "type": "text" + }, + { + "bbox": [ + 293, + 568, + 301, + 579 + ], + "score": 0.85, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 564, + 348, + 582 + ], + "score": 1.0, + "content": "is given by", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 564, + 348, + 582 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 583, + 389, + 605 + ], + "lines": [ + { + "bbox": [ + 221, + 583, + 389, + 605 + ], + "spans": [ + { + "bbox": [ + 221, + 583, + 389, + 605 + ], + "score": 0.93, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\psi ( x _ { \\mathrm { U } } ) = \\arg \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { w e a k } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "interline_equation", + "image_path": "c31488c5f6757187bc88df13313b8e5a2516aeea9537d0e509b588bf88580bbe.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 221, + 583, + 389, + 605 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 615, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 615, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 628 + ], + "score": 1.0, + "content": "Like Pseudo-Label, FixMatch uses confidence thresholding, and unlabeled examples on which the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "score": 1.0, + "content": "model has low confidence have zero loss. Following Sohn et al. (2020), we keep the balance between", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 636, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 275, + 652 + ], + "score": 1.0, + "content": "labeled and unlabeled losses constant at", + "type": "text" + }, + { + "bbox": [ + 275, + 638, + 317, + 650 + ], + "score": 0.92, + "content": "\\lambda ( t ) = \\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 636, + 505, + 652 + ], + "score": 1.0, + "content": ". FixMatch’s original authors justify keeping", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 649, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 125, + 661 + ], + "score": 0.91, + "content": "\\lambda ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 649, + 344, + 660 + ], + "score": 1.0, + "content": "at a fixed magnitude (as opposed to slowly increasing", + "type": "text" + }, + { + "bbox": [ + 344, + 649, + 362, + 661 + ], + "score": 0.92, + "content": "\\lambda ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "as in Pseudo-Label) by noting that", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "most predictions made by FixMatch are initially low confidence, so for sufficiently high confidence", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 147, + 684 + ], + "score": 1.0, + "content": "threshold", + "type": "text" + }, + { + "bbox": [ + 147, + 673, + 154, + 681 + ], + "score": 0.73, + "content": "\\tau", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 671, + 442, + 684 + ], + "score": 1.0, + "content": ", most unlabeled examples have loss zero, keeping the magnitude of", + "type": "text" + }, + { + "bbox": [ + 442, + 671, + 469, + 682 + ], + "score": 0.91, + "content": "L _ { \\mathrm { U } } ( h )", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "initially", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 682, + 470, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 470, + 694 + ], + "score": 1.0, + "content": "small. This magnitude grows over time, providing a natural curriculum (Sohn et al., 2020).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 615, + 506, + 694 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "We endeavored to match our implementation of FixMatch to the formulation of Sohn et al. (2020),", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "except in the use of augmentations for labeled data. Where we have strongly augmented all labeled", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 171, + 733 + ], + "score": 1.0, + "content": "examples using", + "type": "text" + }, + { + "bbox": [ + 172, + 721, + 198, + 733 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 720, + 506, + 733 + ], + "score": 1.0, + "content": "in Algorithm 4, Sohn et al. (2020) explicitly choose to use weak instead of", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 82, + 506, + 94 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 94 + ], + "score": 1.0, + "content": "strong augmentations on the labeled examples. However, our results on DomainNet in Appendix E", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "suggest that using strong instead of weak augmentations for the labeled examples improves per-", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "formance, so we use strong augmentations on the labeled examples in order to allow for a fairer", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 116, + 227, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 227, + 127 + ], + "score": 1.0, + "content": "comparison to other methods.", + "type": "text", + "cross_page": true + } + ], + "index": 3 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 698, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 94 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 94 + ], + "score": 1.0, + "content": "strong augmentations on the labeled examples. However, our results on DomainNet in Appendix E", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "suggest that using strong instead of weak augmentations for the labeled examples improves per-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "formance, so we use strong augmentations on the labeled examples in order to allow for a fairer", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 116, + 227, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 227, + 127 + ], + "score": 1.0, + "content": "comparison to other methods.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "title", + "bbox": [ + 107, + 142, + 205, + 154 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 205, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 205, + 156 + ], + "score": 1.0, + "content": "Algorithm 4: FixMatch", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 105, + 157, + 495, + 196 + ], + "lines": [ + { + "bbox": [ + 102, + 152, + 412, + 177 + ], + "spans": [ + { + "bbox": [ + 102, + 152, + 196, + 177 + ], + "score": 1.0, + "content": "Input: Labeled batch", + "type": "text" + }, + { + "bbox": [ + 196, + 157, + 339, + 173 + ], + "score": 0.91, + "content": "\\{ ( x _ { \\mathrm { L } } ^ { ( i ) } , y _ { \\mathrm { L } } ^ { ( i ) } , d _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 152, + 412, + 177 + ], + "score": 1.0, + "content": ", unlabeled batch", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 137, + 165, + 498, + 192 + ], + "spans": [ + { + "bbox": [ + 137, + 171, + 262, + 186 + ], + "score": 0.88, + "content": "\\{ ( x _ { \\mathrm { U } } ^ { ( i ) } , d _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( \\overline { { 1 } } , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 165, + 381, + 192 + ], + "score": 1.0, + "content": ", weak augmentation function", + "type": "text" + }, + { + "bbox": [ + 382, + 174, + 406, + 185 + ], + "score": 0.91, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 165, + 498, + 192 + ], + "score": 1.0, + "content": ", strong augmentation", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 136, + 184, + 446, + 199 + ], + "spans": [ + { + "bbox": [ + 136, + 184, + 173, + 199 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 173, + 186, + 199, + 197 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 184, + 291, + 199 + ], + "score": 1.0, + "content": ", unlabeled loss weight", + "type": "text" + }, + { + "bbox": [ + 318, + 184, + 407, + 199 + ], + "score": 1.0, + "content": ", confidence threshold", + "type": "text" + }, + { + "bbox": [ + 407, + 185, + 446, + 197 + ], + "score": 0.91, + "content": "\\tau \\in [ 0 , 1 ]", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 98, + 196, + 423, + 220 + ], + "lines": [ + { + "bbox": [ + 97, + 195, + 423, + 210 + ], + "spans": [ + { + "bbox": [ + 97, + 195, + 199, + 210 + ], + "score": 1.0, + "content": "1 Generate pseudolabels", + "type": "text" + }, + { + "bbox": [ + 199, + 196, + 332, + 209 + ], + "score": 0.89, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\arg \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { w e a k } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 195, + 423, + 210 + ], + "score": 1.0, + "content": "for the unlabeled data", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 98, + 209, + 204, + 221 + ], + "spans": [ + { + "bbox": [ + 98, + 209, + 165, + 221 + ], + "score": 1.0, + "content": "2 Update model", + "type": "text" + }, + { + "bbox": [ + 165, + 209, + 172, + 218 + ], + "score": 0.81, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 209, + 204, + 221 + ], + "score": 1.0, + "content": "on loss", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 226, + 464, + 294 + ], + "lines": [ + { + "bbox": [ + 150, + 226, + 464, + 294 + ], + "spans": [ + { + "bbox": [ + 150, + 226, + 464, + 294 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } ) \\big ) } \\\\ & { \\displaystyle + \\frac { \\lambda } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\mathbf { 1 } \\Big \\{ \\mathrm { S o f t m a x } \\Big ( \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ] \\Big ) \\geq \\tau \\Big \\} \\cdot \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) , \\tilde { y } _ { \\mathrm { U } } \\big ) } \\end{array}", + "type": "interline_equation", + "image_path": "860f98d8ccd4bb4eb0f5ff311c7fc22052584b2db41a4968951bb9cab6221135.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 150, + 226, + 464, + 248.66666666666666 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 150, + 248.66666666666666, + 464, + 271.3333333333333 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 150, + 271.3333333333333, + 464, + 294.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 324, + 504, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 336 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate FixMatch on image classification datasets", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "IWILDCAM2020-WILDS, CAMELYON17-WILDS, POVERTYMAP-WILDS, and FMOW-WILDS.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 346, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 360 + ], + "score": 1.0, + "content": "We do not evaluate FixMatch on other datasets because FixMatch relies on enforcing consistency", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 358, + 453, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 453, + 370 + ], + "score": 1.0, + "content": "across data augmentations, which we only define for image datasets (see Appendix C).", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 381, + 505, + 470 + ], + "lines": [ + { + "bbox": [ + 105, + 381, + 506, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 506, + 395 + ], + "score": 1.0, + "content": "Noisy Student. Algorithm 5 describes Noisy Student, proposed by Xie et al. (2020). Unlike", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "Pseudo-Label and FixMatch, which update the model and re-generate new pseudolabels each batch,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "Noisy Student generates pseudolabels, fixes them, and then trains the model until convergence be-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 415, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 427 + ], + "score": 1.0, + "content": "fore generating new pseudolabels. First, an initial teacher model is trained on the labeled data; next,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 426, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 505, + 437 + ], + "score": 1.0, + "content": "the teacher model pseudolabels the unlabeled data, and a student model is trained on the labeled", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "score": 1.0, + "content": "and pseudolabeled data; finally, the student model becomes the new teacher, and the cycle repeats", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 448, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 506, + 460 + ], + "score": 1.0, + "content": "(see Algorithm 5). Each (teacher, student) pair is termed an iteration; we study the results of two", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 459, + 149, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 149, + 471 + ], + "score": 1.0, + "content": "iterations.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 108, + 475, + 504, + 498 + ], + "lines": [ + { + "bbox": [ + 105, + 472, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 491 + ], + "score": 1.0, + "content": "We train Noisy Student using hard pseudolabels, which the teacher generates over weakly aug-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 487, + 169, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 169, + 500 + ], + "score": 1.0, + "content": "mented inputs:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 504, + 400, + 525 + ], + "lines": [ + { + "bbox": [ + 210, + 504, + 400, + 525 + ], + "spans": [ + { + "bbox": [ + 210, + 504, + 400, + 525 + ], + "score": 0.92, + "content": "{ \\tilde { y } } _ { \\mathrm { U } } = \\psi ( x _ { \\mathrm { U } } ) = \\mathop { \\mathrm { a r g m a x } } _ { y } f _ { \\mathrm { t e a c h e r } } \\circ A _ { \\mathrm { w e a k } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "interline_equation", + "image_path": "e8f645df1ef47ad75e86c4f5df86f99bc0548e4142d51645a889333ebb2e1fb3.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 210, + 504, + 400, + 525 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 530, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 530, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 505, + 544 + ], + "score": 1.0, + "content": "While the teacher generates pseudolabels on a weakly augmented data, the student must make both", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "score": 1.0, + "content": "labeled and unlabeled predictions on noisy (i.e., strongly augmented) data. Following Xie et al.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 236, + 565 + ], + "score": 1.0, + "content": "(2020), we add a dropout layer", + "type": "text" + }, + { + "bbox": [ + 236, + 553, + 273, + 564 + ], + "score": 0.87, + "content": "( p = 0 . 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 552, + 505, + 565 + ], + "score": 1.0, + "content": ") before the student’s last layer, randomly corrupting final", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 564, + 504, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 504, + 575 + ], + "score": 1.0, + "content": "feature maps. Students thus are trained to be consistent across both data-based and model-based", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 574, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 369, + 586 + ], + "score": 1.0, + "content": "noise. We denote the model with inserted dropout as Dropout", + "type": "text" + }, + { + "bbox": [ + 369, + 575, + 384, + 586 + ], + "score": 0.62, + "content": "\\mathrm { ~ o ~ } f", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 574, + 505, + 586 + ], + "score": 1.0, + "content": ". Xie et al. (2020) add even", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 586, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 598 + ], + "score": 1.0, + "content": "more model-based noise using stochastic depth; for simplicity, we do not use stochastic depth in our", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 597, + 174, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 597, + 174, + 609 + ], + "score": 1.0, + "content": "implementation.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 504, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 501, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 464, + 626 + ], + "score": 1.0, + "content": "We follow the original paper and fix the balance between labeled and unlabeled losses as", + "type": "text" + }, + { + "bbox": [ + 464, + 613, + 501, + 626 + ], + "score": 0.92, + "content": "\\lambda ( t ) = 1", + "type": "inline_equation" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 624, + 318, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 318, + 637 + ], + "score": 1.0, + "content": "Noisy Student does not use confidence thresholding.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 107, + 641, + 505, + 686 + ], + "lines": [ + { + "bbox": [ + 105, + 640, + 505, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 654 + ], + "score": 1.0, + "content": "Note that Xie et al. (2020) use both dropout and strong data augmentations when training the initial", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 651, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 349, + 665 + ], + "score": 1.0, + "content": "teacher on labeled data. We reuse models from our ERM", + "type": "text" + }, + { + "bbox": [ + 349, + 654, + 357, + 662 + ], + "score": 0.66, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 651, + 505, + 665 + ], + "score": 1.0, + "content": "Data Augmentation experiments as", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "initial teacher models; thus we differ from Xie et al. (2020) in that our initial teachers were trained", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 675, + 356, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 356, + 686 + ], + "score": 1.0, + "content": "with strong augmentations, but not dropout (see Algorithm 5).", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate Noisy Student on all datasets except text datasets", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS and AMAZON-WILDS. For GLOBALWHEAT-WILDS and OGB-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 720, + 410, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 410, + 733 + ], + "score": 1.0, + "content": "MOLPCBA, we run Noisy Student without noise from data augmentations.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + } + ], + "page_idx": 39, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "40", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 127 + ], + "lines": [], + "index": 1.5, + "bbox_fs": [ + 105, + 82, + 506, + 127 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 142, + 205, + 154 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 205, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 205, + 156 + ], + "score": 1.0, + "content": "Algorithm 4: FixMatch", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 105, + 157, + 495, + 196 + ], + "lines": [ + { + "bbox": [ + 102, + 152, + 412, + 177 + ], + "spans": [ + { + "bbox": [ + 102, + 152, + 196, + 177 + ], + "score": 1.0, + "content": "Input: Labeled batch", + "type": "text" + }, + { + "bbox": [ + 196, + 157, + 339, + 173 + ], + "score": 0.91, + "content": "\\{ ( x _ { \\mathrm { L } } ^ { ( i ) } , y _ { \\mathrm { L } } ^ { ( i ) } , d _ { \\mathrm { L } } ^ { ( i ) } ) : i \\in ( 1 , \\cdot \\cdot , n _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 152, + 412, + 177 + ], + "score": 1.0, + "content": ", unlabeled batch", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 137, + 165, + 498, + 192 + ], + "spans": [ + { + "bbox": [ + 137, + 171, + 262, + 186 + ], + "score": 0.88, + "content": "\\{ ( x _ { \\mathrm { U } } ^ { ( i ) } , d _ { \\mathrm { U } } ^ { ( i ) } ) : i \\in ( \\overline { { 1 } } , \\cdot \\cdot \\cdot , n _ { \\mathrm { U } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 165, + 381, + 192 + ], + "score": 1.0, + "content": ", weak augmentation function", + "type": "text" + }, + { + "bbox": [ + 382, + 174, + 406, + 185 + ], + "score": 0.91, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 165, + 498, + 192 + ], + "score": 1.0, + "content": ", strong augmentation", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 136, + 184, + 446, + 199 + ], + "spans": [ + { + "bbox": [ + 136, + 184, + 173, + 199 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 173, + 186, + 199, + 197 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 184, + 291, + 199 + ], + "score": 1.0, + "content": ", unlabeled loss weight", + "type": "text" + }, + { + "bbox": [ + 318, + 184, + 407, + 199 + ], + "score": 1.0, + "content": ", confidence threshold", + "type": "text" + }, + { + "bbox": [ + 407, + 185, + 446, + 197 + ], + "score": 0.91, + "content": "\\tau \\in [ 0 , 1 ]", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 102, + 152, + 498, + 199 + ] + }, + { + "type": "index", + "bbox": [ + 98, + 196, + 423, + 220 + ], + "lines": [ + { + "bbox": [ + 97, + 195, + 423, + 210 + ], + "spans": [ + { + "bbox": [ + 97, + 195, + 199, + 210 + ], + "score": 1.0, + "content": "1 Generate pseudolabels", + "type": "text" + }, + { + "bbox": [ + 199, + 196, + 332, + 209 + ], + "score": 0.89, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\arg \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { w e a k } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 195, + 423, + 210 + ], + "score": 1.0, + "content": "for the unlabeled data", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 98, + 209, + 204, + 221 + ], + "spans": [ + { + "bbox": [ + 98, + 209, + 165, + 221 + ], + "score": 1.0, + "content": "2 Update model", + "type": "text" + }, + { + "bbox": [ + 165, + 209, + 172, + 218 + ], + "score": 0.81, + "content": "h", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 209, + 204, + 221 + ], + "score": 1.0, + "content": "on loss", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + } + ], + "index": 8.5, + "bbox_fs": [ + 97, + 195, + 423, + 221 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 226, + 464, + 294 + ], + "lines": [ + { + "bbox": [ + 150, + 226, + 464, + 294 + ], + "spans": [ + { + "bbox": [ + 150, + 226, + 464, + 294 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\displaystyle \\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } ) \\big ) } \\\\ & { \\displaystyle + \\frac { \\lambda } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\mathbf { 1 } \\Big \\{ \\mathrm { S o f t m a x } \\Big ( \\operatorname* { m a x } _ { y } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) [ y ] \\Big ) \\geq \\tau \\Big \\} \\cdot \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ) , \\tilde { y } _ { \\mathrm { U } } \\big ) } \\end{array}", + "type": "interline_equation", + "image_path": "860f98d8ccd4bb4eb0f5ff311c7fc22052584b2db41a4968951bb9cab6221135.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 150, + 226, + 464, + 248.66666666666666 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 150, + 248.66666666666666, + 464, + 271.3333333333333 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 150, + 271.3333333333333, + 464, + 294.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 324, + 504, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 336 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate FixMatch on image classification datasets", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "IWILDCAM2020-WILDS, CAMELYON17-WILDS, POVERTYMAP-WILDS, and FMOW-WILDS.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 346, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 360 + ], + "score": 1.0, + "content": "We do not evaluate FixMatch on other datasets because FixMatch relies on enforcing consistency", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 358, + 453, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 358, + 453, + 370 + ], + "score": 1.0, + "content": "across data augmentations, which we only define for image datasets (see Appendix C).", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 325, + 506, + 370 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 381, + 505, + 470 + ], + "lines": [ + { + "bbox": [ + 105, + 381, + 506, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 506, + 395 + ], + "score": 1.0, + "content": "Noisy Student. Algorithm 5 describes Noisy Student, proposed by Xie et al. (2020). Unlike", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "Pseudo-Label and FixMatch, which update the model and re-generate new pseudolabels each batch,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "score": 1.0, + "content": "Noisy Student generates pseudolabels, fixes them, and then trains the model until convergence be-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 415, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 427 + ], + "score": 1.0, + "content": "fore generating new pseudolabels. First, an initial teacher model is trained on the labeled data; next,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 426, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 505, + 437 + ], + "score": 1.0, + "content": "the teacher model pseudolabels the unlabeled data, and a student model is trained on the labeled", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "score": 1.0, + "content": "and pseudolabeled data; finally, the student model becomes the new teacher, and the cycle repeats", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 448, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 506, + 460 + ], + "score": 1.0, + "content": "(see Algorithm 5). Each (teacher, student) pair is termed an iteration; we study the results of two", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 459, + 149, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 149, + 471 + ], + "score": 1.0, + "content": "iterations.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 381, + 506, + 471 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 475, + 504, + 498 + ], + "lines": [ + { + "bbox": [ + 105, + 472, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 505, + 491 + ], + "score": 1.0, + "content": "We train Noisy Student using hard pseudolabels, which the teacher generates over weakly aug-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 487, + 169, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 169, + 500 + ], + "score": 1.0, + "content": "mented inputs:", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 472, + 505, + 500 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 210, + 504, + 400, + 525 + ], + "lines": [ + { + "bbox": [ + 210, + 504, + 400, + 525 + ], + "spans": [ + { + "bbox": [ + 210, + 504, + 400, + 525 + ], + "score": 0.92, + "content": "{ \\tilde { y } } _ { \\mathrm { U } } = \\psi ( x _ { \\mathrm { U } } ) = \\mathop { \\mathrm { a r g m a x } } _ { y } f _ { \\mathrm { t e a c h e r } } \\circ A _ { \\mathrm { w e a k } } ( x _ { \\mathrm { U } } ) [ y ]", + "type": "interline_equation", + "image_path": "e8f645df1ef47ad75e86c4f5df86f99bc0548e4142d51645a889333ebb2e1fb3.jpg" + } + ] + } + ], + "index": 27, + "virtual_lines": [ + { + "bbox": [ + 210, + 504, + 400, + 525 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 530, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 530, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 505, + 544 + ], + "score": 1.0, + "content": "While the teacher generates pseudolabels on a weakly augmented data, the student must make both", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 505, + 555 + ], + "score": 1.0, + "content": "labeled and unlabeled predictions on noisy (i.e., strongly augmented) data. Following Xie et al.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 236, + 565 + ], + "score": 1.0, + "content": "(2020), we add a dropout layer", + "type": "text" + }, + { + "bbox": [ + 236, + 553, + 273, + 564 + ], + "score": 0.87, + "content": "( p = 0 . 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 552, + 505, + 565 + ], + "score": 1.0, + "content": ") before the student’s last layer, randomly corrupting final", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 564, + 504, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 504, + 575 + ], + "score": 1.0, + "content": "feature maps. Students thus are trained to be consistent across both data-based and model-based", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 574, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 369, + 586 + ], + "score": 1.0, + "content": "noise. We denote the model with inserted dropout as Dropout", + "type": "text" + }, + { + "bbox": [ + 369, + 575, + 384, + 586 + ], + "score": 0.62, + "content": "\\mathrm { ~ o ~ } f", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 574, + 505, + 586 + ], + "score": 1.0, + "content": ". Xie et al. (2020) add even", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 586, + 505, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 598 + ], + "score": 1.0, + "content": "more model-based noise using stochastic depth; for simplicity, we do not use stochastic depth in our", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 597, + 174, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 597, + 174, + 609 + ], + "score": 1.0, + "content": "implementation.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 530, + 505, + 609 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 613, + 504, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 501, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 464, + 626 + ], + "score": 1.0, + "content": "We follow the original paper and fix the balance between labeled and unlabeled losses as", + "type": "text" + }, + { + "bbox": [ + 464, + 613, + 501, + 626 + ], + "score": 0.92, + "content": "\\lambda ( t ) = 1", + "type": "inline_equation" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 624, + 318, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 318, + 637 + ], + "score": 1.0, + "content": "Noisy Student does not use confidence thresholding.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 612, + 501, + 637 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 641, + 505, + 686 + ], + "lines": [ + { + "bbox": [ + 105, + 640, + 505, + 654 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 654 + ], + "score": 1.0, + "content": "Note that Xie et al. (2020) use both dropout and strong data augmentations when training the initial", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 651, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 349, + 665 + ], + "score": 1.0, + "content": "teacher on labeled data. We reuse models from our ERM", + "type": "text" + }, + { + "bbox": [ + 349, + 654, + 357, + 662 + ], + "score": 0.66, + "content": "^ +", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 651, + 505, + 665 + ], + "score": 1.0, + "content": "Data Augmentation experiments as", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "initial teacher models; thus we differ from Xie et al. (2020) in that our initial teachers were trained", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 675, + 356, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 356, + 686 + ], + "score": 1.0, + "content": "with strong augmentations, but not dropout (see Algorithm 5).", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 640, + 505, + 686 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate Noisy Student on all datasets except text datasets", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS and AMAZON-WILDS. For GLOBALWHEAT-WILDS and OGB-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 720, + 410, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 410, + 733 + ], + "score": 1.0, + "content": "MOLPCBA, we run Noisy Student without noise from data augmentations.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 106, + 698, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 85, + 222, + 97 + ], + "lines": [ + { + "bbox": [ + 106, + 84, + 224, + 99 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 224, + 99 + ], + "score": 1.0, + "content": "Algorithm 5: Noisy Student", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 99, + 504, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 97, + 504, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 201, + 114 + ], + "score": 1.0, + "content": "Input: Labeled dataset", + "type": "text" + }, + { + "bbox": [ + 202, + 99, + 259, + 112 + ], + "score": 0.92, + "content": "\\{ ( x _ { \\mathrm { L } } , y _ { \\mathrm { L } } , d _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 97, + 371, + 114 + ], + "score": 1.0, + "content": "divided into batches of size", + "type": "text" + }, + { + "bbox": [ + 371, + 102, + 383, + 111 + ], + "score": 0.84, + "content": "n _ { \\mathrm { L } }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 97, + 459, + 114 + ], + "score": 1.0, + "content": ", unlabeled dataset", + "type": "text" + }, + { + "bbox": [ + 459, + 98, + 504, + 112 + ], + "score": 0.88, + "content": "\\overline { { \\{ ( x _ { \\mathrm { U } } , d _ { \\mathrm { U } } ) \\} } }", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 135, + 110, + 496, + 123 + ], + "spans": [ + { + "bbox": [ + 135, + 110, + 248, + 123 + ], + "score": 1.0, + "content": "divided into batches of size", + "type": "text" + }, + { + "bbox": [ + 248, + 113, + 261, + 122 + ], + "score": 0.83, + "content": "n _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 110, + 368, + 123 + ], + "score": 1.0, + "content": ", total number of iterations", + "type": "text" + }, + { + "bbox": [ + 368, + 111, + 376, + 120 + ], + "score": 0.82, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 110, + 496, + 123 + ], + "score": 1.0, + "content": ", weak augmentation function", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 137, + 118, + 311, + 137 + ], + "spans": [ + { + "bbox": [ + 137, + 122, + 160, + 133 + ], + "score": 0.87, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 118, + 284, + 137 + ], + "score": 1.0, + "content": ", strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 285, + 122, + 311, + 134 + ], + "score": 0.83, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 99, + 134, + 471, + 157 + ], + "lines": [ + { + "bbox": [ + 96, + 131, + 470, + 149 + ], + "spans": [ + { + "bbox": [ + 96, + 131, + 226, + 149 + ], + "score": 1.0, + "content": "1 Train an initial teacher model", + "type": "text" + }, + { + "bbox": [ + 227, + 133, + 243, + 146 + ], + "score": 0.91, + "content": "f ^ { [ 0 ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 131, + 470, + 149 + ], + "score": 1.0, + "content": "to convergence on labeled examples using the following", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 110, + 144, + 196, + 158 + ], + "spans": [ + { + "bbox": [ + 110, + 144, + 196, + 158 + ], + "score": 1.0, + "content": "batch-wise objective", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 162, + 374, + 195 + ], + "lines": [ + { + "bbox": [ + 241, + 162, + 374, + 195 + ], + "spans": [ + { + "bbox": [ + 241, + 162, + 374, + 195 + ], + "score": 0.95, + "content": "\\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } ) \\big )", + "type": "interline_equation", + "image_path": "7523262d742add3b42d685331b1bcce5f32d4850bb3fbe6e5edc1145dcdecd75.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 241, + 162, + 374, + 178.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 241, + 178.5, + 374, + 195.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 111, + 200, + 239, + 212 + ], + "lines": [ + { + "bbox": [ + 111, + 200, + 239, + 213 + ], + "spans": [ + { + "bbox": [ + 111, + 200, + 163, + 213 + ], + "score": 1.0, + "content": "for iteration", + "type": "text" + }, + { + "bbox": [ + 164, + 200, + 224, + 213 + ], + "score": 0.92, + "content": "s \\in ( 1 , \\cdots , S )", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 200, + 239, + 213 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 97, + 212, + 500, + 249 + ], + "lines": [ + { + "bbox": [ + 98, + 210, + 470, + 226 + ], + "spans": [ + { + "bbox": [ + 98, + 216, + 104, + 222 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 120, + 210, + 236, + 225 + ], + "score": 1.0, + "content": "Generate fixed pseudolabels", + "type": "text" + }, + { + "bbox": [ + 237, + 211, + 378, + 226 + ], + "score": 0.92, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\arg \\operatorname* { m a x } _ { y } f ^ { [ s - 1 ] } \\circ A _ { \\mathrm { w e a k } } ( x _ { \\mathrm { U } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 210, + 470, + 225 + ], + "score": 1.0, + "content": "for the unlabeled data", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 98, + 224, + 499, + 239 + ], + "spans": [ + { + "bbox": [ + 98, + 229, + 104, + 236 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 121, + 224, + 238, + 239 + ], + "score": 1.0, + "content": "Train the next student model", + "type": "text" + }, + { + "bbox": [ + 239, + 225, + 254, + 238 + ], + "score": 0.9, + "content": "f ^ { [ s ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 224, + 499, + 239 + ], + "score": 1.0, + "content": "to convergence on unlabeled and labeled examples using the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 237, + 252, + 250 + ], + "spans": [ + { + "bbox": [ + 126, + 237, + 252, + 250 + ], + "score": 1.0, + "content": "following batch-wise objective", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 254, + 495, + 287 + ], + "lines": [ + { + "bbox": [ + 133, + 254, + 495, + 287 + ], + "spans": [ + { + "bbox": [ + 133, + 254, + 495, + 287 + ], + "score": 0.93, + "content": "\\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( \\mathrm { D r o p o u t \\circ } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } ) \\big ) + \\frac { 1 } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\ell \\big ( \\mathrm { D r o p o u t \\circ } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) , \\tilde { y } _ { \\mathrm { U } } ^ { ( i ) } ) \\big )", + "type": "interline_equation", + "image_path": "5a7a77b3f04ea8c7d06a87371a9bc70deb261c0ce8d85a04e7b4c49c7302316a.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 133, + 254, + 495, + 265.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 133, + 265.0, + 495, + 276.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 133, + 276.0, + 495, + 287.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 325, + 260, + 336 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 261, + 337 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 261, + 337 + ], + "score": 1.0, + "content": "B.4 SELF-SUPERVISION METHODS", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 346, + 505, + 435 + ], + "lines": [ + { + "bbox": [ + 106, + 346, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 359 + ], + "score": 1.0, + "content": "Self-supervised methods learn useful representations by training on unlabeled data via auxiliary", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 358, + 504, + 370 + ], + "spans": [ + { + "bbox": [ + 107, + 358, + 504, + 370 + ], + "score": 1.0, + "content": "“proxy” tasks. Common approaches include reconstruction tasks (Vincent et al., 2008; Erhan et al.,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 369, + 504, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 504, + 380 + ], + "score": 1.0, + "content": "2010; Devlin et al., 2019; Gidaris et al., 2018; Lewis et al., 2020), which remove or corrupt a small", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 380, + 504, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 504, + 392 + ], + "score": 1.0, + "content": "part of each training example and use it as a prediction goal, and contrastive learning (He et al., 2020;", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "score": 1.0, + "content": "Chen et al., 2020b; Caron et al., 2020; Radford et al., 2021b), which aims to learn a representation", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 402, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 414 + ], + "score": 1.0, + "content": "space such that similar example pairs stay close to each other while dissimilar ones are far apart.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "score": 1.0, + "content": "The underlying assumption is that feature encoders that solve the proxy tasks will also perform well", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 424, + 394, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 394, + 436 + ], + "score": 1.0, + "content": "on the downstream supervised task (Lee et al., 2020a; Wei et al., 2021).", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 440, + 505, + 518 + ], + "lines": [ + { + "bbox": [ + 106, + 441, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 505, + 453 + ], + "score": 1.0, + "content": "In our work, we consider two self-supervised methods: SwAV Caron et al. (2020) for images and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 452, + 504, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 504, + 463 + ], + "score": 1.0, + "content": "masked language modeling (Devlin et al., 2019) for text. We use these methods to pre-train models", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 463, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 505, + 474 + ], + "score": 1.0, + "content": "on the unlabeled data. In all cases, we start with the same model initialization used for all of the other", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 473, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 505, + 486 + ], + "score": 1.0, + "content": "algorithms on that dataset; do additional pre-training via self-supervision on the unlabeled data; and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "then initialize a new classifier head and finetune the model via ERM with data augmentation. This", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "score": 1.0, + "content": "follows the procedure in Shen et al. (2021). As a concrete example, for FMOW-WILDS, we use the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 506, + 309, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 309, + 519 + ], + "score": 1.0, + "content": "following procedure to run our ERM experiments:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 127, + 528, + 504, + 576 + ], + "lines": [ + { + "bbox": [ + 129, + 528, + 502, + 541 + ], + "spans": [ + { + "bbox": [ + 129, + 528, + 502, + 541 + ], + "score": 1.0, + "content": "1. Initialize a DenseNet-121 model (Huang et al., 2017) using ImageNet-pretrained weights.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 129, + 545, + 384, + 558 + ], + "spans": [ + { + "bbox": [ + 129, + 545, + 384, + 558 + ], + "score": 1.0, + "content": "2. Finetune the model on labeled data from the source domain.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 128, + 563, + 343, + 576 + ], + "spans": [ + { + "bbox": [ + 128, + 563, + 343, + 576 + ], + "score": 1.0, + "content": "3. Evaluate on held-out data from the target domain.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 585, + 439, + 597 + ], + "lines": [ + { + "bbox": [ + 105, + 583, + 440, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 440, + 599 + ], + "score": 1.0, + "content": "For SwAV, we use the exact same procedure but with the addition of a second step:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 126, + 607, + 502, + 672 + ], + "lines": [ + { + "bbox": [ + 129, + 606, + 502, + 621 + ], + "spans": [ + { + "bbox": [ + 129, + 606, + 502, + 621 + ], + "score": 1.0, + "content": "1. Initialize a DenseNet-121 model (Huang et al., 2017) using ImageNet-pretrained weights.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 128, + 623, + 490, + 639 + ], + "spans": [ + { + "bbox": [ + 128, + 623, + 490, + 639 + ], + "score": 1.0, + "content": "2. Continue pre-training the model with SwAV on unlabeled data from the target domain.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 128, + 641, + 384, + 655 + ], + "spans": [ + { + "bbox": [ + 128, + 641, + 384, + 655 + ], + "score": 1.0, + "content": "3. Finetune the model on labeled data from the source domain.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 128, + 659, + 342, + 672 + ], + "spans": [ + { + "bbox": [ + 128, + 659, + 342, + 672 + ], + "score": 1.0, + "content": "4. Evaluate on held-out data from the target domain.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 108, + 681, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 680, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 506, + 695 + ], + "score": 1.0, + "content": "Similarly, for text datasets, we initialized pre-trained BERT models and then continued pre-training", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 693, + 413, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 413, + 705 + ], + "score": 1.0, + "content": "them using masked language modeling on the unlabeled data in WILDS 2.0.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "We tuned hyperparameters for finetuning, following the exact same procedure and hyperparameters", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 718, + 254, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 254, + 734 + ], + "score": 1.0, + "content": "as for ERM, but not for pre-training.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + } + ], + "page_idx": 40, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 13, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 85, + 222, + 97 + ], + "lines": [ + { + "bbox": [ + 106, + 84, + 224, + 99 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 224, + 99 + ], + "score": 1.0, + "content": "Algorithm 5: Noisy Student", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 99, + 504, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 97, + 504, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 201, + 114 + ], + "score": 1.0, + "content": "Input: Labeled dataset", + "type": "text" + }, + { + "bbox": [ + 202, + 99, + 259, + 112 + ], + "score": 0.92, + "content": "\\{ ( x _ { \\mathrm { L } } , y _ { \\mathrm { L } } , d _ { \\mathrm { L } } ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 97, + 371, + 114 + ], + "score": 1.0, + "content": "divided into batches of size", + "type": "text" + }, + { + "bbox": [ + 371, + 102, + 383, + 111 + ], + "score": 0.84, + "content": "n _ { \\mathrm { L } }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 97, + 459, + 114 + ], + "score": 1.0, + "content": ", unlabeled dataset", + "type": "text" + }, + { + "bbox": [ + 459, + 98, + 504, + 112 + ], + "score": 0.88, + "content": "\\overline { { \\{ ( x _ { \\mathrm { U } } , d _ { \\mathrm { U } } ) \\} } }", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 135, + 110, + 496, + 123 + ], + "spans": [ + { + "bbox": [ + 135, + 110, + 248, + 123 + ], + "score": 1.0, + "content": "divided into batches of size", + "type": "text" + }, + { + "bbox": [ + 248, + 113, + 261, + 122 + ], + "score": 0.83, + "content": "n _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 110, + 368, + 123 + ], + "score": 1.0, + "content": ", total number of iterations", + "type": "text" + }, + { + "bbox": [ + 368, + 111, + 376, + 120 + ], + "score": 0.82, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 110, + 496, + 123 + ], + "score": 1.0, + "content": ", weak augmentation function", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 137, + 118, + 311, + 137 + ], + "spans": [ + { + "bbox": [ + 137, + 122, + 160, + 133 + ], + "score": 0.87, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 118, + 284, + 137 + ], + "score": 1.0, + "content": ", strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 285, + 122, + 311, + 134 + ], + "score": 0.83, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 97, + 504, + 137 + ] + }, + { + "type": "text", + "bbox": [ + 99, + 134, + 471, + 157 + ], + "lines": [ + { + "bbox": [ + 96, + 131, + 470, + 149 + ], + "spans": [ + { + "bbox": [ + 96, + 131, + 226, + 149 + ], + "score": 1.0, + "content": "1 Train an initial teacher model", + "type": "text" + }, + { + "bbox": [ + 227, + 133, + 243, + 146 + ], + "score": 0.91, + "content": "f ^ { [ 0 ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 131, + 470, + 149 + ], + "score": 1.0, + "content": "to convergence on labeled examples using the following", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 110, + 144, + 196, + 158 + ], + "spans": [ + { + "bbox": [ + 110, + 144, + 196, + 158 + ], + "score": 1.0, + "content": "batch-wise objective", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 96, + 131, + 470, + 158 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 162, + 374, + 195 + ], + "lines": [ + { + "bbox": [ + 241, + 162, + 374, + 195 + ], + "spans": [ + { + "bbox": [ + 241, + 162, + 374, + 195 + ], + "score": 0.95, + "content": "\\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } ) \\big )", + "type": "interline_equation", + "image_path": "7523262d742add3b42d685331b1bcce5f32d4850bb3fbe6e5edc1145dcdecd75.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 241, + 162, + 374, + 178.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 241, + 178.5, + 374, + 195.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 111, + 200, + 239, + 212 + ], + "lines": [ + { + "bbox": [ + 111, + 200, + 239, + 213 + ], + "spans": [ + { + "bbox": [ + 111, + 200, + 163, + 213 + ], + "score": 1.0, + "content": "for iteration", + "type": "text" + }, + { + "bbox": [ + 164, + 200, + 224, + 213 + ], + "score": 0.92, + "content": "s \\in ( 1 , \\cdots , S )", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 200, + 239, + 213 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 111, + 200, + 239, + 213 + ] + }, + { + "type": "text", + "bbox": [ + 97, + 212, + 500, + 249 + ], + "lines": [ + { + "bbox": [ + 98, + 210, + 470, + 226 + ], + "spans": [ + { + "bbox": [ + 98, + 216, + 104, + 222 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 120, + 210, + 236, + 225 + ], + "score": 1.0, + "content": "Generate fixed pseudolabels", + "type": "text" + }, + { + "bbox": [ + 237, + 211, + 378, + 226 + ], + "score": 0.92, + "content": "\\tilde { y } _ { \\mathrm { U } } = \\arg \\operatorname* { m a x } _ { y } f ^ { [ s - 1 ] } \\circ A _ { \\mathrm { w e a k } } ( x _ { \\mathrm { U } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 210, + 470, + 225 + ], + "score": 1.0, + "content": "for the unlabeled data", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 98, + 224, + 499, + 239 + ], + "spans": [ + { + "bbox": [ + 98, + 229, + 104, + 236 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 121, + 224, + 238, + 239 + ], + "score": 1.0, + "content": "Train the next student model", + "type": "text" + }, + { + "bbox": [ + 239, + 225, + 254, + 238 + ], + "score": 0.9, + "content": "f ^ { [ s ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 224, + 499, + 239 + ], + "score": 1.0, + "content": "to convergence on unlabeled and labeled examples using the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 237, + 252, + 250 + ], + "spans": [ + { + "bbox": [ + 126, + 237, + 252, + 250 + ], + "score": 1.0, + "content": "following batch-wise objective", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 98, + 210, + 499, + 250 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 254, + 495, + 287 + ], + "lines": [ + { + "bbox": [ + 133, + 254, + 495, + 287 + ], + "spans": [ + { + "bbox": [ + 133, + 254, + 495, + 287 + ], + "score": 0.93, + "content": "\\frac { 1 } { n _ { \\mathrm { L } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { L } } } \\ell \\big ( \\mathrm { D r o p o u t \\circ } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { L } } ^ { ( i ) } ) , y _ { \\mathrm { L } } ^ { ( i ) } ) \\big ) + \\frac { 1 } { n _ { \\mathrm { U } } } \\sum _ { i = 1 } ^ { n _ { \\mathrm { U } } } \\ell \\big ( \\mathrm { D r o p o u t \\circ } h \\circ A _ { \\mathrm { s t r o n g } } ( x _ { \\mathrm { U } } ^ { ( i ) } ) , \\tilde { y } _ { \\mathrm { U } } ^ { ( i ) } ) \\big )", + "type": "interline_equation", + "image_path": "5a7a77b3f04ea8c7d06a87371a9bc70deb261c0ce8d85a04e7b4c49c7302316a.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 133, + 254, + 495, + 265.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 133, + 265.0, + 495, + 276.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 133, + 276.0, + 495, + 287.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 325, + 260, + 336 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 261, + 337 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 261, + 337 + ], + "score": 1.0, + "content": "B.4 SELF-SUPERVISION METHODS", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 346, + 505, + 435 + ], + "lines": [ + { + "bbox": [ + 106, + 346, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 359 + ], + "score": 1.0, + "content": "Self-supervised methods learn useful representations by training on unlabeled data via auxiliary", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 358, + 504, + 370 + ], + "spans": [ + { + "bbox": [ + 107, + 358, + 504, + 370 + ], + "score": 1.0, + "content": "“proxy” tasks. Common approaches include reconstruction tasks (Vincent et al., 2008; Erhan et al.,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 369, + 504, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 504, + 380 + ], + "score": 1.0, + "content": "2010; Devlin et al., 2019; Gidaris et al., 2018; Lewis et al., 2020), which remove or corrupt a small", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 380, + 504, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 504, + 392 + ], + "score": 1.0, + "content": "part of each training example and use it as a prediction goal, and contrastive learning (He et al., 2020;", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "score": 1.0, + "content": "Chen et al., 2020b; Caron et al., 2020; Radford et al., 2021b), which aims to learn a representation", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 402, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 505, + 414 + ], + "score": 1.0, + "content": "space such that similar example pairs stay close to each other while dissimilar ones are far apart.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "score": 1.0, + "content": "The underlying assumption is that feature encoders that solve the proxy tasks will also perform well", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 424, + 394, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 394, + 436 + ], + "score": 1.0, + "content": "on the downstream supervised task (Lee et al., 2020a; Wei et al., 2021).", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 346, + 505, + 436 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 440, + 505, + 518 + ], + "lines": [ + { + "bbox": [ + 106, + 441, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 505, + 453 + ], + "score": 1.0, + "content": "In our work, we consider two self-supervised methods: SwAV Caron et al. (2020) for images and", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 452, + 504, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 452, + 504, + 463 + ], + "score": 1.0, + "content": "masked language modeling (Devlin et al., 2019) for text. We use these methods to pre-train models", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 463, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 505, + 474 + ], + "score": 1.0, + "content": "on the unlabeled data. In all cases, we start with the same model initialization used for all of the other", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 473, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 505, + 486 + ], + "score": 1.0, + "content": "algorithms on that dataset; do additional pre-training via self-supervision on the unlabeled data; and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 497 + ], + "score": 1.0, + "content": "then initialize a new classifier head and finetune the model via ERM with data augmentation. This", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "score": 1.0, + "content": "follows the procedure in Shen et al. (2021). As a concrete example, for FMOW-WILDS, we use the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 506, + 309, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 309, + 519 + ], + "score": 1.0, + "content": "following procedure to run our ERM experiments:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 441, + 505, + 519 + ] + }, + { + "type": "index", + "bbox": [ + 127, + 528, + 504, + 576 + ], + "lines": [ + { + "bbox": [ + 129, + 528, + 502, + 541 + ], + "spans": [ + { + "bbox": [ + 129, + 528, + 502, + 541 + ], + "score": 1.0, + "content": "1. Initialize a DenseNet-121 model (Huang et al., 2017) using ImageNet-pretrained weights.", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 545, + 384, + 558 + ], + "spans": [ + { + "bbox": [ + 129, + 545, + 384, + 558 + ], + "score": 1.0, + "content": "2. Finetune the model on labeled data from the source domain.", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 563, + 343, + 576 + ], + "spans": [ + { + "bbox": [ + 128, + 563, + 343, + 576 + ], + "score": 1.0, + "content": "3. Evaluate on held-out data from the target domain.", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + } + ], + "index": 32, + "bbox_fs": [ + 128, + 528, + 502, + 576 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 585, + 439, + 597 + ], + "lines": [ + { + "bbox": [ + 105, + 583, + 440, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 440, + 599 + ], + "score": 1.0, + "content": "For SwAV, we use the exact same procedure but with the addition of a second step:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 583, + 440, + 599 + ] + }, + { + "type": "index", + "bbox": [ + 126, + 607, + 502, + 672 + ], + "lines": [ + { + "bbox": [ + 129, + 606, + 502, + 621 + ], + "spans": [ + { + "bbox": [ + 129, + 606, + 502, + 621 + ], + "score": 1.0, + "content": "1. Initialize a DenseNet-121 model (Huang et al., 2017) using ImageNet-pretrained weights.", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 623, + 490, + 639 + ], + "spans": [ + { + "bbox": [ + 128, + 623, + 490, + 639 + ], + "score": 1.0, + "content": "2. Continue pre-training the model with SwAV on unlabeled data from the target domain.", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 641, + 384, + 655 + ], + "spans": [ + { + "bbox": [ + 128, + 641, + 384, + 655 + ], + "score": 1.0, + "content": "3. Finetune the model on labeled data from the source domain.", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 659, + 342, + 672 + ], + "spans": [ + { + "bbox": [ + 128, + 659, + 342, + 672 + ], + "score": 1.0, + "content": "4. Evaluate on held-out data from the target domain.", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + } + ], + "index": 36.5, + "bbox_fs": [ + 128, + 606, + 502, + 672 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 681, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 680, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 506, + 695 + ], + "score": 1.0, + "content": "Similarly, for text datasets, we initialized pre-trained BERT models and then continued pre-training", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 693, + 413, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 413, + 705 + ], + "score": 1.0, + "content": "them using masked language modeling on the unlabeled data in WILDS 2.0.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 680, + 506, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "We tuned hyperparameters for finetuning, following the exact same procedure and hyperparameters", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 718, + 254, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 718, + 254, + 734 + ], + "score": 1.0, + "content": "as for ERM, but not for pre-training.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 708, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 504, + 95 + ], + "score": 1.0, + "content": "SwAV. We directly use the public SwAV repository available at https://github.com/", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "facebookresearch/swav. We keep almost all of the hyperparameters used by the original", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "paper for 400 epoch training with batch size 256. However, we make the following changes based", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 380, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 380, + 128 + ], + "score": 1.0, + "content": "on issues and tips from the original authors in the SwAV repository:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 125, + 136, + 505, + 280 + ], + "lines": [ + { + "bbox": [ + 130, + 136, + 494, + 150 + ], + "spans": [ + { + "bbox": [ + 130, + 136, + 474, + 150 + ], + "score": 1.0, + "content": "1. To stabilize training, we opt not to use a queue; this follows the suggestion in issue", + "type": "text" + }, + { + "bbox": [ + 474, + 136, + 491, + 147 + ], + "score": 0.59, + "content": "\\# 6 9", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 136, + 494, + 150 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 128, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 128, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "2. For each dataset, we set the number of prototypes to approximately 10x the number of", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 142, + 162, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 142, + 162, + 506, + 174 + ], + "score": 1.0, + "content": "classes; this follows the suggestion in issue #37. For POVERTYMAP-WILDS, which is a", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 173, + 506, + 186 + ], + "spans": [ + { + "bbox": [ + 141, + 173, + 506, + 186 + ], + "score": 1.0, + "content": "regression problem, we use 1000 prototypes, which displayed more stable training than 10", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 184, + 217, + 198 + ], + "spans": [ + { + "bbox": [ + 141, + 184, + 217, + 198 + ], + "score": 1.0, + "content": "or 100 prototypes.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 128, + 199, + 504, + 212 + ], + "spans": [ + { + "bbox": [ + 128, + 199, + 172, + 212 + ], + "score": 1.0, + "content": "3. We set", + "type": "text" + }, + { + "bbox": [ + 172, + 199, + 210, + 210 + ], + "score": 0.88, + "content": "\\epsilon = 0 . 0 3", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 199, + 504, + 212 + ], + "score": 1.0, + "content": "to avoid representation collapse; this follows the suggestion in the Com-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 209, + 329, + 223 + ], + "spans": [ + { + "bbox": [ + 141, + 209, + 329, + 223 + ], + "score": 1.0, + "content": "mon Issues section of the repository’s readme.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 128, + 225, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 128, + 225, + 506, + 238 + ], + "score": 1.0, + "content": "4. We set the base learning rate via the suggested “linear scaling” rule (issue #37). In other", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 235, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 141, + 235, + 307, + 249 + ], + "score": 1.0, + "content": "words, for total batch size (over GPUs)", + "type": "text" + }, + { + "bbox": [ + 308, + 236, + 336, + 247 + ], + "score": 0.76, + "content": "\\geq 5 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 235, + 505, + 249 + ], + "score": 1.0, + "content": ", the learning rate is scaled linearly. For", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 246, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 141, + 246, + 224, + 260 + ], + "score": 1.0, + "content": "smaller batch sizes", + "type": "text" + }, + { + "bbox": [ + 225, + 248, + 255, + 258 + ], + "score": 0.81, + "content": "( < 5 1 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 246, + 506, + 260 + ], + "score": 1.0, + "content": ", we set the base learning rate at 0.6. We multiply the base", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 258, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 141, + 258, + 209, + 271 + ], + "score": 1.0, + "content": "learning rate by", + "type": "text" + }, + { + "bbox": [ + 209, + 258, + 238, + 269 + ], + "score": 0.88, + "content": "1 0 0 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 258, + 506, + 271 + ], + "score": 1.0, + "content": "to obtain the final learning rate, since each of the base/final pairs", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 269, + 311, + 281 + ], + "spans": [ + { + "bbox": [ + 142, + 269, + 311, + 281 + ], + "score": 1.0, + "content": "that the paper reports differ by that factor.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 109, + 290, + 504, + 312 + ], + "lines": [ + { + "bbox": [ + 106, + 289, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 505, + 303 + ], + "score": 1.0, + "content": "We set the maximum number of epochs to 400 but stop pre-training early when the loss does not", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 300, + 325, + 313 + ], + "spans": [ + { + "bbox": [ + 107, + 300, + 198, + 313 + ], + "score": 1.0, + "content": "decrease by more than", + "type": "text" + }, + { + "bbox": [ + 198, + 301, + 221, + 311 + ], + "score": 0.85, + "content": "0 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 300, + 325, + 313 + ], + "score": 1.0, + "content": "for 5 consecutive epochs.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 107, + 324, + 505, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 505, + 335 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate SwAV on IWILDCAM2020-WILDS, CAMELYON17-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 336, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 505, + 346 + ], + "score": 1.0, + "content": "WILDS, POVERTYMAP-WILDS, and FMOW-WILDS. We do not evaluate SwAV on other datasets", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 345, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 505, + 359 + ], + "score": 1.0, + "content": "because SwAV relies training with data augmentations, which we only define for image datasets (see", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 357, + 164, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 164, + 370 + ], + "score": 1.0, + "content": "Appendix C).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 380, + 505, + 480 + ], + "lines": [ + { + "bbox": [ + 106, + 380, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 506, + 393 + ], + "score": 1.0, + "content": "Masked language modeling (MLM). MLM is a popular self-supervised objective for text data", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 392, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 505, + 403 + ], + "score": 1.0, + "content": "and is commonly used to pre-train model representations (Devlin et al., 2019). Given an unlabeled", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 153, + 415 + ], + "score": 1.0, + "content": "text corpus", + "type": "text" + }, + { + "bbox": [ + 154, + 403, + 200, + 415 + ], + "score": 0.92, + "content": "\\mathcal { X } = \\{ X \\} _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "(e.g., a set of comments for CivilComments; a set of reviews for Amazon),", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 412, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 104, + 412, + 184, + 428 + ], + "score": 1.0, + "content": "a training example", + "type": "text" + }, + { + "bbox": [ + 184, + 414, + 208, + 426 + ], + "score": 0.9, + "content": "( x , y )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 412, + 470, + 428 + ], + "score": 1.0, + "content": "can be generated by randomly masking tokens in each text piece", + "type": "text" + }, + { + "bbox": [ + 470, + 414, + 481, + 424 + ], + "score": 0.76, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 412, + 506, + 428 + ], + "score": 1.0, + "content": "(e.g.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 424, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 127, + 435 + ], + "score": 0.83, + "content": "x =", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 424, + 352, + 438 + ], + "score": 1.0, + "content": "“The [MASK] is the currency [MASK] the UK”;", + "type": "text" + }, + { + "bbox": [ + 352, + 425, + 372, + 437 + ], + "score": 0.59, + "content": "y =", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 424, + 506, + 438 + ], + "score": 1.0, + "content": "(“pound”, “of”)). The model is", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 434, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 314, + 449 + ], + "score": 1.0, + "content": "trained to use its representation of the masked input", + "type": "text" + }, + { + "bbox": [ + 314, + 438, + 321, + 446 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 434, + 438, + 449 + ], + "score": 1.0, + "content": "to predict the original tokens", + "type": "text" + }, + { + "bbox": [ + 439, + 438, + 445, + 447 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 434, + 505, + 449 + ], + "score": 1.0, + "content": "that should go", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "score": 1.0, + "content": "in each mask. The MLM objective encourages the model to learn syntactic and semantic knowledge", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "score": 1.0, + "content": "(e.g., to predict “of”) as well as world knowledge (e.g., to predict “pound”) present in the text corpus", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 466, + 182, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 182, + 481 + ], + "score": 1.0, + "content": "(Guu et al., 2020).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 485, + 505, + 541 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 505, + 496 + ], + "score": 1.0, + "content": "For our implementation, we use DistilBERT (Sanh et al., 2019) as our initial model and pre-train it", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 495, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 510 + ], + "score": 1.0, + "content": "with the MLM objective on the unlabeled data of each task (CivilComments, Amazon). Following", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 417, + 519 + ], + "score": 1.0, + "content": "the original BERT implementation (Devlin et al., 2019), we randomly mask", + "type": "text" + }, + { + "bbox": [ + 418, + 507, + 437, + 518 + ], + "score": 0.86, + "content": "15 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "of the tokens in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 236, + 531 + ], + "score": 1.0, + "content": "each input text piece, of which", + "type": "text" + }, + { + "bbox": [ + 236, + 519, + 256, + 529 + ], + "score": 0.86, + "content": "80 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 518, + 369, + 531 + ], + "score": 1.0, + "content": "are replaced with [MASK],", + "type": "text" + }, + { + "bbox": [ + 369, + 519, + 389, + 529 + ], + "score": 0.86, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 518, + 505, + 531 + ], + "score": 1.0, + "content": "are replaced with a random", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 529, + 410, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 308, + 542 + ], + "score": 1.0, + "content": "token (according to the unigram distribution), and", + "type": "text" + }, + { + "bbox": [ + 308, + 529, + 327, + 540 + ], + "score": 0.86, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 529, + 410, + 542 + ], + "score": 1.0, + "content": "are kept unchanged.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 553, + 504, + 575 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 566 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate masked language modeling on the text datasets", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 563, + 308, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 308, + 576 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS and AMAZON-WILDS.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + } + ], + "page_idx": 41, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "42", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 504, + 95 + ], + "score": 1.0, + "content": "SwAV. We directly use the public SwAV repository available at https://github.com/", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "facebookresearch/swav. We keep almost all of the hyperparameters used by the original", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "paper for 400 epoch training with batch size 256. However, we make the following changes based", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 380, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 380, + 128 + ], + "score": 1.0, + "content": "on issues and tips from the original authors in the SwAV repository:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 81, + 505, + 128 + ] + }, + { + "type": "list", + "bbox": [ + 125, + 136, + 505, + 280 + ], + "lines": [ + { + "bbox": [ + 130, + 136, + 494, + 150 + ], + "spans": [ + { + "bbox": [ + 130, + 136, + 474, + 150 + ], + "score": 1.0, + "content": "1. To stabilize training, we opt not to use a queue; this follows the suggestion in issue", + "type": "text" + }, + { + "bbox": [ + 474, + 136, + 491, + 147 + ], + "score": 0.59, + "content": "\\# 6 9", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 136, + 494, + 150 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 128, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "2. For each dataset, we set the number of prototypes to approximately 10x the number of", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 162, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 142, + 162, + 506, + 174 + ], + "score": 1.0, + "content": "classes; this follows the suggestion in issue #37. For POVERTYMAP-WILDS, which is a", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 173, + 506, + 186 + ], + "spans": [ + { + "bbox": [ + 141, + 173, + 506, + 186 + ], + "score": 1.0, + "content": "regression problem, we use 1000 prototypes, which displayed more stable training than 10", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 184, + 217, + 198 + ], + "spans": [ + { + "bbox": [ + 141, + 184, + 217, + 198 + ], + "score": 1.0, + "content": "or 100 prototypes.", + "type": "text" + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 199, + 504, + 212 + ], + "spans": [ + { + "bbox": [ + 128, + 199, + 172, + 212 + ], + "score": 1.0, + "content": "3. We set", + "type": "text" + }, + { + "bbox": [ + 172, + 199, + 210, + 210 + ], + "score": 0.88, + "content": "\\epsilon = 0 . 0 3", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 199, + 504, + 212 + ], + "score": 1.0, + "content": "to avoid representation collapse; this follows the suggestion in the Com-", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 209, + 329, + 223 + ], + "spans": [ + { + "bbox": [ + 141, + 209, + 329, + 223 + ], + "score": 1.0, + "content": "mon Issues section of the repository’s readme.", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 225, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 128, + 225, + 506, + 238 + ], + "score": 1.0, + "content": "4. We set the base learning rate via the suggested “linear scaling” rule (issue #37). In other", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 235, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 141, + 235, + 307, + 249 + ], + "score": 1.0, + "content": "words, for total batch size (over GPUs)", + "type": "text" + }, + { + "bbox": [ + 308, + 236, + 336, + 247 + ], + "score": 0.76, + "content": "\\geq 5 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 235, + 505, + 249 + ], + "score": 1.0, + "content": ", the learning rate is scaled linearly. For", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 246, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 141, + 246, + 224, + 260 + ], + "score": 1.0, + "content": "smaller batch sizes", + "type": "text" + }, + { + "bbox": [ + 225, + 248, + 255, + 258 + ], + "score": 0.81, + "content": "( < 5 1 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 246, + 506, + 260 + ], + "score": 1.0, + "content": ", we set the base learning rate at 0.6. We multiply the base", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 258, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 141, + 258, + 209, + 271 + ], + "score": 1.0, + "content": "learning rate by", + "type": "text" + }, + { + "bbox": [ + 209, + 258, + 238, + 269 + ], + "score": 0.88, + "content": "1 0 0 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 258, + 506, + 271 + ], + "score": 1.0, + "content": "to obtain the final learning rate, since each of the base/final pairs", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 269, + 311, + 281 + ], + "spans": [ + { + "bbox": [ + 142, + 269, + 311, + 281 + ], + "score": 1.0, + "content": "that the paper reports differ by that factor.", + "type": "text" + } + ], + "index": 15, + "is_list_end_line": true + } + ], + "index": 9.5, + "bbox_fs": [ + 128, + 136, + 506, + 281 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 290, + 504, + 312 + ], + "lines": [ + { + "bbox": [ + 106, + 289, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 505, + 303 + ], + "score": 1.0, + "content": "We set the maximum number of epochs to 400 but stop pre-training early when the loss does not", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 300, + 325, + 313 + ], + "spans": [ + { + "bbox": [ + 107, + 300, + 198, + 313 + ], + "score": 1.0, + "content": "decrease by more than", + "type": "text" + }, + { + "bbox": [ + 198, + 301, + 221, + 311 + ], + "score": 0.85, + "content": "0 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 300, + 325, + 313 + ], + "score": 1.0, + "content": "for 5 consecutive epochs.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 106, + 289, + 505, + 313 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 324, + 505, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 325, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 505, + 335 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate SwAV on IWILDCAM2020-WILDS, CAMELYON17-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 336, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 505, + 346 + ], + "score": 1.0, + "content": "WILDS, POVERTYMAP-WILDS, and FMOW-WILDS. We do not evaluate SwAV on other datasets", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 345, + 505, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 505, + 359 + ], + "score": 1.0, + "content": "because SwAV relies training with data augmentations, which we only define for image datasets (see", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 357, + 164, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 164, + 370 + ], + "score": 1.0, + "content": "Appendix C).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 325, + 505, + 370 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 380, + 505, + 480 + ], + "lines": [ + { + "bbox": [ + 106, + 380, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 506, + 393 + ], + "score": 1.0, + "content": "Masked language modeling (MLM). MLM is a popular self-supervised objective for text data", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 392, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 505, + 403 + ], + "score": 1.0, + "content": "and is commonly used to pre-train model representations (Devlin et al., 2019). Given an unlabeled", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 153, + 415 + ], + "score": 1.0, + "content": "text corpus", + "type": "text" + }, + { + "bbox": [ + 154, + 403, + 200, + 415 + ], + "score": 0.92, + "content": "\\mathcal { X } = \\{ X \\} _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "(e.g., a set of comments for CivilComments; a set of reviews for Amazon),", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 412, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 104, + 412, + 184, + 428 + ], + "score": 1.0, + "content": "a training example", + "type": "text" + }, + { + "bbox": [ + 184, + 414, + 208, + 426 + ], + "score": 0.9, + "content": "( x , y )", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 412, + 470, + 428 + ], + "score": 1.0, + "content": "can be generated by randomly masking tokens in each text piece", + "type": "text" + }, + { + "bbox": [ + 470, + 414, + 481, + 424 + ], + "score": 0.76, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 412, + 506, + 428 + ], + "score": 1.0, + "content": "(e.g.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 424, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 127, + 435 + ], + "score": 0.83, + "content": "x =", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 424, + 352, + 438 + ], + "score": 1.0, + "content": "“The [MASK] is the currency [MASK] the UK”;", + "type": "text" + }, + { + "bbox": [ + 352, + 425, + 372, + 437 + ], + "score": 0.59, + "content": "y =", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 424, + 506, + 438 + ], + "score": 1.0, + "content": "(“pound”, “of”)). The model is", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 434, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 314, + 449 + ], + "score": 1.0, + "content": "trained to use its representation of the masked input", + "type": "text" + }, + { + "bbox": [ + 314, + 438, + 321, + 446 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 434, + 438, + 449 + ], + "score": 1.0, + "content": "to predict the original tokens", + "type": "text" + }, + { + "bbox": [ + 439, + 438, + 445, + 447 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 434, + 505, + 449 + ], + "score": 1.0, + "content": "that should go", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "score": 1.0, + "content": "in each mask. The MLM objective encourages the model to learn syntactic and semantic knowledge", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 471 + ], + "score": 1.0, + "content": "(e.g., to predict “of”) as well as world knowledge (e.g., to predict “pound”) present in the text corpus", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 466, + 182, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 182, + 481 + ], + "score": 1.0, + "content": "(Guu et al., 2020).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 26, + "bbox_fs": [ + 104, + 380, + 506, + 481 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 485, + 505, + 541 + ], + "lines": [ + { + "bbox": [ + 106, + 486, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 505, + 496 + ], + "score": 1.0, + "content": "For our implementation, we use DistilBERT (Sanh et al., 2019) as our initial model and pre-train it", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 495, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 510 + ], + "score": 1.0, + "content": "with the MLM objective on the unlabeled data of each task (CivilComments, Amazon). Following", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 417, + 519 + ], + "score": 1.0, + "content": "the original BERT implementation (Devlin et al., 2019), we randomly mask", + "type": "text" + }, + { + "bbox": [ + 418, + 507, + 437, + 518 + ], + "score": 0.86, + "content": "15 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "of the tokens in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 518, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 236, + 531 + ], + "score": 1.0, + "content": "each input text piece, of which", + "type": "text" + }, + { + "bbox": [ + 236, + 519, + 256, + 529 + ], + "score": 0.86, + "content": "80 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 518, + 369, + 531 + ], + "score": 1.0, + "content": "are replaced with [MASK],", + "type": "text" + }, + { + "bbox": [ + 369, + 519, + 389, + 529 + ], + "score": 0.86, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 518, + 505, + 531 + ], + "score": 1.0, + "content": "are replaced with a random", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 529, + 410, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 308, + 542 + ], + "score": 1.0, + "content": "token (according to the unigram distribution), and", + "type": "text" + }, + { + "bbox": [ + 308, + 529, + 327, + 540 + ], + "score": 0.86, + "content": "10 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 529, + 410, + 542 + ], + "score": 1.0, + "content": "are kept unchanged.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 486, + 505, + 542 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 553, + 504, + 575 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 566 + ], + "score": 1.0, + "content": "APPLICABLE DATASETS. We evaluate masked language modeling on the text datasets", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 563, + 308, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 308, + 576 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS and AMAZON-WILDS.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 552, + 505, + 576 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 242, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 243, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 243, + 96 + ], + "score": 1.0, + "content": "C DATA AUGMENTATION", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 106, + 504, + 128 + ], + "lines": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "In this work, several methods we study leverage data augmentations to encourage generalization", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 118, + 472, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 472, + 129 + ], + "score": 1.0, + "content": "across domains. Below, we provide details on our implementations of these augmentations.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 107, + 141, + 505, + 283 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 505, + 153 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 505, + 153 + ], + "score": 1.0, + "content": "Image classification (IWILDCAM2020-WILDS, CAMELYON17-WILDS, and FMOW-WILDS).", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "We use a consistent set of data augmentations across image datasets IWILDCAM2020-WILDS,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 163, + 505, + 175 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 505, + 175 + ], + "score": 1.0, + "content": "CAMELYON17-WILDS, and FMOW-WILDS. For methods other than SwAV, we define two strengths", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 172, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 272, + 188 + ], + "score": 1.0, + "content": "of data augmentations: a weak function", + "type": "text" + }, + { + "bbox": [ + 272, + 174, + 297, + 185 + ], + "score": 0.91, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 172, + 388, + 188 + ], + "score": 1.0, + "content": "and a strong function", + "type": "text" + }, + { + "bbox": [ + 389, + 174, + 415, + 185 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 172, + 506, + 188 + ], + "score": 1.0, + "content": ", and we specify both", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 184, + 506, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 184, + 383, + 199 + ], + "score": 1.0, + "content": "according to Sohn et al. (2020). The weak augmentation function", + "type": "text" + }, + { + "bbox": [ + 383, + 185, + 407, + 196 + ], + "score": 0.89, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 184, + 506, + 199 + ], + "score": 1.0, + "content": "is a random horizontal", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 195, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 195, + 269, + 210 + ], + "score": 1.0, + "content": "flip. The strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 270, + 196, + 296, + 207 + ], + "score": 0.91, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 195, + 505, + 210 + ], + "score": 1.0, + "content": "is a composition of (i) random horizontal flip, (ii)", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 207, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 218 + ], + "score": 1.0, + "content": "RandAugment (Cubuk et al., 2020), and (iii) Cutout (DeVries & Taylor, 2017). For the exact im-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "plementation of RandAugment, we directly use the implementation of Zhang et al. (2021), which is", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 228, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 506, + 241 + ], + "score": 1.0, + "content": "based on the implementation used by Sohn et al. (2020). This implementation specifies a pool of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "score": 1.0, + "content": "operations and sample magnitudes for each operation uniformly across a pre-specified range. The", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 250, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 263 + ], + "score": 1.0, + "content": "pool of operations includes: autocontrast, brightness, color jitter, contrast, equalize, posterize, rota-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 261, + 505, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 274 + ], + "score": 1.0, + "content": "tion, sharpness, horizontal and vertical shearing, solarize, and horizontal and vertical translations.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 271, + 414, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 146, + 285 + ], + "score": 1.0, + "content": "We apply", + "type": "text" + }, + { + "bbox": [ + 147, + 272, + 175, + 282 + ], + "score": 0.89, + "content": "N = 2", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 271, + 414, + 285 + ], + "score": 1.0, + "content": "random operations for all experiments (see Appendix D.4).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 503, + 311 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "The labeled loss for all methods, including finetuning models pre-trained with SwAV, uses this", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 300, + 228, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 228, + 312 + ], + "score": 1.0, + "content": "strong augmentation function.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 505, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "score": 1.0, + "content": "For SwAV pre-training, we use the data augmentation pipeline used in the original paper (Caron", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "et al., 2020), which is almost identical to the strong data augmentation introduced in SimCLR (Chen", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "et al., 2020a) but with different random crop scales to accommodate the several additional lower-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 349, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 363 + ], + "score": 1.0, + "content": "resolution crops. For each image, the pipeline is the following sequence of random transformations:", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 361, + 387, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 387, + 374 + ], + "score": 1.0, + "content": "resized crop, horizontal flip, color jitter, grayscale, and Gaussian blur.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 384, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 106, + 384, + 504, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 504, + 396 + ], + "score": 1.0, + "content": "POVERTYMAP-WILDS. As POVERTYMAP-WILDS is a dataset of multispectral images, we de-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 396, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 407 + ], + "score": 1.0, + "content": "fine a separate set of data augmentations. For methods other than SwAV, we define two strengths of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 406, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 258, + 420 + ], + "score": 1.0, + "content": "data augmentations: a weak function", + "type": "text" + }, + { + "bbox": [ + 259, + 407, + 283, + 417 + ], + "score": 0.9, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 406, + 373, + 420 + ], + "score": 1.0, + "content": "and a strong function", + "type": "text" + }, + { + "bbox": [ + 374, + 407, + 399, + 418 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 406, + 506, + 420 + ], + "score": 1.0, + "content": ". The weak augmentation", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 416, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 143, + 432 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 143, + 418, + 167, + 429 + ], + "score": 0.9, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 416, + 426, + 432 + ], + "score": 1.0, + "content": "is a random horizontal flip. The strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 426, + 418, + 453, + 430 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 416, + 505, + 432 + ], + "score": 1.0, + "content": "is a compo-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "score": 1.0, + "content": "sition of (i) random horizontal flip, (ii) random affine transformation, (iii) color jitter on the RGB", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 438, + 379, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 379, + 452 + ], + "score": 1.0, + "content": "channels, and (iv) Cutout on all channels (DeVries & Taylor, 2017).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 456, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "score": 1.0, + "content": "We use the same augmentations for SwAV pre-training as above for IWILDCAM2020-WILDS,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 466, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 505, + 480 + ], + "score": 1.0, + "content": "CAMELYON17-WILDS, and FMOW-WILDS, but note that the color jitter module is applied only", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 478, + 194, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 194, + 489 + ], + "score": 1.0, + "content": "to the RGB channels.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 502, + 505, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 502, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 505, + 514 + ], + "score": 1.0, + "content": "Other datasets. We do not define data augmentations for other datasets, i.e., GLOBALWHEAT-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 514, + 504, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 504, + 525 + ], + "score": 1.0, + "content": "WILDS, OGB-MOLPCBA, CIVILCOMMENTS-WILDS, and AMAZON-WILDS. Although", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "GLOBALWHEAT-WILDS is an image dataset and can be transformed using augmentations defined", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 533, + 505, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 505, + 548 + ], + "score": 1.0, + "content": "above, we omit data augmentations for simplicity, because such augmentations would generally re-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 168, + 559 + ], + "score": 1.0, + "content": "quire changing", + "type": "text" + }, + { + "bbox": [ + 169, + 548, + 176, + 558 + ], + "score": 0.75, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 546, + 219, + 559 + ], + "score": 1.0, + "content": "as well as", + "type": "text" + }, + { + "bbox": [ + 219, + 548, + 226, + 556 + ], + "score": 0.71, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "(e.g., random translations on the input image also require translating", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "the bounding box labels). For OGB-MOLPCBA, we omit augmentations because data augmenta-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "score": 1.0, + "content": "tions on graphs are not well developed. CIVILCOMMENTS-WILDS and AMAZON-WILDS are text", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 579, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 505, + 591 + ], + "score": 1.0, + "content": "datasets; although data augmentations have been proposed for text datasets, we do not use these", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 589, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 505, + 603 + ], + "score": 1.0, + "content": "augmentations because training with augmentations is not as standard on text datasets as on image", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 601, + 504, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 504, + 612 + ], + "score": 1.0, + "content": "datasets. For these datasets, methods are benchmarked without augmentations, i.e. we substitute all", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 612, + 292, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 167, + 624 + ], + "score": 1.0, + "content": "occurrences of", + "type": "text" + }, + { + "bbox": [ + 168, + 612, + 191, + 623 + ], + "score": 0.46, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 612, + 194, + 624 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 195, + 612, + 221, + 624 + ], + "score": 0.51, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 612, + 292, + 624 + ], + "score": 1.0, + "content": "with the identity.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 37 + } + ], + "page_idx": 42, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "43", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 242, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 243, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 243, + 96 + ], + "score": 1.0, + "content": "C DATA AUGMENTATION", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 106, + 504, + 128 + ], + "lines": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 106, + 505, + 119 + ], + "score": 1.0, + "content": "In this work, several methods we study leverage data augmentations to encourage generalization", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 118, + 472, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 472, + 129 + ], + "score": 1.0, + "content": "across domains. Below, we provide details on our implementations of these augmentations.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 106, + 505, + 129 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 141, + 505, + 283 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 505, + 153 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 505, + 153 + ], + "score": 1.0, + "content": "Image classification (IWILDCAM2020-WILDS, CAMELYON17-WILDS, and FMOW-WILDS).", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 150, + 506, + 164 + ], + "spans": [ + { + "bbox": [ + 106, + 150, + 506, + 164 + ], + "score": 1.0, + "content": "We use a consistent set of data augmentations across image datasets IWILDCAM2020-WILDS,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 163, + 505, + 175 + ], + "spans": [ + { + "bbox": [ + 106, + 163, + 505, + 175 + ], + "score": 1.0, + "content": "CAMELYON17-WILDS, and FMOW-WILDS. For methods other than SwAV, we define two strengths", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 172, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 272, + 188 + ], + "score": 1.0, + "content": "of data augmentations: a weak function", + "type": "text" + }, + { + "bbox": [ + 272, + 174, + 297, + 185 + ], + "score": 0.91, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 172, + 388, + 188 + ], + "score": 1.0, + "content": "and a strong function", + "type": "text" + }, + { + "bbox": [ + 389, + 174, + 415, + 185 + ], + "score": 0.89, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 172, + 506, + 188 + ], + "score": 1.0, + "content": ", and we specify both", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 184, + 506, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 184, + 383, + 199 + ], + "score": 1.0, + "content": "according to Sohn et al. (2020). The weak augmentation function", + "type": "text" + }, + { + "bbox": [ + 383, + 185, + 407, + 196 + ], + "score": 0.89, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 184, + 506, + 199 + ], + "score": 1.0, + "content": "is a random horizontal", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 195, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 195, + 269, + 210 + ], + "score": 1.0, + "content": "flip. The strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 270, + 196, + 296, + 207 + ], + "score": 0.91, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 195, + 505, + 210 + ], + "score": 1.0, + "content": "is a composition of (i) random horizontal flip, (ii)", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 207, + 505, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 505, + 218 + ], + "score": 1.0, + "content": "RandAugment (Cubuk et al., 2020), and (iii) Cutout (DeVries & Taylor, 2017). For the exact im-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "plementation of RandAugment, we directly use the implementation of Zhang et al. (2021), which is", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 228, + 506, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 506, + 241 + ], + "score": 1.0, + "content": "based on the implementation used by Sohn et al. (2020). This implementation specifies a pool of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "score": 1.0, + "content": "operations and sample magnitudes for each operation uniformly across a pre-specified range. The", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 250, + 505, + 263 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 263 + ], + "score": 1.0, + "content": "pool of operations includes: autocontrast, brightness, color jitter, contrast, equalize, posterize, rota-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 261, + 505, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 274 + ], + "score": 1.0, + "content": "tion, sharpness, horizontal and vertical shearing, solarize, and horizontal and vertical translations.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 271, + 414, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 146, + 285 + ], + "score": 1.0, + "content": "We apply", + "type": "text" + }, + { + "bbox": [ + 147, + 272, + 175, + 282 + ], + "score": 0.89, + "content": "N = 2", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 271, + 414, + 285 + ], + "score": 1.0, + "content": "random operations for all experiments (see Appendix D.4).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 141, + 506, + 285 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 503, + 311 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "The labeled loss for all methods, including finetuning models pre-trained with SwAV, uses this", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 300, + 228, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 228, + 312 + ], + "score": 1.0, + "content": "strong augmentation function.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 288, + 505, + 312 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 505, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "score": 1.0, + "content": "For SwAV pre-training, we use the data augmentation pipeline used in the original paper (Caron", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "et al., 2020), which is almost identical to the strong data augmentation introduced in SimCLR (Chen", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "et al., 2020a) but with different random crop scales to accommodate the several additional lower-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 349, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 363 + ], + "score": 1.0, + "content": "resolution crops. For each image, the pipeline is the following sequence of random transformations:", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 361, + 387, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 387, + 374 + ], + "score": 1.0, + "content": "resized crop, horizontal flip, color jitter, grayscale, and Gaussian blur.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 317, + 506, + 374 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 384, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 106, + 384, + 504, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 504, + 396 + ], + "score": 1.0, + "content": "POVERTYMAP-WILDS. As POVERTYMAP-WILDS is a dataset of multispectral images, we de-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 396, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 505, + 407 + ], + "score": 1.0, + "content": "fine a separate set of data augmentations. For methods other than SwAV, we define two strengths of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 406, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 258, + 420 + ], + "score": 1.0, + "content": "data augmentations: a weak function", + "type": "text" + }, + { + "bbox": [ + 259, + 407, + 283, + 417 + ], + "score": 0.9, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 406, + 373, + 420 + ], + "score": 1.0, + "content": "and a strong function", + "type": "text" + }, + { + "bbox": [ + 374, + 407, + 399, + 418 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 406, + 506, + 420 + ], + "score": 1.0, + "content": ". The weak augmentation", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 416, + 505, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 143, + 432 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 143, + 418, + 167, + 429 + ], + "score": 0.9, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 416, + 426, + 432 + ], + "score": 1.0, + "content": "is a random horizontal flip. The strong augmentation function", + "type": "text" + }, + { + "bbox": [ + 426, + 418, + 453, + 430 + ], + "score": 0.9, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 416, + 505, + 432 + ], + "score": 1.0, + "content": "is a compo-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "score": 1.0, + "content": "sition of (i) random horizontal flip, (ii) random affine transformation, (iii) color jitter on the RGB", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 438, + 379, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 379, + 452 + ], + "score": 1.0, + "content": "channels, and (iv) Cutout on all channels (DeVries & Taylor, 2017).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 384, + 506, + 452 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 456, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "score": 1.0, + "content": "We use the same augmentations for SwAV pre-training as above for IWILDCAM2020-WILDS,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 466, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 505, + 480 + ], + "score": 1.0, + "content": "CAMELYON17-WILDS, and FMOW-WILDS, but note that the color jitter module is applied only", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 478, + 194, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 194, + 489 + ], + "score": 1.0, + "content": "to the RGB channels.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 455, + 505, + 489 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 502, + 505, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 502, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 505, + 514 + ], + "score": 1.0, + "content": "Other datasets. We do not define data augmentations for other datasets, i.e., GLOBALWHEAT-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 514, + 504, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 504, + 525 + ], + "score": 1.0, + "content": "WILDS, OGB-MOLPCBA, CIVILCOMMENTS-WILDS, and AMAZON-WILDS. Although", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "GLOBALWHEAT-WILDS is an image dataset and can be transformed using augmentations defined", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 533, + 505, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 505, + 548 + ], + "score": 1.0, + "content": "above, we omit data augmentations for simplicity, because such augmentations would generally re-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 546, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 168, + 559 + ], + "score": 1.0, + "content": "quire changing", + "type": "text" + }, + { + "bbox": [ + 169, + 548, + 176, + 558 + ], + "score": 0.75, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 546, + 219, + 559 + ], + "score": 1.0, + "content": "as well as", + "type": "text" + }, + { + "bbox": [ + 219, + 548, + 226, + 556 + ], + "score": 0.71, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 546, + 505, + 559 + ], + "score": 1.0, + "content": "(e.g., random translations on the input image also require translating", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "the bounding box labels). For OGB-MOLPCBA, we omit augmentations because data augmenta-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "score": 1.0, + "content": "tions on graphs are not well developed. CIVILCOMMENTS-WILDS and AMAZON-WILDS are text", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 579, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 505, + 591 + ], + "score": 1.0, + "content": "datasets; although data augmentations have been proposed for text datasets, we do not use these", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 589, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 505, + 603 + ], + "score": 1.0, + "content": "augmentations because training with augmentations is not as standard on text datasets as on image", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 601, + 504, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 504, + 612 + ], + "score": 1.0, + "content": "datasets. For these datasets, methods are benchmarked without augmentations, i.e. we substitute all", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 612, + 292, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 167, + 624 + ], + "score": 1.0, + "content": "occurrences of", + "type": "text" + }, + { + "bbox": [ + 168, + 612, + 191, + 623 + ], + "score": 0.46, + "content": "A _ { \\mathrm { w e a k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 612, + 194, + 624 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 195, + 612, + 221, + 624 + ], + "score": 0.51, + "content": "A _ { \\mathrm { s t r o n g } }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 612, + 292, + 624 + ], + "score": 1.0, + "content": "with the identity.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 502, + 506, + 624 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 256, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 257, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 257, + 95 + ], + "score": 1.0, + "content": "D EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 107, + 393, + 118 + ], + "lines": [ + { + "bbox": [ + 106, + 107, + 394, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 107, + 394, + 118 + ], + "score": 1.0, + "content": "D.1 IN-DISTRIBUTION VS. OUT-OF-DISTRIBUTION PERFORMANCE", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 127, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 107, + 127, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 107, + 127, + 505, + 139 + ], + "score": 1.0, + "content": "We report both in-distribution and out-of-distribution performance metrics on all datasets, with the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "score": 1.0, + "content": "exception of OGB-MOLPCBA, which does not have a separate in-distribution test set. Using the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 148, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 162 + ], + "score": 1.0, + "content": "terminology in WILDS (Koh et al., 2021), we consider the train-to-train in-distribution comparison", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 159, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 506, + 174 + ], + "score": 1.0, + "content": "on IWILDCAM2020-WILDS, CAMELYON17-WILDS, FMOW-WILDS, and POVERTYMAP-WILDS,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 171, + 435, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 435, + 184 + ], + "score": 1.0, + "content": "and the average comparison on CIVILCOMMENTS-WILDS and AMAZON-WILDS.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4 + }, + { + "type": "title", + "bbox": [ + 107, + 196, + 241, + 207 + ], + "lines": [ + { + "bbox": [ + 106, + 196, + 241, + 209 + ], + "spans": [ + { + "bbox": [ + 106, + 196, + 241, + 209 + ], + "score": 1.0, + "content": "D.2 MODEL ARCHITECTURES", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 216, + 424, + 228 + ], + "lines": [ + { + "bbox": [ + 105, + 216, + 425, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 425, + 229 + ], + "score": 1.0, + "content": "For all experiments, we use the same models for each dataset as in WILDS 1.0:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 132, + 237, + 418, + 357 + ], + "lines": [ + { + "bbox": [ + 132, + 237, + 365, + 250 + ], + "spans": [ + { + "bbox": [ + 132, + 237, + 365, + 250 + ], + "score": 1.0, + "content": "• IWILDCAM2020-WILDS: ResNet-50 (He et al., 2016).", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 132, + 253, + 385, + 266 + ], + "spans": [ + { + "bbox": [ + 132, + 253, + 385, + 266 + ], + "score": 1.0, + "content": "• CAMELYON17-WILDS: DenseNet-121 (Huang et al., 2017).", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 132, + 268, + 356, + 281 + ], + "spans": [ + { + "bbox": [ + 132, + 268, + 356, + 281 + ], + "score": 1.0, + "content": "• FMOW-WILDS: DenseNet-121 (Huang et al., 2017).", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 132, + 283, + 418, + 296 + ], + "spans": [ + { + "bbox": [ + 132, + 283, + 418, + 296 + ], + "score": 1.0, + "content": "• POVERTYMAP-WILDS: Multi-spectral ResNet-18 (Yeh et al., 2020).", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 132, + 299, + 379, + 311 + ], + "spans": [ + { + "bbox": [ + 132, + 299, + 379, + 311 + ], + "score": 1.0, + "content": "• GLOBALWHEAT-WILDS: Faster-RCNN (Ren et al., 2015).", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 131, + 314, + 412, + 326 + ], + "spans": [ + { + "bbox": [ + 131, + 314, + 412, + 326 + ], + "score": 1.0, + "content": "• OGB-MOLPCBA: Graph Isomorphism Network (Xu et al., 2018).", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 133, + 329, + 382, + 341 + ], + "spans": [ + { + "bbox": [ + 133, + 329, + 382, + 341 + ], + "score": 1.0, + "content": "• CIVILCOMMENTS-WILDS: DistilBERT (Sanh et al., 2019).", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 132, + 345, + 346, + 357 + ], + "spans": [ + { + "bbox": [ + 132, + 345, + 346, + 357 + ], + "score": 1.0, + "content": "• AMAZON-WILDS: DistilBERT (Sanh et al., 2019).", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 108, + 366, + 504, + 410 + ], + "lines": [ + { + "bbox": [ + 106, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "The models for IWILDCAM2020-WILDS, FMOW-WILDS, and GLOBALWHEAT-WILDS were ini-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "score": 1.0, + "content": "tialized with weights pre-trained on ImageNet. Note that models for CAMELYON17-WILDS were", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "not initialized with ImageNet weights. The DistilBERT models were also initialized with pre-trained", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 398, + 264, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 264, + 412 + ], + "score": 1.0, + "content": "weights from the Transformers library.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "title", + "bbox": [ + 107, + 424, + 317, + 435 + ], + "lines": [ + { + "bbox": [ + 106, + 424, + 317, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 317, + 436 + ], + "score": 1.0, + "content": "D.3 BATCH SIZES AND BATCH NORMALIZATION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 444, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 444, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 456 + ], + "score": 1.0, + "content": "For each dataset, we set the total batch size (where a batch contains both labeled and unlabeled data)", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 455, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 505, + 468 + ], + "score": 1.0, + "content": "to the maximum that can fit on 12GB of GPU memory (Table 13). For all the methods that leverage", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 467, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 505, + 479 + ], + "score": 1.0, + "content": "unlabeled data, except the pre-training algorithms, we run with 4 steps of gradient accumulation,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 478, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 160, + 490 + ], + "score": 1.0, + "content": "resulting in a", + "type": "text" + }, + { + "bbox": [ + 160, + 478, + 174, + 488 + ], + "score": 0.86, + "content": "4 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 478, + 505, + 490 + ], + "score": 1.0, + "content": "larger effective batch size. For SwAV pre-training, we run with 4 GPUs in parallel,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 489, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 505, + 501 + ], + "score": 1.0, + "content": "which achieves a similar effect. For masked LM pre-training, we run with the default setting of 256", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "steps of gradient accumulation. These larger batch sizes deviate from the defaults used in the WILDS", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 510, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 522 + ], + "score": 1.0, + "content": "paper (Koh et al., 2021). We use these larger batch sizes because methods that leverage unlabeled", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 522, + 464, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 464, + 533 + ], + "score": 1.0, + "content": "data tend to use larger batch sizes (Sohn et al., 2020; Xie et al., 2020; Caron et al., 2020).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 25.5 + }, + { + "type": "table", + "bbox": [ + 144, + 543, + 466, + 655 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 144, + 543, + 466, + 655 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 144, + 543, + 466, + 655 + ], + "spans": [ + { + "bbox": [ + 144, + 543, + 466, + 655 + ], + "score": 0.979, + "html": "
DatasetWILDS 1.0 batch size WILDS 2.0 batch size
CAMELYON17-WILDS32 168
CIVILCOMMENTS-WILDS16 48
FMOW-WILDS32 72
POVERTYMAP-WILDS64 120
AMAZON-WILDS8 24
IWILDCAM2020-WILDS16 24
OGB-MOLPCBA32 4,096
GLOBALWHEAT-WILDS4 8
", + "type": "table", + "image_path": "58c34035a5561070989b99f9ac679ab9f461303e23e8b8e047c8cc86f2ee884f.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 144, + 543, + 466, + 580.3333333333334 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 144, + 580.3333333333334, + 466, + 617.6666666666667 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 144, + 617.6666666666667, + 466, + 655.0000000000001 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 663, + 504, + 686 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "Table 13: The batch sizes of each dataset from the original WILDS 1.0 paper and the batch sizes", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 673, + 489, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 489, + 687 + ], + "score": 1.0, + "content": "used in WILDS 2.0, which correspond to the maximum that can fit into 12GB of GPU memory.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + } + ], + "index": 32.25 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "For models that use batch normalization, the composition of each batch affects the way in which", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "batch normalization is applied. For CORAL, DANN, and Pseudo-Label, we concatenate the labeled", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "and unlabeled data together in each batch, so the labeled and unlabeled data are jointly normalized.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + } + ], + "page_idx": 43, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 256, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 257, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 257, + 95 + ], + "score": 1.0, + "content": "D EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 107, + 393, + 118 + ], + "lines": [ + { + "bbox": [ + 106, + 107, + 394, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 107, + 394, + 118 + ], + "score": 1.0, + "content": "D.1 IN-DISTRIBUTION VS. OUT-OF-DISTRIBUTION PERFORMANCE", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 127, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 107, + 127, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 107, + 127, + 505, + 139 + ], + "score": 1.0, + "content": "We report both in-distribution and out-of-distribution performance metrics on all datasets, with the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "score": 1.0, + "content": "exception of OGB-MOLPCBA, which does not have a separate in-distribution test set. Using the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 148, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 162 + ], + "score": 1.0, + "content": "terminology in WILDS (Koh et al., 2021), we consider the train-to-train in-distribution comparison", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 159, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 506, + 174 + ], + "score": 1.0, + "content": "on IWILDCAM2020-WILDS, CAMELYON17-WILDS, FMOW-WILDS, and POVERTYMAP-WILDS,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 171, + 435, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 435, + 184 + ], + "score": 1.0, + "content": "and the average comparison on CIVILCOMMENTS-WILDS and AMAZON-WILDS.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 127, + 506, + 184 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 196, + 241, + 207 + ], + "lines": [ + { + "bbox": [ + 106, + 196, + 241, + 209 + ], + "spans": [ + { + "bbox": [ + 106, + 196, + 241, + 209 + ], + "score": 1.0, + "content": "D.2 MODEL ARCHITECTURES", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 216, + 424, + 228 + ], + "lines": [ + { + "bbox": [ + 105, + 216, + 425, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 425, + 229 + ], + "score": 1.0, + "content": "For all experiments, we use the same models for each dataset as in WILDS 1.0:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 216, + 425, + 229 + ] + }, + { + "type": "list", + "bbox": [ + 132, + 237, + 418, + 357 + ], + "lines": [ + { + "bbox": [ + 132, + 237, + 365, + 250 + ], + "spans": [ + { + "bbox": [ + 132, + 237, + 365, + 250 + ], + "score": 1.0, + "content": "• IWILDCAM2020-WILDS: ResNet-50 (He et al., 2016).", + "type": "text" + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 253, + 385, + 266 + ], + "spans": [ + { + "bbox": [ + 132, + 253, + 385, + 266 + ], + "score": 1.0, + "content": "• CAMELYON17-WILDS: DenseNet-121 (Huang et al., 2017).", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 268, + 356, + 281 + ], + "spans": [ + { + "bbox": [ + 132, + 268, + 356, + 281 + ], + "score": 1.0, + "content": "• FMOW-WILDS: DenseNet-121 (Huang et al., 2017).", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 283, + 418, + 296 + ], + "spans": [ + { + "bbox": [ + 132, + 283, + 418, + 296 + ], + "score": 1.0, + "content": "• POVERTYMAP-WILDS: Multi-spectral ResNet-18 (Yeh et al., 2020).", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 299, + 379, + 311 + ], + "spans": [ + { + "bbox": [ + 132, + 299, + 379, + 311 + ], + "score": 1.0, + "content": "• GLOBALWHEAT-WILDS: Faster-RCNN (Ren et al., 2015).", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 131, + 314, + 412, + 326 + ], + "spans": [ + { + "bbox": [ + 131, + 314, + 412, + 326 + ], + "score": 1.0, + "content": "• OGB-MOLPCBA: Graph Isomorphism Network (Xu et al., 2018).", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 133, + 329, + 382, + 341 + ], + "spans": [ + { + "bbox": [ + 133, + 329, + 382, + 341 + ], + "score": 1.0, + "content": "• CIVILCOMMENTS-WILDS: DistilBERT (Sanh et al., 2019).", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 132, + 345, + 346, + 357 + ], + "spans": [ + { + "bbox": [ + 132, + 345, + 346, + 357 + ], + "score": 1.0, + "content": "• AMAZON-WILDS: DistilBERT (Sanh et al., 2019).", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 12.5, + "bbox_fs": [ + 131, + 237, + 418, + 357 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 366, + 504, + 410 + ], + "lines": [ + { + "bbox": [ + 106, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "The models for IWILDCAM2020-WILDS, FMOW-WILDS, and GLOBALWHEAT-WILDS were ini-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "score": 1.0, + "content": "tialized with weights pre-trained on ImageNet. Note that models for CAMELYON17-WILDS were", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "not initialized with ImageNet weights. The DistilBERT models were also initialized with pre-trained", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 398, + 264, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 264, + 412 + ], + "score": 1.0, + "content": "weights from the Transformers library.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 366, + 505, + 412 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 424, + 317, + 435 + ], + "lines": [ + { + "bbox": [ + 106, + 424, + 317, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 317, + 436 + ], + "score": 1.0, + "content": "D.3 BATCH SIZES AND BATCH NORMALIZATION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 444, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 444, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 456 + ], + "score": 1.0, + "content": "For each dataset, we set the total batch size (where a batch contains both labeled and unlabeled data)", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 455, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 505, + 468 + ], + "score": 1.0, + "content": "to the maximum that can fit on 12GB of GPU memory (Table 13). For all the methods that leverage", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 467, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 505, + 479 + ], + "score": 1.0, + "content": "unlabeled data, except the pre-training algorithms, we run with 4 steps of gradient accumulation,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 478, + 505, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 160, + 490 + ], + "score": 1.0, + "content": "resulting in a", + "type": "text" + }, + { + "bbox": [ + 160, + 478, + 174, + 488 + ], + "score": 0.86, + "content": "4 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 478, + 505, + 490 + ], + "score": 1.0, + "content": "larger effective batch size. For SwAV pre-training, we run with 4 GPUs in parallel,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 489, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 505, + 501 + ], + "score": 1.0, + "content": "which achieves a similar effect. For masked LM pre-training, we run with the default setting of 256", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 500, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 505, + 511 + ], + "score": 1.0, + "content": "steps of gradient accumulation. These larger batch sizes deviate from the defaults used in the WILDS", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 510, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 522 + ], + "score": 1.0, + "content": "paper (Koh et al., 2021). We use these larger batch sizes because methods that leverage unlabeled", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 522, + 464, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 464, + 533 + ], + "score": 1.0, + "content": "data tend to use larger batch sizes (Sohn et al., 2020; Xie et al., 2020; Caron et al., 2020).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 444, + 506, + 533 + ] + }, + { + "type": "table", + "bbox": [ + 144, + 543, + 466, + 655 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 144, + 543, + 466, + 655 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 144, + 543, + 466, + 655 + ], + "spans": [ + { + "bbox": [ + 144, + 543, + 466, + 655 + ], + "score": 0.979, + "html": "
DatasetWILDS 1.0 batch size WILDS 2.0 batch size
CAMELYON17-WILDS32 168
CIVILCOMMENTS-WILDS16 48
FMOW-WILDS32 72
POVERTYMAP-WILDS64 120
AMAZON-WILDS8 24
IWILDCAM2020-WILDS16 24
OGB-MOLPCBA32 4,096
GLOBALWHEAT-WILDS4 8
", + "type": "table", + "image_path": "58c34035a5561070989b99f9ac679ab9f461303e23e8b8e047c8cc86f2ee884f.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 144, + 543, + 466, + 580.3333333333334 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 144, + 580.3333333333334, + 466, + 617.6666666666667 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 144, + 617.6666666666667, + 466, + 655.0000000000001 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 663, + 504, + 686 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "Table 13: The batch sizes of each dataset from the original WILDS 1.0 paper and the batch sizes", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 673, + 489, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 489, + 687 + ], + "score": 1.0, + "content": "used in WILDS 2.0, which correspond to the maximum that can fit into 12GB of GPU memory.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + } + ], + "index": 32.25 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "For models that use batch normalization, the composition of each batch affects the way in which", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "batch normalization is applied. For CORAL, DANN, and Pseudo-Label, we concatenate the labeled", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 732 + ], + "score": 1.0, + "content": "and unlabeled data together in each batch, so the labeled and unlabeled data are jointly normalized.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 106, + 699, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 156, + 80, + 453, + 192 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 156, + 80, + 453, + 192 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 156, + 80, + 453, + 192 + ], + "spans": [ + { + "bbox": [ + 156, + 80, + 453, + 192 + ], + "score": 0.973, + "html": "
Dataset\\#epochsERM3:1 ratio7:1 ratio15:1 ratio
IWILDCAM2020-WILDS12632
CAMELYON17-WILDS10532
FMOW-WILDS6030158
POVERTYMAP-WILDS150753819
GLOBALWHEAT-WILDS12632
OGB-MOLPCBA2001005025
CIVILCOMMENTS-WILDS5321
AMAZON-WILDS3211
", + "type": "table", + "image_path": "aa6ade0ebd6aa6d2d2302077a5ab094c847cec152f4462ecc93ecfa6ba4300b9.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 156, + 80, + 453, + 117.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 156, + 117.33333333333334, + 453, + 154.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 156, + 154.66666666666669, + 453, + 192.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 505, + 222 + ], + "lines": [ + { + "bbox": [ + 107, + 200, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 107, + 200, + 505, + 212 + ], + "score": 1.0, + "content": "Table 14: The number of epochs (complete passes over the labeled data) used for each dataset,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 210, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 505, + 223 + ], + "score": 1.0, + "content": "specified for the ERM baseline as well as different ratios of unlabeled to labeled data within a batch.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 248, + 505, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 505, + 260 + ], + "score": 1.0, + "content": "For FixMatch, we jointly normalize the labeled data and the strongly augmented unlabeled data,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "score": 1.0, + "content": "but we separately normalize the weakly augmented unlabeled data in a separate forward pass; we", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 269, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 505, + 282 + ], + "score": 1.0, + "content": "did two forward passes to keep the overall batch sizes consistent with the other algorithms, as in", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 281, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 505, + 293 + ], + "score": 1.0, + "content": "Table 13, while still fitting in GPU memory. For Noisy Student, MLM pre-training, and SwAV pre-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "training, the unlabeled data is processed separately from the labeled data, so each batch of labeled", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 303, + 279, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 279, + 315 + ], + "score": 1.0, + "content": "or unlabeled data is separately normalized.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 108, + 332, + 252, + 343 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 253, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 253, + 344 + ], + "score": 1.0, + "content": "D.4 HYPERPARAMETER TUNING", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 354, + 505, + 432 + ], + "lines": [ + { + "bbox": [ + 105, + 353, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 368 + ], + "score": 1.0, + "content": "We tune each algorithm separately for each dataset by randomly sampling 10 different hyperparame-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 367, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 505, + 378 + ], + "score": 1.0, + "content": "ter configurations within the ranges defined below. We early stop and select the best hyperparameters", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "based on the OOD validation performance, which is computed on the labeled Validation (OOD) data", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "score": 1.0, + "content": "for each dataset; we do not use the labeled Validation (ID) data in our experiments. We then run", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 398, + 504, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 504, + 411 + ], + "score": 1.0, + "content": "replicates using the best hyperparameters. For computational reasons, we do not tune hyperparam-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "score": 1.0, + "content": "eters for the pre-training algorithms, though we tune the finetuning of their resulting pre-trained", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 421, + 174, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 174, + 431 + ], + "score": 1.0, + "content": "models as usual.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 448, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 448, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 505, + 461 + ], + "score": 1.0, + "content": "Learning rates. For all the datasets, except for OGB-MOLPCBA, we multiply the learning rates", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 460, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 472 + ], + "score": 1.0, + "content": "used in WILDS by the ratio of the effective batch size to the original batch size used in WILDS 1.0.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 470, + 506, + 485 + ], + "spans": [ + { + "bbox": [ + 104, + 470, + 373, + 485 + ], + "score": 1.0, + "content": "We center the learning rate grid around this modified learning rate", + "type": "text" + }, + { + "bbox": [ + 373, + 474, + 379, + 482 + ], + "score": 0.67, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 470, + 448, + 485 + ], + "score": 1.0, + "content": ", and search over", + "type": "text" + }, + { + "bbox": [ + 448, + 470, + 501, + 482 + ], + "score": 0.9, + "content": "r \\cdot 1 0 ^ { U ( - 1 , 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 470, + 506, + 485 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 482, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 133, + 496 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 483, + 143, + 492 + ], + "score": 0.79, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 482, + 386, + 496 + ], + "score": 1.0, + "content": "is the uniform distribution. For OGB-MOLPCBA, we pick", + "type": "text" + }, + { + "bbox": [ + 387, + 484, + 393, + 493 + ], + "score": 0.71, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 482, + 506, + 496 + ], + "score": 1.0, + "content": "by multiplying the original", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 493, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 270, + 506 + ], + "score": 1.0, + "content": "learning rate by a factor of 10 instead of", + "type": "text" + }, + { + "bbox": [ + 270, + 493, + 335, + 505 + ], + "score": 0.87, + "content": "4 0 9 6 / 3 2 = 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 493, + 505, + 506 + ], + "score": 1.0, + "content": "(for ERM, which does not have grandient", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 504, + 412, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 412, + 517 + ], + "score": 1.0, + "content": "accumulation), because we found that the latter led to unstable optimization.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 504, + 554 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 119, + 544 + ], + "score": 0.86, + "content": "L _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 532, + 394, + 546 + ], + "score": 1.0, + "content": "-regularization. Across all datasets and methods, we used the same", + "type": "text" + }, + { + "bbox": [ + 395, + 533, + 407, + 544 + ], + "score": 0.88, + "content": "L _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "-regularization strengths", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 543, + 189, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 189, + 555 + ], + "score": 1.0, + "content": "used in WILDS 1.0.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 585 + ], + "score": 1.0, + "content": "Ratio of unlabeled to labeled data in a batch. For all the domain-invariant and self-training", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 582, + 503, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 464, + 594 + ], + "score": 1.0, + "content": "methods, we search over the ratio of unlabeled to labeled data in a batch, using the values", + "type": "text" + }, + { + "bbox": [ + 465, + 583, + 503, + 595 + ], + "score": 0.84, + "content": "\\{ 3 : 1 , 7 :", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 592, + 150, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 145, + 606 + ], + "score": 0.87, + "content": "1 , 1 5 : 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 592, + 150, + 607 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 622, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 506, + 634 + ], + "score": 1.0, + "content": "Number of epochs. We defined an epoch as a complete pass over the labeled data. This means", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 632, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 646 + ], + "score": 1.0, + "content": "that the number of batches / gradient steps taken per epoch varies with the ratio of unlabeled to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "score": 1.0, + "content": "labeled data in a batch, as a higher ratio means that each batch contains fewer labeled examples.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "score": 1.0, + "content": "We adjusted the number of epochs accordingly so that the total amount of compute was similar", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "regardless of the ratio of unlabeled to labeled data in a batch. We allocated roughly twice as much", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 677, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 689 + ], + "score": 1.0, + "content": "compute (i.e., processing twice as many batches) to methods that used unlabeled data, compared to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "the purely-supervised ERM baseline. Overall, we set the number of epochs based on the WILDS 1.0", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "defaults, with some upwards adjustments (due to the different batch sizes and the use of unlabeled", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "data) if we found that the best hyperparameter configuration had not converged on the validation set.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 720, + 348, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 348, + 733 + ], + "score": 1.0, + "content": "Table 14 shows the total number of epochs used per dataset.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 34.5 + } + ], + "page_idx": 44, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 13, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 156, + 80, + 453, + 192 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 156, + 80, + 453, + 192 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 156, + 80, + 453, + 192 + ], + "spans": [ + { + "bbox": [ + 156, + 80, + 453, + 192 + ], + "score": 0.973, + "html": "
Dataset\\#epochsERM3:1 ratio7:1 ratio15:1 ratio
IWILDCAM2020-WILDS12632
CAMELYON17-WILDS10532
FMOW-WILDS6030158
POVERTYMAP-WILDS150753819
GLOBALWHEAT-WILDS12632
OGB-MOLPCBA2001005025
CIVILCOMMENTS-WILDS5321
AMAZON-WILDS3211
", + "type": "table", + "image_path": "aa6ade0ebd6aa6d2d2302077a5ab094c847cec152f4462ecc93ecfa6ba4300b9.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 156, + 80, + 453, + 117.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 156, + 117.33333333333334, + 453, + 154.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 156, + 154.66666666666669, + 453, + 192.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 505, + 222 + ], + "lines": [ + { + "bbox": [ + 107, + 200, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 107, + 200, + 505, + 212 + ], + "score": 1.0, + "content": "Table 14: The number of epochs (complete passes over the labeled data) used for each dataset,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 210, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 505, + 223 + ], + "score": 1.0, + "content": "specified for the ERM baseline as well as different ratios of unlabeled to labeled data within a batch.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 106, + 200, + 505, + 223 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 248, + 505, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 505, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 505, + 260 + ], + "score": 1.0, + "content": "For FixMatch, we jointly normalize the labeled data and the strongly augmented unlabeled data,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 505, + 272 + ], + "score": 1.0, + "content": "but we separately normalize the weakly augmented unlabeled data in a separate forward pass; we", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 269, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 269, + 505, + 282 + ], + "score": 1.0, + "content": "did two forward passes to keep the overall batch sizes consistent with the other algorithms, as in", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 281, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 505, + 293 + ], + "score": 1.0, + "content": "Table 13, while still fitting in GPU memory. For Noisy Student, MLM pre-training, and SwAV pre-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 505, + 304 + ], + "score": 1.0, + "content": "training, the unlabeled data is processed separately from the labeled data, so each batch of labeled", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 303, + 279, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 279, + 315 + ], + "score": 1.0, + "content": "or unlabeled data is separately normalized.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 248, + 505, + 315 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 332, + 252, + 343 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 253, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 253, + 344 + ], + "score": 1.0, + "content": "D.4 HYPERPARAMETER TUNING", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 354, + 505, + 432 + ], + "lines": [ + { + "bbox": [ + 105, + 353, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 368 + ], + "score": 1.0, + "content": "We tune each algorithm separately for each dataset by randomly sampling 10 different hyperparame-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 367, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 505, + 378 + ], + "score": 1.0, + "content": "ter configurations within the ranges defined below. We early stop and select the best hyperparameters", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "based on the OOD validation performance, which is computed on the labeled Validation (OOD) data", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "score": 1.0, + "content": "for each dataset; we do not use the labeled Validation (ID) data in our experiments. We then run", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 398, + 504, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 504, + 411 + ], + "score": 1.0, + "content": "replicates using the best hyperparameters. For computational reasons, we do not tune hyperparam-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "score": 1.0, + "content": "eters for the pre-training algorithms, though we tune the finetuning of their resulting pre-trained", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 421, + 174, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 174, + 431 + ], + "score": 1.0, + "content": "models as usual.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 353, + 505, + 431 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 448, + 505, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 448, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 505, + 461 + ], + "score": 1.0, + "content": "Learning rates. For all the datasets, except for OGB-MOLPCBA, we multiply the learning rates", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 460, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 505, + 472 + ], + "score": 1.0, + "content": "used in WILDS by the ratio of the effective batch size to the original batch size used in WILDS 1.0.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 470, + 506, + 485 + ], + "spans": [ + { + "bbox": [ + 104, + 470, + 373, + 485 + ], + "score": 1.0, + "content": "We center the learning rate grid around this modified learning rate", + "type": "text" + }, + { + "bbox": [ + 373, + 474, + 379, + 482 + ], + "score": 0.67, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 470, + 448, + 485 + ], + "score": 1.0, + "content": ", and search over", + "type": "text" + }, + { + "bbox": [ + 448, + 470, + 501, + 482 + ], + "score": 0.9, + "content": "r \\cdot 1 0 ^ { U ( - 1 , 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 470, + 506, + 485 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 482, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 133, + 496 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 483, + 143, + 492 + ], + "score": 0.79, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 482, + 386, + 496 + ], + "score": 1.0, + "content": "is the uniform distribution. For OGB-MOLPCBA, we pick", + "type": "text" + }, + { + "bbox": [ + 387, + 484, + 393, + 493 + ], + "score": 0.71, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 482, + 506, + 496 + ], + "score": 1.0, + "content": "by multiplying the original", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 493, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 270, + 506 + ], + "score": 1.0, + "content": "learning rate by a factor of 10 instead of", + "type": "text" + }, + { + "bbox": [ + 270, + 493, + 335, + 505 + ], + "score": 0.87, + "content": "4 0 9 6 / 3 2 = 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 493, + 505, + 506 + ], + "score": 1.0, + "content": "(for ERM, which does not have grandient", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 504, + 412, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 412, + 517 + ], + "score": 1.0, + "content": "accumulation), because we found that the latter led to unstable optimization.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 104, + 448, + 506, + 517 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 504, + 554 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 119, + 544 + ], + "score": 0.86, + "content": "L _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 532, + 394, + 546 + ], + "score": 1.0, + "content": "-regularization. Across all datasets and methods, we used the same", + "type": "text" + }, + { + "bbox": [ + 395, + 533, + 407, + 544 + ], + "score": 0.88, + "content": "L _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "-regularization strengths", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 543, + 189, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 189, + 555 + ], + "score": 1.0, + "content": "used in WILDS 1.0.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 532, + 505, + 555 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 585 + ], + "score": 1.0, + "content": "Ratio of unlabeled to labeled data in a batch. For all the domain-invariant and self-training", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 582, + 503, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 464, + 594 + ], + "score": 1.0, + "content": "methods, we search over the ratio of unlabeled to labeled data in a batch, using the values", + "type": "text" + }, + { + "bbox": [ + 465, + 583, + 503, + 595 + ], + "score": 0.84, + "content": "\\{ 3 : 1 , 7 :", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 592, + 150, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 145, + 606 + ], + "score": 0.87, + "content": "1 , 1 5 : 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 592, + 150, + 607 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 570, + 505, + 607 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 622, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 506, + 634 + ], + "score": 1.0, + "content": "Number of epochs. We defined an epoch as a complete pass over the labeled data. This means", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 632, + 506, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 646 + ], + "score": 1.0, + "content": "that the number of batches / gradient steps taken per epoch varies with the ratio of unlabeled to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 656 + ], + "score": 1.0, + "content": "labeled data in a batch, as a higher ratio means that each batch contains fewer labeled examples.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "score": 1.0, + "content": "We adjusted the number of epochs accordingly so that the total amount of compute was similar", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "regardless of the ratio of unlabeled to labeled data in a batch. We allocated roughly twice as much", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 677, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 689 + ], + "score": 1.0, + "content": "compute (i.e., processing twice as many batches) to methods that used unlabeled data, compared to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "the purely-supervised ERM baseline. Overall, we set the number of epochs based on the WILDS 1.0", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "defaults, with some upwards adjustments (due to the different batch sizes and the use of unlabeled", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "data) if we found that the best hyperparameter configuration had not converged on the validation set.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 720, + 348, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 348, + 733 + ], + "score": 1.0, + "content": "Table 14 shows the total number of epochs used per dataset.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 621, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 109, + 82, + 316, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 318, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 318, + 95 + ], + "score": 1.0, + "content": "D.5 ALGORITHM-SPECIFIC HYPERPARAMETERS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 102, + 348, + 115 + ], + "lines": [ + { + "bbox": [ + 106, + 100, + 349, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 349, + 118 + ], + "score": 1.0, + "content": "We tuned the following algorithm-specific hyperparameters:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 126, + 336, + 139 + ], + "lines": [ + { + "bbox": [ + 105, + 126, + 333, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 293, + 141 + ], + "score": 1.0, + "content": "CORAL. We searched over penalty weights", + "type": "text" + }, + { + "bbox": [ + 293, + 126, + 333, + 138 + ], + "score": 0.9, + "content": "1 0 ^ { U ( - 1 , 1 ) }", + "type": "inline_equation" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 151, + 505, + 196 + ], + "lines": [ + { + "bbox": [ + 105, + 150, + 505, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 286, + 165 + ], + "score": 1.0, + "content": "DANN. We searched over penalty weights", + "type": "text" + }, + { + "bbox": [ + 286, + 151, + 326, + 163 + ], + "score": 0.91, + "content": "1 0 ^ { U ( - 1 , 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 150, + 505, + 165 + ], + "score": 1.0, + "content": "and have separate learning rates for the fea-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 163, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 505, + 176 + ], + "score": 1.0, + "content": "turizer, classifier and domain discriminator. We tuned the learning rate for the classifier and domain", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 174, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 505, + 186 + ], + "score": 1.0, + "content": "discriminator, then fixed the learning rate of the featurizer to be a tenth of the learning rate of the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 185, + 147, + 197 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 147, + 197 + ], + "score": 1.0, + "content": "classifier.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 208, + 505, + 242 + ], + "lines": [ + { + "bbox": [ + 105, + 208, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 505, + 222 + ], + "score": 1.0, + "content": "Pseudo-Label, FixMatch, and Noisy Student. We fixed the penalty weight to be 1. For FixMatch", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 340, + 233 + ], + "score": 1.0, + "content": "and Pseudo-Label, we searched over confidence thresholds", + "type": "text" + }, + { + "bbox": [ + 340, + 220, + 391, + 232 + ], + "score": 0.89, + "content": "U ( 0 . 7 , 0 . 9 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 219, + 506, + 233 + ], + "score": 1.0, + "content": ". Noisy Student does not use", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 231, + 201, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 201, + 241 + ], + "score": 1.0, + "content": "a confidence threshold.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 108, + 254, + 504, + 276 + ], + "lines": [ + { + "bbox": [ + 105, + 253, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 505, + 267 + ], + "score": 1.0, + "content": "SwAV. We did not tune SwAV hyperparameters. See Appendix B.4 for a description of the default", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 266, + 199, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 199, + 277 + ], + "score": 1.0, + "content": "hyperparameters used.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 505, + 344 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "Masked language modeling. We did not tune masked LM hyperparameters, opting instead to use", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "score": 1.0, + "content": "default hyperparameters. For both CIVILCOMMENTS-WILDS and AMAZON-WILDS, we pre-trained", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 308, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 311, + 326 + ], + "score": 1.0, + "content": "DistilBERT for 1,000 steps with a learning rate of", + "type": "text" + }, + { + "bbox": [ + 311, + 311, + 333, + 321 + ], + "score": 0.9, + "content": "1 0 ^ { - 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 308, + 505, + 326 + ], + "score": 1.0, + "content": "and a batch size of 8,192 sequences using", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "score": 1.0, + "content": "gradient accumulation. Following WILDS defaults, for CivilComments, we set the max sequence", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 331, + 453, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 453, + 348 + ], + "score": 1.0, + "content": "length to be 300 and for Amazon, 512. We used FP16 training to speed up pretraining.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 358, + 256, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 358, + 257, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 257, + 370 + ], + "score": 1.0, + "content": "D.6 COMPUTE INFRASTRUCTURE", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 378, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 106, + 377, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 487, + 390 + ], + "score": 1.0, + "content": "We ran experiments on a mix of NVIDIA GPUs: V100, K80, GeForce RTX, Titan RTX, Titan", + "type": "text" + }, + { + "bbox": [ + 487, + 379, + 501, + 390 + ], + "score": 0.29, + "content": "\\mathrm { X p }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 377, + 505, + 390 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 389, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 343, + 403 + ], + "score": 1.0, + "content": "and Titan V. SwAV pre-training took approximately 3 days", + "type": "text" + }, + { + "bbox": [ + 344, + 390, + 361, + 400 + ], + "score": 0.49, + "content": "\\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 389, + 506, + 403 + ], + "score": 1.0, + "content": "V100 GPUs for each dataset, while", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 400, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 506, + 414 + ], + "score": 1.0, + "content": "masked LM pre-training took approximately 3 days for a single GPU for each dataset. The other", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 412, + 504, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 504, + 424 + ], + "score": 1.0, + "content": "algorithms took less than a day on a V100 to run. The runtime estimates in Section 6 are estimated", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 421, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 435 + ], + "score": 1.0, + "content": "for V100 GPUs. We used the Weights and Biases platform (Biewald, 2020) to monitor experiments.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20 + } + ], + "page_idx": 45, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 13, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 109, + 82, + 316, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 318, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 318, + 95 + ], + "score": 1.0, + "content": "D.5 ALGORITHM-SPECIFIC HYPERPARAMETERS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 108, + 102, + 348, + 115 + ], + "lines": [ + { + "bbox": [ + 106, + 100, + 349, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 349, + 118 + ], + "score": 1.0, + "content": "We tuned the following algorithm-specific hyperparameters:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 100, + 349, + 118 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 126, + 336, + 139 + ], + "lines": [ + { + "bbox": [ + 105, + 126, + 333, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 293, + 141 + ], + "score": 1.0, + "content": "CORAL. We searched over penalty weights", + "type": "text" + }, + { + "bbox": [ + 293, + 126, + 333, + 138 + ], + "score": 0.9, + "content": "1 0 ^ { U ( - 1 , 1 ) }", + "type": "inline_equation" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 126, + 333, + 141 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 151, + 505, + 196 + ], + "lines": [ + { + "bbox": [ + 105, + 150, + 505, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 286, + 165 + ], + "score": 1.0, + "content": "DANN. We searched over penalty weights", + "type": "text" + }, + { + "bbox": [ + 286, + 151, + 326, + 163 + ], + "score": 0.91, + "content": "1 0 ^ { U ( - 1 , 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 150, + 505, + 165 + ], + "score": 1.0, + "content": "and have separate learning rates for the fea-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 163, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 505, + 176 + ], + "score": 1.0, + "content": "turizer, classifier and domain discriminator. We tuned the learning rate for the classifier and domain", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 174, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 505, + 186 + ], + "score": 1.0, + "content": "discriminator, then fixed the learning rate of the featurizer to be a tenth of the learning rate of the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 185, + 147, + 197 + ], + "spans": [ + { + "bbox": [ + 105, + 185, + 147, + 197 + ], + "score": 1.0, + "content": "classifier.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 150, + 505, + 197 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 208, + 505, + 242 + ], + "lines": [ + { + "bbox": [ + 105, + 208, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 505, + 222 + ], + "score": 1.0, + "content": "Pseudo-Label, FixMatch, and Noisy Student. We fixed the penalty weight to be 1. For FixMatch", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 340, + 233 + ], + "score": 1.0, + "content": "and Pseudo-Label, we searched over confidence thresholds", + "type": "text" + }, + { + "bbox": [ + 340, + 220, + 391, + 232 + ], + "score": 0.89, + "content": "U ( 0 . 7 , 0 . 9 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 219, + 506, + 233 + ], + "score": 1.0, + "content": ". Noisy Student does not use", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 231, + 201, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 201, + 241 + ], + "score": 1.0, + "content": "a confidence threshold.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 208, + 506, + 241 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 254, + 504, + 276 + ], + "lines": [ + { + "bbox": [ + 105, + 253, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 505, + 267 + ], + "score": 1.0, + "content": "SwAV. We did not tune SwAV hyperparameters. See Appendix B.4 for a description of the default", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 266, + 199, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 199, + 277 + ], + "score": 1.0, + "content": "hyperparameters used.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 253, + 505, + 277 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 505, + 344 + ], + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 302 + ], + "score": 1.0, + "content": "Masked language modeling. We did not tune masked LM hyperparameters, opting instead to use", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "score": 1.0, + "content": "default hyperparameters. For both CIVILCOMMENTS-WILDS and AMAZON-WILDS, we pre-trained", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 308, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 311, + 326 + ], + "score": 1.0, + "content": "DistilBERT for 1,000 steps with a learning rate of", + "type": "text" + }, + { + "bbox": [ + 311, + 311, + 333, + 321 + ], + "score": 0.9, + "content": "1 0 ^ { - 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 308, + 505, + 326 + ], + "score": 1.0, + "content": "and a batch size of 8,192 sequences using", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "score": 1.0, + "content": "gradient accumulation. Following WILDS defaults, for CivilComments, we set the max sequence", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 331, + 453, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 453, + 348 + ], + "score": 1.0, + "content": "length to be 300 and for Amazon, 512. We used FP16 training to speed up pretraining.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 288, + 506, + 348 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 358, + 256, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 358, + 257, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 257, + 370 + ], + "score": 1.0, + "content": "D.6 COMPUTE INFRASTRUCTURE", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 378, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 106, + 377, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 487, + 390 + ], + "score": 1.0, + "content": "We ran experiments on a mix of NVIDIA GPUs: V100, K80, GeForce RTX, Titan RTX, Titan", + "type": "text" + }, + { + "bbox": [ + 487, + 379, + 501, + 390 + ], + "score": 0.29, + "content": "\\mathrm { X p }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 377, + 505, + 390 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 389, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 343, + 403 + ], + "score": 1.0, + "content": "and Titan V. SwAV pre-training took approximately 3 days", + "type": "text" + }, + { + "bbox": [ + 344, + 390, + 361, + 400 + ], + "score": 0.49, + "content": "\\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 389, + 506, + 403 + ], + "score": 1.0, + "content": "V100 GPUs for each dataset, while", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 400, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 506, + 414 + ], + "score": 1.0, + "content": "masked LM pre-training took approximately 3 days for a single GPU for each dataset. The other", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 412, + 504, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 504, + 424 + ], + "score": 1.0, + "content": "algorithms took less than a day on a V100 to run. The runtime estimates in Section 6 are estimated", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 421, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 505, + 435 + ], + "score": 1.0, + "content": "for V100 GPUs. We used the Weights and Biases platform (Biewald, 2020) to monitor experiments.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 377, + 506, + 435 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 288, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 291, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 291, + 96 + ], + "score": 1.0, + "content": "E EXPERIMENTS ON DOMAINNET", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 195 + ], + "lines": [ + { + "bbox": [ + 105, + 107, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 107, + 505, + 119 + ], + "score": 1.0, + "content": "Prior work has shown that domain-invariant, self-training, and self-supervised methods can perform", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 118, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 505, + 130 + ], + "score": 1.0, + "content": "well on standard benchmarks for unsupervised domain adaptation. In this section, we describe", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 129, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 129, + 505, + 141 + ], + "score": 1.0, + "content": "our experiments on DomainNet (Peng et al., 2019), a standard unsupervised domain adaptation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 140, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 505, + 152 + ], + "score": 1.0, + "content": "benchmark for object recognition. Our goal was to verify that our training/tuning protocol and our", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 151, + 504, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 151, + 504, + 162 + ], + "score": 1.0, + "content": "implementations of the methods we benchmark in Section 6—which differ slightly from prior work", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 162, + 504, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 504, + 173 + ], + "score": 1.0, + "content": "in the ways described in Section B—still result in models that can perform well on DomainNet.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "score": 1.0, + "content": "Consistent with prior work, the methods we benchmark in Section 6, with the exception of CORAL,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 183, + 396, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 396, + 196 + ], + "score": 1.0, + "content": "all improve over standard ERM training in our DomainNet experiments.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4.5 + }, + { + "type": "title", + "bbox": [ + 107, + 209, + 161, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 208, + 162, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 162, + 222 + ], + "score": 1.0, + "content": "E.1 SETUP", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 230, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "DomainNet is an object recognition dataset with approximately 600,000 images across six different", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 240, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 505, + 254 + ], + "score": 1.0, + "content": "domains: sketch, real, quickdraw, painting, infograph and clipart (Peng et al., 2019). Typically, one", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "score": 1.0, + "content": "of these domains is selected as the source, and another domain as the target for evaluation. In our", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 263, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 222, + 275 + ], + "score": 1.0, + "content": "experiments, we use the real", + "type": "text" + }, + { + "bbox": [ + 222, + 264, + 235, + 273 + ], + "score": 0.75, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 263, + 506, + 275 + ], + "score": 1.0, + "content": "sketch setting for two reasons: it is a common choice in prior work", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 275, + 504, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 504, + 287 + ], + "score": 1.0, + "content": "on DomainNet, and as our models are pre-trained on ImageNet (following (Zhang et al., 2021)),", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 286, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 505, + 298 + ], + "score": 1.0, + "content": "we wanted to use the real domain as the source to be consistent with the realistic photographs used", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "score": 1.0, + "content": "for ImageNet pretraining. While it is common to evaluate methods on multiple pairs of source and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 306, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 506, + 321 + ], + "score": 1.0, + "content": "target domains in DomainNet, in our experiments we only chose one pair, as our goal was only to", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 318, + 253, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 253, + 330 + ], + "score": 1.0, + "content": "verify consistency with prior results.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 342, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 343, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 479, + 354 + ], + "score": 1.0, + "content": "Data. The DomainNet dataset includes train and test splits for each of the domains, with", + "type": "text" + }, + { + "bbox": [ + 479, + 343, + 505, + 353 + ], + "score": 0.81, + "content": "70 \\text{‰}", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 353, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 126, + 364 + ], + "score": 0.84, + "content": "30 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 353, + 505, + 366 + ], + "score": 1.0, + "content": "split between train and test examples. The real domain has 172,947 images total: 120,906", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 364, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 506, + 377 + ], + "score": 1.0, + "content": "images in the train split and 52,041 images in the test split. The target domain, sketch, has a total of", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 375, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 389 + ], + "score": 1.0, + "content": "69,128 images: 48,212 in the train split and 20,916 images are in the test split. We used this data in", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 385, + 184, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 184, + 401 + ], + "score": 1.0, + "content": "the following way:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 130, + 407, + 505, + 506 + ], + "lines": [ + { + "bbox": [ + 129, + 407, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 129, + 407, + 505, + 421 + ], + "score": 1.0, + "content": "1. For training, we used the source training examples (with labels) and the target training", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 418, + 250, + 432 + ], + "spans": [ + { + "bbox": [ + 141, + 418, + 250, + 432 + ], + "score": 1.0, + "content": "examples (without labels).", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 128, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 128, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "2. For validation, we used the same set of target training examples, but with labels; this", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 446, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 141, + 446, + 505, + 458 + ], + "score": 1.0, + "content": "overestimates performance in a true domain adaptation setting (where one would not have", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 142, + 457, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 142, + 457, + 505, + 469 + ], + "score": 1.0, + "content": "labeled target data), but it is a common practice in the literature, and we followed it for", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 141, + 468, + 383, + 480 + ], + "spans": [ + { + "bbox": [ + 141, + 468, + 383, + 480 + ], + "score": 1.0, + "content": "consistency with Jiang et al. (2020) and Zhang et al. (2021).", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 129, + 484, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 129, + 484, + 505, + 496 + ], + "score": 1.0, + "content": "3. For evaluation, we used the source test examples as the in-distribution test set, and the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 495, + 358, + 506 + ], + "spans": [ + { + "bbox": [ + 141, + 495, + 358, + 506 + ], + "score": 1.0, + "content": "target test examples as the out-of-distribution test set.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 519, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 532 + ], + "score": 1.0, + "content": "Hyperparameters and other details. Other experimental details followed our main experiments", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 529, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 542 + ], + "score": 1.0, + "content": "in Section 6. We used the strong and weak data augmentation described for image classification", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 541, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 553 + ], + "score": 1.0, + "content": "in Appendix C. We set the total batch size to 96, which is the maximum that can fit on 12GB of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 551, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 565 + ], + "score": 1.0, + "content": "GPU memory. We tuned hyperparameters with the protocol described in Section D.4. Specifically,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 561, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 215, + 576 + ], + "score": 1.0, + "content": "for all methods, we fixed", + "type": "text" + }, + { + "bbox": [ + 215, + 563, + 228, + 574 + ], + "score": 0.88, + "content": "L _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 561, + 299, + 576 + ], + "score": 1.0, + "content": "-regularization at", + "type": "text" + }, + { + "bbox": [ + 300, + 563, + 322, + 574 + ], + "score": 0.9, + "content": "1 0 ^ { - 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 561, + 506, + 576 + ], + "score": 1.0, + "content": ". We then randomly sampled learning rates", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 573, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 129, + 589 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 129, + 574, + 175, + 585 + ], + "score": 0.9, + "content": "1 0 ^ { U ( - 4 , - 2 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 573, + 506, + 589 + ], + "score": 1.0, + "content": "to train the ERM with data augmentation model. For all other models, we took", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 585, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 506, + 598 + ], + "score": 1.0, + "content": "the best learning rate that we found for the ERM with data augmentation model and searched over", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 597, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 505, + 609 + ], + "score": 1.0, + "content": "one order of magnitude lower and higher from it. As in Zhang et al. (2021), we used a ResNet-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "score": 1.0, + "content": "50 model initialized by pretraining on ImageNet. For SwAV pre-training, instead of following the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 619, + 504, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 504, + 631 + ], + "score": 1.0, + "content": "early-stopping procedure in Appendix B.4, we trained for the full 400 epochs used in Caron et al.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 630, + 504, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 630, + 504, + 642 + ], + "score": 1.0, + "content": "(2020) since the experiment finished relatively quickly compared to the larger WILDS 2.0 datasets.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 37 + }, + { + "type": "title", + "bbox": [ + 107, + 655, + 171, + 667 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 173, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 173, + 668 + ], + "score": 1.0, + "content": "E.2 RESULTS", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 676, + 504, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 324, + 690 + ], + "score": 1.0, + "content": "Table 15 shows the results of our experiments on real", + "type": "text" + }, + { + "bbox": [ + 324, + 678, + 337, + 687 + ], + "score": 0.34, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 676, + 504, + 690 + ], + "score": 1.0, + "content": "sketch. The use of (strong) data augmen-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 687, + 504, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 276, + 699 + ], + "score": 1.0, + "content": "tation improved ERM performance from", + "type": "text" + }, + { + "bbox": [ + 276, + 688, + 303, + 698 + ], + "score": 0.87, + "content": "3 4 . 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 687, + 316, + 699 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 317, + 688, + 344, + 698 + ], + "score": 0.87, + "content": "3 5 . 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 687, + 504, + 699 + ], + "score": 1.0, + "content": ". All unsupervised adaptation methods", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "except CORAL improved over ERM. We also tested the use of strong vs. weak augmentation for la-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "beled examples for both Pseudo-Label and FixMatch, and we found that using strong augmentation", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 720, + 300, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 300, + 733 + ], + "score": 1.0, + "content": "for the labeled examples improves performance.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 46 + } + ], + "page_idx": 46, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 294, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 759 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 13, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 288, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 291, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 291, + 96 + ], + "score": 1.0, + "content": "E EXPERIMENTS ON DOMAINNET", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 195 + ], + "lines": [ + { + "bbox": [ + 105, + 107, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 107, + 505, + 119 + ], + "score": 1.0, + "content": "Prior work has shown that domain-invariant, self-training, and self-supervised methods can perform", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 118, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 505, + 130 + ], + "score": 1.0, + "content": "well on standard benchmarks for unsupervised domain adaptation. In this section, we describe", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 129, + 505, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 129, + 505, + 141 + ], + "score": 1.0, + "content": "our experiments on DomainNet (Peng et al., 2019), a standard unsupervised domain adaptation", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 140, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 505, + 152 + ], + "score": 1.0, + "content": "benchmark for object recognition. Our goal was to verify that our training/tuning protocol and our", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 151, + 504, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 151, + 504, + 162 + ], + "score": 1.0, + "content": "implementations of the methods we benchmark in Section 6—which differ slightly from prior work", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 162, + 504, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 162, + 504, + 173 + ], + "score": 1.0, + "content": "in the ways described in Section B—still result in models that can perform well on DomainNet.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 105, + 172, + 505, + 186 + ], + "score": 1.0, + "content": "Consistent with prior work, the methods we benchmark in Section 6, with the exception of CORAL,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 183, + 396, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 396, + 196 + ], + "score": 1.0, + "content": "all improve over standard ERM training in our DomainNet experiments.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 107, + 505, + 196 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 209, + 161, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 208, + 162, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 162, + 222 + ], + "score": 1.0, + "content": "E.1 SETUP", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 230, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "DomainNet is an object recognition dataset with approximately 600,000 images across six different", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 240, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 505, + 254 + ], + "score": 1.0, + "content": "domains: sketch, real, quickdraw, painting, infograph and clipart (Peng et al., 2019). Typically, one", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "score": 1.0, + "content": "of these domains is selected as the source, and another domain as the target for evaluation. In our", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 263, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 222, + 275 + ], + "score": 1.0, + "content": "experiments, we use the real", + "type": "text" + }, + { + "bbox": [ + 222, + 264, + 235, + 273 + ], + "score": 0.75, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 263, + 506, + 275 + ], + "score": 1.0, + "content": "sketch setting for two reasons: it is a common choice in prior work", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 275, + 504, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 504, + 287 + ], + "score": 1.0, + "content": "on DomainNet, and as our models are pre-trained on ImageNet (following (Zhang et al., 2021)),", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 286, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 505, + 298 + ], + "score": 1.0, + "content": "we wanted to use the real domain as the source to be consistent with the realistic photographs used", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 505, + 308 + ], + "score": 1.0, + "content": "for ImageNet pretraining. While it is common to evaluate methods on multiple pairs of source and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 306, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 506, + 321 + ], + "score": 1.0, + "content": "target domains in DomainNet, in our experiments we only chose one pair, as our goal was only to", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 318, + 253, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 253, + 330 + ], + "score": 1.0, + "content": "verify consistency with prior results.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 231, + 506, + 330 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 342, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 343, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 479, + 354 + ], + "score": 1.0, + "content": "Data. The DomainNet dataset includes train and test splits for each of the domains, with", + "type": "text" + }, + { + "bbox": [ + 479, + 343, + 505, + 353 + ], + "score": 0.81, + "content": "70 \\text{‰}", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 353, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 126, + 364 + ], + "score": 0.84, + "content": "30 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 353, + 505, + 366 + ], + "score": 1.0, + "content": "split between train and test examples. The real domain has 172,947 images total: 120,906", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 364, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 506, + 377 + ], + "score": 1.0, + "content": "images in the train split and 52,041 images in the test split. The target domain, sketch, has a total of", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 375, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 389 + ], + "score": 1.0, + "content": "69,128 images: 48,212 in the train split and 20,916 images are in the test split. We used this data in", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 385, + 184, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 184, + 401 + ], + "score": 1.0, + "content": "the following way:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 343, + 506, + 401 + ] + }, + { + "type": "list", + "bbox": [ + 130, + 407, + 505, + 506 + ], + "lines": [ + { + "bbox": [ + 129, + 407, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 129, + 407, + 505, + 421 + ], + "score": 1.0, + "content": "1. For training, we used the source training examples (with labels) and the target training", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 418, + 250, + 432 + ], + "spans": [ + { + "bbox": [ + 141, + 418, + 250, + 432 + ], + "score": 1.0, + "content": "examples (without labels).", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 128, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "2. For validation, we used the same set of target training examples, but with labels; this", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 446, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 141, + 446, + 505, + 458 + ], + "score": 1.0, + "content": "overestimates performance in a true domain adaptation setting (where one would not have", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 142, + 457, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 142, + 457, + 505, + 469 + ], + "score": 1.0, + "content": "labeled target data), but it is a common practice in the literature, and we followed it for", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 141, + 468, + 383, + 480 + ], + "spans": [ + { + "bbox": [ + 141, + 468, + 383, + 480 + ], + "score": 1.0, + "content": "consistency with Jiang et al. (2020) and Zhang et al. (2021).", + "type": "text" + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 129, + 484, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 129, + 484, + 505, + 496 + ], + "score": 1.0, + "content": "3. For evaluation, we used the source test examples as the in-distribution test set, and the", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 495, + 358, + 506 + ], + "spans": [ + { + "bbox": [ + 141, + 495, + 358, + 506 + ], + "score": 1.0, + "content": "target test examples as the out-of-distribution test set.", + "type": "text" + } + ], + "index": 31, + "is_list_end_line": true + } + ], + "index": 27.5, + "bbox_fs": [ + 128, + 407, + 505, + 506 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 519, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 532 + ], + "score": 1.0, + "content": "Hyperparameters and other details. Other experimental details followed our main experiments", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 529, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 505, + 542 + ], + "score": 1.0, + "content": "in Section 6. We used the strong and weak data augmentation described for image classification", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 541, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 553 + ], + "score": 1.0, + "content": "in Appendix C. We set the total batch size to 96, which is the maximum that can fit on 12GB of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 551, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 565 + ], + "score": 1.0, + "content": "GPU memory. We tuned hyperparameters with the protocol described in Section D.4. Specifically,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 561, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 215, + 576 + ], + "score": 1.0, + "content": "for all methods, we fixed", + "type": "text" + }, + { + "bbox": [ + 215, + 563, + 228, + 574 + ], + "score": 0.88, + "content": "L _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 561, + 299, + 576 + ], + "score": 1.0, + "content": "-regularization at", + "type": "text" + }, + { + "bbox": [ + 300, + 563, + 322, + 574 + ], + "score": 0.9, + "content": "1 0 ^ { - 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 561, + 506, + 576 + ], + "score": 1.0, + "content": ". We then randomly sampled learning rates", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 573, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 129, + 589 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 129, + 574, + 175, + 585 + ], + "score": 0.9, + "content": "1 0 ^ { U ( - 4 , - 2 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 573, + 506, + 589 + ], + "score": 1.0, + "content": "to train the ERM with data augmentation model. For all other models, we took", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 585, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 506, + 598 + ], + "score": 1.0, + "content": "the best learning rate that we found for the ERM with data augmentation model and searched over", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 597, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 505, + 609 + ], + "score": 1.0, + "content": "one order of magnitude lower and higher from it. As in Zhang et al. (2021), we used a ResNet-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 505, + 620 + ], + "score": 1.0, + "content": "50 model initialized by pretraining on ImageNet. For SwAV pre-training, instead of following the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 619, + 504, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 504, + 631 + ], + "score": 1.0, + "content": "early-stopping procedure in Appendix B.4, we trained for the full 400 epochs used in Caron et al.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 630, + 504, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 630, + 504, + 642 + ], + "score": 1.0, + "content": "(2020) since the experiment finished relatively quickly compared to the larger WILDS 2.0 datasets.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 518, + 506, + 642 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 655, + 171, + 667 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 173, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 173, + 668 + ], + "score": 1.0, + "content": "E.2 RESULTS", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 107, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 676, + 504, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 324, + 690 + ], + "score": 1.0, + "content": "Table 15 shows the results of our experiments on real", + "type": "text" + }, + { + "bbox": [ + 324, + 678, + 337, + 687 + ], + "score": 0.34, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 676, + 504, + 690 + ], + "score": 1.0, + "content": "sketch. The use of (strong) data augmen-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 687, + 504, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 276, + 699 + ], + "score": 1.0, + "content": "tation improved ERM performance from", + "type": "text" + }, + { + "bbox": [ + 276, + 688, + 303, + 698 + ], + "score": 0.87, + "content": "3 4 . 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 687, + 316, + 699 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 317, + 688, + 344, + 698 + ], + "score": 0.87, + "content": "3 5 . 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 687, + 504, + 699 + ], + "score": 1.0, + "content": ". All unsupervised adaptation methods", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "except CORAL improved over ERM. We also tested the use of strong vs. weak augmentation for la-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "beled examples for both Pseudo-Label and FixMatch, and we found that using strong augmentation", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 720, + 300, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 300, + 733 + ], + "score": 1.0, + "content": "for the labeled examples improves performance.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 46, + "bbox_fs": [ + 105, + 676, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 139, + 80, + 473, + 218 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 139, + 80, + 473, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 139, + 80, + 473, + 218 + ], + "spans": [ + { + "bbox": [ + 139, + 80, + 473, + 218 + ], + "score": 0.984, + "html": "
In-distribution (real)Out-of-distribution (sketch)
ERM (-data aug) ERM82.6 (0.0) 82.5 (0.3)34.9 (0.2) 35.9 (0.3)
CORAL79.1 (0.4)33.6 (0.6)
DANN77.8 (0.2)39.4 (0.8)
Pseudo-Label79.9 (0.2)
Pseudo-Label (weak aug)79.9 (0.6)36.1 (0.4) 32.0 (0.8)
FixMatch80.8 (0.2)50.2 (0.4)
FixMatch (weak aug)80.1 (0.1)49.3 (0.2)
Noisy Student82.0 (0.3)39.7 (0.2)
SwAV
79.0 (0.3)38.2 (0.4)
", + "type": "table", + "image_path": "c133739a5f00b64fc61ceb1ffd3524798eef2c81b60d434dd3b10cf046139fe7.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 139, + 80, + 473, + 126.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 139, + 126.0, + 473, + 172.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 139, + 172.0, + 473, + 218.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 107, + 227, + 505, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 227, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 505, + 238 + ], + "score": 1.0, + "content": "Table 15: The in-distribution vs. out-of-distribution test performance of each method on DomainNet", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 108, + 238, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 108, + 238, + 168, + 249 + ], + "score": 0.65, + "content": "( r e a l \\to s k e t c h", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 238, + 505, + 250 + ], + "score": 1.0, + "content": "). We also included the results of applying weak instead of strong augmentation on", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 249, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 506, + 261 + ], + "score": 1.0, + "content": "labeled examples for Pseudo-Label and FixMatch. Parentheses show standard deviation across 3", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 260, + 150, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 150, + 272 + ], + "score": 1.0, + "content": "replicates.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 106, + 292, + 505, + 359 + ], + "lines": [ + { + "bbox": [ + 105, + 292, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 505, + 305 + ], + "score": 1.0, + "content": "For DANN, Pseudo-Label, and FixMatch, we compared our results against the results reported in", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 371, + 316 + ], + "score": 1.0, + "content": "Zhang et al. (2021). Performance was similar for DANN (ours,", + "type": "text" + }, + { + "bbox": [ + 371, + 304, + 398, + 314 + ], + "score": 0.86, + "content": "3 9 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 303, + 431, + 316 + ], + "score": 1.0, + "content": "; theirs,", + "type": "text" + }, + { + "bbox": [ + 432, + 304, + 460, + 315 + ], + "score": 0.85, + "content": "4 0 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "). For Fix-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 307, + 327 + ], + "score": 1.0, + "content": "Match, our implementation performs better (ours,", + "type": "text" + }, + { + "bbox": [ + 307, + 315, + 334, + 325 + ], + "score": 0.87, + "content": "5 0 . 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 314, + 365, + 327 + ], + "score": 1.0, + "content": "; theirs,", + "type": "text" + }, + { + "bbox": [ + 366, + 315, + 393, + 326 + ], + "score": 0.86, + "content": "4 5 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "); this is partially due to our", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 324, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 104, + 324, + 505, + 339 + ], + "score": 1.0, + "content": "use of strong instead of weak augmentation for the labeled data, which increases performance by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 336, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 128, + 347 + ], + "score": 0.85, + "content": "0 . 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 336, + 381, + 349 + ], + "score": 1.0, + "content": ". For Pseudo-Label, our implementation performs worse (ours,", + "type": "text" + }, + { + "bbox": [ + 381, + 337, + 408, + 347 + ], + "score": 0.86, + "content": "3 6 . 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 336, + 439, + 349 + ], + "score": 1.0, + "content": "; theirs,", + "type": "text" + }, + { + "bbox": [ + 439, + 337, + 467, + 348 + ], + "score": 0.83, + "content": "4 0 . 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 336, + 505, + 349 + ], + "score": 1.0, + "content": "), and we", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 347, + 325, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 325, + 361 + ], + "score": 1.0, + "content": "believe it is due to variation in hyperparameter tuning.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 365, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 363, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 447, + 378 + ], + "score": 1.0, + "content": "For Noisy Student, Berthelot et al. (2021) reported significantly lower numbers (ours,", + "type": "text" + }, + { + "bbox": [ + 447, + 365, + 474, + 375 + ], + "score": 0.85, + "content": "3 9 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 363, + 505, + 378 + ], + "score": 1.0, + "content": "; theirs,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 135, + 387 + ], + "score": 0.84, + "content": "3 2 . 6 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 375, + 505, + 388 + ], + "score": 1.0, + "content": ". However, this is expected as they trained their models from scratch, whereas we used", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 387, + 226, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 226, + 399 + ], + "score": 1.0, + "content": "ImageNet-pretrained models.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 403, + 505, + 459 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "We were unable to find comparable results in prior work for CORAL and SwAV pretraining on", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 414, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 140, + 427 + ], + "score": 1.0, + "content": "the real", + "type": "text" + }, + { + "bbox": [ + 141, + 415, + 154, + 425 + ], + "score": 0.65, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 414, + 505, + 427 + ], + "score": 1.0, + "content": "sketch split. Prior work has shown that these methods can improve performance on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 425, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 506, + 438 + ], + "score": 1.0, + "content": "other unsupervised adaptation datasets (Sun & Saenko, 2016; Shen et al., 2021). On our DomainNet", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 436, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 506, + 449 + ], + "score": 1.0, + "content": "experiments, we found that SwAV pre-training did improve performance over ERM, though CORAL", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 446, + 182, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 182, + 459 + ], + "score": 1.0, + "content": "did not (Table 15).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18 + } + ], + "page_idx": 47, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "48", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 139, + 80, + 473, + 218 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 139, + 80, + 473, + 218 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 139, + 80, + 473, + 218 + ], + "spans": [ + { + "bbox": [ + 139, + 80, + 473, + 218 + ], + "score": 0.984, + "html": "
In-distribution (real)Out-of-distribution (sketch)
ERM (-data aug) ERM82.6 (0.0) 82.5 (0.3)34.9 (0.2) 35.9 (0.3)
CORAL79.1 (0.4)33.6 (0.6)
DANN77.8 (0.2)39.4 (0.8)
Pseudo-Label79.9 (0.2)
Pseudo-Label (weak aug)79.9 (0.6)36.1 (0.4) 32.0 (0.8)
FixMatch80.8 (0.2)50.2 (0.4)
FixMatch (weak aug)80.1 (0.1)49.3 (0.2)
Noisy Student82.0 (0.3)39.7 (0.2)
SwAV
79.0 (0.3)38.2 (0.4)
", + "type": "table", + "image_path": "c133739a5f00b64fc61ceb1ffd3524798eef2c81b60d434dd3b10cf046139fe7.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 139, + 80, + 473, + 126.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 139, + 126.0, + 473, + 172.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 139, + 172.0, + 473, + 218.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 107, + 227, + 505, + 271 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 227, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 505, + 238 + ], + "score": 1.0, + "content": "Table 15: The in-distribution vs. out-of-distribution test performance of each method on DomainNet", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 108, + 238, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 108, + 238, + 168, + 249 + ], + "score": 0.65, + "content": "( r e a l \\to s k e t c h", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 238, + 505, + 250 + ], + "score": 1.0, + "content": "). We also included the results of applying weak instead of strong augmentation on", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 249, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 506, + 261 + ], + "score": 1.0, + "content": "labeled examples for Pseudo-Label and FixMatch. Parentheses show standard deviation across 3", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 260, + 150, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 150, + 272 + ], + "score": 1.0, + "content": "replicates.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 106, + 292, + 505, + 359 + ], + "lines": [ + { + "bbox": [ + 105, + 292, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 505, + 305 + ], + "score": 1.0, + "content": "For DANN, Pseudo-Label, and FixMatch, we compared our results against the results reported in", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 371, + 316 + ], + "score": 1.0, + "content": "Zhang et al. (2021). Performance was similar for DANN (ours,", + "type": "text" + }, + { + "bbox": [ + 371, + 304, + 398, + 314 + ], + "score": 0.86, + "content": "3 9 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 303, + 431, + 316 + ], + "score": 1.0, + "content": "; theirs,", + "type": "text" + }, + { + "bbox": [ + 432, + 304, + 460, + 315 + ], + "score": 0.85, + "content": "4 0 . 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "). For Fix-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 307, + 327 + ], + "score": 1.0, + "content": "Match, our implementation performs better (ours,", + "type": "text" + }, + { + "bbox": [ + 307, + 315, + 334, + 325 + ], + "score": 0.87, + "content": "5 0 . 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 314, + 365, + 327 + ], + "score": 1.0, + "content": "; theirs,", + "type": "text" + }, + { + "bbox": [ + 366, + 315, + 393, + 326 + ], + "score": 0.86, + "content": "4 5 . 3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "); this is partially due to our", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 324, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 104, + 324, + 505, + 339 + ], + "score": 1.0, + "content": "use of strong instead of weak augmentation for the labeled data, which increases performance by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 336, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 337, + 128, + 347 + ], + "score": 0.85, + "content": "0 . 9 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 336, + 381, + 349 + ], + "score": 1.0, + "content": ". For Pseudo-Label, our implementation performs worse (ours,", + "type": "text" + }, + { + "bbox": [ + 381, + 337, + 408, + 347 + ], + "score": 0.86, + "content": "3 6 . 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 336, + 439, + 349 + ], + "score": 1.0, + "content": "; theirs,", + "type": "text" + }, + { + "bbox": [ + 439, + 337, + 467, + 348 + ], + "score": 0.83, + "content": "4 0 . 6 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 336, + 505, + 349 + ], + "score": 1.0, + "content": "), and we", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 347, + 325, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 325, + 361 + ], + "score": 1.0, + "content": "believe it is due to variation in hyperparameter tuning.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9.5, + "bbox_fs": [ + 104, + 292, + 506, + 361 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 365, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 363, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 447, + 378 + ], + "score": 1.0, + "content": "For Noisy Student, Berthelot et al. (2021) reported significantly lower numbers (ours,", + "type": "text" + }, + { + "bbox": [ + 447, + 365, + 474, + 375 + ], + "score": 0.85, + "content": "3 9 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 363, + 505, + 378 + ], + "score": 1.0, + "content": "; theirs,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 135, + 387 + ], + "score": 0.84, + "content": "3 2 . 6 \\% )", + "type": "inline_equation" + }, + { + "bbox": [ + 136, + 375, + 505, + 388 + ], + "score": 1.0, + "content": ". However, this is expected as they trained their models from scratch, whereas we used", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 387, + 226, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 226, + 399 + ], + "score": 1.0, + "content": "ImageNet-pretrained models.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 363, + 505, + 399 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 403, + 505, + 459 + ], + "lines": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "We were unable to find comparable results in prior work for CORAL and SwAV pretraining on", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 414, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 140, + 427 + ], + "score": 1.0, + "content": "the real", + "type": "text" + }, + { + "bbox": [ + 141, + 415, + 154, + 425 + ], + "score": 0.65, + "content": "", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 414, + 505, + 427 + ], + "score": 1.0, + "content": "sketch split. Prior work has shown that these methods can improve performance on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 425, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 506, + 438 + ], + "score": 1.0, + "content": "other unsupervised adaptation datasets (Sun & Saenko, 2016; Shen et al., 2021). On our DomainNet", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 436, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 506, + 449 + ], + "score": 1.0, + "content": "experiments, we found that SwAV pre-training did improve performance over ERM, though CORAL", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 446, + 182, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 182, + 459 + ], + "score": 1.0, + "content": "did not (Table 15).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 403, + 506, + 459 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 371, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 373, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 373, + 96 + ], + "score": 1.0, + "content": "F FULLY-LABELED ERM EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 108, + 505, + 174 + ], + "lines": [ + { + "bbox": [ + 105, + 107, + 505, + 121 + ], + "spans": [ + { + "bbox": [ + 105, + 107, + 413, + 121 + ], + "score": 1.0, + "content": "The self-training methods we evaluate in Section 5 generate a pseudolabel", + "type": "text" + }, + { + "bbox": [ + 414, + 109, + 425, + 119 + ], + "score": 0.88, + "content": "\\tilde { y } _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 107, + 505, + 121 + ], + "score": 1.0, + "content": "for each unlabeled", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 119, + 505, + 132 + ], + "spans": [ + { + "bbox": [ + 105, + 119, + 143, + 132 + ], + "score": 1.0, + "content": "example", + "type": "text" + }, + { + "bbox": [ + 144, + 120, + 157, + 130 + ], + "score": 0.83, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 119, + 234, + 132 + ], + "score": 1.0, + "content": "and then train on", + "type": "text" + }, + { + "bbox": [ + 235, + 119, + 269, + 131 + ], + "score": 0.93, + "content": "( x _ { \\mathrm { U } } , \\tilde { y } _ { \\mathrm { U } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 119, + 505, + 132 + ], + "score": 1.0, + "content": "as if the pseudolabels were true labels. However, these", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 129, + 506, + 142 + ], + "spans": [ + { + "bbox": [ + 104, + 129, + 506, + 142 + ], + "score": 1.0, + "content": "pseudolabels may not be accurate. In this section, we describe how we ran fully-labeled ERM", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 140, + 505, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 505, + 153 + ], + "score": 1.0, + "content": "experiments using ground truth labels on the “unlabeled” data to establish informal upper bounds", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 505, + 164 + ], + "score": 1.0, + "content": "on how well we might expect a standard self-training approach to perform with perfect pseudolabel", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 163, + 146, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 146, + 176 + ], + "score": 1.0, + "content": "accuracy.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 180, + 505, + 300 + ], + "lines": [ + { + "bbox": [ + 105, + 179, + 505, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 505, + 192 + ], + "score": 1.0, + "content": "For four of our datasets (AMAZON-WILDS, CIVILCOMMENTS-WILDS, IWILDCAM2020-WILDS,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 189, + 505, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 505, + 204 + ], + "score": 1.0, + "content": "and FMOW-WILDS), we curated the “unlabeled” data by taking labeled examples and discarding", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 201, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 214 + ], + "score": 1.0, + "content": "the ground truth labels. For example, all 268,761 of the unlabeled target reviews in AMAZON-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 213, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 505, + 224 + ], + "score": 1.0, + "content": "WILDS actually have associated star ratings; these are available in our data loaders, but in our main", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 224, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 505, + 236 + ], + "score": 1.0, + "content": "experiments we treat these reviews as unlabeled by not loading the star ratings. We evaluated models", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 233, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 506, + 248 + ], + "score": 1.0, + "content": "trained via empirical risk minimization (ERM) on the combination of the standard labeled training", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "score": 1.0, + "content": "set and the unlabeled data with these hidden labels revealed. For example, in AMAZON-WILDS, we", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 256, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 104, + 256, + 505, + 269 + ], + "score": 1.0, + "content": "pool together the labeled source examples as well as the unlabeled target examples with ground truth", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "labels, and evaluate ERM models trained on all of that data. As with all of the other experiments in", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "score": 1.0, + "content": "this paper, we evaluate test performance for all datasets on the labeled target splits, so at no point", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 290, + 281, + 301 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 281, + 301 + ], + "score": 1.0, + "content": "are we training on our actual test examples.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 107, + 317, + 218, + 328 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 219, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 219, + 329 + ], + "score": 1.0, + "content": "F.1 HYPERPARAMETERS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 338, + 505, + 416 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "Pooling labeled and unlabeled data. For all datasets, we pooled labeled source examples with", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "score": 1.0, + "content": "examples from the same “unlabeled” split as in our main experiments (Table 2). We computed gra-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "score": 1.0, + "content": "dients for labeled minibatches and unlabeled minibatches separately, which means that for models", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 372, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 384 + ], + "score": 1.0, + "content": "using batch normalization, the labeled and unlabeled data were normalized separately. However, we", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 382, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 505, + 394 + ], + "score": 1.0, + "content": "fixed the labeled to “unlabeled” batch size ratios to match the overall labeled to unlabeled dataset", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 394, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 405 + ], + "score": 1.0, + "content": "size ratio, so other than the batch normalization effects, the training procedure can be viewed as", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 405, + 341, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 341, + 417 + ], + "score": 1.0, + "content": "running ERM on the pooled labeled and “unlabeled” data.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 431, + 505, + 475 + ], + "lines": [ + { + "bbox": [ + 105, + 431, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 443 + ], + "score": 1.0, + "content": "Number of epochs. With the exception of IWILDCAM2020-WILDS, detailed below, we followed", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 442, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 454 + ], + "score": 1.0, + "content": "the procedure in Appendix D.4 to adjust the number of epochs based on the labeled to unlabeled", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "batch size ratios. This resulted in a similar amount of computation allocated to these fully-labeled", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 464, + 326, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 326, + 476 + ], + "score": 1.0, + "content": "ERM experiments as the other experiments in Table 2.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 491, + 505, + 568 + ], + "lines": [ + { + "bbox": [ + 106, + 489, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 505, + 504 + ], + "score": 1.0, + "content": "Other details. Other experimental details were kept similar to the other experiments in the pa-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "per. Specifically, we tuned each experiment by randomly sampling 10 different hyperparameters", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 512, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 526 + ], + "score": 1.0, + "content": "within the ranges defined in Appendix D.4; the only hyperparameter we tuned in these experiments", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 524, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 505, + 536 + ], + "score": 1.0, + "content": "was the learning rate. We early stopped and selected the best hyperparameters based on the OOD", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "validation performance, and then ran replicates using the best hyperparameters. We also used data", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "augmentation for IWILDCAM2020-WILDS and FMOW-WILDS but not for AMAZON-WILDS and", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 556, + 219, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 219, + 568 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33 + }, + { + "type": "title", + "bbox": [ + 108, + 585, + 249, + 595 + ], + "lines": [ + { + "bbox": [ + 106, + 585, + 251, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 251, + 596 + ], + "score": 1.0, + "content": "F.2 DATASET-SPECIFIC DETAILS", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 606, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 106, + 606, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 505, + 619 + ], + "score": 1.0, + "content": "AMAZON-WILDS. We matched the experiments in Table 2 by training on the unlabeled target data", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 617, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 629 + ], + "score": 1.0, + "content": "(268,761 examples). In addition, we ran a separate experiment where we trained on the unlabeled", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 380, + 641 + ], + "score": 1.0, + "content": "extra data instead of the unlabeled target data, as the former has", + "type": "text" + }, + { + "bbox": [ + 380, + 628, + 400, + 639 + ], + "score": 0.88, + "content": "1 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "the number of examples", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 639, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 652 + ], + "score": 1.0, + "content": "(2,927,841 examples). However, this did not improve performance. Using the unlabeled target data,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 650, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 255, + 663 + ], + "score": 1.0, + "content": "we obtained an average accuracy of", + "type": "text" + }, + { + "bbox": [ + 256, + 650, + 308, + 662 + ], + "score": 0.78, + "content": "7 3 . 6 ( \\pm 0 . 1 ) ", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 650, + 470, + 663 + ], + "score": 1.0, + "content": "and a 10th percentile accuracy of 56.4", + "type": "text" + }, + { + "bbox": [ + 470, + 650, + 501, + 662 + ], + "score": 0.59, + "content": "( \\pm \\ : 0 . 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 650, + 505, + 663 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 409, + 675 + ], + "score": 1.0, + "content": "whereas using the unlabeled extra data, we obtained an average accuracy of", + "type": "text" + }, + { + "bbox": [ + 409, + 661, + 459, + 672 + ], + "score": 0.5, + "content": "7 3 . 1 \\ : ( \\pm 0 . 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 661, + 506, + 675 + ], + "score": 1.0, + "content": "and a 10th", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 672, + 251, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 196, + 685 + ], + "score": 1.0, + "content": "percentile accuracy of", + "type": "text" + }, + { + "bbox": [ + 197, + 672, + 247, + 684 + ], + "score": 0.58, + "content": "5 4 . 7 \\ : ( \\pm 0 . 0 ) ", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 673, + 251, + 685 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS. We used the unlabeled extra split (1,551,515 examples). As in our", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "other experiments on CIVILCOMMENTS-WILDS, we accounted for label imbalance by sampling", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 719, + 362, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 362, + 734 + ], + "score": 1.0, + "content": "class-balanced labeled and “unlabeled” batches during training.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46 + } + ], + "page_idx": 48, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 311, + 759 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 762 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 371, + 93 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 373, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 373, + 96 + ], + "score": 1.0, + "content": "F FULLY-LABELED ERM EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 108, + 505, + 174 + ], + "lines": [ + { + "bbox": [ + 105, + 107, + 505, + 121 + ], + "spans": [ + { + "bbox": [ + 105, + 107, + 413, + 121 + ], + "score": 1.0, + "content": "The self-training methods we evaluate in Section 5 generate a pseudolabel", + "type": "text" + }, + { + "bbox": [ + 414, + 109, + 425, + 119 + ], + "score": 0.88, + "content": "\\tilde { y } _ { \\mathrm { U } }", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 107, + 505, + 121 + ], + "score": 1.0, + "content": "for each unlabeled", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 119, + 505, + 132 + ], + "spans": [ + { + "bbox": [ + 105, + 119, + 143, + 132 + ], + "score": 1.0, + "content": "example", + "type": "text" + }, + { + "bbox": [ + 144, + 120, + 157, + 130 + ], + "score": 0.83, + "content": "x _ { \\mathrm { { U } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 119, + 234, + 132 + ], + "score": 1.0, + "content": "and then train on", + "type": "text" + }, + { + "bbox": [ + 235, + 119, + 269, + 131 + ], + "score": 0.93, + "content": "( x _ { \\mathrm { U } } , \\tilde { y } _ { \\mathrm { U } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 119, + 505, + 132 + ], + "score": 1.0, + "content": "as if the pseudolabels were true labels. However, these", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 129, + 506, + 142 + ], + "spans": [ + { + "bbox": [ + 104, + 129, + 506, + 142 + ], + "score": 1.0, + "content": "pseudolabels may not be accurate. In this section, we describe how we ran fully-labeled ERM", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 140, + 505, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 505, + 153 + ], + "score": 1.0, + "content": "experiments using ground truth labels on the “unlabeled” data to establish informal upper bounds", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 505, + 164 + ], + "score": 1.0, + "content": "on how well we might expect a standard self-training approach to perform with perfect pseudolabel", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 163, + 146, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 146, + 176 + ], + "score": 1.0, + "content": "accuracy.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5, + "bbox_fs": [ + 104, + 107, + 506, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 180, + 505, + 300 + ], + "lines": [ + { + "bbox": [ + 105, + 179, + 505, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 505, + 192 + ], + "score": 1.0, + "content": "For four of our datasets (AMAZON-WILDS, CIVILCOMMENTS-WILDS, IWILDCAM2020-WILDS,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 189, + 505, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 189, + 505, + 204 + ], + "score": 1.0, + "content": "and FMOW-WILDS), we curated the “unlabeled” data by taking labeled examples and discarding", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 201, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 505, + 214 + ], + "score": 1.0, + "content": "the ground truth labels. For example, all 268,761 of the unlabeled target reviews in AMAZON-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 213, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 505, + 224 + ], + "score": 1.0, + "content": "WILDS actually have associated star ratings; these are available in our data loaders, but in our main", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 224, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 505, + 236 + ], + "score": 1.0, + "content": "experiments we treat these reviews as unlabeled by not loading the star ratings. We evaluated models", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 233, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 506, + 248 + ], + "score": 1.0, + "content": "trained via empirical risk minimization (ERM) on the combination of the standard labeled training", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "score": 1.0, + "content": "set and the unlabeled data with these hidden labels revealed. For example, in AMAZON-WILDS, we", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 256, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 104, + 256, + 505, + 269 + ], + "score": 1.0, + "content": "pool together the labeled source examples as well as the unlabeled target examples with ground truth", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "labels, and evaluate ERM models trained on all of that data. As with all of the other experiments in", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "score": 1.0, + "content": "this paper, we evaluate test performance for all datasets on the labeled target splits, so at no point", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 290, + 281, + 301 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 281, + 301 + ], + "score": 1.0, + "content": "are we training on our actual test examples.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 12, + "bbox_fs": [ + 104, + 179, + 506, + 301 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 317, + 218, + 328 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 219, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 219, + 329 + ], + "score": 1.0, + "content": "F.1 HYPERPARAMETERS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 338, + 505, + 416 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "Pooling labeled and unlabeled data. For all datasets, we pooled labeled source examples with", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "score": 1.0, + "content": "examples from the same “unlabeled” split as in our main experiments (Table 2). We computed gra-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 360, + 505, + 373 + ], + "score": 1.0, + "content": "dients for labeled minibatches and unlabeled minibatches separately, which means that for models", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 372, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 384 + ], + "score": 1.0, + "content": "using batch normalization, the labeled and unlabeled data were normalized separately. However, we", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 382, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 505, + 394 + ], + "score": 1.0, + "content": "fixed the labeled to “unlabeled” batch size ratios to match the overall labeled to unlabeled dataset", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 394, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 405 + ], + "score": 1.0, + "content": "size ratio, so other than the batch normalization effects, the training procedure can be viewed as", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 405, + 341, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 341, + 417 + ], + "score": 1.0, + "content": "running ERM on the pooled labeled and “unlabeled” data.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 338, + 505, + 417 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 431, + 505, + 475 + ], + "lines": [ + { + "bbox": [ + 105, + 431, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 443 + ], + "score": 1.0, + "content": "Number of epochs. With the exception of IWILDCAM2020-WILDS, detailed below, we followed", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 442, + 505, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 454 + ], + "score": 1.0, + "content": "the procedure in Appendix D.4 to adjust the number of epochs based on the labeled to unlabeled", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "batch size ratios. This resulted in a similar amount of computation allocated to these fully-labeled", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 464, + 326, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 326, + 476 + ], + "score": 1.0, + "content": "ERM experiments as the other experiments in Table 2.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 431, + 506, + 476 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 491, + 505, + 568 + ], + "lines": [ + { + "bbox": [ + 106, + 489, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 489, + 505, + 504 + ], + "score": 1.0, + "content": "Other details. Other experimental details were kept similar to the other experiments in the pa-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "per. Specifically, we tuned each experiment by randomly sampling 10 different hyperparameters", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 512, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 526 + ], + "score": 1.0, + "content": "within the ranges defined in Appendix D.4; the only hyperparameter we tuned in these experiments", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 524, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 505, + 536 + ], + "score": 1.0, + "content": "was the learning rate. We early stopped and selected the best hyperparameters based on the OOD", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 547 + ], + "score": 1.0, + "content": "validation performance, and then ran replicates using the best hyperparameters. We also used data", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "augmentation for IWILDCAM2020-WILDS and FMOW-WILDS but not for AMAZON-WILDS and", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 556, + 219, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 219, + 568 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 489, + 506, + 568 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 585, + 249, + 595 + ], + "lines": [ + { + "bbox": [ + 106, + 585, + 251, + 596 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 251, + 596 + ], + "score": 1.0, + "content": "F.2 DATASET-SPECIFIC DETAILS", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 606, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 106, + 606, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 505, + 619 + ], + "score": 1.0, + "content": "AMAZON-WILDS. We matched the experiments in Table 2 by training on the unlabeled target data", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 617, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 629 + ], + "score": 1.0, + "content": "(268,761 examples). In addition, we ran a separate experiment where we trained on the unlabeled", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 380, + 641 + ], + "score": 1.0, + "content": "extra data instead of the unlabeled target data, as the former has", + "type": "text" + }, + { + "bbox": [ + 380, + 628, + 400, + 639 + ], + "score": 0.88, + "content": "1 0 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "the number of examples", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 639, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 652 + ], + "score": 1.0, + "content": "(2,927,841 examples). However, this did not improve performance. Using the unlabeled target data,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 650, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 255, + 663 + ], + "score": 1.0, + "content": "we obtained an average accuracy of", + "type": "text" + }, + { + "bbox": [ + 256, + 650, + 308, + 662 + ], + "score": 0.78, + "content": "7 3 . 6 ( \\pm 0 . 1 ) ", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 650, + 470, + 663 + ], + "score": 1.0, + "content": "and a 10th percentile accuracy of 56.4", + "type": "text" + }, + { + "bbox": [ + 470, + 650, + 501, + 662 + ], + "score": 0.59, + "content": "( \\pm \\ : 0 . 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 650, + 505, + 663 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 409, + 675 + ], + "score": 1.0, + "content": "whereas using the unlabeled extra data, we obtained an average accuracy of", + "type": "text" + }, + { + "bbox": [ + 409, + 661, + 459, + 672 + ], + "score": 0.5, + "content": "7 3 . 1 \\ : ( \\pm 0 . 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 661, + 506, + 675 + ], + "score": 1.0, + "content": "and a 10th", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 672, + 251, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 196, + 685 + ], + "score": 1.0, + "content": "percentile accuracy of", + "type": "text" + }, + { + "bbox": [ + 197, + 672, + 247, + 684 + ], + "score": 0.58, + "content": "5 4 . 7 \\ : ( \\pm 0 . 0 ) ", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 673, + 251, + 685 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 606, + 506, + 685 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "CIVILCOMMENTS-WILDS. We used the unlabeled extra split (1,551,515 examples). As in our", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "other experiments on CIVILCOMMENTS-WILDS, we accounted for label imbalance by sampling", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 719, + 362, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 362, + 734 + ], + "score": 1.0, + "content": "class-balanced labeled and “unlabeled” batches during training.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46, + "bbox_fs": [ + 105, + 698, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 148 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "score": 1.0, + "content": "IWILDCAM2020-WILDS. We used the unlabeled extra split. Out of the 819,120 unlabeled extra", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "examples, 108,452 examples have ground truth labels (animal species) that are not present in the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "labeled training and test sets, so we omitted those examples and trained on the remaining 710,668", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "examples. We found that we required twice as many epochs compared to the other unlabeled meth-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "score": 1.0, + "content": "ods for the fully-labeled ERM training to converge, so we doubled the amount of compute allocated", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 343, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 343, + 150 + ], + "score": 1.0, + "content": "to the fully-labeled IWILDCAM2020-WILDS experiments.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 160, + 404, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 159, + 406, + 175 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 406, + 175 + ], + "score": 1.0, + "content": "FMOW-WILDS. We used the unlabeled target split (173,208 examples).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + } + ], + "page_idx": 49, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "50", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 148 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "score": 1.0, + "content": "IWILDCAM2020-WILDS. We used the unlabeled extra split. Out of the 819,120 unlabeled extra", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "examples, 108,452 examples have ground truth labels (animal species) that are not present in the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "labeled training and test sets, so we omitted those examples and trained on the remaining 710,668", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "examples. We found that we required twice as many epochs compared to the other unlabeled meth-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "score": 1.0, + "content": "ods for the fully-labeled ERM training to converge, so we doubled the amount of compute allocated", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 343, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 343, + 150 + ], + "score": 1.0, + "content": "to the fully-labeled IWILDCAM2020-WILDS experiments.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 81, + 506, + 150 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 160, + 404, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 159, + 406, + 175 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 406, + 175 + ], + "score": 1.0, + "content": "FMOW-WILDS. We used the unlabeled target split (173,208 examples).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 159, + 406, + 175 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 398, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 400, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 400, + 96 + ], + "score": 1.0, + "content": "G USING THE WILDS LIBRARY WITH UNLABELED DATA", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 140 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 506, + 118 + ], + "score": 1.0, + "content": "We have extended the existing WILDS library (Koh et al., 2021) to add data loaders for each of the 8", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 116, + 506, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 506, + 130 + ], + "score": 1.0, + "content": "datasets with unlabeled data. These data loaders are compatible with the WILDS 1.0 APIs, allowing", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 385, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 385, + 140 + ], + "score": 1.0, + "content": "the unlabeled data to be accessed in a similar way to the labeled data:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "image", + "bbox": [ + 115, + 159, + 493, + 307 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 159, + 493, + 307 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 159, + 493, + 307 + ], + "spans": [ + { + "bbox": [ + 115, + 159, + 493, + 307 + ], + "score": 0.906, + "type": "image", + "image_path": "918ce9d12404760bf542ff57e8cae97ee7c36cafded0f6782f3daa0654841009.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 115, + 159, + 493, + 208.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 115, + 208.33333333333334, + 493, + 257.6666666666667 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 115, + 257.6666666666667, + 493, + 307.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 162, + 325, + 447, + 336 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 162, + 324, + 448, + 337 + ], + "spans": [ + { + "bbox": [ + 162, + 324, + 448, + 337 + ], + "score": 1.0, + "content": "Figure 3: Example of data loading for both labeled and unlabeled data.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + } + ], + "index": 6.0 + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 505, + 421 + ], + "lines": [ + { + "bbox": [ + 106, + 355, + 504, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 504, + 366 + ], + "score": 1.0, + "content": "As in the existing WILDS library, data downloading is automated. In addition, we implemented", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 365, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 505, + 379 + ], + "score": 1.0, + "content": "CORAL, DANN, Pseudo-Label, FixMatch, and Noisy Student using the existing WILDS interfaces.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 375, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 390 + ], + "score": 1.0, + "content": "This allows developers to easily extend these algorithms and evaluate them in a standardized way", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 387, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 506, + 400 + ], + "score": 1.0, + "content": "on all of the WILDS datasets with unlabeled data. The WILDS repository also contains scripts for", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 398, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 506, + 411 + ], + "score": 1.0, + "content": "masked language model pre-training and for SwAV pre-training, which uses a modified version of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 410, + 410, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 410, + 422 + ], + "score": 1.0, + "content": "the public SwAV repository that can interface with the WILDS data loaders.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5 + } + ], + "page_idx": 50, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2022", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 398, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 400, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 400, + 96 + ], + "score": 1.0, + "content": "G USING THE WILDS LIBRARY WITH UNLABELED DATA", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 505, + 140 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 506, + 118 + ], + "score": 1.0, + "content": "We have extended the existing WILDS library (Koh et al., 2021) to add data loaders for each of the 8", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 116, + 506, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 506, + 130 + ], + "score": 1.0, + "content": "datasets with unlabeled data. These data loaders are compatible with the WILDS 1.0 APIs, allowing", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 385, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 385, + 140 + ], + "score": 1.0, + "content": "the unlabeled data to be accessed in a similar way to the labeled data:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 106, + 506, + 140 + ] + }, + { + "type": "image", + "bbox": [ + 115, + 159, + 493, + 307 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 159, + 493, + 307 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 159, + 493, + 307 + ], + "spans": [ + { + "bbox": [ + 115, + 159, + 493, + 307 + ], + "score": 0.906, + "type": "image", + "image_path": "918ce9d12404760bf542ff57e8cae97ee7c36cafded0f6782f3daa0654841009.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 115, + 159, + 493, + 208.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 115, + 208.33333333333334, + 493, + 257.6666666666667 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 115, + 257.6666666666667, + 493, + 307.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 162, + 325, + 447, + 336 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 162, + 324, + 448, + 337 + ], + "spans": [ + { + "bbox": [ + 162, + 324, + 448, + 337 + ], + "score": 1.0, + "content": "Figure 3: Example of data loading for both labeled and unlabeled data.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + } + ], + "index": 6.0 + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 505, + 421 + ], + "lines": [ + { + "bbox": [ + 106, + 355, + 504, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 504, + 366 + ], + "score": 1.0, + "content": "As in the existing WILDS library, data downloading is automated. In addition, we implemented", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 365, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 365, + 505, + 379 + ], + "score": 1.0, + "content": "CORAL, DANN, Pseudo-Label, FixMatch, and Noisy Student using the existing WILDS interfaces.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 375, + 505, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 390 + ], + "score": 1.0, + "content": "This allows developers to easily extend these algorithms and evaluate them in a standardized way", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 387, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 506, + 400 + ], + "score": 1.0, + "content": "on all of the WILDS datasets with unlabeled data. The WILDS repository also contains scripts for", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 398, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 398, + 506, + 411 + ], + "score": 1.0, + "content": "masked language model pre-training and for SwAV pre-training, which uses a modified version of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 410, + 410, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 410, + 422 + ], + "score": 1.0, + "content": "the public SwAV repository that can interface with the WILDS data loaders.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 355, + 506, + 422 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/dev/zT5T9gHpGI/zT5T9gHpGI.md b/parse/dev/zT5T9gHpGI/zT5T9gHpGI.md new file mode 100644 index 0000000000000000000000000000000000000000..ffba290295267aa0376bcaea7961aa24af01ec21 --- /dev/null +++ b/parse/dev/zT5T9gHpGI/zT5T9gHpGI.md @@ -0,0 +1,927 @@ +# ADVERSARIAL COUNTERFACTUAL ENVIRONMENT MODEL LEARNING + +Anonymous authors Paper under double-blind review + +# ABSTRACT + +A good model for action-effect prediction, i.e., the environment model, is essential for sample-efficient policy learning, in which the agent can take numerous free trials to find good policies. Currently, the model is commonly learned by fitting historical transition data through empirical risk minimization (ERM). However, we discover that simple data fitting can lead to a model that will be totally wrong in guiding policy learning due to the selection bias in offline dataset collection. In this work, we introduce weighted empirical risk minimization (WERM) to handle this problem in model learning. A typical WERM method utilizes inverse propensity scores to re-weight the training data to approximate the target distribution. However, during the policy training, the data distributions of the candidate policies can be various and unknown. Thus, we propose an adversarial weighted empirical risk minimization (AWRM) objective that learns the model with respect to the worst case of the target distributions. We implement AWRM in a sequential decision structure, resulting in the GALILEO model learning algorithm. We also discover that GALILEO is closely related to adversarial model learning, explaining the empirical effectiveness of the latter. We apply GALILEO in synthetic tasks and verify that GALILEO makes accurate predictions on counterfactual data. We finally applied GALILEO in real-world offline policy learning tasks and found that GALILEO significantly improves policy performance in real-world testing. + +# 1 INTRODUCTION + +A good environment model is important for sample-efficient decision-making policy learning techniques like reinforcement learning (RL) (James & Johns, 2016). The agent can take trials with this model to find better policies, then the costly real-world trial-and-errors can be saved (James & Johns, 2016; Yu et al., 2020) or completely waived (Shi et al., 2019). In this process, the core of the models is to answer queries on counterfactual data unbiasedly, that is, given states, correctly answer what might happen if we were to carry out actions unseen in the training data (Levine et al., 2020). + +Requiring counterfactual queries makes the environment model learning essentially different from standard supervised learning (SL) which directly fits the offline dataset. In real-world applications, the offline data is often collected with selection bias, that is, for each state, each action might be chosen unfairly. Seeing the example in Fig. 1(a), to keep the ball following a target line, a behavior policy will use a smaller force when the ball’s location is closer to the target line. When a dataset is collected with selection bias, the association between the (location) states and (force) actions will make SL hard to identify the correct causal relationship of the states and actions to the next states respectively. Then when we query the model with counterfactual data, the predictions might be catastrophic failures. In Fig. 1(c), it mistakes that smaller forces will increase the ball’s next location. + +Generally speaking, the problem corresponds to a challenge of training the model in one dataset but testing in another dataset with a shifted distribution (i.e., the dataset generated by counterfactual queries), which is beyond the SL’s capability as it violates the independent and identically distributed (i.i.d.) assumption. The problem is widely discussed in causal inference for individual treatment effects (ITEs) estimation in many scenarios like patients’ treatment selection (Imbens, 1999; Alaa & van der Schaar, 2018). ITEs are the effects of treatments on individuals, which are measured by treating each individual under a uniform policy and evaluate the effect differences. Practical solutions use weighted empirical risk minimization (WERM) to handle this problem (Jung et al., 2020; Shimodaira, 2000; Hassanpour & Greiner, 2019). In particular, they estimate an inverse propensity score (IPS) to re-weight the training data to approximate the data distribution under a uniform policy. Then a model is trained under the reweighted data distribution. The distribution-shift problem is solved as ITEs estimation and model training are under the same distribution. + +![](images/b68193a7fb3b9605331279446f1bc56a0b45f7422ed57eced22fd9b4e8ab8a5c.jpg) +Figure 1: An example of selection bias and predictions under counterfactual queries. Subfigure (a) shows how the data is collected: a ball locates in a 2D plane whose position is $( x _ { t } , y _ { t } )$ at time $t$ . The ball will move to $( x _ { t + 1 } , y _ { t + 1 } )$ according to $x _ { t + 1 } = x _ { t } + 1$ and $y _ { t + 1 } \sim \mathcal { N } ( y _ { t } + a _ { t } , 2 )$ . Here, $a _ { t }$ is chosen by a control policy $a _ { t } \sim \mathcal { N } ( ( \phi - y _ { t } ) / 1 5 , 0 . 0 5 )$ parameterized by $\phi$ , which tries to keep the ball near the line $y = \phi$ . In Subfigure (a), $\phi$ is set to 62.5. Subfigure (b) shows the collected training data (grey dashed line) and the two learned models’ prediction of the next position of $y$ . All the models discovered the relation that the corresponding next $y$ will be smaller with a larger action. However, the truth is not because the larger $a _ { t }$ causes a smaller $y _ { t + 1 }$ , but the policy selects a small $a _ { t }$ when $y _ { t }$ is close to the target line. When we estimate the response curves by fixing $y _ { t }$ and reassigning action $a _ { t }$ with other actions $a _ { t } + \Delta a$ , where $\Delta a \in [ - 1 , 1 ]$ is a variation of action value, the model of SL will exploit the association and give opposite responses, while in AWRM and its practical implementation GALILEO, the predictions are closer to the ground truths. The result is in Subfigure (c), where the darker a region is, the more samples are fallen in. + +The selection bias can be regarded as an instance of the problem called “distributional shift” in offline model-based RL, which has also received great attention (Levine et al., 2020; Yu et al., 2020; Kidambi et al., 2020; Chen et al., 2021). However, previous methods, where naive supervised learning is used for environment model learning, ignore the problem in environment model learning but handling the problem by suppressing the policy exploration and learning in risky regions. Although these methods have made great progress in many tasks, so far, how to learn a better environment model that can alleviate the problem for faithful offline policy optimization has rarely been discussed. + +In this work, for faithful offline policy optimization, we introduce WERM to environment model learning. The extra challenge of model learning for policy optimization is that we have to query numerous different policies’ feedback besides the uniform policy for finding a good policy. Thus the target data distribution to reweight can be various and unknown. To solve the problem, we propose an objective called adversarial weighted empirical risk minimization (AWRM). AWRM introduces adversarial policies, of which the corresponding counterfactual dataset has the maximal prediction error of the model. For each iteration, the model is learned to be as small prediction risks as possible under the adversarial counterfactual dataset. However, the adversarial counterfactual dataset cannot be obtained in the offline setting, thus we derive an approximation of the counterfactual data distribution queried by the optimal adversarial policy and use a variational representation to give a tractable solution to learn a model from the approximated data distribution. As a result, we derive a practical approach named Generative Adversarial offLIne counterfactuaL Environment mOdel learning (GALILEO) for AWRM. Fig. 2 shows the difference in the prediction errors learned by these algorithms. We also discover that GALILEO is closely related to existing generative-adversarial model learning techniques, explaining the effectiveness of the latter. + +![](images/e21350e3d9b064285fe113e1adaa5132ea73dc8398bcf24bc5da581bdb24fdad.jpg) +Figure 2: An illustration of the prediction error in counterfactual datasets. The prediction risks is measured with mean square error (MSE). The error of SL is small only in training data ( $\phi =$ 62.5) but becomes much larger in the dataset “far away from” the training data. AWRM-oracle selects the oracle worst counterfactual dataset for training for each iteration (pseudocode is in Alg. 1) which reaches small MSE in all datasets and gives correct response curves (Fig. 1(c)). GALILEO approximates the optimal adversarial counterfactual data distribution based on the training data and model. Although the MSE of GALILEO is a bit larger than SL in the training data, in the counterfactual datasets, the MSE is on the same scale as AWRM-oracle. + +Experiments are conducted in two synthetic and two realistic environments. The results in the synthetic environments show that GALILEO can reconstruct correct responses for counterfactual queries. The evaluation results in two realistic environments also demonstrate that GALILEO has better ability in counterfactual query compared with baselines. We finally search for a policy based on the learned model in a real-world online platform. The policy significantly improves performance in concerned business indicators. + +# 2 RELATED WORK + +We give related adversarial algorithms for model learning in the following and leave other related work in Appx. F. GANTIE (Yoon et al., 2018) uses a generator to fill counterfactual outcomes for each data pair and a discriminator to judge the source (treatment group or control group) of the filled data pair. The generator is trained to minimize the output of the discriminator. GANITE is trained until the discriminator cannot determine which of the components is the factual outcome. Bica et al. (2020) propose SCIGAN to extend GANITE to continuous treatment effect estimation (a.k.a., dosage-response estimation) via a hierarchical discriminator architecture. In real-world applications, environment model learning based on Generative Adversarial Imitation Learning (GAIL) has also been adopted for sequential decision-making problems (Ho & Ermon, 2016). GAIL is first proposed for policy imitation (Ho & Ermon, 2016), which uses the imitated policy to generate trajectories by interacting with the environment. The policy is learned with the trajectories through RL which maximizes the cumulative rewards given by the discriminator. Shi et al. (2019); Chen et al. (2019); Shang et al. (2019) use GAIL for environment model learning by regarding the environment model as the generator and the behavior policy as the “environment” in standard GAIL. These studies empirically demonstrate that adversarial model learning algorithms have better generalization ability for counterfactual queries, while our study reveals the connection between adversarial model learning and the WERM through IPS. Our derived practical algorithm GALILEO is closely related to the existing adversarial model learning algorithms, explaining the effectiveness of the latter. + +# 3 PRELIMINARIES + +# 3.1 SINGLE-STEP INDIVIDUALIZED TREATMENT EFFECTS ESTIMATION AND WEIGHTEDEMPIRICAL RISKS MINIMIZATION + +We first introduce individualized treatment effects (ITEs) estimation (Rosenbaum & Rubin, 1983), which can be regarded as the scenario in which the environment model has only a single step. ITEs are typically defined as $I T E ( x ) : = \mathbb { E } [ M ^ { * } ( y | x , 1 ) | A = 1 , X = x ] - \mathbb { E } [ M ^ { * } ( y | x , 0 ) | A = 0 , X = x ]$ where $y$ is the feedback of the environment $M ^ { * } ( y | x , a )$ , $X$ denotes the state vector containing pre-treatment covariates (such as age and weight), $A$ denotes the treatment variable which is the action intervening to the state $X$ , and $A$ should be sampled from a uniform policy. In the twotreatment scenario, $A$ is in $\{ 0 , 1 \}$ where 1 is the action to intervene and 0 is the action to do nothing. A correct ITEs estimation should be done in Randomized Controlled Trials (RCT) in which we have the same probability of samples of $A = 1$ and $A = 0$ for each $X$ . Here we use lowercase $x , a$ and $y$ to denote samples of random variables $X , A$ and $Y$ , and use $x , A$ and $\mathcal { V }$ to denote space of the samples. In practice, we prefer to estimate ITEs under observational studies. In observational studies, datasets are pre-collected from the real world by a behavior policy such as a human-expert policy. In this case, a common approach for estimating the ITEs can be $\hat { I T E } ( x _ { i } ) = a _ { i } ( y _ { i } ^ { F } - M ( x _ { i } , 1 - a _ { i } ) ) + ( 1 - a _ { i } ) ( M ( x _ { i } , 1 - a _ { i } ) - y _ { i } ^ { F } )$ (Shalit et al., 2017) in deterministic prediction, where $x _ { i }$ and $y _ { i } ^ { F }$ denote the covariate and factual feedback of the $i$ -th sample, and $M \in \mathcal { M }$ denotes an approximated feedback model. $\mathcal { M }$ is the space of the model. In this formulation, the training set is an empirical factual data distribution $\bar { P _ { F } } = \{ ( x _ { i } , a _ { i } ) \} _ { i } ^ { n }$ and the testing set is an empirical counterfactual data distribution $P _ { C F } = \{ ( x _ { i } , 1 - a _ { i } ) \} _ { i } ^ { n }$ . If $a$ does not sample from a discrete uniform policy, i.e., the policy has selection bias, $P _ { F }$ and $P _ { C F }$ will be two different distributions, which violate the i.i.d. assumption of standard supervised learning. In stochastic prediction, $\hat { I T E } ( x ) = \mathbb { E } [ M ( y | x , 1 ) ] - \mathbb { E } [ M ( y | x , 0 ) ]$ and the counterfactual distribution for testing is the dataset with action sampling from a uniform policy. + +Generally speaking, in ITEs estimation, the risks of queries under counterfactual data are caused by the gap between the policy in training and testing data distributions. Without further processing, minimizing the empirical risks cannot guarantee the counterfactual-query risks being minimized. Assuming that the policy in training data $\mu$ satisfies $\mu ( a | x ) > 0 , \forall a \stackrel { - } { \in } \mathcal { \bar { A } } , \forall x \in \mathcal { X }$ (often named overlap assumption), a classical solution to handle the above problem is weighted empirical risk minimization (WERM) through an inverse propensity scoring (IPS) term $\omega$ (Shimodaira, 2000; Assaad et al., 2021; Hassanpour & Greiner, 2019; Jung et al., 2020): + +Definition 3.1. The learning objective of WERM through IPS is formulated as + +$$ +\begin{array} { r l } { \underset { M \in \mathcal { M } } { \operatorname* { m i n } } L ( M ) = \underset { M \in \mathcal { M } } { \operatorname* { m i n } } \mathbb { E } _ { { x } , { a } , { y } \sim { p } _ { M ^ { * } } } [ \omega ( { x } , { a } ) \ell ( M ( { y } | { x } , { a } ) , { y } ) ] , } & { } \\ { { s . t . } \quad \omega ( { x } , { a } ) = \frac { \beta ( a | { x } ) } { \mu ( a | { x } ) } , } & { } \end{array} +$$ + +where $\beta$ and $\mu$ denote the policies in testing and training domains, and $p _ { M ^ { * } } ^ { \mu }$ is the joint probability $p _ { M ^ { * } } ^ { \mu } ( x , a , y ) : = \rho _ { 0 } ( x ) \mu ( a | x ) M ^ { * } ( y | x , a )$ in which $\rho _ { 0 } ( x )$ is the distribution of state. $\mathcal { M }$ is the model space. $\ell$ is a loss function for model learning. + +The $\omega$ is also known as importance sampling (IS) weight, which corrects the sampling bias. In this objective, $\omega$ is to align the training data distribution to the testing data. By selecting different $\hat { \omega }$ to approximate $\omega$ to learn the model $M$ , current environment model learning algorithms for ITEs estimation are fallen into the framework. In standard supervised learning and some works for ITEs estimation (Wager & Athey, 2018; Weiss et al., 2015), $\hat { \omega } ( x , a ) = 1$ as the distribution-shift problem is ignored. In Shimodaira (2000); Assaad et al. (2021); Hassanpour & Greiner (2019), $\begin{array} { r } { \omega = \frac { \mathrm { i } } { \hat { \mu } } } \end{array}$ (i.e., $\beta$ a uniform policy) for balancing treatment and control group, where $\hat { \mu }$ is an approximation of behavior policy $\mu$ . Note that it is a reasonable weight in ITEs estimation: ITEs are defined to evaluate the effect of each state between treatment and control behavior under a uniform policy. + +# 3.2 SEQUENTIAL DECISION-MAKING SETTING + +Decision-making processes in a sequential environment are often formulated into Markov Decision Process (MDP) (Sutton & Barto, 1998). MDP depicts an agent interacting with the environment through actions. In the first step, states are sampled from an initial state distribution $x _ { 0 } \sim \rho _ { 0 } ( x )$ . Then at each time-step $t \in \{ 0 , 1 , 2 , \ldots \}$ , the agent takes an action $a _ { t } \in { \mathcal { A } }$ through a policy $\pi ( a _ { t } | x _ { t } ) \in \Pi$ based on the state $x _ { t } \in \mathcal { X }$ , then the agent receives a reward $r _ { t }$ from a reward function $r ( x _ { t } , a _ { t } ) \in$ $\mathbb { R }$ and transits to the next state $x _ { t + 1 }$ given by a transition function $M ^ { \ast } ( x _ { t + 1 } | x _ { t } , a _ { t } )$ built in the environment. $\Pi$ , $\mathcal { X }$ , and $\mathcal { A }$ denote the policy, state, and action spaces. + +# 4 METHOD + +In this section, we first propose a new offline model-learning objective based on Def. 3.1 for policy optimization tasks in Sec. 4.1; In Sec. 4.2, we derive a tractable solution to the proposed objective; Finally, we give a practical implementation in Sec. 4.3. + +# 4.1 PROBLEM FORMULATION + +For offline policy optimization, we require the environment model to have generalization ability in counterfactual queries since we need to query numerous different policies’ correct feedback from $M$ . Referring to the formulation of WERM through IPS in Def. 3.1, policy optimization requires $M$ to minimize counterfactual-query risks under numerous unknown different policies rather than a specific target policy $\beta$ . More specifically, the question is: If $\beta$ is unknown and can be varied, how should we reduce the risks in counterfactual queries? In this article, we call the model learning problem in this setting “counterfactual environment model learning” and propose a new objective to handle the problem. To be compatible with multi-step environment model learning, we first define a generalized WERM through IPS based on Def. 3.1. + +Definition 4.1. Given the MDP transition function $M ^ { * }$ that satisfies $M ^ { * } ( x ^ { \prime } | x , a ) > 0 , \forall x \in \mathcal { X } , \forall a \in$ $\mathcal { A } , \forall x ^ { \prime } \in \mathcal { X }$ and $\mu$ satisfies $\mu ( a | x ) > 0 , \forall a \in \mathcal { A } , \forall x \in \mathcal { X }$ , the learning objective of generalized WERM through IPS is formulated as + +$$ +\begin{array} { r l } & { \underset { M \in \mathcal { M } } { \operatorname* { m i n } } L ( M ) = \underset { M \in \mathcal { M } } { \operatorname* { m i n } } \mathbb { E } _ { x , a , x ^ { \prime } \sim \rho _ { M ^ { * } } ^ { \mu } } [ \omega ( x , a , x ^ { \prime } ) \ell _ { M } ( x , a , x ^ { \prime } ) ] , } \\ & { ~ s . t . ~ \omega ( x , a , x ^ { \prime } ) = \frac { \rho _ { M ^ { * } } ^ { \beta } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \mu } ( x , a , x ^ { \prime } ) } , } \end{array} +$$ + +where $\rho _ { M ^ { * } } ^ { \mu }$ is the training data distribution (collected by policy $\mu$ ), $\rho _ { M ^ { * } } ^ { \beta }$ is the testing data distribution (collected by policy $\beta$ ). We define $\ell _ { M } ( x , a , x ^ { \prime } ) : = \ell ( M ( x ^ { \prime } | x , a ) , x ^ { \prime } ) ^ { - }$ for brevity. + +In an MDP, given any policy $\pi$ , $\rho _ { M ^ { * } } ^ { \pi } ( x , a , x ^ { \prime } ) ~ = ~ \rho _ { M ^ { * } } ^ { \pi } ( x ) \pi ( a | x ) M ^ { * } ( x ^ { \prime } | x , a )$ where $\rho _ { M ^ { \ast } } ^ { \pi } ( x )$ denotes the occupancy measure of $x$ for policy $\pi$ , which can be defined as $\rho _ { M ^ { \ast } } ^ { \pi } ( x ) : =$ $\begin{array} { r } { ( 1 - \gamma ) \mathbb { E } _ { x _ { 0 } \sim \rho _ { 0 } } \left[ \sum _ { t = 0 } ^ { \infty } \dot { \gamma } ^ { t } \mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \right] } \end{array}$ (Sutton $\&$ Barto, 1998; Ho & Ermon, 2016) where $ { \mathrm { P r } } ^ { \pi } \left[ x _ { t } = x | x _ { 0 } , M ^ { * } \right]$ is the state visitation probability that the policy $\pi$ visits $x$ at time-step $t$ by executing in the environment $M ^ { * }$ and starting at the state $x _ { 0 }$ , and $\gamma \in [ 0 , 1 ]$ is the discount factor. Here we also define $\rho _ { M ^ { * } } ^ { \pi } ( x , a ) : = \rho _ { M ^ { * } } ^ { \pi } ( x ) \pi ( a | x )$ for simplicity. In this definition, $\omega$ can be rewritten as: $\begin{array} { r } { \omega ( x , a , x ^ { \prime } ) = \frac { \rho _ { M ^ { * } } ^ { \beta } ( x ) \beta ( a | x ) M ^ { * } ( x ^ { \prime } | x , a ) } { \rho _ { M ^ { * } } ^ { \mu } ( x ) \mu ( a | x ) M ^ { * } ( x ^ { \prime } | x , a ) } = \frac { \rho _ { M ^ { * } } ^ { \beta } ( x , a ) } { \rho _ { M ^ { * } } ^ { \mu } ( x , a ) } } \end{array}$ . In single-step environments, for any policy $\pi$ , ρπM∗ (x) = ρ0(x). Then we have ω(x, a, x′) = ρ0(x)β(a|x)ρ (x)µ(a|x) , and the objective is degraded to Eq. 1. Therefore, Def. 3.1 is a special case of this generalized form. + +Since $\beta$ for counterfactual queries is unknown and can be varied in policy optimization, to reduce the risks in counterfactual queries in this scenario, we introduce adversarial policies which can induce the worst performance of the model predictions and propose to optimize WERM under the adversarial policies. In particular, we propose Adversarial Weighted empirical Risk Minimization (AWRM) based on Def. 4.1 to handle this problem. + +Definition 4.2. Given the MDP transition function $M ^ { * }$ , the learning objective of adversarial Weighted empirical risk minimization through IPS is formulated as + +$$ +\begin{array} { r l } & { \hat { \boldsymbol M } ^ { * } = \underset { M \in \mathcal { M } } { \mathrm { \operatorname* { m i n } } } \underset { \beta \in \Pi } { \mathrm { m a x } } \boldsymbol L ( \rho _ { M ^ { * } } ^ { \beta } , M ) = \underset { M \in \mathcal { M } } { \mathrm { m i n } } \underset { \beta \in \Pi } { \mathrm { m a x } } \mathbb { E } _ { \boldsymbol x , \boldsymbol a , \boldsymbol x ^ { \prime } \sim \boldsymbol \rho _ { M ^ { * } } ^ { \mu } } [ \omega ( \boldsymbol x , \boldsymbol a | \rho _ { M ^ { * } } ^ { \beta } ) \ell _ { M } ( \boldsymbol x , \boldsymbol a , \boldsymbol x ^ { \prime } ) ] , } \\ & { } \\ & { s . t . \quad \omega ( \boldsymbol x , \boldsymbol a | \rho _ { M ^ { * } } ^ { \beta } ) = \frac { \rho _ { M ^ { * } } ^ { \beta } ( \boldsymbol x , \boldsymbol a ) } { \rho _ { M ^ { * } } ^ { \mu } ( \boldsymbol x , \boldsymbol a ) } , } \end{array} +$$ + +where the re-weighting term $\omega ( x , a | \rho _ { M ^ { * } } ^ { \beta } )$ is conditioned on the data distribution $\rho _ { M ^ { * } } ^ { \beta }$ of the adversarial policy $\beta$ . In the following, we will ignore $\rho _ { M ^ { * } } ^ { \beta }$ and use $\omega ( x , a )$ for brevity. + +According to the definition of MDP, $\omega ( x , a , x ^ { \prime } ) = \omega ( x , a )$ since the transition probability in the ratio can be canceled. Eq. 3 minimizes the maximum model loss under all counterfactual data distributions $\rho _ { M ^ { \ast } } ^ { \beta } ( x , a , x ^ { \prime } ) , \beta \in \Pi$ to guarantee the generalization ability for counterfactual data queried by policies in $\Pi$ . + +# 4.2 TRACTABLE AWRM SOLUTION + +In this section, we propose a tractable solution to optimize Eq. 3. The full derivations can be found in the Appx. A. We choose $\ell _ { M }$ to be the negative log-likelihood loss and derive the solution to Eq. 3: + +$$ +L ( \rho _ { M ^ { \ast } } ^ { \beta } , M ) = \mathbb { E } _ { x , a \sim \rho _ { M ^ { \ast } } ^ { \mu } } \left[ \omega ( x , a | \rho _ { M ^ { \ast } } ^ { \beta } ) \mathbb { E } _ { M ^ { \ast } } \left( - \log M ( x ^ { \prime } | x , a ) \right) \right] , +$$ + +where $\mathbb { E } _ { M ^ { * } } \left[ \cdot \right]$ denotes $\mathbb { E } _ { x ^ { \prime } \sim M ^ { * } \left( x ^ { \prime } \mid x , a \right) }$ [·]. The core problem is how to construct the data distribution $\rho _ { M ^ { * } } ^ { \beta ^ { * } }$ of the best-response policy $\beta ^ { * }$ in $M ^ { * }$ as it is costly to get extra data from $M ^ { * }$ in real-world applications. Instead of deriving the optimal $\beta ^ { * }$ , our solution is to offline estimate the optimal adversarial distribution $\rho _ { M ^ { * } } ^ { \beta ^ { * } }$ with respect to $M$ , then we can construct a surrogate objective to optimize $M$ without directly querying the real environment $M ^ { * }$ . + +# 4.2.1 OPTIMAL ADVERSARIAL DATA DISTRIBUTION APPROXIMATION + +Ideally, given any $M$ , it is obvious that the optimal $\beta$ is the one that makes $\rho _ { M ^ { \ast } } ^ { \beta } ( x , a )$ assign all densities to the point that has the largest negative log-likelihood. However, searching for the maximum is impractical, especially in continuous space. To give a relaxed but tractable solution, we add an $L _ { 2 }$ regularizer to the original objective Eq. 3: + +$$ +\operatorname* { m i n } _ { d \in M } \operatorname* { m a x } _ { \beta \in \Pi } \bar { L } ( \rho _ { M ^ { * } } ^ { \beta } , M ) = \operatorname* { m i n } _ { M \in M } \operatorname* { m a x } _ { \beta \in \Pi } \mathbb { E } _ { x , a \sim \rho _ { M ^ { * } } ^ { \mu } } \Big [ \omega ( x , a ) \mathbb { E } _ { M ^ { * } } \big [ - \log M ( x ^ { \prime } | x , a ) \big ] \Big ] - \frac { \alpha } { 2 } \| \rho _ { M ^ { * } } ^ { \beta } ( \cdot , \cdot ) \| _ { 2 } ^ { 2 } . +$$ + +where $\alpha$ denotes the regularization coefficient of $\rho _ { M ^ { * } } ^ { \beta }$ and $\begin{array} { r } { \| \rho _ { M ^ { * } } ^ { \beta } ( \cdot , \cdot ) \| _ { 2 } ^ { 2 } = \int _ { \mathcal { X } , \mathcal { A } } ( \rho _ { M ^ { * } } ^ { \beta } ( x , a ) ) ^ { 2 } \mathrm { d } a \mathrm { d } x . } \end{array}$ Then we can approximate the optimal distribution $\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ via Lemma. 4.3. + +Lemma 4.3. Given any $M$ in $\bar { L } ( \rho _ { M ^ { * } } ^ { \beta } , M )$ , the distribution of the ideal best-response policy $\bar { \beta } ^ { * }$ satisfies: + +$$ +\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } ( x , a ) = \frac { 1 } { \alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) , +$$ + +where $D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) )$ is the Kullback-Leibler $( K L )$ divergence between $M ^ { * } ( \cdot | x , a )$ and $M ( \cdot | x , a )$ , $H _ { M ^ { \ast } } ( x , a )$ denotes the entropy of $M ^ { * } ( \cdot | x , a )$ , and $\alpha _ { M }$ is the regularization coefficient $\alpha$ in Eq. $^ { 4 }$ and also as a normalizer of Eq. 5. + +Intuitively, $\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ has larger densities on the data where the divergence between the approximation model and the real model (i.e., $D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) ) )$ is larger or the stochasticity of the real model (i.e., $H _ { M ^ { * } }$ ) is larger. However, the integral process of $D _ { K L }$ in Eq. 5 is intractable in the offline setting as it explicitly requires the conditional probability function of $M ^ { * }$ . Our solution to solve the problem is utilizing the offline dataset $\mathcal { D } _ { \mathrm { r e a l } }$ as the empirical joint distribution $\rho _ { M ^ { * } } ^ { \mu } ( x , a , x ^ { \prime } )$ and adopting practical techniques for distance estimation on two joint distributions, like GAN (Goodfellow et al., 2014; Nowozin et al., 2016), to approximate Eq. 5. To adopt that solution, we should first transform Eq. 5 into a form under joint distributions. Without loss of generality, we introduce an intermediary policy $\kappa$ , of which $\mu$ can be regarded as a specific instance. Then we have $M ( x ^ { \prime } | x , a ) =$ $\rho _ { M } ^ { \kappa } ( x , a , x ^ { \prime } ) / \rho _ { M } ^ { \kappa } ( x , a )$ for any $M$ if $\rho _ { M } ^ { \kappa } ( x , a ) > 0$ . Assuming $\forall x \in \mathcal { X } , \forall a \in \mathcal { A } , \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) > 0$ if $\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } ( x , a ) > 0$ , which will hold when $\kappa$ overlaps with $\mu$ , then Eq. 5 can transform to: + +$$ +\displaystyle \frac { 1 } { \alpha _ { 0 } ( x , a ) } \Bigg ( \int _ { \mathcal { X } } \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) \log \frac { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M } ^ { \kappa } ( x , a , x ^ { \prime } ) } \mathrm { d } x ^ { \prime } - \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) \left( \log \frac { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } { \rho _ { M } ^ { \kappa } ( x , a ) } - H _ { M ^ { * } } ( x , a ) \right) \Bigg ) , +$$ + +where $\alpha _ { 0 } ( x , a ) = \alpha _ { M } \rho _ { M ^ { * } } ^ { \kappa } ( x , a )$ . We notice that the form $\begin{array} { r } { \rho _ { M ^ { * } } ^ { \kappa } \log \frac { \rho _ { M ^ { * } } ^ { \kappa } } { \rho _ { M } ^ { \kappa } } } \end{array}$ ρκM∗ρκ is the integrated function in reverse $\mathrm { K L }$ divergence, which is an instance of $f$ function in $f$ -divergence (Ali & Silvey, 1966). Replacing that form with $f$ function, we obtain a generalized representation of $\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ : + +$$ +\tilde { \bar { \nu } } _ { M ^ { * } } ^ { * } : = \frac { 1 } { \alpha _ { 0 } ( x , a ) } \left( \int _ { \mathcal { X } } \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) f \left( \frac { \rho _ { M } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } \right) \mathrm { d } x ^ { \prime } - \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) \left( f \left( \frac { \rho _ { M } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } \right) - H _ { M ^ { * } } ( x , a ) \right) \right. +$$ + +to Reviewer bcJS + +where $f : \mathbb { R } _ { + } \to \mathbb { R }$ is a convex and lower semi-continuous (l.s.c.) function. $\bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ gives a generalized representation of the optimal adversarial data distribution to maximize the error of the model. Based on Eq. 6, we have a surrogate objective of AWRM which can avoid querying $M ^ { * }$ to construct $\rho _ { M ^ { * } } ^ { \beta ^ { * } }$ : + +Theorem 4.4. Let $\bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ as the data distribution of the best-response policy $\bar { \beta } ^ { * }$ in Eq. 4 under model $M _ { \theta }$ parameterized by $\theta$ , then we can find the optimal $\theta ^ { * }$ of $\begin{array} { r } { \operatorname* { m i n } _ { \theta } \operatorname* { m a x } _ { \beta \in \Pi } \bar { L } ( \rho _ { M ^ { * } } ^ { \beta } , M _ { \theta } ) } \end{array}$ (Eq. 4) via iteratively optimizing the objective $\begin{array} { r } { \theta _ { t + 1 } = \operatorname* { m i n } _ { \theta } \bar { L } ( \bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } , M _ { \theta } ) } \end{array}$ , where $\bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ is approximated via the last-iteration model $M _ { \theta _ { t } }$ . Based on Corollary A.7, we derive an upper bound objective for $\begin{array} { r } { \operatorname* { m i n } _ { \theta } \bar { L } ( \bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } , M _ { \theta } ) } \end{array}$ : + +$$ +\begin{array} { r } { \iota _ { t + 1 } = \underset { \theta } { \operatorname* { m i n } } \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \kappa } } \left[ \frac { - 1 } { \alpha _ { 0 } ( x , a ) } \log M _ { \theta } ( x ^ { \prime } | x , a ) \left( f \left( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } \right) - f \left( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } \right) + H _ { M ^ { * } } ( x , a ) \right) \right] , } \\ { W ( x , a , x ^ { \prime } ) } \end{array} +$$ + +where $\mathbb { E } _ { \rho _ { M ^ { * } } ^ { \kappa } }$ [·] denotes $\mathbb { E } _ { x , a , x ^ { \prime } \sim \rho _ { M ^ { * } } ^ { \kappa } }$ [·], $f$ is a l.s.c function satisfying $f ^ { \prime } ( x ) \leq 0 , \forall x \in \mathcal { X }$ , and $\alpha _ { 0 } ( x , a ) = \alpha _ { M _ { \theta _ { t } } } \rho _ { M ^ { * } } ^ { \kappa } ( x , a )$ . + +Thm. 4.4 approximately achieve AWRM by using $\kappa$ and a pseudo-reweighting module $W$ . $W$ assigns learning propensities for data points with larger differences between distributions $\rho _ { M _ { \theta _ { t } } } ^ { \kappa }$ and $\rho _ { M ^ { * } } ^ { \kappa }$ . By adjusting the weights, the learning process will exploit subtle errors in any data point, whatever how many proportions it contributes, to correct potential generalization errors on counterfactual data. + +# 4.2.2 TRACTABLE SOLUTION TO THM. 4.4 + +In Thm. 4.4, the term $\begin{array} { r } { f \big ( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } \big ) - f \big ( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } \big ) } \end{array}$ is still intractable. To solve the problem, first, we resort to the first-order approximation of $f$ . Given some $u \in ( 1 - \xi , 1 + \xi ) , \xi > 0$ , we have + +$$ +f ( u ) \approx f ( 1 ) + f ^ { \prime } ( u ) ( u - 1 ) , +$$ + +where $f ^ { \prime }$ is the first-order derivative of $f$ . By Taylor’s formula and the fact that $f ^ { \prime } ( u )$ of the generator function $f$ is bounded in $( 1 - \xi , 1 + \xi )$ , the approximation error is no more than $\mathcal { O } ( \xi ^ { 2 } )$ . Let $\begin{array} { r } { u = \frac { p ( x ) } { q ( x ) } } \end{array}$ in Eq. 7, the pattern $f ( { \frac { p ( x ) } { q ( x ) } } )$ in Thm. 4.4 can be converted to $\begin{array} { r } { f ^ { \prime } ( \frac { p ( x ) } { q ( x ) } ) ( \frac { p ( x ) } { q ( x ) } - 1 ) + f ( 1 ) } \end{array}$ . $\frac { p ( x ) } { q ( x ) }$ can be approximated by sampling from datasets and $f ^ { \prime } \big ( \frac { p ( x ) } { q ( x ) } \big )$ can be approximated by the corresponding variational representation $T _ { \varphi ^ { * } }$ according to Lemma A.9 (Nowozin et al., 2016) (see + +Appendix for details). Based on any specific $f$ -function, we can represent $T _ { \varphi }$ with a discriminator $D _ { \varphi }$ . However, Thm. 4.4 holds only when $\forall x \in { \mathcal { X } }$ , $f ^ { \prime } ( x ) \leq 0$ . It can be verified that the instance $f ( u ) = u \log u - ( u + 1 ) \log ( u + 1 )$ , $T _ { \varphi } ( u ) = \log D _ { \varphi } ( u )$ satisfy this condition (see Tab. 2 in the Appendix). We select that instance for the tractable solution. In summary, by computing $f ( \cdot )$ with first-order approximation (Eq. 7) and leveraging the variational representation $T _ { \varphi ^ { * } } = \log { \hat { D } _ { \varphi ^ { * } } }$ to approximate $f ^ { \prime } \Bigl ( \frac { { { \rho } _ { M _ { \theta _ { t } } } ^ { \kappa } } } { { { \rho } _ { M ^ { * } } ^ { \kappa } } } \Bigr )$ , we can optimize the surrogate objective in Thm. 4.4 via: + +$$ +\begin{array} { r l } { { \iota _ { + 1 } } = \mathop { \operatorname* { m a x } } _ { \theta } \mathbb { E } _ { \rho _ { M _ { \theta } } ^ { \times } } \left[ A _ { \varphi _ { 0 } ^ { * } , \varphi _ { 1 } ^ { * } } ( x , a , x ^ { \prime } ) \log M \theta ( x ^ { \prime } | x , a ) \right] + \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \times } } \left[ \left( H _ { M ^ { * } } ( x , a ) - A _ { \varphi _ { 0 } ^ { * } , \varphi _ { 1 } ^ { * } } ( x , a , x ^ { \prime } ) \right) \log M _ { \theta } ( x ^ { \prime } | x , a ) \right] , } & { } \\ { { s . t . } } & { \varphi _ { 0 } ^ { * } = \mathop { \operatorname* { a r g m a x } } _ { \varphi _ { 0 } ^ { * } } \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \times } } \left[ \log D _ { \varphi _ { 0 } } ( x , a , x ^ { \prime } ) \right] + \mathbb { E } _ { \rho _ { M _ { \theta } } ^ { \times } } \left[ \log ( 1 - D _ { \varphi _ { 0 } } ( x , a , x ^ { \prime } ) ) \right] } & { ( 8 ) } \\ & { \varphi _ { 1 } ^ { * } = \mathop { \operatorname { a r g m a x } } _ { \varphi _ { 1 } } \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \times } } \left[ \log D _ { \varphi _ { 1 } } ( x , a ) \right] + \mathbb { E } _ { \rho _ { M _ { \theta } } ^ { \times } } \left[ \log ( 1 - D _ { \varphi _ { 1 } } ( x , a ) ) \right] , } \end{array} +$$ + +where $A _ { \varphi _ { 0 } ^ { * } , \varphi _ { 1 } ^ { * } } ( x , a , x ^ { \prime } ) \ = \ \log D _ { \varphi _ { 0 } ^ { * } } ( x , a , x ^ { \prime } ) \ - \ \log D _ { \varphi _ { 1 } ^ { * } } ( x , a ) , \ \mathbb { E } _ { \rho _ { M } ^ { \kappa } } [ \cdot ]$ is a simplification of $\mathbb { E } _ { x , a , x ^ { \prime } \sim \rho _ { M } ^ { \kappa } } [ \cdot ]$ and $\varphi _ { 0 }$ and $\varphi _ { 1 }$ are the parameters of $T _ { \varphi _ { 0 } }$ and $T _ { \varphi _ { 1 } }$ . Based on Lemma A.9, we have $\begin{array} { r } { f ^ { \prime } ( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } ) \approx \log D _ { \varphi _ { 0 } ^ { * } } ( x , a , x ^ { \prime } ) } \end{array}$ and $f ^ { \prime } ( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } ) \approx \log D _ { \varphi _ { 1 } ^ { * } } ( x , a )$ . Note that in the process, we ignore the term $\alpha _ { 0 } ( x , a )$ for simplifying the objective. The discussion on the impacts of removing $\alpha _ { 0 } ( x , a )$ is left in App. B. In Eq. 8, the pseudo-reweighting module $W$ in Thm. 4.4 is split into two terms in the RHS of the equation. The first term is a generative adversarial training objective regarding $M _ { \theta }$ as the generator, while the second term is WERM through $H _ { M ^ { * } } - A _ { \varphi _ { 0 } ^ { * } , \varphi _ { 1 } ^ { * } }$ . + +# 4.3 PRACTICAL IMPLEMENTATION + +To give a practical implementation of the solution, extra two assumptions introduced in the process of modeling should be handled: + +First, the approximation of Eq. 7 holds only when $\textstyle { \frac { p ( x ) } { q ( x ) } }$ is close to 1, which might not be satisfied. To handle the problem, we inject a naive supervised learning loss and replace the second term of the objective Eq. 8 when the expectation of $D _ { \varphi _ { 0 } }$ in $\rho _ { M _ { \theta _ { t } } } ^ { \kappa }$ is far away from 0.5 $f ^ { \prime } ( 1 ) = \log 0 . 5 )$ ; + +Second, the overlap assumption of $\cdot$ . In practice, we need to use the real-world data to construct the distribution $\rho _ { M ^ { * } } ^ { \kappa }$ and the generative data to construct $\cdot$ . In the offline model-learning setting, we only have a real-world dataset $\cdot$ collected by the behavior policy $\cdot$ . We can learn a policy $\cdot$ via behavior cloning based on $\mathcal { D }$ (Pomerleau, 1991; Ho & Ermon, 2016) and let $\cdot$ be the policy $\kappa$ . Then we can regard $\mathcal { D }$ as the empirical data distribution of $\rho _ { M ^ { * } } ^ { \kappa }$ and the trajectories collected by $\hat { \mu }$ in the model $M _ { \theta _ { t } }$ as the empirical data distribution of $\rho _ { M _ { \theta _ { t } } } ^ { \kappa }$ . But the assumption $\forall x \in \mathcal { X } , \forall a \in \mathcal { A } , \mu ( a | x ) > 0$ might not be satisfied. To handle the problem, in behavior cloning, we model $\hat { \mu }$ with a Gaussian distribution and constrain the lower bound of the variance with a small value $\epsilon _ { \mu } > 0$ to keep the assumption holding. Besides, we add small Gaussian noises $\mathcal { N } ( 0 , \epsilon _ { D } )$ to the inputs of $D _ { \varphi }$ to handle the mismatch between $\rho _ { M ^ { * } } ^ { \mu }$ and $\rho _ { M ^ { * } } ^ { \hat { \mu } }$ due to $\epsilon _ { \mu }$ . + +In Eq. 8, $H _ { M ^ { * } }$ is unknown in advance. In practice, we use $H _ { M _ { \theta } }$ to estimate it. More specifically, the neural network of $M _ { \theta }$ can be modeled with a Gaussian distribution. The variance of the Gaussian distribution is modeled with global variables $\Sigma$ for each dimension of output. We estimate $H _ { M ^ { * } }$ with the closed-form solution of Gaussian entropy through $\Sigma$ . + +Based on the above techniques, we propose Generative Adversarial offLIne counterfactuaL Environment mOdel learning (GALILEO) for environment model learning. GALILEO can be adopted in both single-step and sequential environment model learning. The detailed implementation and comparison to previous adversarial methods are in Appx. E. + +# 5 EXPERIMENTS + +In this section, we first conduct experiments in synthetic environments (Bica et al., 2020) to verify GALILEO on counterfactual queries 1 and the compatibility of GALILEO in sequential and single-step environments. We select Mean Integrated Square Error MISE $=$ $\begin{array} { r } { \bar { \mathbb { E } } \left[ \int _ { \mathcal { A } } \left( M ^ { * } ( y | x , a ) - \bar { M } ( y | x , a ) \right) ^ { 2 } \mathrm { d } a \right] } \end{array}$ as the metric, which is a commonly used metric to measure the accuracy in counterfactual queries by considering the prediction errors in the whole action space. Then we analyze the benefits of the implementation techniques described in Sec. 4.3 and + +to Reviewer bcJS to Reviewer bcJS + +![](images/1b046d05ed7b7da16cd32af8ae1f8b8b0181d988d9e217a6214be2ede6d9f372.jpg) +Figure 3: Illustration of the performance in GNFC and TCGA. The grey bar denotes the standard error ( $\mathrm { \times 0 . 3 }$ for brevity) of 3 random seeds. + +the problems without them. Finally, we deploy GALILEO in two complex environments: MuJoCo in Gym (Todorov et al., 2012) and a real-world food-delivery platform to test the performance of GALILEO in difficult tasks. The algorithms compared are: (1) Supervised Learning (SL): using standard empirical risk minimization for model learning; (2) Inverse Propensity Weighting (IPW) (Spirtes, 2010): a standard implementation of WERM based IPS; (3) eStimating the effects of Continuous Interventions using GANs (SCIGAN) (Bica et al., 2020): an adversarial algorithms for model learning used for causal effect estimation, which can be roughly regarded as a partial implementation of GALILEO (Refer to Appx. E.2). We give a detailed description in Appx. G.2. + +# 5.1 TEST IN SYNTHETIC ENVIRONMENTS + +Test in sequential environments Since there does not exist a task specifically designed for sequential environment model learning with selection bias, we construct a task, General Negative Feedback Control (GNFC), which can represent a classic type of task with policies having selection bias. Fig. 1(a) is also an example of GNFC. We give detailed motivation, the effect of selection bias, and other details in Appx. G.1.1. We construct tasks on GNFC by adding behavior policies $\mu$ with different scales of uniform noise $U ( - e , e )$ with different probabilities $p$ . In particular, with $e \in \{ 1 . 0 , 0 . 2 , 0 . 0 5 \}$ and $p \in \{ 1 . 0 , 0 . 2 , 0 . 0 5 \}$ , we construct 9 tasks and name them with the format of $^ { \ast } \mathrm { e } ^ { \ast } \mathrm { \_ p ^ { \ast } } ^ { \ast }$ . For example, $\Theta ^ { 1 } \mathrm { \mathrm { - } } \Theta ^ { 0 } \cdot 2$ is the task with behavior policy injecting with $U ( - 1 , 1 )$ with 0.2 probability. The results of GNFC tasks are summarized in Fig. 3(a) and the detailed results can be found in Tab. 8. The results show that the property of the behavior policy (i.e., $e$ and $p$ ) dominates the generalization ability of the baseline algorithms. When $e = 0 . 0 5$ , almost all of the baselines fail and give a completely opposite response curve (see Fig. 4(a) and Appx. H.2). IPW still perform well when $0 . 2 \leq e \leq 1 . 0$ but fails when $e = 0 . 0 5 , p < = 0 . 2$ . We also found that SCIGAN can reach a better performance than other baselines when $e = 0 . 0 5 , p < = 0 . 2$ , but the results in other tasks are unstable. GALILEO is the only algorithm that is robust to the selection bias and outputs correct response curves in all of the tasks. Based on the experiment, we also indicate that the commonly used overlap assumption is unreasonable to a certain extent especially in real-world applications since it is impractical to inject noises into the whole action space. The problem of overlap assumption being violated should be taken into consideration otherwise the algorithm will be hard to use in practice if it is sensitive to the noise range. + +Test in single-step environments Previous experiments on counterfactual environment model learning are based on single-step semi-synthetic data simulation (Bica et al., 2020). Since GALILEO is compatible with single-step environment model learning, we select the same task named TCGA in Bica et al. (2020) to test GALILEO. Based on three synthetic response functions in TCGA, we construct 9 tasks by choosing different parameters of selection bias on $\mu$ which is constructed with beta distribution, and design a coefficient $c$ to control the selection bias of the beta distribution. We name the tasks with the format of “t?_bias?”. For example, t1_bias2 is the task with the first response functions and $c = 2$ . The detail of TCGA is in Appx. G.1.2. The results of TCGA tasks are summarized in Fig. 3(b) and the detailed results can be found in Tab. 9 in Appendix. We found the phenomenon in this experiment is similar to the one in GNFC, which demonstrates the compatibility of GALILEO to single-step environments. We also found that the results of IPW are unstable in this experiment. It might be because the behavior policy is modeled with beta distribution while the propensity score $\hat { \mu }$ is modeled with Gaussian distribution. Since IPW directly reweight loss function with $\stackrel { \mathrm { ~ \tiny ~ 1 ~ } } { \hat { \mu } }$ , the results are sensitive to the error on $\hat { \mu }$ . GALILEO also models $\hat { \mu }$ with Gaussian distribution but the results are more stable since GALILEO does not re-weight through $\hat { \mu }$ explicitly. + +![](images/2f2933e9556a0a707044a2d135038204ddd64c137dda2d4f04c33d29fb14eee6.jpg) +Figure 4: Illustration of the averaged response curves. + +Table 1: Results of policy performance directly optimized through standard SAC (Haarnoja et al., 2018) using the learned dynamics models and deployed in MuJoCo environments. MAX-RETURN is the policy performance of SAC in the MuJoCo environments, and “avg. norm.” is the averaged normalized return of the policies in the 9 tasks, where the returns are normalized to lie between 0 and 100, where a score of 0 corresponds to the worst policy, and 100 corresponds to MAX-RETURN. + +
TaskHopperWalker2dHalfCheetah|avg. norm.
HorizonH=10H=20H=40H=10H=20H=40H=10H=20H=40/
GALILEO13.0 ± 0.133.2 ± 0.153.5± 1.211.7±0.229.9 ± 0.361.2 ± 3.40.7±0.2-1.1 ± 0.2-14.2 ± 1.451.1
SL4.8±0.53.0±0.24.6±0.210.7±0.220.1±0.837.5±6.70.4±0.5-1.1± 0.6-13.2 ± 0.321.1
IPW5.9 ±0.74.1 ± 0.65.9±0.24.7 ± 1.12.8 ±3.914.5 ± 1.41.6± 0.20.5 ± 0.8-11.3 ± 0.919.7
SCIGAN12.7 ± 0.129.2 ± 0.646.2 ±5.28.4 ± 0.59.1 ± 1.71.0 ± 5.81.2 ± 0.3-0.3 ±1.0-11.4 ± 0.341.8
MAX-RETURN13.2 ±0.033.3±0.271.0±0.514.9 ± 1.360.7 ± 11.1221.1 ± 8.92.6±0.113.3 ± 1.149.1 ± 2.3100.0
+ +Response curve visualization We plot the averaged response curves which are constructed by equidistantly sampling action from the action space and averaging the feedback of the states in the dataset as the averaged response. Parts of the results in Fig. 4 (all curves can be seen in Appx. H.2). For those tasks where baselines fail in reconstructing response curves, GALILEO not only reaches a better MISE score but reconstructs almost exact responses. + +Ablation studies In Sec. 4.3, we introduce several techniques to develop a practical GALILEO algorithm. Based on task $\mathtt { e 0 } . 2 _ { - \mathtt { p } 0 . 0 5 }$ of GNFC, we give the ablation studies to investigate the effects of these techniques. As the main-body space is limited, we leave the results in Appx. H.3. + +# 5.2 TEST IN COMPLEX ENVIRONMENTS + +In MuJoCo tasks MuJoCo is a benchmark task in Gym (Todorov et al., 2012; Brockman et al., 2016) where we need to control a robot with specific dynamics to complete some tasks (e.g., standing or running). We select 3 environment from D4RL (Fu et al., 2020) to construct our model learning tasks. We compare it with a standard transition model learning algorithm used in the previous offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020), which is a variant of supervised learning. We name the method OFF-SL. Besides, we also implement IPW and SCIGAN as the baselines. In D4RL benchmark, only the “medium” tasks is collected with a fixed policy, i.e., the behavior policy is with 1/3 performance to the expert policy), which is most matching to our proposed problem. So we train models in datasets HalfCheetah-medium, Walker2d-medium, and Hopper-medium. We trained the models with the same gradient steps and saved the models. + +We first verify the generalization ability of the models by adopting them into offline model-based RL. Instead of designing sophisticated tricks to suppress policy exploration and learning in risky regions as current offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020) do, we just use the standard SAC algorithm Haarnoja et al. (2018) to exploit the models for policy learning to strictly verify the ability of the models. Unfortunately, we found that the compounding error will still be inevitably large in the 1,000-step rollout, which is the standard horizon in MuJoCo tasks, leading all models to fail to derive a reasonable policy. To better verify the effects of models on policy optimization, we learn and evaluate the policies with three smaller horizons: $H \in \{ 1 0 , 2 0 , 4 0 \}$ . + +The results are listed in Tab. 1. We first averaged the normalized return (refer to “avg. norm.”) under each task, and we can see that the policy obtained by GALILEO is significantly higher than other models (the improvements are $24 \%$ to $161 \%$ ). At the same time, we found that SCIGAN performed better in policy learning, while IPW performed similarly to SL. This is in line with our expectations, since IPW only considers the uniform policy as the target policy for debiasing, while policy optimization requires querying a wide variety of policies. Minimizing the prediction risks only under a uniform policy cannot yield a good environment model for policy optimization. Besides, in IPW, the cumulative effects of policy on the state distribution are ignored. On the other hand, SCIGAN, as a partial implementation of GALILEO (refer to Appx. E.2), also roughly achieves AWRM and considers the cumulative effects of policy on the state distribution, so its overall performance is better; In addition, we find that GALILEO achieves significant improvement in 6 of the 9 tasks. But in HalfCheetah, IPW works slightly better. However, compared with MAX-RETURN, it can be found that all methods fail to derive reasonable policies because their policies’ performances are far away from the optimal policy. By further visualizing the trajectories, we found that all the learned policies just keep the cheetah standing in the same place or even going backward. This phenomenon is also similar to the results in MOPO (Yu et al., 2020). In MOPO’s experiment in the medium datasets, the truncated-rollout horizon used in Walker and Hopper for policy training is set to 5, while HalfCheetah has to be set to the minimal value: 1. These phenomena indicate that HalfCheetah may still have unknown problems, resulting in the generalization bottleneck of the models. Besides, we also test the prediction error of the learned model in corresponding unseen “expert” and “medium-replay” datasets. The detailed results are in Appx. H.5. + +![](images/48ff8ddb26f31d38f842a347595907ec3933c9a964af1fd467fd13b1ca24d29a.jpg) +Figure 5: An illustration of the performance in BAT tasks. In Fig. 5(a) demonstrate the averaged response curves of the SL and GALILEO model in City A. It is expected to be monotonically increasing through our prior knowledge. In Fig. 5(b), the model with larger areas above the “random” line makes better predictions in randomized-controlled-trials data (Betlei et al., 2020). Fig. 5(c) shows the daily responses in the A/B test in City A. The complete results are in Appx. H.6. + +In a real-world platform We finally deploy GALILEO in a real-world large-scale food-delivery platform. The goal of the platform is to balance the demand from take-out food orders and the supply of delivery clerks, i.e., helping delivery clerks fulfill more orders by giving reasonable strategies. We focus on a Budget Allocation task to the Time period (BAT) in the platform (see Appx. G.1.3 for details). The goal of the BAT task is to handle the imbalance problem between the demanded orders from customers and the supply of delivery clerks in different time periods by allocating reasonable allowances to those time periods. The core challenge of the environment model learning in BAT tasks is similar to the challenge in Fig. 1. Specifically, the behavior policy in BAT tasks is a human-expert policy, which tends to increase the budget of allowance in the time periods with a lower supply of delivery clerks, otherwise tends to decrease the budget (Fig. 12 gives a real-data instance of this phenomenon). + +We first learn a model to predict the supply of delivery clerks (measured by fulfilled order amount) on given allowances. Although the SL model can efficiently fit the offline data, the tendency of the response curve is easily to be incorrect. As can be seen in Fig. 5(a), with a larger budget of allowance, the prediction of the supply is decreased in SL, which obviously goes against our prior knowledge. This is because, in the offline dataset, the corresponding supply will be smaller when the allowance is larger. It is conceivable that if we learn a policy through the model of SL, the optimal solution is canceling all of the allowances, which is obviously incorrect in practice. On the other hand, the tendency of GALILEO’s response is correct. Fig. 13 plots all the results in 6 cities. + +Second, we conduct randomized controlled trials (RCT) in one of the testing cities. Using the RCT samples, we can evaluate the generalization ability of the model predictions via Area Under the Uplift Curve (AUUC) (Betlei et al., 2020), which measure the correctness of the sort order of the model prediction in RCT samples. The AUUC further show that GALILEO gives a reasonable sort order on the supply prediction (see Fig. 5(b)) while the standard SL technique fails to complete this task. + +Finally, we search for the optimal policy via the cross-entropy method planner (Hafner et al., 2019) based on the learned model and deploy the policy in a real-world platform. The results of A/B test in City A is shown in Fig. 5(c). It can be seen that after the day of the A/B test, the treatment group (deploying our policy) significant improve the five-minute order-taken rate than the baseline policy (the same as the behavior policy). In summary, the policy improves the supply from 0.14 to 1.63 percentage points to the behavior policies in the 6 cities. The details of these results are in Appx. H.6. + +# 6 DISCUSSION AND FUTURE WORK + +In this work, we propose AWRM which handles the generalization challenges of the counterfactual environment model learning. By theoretical modeling, we give a tractable solution to handle AWRM and propose GALILEO. GALILEO is verified in synthetic environments, complex robot control tasks, and a real-world platform, and shows great generalization ability on counterfactual queries. + +Giving correct answers to counterfactual queries is important for policy learning. We hope the work can inspire researchers to develop more powerful tools for counterfactual environment model learning. The current limitation lies in: There are several simplifications in the theoretical modeling process (further discussion is in Appx. B), which can be modeled more elaborately . Besides, experiments on MuJoCo indicate that these tasks are still challenging to give correct predictions on counterfactual data. These should also be further investigated in future work. + +# REFERENCES + +Ahmed M. Alaa and Mihaela van der Schaar. Limits of estimating heterogeneous treatment effects: Guidelines for practical algorithm design. In Proceedings of the 35th International Conference on Machine Learning, volume 80, pp. 129–138, Stockholm, Sweden, 2018. + +Syed Mumtaz Ali and Samuel D Silvey. A general class of coefficients of divergence of one distribution from another. Journal of the Royal Statistical Society: Series B (Methodological), 28 (1):131–142, 1966. + +Serge Assaad, Shuxi Zeng, Chenyang Tao, Shounak Datta, Nikhil Mehta, Ricardo Henao, Fan Li, and Lawrence Carin. Counterfactual representation learning with balancing weights. In Arindam Banerjee and Kenji Fukumizu (eds.), The 24th International Conference on Artificial Intelligence and Statistics, volume 130 of Proceedings of Machine Learning Research, pp. 1972–1980, Virtual Event, 2021. PMLR. + +Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. Analysis of representations for domain adaptation. In Bernhard Schölkopf, John C. Platt, and Thomas Hofmann (eds.), Advances in Neural Information Processing Systems 19, pp. 137–144, British Columbia, Canada, 2006. MIT Press. + +Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wortman Vaughan. A theory of learning from different domains. Mach. Learn., 79(1-2):151–175, 2010. + +Artem Betlei, Eustache Diemert, and Massih-Reza Amini. Treatment targeting by AUUC maximization with generalization guarantees. CoRR, abs/2012.09897, 2020. + +Ioana Bica, James Jordon, and Mihaela van der Schaar. Estimating the effects of continuous-valued interventions using generative adversarial networks. In Advances in Neural Information Processing Systems 33, virtual event, 2020. + +Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. Openai gym. CoRR, abs/1606.01540, 2016. + +Lars Buesing, Theophane Weber, Yori Zwols, Nicolas Heess, Sébastien Racanière, Arthur Guez, and Jean-Baptiste Lespiau. Woulda, coulda, shoulda: Counterfactually-guided policy search. In $7 t h$ International Conference on Learning Representations, New Orleans, LA, 2019. + +Jonathon Byrd and Zachary Chase Lipton. What is the effect of importance weighting in deep learning? In Kamalika Chaudhuri and Ruslan Salakhutdinov (eds.), Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pp. 872–881, Long Beach, CA, 2019. PMLR. + +Xinshi Chen, Shuang Li, Hui Li, Shaohua Jiang, Yuan Qi, and Le Song. Generative adversarial user model for reinforcement learning based recommendation system. In Proceedings of the 36th International Conference on Machine Learning, pp. 1052–1061, Long Beach, CA, 2019. + +Xiong-Hui Chen, Yang Yu, Qingyang Li, Fan-Ming Luo, Zhiwei Tony Qin, Shang Wenjie, and Jieping Ye. Offline model-based adaptable policy learning. In Advances in Neural Information Processing Systems 34 (NeurIPS’21), Virtual Conference, 2021. + +Corinna Cortes, Yishay Mansour, and Mehryar Mohri. Learning bounds for importance weighting. In John D. Lafferty, Christopher K. I. Williams, John Shawe-Taylor, Richard S. Zemel, and Aron Culotta (eds.), Advances in Neural Information Processing Systems 23, pp. 442–450, British Columbia, Canada, 2010. Curran Associates, Inc. + +Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4RL: datasets for deep data-driven reinforcement learning. CoRR, abs/2004.07219, 2020. + +Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron C. Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in Neural Information Processing Systems 27, pp. 2672–2680, Montréal, Canada, 2014. + +Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft Actor-Critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. In Proceedings of the 35th International Conference on Machine Learning, pp. 1856–1865, Stockholmsmässan, Sweden, 2018. + +Danijar Hafner, Timothy P. Lillicrap, Ian Fischer, Ruben Villegas, David Ha, Honglak Lee, and James Davidson. Learning latent dynamics for planning from pixels. In Kamalika Chaudhuri and Ruslan Salakhutdinov (eds.), Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pp. 2555–2565, Long Beach, CA, 2019. PMLR. + +Negar Hassanpour and Russell Greiner. Counterfactual regression with importance sampling weights. In Proceedings of the 28th International Joint Conference on Artificial Intelligence, pp. 5880–5887, Macao, China, 2019. + +J.B. Hiriart-Urruty and C. Lemaréchal. Fundamentals of Convex Analysis. 2001. ISBN 9783642564680. + +Jonathan Ho and Stefano Ermon. Generative adversarial imitation learning. In Advances in Neural Information Processing Systems 29, pp. 4565–4573, Barcelona, Spain, 2016. + +Guido Imbens. The role of the propensity score in estimating dose-response functions. Econometrics eJournal, 1999. + +Edward L Ionides. Truncated importance sampling. Journal of Computational and Graphical Statistics, 17(2):295–311, 2008. + +Stephen James and Edward Johns. 3D simulation for robot arm control with deep Q-learning. CoRR, abs/1609.03759, 2016. + +Fredrik D Johansson, Nathan Kallus, Uri Shalit, and David Sontag. Learning weighted representations for generalization across designs. arXiv preprint arXiv:1802.08598, 2018. + +Yonghan Jung, Jin Tian, and Elias Bareinboim. Learning causal effects via weighted empirical risk minimization. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan, and Hsuan-Tien Lin (eds.), Advances in Neural Information Processing Systems 33, Virtual Event, 2020. + +Rahul Kidambi, Aravind Rajeswaran, Praneeth Netrapalli, and Thorsten Joachims. MOReL : Modelbased offline reinforcement learning. CoRR, abs/2005.05951, 2020. + +Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tutorial, review, and perspectives on open problems. CoRR, abs/2005.01643, 2020. + +Yao Liu, Adith Swaminathan, Alekh Agarwal, and Emma Brunskill. Provably good batch off-policy reinforcement learning without great exploration. In Advances in Neural Information Processing Systems 33, virtual event, 2020. + +XuanLong Nguyen, Martin J. Wainwright, and Michael I. Jordan. Estimating divergence functionals and the likelihood ratio by convex risk minimization. IEEE Trans. Inf. Theory, 56(11):5847–5861, 2010. + +Sebastian Nowozin, Botond Cseke, and Ryota Tomioka. f-GAN: Training generative neural samplers using variational divergence minimization. In Advances in Neural Information Processing Systems 29, pp. 271–279, Barcelona, Spain, 2016. + +Silviu Pitis, Elliot Creager, and Animesh Garg. Counterfactual data augmentation using locally factored dynamics. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan, and Hsuan-Tien Lin (eds.), Advances in Neural Information Processing Systems 33, virtual, 2020. + +Dean Pomerleau. Efficient training of artificial neural networks for autonomous navigation. Neural Computation, 3(1):88–97, 1991. + +Joaquin Quinonero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D Lawrence. Dataset shift in machine learning. Mit Press, 2008. + +Paul R. Rosenbaum and Donald B. Rubin. The central role of the propensity score in observational studies for causal effects. Biometrika, 70:41–55, 1983. + +John Schulman, Sergey Levine, Pieter Abbeel, Michael I. Jordan, and Philipp Moritz. Trust region policy optimization. In Proceedings of the 32nd International Conference on Machine Learning, volume 37, pp. 1889–1897, Lille, France, 2015. + +John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. CoRR, abs/1707.06347, 2017. + +Uri Shalit, Fredrik D. Johansson, and David A. Sontag. Estimating individual treatment effect: generalization bounds and algorithms. In Proceedings of the 34th International Conference on Machine Learning, volume 70, pp. 3076–3085, Sydney, Australia, 2017. + +Wenjie Shang, Yang Yu, Qingyang Li, Zhiwei (Tony) Qin, Yiping Meng, and Jieping Ye. Environment reconstruction with hidden confounders for reinforcement learning based recommendation. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 566–576, Anchorage, AK, 2019. + +Jing-Cheng Shi, Yang Yu, Qing Da, Shi-Yong Chen, and Anxiang Zeng. Virtual-taobao: Virtualizing real-world online retail environment for reinforcement learning. In The 33th AAAI Conference on Artificial Intelligence, pp. 4902–4909, Honolulu, Hawaii, 2019. + +Hidetoshi Shimodaira. Improving predictive inference under covariate shift by weighting the loglikelihood function. Journal of statistical planning and inference, 90(2):227–244, 2000. + +Sumedh A. Sontakke, Arash Mehrjou, Laurent Itti, and Bernhard Schölkopf. Causal curiosity: RL agents discovering self-supervised experiments for causal representation learning. In Marina Meila and Tong Zhang (eds.), Proceedings of the 38th International Conference on Machine Learning, volume 139 of Proceedings of Machine Learning Research, pp. 9848–9858, Virtual Event, 2021. PMLR. + +Peter Spirtes. Introduction to causal inference. J. Mach. Learn. Res., 11:1643–1662, aug 2010. ISSN 1532-4435. + +Richard S. Sutton and Andrew G. Barto. Reinforcement learning - an introduction. Adaptive computation and machine learning. MIT Press, 1998. + +Emanuel Todorov, Tom Erez, and Yuval Tassa. MuJoCo: A physics engine for model-based control. In Proceedings of the 24th IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 5026–5033, Vilamoura, Portugal, 2012. + +Stefan Wager and Susan Athey. Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association, 113(523):1228–1242, 2018. + +Jeremy C. Weiss, Finn Kuusisto, Kendrick Boyd, Jie Liu, and David Page. Machine learning for treatment assignment: Improving individualized risk attribution. In American Medical Informatics Association Annual Symposium, San Francisco, CA, 2015. + +Jinsung Yoon, James Jordon, and Mihaela van der Schaar. GANITE: estimation of individualized treatment effects using generative adversarial nets. In 6th International Conference on Learning Representations, 2018. + +Tianhe Yu, Garrett Thomas, Lantao Yu, Stefano Ermon, James Y. Zou, Sergey Levine, Chelsea Finn, and Tengyu Ma. MOPO: model-based offline policy optimization. In Advances in Neural Information Processing Systems 33, virtual event, 2020. + +# Appendix + +# Table of Contents + +# A Proof of Theoretical Results 14 + +A.1 Proof of Lemma 4.3 15 +A.2 Proof of Eq. 6 16 +A.3 Proof of Thm. 4.4 17 +A.4 Proof of the Tractable Solution 20 +B Discussion of the Theoretical Results 22 +C Societal Impact 22 +D AWRM-oracle Pseudocode 23 + +# E Implementation 23 + +E.1 Details of the GALILEO Implementation 23 +E.2 Connection with Previous Adversarial Algorithms 25 + +# F Additional Related Work 26 + +# Experiment Details 27 + +G.1 Settings 27 +G.2 Baseline Algorithms 29 +G.3 Hyper-parameters 30 +G.4 Computation Resources 30 + +# H Additional Results 31 + +H.1 All of the Result Table . 31 +H.2 Averaged Response Curves 31 +H.3 Ablation Studies 31 +H.4 Worst-Case Prediction Error 32 +H.5 Detailed Results in the MuJoCo Tasks 32 +H.6 Detailed Results in the BAT Task 34 + +# A PROOF OF THEORETICAL RESULTS + +The overall pipeline to model the tractable solution to AWRM is given in Fig. 6. + +In the proof section, we replace the notation of $\mathbb { E }$ with an integral for brevity. Now we rewrite the original objective $\bar { L } ( \rho _ { M ^ { * } } ^ { \beta } , \bar { M } )$ as: + +$$ +\operatorname* { m i n } _ { \substack { M \in M } } \operatorname* { m a x } _ { \beta \in \Pi } \int _ { \mathcal { X } , \mathcal { A } } \rho _ { M ^ { * } } ^ { \mu } ( x , a ) \omega ( x , a ) \int _ { \mathcal { X } } M ^ { * } ( x ^ { \prime } | x , a ) \left( - \log M ( x ^ { \prime } | x , a ) \right) \mathrm { d } x ^ { \prime } \mathrm { d } a \mathrm { d } x - \frac { \alpha } { 2 } \| \rho _ { M ^ { * } } ^ { \beta } ( \cdot , \cdot ) \| _ { 2 } ^ { 2 } , +$$ + +where $\begin{array} { r } { \omega ( x , a ) = \frac { \rho _ { M ^ { * } } ^ { \beta } ( x , a ) } { \rho _ { M ^ { * } } ^ { \mu } ( x , a ) } } \end{array}$ and $\begin{array} { r } { \| \rho _ { M ^ { * } } ^ { \beta } ( \cdot , \cdot ) \| _ { 2 } ^ { 2 } = \int _ { \mathcal { X } , \mathcal { A } } \rho _ { M ^ { * } } ^ { \beta } ( x , a ) ^ { 2 } \mathrm { d } a \mathrm { d } x } \end{array}$ , which is the squared $l _ { 2 }$ -norm. In an MDP, given any policy $\pi$ , $\rho _ { M ^ { * } } ^ { \pi } ( x , a , x ^ { \prime } ) = \rho _ { M ^ { * } } ^ { \pi } ( x ) \pi ( a | x ) M ^ { * } ( x ^ { \prime } | x , a )$ where $\rho _ { M ^ { * } } ^ { \pi } ( x )$ denotes the occupancy measure of $x$ for policy $\pi$ , which can be defined (Sutton $\&$ Barto, 1998; Ho & Ermon, 2016) as $\begin{array} { r } { \rho _ { M ^ { * } } ^ { \pi } ( x ) : = ( 1 - \gamma ) \mathbb { E } _ { x _ { 0 } \sim \rho _ { 0 } } \left[ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } \mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \right] } \end{array}$ where $ { \mathrm { P r } } ^ { \pi } \left[ x _ { t } = x | x _ { 0 } , M ^ { * } \right]$ is the state visitation probability that $\pi$ starts at state $x _ { 0 }$ in model $M ^ { * }$ and receive $x$ at timestep $t$ and $\gamma \in [ 0 , 1 ]$ is the discount factor. + +![](images/b8d8ca5df420de7f53155d64557ee7e5645c8c107cdf9e997b057385c9255a79.jpg) +Figure 6: The overall pipeline to model the tractable solution to AWRM. $f$ is a generator function defined by $f$ -divergence (Nowozin et al., 2016). $\kappa$ is an intermediary policy introduced in the estimation. + +# A.1 PROOF OF LEMMA 4.3 + +For better readability, we first rewrite Lemma 4.3 as follows: + +Lemma A.1. Given any $M$ in $\bar { L } ( \rho _ { M ^ { * } } ^ { \beta } , M )$ , the distribution of the ideal best-response policy $\bar { \beta } ^ { * }$ satisfies: + +$$ +\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } ( x , a ) = \frac { 1 } { \alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) , +$$ + +where $D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) )$ is the Kullback-Leibler $( K L )$ divergence between $M ^ { * } ( \cdot | x , a )$ and $M ( \cdot | x , a )$ , $H _ { M ^ { \ast } } ( x , a )$ denotes the entropy of $M ^ { * } ( \cdot | x , a )$ , where $D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) )$ is the Kullback-Leibler $( K L )$ divergence between $M ^ { * } ( \cdot | x , a )$ and $M ( \cdot | x , a )$ , $H _ { M ^ { \ast } } ( x , a )$ denotes the entropy of $M ^ { * } ( \cdot | x , a )$ , and $\alpha _ { M }$ is the regularization coefficient $\alpha$ in Eq. 9 and also as a normalizer. + +Proof. Given a transition function $M$ of an MDP, the distribution of the best-response policy $\beta ^ { * }$ satisfies: + +$$ +\begin{array} { r l } { \rho _ { \mu \mu ^ { * } } ^ { * } } & { = \operatorname* { s u g m a x } _ { \hat { \rho } _ { \nu \mu ^ { * } } ^ { 0 } , } \int _ { x , z } \rho _ { \mu \nu } ^ { * } ( z , \omega ) \omega ( x , \omega ) \int _ { x } d x ^ { * } ( x ^ { * } | x , a ) \left( - \log { \mathcal { A } ( x ^ { * } | x , a ) } \right) \mathrm { d } x ^ { * } \mathrm { d } \alpha \mathrm { d } x - \frac { \alpha } { 2 } \| \rho _ { \nu \mu } ^ { 2 } ( z , \omega ) } \\ & { = \operatorname* { s u g m a x } _ { \hat { \rho } _ { \nu \mu ^ { * } } ^ { 0 } , } \int _ { x , z } \rho _ { \mu \nu } ^ { * } ( z , \omega ) \int _ { x } \underline { { \mathcal { A } } } ^ { \mathcal { A } ^ { * } } ( x ^ { * } | x , a ) \left( - \log { \mathcal { A } ( x ^ { * } | x , a ) } \right) \mathrm { d } x ^ { * } \mathrm { d } \alpha \mathrm { d } x - \frac { \alpha } { 2 } \| \rho _ { \mu \nu } ^ { 3 } d \cdot ( x ^ { * } ) \| _ { 2 } ^ { 2 } } \\ & { = \operatorname* { s u g m a x } _ { \hat { \rho } _ { \nu \mu ^ { * } } ^ { 0 } , } \frac { 2 } { \alpha } \int _ { x , z } \rho _ { \mu \nu } ^ { * } ( z , \omega ) \| \phi ( x , a ) \mathrm { d } \alpha \mathrm { d } x - \| \rho _ { \mu \nu } ^ { 3 } d \cdot ( x ^ { * } ) \| _ { 2 } ^ { 2 } } \\ & { = \operatorname* { s u g m a x } _ { \hat { \rho } _ { \nu \mu ^ { * } } ^ { 0 } , } \alpha \int _ { x , z } \rho _ { \mu \nu } ^ { * } ( z , \omega ) g ( x , a ) \mathrm { d } \alpha \mathrm { d } x - \| \rho _ { \nu \mu ^ { * } } ^ { 3 } ( z , \nu ) \| _ { 2 } ^ { 2 } } \\ & = \operatorname* { m a x } _ { \hat { \rho } _ { \nu \mu ^ { * } } ^ { 0 } , } \alpha \int _ { x , z } \rho _ { \mu \nu } ^ { * } ( x , a ) g ( x , a ) \mathrm { d } \alpha \mathrm { d } x - \| \rho _ { \mu \nu } ^ { 3 } ( z , \omega ) \| _ { 2 } ^ { 2 } - \frac \end{array} +$$ + +We know that the occupancy measure $\rho _ { M ^ { * } } ^ { \beta }$ is a density function with a constraint $\begin{array} { r } { \int _ { \mathcal { X } } \int _ { A } \rho _ { M ^ { * } } ^ { \beta } ( x , a ) \mathrm { d } a \mathrm { d } x = 1 } \end{array}$ . Assuming the occupancy measure $\rho _ { M ^ { * } } ^ { \beta }$ has an upper bound $c$ , that is $0 ~ \leq ~ \rho _ { M ^ { * } } ^ { \beta } ( x , a ) ~ \leq ~ c , \forall a ~ \in ~ { \mathcal { A } } , \forall x ~ \in ~ \mathcal { X }$ , constructing a regularization coefficient $\alpha _ { M } =$ $\begin{array} { r } { \int _ { \mathcal { X } } \int _ { \mathcal { A } } ( D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) \mathrm { d } x \mathrm { d } a } \end{array}$ as a constant value given any $M$ , then we have + +$$ +\begin{array} { r l } & { \rho _ { M ^ { * } } ^ { \beta ^ { * } } ( x , a ) = \frac { g ( x , a ) } { \alpha _ { M } } } \\ & { \qquad = \frac { \int _ { \mathcal { X } } M ^ { * } ( x ^ { \prime } | x , a ) \log \frac { M ^ { * } ( x ^ { \prime } | x , a ) } { M ( x ^ { \prime } | x , a ) } \mathrm { d } x - \int _ { \mathcal { X } } M ^ { * } ( x ^ { \prime } | x , a ) \log M ^ { * } ( x ^ { \prime } | x , a ) \mathrm { d } x } { \alpha _ { M } } } \\ & { \qquad = \frac { D _ { K L } \left( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) \right) + H _ { M ^ { * } } ( x , a ) } { \alpha _ { M } } } \\ & { \qquad \propto \Big ( D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) \Big ) , } \end{array} +$$ + +which is the optimal density function of Eq. 9 with $\alpha = \alpha _ { M }$ . + +Note that in some particular $M ^ { * }$ , we still cannot construct a $\beta$ that can generate an occupancy specified by $g ( x , a ) / \alpha _ { M }$ for any $M$ . We can only claim the distribution of the ideal best-response policy $\bar { \beta } ^ { * }$ satisfies: + +$$ +\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } ( x , a ) = \frac { 1 } { \alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) , +$$ + +where $\alpha _ { M }$ is a normalizer that $\begin{array} { r } { \alpha _ { M } = \int _ { \mathcal { X } } \int _ { \mathcal { A } } ( D _ { K L } ( M ^ { * } ( \cdot | x , a ) , M ( \cdot | \underline { { x } } , a ) ) + H _ { M ^ { * } } ( x , a ) ) \mathrm { d } x \mathrm { d } a } \end{array}$ . We give a discussion of the rationality of the ideal best-response policy $\bar { \beta } ^ { * }$ as a replacement of the real best-response policy $\beta ^ { * }$ in Remark A.2. + +Remark A.2. The optimal solution Eq. 11 relies on $g ( x , a )$ . In some particular $M ^ { * }$ , it is intractable to derive a $\beta$ that can generate an occupancy specified by $g ( x , a ) / \alpha _ { M }$ . Consider the following case: a state $x _ { 1 }$ in $M ^ { * }$ might be harder to reach than another state $x _ { 2 }$ , e.g., $M ^ { \ast } ( x _ { 1 } | x , a ) \ <$ $M ^ { * } ( x _ { 2 } | x , a ) , \forall x \in \mathcal { X } , \forall a \in \mathcal { A }$ , then it is impossible to find a $\beta$ that the occupancy satisfies $\rho _ { M ^ { * } } ^ { \beta } ( x _ { 1 } , a ) > \rho _ { M ^ { * } } ^ { \beta } ( x _ { 2 } , a )$ . In this case, Eq. 11 can be a sub-optimal solution. Since this work focuses on task-agnostic solution derivation while the solution to the above problem should rely on the specific description of $M ^ { * }$ , we leave it as future work. However, we point out that Eq. 11 is $a$ reasonable re-weighting term even as a sub-optimum: $\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ gives larger densities on the data where the distribution distance between the approximation model and the real model (i.e., $D _ { K L } ( M ^ { * } , M ) )$ is larger or the stochasticity of the real model (i.e., $H _ { M ^ { * } }$ ) is larger. + +# A.2 PROOF OF EQ. 6 + +The integral process of $D _ { K L }$ in Eq. 5 is intractable in the offline setting as it explicitly requires the conditional probability function of $M ^ { * }$ . Our motivation for the tractable solution is utilizing the offline dataset $\mathcal { D } _ { \mathrm { r e a l } }$ as the empirical joint distribution $\rho _ { M ^ { * } } ^ { \mu } ( x , a , x ^ { \prime } )$ and adopting practical techniques for distance estimation on two joint distributions, like GAN (Goodfellow et al., 2014; Nowozin et al., 2016), to approximate Eq. 5. To adopt that solution, we should first transform Eq. 5 into a form under joint distributions. Without loss of generality, we introduce an intermediary policy $\kappa$ , of which $\mu$ can be regarded as a specific instance. Then we have $M ( x ^ { \prime } | x , a ) = \rho _ { M } ^ { \kappa } ( x , a , x ^ { \prime } ) / \rho _ { M } ^ { \kappa } ( x , a )$ for any $M$ if $\rho _ { M } ^ { \kappa } ( x , a ) > 0$ . Assuming $\forall x \in \mathcal { X } , \forall a \in A , \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) > 0$ if $\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } ( x , a ) > 0$ , which will hold when $\kappa$ overlaps with $\mu$ , then Eq. 5 can transform to: + +$$ +\begin{array} { r l } { \frac { d } { d t } } & { ( z , z , n ) = \frac { 1 } { n + 1 } ( 2 \frac { n + 1 } { n + 1 } , \frac { 1 } { n + 1 } ; \frac { 1 } { n + 2 } , z ) ! \log ( \log \frac { \mathcal { N } _ { \mathrm { e f f } } ( z , z , n ) } { \mathcal { N } _ { \mathrm { e f f } } ( z , z , n ) } ) \log ( \log \mathcal { N } _ { \mathrm { e f f } } ( z , z , n ) ) } \\ & { - \frac { 1 } { n + 1 } \int _ { x } d \gamma \exp ( z , z , n ) ( \log \mathcal { N } _ { \mathrm { e f f } } ( z , z , n ) ) \log \log ( \log \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) ) \mathrm { d } x ^ { \prime } } \\ & { = \frac { 1 } { n + 1 } \int _ { x } d \gamma \exp ( z , \log \mathcal { N } _ { \mathrm { e f f } } ( z , n ) ) \mathrm { d } x ^ { \prime } \exp ( z , \log \mathcal { N } _ { \mathrm { e f f } } ( z , n ) ) \exp ( z , \log \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) ) \mathrm { d } x ^ { \prime } } \\ & { = \frac { 1 } { n + 1 } \int _ { x } d \gamma \exp ( z , \log \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) ) \mathrm { d } x ^ { \prime } \exp ( \log \frac { \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) } { \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) } + \log \frac { \mathcal { N } _ { \mathrm { e f f } } ( z , z , n ) } { \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) } - \log \mathcal { N } _ { \mathrm { e f f } } ( z , z , n ) ) \mathrm { d } y ^ { \prime } } \\ & = \frac { 1 } { n + 1 } \int _ { x } d \gamma \exp ( z , \log \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) ) \exp ( \log \frac { \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) } { \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) } + \log \frac { \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) } { \mathcal { N } _ { \mathrm { e f f } } ( z , n , z ) } - \log \mathcal { N } _ { \mathrm { e f f } } ( z , n , \end{array} +$$ + +where $\alpha _ { 0 } ( x , a ) = \alpha _ { M } \rho _ { M ^ { \ast } } ^ { \kappa } ( x , a )$ + +Definition A.3 ( $f$ -divergence). Given two distributions $P$ and $Q$ , two absolutely continuous density functions $p$ and $q$ with respect to a base measure dx defined on the domain $\mathcal { X }$ , we define the $f$ -divergence (Nowozin et al., 2016), + +$$ +D _ { f } ( P \| Q ) = \int _ { \mathcal X } q ( x ) f \left( \frac { p ( x ) } { q ( x ) } \right) \mathrm { d } x , +$$ + +where the generator function $f : \mathbb { R } _ { + } \to \mathbb { R }$ is a convex, lower-semicontinuous function. + +We notice that the terms $\begin{array} { r } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) \log \frac { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M } ^ { \kappa } ( x , a , x ^ { \prime } ) } } \end{array}$ and $\begin{array} { r } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) \log \frac { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } { \rho _ { M } ^ { \kappa } ( x , a ) } } \end{array}$ are the integrated functions in reverse KL divergence, which is an instance of $f$ function in $f$ -divergence (See ReverseKL divergence of Tab.1 in (Nowozin et al., 2016) for more details). Replacing that form $q \log { \frac { q } { p } }$ with $q f ( { \underline { { p } } } )$ , we obtain a generalized representation of $\rho _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ : + +$$ +\tilde { \bar { \nu } } _ { M ^ { * } } ^ { * } : = \frac { 1 } { \alpha _ { 0 } ( x , a ) } \Bigg ( \int _ { \mathcal { X } } \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) f \left( \frac { \rho _ { M } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } \right) \mathrm { d } x ^ { \prime } - \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) \left( f \left( \frac { \rho _ { M } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } \right) - H _ { M ^ { * } } ( x , a ) \right) +$$ + +# A.3 PROOF OF THM. 4.4 + +We first introduce several useful lemmas for the proof. + +Lemma A.4. Rearrangement inequality The rearrangement inequality states that, for two sequences $a _ { 1 } \geq a _ { 2 } \geq . . . \geq a _ { n }$ and $b _ { 1 } \geq b _ { 2 } \geq . . . \geq b _ { n }$ , the inequalities + +$$ +a _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \cdot \cdot \cdot + a _ { n } b _ { n } \geq a _ { 1 } b _ { \pi ( 1 ) } + a _ { 2 } b _ { \pi ( 2 ) } + \cdot \cdot \cdot + a _ { n } b _ { \pi ( n ) } \geq a _ { 1 } b _ { n } + a _ { 2 } b _ { n - 1 } + \cdot \cdot \cdot + a _ { n } b _ { 1 } +$$ + +hold, where $\pi ( 1 ) , \pi ( 2 ) , \ldots , \pi ( n )$ is any permutation of $1 , 2 , \ldots , n$ . + +Lemma A.5. For two sequences $a _ { 1 } \geq a _ { 2 } \geq . . . \geq a _ { n }$ and $b _ { 1 } \geq b _ { 2 } \geq . . . \geq b _ { n }$ , the inequalities + +$$ +\sum _ { i = 1 } ^ { n } { \frac { 1 } { n } } a _ { i } b _ { i } \geq \sum _ { i = 1 } ^ { n } { \frac { 1 } { n } } a _ { i } \sum { \frac { 1 } { n } } b _ { i } +$$ + +hold. + +Proof. By rearrangement inequality, we have + +$$ +\begin{array} { l } { { \displaystyle \sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \geq a _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \cdots + a _ { n } b _ { n } } } \\ { { \displaystyle \sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \geq a _ { 1 } b _ { 2 } + a _ { 2 } b _ { 3 } + \cdots + a _ { n } b _ { 1 } } } \\ { { \displaystyle \sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \cdots + a _ { n } b _ { 2 } } } \\ { { \displaystyle \sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \cdots + a _ { n } b _ { 2 } } } \\ { { \displaystyle \qquad \vdots \qquad } } \\ { { \displaystyle \sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \geq a _ { 1 } b _ { n } + a _ { 2 } b _ { 1 } + \cdots + a _ { n } b _ { n - 1 } } } \end{array} +$$ + +Then we have + +$$ +\begin{array} { l } { { \displaystyle n \sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \geq \sum _ { i = 1 } ^ { n } a _ { i } \sum _ { i = 1 } ^ { n } b _ { i } } } \\ { { \displaystyle \sum _ { i = 1 } ^ { n } \frac { 1 } { n } a _ { i } b _ { i } \geq \sum _ { i = 1 } ^ { n } \frac { 1 } { n } a _ { i } \sum _ { n } \frac { 1 } { n } b _ { i } } } \end{array} +$$ + +Now we extend Lemma A.5 into the continuous integral scenario: + +Lemma A.6. Given $\mathcal { X } \subset \mathbb { R } ,$ , for two functions $f : \mathcal { X } \to \mathbb { R }$ and $g : \mathcal { X } \mathbb { R }$ that $f \left( x \right) \geq f \left( y \right)$ if and only if $g ( x ) \geq g ( y )$ , $\forall x , y \in { \mathcal { X } }$ , the inequality + +$$ +\int _ { \mathcal { X } } p ( x ) f ( x ) g ( x ) \mathrm { d } x \geq \int _ { \mathcal { X } } p ( x ) f ( x ) \mathrm { d } x \int _ { \mathcal { X } } p ( x ) g ( x ) \mathrm { d } x +$$ + +holds, where $p : \mathcal { X } \mathbb { R }$ and $p ( x ) > 0 , \forall x \in \mathcal { X }$ and $\begin{array} { r } { \int _ { \mathcal { X } } p ( x ) \mathrm { d } x = 1 } \end{array}$ . + +Proof. Since $( f ( x ) - f ( y ) ) ( g ( x ) - g ( y ) ) \geq 0 , \forall x , y \in \mathcal { X }$ , we have + +$$ +\begin{array} { r l } & { \int _ { \infty } ^ { \infty } \int _ { \infty } ^ { \infty } \mathcal { \bar { E } } _ { x } ^ { x } \Bigl ( \mathrm { d } \xi \Bigr ( x \mathrm { S p } ( \xi ) + \mathrm { i } \omega ( 2 \xi ) \mathrm { S p } ( \xi ) + \mathrm { i } \omega \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) } \\ & { \int _ { \infty } ^ { \infty } \int _ { \infty } ^ { \infty } \mathcal { \bar { E } } _ { x } ^ { x } \Bigl ( \mathrm { d } \xi \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) } \\ & { \int _ { \infty } ^ { \infty } \int _ { \infty } ^ { \infty } \mathcal { \bar { E } } _ { x } ^ { x } \Bigl ( \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) \mathrm { S p } ( \xi ) } \\ & \int _ { \infty } ^ { \infty } \int _ { \infty } ^ { \infty } \int _ { \infty } ^ { \infty } \int _ { \infty } ^ { \infty } \int _ { \infty } ^ \end{array} +$$ + +Corollary A.7. Let g( p(x)q(x) ) $\begin{array} { r } { g ( \frac { p ( x ) } { q ( x ) } ) = - \log \frac { p ( x ) } { q ( x ) } } \end{array}$ where $p ( x ) > 0 , \forall x \in \mathcal { X }$ and $q ( x ) > 0 , \forall x \in \mathcal { X } _ { }$ , for $\upsilon > 0$ , the inequality + +$$ +\int _ { \mathcal { X } } q ( x ) f ( v \frac { p ( x ) } { q ( x ) } ) g ( \frac { p ( x ) } { q ( x ) } ) \mathrm { d } x \geq \int _ { \mathcal { X } } q ( x ) f ( v \frac { p ( x ) } { q ( x ) } ) \mathrm { d } x \int _ { \mathcal { X } } q ( x ) g ( \frac { p ( x ) } { q ( x ) } ) \mathrm { d } x , +$$ + +holds if $f ^ { \prime } ( x ) \leq 0 , \forall x \in \mathcal { X }$ . It is not always satisfied for $f$ functions of $f$ -divergence. We list a comparison of $f$ on that condition in Tab. 2. + +Proof. $\begin{array} { r } { g ^ { \prime } ( x ) = - \log x = - \frac { 1 } { x } < 0 , \forall x \in \mathcal { X } } \end{array}$ . Suppose $f ^ { \prime } ( x ) \leq 0 , \forall x \in \mathcal { X }$ , we have $f \left( x \right) \geq f \left( y \right)$ if and only if $g ( \frac { p ( y ) } { q ( y ) } ) , \forall x , y \in \mathcal { X }$ $g ( x ) \geq g ( y ) , \forall x , y \in \mathcal { X }$ holds for all $\upsilon > 0$ holds. Thus . By defining $\begin{array} { r } { f ( v \frac { p ( x ) } { q ( x ) } ) \geq f ( v \frac { p ( y ) } { q ( y ) } ) } \end{array}$ $\begin{array} { r } { F ( x ) = f ( v \frac { p ( x ) } { q ( x ) } ) ) } \end{array}$ and if and only if $\begin{array} { r } { G ( x ) = g ( \frac { p ( x ) } { q ( x ) } ) } \end{array}$ $g \big ( \frac { p ( x ) } { q ( x ) } \big ) \geq$ and using Lemma A.6, we have: + +$$ +\int _ { \mathcal { X } } q ( x ) F ( x ) G ( x ) \mathrm { d } x \geq \int _ { \mathcal { X } } q ( x ) F ( x ) \mathrm { d } x \int _ { \mathcal { X } } q ( x ) G ( x ) \mathrm { d } x . +$$ + +Then we know + +$$ +\int _ { \mathcal { X } } q ( x ) f ( v \frac { p ( x ) } { q ( x ) } ) g ( \frac { p ( x ) } { q ( x ) } ) \mathrm { d } x \geq \int _ { \mathcal { X } } q ( x ) f ( v \frac { p ( x ) } { q ( x ) } ) \mathrm { d } x \int _ { \mathcal { X } } q ( x ) g ( \frac { p ( x ) } { q ( x ) } ) \mathrm { d } x +$$ + +holds. + +Table 2: Properties of $f ^ { \prime } ( x ) \leq 0 , \forall x \in \mathcal { X }$ for $f$ -divergences. + +
NameGenerator function f(x)If f'(x)≤0,∀x∈X
Kullback-LeiblerReverse KLPearson x²Squared HellingerJensen-ShannonGANx logxFalseTrueFalseFalseFalseTrue
-logx(x-1)²(√x-1)²-(x+1)log1++xlog x2xlogx- (𝑥+1)log(𝑥+1)
+ +Now, we prove Thm. 4.4. For better readability, we first rewrite Thm. 4.4 as follows: + +Theorem A.8. Let $\bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ as the data distribution of the best-response policy $\bar { \beta } ^ { * }$ in Eq. 4 under model $M _ { \theta }$ parameterized by $\theta$ , then we can find the optimal $\theta ^ { * }$ o $\begin{array} { r } \mathsf { \Gamma } _ { \mathsf { m i n } _ { \theta } \operatorname* { m a x } _ { \beta \in \Pi } \bar { L } ( \rho _ { M ^ { * } } ^ { \beta } , M _ { \theta } ) } \end{array}$ (Eq. 4) via iteratively optimizing the objective $\theta _ { t + 1 } = \mathrm { m i n } _ { \theta } \bar { L } ( \bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } , M _ { \theta } )$ , where $\bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ is approximated via the last-iteration model $M _ { \theta _ { t } }$ . Based on Corollary A.7, we have an upper bound objective for $\begin{array} { r } { \operatorname* { m i n } _ { \theta } \bar { L } ( \bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } , M _ { \theta } ) } \end{array}$ and derive the following objective + +$$ +\mathfrak { t } _ { t + 1 } = \arg \operatorname* { m a x } _ { \theta } \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \times } } \left[ \frac { 1 } { \alpha _ { 0 } ( x , a ) } \log M _ { \theta } ( x ^ { \prime } | x , a ) \underbrace { \left( f \left( \frac { \rho _ { M _ { \theta } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } \right) - f \left( \frac { \rho _ { M _ { \theta } } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } \right) + H _ { M ^ { * } } ( x , a ) \right) } _ { W ( x , a , x ^ { \prime } ) } \right] . +$$ + +where $\alpha _ { 0 } ( x , a ) = \alpha _ { M _ { \theta _ { t } } } \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) , \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \kappa } } [ \cdot ]$ ] denotes $\mathbb { E } _ { x , a , x ^ { \prime } \sim \rho _ { M ^ { * } } ^ { \kappa } }$ [·], $f$ is the generator function in $f$ -divergence which satisfies $f ^ { \prime } ( x ) \leq 0 , \forall x \in \mathcal { X }$ , and $\theta$ is the parameters of $M$ . $M _ { \theta _ { t } }$ denotes $a$ probability function with the same parameters as the learned model (i.e., $\bar { \theta } = \theta$ ) but the parameter is fixed and only used for sampling. + +Proof. Let $\bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ as the data distribution of the best-response policy $\bar { \beta } ^ { * }$ in Eq. 4 under model $M _ { \theta }$ parameterized by $\theta$ , then we can find the optimal $\theta _ { t + 1 }$ of $\begin{array} { r } { \operatorname* { m i n } _ { \theta } \operatorname* { m a x } _ { \beta \in \Pi } \bar { L } ( \rho _ { M ^ { * } } ^ { \beta } , M _ { \theta } ) } \end{array}$ (Eq. 4) via iteratively optimizing the objective $\begin{array} { r } { \theta _ { t + 1 } = \operatorname* { m i n } _ { \theta } \bar { L } ( \bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } } , M _ { \theta } ) } \end{array}$ , where $\bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ is approximated via the last-iteration model $\bar { M } _ { \theta _ { t } }$ : + +$$ +\begin{array} { r l } & \quad \mathrm { w r e r } \int _ { - \infty } ^ { \infty } \frac { 1 } { \mathrm { d } { x } } \frac { \mathrm { d } } { \mathrm { d } { x } ^ { 2 } \mathrm { x } ^ { 2 } \mathrm { x } ^ { 3 } } \left( \int _ { - \infty } ^ { \infty } \mathrm { d } { x } \omega \mathrm { e r } \omega ^ { 2 } \mathrm { d } { x } ^ { 2 } \mathrm { e } ^ { \mathrm { i } \omega _ { 2 } \mathrm { x } ^ { 2 } \mathrm { x } ^ { 2 } \mathrm { x } ^ { 3 } } \right) + \int _ { - \infty } ^ { \infty } \mathrm { d } { x } ^ { 2 } \mathrm { e } ^ { \mathrm { i } \omega _ { 3 } \mathrm { x } ^ { 2 } \mathrm { x } ^ { 3 } \mathrm { x } ^ { 3 } \mathrm { x } ^ { 3 } \mathrm { x } ^ { 4 } \mathrm { x } ^ { 4 } \mathrm { x } ^ { 2 } \mathrm { x } ^ { 4 } \mathrm { x } ^ { 4 } \mathrm { x } ^ { 3 } } \\ & \qquad - \mathrm { e } ^ \mathrm { i } \omega _ { 3 } \mathrm { x } \mathrm { a n } \mathrm { x } ^ { 3 } \mathrm { y } ^ { 4 } \left( \frac { \sqrt { 3 } } { \mathrm { d } { x } ^ { 2 } \mathrm { x } ^ { 3 } \mathrm { x } ^ { 4 } \mathrm { x } ^ { 3 } \mathrm { y } ^ { 2 } \right) } - \mathrm { e } ^ { \mathrm { i } \omega _ { 3 } \mathrm { x } \mathrm { a n } \mathrm { y } ^ { 4 } \mathrm { y } ^ { 4 } \mathrm { x } ^ { 3 } \mathrm { x } ^ { 3 } \mathrm { x } ^ { 4 } \mathrm { y } ^ { 4 } \mathrm { x } ^ { 4 } \mathrm { x } ^ { 3 } \mathrm { x } ^ { 4 } \mathrm { y } ^ { 4 } \mathrm { x } ^ { 4 } \mathrm { x } ^ { 3 } \mathrm { x } ^ { 4 } \mathrm { x } ^ { 3 } } \\ & \qquad - \mathrm { e } ^ \mathrm { i } \omega _ { 3 } \mathrm { x } \mathrm { a n } \mathrm { x } ^ { 3 } \mathrm { y } ^ { 4 } \left( \frac { \sqrt { 3 } } { \mathrm { d } { x } ^ { 2 } \mathrm { x } ^ { 3 } \mathrm { x } ^ { 4 } \mathrm { x } ^ { 3 } \mathrm { y } ^ { 2 } \right) } - \mathrm { e } ^ \mathrm { i } \omega _ { 3 } \mathrm { x } ^ { 3 } \mathrm { y } ^ { 4 } \mathrm { x } ^ { 3 } \mathrm { y } ^ { 4 } \mathrm y \end{array} +$$ + +where $M _ { \theta _ { t } }$ is introduced to approximate the term $\bar { \rho } _ { M ^ { * } } ^ { \bar { \beta } ^ { * } }$ and fixed when optimizing $\theta$ . In Eq. 16, $\| \rho _ { M ^ { * } } ^ { \beta } ( \cdot , \cdot ) \| _ { 2 } ^ { 2 }$ for Eq. 9 is eliminated as it does not contribute to the gradient of $\theta$ . Assume $f ^ { \prime } ( x ) \leq$ $0 , \forall x \in \mathcal { X }$ , let $\begin{array} { r } { v ( x , a ) : = \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } > 0 } \end{array}$ , $p ( x ^ { \prime } | x , a ) = M _ { \theta } ( x ^ { \prime } | x , a )$ , and $q ( x ^ { \prime } | x , a ) = M ^ { * } ( x ^ { \prime } | x , a )$ , the first inequality can be derived by adopting Corollary A.7 and eliminating the first $H _ { M ^ { * } }$ since it does not contribute to the gradient of $\theta$ . + +# A.4 PROOF OF THE TRACTABLE SOLUTION + +Now we are ready to prove the tractable solution: + +Proof. The core challenge is that the term $\begin{array} { r } { f \big ( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } \big ) - f \big ( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } \big ) } \end{array}$ is still intractable. In the following, we give a tractable solution to Thm. 4.4. First, we resort to the first-order approximation. Given some $u \bar { \in } ( 1 - \xi , 1 + \xi ) , \xi > 0$ , we have + +$$ +f ( u ) \approx f ( 1 ) + f ^ { \prime } ( u ) ( u - 1 ) , +$$ + +where $f ^ { \prime }$ is the first-order derivative of $f$ . By Taylor’s formula and the fact that $f ^ { \prime } ( u )$ of the generator function $f$ is bounded in $( 1 - \xi , 1 + \xi )$ , the approximation error is no more than $\mathcal { O } ( \xi ^ { 2 } )$ . Substituting $u$ with $\textstyle { \frac { p ( x ) } { q ( x ) } }$ in Eq. 18, the pattern $f ( { \frac { p ( x ) } { q ( x ) } } )$ in Eq. 17 can be converted to $\begin{array} { r } { \frac { p ( x ) } { q ( x ) } f ^ { \prime } ( \frac { p ( x ) } { q ( x ) } ) - f ^ { \prime } ( \frac { p ( x ) } { q ( x ) } ) + f ( 1 ) } \end{array}$ , + +then we have: + +$$ +\begin{array} { r l } { \gamma _ { 1 } = \gamma _ { 2 } \gamma _ { 2 } \gamma _ { 3 } = \gamma _ { 3 } \gamma _ { 3 } \gamma _ { 4 } } & { \gamma _ { 2 } \gamma _ { 3 } = \gamma _ { 1 } \gamma _ { 3 } \gamma _ { 3 } , } \\ & { \qquad \quad + \frac { \gamma _ { 1 } \gamma _ { 1 } \gamma _ { 2 } } { \mu _ { 1 } \mu _ { 2 } \mu _ { 3 } } \gamma _ { 1 } ( \frac { \gamma _ { 2 } \gamma _ { 3 } \gamma _ { 3 } } { \mu _ { 2 } } + \frac { \gamma _ { 1 } \gamma _ { 2 } \gamma _ { 3 } } { \mu _ { 3 } } ) \gamma _ { 3 } ( \gamma _ { 3 } + \gamma _ { 1 } \gamma _ { 2 } \gamma _ { 3 } + \gamma _ { 2 } \gamma _ { 3 } ) } \\ & { \qquad \quad + \frac { \gamma _ { 1 } \gamma _ { 1 } \gamma _ { 3 } } { \mu _ { 2 } \mu _ { 3 } } \gamma _ { 1 } ( \frac { \gamma _ { 2 } \gamma _ { 3 } \gamma _ { 3 } } { \mu _ { 2 } } + \frac { \gamma _ { 1 } \gamma _ { 2 } \gamma _ { 3 } } { \mu _ { 3 } } ) \gamma _ { 3 } ( \gamma _ { 3 } + \gamma _ { 1 } \gamma _ { 2 } \gamma _ { 3 } + \gamma _ { 2 } \gamma _ { 3 } ) } \\ & { \qquad \quad \times ( \gamma _ { 2 } + \gamma _ { 1 } \gamma _ { 3 } \gamma _ { 3 } + \gamma _ { 1 } \gamma _ { 2 } \gamma _ { 3 } + \gamma _ { 1 } \gamma _ { 3 } \gamma _ { 3 } \gamma _ { 3 } ) \gamma _ { 4 } \exp ( - \gamma _ { 2 } \gamma _ { 3 } \gamma _ { 3 } \gamma _ { 4 } + \gamma _ { 2 } \gamma _ { 3 } ) } \\ & { \qquad \quad \times \exp ( - \gamma _ { 1 } \gamma _ { 1 } \gamma _ { 2 } \gamma _ { 3 } \gamma _ { 3 } ) \gamma _ { 3 } ( \gamma _ { 3 } + \gamma _ { 2 } \gamma _ { 3 } \gamma _ { 4 } ) \gamma _ { 4 } \exp ( - \gamma _ { 2 } \gamma _ { 3 } \gamma _ { 2 } + \gamma _ { 3 } \gamma _ { 4 } ) } \\ & \qquad \quad \times \exp ( - \gamma _ { 1 } \gamma _ { 3 } \gamma _ { 3 } \gamma _ { 4 } ) \gamma _ { 5 } ( \gamma _ 3 \end{array} +$$ + +Note +but we choose to keep it and ignore +an empirical data distribution through data collected by $\kappa$ in $M ^ { * }$ directly, rather than from a uniform distribution which is harder to be generated. Although keeping $\rho _ { M ^ { * } } ^ { \kappa } ( x , a )$ incurs extra bias in theory, the results in our experiments show that it has not made significant negative effects in practice. We leave this part of modeling in future work. In particular, by ignoring $\bar { \alpha _ { 0 } } ( x , a )$ , we have: + +$$ += \arg \operatorname* { m a x } _ { \theta } \int _ { \chi , A , \chi } \rho _ { \mathbb { M } _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ) \left( f ^ { \prime } \left( \frac { \rho _ { \mathbb { M } _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { \mathbb { M } ^ { + } } ^ { \kappa } ( x , a , x ^ { \prime } ) } \right) - f ^ { \prime } \left( \frac { \rho _ { \mathbb { M } _ { \theta _ { t } } } ^ { \kappa } ( x , a ) } { \rho _ { \mathbb { M } ^ { + } } ^ { \kappa } ( x , a ) } \right) \right) \log M _ { \theta } ( x ^ { \prime } | x , a ) \mathrm { d } x ^ { \prime } \mathrm { d } a \mathrm { d } x + +$$ + +$$ +\int _ { \boldsymbol { x } , A , \boldsymbol { \tilde { x } } } \rho _ { \mathrm { M } ^ { * } } ^ { \kappa } ( \boldsymbol { x } , \boldsymbol { a } , \boldsymbol { x } ^ { \prime } ) \left( f ^ { \prime } \left( \frac { \rho _ { \mathrm { M } \theta _ { t } } ^ { \kappa } ( \boldsymbol { x } , \boldsymbol { a } ) } { \rho _ { \mathrm { M } ^ { * } } ^ { \kappa } ( \boldsymbol { x } , \boldsymbol { a } ) } \right) - f ^ { \prime } \left( \frac { \rho _ { \mathrm { M } \theta _ { t } } ^ { \kappa } ( \boldsymbol { x } , \boldsymbol { a } , \boldsymbol { x } ^ { \prime } ) } { \rho _ { \mathrm { M } ^ { * } } ^ { \kappa } ( \boldsymbol { x } , \boldsymbol { a } , \boldsymbol { x } ^ { \prime } ) } \right) + H _ { \mathsf { M } ^ { * } } ( \boldsymbol { x } , \boldsymbol { a } ) \right) \log M _ { \theta } ( \boldsymbol { x } ^ { \prime } | \boldsymbol { x } , \boldsymbol { a } ) \mathrm { d } \boldsymbol { x } ^ { \prime } \mathrm { d } \boldsymbol { a } \mathrm { d } \boldsymbol { x } +$$ + +We can estimate $f ^ { \prime } \left( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } \right)$ and $f ^ { \prime } \left( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } \right)$ through Lemma A.9. + +Lemma A.9 $( f ^ { \prime } ( { \frac { p } { q } } )$ estimation (Nguyen et al., 2010)). Given a function $T _ { \varphi } : \mathcal { X } \to \mathbb { R }$ parameterized by $\varphi \in \Phi$ , if $f$ is convex and lower semi-continuous, by finding the maximum point of $\varphi$ in the following objective: + +$$ +\varphi ^ { * } = \arg \operatorname* { m a x } _ { \varphi } \mathbb { E } _ { x \sim p ( x ) } \left[ T _ { \varphi } ( x ) \right] - \mathbb { E } _ { x \sim q ( x ) } \left[ f ^ { * } ( T _ { \varphi } ( x ) ) \right] , +$$ + +we have $f ^ { \prime } ( \frac { p ( x ) } { q ( x ) } ) = T _ { \varphi ^ { * } } ( x )$ . $f ^ { * }$ is Fenchel conjugate of $f$ (Hiriart-Urruty & Lemaréchal, 2001). + +In particular, + +$$ +\begin{array} { r l } & { \varphi _ { 0 } ^ { * } = \underset { \varphi _ { 0 } } { \arg \operatorname* { m a x } } ~ \mathbb { E } _ { x , a , x ^ { \prime } \sim \rho _ { M ^ { * } } ^ { * } } \left[ T _ { \varphi _ { 0 } } ( x , a , x ^ { \prime } ) \right] - \mathbb { E } _ { x , a , x ^ { \prime } \sim \rho _ { M _ { \theta _ { t } } } ^ { * } } \left[ f ^ { * } ( T _ { \varphi _ { 0 } } ( x , a , x ^ { \prime } ) ) \right] } \\ & { \varphi _ { 1 } ^ { * } = \underset { \varphi _ { 1 } } { \arg \operatorname* { m a x } } ~ \mathbb { E } _ { x , a \sim \rho _ { M ^ { * } } ^ { * } } \left[ T _ { \varphi _ { 1 } } ( x , a ) \right] - \mathbb { E } _ { x , a \sim \rho _ { M _ { \theta _ { t } } } ^ { \kappa } } \left[ f ^ { * } ( T _ { \varphi _ { 1 } } ( x , a ) ) \right] , } \end{array} +$$ + +then we have $\begin{array} { r } { f ^ { \prime } \left( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } ) } \right) \approx T _ { \varphi _ { 0 } ^ { * } } ( x , a , x ^ { \prime } ) } \end{array}$ and $\begin{array} { r } { f ^ { \prime } \left( \frac { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) } { \rho _ { M ^ { * } } ^ { \kappa } ( x , a ) } \right) \approx T _ { \varphi _ { 1 } ^ { * } } ( x , a ) } \end{array}$ . Given $\varphi _ { 0 } ^ { * }$ and $\varphi _ { 1 } ^ { * }$ , let $A _ { \varphi _ { 0 } ^ { * } , \varphi _ { 1 } ^ { * } } ( x , \dot { a } , x ^ { \prime } ) = T _ { \varphi _ { 0 } ^ { * } } ( x , a , x ^ { \prime } ) - T _ { \varphi _ { 1 } ^ { * } } ( x , a ) ,$ , then we can optimize $\theta$ via: + +$$ +\begin{array} { r l } & { \mathsf { \Gamma } _ { t + 1 } = \arg \operatorname* { m a x } _ { \theta } \displaystyle \int _ { \mathcal X , A , \mathcal X } \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ) \left( T _ { \varphi _ { 0 } ^ { \star } } ( x , a , x ^ { \prime } ) - T _ { \varphi _ { 1 } ^ { \star } } ( x , a ) \right) \log M _ { \theta } ( x ^ { \prime } | x , a ) \mathrm { d } x ^ { \prime } \mathrm { d } a \mathrm { d } x + } \\ & { \qquad \displaystyle \int _ { \mathcal X , A , \mathcal X } \rho _ { M ^ { \star } } ^ { \kappa } ( x , a , x ^ { \prime } ) \left( T _ { \varphi _ { 1 } ^ { \star } } ( x , a ) - T _ { \varphi _ { 0 } ^ { \star } } ( x , a , x ^ { \prime } ) + H _ { M ^ { \star } } ( x , a ) \right) \log M _ { \theta } ( x ^ { \prime } | x , a ) \mathrm { d } x ^ { \prime } \mathrm { d } a \mathrm { d } x ^ { \prime } } \\ & { = \arg \operatorname* { m a x } _ { \theta } \displaystyle \int _ { \mathcal X , A , \mathcal X } \rho _ { M _ { \theta } } ^ { \kappa } ( x , a , x ) A _ { \varphi _ { 0 } ^ { \star } , \varphi _ { 1 } ^ { \star } } ( x , a , x ^ { \prime } ) \log M _ { \theta } ( x ^ { \prime } | x , a ) \mathrm { d } x ^ { \prime } \mathrm { d } a \mathrm { d } x + } \\ & { \qquad \displaystyle \int _ { \mathcal X , A , \mathcal X } \rho _ { M ^ { \star } } ^ { \kappa } ( x , a , x ^ { \prime } ) ( - A _ { \varphi _ { 0 } ^ { \star } , \varphi _ { 1 } ^ { \star } } ( x , a , x ^ { \prime } ) + H _ { M ^ { \star } } ( x , a ) ) \log M _ { \theta } ( x ^ { \prime } | x , a ) \mathrm { d } x ^ { \prime } \mathrm { d } a \mathrm { d } x . } \end{array} +$$ + +Based on the specific $f$ -divergence, we can represent $T$ and $f ^ { * } ( T )$ with a discriminator $D _ { \varphi }$ . It can be verified that $f ( u ) = u \log u - ( u + 1 ) \log ( u + 1 )$ , $T _ { \varphi } ( u ) = \log D _ { \varphi } ( u )$ , and $f ^ { * } ( T _ { \varphi } ( u ) ) =$ $- \log ( 1 - D _ { \varphi } ( u ) )$ proposed in Nowozin et al. (2016) satisfies the condition $f ^ { \prime } ( x ) \leq 0 , \forall x \in \mathcal { X }$ (see Tab. 2). We select the former in the implementation and convert the tractable solution to: + +$$ +\begin{array} { r l } & { _ { t + 1 } = \arg \operatorname* { m a x } _ { \theta } \mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \times } } \left[ A _ { \varphi _ { 0 } ^ { \star } , \varphi _ { 1 } ^ { \star } } ( x , a , x ^ { \prime } ) \log M _ { \theta } ( x ^ { \prime } | x , a ) \right] + \mathbb { E } _ { \rho _ { M ^ { \star } } ^ { \times } } \left[ \left( H _ { M ^ { \star } } ( x , a ) - A _ { \varphi _ { 0 } ^ { \star } , \varphi _ { 1 } ^ { \star } } ( x , a , x ^ { \prime } ) \right) \log M _ { \theta } ( x ^ { \prime } | x , a ) \right] } \\ { \therefore } & { _ { \varphi _ { 0 } ^ { \star } } = \arg \operatorname* { m a x } _ { \theta _ { 0 } } \mathbb { E } _ { \rho _ { M ^ { \star } } ^ { \times } } \left[ \log D _ { \varphi _ { 0 } } ( x , a , x ^ { \prime } ) \right] + \mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \times } } \left[ \log ( 1 - D _ { \varphi _ { 0 } } ( x , a , x ^ { \prime } ) ) \right] } \\ & { _ { \varphi _ { 1 } ^ { \star } } = \arg \operatorname* { m a x } _ { \varphi _ { 1 } } \mathbb { E } _ { \rho _ { M ^ { \star } } ^ { \times } } \left[ \log D _ { \varphi _ { 1 } } ( x , a ) \right] + \mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \times } } \left[ \log ( 1 - D _ { \varphi _ { 1 } } ( x , a ) ) \right] , } \end{array} +$$ + +where Aφ∗0 ,φ∗1 (x, a, x′) = log Dφ∗0 (x, a, x′) − log Dφ∗1 (x, a), EρκMθt [· is a simplification of $\mathbb { E } _ { x , a , x ^ { \prime } \sim \rho _ { M _ { \theta _ { t } } } ^ { \kappa } } [ \cdot$ ]. + +# B DISCUSSION OF THE THEORETICAL RESULTS + +We summarize the limitations of current theoretical results and future work as follows: + +1. As discussed in Remark A.2, the solution Eq. 11 relies on $\rho _ { M ^ { * } } ^ { \beta } ( x , a ) \in [ 0 , c ] , \forall a \in \mathcal { A } , \forall x \in$ $\mathcal { X }$ . In some particular $M ^ { * }$ , it is intractable to derive a $\beta$ that can generate an occupancy specified by $g ( x , a ) / \alpha _ { M }$ . If more knowledge of $M ^ { * }$ or $\beta ^ { * }$ is provided or some mild assumptions can be made on the properties of $M ^ { * }$ or $\beta ^ { * }$ , we may model $\rho$ in a more sophisticated approach to alleviating the above problem. +2. In the tractable solution derivation, we ignore the term $\alpha _ { 0 } ( x , a ) = \alpha _ { M _ { \theta _ { t } } } \rho _ { M ^ { * } } ^ { \kappa } ( x , a )$ (See Eq. 20). The benefit is that $\rho _ { M ^ { * } } ^ { \kappa } ( x , a , x ^ { \prime } )$ in the tractable solution can be estimated through offline datasets directly. Although the results in our experiments show that it does not produce significant negative effects in these tasks, ignoring $\rho _ { M ^ { * } } ^ { \kappa } ( x , a )$ indeed incurs extra bias in theory. In future work, techniques for estimating $\rho _ { M ^ { * } } ^ { \kappa } ( x , a )$ (Liu et al., 2020) can be incorporated to correct the bias. On the other hand, $\alpha _ { M _ { \theta _ { t } } }$ is also ignored in the process. $\alpha _ { M _ { \theta _ { t } } }$ can be regarded as a global rescaling term of the final objective Eq. 20. Intuitively, it constructs an adaptive learning rate for Eq. 20, which increases the step size when the model is better fitted and decreases the step size otherwise. It can be considered to further improve the learning process in future work, e.g., cooperating with empirical risk minimization by balancing the weights of the two objectives through $\alpha _ { M _ { \theta _ { t } } }$ . + +# C SOCIETAL IMPACT + +This work studies a method toward counterfactual environment model learning. Reconstructing an accurate environment of the real world will promote the wide adoption of decision-making policy optimization methods in real life, enhancing our daily experience. We are aware that decision-making policy in some domains like recommendation systems that interact with customers may have risks of causing price discrimination and misleading customers if inappropriately used. A promising way to reduce the risk is to introduce fairness into policy optimization and rules to constrain the actions (Also see our policy design in Sec. G.1.3). We are involved in and advocating research in such directions. We believe that business organizations would like to embrace fair systems that can ultimately bring long-term financial benefits by providing a better user experience. + +# D AWRM-ORACLE PSEUDOCODE + +We list the pseudocode of AWRM-oracle in Alg. 1. + +# Algorithm 1 AWRM with Oracle Counterfactual Datasets + +Input: $\Phi$ : policy space; $N$ : total iterations + +# Process: + +1: Generate counterfactual datasets $\{ \mathcal { D } _ { \pi _ { \phi } } \}$ for all adversarial policies $\pi _ { \phi } , \phi \in \Phi$ +2: Initialize an environment model $M _ { \theta }$ +3: for $\mathrm { i } = 1 { : } \mathrm { N }$ do +4: Select $\mathcal { D } _ { \pi _ { \phi } }$ with worst prediction errors through $M _ { \theta }$ from $\{ \mathcal { D } _ { \pi _ { \phi } } \}$ +5: Optimize $M _ { \theta }$ with standard supervised learning based on $\mathcal { D } _ { \pi _ { \phi } }$ +6: end for + +# E IMPLEMENTATION + +# E.1 DETAILS OF THE GALILEO IMPLEMENTATION + +The approximation of Eq. 18 holds only when $\textstyle { \frac { p ( x ) } { q ( x ) } }$ is close to 1, which might not be satisfied. To handle the problem, we inject a standard supervised learning loss + +$$ +\arg \operatorname* { m a x } _ { \theta } \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \kappa } } \left[ \log M _ { \theta } ( x ^ { \prime } | x , a ) \right] +$$ + +to replace the second term of the above objective when the output probability of $D$ is far away from 0.5 $\bar { f } ^ { \prime } ( 1 ) = \log { 0 . 5 } )$ ). + +In the offline model-learning setting, we only have a real-world dataset $\mathcal { D }$ collected by the behavior policy $\mu$ . We learn a policy $\hat { \mu } \approx \bar { \mu }$ via behavior cloning with $\mathcal { D }$ (Pomerleau, 1991; Ho & Ermon, 2016) and let $\hat { \mu }$ be the policy $\kappa$ . We regard $\mathcal { D }$ as the empirical data distribution of $\rho _ { M ^ { * } } ^ { \kappa }$ and the trajectories collected by $\hat { \mu }$ in the model $M _ { \theta _ { t } }$ as the empirical data distribution of $\rho _ { M _ { \theta _ { t } } } ^ { \kappa }$ . But the assumption $\forall x \in \mathcal { X } , \forall a \in \mathcal { A } , \mu ( a | x ) > 0$ might not be satisfied. In behavior cloning, we model $\hat { \mu }$ with a Gaussian distribution and constrain the lower bound of the variance with a small value $\epsilon _ { \mu } > 0$ to keep the assumption holding. Besides, we add small Gaussian noises $\mathbf { u } \sim \mathcal { N } ( 0 , \epsilon _ { D } )$ to the inputs of $D _ { \varphi }$ to handle the mismatch between $\rho _ { M ^ { * } } ^ { \mu }$ and $\rho _ { M ^ { * } } ^ { \hat { \mu } }$ due to $\epsilon _ { \mu }$ . In particular, for $\varphi _ { 0 }$ and $\varphi _ { 1 }$ learning, we have: + +$$ +\begin{array} { r l } & { \gamma _ { 0 } ^ { * } = \arg \underset { \varphi _ { 0 } } { \operatorname* { m a x } } \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \mathrm { s } } , \mathrm { ~ u } } \left[ \log D _ { \varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \prime } + u _ { x ^ { \prime } } ) \right] + \mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \mathrm { s } } , \mathrm { ~ u } } \left[ \log ( 1 - D _ { \varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \prime } + u _ { x ^ { \prime } } ) ) \right] } \\ & { \gamma _ { 1 } ^ { * } = \arg \underset { \varphi _ { 1 } } { \operatorname* { m a x } } \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \mathrm { s } } , \mathrm { ~ u } } \left[ \log D _ { \varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) \right] + \mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \mathrm { s } } , \mathrm { ~ u } } \left[ \log ( 1 - D _ { \varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) \right] , } \end{array} +$$ + +where EρκMθ , is a simplification of $\mathbb { E } _ { x , a , x ^ { \prime } \sim \rho _ { M _ { \theta _ { t } } } ^ { \kappa } , \mathbf { u } \sim \mathcal { N } ( 0 , \epsilon _ { D } ) } [ \cdot ]$ and $\mathbf { u } = [ u _ { x } , u _ { a } , u _ { x ^ { \prime } } ] .$ + +On the other hand, we notice that the first term in Eq. 21 is similar to the objective of GAIL (Ho & Ermon, 2016) by regarding $M _ { \theta }$ as the policy to learn and $\kappa$ as the environment to generate data. For better capability in sequential environment model learning, here we introduce some practical tricks inspired by GAIL for model learning (Shi et al., 2019; Shang et al., 2019): we introduce an MDP for $\kappa$ and $M _ { \theta }$ , where the reward is defined by the discriminator $D$ , i.e., $r ( x , a , x ^ { \prime } ) ~ = ~ \log D ( x , a , x ^ { \prime } )$ . $M _ { \theta }$ is learned to maximize the cumulative rewards. With advanced policy gradient methods (Schulman et al., 2015; 2017), the objective is converted to maxθ $\left[ \bar { A _ { \varphi _ { 0 } ^ { * } , \varphi _ { 1 } ^ { * } } } ( x , a , x ^ { \prime } ) \log M _ { \theta } ( x , a , x ^ { \prime } ) \right]$ , where ${ \cal A } \ = \ Q _ { M _ { \theta _ { t } } } ^ { \kappa } \ - \ V _ { M _ { \theta _ { t } } } ^ { \kappa }$ , + +# Algorithm 2 GALILEO pseudocode + +# + +$\mathcal { D } _ { \mathrm { r e a l } }$ : offline dataset sampled from $\rho _ { M ^ { * } } ^ { \mu }$ where $\mu$ is the behavior policy; +$N$ : total iterations; + +# Process: + +1: Approximate a behavior policy $\hat { \mu }$ via behavior cloning +2: Initialize an environment model $M _ { \theta _ { 1 } }$ +3: for $t = 1 : N$ do +4: Use $\hat { \mu }$ to generate a dataset $\mathcal { D } _ { \mathrm { g e n } }$ with the model $M _ { \theta _ { t } }$ +5: Update the discriminators $D _ { \varphi _ { 0 } }$ and $D _ { \varphi _ { 1 } }$ via Eq. 26 and Eq. 27 respectively, where $\rho _ { M _ { \theta _ { t } } } ^ { \hat { \mu } }$ is +estimated by $\mathcal { D } _ { \mathrm { g e n } }$ and $\rho _ { M ^ { * } } ^ { \mu }$ is estimated by $\mathcal { D } _ { \mathrm { r e a l } }$ +6: Update $Q$ and $V$ via Eq. 24 and Eq. 25 through $\mathcal { D } _ { \mathrm { g e n } }$ , $D _ { \varphi _ { 0 } }$ , and $D _ { \varphi _ { 1 } }$ +7: Update the model $M _ { \theta _ { t } }$ via the first term of Eq. 23, which is implemented with a standard +policy gradient method like TRPO (Schulman et al., 2015) or PPO (Schulman et al., 2017). +Record the policy gradient gpg +8: if $p _ { 0 } < \mathbb { E } _ { \mathcal { D } _ { \mathrm { g e n } } } \left[ D _ { \varphi _ { 0 } } ( x _ { t } , \hat { a _ { t } } , x _ { t + 1 } ) \right] < p _ { 1 }$ then +9: Compute the gradient of $M _ { \theta _ { t } }$ via the second term of Eq. 23 and record it as $g _ { \mathrm { s l } }$ +10: else +11: Compute the gradient of $M _ { \theta _ { t } }$ via Eq. 22 and record it as $g _ { \mathrm { s l } }$ +12: end if +13: Rescale $g _ { \mathrm { s l } }$ via Eq. 28 +14: Update the model $M _ { \theta _ { t } }$ via the gradient $g _ { \mathrm { s l } }$ and obtain $M _ { \theta _ { t + 1 } }$ + +$\begin{array} { r } { Q _ { M _ { \bar { \theta } } } ^ { \kappa } ( x , a , x ^ { \prime } ) = { \mathbb E } \left[ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } r ( x _ { t } , a _ { t } , x _ { t + 1 } ) \mid ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \prime } ) , \kappa , M _ { \theta _ { t } } \right] } \end{array}$ , and ${ \cal V } _ { M _ { \bar { \theta } } } ^ { \kappa } ( x , a ) =$ $\mathbb { E } _ { M _ { \bar { \theta } } } \left[ Q _ { M _ { \bar { \theta } } } ^ { \kappa } ( x , a , x ^ { \prime } ) \right]$ . $A$ in Eq. 21 can also be constructed similarly. Although it looks unnecessary in theory since the one-step optimal model $M _ { \theta }$ is the global optimal model in this setting, the technique is helpful in practice as it makes $A$ more sensitive to the compounding effect of one-step prediction errors: we would consider the cumulative effects of prediction errors induced by multi-step transitions in environments. In particular, to consider the cumulative effmulti-step of transitions in environments, we overwrite function $A _ { \varphi _ { 0 } ^ { * } , \varphi _ { 1 } ^ { * } }$ preas $A _ { \varphi _ { 0 } ^ { \ast } , \varphi _ { 1 } ^ { \ast } } = Q _ { M _ { \theta _ { t } } } ^ { \kappa } - V _ { M _ { \theta _ { t } } } ^ { \kappa }$ , where $\begin{array} { r } { Q _ { M _ { \theta _ { \star } } } ^ { \kappa } ( x , a , x ^ { \prime } ) = \mathbb { E } \left[ \sum _ { t } ^ { \infty } \gamma ^ { t } \log D _ { \varphi _ { 0 } ^ { \star } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \prime } ) , \kappa , \bar { M } _ { \theta _ { t } } \right] } \end{array}$ and $\begin{array} { r } { V _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) = \mathbb { E } \left[ \sum _ { t } ^ { \infty } \gamma ^ { t } \log D _ { \varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \kappa , M _ { \theta _ { t } } \right] , } \end{array}$ . To give an algorithm for singlestep environment model learning, we can just set $\gamma$ in $Q$ and $V$ to 0. + +![](images/250eec592a0278e03dcb9c611ad65d4003d340434d2a84445af36733f46969b9.jpg) +15: end for +Figure 7: Illustration of the workflow of the GALILEO algorithm. + +By adopting the above implementation techniques, we convert the objective into the following formulation + +$$ +t + 1 = \arg \operatorname* { m a x } _ { \theta } \mathbb { E } _ { \rho _ { M _ { \theta } } ^ { \star } } \left[ A _ { \varphi _ { 0 } ^ { \star } , \varphi _ { 1 } ^ { \star } } ( x , a , x ^ { \prime } ) \log M _ { \theta } ( x ^ { \prime } | x , a ) \right] + \mathbb { E } _ { \rho _ { M ^ { \star } } ^ { \star } } \left[ \left( H _ { M ^ { \star } } ( x , a ) - A _ { \varphi _ { 0 } ^ { \star } , \varphi _ { 1 } ^ { \star } } ( x , a , x ^ { \prime } ) \right) \log M _ { \theta } ( x ^ { \prime } | x , a ) \right] . +$$ + +s.t. + +$$ +\begin{array} { r l } & { Q _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) = \mathbb E [ \displaystyle \sum _ { t } ^ { \infty } \gamma ^ { t } \log D _ { \varphi _ { 0 } ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \prime } ) , \kappa , M _ { \theta _ { t } } ] \qquad ( 2 4 ) } \\ & { V _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) = \mathbb E [ \displaystyle \sum _ { t } ^ { \infty } \gamma ^ { t } \log D _ { \varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \kappa , M _ { \theta _ { t } } ] } \\ & { \varphi _ { 0 } ^ { * } = \arg \operatorname* { m a x } _ { \varphi _ { 0 } ^ { * } } \mathbb E _ { \rho _ { M ^ { * } } ^ { \kappa } , \textbf { u } } [ \log D _ { \varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \prime } + u _ { x ^ { \prime } } ) ] + \mathbb E _ { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } , \textbf { u } } [ \log ( 1 - D _ { \varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \prime } ) ] ] } \\ & { \varphi _ { 1 } ^ { * } = \arg \operatorname* { m a x } _ { \varphi _ { 1 } ^ { * } } \mathbb E _ { \rho _ { M ^ { * } } ^ { \kappa } , \textbf { u } } [ \log D _ { \varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ] + \mathbb E _ { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } , \textbf { u } } [ \log ( 1 - D _ { \varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) ] , } \end{array} +$$ + +where $A _ { \varphi _ { 0 } ^ { * } , \varphi _ { 1 } ^ { * } } ( x , \underline { { a } } , x ^ { \prime } ) = Q _ { M _ { \theta } } ^ { \kappa } ( x , a , \underline { { x } } ^ { \prime } ) - V _ { M _ { \theta } } ^ { \kappa } ( x , \underline { { a } } )$ . In practice, GALILEO optimizes the first term of Eq. 23 with conservative policy gradient algorithms (e.g., PPO (Schulman et al., 2017) or TRPO (Schulman et al., 2015)) to avoid unreliable gradients for model improvements. Eq. 26 and Eq. 27 are optimized with supervised learning. The second term of Eq. 23 is optimized with supervised learning with a re-weighting term $- A _ { \varphi _ { 0 } ^ { * } , \varphi _ { 1 } ^ { * } } + H _ { M ^ { * } }$ . Since $H _ { M ^ { * } }$ is unknown, we use $H _ { M _ { \theta } }$ to estimate it. When the mean output probability of a batch of data is larger than 0.6 or small than 0.4, we replace the second term of Eq. 23 with a standard supervised learning in Eq. 22. Besides, unreliable gradients also exist in the process of optimizing the second term of Eq. 23. In our implementation, we use the scale of policy gradients to constrain the gradients of the second term of Eq. 23. In particular, we first compute the $l _ { 2 }$ -norm of the gradient of the first term of Eq. 23 via conservative policy gradient algorithms, named $| | g _ { \mathrm { p g } } | | _ { 2 }$ . Then we compute the $l _ { 2 }$ -norm of the gradient of the second term of Eq. 23, name $| | g _ { \mathrm { s l } } | | _ { 2 }$ . Finally, we rescale the gradients of the second term $g _ { \mathrm { s l } }$ by + +$$ +g _ { \mathrm { s l } } g _ { \mathrm { s l } } \frac { \vert \vert g _ { \mathrm { p g } } \vert \vert _ { 2 } } { \operatorname* { m a x } \{ \vert \vert g _ { \mathrm { p g } } \vert \vert _ { 2 } , \vert \vert g _ { \mathrm { s l } } \vert \vert _ { 2 } \} } . +$$ + +For each iteration, Eq. 23, Eq. 26, and Eq. 27 are trained with certain steps (See Tab. 5) following the same framework as GAIL. Based on the above techniques, we summarize the pseudocode of GALILEO in Alg. 2, where $p _ { 0 }$ and $p _ { 1 }$ are set to 0.4 and 0.6 in all of our experiments. The overall architecture is shown in Fig. 7. + +to Reviewer bcJS + +# E.2 CONNECTION WITH PREVIOUS ADVERSARIAL ALGORITHMS + +Standard GAN (Goodfellow et al., 2014) can be regarded as a partial implementation including the first term of Eq. 23 and Eq. 26 by degrading them into the single-step scenario. In the context of GALILEO, the objective of GAN is + +$$ +\begin{array} { r l } & { \theta _ { t + 1 } = \underset { \theta } { \arg \operatorname* { m a x } } ~ \mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } } \left[ A _ { \varphi ^ { * } } ( x , a , x ^ { \prime } ) \log M _ { \theta } ( x ^ { \prime } | x , a ) \right] } \\ & { s . t . \quad \varphi * = \underset { \varphi } { \arg \operatorname* { m a x } } ~ \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \kappa } } \left[ \log D _ { \varphi } ( x , a , x ^ { \prime } ) \right] + \mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } } \left[ \log ( 1 - D _ { \varphi } ( x , a , x ^ { \prime } ) ) \right] , } \end{array} +$$ + +where $\begin{array} { r c l } { { A _ { \varphi ^ { * } } ( x , a , x ^ { \prime } ) } } & { { = } } & { { \log D _ { \varphi ^ { * } } ( x , a , x ^ { \prime } ) } } \end{array}$ . In the single-step scenario, $\begin{array} { r l } { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) } & { { } = } \end{array}$ $\rho _ { 0 } ( x ) \kappa ( a | x ) M _ { \theta _ { t } } ( x ^ { \prime } | a , x )$ . The term $\mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } } \left[ A _ { \varphi ^ { * } } ( x , a , x ^ { \prime } ) \log M _ { \theta } ( x ^ { \prime } | x , a ) \right]$ can convert to $\mathbb { E } _ { \rho _ { M _ { \theta } } ^ { \kappa } } \left[ \log D _ { \varphi ^ { * } } ( x , a , x ^ { \prime } ) \right]$ by replacing the gradient of $M _ { \theta _ { t } } ( x ^ { \prime } | x , a ) \nabla _ { \theta } \log M _ { \theta } ( x ^ { \prime } | x , a )$ with $\nabla _ { \boldsymbol { \theta } } M _ { \boldsymbol { \theta } } ( \boldsymbol { x } ^ { \prime } | \boldsymbol { x } , a )$ (Sutton & Barto, 1998). Previous algorithms like GANITE (Yoon et al., 2018) and SCIGAN (Bica et al., 2020) can be regarded as variants of the above training framework. + +The first term of Eq. 23 and Eq. 26 are similar to the objective of GAIL by regarding $M _ { \theta }$ as the “policy” to imitate and $\hat { \mu }$ as the “environment” to collect data. In the context of GALILEO, the + +objective of GAIL is: + +$$ +\begin{array} { r l } { \displaystyle \theta _ { t + 1 } = \arg \operatorname* { m a x } _ { \theta } \mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } } \left[ A _ { \varphi ^ { * } } ( x , a , x ^ { \prime } ) \log M _ { \theta } ( x ^ { \prime } | x , a ) \right] } & { } \\ { \displaystyle s . t . } & { Q _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a , x ^ { \prime } ) = \mathbb { E } \left[ \sum _ { t } ^ { \infty } \gamma ^ { t } \log D _ { \varphi ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \prime } ) , \kappa , M _ { \theta _ { t } } \right] } \\ { \displaystyle \varphi ^ { * } = \arg \operatorname* { m a x } _ { \varphi } \mathbb { E } _ { \rho _ { M ^ { * } } ^ { \kappa } } \left[ \log D _ { \varphi } ( x , a , x ^ { \prime } ) \right] + \mathbb { E } _ { \rho _ { M _ { \theta _ { t } } } ^ { \kappa } } \left[ \log ( 1 - D _ { \varphi } ( x , a , x ^ { \prime } ) ) \right] , } \end{array} +$$ + +where $A _ { \varphi ^ { * } } ( x , a , x ^ { \prime } ) = Q _ { M _ { \theta } } ^ { \kappa } ( x , a , x ^ { \prime } ) - V _ { M _ { \theta } } ^ { \kappa } ( x , a )$ and $V _ { M _ { \theta _ { t } } } ^ { \kappa } ( x , a ) = \mathbb { E } _ { M _ { \theta _ { t } } ( x , a ) } \left[ Q ^ { \kappa } ( x , a , x ^ { \prime } ) \right] ,$ + +# F ADDITIONAL RELATED WORK + +Our primitive objective is inspired by weighted empirical risk minimization (WERM) based on inverse propensity score (IPS). WERM is originally proposed to solve the generalization problem of domain adaptation in machine learning literature. For instance, we would like to train a predictor $M ( y | x )$ in a domain with distribution $P _ { \mathrm { t r a i n } } ( x )$ to minimize the prediction risks in the domain with distribution $P _ { \mathrm { { t e s t } } } ( x )$ , where $P _ { \mathrm { t e s t } } \neq P _ { \mathrm { t e s t } }$ . To solve the problem, we can train a weighted objective $\begin{array} { r } { \operatorname* { m a x } _ { M } \mathbb { E } _ { x \sim P _ { \mathrm { t r a i n } } } [ \frac { P _ { \mathrm { t e s t } } ( x ) } { P _ { \mathrm { t r a i n } } ( x ) } \log M ( y | x ) ] } \end{array}$ , which is called weighted empirical risk minimization methods (Ben-David et al., 2006; 2010; Cortes et al., 2010; Byrd & Lipton, 2019; Quinonero-Candela et al., 2008). These results have been extended and applied to causal inference, where the predictor is required to be generalized from the data distribution in observational studies (source domain) to the data distribution in randomized controlled trials (target domain) (Shimodaira, 2000; Assaad et al., 2021; Hassanpour $\&$ Greiner, 2019; Jung et al., 2020; Johansson et al., 2018). In this case, the input features include a state $x$ (a.k.a. covariates) and an action $a$ (a.k.a. treatment variable) which is sampled from a policy. We often assume the distribution of $x$ , $P ( x )$ is consistent between the source domain and the test domain, then we have Ptest(x)Ptrain(x) = P (x)β(a|x)P (x)µ(a|x) , where $\mu$ and $\beta$ are the policies in source and target domains respectively. In Shimodaira (2000); Assaad et al. (2021); Hassanpour $\&$ Greiner (2019), the policy in randomized controlled trials is modeled as a uniform policy, then $\begin{array} { r } { \frac { P _ { \mathrm { t e s t } } ( x ) } { P _ { \mathrm { t r a i n } } ( x ) } = \frac { P ( x ) \beta ( a | x ) } { P ( x ) \mu ( a | x ) } = \frac { \beta ( a | x ) } { \mu ( a | x ) } \propto \frac { 1 } { \mu ( a | x ) } } \end{array}$ . ${ \frac { 1 } { \mu ( a | x ) } }$ is also known as inverse propensity score (IPS). In Johansson et al. (2018), it assumes that the policy in the target domain is predefined as $\beta ( a | x )$ before environment model learning, then it uses $\frac { \beta } { \mu }$ as the IPS. The differences between AWRM and previous works are fallen in two aspects: (1) We consider the distribution-shift problem in the sequential decision-making scenario. In this scenario, we not only consider the action distribution mismatching between the behavior policy $\mu$ and the policy to evaluation $\beta$ , but also the follow-up effects of policies to the state distribution; (2) For faithful offline policy optimization, we require the environment model to have generalization ability in numerous different policies. The objective of AWRM is proposed to guarantee the generalization ability of $M$ in numerous different policies instead of a specific policy. + +On a different thread, there are also studies that bring counterfactual inference techniques of causal inference into model-based RL (Buesing et al., 2019; Pitis et al., 2020; Sontakke et al., 2021). These works consider that the transition function is relevant to some hidden noise variables and use Pearl-style structural causal models (SCMs), which is a directed acyclic graphs to define the causality of nodes in an environment, to handle the problem. SCMs can help RL in different ways: Buesing et al. (2019) approximate the posterior of the noise variables based on the observation of data, and environment models are learned based on the inferred noises. The generalization ability is improved if we can infer the correct value of the noise variables. Pitis et al. (2020) discover several local causal structural models of a global environment model, then data augmentation strategies by leveraging these local structures to generate counterfactual experiences. Sontakke et al. (2021) proposes a representation learning technique for causal factors, which is an instance of the hidden noise variables, in partially observable Markov decision processes (POMDPs). With the learned representation of causal factors, the performance of policy learning and transfer in downstream tasks will be improved. Instead of considering the hidden noise variables in the environments, our study considers the environment model learning problem in the fully observed setting and focuses on unbiased causal effect estimation in the offline dataset under behavior policies collected with selection bias. + +In offline model-based RL, the problem is called distribution shift (Yu et al., 2020; Levine et al., 2020; Chen et al., 2021) which has received great attentions. However, previous algorithms do not handle the model learning challenge directly but propose techniques to suppress policy sampling and learning in risky regions (Yu et al., 2020; Kidambi et al., 2020). Although these algorithms have made great progress in offline policy optimization in many tasks, so far, how to learn a better environment model in this scenario has rarely been discussed. + +# G EXPERIMENT DETAILS + +# G.1 SETTINGS + +G.1.1 GENERAL NEGATIVE FEEDBACK CONTROL (GNFC) + +The design of GNFC is inspired by a classic type of scenario that behavior policies $\mu$ have selection bias and easily lead to counterfactual risks: For some internet platforms, we would like to allocate budgets to a set of targets (e.g., customers or cities) to increase the engagement of the targets in the platforms. Our task is to train a model to predict targets’ feedback on engagement given targets’ features and allocated budgets. + +In these tasks, for better benefits, the online working policy (i.e., the behavior policy) will tend to cut down the budgets if targets have better engagement, otherwise, the budgets might be increased. The risk of counterfactual environment model learning in the task is that: the object with better historical engagement will be sent to smaller budgets because of the selection bias of the behavior policies, then the model might exploit this correlation for learning and get a conclusion that: increasing budgets will reduce the targets’ engagement, which violates the real causality. We construct an environment and a behavior policy to mimic the above process. In particular, the behavior policy $\mu _ { G N F C }$ is + +$$ +\mu _ { G N F C } ( x ) = \frac { ( 6 2 . 5 - \mathrm { m e a n } ( x ) ) } { 1 5 } + \epsilon , +$$ + +where $\epsilon$ is a sample noise, which will be discussed later. The environment includes two parts: + +(1) response function $M _ { 1 } ( y | x , a )$ : + +$$ +M _ { 1 } ( y | x , a ) = \mathcal { N } ( \mathrm { m e a n } ( x ) + a , 2 ) +$$ + +(2) mapping function $M _ { 2 } ( x ^ { \prime } | x , y )$ + +$$ +M _ { 2 } ( x ^ { \prime } | x , a , y ) = y - \mathrm { m e a n } ( x ) + x +$$ + +The transition function $M ^ { * }$ is a composite of $M ^ { * } ( x ^ { \prime } | x , a ) = M _ { 2 } ( x ^ { \prime } | x , a , M _ { 1 } ( y | x , a ) )$ . The behavior policies have selection bias: the actions taken are negatively correlated with the states, as illustrated in Fig. 8(a) and Fig. 8(b). We control the difficulty of distinguishing the correct causality of $x , a ,$ , and $y$ by designing different strategies of noise sampling on $\epsilon$ . In principle, with a larger number or more pronounced disturbances, there are more samples violating the correlation between $x$ and $a$ , then more samples can be used to find the correct causality. Therefore, we can control the difficulty of counterfactual environment model learning by controlling the strength of disturbance. In particular, we sample $\epsilon$ from a uniform distribution $U ( - e , e )$ with probability $p$ . That is, $\epsilon = 0$ with probability $1 - p$ and $\epsilon \sim U ( - e , e )$ with probability $p$ . Then with larger $p$ , there are more samples in the dataset violating the negative correlation (i.e., $\mu _ { G N F C } ,$ ), and with larger $e$ , the difference of the feedback will be more obvious. By selecting different $e$ and $p$ , we can construct different tasks to verify the effectiveness and ability of the counterfactual environment model learning algorithm. + +# G.1.2 THE CANCER GENOMIC ATLAS (TCGA) + +The Cancer Genomic Atlas (TCGA) is a project that has profiled and analyzed large numbers of human tumors to discover molecular aberrations at the DNA, RNA, protein, and epigenetic levels. The resulting rich data provide a significant opportunity to accelerate our understanding of the molecular basis of cancer. We obtain features, $\mathbf { x }$ , from the TCGA dataset and consider three continuous treatments as done in SCIGAN (Bica et al., 2020). Each treatment, $a$ , is associated with a set of parameters, $\mathbf { v } _ { 1 }$ , $\mathbf { v } _ { 2 }$ , $\mathbf { v } _ { 3 }$ , that are sampled randomly by sampling a vector from a standard normal distribution and scaling it with its norm. We assign interventions by sampling a treatment, $a$ , from a beta distribution, $a \mid \mathbf { x } \sim \operatorname { B e t a } \left( \alpha , \beta \right)$ . $\alpha \geq 1$ controls the sampling bias and $\textstyle { \beta = { \frac { \alpha - 1 } { a ^ { * } } } + 2 - \alpha }$ where $a ^ { * }$ is the optimal treatment. This setting of $\beta$ ensures that the mode of Beta $( \alpha , \tilde { \beta } )$ is $a ^ { * }$ . + +![](images/55abe1b08792484f5a786bb75b1b3aab1b1d67d16735214a33fde14020f480e8.jpg) +Figure 8: Illustration of information about the collected dataset in GNFC. Each color of the line denotes one of the collected trajectories. The $\mathrm { X }$ -axis denotes the timestep of a trajectory. + +The calculation of treatment response and optimal treatment are shown in Table 3. + +Table 3: Treatment response used to generate semi-synthetic outcomes for patient features $\mathbf { x }$ . In the experiments, we set $C = 1 0$ . + +
TreatmentTreatment ResponseOptimal treatment
1f1(x,a1)=C((v)Tx+12(ν)T xa1-12(ν)T x²)(v)x a= 2(v3)x
2a2))(v)x 2(v)x
3(v2)x) f3(x,a3)=C((vi)Tx+12a3(a3 -b)²,where b=0.75
+ +We conduct experiments on three different treatments separately and change the value of bias $\alpha$ to assess the robustness of different methods to treatment bias. When the bias of treatment is large, which means $\alpha$ is large, the training set contains data with a strong bias on treatment so it would be difficult for models to appropriately predict the treatment responses out of the distribution of training data. + +# G.1.3 BUDGET ALLOCATION TASK TO THE TIME PERIOD (BAT) + +We deploy GALILEO in a real-world large-scale food-delivery platform. The platform contains various food stores, and food delivery clerks. The overall workflow is as follows: the platform presents the nearby food stores to the customers and the customers make orders, i.e., purchase take-out foods from some stores on the platform. The food delivery clerks can select orders from the platform to fulfill. After an order is selected to fulfill, the delivery clerks will take the ordered take-out foods from the stores and then send the food to the customers. The platform will pay the delivery clerks (mainly in proportion to the distance between the store and the customers’ location) once the orders are fulfilled. An illustration of the workflow can be found in Fig. 9. + +However, there is an imbalance problem between the demanded orders from customers and the supply of delivery clerks to fulfill these orders. For example, at peak times like lunchtime, there will be many more demanded orders than at other times, and the existed delivery clerks might not be able to fulfill all of these orders timely. The goal of the Budget Allocation task to the Time period (BAT) is to handle the imbalance problem in time periods by sending reasonable allowances to different time periods. More precisely, the goal of BAT is to make all orders (i.e., the demand) sent in different time periods can be fulfilled (i.e., the supply) timely. + +![](images/b1e7c25ec14554cb2a78317429793502eb19c5fd3212ebf07b308c2649c646bd.jpg) +Figure 9: Illustration of the workflow of the food-delivery platform. + +To handle the imbalance problem in different time periods, in the platform, the orders in different time periods $t \in [ 0 , 1 , 2 . . . , 2 3 ]$ will be allocated with different allowances $c \in \mathcal { N } ^ { + }$ . For example, at $1 0 \mathrm { A . M }$ . (i.e., $t = 1 0$ ), we add $0 . 5 \ S$ (i.e., $c = 0 . 5$ ) allowances to all of the demanded orders. From $1 0 \mathrm { \ A . M }$ . to 11 A.M., the delivery clerks who take orders and send food to customers will receive extra allowances. Specifically, if the platform pays the delivery clerks $2 \mathfrak { H }$ for fulfilling the order, now he/she will receive $2 . 5 \ S$ . For each day, the budget of allowance $C$ is fixed. We should find the best budget allocation policy $\pi ^ { * } ( c | t )$ of the limited budget $C$ to make as many orders as possible can be taken timely. + +To find the policy, we first learn a model to reconstruct the response of allowance for each delivery clerk $\hat { M } \big ( y _ { t + 1 } \vert s _ { t } , p _ { t } , c _ { t } \big )$ , where $y _ { t + 1 }$ is the taken orders of the delivery clerks in state $s _ { t } , c _ { t }$ is the allowances, $p _ { t }$ denotes static features of the time period $t$ . In particular, the state $s _ { t }$ includes historical order-taken information of the delivery clerks, current orders information, the feature of weather, city information, and so on. Then we use a rule-based mapping function $f$ to fill the complete next time-period states, i.e., $s _ { t + 1 } = f ( s _ { t } , p _ { t } , c _ { t } , y _ { t + 1 } )$ . Here we define the composition of the above functions $\hat { M }$ and $f$ as $\hat { M } _ { f }$ . Finally, we learn a budget allocation policy based on the learned model. For each day, the policy we would like to find is: + +$$ +\begin{array} { r } { \underset { \pi } { \operatorname* { m a x } } \mathbb { E } _ { s _ { 0 } \sim { \cal S } } \left[ \displaystyle \sum _ { t = 0 } ^ { 2 3 } y _ { t } \vert \hat { M } _ { f } , \pi \right] , } \\ { \mathrm { s . t . , } \displaystyle \sum _ { t , s \in { \cal S } } c _ { t } y _ { t } \leq C } \end{array} +$$ + +In our experiment, we evaluate the degree of balancing between demand and supply by computing the averaged five-minute order-taken rate, that is the percentage of orders picked up within five minutes. Note that the behavior policy is fixed for the long term in this application. So we directly use the data replay with a small scale of noise (See Tab. 5) to reconstruct the behavior policy for model learning in GALILEO. + +Also note that although we model the response for each delivery clerk, for fairness, the budget allocation policy is just determining the allowance of each time period $t$ and keeps the allowance to each delivery clerk $s$ the same. + +# G.2 BASELINE ALGORITHMS + +The algorithm we compared are: (1) Supervised Learning (SL): training a environment model to minimize the expectation of prediction error, without considering the counterfactual risks; (2) inverse propensity weighting (IPW) (Spirtes, 2010): a practical way to balance the selection bias by re-weighting. It can be regarded as $\begin{array} { r } { \dot { \omega } = \frac { 1 } { \hat { \mu } } } \end{array}$ , where $\hat { \mu }$ is another model learned to approximate the behavior policy; (3) SCIGAN: a recent proposed adversarial algorithm for model learning for continuous-valued interventions (Bica et al., 2020). All of the baselines algorithms are implemented with the same capacity of neural networks (See Tab. 5). + +# G.2.1 SUPERVISED LEARNING (SL) + +As a baseline, we train a multilayer perceptron model to directly predict the response of different treatments, without considering the counterfactual risks. We use mean square error to estimate the perfor ance of our model so that the loss function can be expressed as $\begin{array} { r } { \boldsymbol { M } \boldsymbol { \bar { S } } \boldsymbol { E } = \frac { 1 } { n } \sum _ { i = 1 } ^ { n } \left( y _ { i } - \hat { y } _ { i } \right) ^ { 2 } } \end{array}$ $n$ $\hat { y }$ is the predicted response. In practice, we train our SL models using Adam optimizer and the initial learning rate $3 e ^ { - \bar { 4 } }$ on both datasets TCGA and GNFC. The architecture of the neural networks is listed in Tab. 5. + +# G.2.2 INVERSE PROPENSITY WEIGHTING (IPW) + +Inverse propensity weighting (Spirtes, 2010) is an approach where the treatment outcome model uses sample weights to balance the selection bias by re-weighting. The weights are defined as the inverse propensity of actually getting the treatment, which can be expressed as $\dot { \overline { { { \hat { \mu } } ( a | x ) } } }$ , where $x$ stands for the feature vectors in a dataset, $a$ is the corresponding action and ${ \hat { \mu } } ( a | x )$ indicates the action taken probability of $a$ given the features $x$ within the dataset. $\hat { \mu }$ is learned with standard supervised learning. Standard IPW leads to large weights for the points with small sampling probabilities and finally makes the learning process unstable. We solve the problem by clipping the propensity score: $\hat { \mu } \gets \operatorname* { m i n } ( \hat { \mu } , 0 . 0 5 )$ , which is common used in existing studies (Ionides, 2008). The loss function can thus be expressed as 1n Pni=1 1µˆ(ai|xi) $\begin{array} { r } { { \frac { 1 } { n } } \sum _ { i = 1 } ^ { n } { \frac { 1 } { { \hat { \mu } } \left( a _ { i } | x _ { i } \right) } } \left( y _ { i } - \hat { y } _ { i } \right) ^ { 2 } } \end{array}$ . The architecture of the neural networks is listed in Tab. 5. + +# G.2.3 SCIGAN + +SCIGAN (Bica et al., 2020) is a model that uses generative adversarial networks to learn the data distribution of the counterfactual outcomes and thus generate individualized response curves. SCIGAN does not place any restrictions on the form of the treatment-does response functions and is capable of estimating patient outcomes for multiple treatments, each with an associated parameter. SCIGAN first trains a generator to generate response curves for each sample within the training dataset. The learned generator can then be used to train an inference network using standard supervised methods. For fair comparison, we increase the number of parameters for the the open-source version of SCIGAN so that the SCIGAN model can have same order of magnitude of network parameters as GALILEO. In addition, we also finetune the hyperparameters (Tab. 4) of the enlarged SCIGAN to realize its full strength. We set num_dosage_samples 9 and $\lambda = 1 0$ . + +Table 4: Table of hyper-parameters for SCIGAN. + +
ParameterValues
Number of samples3,5,7,9, 11
入0.1,1,10, 20
+ +# G.3 HYPER-PARAMETERS + +We list the hyper-parameter of GALILEO in Tab. 5. + +# G.4 COMPUTATION RESOURCES + +We use one Tesla V100 PCIe 32GB GPU and a 32-core Intel(R) Xeon(R) Gold 5118 CPU @ 2.30GHz to train all of our model. + +Table 5: Table of hyper-parameters for all of the tasks. + +
ParameterGNFCTAGCMuJoCoBAT
hidden layers of all neural networks4455
hidden units of all neural networks256256512512
collect samples for each time of model update500050004000096000
batch size of discriminators500050004000080000
horizon50110048 (half-hours)
∈μ(also ∈D)0.0050.010.05 (0.1 for walker2d)0.05
times for discriminator update2215
times for model update11220
times for supervised learning update11420
learning rate for supervised learning1e-51e-53e-41e-5
Y0.990.00.990.99
clip-ratioNANNANNAN0.1
max DkL0.0010.0010.001NAN
optimization algorithm (the first term of Eq.23)TRPOTRPOTRPOPPO
+ +# H ADDITIONAL RESULTS + +# H.1 ALL OF THE RESULT TABLE + +We give the result of CNFC in Tab. 8, TCGA in Tab. 9, BAT in Tab. 7, and MuJoCo in Tab. 6. + +# H.2 AVERAGED RESPONSE CURVES + +We give the averaged responses for all of the tasks and the algorithms in Fig. 16 to Fig. 23. We randomly select $20 \%$ of the states in the dataset and equidistantly sample actions from the action space for each sampled state, and plot the averaged predicted feedback of each action. The real response is slightly different among different figure as the randomly-selected states for testing is different. We sample 9 points in GNFC tasks and 33 points in TAGC tasks for plotting. + +# H.3 ABLATION STUDIES + +In Sec. 4.3 and Appx. E.1, we introduce several techniques to develop a practical GALILEO algorithm. Based on task $\mathtt { e 0 } . 2 _ { - \mathtt { p } 0 . 0 5 }$ in GNFC, we give the ablation studies to investigate the effects of these techniques. We first compare two variants that do not handle the assumptions violation problems: (1) NO_INJECT_NOISE: set $\epsilon _ { \mu }$ and $\epsilon _ { D }$ to zero, which makes the overlap assumption not satisfied;; (2) SINGLE_SL: without replacing the second term in Eq. 8 with standard supervised learning even when the output probability of $D$ is far away from 0.5. Besides, we introduced several tricks inspired by GAIL and give a comparison of these tricks and GAIL: (3) ONE_STEP: use one-step reward instead of cumulative rewards (i.e., Q and V; see Eq. 24 and Eq. 25) for re-weighting, which is implemented by set $\gamma$ to 0; (4) SINGE_DIS: remove $T _ { \varphi _ { 1 } ^ { * } } ( x , a )$ and replace it with $\mathbb { E } _ { M _ { \theta } } \left[ T _ { \varphi _ { 0 } ^ { * } } ( x , a , x ^ { \prime } ) \right]$ , which is inspired by GAIL that uses a value function as a baseline instead of using another discriminator; (5) PURE_GAIL: + +![](images/bd32c544f33fa909bf10ef2d6bc1392ce066275899883232e0b00bb79348eae2.jpg) +Figure 10: Illustration of the ablation studies. The error bars are the standard error. + +remove the second term in Eq. 8. It can be regarded as a naive adoption of GAIL and a partial implementation of GALILEO. + +We summarize the results in Fig. 10. Based on the results of NO_INJECT_NOISE and SINGLE_SL, we can see that handling the assumption violation problems is important and will increase the ability on counterfactual queries. The results of PURE_GAIL tell us that the partial implementation of GALILEO is not enough to give stable predictions on counterfactual data; On the other hand, the result of ONE_STEP also demonstrates that embedding the cumulative error of one-step prediction is helpful for GALILEO training; Finally, we also found that SINGLE_DIS nearly has almost no effect on the results. It suggests that, empirically, we can use $\mathbb { E } _ { M _ { \theta } } \left[ T _ { \varphi _ { 0 } ^ { * } } ( x , a , x ^ { \prime } ) \right]$ as a replacement for $T _ { \varphi _ { 1 } ^ { * } } ( x , a )$ , which can reduce the computation costs of the extra discriminator training. + +Table 6: The root mean square errors on MuJoCo tasks. We bold the lowest error for each task. “medium” dataset is used for training, while “expert” and “medium-replay” datasets are just used for testing. $\pm$ follows the standard deviation of three seeds. + +
TASKHalfCheetah
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.378 ± 0.0032.287± 0.0051.411 ± 0.037
OFF-SL0.404 ± 0.0013.311 ± 0.0552.246 ± 0.016
TASKWalker2d
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.49 ± 0.001.514 ± 0.0020.968 ± 0.004
OFF-SL0.467 ± 0.0041.825 ± 0.0611.239 ± 0.004
TASKHopper
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.037 ± 0.0020.322 ± 0.0360.408 ± 0.003
OFF-SL0.034 ± 0.0010.464 ± 0.0210.574 ± 0.008
+ +# H.4 WORST-CASE PREDICTION ERROR + +In theory, GALILEO increases the generalization ability by focusing on the worst-case samples’ training to achieve AWRM. To demonstrate the property, we propose a new metric named Mean-Max Square Error (MMSE): $\begin{array} { r } { \bar { \mathfrak { z } } \left[ \operatorname* { m a x } _ { a \in \mathcal { A } } \left( M ^ { * } ( x ^ { \prime } | x , a ) - \bar { M } ( x ^ { \prime } | x , a ) \right) ^ { 2 } \right] } \end{array}$ and give the results of MMSE for GNFC in Tab. 10 and for TCGA in Tab. 11. + +# H.5 DETAILED RESULTS IN THE MUJOCO TASKS + +We select 3 environments from D4RL (Fu et al., 2020) to construct our model learning tasks. We compare it with a typical transition model learning algorithm used in the previous offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020), which is a variant of standard supervised learning. We name the method OFF-SL. We train models in datasets HalfCheetah-medium, Walker2dmedium, and Hopper-medium, which are collected by a behavior policy with 1/3 performance to the expert policy, then we test them in the corresponding expert dataset. We plot the converged results and learning curves of GALILEO and OFF-SL in three MuJoCo tasks in Tab. 6 and Fig. 11 respectively. + +In Fig. 11, we can see that both OFF-SL and GALILEO perform well in the training datasets. OFF-SL can even reach a bit lower error in halfcheetah and walker2d. However, when we verify the models through “expert” and “medium-replay” datasets, which are collected by other policies, the performance of GALILEO is significantly more stable and better than OFF-SL. As the training continues, OFF-SL even gets worse and worse. In summary, GALILEO reaches significantly better performances in the expert dataset: the averaged declines of root MSE in three environments are $5 6 . 5 \%$ , $4 9 . 2 \%$ , and $3 4 . 8 \%$ . However, whether in GALILEO or OFF-SL, the performance for testing is at least $2 \mathbf { x }$ worse than in the training dataset. The phenomenon indicates that although GALILEO can make better performances for counterfactual queries, the risks of using the models are still large and still challenging to be further solved. + +![](images/5547f6091a98561bc79b830999d27857990e83ebd0ac546794410f3a7f6be8af.jpg) +Figure 11: Illustration of learning curves of the MuJoCo Tasks. The X-axis record the steps of the environment model update, and the Y-axis is the corresponding prediction error. The figures with titles ending in “(train)” means the dataset is used for training while the titles ending in “(test)” means the dataset is just used for testing. The solid curves are the mean reward and the shadow is the standard error of three seeds. + +![](images/7fc39a335c2cd42ff20c40f5c27345da54078256b068d1373bfc35f96bdfa28c.jpg) +Figure 12: Illustration of relationship between user feedback and the actions of the offline dataset in the real-world food-delivery platform. + +# H.6 DETAILED RESULTS IN THE BAT TASK + +The core challenge of the environment model learning in BAT tasks is similar to the challenge in Fig. 1. Specifically, the behavior policy in BAT tasks is a human-expert policy, which will tend to increase the budget of allowance in the time periods with a lower supply of delivery clerks, otherwise will decrease the budget (Fig. 12 gives an instance of this phenomenon in the real data). + +Since there is no oracle environment model for querying, we have to describe the results with other metrics. + +![](images/b869771cf06ee02caa745de9087c18bd06b71c1215b64025a89e0777903965ba.jpg) +Figure 13: Illustration of the response curves in the 6 cities. Although the ground-truth curves are unknown, through human expert knowledge, we know that it is expected to be monotonically increasing . + +First, we review whether the tendency of the response curve is consistent. In this application, with a larger budget of allowance, the supply will not be decreased. As can be seen in Fig. 13, the tendency of GALILEO’s response is valid in 6 cities but almost all of the models of SL give opposite directions to the response. If we learn a policy through the model of SL, the optimal solution is canceling all of the allowances, which is obviously incorrect in practice. + +Second, we conduct randomized controlled trials (RCT) in one of the testing cities. Using the RCT samples, we can evaluate the correctness of the sort order of the model predictions via Area Under the Uplift Curve (AUUC) (Betlei et al., 2020). To plot AUUC, we first sort the RCT samples based on the predicted treatment effects. Then the cumulative treatment effects are computed by scanning the sorted sample list. If the sort order of the model predictions is better, the sample with larger treatment effects will be computed early. Then the area of AUUC will be larger than the one via a random sorting strategy. The result of AUUC show GALILEO gives a reasonable sorting to the RCT samples (see Fig. 14). + +![](images/eb37f5b144b993f4cd093333bca2ea018eace1f580556caafc4eb6ea290aeb5e.jpg) +Figure 14: Illustration of the AUUC result for BAT. + +Finally, we search for the optimal policy via the cross-entropy method planner (Hafner et al., 2019) based on the learned model. We test the online supply improvement in 6 cities. The algorithm compared is a human-expert policy, which is also the behavior policy of the offline datasets. We conduct online A/B tests for each of the cities. For each test, we randomly split a city into two partitions, one is for deploying the optimal policy learned from the GALILEO model, and the other is as a control group, which keeps the human-expert policy as before. Before the intervention, we collect 10 days’ observation data and compute the averaged five-minute order-taken rates as the baselines of the treatment and control group, named $b ^ { t }$ and $b ^ { c }$ respectively. Then we start intervention and observe the five-minute order-taken rate in the following 14 days for the two groups. The results of the treatment and control groups are $y _ { i } ^ { t }$ and $y _ { i } ^ { c }$ respectively, where $i$ denotes the $i$ -th day of the deployment. The percentage points of the supply improvement are computed via difference-in-difference (DID): + +$$ +\frac { \sum _ { i } ^ { T } ( y _ { i } ^ { t } - b ^ { t } ) - ( y _ { i } ^ { c } - b ^ { c } ) } { T } \times 1 0 0 , +$$ + +where $T$ is the total days of the intervention and $T = 1 4$ in our experiments. + +Table 7: Results on BAT. We use City- $\mathbf { X }$ to denote the experiments on different cities. “pp” is an abbreviation of percentage points on the supply improvement. + +
targetCity-ACity-BCity-C
supply improvement+1.63pp+0.79pp+0.27pp
targetCity-DCity-ECity-F
supply improvement+0.2pp+0.14pp+0.41pp
+ +The results are summarized in Tab. 7. The online experiment is conducted in 14 days and the results show that the policy learned with GALILEO can make better (the supply improvements are from 0.14 to 1.63 percentage points) budget allocation than the behavior policies in all the testing cities. We give detailed results which record the supply difference between the treatment group and the control group in Fig. 15. + +![](images/1a2796121e8ea865a6cc920895fd3318ac583274799e2549cdb6430e7ce60dd6.jpg) +Figure 15: Illustration of the daily responses in the A/B test in the 6 cities. + +Table 8: $\sqrt { M I S E }$ results on GNFC. We bold the lowest error for each task. $\pm$ is the standard deviation of three random seeds. + +
e1_p1e0.2_p1e0.05_p1
GALILEO5.17 ± 0.064.73 ± 0.134.70 ± 0.02
SL5.15 ± 0.234.73 ± 0.3123.64 ± 4.86
IPW5.22 ± 0.095.50 ± 0.015.02 ± 0.07
SCIGAN7.05 ± 0.526.58 ± 0.5818.55 ± 3.50
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO5.03 ± 0.094.72 ± 0.054.87 ± 0.15
SL5.21 ± 0.636.74 ± 0.1533.52 ±1.32
IPW5.27 ± 0.055.69 ± 0.0020.23 ± 0.45
SCIGAN16.07 ± 0.2712.07 ± 1.9319.27 ± 10.72
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO5.23 ± 0.415.01 ± 0.086.17 ± 0.33
SL5.89 ± 0.8814.25 ± 3.4837.50 ± 2.29
IPW5.21 ± 0.015.52 ± 0.4431.95 ± 0.05
SCIGAN11.50 ± 7.7613.05 ± 4.1925.74 ± 8.30
+ +Table 9: $\sqrt { M I S E }$ results on TCGA. We bold the lowest error for each task. $\pm$ is the standard deviation of three random seeds. + +
t0_bias_2.0 t0_bias_s_20.0 t0_bias_50.0
GALILEO SL0.34 ± 0.05 0.67 ± 0.13 0.38 ± 0.13 1.50 ± 0.312.04 ± 0.12 3.06 ± 0.65
IPW SCIGAN6.57 ± 1.16 0.74 ± 0.056.88 ± 0.30 5.84 ± 0.71 2.74 ± 0.35 3.19 ± 0.09
t1_bias_2.0t1_bias_6.0 t1_bias_8.0
GALILEO SL0.43 ± 0.05 0.47 ± 0.050.25 ± 0.02 0.21 ± 0.04 1.33 ± 0.97 1.18 ± 0.73
IPW3.67 ± 2.370.54 ± 0.13 2.69 ± 1.17
SCIGAN0.45 ± 0.251.08 ± 1.04 1.01 ± 0.77
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO
1.46 ± 0.090.85 ± 0.04 0.46 ± 0.01
SL IPW0.81 ± 0.143.74 ± 2.04 3.59 ± 0.14
+ +Table 10: $\sqrt { M M S E }$ results on GNFC. We bold the lowest error for each task. $\pm$ is the standard deviation of three random seeds. + +
e1_p1e0.2_p1e0.05_p1
GALILEO3.86 ± 0.033.99 ± 0.014.07 ± 0.03
SL5.73 ± 0.335.80 ± 0.2818.78 ± 3.13
IPW4.02 ± 0.054.15 ± 0.1222.66 ± 0.33
SCIGAN8.84 ± 0.5412.62 ± 2.1724.21 ± 5.20
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO4.13 ± 0.104.11 ± 0.154.21 ± 0.15
SL5.87 ± 0.437.44 ± 1.1329.13 ± 3.44
IPW4.12 ± 0.026.12 ± 0.4830.96 ± 0.17
SCIGAN12.87 ± 3.0214.59 ± 2.1324.57 ± 3.00
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO4.39 ± 0.204.34 ± 0.205.26 ± 0.29
SL6.12 ± 0.4314.88 ± 4.4130.81 ± 1.69
IPW13.60 ± 7.8326.27 ± 2.6732.55 ± 0.12
SCIGAN9.19 ± 1.0415.08 ± 1.2617.52 ± 0.02
+ +Table 11: $\sqrt { M M S E }$ results on TCGA. We bold the lowest error for each task. $\pm$ is the standard deviation of three random seeds. + +
t0_bias_2.0t0_bias_20.0 tO_bias_50.0
GALILEO1.56 ± 0.041.96 ± 0.53 3.16 ± 0.13
SL1.92 ± 0.672.31 ± 0.19 5.11 ± 0.66
IPW7.42 ± 0.465.36 ± 0.96 5.38 ± 1.24
SCIGAN2.11 ± 0.475.23 ±0.27 5.59 ± 1.02
tl_bias_2.0tl_bias_6.0 tl_bias_8.0
GALILEO1.43 ± 0.061.09 ± 0.05 1.36 ± 0.36
SL1.12 ± 0.153.65 ± 1.91 3.96 ± 1.81
IPW1.14 ± 0.11 0.90 ± 0.092.04 ± 0.99
SCIGAN3.32 ± 0.88 4.74 ± 2.125.17 ± 2.42
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO3.77 ± 0.35 3.99 ± 0.402.08 ± 0.60
SL2.70 ± 0.67 8.33 ± 5.059.70 ± 3.12
IPW2.92 ± 0.15 3.90 ± 0.174.47 ± 2.16
SCIGAN3.82 ± 2.12 1.83 ± 1.493.62 ± 4.9
+ +![](images/0f4a66ef4baf70e73f3c6d49b3d583ae88a89e44b95edef40bd86d6ec7c7718d.jpg) +Figure 16: Illustration of the averaged response curves of Supervised Learning (SL) in TCGA. + +![](images/2abae9ea1c75f9d0de485507d99f7e3fc9e0798cab13f63e1ba494cd5a148ac7.jpg) +Figure 17: Illustration of the averaged response curves of Supervised Learning (SL) in GNFC. + +![](images/e95af2c71f12750c1e7632c99166d6e0f6065e19f98d35100fc5980516143630.jpg) +Figure 18: Illustration of the averaged response curves of Inverse Propensity Weighting (IPW) in TCGA. + +![](images/cef37522bd881e3dbc177930bf4e98a750e94226e9d0ed7fc4238330fda9b47a.jpg) +Figure 19: Illustration of the averaged response curves of Inverse Propensity Weighting (IPW) in GNFC. + +![](images/fa953ff9d5f43062141219c074a774984b11e3b159955de3d0d2558401086b92.jpg) +Figure 20: Illustration of the averaged response curves of SCIGAN in TCGA. + +![](images/f7f3fe5f9e09c0042241c841726d9b8c5004cb2f7cb0769552686dc09c60b91f.jpg) +Figure 21: Illustration of the averaged response curves of SCIGAN in GNFC. + +![](images/f4bdbf1e7453b14d1e9a36552e1a5b2ea53c70db26825936362204bac36ee869.jpg) +Figure 22: Illustration of the averaged response curves of GALILEO in TCGA. + +![](images/2290b75ec5c556d9f23c1eec2979a5a76f05ef08e2d626f32314d745cbc6a3c0.jpg) +Figure 23: Illustration of the averaged response curves of GALILEO in GNFC. \ No newline at end of file diff --git a/parse/dev/zT5T9gHpGI/zT5T9gHpGI_content_list.json b/parse/dev/zT5T9gHpGI/zT5T9gHpGI_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..da8dbc9b7eeb6e47ae5b9021a8f7fa2ffc79e206 --- /dev/null +++ b/parse/dev/zT5T9gHpGI/zT5T9gHpGI_content_list.json @@ -0,0 +1,4555 @@ +[ + { + "type": "text", + "text": "ADVERSARIAL COUNTERFACTUAL ENVIRONMENT MODEL LEARNING ", + "text_level": 1, + "bbox": [ + 176, + 99, + 823, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Anonymous authors Paper under double-blind review ", + "bbox": [ + 183, + 170, + 398, + 198 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 210, + 544, + 226 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "A good model for action-effect prediction, i.e., the environment model, is essential for sample-efficient policy learning, in which the agent can take numerous free trials to find good policies. Currently, the model is commonly learned by fitting historical transition data through empirical risk minimization (ERM). However, we discover that simple data fitting can lead to a model that will be totally wrong in guiding policy learning due to the selection bias in offline dataset collection. In this work, we introduce weighted empirical risk minimization (WERM) to handle this problem in model learning. A typical WERM method utilizes inverse propensity scores to re-weight the training data to approximate the target distribution. However, during the policy training, the data distributions of the candidate policies can be various and unknown. Thus, we propose an adversarial weighted empirical risk minimization (AWRM) objective that learns the model with respect to the worst case of the target distributions. We implement AWRM in a sequential decision structure, resulting in the GALILEO model learning algorithm. We also discover that GALILEO is closely related to adversarial model learning, explaining the empirical effectiveness of the latter. We apply GALILEO in synthetic tasks and verify that GALILEO makes accurate predictions on counterfactual data. We finally applied GALILEO in real-world offline policy learning tasks and found that GALILEO significantly improves policy performance in real-world testing. ", + "bbox": [ + 233, + 242, + 766, + 506 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 517, + 336, + 531 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "A good environment model is important for sample-efficient decision-making policy learning techniques like reinforcement learning (RL) (James & Johns, 2016). The agent can take trials with this model to find better policies, then the costly real-world trial-and-errors can be saved (James & Johns, 2016; Yu et al., 2020) or completely waived (Shi et al., 2019). In this process, the core of the models is to answer queries on counterfactual data unbiasedly, that is, given states, correctly answer what might happen if we were to carry out actions unseen in the training data (Levine et al., 2020). ", + "bbox": [ + 174, + 535, + 825, + 617 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Requiring counterfactual queries makes the environment model learning essentially different from standard supervised learning (SL) which directly fits the offline dataset. In real-world applications, the offline data is often collected with selection bias, that is, for each state, each action might be chosen unfairly. Seeing the example in Fig. 1(a), to keep the ball following a target line, a behavior policy will use a smaller force when the ball’s location is closer to the target line. When a dataset is collected with selection bias, the association between the (location) states and (force) actions will make SL hard to identify the correct causal relationship of the states and actions to the next states respectively. Then when we query the model with counterfactual data, the predictions might be catastrophic failures. In Fig. 1(c), it mistakes that smaller forces will increase the ball’s next location. ", + "bbox": [ + 174, + 625, + 825, + 750 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Generally speaking, the problem corresponds to a challenge of training the model in one dataset but testing in another dataset with a shifted distribution (i.e., the dataset generated by counterfactual queries), which is beyond the SL’s capability as it violates the independent and identically distributed (i.i.d.) assumption. The problem is widely discussed in causal inference for individual treatment effects (ITEs) estimation in many scenarios like patients’ treatment selection (Imbens, 1999; Alaa & van der Schaar, 2018). ITEs are the effects of treatments on individuals, which are measured by treating each individual under a uniform policy and evaluate the effect differences. Practical solutions use weighted empirical risk minimization (WERM) to handle this problem (Jung et al., 2020; Shimodaira, 2000; Hassanpour & Greiner, 2019). In particular, they estimate an inverse propensity score (IPS) to re-weight the training data to approximate the data distribution under a uniform policy. Then a model is trained under the reweighted data distribution. The distribution-shift problem is solved as ITEs estimation and model training are under the same distribution. ", + "bbox": [ + 174, + 757, + 825, + 922 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/b68193a7fb3b9605331279446f1bc56a0b45f7422ed57eced22fd9b4e8ab8a5c.jpg", + "image_caption": [ + "Figure 1: An example of selection bias and predictions under counterfactual queries. Subfigure (a) shows how the data is collected: a ball locates in a 2D plane whose position is $( x _ { t } , y _ { t } )$ at time $t$ . The ball will move to $( x _ { t + 1 } , y _ { t + 1 } )$ according to $x _ { t + 1 } = x _ { t } + 1$ and $y _ { t + 1 } \\sim \\mathcal { N } ( y _ { t } + a _ { t } , 2 )$ . Here, $a _ { t }$ is chosen by a control policy $a _ { t } \\sim \\mathcal { N } ( ( \\phi - y _ { t } ) / 1 5 , 0 . 0 5 )$ parameterized by $\\phi$ , which tries to keep the ball near the line $y = \\phi$ . In Subfigure (a), $\\phi$ is set to 62.5. Subfigure (b) shows the collected training data (grey dashed line) and the two learned models’ prediction of the next position of $y$ . All the models discovered the relation that the corresponding next $y$ will be smaller with a larger action. However, the truth is not because the larger $a _ { t }$ causes a smaller $y _ { t + 1 }$ , but the policy selects a small $a _ { t }$ when $y _ { t }$ is close to the target line. When we estimate the response curves by fixing $y _ { t }$ and reassigning action $a _ { t }$ with other actions $a _ { t } + \\Delta a$ , where $\\Delta a \\in [ - 1 , 1 ]$ is a variation of action value, the model of SL will exploit the association and give opposite responses, while in AWRM and its practical implementation GALILEO, the predictions are closer to the ground truths. The result is in Subfigure (c), where the darker a region is, the more samples are fallen in. " + ], + "image_footnote": [], + "bbox": [ + 194, + 53, + 787, + 202 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "The selection bias can be regarded as an instance of the problem called “distributional shift” in offline model-based RL, which has also received great attention (Levine et al., 2020; Yu et al., 2020; Kidambi et al., 2020; Chen et al., 2021). However, previous methods, where naive supervised learning is used for environment model learning, ignore the problem in environment model learning but handling the problem by suppressing the policy exploration and learning in risky regions. Although these methods have made great progress in many tasks, so far, how to learn a better environment model that can alleviate the problem for faithful offline policy optimization has rarely been discussed. ", + "bbox": [ + 173, + 396, + 825, + 452 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 453, + 565, + 522 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this work, for faithful offline policy optimization, we introduce WERM to environment model learning. The extra challenge of model learning for policy optimization is that we have to query numerous different policies’ feedback besides the uniform policy for finding a good policy. Thus the target data distribution to reweight can be various and unknown. To solve the problem, we propose an objective called adversarial weighted empirical risk minimization (AWRM). AWRM introduces adversarial policies, of which the corresponding counterfactual dataset has the maximal prediction error of the model. For each iteration, the model is learned to be as small prediction risks as possible under the adversarial counterfactual dataset. However, the adversarial counterfactual dataset cannot be obtained in the offline setting, thus we derive an approximation of the counterfactual data distribution queried by the optimal adversarial policy and use a variational representation to give a tractable solution to learn a model from the approximated data distribution. As a result, we derive a practical approach named Generative Adversarial offLIne counterfactuaL Environment mOdel learning (GALILEO) for AWRM. Fig. 2 shows the difference in the prediction errors learned by these algorithms. We also discover that GALILEO is closely related to existing generative-adversarial model learning techniques, explaining the effectiveness of the latter. ", + "bbox": [ + 174, + 529, + 565, + 875 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/e21350e3d9b064285fe113e1adaa5132ea73dc8398bcf24bc5da581bdb24fdad.jpg", + "image_caption": [ + "Figure 2: An illustration of the prediction error in counterfactual datasets. The prediction risks is measured with mean square error (MSE). The error of SL is small only in training data ( $\\phi =$ 62.5) but becomes much larger in the dataset “far away from” the training data. AWRM-oracle selects the oracle worst counterfactual dataset for training for each iteration (pseudocode is in Alg. 1) which reaches small MSE in all datasets and gives correct response curves (Fig. 1(c)). GALILEO approximates the optimal adversarial counterfactual data distribution based on the training data and model. Although the MSE of GALILEO is a bit larger than SL in the training data, in the counterfactual datasets, the MSE is on the same scale as AWRM-oracle. " + ], + "image_footnote": [], + "bbox": [ + 580, + 460, + 805, + 588 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Experiments are conducted in two synthetic and two realistic environments. The results in the synthetic environments show that GALILEO can reconstruct correct responses for counterfactual queries. The evaluation results in two realistic environments also demonstrate that GALILEO has better ability in counterfactual query compared with baselines. We finally search for a policy based on the learned model in a real-world online platform. The policy significantly improves performance in concerned business indicators. ", + "bbox": [ + 176, + 882, + 823, + 924 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 176, + 103, + 823, + 145 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2 RELATED WORK ", + "text_level": 1, + "bbox": [ + 176, + 155, + 344, + 171 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We give related adversarial algorithms for model learning in the following and leave other related work in Appx. F. GANTIE (Yoon et al., 2018) uses a generator to fill counterfactual outcomes for each data pair and a discriminator to judge the source (treatment group or control group) of the filled data pair. The generator is trained to minimize the output of the discriminator. GANITE is trained until the discriminator cannot determine which of the components is the factual outcome. Bica et al. (2020) propose SCIGAN to extend GANITE to continuous treatment effect estimation (a.k.a., dosage-response estimation) via a hierarchical discriminator architecture. In real-world applications, environment model learning based on Generative Adversarial Imitation Learning (GAIL) has also been adopted for sequential decision-making problems (Ho & Ermon, 2016). GAIL is first proposed for policy imitation (Ho & Ermon, 2016), which uses the imitated policy to generate trajectories by interacting with the environment. The policy is learned with the trajectories through RL which maximizes the cumulative rewards given by the discriminator. Shi et al. (2019); Chen et al. (2019); Shang et al. (2019) use GAIL for environment model learning by regarding the environment model as the generator and the behavior policy as the “environment” in standard GAIL. These studies empirically demonstrate that adversarial model learning algorithms have better generalization ability for counterfactual queries, while our study reveals the connection between adversarial model learning and the WERM through IPS. Our derived practical algorithm GALILEO is closely related to the existing adversarial model learning algorithms, explaining the effectiveness of the latter. ", + "bbox": [ + 174, + 184, + 825, + 433 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 PRELIMINARIES ", + "text_level": 1, + "bbox": [ + 176, + 449, + 339, + 465 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.1 SINGLE-STEP INDIVIDUALIZED TREATMENT EFFECTS ESTIMATION AND WEIGHTEDEMPIRICAL RISKS MINIMIZATION", + "text_level": 1, + "bbox": [ + 178, + 474, + 797, + 502 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We first introduce individualized treatment effects (ITEs) estimation (Rosenbaum & Rubin, 1983), which can be regarded as the scenario in which the environment model has only a single step. ITEs are typically defined as $I T E ( x ) : = \\mathbb { E } [ M ^ { * } ( y | x , 1 ) | A = 1 , X = x ] - \\mathbb { E } [ M ^ { * } ( y | x , 0 ) | A = 0 , X = x ]$ where $y$ is the feedback of the environment $M ^ { * } ( y | x , a )$ , $X$ denotes the state vector containing pre-treatment covariates (such as age and weight), $A$ denotes the treatment variable which is the action intervening to the state $X$ , and $A$ should be sampled from a uniform policy. In the twotreatment scenario, $A$ is in $\\{ 0 , 1 \\}$ where 1 is the action to intervene and 0 is the action to do nothing. A correct ITEs estimation should be done in Randomized Controlled Trials (RCT) in which we have the same probability of samples of $A = 1$ and $A = 0$ for each $X$ . Here we use lowercase $x , a$ and $y$ to denote samples of random variables $X , A$ and $Y$ , and use $x , A$ and $\\mathcal { V }$ to denote space of the samples. In practice, we prefer to estimate ITEs under observational studies. In observational studies, datasets are pre-collected from the real world by a behavior policy such as a human-expert policy. In this case, a common approach for estimating the ITEs can be $\\hat { I T E } ( x _ { i } ) = a _ { i } ( y _ { i } ^ { F } - M ( x _ { i } , 1 - a _ { i } ) ) + ( 1 - a _ { i } ) ( M ( x _ { i } , 1 - a _ { i } ) - y _ { i } ^ { F } )$ (Shalit et al., 2017) in deterministic prediction, where $x _ { i }$ and $y _ { i } ^ { F }$ denote the covariate and factual feedback of the $i$ -th sample, and $M \\in \\mathcal { M }$ denotes an approximated feedback model. $\\mathcal { M }$ is the space of the model. In this formulation, the training set is an empirical factual data distribution $\\bar { P _ { F } } = \\{ ( x _ { i } , a _ { i } ) \\} _ { i } ^ { n }$ and the testing set is an empirical counterfactual data distribution $P _ { C F } = \\{ ( x _ { i } , 1 - a _ { i } ) \\} _ { i } ^ { n }$ . If $a$ does not sample from a discrete uniform policy, i.e., the policy has selection bias, $P _ { F }$ and $P _ { C F }$ will be two different distributions, which violate the i.i.d. assumption of standard supervised learning. In stochastic prediction, $\\hat { I T E } ( x ) = \\mathbb { E } [ M ( y | x , 1 ) ] - \\mathbb { E } [ M ( y | x , 0 ) ]$ and the counterfactual distribution for testing is the dataset with action sampling from a uniform policy. ", + "bbox": [ + 173, + 508, + 825, + 820 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Generally speaking, in ITEs estimation, the risks of queries under counterfactual data are caused by the gap between the policy in training and testing data distributions. Without further processing, minimizing the empirical risks cannot guarantee the counterfactual-query risks being minimized. Assuming that the policy in training data $\\mu$ satisfies $\\mu ( a | x ) > 0 , \\forall a \\stackrel { - } { \\in } \\mathcal { \\bar { A } } , \\forall x \\in \\mathcal { X }$ (often named overlap assumption), a classical solution to handle the above problem is weighted empirical risk minimization (WERM) through an inverse propensity scoring (IPS) term $\\omega$ (Shimodaira, 2000; Assaad et al., 2021; Hassanpour & Greiner, 2019; Jung et al., 2020): ", + "bbox": [ + 174, + 827, + 825, + 924 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Definition 3.1. The learning objective of WERM through IPS is formulated as ", + "bbox": [ + 171, + 103, + 689, + 118 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/84b11ff30828b8a64799e15e9dedc010134f03df389f6ff864ac494a2de8e6c2.jpg", + "text": "$$\n\\begin{array} { r l } { \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } L ( M ) = \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } \\mathbb { E } _ { { x } , { a } , { y } \\sim { p } _ { M ^ { * } } } [ \\omega ( { x } , { a } ) \\ell ( M ( { y } | { x } , { a } ) , { y } ) ] , } & { } \\\\ { { s . t . } \\quad \\omega ( { x } , { a } ) = \\frac { \\beta ( a | { x } ) } { \\mu ( a | { x } ) } , } & { } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 302, + 122, + 692, + 184 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $\\beta$ and $\\mu$ denote the policies in testing and training domains, and $p _ { M ^ { * } } ^ { \\mu }$ is the joint probability $p _ { M ^ { * } } ^ { \\mu } ( x , a , y ) : = \\rho _ { 0 } ( x ) \\mu ( a | x ) M ^ { * } ( y | x , a )$ in which $\\rho _ { 0 } ( x )$ is the distribution of state. $\\mathcal { M }$ is the model space. $\\ell$ is a loss function for model learning. ", + "bbox": [ + 174, + 188, + 823, + 231 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The $\\omega$ is also known as importance sampling (IS) weight, which corrects the sampling bias. In this objective, $\\omega$ is to align the training data distribution to the testing data. By selecting different $\\hat { \\omega }$ to approximate $\\omega$ to learn the model $M$ , current environment model learning algorithms for ITEs estimation are fallen into the framework. In standard supervised learning and some works for ITEs estimation (Wager & Athey, 2018; Weiss et al., 2015), $\\hat { \\omega } ( x , a ) = 1$ as the distribution-shift problem is ignored. In Shimodaira (2000); Assaad et al. (2021); Hassanpour & Greiner (2019), $\\begin{array} { r } { \\omega = \\frac { \\mathrm { i } } { \\hat { \\mu } } } \\end{array}$ (i.e., $\\beta$ a uniform policy) for balancing treatment and control group, where $\\hat { \\mu }$ is an approximation of behavior policy $\\mu$ . Note that it is a reasonable weight in ITEs estimation: ITEs are defined to evaluate the effect of each state between treatment and control behavior under a uniform policy. ", + "bbox": [ + 173, + 241, + 825, + 369 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.2 SEQUENTIAL DECISION-MAKING SETTING ", + "text_level": 1, + "bbox": [ + 174, + 375, + 511, + 388 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Decision-making processes in a sequential environment are often formulated into Markov Decision Process (MDP) (Sutton & Barto, 1998). MDP depicts an agent interacting with the environment through actions. In the first step, states are sampled from an initial state distribution $x _ { 0 } \\sim \\rho _ { 0 } ( x )$ . Then at each time-step $t \\in \\{ 0 , 1 , 2 , \\ldots \\}$ , the agent takes an action $a _ { t } \\in { \\mathcal { A } }$ through a policy $\\pi ( a _ { t } | x _ { t } ) \\in \\Pi$ based on the state $x _ { t } \\in \\mathcal { X }$ , then the agent receives a reward $r _ { t }$ from a reward function $r ( x _ { t } , a _ { t } ) \\in$ $\\mathbb { R }$ and transits to the next state $x _ { t + 1 }$ given by a transition function $M ^ { \\ast } ( x _ { t + 1 } | x _ { t } , a _ { t } )$ built in the environment. $\\Pi$ , $\\mathcal { X }$ , and $\\mathcal { A }$ denote the policy, state, and action spaces. ", + "bbox": [ + 173, + 393, + 825, + 491 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4 METHOD ", + "text_level": 1, + "bbox": [ + 174, + 500, + 282, + 516 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In this section, we first propose a new offline model-learning objective based on Def. 3.1 for policy optimization tasks in Sec. 4.1; In Sec. 4.2, we derive a tractable solution to the proposed objective; Finally, we give a practical implementation in Sec. 4.3. ", + "bbox": [ + 174, + 521, + 825, + 563 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4.1 PROBLEM FORMULATION ", + "text_level": 1, + "bbox": [ + 174, + 573, + 392, + 587 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "For offline policy optimization, we require the environment model to have generalization ability in counterfactual queries since we need to query numerous different policies’ correct feedback from $M$ . Referring to the formulation of WERM through IPS in Def. 3.1, policy optimization requires $M$ to minimize counterfactual-query risks under numerous unknown different policies rather than a specific target policy $\\beta$ . More specifically, the question is: If $\\beta$ is unknown and can be varied, how should we reduce the risks in counterfactual queries? In this article, we call the model learning problem in this setting “counterfactual environment model learning” and propose a new objective to handle the problem. To be compatible with multi-step environment model learning, we first define a generalized WERM through IPS based on Def. 3.1. ", + "bbox": [ + 173, + 594, + 826, + 719 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Definition 4.1. Given the MDP transition function $M ^ { * }$ that satisfies $M ^ { * } ( x ^ { \\prime } | x , a ) > 0 , \\forall x \\in \\mathcal { X } , \\forall a \\in$ $\\mathcal { A } , \\forall x ^ { \\prime } \\in \\mathcal { X }$ and $\\mu$ satisfies $\\mu ( a | x ) > 0 , \\forall a \\in \\mathcal { A } , \\forall x \\in \\mathcal { X }$ , the learning objective of generalized WERM through IPS is formulated as ", + "bbox": [ + 174, + 723, + 825, + 766 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/6b922044bab2cecce8ce3cfa5b9946e9dc58b17f41c4fcc348163596a1c8d808.jpg", + "text": "$$\n\\begin{array} { r l } & { \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } L ( M ) = \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\mu } } [ \\omega ( x , a , x ^ { \\prime } ) \\ell _ { M } ( x , a , x ^ { \\prime } ) ] , } \\\\ & { ~ s . t . ~ \\omega ( x , a , x ^ { \\prime } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } ) } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 302, + 770, + 694, + 835 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $\\rho _ { M ^ { * } } ^ { \\mu }$ is the training data distribution (collected by policy $\\mu$ ), $\\rho _ { M ^ { * } } ^ { \\beta }$ is the testing data distribution (collected by policy $\\beta$ ). We define $\\ell _ { M } ( x , a , x ^ { \\prime } ) : = \\ell ( M ( x ^ { \\prime } | x , a ) , x ^ { \\prime } ) ^ { - }$ for brevity. ", + "bbox": [ + 169, + 840, + 825, + 872 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In an MDP, given any policy $\\pi$ , $\\rho _ { M ^ { * } } ^ { \\pi } ( x , a , x ^ { \\prime } ) ~ = ~ \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a )$ where $\\rho _ { M ^ { \\ast } } ^ { \\pi } ( x )$ denotes the occupancy measure of $x$ for policy $\\pi$ , which can be defined as $\\rho _ { M ^ { \\ast } } ^ { \\pi } ( x ) : =$ $\\begin{array} { r } { ( 1 - \\gamma ) \\mathbb { E } _ { x _ { 0 } \\sim \\rho _ { 0 } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\dot { \\gamma } ^ { t } \\mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \\right] } \\end{array}$ (Sutton $\\&$ Barto, 1998; Ho & Ermon, 2016) where $ { \\mathrm { P r } } ^ { \\pi } \\left[ x _ { t } = x | x _ { 0 } , M ^ { * } \\right]$ is the state visitation probability that the policy $\\pi$ visits $x$ at time-step $t$ by executing in the environment $M ^ { * }$ and starting at the state $x _ { 0 }$ , and $\\gamma \\in [ 0 , 1 ]$ is the discount factor. Here we also define $\\rho _ { M ^ { * } } ^ { \\pi } ( x , a ) : = \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x )$ for simplicity. In this definition, $\\omega$ can be rewritten as: $\\begin{array} { r } { \\omega ( x , a , x ^ { \\prime } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x ) \\beta ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x ) \\mu ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a ) } = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) } } \\end{array}$ . In single-step environments, for any policy $\\pi$ , ρπM∗ (x) = ρ0(x). Then we have ω(x, a, x′) = ρ0(x)β(a|x)ρ (x)µ(a|x) , and the objective is degraded to Eq. 1. Therefore, Def. 3.1 is a special case of this generalized form. ", + "bbox": [ + 173, + 881, + 825, + 926 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 102, + 826, + 205 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Since $\\beta$ for counterfactual queries is unknown and can be varied in policy optimization, to reduce the risks in counterfactual queries in this scenario, we introduce adversarial policies which can induce the worst performance of the model predictions and propose to optimize WERM under the adversarial policies. In particular, we propose Adversarial Weighted empirical Risk Minimization (AWRM) based on Def. 4.1 to handle this problem. ", + "bbox": [ + 173, + 212, + 825, + 281 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Definition 4.2. Given the MDP transition function $M ^ { * }$ , the learning objective of adversarial Weighted empirical risk minimization through IPS is formulated as ", + "bbox": [ + 171, + 284, + 825, + 313 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/39944d2b5a594f42b7761eb3e27008f21b546a70714b76f8a617a7dfda623ea0.jpg", + "text": "$$\n\\begin{array} { r l } & { \\hat { \\boldsymbol M } ^ { * } = \\underset { M \\in \\mathcal { M } } { \\mathrm { \\operatorname* { m i n } } } \\underset { \\beta \\in \\Pi } { \\mathrm { m a x } } \\boldsymbol L ( \\rho _ { M ^ { * } } ^ { \\beta } , M ) = \\underset { M \\in \\mathcal { M } } { \\mathrm { m i n } } \\underset { \\beta \\in \\Pi } { \\mathrm { m a x } } \\mathbb { E } _ { \\boldsymbol x , \\boldsymbol a , \\boldsymbol x ^ { \\prime } \\sim \\boldsymbol \\rho _ { M ^ { * } } ^ { \\mu } } [ \\omega ( \\boldsymbol x , \\boldsymbol a | \\rho _ { M ^ { * } } ^ { \\beta } ) \\ell _ { M } ( \\boldsymbol x , \\boldsymbol a , \\boldsymbol x ^ { \\prime } ) ] , } \\\\ & { } \\\\ & { s . t . \\quad \\omega ( \\boldsymbol x , \\boldsymbol a | \\rho _ { M ^ { * } } ^ { \\beta } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( \\boldsymbol x , \\boldsymbol a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( \\boldsymbol x , \\boldsymbol a ) } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 218, + 315, + 776, + 381 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where the re-weighting term $\\omega ( x , a | \\rho _ { M ^ { * } } ^ { \\beta } )$ is conditioned on the data distribution $\\rho _ { M ^ { * } } ^ { \\beta }$ of the adversarial policy $\\beta$ . In the following, we will ignore $\\rho _ { M ^ { * } } ^ { \\beta }$ and use $\\omega ( x , a )$ for brevity. ", + "bbox": [ + 173, + 385, + 823, + 420 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "According to the definition of MDP, $\\omega ( x , a , x ^ { \\prime } ) = \\omega ( x , a )$ since the transition probability in the ratio can be canceled. Eq. 3 minimizes the maximum model loss under all counterfactual data distributions $\\rho _ { M ^ { \\ast } } ^ { \\beta } ( x , a , x ^ { \\prime } ) , \\beta \\in \\Pi$ to guarantee the generalization ability for counterfactual data queried by policies in $\\Pi$ . ", + "bbox": [ + 174, + 428, + 825, + 487 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.2 TRACTABLE AWRM SOLUTION ", + "text_level": 1, + "bbox": [ + 174, + 496, + 434, + 510 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In this section, we propose a tractable solution to optimize Eq. 3. The full derivations can be found in the Appx. A. We choose $\\ell _ { M }$ to be the negative log-likelihood loss and derive the solution to Eq. 3: ", + "bbox": [ + 173, + 517, + 823, + 546 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/072350e4aea7ec2b69e997e13744c49a14155f1d140c22d93a2707e2f9fb329f.jpg", + "text": "$$\nL ( \\rho _ { M ^ { \\ast } } ^ { \\beta } , M ) = \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { \\ast } } ^ { \\mu } } \\left[ \\omega ( x , a | \\rho _ { M ^ { \\ast } } ^ { \\beta } ) \\mathbb { E } _ { M ^ { \\ast } } \\left( - \\log M ( x ^ { \\prime } | x , a ) \\right) \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 279, + 549, + 715, + 577 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\mathbb { E } _ { M ^ { * } } \\left[ \\cdot \\right]$ denotes $\\mathbb { E } _ { x ^ { \\prime } \\sim M ^ { * } \\left( x ^ { \\prime } \\mid x , a \\right) }$ [·]. The core problem is how to construct the data distribution $\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }$ of the best-response policy $\\beta ^ { * }$ in $M ^ { * }$ as it is costly to get extra data from $M ^ { * }$ in real-world applications. Instead of deriving the optimal $\\beta ^ { * }$ , our solution is to offline estimate the optimal adversarial distribution $\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }$ with respect to $M$ , then we can construct a surrogate objective to optimize $M$ without directly querying the real environment $M ^ { * }$ . ", + "bbox": [ + 173, + 578, + 825, + 657 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.2.1 OPTIMAL ADVERSARIAL DATA DISTRIBUTION APPROXIMATION ", + "text_level": 1, + "bbox": [ + 174, + 660, + 676, + 674 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Ideally, given any $M$ , it is obvious that the optimal $\\beta$ is the one that makes $\\rho _ { M ^ { \\ast } } ^ { \\beta } ( x , a )$ assign all densities to the point that has the largest negative log-likelihood. However, searching for the maximum is impractical, especially in continuous space. To give a relaxed but tractable solution, we add an $L _ { 2 }$ regularizer to the original objective Eq. 3: ", + "bbox": [ + 173, + 678, + 826, + 734 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/92654ebd7b17c680d8289ce5867177b202ca333373c1c15e4f9dfc0f08704048.jpg", + "text": "$$\n\\operatorname* { m i n } _ { d \\in M } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M ) = \\operatorname* { m i n } _ { M \\in M } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { * } } ^ { \\mu } } \\Big [ \\omega ( x , a ) \\mathbb { E } _ { M ^ { * } } \\big [ - \\log M ( x ^ { \\prime } | x , a ) \\big ] \\Big ] - \\frac { \\alpha } { 2 } \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 181, + 736, + 838, + 765 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\alpha$ denotes the regularization coefficient of $\\rho _ { M ^ { * } } ^ { \\beta }$ and $\\begin{array} { r } { \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } = \\int _ { \\mathcal { X } , \\mathcal { A } } ( \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ) ^ { 2 } \\mathrm { d } a \\mathrm { d } x . } \\end{array}$ Then we can approximate the optimal distribution $\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ via Lemma. 4.3. ", + "bbox": [ + 174, + 784, + 826, + 824 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Lemma 4.3. Given any $M$ in $\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M )$ , the distribution of the ideal best-response policy $\\bar { \\beta } ^ { * }$ satisfies: ", + "bbox": [ + 169, + 827, + 823, + 854 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/37dc1f9700a687f33b539434a21e83ea1994226b49396acf0eefa63e62814c23.jpg", + "text": "$$\n\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ,\n$$", + "text_format": "latex", + "bbox": [ + 302, + 853, + 694, + 881 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )$ is the Kullback-Leibler $( K L )$ divergence between $M ^ { * } ( \\cdot | x , a )$ and $M ( \\cdot | x , a )$ , $H _ { M ^ { \\ast } } ( x , a )$ denotes the entropy of $M ^ { * } ( \\cdot | x , a )$ , and $\\alpha _ { M }$ is the regularization coefficient $\\alpha$ in Eq. $^ { 4 }$ and also as a normalizer of Eq. 5. ", + "bbox": [ + 176, + 882, + 825, + 924 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Intuitively, $\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ has larger densities on the data where the divergence between the approximation model and the real model (i.e., $D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) )$ is larger or the stochasticity of the real model (i.e., $H _ { M ^ { * } }$ ) is larger. However, the integral process of $D _ { K L }$ in Eq. 5 is intractable in the offline setting as it explicitly requires the conditional probability function of $M ^ { * }$ . Our solution to solve the problem is utilizing the offline dataset $\\mathcal { D } _ { \\mathrm { r e a l } }$ as the empirical joint distribution $\\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } )$ and adopting practical techniques for distance estimation on two joint distributions, like GAN (Goodfellow et al., 2014; Nowozin et al., 2016), to approximate Eq. 5. To adopt that solution, we should first transform Eq. 5 into a form under joint distributions. Without loss of generality, we introduce an intermediary policy $\\kappa$ , of which $\\mu$ can be regarded as a specific instance. Then we have $M ( x ^ { \\prime } | x , a ) =$ $\\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) / \\rho _ { M } ^ { \\kappa } ( x , a )$ for any $M$ if $\\rho _ { M } ^ { \\kappa } ( x , a ) > 0$ . Assuming $\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) > 0$ if $\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) > 0$ , which will hold when $\\kappa$ overlaps with $\\mu$ , then Eq. 5 can transform to: ", + "bbox": [ + 173, + 101, + 826, + 255 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/78af9ac575fb1379ea94dfaa342e4030712e344f1b8af727eb02a909a481a75c.jpg", + "text": "$$\n\\displaystyle \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\Bigg ( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } { \\rho _ { M } ^ { \\kappa } ( x , a ) } - H _ { M ^ { * } } ( x , a ) \\right) \\Bigg ) ,\n$$", + "text_format": "latex", + "bbox": [ + 194, + 260, + 803, + 299 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $\\alpha _ { 0 } ( x , a ) = \\alpha _ { M } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )$ . We notice that the form $\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } } { \\rho _ { M } ^ { \\kappa } } } \\end{array}$ ρκM∗ρκ is the integrated function in reverse $\\mathrm { K L }$ divergence, which is an instance of $f$ function in $f$ -divergence (Ali & Silvey, 1966). Replacing that form with $f$ function, we obtain a generalized representation of $\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ : ", + "bbox": [ + 174, + 306, + 826, + 359 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/43d2dd2f890e8f95bed4967ff118330b972404d5f4c844faa96a213360d47f59.jpg", + "text": "$$\n\\tilde { \\bar { \\nu } } _ { M ^ { * } } ^ { * } : = \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\left( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) - H _ { M ^ { * } } ( x , a ) \\right) \\right.\n$$", + "text_format": "latex", + "bbox": [ + 181, + 364, + 839, + 404 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "to Reviewer bcJS ", + "bbox": [ + 838, + 207, + 954, + 220 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $f : \\mathbb { R } _ { + } \\to \\mathbb { R }$ is a convex and lower semi-continuous (l.s.c.) function. $\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ gives a generalized representation of the optimal adversarial data distribution to maximize the error of the model. Based on Eq. 6, we have a surrogate objective of AWRM which can avoid querying $M ^ { * }$ to construct $\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }$ : ", + "bbox": [ + 173, + 424, + 826, + 473 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Theorem 4.4. Let $\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ as the data distribution of the best-response policy $\\bar { \\beta } ^ { * }$ in Eq. 4 under model $M _ { \\theta }$ parameterized by $\\theta$ , then we can find the optimal $\\theta ^ { * }$ of $\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}$ (Eq. 4) via iteratively optimizing the objective $\\begin{array} { r } { \\theta _ { t + 1 } = \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}$ , where $\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ is approximated via the last-iteration model $M _ { \\theta _ { t } }$ . Based on Corollary A.7, we derive an upper bound objective for $\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}$ : ", + "bbox": [ + 173, + 477, + 826, + 563 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/300a71d94016b5fd8af5a721594f0ad5429351b188668424c87f05a5a0abc29e.jpg", + "text": "$$\n\\begin{array} { r } { \\iota _ { t + 1 } = \\underset { \\theta } { \\operatorname* { m i n } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\frac { - 1 } { \\alpha _ { 0 } ( x , a ) } \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\left( f \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) + H _ { M ^ { * } } ( x , a ) \\right) \\right] , } \\\\ { W ( x , a , x ^ { \\prime } ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 569, + 826, + 626 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $\\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } }$ [·] denotes $\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\kappa } }$ [·], $f$ is a l.s.c function satisfying $f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }$ , and $\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )$ . ", + "bbox": [ + 176, + 631, + 825, + 665 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Thm. 4.4 approximately achieve AWRM by using $\\kappa$ and a pseudo-reweighting module $W$ . $W$ assigns learning propensities for data points with larger differences between distributions $\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }$ and $\\rho _ { M ^ { * } } ^ { \\kappa }$ . By adjusting the weights, the learning process will exploit subtle errors in any data point, whatever how many proportions it contributes, to correct potential generalization errors on counterfactual data. ", + "bbox": [ + 173, + 672, + 825, + 733 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.2.2 TRACTABLE SOLUTION TO THM. 4.4 ", + "text_level": 1, + "bbox": [ + 174, + 746, + 482, + 762 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In Thm. 4.4, the term $\\begin{array} { r } { f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\big ) - f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\big ) } \\end{array}$ is still intractable. To solve the problem, first, we resort to the first-order approximation of $f$ . Given some $u \\in ( 1 - \\xi , 1 + \\xi ) , \\xi > 0$ , we have ", + "bbox": [ + 173, + 768, + 826, + 810 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/2b517abc4c2eca59111a5c29db0c8e310899041f086df0a62064d90652b53b69.jpg", + "text": "$$\nf ( u ) \\approx f ( 1 ) + f ^ { \\prime } ( u ) ( u - 1 ) ,\n$$", + "text_format": "latex", + "bbox": [ + 398, + 815, + 598, + 833 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $f ^ { \\prime }$ is the first-order derivative of $f$ . By Taylor’s formula and the fact that $f ^ { \\prime } ( u )$ of the generator function $f$ is bounded in $( 1 - \\xi , 1 + \\xi )$ , the approximation error is no more than $\\mathcal { O } ( \\xi ^ { 2 } )$ . Let $\\begin{array} { r } { u = \\frac { p ( x ) } { q ( x ) } } \\end{array}$ in Eq. 7, the pattern $f ( { \\frac { p ( x ) } { q ( x ) } } )$ in Thm. 4.4 can be converted to $\\begin{array} { r } { f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) ( \\frac { p ( x ) } { q ( x ) } - 1 ) + f ( 1 ) } \\end{array}$ . $\\frac { p ( x ) } { q ( x ) }$ can be approximated by sampling from datasets and $f ^ { \\prime } \\big ( \\frac { p ( x ) } { q ( x ) } \\big )$ can be approximated by the corresponding variational representation $T _ { \\varphi ^ { * } }$ according to Lemma A.9 (Nowozin et al., 2016) (see ", + "bbox": [ + 173, + 839, + 826, + 925 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Appendix for details). Based on any specific $f$ -function, we can represent $T _ { \\varphi }$ with a discriminator $D _ { \\varphi }$ . However, Thm. 4.4 holds only when $\\forall x \\in { \\mathcal { X } }$ , $f ^ { \\prime } ( x ) \\leq 0$ . It can be verified that the instance $f ( u ) = u \\log u - ( u + 1 ) \\log ( u + 1 )$ , $T _ { \\varphi } ( u ) = \\log D _ { \\varphi } ( u )$ satisfy this condition (see Tab. 2 in the Appendix). We select that instance for the tractable solution. In summary, by computing $f ( \\cdot )$ with first-order approximation (Eq. 7) and leveraging the variational representation $T _ { \\varphi ^ { * } } = \\log { \\hat { D } _ { \\varphi ^ { * } } }$ to approximate $f ^ { \\prime } \\Bigl ( \\frac { { { \\rho } _ { M _ { \\theta _ { t } } } ^ { \\kappa } } } { { { \\rho } _ { M ^ { * } } ^ { \\kappa } } } \\Bigr )$ , we can optimize the surrogate objective in Thm. 4.4 via: ", + "bbox": [ + 173, + 103, + 826, + 200 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/7a43c0d6a6b17809cafa4cd97f6788b226db8b13f0b3e94e29288ac22d2e93fc.jpg", + "text": "$$\n\\begin{array} { r l } { { \\iota _ { + 1 } } = \\mathop { \\operatorname* { m a x } } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\log M \\theta ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\left( H _ { M ^ { * } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] , } & { } \\\\ { { s . t . } } & { \\varphi _ { 0 } ^ { * } = \\mathop { \\operatorname* { a r g m a x } } _ { \\varphi _ { 0 } ^ { * } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } & { ( 8 ) } \\\\ & { \\varphi _ { 1 } ^ { * } = \\mathop { \\operatorname { a r g m a x } } _ { \\varphi _ { 1 } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 171, + 204, + 874, + 275 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where $A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ = \\ \\log D _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ - \\ \\log D _ { \\varphi _ { 1 } ^ { * } } ( x , a ) , \\ \\mathbb { E } _ { \\rho _ { M } ^ { \\kappa } } [ \\cdot ]$ is a simplification of $\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M } ^ { \\kappa } } [ \\cdot ]$ and $\\varphi _ { 0 }$ and $\\varphi _ { 1 }$ are the parameters of $T _ { \\varphi _ { 0 } }$ and $T _ { \\varphi _ { 1 } }$ . Based on Lemma A.9, we have $\\begin{array} { r } { f ^ { \\prime } ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } ) \\approx \\log D _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) } \\end{array}$ and $f ^ { \\prime } ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } ) \\approx \\log D _ { \\varphi _ { 1 } ^ { * } } ( x , a )$ . Note that in the process, we ignore the term $\\alpha _ { 0 } ( x , a )$ for simplifying the objective. The discussion on the impacts of removing $\\alpha _ { 0 } ( x , a )$ is left in App. B. In Eq. 8, the pseudo-reweighting module $W$ in Thm. 4.4 is split into two terms in the RHS of the equation. The first term is a generative adversarial training objective regarding $M _ { \\theta }$ as the generator, while the second term is WERM through $H _ { M ^ { * } } - A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } }$ . ", + "bbox": [ + 173, + 280, + 826, + 396 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.3 PRACTICAL IMPLEMENTATION ", + "text_level": 1, + "bbox": [ + 176, + 405, + 426, + 419 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "To give a practical implementation of the solution, extra two assumptions introduced in the process of modeling should be handled: ", + "bbox": [ + 173, + 426, + 823, + 454 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "First, the approximation of Eq. 7 holds only when $\\textstyle { \\frac { p ( x ) } { q ( x ) } }$ is close to 1, which might not be satisfied. To handle the problem, we inject a naive supervised learning loss and replace the second term of the objective Eq. 8 when the expectation of $D _ { \\varphi _ { 0 } }$ in $\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }$ is far away from 0.5 $f ^ { \\prime } ( 1 ) = \\log 0 . 5 )$ ; ", + "bbox": [ + 174, + 455, + 825, + 503 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Second, the overlap assumption of $\\cdot$ . In practice, we need to use the real-world data to construct the distribution $\\rho _ { M ^ { * } } ^ { \\kappa }$ and the generative data to construct $\\cdot$ . In the offline model-learning setting, we only have a real-world dataset $\\cdot$ collected by the behavior policy $\\cdot$ . We can learn a policy $\\cdot$ via behavior cloning based on $\\mathcal { D }$ (Pomerleau, 1991; Ho & Ermon, 2016) and let $\\cdot$ be the policy $\\kappa$ . Then we can regard $\\mathcal { D }$ as the empirical data distribution of $\\rho _ { M ^ { * } } ^ { \\kappa }$ and the trajectories collected by $\\hat { \\mu }$ in the model $M _ { \\theta _ { t } }$ as the empirical data distribution of $\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }$ . But the assumption $\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\mu ( a | x ) > 0$ might not be satisfied. To handle the problem, in behavior cloning, we model $\\hat { \\mu }$ with a Gaussian distribution and constrain the lower bound of the variance with a small value $\\epsilon _ { \\mu } > 0$ to keep the assumption holding. Besides, we add small Gaussian noises $\\mathcal { N } ( 0 , \\epsilon _ { D } )$ to the inputs of $D _ { \\varphi }$ to handle the mismatch between $\\rho _ { M ^ { * } } ^ { \\mu }$ and $\\rho _ { M ^ { * } } ^ { \\hat { \\mu } }$ due to $\\epsilon _ { \\mu }$ . ", + "bbox": [ + 173, + 505, + 825, + 654 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In Eq. 8, $H _ { M ^ { * } }$ is unknown in advance. In practice, we use $H _ { M _ { \\theta } }$ to estimate it. More specifically, the neural network of $M _ { \\theta }$ can be modeled with a Gaussian distribution. The variance of the Gaussian distribution is modeled with global variables $\\Sigma$ for each dimension of output. We estimate $H _ { M ^ { * } }$ with the closed-form solution of Gaussian entropy through $\\Sigma$ . ", + "bbox": [ + 174, + 659, + 825, + 715 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Based on the above techniques, we propose Generative Adversarial offLIne counterfactuaL Environment mOdel learning (GALILEO) for environment model learning. GALILEO can be adopted in both single-step and sequential environment model learning. The detailed implementation and comparison to previous adversarial methods are in Appx. E. ", + "bbox": [ + 174, + 722, + 825, + 779 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5 EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 174, + 784, + 328, + 799 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In this section, we first conduct experiments in synthetic environments (Bica et al., 2020) to verify GALILEO on counterfactual queries 1 and the compatibility of GALILEO in sequential and single-step environments. We select Mean Integrated Square Error MISE $=$ $\\begin{array} { r } { \\bar { \\mathbb { E } } \\left[ \\int _ { \\mathcal { A } } \\left( M ^ { * } ( y | x , a ) - \\bar { M } ( y | x , a ) \\right) ^ { 2 } \\mathrm { d } a \\right] } \\end{array}$ as the metric, which is a commonly used metric to measure the accuracy in counterfactual queries by considering the prediction errors in the whole action space. Then we analyze the benefits of the implementation techniques described in Sec. 4.3 and ", + "bbox": [ + 173, + 808, + 826, + 901 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "to Reviewer bcJS to Reviewer bcJS ", + "bbox": [ + 838, + 563, + 954, + 577 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 838, + 700, + 954, + 714 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/1b046d05ed7b7da16cd32af8ae1f8b8b0181d988d9e217a6214be2ede6d9f372.jpg", + "image_caption": [ + "Figure 3: Illustration of the performance in GNFC and TCGA. The grey bar denotes the standard error ( $\\mathrm { \\times 0 . 3 }$ for brevity) of 3 random seeds. " + ], + "image_footnote": [], + "bbox": [ + 194, + 55, + 795, + 195 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "the problems without them. Finally, we deploy GALILEO in two complex environments: MuJoCo in Gym (Todorov et al., 2012) and a real-world food-delivery platform to test the performance of GALILEO in difficult tasks. The algorithms compared are: (1) Supervised Learning (SL): using standard empirical risk minimization for model learning; (2) Inverse Propensity Weighting (IPW) (Spirtes, 2010): a standard implementation of WERM based IPS; (3) eStimating the effects of Continuous Interventions using GANs (SCIGAN) (Bica et al., 2020): an adversarial algorithms for model learning used for causal effect estimation, which can be roughly regarded as a partial implementation of GALILEO (Refer to Appx. E.2). We give a detailed description in Appx. G.2. ", + "bbox": [ + 173, + 236, + 825, + 348 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5.1 TEST IN SYNTHETIC ENVIRONMENTS ", + "text_level": 1, + "bbox": [ + 176, + 364, + 473, + 378 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Test in sequential environments Since there does not exist a task specifically designed for sequential environment model learning with selection bias, we construct a task, General Negative Feedback Control (GNFC), which can represent a classic type of task with policies having selection bias. Fig. 1(a) is also an example of GNFC. We give detailed motivation, the effect of selection bias, and other details in Appx. G.1.1. We construct tasks on GNFC by adding behavior policies $\\mu$ with different scales of uniform noise $U ( - e , e )$ with different probabilities $p$ . In particular, with $e \\in \\{ 1 . 0 , 0 . 2 , 0 . 0 5 \\}$ and $p \\in \\{ 1 . 0 , 0 . 2 , 0 . 0 5 \\}$ , we construct 9 tasks and name them with the format of $^ { \\ast } \\mathrm { e } ^ { \\ast } \\mathrm { \\_ p ^ { \\ast } } ^ { \\ast }$ . For example, $\\Theta ^ { 1 } \\mathrm { \\mathrm { - } } \\Theta ^ { 0 } \\cdot 2$ is the task with behavior policy injecting with $U ( - 1 , 1 )$ with 0.2 probability. The results of GNFC tasks are summarized in Fig. 3(a) and the detailed results can be found in Tab. 8. The results show that the property of the behavior policy (i.e., $e$ and $p$ ) dominates the generalization ability of the baseline algorithms. When $e = 0 . 0 5$ , almost all of the baselines fail and give a completely opposite response curve (see Fig. 4(a) and Appx. H.2). IPW still perform well when $0 . 2 \\leq e \\leq 1 . 0$ but fails when $e = 0 . 0 5 , p < = 0 . 2$ . We also found that SCIGAN can reach a better performance than other baselines when $e = 0 . 0 5 , p < = 0 . 2$ , but the results in other tasks are unstable. GALILEO is the only algorithm that is robust to the selection bias and outputs correct response curves in all of the tasks. Based on the experiment, we also indicate that the commonly used overlap assumption is unreasonable to a certain extent especially in real-world applications since it is impractical to inject noises into the whole action space. The problem of overlap assumption being violated should be taken into consideration otherwise the algorithm will be hard to use in practice if it is sensitive to the noise range. ", + "bbox": [ + 173, + 385, + 825, + 662 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Test in single-step environments Previous experiments on counterfactual environment model learning are based on single-step semi-synthetic data simulation (Bica et al., 2020). Since GALILEO is compatible with single-step environment model learning, we select the same task named TCGA in Bica et al. (2020) to test GALILEO. Based on three synthetic response functions in TCGA, we construct 9 tasks by choosing different parameters of selection bias on $\\mu$ which is constructed with beta distribution, and design a coefficient $c$ to control the selection bias of the beta distribution. We name the tasks with the format of “t?_bias?”. For example, t1_bias2 is the task with the first response functions and $c = 2$ . The detail of TCGA is in Appx. G.1.2. The results of TCGA tasks are summarized in Fig. 3(b) and the detailed results can be found in Tab. 9 in Appendix. We found the phenomenon in this experiment is similar to the one in GNFC, which demonstrates the compatibility of GALILEO to single-step environments. We also found that the results of IPW are unstable in this experiment. It might be because the behavior policy is modeled with beta distribution while the propensity score $\\hat { \\mu }$ is modeled with Gaussian distribution. Since IPW directly reweight loss function with $\\stackrel { \\mathrm { ~ \\tiny ~ 1 ~ } } { \\hat { \\mu } }$ , the results are sensitive to the error on $\\hat { \\mu }$ . GALILEO also models $\\hat { \\mu }$ with Gaussian distribution but the results are more stable since GALILEO does not re-weight through $\\hat { \\mu }$ explicitly. ", + "bbox": [ + 174, + 674, + 550, + 827 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/2f2933e9556a0a707044a2d135038204ddd64c137dda2d4f04c33d29fb14eee6.jpg", + "image_caption": [ + "Figure 4: Illustration of the averaged response curves. " + ], + "image_footnote": [], + "bbox": [ + 568, + 659, + 816, + 784 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 827, + 825, + 924 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/8ea8b67bfcc3d24ffbedb995e97b37b863570855e8aa4a67d2ca9de37106e6f8.jpg", + "table_caption": [ + "Table 1: Results of policy performance directly optimized through standard SAC (Haarnoja et al., 2018) using the learned dynamics models and deployed in MuJoCo environments. MAX-RETURN is the policy performance of SAC in the MuJoCo environments, and “avg. norm.” is the averaged normalized return of the policies in the 9 tasks, where the returns are normalized to lie between 0 and 100, where a score of 0 corresponds to the worst policy, and 100 corresponds to MAX-RETURN. " + ], + "table_footnote": [], + "table_body": "
TaskHopperWalker2dHalfCheetah|avg. norm.
HorizonH=10H=20H=40H=10H=20H=40H=10H=20H=40/
GALILEO13.0 ± 0.133.2 ± 0.153.5± 1.211.7±0.229.9 ± 0.361.2 ± 3.40.7±0.2-1.1 ± 0.2-14.2 ± 1.451.1
SL4.8±0.53.0±0.24.6±0.210.7±0.220.1±0.837.5±6.70.4±0.5-1.1± 0.6-13.2 ± 0.321.1
IPW5.9 ±0.74.1 ± 0.65.9±0.24.7 ± 1.12.8 ±3.914.5 ± 1.41.6± 0.20.5 ± 0.8-11.3 ± 0.919.7
SCIGAN12.7 ± 0.129.2 ± 0.646.2 ±5.28.4 ± 0.59.1 ± 1.71.0 ± 5.81.2 ± 0.3-0.3 ±1.0-11.4 ± 0.341.8
MAX-RETURN13.2 ±0.033.3±0.271.0±0.514.9 ± 1.360.7 ± 11.1221.1 ± 8.92.6±0.113.3 ± 1.149.1 ± 2.3100.0
", + "bbox": [ + 176, + 128, + 821, + 212 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 220, + 821, + 250 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Response curve visualization We plot the averaged response curves which are constructed by equidistantly sampling action from the action space and averaging the feedback of the states in the dataset as the averaged response. Parts of the results in Fig. 4 (all curves can be seen in Appx. H.2). For those tasks where baselines fail in reconstructing response curves, GALILEO not only reaches a better MISE score but reconstructs almost exact responses. ", + "bbox": [ + 173, + 253, + 825, + 324 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Ablation studies In Sec. 4.3, we introduce several techniques to develop a practical GALILEO algorithm. Based on task $\\mathtt { e 0 } . 2 _ { - \\mathtt { p } 0 . 0 5 }$ of GNFC, we give the ablation studies to investigate the effects of these techniques. As the main-body space is limited, we leave the results in Appx. H.3. ", + "bbox": [ + 176, + 328, + 823, + 371 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "5.2 TEST IN COMPLEX ENVIRONMENTS ", + "text_level": 1, + "bbox": [ + 176, + 377, + 462, + 392 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In MuJoCo tasks MuJoCo is a benchmark task in Gym (Todorov et al., 2012; Brockman et al., 2016) where we need to control a robot with specific dynamics to complete some tasks (e.g., standing or running). We select 3 environment from D4RL (Fu et al., 2020) to construct our model learning tasks. We compare it with a standard transition model learning algorithm used in the previous offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020), which is a variant of supervised learning. We name the method OFF-SL. Besides, we also implement IPW and SCIGAN as the baselines. In D4RL benchmark, only the “medium” tasks is collected with a fixed policy, i.e., the behavior policy is with 1/3 performance to the expert policy), which is most matching to our proposed problem. So we train models in datasets HalfCheetah-medium, Walker2d-medium, and Hopper-medium. We trained the models with the same gradient steps and saved the models. ", + "bbox": [ + 174, + 395, + 825, + 535 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We first verify the generalization ability of the models by adopting them into offline model-based RL. Instead of designing sophisticated tricks to suppress policy exploration and learning in risky regions as current offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020) do, we just use the standard SAC algorithm Haarnoja et al. (2018) to exploit the models for policy learning to strictly verify the ability of the models. Unfortunately, we found that the compounding error will still be inevitably large in the 1,000-step rollout, which is the standard horizon in MuJoCo tasks, leading all models to fail to derive a reasonable policy. To better verify the effects of models on policy optimization, we learn and evaluate the policies with three smaller horizons: $H \\in \\{ 1 0 , 2 0 , 4 0 \\}$ . ", + "bbox": [ + 173, + 541, + 825, + 654 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "The results are listed in Tab. 1. We first averaged the normalized return (refer to “avg. norm.”) under each task, and we can see that the policy obtained by GALILEO is significantly higher than other models (the improvements are $24 \\%$ to $161 \\%$ ). At the same time, we found that SCIGAN performed better in policy learning, while IPW performed similarly to SL. This is in line with our expectations, since IPW only considers the uniform policy as the target policy for debiasing, while policy optimization requires querying a wide variety of policies. Minimizing the prediction risks only under a uniform policy cannot yield a good environment model for policy optimization. Besides, in IPW, the cumulative effects of policy on the state distribution are ignored. On the other hand, SCIGAN, as a partial implementation of GALILEO (refer to Appx. E.2), also roughly achieves AWRM and considers the cumulative effects of policy on the state distribution, so its overall performance is better; In addition, we find that GALILEO achieves significant improvement in 6 of the 9 tasks. But in HalfCheetah, IPW works slightly better. However, compared with MAX-RETURN, it can be found that all methods fail to derive reasonable policies because their policies’ performances are far away from the optimal policy. By further visualizing the trajectories, we found that all the learned policies just keep the cheetah standing in the same place or even going backward. This phenomenon is also similar to the results in MOPO (Yu et al., 2020). In MOPO’s experiment in the medium datasets, the truncated-rollout horizon used in Walker and Hopper for policy training is set to 5, while HalfCheetah has to be set to the minimal value: 1. These phenomena indicate that HalfCheetah may still have unknown problems, resulting in the generalization bottleneck of the models. Besides, we also test the prediction error of the learned model in corresponding unseen “expert” and “medium-replay” datasets. The detailed results are in Appx. H.5. ", + "bbox": [ + 173, + 659, + 825, + 924 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/48ff8ddb26f31d38f842a347595907ec3933c9a964af1fd467fd13b1ca24d29a.jpg", + "image_caption": [ + "Figure 5: An illustration of the performance in BAT tasks. In Fig. 5(a) demonstrate the averaged response curves of the SL and GALILEO model in City A. It is expected to be monotonically increasing through our prior knowledge. In Fig. 5(b), the model with larger areas above the “random” line makes better predictions in randomized-controlled-trials data (Betlei et al., 2020). Fig. 5(c) shows the daily responses in the A/B test in City A. The complete results are in Appx. H.6. " + ], + "image_footnote": [], + "bbox": [ + 194, + 55, + 799, + 194 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 271, + 821, + 299 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In a real-world platform We finally deploy GALILEO in a real-world large-scale food-delivery platform. The goal of the platform is to balance the demand from take-out food orders and the supply of delivery clerks, i.e., helping delivery clerks fulfill more orders by giving reasonable strategies. We focus on a Budget Allocation task to the Time period (BAT) in the platform (see Appx. G.1.3 for details). The goal of the BAT task is to handle the imbalance problem between the demanded orders from customers and the supply of delivery clerks in different time periods by allocating reasonable allowances to those time periods. The core challenge of the environment model learning in BAT tasks is similar to the challenge in Fig. 1. Specifically, the behavior policy in BAT tasks is a human-expert policy, which tends to increase the budget of allowance in the time periods with a lower supply of delivery clerks, otherwise tends to decrease the budget (Fig. 12 gives a real-data instance of this phenomenon). ", + "bbox": [ + 174, + 308, + 825, + 460 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We first learn a model to predict the supply of delivery clerks (measured by fulfilled order amount) on given allowances. Although the SL model can efficiently fit the offline data, the tendency of the response curve is easily to be incorrect. As can be seen in Fig. 5(a), with a larger budget of allowance, the prediction of the supply is decreased in SL, which obviously goes against our prior knowledge. This is because, in the offline dataset, the corresponding supply will be smaller when the allowance is larger. It is conceivable that if we learn a policy through the model of SL, the optimal solution is canceling all of the allowances, which is obviously incorrect in practice. On the other hand, the tendency of GALILEO’s response is correct. Fig. 13 plots all the results in 6 cities. ", + "bbox": [ + 173, + 468, + 825, + 579 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Second, we conduct randomized controlled trials (RCT) in one of the testing cities. Using the RCT samples, we can evaluate the generalization ability of the model predictions via Area Under the Uplift Curve (AUUC) (Betlei et al., 2020), which measure the correctness of the sort order of the model prediction in RCT samples. The AUUC further show that GALILEO gives a reasonable sort order on the supply prediction (see Fig. 5(b)) while the standard SL technique fails to complete this task. ", + "bbox": [ + 174, + 585, + 823, + 656 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Finally, we search for the optimal policy via the cross-entropy method planner (Hafner et al., 2019) based on the learned model and deploy the policy in a real-world platform. The results of A/B test in City A is shown in Fig. 5(c). It can be seen that after the day of the A/B test, the treatment group (deploying our policy) significant improve the five-minute order-taken rate than the baseline policy (the same as the behavior policy). In summary, the policy improves the supply from 0.14 to 1.63 percentage points to the behavior policies in the 6 cities. The details of these results are in Appx. H.6. ", + "bbox": [ + 173, + 662, + 825, + 746 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "6 DISCUSSION AND FUTURE WORK ", + "text_level": 1, + "bbox": [ + 176, + 756, + 485, + 772 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In this work, we propose AWRM which handles the generalization challenges of the counterfactual environment model learning. By theoretical modeling, we give a tractable solution to handle AWRM and propose GALILEO. GALILEO is verified in synthetic environments, complex robot control tasks, and a real-world platform, and shows great generalization ability on counterfactual queries. ", + "bbox": [ + 174, + 777, + 825, + 833 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Giving correct answers to counterfactual queries is important for policy learning. We hope the work can inspire researchers to develop more powerful tools for counterfactual environment model learning. The current limitation lies in: There are several simplifications in the theoretical modeling process (further discussion is in Appx. B), which can be modeled more elaborately . Besides, experiments on MuJoCo indicate that these tasks are still challenging to give correct predictions on counterfactual data. These should also be further investigated in future work. ", + "bbox": [ + 174, + 840, + 825, + 924 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 176, + 102, + 285, + 117 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ahmed M. Alaa and Mihaela van der Schaar. Limits of estimating heterogeneous treatment effects: Guidelines for practical algorithm design. In Proceedings of the 35th International Conference on Machine Learning, volume 80, pp. 129–138, Stockholm, Sweden, 2018. ", + "bbox": [ + 174, + 126, + 826, + 167 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Syed Mumtaz Ali and Samuel D Silvey. A general class of coefficients of divergence of one distribution from another. Journal of the Royal Statistical Society: Series B (Methodological), 28 (1):131–142, 1966. ", + "bbox": [ + 174, + 179, + 823, + 222 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Serge Assaad, Shuxi Zeng, Chenyang Tao, Shounak Datta, Nikhil Mehta, Ricardo Henao, Fan Li, and Lawrence Carin. Counterfactual representation learning with balancing weights. In Arindam Banerjee and Kenji Fukumizu (eds.), The 24th International Conference on Artificial Intelligence and Statistics, volume 130 of Proceedings of Machine Learning Research, pp. 1972–1980, Virtual Event, 2021. PMLR. ", + "bbox": [ + 174, + 232, + 825, + 301 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. Analysis of representations for domain adaptation. In Bernhard Schölkopf, John C. Platt, and Thomas Hofmann (eds.), Advances in Neural Information Processing Systems 19, pp. 137–144, British Columbia, Canada, 2006. MIT Press. ", + "bbox": [ + 174, + 313, + 825, + 368 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wortman Vaughan. A theory of learning from different domains. Mach. Learn., 79(1-2):151–175, 2010. ", + "bbox": [ + 173, + 378, + 823, + 409 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Artem Betlei, Eustache Diemert, and Massih-Reza Amini. Treatment targeting by AUUC maximization with generalization guarantees. CoRR, abs/2012.09897, 2020. ", + "bbox": [ + 171, + 419, + 825, + 446 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ioana Bica, James Jordon, and Mihaela van der Schaar. Estimating the effects of continuous-valued interventions using generative adversarial networks. In Advances in Neural Information Processing Systems 33, virtual event, 2020. ", + "bbox": [ + 174, + 457, + 825, + 500 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. Openai gym. CoRR, abs/1606.01540, 2016. ", + "bbox": [ + 173, + 511, + 823, + 540 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Lars Buesing, Theophane Weber, Yori Zwols, Nicolas Heess, Sébastien Racanière, Arthur Guez, and Jean-Baptiste Lespiau. Woulda, coulda, shoulda: Counterfactually-guided policy search. In $7 t h$ International Conference on Learning Representations, New Orleans, LA, 2019. ", + "bbox": [ + 173, + 549, + 826, + 592 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jonathon Byrd and Zachary Chase Lipton. What is the effect of importance weighting in deep learning? In Kamalika Chaudhuri and Ruslan Salakhutdinov (eds.), Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pp. 872–881, Long Beach, CA, 2019. PMLR. ", + "bbox": [ + 173, + 603, + 825, + 660 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Xinshi Chen, Shuang Li, Hui Li, Shaohua Jiang, Yuan Qi, and Le Song. Generative adversarial user model for reinforcement learning based recommendation system. In Proceedings of the 36th International Conference on Machine Learning, pp. 1052–1061, Long Beach, CA, 2019. ", + "bbox": [ + 174, + 670, + 825, + 712 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Xiong-Hui Chen, Yang Yu, Qingyang Li, Fan-Ming Luo, Zhiwei Tony Qin, Shang Wenjie, and Jieping Ye. Offline model-based adaptable policy learning. In Advances in Neural Information Processing Systems 34 (NeurIPS’21), Virtual Conference, 2021. ", + "bbox": [ + 174, + 722, + 825, + 765 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Corinna Cortes, Yishay Mansour, and Mehryar Mohri. Learning bounds for importance weighting. In John D. Lafferty, Christopher K. I. Williams, John Shawe-Taylor, Richard S. Zemel, and Aron Culotta (eds.), Advances in Neural Information Processing Systems 23, pp. 442–450, British Columbia, Canada, 2010. Curran Associates, Inc. ", + "bbox": [ + 173, + 775, + 825, + 832 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4RL: datasets for deep data-driven reinforcement learning. CoRR, abs/2004.07219, 2020. ", + "bbox": [ + 168, + 842, + 823, + 871 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron C. Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in Neural Information Processing Systems 27, pp. 2672–2680, Montréal, Canada, 2014. ", + "bbox": [ + 174, + 881, + 825, + 924 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft Actor-Critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. In Proceedings of the 35th International Conference on Machine Learning, pp. 1856–1865, Stockholmsmässan, Sweden, 2018. ", + "bbox": [ + 174, + 103, + 825, + 159 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Danijar Hafner, Timothy P. Lillicrap, Ian Fischer, Ruben Villegas, David Ha, Honglak Lee, and James Davidson. Learning latent dynamics for planning from pixels. In Kamalika Chaudhuri and Ruslan Salakhutdinov (eds.), Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pp. 2555–2565, Long Beach, CA, 2019. PMLR. ", + "bbox": [ + 174, + 169, + 826, + 238 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Negar Hassanpour and Russell Greiner. Counterfactual regression with importance sampling weights. In Proceedings of the 28th International Joint Conference on Artificial Intelligence, pp. 5880–5887, Macao, China, 2019. ", + "bbox": [ + 173, + 250, + 826, + 291 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "J.B. Hiriart-Urruty and C. Lemaréchal. Fundamentals of Convex Analysis. 2001. ISBN 9783642564680. ", + "bbox": [ + 173, + 300, + 825, + 330 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Jonathan Ho and Stefano Ermon. Generative adversarial imitation learning. In Advances in Neural Information Processing Systems 29, pp. 4565–4573, Barcelona, Spain, 2016. ", + "bbox": [ + 171, + 339, + 823, + 369 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Guido Imbens. The role of the propensity score in estimating dose-response functions. Econometrics eJournal, 1999. ", + "bbox": [ + 173, + 377, + 823, + 406 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Edward L Ionides. Truncated importance sampling. Journal of Computational and Graphical Statistics, 17(2):295–311, 2008. ", + "bbox": [ + 173, + 416, + 823, + 444 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Stephen James and Edward Johns. 3D simulation for robot arm control with deep Q-learning. CoRR, abs/1609.03759, 2016. ", + "bbox": [ + 173, + 454, + 825, + 483 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Fredrik D Johansson, Nathan Kallus, Uri Shalit, and David Sontag. Learning weighted representations for generalization across designs. arXiv preprint arXiv:1802.08598, 2018. ", + "bbox": [ + 171, + 492, + 823, + 521 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Yonghan Jung, Jin Tian, and Elias Bareinboim. Learning causal effects via weighted empirical risk minimization. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan, and Hsuan-Tien Lin (eds.), Advances in Neural Information Processing Systems 33, Virtual Event, 2020. ", + "bbox": [ + 174, + 530, + 826, + 587 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Rahul Kidambi, Aravind Rajeswaran, Praneeth Netrapalli, and Thorsten Joachims. MOReL : Modelbased offline reinforcement learning. CoRR, abs/2005.05951, 2020. ", + "bbox": [ + 173, + 595, + 825, + 626 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tutorial, review, and perspectives on open problems. CoRR, abs/2005.01643, 2020. ", + "bbox": [ + 173, + 633, + 826, + 664 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Yao Liu, Adith Swaminathan, Alekh Agarwal, and Emma Brunskill. Provably good batch off-policy reinforcement learning without great exploration. In Advances in Neural Information Processing Systems 33, virtual event, 2020. ", + "bbox": [ + 174, + 672, + 825, + 715 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "XuanLong Nguyen, Martin J. Wainwright, and Michael I. Jordan. Estimating divergence functionals and the likelihood ratio by convex risk minimization. IEEE Trans. Inf. Theory, 56(11):5847–5861, 2010. ", + "bbox": [ + 174, + 724, + 826, + 767 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Sebastian Nowozin, Botond Cseke, and Ryota Tomioka. f-GAN: Training generative neural samplers using variational divergence minimization. In Advances in Neural Information Processing Systems 29, pp. 271–279, Barcelona, Spain, 2016. ", + "bbox": [ + 176, + 776, + 823, + 819 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Silviu Pitis, Elliot Creager, and Animesh Garg. Counterfactual data augmentation using locally factored dynamics. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan, and Hsuan-Tien Lin (eds.), Advances in Neural Information Processing Systems 33, virtual, 2020. ", + "bbox": [ + 174, + 829, + 825, + 886 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Dean Pomerleau. Efficient training of artificial neural networks for autonomous navigation. Neural Computation, 3(1):88–97, 1991. ", + "bbox": [ + 178, + 895, + 823, + 924 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Joaquin Quinonero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D Lawrence. Dataset shift in machine learning. Mit Press, 2008. ", + "bbox": [ + 169, + 103, + 823, + 132 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Paul R. Rosenbaum and Donald B. Rubin. The central role of the propensity score in observational studies for causal effects. Biometrika, 70:41–55, 1983. ", + "bbox": [ + 171, + 141, + 823, + 170 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "John Schulman, Sergey Levine, Pieter Abbeel, Michael I. Jordan, and Philipp Moritz. Trust region policy optimization. In Proceedings of the 32nd International Conference on Machine Learning, volume 37, pp. 1889–1897, Lille, France, 2015. ", + "bbox": [ + 176, + 178, + 825, + 222 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms. CoRR, abs/1707.06347, 2017. ", + "bbox": [ + 171, + 229, + 823, + 260 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Uri Shalit, Fredrik D. Johansson, and David A. Sontag. Estimating individual treatment effect: generalization bounds and algorithms. In Proceedings of the 34th International Conference on Machine Learning, volume 70, pp. 3076–3085, Sydney, Australia, 2017. ", + "bbox": [ + 174, + 267, + 823, + 311 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Wenjie Shang, Yang Yu, Qingyang Li, Zhiwei (Tony) Qin, Yiping Meng, and Jieping Ye. Environment reconstruction with hidden confounders for reinforcement learning based recommendation. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 566–576, Anchorage, AK, 2019. ", + "bbox": [ + 174, + 319, + 825, + 376 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Jing-Cheng Shi, Yang Yu, Qing Da, Shi-Yong Chen, and Anxiang Zeng. Virtual-taobao: Virtualizing real-world online retail environment for reinforcement learning. In The 33th AAAI Conference on Artificial Intelligence, pp. 4902–4909, Honolulu, Hawaii, 2019. ", + "bbox": [ + 173, + 385, + 825, + 428 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Hidetoshi Shimodaira. Improving predictive inference under covariate shift by weighting the loglikelihood function. Journal of statistical planning and inference, 90(2):227–244, 2000. ", + "bbox": [ + 173, + 436, + 823, + 467 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Sumedh A. Sontakke, Arash Mehrjou, Laurent Itti, and Bernhard Schölkopf. Causal curiosity: RL agents discovering self-supervised experiments for causal representation learning. In Marina Meila and Tong Zhang (eds.), Proceedings of the 38th International Conference on Machine Learning, volume 139 of Proceedings of Machine Learning Research, pp. 9848–9858, Virtual Event, 2021. PMLR. ", + "bbox": [ + 173, + 473, + 826, + 545 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Peter Spirtes. Introduction to causal inference. J. Mach. Learn. Res., 11:1643–1662, aug 2010. ISSN 1532-4435. ", + "bbox": [ + 171, + 553, + 825, + 583 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Richard S. Sutton and Andrew G. Barto. Reinforcement learning - an introduction. Adaptive computation and machine learning. MIT Press, 1998. ", + "bbox": [ + 171, + 590, + 825, + 621 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Emanuel Todorov, Tom Erez, and Yuval Tassa. MuJoCo: A physics engine for model-based control. In Proceedings of the 24th IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 5026–5033, Vilamoura, Portugal, 2012. ", + "bbox": [ + 174, + 628, + 823, + 672 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Stefan Wager and Susan Athey. Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association, 113(523):1228–1242, 2018. ", + "bbox": [ + 171, + 680, + 825, + 710 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Jeremy C. Weiss, Finn Kuusisto, Kendrick Boyd, Jie Liu, and David Page. Machine learning for treatment assignment: Improving individualized risk attribution. In American Medical Informatics Association Annual Symposium, San Francisco, CA, 2015. ", + "bbox": [ + 174, + 718, + 825, + 761 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Jinsung Yoon, James Jordon, and Mihaela van der Schaar. GANITE: estimation of individualized treatment effects using generative adversarial nets. In 6th International Conference on Learning Representations, 2018. ", + "bbox": [ + 174, + 770, + 825, + 813 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Tianhe Yu, Garrett Thomas, Lantao Yu, Stefano Ermon, James Y. Zou, Sergey Levine, Chelsea Finn, and Tengyu Ma. MOPO: model-based offline policy optimization. In Advances in Neural Information Processing Systems 33, virtual event, 2020. ", + "bbox": [ + 174, + 821, + 825, + 864 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Appendix ", + "text_level": 1, + "bbox": [ + 174, + 131, + 315, + 157 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Table of Contents ", + "text_level": 1, + "bbox": [ + 174, + 184, + 352, + 203 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "A Proof of Theoretical Results 14 ", + "text_level": 1, + "bbox": [ + 207, + 212, + 787, + 223 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "A.1 Proof of Lemma 4.3 15 \nA.2 Proof of Eq. 6 16 \nA.3 Proof of Thm. 4.4 17 \nA.4 Proof of the Tractable Solution 20 \nB Discussion of the Theoretical Results 22 \nC Societal Impact 22 \nD AWRM-oracle Pseudocode 23 ", + "bbox": [ + 233, + 226, + 787, + 284 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "", + "bbox": [ + 214, + 295, + 787, + 362 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "E Implementation 23 ", + "text_level": 1, + "bbox": [ + 214, + 375, + 784, + 387 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "E.1 Details of the GALILEO Implementation 23 \nE.2 Connection with Previous Adversarial Algorithms 25 ", + "bbox": [ + 232, + 388, + 784, + 417 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "F Additional Related Work 26 ", + "text_level": 1, + "bbox": [ + 214, + 430, + 781, + 443 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "", + "bbox": [ + 766, + 430, + 785, + 443 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Experiment Details 27 ", + "text_level": 1, + "bbox": [ + 210, + 457, + 787, + 469 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "G.1 Settings 27 \nG.2 Baseline Algorithms 29 \nG.3 Hyper-parameters 30 \nG.4 Computation Resources 30 ", + "bbox": [ + 235, + 470, + 787, + 530 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "H Additional Results 31 ", + "text_level": 1, + "bbox": [ + 210, + 540, + 787, + 554 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "H.1 All of the Result Table . 31 \nH.2 Averaged Response Curves 31 \nH.3 Ablation Studies 31 \nH.4 Worst-Case Prediction Error 32 \nH.5 Detailed Results in the MuJoCo Tasks 32 \nH.6 Detailed Results in the BAT Task 34 ", + "bbox": [ + 235, + 556, + 787, + 645 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "A PROOF OF THEORETICAL RESULTS ", + "text_level": 1, + "bbox": [ + 174, + 679, + 501, + 696 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "The overall pipeline to model the tractable solution to AWRM is given in Fig. 6. ", + "bbox": [ + 174, + 710, + 699, + 727 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "In the proof section, we replace the notation of $\\mathbb { E }$ with an integral for brevity. Now we rewrite the original objective $\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , \\bar { M } )$ as: ", + "bbox": [ + 174, + 732, + 820, + 763 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/1c0ae0f1ffd4e5eca34900d7c57ddb97f2a842bdeece077cf33f4bbb2a5b6d08.jpg", + "text": "$$\n\\operatorname* { m i n } _ { \\substack { M \\in M } } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\int _ { \\mathcal { X } , \\mathcal { A } } \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) \\omega ( x , a ) \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\left( - \\log M ( x ^ { \\prime } | x , a ) \\right) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } ,\n$$", + "text_format": "latex", + "bbox": [ + 181, + 767, + 823, + 801 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "where $\\begin{array} { r } { \\omega ( x , a ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) } } \\end{array}$ and $\\begin{array} { r } { \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } = \\int _ { \\mathcal { X } , \\mathcal { A } } \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ^ { 2 } \\mathrm { d } a \\mathrm { d } x } \\end{array}$ , which is the squared $l _ { 2 }$ -norm. In an MDP, given any policy $\\pi$ , $\\rho _ { M ^ { * } } ^ { \\pi } ( x , a , x ^ { \\prime } ) = \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a )$ where $\\rho _ { M ^ { * } } ^ { \\pi } ( x )$ denotes the occupancy measure of $x$ for policy $\\pi$ , which can be defined (Sutton $\\&$ Barto, 1998; Ho & Ermon, 2016) as $\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\pi } ( x ) : = ( 1 - \\gamma ) \\mathbb { E } _ { x _ { 0 } \\sim \\rho _ { 0 } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \\right] } \\end{array}$ where $ { \\mathrm { P r } } ^ { \\pi } \\left[ x _ { t } = x | x _ { 0 } , M ^ { * } \\right]$ is the state visitation probability that $\\pi$ starts at state $x _ { 0 }$ in model $M ^ { * }$ and receive $x$ at timestep $t$ and $\\gamma \\in [ 0 , 1 ]$ is the discount factor. ", + "bbox": [ + 173, + 829, + 826, + 924 + ], + "page_idx": 13 + }, + { + "type": "image", + "img_path": "images/b8d8ca5df420de7f53155d64557ee7e5645c8c107cdf9e997b057385c9255a79.jpg", + "image_caption": [ + "Figure 6: The overall pipeline to model the tractable solution to AWRM. $f$ is a generator function defined by $f$ -divergence (Nowozin et al., 2016). $\\kappa$ is an intermediary policy introduced in the estimation. " + ], + "image_footnote": [], + "bbox": [ + 282, + 101, + 715, + 338 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "A.1 PROOF OF LEMMA 4.3 ", + "text_level": 1, + "bbox": [ + 176, + 420, + 374, + 435 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "For better readability, we first rewrite Lemma 4.3 as follows: ", + "bbox": [ + 173, + 446, + 573, + 463 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Lemma A.1. Given any $M$ in $\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M )$ , the distribution of the ideal best-response policy $\\bar { \\beta } ^ { * }$ satisfies: ", + "bbox": [ + 171, + 468, + 821, + 500 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/55d6a8f2e6526f3bd013bea3c86052f10a04d6c901b66db1ff872d32038fd387.jpg", + "text": "$$\n\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ,\n$$", + "text_format": "latex", + "bbox": [ + 287, + 501, + 709, + 534 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "where $D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )$ is the Kullback-Leibler $( K L )$ divergence between $M ^ { * } ( \\cdot | x , a )$ and $M ( \\cdot | x , a )$ , $H _ { M ^ { \\ast } } ( x , a )$ denotes the entropy of $M ^ { * } ( \\cdot | x , a )$ , where $D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )$ is the Kullback-Leibler $( K L )$ divergence between $M ^ { * } ( \\cdot | x , a )$ and $M ( \\cdot | x , a )$ , $H _ { M ^ { \\ast } } ( x , a )$ denotes the entropy of $M ^ { * } ( \\cdot | x , a )$ , and $\\alpha _ { M }$ is the regularization coefficient $\\alpha$ in Eq. 9 and also as a normalizer. ", + "bbox": [ + 173, + 540, + 826, + 599 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Proof. Given a transition function $M$ of an MDP, the distribution of the best-response policy $\\beta ^ { * }$ satisfies: ", + "bbox": [ + 173, + 622, + 825, + 652 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/42a5873062af4a9b91baad6a11c887370a4afab3a398ab7be1cec5015ac74585.jpg", + "text": "$$\n\\begin{array} { r l } { \\rho _ { \\mu \\mu ^ { * } } ^ { * } } & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\omega ( x , \\omega ) \\int _ { x } d x ^ { * } ( x ^ { * } | x , a ) \\left( - \\log { \\mathcal { A } ( x ^ { * } | x , a ) } \\right) \\mathrm { d } x ^ { * } \\mathrm { d } \\alpha \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { \\nu \\mu } ^ { 2 } ( z , \\omega ) } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\int _ { x } \\underline { { \\mathcal { A } } } ^ { \\mathcal { A } ^ { * } } ( x ^ { * } | x , a ) \\left( - \\log { \\mathcal { A } ( x ^ { * } | x , a ) } \\right) \\mathrm { d } x ^ { * } \\mathrm { d } \\alpha \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { \\mu \\nu } ^ { 3 } d \\cdot ( x ^ { * } ) \\| _ { 2 } ^ { 2 } } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\frac { 2 } { \\alpha } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\| \\phi ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\mu \\nu } ^ { 3 } d \\cdot ( x ^ { * } ) \\| _ { 2 } ^ { 2 } } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\alpha \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) g ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\nu \\mu ^ { * } } ^ { 3 } ( z , \\nu ) \\| _ { 2 } ^ { 2 } } \\\\ & = \\operatorname* { m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\alpha \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( x , a ) g ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\mu \\nu } ^ { 3 } ( z , \\omega ) \\| _ { 2 } ^ { 2 } - \\frac \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 166, + 680, + 816, + 928 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "We know that the occupancy measure $\\rho _ { M ^ { * } } ^ { \\beta }$ is a density function with a constraint $\\begin{array} { r } { \\int _ { \\mathcal { X } } \\int _ { A } \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) \\mathrm { d } a \\mathrm { d } x = 1 } \\end{array}$ . Assuming the occupancy measure $\\rho _ { M ^ { * } } ^ { \\beta }$ has an upper bound $c$ , that is $0 ~ \\leq ~ \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ~ \\leq ~ c , \\forall a ~ \\in ~ { \\mathcal { A } } , \\forall x ~ \\in ~ \\mathcal { X }$ , constructing a regularization coefficient $\\alpha _ { M } =$ $\\begin{array} { r } { \\int _ { \\mathcal { X } } \\int _ { \\mathcal { A } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) \\mathrm { d } x \\mathrm { d } a } \\end{array}$ as a constant value given any $M$ , then we have ", + "bbox": [ + 173, + 101, + 825, + 181 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/1da27a213d4fe3fa509876db24b86328155cda21a6b3900e45f8355fa7c7d562.jpg", + "text": "$$\n\\begin{array} { r l } & { \\rho _ { M ^ { * } } ^ { \\beta ^ { * } } ( x , a ) = \\frac { g ( x , a ) } { \\alpha _ { M } } } \\\\ & { \\qquad = \\frac { \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\log \\frac { M ^ { * } ( x ^ { \\prime } | x , a ) } { M ( x ^ { \\prime } | x , a ) } \\mathrm { d } x - \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\log M ^ { * } ( x ^ { \\prime } | x , a ) \\mathrm { d } x } { \\alpha _ { M } } } \\\\ & { \\qquad = \\frac { D _ { K L } \\left( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) \\right) + H _ { M ^ { * } } ( x , a ) } { \\alpha _ { M } } } \\\\ & { \\qquad \\propto \\Big ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) \\Big ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 215, + 195, + 782, + 332 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "which is the optimal density function of Eq. 9 with $\\alpha = \\alpha _ { M }$ . ", + "bbox": [ + 173, + 347, + 573, + 362 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Note that in some particular $M ^ { * }$ , we still cannot construct a $\\beta$ that can generate an occupancy specified by $g ( x , a ) / \\alpha _ { M }$ for any $M$ . We can only claim the distribution of the ideal best-response policy $\\bar { \\beta } ^ { * }$ satisfies: ", + "bbox": [ + 174, + 368, + 825, + 411 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/131c9bcd79334176a65e290aa8869ff310b09911deb3d3a4a099405db6f0b10f.jpg", + "text": "$$\n\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ,\n$$", + "text_format": "latex", + "bbox": [ + 287, + 428, + 710, + 460 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "where $\\alpha _ { M }$ is a normalizer that $\\begin{array} { r } { \\alpha _ { M } = \\int _ { \\mathcal { X } } \\int _ { \\mathcal { A } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | \\underline { { x } } , a ) ) + H _ { M ^ { * } } ( x , a ) ) \\mathrm { d } x \\mathrm { d } a } \\end{array}$ . We give a discussion of the rationality of the ideal best-response policy $\\bar { \\beta } ^ { * }$ as a replacement of the real best-response policy $\\beta ^ { * }$ in Remark A.2. ", + "bbox": [ + 174, + 477, + 825, + 523 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Remark A.2. The optimal solution Eq. 11 relies on $g ( x , a )$ . In some particular $M ^ { * }$ , it is intractable to derive a $\\beta$ that can generate an occupancy specified by $g ( x , a ) / \\alpha _ { M }$ . Consider the following case: a state $x _ { 1 }$ in $M ^ { * }$ might be harder to reach than another state $x _ { 2 }$ , e.g., $M ^ { \\ast } ( x _ { 1 } | x , a ) \\ <$ $M ^ { * } ( x _ { 2 } | x , a ) , \\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A }$ , then it is impossible to find a $\\beta$ that the occupancy satisfies $\\rho _ { M ^ { * } } ^ { \\beta } ( x _ { 1 } , a ) > \\rho _ { M ^ { * } } ^ { \\beta } ( x _ { 2 } , a )$ . In this case, Eq. 11 can be a sub-optimal solution. Since this work focuses on task-agnostic solution derivation while the solution to the above problem should rely on the specific description of $M ^ { * }$ , we leave it as future work. However, we point out that Eq. 11 is $a$ reasonable re-weighting term even as a sub-optimum: $\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ gives larger densities on the data where the distribution distance between the approximation model and the real model (i.e., $D _ { K L } ( M ^ { * } , M ) )$ is larger or the stochasticity of the real model (i.e., $H _ { M ^ { * } }$ ) is larger. ", + "bbox": [ + 173, + 587, + 826, + 734 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "A.2 PROOF OF EQ. 6 ", + "text_level": 1, + "bbox": [ + 176, + 762, + 331, + 777 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "The integral process of $D _ { K L }$ in Eq. 5 is intractable in the offline setting as it explicitly requires the conditional probability function of $M ^ { * }$ . Our motivation for the tractable solution is utilizing the offline dataset $\\mathcal { D } _ { \\mathrm { r e a l } }$ as the empirical joint distribution $\\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } )$ and adopting practical techniques for distance estimation on two joint distributions, like GAN (Goodfellow et al., 2014; Nowozin et al., 2016), to approximate Eq. 5. To adopt that solution, we should first transform Eq. 5 into a form under joint distributions. Without loss of generality, we introduce an intermediary policy $\\kappa$ , of which $\\mu$ can be regarded as a specific instance. Then we have $M ( x ^ { \\prime } | x , a ) = \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) / \\rho _ { M } ^ { \\kappa } ( x , a )$ for any $M$ if $\\rho _ { M } ^ { \\kappa } ( x , a ) > 0$ . Assuming $\\forall x \\in \\mathcal { X } , \\forall a \\in A , \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) > 0$ if $\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) > 0$ , which will hold when $\\kappa$ overlaps with $\\mu$ , then Eq. 5 can transform to: ", + "bbox": [ + 173, + 792, + 826, + 924 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/603aa333161db1e721e5232fd0062b09eeebf1530eef988de2c1ca9fa3679555.jpg", + "text": "$$\n\\begin{array} { r l } { \\frac { d } { d t } } & { ( z , z , n ) = \\frac { 1 } { n + 1 } ( 2 \\frac { n + 1 } { n + 1 } , \\frac { 1 } { n + 1 } ; \\frac { 1 } { n + 2 } , z ) ! \\log ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } ) \\log ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) } \\\\ & { - \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , z , n ) ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) \\log \\log ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n ) ) \\mathrm { d } x ^ { \\prime } \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n ) ) \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } \\exp ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } + \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } - \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) \\mathrm { d } y ^ { \\prime } } \\\\ & = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\exp ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } + \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } - \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 119, + 839, + 426 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "where $\\alpha _ { 0 } ( x , a ) = \\alpha _ { M } \\rho _ { M ^ { \\ast } } ^ { \\kappa } ( x , a )$ ", + "bbox": [ + 173, + 438, + 393, + 454 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Definition A.3 ( $f$ -divergence). Given two distributions $P$ and $Q$ , two absolutely continuous density functions $p$ and $q$ with respect to a base measure dx defined on the domain $\\mathcal { X }$ , we define the $f$ -divergence (Nowozin et al., 2016), ", + "bbox": [ + 173, + 457, + 825, + 500 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/35c46f2fda671505c0299be9d24dc9d673a9f6045c71c952c2e221eab3d4810b.jpg", + "text": "$$\nD _ { f } ( P \\| Q ) = \\int _ { \\mathcal X } q ( x ) f \\left( \\frac { p ( x ) } { q ( x ) } \\right) \\mathrm { d } x ,\n$$", + "text_format": "latex", + "bbox": [ + 377, + 506, + 620, + 541 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "where the generator function $f : \\mathbb { R } _ { + } \\to \\mathbb { R }$ is a convex, lower-semicontinuous function. ", + "bbox": [ + 173, + 546, + 738, + 561 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "We notice that the terms $\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } } \\end{array}$ and $\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } { \\rho _ { M } ^ { \\kappa } ( x , a ) } } \\end{array}$ are the integrated functions in reverse KL divergence, which is an instance of $f$ function in $f$ -divergence (See ReverseKL divergence of Tab.1 in (Nowozin et al., 2016) for more details). Replacing that form $q \\log { \\frac { q } { p } }$ with $q f ( { \\underline { { p } } } )$ , we obtain a generalized representation of $\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ : ", + "bbox": [ + 173, + 571, + 828, + 643 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/beac113262e3629c20136f81ade1f37648176e45e901733b49b520980dec2112.jpg", + "text": "$$\n\\tilde { \\bar { \\nu } } _ { M ^ { * } } ^ { * } : = \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\Bigg ( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) - H _ { M ^ { * } } ( x , a ) \\right)\n$$", + "text_format": "latex", + "bbox": [ + 181, + 651, + 839, + 690 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "A.3 PROOF OF THM. 4.4 ", + "text_level": 1, + "bbox": [ + 174, + 718, + 357, + 733 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "We first introduce several useful lemmas for the proof. ", + "bbox": [ + 173, + 744, + 532, + 760 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Lemma A.4. Rearrangement inequality The rearrangement inequality states that, for two sequences $a _ { 1 } \\geq a _ { 2 } \\geq . . . \\geq a _ { n }$ and $b _ { 1 } \\geq b _ { 2 } \\geq . . . \\geq b _ { n }$ , the inequalities ", + "bbox": [ + 171, + 763, + 823, + 792 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/9ebcb262c29a009d3359e82dcd3b1f83c045ef4f7c6892244fad87fa1d23b207.jpg", + "text": "$$\na _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \\cdot \\cdot \\cdot + a _ { n } b _ { n } \\geq a _ { 1 } b _ { \\pi ( 1 ) } + a _ { 2 } b _ { \\pi ( 2 ) } + \\cdot \\cdot \\cdot + a _ { n } b _ { \\pi ( n ) } \\geq a _ { 1 } b _ { n } + a _ { 2 } b _ { n - 1 } + \\cdot \\cdot \\cdot + a _ { n } b _ { 1 }\n$$", + "text_format": "latex", + "bbox": [ + 181, + 797, + 818, + 816 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "hold, where $\\pi ( 1 ) , \\pi ( 2 ) , \\ldots , \\pi ( n )$ is any permutation of $1 , 2 , \\ldots , n$ . ", + "bbox": [ + 173, + 821, + 617, + 839 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Lemma A.5. For two sequences $a _ { 1 } \\geq a _ { 2 } \\geq . . . \\geq a _ { n }$ and $b _ { 1 } \\geq b _ { 2 } \\geq . . . \\geq b _ { n }$ , the inequalities ", + "bbox": [ + 171, + 842, + 799, + 858 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/a9c1c632cb3afc7394107043c403f4aeb12562063d12e4044609df809363804a.jpg", + "text": "$$\n\\sum _ { i = 1 } ^ { n } { \\frac { 1 } { n } } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } { \\frac { 1 } { n } } a _ { i } \\sum { \\frac { 1 } { n } } b _ { i }\n$$", + "text_format": "latex", + "bbox": [ + 397, + 862, + 599, + 904 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "hold. ", + "bbox": [ + 173, + 909, + 209, + 924 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Proof. By rearrangement inequality, we have ", + "bbox": [ + 174, + 103, + 473, + 119 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/54f0537a515b0381540541728acfc06db53a99a7806d3346c8b48a13ebe7b7cf.jpg", + "text": "$$\n\\begin{array} { l } { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \\cdots + a _ { n } b _ { n } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 2 } + a _ { 2 } b _ { 3 } + \\cdots + a _ { n } b _ { 1 } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \\cdots + a _ { n } b _ { 2 } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \\cdots + a _ { n } b _ { 2 } } } \\\\ { { \\displaystyle \\qquad \\vdots \\qquad } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { n } + a _ { 2 } b _ { 1 } + \\cdots + a _ { n } b _ { n - 1 } } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 361, + 122, + 632, + 319 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Then we have ", + "bbox": [ + 173, + 319, + 267, + 333 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/118a0a4436f20ba33227d29825c64f45d04c5eec5a70f85acc6854cdfac4dad9.jpg", + "text": "$$\n\\begin{array} { l } { { \\displaystyle n \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } a _ { i } \\sum _ { i = 1 } ^ { n } b _ { i } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } \\frac { 1 } { n } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } \\frac { 1 } { n } a _ { i } \\sum _ { n } \\frac { 1 } { n } b _ { i } } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 397, + 334, + 599, + 420 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Now we extend Lemma A.5 into the continuous integral scenario: ", + "bbox": [ + 173, + 449, + 606, + 465 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Lemma A.6. Given $\\mathcal { X } \\subset \\mathbb { R } ,$ , for two functions $f : \\mathcal { X } \\to \\mathbb { R }$ and $g : \\mathcal { X } \\mathbb { R }$ that $f \\left( x \\right) \\geq f \\left( y \\right)$ if and only if $g ( x ) \\geq g ( y )$ , $\\forall x , y \\in { \\mathcal { X } }$ , the inequality ", + "bbox": [ + 174, + 467, + 825, + 497 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/072e0bb93a1ea14e45290006942c024fa487618183a385738ddaa285f223fa11.jpg", + "text": "$$\n\\int _ { \\mathcal { X } } p ( x ) f ( x ) g ( x ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } p ( x ) f ( x ) \\mathrm { d } x \\int _ { \\mathcal { X } } p ( x ) g ( x ) \\mathrm { d } x\n$$", + "text_format": "latex", + "bbox": [ + 313, + 500, + 684, + 532 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "holds, where $p : \\mathcal { X } \\mathbb { R }$ and $p ( x ) > 0 , \\forall x \\in \\mathcal { X }$ and $\\begin{array} { r } { \\int _ { \\mathcal { X } } p ( x ) \\mathrm { d } x = 1 } \\end{array}$ . ", + "bbox": [ + 176, + 536, + 624, + 553 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Proof. Since $( f ( x ) - f ( y ) ) ( g ( x ) - g ( y ) ) \\geq 0 , \\forall x , y \\in \\mathcal { X }$ , we have ", + "bbox": [ + 173, + 566, + 620, + 584 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/5bcefd60596a1023d02424dcb1c80888eb8012c6a9d062c03269c6330e84701f.jpg", + "text": "$$\n\\begin{array} { r l } & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { d } \\xi \\Bigr ( x \\mathrm { S p } ( \\xi ) + \\mathrm { i } \\omega ( 2 \\xi ) \\mathrm { S p } ( \\xi ) + \\mathrm { i } \\omega \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { d } \\xi \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 173, + 606, + 839, + 926 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Corollary A.7. Let g( p(x)q(x) ) $\\begin{array} { r } { g ( \\frac { p ( x ) } { q ( x ) } ) = - \\log \\frac { p ( x ) } { q ( x ) } } \\end{array}$ where $p ( x ) > 0 , \\forall x \\in \\mathcal { X }$ and $q ( x ) > 0 , \\forall x \\in \\mathcal { X } _ { }$ , for $\\upsilon > 0$ , the inequality ", + "bbox": [ + 174, + 126, + 826, + 160 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/be15a1d06758081716993cce8840fa46cc7b03706dd260cab3e53516d187e7a5.jpg", + "text": "$$\n\\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x ,\n$$", + "text_format": "latex", + "bbox": [ + 251, + 166, + 741, + 202 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "holds if $f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }$ . It is not always satisfied for $f$ functions of $f$ -divergence. We list a comparison of $f$ on that condition in Tab. 2. ", + "bbox": [ + 173, + 208, + 825, + 238 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Proof. $\\begin{array} { r } { g ^ { \\prime } ( x ) = - \\log x = - \\frac { 1 } { x } < 0 , \\forall x \\in \\mathcal { X } } \\end{array}$ . Suppose $f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }$ , we have $f \\left( x \\right) \\geq f \\left( y \\right)$ if and only if $g ( \\frac { p ( y ) } { q ( y ) } ) , \\forall x , y \\in \\mathcal { X }$ $g ( x ) \\geq g ( y ) , \\forall x , y \\in \\mathcal { X }$ holds for all $\\upsilon > 0$ holds. Thus . By defining $\\begin{array} { r } { f ( v \\frac { p ( x ) } { q ( x ) } ) \\geq f ( v \\frac { p ( y ) } { q ( y ) } ) } \\end{array}$ $\\begin{array} { r } { F ( x ) = f ( v \\frac { p ( x ) } { q ( x ) } ) ) } \\end{array}$ and if and only if $\\begin{array} { r } { G ( x ) = g ( \\frac { p ( x ) } { q ( x ) } ) } \\end{array}$ $g \\big ( \\frac { p ( x ) } { q ( x ) } \\big ) \\geq$ and using Lemma A.6, we have: ", + "bbox": [ + 173, + 252, + 826, + 327 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/c3950b114a613cb86264afb680420911ff7351973e16a1ff704469cd03c1a77b.jpg", + "text": "$$\n\\int _ { \\mathcal { X } } q ( x ) F ( x ) G ( x ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) F ( x ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) G ( x ) \\mathrm { d } x .\n$$", + "text_format": "latex", + "bbox": [ + 303, + 333, + 691, + 366 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Then we know ", + "bbox": [ + 173, + 372, + 272, + 386 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/fcb2558754c0452cee69dad339c5d1b39548a5bfe344d27640856e105bdf7c58.jpg", + "text": "$$\n\\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x\n$$", + "text_format": "latex", + "bbox": [ + 254, + 392, + 741, + 426 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "holds. ", + "bbox": [ + 173, + 433, + 215, + 446 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/59d020305dc954f7d90ff9b3ad9b6413b635a52d1e47ee126153a040b69b4ab4.jpg", + "table_caption": [ + "Table 2: Properties of $f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }$ for $f$ -divergences. " + ], + "table_footnote": [], + "table_body": "
NameGenerator function f(x)If f'(x)≤0,∀x∈X
Kullback-LeiblerReverse KLPearson x²Squared HellingerJensen-ShannonGANx logxFalseTrueFalseFalseFalseTrue
-logx(x-1)²(√x-1)²-(x+1)log1++xlog x2xlogx- (𝑥+1)log(𝑥+1)
", + "bbox": [ + 240, + 497, + 753, + 612 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Now, we prove Thm. 4.4. For better readability, we first rewrite Thm. 4.4 as follows: ", + "bbox": [ + 173, + 628, + 728, + 645 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Theorem A.8. Let $\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ as the data distribution of the best-response policy $\\bar { \\beta } ^ { * }$ in Eq. 4 under model $M _ { \\theta }$ parameterized by $\\theta$ , then we can find the optimal $\\theta ^ { * }$ o $\\begin{array} { r } \\mathsf { \\Gamma } _ { \\mathsf { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}$ (Eq. 4) via iteratively optimizing the objective $\\theta _ { t + 1 } = \\mathrm { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } )$ , where $\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ is approximated via the last-iteration model $M _ { \\theta _ { t } }$ . Based on Corollary A.7, we have an upper bound objective for $\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}$ and derive the following objective ", + "bbox": [ + 173, + 648, + 830, + 736 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/529b1ff22fa183e3b27afd2d293ac1cfa96e26643dba363fca5f8a97233c05cf.jpg", + "text": "$$\n\\mathfrak { t } _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\underbrace { \\left( f \\left( \\frac { \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f \\left( \\frac { \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) + H _ { M ^ { * } } ( x , a ) \\right) } _ { W ( x , a , x ^ { \\prime } ) } \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 181, + 742, + 839, + 804 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "where $\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) , \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } [ \\cdot ]$ ] denotes $\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\kappa } }$ [·], $f$ is the generator function in $f$ -divergence which satisfies $f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }$ , and $\\theta$ is the parameters of $M$ . $M _ { \\theta _ { t } }$ denotes $a$ probability function with the same parameters as the learned model (i.e., $\\bar { \\theta } = \\theta$ ) but the parameter is fixed and only used for sampling. ", + "bbox": [ + 173, + 810, + 825, + 871 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Proof. Let $\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ as the data distribution of the best-response policy $\\bar { \\beta } ^ { * }$ in Eq. 4 under model $M _ { \\theta }$ parameterized by $\\theta$ , then we can find the optimal $\\theta _ { t + 1 }$ of $\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}$ (Eq. 4) via iteratively optimizing the objective $\\begin{array} { r } { \\theta _ { t + 1 } = \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}$ , where $\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ is approximated via the last-iteration model $\\bar { M } _ { \\theta _ { t } }$ : ", + "bbox": [ + 174, + 888, + 825, + 926 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 101, + 823, + 133 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/55ba95f92ddc0d86b32bbc4155ee24719fbebd61dd6aca09bbcaba8307bf06aa.jpg", + "text": "$$\n\\begin{array} { r l } & \\quad \\mathrm { w r e r } \\int _ { - \\infty } ^ { \\infty } \\frac { 1 } { \\mathrm { d } { x } } \\frac { \\mathrm { d } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } } \\left( \\int _ { - \\infty } ^ { \\infty } \\mathrm { d } { x } \\omega \\mathrm { e r } \\omega ^ { 2 } \\mathrm { d } { x } ^ { 2 } \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } } \\right) + \\int _ { - \\infty } ^ { \\infty } \\mathrm { d } { x } ^ { 2 } \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 3 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } } \\\\ & \\qquad - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\left( \\frac { \\sqrt { 3 } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 2 } \\right) } - \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { y } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } } \\\\ & \\qquad - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\left( \\frac { \\sqrt { 3 } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 2 } \\right) } - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\mathrm y \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 202, + 179, + 861, + 554 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "where $M _ { \\theta _ { t } }$ is introduced to approximate the term $\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }$ and fixed when optimizing $\\theta$ . In Eq. 16, $\\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 }$ for Eq. 9 is eliminated as it does not contribute to the gradient of $\\theta$ . Assume $f ^ { \\prime } ( x ) \\leq$ $0 , \\forall x \\in \\mathcal { X }$ , let $\\begin{array} { r } { v ( x , a ) : = \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } > 0 } \\end{array}$ , $p ( x ^ { \\prime } | x , a ) = M _ { \\theta } ( x ^ { \\prime } | x , a )$ , and $q ( x ^ { \\prime } | x , a ) = M ^ { * } ( x ^ { \\prime } | x , a )$ , the first inequality can be derived by adopting Corollary A.7 and eliminating the first $H _ { M ^ { * } }$ since it does not contribute to the gradient of $\\theta$ . ", + "bbox": [ + 173, + 577, + 826, + 666 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "A.4 PROOF OF THE TRACTABLE SOLUTION ", + "text_level": 1, + "bbox": [ + 174, + 708, + 486, + 723 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Now we are ready to prove the tractable solution: ", + "bbox": [ + 173, + 736, + 496, + 752 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Proof. The core challenge is that the term $\\begin{array} { r } { f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\big ) - f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\big ) } \\end{array}$ is still intractable. In the following, we give a tractable solution to Thm. 4.4. First, we resort to the first-order approximation. Given some $u \\bar { \\in } ( 1 - \\xi , 1 + \\xi ) , \\xi > 0$ , we have ", + "bbox": [ + 173, + 781, + 826, + 837 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/6242018bc8440a3db08d4d5a227d5ee5d932127bd457b4e7187ada26e7bb0585.jpg", + "text": "$$\nf ( u ) \\approx f ( 1 ) + f ^ { \\prime } ( u ) ( u - 1 ) ,\n$$", + "text_format": "latex", + "bbox": [ + 400, + 847, + 598, + 866 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "where $f ^ { \\prime }$ is the first-order derivative of $f$ . By Taylor’s formula and the fact that $f ^ { \\prime } ( u )$ of the generator function $f$ is bounded in $( 1 - \\xi , 1 + \\xi )$ , the approximation error is no more than $\\mathcal { O } ( \\xi ^ { 2 } )$ . Substituting $u$ with $\\textstyle { \\frac { p ( x ) } { q ( x ) } }$ in Eq. 18, the pattern $f ( { \\frac { p ( x ) } { q ( x ) } } )$ in Eq. 17 can be converted to $\\begin{array} { r } { \\frac { p ( x ) } { q ( x ) } f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) - f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) + f ( 1 ) } \\end{array}$ , ", + "bbox": [ + 173, + 876, + 826, + 928 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "then we have: ", + "bbox": [ + 173, + 103, + 266, + 117 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/a1bccf439183ce51f0a5ff2f3520c4d9de04322e74aaf72ed2fe05144d6d5327.jpg", + "text": "$$\n\\begin{array} { r l } { \\gamma _ { 1 } = \\gamma _ { 2 } \\gamma _ { 2 } \\gamma _ { 3 } = \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } } & { \\gamma _ { 2 } \\gamma _ { 3 } = \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } , } \\\\ & { \\qquad \\quad + \\frac { \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 2 } } { \\mu _ { 1 } \\mu _ { 2 } \\mu _ { 3 } } \\gamma _ { 1 } ( \\frac { \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } } { \\mu _ { 2 } } + \\frac { \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } } { \\mu _ { 3 } } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad + \\frac { \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 3 } } { \\mu _ { 2 } \\mu _ { 3 } } \\gamma _ { 1 } ( \\frac { \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } } { \\mu _ { 2 } } + \\frac { \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } } { \\mu _ { 3 } } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad \\times ( \\gamma _ { 2 } + \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 3 } ) \\gamma _ { 4 } \\exp ( - \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad \\times \\exp ( - \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 4 } ) \\gamma _ { 4 } \\exp ( - \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 2 } + \\gamma _ { 3 } \\gamma _ { 4 } ) } \\\\ & \\qquad \\quad \\times \\exp ( - \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } ) \\gamma _ { 5 } ( \\gamma _ 3 \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 214, + 141, + 856, + 536 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Note \nbut we choose to keep it and ignore \nan empirical data distribution through data collected by $\\kappa$ in $M ^ { * }$ directly, rather than from a uniform distribution which is harder to be generated. Although keeping $\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )$ incurs extra bias in theory, the results in our experiments show that it has not made significant negative effects in practice. We leave this part of modeling in future work. In particular, by ignoring $\\bar { \\alpha _ { 0 } } ( x , a )$ , we have: ", + "bbox": [ + 173, + 508, + 828, + 590 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/831acb50e54dfa42d165c4f521955a089e6c78344de0fd7cb2167e4d717c64d8.jpg", + "text": "$$\n= \\arg \\operatorname* { m a x } _ { \\theta } \\int _ { \\chi , A , \\chi } \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ) \\left( f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { \\mathbb { M } ^ { + } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { \\mathbb { M } ^ { + } } ^ { \\kappa } ( x , a ) } \\right) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + \n$$", + "text_format": "latex", + "bbox": [ + 202, + 593, + 856, + 631 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/f3421198220e4fe5c5c7ce878300a2090d39b0826d8aac72a940954a9e24c7da.jpg", + "text": "$$\n\\int _ { \\boldsymbol { x } , A , \\boldsymbol { \\tilde { x } } } \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) \\left( f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathrm { M } \\theta _ { t } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } ) } { \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } ) } \\right) - f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathrm { M } \\theta _ { t } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) } { \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) } \\right) + H _ { \\mathsf { M } ^ { * } } ( \\boldsymbol { x } , \\boldsymbol { a } ) \\right) \\log M _ { \\theta } ( \\boldsymbol { x } ^ { \\prime } | \\boldsymbol { x } , \\boldsymbol { a } ) \\mathrm { d } \\boldsymbol { x } ^ { \\prime } \\mathrm { d } \\boldsymbol { a } \\mathrm { d } \\boldsymbol { x }\n$$", + "text_format": "latex", + "bbox": [ + 271, + 645, + 926, + 684 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "We can estimate $f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right)$ and $f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right)$ through Lemma A.9. ", + "bbox": [ + 173, + 707, + 681, + 741 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Lemma A.9 $( f ^ { \\prime } ( { \\frac { p } { q } } )$ estimation (Nguyen et al., 2010)). Given a function $T _ { \\varphi } : \\mathcal { X } \\to \\mathbb { R }$ parameterized by $\\varphi \\in \\Phi$ , if $f$ is convex and lower semi-continuous, by finding the maximum point of $\\varphi$ in the following objective: ", + "bbox": [ + 173, + 746, + 828, + 791 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/b48098841ee718edf4032c358c28ad998611761eb4b579b8043f492fb1df5bfe.jpg", + "text": "$$\n\\varphi ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi } \\mathbb { E } _ { x \\sim p ( x ) } \\left[ T _ { \\varphi } ( x ) \\right] - \\mathbb { E } _ { x \\sim q ( x ) } \\left[ f ^ { * } ( T _ { \\varphi } ( x ) ) \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 312, + 794, + 681, + 816 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "we have $f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) = T _ { \\varphi ^ { * } } ( x )$ . $f ^ { * }$ is Fenchel conjugate of $f$ (Hiriart-Urruty & Lemaréchal, 2001). ", + "bbox": [ + 171, + 821, + 805, + 842 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "In particular, ", + "bbox": [ + 173, + 856, + 259, + 869 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/281b23db492dda2f3a6b7786173739c9c5791e5c7e320a031be4d318ee9706eb.jpg", + "text": "$$\n\\begin{array} { r l } & { \\varphi _ { 0 } ^ { * } = \\underset { \\varphi _ { 0 } } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { * } } \\left[ T _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] - \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { * } } \\left[ f ^ { * } ( T _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } \\\\ & { \\varphi _ { 1 } ^ { * } = \\underset { \\varphi _ { 1 } } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { * } } ^ { * } } \\left[ T _ { \\varphi _ { 1 } } ( x , a ) \\right] - \\mathbb { E } _ { x , a \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ f ^ { * } ( T _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 240, + 871, + 756, + 922 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "then we have $\\begin{array} { r } { f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\approx T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) } \\end{array}$ and $\\begin{array} { r } { f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) \\approx T _ { \\varphi _ { 1 } ^ { * } } ( x , a ) } \\end{array}$ . Given $\\varphi _ { 0 } ^ { * }$ and $\\varphi _ { 1 } ^ { * }$ , let $A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , \\dot { a } , x ^ { \\prime } ) = T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) - T _ { \\varphi _ { 1 } ^ { * } } ( x , a ) ,$ , then we can optimize $\\theta$ via: ", + "bbox": [ + 173, + 99, + 828, + 148 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/316a26a7d4a9506aaa976710d2906b3c44f9f15b0634f1cf4aa65219802c1038.jpg", + "text": "$$\n\\begin{array} { r l } & { \\mathsf { \\Gamma } _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ) \\left( T _ { \\varphi _ { 0 } ^ { \\star } } ( x , a , x ^ { \\prime } ) - T _ { \\varphi _ { 1 } ^ { \\star } } ( x , a ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + } \\\\ & { \\qquad \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M ^ { \\star } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\left( T _ { \\varphi _ { 1 } ^ { \\star } } ( x , a ) - T _ { \\varphi _ { 0 } ^ { \\star } } ( x , a , x ^ { \\prime } ) + H _ { M ^ { \\star } } ( x , a ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\arg \\operatorname* { m a x } _ { \\theta } \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ) A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + } \\\\ & { \\qquad \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M ^ { \\star } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) ( - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) + H _ { M ^ { \\star } } ( x , a ) ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 156, + 839, + 297 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Based on the specific $f$ -divergence, we can represent $T$ and $f ^ { * } ( T )$ with a discriminator $D _ { \\varphi }$ . It can be verified that $f ( u ) = u \\log u - ( u + 1 ) \\log ( u + 1 )$ , $T _ { \\varphi } ( u ) = \\log D _ { \\varphi } ( u )$ , and $f ^ { * } ( T _ { \\varphi } ( u ) ) =$ $- \\log ( 1 - D _ { \\varphi } ( u ) )$ proposed in Nowozin et al. (2016) satisfies the condition $f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }$ (see Tab. 2). We select the former in the implementation and convert the tractable solution to: ", + "bbox": [ + 173, + 309, + 826, + 366 + ], + "page_idx": 21 + }, + { + "type": "equation", + "img_path": "images/913cd2a525614fb7f942e419a24666f6fec2c71503368bd7a24a6117750e8fd9.jpg", + "text": "$$\n\\begin{array} { r l } & { _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\left( H _ { M ^ { \\star } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } \\\\ { \\therefore } & { _ { \\varphi _ { 0 } ^ { \\star } } = \\arg \\operatorname* { m a x } _ { \\theta _ { 0 } } \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } \\\\ & { _ { \\varphi _ { 1 } ^ { \\star } } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 1 } } \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 166, + 369, + 897, + 443 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "where Aφ∗0 ,φ∗1 (x, a, x′) = log Dφ∗0 (x, a, x′) − log Dφ∗1 (x, a), EρκMθt [· is a simplification of $\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } [ \\cdot$ ]. ", + "bbox": [ + 174, + 462, + 826, + 502 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "B DISCUSSION OF THE THEORETICAL RESULTS ", + "text_level": 1, + "bbox": [ + 174, + 542, + 584, + 559 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "We summarize the limitations of current theoretical results and future work as follows: ", + "bbox": [ + 173, + 574, + 740, + 589 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "1. As discussed in Remark A.2, the solution Eq. 11 relies on $\\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) \\in [ 0 , c ] , \\forall a \\in \\mathcal { A } , \\forall x \\in$ $\\mathcal { X }$ . In some particular $M ^ { * }$ , it is intractable to derive a $\\beta$ that can generate an occupancy specified by $g ( x , a ) / \\alpha _ { M }$ . If more knowledge of $M ^ { * }$ or $\\beta ^ { * }$ is provided or some mild assumptions can be made on the properties of $M ^ { * }$ or $\\beta ^ { * }$ , we may model $\\rho$ in a more sophisticated approach to alleviating the above problem. \n2. In the tractable solution derivation, we ignore the term $\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )$ (See Eq. 20). The benefit is that $\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } )$ in the tractable solution can be estimated through offline datasets directly. Although the results in our experiments show that it does not produce significant negative effects in these tasks, ignoring $\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )$ indeed incurs extra bias in theory. In future work, techniques for estimating $\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )$ (Liu et al., 2020) can be incorporated to correct the bias. On the other hand, $\\alpha _ { M _ { \\theta _ { t } } }$ is also ignored in the process. $\\alpha _ { M _ { \\theta _ { t } } }$ can be regarded as a global rescaling term of the final objective Eq. 20. Intuitively, it constructs an adaptive learning rate for Eq. 20, which increases the step size when the model is better fitted and decreases the step size otherwise. It can be considered to further improve the learning process in future work, e.g., cooperating with empirical risk minimization by balancing the weights of the two objectives through $\\alpha _ { M _ { \\theta _ { t } } }$ . ", + "bbox": [ + 210, + 599, + 825, + 832 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "C SOCIETAL IMPACT ", + "text_level": 1, + "bbox": [ + 174, + 849, + 362, + 866 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "This work studies a method toward counterfactual environment model learning. Reconstructing an accurate environment of the real world will promote the wide adoption of decision-making policy optimization methods in real life, enhancing our daily experience. We are aware that decision-making policy in some domains like recommendation systems that interact with customers may have risks of causing price discrimination and misleading customers if inappropriately used. A promising way to reduce the risk is to introduce fairness into policy optimization and rules to constrain the actions (Also see our policy design in Sec. G.1.3). We are involved in and advocating research in such directions. We believe that business organizations would like to embrace fair systems that can ultimately bring long-term financial benefits by providing a better user experience. ", + "bbox": [ + 176, + 882, + 823, + 924 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 103, + 825, + 188 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "D AWRM-ORACLE PSEUDOCODE ", + "text_level": 1, + "bbox": [ + 176, + 207, + 470, + 223 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "We list the pseudocode of AWRM-oracle in Alg. 1. ", + "bbox": [ + 173, + 238, + 508, + 253 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Algorithm 1 AWRM with Oracle Counterfactual Datasets ", + "text_level": 1, + "bbox": [ + 174, + 267, + 555, + 282 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Input: $\\Phi$ : policy space; $N$ : total iterations ", + "bbox": [ + 173, + 286, + 406, + 314 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Process: ", + "text_level": 1, + "bbox": [ + 174, + 314, + 232, + 327 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "1: Generate counterfactual datasets $\\{ \\mathcal { D } _ { \\pi _ { \\phi } } \\}$ for all adversarial policies $\\pi _ { \\phi } , \\phi \\in \\Phi$ \n2: Initialize an environment model $M _ { \\theta }$ \n3: for $\\mathrm { i } = 1 { : } \\mathrm { N }$ do \n4: Select $\\mathcal { D } _ { \\pi _ { \\phi } }$ with worst prediction errors through $M _ { \\theta }$ from $\\{ \\mathcal { D } _ { \\pi _ { \\phi } } \\}$ \n5: Optimize $M _ { \\theta }$ with standard supervised learning based on $\\mathcal { D } _ { \\pi _ { \\phi } }$ \n6: end for ", + "bbox": [ + 179, + 332, + 710, + 416 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "E IMPLEMENTATION ", + "text_level": 1, + "bbox": [ + 174, + 446, + 361, + 463 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "E.1 DETAILS OF THE GALILEO IMPLEMENTATION ", + "text_level": 1, + "bbox": [ + 174, + 477, + 544, + 493 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "The approximation of Eq. 18 holds only when $\\textstyle { \\frac { p ( x ) } { q ( x ) } }$ is close to 1, which might not be satisfied. To handle the problem, we inject a standard supervised learning loss ", + "bbox": [ + 171, + 501, + 823, + 535 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/8ba42c7940c2ea542bde8b4314180594bc5b9ef185f5b6ac16999cb1a17f4b2f.jpg", + "text": "$$\n\\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right]\n$$", + "text_format": "latex", + "bbox": [ + 390, + 540, + 606, + 563 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "to replace the second term of the above objective when the output probability of $D$ is far away from 0.5 $\\bar { f } ^ { \\prime } ( 1 ) = \\log { 0 . 5 } )$ ). ", + "bbox": [ + 174, + 569, + 825, + 598 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "In the offline model-learning setting, we only have a real-world dataset $\\mathcal { D }$ collected by the behavior policy $\\mu$ . We learn a policy $\\hat { \\mu } \\approx \\bar { \\mu }$ via behavior cloning with $\\mathcal { D }$ (Pomerleau, 1991; Ho & Ermon, 2016) and let $\\hat { \\mu }$ be the policy $\\kappa$ . We regard $\\mathcal { D }$ as the empirical data distribution of $\\rho _ { M ^ { * } } ^ { \\kappa }$ and the trajectories collected by $\\hat { \\mu }$ in the model $M _ { \\theta _ { t } }$ as the empirical data distribution of $\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }$ . But the assumption $\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\mu ( a | x ) > 0$ might not be satisfied. In behavior cloning, we model $\\hat { \\mu }$ with a Gaussian distribution and constrain the lower bound of the variance with a small value $\\epsilon _ { \\mu } > 0$ to keep the assumption holding. Besides, we add small Gaussian noises $\\mathbf { u } \\sim \\mathcal { N } ( 0 , \\epsilon _ { D } )$ to the inputs of $D _ { \\varphi }$ to handle the mismatch between $\\rho _ { M ^ { * } } ^ { \\mu }$ and $\\rho _ { M ^ { * } } ^ { \\hat { \\mu } }$ due to $\\epsilon _ { \\mu }$ . In particular, for $\\varphi _ { 0 }$ and $\\varphi _ { 1 }$ learning, we have: ", + "bbox": [ + 173, + 602, + 825, + 729 + ], + "page_idx": 22 + }, + { + "type": "equation", + "img_path": "images/78d40402d4911316f7024794630711f2a1bed4b1c9fce5c830274a47b7d591a6.jpg", + "text": "$$\n\\begin{array} { r l } & { \\gamma _ { 0 } ^ { * } = \\arg \\underset { \\varphi _ { 0 } } { \\operatorname* { m a x } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) ) \\right] } \\\\ & { \\gamma _ { 1 } ^ { * } = \\arg \\underset { \\varphi _ { 1 } } { \\operatorname* { m a x } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) \\right] , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 731, + 839, + 780 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "where EρκMθ , is a simplification of $\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\mathbf { u } \\sim \\mathcal { N } ( 0 , \\epsilon _ { D } ) } [ \\cdot ]$ and $\\mathbf { u } = [ u _ { x } , u _ { a } , u _ { x ^ { \\prime } } ] .$ ", + "bbox": [ + 176, + 785, + 746, + 808 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "On the other hand, we notice that the first term in Eq. 21 is similar to the objective of GAIL (Ho & Ermon, 2016) by regarding $M _ { \\theta }$ as the policy to learn and $\\kappa$ as the environment to generate data. For better capability in sequential environment model learning, here we introduce some practical tricks inspired by GAIL for model learning (Shi et al., 2019; Shang et al., 2019): we introduce an MDP for $\\kappa$ and $M _ { \\theta }$ , where the reward is defined by the discriminator $D$ , i.e., $r ( x , a , x ^ { \\prime } ) ~ = ~ \\log D ( x , a , x ^ { \\prime } )$ . $M _ { \\theta }$ is learned to maximize the cumulative rewards. With advanced policy gradient methods (Schulman et al., 2015; 2017), the objective is converted to maxθ $\\left[ \\bar { A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x , a , x ^ { \\prime } ) \\right]$ , where ${ \\cal A } \\ = \\ Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } \\ - \\ V _ { M _ { \\theta _ { t } } } ^ { \\kappa }$ , ", + "bbox": [ + 173, + 811, + 826, + 926 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Algorithm 2 GALILEO pseudocode ", + "text_level": 1, + "bbox": [ + 176, + 103, + 415, + 118 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "", + "text_level": 1, + "bbox": [ + 174, + 121, + 218, + 133 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "$\\mathcal { D } _ { \\mathrm { r e a l } }$ : offline dataset sampled from $\\rho _ { M ^ { * } } ^ { \\mu }$ where $\\mu$ is the behavior policy; \n$N$ : total iterations; ", + "bbox": [ + 173, + 133, + 647, + 161 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Process: ", + "text_level": 1, + "bbox": [ + 173, + 162, + 232, + 175 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "1: Approximate a behavior policy $\\hat { \\mu }$ via behavior cloning \n2: Initialize an environment model $M _ { \\theta _ { 1 } }$ \n3: for $t = 1 : N$ do \n4: Use $\\hat { \\mu }$ to generate a dataset $\\mathcal { D } _ { \\mathrm { g e n } }$ with the model $M _ { \\theta _ { t } }$ \n5: Update the discriminators $D _ { \\varphi _ { 0 } }$ and $D _ { \\varphi _ { 1 } }$ via Eq. 26 and Eq. 27 respectively, where $\\rho _ { M _ { \\theta _ { t } } } ^ { \\hat { \\mu } }$ is \nestimated by $\\mathcal { D } _ { \\mathrm { g e n } }$ and $\\rho _ { M ^ { * } } ^ { \\mu }$ is estimated by $\\mathcal { D } _ { \\mathrm { r e a l } }$ \n6: Update $Q$ and $V$ via Eq. 24 and Eq. 25 through $\\mathcal { D } _ { \\mathrm { g e n } }$ , $D _ { \\varphi _ { 0 } }$ , and $D _ { \\varphi _ { 1 } }$ \n7: Update the model $M _ { \\theta _ { t } }$ via the first term of Eq. 23, which is implemented with a standard \npolicy gradient method like TRPO (Schulman et al., 2015) or PPO (Schulman et al., 2017). \nRecord the policy gradient gpg \n8: if $p _ { 0 } < \\mathbb { E } _ { \\mathcal { D } _ { \\mathrm { g e n } } } \\left[ D _ { \\varphi _ { 0 } } ( x _ { t } , \\hat { a _ { t } } , x _ { t + 1 } ) \\right] < p _ { 1 }$ then \n9: Compute the gradient of $M _ { \\theta _ { t } }$ via the second term of Eq. 23 and record it as $g _ { \\mathrm { s l } }$ \n10: else \n11: Compute the gradient of $M _ { \\theta _ { t } }$ via Eq. 22 and record it as $g _ { \\mathrm { s l } }$ \n12: end if \n13: Rescale $g _ { \\mathrm { s l } }$ via Eq. 28 \n14: Update the model $M _ { \\theta _ { t } }$ via the gradient $g _ { \\mathrm { s l } }$ and obtain $M _ { \\theta _ { t + 1 } }$ ", + "bbox": [ + 179, + 183, + 826, + 433 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "$\\begin{array} { r } { Q _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = { \\mathbb E } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r ( x _ { t } , a _ { t } , x _ { t + 1 } ) \\mid ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } \\right] } \\end{array}$ , and ${ \\cal V } _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a ) =$ $\\mathbb { E } _ { M _ { \\bar { \\theta } } } \\left[ Q _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\right]$ . $A$ in Eq. 21 can also be constructed similarly. Although it looks unnecessary in theory since the one-step optimal model $M _ { \\theta }$ is the global optimal model in this setting, the technique is helpful in practice as it makes $A$ more sensitive to the compounding effect of one-step prediction errors: we would consider the cumulative effects of prediction errors induced by multi-step transitions in environments. In particular, to consider the cumulative effmulti-step of transitions in environments, we overwrite function $A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } }$ preas $A _ { \\varphi _ { 0 } ^ { \\ast } , \\varphi _ { 1 } ^ { \\ast } } = Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } - V _ { M _ { \\theta _ { t } } } ^ { \\kappa }$ , where $\\begin{array} { r } { Q _ { M _ { \\theta _ { \\star } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 0 } ^ { \\star } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , \\bar { M } _ { \\theta _ { t } } \\right] } \\end{array}$ and $\\begin{array} { r } { V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \\kappa , M _ { \\theta _ { t } } \\right] , } \\end{array}$ . To give an algorithm for singlestep environment model learning, we can just set $\\gamma$ in $Q$ and $V$ to 0. ", + "bbox": [ + 173, + 492, + 826, + 657 + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/250eec592a0278e03dcb9c611ad65d4003d340434d2a84445af36733f46969b9.jpg", + "image_caption": [ + "15: end for ", + "Figure 7: Illustration of the workflow of the GALILEO algorithm. " + ], + "image_footnote": [], + "bbox": [ + 179, + 702, + 821, + 883 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "By adopting the above implementation techniques, we convert the objective into the following formulation ", + "bbox": [ + 171, + 103, + 825, + 132 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/6a73c9667b1b03e9f317568a2a21065bca1ae266f00327e9f119736f1381d336.jpg", + "text": "$$\nt + 1 = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\star } } \\left[ A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\star } } \\left[ \\left( H _ { M ^ { \\star } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 202, + 141, + 861, + 162 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "s.t. ", + "bbox": [ + 174, + 198, + 196, + 209 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/c0d58b24005654f7799b4139385c1b768d7080cb3287d2e0e505d2fca3235f30.jpg", + "text": "$$\n\\begin{array} { r l } & { Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb E [ \\displaystyle \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 0 } ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } ] \\qquad ( 2 4 ) } \\\\ & { V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb E [ \\displaystyle \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \\kappa , M _ { \\theta _ { t } } ] } \\\\ & { \\varphi _ { 0 } ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 0 } ^ { * } } \\mathbb E _ { \\rho _ { M ^ { * } } ^ { \\kappa } , \\textbf { u } } [ \\log D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) ] + \\mathbb E _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\textbf { u } } [ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } ) ] ] } \\\\ & { \\varphi _ { 1 } ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 1 } ^ { * } } \\mathbb E _ { \\rho _ { M ^ { * } } ^ { \\kappa } , \\textbf { u } } [ \\log D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ] + \\mathbb E _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\textbf { u } } [ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) ] , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 235, + 184, + 882, + 325 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "where $A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , \\underline { { a } } , x ^ { \\prime } ) = Q _ { M _ { \\theta } } ^ { \\kappa } ( x , a , \\underline { { x } } ^ { \\prime } ) - V _ { M _ { \\theta } } ^ { \\kappa } ( x , \\underline { { a } } )$ . In practice, GALILEO optimizes the first term of Eq. 23 with conservative policy gradient algorithms (e.g., PPO (Schulman et al., 2017) or TRPO (Schulman et al., 2015)) to avoid unreliable gradients for model improvements. Eq. 26 and Eq. 27 are optimized with supervised learning. The second term of Eq. 23 is optimized with supervised learning with a re-weighting term $- A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } + H _ { M ^ { * } }$ . Since $H _ { M ^ { * } }$ is unknown, we use $H _ { M _ { \\theta } }$ to estimate it. When the mean output probability of a batch of data is larger than 0.6 or small than 0.4, we replace the second term of Eq. 23 with a standard supervised learning in Eq. 22. Besides, unreliable gradients also exist in the process of optimizing the second term of Eq. 23. In our implementation, we use the scale of policy gradients to constrain the gradients of the second term of Eq. 23. In particular, we first compute the $l _ { 2 }$ -norm of the gradient of the first term of Eq. 23 via conservative policy gradient algorithms, named $| | g _ { \\mathrm { p g } } | | _ { 2 }$ . Then we compute the $l _ { 2 }$ -norm of the gradient of the second term of Eq. 23, name $| | g _ { \\mathrm { s l } } | | _ { 2 }$ . Finally, we rescale the gradients of the second term $g _ { \\mathrm { s l } }$ by ", + "bbox": [ + 173, + 348, + 826, + 530 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/b8ea05dedc4d1c7fb1f382c89b84db33367cdb5848c41c2fddafbb413d0abf1c.jpg", + "text": "$$\ng _ { \\mathrm { s l } } g _ { \\mathrm { s l } } \\frac { \\vert \\vert g _ { \\mathrm { p g } } \\vert \\vert _ { 2 } } { \\operatorname* { m a x } \\{ \\vert \\vert g _ { \\mathrm { p g } } \\vert \\vert _ { 2 } , \\vert \\vert g _ { \\mathrm { s l } } \\vert \\vert _ { 2 } \\} } .\n$$", + "text_format": "latex", + "bbox": [ + 390, + 541, + 607, + 575 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "For each iteration, Eq. 23, Eq. 26, and Eq. 27 are trained with certain steps (See Tab. 5) following the same framework as GAIL. Based on the above techniques, we summarize the pseudocode of GALILEO in Alg. 2, where $p _ { 0 }$ and $p _ { 1 }$ are set to 0.4 and 0.6 in all of our experiments. The overall architecture is shown in Fig. 7. ", + "bbox": [ + 173, + 585, + 825, + 643 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "to Reviewer bcJS ", + "bbox": [ + 838, + 628, + 954, + 642 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "E.2 CONNECTION WITH PREVIOUS ADVERSARIAL ALGORITHMS ", + "text_level": 1, + "bbox": [ + 173, + 665, + 635, + 680 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "Standard GAN (Goodfellow et al., 2014) can be regarded as a partial implementation including the first term of Eq. 23 and Eq. 26 by degrading them into the single-step scenario. In the context of GALILEO, the objective of GAN is ", + "bbox": [ + 176, + 693, + 825, + 736 + ], + "page_idx": 24 + }, + { + "type": "equation", + "img_path": "images/d97b5374a96162d6f13436b835ca976cd3c134e79756321353bb6471be8747ec.jpg", + "text": "$$\n\\begin{array} { r l } & { \\theta _ { t + 1 } = \\underset { \\theta } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } \\\\ & { s . t . \\quad \\varphi * = \\underset { \\varphi } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ \\log ( 1 - D _ { \\varphi } ( x , a , x ^ { \\prime } ) ) \\right] , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 236, + 746, + 759, + 796 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "where $\\begin{array} { r c l } { { A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) } } & { { = } } & { { \\log D _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) } } \\end{array}$ . In the single-step scenario, $\\begin{array} { r l } { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } & { { } = } \\end{array}$ $\\rho _ { 0 } ( x ) \\kappa ( a | x ) M _ { \\theta _ { t } } ( x ^ { \\prime } | a , x )$ . The term $\\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right]$ can convert to $\\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\right]$ by replacing the gradient of $M _ { \\theta _ { t } } ( x ^ { \\prime } | x , a ) \\nabla _ { \\theta } \\log M _ { \\theta } ( x ^ { \\prime } | x , a )$ with $\\nabla _ { \\boldsymbol { \\theta } } M _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } ^ { \\prime } | \\boldsymbol { x } , a )$ (Sutton & Barto, 1998). Previous algorithms like GANITE (Yoon et al., 2018) and SCIGAN (Bica et al., 2020) can be regarded as variants of the above training framework. ", + "bbox": [ + 173, + 806, + 825, + 892 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "The first term of Eq. 23 and Eq. 26 are similar to the objective of GAIL by regarding $M _ { \\theta }$ as the “policy” to imitate and $\\hat { \\mu }$ as the “environment” to collect data. In the context of GALILEO, the ", + "bbox": [ + 171, + 896, + 825, + 924 + ], + "page_idx": 24 + }, + { + "type": "text", + "text": "objective of GAIL is: ", + "bbox": [ + 173, + 103, + 315, + 118 + ], + "page_idx": 25 + }, + { + "type": "equation", + "img_path": "images/dcc7cb0b1e1425d35c6360b351bcb8b8deac7e8830c1f531755e2409a6b68632.jpg", + "text": "$$\n\\begin{array} { r l } { \\displaystyle \\theta _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } & { } \\\\ { \\displaystyle s . t . } & { Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } \\right] } \\\\ { \\displaystyle \\varphi ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ \\log ( 1 - D _ { \\varphi } ( x , a , x ^ { \\prime } ) ) \\right] , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 217, + 127, + 781, + 215 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "where $A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) = Q _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) - V _ { M _ { \\theta } } ^ { \\kappa } ( x , a )$ and $V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb { E } _ { M _ { \\theta _ { t } } ( x , a ) } \\left[ Q ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\right] ,$ ", + "bbox": [ + 173, + 223, + 797, + 243 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "F ADDITIONAL RELATED WORK ", + "text_level": 1, + "bbox": [ + 176, + 265, + 459, + 281 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "Our primitive objective is inspired by weighted empirical risk minimization (WERM) based on inverse propensity score (IPS). WERM is originally proposed to solve the generalization problem of domain adaptation in machine learning literature. For instance, we would like to train a predictor $M ( y | x )$ in a domain with distribution $P _ { \\mathrm { t r a i n } } ( x )$ to minimize the prediction risks in the domain with distribution $P _ { \\mathrm { { t e s t } } } ( x )$ , where $P _ { \\mathrm { t e s t } } \\neq P _ { \\mathrm { t e s t } }$ . To solve the problem, we can train a weighted objective $\\begin{array} { r } { \\operatorname* { m a x } _ { M } \\mathbb { E } _ { x \\sim P _ { \\mathrm { t r a i n } } } [ \\frac { P _ { \\mathrm { t e s t } } ( x ) } { P _ { \\mathrm { t r a i n } } ( x ) } \\log M ( y | x ) ] } \\end{array}$ , which is called weighted empirical risk minimization methods (Ben-David et al., 2006; 2010; Cortes et al., 2010; Byrd & Lipton, 2019; Quinonero-Candela et al., 2008). These results have been extended and applied to causal inference, where the predictor is required to be generalized from the data distribution in observational studies (source domain) to the data distribution in randomized controlled trials (target domain) (Shimodaira, 2000; Assaad et al., 2021; Hassanpour $\\&$ Greiner, 2019; Jung et al., 2020; Johansson et al., 2018). In this case, the input features include a state $x$ (a.k.a. covariates) and an action $a$ (a.k.a. treatment variable) which is sampled from a policy. We often assume the distribution of $x$ , $P ( x )$ is consistent between the source domain and the test domain, then we have Ptest(x)Ptrain(x) = P (x)β(a|x)P (x)µ(a|x) , where $\\mu$ and $\\beta$ are the policies in source and target domains respectively. In Shimodaira (2000); Assaad et al. (2021); Hassanpour $\\&$ Greiner (2019), the policy in randomized controlled trials is modeled as a uniform policy, then $\\begin{array} { r } { \\frac { P _ { \\mathrm { t e s t } } ( x ) } { P _ { \\mathrm { t r a i n } } ( x ) } = \\frac { P ( x ) \\beta ( a | x ) } { P ( x ) \\mu ( a | x ) } = \\frac { \\beta ( a | x ) } { \\mu ( a | x ) } \\propto \\frac { 1 } { \\mu ( a | x ) } } \\end{array}$ . ${ \\frac { 1 } { \\mu ( a | x ) } }$ is also known as inverse propensity score (IPS). In Johansson et al. (2018), it assumes that the policy in the target domain is predefined as $\\beta ( a | x )$ before environment model learning, then it uses $\\frac { \\beta } { \\mu }$ as the IPS. The differences between AWRM and previous works are fallen in two aspects: (1) We consider the distribution-shift problem in the sequential decision-making scenario. In this scenario, we not only consider the action distribution mismatching between the behavior policy $\\mu$ and the policy to evaluation $\\beta$ , but also the follow-up effects of policies to the state distribution; (2) For faithful offline policy optimization, we require the environment model to have generalization ability in numerous different policies. The objective of AWRM is proposed to guarantee the generalization ability of $M$ in numerous different policies instead of a specific policy. ", + "bbox": [ + 173, + 300, + 825, + 681 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "On a different thread, there are also studies that bring counterfactual inference techniques of causal inference into model-based RL (Buesing et al., 2019; Pitis et al., 2020; Sontakke et al., 2021). These works consider that the transition function is relevant to some hidden noise variables and use Pearl-style structural causal models (SCMs), which is a directed acyclic graphs to define the causality of nodes in an environment, to handle the problem. SCMs can help RL in different ways: Buesing et al. (2019) approximate the posterior of the noise variables based on the observation of data, and environment models are learned based on the inferred noises. The generalization ability is improved if we can infer the correct value of the noise variables. Pitis et al. (2020) discover several local causal structural models of a global environment model, then data augmentation strategies by leveraging these local structures to generate counterfactual experiences. Sontakke et al. (2021) proposes a representation learning technique for causal factors, which is an instance of the hidden noise variables, in partially observable Markov decision processes (POMDPs). With the learned representation of causal factors, the performance of policy learning and transfer in downstream tasks will be improved. Instead of considering the hidden noise variables in the environments, our study considers the environment model learning problem in the fully observed setting and focuses on unbiased causal effect estimation in the offline dataset under behavior policies collected with selection bias. ", + "bbox": [ + 173, + 688, + 825, + 924 + ], + "page_idx": 25 + }, + { + "type": "text", + "text": "In offline model-based RL, the problem is called distribution shift (Yu et al., 2020; Levine et al., 2020; Chen et al., 2021) which has received great attentions. However, previous algorithms do not handle the model learning challenge directly but propose techniques to suppress policy sampling and learning in risky regions (Yu et al., 2020; Kidambi et al., 2020). Although these algorithms have made great progress in offline policy optimization in many tasks, so far, how to learn a better environment model in this scenario has rarely been discussed. ", + "bbox": [ + 174, + 103, + 825, + 188 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "G EXPERIMENT DETAILS ", + "text_level": 1, + "bbox": [ + 176, + 207, + 400, + 223 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "G.1 SETTINGS ", + "text_level": 1, + "bbox": [ + 174, + 238, + 289, + 252 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "G.1.1 GENERAL NEGATIVE FEEDBACK CONTROL (GNFC) ", + "bbox": [ + 174, + 263, + 594, + 279 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "The design of GNFC is inspired by a classic type of scenario that behavior policies $\\mu$ have selection bias and easily lead to counterfactual risks: For some internet platforms, we would like to allocate budgets to a set of targets (e.g., customers or cities) to increase the engagement of the targets in the platforms. Our task is to train a model to predict targets’ feedback on engagement given targets’ features and allocated budgets. ", + "bbox": [ + 174, + 289, + 825, + 359 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "In these tasks, for better benefits, the online working policy (i.e., the behavior policy) will tend to cut down the budgets if targets have better engagement, otherwise, the budgets might be increased. The risk of counterfactual environment model learning in the task is that: the object with better historical engagement will be sent to smaller budgets because of the selection bias of the behavior policies, then the model might exploit this correlation for learning and get a conclusion that: increasing budgets will reduce the targets’ engagement, which violates the real causality. We construct an environment and a behavior policy to mimic the above process. In particular, the behavior policy $\\mu _ { G N F C }$ is ", + "bbox": [ + 173, + 366, + 825, + 464 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/7366a38ced3d0e1a19ebf10b54b8fb13efb412b10f6569aee91727b2cee843e6.jpg", + "text": "$$\n\\mu _ { G N F C } ( x ) = \\frac { ( 6 2 . 5 - \\mathrm { m e a n } ( x ) ) } { 1 5 } + \\epsilon ,\n$$", + "text_format": "latex", + "bbox": [ + 370, + 468, + 625, + 501 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "where $\\epsilon$ is a sample noise, which will be discussed later. The environment includes two parts: ", + "bbox": [ + 173, + 505, + 782, + 520 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "(1) response function $M _ { 1 } ( y | x , a )$ : ", + "bbox": [ + 174, + 525, + 397, + 540 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/aeabc49d314cce4db29df3c871644034fe0c1b6bc9e8e50cbea5b151dbafa052.jpg", + "text": "$$\nM _ { 1 } ( y | x , a ) = \\mathcal { N } ( \\mathrm { m e a n } ( x ) + a , 2 )\n$$", + "text_format": "latex", + "bbox": [ + 382, + 546, + 616, + 564 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "(2) mapping function $M _ { 2 } ( x ^ { \\prime } | x , y )$ ", + "bbox": [ + 174, + 577, + 401, + 592 + ], + "page_idx": 26 + }, + { + "type": "equation", + "img_path": "images/be029771be3f70c87d8809389293e5a4aa3ecb5df20f4ec1c492ac7a5ce34ddf.jpg", + "text": "$$\nM _ { 2 } ( x ^ { \\prime } | x , a , y ) = y - \\mathrm { m e a n } ( x ) + x\n$$", + "text_format": "latex", + "bbox": [ + 379, + 597, + 619, + 614 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "The transition function $M ^ { * }$ is a composite of $M ^ { * } ( x ^ { \\prime } | x , a ) = M _ { 2 } ( x ^ { \\prime } | x , a , M _ { 1 } ( y | x , a ) )$ . The behavior policies have selection bias: the actions taken are negatively correlated with the states, as illustrated in Fig. 8(a) and Fig. 8(b). We control the difficulty of distinguishing the correct causality of $x , a ,$ , and $y$ by designing different strategies of noise sampling on $\\epsilon$ . In principle, with a larger number or more pronounced disturbances, there are more samples violating the correlation between $x$ and $a$ , then more samples can be used to find the correct causality. Therefore, we can control the difficulty of counterfactual environment model learning by controlling the strength of disturbance. In particular, we sample $\\epsilon$ from a uniform distribution $U ( - e , e )$ with probability $p$ . That is, $\\epsilon = 0$ with probability $1 - p$ and $\\epsilon \\sim U ( - e , e )$ with probability $p$ . Then with larger $p$ , there are more samples in the dataset violating the negative correlation (i.e., $\\mu _ { G N F C } ,$ ), and with larger $e$ , the difference of the feedback will be more obvious. By selecting different $e$ and $p$ , we can construct different tasks to verify the effectiveness and ability of the counterfactual environment model learning algorithm. ", + "bbox": [ + 173, + 619, + 826, + 787 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "G.1.2 THE CANCER GENOMIC ATLAS (TCGA) ", + "text_level": 1, + "bbox": [ + 174, + 801, + 514, + 815 + ], + "page_idx": 26 + }, + { + "type": "text", + "text": "The Cancer Genomic Atlas (TCGA) is a project that has profiled and analyzed large numbers of human tumors to discover molecular aberrations at the DNA, RNA, protein, and epigenetic levels. The resulting rich data provide a significant opportunity to accelerate our understanding of the molecular basis of cancer. We obtain features, $\\mathbf { x }$ , from the TCGA dataset and consider three continuous treatments as done in SCIGAN (Bica et al., 2020). Each treatment, $a$ , is associated with a set of parameters, $\\mathbf { v } _ { 1 }$ , $\\mathbf { v } _ { 2 }$ , $\\mathbf { v } _ { 3 }$ , that are sampled randomly by sampling a vector from a standard normal distribution and scaling it with its norm. We assign interventions by sampling a treatment, $a$ , from a beta distribution, $a \\mid \\mathbf { x } \\sim \\operatorname { B e t a } \\left( \\alpha , \\beta \\right)$ . $\\alpha \\geq 1$ controls the sampling bias and $\\textstyle { \\beta = { \\frac { \\alpha - 1 } { a ^ { * } } } + 2 - \\alpha }$ where $a ^ { * }$ is the optimal treatment. This setting of $\\beta$ ensures that the mode of Beta $( \\alpha , \\tilde { \\beta } )$ is $a ^ { * }$ . ", + "bbox": [ + 174, + 825, + 825, + 924 + ], + "page_idx": 26 + }, + { + "type": "image", + "img_path": "images/55abe1b08792484f5a786bb75b1b3aab1b1d67d16735214a33fde14020f480e8.jpg", + "image_caption": [ + "Figure 8: Illustration of information about the collected dataset in GNFC. Each color of the line denotes one of the collected trajectories. The $\\mathrm { X }$ -axis denotes the timestep of a trajectory. " + ], + "image_footnote": [], + "bbox": [ + 236, + 123, + 741, + 276 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 352, + 825, + 381 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "The calculation of treatment response and optimal treatment are shown in Table 3. ", + "bbox": [ + 176, + 387, + 710, + 401 + ], + "page_idx": 27 + }, + { + "type": "table", + "img_path": "images/c7b8468ffd79d11fa3b999eb55c75879cc969434cf2cccead0c129b634700ed5.jpg", + "table_caption": [ + "Table 3: Treatment response used to generate semi-synthetic outcomes for patient features $\\mathbf { x }$ . In the experiments, we set $C = 1 0$ . " + ], + "table_footnote": [], + "table_body": "
TreatmentTreatment ResponseOptimal treatment
1f1(x,a1)=C((v)Tx+12(ν)T xa1-12(ν)T x²)(v)x a= 2(v3)x
2a2))(v)x 2(v)x
3(v2)x) f3(x,a3)=C((vi)Tx+12a3(a3 -b)²,where b=0.75
", + "bbox": [ + 173, + 459, + 820, + 571 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "We conduct experiments on three different treatments separately and change the value of bias $\\alpha$ to assess the robustness of different methods to treatment bias. When the bias of treatment is large, which means $\\alpha$ is large, the training set contains data with a strong bias on treatment so it would be difficult for models to appropriately predict the treatment responses out of the distribution of training data. ", + "bbox": [ + 173, + 590, + 825, + 660 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "G.1.3 BUDGET ALLOCATION TASK TO THE TIME PERIOD (BAT) ", + "text_level": 1, + "bbox": [ + 174, + 681, + 630, + 695 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "We deploy GALILEO in a real-world large-scale food-delivery platform. The platform contains various food stores, and food delivery clerks. The overall workflow is as follows: the platform presents the nearby food stores to the customers and the customers make orders, i.e., purchase take-out foods from some stores on the platform. The food delivery clerks can select orders from the platform to fulfill. After an order is selected to fulfill, the delivery clerks will take the ordered take-out foods from the stores and then send the food to the customers. The platform will pay the delivery clerks (mainly in proportion to the distance between the store and the customers’ location) once the orders are fulfilled. An illustration of the workflow can be found in Fig. 9. ", + "bbox": [ + 174, + 707, + 825, + 819 + ], + "page_idx": 27 + }, + { + "type": "text", + "text": "However, there is an imbalance problem between the demanded orders from customers and the supply of delivery clerks to fulfill these orders. For example, at peak times like lunchtime, there will be many more demanded orders than at other times, and the existed delivery clerks might not be able to fulfill all of these orders timely. The goal of the Budget Allocation task to the Time period (BAT) is to handle the imbalance problem in time periods by sending reasonable allowances to different time periods. More precisely, the goal of BAT is to make all orders (i.e., the demand) sent in different time periods can be fulfilled (i.e., the supply) timely. ", + "bbox": [ + 174, + 825, + 825, + 924 + ], + "page_idx": 27 + }, + { + "type": "image", + "img_path": "images/b1e7c25ec14554cb2a78317429793502eb19c5fd3212ebf07b308c2649c646bd.jpg", + "image_caption": [ + "Figure 9: Illustration of the workflow of the food-delivery platform. " + ], + "image_footnote": [], + "bbox": [ + 267, + 103, + 727, + 318 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "To handle the imbalance problem in different time periods, in the platform, the orders in different time periods $t \\in [ 0 , 1 , 2 . . . , 2 3 ]$ will be allocated with different allowances $c \\in \\mathcal { N } ^ { + }$ . For example, at $1 0 \\mathrm { A . M }$ . (i.e., $t = 1 0$ ), we add $0 . 5 \\ S$ (i.e., $c = 0 . 5$ ) allowances to all of the demanded orders. From $1 0 \\mathrm { \\ A . M }$ . to 11 A.M., the delivery clerks who take orders and send food to customers will receive extra allowances. Specifically, if the platform pays the delivery clerks $2 \\mathfrak { H }$ for fulfilling the order, now he/she will receive $2 . 5 \\ S$ . For each day, the budget of allowance $C$ is fixed. We should find the best budget allocation policy $\\pi ^ { * } ( c | t )$ of the limited budget $C$ to make as many orders as possible can be taken timely. ", + "bbox": [ + 173, + 376, + 825, + 488 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "To find the policy, we first learn a model to reconstruct the response of allowance for each delivery clerk $\\hat { M } \\big ( y _ { t + 1 } \\vert s _ { t } , p _ { t } , c _ { t } \\big )$ , where $y _ { t + 1 }$ is the taken orders of the delivery clerks in state $s _ { t } , c _ { t }$ is the allowances, $p _ { t }$ denotes static features of the time period $t$ . In particular, the state $s _ { t }$ includes historical order-taken information of the delivery clerks, current orders information, the feature of weather, city information, and so on. Then we use a rule-based mapping function $f$ to fill the complete next time-period states, i.e., $s _ { t + 1 } = f ( s _ { t } , p _ { t } , c _ { t } , y _ { t + 1 } )$ . Here we define the composition of the above functions $\\hat { M }$ and $f$ as $\\hat { M } _ { f }$ . Finally, we learn a budget allocation policy based on the learned model. For each day, the policy we would like to find is: ", + "bbox": [ + 173, + 493, + 826, + 611 + ], + "page_idx": 28 + }, + { + "type": "equation", + "img_path": "images/4ed7580e27e1590c372d6def6b135f8cbcccb6b39699b0f23d3340c6516875e9.jpg", + "text": "$$\n\\begin{array} { r } { \\underset { \\pi } { \\operatorname* { m a x } } \\mathbb { E } _ { s _ { 0 } \\sim { \\cal S } } \\left[ \\displaystyle \\sum _ { t = 0 } ^ { 2 3 } y _ { t } \\vert \\hat { M } _ { f } , \\pi \\right] , } \\\\ { \\mathrm { s . t . , } \\displaystyle \\sum _ { t , s \\in { \\cal S } } c _ { t } y _ { t } \\leq C } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 400, + 618, + 598, + 698 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "In our experiment, we evaluate the degree of balancing between demand and supply by computing the averaged five-minute order-taken rate, that is the percentage of orders picked up within five minutes. Note that the behavior policy is fixed for the long term in this application. So we directly use the data replay with a small scale of noise (See Tab. 5) to reconstruct the behavior policy for model learning in GALILEO. ", + "bbox": [ + 174, + 703, + 825, + 773 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "Also note that although we model the response for each delivery clerk, for fairness, the budget allocation policy is just determining the allowance of each time period $t$ and keeps the allowance to each delivery clerk $s$ the same. ", + "bbox": [ + 174, + 780, + 825, + 824 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "G.2 BASELINE ALGORITHMS", + "text_level": 1, + "bbox": [ + 174, + 842, + 390, + 856 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "The algorithm we compared are: (1) Supervised Learning (SL): training a environment model to minimize the expectation of prediction error, without considering the counterfactual risks; (2) inverse propensity weighting (IPW) (Spirtes, 2010): a practical way to balance the selection bias by re-weighting. It can be regarded as $\\begin{array} { r } { \\dot { \\omega } = \\frac { 1 } { \\hat { \\mu } } } \\end{array}$ , where $\\hat { \\mu }$ is another model learned to approximate the behavior policy; (3) SCIGAN: a recent proposed adversarial algorithm for model learning for continuous-valued interventions (Bica et al., 2020). All of the baselines algorithms are implemented with the same capacity of neural networks (See Tab. 5). ", + "bbox": [ + 174, + 867, + 825, + 926 + ], + "page_idx": 28 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 103, + 825, + 146 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "G.2.1 SUPERVISED LEARNING (SL) ", + "text_level": 1, + "bbox": [ + 176, + 166, + 436, + 181 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "As a baseline, we train a multilayer perceptron model to directly predict the response of different treatments, without considering the counterfactual risks. We use mean square error to estimate the perfor ance of our model so that the loss function can be expressed as $\\begin{array} { r } { \\boldsymbol { M } \\boldsymbol { \\bar { S } } \\boldsymbol { E } = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\left( y _ { i } - \\hat { y } _ { i } \\right) ^ { 2 } } \\end{array}$ $n$ $\\hat { y }$ is the predicted response. In practice, we train our SL models using Adam optimizer and the initial learning rate $3 e ^ { - \\bar { 4 } }$ on both datasets TCGA and GNFC. The architecture of the neural networks is listed in Tab. 5. ", + "bbox": [ + 174, + 194, + 826, + 280 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "G.2.2 INVERSE PROPENSITY WEIGHTING (IPW) ", + "text_level": 1, + "bbox": [ + 174, + 301, + 524, + 315 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "Inverse propensity weighting (Spirtes, 2010) is an approach where the treatment outcome model uses sample weights to balance the selection bias by re-weighting. The weights are defined as the inverse propensity of actually getting the treatment, which can be expressed as $\\dot { \\overline { { { \\hat { \\mu } } ( a | x ) } } }$ , where $x$ stands for the feature vectors in a dataset, $a$ is the corresponding action and ${ \\hat { \\mu } } ( a | x )$ indicates the action taken probability of $a$ given the features $x$ within the dataset. $\\hat { \\mu }$ is learned with standard supervised learning. Standard IPW leads to large weights for the points with small sampling probabilities and finally makes the learning process unstable. We solve the problem by clipping the propensity score: $\\hat { \\mu } \\gets \\operatorname* { m i n } ( \\hat { \\mu } , 0 . 0 5 )$ , which is common used in existing studies (Ionides, 2008). The loss function can thus be expressed as 1n Pni=1 1µˆ(ai|xi) $\\begin{array} { r } { { \\frac { 1 } { n } } \\sum _ { i = 1 } ^ { n } { \\frac { 1 } { { \\hat { \\mu } } \\left( a _ { i } | x _ { i } \\right) } } \\left( y _ { i } - \\hat { y } _ { i } \\right) ^ { 2 } } \\end{array}$ . The architecture of the neural networks is listed in Tab. 5. ", + "bbox": [ + 173, + 328, + 825, + 476 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "G.2.3 SCIGAN ", + "text_level": 1, + "bbox": [ + 174, + 496, + 297, + 511 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "SCIGAN (Bica et al., 2020) is a model that uses generative adversarial networks to learn the data distribution of the counterfactual outcomes and thus generate individualized response curves. SCIGAN does not place any restrictions on the form of the treatment-does response functions and is capable of estimating patient outcomes for multiple treatments, each with an associated parameter. SCIGAN first trains a generator to generate response curves for each sample within the training dataset. The learned generator can then be used to train an inference network using standard supervised methods. For fair comparison, we increase the number of parameters for the the open-source version of SCIGAN so that the SCIGAN model can have same order of magnitude of network parameters as GALILEO. In addition, we also finetune the hyperparameters (Tab. 4) of the enlarged SCIGAN to realize its full strength. We set num_dosage_samples 9 and $\\lambda = 1 0$ . ", + "bbox": [ + 174, + 523, + 825, + 662 + ], + "page_idx": 29 + }, + { + "type": "table", + "img_path": "images/8ec5dc6c111badbd1e205f3b3f1e840f45d4e74ac912ed430f8883e738032a52.jpg", + "table_caption": [ + "Table 4: Table of hyper-parameters for SCIGAN. " + ], + "table_footnote": [], + "table_body": "
ParameterValues
Number of samples3,5,7,9, 11
入0.1,1,10, 20
", + "bbox": [ + 367, + 708, + 625, + 765 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "G.3 HYPER-PARAMETERS ", + "text_level": 1, + "bbox": [ + 174, + 800, + 367, + 815 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "We list the hyper-parameter of GALILEO in Tab. 5. ", + "bbox": [ + 176, + 829, + 511, + 844 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "G.4 COMPUTATION RESOURCES ", + "text_level": 1, + "bbox": [ + 176, + 866, + 411, + 881 + ], + "page_idx": 29 + }, + { + "type": "text", + "text": "We use one Tesla V100 PCIe 32GB GPU and a 32-core Intel(R) Xeon(R) Gold 5118 CPU @ 2.30GHz to train all of our model. ", + "bbox": [ + 174, + 895, + 823, + 922 + ], + "page_idx": 29 + }, + { + "type": "table", + "img_path": "images/61ef1cf2a80485a84bf0e9655791dc7d13baf5a81d841666a2976ef65838b5ec.jpg", + "table_caption": [ + "Table 5: Table of hyper-parameters for all of the tasks. " + ], + "table_footnote": [], + "table_body": "
ParameterGNFCTAGCMuJoCoBAT
hidden layers of all neural networks4455
hidden units of all neural networks256256512512
collect samples for each time of model update500050004000096000
batch size of discriminators500050004000080000
horizon50110048 (half-hours)
∈μ(also ∈D)0.0050.010.05 (0.1 for walker2d)0.05
times for discriminator update2215
times for model update11220
times for supervised learning update11420
learning rate for supervised learning1e-51e-53e-41e-5
Y0.990.00.990.99
clip-ratioNANNANNAN0.1
max DkL0.0010.0010.001NAN
optimization algorithm (the first term of Eq.23)TRPOTRPOTRPOPPO
", + "bbox": [ + 176, + 128, + 926, + 349 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "H ADDITIONAL RESULTS ", + "text_level": 1, + "bbox": [ + 176, + 376, + 401, + 392 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "H.1 ALL OF THE RESULT TABLE ", + "text_level": 1, + "bbox": [ + 176, + 409, + 413, + 422 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "We give the result of CNFC in Tab. 8, TCGA in Tab. 9, BAT in Tab. 7, and MuJoCo in Tab. 6. ", + "bbox": [ + 171, + 435, + 784, + 450 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "H.2 AVERAGED RESPONSE CURVES ", + "text_level": 1, + "bbox": [ + 176, + 468, + 436, + 483 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "We give the averaged responses for all of the tasks and the algorithms in Fig. 16 to Fig. 23. We randomly select $20 \\%$ of the states in the dataset and equidistantly sample actions from the action space for each sampled state, and plot the averaged predicted feedback of each action. The real response is slightly different among different figure as the randomly-selected states for testing is different. We sample 9 points in GNFC tasks and 33 points in TAGC tasks for plotting. ", + "bbox": [ + 174, + 496, + 826, + 565 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "H.3 ABLATION STUDIES ", + "text_level": 1, + "bbox": [ + 176, + 584, + 356, + 598 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "In Sec. 4.3 and Appx. E.1, we introduce several techniques to develop a practical GALILEO algorithm. Based on task $\\mathtt { e 0 } . 2 _ { - \\mathtt { p } 0 . 0 5 }$ in GNFC, we give the ablation studies to investigate the effects of these techniques. We first compare two variants that do not handle the assumptions violation problems: (1) NO_INJECT_NOISE: set $\\epsilon _ { \\mu }$ and $\\epsilon _ { D }$ to zero, which makes the overlap assumption not satisfied;; (2) SINGLE_SL: without replacing the second term in Eq. 8 with standard supervised learning even when the output probability of $D$ is far away from 0.5. Besides, we introduced several tricks inspired by GAIL and give a comparison of these tricks and GAIL: (3) ONE_STEP: use one-step reward instead of cumulative rewards (i.e., Q and V; see Eq. 24 and Eq. 25) for re-weighting, which is implemented by set $\\gamma$ to 0; (4) SINGE_DIS: remove $T _ { \\varphi _ { 1 } ^ { * } } ( x , a )$ and replace it with $\\mathbb { E } _ { M _ { \\theta } } \\left[ T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right]$ , which is inspired by GAIL that uses a value function as a baseline instead of using another discriminator; (5) PURE_GAIL: ", + "bbox": [ + 174, + 611, + 550, + 859 + ], + "page_idx": 30 + }, + { + "type": "image", + "img_path": "images/bd32c544f33fa909bf10ef2d6bc1392ce066275899883232e0b00bb79348eae2.jpg", + "image_caption": [ + "Figure 10: Illustration of the ablation studies. The error bars are the standard error. " + ], + "image_footnote": [], + "bbox": [ + 563, + 627, + 818, + 800 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "remove the second term in Eq. 8. It can be regarded as a naive adoption of GAIL and a partial implementation of GALILEO. ", + "bbox": [ + 176, + 861, + 823, + 888 + ], + "page_idx": 30 + }, + { + "type": "text", + "text": "We summarize the results in Fig. 10. Based on the results of NO_INJECT_NOISE and SINGLE_SL, we can see that handling the assumption violation problems is important and will increase the ability on counterfactual queries. The results of PURE_GAIL tell us that the partial implementation of GALILEO is not enough to give stable predictions on counterfactual data; On the other hand, the result of ONE_STEP also demonstrates that embedding the cumulative error of one-step prediction is helpful for GALILEO training; Finally, we also found that SINGLE_DIS nearly has almost no effect on the results. It suggests that, empirically, we can use $\\mathbb { E } _ { M _ { \\theta } } \\left[ T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right]$ as a replacement for $T _ { \\varphi _ { 1 } ^ { * } } ( x , a )$ , which can reduce the computation costs of the extra discriminator training. ", + "bbox": [ + 174, + 895, + 821, + 924 + ], + "page_idx": 30 + }, + { + "type": "table", + "img_path": "images/38ea1f1978b9e9b624ac794e28682b08e25655d901960c5c6e9f60cfe4af3b80.jpg", + "table_caption": [ + "Table 6: The root mean square errors on MuJoCo tasks. We bold the lowest error for each task. “medium” dataset is used for training, while “expert” and “medium-replay” datasets are just used for testing. $\\pm$ follows the standard deviation of three seeds. " + ], + "table_footnote": [], + "table_body": "
TASKHalfCheetah
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.378 ± 0.0032.287± 0.0051.411 ± 0.037
OFF-SL0.404 ± 0.0013.311 ± 0.0552.246 ± 0.016
TASKWalker2d
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.49 ± 0.001.514 ± 0.0020.968 ± 0.004
OFF-SL0.467 ± 0.0041.825 ± 0.0611.239 ± 0.004
TASKHopper
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.037 ± 0.0020.322 ± 0.0360.408 ± 0.003
OFF-SL0.034 ± 0.0010.464 ± 0.0210.574 ± 0.008
", + "bbox": [ + 253, + 118, + 740, + 351 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 364, + 826, + 450 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "H.4 WORST-CASE PREDICTION ERROR ", + "text_level": 1, + "bbox": [ + 176, + 467, + 459, + 482 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "In theory, GALILEO increases the generalization ability by focusing on the worst-case samples’ training to achieve AWRM. To demonstrate the property, we propose a new metric named Mean-Max Square Error (MMSE): $\\begin{array} { r } { \\bar { \\mathfrak { z } } \\left[ \\operatorname* { m a x } _ { a \\in \\mathcal { A } } \\left( M ^ { * } ( x ^ { \\prime } | x , a ) - \\bar { M } ( x ^ { \\prime } | x , a ) \\right) ^ { 2 } \\right] } \\end{array}$ and give the results of MMSE for GNFC in Tab. 10 and for TCGA in Tab. 11. ", + "bbox": [ + 174, + 493, + 826, + 559 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "H.5 DETAILED RESULTS IN THE MUJOCO TASKS ", + "text_level": 1, + "bbox": [ + 174, + 574, + 529, + 590 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "We select 3 environments from D4RL (Fu et al., 2020) to construct our model learning tasks. We compare it with a typical transition model learning algorithm used in the previous offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020), which is a variant of standard supervised learning. We name the method OFF-SL. We train models in datasets HalfCheetah-medium, Walker2dmedium, and Hopper-medium, which are collected by a behavior policy with 1/3 performance to the expert policy, then we test them in the corresponding expert dataset. We plot the converged results and learning curves of GALILEO and OFF-SL in three MuJoCo tasks in Tab. 6 and Fig. 11 respectively. ", + "bbox": [ + 173, + 601, + 825, + 713 + ], + "page_idx": 31 + }, + { + "type": "text", + "text": "In Fig. 11, we can see that both OFF-SL and GALILEO perform well in the training datasets. OFF-SL can even reach a bit lower error in halfcheetah and walker2d. However, when we verify the models through “expert” and “medium-replay” datasets, which are collected by other policies, the performance of GALILEO is significantly more stable and better than OFF-SL. As the training continues, OFF-SL even gets worse and worse. In summary, GALILEO reaches significantly better performances in the expert dataset: the averaged declines of root MSE in three environments are $5 6 . 5 \\%$ , $4 9 . 2 \\%$ , and $3 4 . 8 \\%$ . However, whether in GALILEO or OFF-SL, the performance for testing is at least $2 \\mathbf { x }$ worse than in the training dataset. The phenomenon indicates that although GALILEO can make better performances for counterfactual queries, the risks of using the models are still large and still challenging to be further solved. ", + "bbox": [ + 173, + 719, + 825, + 858 + ], + "page_idx": 31 + }, + { + "type": "image", + "img_path": "images/5547f6091a98561bc79b830999d27857990e83ebd0ac546794410f3a7f6be8af.jpg", + "image_caption": [ + "Figure 11: Illustration of learning curves of the MuJoCo Tasks. The X-axis record the steps of the environment model update, and the Y-axis is the corresponding prediction error. The figures with titles ending in “(train)” means the dataset is used for training while the titles ending in “(test)” means the dataset is just used for testing. The solid curves are the mean reward and the shadow is the standard error of three seeds. " + ], + "image_footnote": [], + "bbox": [ + 184, + 233, + 810, + 710 + ], + "page_idx": 32 + }, + { + "type": "image", + "img_path": "images/7fc39a335c2cd42ff20c40f5c27345da54078256b068d1373bfc35f96bdfa28c.jpg", + "image_caption": [ + "Figure 12: Illustration of relationship between user feedback and the actions of the offline dataset in the real-world food-delivery platform. " + ], + "image_footnote": [], + "bbox": [ + 369, + 103, + 625, + 252 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "H.6 DETAILED RESULTS IN THE BAT TASK ", + "text_level": 1, + "bbox": [ + 176, + 324, + 486, + 339 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "The core challenge of the environment model learning in BAT tasks is similar to the challenge in Fig. 1. Specifically, the behavior policy in BAT tasks is a human-expert policy, which will tend to increase the budget of allowance in the time periods with a lower supply of delivery clerks, otherwise will decrease the budget (Fig. 12 gives an instance of this phenomenon in the real data). ", + "bbox": [ + 173, + 352, + 825, + 409 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Since there is no oracle environment model for querying, we have to describe the results with other metrics. ", + "bbox": [ + 174, + 415, + 823, + 444 + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/b869771cf06ee02caa745de9087c18bd06b71c1215b64025a89e0777903965ba.jpg", + "image_caption": [ + "Figure 13: Illustration of the response curves in the 6 cities. Although the ground-truth curves are unknown, through human expert knowledge, we know that it is expected to be monotonically increasing . " + ], + "image_footnote": [], + "bbox": [ + 189, + 462, + 808, + 742 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "First, we review whether the tendency of the response curve is consistent. In this application, with a larger budget of allowance, the supply will not be decreased. As can be seen in Fig. 13, the tendency of GALILEO’s response is valid in 6 cities but almost all of the models of SL give opposite directions to the response. If we learn a policy through the model of SL, the optimal solution is canceling all of the allowances, which is obviously incorrect in practice. ", + "bbox": [ + 174, + 818, + 825, + 888 + ], + "page_idx": 33 + }, + { + "type": "text", + "text": "Second, we conduct randomized controlled trials (RCT) in one of the testing cities. Using the RCT samples, we can evaluate the correctness of the sort order of the model predictions via Area Under the Uplift Curve (AUUC) (Betlei et al., 2020). To plot AUUC, we first sort the RCT samples based on the predicted treatment effects. Then the cumulative treatment effects are computed by scanning the sorted sample list. If the sort order of the model predictions is better, the sample with larger treatment effects will be computed early. Then the area of AUUC will be larger than the one via a random sorting strategy. The result of AUUC show GALILEO gives a reasonable sorting to the RCT samples (see Fig. 14). ", + "bbox": [ + 173, + 895, + 823, + 924 + ], + "page_idx": 33 + }, + { + "type": "image", + "img_path": "images/eb37f5b144b993f4cd093333bca2ea018eace1f580556caafc4eb6ea290aeb5e.jpg", + "image_caption": [ + "Figure 14: Illustration of the AUUC result for BAT. " + ], + "image_footnote": [], + "bbox": [ + 369, + 103, + 627, + 263 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 320, + 826, + 405 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "Finally, we search for the optimal policy via the cross-entropy method planner (Hafner et al., 2019) based on the learned model. We test the online supply improvement in 6 cities. The algorithm compared is a human-expert policy, which is also the behavior policy of the offline datasets. We conduct online A/B tests for each of the cities. For each test, we randomly split a city into two partitions, one is for deploying the optimal policy learned from the GALILEO model, and the other is as a control group, which keeps the human-expert policy as before. Before the intervention, we collect 10 days’ observation data and compute the averaged five-minute order-taken rates as the baselines of the treatment and control group, named $b ^ { t }$ and $b ^ { c }$ respectively. Then we start intervention and observe the five-minute order-taken rate in the following 14 days for the two groups. The results of the treatment and control groups are $y _ { i } ^ { t }$ and $y _ { i } ^ { c }$ respectively, where $i$ denotes the $i$ -th day of the deployment. The percentage points of the supply improvement are computed via difference-in-difference (DID): ", + "bbox": [ + 173, + 411, + 826, + 565 + ], + "page_idx": 34 + }, + { + "type": "equation", + "img_path": "images/2cb9d2f90f71f871223efc3b44a5dd77d13e117cc1308406538d937414424e6d.jpg", + "text": "$$\n\\frac { \\sum _ { i } ^ { T } ( y _ { i } ^ { t } - b ^ { t } ) - ( y _ { i } ^ { c } - b ^ { c } ) } { T } \\times 1 0 0 ,\n$$", + "text_format": "latex", + "bbox": [ + 387, + 570, + 611, + 606 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "where $T$ is the total days of the intervention and $T = 1 4$ in our experiments. ", + "bbox": [ + 173, + 611, + 671, + 626 + ], + "page_idx": 34 + }, + { + "type": "table", + "img_path": "images/f1acd02301fdd4d8b62e1604fd4e54f1d26dc6146602254dcedde6d5575f914d.jpg", + "table_caption": [ + "Table 7: Results on BAT. We use City- $\\mathbf { X }$ to denote the experiments on different cities. “pp” is an abbreviation of percentage points on the supply improvement. " + ], + "table_footnote": [], + "table_body": "
targetCity-ACity-BCity-C
supply improvement+1.63pp+0.79pp+0.27pp
targetCity-DCity-ECity-F
supply improvement+0.2pp+0.14pp+0.41pp
", + "bbox": [ + 305, + 679, + 686, + 765 + ], + "page_idx": 34 + }, + { + "type": "text", + "text": "The results are summarized in Tab. 7. The online experiment is conducted in 14 days and the results show that the policy learned with GALILEO can make better (the supply improvements are from 0.14 to 1.63 percentage points) budget allocation than the behavior policies in all the testing cities. We give detailed results which record the supply difference between the treatment group and the control group in Fig. 15. ", + "bbox": [ + 173, + 779, + 825, + 851 + ], + "page_idx": 34 + }, + { + "type": "image", + "img_path": "images/1a2796121e8ea865a6cc920895fd3318ac583274799e2549cdb6430e7ce60dd6.jpg", + "image_caption": [ + "Figure 15: Illustration of the daily responses in the A/B test in the 6 cities. " + ], + "image_footnote": [], + "bbox": [ + 191, + 351, + 807, + 647 + ], + "page_idx": 35 + }, + { + "type": "table", + "img_path": "images/94d6e978af6f49e3cb43fe74b880b37bd0df7a36aceed850b0753e61885b8634.jpg", + "table_caption": [ + "Table 8: $\\sqrt { M I S E }$ results on GNFC. We bold the lowest error for each task. $\\pm$ is the standard deviation of three random seeds. " + ], + "table_footnote": [], + "table_body": "
e1_p1e0.2_p1e0.05_p1
GALILEO5.17 ± 0.064.73 ± 0.134.70 ± 0.02
SL5.15 ± 0.234.73 ± 0.3123.64 ± 4.86
IPW5.22 ± 0.095.50 ± 0.015.02 ± 0.07
SCIGAN7.05 ± 0.526.58 ± 0.5818.55 ± 3.50
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO5.03 ± 0.094.72 ± 0.054.87 ± 0.15
SL5.21 ± 0.636.74 ± 0.1533.52 ±1.32
IPW5.27 ± 0.055.69 ± 0.0020.23 ± 0.45
SCIGAN16.07 ± 0.2712.07 ± 1.9319.27 ± 10.72
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO5.23 ± 0.415.01 ± 0.086.17 ± 0.33
SL5.89 ± 0.8814.25 ± 3.4837.50 ± 2.29
IPW5.21 ± 0.015.52 ± 0.4431.95 ± 0.05
SCIGAN11.50 ± 7.7613.05 ± 4.1925.74 ± 8.30
", + "bbox": [ + 289, + 142, + 709, + 396 + ], + "page_idx": 36 + }, + { + "type": "table", + "img_path": "images/44b6289cb60800febc0eeed8fbf9b650658828a2122f74672b8671c30af9c4d0.jpg", + "table_caption": [ + "Table 9: $\\sqrt { M I S E }$ results on TCGA. We bold the lowest error for each task. $\\pm$ is the standard deviation of three random seeds. " + ], + "table_footnote": [], + "table_body": "
t0_bias_2.0 t0_bias_s_20.0 t0_bias_50.0
GALILEO SL0.34 ± 0.05 0.67 ± 0.13 0.38 ± 0.13 1.50 ± 0.312.04 ± 0.12 3.06 ± 0.65
IPW SCIGAN6.57 ± 1.16 0.74 ± 0.056.88 ± 0.30 5.84 ± 0.71 2.74 ± 0.35 3.19 ± 0.09
t1_bias_2.0t1_bias_6.0 t1_bias_8.0
GALILEO SL0.43 ± 0.05 0.47 ± 0.050.25 ± 0.02 0.21 ± 0.04 1.33 ± 0.97 1.18 ± 0.73
IPW3.67 ± 2.370.54 ± 0.13 2.69 ± 1.17
SCIGAN0.45 ± 0.251.08 ± 1.04 1.01 ± 0.77
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO
1.46 ± 0.090.85 ± 0.04 0.46 ± 0.01
SL IPW0.81 ± 0.143.74 ± 2.04 3.59 ± 0.14
", + "bbox": [ + 299, + 450, + 697, + 703 + ], + "page_idx": 36 + }, + { + "type": "table", + "img_path": "images/7c570ae7c458cd9f5d3077904feab0408ba171a8ed0836f36684ec0bea2ee2c6.jpg", + "table_caption": [ + "Table 10: $\\sqrt { M M S E }$ results on GNFC. We bold the lowest error for each task. $\\pm$ is the standard deviation of three random seeds. " + ], + "table_footnote": [], + "table_body": "
e1_p1e0.2_p1e0.05_p1
GALILEO3.86 ± 0.033.99 ± 0.014.07 ± 0.03
SL5.73 ± 0.335.80 ± 0.2818.78 ± 3.13
IPW4.02 ± 0.054.15 ± 0.1222.66 ± 0.33
SCIGAN8.84 ± 0.5412.62 ± 2.1724.21 ± 5.20
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO4.13 ± 0.104.11 ± 0.154.21 ± 0.15
SL5.87 ± 0.437.44 ± 1.1329.13 ± 3.44
IPW4.12 ± 0.026.12 ± 0.4830.96 ± 0.17
SCIGAN12.87 ± 3.0214.59 ± 2.1324.57 ± 3.00
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO4.39 ± 0.204.34 ± 0.205.26 ± 0.29
SL6.12 ± 0.4314.88 ± 4.4130.81 ± 1.69
IPW13.60 ± 7.8326.27 ± 2.6732.55 ± 0.12
SCIGAN9.19 ± 1.0415.08 ± 1.2617.52 ± 0.02
", + "bbox": [ + 294, + 199, + 705, + 452 + ], + "page_idx": 37 + }, + { + "type": "table", + "img_path": "images/7f0576f822a959d36f6b1cef1e840ad20fc9bb83231043ef7d8337d604ea2f7a.jpg", + "table_caption": [ + "Table 11: $\\sqrt { M M S E }$ results on TCGA. We bold the lowest error for each task. $\\pm$ is the standard deviation of three random seeds. " + ], + "table_footnote": [], + "table_body": "
t0_bias_2.0t0_bias_20.0 tO_bias_50.0
GALILEO1.56 ± 0.041.96 ± 0.53 3.16 ± 0.13
SL1.92 ± 0.672.31 ± 0.19 5.11 ± 0.66
IPW7.42 ± 0.465.36 ± 0.96 5.38 ± 1.24
SCIGAN2.11 ± 0.475.23 ±0.27 5.59 ± 1.02
tl_bias_2.0tl_bias_6.0 tl_bias_8.0
GALILEO1.43 ± 0.061.09 ± 0.05 1.36 ± 0.36
SL1.12 ± 0.153.65 ± 1.91 3.96 ± 1.81
IPW1.14 ± 0.11 0.90 ± 0.092.04 ± 0.99
SCIGAN3.32 ± 0.88 4.74 ± 2.125.17 ± 2.42
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO3.77 ± 0.35 3.99 ± 0.402.08 ± 0.60
SL2.70 ± 0.67 8.33 ± 5.059.70 ± 3.12
IPW2.92 ± 0.15 3.90 ± 0.174.47 ± 2.16
SCIGAN3.82 ± 2.12 1.83 ± 1.493.62 ± 4.9
", + "bbox": [ + 299, + 611, + 697, + 866 + ], + "page_idx": 37 + }, + { + "type": "image", + "img_path": "images/0f4a66ef4baf70e73f3c6d49b3d583ae88a89e44b95edef40bd86d6ec7c7718d.jpg", + "image_caption": [ + "Figure 16: Illustration of the averaged response curves of Supervised Learning (SL) in TCGA. " + ], + "image_footnote": [], + "bbox": [ + 191, + 304, + 807, + 696 + ], + "page_idx": 38 + }, + { + "type": "image", + "img_path": "images/2abae9ea1c75f9d0de485507d99f7e3fc9e0798cab13f63e1ba494cd5a148ac7.jpg", + "image_caption": [ + "Figure 17: Illustration of the averaged response curves of Supervised Learning (SL) in GNFC. " + ], + "image_footnote": [], + "bbox": [ + 189, + 303, + 805, + 698 + ], + "page_idx": 39 + }, + { + "type": "image", + "img_path": "images/e95af2c71f12750c1e7632c99166d6e0f6065e19f98d35100fc5980516143630.jpg", + "image_caption": [ + "Figure 18: Illustration of the averaged response curves of Inverse Propensity Weighting (IPW) in TCGA. " + ], + "image_footnote": [], + "bbox": [ + 187, + 281, + 810, + 707 + ], + "page_idx": 40 + }, + { + "type": "image", + "img_path": "images/cef37522bd881e3dbc177930bf4e98a750e94226e9d0ed7fc4238330fda9b47a.jpg", + "image_caption": [ + "Figure 19: Illustration of the averaged response curves of Inverse Propensity Weighting (IPW) in GNFC. " + ], + "image_footnote": [], + "bbox": [ + 187, + 284, + 808, + 705 + ], + "page_idx": 41 + }, + { + "type": "image", + "img_path": "images/fa953ff9d5f43062141219c074a774984b11e3b159955de3d0d2558401086b92.jpg", + "image_caption": [ + "Figure 20: Illustration of the averaged response curves of SCIGAN in TCGA. " + ], + "image_footnote": [], + "bbox": [ + 189, + 304, + 807, + 696 + ], + "page_idx": 42 + }, + { + "type": "image", + "img_path": "images/f7f3fe5f9e09c0042241c841726d9b8c5004cb2f7cb0769552686dc09c60b91f.jpg", + "image_caption": [ + "Figure 21: Illustration of the averaged response curves of SCIGAN in GNFC. " + ], + "image_footnote": [], + "bbox": [ + 189, + 303, + 805, + 699 + ], + "page_idx": 43 + }, + { + "type": "image", + "img_path": "images/f4bdbf1e7453b14d1e9a36552e1a5b2ea53c70db26825936362204bac36ee869.jpg", + "image_caption": [ + "Figure 22: Illustration of the averaged response curves of GALILEO in TCGA. " + ], + "image_footnote": [], + "bbox": [ + 186, + 292, + 808, + 707 + ], + "page_idx": 44 + }, + { + "type": "image", + "img_path": "images/2290b75ec5c556d9f23c1eec2979a5a76f05ef08e2d626f32314d745cbc6a3c0.jpg", + "image_caption": [ + "Figure 23: Illustration of the averaged response curves of GALILEO in GNFC. " + ], + "image_footnote": [], + "bbox": [ + 187, + 291, + 808, + 709 + ], + "page_idx": 45 + } +] \ No newline at end of file diff --git a/parse/dev/zT5T9gHpGI/zT5T9gHpGI_middle.json b/parse/dev/zT5T9gHpGI/zT5T9gHpGI_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..ecd08b0ad2c374992c3c07df961fddcf448ec4f8 --- /dev/null +++ b/parse/dev/zT5T9gHpGI/zT5T9gHpGI_middle.json @@ -0,0 +1,114622 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 79, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 506, + 98 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 506, + 98 + ], + "score": 1.0, + "content": "ADVERSARIAL COUNTERFACTUAL ENVIRONMENT", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 98, + 249, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 249, + 118 + ], + "score": 1.0, + "content": "MODEL LEARNING", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 135, + 244, + 157 + ], + "lines": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 278, + 167, + 333, + 179 + ], + "lines": [ + { + "bbox": [ + 277, + 166, + 335, + 180 + ], + "spans": [ + { + "bbox": [ + 277, + 166, + 335, + 180 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 192, + 469, + 401 + ], + "lines": [ + { + "bbox": [ + 141, + 192, + 469, + 205 + ], + "spans": [ + { + "bbox": [ + 141, + 192, + 469, + 205 + ], + "score": 1.0, + "content": "A good model for action-effect prediction, i.e., the environment model, is essential", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 204, + 469, + 216 + ], + "spans": [ + { + "bbox": [ + 141, + 204, + 469, + 216 + ], + "score": 1.0, + "content": "for sample-efficient policy learning, in which the agent can take numerous free", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 213, + 469, + 228 + ], + "spans": [ + { + "bbox": [ + 141, + 213, + 469, + 228 + ], + "score": 1.0, + "content": "trials to find good policies. Currently, the model is commonly learned by fitting", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 225, + 470, + 239 + ], + "spans": [ + { + "bbox": [ + 141, + 225, + 470, + 239 + ], + "score": 1.0, + "content": "historical transition data through empirical risk minimization (ERM). However, we", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 236, + 469, + 249 + ], + "spans": [ + { + "bbox": [ + 141, + 236, + 469, + 249 + ], + "score": 1.0, + "content": "discover that simple data fitting can lead to a model that will be totally wrong in", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 248, + 470, + 259 + ], + "spans": [ + { + "bbox": [ + 141, + 248, + 470, + 259 + ], + "score": 1.0, + "content": "guiding policy learning due to the selection bias in offline dataset collection. In this", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 258, + 470, + 271 + ], + "spans": [ + { + "bbox": [ + 141, + 258, + 470, + 271 + ], + "score": 1.0, + "content": "work, we introduce weighted empirical risk minimization (WERM) to handle this", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 269, + 469, + 283 + ], + "spans": [ + { + "bbox": [ + 141, + 269, + 469, + 283 + ], + "score": 1.0, + "content": "problem in model learning. A typical WERM method utilizes inverse propensity", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 280, + 470, + 292 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 470, + 292 + ], + "score": 1.0, + "content": "scores to re-weight the training data to approximate the target distribution. However,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 291, + 469, + 304 + ], + "spans": [ + { + "bbox": [ + 141, + 291, + 469, + 304 + ], + "score": 1.0, + "content": "during the policy training, the data distributions of the candidate policies can be", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 302, + 470, + 315 + ], + "spans": [ + { + "bbox": [ + 141, + 302, + 470, + 315 + ], + "score": 1.0, + "content": "various and unknown. Thus, we propose an adversarial weighted empirical risk", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 313, + 470, + 326 + ], + "spans": [ + { + "bbox": [ + 141, + 313, + 470, + 326 + ], + "score": 1.0, + "content": "minimization (AWRM) objective that learns the model with respect to the worst", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 325, + 469, + 336 + ], + "spans": [ + { + "bbox": [ + 141, + 325, + 469, + 336 + ], + "score": 1.0, + "content": "case of the target distributions. We implement AWRM in a sequential decision", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 335, + 470, + 347 + ], + "spans": [ + { + "bbox": [ + 141, + 335, + 470, + 347 + ], + "score": 1.0, + "content": "structure, resulting in the GALILEO model learning algorithm. We also discover", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 346, + 469, + 358 + ], + "spans": [ + { + "bbox": [ + 141, + 346, + 469, + 358 + ], + "score": 1.0, + "content": "that GALILEO is closely related to adversarial model learning, explaining the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 357, + 470, + 369 + ], + "spans": [ + { + "bbox": [ + 141, + 357, + 470, + 369 + ], + "score": 1.0, + "content": "empirical effectiveness of the latter. We apply GALILEO in synthetic tasks and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 368, + 470, + 380 + ], + "spans": [ + { + "bbox": [ + 141, + 368, + 470, + 380 + ], + "score": 1.0, + "content": "verify that GALILEO makes accurate predictions on counterfactual data. We", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 379, + 470, + 391 + ], + "spans": [ + { + "bbox": [ + 142, + 379, + 470, + 391 + ], + "score": 1.0, + "content": "finally applied GALILEO in real-world offline policy learning tasks and found that", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 389, + 445, + 403 + ], + "spans": [ + { + "bbox": [ + 141, + 389, + 445, + 403 + ], + "score": 1.0, + "content": "GALILEO significantly improves policy performance in real-world testing.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 108, + 410, + 206, + 421 + ], + "lines": [ + { + "bbox": [ + 105, + 408, + 208, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 208, + 424 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 435 + ], + "score": 1.0, + "content": "A good environment model is important for sample-efficient decision-making policy learning tech-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 434, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 446 + ], + "score": 1.0, + "content": "niques like reinforcement learning (RL) (James & Johns, 2016). The agent can take trials with this", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 446, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 506, + 458 + ], + "score": 1.0, + "content": "model to find better policies, then the costly real-world trial-and-errors can be saved (James & Johns,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 455, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 506, + 469 + ], + "score": 1.0, + "content": "2016; Yu et al., 2020) or completely waived (Shi et al., 2019). In this process, the core of the models", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "is to answer queries on counterfactual data unbiasedly, that is, given states, correctly answer what", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 478, + 480, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 480, + 491 + ], + "score": 1.0, + "content": "might happen if we were to carry out actions unseen in the training data (Levine et al., 2020).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "score": 1.0, + "content": "Requiring counterfactual queries makes the environment model learning essentially different from", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "standard supervised learning (SL) which directly fits the offline dataset. In real-world applications,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "the offline data is often collected with selection bias, that is, for each state, each action might be", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 528, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 506, + 541 + ], + "score": 1.0, + "content": "chosen unfairly. Seeing the example in Fig. 1(a), to keep the ball following a target line, a behavior", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 539, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 506, + 552 + ], + "score": 1.0, + "content": "policy will use a smaller force when the ball’s location is closer to the target line. When a dataset is", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 551, + 504, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 504, + 562 + ], + "score": 1.0, + "content": "collected with selection bias, the association between the (location) states and (force) actions will", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "make SL hard to identify the correct causal relationship of the states and actions to the next states", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "respectively. Then when we query the model with counterfactual data, the predictions might be", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 583, + 507, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 507, + 595 + ], + "score": 1.0, + "content": "catastrophic failures. In Fig. 1(c), it mistakes that smaller forces will increase the ball’s next location.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 599, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 506, + 613 + ], + "score": 1.0, + "content": "Generally speaking, the problem corresponds to a challenge of training the model in one dataset but", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "testing in another dataset with a shifted distribution (i.e., the dataset generated by counterfactual", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 622, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 634 + ], + "score": 1.0, + "content": "queries), which is beyond the SL’s capability as it violates the independent and identically distributed", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "score": 1.0, + "content": "(i.i.d.) assumption. The problem is widely discussed in causal inference for individual treatment", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "score": 1.0, + "content": "effects (ITEs) estimation in many scenarios like patients’ treatment selection (Imbens, 1999; Alaa", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "& van der Schaar, 2018). ITEs are the effects of treatments on individuals, which are measured", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 665, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 677 + ], + "score": 1.0, + "content": "by treating each individual under a uniform policy and evaluate the effect differences. Practical", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "solutions use weighted empirical risk minimization (WERM) to handle this problem (Jung et al., 2020;", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "Shimodaira, 2000; Hassanpour & Greiner, 2019). In particular, they estimate an inverse propensity", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 698, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 507, + 712 + ], + "score": 1.0, + "content": "score (IPS) to re-weight the training data to approximate the data distribution under a uniform policy.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Then a model is trained under the reweighted data distribution. The distribution-shift problem is", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 721, + 417, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 417, + 732 + ], + "score": 1.0, + "content": "solved as ITEs estimation and model training are under the same distribution.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 45.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 79, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 78, + 506, + 98 + ], + "spans": [ + { + "bbox": [ + 105, + 78, + 506, + 98 + ], + "score": 1.0, + "content": "ADVERSARIAL COUNTERFACTUAL ENVIRONMENT", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 98, + 249, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 249, + 118 + ], + "score": 1.0, + "content": "MODEL LEARNING", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 135, + 244, + 157 + ], + "lines": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 111, + 135, + 245, + 158 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 167, + 333, + 179 + ], + "lines": [ + { + "bbox": [ + 277, + 166, + 335, + 180 + ], + "spans": [ + { + "bbox": [ + 277, + 166, + 335, + 180 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 192, + 469, + 401 + ], + "lines": [ + { + "bbox": [ + 141, + 192, + 469, + 205 + ], + "spans": [ + { + "bbox": [ + 141, + 192, + 469, + 205 + ], + "score": 1.0, + "content": "A good model for action-effect prediction, i.e., the environment model, is essential", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 204, + 469, + 216 + ], + "spans": [ + { + "bbox": [ + 141, + 204, + 469, + 216 + ], + "score": 1.0, + "content": "for sample-efficient policy learning, in which the agent can take numerous free", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 213, + 469, + 228 + ], + "spans": [ + { + "bbox": [ + 141, + 213, + 469, + 228 + ], + "score": 1.0, + "content": "trials to find good policies. Currently, the model is commonly learned by fitting", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 225, + 470, + 239 + ], + "spans": [ + { + "bbox": [ + 141, + 225, + 470, + 239 + ], + "score": 1.0, + "content": "historical transition data through empirical risk minimization (ERM). However, we", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 236, + 469, + 249 + ], + "spans": [ + { + "bbox": [ + 141, + 236, + 469, + 249 + ], + "score": 1.0, + "content": "discover that simple data fitting can lead to a model that will be totally wrong in", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 248, + 470, + 259 + ], + "spans": [ + { + "bbox": [ + 141, + 248, + 470, + 259 + ], + "score": 1.0, + "content": "guiding policy learning due to the selection bias in offline dataset collection. In this", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 258, + 470, + 271 + ], + "spans": [ + { + "bbox": [ + 141, + 258, + 470, + 271 + ], + "score": 1.0, + "content": "work, we introduce weighted empirical risk minimization (WERM) to handle this", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 269, + 469, + 283 + ], + "spans": [ + { + "bbox": [ + 141, + 269, + 469, + 283 + ], + "score": 1.0, + "content": "problem in model learning. A typical WERM method utilizes inverse propensity", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 280, + 470, + 292 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 470, + 292 + ], + "score": 1.0, + "content": "scores to re-weight the training data to approximate the target distribution. However,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 291, + 469, + 304 + ], + "spans": [ + { + "bbox": [ + 141, + 291, + 469, + 304 + ], + "score": 1.0, + "content": "during the policy training, the data distributions of the candidate policies can be", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 302, + 470, + 315 + ], + "spans": [ + { + "bbox": [ + 141, + 302, + 470, + 315 + ], + "score": 1.0, + "content": "various and unknown. Thus, we propose an adversarial weighted empirical risk", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 313, + 470, + 326 + ], + "spans": [ + { + "bbox": [ + 141, + 313, + 470, + 326 + ], + "score": 1.0, + "content": "minimization (AWRM) objective that learns the model with respect to the worst", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 325, + 469, + 336 + ], + "spans": [ + { + "bbox": [ + 141, + 325, + 469, + 336 + ], + "score": 1.0, + "content": "case of the target distributions. We implement AWRM in a sequential decision", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 335, + 470, + 347 + ], + "spans": [ + { + "bbox": [ + 141, + 335, + 470, + 347 + ], + "score": 1.0, + "content": "structure, resulting in the GALILEO model learning algorithm. We also discover", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 346, + 469, + 358 + ], + "spans": [ + { + "bbox": [ + 141, + 346, + 469, + 358 + ], + "score": 1.0, + "content": "that GALILEO is closely related to adversarial model learning, explaining the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 357, + 470, + 369 + ], + "spans": [ + { + "bbox": [ + 141, + 357, + 470, + 369 + ], + "score": 1.0, + "content": "empirical effectiveness of the latter. We apply GALILEO in synthetic tasks and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 368, + 470, + 380 + ], + "spans": [ + { + "bbox": [ + 141, + 368, + 470, + 380 + ], + "score": 1.0, + "content": "verify that GALILEO makes accurate predictions on counterfactual data. We", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 379, + 470, + 391 + ], + "spans": [ + { + "bbox": [ + 142, + 379, + 470, + 391 + ], + "score": 1.0, + "content": "finally applied GALILEO in real-world offline policy learning tasks and found that", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 389, + 445, + 403 + ], + "spans": [ + { + "bbox": [ + 141, + 389, + 445, + 403 + ], + "score": 1.0, + "content": "GALILEO significantly improves policy performance in real-world testing.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 14, + "bbox_fs": [ + 141, + 192, + 470, + 403 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 410, + 206, + 421 + ], + "lines": [ + { + "bbox": [ + 105, + 408, + 208, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 208, + 424 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 435 + ], + "score": 1.0, + "content": "A good environment model is important for sample-efficient decision-making policy learning tech-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 434, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 446 + ], + "score": 1.0, + "content": "niques like reinforcement learning (RL) (James & Johns, 2016). The agent can take trials with this", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 446, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 506, + 458 + ], + "score": 1.0, + "content": "model to find better policies, then the costly real-world trial-and-errors can be saved (James & Johns,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 455, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 506, + 469 + ], + "score": 1.0, + "content": "2016; Yu et al., 2020) or completely waived (Shi et al., 2019). In this process, the core of the models", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "is to answer queries on counterfactual data unbiasedly, that is, given states, correctly answer what", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 478, + 480, + 491 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 480, + 491 + ], + "score": 1.0, + "content": "might happen if we were to carry out actions unseen in the training data (Levine et al., 2020).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 423, + 506, + 491 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 505, + 507 + ], + "score": 1.0, + "content": "Requiring counterfactual queries makes the environment model learning essentially different from", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 506, + 519 + ], + "score": 1.0, + "content": "standard supervised learning (SL) which directly fits the offline dataset. In real-world applications,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "the offline data is often collected with selection bias, that is, for each state, each action might be", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 528, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 506, + 541 + ], + "score": 1.0, + "content": "chosen unfairly. Seeing the example in Fig. 1(a), to keep the ball following a target line, a behavior", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 539, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 506, + 552 + ], + "score": 1.0, + "content": "policy will use a smaller force when the ball’s location is closer to the target line. When a dataset is", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 551, + 504, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 504, + 562 + ], + "score": 1.0, + "content": "collected with selection bias, the association between the (location) states and (force) actions will", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 506, + 574 + ], + "score": 1.0, + "content": "make SL hard to identify the correct causal relationship of the states and actions to the next states", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "respectively. Then when we query the model with counterfactual data, the predictions might be", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 583, + 507, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 507, + 595 + ], + "score": 1.0, + "content": "catastrophic failures. In Fig. 1(c), it mistakes that smaller forces will increase the ball’s next location.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 496, + 507, + 595 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 599, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 506, + 613 + ], + "score": 1.0, + "content": "Generally speaking, the problem corresponds to a challenge of training the model in one dataset but", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "testing in another dataset with a shifted distribution (i.e., the dataset generated by counterfactual", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 622, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 634 + ], + "score": 1.0, + "content": "queries), which is beyond the SL’s capability as it violates the independent and identically distributed", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "score": 1.0, + "content": "(i.i.d.) assumption. The problem is widely discussed in causal inference for individual treatment", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 656 + ], + "score": 1.0, + "content": "effects (ITEs) estimation in many scenarios like patients’ treatment selection (Imbens, 1999; Alaa", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "& van der Schaar, 2018). ITEs are the effects of treatments on individuals, which are measured", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 665, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 677 + ], + "score": 1.0, + "content": "by treating each individual under a uniform policy and evaluate the effect differences. Practical", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "solutions use weighted empirical risk minimization (WERM) to handle this problem (Jung et al., 2020;", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "Shimodaira, 2000; Hassanpour & Greiner, 2019). In particular, they estimate an inverse propensity", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 698, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 507, + 712 + ], + "score": 1.0, + "content": "score (IPS) to re-weight the training data to approximate the data distribution under a uniform policy.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Then a model is trained under the reweighted data distribution. The distribution-shift problem is", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 721, + 417, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 417, + 732 + ], + "score": 1.0, + "content": "solved as ITEs estimation and model training are under the same distribution.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 599, + 507, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 42, + 482, + 160 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 42, + 482, + 160 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 42, + 482, + 160 + ], + "spans": [ + { + "bbox": [ + 119, + 42, + 482, + 160 + ], + "score": 0.971, + "type": "image", + "image_path": "b68193a7fb3b9605331279446f1bc56a0b45f7422ed57eced22fd9b4e8ab8a5c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 42, + 482, + 81.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 81.33333333333334, + 482, + 120.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 120.66666666666669, + 482, + 160.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 164, + 506, + 308 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "Figure 1: An example of selection bias and predictions under counterfactual queries. Subfigure (a)", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 416, + 189 + ], + "score": 1.0, + "content": "shows how the data is collected: a ball locates in a 2D plane whose position is", + "type": "text" + }, + { + "bbox": [ + 417, + 176, + 447, + 187 + ], + "score": 0.9, + "content": "( x _ { t } , y _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 174, + 478, + 189 + ], + "score": 1.0, + "content": "at time", + "type": "text" + }, + { + "bbox": [ + 478, + 176, + 483, + 185 + ], + "score": 0.61, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 174, + 506, + 189 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 186, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 178, + 200 + ], + "score": 1.0, + "content": "ball will move to", + "type": "text" + }, + { + "bbox": [ + 179, + 186, + 229, + 198 + ], + "score": 0.92, + "content": "( x _ { t + 1 } , y _ { t + 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 186, + 284, + 200 + ], + "score": 1.0, + "content": "according to", + "type": "text" + }, + { + "bbox": [ + 284, + 187, + 345, + 198 + ], + "score": 0.9, + "content": "x _ { t + 1 } = x _ { t } + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 186, + 364, + 200 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 364, + 186, + 455, + 198 + ], + "score": 0.91, + "content": "y _ { t + 1 } \\sim \\mathcal { N } ( y _ { t } + a _ { t } , 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 186, + 484, + 200 + ], + "score": 1.0, + "content": ". Here,", + "type": "text" + }, + { + "bbox": [ + 484, + 188, + 494, + 197 + ], + "score": 0.85, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 186, + 506, + 200 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 196, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 215, + 211 + ], + "score": 1.0, + "content": "chosen by a control policy", + "type": "text" + }, + { + "bbox": [ + 215, + 197, + 326, + 209 + ], + "score": 0.91, + "content": "a _ { t } \\sim \\mathcal { N } ( ( \\phi - y _ { t } ) / 1 5 , 0 . 0 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 196, + 400, + 211 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 401, + 198, + 408, + 209 + ], + "score": 0.83, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 196, + 506, + 211 + ], + "score": 1.0, + "content": ", which tries to keep the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 207, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 175, + 222 + ], + "score": 1.0, + "content": "ball near the line", + "type": "text" + }, + { + "bbox": [ + 175, + 209, + 201, + 220 + ], + "score": 0.92, + "content": "y = \\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 207, + 272, + 222 + ], + "score": 1.0, + "content": ". In Subfigure (a),", + "type": "text" + }, + { + "bbox": [ + 273, + 209, + 280, + 220 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 207, + 506, + 222 + ], + "score": 1.0, + "content": "is set to 62.5. Subfigure (b) shows the collected training", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 461, + 232 + ], + "score": 1.0, + "content": "data (grey dashed line) and the two learned models’ prediction of the next position of", + "type": "text" + }, + { + "bbox": [ + 462, + 221, + 468, + 231 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 219, + 505, + 232 + ], + "score": 1.0, + "content": ". All the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 229, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 349, + 244 + ], + "score": 1.0, + "content": "models discovered the relation that the corresponding next", + "type": "text" + }, + { + "bbox": [ + 349, + 232, + 357, + 242 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 229, + 506, + 244 + ], + "score": 1.0, + "content": "will be smaller with a larger action.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 298, + 254 + ], + "score": 1.0, + "content": "However, the truth is not because the larger", + "type": "text" + }, + { + "bbox": [ + 298, + 243, + 308, + 252 + ], + "score": 0.84, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 240, + 381, + 254 + ], + "score": 1.0, + "content": "causes a smaller", + "type": "text" + }, + { + "bbox": [ + 382, + 242, + 401, + 253 + ], + "score": 0.87, + "content": "y _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 240, + 506, + 254 + ], + "score": 1.0, + "content": ", but the policy selects a", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 252, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 131, + 265 + ], + "score": 1.0, + "content": "small", + "type": "text" + }, + { + "bbox": [ + 131, + 253, + 141, + 263 + ], + "score": 0.84, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 252, + 167, + 265 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 167, + 253, + 177, + 263 + ], + "score": 0.83, + "content": "y _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 252, + 477, + 265 + ], + "score": 1.0, + "content": "is close to the target line. When we estimate the response curves by fixing", + "type": "text" + }, + { + "bbox": [ + 477, + 254, + 487, + 264 + ], + "score": 0.82, + "content": "y _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 252, + 506, + 265 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 263, + 506, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 181, + 276 + ], + "score": 1.0, + "content": "reassigning action", + "type": "text" + }, + { + "bbox": [ + 181, + 264, + 191, + 274 + ], + "score": 0.84, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 263, + 266, + 276 + ], + "score": 1.0, + "content": "with other actions", + "type": "text" + }, + { + "bbox": [ + 266, + 263, + 302, + 274 + ], + "score": 0.91, + "content": "a _ { t } + \\Delta a", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 263, + 333, + 276 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 333, + 263, + 387, + 275 + ], + "score": 0.93, + "content": "\\Delta a \\in [ - 1 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 263, + 506, + 276 + ], + "score": 1.0, + "content": "is a variation of action value,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 273, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 506, + 287 + ], + "score": 1.0, + "content": "the model of SL will exploit the association and give opposite responses, while in AWRM and its", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 285, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 506, + 297 + ], + "score": 1.0, + "content": "practical implementation GALILEO, the predictions are closer to the ground truths. The result is in", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 296, + 404, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 296, + 404, + 309 + ], + "score": 1.0, + "content": "Subfigure (c), where the darker a region is, the more samples are fallen in.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 9 + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 106, + 314, + 505, + 358 + ], + "lines": [ + { + "bbox": [ + 104, + 313, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 313, + 506, + 326 + ], + "score": 1.0, + "content": "The selection bias can be regarded as an instance of the problem called “distributional shift” in", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 325, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 336 + ], + "score": 1.0, + "content": "offline model-based RL, which has also received great attention (Levine et al., 2020; Yu et al., 2020;", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "Kidambi et al., 2020; Chen et al., 2021). However, previous methods, where naive supervised", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 346, + 501, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 501, + 360 + ], + "score": 1.0, + "content": "learning is used for environment model learning, ignore the problem in environment model learning", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 359, + 346, + 414 + ], + "lines": [ + { + "bbox": [ + 106, + 357, + 347, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 347, + 370 + ], + "score": 1.0, + "content": "but handling the problem by suppressing the policy explo-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 369, + 347, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 347, + 380 + ], + "score": 1.0, + "content": "ration and learning in risky regions. Although these methods", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 380, + 347, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 347, + 392 + ], + "score": 1.0, + "content": "have made great progress in many tasks, so far, how to learn", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 390, + 346, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 346, + 401 + ], + "score": 1.0, + "content": "a better environment model that can alleviate the problem for", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 401, + 347, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 347, + 414 + ], + "score": 1.0, + "content": "faithful offline policy optimization has rarely been discussed.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 419, + 346, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 347, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 347, + 431 + ], + "score": 1.0, + "content": "In this work, for faithful offline policy optimization, we in-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 429, + 346, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 346, + 442 + ], + "score": 1.0, + "content": "troduce WERM to environment model learning. The extra", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 441, + 347, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 347, + 453 + ], + "score": 1.0, + "content": "challenge of model learning for policy optimization is that we", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 451, + 347, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 347, + 464 + ], + "score": 1.0, + "content": "have to query numerous different policies’ feedback besides", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 462, + 347, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 347, + 475 + ], + "score": 1.0, + "content": "the uniform policy for finding a good policy. Thus the target", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 473, + 347, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 347, + 486 + ], + "score": 1.0, + "content": "data distribution to reweight can be various and unknown. To", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 485, + 347, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 347, + 497 + ], + "score": 1.0, + "content": "solve the problem, we propose an objective called adversar-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 496, + 346, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 346, + 507 + ], + "score": 1.0, + "content": "ial weighted empirical risk minimization (AWRM). AWRM", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 506, + 347, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 347, + 519 + ], + "score": 1.0, + "content": "introduces adversarial policies, of which the corresponding", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 518, + 347, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 347, + 529 + ], + "score": 1.0, + "content": "counterfactual dataset has the maximal prediction error of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 528, + 347, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 347, + 540 + ], + "score": 1.0, + "content": "the model. For each iteration, the model is learned to be", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 540, + 347, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 347, + 550 + ], + "score": 1.0, + "content": "as small prediction risks as possible under the adversarial", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 551, + 347, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 347, + 561 + ], + "score": 1.0, + "content": "counterfactual dataset. However, the adversarial counterfac-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 562, + 347, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 347, + 573 + ], + "score": 1.0, + "content": "tual dataset cannot be obtained in the offline setting, thus", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 573, + 348, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 348, + 583 + ], + "score": 1.0, + "content": "we derive an approximation of the counterfactual data dis-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 583, + 346, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 346, + 595 + ], + "score": 1.0, + "content": "tribution queried by the optimal adversarial policy and use", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 595, + 347, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 347, + 606 + ], + "score": 1.0, + "content": "a variational representation to give a tractable solution to", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 605, + 347, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 347, + 616 + ], + "score": 1.0, + "content": "learn a model from the approximated data distribution. As", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 617, + 347, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 347, + 628 + ], + "score": 1.0, + "content": "a result, we derive a practical approach named Generative", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 627, + 347, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 347, + 638 + ], + "score": 1.0, + "content": "Adversarial offLIne counterfactuaL Environment mOdel", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 638, + 347, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 347, + 650 + ], + "score": 1.0, + "content": "learning (GALILEO) for AWRM. Fig. 2 shows the differ-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 649, + 346, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 346, + 660 + ], + "score": 1.0, + "content": "ence in the prediction errors learned by these algorithms. We", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 658, + 347, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 347, + 673 + ], + "score": 1.0, + "content": "also discover that GALILEO is closely related to existing", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 669, + 347, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 347, + 684 + ], + "score": 1.0, + "content": "generative-adversarial model learning techniques, explaining", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 681, + 225, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 225, + 693 + ], + "score": 1.0, + "content": "the effectiveness of the latter.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 37 + }, + { + "type": "image", + "bbox": [ + 355, + 365, + 493, + 466 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 355, + 365, + 493, + 466 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 355, + 365, + 493, + 466 + ], + "spans": [ + { + "bbox": [ + 355, + 365, + 493, + 466 + ], + "score": 0.888, + "type": "image", + "image_path": "e21350e3d9b064285fe113e1adaa5132ea73dc8398bcf24bc5da581bdb24fdad.jpg" + } + ] + } + ], + "index": 50.5, + "virtual_lines": [ + { + "bbox": [ + 355, + 365, + 493, + 415.5 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 355, + 415.5, + 493, + 466.0 + ], + "spans": [], + "index": 51 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 354, + 480, + 505, + 698 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 354, + 479, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 354, + 479, + 506, + 491 + ], + "score": 1.0, + "content": "Figure 2: An illustration of the predic-", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 353, + 491, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 353, + 491, + 506, + 502 + ], + "score": 1.0, + "content": "tion error in counterfactual datasets.", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 353, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 353, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "The prediction risks is measured with", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 353, + 513, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 353, + 513, + 505, + 523 + ], + "score": 1.0, + "content": "mean square error (MSE). The error of", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 353, + 523, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 353, + 523, + 486, + 535 + ], + "score": 1.0, + "content": "SL is small only in training data (", + "type": "text" + }, + { + "bbox": [ + 486, + 524, + 506, + 535 + ], + "score": 0.86, + "content": "\\phi =", + "type": "inline_equation" + } + ], + "index": 56 + }, + { + "bbox": [ + 353, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 353, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "62.5) but becomes much larger in the", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 353, + 544, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 353, + 544, + 506, + 558 + ], + "score": 1.0, + "content": "dataset “far away from” the training", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 354, + 556, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 354, + 556, + 505, + 567 + ], + "score": 1.0, + "content": "data. AWRM-oracle selects the oracle", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 353, + 568, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 353, + 568, + 506, + 578 + ], + "score": 1.0, + "content": "worst counterfactual dataset for train-", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 354, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 354, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "ing for each iteration (pseudocode is", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 353, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 353, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "in Alg. 1) which reaches small MSE in", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 353, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 353, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "all datasets and gives correct response", + "type": "text" + } + ], + "index": 63 + }, + { + "bbox": [ + 353, + 611, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 353, + 611, + 506, + 623 + ], + "score": 1.0, + "content": "curves (Fig. 1(c)). GALILEO approxi-", + "type": "text" + } + ], + "index": 64 + }, + { + "bbox": [ + 354, + 623, + 506, + 633 + ], + "spans": [ + { + "bbox": [ + 354, + 623, + 506, + 633 + ], + "score": 1.0, + "content": "mates the optimal adversarial counter-", + "type": "text" + } + ], + "index": 65 + }, + { + "bbox": [ + 354, + 633, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 354, + 633, + 505, + 644 + ], + "score": 1.0, + "content": "factual data distribution based on the", + "type": "text" + } + ], + "index": 66 + }, + { + "bbox": [ + 354, + 644, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 354, + 644, + 505, + 655 + ], + "score": 1.0, + "content": "training data and model. Although the", + "type": "text" + } + ], + "index": 67 + }, + { + "bbox": [ + 353, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 353, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "MSE of GALILEO is a bit larger than", + "type": "text" + } + ], + "index": 68 + }, + { + "bbox": [ + 353, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 353, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "SL in the training data, in the coun-", + "type": "text" + } + ], + "index": 69 + }, + { + "bbox": [ + 354, + 677, + 505, + 688 + ], + "spans": [ + { + "bbox": [ + 354, + 677, + 505, + 688 + ], + "score": 1.0, + "content": "terfactual datasets, the MSE is on the", + "type": "text" + } + ], + "index": 70 + }, + { + "bbox": [ + 353, + 688, + 473, + 699 + ], + "spans": [ + { + "bbox": [ + 353, + 688, + 473, + 699 + ], + "score": 1.0, + "content": "same scale as AWRM-oracle.", + "type": "text" + } + ], + "index": 71 + } + ], + "index": 61.5 + } + ], + "index": 56.0 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Experiments are conducted in two synthetic and two realistic environments. The results in the", + "type": "text" + } + ], + "index": 72 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "synthetic environments show that GALILEO can reconstruct correct responses for counterfactual", + "type": "text" + } + ], + "index": 73 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "queries. The evaluation results in two realistic environments also demonstrate that GALILEO has", + "type": "text" + } + ], + "index": 74 + } + ], + "index": 73 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 513, + 348, + 585, + 358 + ], + "lines": [ + { + "bbox": [ + 512, + 346, + 586, + 360 + ], + "spans": [ + { + "bbox": [ + 512, + 346, + 586, + 360 + ], + "score": 1.0, + "content": "to Reviewer rQ79", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 42, + 482, + 160 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 42, + 482, + 160 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 42, + 482, + 160 + ], + "spans": [ + { + "bbox": [ + 119, + 42, + 482, + 160 + ], + "score": 0.971, + "type": "image", + "image_path": "b68193a7fb3b9605331279446f1bc56a0b45f7422ed57eced22fd9b4e8ab8a5c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 42, + 482, + 81.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 81.33333333333334, + 482, + 120.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 120.66666666666669, + 482, + 160.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 164, + 506, + 308 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "Figure 1: An example of selection bias and predictions under counterfactual queries. Subfigure (a)", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 416, + 189 + ], + "score": 1.0, + "content": "shows how the data is collected: a ball locates in a 2D plane whose position is", + "type": "text" + }, + { + "bbox": [ + 417, + 176, + 447, + 187 + ], + "score": 0.9, + "content": "( x _ { t } , y _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 174, + 478, + 189 + ], + "score": 1.0, + "content": "at time", + "type": "text" + }, + { + "bbox": [ + 478, + 176, + 483, + 185 + ], + "score": 0.61, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 174, + 506, + 189 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 186, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 178, + 200 + ], + "score": 1.0, + "content": "ball will move to", + "type": "text" + }, + { + "bbox": [ + 179, + 186, + 229, + 198 + ], + "score": 0.92, + "content": "( x _ { t + 1 } , y _ { t + 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 186, + 284, + 200 + ], + "score": 1.0, + "content": "according to", + "type": "text" + }, + { + "bbox": [ + 284, + 187, + 345, + 198 + ], + "score": 0.9, + "content": "x _ { t + 1 } = x _ { t } + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 186, + 364, + 200 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 364, + 186, + 455, + 198 + ], + "score": 0.91, + "content": "y _ { t + 1 } \\sim \\mathcal { N } ( y _ { t } + a _ { t } , 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 186, + 484, + 200 + ], + "score": 1.0, + "content": ". Here,", + "type": "text" + }, + { + "bbox": [ + 484, + 188, + 494, + 197 + ], + "score": 0.85, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 186, + 506, + 200 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 196, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 215, + 211 + ], + "score": 1.0, + "content": "chosen by a control policy", + "type": "text" + }, + { + "bbox": [ + 215, + 197, + 326, + 209 + ], + "score": 0.91, + "content": "a _ { t } \\sim \\mathcal { N } ( ( \\phi - y _ { t } ) / 1 5 , 0 . 0 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 196, + 400, + 211 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 401, + 198, + 408, + 209 + ], + "score": 0.83, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 196, + 506, + 211 + ], + "score": 1.0, + "content": ", which tries to keep the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 207, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 175, + 222 + ], + "score": 1.0, + "content": "ball near the line", + "type": "text" + }, + { + "bbox": [ + 175, + 209, + 201, + 220 + ], + "score": 0.92, + "content": "y = \\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 207, + 272, + 222 + ], + "score": 1.0, + "content": ". In Subfigure (a),", + "type": "text" + }, + { + "bbox": [ + 273, + 209, + 280, + 220 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 207, + 506, + 222 + ], + "score": 1.0, + "content": "is set to 62.5. Subfigure (b) shows the collected training", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 461, + 232 + ], + "score": 1.0, + "content": "data (grey dashed line) and the two learned models’ prediction of the next position of", + "type": "text" + }, + { + "bbox": [ + 462, + 221, + 468, + 231 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 219, + 505, + 232 + ], + "score": 1.0, + "content": ". All the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 229, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 349, + 244 + ], + "score": 1.0, + "content": "models discovered the relation that the corresponding next", + "type": "text" + }, + { + "bbox": [ + 349, + 232, + 357, + 242 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 229, + 506, + 244 + ], + "score": 1.0, + "content": "will be smaller with a larger action.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 298, + 254 + ], + "score": 1.0, + "content": "However, the truth is not because the larger", + "type": "text" + }, + { + "bbox": [ + 298, + 243, + 308, + 252 + ], + "score": 0.84, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 240, + 381, + 254 + ], + "score": 1.0, + "content": "causes a smaller", + "type": "text" + }, + { + "bbox": [ + 382, + 242, + 401, + 253 + ], + "score": 0.87, + "content": "y _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 240, + 506, + 254 + ], + "score": 1.0, + "content": ", but the policy selects a", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 252, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 131, + 265 + ], + "score": 1.0, + "content": "small", + "type": "text" + }, + { + "bbox": [ + 131, + 253, + 141, + 263 + ], + "score": 0.84, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 252, + 167, + 265 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 167, + 253, + 177, + 263 + ], + "score": 0.83, + "content": "y _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 252, + 477, + 265 + ], + "score": 1.0, + "content": "is close to the target line. When we estimate the response curves by fixing", + "type": "text" + }, + { + "bbox": [ + 477, + 254, + 487, + 264 + ], + "score": 0.82, + "content": "y _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 252, + 506, + 265 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 263, + 506, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 181, + 276 + ], + "score": 1.0, + "content": "reassigning action", + "type": "text" + }, + { + "bbox": [ + 181, + 264, + 191, + 274 + ], + "score": 0.84, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 263, + 266, + 276 + ], + "score": 1.0, + "content": "with other actions", + "type": "text" + }, + { + "bbox": [ + 266, + 263, + 302, + 274 + ], + "score": 0.91, + "content": "a _ { t } + \\Delta a", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 263, + 333, + 276 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 333, + 263, + 387, + 275 + ], + "score": 0.93, + "content": "\\Delta a \\in [ - 1 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 263, + 506, + 276 + ], + "score": 1.0, + "content": "is a variation of action value,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 273, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 506, + 287 + ], + "score": 1.0, + "content": "the model of SL will exploit the association and give opposite responses, while in AWRM and its", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 285, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 506, + 297 + ], + "score": 1.0, + "content": "practical implementation GALILEO, the predictions are closer to the ground truths. The result is in", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 296, + 404, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 296, + 404, + 309 + ], + "score": 1.0, + "content": "Subfigure (c), where the darker a region is, the more samples are fallen in.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 9 + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 106, + 314, + 505, + 358 + ], + "lines": [ + { + "bbox": [ + 104, + 313, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 313, + 506, + 326 + ], + "score": 1.0, + "content": "The selection bias can be regarded as an instance of the problem called “distributional shift” in", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 325, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 336 + ], + "score": 1.0, + "content": "offline model-based RL, which has also received great attention (Levine et al., 2020; Yu et al., 2020;", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 335, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 335, + 505, + 348 + ], + "score": 1.0, + "content": "Kidambi et al., 2020; Chen et al., 2021). However, previous methods, where naive supervised", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 346, + 501, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 346, + 501, + 360 + ], + "score": 1.0, + "content": "learning is used for environment model learning, ignore the problem in environment model learning", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 357, + 347, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 347, + 370 + ], + "score": 1.0, + "content": "but handling the problem by suppressing the policy explo-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 369, + 347, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 347, + 380 + ], + "score": 1.0, + "content": "ration and learning in risky regions. Although these methods", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 380, + 347, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 347, + 392 + ], + "score": 1.0, + "content": "have made great progress in many tasks, so far, how to learn", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 390, + 346, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 346, + 401 + ], + "score": 1.0, + "content": "a better environment model that can alleviate the problem for", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 401, + 347, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 347, + 414 + ], + "score": 1.0, + "content": "faithful offline policy optimization has rarely been discussed.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 17.5, + "bbox_fs": [ + 104, + 313, + 506, + 360 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 359, + 346, + 414 + ], + "lines": [], + "index": 22, + "bbox_fs": [ + 106, + 357, + 347, + 414 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 419, + 346, + 693 + ], + "lines": [ + { + "bbox": [ + 105, + 418, + 347, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 418, + 347, + 431 + ], + "score": 1.0, + "content": "In this work, for faithful offline policy optimization, we in-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 429, + 346, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 346, + 442 + ], + "score": 1.0, + "content": "troduce WERM to environment model learning. The extra", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 441, + 347, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 347, + 453 + ], + "score": 1.0, + "content": "challenge of model learning for policy optimization is that we", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 451, + 347, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 347, + 464 + ], + "score": 1.0, + "content": "have to query numerous different policies’ feedback besides", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 462, + 347, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 347, + 475 + ], + "score": 1.0, + "content": "the uniform policy for finding a good policy. Thus the target", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 473, + 347, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 347, + 486 + ], + "score": 1.0, + "content": "data distribution to reweight can be various and unknown. To", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 485, + 347, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 347, + 497 + ], + "score": 1.0, + "content": "solve the problem, we propose an objective called adversar-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 496, + 346, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 346, + 507 + ], + "score": 1.0, + "content": "ial weighted empirical risk minimization (AWRM). AWRM", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 506, + 347, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 347, + 519 + ], + "score": 1.0, + "content": "introduces adversarial policies, of which the corresponding", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 518, + 347, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 347, + 529 + ], + "score": 1.0, + "content": "counterfactual dataset has the maximal prediction error of", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 528, + 347, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 347, + 540 + ], + "score": 1.0, + "content": "the model. For each iteration, the model is learned to be", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 540, + 347, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 347, + 550 + ], + "score": 1.0, + "content": "as small prediction risks as possible under the adversarial", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 551, + 347, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 347, + 561 + ], + "score": 1.0, + "content": "counterfactual dataset. However, the adversarial counterfac-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 562, + 347, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 347, + 573 + ], + "score": 1.0, + "content": "tual dataset cannot be obtained in the offline setting, thus", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 573, + 348, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 348, + 583 + ], + "score": 1.0, + "content": "we derive an approximation of the counterfactual data dis-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 583, + 346, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 346, + 595 + ], + "score": 1.0, + "content": "tribution queried by the optimal adversarial policy and use", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 595, + 347, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 347, + 606 + ], + "score": 1.0, + "content": "a variational representation to give a tractable solution to", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 605, + 347, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 347, + 616 + ], + "score": 1.0, + "content": "learn a model from the approximated data distribution. As", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 617, + 347, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 347, + 628 + ], + "score": 1.0, + "content": "a result, we derive a practical approach named Generative", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 627, + 347, + 638 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 347, + 638 + ], + "score": 1.0, + "content": "Adversarial offLIne counterfactuaL Environment mOdel", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 638, + 347, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 347, + 650 + ], + "score": 1.0, + "content": "learning (GALILEO) for AWRM. Fig. 2 shows the differ-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 649, + 346, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 346, + 660 + ], + "score": 1.0, + "content": "ence in the prediction errors learned by these algorithms. We", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 658, + 347, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 347, + 673 + ], + "score": 1.0, + "content": "also discover that GALILEO is closely related to existing", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 669, + 347, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 347, + 684 + ], + "score": 1.0, + "content": "generative-adversarial model learning techniques, explaining", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 681, + 225, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 225, + 693 + ], + "score": 1.0, + "content": "the effectiveness of the latter.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 418, + 348, + 693 + ] + }, + { + "type": "image", + "bbox": [ + 355, + 365, + 493, + 466 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 355, + 365, + 493, + 466 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 355, + 365, + 493, + 466 + ], + "spans": [ + { + "bbox": [ + 355, + 365, + 493, + 466 + ], + "score": 0.888, + "type": "image", + "image_path": "e21350e3d9b064285fe113e1adaa5132ea73dc8398bcf24bc5da581bdb24fdad.jpg" + } + ] + } + ], + "index": 50.5, + "virtual_lines": [ + { + "bbox": [ + 355, + 365, + 493, + 415.5 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 355, + 415.5, + 493, + 466.0 + ], + "spans": [], + "index": 51 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 354, + 480, + 505, + 698 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 354, + 479, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 354, + 479, + 506, + 491 + ], + "score": 1.0, + "content": "Figure 2: An illustration of the predic-", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 353, + 491, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 353, + 491, + 506, + 502 + ], + "score": 1.0, + "content": "tion error in counterfactual datasets.", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 353, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 353, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "The prediction risks is measured with", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 353, + 513, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 353, + 513, + 505, + 523 + ], + "score": 1.0, + "content": "mean square error (MSE). The error of", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 353, + 523, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 353, + 523, + 486, + 535 + ], + "score": 1.0, + "content": "SL is small only in training data (", + "type": "text" + }, + { + "bbox": [ + 486, + 524, + 506, + 535 + ], + "score": 0.86, + "content": "\\phi =", + "type": "inline_equation" + } + ], + "index": 56 + }, + { + "bbox": [ + 353, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 353, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "62.5) but becomes much larger in the", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 353, + 544, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 353, + 544, + 506, + 558 + ], + "score": 1.0, + "content": "dataset “far away from” the training", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 354, + 556, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 354, + 556, + 505, + 567 + ], + "score": 1.0, + "content": "data. AWRM-oracle selects the oracle", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 353, + 568, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 353, + 568, + 506, + 578 + ], + "score": 1.0, + "content": "worst counterfactual dataset for train-", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 354, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 354, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "ing for each iteration (pseudocode is", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 353, + 590, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 353, + 590, + 505, + 601 + ], + "score": 1.0, + "content": "in Alg. 1) which reaches small MSE in", + "type": "text" + } + ], + "index": 62 + }, + { + "bbox": [ + 353, + 600, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 353, + 600, + 506, + 613 + ], + "score": 1.0, + "content": "all datasets and gives correct response", + "type": "text" + } + ], + "index": 63 + }, + { + "bbox": [ + 353, + 611, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 353, + 611, + 506, + 623 + ], + "score": 1.0, + "content": "curves (Fig. 1(c)). GALILEO approxi-", + "type": "text" + } + ], + "index": 64 + }, + { + "bbox": [ + 354, + 623, + 506, + 633 + ], + "spans": [ + { + "bbox": [ + 354, + 623, + 506, + 633 + ], + "score": 1.0, + "content": "mates the optimal adversarial counter-", + "type": "text" + } + ], + "index": 65 + }, + { + "bbox": [ + 354, + 633, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 354, + 633, + 505, + 644 + ], + "score": 1.0, + "content": "factual data distribution based on the", + "type": "text" + } + ], + "index": 66 + }, + { + "bbox": [ + 354, + 644, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 354, + 644, + 505, + 655 + ], + "score": 1.0, + "content": "training data and model. Although the", + "type": "text" + } + ], + "index": 67 + }, + { + "bbox": [ + 353, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 353, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "MSE of GALILEO is a bit larger than", + "type": "text" + } + ], + "index": 68 + }, + { + "bbox": [ + 353, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 353, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "SL in the training data, in the coun-", + "type": "text" + } + ], + "index": 69 + }, + { + "bbox": [ + 354, + 677, + 505, + 688 + ], + "spans": [ + { + "bbox": [ + 354, + 677, + 505, + 688 + ], + "score": 1.0, + "content": "terfactual datasets, the MSE is on the", + "type": "text" + } + ], + "index": 70 + }, + { + "bbox": [ + 353, + 688, + 473, + 699 + ], + "spans": [ + { + "bbox": [ + 353, + 688, + 473, + 699 + ], + "score": 1.0, + "content": "same scale as AWRM-oracle.", + "type": "text" + } + ], + "index": 71 + } + ], + "index": 61.5 + } + ], + "index": 56.0 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "Experiments are conducted in two synthetic and two realistic environments. The results in the", + "type": "text" + } + ], + "index": 72 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "synthetic environments show that GALILEO can reconstruct correct responses for counterfactual", + "type": "text" + } + ], + "index": 73 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "queries. The evaluation results in two realistic environments also demonstrate that GALILEO has", + "type": "text" + } + ], + "index": 74 + }, + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "better ability in counterfactual query compared with baselines. We finally search for a policy based", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "on the learned model in a real-world online platform. The policy significantly improves performance", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 240, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 240, + 116 + ], + "score": 1.0, + "content": "in concerned business indicators.", + "type": "text", + "cross_page": true + } + ], + "index": 2 + } + ], + "index": 73, + "bbox_fs": [ + 106, + 699, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 115 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "better ability in counterfactual query compared with baselines. We finally search for a policy based", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "on the learned model in a real-world online platform. The policy significantly improves performance", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 240, + 116 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 240, + 116 + ], + "score": 1.0, + "content": "in concerned business indicators.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 108, + 123, + 211, + 136 + ], + "lines": [ + { + "bbox": [ + 104, + 122, + 213, + 138 + ], + "spans": [ + { + "bbox": [ + 104, + 122, + 213, + 138 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 146, + 505, + 343 + ], + "lines": [ + { + "bbox": [ + 106, + 145, + 506, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 506, + 158 + ], + "score": 1.0, + "content": "We give related adversarial algorithms for model learning in the following and leave other related", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 156, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 506, + 168 + ], + "score": 1.0, + "content": "work in Appx. F. GANTIE (Yoon et al., 2018) uses a generator to fill counterfactual outcomes for", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 167, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 505, + 180 + ], + "score": 1.0, + "content": "each data pair and a discriminator to judge the source (treatment group or control group) of the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 179, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 179, + 505, + 190 + ], + "score": 1.0, + "content": "filled data pair. The generator is trained to minimize the output of the discriminator. GANITE is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 189, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 506, + 202 + ], + "score": 1.0, + "content": "trained until the discriminator cannot determine which of the components is the factual outcome. Bica", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "score": 1.0, + "content": "et al. (2020) propose SCIGAN to extend GANITE to continuous treatment effect estimation (a.k.a.,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 211, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 506, + 224 + ], + "score": 1.0, + "content": "dosage-response estimation) via a hierarchical discriminator architecture. In real-world applications,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "score": 1.0, + "content": "environment model learning based on Generative Adversarial Imitation Learning (GAIL) has also", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "score": 1.0, + "content": "been adopted for sequential decision-making problems (Ho & Ermon, 2016). GAIL is first proposed", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "score": 1.0, + "content": "for policy imitation (Ho & Ermon, 2016), which uses the imitated policy to generate trajectories", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 255, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 268 + ], + "score": 1.0, + "content": "by interacting with the environment. The policy is learned with the trajectories through RL which", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 265, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 104, + 265, + 506, + 280 + ], + "score": 1.0, + "content": "maximizes the cumulative rewards given by the discriminator. Shi et al. (2019); Chen et al. (2019);", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 277, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 506, + 290 + ], + "score": 1.0, + "content": "Shang et al. (2019) use GAIL for environment model learning by regarding the environment model", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 288, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 300 + ], + "score": 1.0, + "content": "as the generator and the behavior policy as the “environment” in standard GAIL. These studies", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "empirically demonstrate that adversarial model learning algorithms have better generalization ability", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 308, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 104, + 308, + 506, + 324 + ], + "score": 1.0, + "content": "for counterfactual queries, while our study reveals the connection between adversarial model learning", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 321, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 506, + 333 + ], + "score": 1.0, + "content": "and the WERM through IPS. Our derived practical algorithm GALILEO is closely related to the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 332, + 460, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 460, + 344 + ], + "score": 1.0, + "content": "existing adversarial model learning algorithms, explaining the effectiveness of the latter.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 12.5 + }, + { + "type": "title", + "bbox": [ + 108, + 356, + 208, + 369 + ], + "lines": [ + { + "bbox": [ + 104, + 355, + 209, + 371 + ], + "spans": [ + { + "bbox": [ + 104, + 355, + 209, + 371 + ], + "score": 1.0, + "content": "3 PRELIMINARIES", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 109, + 376, + 488, + 398 + ], + "lines": [ + { + "bbox": [ + 107, + 375, + 488, + 388 + ], + "spans": [ + { + "bbox": [ + 107, + 375, + 488, + 388 + ], + "score": 1.0, + "content": "3.1 SINGLE-STEP INDIVIDUALIZED TREATMENT EFFECTS ESTIMATION AND WEIGHTED", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 130, + 387, + 277, + 399 + ], + "spans": [ + { + "bbox": [ + 130, + 387, + 277, + 399 + ], + "score": 1.0, + "content": "EMPIRICAL RISKS MINIMIZATION", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 403, + 505, + 650 + ], + "lines": [ + { + "bbox": [ + 105, + 403, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 506, + 416 + ], + "score": 1.0, + "content": "We first introduce individualized treatment effects (ITEs) estimation (Rosenbaum & Rubin, 1983),", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 414, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 506, + 427 + ], + "score": 1.0, + "content": "which can be regarded as the scenario in which the environment model has only a single step. ITEs", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 425, + 503, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 203, + 440 + ], + "score": 1.0, + "content": "are typically defined as", + "type": "text" + }, + { + "bbox": [ + 203, + 426, + 503, + 438 + ], + "score": 0.9, + "content": "I T E ( x ) : = \\mathbb { E } [ M ^ { * } ( y | x , 1 ) | A = 1 , X = x ] - \\mathbb { E } [ M ^ { * } ( y | x , 0 ) | A = 0 , X = x ]", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 436, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 134, + 451 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 439, + 141, + 449 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 436, + 293, + 451 + ], + "score": 1.0, + "content": "is the feedback of the environment", + "type": "text" + }, + { + "bbox": [ + 293, + 438, + 340, + 449 + ], + "score": 0.89, + "content": "M ^ { * } ( y | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 436, + 345, + 451 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 345, + 438, + 356, + 447 + ], + "score": 0.68, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 436, + 506, + 451 + ], + "score": 1.0, + "content": "denotes the state vector containing", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 448, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 316, + 461 + ], + "score": 1.0, + "content": "pre-treatment covariates (such as age and weight),", + "type": "text" + }, + { + "bbox": [ + 316, + 449, + 325, + 459 + ], + "score": 0.66, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 448, + 506, + 461 + ], + "score": 1.0, + "content": "denotes the treatment variable which is the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 233, + 472 + ], + "score": 1.0, + "content": "action intervening to the state", + "type": "text" + }, + { + "bbox": [ + 234, + 460, + 243, + 469 + ], + "score": 0.81, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 459, + 266, + 472 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 267, + 460, + 275, + 469 + ], + "score": 0.76, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 459, + 506, + 472 + ], + "score": 1.0, + "content": "should be sampled from a uniform policy. In the two-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 189, + 483 + ], + "score": 1.0, + "content": "treatment scenario,", + "type": "text" + }, + { + "bbox": [ + 190, + 470, + 198, + 480 + ], + "score": 0.7, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 469, + 223, + 483 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 223, + 470, + 249, + 482 + ], + "score": 0.93, + "content": "\\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "where 1 is the action to intervene and 0 is the action to do", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 480, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 506, + 493 + ], + "score": 1.0, + "content": "nothing. A correct ITEs estimation should be done in Randomized Controlled Trials (RCT) in", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 323, + 504 + ], + "score": 1.0, + "content": "which we have the same probability of samples of", + "type": "text" + }, + { + "bbox": [ + 324, + 492, + 356, + 502 + ], + "score": 0.91, + "content": "A = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 491, + 376, + 504 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 376, + 492, + 408, + 502 + ], + "score": 0.9, + "content": "A = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 491, + 448, + 504 + ], + "score": 1.0, + "content": "for each", + "type": "text" + }, + { + "bbox": [ + 448, + 492, + 459, + 502 + ], + "score": 0.77, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 491, + 506, + 504 + ], + "score": 1.0, + "content": ". Here we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 503, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 166, + 515 + ], + "score": 1.0, + "content": "use lowercase", + "type": "text" + }, + { + "bbox": [ + 167, + 505, + 183, + 514 + ], + "score": 0.87, + "content": "x , a", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 503, + 203, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 203, + 505, + 210, + 514 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 503, + 374, + 515 + ], + "score": 1.0, + "content": "to denote samples of random variables", + "type": "text" + }, + { + "bbox": [ + 374, + 503, + 396, + 514 + ], + "score": 0.89, + "content": "X , A", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 503, + 415, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 415, + 503, + 425, + 513 + ], + "score": 0.74, + "content": "Y", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 503, + 463, + 515 + ], + "score": 1.0, + "content": ", and use", + "type": "text" + }, + { + "bbox": [ + 464, + 503, + 486, + 514 + ], + "score": 0.88, + "content": "x , A", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 503, + 505, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 514, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 116, + 525 + ], + "score": 0.8, + "content": "\\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 514, + 505, + 525 + ], + "score": 1.0, + "content": "to denote space of the samples. In practice, we prefer to estimate ITEs under observational", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "score": 1.0, + "content": "studies. In observational studies, datasets are pre-collected from the real world by a behavior", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 536, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 505, + 549 + ], + "score": 1.0, + "content": "policy such as a human-expert policy. In this case, a common approach for estimating the ITEs", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 546, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 104, + 546, + 135, + 562 + ], + "score": 1.0, + "content": "can be", + "type": "text" + }, + { + "bbox": [ + 135, + 547, + 420, + 560 + ], + "score": 0.88, + "content": "\\hat { I T E } ( x _ { i } ) = a _ { i } ( y _ { i } ^ { F } - M ( x _ { i } , 1 - a _ { i } ) ) + ( 1 - a _ { i } ) ( M ( x _ { i } , 1 - a _ { i } ) - y _ { i } ^ { F } )", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 546, + 506, + 562 + ], + "score": 1.0, + "content": "(Shalit et al., 2017)", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 246, + 573 + ], + "score": 1.0, + "content": "in deterministic prediction, where", + "type": "text" + }, + { + "bbox": [ + 246, + 562, + 257, + 571 + ], + "score": 0.85, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 559, + 275, + 573 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 276, + 560, + 289, + 572 + ], + "score": 0.9, + "content": "y _ { i } ^ { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 559, + 487, + 573 + ], + "score": 1.0, + "content": "denote the covariate and factual feedback of the", + "type": "text" + }, + { + "bbox": [ + 488, + 561, + 492, + 570 + ], + "score": 0.69, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 559, + 506, + 573 + ], + "score": 1.0, + "content": "-th", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 570, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 158, + 584 + ], + "score": 1.0, + "content": "sample, and", + "type": "text" + }, + { + "bbox": [ + 158, + 571, + 195, + 582 + ], + "score": 0.91, + "content": "M \\in \\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 570, + 372, + 584 + ], + "score": 1.0, + "content": "denotes an approximated feedback model.", + "type": "text" + }, + { + "bbox": [ + 372, + 571, + 385, + 581 + ], + "score": 0.8, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 570, + 505, + 584 + ], + "score": 1.0, + "content": "is the space of the model. In", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 579, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 104, + 579, + 410, + 596 + ], + "score": 1.0, + "content": "this formulation, the training set is an empirical factual data distribution", + "type": "text" + }, + { + "bbox": [ + 410, + 582, + 486, + 594 + ], + "score": 0.92, + "content": "\\bar { P _ { F } } = \\{ ( x _ { i } , a _ { i } ) \\} _ { i } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 579, + 506, + 596 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 591, + 506, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 361, + 607 + ], + "score": 1.0, + "content": "the testing set is an empirical counterfactual data distribution", + "type": "text" + }, + { + "bbox": [ + 361, + 593, + 460, + 605 + ], + "score": 0.92, + "content": "P _ { C F } = \\{ ( x _ { i } , 1 - a _ { i } ) \\} _ { i } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 591, + 475, + 607 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 476, + 595, + 482, + 603 + ], + "score": 0.69, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 591, + 506, + 607 + ], + "score": 1.0, + "content": "does", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 603, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 419, + 616 + ], + "score": 1.0, + "content": "not sample from a discrete uniform policy, i.e., the policy has selection bias,", + "type": "text" + }, + { + "bbox": [ + 420, + 605, + 434, + 615 + ], + "score": 0.86, + "content": "P _ { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 603, + 452, + 616 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 453, + 605, + 473, + 615 + ], + "score": 0.9, + "content": "P _ { C F }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 603, + 506, + 616 + ], + "score": 1.0, + "content": "will be", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "two different distributions, which violate the i.i.d. assumption of standard supervised learning. In", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 626, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 191, + 640 + ], + "score": 1.0, + "content": "stochastic prediction,", + "type": "text" + }, + { + "bbox": [ + 192, + 626, + 354, + 639 + ], + "score": 0.92, + "content": "\\hat { I T E } ( x ) = \\mathbb { E } [ M ( y | x , 1 ) ] - \\mathbb { E } [ M ( y | x , 0 ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 626, + 506, + 640 + ], + "score": 1.0, + "content": "and the counterfactual distribution for", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 637, + 368, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 368, + 651 + ], + "score": 1.0, + "content": "testing is the dataset with action sampling from a uniform policy.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "Generally speaking, in ITEs estimation, the risks of queries under counterfactual data are caused", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 664, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 506, + 679 + ], + "score": 1.0, + "content": "by the gap between the policy in training and testing data distributions. Without further processing,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "minimizing the empirical risks cannot guarantee the counterfactual-query risks being minimized.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 278, + 699 + ], + "score": 1.0, + "content": "Assuming that the policy in training data", + "type": "text" + }, + { + "bbox": [ + 278, + 690, + 285, + 699 + ], + "score": 0.76, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 688, + 322, + 699 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 322, + 687, + 447, + 700 + ], + "score": 0.92, + "content": "\\mu ( a | x ) > 0 , \\forall a \\stackrel { - } { \\in } \\mathcal { \\bar { A } } , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 688, + 506, + 699 + ], + "score": 1.0, + "content": "(often named", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "overlap assumption), a classical solution to handle the above problem is weighted empirical risk", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 414, + 722 + ], + "score": 1.0, + "content": "minimization (WERM) through an inverse propensity scoring (IPS) term", + "type": "text" + }, + { + "bbox": [ + 414, + 712, + 422, + 720 + ], + "score": 0.69, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "(Shimodaira, 2000;", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 719, + 381, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 381, + 732 + ], + "score": 1.0, + "content": "Assaad et al., 2021; Hassanpour & Greiner, 2019; Jung et al., 2020):", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 50 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 504, + 115 + ], + "lines": [], + "index": 1, + "bbox_fs": [ + 105, + 82, + 506, + 116 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 123, + 211, + 136 + ], + "lines": [ + { + "bbox": [ + 104, + 122, + 213, + 138 + ], + "spans": [ + { + "bbox": [ + 104, + 122, + 213, + 138 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 146, + 505, + 343 + ], + "lines": [ + { + "bbox": [ + 106, + 145, + 506, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 145, + 506, + 158 + ], + "score": 1.0, + "content": "We give related adversarial algorithms for model learning in the following and leave other related", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 156, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 506, + 168 + ], + "score": 1.0, + "content": "work in Appx. F. GANTIE (Yoon et al., 2018) uses a generator to fill counterfactual outcomes for", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 167, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 505, + 180 + ], + "score": 1.0, + "content": "each data pair and a discriminator to judge the source (treatment group or control group) of the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 179, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 179, + 505, + 190 + ], + "score": 1.0, + "content": "filled data pair. The generator is trained to minimize the output of the discriminator. GANITE is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 189, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 506, + 202 + ], + "score": 1.0, + "content": "trained until the discriminator cannot determine which of the components is the factual outcome. Bica", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 506, + 213 + ], + "score": 1.0, + "content": "et al. (2020) propose SCIGAN to extend GANITE to continuous treatment effect estimation (a.k.a.,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 211, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 506, + 224 + ], + "score": 1.0, + "content": "dosage-response estimation) via a hierarchical discriminator architecture. In real-world applications,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 105, + 222, + 505, + 235 + ], + "score": 1.0, + "content": "environment model learning based on Generative Adversarial Imitation Learning (GAIL) has also", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 106, + 233, + 505, + 246 + ], + "score": 1.0, + "content": "been adopted for sequential decision-making problems (Ho & Ermon, 2016). GAIL is first proposed", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 505, + 256 + ], + "score": 1.0, + "content": "for policy imitation (Ho & Ermon, 2016), which uses the imitated policy to generate trajectories", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 255, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 268 + ], + "score": 1.0, + "content": "by interacting with the environment. The policy is learned with the trajectories through RL which", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 265, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 104, + 265, + 506, + 280 + ], + "score": 1.0, + "content": "maximizes the cumulative rewards given by the discriminator. Shi et al. (2019); Chen et al. (2019);", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 277, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 506, + 290 + ], + "score": 1.0, + "content": "Shang et al. (2019) use GAIL for environment model learning by regarding the environment model", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 288, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 300 + ], + "score": 1.0, + "content": "as the generator and the behavior policy as the “environment” in standard GAIL. These studies", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "empirically demonstrate that adversarial model learning algorithms have better generalization ability", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 308, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 104, + 308, + 506, + 324 + ], + "score": 1.0, + "content": "for counterfactual queries, while our study reveals the connection between adversarial model learning", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 321, + 506, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 321, + 506, + 333 + ], + "score": 1.0, + "content": "and the WERM through IPS. Our derived practical algorithm GALILEO is closely related to the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 332, + 460, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 332, + 460, + 344 + ], + "score": 1.0, + "content": "existing adversarial model learning algorithms, explaining the effectiveness of the latter.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 12.5, + "bbox_fs": [ + 104, + 145, + 506, + 344 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 356, + 208, + 369 + ], + "lines": [ + { + "bbox": [ + 104, + 355, + 209, + 371 + ], + "spans": [ + { + "bbox": [ + 104, + 355, + 209, + 371 + ], + "score": 1.0, + "content": "3 PRELIMINARIES", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 109, + 376, + 488, + 398 + ], + "lines": [ + { + "bbox": [ + 107, + 375, + 488, + 388 + ], + "spans": [ + { + "bbox": [ + 107, + 375, + 488, + 388 + ], + "score": 1.0, + "content": "3.1 SINGLE-STEP INDIVIDUALIZED TREATMENT EFFECTS ESTIMATION AND WEIGHTED", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 130, + 387, + 277, + 399 + ], + "spans": [ + { + "bbox": [ + 130, + 387, + 277, + 399 + ], + "score": 1.0, + "content": "EMPIRICAL RISKS MINIMIZATION", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 403, + 505, + 650 + ], + "lines": [ + { + "bbox": [ + 105, + 403, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 506, + 416 + ], + "score": 1.0, + "content": "We first introduce individualized treatment effects (ITEs) estimation (Rosenbaum & Rubin, 1983),", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 414, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 506, + 427 + ], + "score": 1.0, + "content": "which can be regarded as the scenario in which the environment model has only a single step. ITEs", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 425, + 503, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 203, + 440 + ], + "score": 1.0, + "content": "are typically defined as", + "type": "text" + }, + { + "bbox": [ + 203, + 426, + 503, + 438 + ], + "score": 0.9, + "content": "I T E ( x ) : = \\mathbb { E } [ M ^ { * } ( y | x , 1 ) | A = 1 , X = x ] - \\mathbb { E } [ M ^ { * } ( y | x , 0 ) | A = 0 , X = x ]", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 436, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 134, + 451 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 439, + 141, + 449 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 436, + 293, + 451 + ], + "score": 1.0, + "content": "is the feedback of the environment", + "type": "text" + }, + { + "bbox": [ + 293, + 438, + 340, + 449 + ], + "score": 0.89, + "content": "M ^ { * } ( y | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 436, + 345, + 451 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 345, + 438, + 356, + 447 + ], + "score": 0.68, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 436, + 506, + 451 + ], + "score": 1.0, + "content": "denotes the state vector containing", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 448, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 316, + 461 + ], + "score": 1.0, + "content": "pre-treatment covariates (such as age and weight),", + "type": "text" + }, + { + "bbox": [ + 316, + 449, + 325, + 459 + ], + "score": 0.66, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 448, + 506, + 461 + ], + "score": 1.0, + "content": "denotes the treatment variable which is the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 233, + 472 + ], + "score": 1.0, + "content": "action intervening to the state", + "type": "text" + }, + { + "bbox": [ + 234, + 460, + 243, + 469 + ], + "score": 0.81, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 459, + 266, + 472 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 267, + 460, + 275, + 469 + ], + "score": 0.76, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 459, + 506, + 472 + ], + "score": 1.0, + "content": "should be sampled from a uniform policy. In the two-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 189, + 483 + ], + "score": 1.0, + "content": "treatment scenario,", + "type": "text" + }, + { + "bbox": [ + 190, + 470, + 198, + 480 + ], + "score": 0.7, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 469, + 223, + 483 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 223, + 470, + 249, + 482 + ], + "score": 0.93, + "content": "\\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "where 1 is the action to intervene and 0 is the action to do", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 480, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 506, + 493 + ], + "score": 1.0, + "content": "nothing. A correct ITEs estimation should be done in Randomized Controlled Trials (RCT) in", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 323, + 504 + ], + "score": 1.0, + "content": "which we have the same probability of samples of", + "type": "text" + }, + { + "bbox": [ + 324, + 492, + 356, + 502 + ], + "score": 0.91, + "content": "A = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 491, + 376, + 504 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 376, + 492, + 408, + 502 + ], + "score": 0.9, + "content": "A = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 491, + 448, + 504 + ], + "score": 1.0, + "content": "for each", + "type": "text" + }, + { + "bbox": [ + 448, + 492, + 459, + 502 + ], + "score": 0.77, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 491, + 506, + 504 + ], + "score": 1.0, + "content": ". Here we", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 503, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 166, + 515 + ], + "score": 1.0, + "content": "use lowercase", + "type": "text" + }, + { + "bbox": [ + 167, + 505, + 183, + 514 + ], + "score": 0.87, + "content": "x , a", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 503, + 203, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 203, + 505, + 210, + 514 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 503, + 374, + 515 + ], + "score": 1.0, + "content": "to denote samples of random variables", + "type": "text" + }, + { + "bbox": [ + 374, + 503, + 396, + 514 + ], + "score": 0.89, + "content": "X , A", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 503, + 415, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 415, + 503, + 425, + 513 + ], + "score": 0.74, + "content": "Y", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 503, + 463, + 515 + ], + "score": 1.0, + "content": ", and use", + "type": "text" + }, + { + "bbox": [ + 464, + 503, + 486, + 514 + ], + "score": 0.88, + "content": "x , A", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 503, + 505, + 515 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 514, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 116, + 525 + ], + "score": 0.8, + "content": "\\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 514, + 505, + 525 + ], + "score": 1.0, + "content": "to denote space of the samples. In practice, we prefer to estimate ITEs under observational", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "score": 1.0, + "content": "studies. In observational studies, datasets are pre-collected from the real world by a behavior", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 536, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 505, + 549 + ], + "score": 1.0, + "content": "policy such as a human-expert policy. In this case, a common approach for estimating the ITEs", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 546, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 104, + 546, + 135, + 562 + ], + "score": 1.0, + "content": "can be", + "type": "text" + }, + { + "bbox": [ + 135, + 547, + 420, + 560 + ], + "score": 0.88, + "content": "\\hat { I T E } ( x _ { i } ) = a _ { i } ( y _ { i } ^ { F } - M ( x _ { i } , 1 - a _ { i } ) ) + ( 1 - a _ { i } ) ( M ( x _ { i } , 1 - a _ { i } ) - y _ { i } ^ { F } )", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 546, + 506, + 562 + ], + "score": 1.0, + "content": "(Shalit et al., 2017)", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 559, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 559, + 246, + 573 + ], + "score": 1.0, + "content": "in deterministic prediction, where", + "type": "text" + }, + { + "bbox": [ + 246, + 562, + 257, + 571 + ], + "score": 0.85, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 559, + 275, + 573 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 276, + 560, + 289, + 572 + ], + "score": 0.9, + "content": "y _ { i } ^ { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 559, + 487, + 573 + ], + "score": 1.0, + "content": "denote the covariate and factual feedback of the", + "type": "text" + }, + { + "bbox": [ + 488, + 561, + 492, + 570 + ], + "score": 0.69, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 559, + 506, + 573 + ], + "score": 1.0, + "content": "-th", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 570, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 158, + 584 + ], + "score": 1.0, + "content": "sample, and", + "type": "text" + }, + { + "bbox": [ + 158, + 571, + 195, + 582 + ], + "score": 0.91, + "content": "M \\in \\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 570, + 372, + 584 + ], + "score": 1.0, + "content": "denotes an approximated feedback model.", + "type": "text" + }, + { + "bbox": [ + 372, + 571, + 385, + 581 + ], + "score": 0.8, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 570, + 505, + 584 + ], + "score": 1.0, + "content": "is the space of the model. In", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 579, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 104, + 579, + 410, + 596 + ], + "score": 1.0, + "content": "this formulation, the training set is an empirical factual data distribution", + "type": "text" + }, + { + "bbox": [ + 410, + 582, + 486, + 594 + ], + "score": 0.92, + "content": "\\bar { P _ { F } } = \\{ ( x _ { i } , a _ { i } ) \\} _ { i } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 579, + 506, + 596 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 591, + 506, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 361, + 607 + ], + "score": 1.0, + "content": "the testing set is an empirical counterfactual data distribution", + "type": "text" + }, + { + "bbox": [ + 361, + 593, + 460, + 605 + ], + "score": 0.92, + "content": "P _ { C F } = \\{ ( x _ { i } , 1 - a _ { i } ) \\} _ { i } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 591, + 475, + 607 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 476, + 595, + 482, + 603 + ], + "score": 0.69, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 591, + 506, + 607 + ], + "score": 1.0, + "content": "does", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 603, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 419, + 616 + ], + "score": 1.0, + "content": "not sample from a discrete uniform policy, i.e., the policy has selection bias,", + "type": "text" + }, + { + "bbox": [ + 420, + 605, + 434, + 615 + ], + "score": 0.86, + "content": "P _ { F }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 603, + 452, + 616 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 453, + 605, + 473, + 615 + ], + "score": 0.9, + "content": "P _ { C F }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 603, + 506, + 616 + ], + "score": 1.0, + "content": "will be", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "two different distributions, which violate the i.i.d. assumption of standard supervised learning. In", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 626, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 191, + 640 + ], + "score": 1.0, + "content": "stochastic prediction,", + "type": "text" + }, + { + "bbox": [ + 192, + 626, + 354, + 639 + ], + "score": 0.92, + "content": "\\hat { I T E } ( x ) = \\mathbb { E } [ M ( y | x , 1 ) ] - \\mathbb { E } [ M ( y | x , 0 ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 626, + 506, + 640 + ], + "score": 1.0, + "content": "and the counterfactual distribution for", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 637, + 368, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 368, + 651 + ], + "score": 1.0, + "content": "testing is the dataset with action sampling from a uniform policy.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 35.5, + "bbox_fs": [ + 104, + 403, + 506, + 651 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "Generally speaking, in ITEs estimation, the risks of queries under counterfactual data are caused", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 664, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 506, + 679 + ], + "score": 1.0, + "content": "by the gap between the policy in training and testing data distributions. Without further processing,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "minimizing the empirical risks cannot guarantee the counterfactual-query risks being minimized.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 278, + 699 + ], + "score": 1.0, + "content": "Assuming that the policy in training data", + "type": "text" + }, + { + "bbox": [ + 278, + 690, + 285, + 699 + ], + "score": 0.76, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 688, + 322, + 699 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 322, + 687, + 447, + 700 + ], + "score": 0.92, + "content": "\\mu ( a | x ) > 0 , \\forall a \\stackrel { - } { \\in } \\mathcal { \\bar { A } } , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 688, + 506, + 699 + ], + "score": 1.0, + "content": "(often named", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "overlap assumption), a classical solution to handle the above problem is weighted empirical risk", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 414, + 722 + ], + "score": 1.0, + "content": "minimization (WERM) through an inverse propensity scoring (IPS) term", + "type": "text" + }, + { + "bbox": [ + 414, + 712, + 422, + 720 + ], + "score": 0.69, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "(Shimodaira, 2000;", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 719, + 381, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 719, + 381, + 732 + ], + "score": 1.0, + "content": "Assaad et al., 2021; Hassanpour & Greiner, 2019; Jung et al., 2020):", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 50, + "bbox_fs": [ + 105, + 654, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 422, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 423, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 423, + 97 + ], + "score": 1.0, + "content": "Definition 3.1. The learning objective of WERM through IPS is formulated as", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 97, + 424, + 146 + ], + "lines": [ + { + "bbox": [ + 185, + 97, + 424, + 146 + ], + "spans": [ + { + "bbox": [ + 185, + 97, + 424, + 146 + ], + "score": 0.92, + "content": "\\begin{array} { r l } { \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } L ( M ) = \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } \\mathbb { E } _ { { x } , { a } , { y } \\sim { p } _ { M ^ { * } } } [ \\omega ( { x } , { a } ) \\ell ( M ( { y } | { x } , { a } ) , { y } ) ] , } & { } \\\\ { { s . t . } \\quad \\omega ( { x } , { a } ) = \\frac { \\beta ( a | { x } ) } { \\mu ( a | { x } ) } , } & { } \\end{array}", + "type": "interline_equation", + "image_path": "84b11ff30828b8a64799e15e9dedc010134f03df389f6ff864ac494a2de8e6c2.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 185, + 97, + 424, + 113.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 185, + 113.33333333333333, + 424, + 129.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 185, + 129.66666666666666, + 424, + 146.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 149, + 504, + 183 + ], + "lines": [ + { + "bbox": [ + 105, + 148, + 506, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 133, + 163 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 150, + 141, + 160 + ], + "score": 0.79, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 148, + 159, + 163 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 159, + 151, + 167, + 160 + ], + "score": 0.7, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 148, + 393, + 163 + ], + "score": 1.0, + "content": "denote the policies in testing and training domains, and", + "type": "text" + }, + { + "bbox": [ + 393, + 149, + 412, + 162 + ], + "score": 0.92, + "content": "p _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 148, + 506, + 163 + ], + "score": 1.0, + "content": "is the joint probability", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 159, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 271, + 172 + ], + "score": 0.9, + "content": "p _ { M ^ { * } } ^ { \\mu } ( x , a , y ) : = \\rho _ { 0 } ( x ) \\mu ( a | x ) M ^ { * } ( y | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 159, + 308, + 174 + ], + "score": 1.0, + "content": "in which", + "type": "text" + }, + { + "bbox": [ + 309, + 160, + 333, + 172 + ], + "score": 0.92, + "content": "\\rho _ { 0 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 159, + 441, + 174 + ], + "score": 1.0, + "content": "is the distribution of state.", + "type": "text" + }, + { + "bbox": [ + 441, + 161, + 454, + 170 + ], + "score": 0.74, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 159, + 506, + 174 + ], + "score": 1.0, + "content": "is the model", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 171, + 291, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 134, + 185 + ], + "score": 1.0, + "content": "space.", + "type": "text" + }, + { + "bbox": [ + 135, + 173, + 140, + 181 + ], + "score": 0.53, + "content": "\\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 171, + 291, + 185 + ], + "score": 1.0, + "content": "is a loss function for model learning.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 191, + 505, + 293 + ], + "lines": [ + { + "bbox": [ + 105, + 191, + 505, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 124, + 204 + ], + "score": 1.0, + "content": "The", + "type": "text" + }, + { + "bbox": [ + 124, + 193, + 132, + 202 + ], + "score": 0.79, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 191, + 505, + 204 + ], + "score": 1.0, + "content": "is also known as importance sampling (IS) weight, which corrects the sampling bias. In this", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 203, + 504, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 149, + 214 + ], + "score": 1.0, + "content": "objective,", + "type": "text" + }, + { + "bbox": [ + 149, + 204, + 157, + 212 + ], + "score": 0.74, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 203, + 496, + 214 + ], + "score": 1.0, + "content": "is to align the training data distribution to the testing data. By selecting different", + "type": "text" + }, + { + "bbox": [ + 496, + 203, + 504, + 212 + ], + "score": 0.79, + "content": "\\hat { \\omega }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 214, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 170, + 226 + ], + "score": 1.0, + "content": "to approximate", + "type": "text" + }, + { + "bbox": [ + 171, + 216, + 179, + 223 + ], + "score": 0.75, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 214, + 257, + 226 + ], + "score": 1.0, + "content": "to learn the model", + "type": "text" + }, + { + "bbox": [ + 257, + 214, + 269, + 224 + ], + "score": 0.77, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 214, + 505, + 226 + ], + "score": 1.0, + "content": ", current environment model learning algorithms for ITEs", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 225, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 505, + 236 + ], + "score": 1.0, + "content": "estimation are fallen into the framework. In standard supervised learning and some works for ITEs", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 235, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 325, + 248 + ], + "score": 1.0, + "content": "estimation (Wager & Athey, 2018; Weiss et al., 2015),", + "type": "text" + }, + { + "bbox": [ + 325, + 235, + 374, + 248 + ], + "score": 0.92, + "content": "\\hat { \\omega } ( x , a ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 236, + 505, + 248 + ], + "score": 1.0, + "content": "as the distribution-shift problem", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 245, + 504, + 261 + ], + "spans": [ + { + "bbox": [ + 104, + 245, + 446, + 260 + ], + "score": 1.0, + "content": "is ignored. In Shimodaira (2000); Assaad et al. (2021); Hassanpour & Greiner (2019),", + "type": "text" + }, + { + "bbox": [ + 447, + 245, + 475, + 261 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\omega = \\frac { \\mathrm { i } } { \\hat { \\mu } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 245, + 496, + 260 + ], + "score": 1.0, + "content": "(i.e.,", + "type": "text" + }, + { + "bbox": [ + 496, + 246, + 504, + 258 + ], + "score": 0.82, + "content": "\\beta", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 258, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 370, + 272 + ], + "score": 1.0, + "content": "a uniform policy) for balancing treatment and control group, where", + "type": "text" + }, + { + "bbox": [ + 370, + 259, + 378, + 270 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 258, + 506, + 272 + ], + "score": 1.0, + "content": "is an approximation of behavior", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 134, + 282 + ], + "score": 1.0, + "content": "policy", + "type": "text" + }, + { + "bbox": [ + 135, + 272, + 142, + 281 + ], + "score": 0.73, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 271, + 505, + 282 + ], + "score": 1.0, + "content": ". Note that it is a reasonable weight in ITEs estimation: ITEs are defined to evaluate the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 280, + 440, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 440, + 293 + ], + "score": 1.0, + "content": "effect of each state between treatment and control behavior under a uniform policy.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 107, + 297, + 313, + 308 + ], + "lines": [ + { + "bbox": [ + 106, + 297, + 313, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 313, + 309 + ], + "score": 1.0, + "content": "3.2 SEQUENTIAL DECISION-MAKING SETTING", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 312, + 505, + 389 + ], + "lines": [ + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "score": 1.0, + "content": "Decision-making processes in a sequential environment are often formulated into Markov Decision", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "Process (MDP) (Sutton & Barto, 1998). MDP depicts an agent interacting with the environment", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 430, + 346 + ], + "score": 1.0, + "content": "through actions. In the first step, states are sampled from an initial state distribution", + "type": "text" + }, + { + "bbox": [ + 430, + 334, + 478, + 345 + ], + "score": 0.91, + "content": "x _ { 0 } \\sim \\rho _ { 0 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 334, + 505, + 346 + ], + "score": 1.0, + "content": ". Then", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 344, + 504, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 178, + 357 + ], + "score": 1.0, + "content": "at each time-step", + "type": "text" + }, + { + "bbox": [ + 178, + 345, + 241, + 357 + ], + "score": 0.92, + "content": "t \\in \\{ 0 , 1 , 2 , \\ldots \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 344, + 347, + 357 + ], + "score": 1.0, + "content": ", the agent takes an action", + "type": "text" + }, + { + "bbox": [ + 348, + 345, + 378, + 356 + ], + "score": 0.92, + "content": "a _ { t } \\in { \\mathcal { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 344, + 448, + 357 + ], + "score": 1.0, + "content": "through a policy", + "type": "text" + }, + { + "bbox": [ + 449, + 345, + 504, + 357 + ], + "score": 0.91, + "content": "\\pi ( a _ { t } | x _ { t } ) \\in \\Pi", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 181, + 369 + ], + "score": 1.0, + "content": "based on the state", + "type": "text" + }, + { + "bbox": [ + 182, + 357, + 213, + 367 + ], + "score": 0.91, + "content": "x _ { t } \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 356, + 350, + 369 + ], + "score": 1.0, + "content": ", then the agent receives a reward", + "type": "text" + }, + { + "bbox": [ + 351, + 358, + 360, + 367 + ], + "score": 0.84, + "content": "r _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 356, + 458, + 369 + ], + "score": 1.0, + "content": "from a reward function", + "type": "text" + }, + { + "bbox": [ + 458, + 357, + 505, + 368 + ], + "score": 0.89, + "content": "r ( x _ { t } , a _ { t } ) \\in", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 366, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 115, + 377 + ], + "score": 0.79, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 366, + 237, + 380 + ], + "score": 1.0, + "content": "and transits to the next state", + "type": "text" + }, + { + "bbox": [ + 237, + 369, + 258, + 379 + ], + "score": 0.89, + "content": "x _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 366, + 386, + 380 + ], + "score": 1.0, + "content": "given by a transition function", + "type": "text" + }, + { + "bbox": [ + 387, + 367, + 455, + 379 + ], + "score": 0.93, + "content": "M ^ { \\ast } ( x _ { t + 1 } | x _ { t } , a _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 366, + 505, + 380 + ], + "score": 1.0, + "content": "built in the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 378, + 386, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 162, + 390 + ], + "score": 1.0, + "content": "environment.", + "type": "text" + }, + { + "bbox": [ + 162, + 378, + 171, + 388 + ], + "score": 0.26, + "content": "\\Pi", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 378, + 174, + 390 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 174, + 378, + 184, + 388 + ], + "score": 0.65, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 378, + 205, + 390 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 205, + 378, + 214, + 388 + ], + "score": 0.78, + "content": "\\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 378, + 386, + 390 + ], + "score": 1.0, + "content": "denote the policy, state, and action spaces.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 107, + 396, + 173, + 409 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 174, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 174, + 411 + ], + "score": 1.0, + "content": "4 METHOD", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 413, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 411, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 505, + 426 + ], + "score": 1.0, + "content": "In this section, we first propose a new offline model-learning objective based on Def. 3.1 for policy", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 423, + 507, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 507, + 437 + ], + "score": 1.0, + "content": "optimization tasks in Sec. 4.1; In Sec. 4.2, we derive a tractable solution to the proposed objective;", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 434, + 329, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 329, + 448 + ], + "score": 1.0, + "content": "Finally, we give a practical implementation in Sec. 4.3.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "title", + "bbox": [ + 107, + 454, + 240, + 465 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 241, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 241, + 466 + ], + "score": 1.0, + "content": "4.1 PROBLEM FORMULATION", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 471, + 506, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "For offline policy optimization, we require the environment model to have generalization ability in", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 482, + 507, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 491, + 495 + ], + "score": 1.0, + "content": "counterfactual queries since we need to query numerous different policies’ correct feedback from", + "type": "text" + }, + { + "bbox": [ + 491, + 483, + 503, + 493 + ], + "score": 0.76, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 482, + 507, + 495 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 481, + 506 + ], + "score": 1.0, + "content": "Referring to the formulation of WERM through IPS in Def. 3.1, policy optimization requires", + "type": "text" + }, + { + "bbox": [ + 482, + 494, + 493, + 504 + ], + "score": 0.77, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 493, + 506, + 506 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 505, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 505, + 516 + ], + "score": 1.0, + "content": "minimize counterfactual-query risks under numerous unknown different policies rather than a specific", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 515, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 160, + 528 + ], + "score": 1.0, + "content": "target policy", + "type": "text" + }, + { + "bbox": [ + 160, + 516, + 167, + 527 + ], + "score": 0.8, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 515, + 321, + 528 + ], + "score": 1.0, + "content": ". More specifically, the question is: If", + "type": "text" + }, + { + "bbox": [ + 321, + 516, + 329, + 527 + ], + "score": 0.85, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 515, + 505, + 528 + ], + "score": 1.0, + "content": "is unknown and can be varied, how should", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 526, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 506, + 540 + ], + "score": 1.0, + "content": "we reduce the risks in counterfactual queries? In this article, we call the model learning problem in", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 537, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 506, + 550 + ], + "score": 1.0, + "content": "this setting “counterfactual environment model learning” and propose a new objective to handle the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 547, + 506, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 506, + 561 + ], + "score": 1.0, + "content": "problem. To be compatible with multi-step environment model learning, we first define a generalized", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 558, + 264, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 264, + 572 + ], + "score": 1.0, + "content": "WERM through IPS based on Def. 3.1.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 573, + 505, + 607 + ], + "lines": [ + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 306, + 586 + ], + "score": 1.0, + "content": "Definition 4.1. Given the MDP transition function", + "type": "text" + }, + { + "bbox": [ + 306, + 574, + 322, + 584 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 573, + 374, + 586 + ], + "score": 1.0, + "content": "that satisfies", + "type": "text" + }, + { + "bbox": [ + 374, + 574, + 506, + 586 + ], + "score": 0.92, + "content": "M ^ { * } ( x ^ { \\prime } | x , a ) > 0 , \\forall x \\in \\mathcal { X } , \\forall a \\in", + "type": "inline_equation" + } + ], + "index": 38 + }, + { + "bbox": [ + 107, + 584, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 107, + 585, + 157, + 596 + ], + "score": 0.9, + "content": "\\mathcal { A } , \\forall x ^ { \\prime } \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 584, + 177, + 598 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 177, + 586, + 185, + 596 + ], + "score": 0.8, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 584, + 222, + 598 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 222, + 585, + 347, + 597 + ], + "score": 0.91, + "content": "\\mu ( a | x ) > 0 , \\forall a \\in \\mathcal { A } , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 584, + 506, + 598 + ], + "score": 1.0, + "content": ", the learning objective of generalized", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 595, + 254, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 254, + 609 + ], + "score": 1.0, + "content": "WERM through IPS is formulated as", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 610, + 425, + 662 + ], + "lines": [ + { + "bbox": [ + 185, + 610, + 425, + 662 + ], + "spans": [ + { + "bbox": [ + 185, + 610, + 425, + 662 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } L ( M ) = \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\mu } } [ \\omega ( x , a , x ^ { \\prime } ) \\ell _ { M } ( x , a , x ^ { \\prime } ) ] , } \\\\ & { ~ s . t . ~ \\omega ( x , a , x ^ { \\prime } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } ) } , } \\end{array}", + "type": "interline_equation", + "image_path": "6b922044bab2cecce8ce3cfa5b9946e9dc58b17f41c4fcc348163596a1c8d808.jpg" + } + ] + } + ], + "index": 42.5, + "virtual_lines": [ + { + "bbox": [ + 185, + 610, + 425, + 623.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 185, + 623.0, + 425, + 636.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 185, + 636.0, + 425, + 649.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 185, + 649.0, + 425, + 662.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 666, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 664, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 132, + 682 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 667, + 151, + 680 + ], + "score": 0.89, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 664, + 353, + 682 + ], + "score": 1.0, + "content": "is the training data distribution (collected by policy", + "type": "text" + }, + { + "bbox": [ + 353, + 669, + 360, + 678 + ], + "score": 0.33, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 664, + 366, + 682 + ], + "score": 1.0, + "content": "),", + "type": "text" + }, + { + "bbox": [ + 367, + 665, + 385, + 679 + ], + "score": 0.88, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 664, + 506, + 682 + ], + "score": 1.0, + "content": "is the testing data distribution", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 678, + 424, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 186, + 691 + ], + "score": 1.0, + "content": "(collected by policy", + "type": "text" + }, + { + "bbox": [ + 186, + 679, + 194, + 690 + ], + "score": 0.76, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 678, + 242, + 691 + ], + "score": 1.0, + "content": "). We define", + "type": "text" + }, + { + "bbox": [ + 242, + 678, + 379, + 690 + ], + "score": 0.88, + "content": "\\ell _ { M } ( x , a , x ^ { \\prime } ) : = \\ell ( M ( x ^ { \\prime } | x , a ) , x ^ { \\prime } ) ^ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 678, + 424, + 691 + ], + "score": 1.0, + "content": "for brevity.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + }, + { + "type": "text", + "bbox": [ + 106, + 698, + 505, + 734 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 238, + 712 + ], + "score": 1.0, + "content": "In an MDP, given any policy", + "type": "text" + }, + { + "bbox": [ + 239, + 701, + 246, + 709 + ], + "score": 0.56, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 697, + 251, + 712 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 252, + 698, + 439, + 711 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\pi } ( x , a , x ^ { \\prime } ) ~ = ~ \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 697, + 471, + 712 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 472, + 699, + 505, + 711 + ], + "score": 0.88, + "content": "\\rho _ { M ^ { \\ast } } ^ { \\pi } ( x )", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 104, + 707, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 104, + 707, + 263, + 724 + ], + "score": 1.0, + "content": "denotes the occupancy measure of", + "type": "text" + }, + { + "bbox": [ + 263, + 712, + 270, + 720 + ], + "score": 0.81, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 707, + 322, + 724 + ], + "score": 1.0, + "content": "for policy", + "type": "text" + }, + { + "bbox": [ + 322, + 713, + 330, + 720 + ], + "score": 0.77, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 707, + 451, + 724 + ], + "score": 1.0, + "content": ", which can be defined as", + "type": "text" + }, + { + "bbox": [ + 452, + 711, + 505, + 722 + ], + "score": 0.87, + "content": "\\rho _ { M ^ { \\ast } } ^ { \\pi } ( x ) : =", + "type": "inline_equation" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 718, + 507, + 735 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 289, + 734 + ], + "score": 0.92, + "content": "\\begin{array} { r } { ( 1 - \\gamma ) \\mathbb { E } _ { x _ { 0 } \\sim \\rho _ { 0 } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\dot { \\gamma } ^ { t } \\mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 718, + 325, + 735 + ], + "score": 1.0, + "content": "(Sutton", + "type": "text" + }, + { + "bbox": [ + 325, + 722, + 334, + 731 + ], + "score": 0.49, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 718, + 507, + 735 + ], + "score": 1.0, + "content": "Barto, 1998; Ho & Ermon, 2016) where", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 422, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 423, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 423, + 97 + ], + "score": 1.0, + "content": "Definition 3.1. The learning objective of WERM through IPS is formulated as", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 105, + 80, + 423, + 97 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 97, + 424, + 146 + ], + "lines": [ + { + "bbox": [ + 185, + 97, + 424, + 146 + ], + "spans": [ + { + "bbox": [ + 185, + 97, + 424, + 146 + ], + "score": 0.92, + "content": "\\begin{array} { r l } { \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } L ( M ) = \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } \\mathbb { E } _ { { x } , { a } , { y } \\sim { p } _ { M ^ { * } } } [ \\omega ( { x } , { a } ) \\ell ( M ( { y } | { x } , { a } ) , { y } ) ] , } & { } \\\\ { { s . t . } \\quad \\omega ( { x } , { a } ) = \\frac { \\beta ( a | { x } ) } { \\mu ( a | { x } ) } , } & { } \\end{array}", + "type": "interline_equation", + "image_path": "84b11ff30828b8a64799e15e9dedc010134f03df389f6ff864ac494a2de8e6c2.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 185, + 97, + 424, + 113.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 185, + 113.33333333333333, + 424, + 129.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 185, + 129.66666666666666, + 424, + 146.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 149, + 504, + 183 + ], + "lines": [ + { + "bbox": [ + 105, + 148, + 506, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 133, + 163 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 150, + 141, + 160 + ], + "score": 0.79, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 148, + 159, + 163 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 159, + 151, + 167, + 160 + ], + "score": 0.7, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 148, + 393, + 163 + ], + "score": 1.0, + "content": "denote the policies in testing and training domains, and", + "type": "text" + }, + { + "bbox": [ + 393, + 149, + 412, + 162 + ], + "score": 0.92, + "content": "p _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 148, + 506, + 163 + ], + "score": 1.0, + "content": "is the joint probability", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 159, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 271, + 172 + ], + "score": 0.9, + "content": "p _ { M ^ { * } } ^ { \\mu } ( x , a , y ) : = \\rho _ { 0 } ( x ) \\mu ( a | x ) M ^ { * } ( y | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 159, + 308, + 174 + ], + "score": 1.0, + "content": "in which", + "type": "text" + }, + { + "bbox": [ + 309, + 160, + 333, + 172 + ], + "score": 0.92, + "content": "\\rho _ { 0 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 159, + 441, + 174 + ], + "score": 1.0, + "content": "is the distribution of state.", + "type": "text" + }, + { + "bbox": [ + 441, + 161, + 454, + 170 + ], + "score": 0.74, + "content": "\\mathcal { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 159, + 506, + 174 + ], + "score": 1.0, + "content": "is the model", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 171, + 291, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 134, + 185 + ], + "score": 1.0, + "content": "space.", + "type": "text" + }, + { + "bbox": [ + 135, + 173, + 140, + 181 + ], + "score": 0.53, + "content": "\\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 171, + 291, + 185 + ], + "score": 1.0, + "content": "is a loss function for model learning.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 148, + 506, + 185 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 191, + 505, + 293 + ], + "lines": [ + { + "bbox": [ + 105, + 191, + 505, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 124, + 204 + ], + "score": 1.0, + "content": "The", + "type": "text" + }, + { + "bbox": [ + 124, + 193, + 132, + 202 + ], + "score": 0.79, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 191, + 505, + 204 + ], + "score": 1.0, + "content": "is also known as importance sampling (IS) weight, which corrects the sampling bias. In this", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 203, + 504, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 149, + 214 + ], + "score": 1.0, + "content": "objective,", + "type": "text" + }, + { + "bbox": [ + 149, + 204, + 157, + 212 + ], + "score": 0.74, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 203, + 496, + 214 + ], + "score": 1.0, + "content": "is to align the training data distribution to the testing data. By selecting different", + "type": "text" + }, + { + "bbox": [ + 496, + 203, + 504, + 212 + ], + "score": 0.79, + "content": "\\hat { \\omega }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 214, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 170, + 226 + ], + "score": 1.0, + "content": "to approximate", + "type": "text" + }, + { + "bbox": [ + 171, + 216, + 179, + 223 + ], + "score": 0.75, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 214, + 257, + 226 + ], + "score": 1.0, + "content": "to learn the model", + "type": "text" + }, + { + "bbox": [ + 257, + 214, + 269, + 224 + ], + "score": 0.77, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 214, + 505, + 226 + ], + "score": 1.0, + "content": ", current environment model learning algorithms for ITEs", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 225, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 505, + 236 + ], + "score": 1.0, + "content": "estimation are fallen into the framework. In standard supervised learning and some works for ITEs", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 235, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 325, + 248 + ], + "score": 1.0, + "content": "estimation (Wager & Athey, 2018; Weiss et al., 2015),", + "type": "text" + }, + { + "bbox": [ + 325, + 235, + 374, + 248 + ], + "score": 0.92, + "content": "\\hat { \\omega } ( x , a ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 236, + 505, + 248 + ], + "score": 1.0, + "content": "as the distribution-shift problem", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 245, + 504, + 261 + ], + "spans": [ + { + "bbox": [ + 104, + 245, + 446, + 260 + ], + "score": 1.0, + "content": "is ignored. In Shimodaira (2000); Assaad et al. (2021); Hassanpour & Greiner (2019),", + "type": "text" + }, + { + "bbox": [ + 447, + 245, + 475, + 261 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\omega = \\frac { \\mathrm { i } } { \\hat { \\mu } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 245, + 496, + 260 + ], + "score": 1.0, + "content": "(i.e.,", + "type": "text" + }, + { + "bbox": [ + 496, + 246, + 504, + 258 + ], + "score": 0.82, + "content": "\\beta", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 258, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 370, + 272 + ], + "score": 1.0, + "content": "a uniform policy) for balancing treatment and control group, where", + "type": "text" + }, + { + "bbox": [ + 370, + 259, + 378, + 270 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 258, + 506, + 272 + ], + "score": 1.0, + "content": "is an approximation of behavior", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 134, + 282 + ], + "score": 1.0, + "content": "policy", + "type": "text" + }, + { + "bbox": [ + 135, + 272, + 142, + 281 + ], + "score": 0.73, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 271, + 505, + 282 + ], + "score": 1.0, + "content": ". Note that it is a reasonable weight in ITEs estimation: ITEs are defined to evaluate the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 280, + 440, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 440, + 293 + ], + "score": 1.0, + "content": "effect of each state between treatment and control behavior under a uniform policy.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 11, + "bbox_fs": [ + 104, + 191, + 506, + 293 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 297, + 313, + 308 + ], + "lines": [ + { + "bbox": [ + 106, + 297, + 313, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 313, + 309 + ], + "score": 1.0, + "content": "3.2 SEQUENTIAL DECISION-MAKING SETTING", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 312, + 505, + 389 + ], + "lines": [ + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 325 + ], + "score": 1.0, + "content": "Decision-making processes in a sequential environment are often formulated into Markov Decision", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "Process (MDP) (Sutton & Barto, 1998). MDP depicts an agent interacting with the environment", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 430, + 346 + ], + "score": 1.0, + "content": "through actions. In the first step, states are sampled from an initial state distribution", + "type": "text" + }, + { + "bbox": [ + 430, + 334, + 478, + 345 + ], + "score": 0.91, + "content": "x _ { 0 } \\sim \\rho _ { 0 } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 334, + 505, + 346 + ], + "score": 1.0, + "content": ". Then", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 344, + 504, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 178, + 357 + ], + "score": 1.0, + "content": "at each time-step", + "type": "text" + }, + { + "bbox": [ + 178, + 345, + 241, + 357 + ], + "score": 0.92, + "content": "t \\in \\{ 0 , 1 , 2 , \\ldots \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 344, + 347, + 357 + ], + "score": 1.0, + "content": ", the agent takes an action", + "type": "text" + }, + { + "bbox": [ + 348, + 345, + 378, + 356 + ], + "score": 0.92, + "content": "a _ { t } \\in { \\mathcal { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 344, + 448, + 357 + ], + "score": 1.0, + "content": "through a policy", + "type": "text" + }, + { + "bbox": [ + 449, + 345, + 504, + 357 + ], + "score": 0.91, + "content": "\\pi ( a _ { t } | x _ { t } ) \\in \\Pi", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 181, + 369 + ], + "score": 1.0, + "content": "based on the state", + "type": "text" + }, + { + "bbox": [ + 182, + 357, + 213, + 367 + ], + "score": 0.91, + "content": "x _ { t } \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 356, + 350, + 369 + ], + "score": 1.0, + "content": ", then the agent receives a reward", + "type": "text" + }, + { + "bbox": [ + 351, + 358, + 360, + 367 + ], + "score": 0.84, + "content": "r _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 356, + 458, + 369 + ], + "score": 1.0, + "content": "from a reward function", + "type": "text" + }, + { + "bbox": [ + 458, + 357, + 505, + 368 + ], + "score": 0.89, + "content": "r ( x _ { t } , a _ { t } ) \\in", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 366, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 115, + 377 + ], + "score": 0.79, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 366, + 237, + 380 + ], + "score": 1.0, + "content": "and transits to the next state", + "type": "text" + }, + { + "bbox": [ + 237, + 369, + 258, + 379 + ], + "score": 0.89, + "content": "x _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 366, + 386, + 380 + ], + "score": 1.0, + "content": "given by a transition function", + "type": "text" + }, + { + "bbox": [ + 387, + 367, + 455, + 379 + ], + "score": 0.93, + "content": "M ^ { \\ast } ( x _ { t + 1 } | x _ { t } , a _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 366, + 505, + 380 + ], + "score": 1.0, + "content": "built in the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 378, + 386, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 162, + 390 + ], + "score": 1.0, + "content": "environment.", + "type": "text" + }, + { + "bbox": [ + 162, + 378, + 171, + 388 + ], + "score": 0.26, + "content": "\\Pi", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 378, + 174, + 390 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 174, + 378, + 184, + 388 + ], + "score": 0.65, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 378, + 205, + 390 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 205, + 378, + 214, + 388 + ], + "score": 0.78, + "content": "\\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 378, + 386, + 390 + ], + "score": 1.0, + "content": "denote the policy, state, and action spaces.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 311, + 506, + 390 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 396, + 173, + 409 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 174, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 174, + 411 + ], + "score": 1.0, + "content": "4 METHOD", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 413, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 411, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 505, + 426 + ], + "score": 1.0, + "content": "In this section, we first propose a new offline model-learning objective based on Def. 3.1 for policy", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 423, + 507, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 507, + 437 + ], + "score": 1.0, + "content": "optimization tasks in Sec. 4.1; In Sec. 4.2, we derive a tractable solution to the proposed objective;", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 434, + 329, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 329, + 448 + ], + "score": 1.0, + "content": "Finally, we give a practical implementation in Sec. 4.3.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 411, + 507, + 448 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 454, + 240, + 465 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 241, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 241, + 466 + ], + "score": 1.0, + "content": "4.1 PROBLEM FORMULATION", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 471, + 506, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 506, + 484 + ], + "score": 1.0, + "content": "For offline policy optimization, we require the environment model to have generalization ability in", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 482, + 507, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 491, + 495 + ], + "score": 1.0, + "content": "counterfactual queries since we need to query numerous different policies’ correct feedback from", + "type": "text" + }, + { + "bbox": [ + 491, + 483, + 503, + 493 + ], + "score": 0.76, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 482, + 507, + 495 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 493, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 481, + 506 + ], + "score": 1.0, + "content": "Referring to the formulation of WERM through IPS in Def. 3.1, policy optimization requires", + "type": "text" + }, + { + "bbox": [ + 482, + 494, + 493, + 504 + ], + "score": 0.77, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 493, + 506, + 506 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 505, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 106, + 505, + 505, + 516 + ], + "score": 1.0, + "content": "minimize counterfactual-query risks under numerous unknown different policies rather than a specific", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 515, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 160, + 528 + ], + "score": 1.0, + "content": "target policy", + "type": "text" + }, + { + "bbox": [ + 160, + 516, + 167, + 527 + ], + "score": 0.8, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 515, + 321, + 528 + ], + "score": 1.0, + "content": ". More specifically, the question is: If", + "type": "text" + }, + { + "bbox": [ + 321, + 516, + 329, + 527 + ], + "score": 0.85, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 515, + 505, + 528 + ], + "score": 1.0, + "content": "is unknown and can be varied, how should", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 526, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 506, + 540 + ], + "score": 1.0, + "content": "we reduce the risks in counterfactual queries? In this article, we call the model learning problem in", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 537, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 506, + 550 + ], + "score": 1.0, + "content": "this setting “counterfactual environment model learning” and propose a new objective to handle the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 547, + 506, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 506, + 561 + ], + "score": 1.0, + "content": "problem. To be compatible with multi-step environment model learning, we first define a generalized", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 558, + 264, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 264, + 572 + ], + "score": 1.0, + "content": "WERM through IPS based on Def. 3.1.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 471, + 507, + 572 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 573, + 505, + 607 + ], + "lines": [ + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 306, + 586 + ], + "score": 1.0, + "content": "Definition 4.1. Given the MDP transition function", + "type": "text" + }, + { + "bbox": [ + 306, + 574, + 322, + 584 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 573, + 374, + 586 + ], + "score": 1.0, + "content": "that satisfies", + "type": "text" + }, + { + "bbox": [ + 374, + 574, + 506, + 586 + ], + "score": 0.92, + "content": "M ^ { * } ( x ^ { \\prime } | x , a ) > 0 , \\forall x \\in \\mathcal { X } , \\forall a \\in", + "type": "inline_equation" + } + ], + "index": 38 + }, + { + "bbox": [ + 107, + 584, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 107, + 585, + 157, + 596 + ], + "score": 0.9, + "content": "\\mathcal { A } , \\forall x ^ { \\prime } \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 584, + 177, + 598 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 177, + 586, + 185, + 596 + ], + "score": 0.8, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 584, + 222, + 598 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 222, + 585, + 347, + 597 + ], + "score": 0.91, + "content": "\\mu ( a | x ) > 0 , \\forall a \\in \\mathcal { A } , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 584, + 506, + 598 + ], + "score": 1.0, + "content": ", the learning objective of generalized", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 595, + 254, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 254, + 609 + ], + "score": 1.0, + "content": "WERM through IPS is formulated as", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 573, + 506, + 609 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 610, + 425, + 662 + ], + "lines": [ + { + "bbox": [ + 185, + 610, + 425, + 662 + ], + "spans": [ + { + "bbox": [ + 185, + 610, + 425, + 662 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } L ( M ) = \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\mu } } [ \\omega ( x , a , x ^ { \\prime } ) \\ell _ { M } ( x , a , x ^ { \\prime } ) ] , } \\\\ & { ~ s . t . ~ \\omega ( x , a , x ^ { \\prime } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } ) } , } \\end{array}", + "type": "interline_equation", + "image_path": "6b922044bab2cecce8ce3cfa5b9946e9dc58b17f41c4fcc348163596a1c8d808.jpg" + } + ] + } + ], + "index": 42.5, + "virtual_lines": [ + { + "bbox": [ + 185, + 610, + 425, + 623.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 185, + 623.0, + 425, + 636.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 185, + 636.0, + 425, + 649.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 185, + 649.0, + 425, + 662.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 666, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 664, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 132, + 682 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 667, + 151, + 680 + ], + "score": 0.89, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 664, + 353, + 682 + ], + "score": 1.0, + "content": "is the training data distribution (collected by policy", + "type": "text" + }, + { + "bbox": [ + 353, + 669, + 360, + 678 + ], + "score": 0.33, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 664, + 366, + 682 + ], + "score": 1.0, + "content": "),", + "type": "text" + }, + { + "bbox": [ + 367, + 665, + 385, + 679 + ], + "score": 0.88, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 664, + 506, + 682 + ], + "score": 1.0, + "content": "is the testing data distribution", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 678, + 424, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 186, + 691 + ], + "score": 1.0, + "content": "(collected by policy", + "type": "text" + }, + { + "bbox": [ + 186, + 679, + 194, + 690 + ], + "score": 0.76, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 678, + 242, + 691 + ], + "score": 1.0, + "content": "). We define", + "type": "text" + }, + { + "bbox": [ + 242, + 678, + 379, + 690 + ], + "score": 0.88, + "content": "\\ell _ { M } ( x , a , x ^ { \\prime } ) : = \\ell ( M ( x ^ { \\prime } | x , a ) , x ^ { \\prime } ) ^ { - }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 678, + 424, + 691 + ], + "score": 1.0, + "content": "for brevity.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 664, + 506, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 698, + 505, + 734 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 238, + 712 + ], + "score": 1.0, + "content": "In an MDP, given any policy", + "type": "text" + }, + { + "bbox": [ + 239, + 701, + 246, + 709 + ], + "score": 0.56, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 697, + 251, + 712 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 252, + 698, + 439, + 711 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\pi } ( x , a , x ^ { \\prime } ) ~ = ~ \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 697, + 471, + 712 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 472, + 699, + 505, + 711 + ], + "score": 0.88, + "content": "\\rho _ { M ^ { \\ast } } ^ { \\pi } ( x )", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 104, + 707, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 104, + 707, + 263, + 724 + ], + "score": 1.0, + "content": "denotes the occupancy measure of", + "type": "text" + }, + { + "bbox": [ + 263, + 712, + 270, + 720 + ], + "score": 0.81, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 707, + 322, + 724 + ], + "score": 1.0, + "content": "for policy", + "type": "text" + }, + { + "bbox": [ + 322, + 713, + 330, + 720 + ], + "score": 0.77, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 707, + 451, + 724 + ], + "score": 1.0, + "content": ", which can be defined as", + "type": "text" + }, + { + "bbox": [ + 452, + 711, + 505, + 722 + ], + "score": 0.87, + "content": "\\rho _ { M ^ { \\ast } } ^ { \\pi } ( x ) : =", + "type": "inline_equation" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 718, + 507, + 735 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 289, + 734 + ], + "score": 0.92, + "content": "\\begin{array} { r } { ( 1 - \\gamma ) \\mathbb { E } _ { x _ { 0 } \\sim \\rho _ { 0 } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\dot { \\gamma } ^ { t } \\mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 718, + 325, + 735 + ], + "score": 1.0, + "content": "(Sutton", + "type": "text" + }, + { + "bbox": [ + 325, + 722, + 334, + 731 + ], + "score": 0.49, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 718, + 507, + 735 + ], + "score": 1.0, + "content": "Barto, 1998; Ho & Ermon, 2016) where", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 80, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 192, + 95 + ], + "score": 0.92, + "content": " { \\mathrm { P r } } ^ { \\pi } \\left[ x _ { t } = x | x _ { 0 } , M ^ { * } \\right]", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 192, + 80, + 392, + 96 + ], + "score": 1.0, + "content": "is the state visitation probability that the policy", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 392, + 85, + 399, + 92 + ], + "score": 0.74, + "content": "\\pi", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 400, + 80, + 425, + 96 + ], + "score": 1.0, + "content": "visits", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 426, + 85, + 433, + 92 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 433, + 80, + 486, + 96 + ], + "score": 1.0, + "content": "at time-step", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 486, + 84, + 491, + 92 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 491, + 80, + 506, + 96 + ], + "score": 1.0, + "content": "by", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 507, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 228, + 106 + ], + "score": 1.0, + "content": "executing in the environment", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 228, + 94, + 244, + 104 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 244, + 92, + 343, + 106 + ], + "score": 1.0, + "content": "and starting at the state", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 343, + 95, + 354, + 105 + ], + "score": 0.84, + "content": "x _ { 0 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 354, + 92, + 375, + 106 + ], + "score": 1.0, + "content": ", and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 376, + 94, + 415, + 106 + ], + "score": 0.92, + "content": "\\gamma \\in [ 0 , 1 ]", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 416, + 92, + 507, + 106 + ], + "score": 1.0, + "content": "is the discount factor.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 186, + 118 + ], + "score": 1.0, + "content": "Here we also define", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 186, + 105, + 303, + 117 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\pi } ( x , a ) : = \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 304, + 104, + 431, + 118 + ], + "score": 1.0, + "content": "for simplicity. In this definition,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 431, + 106, + 439, + 114 + ], + "score": 0.75, + "content": "\\omega", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 439, + 104, + 506, + 118 + ], + "score": 1.0, + "content": "can be rewritten", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 112, + 507, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 121, + 132 + ], + "score": 1.0, + "content": "as:", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 121, + 117, + 317, + 137 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\omega ( x , a , x ^ { \\prime } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x ) \\beta ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x ) \\mu ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a ) } = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) } } \\end{array}", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 317, + 112, + 495, + 135 + ], + "score": 1.0, + "content": ". In single-step environments, for any policy", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 496, + 123, + 502, + 131 + ], + "score": 0.72, + "content": "\\pi", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 503, + 112, + 507, + 135 + ], + "score": 1.0, + "content": ",", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 134, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 134, + 340, + 153 + ], + "score": 1.0, + "content": "ρπM∗ (x) = ρ0(x). Then we have ω(x, a, x′) = ρ0(x)β(a|x)ρ (x)µ(a|x)", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 375, + 135, + 505, + 155 + ], + "score": 1.0, + "content": ", and the objective is degraded to", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 379, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 379, + 164 + ], + "score": 1.0, + "content": "Eq. 1. Therefore, Def. 3.1 is a special case of this generalized form.", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 48, + "bbox_fs": [ + 104, + 697, + 507, + 735 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 506, + 163 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 192, + 95 + ], + "score": 0.92, + "content": " { \\mathrm { P r } } ^ { \\pi } \\left[ x _ { t } = x | x _ { 0 } , M ^ { * } \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 80, + 392, + 96 + ], + "score": 1.0, + "content": "is the state visitation probability that the policy", + "type": "text" + }, + { + "bbox": [ + 392, + 85, + 399, + 92 + ], + "score": 0.74, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 80, + 425, + 96 + ], + "score": 1.0, + "content": "visits", + "type": "text" + }, + { + "bbox": [ + 426, + 85, + 433, + 92 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 80, + 486, + 96 + ], + "score": 1.0, + "content": "at time-step", + "type": "text" + }, + { + "bbox": [ + 486, + 84, + 491, + 92 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 80, + 506, + 96 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 507, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 228, + 106 + ], + "score": 1.0, + "content": "executing in the environment", + "type": "text" + }, + { + "bbox": [ + 228, + 94, + 244, + 104 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 92, + 343, + 106 + ], + "score": 1.0, + "content": "and starting at the state", + "type": "text" + }, + { + "bbox": [ + 343, + 95, + 354, + 105 + ], + "score": 0.84, + "content": "x _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 92, + 375, + 106 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 376, + 94, + 415, + 106 + ], + "score": 0.92, + "content": "\\gamma \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 92, + 507, + 106 + ], + "score": 1.0, + "content": "is the discount factor.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 186, + 118 + ], + "score": 1.0, + "content": "Here we also define", + "type": "text" + }, + { + "bbox": [ + 186, + 105, + 303, + 117 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\pi } ( x , a ) : = \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 104, + 431, + 118 + ], + "score": 1.0, + "content": "for simplicity. In this definition,", + "type": "text" + }, + { + "bbox": [ + 431, + 106, + 439, + 114 + ], + "score": 0.75, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 104, + 506, + 118 + ], + "score": 1.0, + "content": "can be rewritten", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 112, + 507, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 121, + 132 + ], + "score": 1.0, + "content": "as:", + "type": "text" + }, + { + "bbox": [ + 121, + 117, + 317, + 137 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\omega ( x , a , x ^ { \\prime } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x ) \\beta ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x ) \\mu ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a ) } = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 112, + 495, + 135 + ], + "score": 1.0, + "content": ". In single-step environments, for any policy", + "type": "text" + }, + { + "bbox": [ + 496, + 123, + 502, + 131 + ], + "score": 0.72, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 112, + 507, + 135 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 134, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 134, + 340, + 153 + ], + "score": 1.0, + "content": "ρπM∗ (x) = ρ0(x). Then we have ω(x, a, x′) = ρ0(x)β(a|x)ρ (x)µ(a|x)", + "type": "text" + }, + { + "bbox": [ + 375, + 135, + 505, + 155 + ], + "score": 1.0, + "content": ", and the objective is degraded to", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 151, + 379, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 379, + 164 + ], + "score": 1.0, + "content": "Eq. 1. Therefore, Def. 3.1 is a special case of this generalized form.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 168, + 505, + 223 + ], + "lines": [ + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 168, + 130, + 180 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 131, + 169, + 138, + 180 + ], + "score": 0.83, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 168, + 505, + 180 + ], + "score": 1.0, + "content": "for counterfactual queries is unknown and can be varied in policy optimization, to reduce the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 178, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 505, + 191 + ], + "score": 1.0, + "content": "risks in counterfactual queries in this scenario, we introduce adversarial policies which can induce the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 190, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 506, + 202 + ], + "score": 1.0, + "content": "worst performance of the model predictions and propose to optimize WERM under the adversarial", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 201, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 506, + 213 + ], + "score": 1.0, + "content": "policies. In particular, we propose Adversarial Weighted empirical Risk Minimization (AWRM)", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 212, + 272, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 272, + 224 + ], + "score": 1.0, + "content": "based on Def. 4.1 to handle this problem.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 105, + 225, + 505, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 304, + 239 + ], + "score": 1.0, + "content": "Definition 4.2. Given the MDP transition function", + "type": "text" + }, + { + "bbox": [ + 305, + 226, + 320, + 236 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 225, + 506, + 239 + ], + "score": 1.0, + "content": ", the learning objective of adversarial Weighted", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 237, + 336, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 336, + 249 + ], + "score": 1.0, + "content": "empirical risk minimization through IPS is formulated as", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "interline_equation", + "bbox": [ + 134, + 250, + 475, + 302 + ], + "lines": [ + { + "bbox": [ + 134, + 250, + 475, + 302 + ], + "spans": [ + { + "bbox": [ + 134, + 250, + 475, + 302 + ], + "score": 0.89, + "content": "\\begin{array} { r l } & { \\hat { \\boldsymbol M } ^ { * } = \\underset { M \\in \\mathcal { M } } { \\mathrm { \\operatorname* { m i n } } } \\underset { \\beta \\in \\Pi } { \\mathrm { m a x } } \\boldsymbol L ( \\rho _ { M ^ { * } } ^ { \\beta } , M ) = \\underset { M \\in \\mathcal { M } } { \\mathrm { m i n } } \\underset { \\beta \\in \\Pi } { \\mathrm { m a x } } \\mathbb { E } _ { \\boldsymbol x , \\boldsymbol a , \\boldsymbol x ^ { \\prime } \\sim \\boldsymbol \\rho _ { M ^ { * } } ^ { \\mu } } [ \\omega ( \\boldsymbol x , \\boldsymbol a | \\rho _ { M ^ { * } } ^ { \\beta } ) \\ell _ { M } ( \\boldsymbol x , \\boldsymbol a , \\boldsymbol x ^ { \\prime } ) ] , } \\\\ & { } \\\\ & { s . t . \\quad \\omega ( \\boldsymbol x , \\boldsymbol a | \\rho _ { M ^ { * } } ^ { \\beta } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( \\boldsymbol x , \\boldsymbol a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( \\boldsymbol x , \\boldsymbol a ) } , } \\end{array}", + "type": "interline_equation", + "image_path": "39944d2b5a594f42b7761eb3e27008f21b546a70714b76f8a617a7dfda623ea0.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 134, + 250, + 475, + 267.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 134, + 267.3333333333333, + 475, + 284.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 134, + 284.66666666666663, + 475, + 301.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 305, + 504, + 333 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 223, + 322 + ], + "score": 1.0, + "content": "where the re-weighting term", + "type": "text" + }, + { + "bbox": [ + 223, + 304, + 274, + 319 + ], + "score": 0.93, + "content": "\\omega ( x , a | \\rho _ { M ^ { * } } ^ { \\beta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 303, + 432, + 322 + ], + "score": 1.0, + "content": "is conditioned on the data distribution", + "type": "text" + }, + { + "bbox": [ + 433, + 304, + 451, + 319 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 303, + 506, + 322 + ], + "score": 1.0, + "content": "of the adver-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 318, + 430, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 159, + 335 + ], + "score": 1.0, + "content": "sarial policy", + "type": "text" + }, + { + "bbox": [ + 159, + 321, + 167, + 331 + ], + "score": 0.68, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 318, + 298, + 335 + ], + "score": 1.0, + "content": ". In the following, we will ignore", + "type": "text" + }, + { + "bbox": [ + 298, + 318, + 317, + 333 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 318, + 352, + 335 + ], + "score": 1.0, + "content": "and use", + "type": "text" + }, + { + "bbox": [ + 352, + 320, + 382, + 332 + ], + "score": 0.92, + "content": "\\omega ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 318, + 430, + 335 + ], + "score": 1.0, + "content": "for brevity.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 107, + 339, + 505, + 386 + ], + "lines": [ + { + "bbox": [ + 106, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 259, + 352 + ], + "score": 1.0, + "content": "According to the definition of MDP,", + "type": "text" + }, + { + "bbox": [ + 259, + 339, + 348, + 352 + ], + "score": 0.93, + "content": "\\omega ( x , a , x ^ { \\prime } ) = \\omega ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "since the transition probability in the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 350, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 506, + 363 + ], + "score": 1.0, + "content": "ratio can be canceled. Eq. 3 minimizes the maximum model loss under all counterfactual data", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 362, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 160, + 378 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 161, + 362, + 248, + 376 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { \\ast } } ^ { \\beta } ( x , a , x ^ { \\prime } ) , \\beta \\in \\Pi", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 362, + 506, + 378 + ], + "score": 1.0, + "content": "to guarantee the generalization ability for counterfactual data", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 375, + 208, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 195, + 386 + ], + "score": 1.0, + "content": "queried by policies in", + "type": "text" + }, + { + "bbox": [ + 195, + 376, + 204, + 384 + ], + "score": 0.72, + "content": "\\Pi", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 375, + 208, + 386 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "title", + "bbox": [ + 107, + 393, + 266, + 404 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 268, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 268, + 406 + ], + "score": 1.0, + "content": "4.2 TRACTABLE AWRM SOLUTION", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 410, + 504, + 433 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "In this section, we propose a tractable solution to optimize Eq. 3. The full derivations can be found in", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 422, + 501, + 434 + ], + "spans": [ + { + "bbox": [ + 107, + 422, + 206, + 434 + ], + "score": 1.0, + "content": "the Appx. A. We choose", + "type": "text" + }, + { + "bbox": [ + 206, + 423, + 220, + 433 + ], + "score": 0.9, + "content": "\\ell _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 422, + 501, + 434 + ], + "score": 1.0, + "content": "to be the negative log-likelihood loss and derive the solution to Eq. 3:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "interline_equation", + "bbox": [ + 171, + 435, + 438, + 457 + ], + "lines": [ + { + "bbox": [ + 171, + 435, + 438, + 457 + ], + "spans": [ + { + "bbox": [ + 171, + 435, + 438, + 457 + ], + "score": 0.91, + "content": "L ( \\rho _ { M ^ { \\ast } } ^ { \\beta } , M ) = \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { \\ast } } ^ { \\mu } } \\left[ \\omega ( x , a | \\rho _ { M ^ { \\ast } } ^ { \\beta } ) \\mathbb { E } _ { M ^ { \\ast } } \\left( - \\log M ( x ^ { \\prime } | x , a ) \\right) \\right] ,", + "type": "interline_equation", + "image_path": "072350e4aea7ec2b69e997e13744c49a14155f1d140c22d93a2707e2f9fb329f.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 171, + 435, + 438, + 457 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 458, + 505, + 521 + ], + "lines": [ + { + "bbox": [ + 105, + 457, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 132, + 473 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 459, + 164, + 471 + ], + "score": 0.78, + "content": "\\mathbb { E } _ { M ^ { * } } \\left[ \\cdot \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 457, + 198, + 473 + ], + "score": 1.0, + "content": "denotes", + "type": "text" + }, + { + "bbox": [ + 198, + 460, + 258, + 472 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { x ^ { \\prime } \\sim M ^ { * } \\left( x ^ { \\prime } \\mid x , a \\right) }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 457, + 506, + 473 + ], + "score": 1.0, + "content": "[·]. The core problem is how to construct the data distribution", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 471, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 107, + 471, + 125, + 486 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 471, + 239, + 486 + ], + "score": 1.0, + "content": "of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 240, + 474, + 252, + 485 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 471, + 264, + 486 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 264, + 474, + 280, + 484 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 471, + 430, + 486 + ], + "score": 1.0, + "content": "as it is costly to get extra data from", + "type": "text" + }, + { + "bbox": [ + 431, + 474, + 447, + 484 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 471, + 506, + 486 + ], + "score": 1.0, + "content": "in real-world", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 104, + 484, + 297, + 497 + ], + "score": 1.0, + "content": "applications. Instead of deriving the optimal", + "type": "text" + }, + { + "bbox": [ + 297, + 485, + 309, + 496 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 484, + 506, + 497 + ], + "score": 1.0, + "content": ", our solution is to offline estimate the optimal", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 495, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 104, + 495, + 205, + 513 + ], + "score": 1.0, + "content": "adversarial distribution", + "type": "text" + }, + { + "bbox": [ + 205, + 496, + 224, + 510 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 495, + 292, + 513 + ], + "score": 1.0, + "content": "with respect to", + "type": "text" + }, + { + "bbox": [ + 292, + 498, + 304, + 508 + ], + "score": 0.84, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 495, + 506, + 513 + ], + "score": 1.0, + "content": ", then we can construct a surrogate objective to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 509, + 366, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 144, + 521 + ], + "score": 1.0, + "content": "optimize", + "type": "text" + }, + { + "bbox": [ + 144, + 509, + 156, + 519 + ], + "score": 0.75, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 509, + 345, + 521 + ], + "score": 1.0, + "content": "without directly querying the real environment", + "type": "text" + }, + { + "bbox": [ + 345, + 509, + 362, + 519 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 509, + 366, + 521 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28 + }, + { + "type": "title", + "bbox": [ + 107, + 523, + 414, + 534 + ], + "lines": [ + { + "bbox": [ + 106, + 522, + 414, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 414, + 536 + ], + "score": 1.0, + "content": "4.2.1 OPTIMAL ADVERSARIAL DATA DISTRIBUTION APPROXIMATION", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 537, + 506, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 182, + 550 + ], + "score": 1.0, + "content": "Ideally, given any", + "type": "text" + }, + { + "bbox": [ + 183, + 537, + 195, + 547 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 536, + 320, + 550 + ], + "score": 1.0, + "content": ", it is obvious that the optimal", + "type": "text" + }, + { + "bbox": [ + 320, + 537, + 328, + 549 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 536, + 420, + 550 + ], + "score": 1.0, + "content": "is the one that makes", + "type": "text" + }, + { + "bbox": [ + 420, + 535, + 462, + 550 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { \\ast } } ^ { \\beta } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 536, + 506, + 550 + ], + "score": 1.0, + "content": "assign all", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 548, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 106, + 548, + 505, + 560 + ], + "score": 1.0, + "content": "densities to the point that has the largest negative log-likelihood. However, searching for the maximum", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 560, + 504, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 492, + 571 + ], + "score": 1.0, + "content": "is impractical, especially in continuous space. To give a relaxed but tractable solution, we add an", + "type": "text" + }, + { + "bbox": [ + 492, + 561, + 504, + 570 + ], + "score": 0.88, + "content": "L _ { 2 }", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 570, + 276, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 276, + 583 + ], + "score": 1.0, + "content": "regularizer to the original objective Eq. 3:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 583, + 513, + 606 + ], + "lines": [ + { + "bbox": [ + 111, + 583, + 513, + 606 + ], + "spans": [ + { + "bbox": [ + 111, + 583, + 513, + 606 + ], + "score": 0.92, + "content": "\\operatorname* { m i n } _ { d \\in M } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M ) = \\operatorname* { m i n } _ { M \\in M } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { * } } ^ { \\mu } } \\Big [ \\omega ( x , a ) \\mathbb { E } _ { M ^ { * } } \\big [ - \\log M ( x ^ { \\prime } | x , a ) \\big ] \\Big ] - \\frac { \\alpha } { 2 } \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } .", + "type": "interline_equation", + "image_path": "92654ebd7b17c680d8289ce5867177b202ca333373c1c15e4f9dfc0f08704048.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 111, + 583, + 513, + 606 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 621, + 506, + 653 + ], + "lines": [ + { + "bbox": [ + 103, + 618, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 103, + 618, + 133, + 639 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 626, + 141, + 633 + ], + "score": 0.78, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 618, + 304, + 639 + ], + "score": 1.0, + "content": "denotes the regularization coefficient of", + "type": "text" + }, + { + "bbox": [ + 304, + 622, + 323, + 636 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 618, + 342, + 639 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 342, + 621, + 504, + 638 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } = \\int _ { \\mathcal { X } , \\mathcal { A } } ( \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ) ^ { 2 } \\mathrm { d } a \\mathrm { d } x . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 37 + }, + { + "bbox": [ + 101, + 630, + 401, + 658 + ], + "spans": [ + { + "bbox": [ + 101, + 630, + 307, + 658 + ], + "score": 1.0, + "content": "Then we can approximate the optimal distribution", + "type": "text" + }, + { + "bbox": [ + 308, + 637, + 326, + 653 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 630, + 401, + 658 + ], + "score": 1.0, + "content": "via Lemma. 4.3.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 104, + 655, + 504, + 677 + ], + "lines": [ + { + "bbox": [ + 104, + 654, + 503, + 671 + ], + "spans": [ + { + "bbox": [ + 104, + 654, + 209, + 671 + ], + "score": 1.0, + "content": "Lemma 4.3. Given any", + "type": "text" + }, + { + "bbox": [ + 209, + 656, + 221, + 667 + ], + "score": 0.75, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 654, + 234, + 671 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 235, + 654, + 283, + 669 + ], + "score": 0.93, + "content": "\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M )", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 654, + 492, + 671 + ], + "score": 1.0, + "content": ", the distribution of the ideal best-response policy", + "type": "text" + }, + { + "bbox": [ + 492, + 657, + 503, + 668 + ], + "score": 0.89, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 665, + 145, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 145, + 680 + ], + "score": 1.0, + "content": "satisfies:", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 676, + 425, + 698 + ], + "lines": [ + { + "bbox": [ + 185, + 676, + 425, + 698 + ], + "spans": [ + { + "bbox": [ + 185, + 676, + 425, + 698 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ,", + "type": "interline_equation", + "image_path": "37dc1f9700a687f33b539434a21e83ea1994226b49396acf0eefa63e62814c23.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 185, + 676, + 425, + 698 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 108, + 696, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 108, + 696, + 137, + 713 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 138, + 699, + 255, + 711 + ], + "score": 0.89, + "content": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 696, + 355, + 713 + ], + "score": 1.0, + "content": "is the Kullback-Leibler", + "type": "text" + }, + { + "bbox": [ + 355, + 700, + 374, + 710 + ], + "score": 0.68, + "content": "( K L )", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 696, + 460, + 713 + ], + "score": 1.0, + "content": "divergence between", + "type": "text" + }, + { + "bbox": [ + 460, + 699, + 505, + 711 + ], + "score": 0.91, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 124, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 711, + 164, + 722 + ], + "score": 0.66, + "content": "M ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 708, + 167, + 723 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 168, + 711, + 212, + 722 + ], + "score": 0.78, + "content": "H _ { M ^ { \\ast } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 708, + 301, + 723 + ], + "score": 1.0, + "content": "denotes the entropy of", + "type": "text" + }, + { + "bbox": [ + 301, + 710, + 345, + 722 + ], + "score": 0.93, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 708, + 366, + 723 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 366, + 711, + 382, + 721 + ], + "score": 0.85, + "content": "\\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "is the regularization coefficient", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 107, + 720, + 288, + 733 + ], + "spans": [ + { + "bbox": [ + 107, + 723, + 114, + 730 + ], + "score": 0.67, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 720, + 141, + 733 + ], + "score": 1.0, + "content": "in Eq.", + "type": "text" + }, + { + "bbox": [ + 142, + 722, + 148, + 730 + ], + "score": 0.28, + "content": "^ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 720, + 288, + 733 + ], + "score": 1.0, + "content": "and also as a normalizer of Eq. 5.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 506, + 163 + ], + "lines": [], + "index": 2.5, + "bbox_fs": [ + 104, + 80, + 507, + 164 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 168, + 505, + 223 + ], + "lines": [ + { + "bbox": [ + 106, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 106, + 168, + 130, + 180 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 131, + 169, + 138, + 180 + ], + "score": 0.83, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 168, + 505, + 180 + ], + "score": 1.0, + "content": "for counterfactual queries is unknown and can be varied in policy optimization, to reduce the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 178, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 105, + 178, + 505, + 191 + ], + "score": 1.0, + "content": "risks in counterfactual queries in this scenario, we introduce adversarial policies which can induce the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 190, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 506, + 202 + ], + "score": 1.0, + "content": "worst performance of the model predictions and propose to optimize WERM under the adversarial", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 201, + 506, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 506, + 213 + ], + "score": 1.0, + "content": "policies. In particular, we propose Adversarial Weighted empirical Risk Minimization (AWRM)", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 212, + 272, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 272, + 224 + ], + "score": 1.0, + "content": "based on Def. 4.1 to handle this problem.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 168, + 506, + 224 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 225, + 505, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 304, + 239 + ], + "score": 1.0, + "content": "Definition 4.2. Given the MDP transition function", + "type": "text" + }, + { + "bbox": [ + 305, + 226, + 320, + 236 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 225, + 506, + 239 + ], + "score": 1.0, + "content": ", the learning objective of adversarial Weighted", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 237, + 336, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 336, + 249 + ], + "score": 1.0, + "content": "empirical risk minimization through IPS is formulated as", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 225, + 506, + 249 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 134, + 250, + 475, + 302 + ], + "lines": [ + { + "bbox": [ + 134, + 250, + 475, + 302 + ], + "spans": [ + { + "bbox": [ + 134, + 250, + 475, + 302 + ], + "score": 0.89, + "content": "\\begin{array} { r l } & { \\hat { \\boldsymbol M } ^ { * } = \\underset { M \\in \\mathcal { M } } { \\mathrm { \\operatorname* { m i n } } } \\underset { \\beta \\in \\Pi } { \\mathrm { m a x } } \\boldsymbol L ( \\rho _ { M ^ { * } } ^ { \\beta } , M ) = \\underset { M \\in \\mathcal { M } } { \\mathrm { m i n } } \\underset { \\beta \\in \\Pi } { \\mathrm { m a x } } \\mathbb { E } _ { \\boldsymbol x , \\boldsymbol a , \\boldsymbol x ^ { \\prime } \\sim \\boldsymbol \\rho _ { M ^ { * } } ^ { \\mu } } [ \\omega ( \\boldsymbol x , \\boldsymbol a | \\rho _ { M ^ { * } } ^ { \\beta } ) \\ell _ { M } ( \\boldsymbol x , \\boldsymbol a , \\boldsymbol x ^ { \\prime } ) ] , } \\\\ & { } \\\\ & { s . t . \\quad \\omega ( \\boldsymbol x , \\boldsymbol a | \\rho _ { M ^ { * } } ^ { \\beta } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( \\boldsymbol x , \\boldsymbol a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( \\boldsymbol x , \\boldsymbol a ) } , } \\end{array}", + "type": "interline_equation", + "image_path": "39944d2b5a594f42b7761eb3e27008f21b546a70714b76f8a617a7dfda623ea0.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 134, + 250, + 475, + 267.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 134, + 267.3333333333333, + 475, + 284.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 134, + 284.66666666666663, + 475, + 301.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 305, + 504, + 333 + ], + "lines": [ + { + "bbox": [ + 105, + 303, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 223, + 322 + ], + "score": 1.0, + "content": "where the re-weighting term", + "type": "text" + }, + { + "bbox": [ + 223, + 304, + 274, + 319 + ], + "score": 0.93, + "content": "\\omega ( x , a | \\rho _ { M ^ { * } } ^ { \\beta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 303, + 432, + 322 + ], + "score": 1.0, + "content": "is conditioned on the data distribution", + "type": "text" + }, + { + "bbox": [ + 433, + 304, + 451, + 319 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 303, + 506, + 322 + ], + "score": 1.0, + "content": "of the adver-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 318, + 430, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 159, + 335 + ], + "score": 1.0, + "content": "sarial policy", + "type": "text" + }, + { + "bbox": [ + 159, + 321, + 167, + 331 + ], + "score": 0.68, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 318, + 298, + 335 + ], + "score": 1.0, + "content": ". In the following, we will ignore", + "type": "text" + }, + { + "bbox": [ + 298, + 318, + 317, + 333 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 318, + 352, + 335 + ], + "score": 1.0, + "content": "and use", + "type": "text" + }, + { + "bbox": [ + 352, + 320, + 382, + 332 + ], + "score": 0.92, + "content": "\\omega ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 318, + 430, + 335 + ], + "score": 1.0, + "content": "for brevity.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 303, + 506, + 335 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 339, + 505, + 386 + ], + "lines": [ + { + "bbox": [ + 106, + 339, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 259, + 352 + ], + "score": 1.0, + "content": "According to the definition of MDP,", + "type": "text" + }, + { + "bbox": [ + 259, + 339, + 348, + 352 + ], + "score": 0.93, + "content": "\\omega ( x , a , x ^ { \\prime } ) = \\omega ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 339, + 505, + 352 + ], + "score": 1.0, + "content": "since the transition probability in the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 350, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 506, + 363 + ], + "score": 1.0, + "content": "ratio can be canceled. Eq. 3 minimizes the maximum model loss under all counterfactual data", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 362, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 160, + 378 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 161, + 362, + 248, + 376 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { \\ast } } ^ { \\beta } ( x , a , x ^ { \\prime } ) , \\beta \\in \\Pi", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 362, + 506, + 378 + ], + "score": 1.0, + "content": "to guarantee the generalization ability for counterfactual data", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 375, + 208, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 195, + 386 + ], + "score": 1.0, + "content": "queried by policies in", + "type": "text" + }, + { + "bbox": [ + 195, + 376, + 204, + 384 + ], + "score": 0.72, + "content": "\\Pi", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 375, + 208, + 386 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 339, + 506, + 386 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 393, + 266, + 404 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 268, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 268, + 406 + ], + "score": 1.0, + "content": "4.2 TRACTABLE AWRM SOLUTION", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 410, + 504, + 433 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "In this section, we propose a tractable solution to optimize Eq. 3. The full derivations can be found in", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 422, + 501, + 434 + ], + "spans": [ + { + "bbox": [ + 107, + 422, + 206, + 434 + ], + "score": 1.0, + "content": "the Appx. A. We choose", + "type": "text" + }, + { + "bbox": [ + 206, + 423, + 220, + 433 + ], + "score": 0.9, + "content": "\\ell _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 422, + 501, + 434 + ], + "score": 1.0, + "content": "to be the negative log-likelihood loss and derive the solution to Eq. 3:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 410, + 505, + 434 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 171, + 435, + 438, + 457 + ], + "lines": [ + { + "bbox": [ + 171, + 435, + 438, + 457 + ], + "spans": [ + { + "bbox": [ + 171, + 435, + 438, + 457 + ], + "score": 0.91, + "content": "L ( \\rho _ { M ^ { \\ast } } ^ { \\beta } , M ) = \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { \\ast } } ^ { \\mu } } \\left[ \\omega ( x , a | \\rho _ { M ^ { \\ast } } ^ { \\beta } ) \\mathbb { E } _ { M ^ { \\ast } } \\left( - \\log M ( x ^ { \\prime } | x , a ) \\right) \\right] ,", + "type": "interline_equation", + "image_path": "072350e4aea7ec2b69e997e13744c49a14155f1d140c22d93a2707e2f9fb329f.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 171, + 435, + 438, + 457 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 458, + 505, + 521 + ], + "lines": [ + { + "bbox": [ + 105, + 457, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 132, + 473 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 459, + 164, + 471 + ], + "score": 0.78, + "content": "\\mathbb { E } _ { M ^ { * } } \\left[ \\cdot \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 457, + 198, + 473 + ], + "score": 1.0, + "content": "denotes", + "type": "text" + }, + { + "bbox": [ + 198, + 460, + 258, + 472 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { x ^ { \\prime } \\sim M ^ { * } \\left( x ^ { \\prime } \\mid x , a \\right) }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 457, + 506, + 473 + ], + "score": 1.0, + "content": "[·]. The core problem is how to construct the data distribution", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 471, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 107, + 471, + 125, + 486 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 471, + 239, + 486 + ], + "score": 1.0, + "content": "of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 240, + 474, + 252, + 485 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 471, + 264, + 486 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 264, + 474, + 280, + 484 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 471, + 430, + 486 + ], + "score": 1.0, + "content": "as it is costly to get extra data from", + "type": "text" + }, + { + "bbox": [ + 431, + 474, + 447, + 484 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 471, + 506, + 486 + ], + "score": 1.0, + "content": "in real-world", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 104, + 484, + 297, + 497 + ], + "score": 1.0, + "content": "applications. Instead of deriving the optimal", + "type": "text" + }, + { + "bbox": [ + 297, + 485, + 309, + 496 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 484, + 506, + 497 + ], + "score": 1.0, + "content": ", our solution is to offline estimate the optimal", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 495, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 104, + 495, + 205, + 513 + ], + "score": 1.0, + "content": "adversarial distribution", + "type": "text" + }, + { + "bbox": [ + 205, + 496, + 224, + 510 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 495, + 292, + 513 + ], + "score": 1.0, + "content": "with respect to", + "type": "text" + }, + { + "bbox": [ + 292, + 498, + 304, + 508 + ], + "score": 0.84, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 495, + 506, + 513 + ], + "score": 1.0, + "content": ", then we can construct a surrogate objective to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 509, + 366, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 144, + 521 + ], + "score": 1.0, + "content": "optimize", + "type": "text" + }, + { + "bbox": [ + 144, + 509, + 156, + 519 + ], + "score": 0.75, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 509, + 345, + 521 + ], + "score": 1.0, + "content": "without directly querying the real environment", + "type": "text" + }, + { + "bbox": [ + 345, + 509, + 362, + 519 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 509, + 366, + 521 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28, + "bbox_fs": [ + 104, + 457, + 506, + 521 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 523, + 414, + 534 + ], + "lines": [ + { + "bbox": [ + 106, + 522, + 414, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 414, + 536 + ], + "score": 1.0, + "content": "4.2.1 OPTIMAL ADVERSARIAL DATA DISTRIBUTION APPROXIMATION", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 537, + 506, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 182, + 550 + ], + "score": 1.0, + "content": "Ideally, given any", + "type": "text" + }, + { + "bbox": [ + 183, + 537, + 195, + 547 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 536, + 320, + 550 + ], + "score": 1.0, + "content": ", it is obvious that the optimal", + "type": "text" + }, + { + "bbox": [ + 320, + 537, + 328, + 549 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 536, + 420, + 550 + ], + "score": 1.0, + "content": "is the one that makes", + "type": "text" + }, + { + "bbox": [ + 420, + 535, + 462, + 550 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { \\ast } } ^ { \\beta } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 536, + 506, + 550 + ], + "score": 1.0, + "content": "assign all", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 548, + 505, + 560 + ], + "spans": [ + { + "bbox": [ + 106, + 548, + 505, + 560 + ], + "score": 1.0, + "content": "densities to the point that has the largest negative log-likelihood. However, searching for the maximum", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 560, + 504, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 492, + 571 + ], + "score": 1.0, + "content": "is impractical, especially in continuous space. To give a relaxed but tractable solution, we add an", + "type": "text" + }, + { + "bbox": [ + 492, + 561, + 504, + 570 + ], + "score": 0.88, + "content": "L _ { 2 }", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 570, + 276, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 276, + 583 + ], + "score": 1.0, + "content": "regularizer to the original objective Eq. 3:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 535, + 506, + 583 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 583, + 513, + 606 + ], + "lines": [ + { + "bbox": [ + 111, + 583, + 513, + 606 + ], + "spans": [ + { + "bbox": [ + 111, + 583, + 513, + 606 + ], + "score": 0.92, + "content": "\\operatorname* { m i n } _ { d \\in M } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M ) = \\operatorname* { m i n } _ { M \\in M } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { * } } ^ { \\mu } } \\Big [ \\omega ( x , a ) \\mathbb { E } _ { M ^ { * } } \\big [ - \\log M ( x ^ { \\prime } | x , a ) \\big ] \\Big ] - \\frac { \\alpha } { 2 } \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } .", + "type": "interline_equation", + "image_path": "92654ebd7b17c680d8289ce5867177b202ca333373c1c15e4f9dfc0f08704048.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 111, + 583, + 513, + 606 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 621, + 506, + 653 + ], + "lines": [ + { + "bbox": [ + 103, + 618, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 103, + 618, + 133, + 639 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 626, + 141, + 633 + ], + "score": 0.78, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 618, + 304, + 639 + ], + "score": 1.0, + "content": "denotes the regularization coefficient of", + "type": "text" + }, + { + "bbox": [ + 304, + 622, + 323, + 636 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 618, + 342, + 639 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 342, + 621, + 504, + 638 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } = \\int _ { \\mathcal { X } , \\mathcal { A } } ( \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ) ^ { 2 } \\mathrm { d } a \\mathrm { d } x . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 37 + }, + { + "bbox": [ + 101, + 630, + 401, + 658 + ], + "spans": [ + { + "bbox": [ + 101, + 630, + 307, + 658 + ], + "score": 1.0, + "content": "Then we can approximate the optimal distribution", + "type": "text" + }, + { + "bbox": [ + 308, + 637, + 326, + 653 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 630, + 401, + 658 + ], + "score": 1.0, + "content": "via Lemma. 4.3.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 101, + 618, + 504, + 658 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 655, + 504, + 677 + ], + "lines": [ + { + "bbox": [ + 104, + 654, + 503, + 671 + ], + "spans": [ + { + "bbox": [ + 104, + 654, + 209, + 671 + ], + "score": 1.0, + "content": "Lemma 4.3. Given any", + "type": "text" + }, + { + "bbox": [ + 209, + 656, + 221, + 667 + ], + "score": 0.75, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 654, + 234, + 671 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 235, + 654, + 283, + 669 + ], + "score": 0.93, + "content": "\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M )", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 654, + 492, + 671 + ], + "score": 1.0, + "content": ", the distribution of the ideal best-response policy", + "type": "text" + }, + { + "bbox": [ + 492, + 657, + 503, + 668 + ], + "score": 0.89, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 665, + 145, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 145, + 680 + ], + "score": 1.0, + "content": "satisfies:", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 104, + 654, + 503, + 680 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 676, + 425, + 698 + ], + "lines": [ + { + "bbox": [ + 185, + 676, + 425, + 698 + ], + "spans": [ + { + "bbox": [ + 185, + 676, + 425, + 698 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ,", + "type": "interline_equation", + "image_path": "37dc1f9700a687f33b539434a21e83ea1994226b49396acf0eefa63e62814c23.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 185, + 676, + 425, + 698 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 108, + 696, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 108, + 696, + 137, + 713 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 138, + 699, + 255, + 711 + ], + "score": 0.89, + "content": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 696, + 355, + 713 + ], + "score": 1.0, + "content": "is the Kullback-Leibler", + "type": "text" + }, + { + "bbox": [ + 355, + 700, + 374, + 710 + ], + "score": 0.68, + "content": "( K L )", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 696, + 460, + 713 + ], + "score": 1.0, + "content": "divergence between", + "type": "text" + }, + { + "bbox": [ + 460, + 699, + 505, + 711 + ], + "score": 0.91, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 124, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 711, + 164, + 722 + ], + "score": 0.66, + "content": "M ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 708, + 167, + 723 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 168, + 711, + 212, + 722 + ], + "score": 0.78, + "content": "H _ { M ^ { \\ast } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 708, + 301, + 723 + ], + "score": 1.0, + "content": "denotes the entropy of", + "type": "text" + }, + { + "bbox": [ + 301, + 710, + 345, + 722 + ], + "score": 0.93, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 708, + 366, + 723 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 366, + 711, + 382, + 721 + ], + "score": 0.85, + "content": "\\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "is the regularization coefficient", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 107, + 720, + 288, + 733 + ], + "spans": [ + { + "bbox": [ + 107, + 723, + 114, + 730 + ], + "score": 0.67, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 720, + 141, + 733 + ], + "score": 1.0, + "content": "in Eq.", + "type": "text" + }, + { + "bbox": [ + 142, + 722, + 148, + 730 + ], + "score": 0.28, + "content": "^ { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 720, + 288, + 733 + ], + "score": 1.0, + "content": "and also as a normalizer of Eq. 5.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 696, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 506, + 202 + ], + "lines": [ + { + "bbox": [ + 104, + 80, + 507, + 98 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 153, + 98 + ], + "score": 1.0, + "content": "Intuitively,", + "type": "text" + }, + { + "bbox": [ + 153, + 80, + 172, + 96 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 81, + 507, + 98 + ], + "score": 1.0, + "content": "has larger densities on the data where the divergence between the approximation", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 106, + 95, + 229, + 107 + ], + "score": 1.0, + "content": "model and the real model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 229, + 94, + 351, + 107 + ], + "score": 0.88, + "content": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 95, + 506, + 107 + ], + "score": 1.0, + "content": "is larger or the stochasticity of the real", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 153, + 117 + ], + "score": 1.0, + "content": "model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 154, + 105, + 176, + 115 + ], + "score": 0.88, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 104, + 344, + 117 + ], + "score": 1.0, + "content": ") is larger. However, the integral process of", + "type": "text" + }, + { + "bbox": [ + 345, + 105, + 367, + 115 + ], + "score": 0.9, + "content": "D _ { K L }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "in Eq. 5 is intractable in the offline", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 393, + 127 + ], + "score": 1.0, + "content": "setting as it explicitly requires the conditional probability function of", + "type": "text" + }, + { + "bbox": [ + 394, + 115, + 410, + 124 + ], + "score": 0.89, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 114, + 506, + 127 + ], + "score": 1.0, + "content": ". Our solution to solve", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 123, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 274, + 138 + ], + "score": 1.0, + "content": "the problem is utilizing the offline dataset", + "type": "text" + }, + { + "bbox": [ + 274, + 125, + 296, + 136 + ], + "score": 0.91, + "content": "\\mathcal { D } _ { \\mathrm { r e a l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 123, + 432, + 138 + ], + "score": 1.0, + "content": "as the empirical joint distribution", + "type": "text" + }, + { + "bbox": [ + 432, + 124, + 487, + 137 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 123, + 506, + 138 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 135, + 506, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 146 + ], + "score": 1.0, + "content": "adopting practical techniques for distance estimation on two joint distributions, like GAN (Goodfellow", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 144, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 506, + 156 + ], + "score": 1.0, + "content": "et al., 2014; Nowozin et al., 2016), to approximate Eq. 5. To adopt that solution, we should first", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "transform Eq. 5 into a form under joint distributions. Without loss of generality, we introduce an", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 164, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 185, + 176 + ], + "score": 1.0, + "content": "intermediary policy", + "type": "text" + }, + { + "bbox": [ + 186, + 169, + 192, + 174 + ], + "score": 0.7, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 164, + 232, + 176 + ], + "score": 1.0, + "content": ", of which", + "type": "text" + }, + { + "bbox": [ + 232, + 167, + 240, + 176 + ], + "score": 0.82, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 164, + 447, + 176 + ], + "score": 1.0, + "content": "can be regarded as a specific instance. Then we have", + "type": "text" + }, + { + "bbox": [ + 447, + 164, + 505, + 176 + ], + "score": 0.9, + "content": "M ( x ^ { \\prime } | x , a ) =", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 174, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 107, + 175, + 200, + 187 + ], + "score": 0.86, + "content": "\\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) / \\rho _ { M } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 174, + 233, + 188 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 233, + 176, + 245, + 186 + ], + "score": 0.82, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 174, + 255, + 188 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 255, + 175, + 312, + 188 + ], + "score": 0.93, + "content": "\\rho _ { M } ^ { \\kappa } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 174, + 360, + 188 + ], + "score": 1.0, + "content": ". Assuming", + "type": "text" + }, + { + "bbox": [ + 361, + 176, + 495, + 187 + ], + "score": 0.89, + "content": "\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 174, + 506, + 188 + ], + "score": 1.0, + "content": "if", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 184, + 452, + 205 + ], + "spans": [ + { + "bbox": [ + 107, + 187, + 167, + 202 + ], + "score": 0.88, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 184, + 260, + 205 + ], + "score": 1.0, + "content": ", which will hold when", + "type": "text" + }, + { + "bbox": [ + 261, + 192, + 267, + 200 + ], + "score": 0.74, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 184, + 324, + 205 + ], + "score": 1.0, + "content": "overlaps with", + "type": "text" + }, + { + "bbox": [ + 325, + 191, + 332, + 201 + ], + "score": 0.79, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 184, + 452, + 205 + ], + "score": 1.0, + "content": ", then Eq. 5 can transform to:", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 119, + 206, + 492, + 237 + ], + "lines": [ + { + "bbox": [ + 119, + 206, + 492, + 237 + ], + "spans": [ + { + "bbox": [ + 119, + 206, + 492, + 237 + ], + "score": 0.9, + "content": "\\displaystyle \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\Bigg ( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } { \\rho _ { M } ^ { \\kappa } ( x , a ) } - H _ { M ^ { * } } ( x , a ) \\right) \\Bigg ) ,", + "type": "interline_equation", + "image_path": "78af9ac575fb1379ea94dfaa342e4030712e344f1b8af727eb02a909a481a75c.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 119, + 206, + 492, + 216.33333333333334 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 119, + 216.33333333333334, + 492, + 226.66666666666669 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 119, + 226.66666666666669, + 492, + 237.00000000000003 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 243, + 506, + 285 + ], + "lines": [ + { + "bbox": [ + 106, + 241, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 137, + 262 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 137, + 244, + 242, + 257 + ], + "score": 0.94, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 241, + 346, + 262 + ], + "score": 1.0, + "content": ". We notice that the form", + "type": "text" + }, + { + "bbox": [ + 346, + 242, + 399, + 261 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } } { \\rho _ { M } ^ { \\kappa } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 241, + 506, + 259 + ], + "score": 1.0, + "content": "ρκM∗ρκ is the integrated function", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 258, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 148, + 272 + ], + "score": 1.0, + "content": "in reverse", + "type": "text" + }, + { + "bbox": [ + 148, + 259, + 163, + 270 + ], + "score": 0.32, + "content": "\\mathrm { K L }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 258, + 308, + 272 + ], + "score": 1.0, + "content": "divergence, which is an instance of", + "type": "text" + }, + { + "bbox": [ + 308, + 259, + 315, + 271 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 258, + 363, + 272 + ], + "score": 1.0, + "content": "function in", + "type": "text" + }, + { + "bbox": [ + 363, + 260, + 370, + 271 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 258, + 506, + 272 + ], + "score": 1.0, + "content": "-divergence (Ali & Silvey, 1966).", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 268, + 447, + 288 + ], + "spans": [ + { + "bbox": [ + 104, + 268, + 209, + 288 + ], + "score": 1.0, + "content": "Replacing that form with", + "type": "text" + }, + { + "bbox": [ + 209, + 273, + 216, + 284 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 268, + 422, + 288 + ], + "score": 1.0, + "content": "function, we obtain a generalized representation of", + "type": "text" + }, + { + "bbox": [ + 423, + 270, + 441, + 285 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 268, + 447, + 288 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 289, + 514, + 320 + ], + "lines": [ + { + "bbox": [ + 111, + 289, + 514, + 320 + ], + "spans": [ + { + "bbox": [ + 111, + 289, + 514, + 320 + ], + "score": 0.89, + "content": "\\tilde { \\bar { \\nu } } _ { M ^ { * } } ^ { * } : = \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\left( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) - H _ { M ^ { * } } ( x , a ) \\right) \\right.", + "type": "interline_equation", + "image_path": "43d2dd2f890e8f95bed4967ff118330b972404d5f4c844faa96a213360d47f59.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 111, + 289, + 514, + 299.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 111, + 299.3333333333333, + 514, + 309.66666666666663 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 111, + 309.66666666666663, + 514, + 319.99999999999994 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 513, + 164, + 584, + 175 + ], + "lines": [ + { + "bbox": [ + 512, + 163, + 585, + 176 + ], + "spans": [ + { + "bbox": [ + 512, + 163, + 585, + 176 + ], + "score": 1.0, + "content": "to Reviewer bcJS", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 336, + 506, + 375 + ], + "lines": [ + { + "bbox": [ + 107, + 334, + 507, + 355 + ], + "spans": [ + { + "bbox": [ + 107, + 334, + 135, + 355 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 339, + 187, + 351 + ], + "score": 0.92, + "content": "f : \\mathbb { R } _ { + } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 334, + 407, + 355 + ], + "score": 1.0, + "content": "is a convex and lower semi-continuous (l.s.c.) function.", + "type": "text" + }, + { + "bbox": [ + 408, + 336, + 426, + 352 + ], + "score": 0.93, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 334, + 507, + 355 + ], + "score": 1.0, + "content": "gives a generalized", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "score": 1.0, + "content": "representation of the optimal adversarial data distribution to maximize the error of the model. Based", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 361, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 104, + 361, + 415, + 379 + ], + "score": 1.0, + "content": "on Eq. 6, we have a surrogate objective of AWRM which can avoid querying", + "type": "text" + }, + { + "bbox": [ + 415, + 363, + 431, + 374 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 361, + 482, + 379 + ], + "score": 1.0, + "content": "to construct", + "type": "text" + }, + { + "bbox": [ + 482, + 361, + 501, + 376 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 361, + 506, + 379 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 378, + 506, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 378, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 182, + 397 + ], + "score": 1.0, + "content": "Theorem 4.4. Let", + "type": "text" + }, + { + "bbox": [ + 182, + 378, + 201, + 394 + ], + "score": 0.91, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 379, + 406, + 397 + ], + "score": 1.0, + "content": "as the data distribution of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 406, + 381, + 417, + 393 + ], + "score": 0.88, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 379, + 506, + 397 + ], + "score": 1.0, + "content": "in Eq. 4 under model", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 394, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 107, + 395, + 122, + 407 + ], + "score": 0.67, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 394, + 199, + 408 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 200, + 396, + 205, + 405 + ], + "score": 0.69, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 394, + 334, + 408 + ], + "score": 1.0, + "content": ", then we can find the optimal", + "type": "text" + }, + { + "bbox": [ + 335, + 396, + 345, + 406 + ], + "score": 0.83, + "content": "\\theta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 394, + 359, + 408 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 359, + 394, + 471, + 408 + ], + "score": 0.85, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 394, + 506, + 408 + ], + "score": 1.0, + "content": "(Eq. 4)", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 406, + 507, + 425 + ], + "spans": [ + { + "bbox": [ + 104, + 406, + 264, + 425 + ], + "score": 1.0, + "content": "via iteratively optimizing the objective", + "type": "text" + }, + { + "bbox": [ + 264, + 407, + 371, + 422 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\theta _ { t + 1 } = \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 406, + 402, + 425 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 402, + 407, + 420, + 422 + ], + "score": 0.91, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 406, + 507, + 425 + ], + "score": 1.0, + "content": "is approximated via", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 207, + 433 + ], + "score": 1.0, + "content": "the last-iteration model", + "type": "text" + }, + { + "bbox": [ + 207, + 420, + 225, + 432 + ], + "score": 0.91, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 420, + 506, + 433 + ], + "score": 1.0, + "content": ". Based on Corollary A.7, we derive an upper bound objective for", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 431, + 187, + 448 + ], + "spans": [ + { + "bbox": [ + 107, + 431, + 181, + 447 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 432, + 187, + 448 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 451, + 506, + 496 + ], + "lines": [ + { + "bbox": [ + 111, + 451, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 111, + 451, + 506, + 496 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\iota _ { t + 1 } = \\underset { \\theta } { \\operatorname* { m i n } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\frac { - 1 } { \\alpha _ { 0 } ( x , a ) } \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\left( f \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) + H _ { M ^ { * } } ( x , a ) \\right) \\right] , } \\\\ { W ( x , a , x ^ { \\prime } ) } \\end{array}", + "type": "interline_equation", + "image_path": "300a71d94016b5fd8af5a721594f0ad5429351b188668424c87f05a5a0abc29e.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 111, + 451, + 506, + 466.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 111, + 466.0, + 506, + 481.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 111, + 481.0, + 506, + 496.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 500, + 505, + 527 + ], + "lines": [ + { + "bbox": [ + 108, + 497, + 507, + 515 + ], + "spans": [ + { + "bbox": [ + 108, + 497, + 137, + 515 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 137, + 501, + 160, + 514 + ], + "score": 0.84, + "content": "\\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 497, + 208, + 515 + ], + "score": 1.0, + "content": "[·] denotes", + "type": "text" + }, + { + "bbox": [ + 208, + 501, + 257, + 514 + ], + "score": 0.86, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\kappa } }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 497, + 273, + 515 + ], + "score": 1.0, + "content": "[·],", + "type": "text" + }, + { + "bbox": [ + 273, + 501, + 281, + 513 + ], + "score": 0.81, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 497, + 399, + 515 + ], + "score": 1.0, + "content": "is a l.s.c function satisfying", + "type": "text" + }, + { + "bbox": [ + 399, + 500, + 482, + 513 + ], + "score": 0.94, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 497, + 507, + 515 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 107, + 512, + 221, + 527 + ], + "spans": [ + { + "bbox": [ + 107, + 514, + 218, + 527 + ], + "score": 0.91, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 512, + 221, + 527 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 533, + 505, + 581 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 302, + 547 + ], + "score": 1.0, + "content": "Thm. 4.4 approximately achieve AWRM by using", + "type": "text" + }, + { + "bbox": [ + 303, + 536, + 310, + 544 + ], + "score": 0.75, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 533, + 445, + 547 + ], + "score": 1.0, + "content": "and a pseudo-reweighting module", + "type": "text" + }, + { + "bbox": [ + 445, + 534, + 457, + 544 + ], + "score": 0.46, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 533, + 461, + 547 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 462, + 534, + 473, + 544 + ], + "score": 0.31, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 533, + 505, + 547 + ], + "score": 1.0, + "content": "assigns", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 544, + 506, + 560 + ], + "spans": [ + { + "bbox": [ + 104, + 544, + 428, + 560 + ], + "score": 1.0, + "content": "learning propensities for data points with larger differences between distributions", + "type": "text" + }, + { + "bbox": [ + 428, + 546, + 449, + 560 + ], + "score": 0.92, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 544, + 468, + 560 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 468, + 546, + 487, + 558 + ], + "score": 0.89, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 544, + 506, + 560 + ], + "score": 1.0, + "content": ". By", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "adjusting the weights, the learning process will exploit subtle errors in any data point, whatever how", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 569, + 492, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 492, + 581 + ], + "score": 1.0, + "content": "many proportions it contributes, to correct potential generalization errors on counterfactual data.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "title", + "bbox": [ + 107, + 591, + 295, + 604 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 297, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 297, + 604 + ], + "score": 1.0, + "content": "4.2.2 TRACTABLE SOLUTION TO THM. 4.4", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 609, + 506, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 608, + 507, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 195, + 634 + ], + "score": 1.0, + "content": "In Thm. 4.4, the term", + "type": "text" + }, + { + "bbox": [ + 195, + 610, + 320, + 631 + ], + "score": 0.94, + "content": "\\begin{array} { r } { f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\big ) - f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\big ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 615, + 507, + 628 + ], + "score": 1.0, + "content": "is still intractable. To solve the problem, first,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 630, + 487, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 284, + 642 + ], + "score": 1.0, + "content": "we resort to the first-order approximation of", + "type": "text" + }, + { + "bbox": [ + 285, + 632, + 291, + 642 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 630, + 345, + 642 + ], + "score": 1.0, + "content": ". Given some", + "type": "text" + }, + { + "bbox": [ + 346, + 630, + 448, + 642 + ], + "score": 0.91, + "content": "u \\in ( 1 - \\xi , 1 + \\xi ) , \\xi > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 630, + 487, + 642 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.0 + }, + { + "type": "interline_equation", + "bbox": [ + 244, + 646, + 366, + 660 + ], + "lines": [ + { + "bbox": [ + 244, + 646, + 366, + 660 + ], + "spans": [ + { + "bbox": [ + 244, + 646, + 366, + 660 + ], + "score": 0.92, + "content": "f ( u ) \\approx f ( 1 ) + f ^ { \\prime } ( u ) ( u - 1 ) ,", + "type": "interline_equation", + "image_path": "2b517abc4c2eca59111a5c29db0c8e310899041f086df0a62064d90652b53b69.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 244, + 646, + 366, + 660 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 506, + 733 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 132, + 678 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 666, + 142, + 678 + ], + "score": 0.88, + "content": "f ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 665, + 259, + 678 + ], + "score": 1.0, + "content": "is the first-order derivative of", + "type": "text" + }, + { + "bbox": [ + 259, + 667, + 266, + 677 + ], + "score": 0.81, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 665, + 416, + 678 + ], + "score": 1.0, + "content": ". By Taylor’s formula and the fact that", + "type": "text" + }, + { + "bbox": [ + 417, + 666, + 440, + 678 + ], + "score": 0.93, + "content": "f ^ { \\prime } ( u )", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "of the generator", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 144, + 690 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 144, + 678, + 151, + 689 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 677, + 215, + 690 + ], + "score": 1.0, + "content": "is bounded in", + "type": "text" + }, + { + "bbox": [ + 216, + 678, + 275, + 690 + ], + "score": 0.89, + "content": "( 1 - \\xi , 1 + \\xi )", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 677, + 454, + 690 + ], + "score": 1.0, + "content": ", the approximation error is no more than", + "type": "text" + }, + { + "bbox": [ + 455, + 677, + 480, + 690 + ], + "score": 0.9, + "content": "\\mathcal { O } ( \\xi ^ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 677, + 506, + 690 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 685, + 506, + 709 + ], + "spans": [ + { + "bbox": [ + 107, + 689, + 144, + 706 + ], + "score": 0.93, + "content": "\\begin{array} { r } { u = \\frac { p ( x ) } { q ( x ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 685, + 229, + 709 + ], + "score": 1.0, + "content": "in Eq. 7, the pattern", + "type": "text" + }, + { + "bbox": [ + 230, + 690, + 261, + 706 + ], + "score": 0.94, + "content": "f ( { \\frac { p ( x ) } { q ( x ) } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 685, + 394, + 709 + ], + "score": 1.0, + "content": "in Thm. 4.4 can be converted to", + "type": "text" + }, + { + "bbox": [ + 395, + 689, + 503, + 706 + ], + "score": 0.93, + "content": "\\begin{array} { r } { f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) ( \\frac { p ( x ) } { q ( x ) } - 1 ) + f ( 1 ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 685, + 506, + 709 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 108, + 703, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 108, + 705, + 125, + 722 + ], + "score": 0.91, + "content": "\\frac { p ( x ) } { q ( x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 703, + 349, + 723 + ], + "score": 1.0, + "content": "can be approximated by sampling from datasets and", + "type": "text" + }, + { + "bbox": [ + 349, + 706, + 384, + 722 + ], + "score": 0.93, + "content": "f ^ { \\prime } \\big ( \\frac { p ( x ) } { q ( x ) } \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 703, + 505, + 723 + ], + "score": 1.0, + "content": "can be approximated by the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 272, + 733 + ], + "score": 1.0, + "content": "corresponding variational representation", + "type": "text" + }, + { + "bbox": [ + 272, + 721, + 288, + 733 + ], + "score": 0.87, + "content": "T _ { \\varphi ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "according to Lemma A.9 (Nowozin et al., 2016) (see", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 506, + 202 + ], + "lines": [ + { + "bbox": [ + 104, + 80, + 507, + 98 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 153, + 98 + ], + "score": 1.0, + "content": "Intuitively,", + "type": "text" + }, + { + "bbox": [ + 153, + 80, + 172, + 96 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 81, + 507, + 98 + ], + "score": 1.0, + "content": "has larger densities on the data where the divergence between the approximation", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 106, + 95, + 229, + 107 + ], + "score": 1.0, + "content": "model and the real model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 229, + 94, + 351, + 107 + ], + "score": 0.88, + "content": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 95, + 506, + 107 + ], + "score": 1.0, + "content": "is larger or the stochasticity of the real", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 153, + 117 + ], + "score": 1.0, + "content": "model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 154, + 105, + 176, + 115 + ], + "score": 0.88, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 104, + 344, + 117 + ], + "score": 1.0, + "content": ") is larger. However, the integral process of", + "type": "text" + }, + { + "bbox": [ + 345, + 105, + 367, + 115 + ], + "score": 0.9, + "content": "D _ { K L }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "in Eq. 5 is intractable in the offline", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 393, + 127 + ], + "score": 1.0, + "content": "setting as it explicitly requires the conditional probability function of", + "type": "text" + }, + { + "bbox": [ + 394, + 115, + 410, + 124 + ], + "score": 0.89, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 114, + 506, + 127 + ], + "score": 1.0, + "content": ". Our solution to solve", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 123, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 123, + 274, + 138 + ], + "score": 1.0, + "content": "the problem is utilizing the offline dataset", + "type": "text" + }, + { + "bbox": [ + 274, + 125, + 296, + 136 + ], + "score": 0.91, + "content": "\\mathcal { D } _ { \\mathrm { r e a l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 123, + 432, + 138 + ], + "score": 1.0, + "content": "as the empirical joint distribution", + "type": "text" + }, + { + "bbox": [ + 432, + 124, + 487, + 137 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 123, + 506, + 138 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 135, + 506, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 146 + ], + "score": 1.0, + "content": "adopting practical techniques for distance estimation on two joint distributions, like GAN (Goodfellow", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 144, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 506, + 156 + ], + "score": 1.0, + "content": "et al., 2014; Nowozin et al., 2016), to approximate Eq. 5. To adopt that solution, we should first", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "transform Eq. 5 into a form under joint distributions. Without loss of generality, we introduce an", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 164, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 185, + 176 + ], + "score": 1.0, + "content": "intermediary policy", + "type": "text" + }, + { + "bbox": [ + 186, + 169, + 192, + 174 + ], + "score": 0.7, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 164, + 232, + 176 + ], + "score": 1.0, + "content": ", of which", + "type": "text" + }, + { + "bbox": [ + 232, + 167, + 240, + 176 + ], + "score": 0.82, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 164, + 447, + 176 + ], + "score": 1.0, + "content": "can be regarded as a specific instance. Then we have", + "type": "text" + }, + { + "bbox": [ + 447, + 164, + 505, + 176 + ], + "score": 0.9, + "content": "M ( x ^ { \\prime } | x , a ) =", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 174, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 107, + 175, + 200, + 187 + ], + "score": 0.86, + "content": "\\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) / \\rho _ { M } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 174, + 233, + 188 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 233, + 176, + 245, + 186 + ], + "score": 0.82, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 174, + 255, + 188 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 255, + 175, + 312, + 188 + ], + "score": 0.93, + "content": "\\rho _ { M } ^ { \\kappa } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 174, + 360, + 188 + ], + "score": 1.0, + "content": ". Assuming", + "type": "text" + }, + { + "bbox": [ + 361, + 176, + 495, + 187 + ], + "score": 0.89, + "content": "\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 174, + 506, + 188 + ], + "score": 1.0, + "content": "if", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 184, + 452, + 205 + ], + "spans": [ + { + "bbox": [ + 107, + 187, + 167, + 202 + ], + "score": 0.88, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 184, + 260, + 205 + ], + "score": 1.0, + "content": ", which will hold when", + "type": "text" + }, + { + "bbox": [ + 261, + 192, + 267, + 200 + ], + "score": 0.74, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 184, + 324, + 205 + ], + "score": 1.0, + "content": "overlaps with", + "type": "text" + }, + { + "bbox": [ + 325, + 191, + 332, + 201 + ], + "score": 0.79, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 184, + 452, + 205 + ], + "score": 1.0, + "content": ", then Eq. 5 can transform to:", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 5, + "bbox_fs": [ + 104, + 80, + 507, + 205 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 119, + 206, + 492, + 237 + ], + "lines": [ + { + "bbox": [ + 119, + 206, + 492, + 237 + ], + "spans": [ + { + "bbox": [ + 119, + 206, + 492, + 237 + ], + "score": 0.9, + "content": "\\displaystyle \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\Bigg ( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } { \\rho _ { M } ^ { \\kappa } ( x , a ) } - H _ { M ^ { * } } ( x , a ) \\right) \\Bigg ) ,", + "type": "interline_equation", + "image_path": "78af9ac575fb1379ea94dfaa342e4030712e344f1b8af727eb02a909a481a75c.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 119, + 206, + 492, + 216.33333333333334 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 119, + 216.33333333333334, + 492, + 226.66666666666669 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 119, + 226.66666666666669, + 492, + 237.00000000000003 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 243, + 506, + 285 + ], + "lines": [ + { + "bbox": [ + 106, + 241, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 137, + 262 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 137, + 244, + 242, + 257 + ], + "score": 0.94, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 241, + 346, + 262 + ], + "score": 1.0, + "content": ". We notice that the form", + "type": "text" + }, + { + "bbox": [ + 346, + 242, + 399, + 261 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } } { \\rho _ { M } ^ { \\kappa } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 241, + 506, + 259 + ], + "score": 1.0, + "content": "ρκM∗ρκ is the integrated function", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 258, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 148, + 272 + ], + "score": 1.0, + "content": "in reverse", + "type": "text" + }, + { + "bbox": [ + 148, + 259, + 163, + 270 + ], + "score": 0.32, + "content": "\\mathrm { K L }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 258, + 308, + 272 + ], + "score": 1.0, + "content": "divergence, which is an instance of", + "type": "text" + }, + { + "bbox": [ + 308, + 259, + 315, + 271 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 258, + 363, + 272 + ], + "score": 1.0, + "content": "function in", + "type": "text" + }, + { + "bbox": [ + 363, + 260, + 370, + 271 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 258, + 506, + 272 + ], + "score": 1.0, + "content": "-divergence (Ali & Silvey, 1966).", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 268, + 447, + 288 + ], + "spans": [ + { + "bbox": [ + 104, + 268, + 209, + 288 + ], + "score": 1.0, + "content": "Replacing that form with", + "type": "text" + }, + { + "bbox": [ + 209, + 273, + 216, + 284 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 268, + 422, + 288 + ], + "score": 1.0, + "content": "function, we obtain a generalized representation of", + "type": "text" + }, + { + "bbox": [ + 423, + 270, + 441, + 285 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 268, + 447, + 288 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 104, + 241, + 506, + 288 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 289, + 514, + 320 + ], + "lines": [ + { + "bbox": [ + 111, + 289, + 514, + 320 + ], + "spans": [ + { + "bbox": [ + 111, + 289, + 514, + 320 + ], + "score": 0.89, + "content": "\\tilde { \\bar { \\nu } } _ { M ^ { * } } ^ { * } : = \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\left( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) - H _ { M ^ { * } } ( x , a ) \\right) \\right.", + "type": "interline_equation", + "image_path": "43d2dd2f890e8f95bed4967ff118330b972404d5f4c844faa96a213360d47f59.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 111, + 289, + 514, + 299.3333333333333 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 111, + 299.3333333333333, + 514, + 309.66666666666663 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 111, + 309.66666666666663, + 514, + 319.99999999999994 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 513, + 164, + 584, + 175 + ], + "lines": [ + { + "bbox": [ + 512, + 163, + 585, + 176 + ], + "spans": [ + { + "bbox": [ + 512, + 163, + 585, + 176 + ], + "score": 1.0, + "content": "to Reviewer bcJS", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 512, + 163, + 585, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 336, + 506, + 375 + ], + "lines": [ + { + "bbox": [ + 107, + 334, + 507, + 355 + ], + "spans": [ + { + "bbox": [ + 107, + 334, + 135, + 355 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 339, + 187, + 351 + ], + "score": 0.92, + "content": "f : \\mathbb { R } _ { + } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 334, + 407, + 355 + ], + "score": 1.0, + "content": "is a convex and lower semi-continuous (l.s.c.) function.", + "type": "text" + }, + { + "bbox": [ + 408, + 336, + 426, + 352 + ], + "score": 0.93, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 334, + 507, + 355 + ], + "score": 1.0, + "content": "gives a generalized", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "score": 1.0, + "content": "representation of the optimal adversarial data distribution to maximize the error of the model. Based", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 361, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 104, + 361, + 415, + 379 + ], + "score": 1.0, + "content": "on Eq. 6, we have a surrogate objective of AWRM which can avoid querying", + "type": "text" + }, + { + "bbox": [ + 415, + 363, + 431, + 374 + ], + "score": 0.87, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 361, + 482, + 379 + ], + "score": 1.0, + "content": "to construct", + "type": "text" + }, + { + "bbox": [ + 482, + 361, + 501, + 376 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 361, + 506, + 379 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 22, + "bbox_fs": [ + 104, + 334, + 507, + 379 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 378, + 506, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 378, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 182, + 397 + ], + "score": 1.0, + "content": "Theorem 4.4. Let", + "type": "text" + }, + { + "bbox": [ + 182, + 378, + 201, + 394 + ], + "score": 0.91, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 379, + 406, + 397 + ], + "score": 1.0, + "content": "as the data distribution of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 406, + 381, + 417, + 393 + ], + "score": 0.88, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 379, + 506, + 397 + ], + "score": 1.0, + "content": "in Eq. 4 under model", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 394, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 107, + 395, + 122, + 407 + ], + "score": 0.67, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 394, + 199, + 408 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 200, + 396, + 205, + 405 + ], + "score": 0.69, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 394, + 334, + 408 + ], + "score": 1.0, + "content": ", then we can find the optimal", + "type": "text" + }, + { + "bbox": [ + 335, + 396, + 345, + 406 + ], + "score": 0.83, + "content": "\\theta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 394, + 359, + 408 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 359, + 394, + 471, + 408 + ], + "score": 0.85, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 394, + 506, + 408 + ], + "score": 1.0, + "content": "(Eq. 4)", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 406, + 507, + 425 + ], + "spans": [ + { + "bbox": [ + 104, + 406, + 264, + 425 + ], + "score": 1.0, + "content": "via iteratively optimizing the objective", + "type": "text" + }, + { + "bbox": [ + 264, + 407, + 371, + 422 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\theta _ { t + 1 } = \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 406, + 402, + 425 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 402, + 407, + 420, + 422 + ], + "score": 0.91, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 406, + 507, + 425 + ], + "score": 1.0, + "content": "is approximated via", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 207, + 433 + ], + "score": 1.0, + "content": "the last-iteration model", + "type": "text" + }, + { + "bbox": [ + 207, + 420, + 225, + 432 + ], + "score": 0.91, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 420, + 506, + 433 + ], + "score": 1.0, + "content": ". Based on Corollary A.7, we derive an upper bound objective for", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 431, + 187, + 448 + ], + "spans": [ + { + "bbox": [ + 107, + 431, + 181, + 447 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 432, + 187, + 448 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 104, + 378, + 507, + 448 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 451, + 506, + 496 + ], + "lines": [ + { + "bbox": [ + 111, + 451, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 111, + 451, + 506, + 496 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\iota _ { t + 1 } = \\underset { \\theta } { \\operatorname* { m i n } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\frac { - 1 } { \\alpha _ { 0 } ( x , a ) } \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\left( f \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) + H _ { M ^ { * } } ( x , a ) \\right) \\right] , } \\\\ { W ( x , a , x ^ { \\prime } ) } \\end{array}", + "type": "interline_equation", + "image_path": "300a71d94016b5fd8af5a721594f0ad5429351b188668424c87f05a5a0abc29e.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 111, + 451, + 506, + 466.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 111, + 466.0, + 506, + 481.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 111, + 481.0, + 506, + 496.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 500, + 505, + 527 + ], + "lines": [ + { + "bbox": [ + 108, + 497, + 507, + 515 + ], + "spans": [ + { + "bbox": [ + 108, + 497, + 137, + 515 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 137, + 501, + 160, + 514 + ], + "score": 0.84, + "content": "\\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 497, + 208, + 515 + ], + "score": 1.0, + "content": "[·] denotes", + "type": "text" + }, + { + "bbox": [ + 208, + 501, + 257, + 514 + ], + "score": 0.86, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\kappa } }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 497, + 273, + 515 + ], + "score": 1.0, + "content": "[·],", + "type": "text" + }, + { + "bbox": [ + 273, + 501, + 281, + 513 + ], + "score": 0.81, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 497, + 399, + 515 + ], + "score": 1.0, + "content": "is a l.s.c function satisfying", + "type": "text" + }, + { + "bbox": [ + 399, + 500, + 482, + 513 + ], + "score": 0.94, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 497, + 507, + 515 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 107, + 512, + 221, + 527 + ], + "spans": [ + { + "bbox": [ + 107, + 514, + 218, + 527 + ], + "score": 0.91, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 512, + 221, + 527 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 107, + 497, + 507, + 527 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 533, + 505, + 581 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 302, + 547 + ], + "score": 1.0, + "content": "Thm. 4.4 approximately achieve AWRM by using", + "type": "text" + }, + { + "bbox": [ + 303, + 536, + 310, + 544 + ], + "score": 0.75, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 533, + 445, + 547 + ], + "score": 1.0, + "content": "and a pseudo-reweighting module", + "type": "text" + }, + { + "bbox": [ + 445, + 534, + 457, + 544 + ], + "score": 0.46, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 533, + 461, + 547 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 462, + 534, + 473, + 544 + ], + "score": 0.31, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 533, + 505, + 547 + ], + "score": 1.0, + "content": "assigns", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 544, + 506, + 560 + ], + "spans": [ + { + "bbox": [ + 104, + 544, + 428, + 560 + ], + "score": 1.0, + "content": "learning propensities for data points with larger differences between distributions", + "type": "text" + }, + { + "bbox": [ + 428, + 546, + 449, + 560 + ], + "score": 0.92, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 544, + 468, + 560 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 468, + 546, + 487, + 558 + ], + "score": 0.89, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 544, + 506, + 560 + ], + "score": 1.0, + "content": ". By", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "adjusting the weights, the learning process will exploit subtle errors in any data point, whatever how", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 569, + 492, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 492, + 581 + ], + "score": 1.0, + "content": "many proportions it contributes, to correct potential generalization errors on counterfactual data.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 104, + 533, + 506, + 581 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 591, + 295, + 604 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 297, + 604 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 297, + 604 + ], + "score": 1.0, + "content": "4.2.2 TRACTABLE SOLUTION TO THM. 4.4", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 609, + 506, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 608, + 507, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 608, + 195, + 634 + ], + "score": 1.0, + "content": "In Thm. 4.4, the term", + "type": "text" + }, + { + "bbox": [ + 195, + 610, + 320, + 631 + ], + "score": 0.94, + "content": "\\begin{array} { r } { f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\big ) - f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\big ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 615, + 507, + 628 + ], + "score": 1.0, + "content": "is still intractable. To solve the problem, first,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 630, + 487, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 284, + 642 + ], + "score": 1.0, + "content": "we resort to the first-order approximation of", + "type": "text" + }, + { + "bbox": [ + 285, + 632, + 291, + 642 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 630, + 345, + 642 + ], + "score": 1.0, + "content": ". Given some", + "type": "text" + }, + { + "bbox": [ + 346, + 630, + 448, + 642 + ], + "score": 0.91, + "content": "u \\in ( 1 - \\xi , 1 + \\xi ) , \\xi > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 630, + 487, + 642 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.0, + "bbox_fs": [ + 105, + 608, + 507, + 642 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 244, + 646, + 366, + 660 + ], + "lines": [ + { + "bbox": [ + 244, + 646, + 366, + 660 + ], + "spans": [ + { + "bbox": [ + 244, + 646, + 366, + 660 + ], + "score": 0.92, + "content": "f ( u ) \\approx f ( 1 ) + f ^ { \\prime } ( u ) ( u - 1 ) ,", + "type": "interline_equation", + "image_path": "2b517abc4c2eca59111a5c29db0c8e310899041f086df0a62064d90652b53b69.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 244, + 646, + 366, + 660 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 506, + 733 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 132, + 678 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 666, + 142, + 678 + ], + "score": 0.88, + "content": "f ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 665, + 259, + 678 + ], + "score": 1.0, + "content": "is the first-order derivative of", + "type": "text" + }, + { + "bbox": [ + 259, + 667, + 266, + 677 + ], + "score": 0.81, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 665, + 416, + 678 + ], + "score": 1.0, + "content": ". By Taylor’s formula and the fact that", + "type": "text" + }, + { + "bbox": [ + 417, + 666, + 440, + 678 + ], + "score": 0.93, + "content": "f ^ { \\prime } ( u )", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "of the generator", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 144, + 690 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 144, + 678, + 151, + 689 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 677, + 215, + 690 + ], + "score": 1.0, + "content": "is bounded in", + "type": "text" + }, + { + "bbox": [ + 216, + 678, + 275, + 690 + ], + "score": 0.89, + "content": "( 1 - \\xi , 1 + \\xi )", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 677, + 454, + 690 + ], + "score": 1.0, + "content": ", the approximation error is no more than", + "type": "text" + }, + { + "bbox": [ + 455, + 677, + 480, + 690 + ], + "score": 0.9, + "content": "\\mathcal { O } ( \\xi ^ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 677, + 506, + 690 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 685, + 506, + 709 + ], + "spans": [ + { + "bbox": [ + 107, + 689, + 144, + 706 + ], + "score": 0.93, + "content": "\\begin{array} { r } { u = \\frac { p ( x ) } { q ( x ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 685, + 229, + 709 + ], + "score": 1.0, + "content": "in Eq. 7, the pattern", + "type": "text" + }, + { + "bbox": [ + 230, + 690, + 261, + 706 + ], + "score": 0.94, + "content": "f ( { \\frac { p ( x ) } { q ( x ) } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 685, + 394, + 709 + ], + "score": 1.0, + "content": "in Thm. 4.4 can be converted to", + "type": "text" + }, + { + "bbox": [ + 395, + 689, + 503, + 706 + ], + "score": 0.93, + "content": "\\begin{array} { r } { f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) ( \\frac { p ( x ) } { q ( x ) } - 1 ) + f ( 1 ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 685, + 506, + 709 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 108, + 703, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 108, + 705, + 125, + 722 + ], + "score": 0.91, + "content": "\\frac { p ( x ) } { q ( x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 703, + 349, + 723 + ], + "score": 1.0, + "content": "can be approximated by sampling from datasets and", + "type": "text" + }, + { + "bbox": [ + 349, + 706, + 384, + 722 + ], + "score": 0.93, + "content": "f ^ { \\prime } \\big ( \\frac { p ( x ) } { q ( x ) } \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 703, + 505, + 723 + ], + "score": 1.0, + "content": "can be approximated by the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 272, + 733 + ], + "score": 1.0, + "content": "corresponding variational representation", + "type": "text" + }, + { + "bbox": [ + 272, + 721, + 288, + 733 + ], + "score": 0.87, + "content": "T _ { \\varphi ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "according to Lemma A.9 (Nowozin et al., 2016) (see", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 665, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 506, + 159 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 288, + 96 + ], + "score": 1.0, + "content": "Appendix for details). Based on any specific", + "type": "text" + }, + { + "bbox": [ + 289, + 83, + 295, + 94 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 81, + 407, + 96 + ], + "score": 1.0, + "content": "-function, we can represent", + "type": "text" + }, + { + "bbox": [ + 407, + 83, + 420, + 95 + ], + "score": 0.89, + "content": "T _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "with a discriminator", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 107, + 94, + 121, + 106 + ], + "score": 0.86, + "content": "D _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 93, + 278, + 107 + ], + "score": 1.0, + "content": ". However, Thm. 4.4 holds only when", + "type": "text" + }, + { + "bbox": [ + 279, + 94, + 313, + 105 + ], + "score": 0.78, + "content": "\\forall x \\in { \\mathcal { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 93, + 316, + 107 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 317, + 94, + 359, + 105 + ], + "score": 0.71, + "content": "f ^ { \\prime } ( x ) \\leq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 93, + 506, + 107 + ], + "score": 1.0, + "content": ". It can be verified that the instance", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 257, + 117 + ], + "score": 0.88, + "content": "f ( u ) = u \\log u - ( u + 1 ) \\log ( u + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 104, + 262, + 119 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 262, + 105, + 344, + 118 + ], + "score": 0.89, + "content": "T _ { \\varphi } ( u ) = \\log D _ { \\varphi } ( u )", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 104, + 506, + 119 + ], + "score": 1.0, + "content": "satisfy this condition (see Tab. 2 in the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 116, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 466, + 129 + ], + "score": 1.0, + "content": "Appendix). We select that instance for the tractable solution. In summary, by computing", + "type": "text" + }, + { + "bbox": [ + 466, + 116, + 483, + 128 + ], + "score": 0.89, + "content": "f ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 116, + 505, + 129 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 429, + 140 + ], + "score": 1.0, + "content": "first-order approximation (Eq. 7) and leveraging the variational representation", + "type": "text" + }, + { + "bbox": [ + 429, + 127, + 492, + 139 + ], + "score": 0.92, + "content": "T _ { \\varphi ^ { * } } = \\log { \\hat { D } _ { \\varphi ^ { * } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 138, + 428, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 159, + 156 + ], + "score": 1.0, + "content": "approximate", + "type": "text" + }, + { + "bbox": [ + 159, + 138, + 197, + 159 + ], + "score": 0.94, + "content": "f ^ { \\prime } \\Bigl ( \\frac { { { \\rho } _ { M _ { \\theta _ { t } } } ^ { \\kappa } } } { { { \\rho } _ { M ^ { * } } ^ { \\kappa } } } \\Bigr )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 144, + 428, + 156 + ], + "score": 1.0, + "content": ", we can optimize the surrogate objective in Thm. 4.4 via:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "interline_equation", + "bbox": [ + 105, + 162, + 535, + 218 + ], + "lines": [ + { + "bbox": [ + 111, + 162, + 514, + 218 + ], + "spans": [ + { + "bbox": [ + 111, + 162, + 514, + 218 + ], + "score": 0.88, + "content": "\\begin{array} { r l } { { \\iota _ { + 1 } } = \\mathop { \\operatorname* { m a x } } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\log M \\theta ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\left( H _ { M ^ { * } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] , } & { } \\\\ { { s . t . } } & { \\varphi _ { 0 } ^ { * } = \\mathop { \\operatorname* { a r g m a x } } _ { \\varphi _ { 0 } ^ { * } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } & { ( 8 ) } \\\\ & { \\varphi _ { 1 } ^ { * } = \\mathop { \\operatorname { a r g m a x } } _ { \\varphi _ { 1 } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "7a43c0d6a6b17809cafa4cd97f6788b226db8b13f0b3e94e29288ac22d2e93fc.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 105, + 162, + 535, + 180.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 105, + 180.66666666666666, + 535, + 199.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 105, + 199.33333333333331, + 535, + 217.99999999999997 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 222, + 506, + 314 + ], + "lines": [ + { + "bbox": [ + 110, + 221, + 507, + 238 + ], + "spans": [ + { + "bbox": [ + 110, + 221, + 141, + 238 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 142, + 223, + 410, + 236 + ], + "score": 0.89, + "content": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ = \\ \\log D _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ - \\ \\log D _ { \\varphi _ { 1 } ^ { * } } ( x , a ) , \\ \\mathbb { E } _ { \\rho _ { M } ^ { \\kappa } } [ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 221, + 507, + 238 + ], + "score": 1.0, + "content": "is a simplification of", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 234, + 507, + 250 + ], + "spans": [ + { + "bbox": [ + 107, + 236, + 162, + 249 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M } ^ { \\kappa } } [ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 234, + 181, + 250 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 181, + 237, + 193, + 247 + ], + "score": 0.82, + "content": "\\varphi _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 234, + 212, + 250 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 213, + 237, + 225, + 247 + ], + "score": 0.85, + "content": "\\varphi _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 234, + 316, + 250 + ], + "score": 1.0, + "content": "are the parameters of", + "type": "text" + }, + { + "bbox": [ + 316, + 236, + 332, + 248 + ], + "score": 0.9, + "content": "T _ { \\varphi _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 234, + 351, + 250 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 352, + 236, + 367, + 248 + ], + "score": 0.9, + "content": "T _ { \\varphi _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 234, + 507, + 250 + ], + "score": 1.0, + "content": ". Based on Lemma A.9, we have", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 244, + 507, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 254, + 270 + ], + "score": 0.92, + "content": "\\begin{array} { r } { f ^ { \\prime } ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } ) \\approx \\log D _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 244, + 274, + 271 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 274, + 249, + 399, + 270 + ], + "score": 0.93, + "content": "f ^ { \\prime } ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } ) \\approx \\log D _ { \\varphi _ { 1 } ^ { * } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 244, + 507, + 271 + ], + "score": 1.0, + "content": ". Note that in the process,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 267, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 104, + 267, + 182, + 283 + ], + "score": 1.0, + "content": "we ignore the term", + "type": "text" + }, + { + "bbox": [ + 182, + 269, + 217, + 281 + ], + "score": 0.9, + "content": "\\alpha _ { 0 } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 267, + 506, + 283 + ], + "score": 1.0, + "content": "for simplifying the objective. The discussion on the impacts of removing", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 107, + 280, + 141, + 292 + ], + "score": 0.92, + "content": "\\alpha _ { 0 } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 279, + 390, + 293 + ], + "score": 1.0, + "content": "is left in App. B. In Eq. 8, the pseudo-reweighting module", + "type": "text" + }, + { + "bbox": [ + 390, + 280, + 402, + 290 + ], + "score": 0.72, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "in Thm. 4.4 is split into", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 290, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 505, + 303 + ], + "score": 1.0, + "content": "two terms in the RHS of the equation. The first term is a generative adversarial training objective", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 298, + 467, + 318 + ], + "spans": [ + { + "bbox": [ + 104, + 298, + 147, + 318 + ], + "score": 1.0, + "content": "regarding", + "type": "text" + }, + { + "bbox": [ + 147, + 302, + 162, + 313 + ], + "score": 0.88, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 298, + 398, + 318 + ], + "score": 1.0, + "content": "as the generator, while the second term is WERM through", + "type": "text" + }, + { + "bbox": [ + 398, + 302, + 461, + 315 + ], + "score": 0.93, + "content": "H _ { M ^ { * } } - A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 298, + 467, + 318 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12 + }, + { + "type": "title", + "bbox": [ + 108, + 321, + 261, + 332 + ], + "lines": [ + { + "bbox": [ + 106, + 321, + 262, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 262, + 333 + ], + "score": 1.0, + "content": "4.3 PRACTICAL IMPLEMENTATION", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 504, + 360 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "score": 1.0, + "content": "To give a practical implementation of the solution, extra two assumptions introduced in the process", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 350, + 235, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 235, + 361 + ], + "score": 1.0, + "content": "of modeling should be handled:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 107, + 361, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 105, + 361, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 306, + 379 + ], + "score": 1.0, + "content": "First, the approximation of Eq. 7 holds only when", + "type": "text" + }, + { + "bbox": [ + 307, + 361, + 324, + 378 + ], + "score": 0.93, + "content": "\\textstyle { \\frac { p ( x ) } { q ( x ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 361, + 506, + 379 + ], + "score": 1.0, + "content": "is close to 1, which might not be satisfied. To", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 375, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 389 + ], + "score": 1.0, + "content": "handle the problem, we inject a naive supervised learning loss and replace the second term of the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 385, + 478, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 266, + 402 + ], + "score": 1.0, + "content": "objective Eq. 8 when the expectation of", + "type": "text" + }, + { + "bbox": [ + 267, + 387, + 285, + 399 + ], + "score": 0.89, + "content": "D _ { \\varphi _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 385, + 297, + 402 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 297, + 387, + 318, + 401 + ], + "score": 0.9, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 385, + 406, + 402 + ], + "score": 1.0, + "content": "is far away from 0.5", + "type": "text" + }, + { + "bbox": [ + 406, + 387, + 471, + 399 + ], + "score": 0.87, + "content": "f ^ { \\prime } ( 1 ) = \\log 0 . 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 385, + 478, + 402 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 400, + 505, + 518 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 251, + 412 + ], + "score": 1.0, + "content": "Second, the overlap assumption of", + "type": "text" + }, + { + "bbox": [ + 251, + 402, + 258, + 410 + ], + "score": 0.7, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 398, + 506, + 412 + ], + "score": 1.0, + "content": ". In practice, we need to use the real-world data to construct", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 409, + 507, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 169, + 425 + ], + "score": 1.0, + "content": "the distribution", + "type": "text" + }, + { + "bbox": [ + 169, + 412, + 188, + 423 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 409, + 332, + 425 + ], + "score": 1.0, + "content": "and the generative data to construct", + "type": "text" + }, + { + "bbox": [ + 333, + 411, + 353, + 425 + ], + "score": 0.9, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 409, + 507, + 425 + ], + "score": 1.0, + "content": ". In the offline model-learning setting,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 423, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 250, + 436 + ], + "score": 1.0, + "content": "we only have a real-world dataset", + "type": "text" + }, + { + "bbox": [ + 250, + 424, + 259, + 434 + ], + "score": 0.81, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 423, + 398, + 436 + ], + "score": 1.0, + "content": "collected by the behavior policy", + "type": "text" + }, + { + "bbox": [ + 398, + 426, + 405, + 435 + ], + "score": 0.75, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 423, + 505, + 436 + ], + "score": 1.0, + "content": ". We can learn a policy", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 433, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 107, + 436, + 135, + 446 + ], + "score": 0.89, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 433, + 262, + 447 + ], + "score": 1.0, + "content": "via behavior cloning based on", + "type": "text" + }, + { + "bbox": [ + 263, + 435, + 272, + 444 + ], + "score": 0.78, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 433, + 468, + 447 + ], + "score": 1.0, + "content": "(Pomerleau, 1991; Ho & Ermon, 2016) and let", + "type": "text" + }, + { + "bbox": [ + 469, + 435, + 476, + 446 + ], + "score": 0.82, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 433, + 505, + 447 + ], + "score": 1.0, + "content": "be the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 444, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 135, + 459 + ], + "score": 1.0, + "content": "policy", + "type": "text" + }, + { + "bbox": [ + 136, + 449, + 142, + 456 + ], + "score": 0.7, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 444, + 236, + 459 + ], + "score": 1.0, + "content": ". Then we can regard", + "type": "text" + }, + { + "bbox": [ + 237, + 446, + 246, + 456 + ], + "score": 0.82, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 444, + 401, + 459 + ], + "score": 1.0, + "content": "as the empirical data distribution of", + "type": "text" + }, + { + "bbox": [ + 402, + 446, + 420, + 458 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 444, + 505, + 459 + ], + "score": 1.0, + "content": "and the trajectories", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 455, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 104, + 455, + 159, + 473 + ], + "score": 1.0, + "content": "collected by", + "type": "text" + }, + { + "bbox": [ + 160, + 457, + 167, + 468 + ], + "score": 0.82, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 455, + 224, + 473 + ], + "score": 1.0, + "content": "in the model", + "type": "text" + }, + { + "bbox": [ + 224, + 457, + 242, + 469 + ], + "score": 0.92, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 455, + 394, + 473 + ], + "score": 1.0, + "content": "as the empirical data distribution of", + "type": "text" + }, + { + "bbox": [ + 395, + 458, + 416, + 471 + ], + "score": 0.91, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 455, + 506, + 473 + ], + "score": 1.0, + "content": ". But the assumption", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 228, + 482 + ], + "score": 0.83, + "content": "\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\mu ( a | x ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "might not be satisfied. To handle the problem, in behavior cloning,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 481, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 147, + 493 + ], + "score": 1.0, + "content": "we model", + "type": "text" + }, + { + "bbox": [ + 148, + 482, + 155, + 492 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 481, + 505, + 493 + ], + "score": 1.0, + "content": "with a Gaussian distribution and constrain the lower bound of the variance with a small", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 491, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 130, + 505 + ], + "score": 1.0, + "content": "value", + "type": "text" + }, + { + "bbox": [ + 131, + 493, + 160, + 505 + ], + "score": 0.92, + "content": "\\epsilon _ { \\mu } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 491, + 454, + 505 + ], + "score": 1.0, + "content": "to keep the assumption holding. Besides, we add small Gaussian noises", + "type": "text" + }, + { + "bbox": [ + 454, + 492, + 493, + 504 + ], + "score": 0.93, + "content": "\\mathcal { N } ( 0 , \\epsilon _ { D } )", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 491, + 505, + 505 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 503, + 406, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 159, + 520 + ], + "score": 1.0, + "content": "the inputs of", + "type": "text" + }, + { + "bbox": [ + 159, + 506, + 174, + 518 + ], + "score": 0.9, + "content": "D _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 503, + 306, + 520 + ], + "score": 1.0, + "content": "to handle the mismatch between", + "type": "text" + }, + { + "bbox": [ + 307, + 505, + 325, + 518 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 503, + 344, + 520 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 344, + 504, + 362, + 518 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\hat { \\mu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 503, + 391, + 520 + ], + "score": 1.0, + "content": "due to", + "type": "text" + }, + { + "bbox": [ + 392, + 507, + 402, + 518 + ], + "score": 0.83, + "content": "\\epsilon _ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 503, + 406, + 520 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 505, + 567 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 143, + 535 + ], + "score": 1.0, + "content": "In Eq. 8,", + "type": "text" + }, + { + "bbox": [ + 143, + 523, + 165, + 534 + ], + "score": 0.88, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 522, + 338, + 535 + ], + "score": 1.0, + "content": "is unknown in advance. In practice, we use", + "type": "text" + }, + { + "bbox": [ + 339, + 524, + 360, + 534 + ], + "score": 0.89, + "content": "H _ { M _ { \\theta } }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 522, + 506, + 535 + ], + "score": 1.0, + "content": "to estimate it. More specifically, the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 533, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 181, + 545 + ], + "score": 1.0, + "content": "neural network of", + "type": "text" + }, + { + "bbox": [ + 181, + 534, + 196, + 545 + ], + "score": 0.88, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 533, + 505, + 545 + ], + "score": 1.0, + "content": "can be modeled with a Gaussian distribution. The variance of the Gaussian", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 283, + 556 + ], + "score": 1.0, + "content": "distribution is modeled with global variables", + "type": "text" + }, + { + "bbox": [ + 284, + 546, + 291, + 554 + ], + "score": 0.85, + "content": "\\Sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 543, + 461, + 556 + ], + "score": 1.0, + "content": "for each dimension of output. We estimate", + "type": "text" + }, + { + "bbox": [ + 462, + 545, + 483, + 555 + ], + "score": 0.89, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 554, + 335, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 323, + 567 + ], + "score": 1.0, + "content": "the closed-form solution of Gaussian entropy through", + "type": "text" + }, + { + "bbox": [ + 323, + 557, + 331, + 565 + ], + "score": 0.84, + "content": "\\Sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 554, + 335, + 567 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 617 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "Based on the above techniques, we propose Generative Adversarial offLIne counterfactuaL", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "Environment mOdel learning (GALILEO) for environment model learning. GALILEO can be", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 595, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 606 + ], + "score": 1.0, + "content": "adopted in both single-step and sequential environment model learning. The detailed implementation", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 605, + 365, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 365, + 618 + ], + "score": 1.0, + "content": "and comparison to previous adversarial methods are in Appx. E.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5 + }, + { + "type": "title", + "bbox": [ + 107, + 621, + 201, + 633 + ], + "lines": [ + { + "bbox": [ + 104, + 619, + 202, + 636 + ], + "spans": [ + { + "bbox": [ + 104, + 619, + 202, + 636 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 106, + 640, + 506, + 714 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 653 + ], + "score": 1.0, + "content": "In this section, we first conduct experiments in synthetic environments (Bica et al., 2020)", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 651, + 507, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 507, + 663 + ], + "score": 1.0, + "content": "to verify GALILEO on counterfactual queries 1 and the compatibility of GALILEO in se-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 662, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 493, + 674 + ], + "score": 1.0, + "content": "quential and single-step environments. We select Mean Integrated Square Error MISE", + "type": "text" + }, + { + "bbox": [ + 493, + 663, + 505, + 673 + ], + "score": 0.66, + "content": "=", + "type": "inline_equation" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 673, + 508, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 264, + 693 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\bar { \\mathbb { E } } \\left[ \\int _ { \\mathcal { A } } \\left( M ^ { * } ( y | x , a ) - \\bar { M } ( y | x , a ) \\right) ^ { 2 } \\mathrm { d } a \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 673, + 508, + 693 + ], + "score": 1.0, + "content": "as the metric, which is a commonly used metric to mea-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "sure the accuracy in counterfactual queries by considering the prediction errors in the whole action", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 702, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 505, + 714 + ], + "score": 1.0, + "content": "space. Then we analyze the benefits of the implementation techniques described in Sec. 4.3 and", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 513, + 446, + 584, + 457 + ], + "lines": [ + { + "bbox": [ + 513, + 445, + 585, + 458 + ], + "spans": [ + { + "bbox": [ + 513, + 445, + 585, + 458 + ], + "score": 1.0, + "content": "to Reviewer bcJS", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47 + }, + { + "type": "text", + "bbox": [ + 513, + 555, + 584, + 566 + ], + "lines": [ + { + "bbox": [ + 513, + 555, + 585, + 567 + ], + "spans": [ + { + "bbox": [ + 513, + 555, + 585, + 567 + ], + "score": 1.0, + "content": "to Reviewer bcJS", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 48 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 119, + 721, + 314, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 720, + 315, + 733 + ], + "spans": [ + { + "bbox": [ + 119, + 720, + 315, + 733 + ], + "score": 1.0, + "content": "1the code will be released after the paper is published.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 506, + 159 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 288, + 96 + ], + "score": 1.0, + "content": "Appendix for details). Based on any specific", + "type": "text" + }, + { + "bbox": [ + 289, + 83, + 295, + 94 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 81, + 407, + 96 + ], + "score": 1.0, + "content": "-function, we can represent", + "type": "text" + }, + { + "bbox": [ + 407, + 83, + 420, + 95 + ], + "score": 0.89, + "content": "T _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 81, + 506, + 96 + ], + "score": 1.0, + "content": "with a discriminator", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 107, + 94, + 121, + 106 + ], + "score": 0.86, + "content": "D _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 93, + 278, + 107 + ], + "score": 1.0, + "content": ". However, Thm. 4.4 holds only when", + "type": "text" + }, + { + "bbox": [ + 279, + 94, + 313, + 105 + ], + "score": 0.78, + "content": "\\forall x \\in { \\mathcal { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 93, + 316, + 107 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 317, + 94, + 359, + 105 + ], + "score": 0.71, + "content": "f ^ { \\prime } ( x ) \\leq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 93, + 506, + 107 + ], + "score": 1.0, + "content": ". It can be verified that the instance", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 257, + 117 + ], + "score": 0.88, + "content": "f ( u ) = u \\log u - ( u + 1 ) \\log ( u + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 104, + 262, + 119 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 262, + 105, + 344, + 118 + ], + "score": 0.89, + "content": "T _ { \\varphi } ( u ) = \\log D _ { \\varphi } ( u )", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 104, + 506, + 119 + ], + "score": 1.0, + "content": "satisfy this condition (see Tab. 2 in the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 116, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 466, + 129 + ], + "score": 1.0, + "content": "Appendix). We select that instance for the tractable solution. In summary, by computing", + "type": "text" + }, + { + "bbox": [ + 466, + 116, + 483, + 128 + ], + "score": 0.89, + "content": "f ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 116, + 505, + 129 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 429, + 140 + ], + "score": 1.0, + "content": "first-order approximation (Eq. 7) and leveraging the variational representation", + "type": "text" + }, + { + "bbox": [ + 429, + 127, + 492, + 139 + ], + "score": 0.92, + "content": "T _ { \\varphi ^ { * } } = \\log { \\hat { D } _ { \\varphi ^ { * } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 126, + 506, + 140 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 138, + 428, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 159, + 156 + ], + "score": 1.0, + "content": "approximate", + "type": "text" + }, + { + "bbox": [ + 159, + 138, + 197, + 159 + ], + "score": 0.94, + "content": "f ^ { \\prime } \\Bigl ( \\frac { { { \\rho } _ { M _ { \\theta _ { t } } } ^ { \\kappa } } } { { { \\rho } _ { M ^ { * } } ^ { \\kappa } } } \\Bigr )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 144, + 428, + 156 + ], + "score": 1.0, + "content": ", we can optimize the surrogate objective in Thm. 4.4 via:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 81, + 506, + 159 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 105, + 162, + 535, + 218 + ], + "lines": [ + { + "bbox": [ + 111, + 162, + 514, + 218 + ], + "spans": [ + { + "bbox": [ + 111, + 162, + 514, + 218 + ], + "score": 0.88, + "content": "\\begin{array} { r l } { { \\iota _ { + 1 } } = \\mathop { \\operatorname* { m a x } } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\log M \\theta ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\left( H _ { M ^ { * } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] , } & { } \\\\ { { s . t . } } & { \\varphi _ { 0 } ^ { * } = \\mathop { \\operatorname* { a r g m a x } } _ { \\varphi _ { 0 } ^ { * } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } & { ( 8 ) } \\\\ & { \\varphi _ { 1 } ^ { * } = \\mathop { \\operatorname { a r g m a x } } _ { \\varphi _ { 1 } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "7a43c0d6a6b17809cafa4cd97f6788b226db8b13f0b3e94e29288ac22d2e93fc.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 105, + 162, + 535, + 180.66666666666666 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 105, + 180.66666666666666, + 535, + 199.33333333333331 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 105, + 199.33333333333331, + 535, + 217.99999999999997 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 222, + 506, + 314 + ], + "lines": [ + { + "bbox": [ + 110, + 221, + 507, + 238 + ], + "spans": [ + { + "bbox": [ + 110, + 221, + 141, + 238 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 142, + 223, + 410, + 236 + ], + "score": 0.89, + "content": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ = \\ \\log D _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ - \\ \\log D _ { \\varphi _ { 1 } ^ { * } } ( x , a ) , \\ \\mathbb { E } _ { \\rho _ { M } ^ { \\kappa } } [ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 221, + 507, + 238 + ], + "score": 1.0, + "content": "is a simplification of", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 234, + 507, + 250 + ], + "spans": [ + { + "bbox": [ + 107, + 236, + 162, + 249 + ], + "score": 0.91, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M } ^ { \\kappa } } [ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 234, + 181, + 250 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 181, + 237, + 193, + 247 + ], + "score": 0.82, + "content": "\\varphi _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 234, + 212, + 250 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 213, + 237, + 225, + 247 + ], + "score": 0.85, + "content": "\\varphi _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 234, + 316, + 250 + ], + "score": 1.0, + "content": "are the parameters of", + "type": "text" + }, + { + "bbox": [ + 316, + 236, + 332, + 248 + ], + "score": 0.9, + "content": "T _ { \\varphi _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 234, + 351, + 250 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 352, + 236, + 367, + 248 + ], + "score": 0.9, + "content": "T _ { \\varphi _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 234, + 507, + 250 + ], + "score": 1.0, + "content": ". Based on Lemma A.9, we have", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 244, + 507, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 254, + 270 + ], + "score": 0.92, + "content": "\\begin{array} { r } { f ^ { \\prime } ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } ) \\approx \\log D _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 244, + 274, + 271 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 274, + 249, + 399, + 270 + ], + "score": 0.93, + "content": "f ^ { \\prime } ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } ) \\approx \\log D _ { \\varphi _ { 1 } ^ { * } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 244, + 507, + 271 + ], + "score": 1.0, + "content": ". Note that in the process,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 267, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 104, + 267, + 182, + 283 + ], + "score": 1.0, + "content": "we ignore the term", + "type": "text" + }, + { + "bbox": [ + 182, + 269, + 217, + 281 + ], + "score": 0.9, + "content": "\\alpha _ { 0 } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 267, + 506, + 283 + ], + "score": 1.0, + "content": "for simplifying the objective. The discussion on the impacts of removing", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 279, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 107, + 280, + 141, + 292 + ], + "score": 0.92, + "content": "\\alpha _ { 0 } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 279, + 390, + 293 + ], + "score": 1.0, + "content": "is left in App. B. In Eq. 8, the pseudo-reweighting module", + "type": "text" + }, + { + "bbox": [ + 390, + 280, + 402, + 290 + ], + "score": 0.72, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 279, + 506, + 293 + ], + "score": 1.0, + "content": "in Thm. 4.4 is split into", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 290, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 505, + 303 + ], + "score": 1.0, + "content": "two terms in the RHS of the equation. The first term is a generative adversarial training objective", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 298, + 467, + 318 + ], + "spans": [ + { + "bbox": [ + 104, + 298, + 147, + 318 + ], + "score": 1.0, + "content": "regarding", + "type": "text" + }, + { + "bbox": [ + 147, + 302, + 162, + 313 + ], + "score": 0.88, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 298, + 398, + 318 + ], + "score": 1.0, + "content": "as the generator, while the second term is WERM through", + "type": "text" + }, + { + "bbox": [ + 398, + 302, + 461, + 315 + ], + "score": 0.93, + "content": "H _ { M ^ { * } } - A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 298, + 467, + 318 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12, + "bbox_fs": [ + 104, + 221, + 507, + 318 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 321, + 261, + 332 + ], + "lines": [ + { + "bbox": [ + 106, + 321, + 262, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 262, + 333 + ], + "score": 1.0, + "content": "4.3 PRACTICAL IMPLEMENTATION", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 338, + 504, + 360 + ], + "lines": [ + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "score": 1.0, + "content": "To give a practical implementation of the solution, extra two assumptions introduced in the process", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 350, + 235, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 235, + 361 + ], + "score": 1.0, + "content": "of modeling should be handled:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 338, + 505, + 361 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 361, + 505, + 399 + ], + "lines": [ + { + "bbox": [ + 105, + 361, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 306, + 379 + ], + "score": 1.0, + "content": "First, the approximation of Eq. 7 holds only when", + "type": "text" + }, + { + "bbox": [ + 307, + 361, + 324, + 378 + ], + "score": 0.93, + "content": "\\textstyle { \\frac { p ( x ) } { q ( x ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 361, + 506, + 379 + ], + "score": 1.0, + "content": "is close to 1, which might not be satisfied. To", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 375, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 505, + 389 + ], + "score": 1.0, + "content": "handle the problem, we inject a naive supervised learning loss and replace the second term of the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 385, + 478, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 266, + 402 + ], + "score": 1.0, + "content": "objective Eq. 8 when the expectation of", + "type": "text" + }, + { + "bbox": [ + 267, + 387, + 285, + 399 + ], + "score": 0.89, + "content": "D _ { \\varphi _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 385, + 297, + 402 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 297, + 387, + 318, + 401 + ], + "score": 0.9, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 385, + 406, + 402 + ], + "score": 1.0, + "content": "is far away from 0.5", + "type": "text" + }, + { + "bbox": [ + 406, + 387, + 471, + 399 + ], + "score": 0.87, + "content": "f ^ { \\prime } ( 1 ) = \\log 0 . 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 385, + 478, + 402 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 361, + 506, + 402 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 400, + 505, + 518 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 251, + 412 + ], + "score": 1.0, + "content": "Second, the overlap assumption of", + "type": "text" + }, + { + "bbox": [ + 251, + 402, + 258, + 410 + ], + "score": 0.7, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 398, + 506, + 412 + ], + "score": 1.0, + "content": ". In practice, we need to use the real-world data to construct", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 409, + 507, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 169, + 425 + ], + "score": 1.0, + "content": "the distribution", + "type": "text" + }, + { + "bbox": [ + 169, + 412, + 188, + 423 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 409, + 332, + 425 + ], + "score": 1.0, + "content": "and the generative data to construct", + "type": "text" + }, + { + "bbox": [ + 333, + 411, + 353, + 425 + ], + "score": 0.9, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 409, + 507, + 425 + ], + "score": 1.0, + "content": ". In the offline model-learning setting,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 423, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 250, + 436 + ], + "score": 1.0, + "content": "we only have a real-world dataset", + "type": "text" + }, + { + "bbox": [ + 250, + 424, + 259, + 434 + ], + "score": 0.81, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 423, + 398, + 436 + ], + "score": 1.0, + "content": "collected by the behavior policy", + "type": "text" + }, + { + "bbox": [ + 398, + 426, + 405, + 435 + ], + "score": 0.75, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 423, + 505, + 436 + ], + "score": 1.0, + "content": ". We can learn a policy", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 433, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 107, + 436, + 135, + 446 + ], + "score": 0.89, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 433, + 262, + 447 + ], + "score": 1.0, + "content": "via behavior cloning based on", + "type": "text" + }, + { + "bbox": [ + 263, + 435, + 272, + 444 + ], + "score": 0.78, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 433, + 468, + 447 + ], + "score": 1.0, + "content": "(Pomerleau, 1991; Ho & Ermon, 2016) and let", + "type": "text" + }, + { + "bbox": [ + 469, + 435, + 476, + 446 + ], + "score": 0.82, + "content": "\\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 433, + 505, + 447 + ], + "score": 1.0, + "content": "be the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 444, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 135, + 459 + ], + "score": 1.0, + "content": "policy", + "type": "text" + }, + { + "bbox": [ + 136, + 449, + 142, + 456 + ], + "score": 0.7, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 444, + 236, + 459 + ], + "score": 1.0, + "content": ". Then we can regard", + "type": "text" + }, + { + "bbox": [ + 237, + 446, + 246, + 456 + ], + "score": 0.82, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 444, + 401, + 459 + ], + "score": 1.0, + "content": "as the empirical data distribution of", + "type": "text" + }, + { + "bbox": [ + 402, + 446, + 420, + 458 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 444, + 505, + 459 + ], + "score": 1.0, + "content": "and the trajectories", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 455, + 506, + 473 + ], + "spans": [ + { + "bbox": [ + 104, + 455, + 159, + 473 + ], + "score": 1.0, + "content": "collected by", + "type": "text" + }, + { + "bbox": [ + 160, + 457, + 167, + 468 + ], + "score": 0.82, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 455, + 224, + 473 + ], + "score": 1.0, + "content": "in the model", + "type": "text" + }, + { + "bbox": [ + 224, + 457, + 242, + 469 + ], + "score": 0.92, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 455, + 394, + 473 + ], + "score": 1.0, + "content": "as the empirical data distribution of", + "type": "text" + }, + { + "bbox": [ + 395, + 458, + 416, + 471 + ], + "score": 0.91, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 455, + 506, + 473 + ], + "score": 1.0, + "content": ". But the assumption", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 228, + 482 + ], + "score": 0.83, + "content": "\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\mu ( a | x ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "might not be satisfied. To handle the problem, in behavior cloning,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 481, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 147, + 493 + ], + "score": 1.0, + "content": "we model", + "type": "text" + }, + { + "bbox": [ + 148, + 482, + 155, + 492 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 481, + 505, + 493 + ], + "score": 1.0, + "content": "with a Gaussian distribution and constrain the lower bound of the variance with a small", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 491, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 130, + 505 + ], + "score": 1.0, + "content": "value", + "type": "text" + }, + { + "bbox": [ + 131, + 493, + 160, + 505 + ], + "score": 0.92, + "content": "\\epsilon _ { \\mu } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 491, + 454, + 505 + ], + "score": 1.0, + "content": "to keep the assumption holding. Besides, we add small Gaussian noises", + "type": "text" + }, + { + "bbox": [ + 454, + 492, + 493, + 504 + ], + "score": 0.93, + "content": "\\mathcal { N } ( 0 , \\epsilon _ { D } )", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 491, + 505, + 505 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 503, + 406, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 159, + 520 + ], + "score": 1.0, + "content": "the inputs of", + "type": "text" + }, + { + "bbox": [ + 159, + 506, + 174, + 518 + ], + "score": 0.9, + "content": "D _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 503, + 306, + 520 + ], + "score": 1.0, + "content": "to handle the mismatch between", + "type": "text" + }, + { + "bbox": [ + 307, + 505, + 325, + 518 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 503, + 344, + 520 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 344, + 504, + 362, + 518 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\hat { \\mu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 503, + 391, + 520 + ], + "score": 1.0, + "content": "due to", + "type": "text" + }, + { + "bbox": [ + 392, + 507, + 402, + 518 + ], + "score": 0.83, + "content": "\\epsilon _ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 503, + 406, + 520 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26.5, + "bbox_fs": [ + 104, + 398, + 507, + 520 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 522, + 505, + 567 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 143, + 535 + ], + "score": 1.0, + "content": "In Eq. 8,", + "type": "text" + }, + { + "bbox": [ + 143, + 523, + 165, + 534 + ], + "score": 0.88, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 522, + 338, + 535 + ], + "score": 1.0, + "content": "is unknown in advance. In practice, we use", + "type": "text" + }, + { + "bbox": [ + 339, + 524, + 360, + 534 + ], + "score": 0.89, + "content": "H _ { M _ { \\theta } }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 522, + 506, + 535 + ], + "score": 1.0, + "content": "to estimate it. More specifically, the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 533, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 181, + 545 + ], + "score": 1.0, + "content": "neural network of", + "type": "text" + }, + { + "bbox": [ + 181, + 534, + 196, + 545 + ], + "score": 0.88, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 533, + 505, + 545 + ], + "score": 1.0, + "content": "can be modeled with a Gaussian distribution. The variance of the Gaussian", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 283, + 556 + ], + "score": 1.0, + "content": "distribution is modeled with global variables", + "type": "text" + }, + { + "bbox": [ + 284, + 546, + 291, + 554 + ], + "score": 0.85, + "content": "\\Sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 543, + 461, + 556 + ], + "score": 1.0, + "content": "for each dimension of output. We estimate", + "type": "text" + }, + { + "bbox": [ + 462, + 545, + 483, + 555 + ], + "score": 0.89, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 554, + 335, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 323, + 567 + ], + "score": 1.0, + "content": "the closed-form solution of Gaussian entropy through", + "type": "text" + }, + { + "bbox": [ + 323, + 557, + 331, + 565 + ], + "score": 0.84, + "content": "\\Sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 554, + 335, + 567 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 522, + 506, + 567 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 617 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "Based on the above techniques, we propose Generative Adversarial offLIne counterfactuaL", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "Environment mOdel learning (GALILEO) for environment model learning. GALILEO can be", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 595, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 606 + ], + "score": 1.0, + "content": "adopted in both single-step and sequential environment model learning. The detailed implementation", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 605, + 365, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 365, + 618 + ], + "score": 1.0, + "content": "and comparison to previous adversarial methods are in Appx. E.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 572, + 506, + 618 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 621, + 201, + 633 + ], + "lines": [ + { + "bbox": [ + 104, + 619, + 202, + 636 + ], + "spans": [ + { + "bbox": [ + 104, + 619, + 202, + 636 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 106, + 640, + 506, + 714 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 653 + ], + "score": 1.0, + "content": "In this section, we first conduct experiments in synthetic environments (Bica et al., 2020)", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 651, + 507, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 507, + 663 + ], + "score": 1.0, + "content": "to verify GALILEO on counterfactual queries 1 and the compatibility of GALILEO in se-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 662, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 493, + 674 + ], + "score": 1.0, + "content": "quential and single-step environments. We select Mean Integrated Square Error MISE", + "type": "text" + }, + { + "bbox": [ + 493, + 663, + 505, + 673 + ], + "score": 0.66, + "content": "=", + "type": "inline_equation" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 673, + 508, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 264, + 693 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\bar { \\mathbb { E } } \\left[ \\int _ { \\mathcal { A } } \\left( M ^ { * } ( y | x , a ) - \\bar { M } ( y | x , a ) \\right) ^ { 2 } \\mathrm { d } a \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 673, + 508, + 693 + ], + "score": 1.0, + "content": "as the metric, which is a commonly used metric to mea-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "sure the accuracy in counterfactual queries by considering the prediction errors in the whole action", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 702, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 505, + 714 + ], + "score": 1.0, + "content": "space. Then we analyze the benefits of the implementation techniques described in Sec. 4.3 and", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 639, + 508, + 714 + ] + }, + { + "type": "text", + "bbox": [ + 513, + 446, + 584, + 457 + ], + "lines": [ + { + "bbox": [ + 513, + 445, + 585, + 458 + ], + "spans": [ + { + "bbox": [ + 513, + 445, + 585, + 458 + ], + "score": 1.0, + "content": "to Reviewer bcJS", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 513, + 555, + 585, + 567 + ], + "spans": [ + { + "bbox": [ + 513, + 555, + 585, + 567 + ], + "score": 1.0, + "content": "to Reviewer bcJS", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47, + "bbox_fs": [ + 513, + 445, + 585, + 458 + ] + }, + { + "type": "text", + "bbox": [ + 513, + 555, + 584, + 566 + ], + "lines": [], + "index": 48, + "bbox_fs": [ + 513, + 555, + 585, + 567 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 44, + 487, + 155 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 44, + 487, + 155 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 44, + 487, + 155 + ], + "spans": [ + { + "bbox": [ + 119, + 44, + 487, + 155 + ], + "score": 0.969, + "type": "image", + "image_path": "1b046d05ed7b7da16cd32af8ae1f8b8b0181d988d9e217a6214be2ede6d9f372.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 44, + 487, + 81.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 81.0, + 487, + 118.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 118.0, + 487, + 155.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 157, + 504, + 180 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 156, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 106, + 156, + 505, + 169 + ], + "score": 1.0, + "content": "Figure 3: Illustration of the performance in GNFC and TCGA. The grey bar denotes the standard", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 168, + 281, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 132, + 180 + ], + "score": 1.0, + "content": "error (", + "type": "text" + }, + { + "bbox": [ + 132, + 168, + 154, + 179 + ], + "score": 0.85, + "content": "\\mathrm { \\times 0 . 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 168, + 281, + 180 + ], + "score": 1.0, + "content": "for brevity) of 3 random seeds.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 187, + 505, + 276 + ], + "lines": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "score": 1.0, + "content": "the problems without them. Finally, we deploy GALILEO in two complex environments: MuJoCo", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 199, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 210 + ], + "score": 1.0, + "content": "in Gym (Todorov et al., 2012) and a real-world food-delivery platform to test the performance of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 208, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 208, + 506, + 222 + ], + "score": 1.0, + "content": "GALILEO in difficult tasks. The algorithms compared are: (1) Supervised Learning (SL): us-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "score": 1.0, + "content": "ing standard empirical risk minimization for model learning; (2) Inverse Propensity Weighting", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "(IPW) (Spirtes, 2010): a standard implementation of WERM based IPS; (3) eStimating the effects", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 243, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 253 + ], + "score": 1.0, + "content": "of Continuous Interventions using GANs (SCIGAN) (Bica et al., 2020): an adversarial algorithms", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 254, + 504, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 504, + 265 + ], + "score": 1.0, + "content": "for model learning used for causal effect estimation, which can be roughly regarded as a partial", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 263, + 495, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 495, + 276 + ], + "score": 1.0, + "content": "implementation of GALILEO (Refer to Appx. E.2). We give a detailed description in Appx. G.2.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 108, + 289, + 290, + 300 + ], + "lines": [ + { + "bbox": [ + 106, + 288, + 291, + 301 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 291, + 301 + ], + "score": 1.0, + "content": "5.1 TEST IN SYNTHETIC ENVIRONMENTS", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 305, + 505, + 525 + ], + "lines": [ + { + "bbox": [ + 105, + 304, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 319 + ], + "score": 1.0, + "content": "Test in sequential environments Since there does not exist a task specifically designed for se-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "quential environment model learning with selection bias, we construct a task, General Negative", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "Feedback Control (GNFC), which can represent a classic type of task with policies having selection", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "bias. Fig. 1(a) is also an example of GNFC. We give detailed motivation, the effect of selection", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "bias, and other details in Appx. G.1.1. We construct tasks on GNFC by adding behavior policies", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 114, + 372 + ], + "score": 0.76, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 360, + 269, + 373 + ], + "score": 1.0, + "content": "with different scales of uniform noise", + "type": "text" + }, + { + "bbox": [ + 270, + 360, + 307, + 372 + ], + "score": 0.93, + "content": "U ( - e , e )", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 360, + 419, + 373 + ], + "score": 1.0, + "content": "with different probabilities", + "type": "text" + }, + { + "bbox": [ + 419, + 362, + 426, + 372 + ], + "score": 0.75, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 360, + 505, + 373 + ], + "score": 1.0, + "content": ". In particular, with", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 107, + 371, + 186, + 383 + ], + "score": 0.91, + "content": "e \\in \\{ 1 . 0 , 0 . 2 , 0 . 0 5 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 371, + 204, + 384 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 204, + 371, + 285, + 383 + ], + "score": 0.91, + "content": "p \\in \\{ 1 . 0 , 0 . 2 , 0 . 0 5 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 371, + 505, + 384 + ], + "score": 1.0, + "content": ", we construct 9 tasks and name them with the format of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 382, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 139, + 394 + ], + "score": 0.85, + "content": "^ { \\ast } \\mathrm { e } ^ { \\ast } \\mathrm { \\_ p ^ { \\ast } } ^ { \\ast }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 382, + 199, + 394 + ], + "score": 1.0, + "content": ". For example,", + "type": "text" + }, + { + "bbox": [ + 199, + 384, + 242, + 394 + ], + "score": 0.86, + "content": "\\Theta ^ { 1 } \\mathrm { \\mathrm { - } } \\Theta ^ { 0 } \\cdot 2", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 382, + 429, + 394 + ], + "score": 1.0, + "content": "is the task with behavior policy injecting with", + "type": "text" + }, + { + "bbox": [ + 430, + 383, + 468, + 394 + ], + "score": 0.92, + "content": "U ( - 1 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 382, + 506, + 394 + ], + "score": 1.0, + "content": "with 0.2", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 104, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "probability. The results of GNFC tasks are summarized in Fig. 3(a) and the detailed results can be", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 425, + 417 + ], + "score": 1.0, + "content": "found in Tab. 8. The results show that the property of the behavior policy (i.e.,", + "type": "text" + }, + { + "bbox": [ + 426, + 406, + 432, + 414 + ], + "score": 0.63, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 403, + 450, + 417 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 450, + 406, + 456, + 416 + ], + "score": 0.72, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 403, + 505, + 417 + ], + "score": 1.0, + "content": ") dominates", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 342, + 428 + ], + "score": 1.0, + "content": "the generalization ability of the baseline algorithms. When", + "type": "text" + }, + { + "bbox": [ + 343, + 416, + 380, + 426 + ], + "score": 0.89, + "content": "e = 0 . 0 5", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 415, + 505, + 428 + ], + "score": 1.0, + "content": ", almost all of the baselines fail", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "and give a completely opposite response curve (see Fig. 4(a) and Appx. H.2). IPW still perform well", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 437, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 131, + 449 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 131, + 437, + 192, + 448 + ], + "score": 0.91, + "content": "0 . 2 \\leq e \\leq 1 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 437, + 254, + 449 + ], + "score": 1.0, + "content": "but fails when", + "type": "text" + }, + { + "bbox": [ + 254, + 438, + 337, + 448 + ], + "score": 0.86, + "content": "e = 0 . 0 5 , p < = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 437, + 505, + 449 + ], + "score": 1.0, + "content": ". We also found that SCIGAN can reach", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 302, + 460 + ], + "score": 1.0, + "content": "a better performance than other baselines when", + "type": "text" + }, + { + "bbox": [ + 302, + 449, + 384, + 460 + ], + "score": 0.9, + "content": "e = 0 . 0 5 , p < = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 448, + 505, + 460 + ], + "score": 1.0, + "content": ", but the results in other tasks", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 459, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 506, + 471 + ], + "score": 1.0, + "content": "are unstable. GALILEO is the only algorithm that is robust to the selection bias and outputs correct", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 470, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 506, + 482 + ], + "score": 1.0, + "content": "response curves in all of the tasks. Based on the experiment, we also indicate that the commonly used", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "score": 1.0, + "content": "overlap assumption is unreasonable to a certain extent especially in real-world applications since it is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 490, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 506, + 506 + ], + "score": 1.0, + "content": "impractical to inject noises into the whole action space. The problem of overlap assumption being", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "violated should be taken into consideration otherwise the algorithm will be hard to use in practice if", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 513, + 236, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 236, + 527 + ], + "score": 1.0, + "content": "it is sensitive to the noise range.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 534, + 337, + 655 + ], + "lines": [ + { + "bbox": [ + 106, + 534, + 337, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 337, + 546 + ], + "score": 1.0, + "content": "Test in single-step environments Previous experiments", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 546, + 336, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 336, + 556 + ], + "score": 1.0, + "content": "on counterfactual environment model learning are based", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 556, + 338, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 338, + 568 + ], + "score": 1.0, + "content": "on single-step semi-synthetic data simulation (Bica et al.,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 567, + 338, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 338, + 579 + ], + "score": 1.0, + "content": "2020). Since GALILEO is compatible with single-step en-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 578, + 337, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 337, + 589 + ], + "score": 1.0, + "content": "vironment model learning, we select the same task named", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 588, + 337, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 337, + 600 + ], + "score": 1.0, + "content": "TCGA in Bica et al. (2020) to test GALILEO. Based on", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 600, + 337, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 337, + 612 + ], + "score": 1.0, + "content": "three synthetic response functions in TCGA, we construct", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 610, + 337, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 337, + 622 + ], + "score": 1.0, + "content": "9 tasks by choosing different parameters of selection bias", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 622, + 337, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 119, + 632 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 119, + 623, + 127, + 633 + ], + "score": 0.8, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 622, + 337, + 632 + ], + "score": 1.0, + "content": "which is constructed with beta distribution, and de-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 633, + 336, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 181, + 644 + ], + "score": 1.0, + "content": "sign a coefficient", + "type": "text" + }, + { + "bbox": [ + 181, + 635, + 188, + 643 + ], + "score": 0.64, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 633, + 336, + 644 + ], + "score": 1.0, + "content": "to control the selection bias of the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 644, + 337, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 337, + 655 + ], + "score": 1.0, + "content": "beta distribution. We name the tasks with the format of", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 39 + }, + { + "type": "image", + "bbox": [ + 348, + 522, + 500, + 621 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 348, + 522, + 500, + 621 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 348, + 522, + 500, + 621 + ], + "spans": [ + { + "bbox": [ + 348, + 522, + 500, + 621 + ], + "score": 0.961, + "type": "image", + "image_path": "2f2933e9556a0a707044a2d135038204ddd64c137dda2d4f04c33d29fb14eee6.jpg" + } + ] + } + ], + "index": 43.5, + "virtual_lines": [ + { + "bbox": [ + 348, + 522, + 500, + 571.5 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 348, + 571.5, + 500, + 621.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 343, + 633, + 505, + 655 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 343, + 631, + 507, + 646 + ], + "spans": [ + { + "bbox": [ + 343, + 631, + 507, + 646 + ], + "score": 1.0, + "content": "Figure 4: Illustration of the averaged re-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 343, + 645, + 405, + 655 + ], + "spans": [ + { + "bbox": [ + 343, + 645, + 405, + 655 + ], + "score": 1.0, + "content": "sponse curves.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5 + } + ], + "index": 45.5 + }, + { + "type": "text", + "bbox": [ + 106, + 655, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 456, + 667 + ], + "score": 1.0, + "content": "“t?_bias?”. For example, t1_bias2 is the task with the first response functions and", + "type": "text" + }, + { + "bbox": [ + 456, + 655, + 482, + 665 + ], + "score": 0.88, + "content": "c = 2", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 654, + 505, + 667 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "detail of TCGA is in Appx. G.1.2. The results of TCGA tasks are summarized in Fig. 3(b) and the", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "detailed results can be found in Tab. 9 in Appendix. We found the phenomenon in this experiment", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 104, + 686, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 104, + 686, + 505, + 702 + ], + "score": 1.0, + "content": "is similar to the one in GNFC, which demonstrates the compatibility of GALILEO to single-step", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "environments. We also found that the results of IPW are unstable in this experiment. It might be", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 451, + 722 + ], + "score": 1.0, + "content": "because the behavior policy is modeled with beta distribution while the propensity score", + "type": "text" + }, + { + "bbox": [ + 452, + 710, + 459, + 721 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "is modeled", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 720, + 505, + 735 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 400, + 733 + ], + "score": 1.0, + "content": "with Gaussian distribution. Since IPW directly reweight loss function with", + "type": "text" + }, + { + "bbox": [ + 401, + 720, + 408, + 735 + ], + "score": 0.86, + "content": "\\stackrel { \\mathrm { ~ \\tiny ~ 1 ~ } } { \\hat { \\mu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 721, + 505, + 733 + ], + "score": 1.0, + "content": ", the results are sensitive", + "type": "text" + } + ], + "index": 55 + } + ], + "index": 52 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 514, + 264, + 585, + 275 + ], + "lines": [ + { + "bbox": [ + 512, + 263, + 586, + 277 + ], + "spans": [ + { + "bbox": [ + 512, + 263, + 586, + 277 + ], + "score": 1.0, + "content": "to Reviewer rQ79", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 44, + 487, + 155 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 44, + 487, + 155 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 44, + 487, + 155 + ], + "spans": [ + { + "bbox": [ + 119, + 44, + 487, + 155 + ], + "score": 0.969, + "type": "image", + "image_path": "1b046d05ed7b7da16cd32af8ae1f8b8b0181d988d9e217a6214be2ede6d9f372.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 44, + 487, + 81.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 81.0, + 487, + 118.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 118.0, + 487, + 155.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 157, + 504, + 180 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 156, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 106, + 156, + 505, + 169 + ], + "score": 1.0, + "content": "Figure 3: Illustration of the performance in GNFC and TCGA. The grey bar denotes the standard", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 168, + 281, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 132, + 180 + ], + "score": 1.0, + "content": "error (", + "type": "text" + }, + { + "bbox": [ + 132, + 168, + 154, + 179 + ], + "score": 0.85, + "content": "\\mathrm { \\times 0 . 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 168, + 281, + 180 + ], + "score": 1.0, + "content": "for brevity) of 3 random seeds.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 187, + 505, + 276 + ], + "lines": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "score": 1.0, + "content": "the problems without them. Finally, we deploy GALILEO in two complex environments: MuJoCo", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 199, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 210 + ], + "score": 1.0, + "content": "in Gym (Todorov et al., 2012) and a real-world food-delivery platform to test the performance of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 208, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 208, + 506, + 222 + ], + "score": 1.0, + "content": "GALILEO in difficult tasks. The algorithms compared are: (1) Supervised Learning (SL): us-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 506, + 233 + ], + "score": 1.0, + "content": "ing standard empirical risk minimization for model learning; (2) Inverse Propensity Weighting", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "(IPW) (Spirtes, 2010): a standard implementation of WERM based IPS; (3) eStimating the effects", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 243, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 253 + ], + "score": 1.0, + "content": "of Continuous Interventions using GANs (SCIGAN) (Bica et al., 2020): an adversarial algorithms", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 254, + 504, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 504, + 265 + ], + "score": 1.0, + "content": "for model learning used for causal effect estimation, which can be roughly regarded as a partial", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 263, + 495, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 495, + 276 + ], + "score": 1.0, + "content": "implementation of GALILEO (Refer to Appx. E.2). We give a detailed description in Appx. G.2.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 187, + 506, + 276 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 289, + 290, + 300 + ], + "lines": [ + { + "bbox": [ + 106, + 288, + 291, + 301 + ], + "spans": [ + { + "bbox": [ + 106, + 288, + 291, + 301 + ], + "score": 1.0, + "content": "5.1 TEST IN SYNTHETIC ENVIRONMENTS", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 305, + 505, + 525 + ], + "lines": [ + { + "bbox": [ + 105, + 304, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 319 + ], + "score": 1.0, + "content": "Test in sequential environments Since there does not exist a task specifically designed for se-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "quential environment model learning with selection bias, we construct a task, General Negative", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "Feedback Control (GNFC), which can represent a classic type of task with policies having selection", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 505, + 351 + ], + "score": 1.0, + "content": "bias. Fig. 1(a) is also an example of GNFC. We give detailed motivation, the effect of selection", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 505, + 362 + ], + "score": 1.0, + "content": "bias, and other details in Appx. G.1.1. We construct tasks on GNFC by adding behavior policies", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 360, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 114, + 372 + ], + "score": 0.76, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 360, + 269, + 373 + ], + "score": 1.0, + "content": "with different scales of uniform noise", + "type": "text" + }, + { + "bbox": [ + 270, + 360, + 307, + 372 + ], + "score": 0.93, + "content": "U ( - e , e )", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 360, + 419, + 373 + ], + "score": 1.0, + "content": "with different probabilities", + "type": "text" + }, + { + "bbox": [ + 419, + 362, + 426, + 372 + ], + "score": 0.75, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 360, + 505, + 373 + ], + "score": 1.0, + "content": ". In particular, with", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 107, + 371, + 186, + 383 + ], + "score": 0.91, + "content": "e \\in \\{ 1 . 0 , 0 . 2 , 0 . 0 5 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 371, + 204, + 384 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 204, + 371, + 285, + 383 + ], + "score": 0.91, + "content": "p \\in \\{ 1 . 0 , 0 . 2 , 0 . 0 5 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 371, + 505, + 384 + ], + "score": 1.0, + "content": ", we construct 9 tasks and name them with the format of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 382, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 139, + 394 + ], + "score": 0.85, + "content": "^ { \\ast } \\mathrm { e } ^ { \\ast } \\mathrm { \\_ p ^ { \\ast } } ^ { \\ast }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 382, + 199, + 394 + ], + "score": 1.0, + "content": ". For example,", + "type": "text" + }, + { + "bbox": [ + 199, + 384, + 242, + 394 + ], + "score": 0.86, + "content": "\\Theta ^ { 1 } \\mathrm { \\mathrm { - } } \\Theta ^ { 0 } \\cdot 2", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 382, + 429, + 394 + ], + "score": 1.0, + "content": "is the task with behavior policy injecting with", + "type": "text" + }, + { + "bbox": [ + 430, + 383, + 468, + 394 + ], + "score": 0.92, + "content": "U ( - 1 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 382, + 506, + 394 + ], + "score": 1.0, + "content": "with 0.2", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 104, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "probability. The results of GNFC tasks are summarized in Fig. 3(a) and the detailed results can be", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 403, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 425, + 417 + ], + "score": 1.0, + "content": "found in Tab. 8. The results show that the property of the behavior policy (i.e.,", + "type": "text" + }, + { + "bbox": [ + 426, + 406, + 432, + 414 + ], + "score": 0.63, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 403, + 450, + 417 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 450, + 406, + 456, + 416 + ], + "score": 0.72, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 403, + 505, + 417 + ], + "score": 1.0, + "content": ") dominates", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 415, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 342, + 428 + ], + "score": 1.0, + "content": "the generalization ability of the baseline algorithms. When", + "type": "text" + }, + { + "bbox": [ + 343, + 416, + 380, + 426 + ], + "score": 0.89, + "content": "e = 0 . 0 5", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 415, + 505, + 428 + ], + "score": 1.0, + "content": ", almost all of the baselines fail", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 438 + ], + "score": 1.0, + "content": "and give a completely opposite response curve (see Fig. 4(a) and Appx. H.2). IPW still perform well", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 437, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 131, + 449 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 131, + 437, + 192, + 448 + ], + "score": 0.91, + "content": "0 . 2 \\leq e \\leq 1 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 437, + 254, + 449 + ], + "score": 1.0, + "content": "but fails when", + "type": "text" + }, + { + "bbox": [ + 254, + 438, + 337, + 448 + ], + "score": 0.86, + "content": "e = 0 . 0 5 , p < = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 437, + 505, + 449 + ], + "score": 1.0, + "content": ". We also found that SCIGAN can reach", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 302, + 460 + ], + "score": 1.0, + "content": "a better performance than other baselines when", + "type": "text" + }, + { + "bbox": [ + 302, + 449, + 384, + 460 + ], + "score": 0.9, + "content": "e = 0 . 0 5 , p < = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 448, + 505, + 460 + ], + "score": 1.0, + "content": ", but the results in other tasks", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 459, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 506, + 471 + ], + "score": 1.0, + "content": "are unstable. GALILEO is the only algorithm that is robust to the selection bias and outputs correct", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 470, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 506, + 482 + ], + "score": 1.0, + "content": "response curves in all of the tasks. Based on the experiment, we also indicate that the commonly used", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "score": 1.0, + "content": "overlap assumption is unreasonable to a certain extent especially in real-world applications since it is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 490, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 506, + 506 + ], + "score": 1.0, + "content": "impractical to inject noises into the whole action space. The problem of overlap assumption being", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "violated should be taken into consideration otherwise the algorithm will be hard to use in practice if", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 513, + 236, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 236, + 527 + ], + "score": 1.0, + "content": "it is sensitive to the noise range.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 23.5, + "bbox_fs": [ + 104, + 304, + 506, + 527 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 534, + 337, + 655 + ], + "lines": [ + { + "bbox": [ + 106, + 534, + 337, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 337, + 546 + ], + "score": 1.0, + "content": "Test in single-step environments Previous experiments", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 546, + 336, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 336, + 556 + ], + "score": 1.0, + "content": "on counterfactual environment model learning are based", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 556, + 338, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 338, + 568 + ], + "score": 1.0, + "content": "on single-step semi-synthetic data simulation (Bica et al.,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 567, + 338, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 338, + 579 + ], + "score": 1.0, + "content": "2020). Since GALILEO is compatible with single-step en-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 578, + 337, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 337, + 589 + ], + "score": 1.0, + "content": "vironment model learning, we select the same task named", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 588, + 337, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 337, + 600 + ], + "score": 1.0, + "content": "TCGA in Bica et al. (2020) to test GALILEO. Based on", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 600, + 337, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 337, + 612 + ], + "score": 1.0, + "content": "three synthetic response functions in TCGA, we construct", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 610, + 337, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 337, + 622 + ], + "score": 1.0, + "content": "9 tasks by choosing different parameters of selection bias", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 622, + 337, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 119, + 632 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 119, + 623, + 127, + 633 + ], + "score": 0.8, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 622, + 337, + 632 + ], + "score": 1.0, + "content": "which is constructed with beta distribution, and de-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 633, + 336, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 181, + 644 + ], + "score": 1.0, + "content": "sign a coefficient", + "type": "text" + }, + { + "bbox": [ + 181, + 635, + 188, + 643 + ], + "score": 0.64, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 633, + 336, + 644 + ], + "score": 1.0, + "content": "to control the selection bias of the", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 644, + 337, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 337, + 655 + ], + "score": 1.0, + "content": "beta distribution. We name the tasks with the format of", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 456, + 667 + ], + "score": 1.0, + "content": "“t?_bias?”. For example, t1_bias2 is the task with the first response functions and", + "type": "text" + }, + { + "bbox": [ + 456, + 655, + 482, + 665 + ], + "score": 0.88, + "content": "c = 2", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 654, + 505, + 667 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "detail of TCGA is in Appx. G.1.2. The results of TCGA tasks are summarized in Fig. 3(b) and the", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "detailed results can be found in Tab. 9 in Appendix. We found the phenomenon in this experiment", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 104, + 686, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 104, + 686, + 505, + 702 + ], + "score": 1.0, + "content": "is similar to the one in GNFC, which demonstrates the compatibility of GALILEO to single-step", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "environments. We also found that the results of IPW are unstable in this experiment. It might be", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 451, + 722 + ], + "score": 1.0, + "content": "because the behavior policy is modeled with beta distribution while the propensity score", + "type": "text" + }, + { + "bbox": [ + 452, + 710, + 459, + 721 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "is modeled", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 720, + 505, + 735 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 400, + 733 + ], + "score": 1.0, + "content": "with Gaussian distribution. Since IPW directly reweight loss function with", + "type": "text" + }, + { + "bbox": [ + 401, + 720, + 408, + 735 + ], + "score": 0.86, + "content": "\\stackrel { \\mathrm { ~ \\tiny ~ 1 ~ } } { \\hat { \\mu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 721, + 505, + 733 + ], + "score": 1.0, + "content": ", the results are sensitive", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 106, + 176, + 504, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 164, + 187 + ], + "score": 1.0, + "content": "to the error on", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 165, + 176, + 172, + 187 + ], + "score": 0.78, + "content": "\\hat { \\mu }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 172, + 176, + 270, + 187 + ], + "score": 1.0, + "content": ". GALILEO also models", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 271, + 176, + 278, + 187 + ], + "score": 0.82, + "content": "\\hat { \\mu }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 279, + 176, + 504, + 187 + ], + "score": 1.0, + "content": "with Gaussian distribution but the results are more stable", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 186, + 335, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 285, + 199 + ], + "score": 1.0, + "content": "since GALILEO does not re-weight through", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 285, + 187, + 293, + 198 + ], + "score": 0.81, + "content": "\\hat { \\mu }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 293, + 186, + 335, + 199 + ], + "score": 1.0, + "content": "explicitly.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 534, + 338, + 655 + ] + }, + { + "type": "image", + "bbox": [ + 348, + 522, + 500, + 621 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 348, + 522, + 500, + 621 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 348, + 522, + 500, + 621 + ], + "spans": [ + { + "bbox": [ + 348, + 522, + 500, + 621 + ], + "score": 0.961, + "type": "image", + "image_path": "2f2933e9556a0a707044a2d135038204ddd64c137dda2d4f04c33d29fb14eee6.jpg" + } + ] + } + ], + "index": 43.5, + "virtual_lines": [ + { + "bbox": [ + 348, + 522, + 500, + 571.5 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 348, + 571.5, + 500, + 621.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 343, + 633, + 505, + 655 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 343, + 631, + 507, + 646 + ], + "spans": [ + { + "bbox": [ + 343, + 631, + 507, + 646 + ], + "score": 1.0, + "content": "Figure 4: Illustration of the averaged re-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 343, + 645, + 405, + 655 + ], + "spans": [ + { + "bbox": [ + 343, + 645, + 405, + 655 + ], + "score": 1.0, + "content": "sponse curves.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5 + } + ], + "index": 45.5 + }, + { + "type": "text", + "bbox": [ + 106, + 655, + 505, + 732 + ], + "lines": [], + "index": 52, + "bbox_fs": [ + 104, + 654, + 505, + 735 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 102, + 503, + 168 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 46, + 505, + 102 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 45, + 506, + 59 + ], + "spans": [ + { + "bbox": [ + 105, + 45, + 506, + 59 + ], + "score": 1.0, + "content": "Table 1: Results of policy performance directly optimized through standard SAC (Haarnoja et al.,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 56, + 505, + 69 + ], + "spans": [ + { + "bbox": [ + 105, + 56, + 505, + 69 + ], + "score": 1.0, + "content": "2018) using the learned dynamics models and deployed in MuJoCo environments. MAX-RETURN", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 68, + 505, + 81 + ], + "spans": [ + { + "bbox": [ + 105, + 68, + 505, + 81 + ], + "score": 1.0, + "content": "is the policy performance of SAC in the MuJoCo environments, and “avg. norm.” is the averaged", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 79, + 505, + 91 + ], + "spans": [ + { + "bbox": [ + 106, + 79, + 505, + 91 + ], + "score": 1.0, + "content": "normalized return of the policies in the 9 tasks, where the returns are normalized to lie between 0 and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 90, + 497, + 103 + ], + "spans": [ + { + "bbox": [ + 105, + 90, + 497, + 103 + ], + "score": 1.0, + "content": "100, where a score of 0 corresponds to the worst policy, and 100 corresponds to MAX-RETURN.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 108, + 102, + 503, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 102, + 503, + 168 + ], + "spans": [ + { + "bbox": [ + 108, + 102, + 503, + 168 + ], + "score": 0.979, + "html": "
TaskHopperWalker2dHalfCheetah|avg. norm.
HorizonH=10H=20H=40H=10H=20H=40H=10H=20H=40/
GALILEO13.0 ± 0.133.2 ± 0.153.5± 1.211.7±0.229.9 ± 0.361.2 ± 3.40.7±0.2-1.1 ± 0.2-14.2 ± 1.451.1
SL4.8±0.53.0±0.24.6±0.210.7±0.220.1±0.837.5±6.70.4±0.5-1.1± 0.6-13.2 ± 0.321.1
IPW5.9 ±0.74.1 ± 0.65.9±0.24.7 ± 1.12.8 ±3.914.5 ± 1.41.6± 0.20.5 ± 0.8-11.3 ± 0.919.7
SCIGAN12.7 ± 0.129.2 ± 0.646.2 ±5.28.4 ± 0.59.1 ± 1.71.0 ± 5.81.2 ± 0.3-0.3 ±1.0-11.4 ± 0.341.8
MAX-RETURN13.2 ±0.033.3±0.271.0±0.514.9 ± 1.360.7 ± 11.1221.1 ± 8.92.6±0.113.3 ± 1.149.1 ± 2.3100.0
", + "type": "table", + "image_path": "8ea8b67bfcc3d24ffbedb995e97b37b863570855e8aa4a67d2ca9de37106e6f8.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 108, + 102, + 503, + 124.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 108, + 124.0, + 503, + 146.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 108, + 146.0, + 503, + 168.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "text", + "bbox": [ + 107, + 175, + 503, + 198 + ], + "lines": [ + { + "bbox": [ + 106, + 176, + 504, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 164, + 187 + ], + "score": 1.0, + "content": "to the error on", + "type": "text" + }, + { + "bbox": [ + 165, + 176, + 172, + 187 + ], + "score": 0.78, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 176, + 270, + 187 + ], + "score": 1.0, + "content": ". GALILEO also models", + "type": "text" + }, + { + "bbox": [ + 271, + 176, + 278, + 187 + ], + "score": 0.82, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 176, + 504, + 187 + ], + "score": 1.0, + "content": "with Gaussian distribution but the results are more stable", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 186, + 335, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 186, + 285, + 199 + ], + "score": 1.0, + "content": "since GALILEO does not re-weight through", + "type": "text" + }, + { + "bbox": [ + 285, + 187, + 293, + 198 + ], + "score": 0.81, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 186, + 335, + 199 + ], + "score": 1.0, + "content": "explicitly.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 106, + 201, + 505, + 257 + ], + "lines": [ + { + "bbox": [ + 106, + 202, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 202, + 505, + 214 + ], + "score": 1.0, + "content": "Response curve visualization We plot the averaged response curves which are constructed by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 212, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 505, + 225 + ], + "score": 1.0, + "content": "equidistantly sampling action from the action space and averaging the feedback of the states in the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "dataset as the averaged response. Parts of the results in Fig. 4 (all curves can be seen in Appx. H.2).", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 234, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 506, + 248 + ], + "score": 1.0, + "content": "For those tasks where baselines fail in reconstructing response curves, GALILEO not only reaches a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 244, + 342, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 342, + 258 + ], + "score": 1.0, + "content": "better MISE score but reconstructs almost exact responses.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 108, + 260, + 504, + 294 + ], + "lines": [ + { + "bbox": [ + 106, + 260, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 505, + 273 + ], + "score": 1.0, + "content": "Ablation studies In Sec. 4.3, we introduce several techniques to develop a practical GALILEO", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 211, + 284 + ], + "score": 1.0, + "content": "algorithm. Based on task", + "type": "text" + }, + { + "bbox": [ + 212, + 272, + 272, + 283 + ], + "score": 0.85, + "content": "\\mathtt { e 0 } . 2 _ { - \\mathtt { p } 0 . 0 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 272, + 505, + 284 + ], + "score": 1.0, + "content": "of GNFC, we give the ablation studies to investigate the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 282, + 496, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 496, + 296 + ], + "score": 1.0, + "content": "effects of these techniques. As the main-body space is limited, we leave the results in Appx. H.3.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 108, + 299, + 283, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 298, + 284, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 284, + 312 + ], + "score": 1.0, + "content": "5.2 TEST IN COMPLEX ENVIRONMENTS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 313, + 505, + 424 + ], + "lines": [ + { + "bbox": [ + 106, + 314, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 505, + 325 + ], + "score": 1.0, + "content": "In MuJoCo tasks MuJoCo is a benchmark task in Gym (Todorov et al., 2012; Brockman et al.,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 324, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 505, + 338 + ], + "score": 1.0, + "content": "2016) where we need to control a robot with specific dynamics to complete some tasks (e.g., standing", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 334, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 104, + 334, + 506, + 351 + ], + "score": 1.0, + "content": "or running). We select 3 environment from D4RL (Fu et al., 2020) to construct our model learning", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 347, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 506, + 360 + ], + "score": 1.0, + "content": "tasks. We compare it with a standard transition model learning algorithm used in the previous", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 357, + 506, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 506, + 370 + ], + "score": 1.0, + "content": "offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020), which is a variant of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 369, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 505, + 381 + ], + "score": 1.0, + "content": "supervised learning. We name the method OFF-SL. Besides, we also implement IPW and SCIGAN", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 378, + 507, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 507, + 393 + ], + "score": 1.0, + "content": "as the baselines. In D4RL benchmark, only the “medium” tasks is collected with a fixed policy, i.e.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "the behavior policy is with 1/3 performance to the expert policy), which is most matching to our", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 401, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 506, + 415 + ], + "score": 1.0, + "content": "proposed problem. So we train models in datasets HalfCheetah-medium, Walker2d-medium, and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 412, + 474, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 474, + 425 + ], + "score": 1.0, + "content": "Hopper-medium. We trained the models with the same gradient steps and saved the models.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 429, + 505, + 518 + ], + "lines": [ + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "We first verify the generalization ability of the models by adopting them into offline model-based", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "RL. Instead of designing sophisticated tricks to suppress policy exploration and learning in risky", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "regions as current offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020) do, we", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "score": 1.0, + "content": "just use the standard SAC algorithm Haarnoja et al. (2018) to exploit the models for policy learning", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 473, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 505, + 486 + ], + "score": 1.0, + "content": "to strictly verify the ability of the models. Unfortunately, we found that the compounding error will", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "score": 1.0, + "content": "still be inevitably large in the 1,000-step rollout, which is the standard horizon in MuJoCo tasks,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "leading all models to fail to derive a reasonable policy. To better verify the effects of models on policy", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 505, + 488, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 413, + 519 + ], + "score": 1.0, + "content": "optimization, we learn and evaluate the policies with three smaller horizons:", + "type": "text" + }, + { + "bbox": [ + 414, + 507, + 484, + 519 + ], + "score": 0.91, + "content": "H \\in \\{ 1 0 , 2 0 , 4 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 505, + 488, + 519 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 106, + 522, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 523, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 534 + ], + "score": 1.0, + "content": "The results are listed in Tab. 1. We first averaged the normalized return (refer to “avg. norm.”)", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "under each task, and we can see that the policy obtained by GALILEO is significantly higher than", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 258, + 556 + ], + "score": 1.0, + "content": "other models (the improvements are", + "type": "text" + }, + { + "bbox": [ + 259, + 545, + 279, + 556 + ], + "score": 0.86, + "content": "24 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 545, + 292, + 556 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 292, + 545, + 317, + 556 + ], + "score": 0.87, + "content": "161 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 545, + 505, + 556 + ], + "score": 1.0, + "content": "). At the same time, we found that SCIGAN", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "score": 1.0, + "content": "performed better in policy learning, while IPW performed similarly to SL. This is in line with", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 566, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 506, + 581 + ], + "score": 1.0, + "content": "our expectations, since IPW only considers the uniform policy as the target policy for debiasing,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 578, + 504, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 504, + 590 + ], + "score": 1.0, + "content": "while policy optimization requires querying a wide variety of policies. Minimizing the prediction", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 588, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 601 + ], + "score": 1.0, + "content": "risks only under a uniform policy cannot yield a good environment model for policy optimization.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "Besides, in IPW, the cumulative effects of policy on the state distribution are ignored. On the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "other hand, SCIGAN, as a partial implementation of GALILEO (refer to Appx. E.2), also roughly", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "score": 1.0, + "content": "achieves AWRM and considers the cumulative effects of policy on the state distribution, so its overall", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 633, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 505, + 644 + ], + "score": 1.0, + "content": "performance is better; In addition, we find that GALILEO achieves significant improvement in 6 of", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 644, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 505, + 655 + ], + "score": 1.0, + "content": "the 9 tasks. But in HalfCheetah, IPW works slightly better. However, compared with MAX-RETURN,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "it can be found that all methods fail to derive reasonable policies because their policies’ performances", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "are far away from the optimal policy. By further visualizing the trajectories, we found that all the", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "learned policies just keep the cheetah standing in the same place or even going backward. This", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "phenomenon is also similar to the results in MOPO (Yu et al., 2020). In MOPO’s experiment in", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "the medium datasets, the truncated-rollout horizon used in Walker and Hopper for policy training", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "is set to 5, while HalfCheetah has to be set to the minimal value: 1. These phenomena indicate", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "that HalfCheetah may still have unknown problems, resulting in the generalization bottleneck of", + "type": "text" + } + ], + "index": 55 + } + ], + "index": 46 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 513, + 413, + 586, + 435 + ], + "lines": [ + { + "bbox": [ + 513, + 412, + 587, + 424 + ], + "spans": [ + { + "bbox": [ + 513, + 412, + 587, + 424 + ], + "score": 1.0, + "content": "to Reviewer bcJS,", + "type": "text" + } + ] + }, + { + "bbox": [ + 511, + 423, + 577, + 436 + ], + "spans": [ + { + "bbox": [ + 511, + 423, + 577, + 436 + ], + "score": 1.0, + "content": "j5p5, and rQ79", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 102, + 503, + 168 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 46, + 505, + 102 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 45, + 506, + 59 + ], + "spans": [ + { + "bbox": [ + 105, + 45, + 506, + 59 + ], + "score": 1.0, + "content": "Table 1: Results of policy performance directly optimized through standard SAC (Haarnoja et al.,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 56, + 505, + 69 + ], + "spans": [ + { + "bbox": [ + 105, + 56, + 505, + 69 + ], + "score": 1.0, + "content": "2018) using the learned dynamics models and deployed in MuJoCo environments. MAX-RETURN", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 68, + 505, + 81 + ], + "spans": [ + { + "bbox": [ + 105, + 68, + 505, + 81 + ], + "score": 1.0, + "content": "is the policy performance of SAC in the MuJoCo environments, and “avg. norm.” is the averaged", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 79, + 505, + 91 + ], + "spans": [ + { + "bbox": [ + 106, + 79, + 505, + 91 + ], + "score": 1.0, + "content": "normalized return of the policies in the 9 tasks, where the returns are normalized to lie between 0 and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 90, + 497, + 103 + ], + "spans": [ + { + "bbox": [ + 105, + 90, + 497, + 103 + ], + "score": 1.0, + "content": "100, where a score of 0 corresponds to the worst policy, and 100 corresponds to MAX-RETURN.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "table_body", + "bbox": [ + 108, + 102, + 503, + 168 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 102, + 503, + 168 + ], + "spans": [ + { + "bbox": [ + 108, + 102, + 503, + 168 + ], + "score": 0.979, + "html": "
TaskHopperWalker2dHalfCheetah|avg. norm.
HorizonH=10H=20H=40H=10H=20H=40H=10H=20H=40/
GALILEO13.0 ± 0.133.2 ± 0.153.5± 1.211.7±0.229.9 ± 0.361.2 ± 3.40.7±0.2-1.1 ± 0.2-14.2 ± 1.451.1
SL4.8±0.53.0±0.24.6±0.210.7±0.220.1±0.837.5±6.70.4±0.5-1.1± 0.6-13.2 ± 0.321.1
IPW5.9 ±0.74.1 ± 0.65.9±0.24.7 ± 1.12.8 ±3.914.5 ± 1.41.6± 0.20.5 ± 0.8-11.3 ± 0.919.7
SCIGAN12.7 ± 0.129.2 ± 0.646.2 ±5.28.4 ± 0.59.1 ± 1.71.0 ± 5.81.2 ± 0.3-0.3 ±1.0-11.4 ± 0.341.8
MAX-RETURN13.2 ±0.033.3±0.271.0±0.514.9 ± 1.360.7 ± 11.1221.1 ± 8.92.6±0.113.3 ± 1.149.1 ± 2.3100.0
", + "type": "table", + "image_path": "8ea8b67bfcc3d24ffbedb995e97b37b863570855e8aa4a67d2ca9de37106e6f8.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 108, + 102, + 503, + 124.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 108, + 124.0, + 503, + 146.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 108, + 146.0, + 503, + 168.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 4.0 + }, + { + "type": "text", + "bbox": [ + 107, + 175, + 503, + 198 + ], + "lines": [], + "index": 8.5, + "bbox_fs": [ + 106, + 176, + 504, + 199 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 201, + 505, + 257 + ], + "lines": [ + { + "bbox": [ + 106, + 202, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 202, + 505, + 214 + ], + "score": 1.0, + "content": "Response curve visualization We plot the averaged response curves which are constructed by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 212, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 505, + 225 + ], + "score": 1.0, + "content": "equidistantly sampling action from the action space and averaging the feedback of the states in the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "dataset as the averaged response. Parts of the results in Fig. 4 (all curves can be seen in Appx. H.2).", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 234, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 506, + 248 + ], + "score": 1.0, + "content": "For those tasks where baselines fail in reconstructing response curves, GALILEO not only reaches a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 244, + 342, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 244, + 342, + 258 + ], + "score": 1.0, + "content": "better MISE score but reconstructs almost exact responses.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 202, + 506, + 258 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 260, + 504, + 294 + ], + "lines": [ + { + "bbox": [ + 106, + 260, + 505, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 505, + 273 + ], + "score": 1.0, + "content": "Ablation studies In Sec. 4.3, we introduce several techniques to develop a practical GALILEO", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 211, + 284 + ], + "score": 1.0, + "content": "algorithm. Based on task", + "type": "text" + }, + { + "bbox": [ + 212, + 272, + 272, + 283 + ], + "score": 0.85, + "content": "\\mathtt { e 0 } . 2 _ { - \\mathtt { p } 0 . 0 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 272, + 505, + 284 + ], + "score": 1.0, + "content": "of GNFC, we give the ablation studies to investigate the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 282, + 496, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 496, + 296 + ], + "score": 1.0, + "content": "effects of these techniques. As the main-body space is limited, we leave the results in Appx. H.3.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 106, + 260, + 505, + 296 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 299, + 283, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 298, + 284, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 284, + 312 + ], + "score": 1.0, + "content": "5.2 TEST IN COMPLEX ENVIRONMENTS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 313, + 505, + 424 + ], + "lines": [ + { + "bbox": [ + 106, + 314, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 505, + 325 + ], + "score": 1.0, + "content": "In MuJoCo tasks MuJoCo is a benchmark task in Gym (Todorov et al., 2012; Brockman et al.,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 324, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 505, + 338 + ], + "score": 1.0, + "content": "2016) where we need to control a robot with specific dynamics to complete some tasks (e.g., standing", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 334, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 104, + 334, + 506, + 351 + ], + "score": 1.0, + "content": "or running). We select 3 environment from D4RL (Fu et al., 2020) to construct our model learning", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 347, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 506, + 360 + ], + "score": 1.0, + "content": "tasks. We compare it with a standard transition model learning algorithm used in the previous", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 357, + 506, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 506, + 370 + ], + "score": 1.0, + "content": "offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020), which is a variant of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 369, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 505, + 381 + ], + "score": 1.0, + "content": "supervised learning. We name the method OFF-SL. Besides, we also implement IPW and SCIGAN", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 378, + 507, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 507, + 393 + ], + "score": 1.0, + "content": "as the baselines. In D4RL benchmark, only the “medium” tasks is collected with a fixed policy, i.e.,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "the behavior policy is with 1/3 performance to the expert policy), which is most matching to our", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 401, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 506, + 415 + ], + "score": 1.0, + "content": "proposed problem. So we train models in datasets HalfCheetah-medium, Walker2d-medium, and", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 412, + 474, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 474, + 425 + ], + "score": 1.0, + "content": "Hopper-medium. We trained the models with the same gradient steps and saved the models.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23.5, + "bbox_fs": [ + 104, + 314, + 507, + 425 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 429, + 505, + 518 + ], + "lines": [ + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 505, + 441 + ], + "score": 1.0, + "content": "We first verify the generalization ability of the models by adopting them into offline model-based", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "RL. Instead of designing sophisticated tricks to suppress policy exploration and learning in risky", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 106, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "regions as current offline model-based RL algorithms (Yu et al., 2020; Kidambi et al., 2020) do, we", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 505, + 475 + ], + "score": 1.0, + "content": "just use the standard SAC algorithm Haarnoja et al. (2018) to exploit the models for policy learning", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 473, + 505, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 505, + 486 + ], + "score": 1.0, + "content": "to strictly verify the ability of the models. Unfortunately, we found that the compounding error will", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 505, + 496 + ], + "score": 1.0, + "content": "still be inevitably large in the 1,000-step rollout, which is the standard horizon in MuJoCo tasks,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 507 + ], + "score": 1.0, + "content": "leading all models to fail to derive a reasonable policy. To better verify the effects of models on policy", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 505, + 488, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 413, + 519 + ], + "score": 1.0, + "content": "optimization, we learn and evaluate the policies with three smaller horizons:", + "type": "text" + }, + { + "bbox": [ + 414, + 507, + 484, + 519 + ], + "score": 0.91, + "content": "H \\in \\{ 1 0 , 2 0 , 4 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 505, + 488, + 519 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 429, + 505, + 519 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 522, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 523, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 505, + 534 + ], + "score": 1.0, + "content": "The results are listed in Tab. 1. We first averaged the normalized return (refer to “avg. norm.”)", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "under each task, and we can see that the policy obtained by GALILEO is significantly higher than", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 545, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 258, + 556 + ], + "score": 1.0, + "content": "other models (the improvements are", + "type": "text" + }, + { + "bbox": [ + 259, + 545, + 279, + 556 + ], + "score": 0.86, + "content": "24 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 545, + 292, + 556 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 292, + 545, + 317, + 556 + ], + "score": 0.87, + "content": "161 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 545, + 505, + 556 + ], + "score": 1.0, + "content": "). At the same time, we found that SCIGAN", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 568 + ], + "score": 1.0, + "content": "performed better in policy learning, while IPW performed similarly to SL. This is in line with", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 566, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 506, + 581 + ], + "score": 1.0, + "content": "our expectations, since IPW only considers the uniform policy as the target policy for debiasing,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 578, + 504, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 504, + 590 + ], + "score": 1.0, + "content": "while policy optimization requires querying a wide variety of policies. Minimizing the prediction", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 588, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 506, + 601 + ], + "score": 1.0, + "content": "risks only under a uniform policy cannot yield a good environment model for policy optimization.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "Besides, in IPW, the cumulative effects of policy on the state distribution are ignored. On the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "other hand, SCIGAN, as a partial implementation of GALILEO (refer to Appx. E.2), also roughly", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 505, + 633 + ], + "score": 1.0, + "content": "achieves AWRM and considers the cumulative effects of policy on the state distribution, so its overall", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 633, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 505, + 644 + ], + "score": 1.0, + "content": "performance is better; In addition, we find that GALILEO achieves significant improvement in 6 of", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 644, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 505, + 655 + ], + "score": 1.0, + "content": "the 9 tasks. But in HalfCheetah, IPW works slightly better. However, compared with MAX-RETURN,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "it can be found that all methods fail to derive reasonable policies because their policies’ performances", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "are far away from the optimal policy. By further visualizing the trajectories, we found that all the", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "learned policies just keep the cheetah standing in the same place or even going backward. This", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "phenomenon is also similar to the results in MOPO (Yu et al., 2020). In MOPO’s experiment in", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "the medium datasets, the truncated-rollout horizon used in Walker and Hopper for policy training", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "is set to 5, while HalfCheetah has to be set to the minimal value: 1. These phenomena indicate", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "that HalfCheetah may still have unknown problems, resulting in the generalization bottleneck of", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 213, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 228 + ], + "score": 1.0, + "content": "the models. Besides, we also test the prediction error of the learned model in corresponding unseen", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 225, + 420, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 420, + 239 + ], + "score": 1.0, + "content": "“expert” and “medium-replay” datasets. The detailed results are in Appx. H.5.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + } + ], + "index": 46, + "bbox_fs": [ + 105, + 523, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 44, + 489, + 154 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 44, + 489, + 154 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 44, + 489, + 154 + ], + "spans": [ + { + "bbox": [ + 119, + 44, + 489, + 154 + ], + "score": 0.972, + "type": "image", + "image_path": "48ff8ddb26f31d38f842a347595907ec3933c9a964af1fd467fd13b1ca24d29a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 44, + 489, + 80.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 80.66666666666666, + 489, + 117.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 117.33333333333331, + 489, + 153.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 156, + 506, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 156, + 504, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 156, + 504, + 168 + ], + "score": 1.0, + "content": "Figure 5: An illustration of the performance in BAT tasks. In Fig. 5(a) demonstrate the averaged", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 167, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 506, + 180 + ], + "score": 1.0, + "content": "response curves of the SL and GALILEO model in City A. It is expected to be monotonically", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "score": 1.0, + "content": "increasing through our prior knowledge. In Fig. 5(b), the model with larger areas above the “random”", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 189, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 506, + 202 + ], + "score": 1.0, + "content": "line makes better predictions in randomized-controlled-trials data (Betlei et al., 2020). Fig. 5(c)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 199, + 472, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 472, + 213 + ], + "score": 1.0, + "content": "shows the daily responses in the A/B test in City A. The complete results are in Appx. H.6.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 215, + 503, + 237 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 228 + ], + "score": 1.0, + "content": "the models. Besides, we also test the prediction error of the learned model in corresponding unseen", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 225, + 420, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 420, + 239 + ], + "score": 1.0, + "content": "“expert” and “medium-replay” datasets. The detailed results are in Appx. H.5.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 244, + 505, + 365 + ], + "lines": [ + { + "bbox": [ + 105, + 243, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 505, + 257 + ], + "score": 1.0, + "content": "In a real-world platform We finally deploy GALILEO in a real-world large-scale food-delivery", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 255, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 505, + 267 + ], + "score": 1.0, + "content": "platform. The goal of the platform is to balance the demand from take-out food orders and the supply", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 266, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 505, + 278 + ], + "score": 1.0, + "content": "of delivery clerks, i.e., helping delivery clerks fulfill more orders by giving reasonable strategies. We", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "focus on a Budget Allocation task to the Time period (BAT) in the platform (see Appx. G.1.3 for", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 288, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 300 + ], + "score": 1.0, + "content": "details). The goal of the BAT task is to handle the imbalance problem between the demanded orders", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "from customers and the supply of delivery clerks in different time periods by allocating reasonable", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 309, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 505, + 322 + ], + "score": 1.0, + "content": "allowances to those time periods. The core challenge of the environment model learning in BAT tasks", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 322, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 505, + 333 + ], + "score": 1.0, + "content": "is similar to the challenge in Fig. 1. Specifically, the behavior policy in BAT tasks is a human-expert", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 331, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 506, + 345 + ], + "score": 1.0, + "content": "policy, which tends to increase the budget of allowance in the time periods with a lower supply of", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 343, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 505, + 355 + ], + "score": 1.0, + "content": "delivery clerks, otherwise tends to decrease the budget (Fig. 12 gives a real-data instance of this", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 354, + 165, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 165, + 365 + ], + "score": 1.0, + "content": "phenomenon).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 371, + 505, + 459 + ], + "lines": [ + { + "bbox": [ + 106, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "We first learn a model to predict the supply of delivery clerks (measured by fulfilled order amount)", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 381, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 394 + ], + "score": 1.0, + "content": "on given allowances. Although the SL model can efficiently fit the offline data, the tendency of the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 392, + 507, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 507, + 406 + ], + "score": 1.0, + "content": "response curve is easily to be incorrect. As can be seen in Fig. 5(a), with a larger budget of allowance,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 403, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 506, + 416 + ], + "score": 1.0, + "content": "the prediction of the supply is decreased in SL, which obviously goes against our prior knowledge.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 414, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 426 + ], + "score": 1.0, + "content": "This is because, in the offline dataset, the corresponding supply will be smaller when the allowance", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "is larger. It is conceivable that if we learn a policy through the model of SL, the optimal solution", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 437, + 504, + 448 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 504, + 448 + ], + "score": 1.0, + "content": "is canceling all of the allowances, which is obviously incorrect in practice. On the other hand, the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 447, + 439, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 439, + 460 + ], + "score": 1.0, + "content": "tendency of GALILEO’s response is correct. Fig. 13 plots all the results in 6 cities.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 107, + 464, + 504, + 520 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 476 + ], + "score": 1.0, + "content": "Second, we conduct randomized controlled trials (RCT) in one of the testing cities. Using the RCT", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "samples, we can evaluate the generalization ability of the model predictions via Area Under the Uplift", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "score": 1.0, + "content": "Curve (AUUC) (Betlei et al., 2020), which measure the correctness of the sort order of the model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 496, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 506, + 511 + ], + "score": 1.0, + "content": "prediction in RCT samples. The AUUC further show that GALILEO gives a reasonable sort order on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 508, + 490, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 490, + 521 + ], + "score": 1.0, + "content": "the supply prediction (see Fig. 5(b)) while the standard SL technique fails to complete this task.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 525, + 505, + 591 + ], + "lines": [ + { + "bbox": [ + 106, + 525, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 506, + 538 + ], + "score": 1.0, + "content": "Finally, we search for the optimal policy via the cross-entropy method planner (Hafner et al., 2019)", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 536, + 506, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 506, + 549 + ], + "score": 1.0, + "content": "based on the learned model and deploy the policy in a real-world platform. The results of A/B test in", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 545, + 506, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 506, + 561 + ], + "score": 1.0, + "content": "City A is shown in Fig. 5(c). It can be seen that after the day of the A/B test, the treatment group", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 558, + 506, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 506, + 571 + ], + "score": 1.0, + "content": "(deploying our policy) significant improve the five-minute order-taken rate than the baseline policy", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 568, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 506, + 582 + ], + "score": 1.0, + "content": "(the same as the behavior policy). In summary, the policy improves the supply from 0.14 to 1.63", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 580, + 507, + 593 + ], + "spans": [ + { + "bbox": [ + 104, + 580, + 507, + 593 + ], + "score": 1.0, + "content": "percentage points to the behavior policies in the 6 cities. The details of these results are in Appx. H.6.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5 + }, + { + "type": "title", + "bbox": [ + 108, + 599, + 297, + 612 + ], + "lines": [ + { + "bbox": [ + 105, + 597, + 299, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 299, + 614 + ], + "score": 1.0, + "content": "6 DISCUSSION AND FUTURE WORK", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 616, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "In this work, we propose AWRM which handles the generalization challenges of the counterfactual", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "environment model learning. By theoretical modeling, we give a tractable solution to handle AWRM", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 638, + 507, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 507, + 651 + ], + "score": 1.0, + "content": "and propose GALILEO. GALILEO is verified in synthetic environments, complex robot control tasks,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 649, + 472, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 472, + 662 + ], + "score": 1.0, + "content": "and a real-world platform, and shows great generalization ability on counterfactual queries.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5 + }, + { + "type": "text", + "bbox": [ + 107, + 666, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "Giving correct answers to counterfactual queries is important for policy learning. We hope the work", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 104, + 675, + 507, + 691 + ], + "spans": [ + { + "bbox": [ + 104, + 675, + 507, + 691 + ], + "score": 1.0, + "content": "can inspire researchers to develop more powerful tools for counterfactual environment model learning.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "The current limitation lies in: There are several simplifications in the theoretical modeling process", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "(further discussion is in Appx. B), which can be modeled more elaborately . Besides, experiments on", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "MuJoCo indicate that these tasks are still challenging to give correct predictions on counterfactual", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 720, + 356, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 356, + 732 + ], + "score": 1.0, + "content": "data. These should also be further investigated in future work.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 47.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 513, + 448, + 585, + 471 + ], + "lines": [ + { + "bbox": [ + 513, + 447, + 586, + 460 + ], + "spans": [ + { + "bbox": [ + 513, + 447, + 586, + 460 + ], + "score": 1.0, + "content": "to Reviewer rQ79", + "type": "text" + } + ] + }, + { + "bbox": [ + 512, + 457, + 552, + 473 + ], + "spans": [ + { + "bbox": [ + 512, + 457, + 552, + 473 + ], + "score": 1.0, + "content": "and j5p5", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 119, + 44, + 489, + 154 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 119, + 44, + 489, + 154 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 44, + 489, + 154 + ], + "spans": [ + { + "bbox": [ + 119, + 44, + 489, + 154 + ], + "score": 0.972, + "type": "image", + "image_path": "48ff8ddb26f31d38f842a347595907ec3933c9a964af1fd467fd13b1ca24d29a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 119, + 44, + 489, + 80.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 119, + 80.66666666666666, + 489, + 117.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 119, + 117.33333333333331, + 489, + 153.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 156, + 506, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 156, + 504, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 156, + 504, + 168 + ], + "score": 1.0, + "content": "Figure 5: An illustration of the performance in BAT tasks. In Fig. 5(a) demonstrate the averaged", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 167, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 506, + 180 + ], + "score": 1.0, + "content": "response curves of the SL and GALILEO model in City A. It is expected to be monotonically", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "score": 1.0, + "content": "increasing through our prior knowledge. In Fig. 5(b), the model with larger areas above the “random”", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 189, + 506, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 506, + 202 + ], + "score": 1.0, + "content": "line makes better predictions in randomized-controlled-trials data (Betlei et al., 2020). Fig. 5(c)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 199, + 472, + 213 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 472, + 213 + ], + "score": 1.0, + "content": "shows the daily responses in the A/B test in City A. The complete results are in Appx. H.6.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 215, + 503, + 237 + ], + "lines": [], + "index": 8.5, + "bbox_fs": [ + 105, + 213, + 505, + 239 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 244, + 505, + 365 + ], + "lines": [ + { + "bbox": [ + 105, + 243, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 505, + 257 + ], + "score": 1.0, + "content": "In a real-world platform We finally deploy GALILEO in a real-world large-scale food-delivery", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 255, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 505, + 267 + ], + "score": 1.0, + "content": "platform. The goal of the platform is to balance the demand from take-out food orders and the supply", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 266, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 505, + 278 + ], + "score": 1.0, + "content": "of delivery clerks, i.e., helping delivery clerks fulfill more orders by giving reasonable strategies. We", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 505, + 289 + ], + "score": 1.0, + "content": "focus on a Budget Allocation task to the Time period (BAT) in the platform (see Appx. G.1.3 for", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 288, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 300 + ], + "score": 1.0, + "content": "details). The goal of the BAT task is to handle the imbalance problem between the demanded orders", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "from customers and the supply of delivery clerks in different time periods by allocating reasonable", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 309, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 505, + 322 + ], + "score": 1.0, + "content": "allowances to those time periods. The core challenge of the environment model learning in BAT tasks", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 322, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 505, + 333 + ], + "score": 1.0, + "content": "is similar to the challenge in Fig. 1. Specifically, the behavior policy in BAT tasks is a human-expert", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 331, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 506, + 345 + ], + "score": 1.0, + "content": "policy, which tends to increase the budget of allowance in the time periods with a lower supply of", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 343, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 505, + 355 + ], + "score": 1.0, + "content": "delivery clerks, otherwise tends to decrease the budget (Fig. 12 gives a real-data instance of this", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 354, + 165, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 165, + 365 + ], + "score": 1.0, + "content": "phenomenon).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 243, + 506, + 365 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 371, + 505, + 459 + ], + "lines": [ + { + "bbox": [ + 106, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "We first learn a model to predict the supply of delivery clerks (measured by fulfilled order amount)", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 381, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 394 + ], + "score": 1.0, + "content": "on given allowances. Although the SL model can efficiently fit the offline data, the tendency of the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 392, + 507, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 507, + 406 + ], + "score": 1.0, + "content": "response curve is easily to be incorrect. As can be seen in Fig. 5(a), with a larger budget of allowance,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 403, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 506, + 416 + ], + "score": 1.0, + "content": "the prediction of the supply is decreased in SL, which obviously goes against our prior knowledge.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 414, + 505, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 426 + ], + "score": 1.0, + "content": "This is because, in the offline dataset, the corresponding supply will be smaller when the allowance", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 437 + ], + "score": 1.0, + "content": "is larger. It is conceivable that if we learn a policy through the model of SL, the optimal solution", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 437, + 504, + 448 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 504, + 448 + ], + "score": 1.0, + "content": "is canceling all of the allowances, which is obviously incorrect in practice. On the other hand, the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 447, + 439, + 460 + ], + "spans": [ + { + "bbox": [ + 106, + 447, + 439, + 460 + ], + "score": 1.0, + "content": "tendency of GALILEO’s response is correct. Fig. 13 plots all the results in 6 cities.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 370, + 507, + 460 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 464, + 504, + 520 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 476 + ], + "score": 1.0, + "content": "Second, we conduct randomized controlled trials (RCT) in one of the testing cities. Using the RCT", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "samples, we can evaluate the generalization ability of the model predictions via Area Under the Uplift", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "score": 1.0, + "content": "Curve (AUUC) (Betlei et al., 2020), which measure the correctness of the sort order of the model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 496, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 506, + 511 + ], + "score": 1.0, + "content": "prediction in RCT samples. The AUUC further show that GALILEO gives a reasonable sort order on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 508, + 490, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 490, + 521 + ], + "score": 1.0, + "content": "the supply prediction (see Fig. 5(b)) while the standard SL technique fails to complete this task.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 464, + 506, + 521 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 525, + 505, + 591 + ], + "lines": [ + { + "bbox": [ + 106, + 525, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 525, + 506, + 538 + ], + "score": 1.0, + "content": "Finally, we search for the optimal policy via the cross-entropy method planner (Hafner et al., 2019)", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 536, + 506, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 506, + 549 + ], + "score": 1.0, + "content": "based on the learned model and deploy the policy in a real-world platform. The results of A/B test in", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 545, + 506, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 506, + 561 + ], + "score": 1.0, + "content": "City A is shown in Fig. 5(c). It can be seen that after the day of the A/B test, the treatment group", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 558, + 506, + 571 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 506, + 571 + ], + "score": 1.0, + "content": "(deploying our policy) significant improve the five-minute order-taken rate than the baseline policy", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 568, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 506, + 582 + ], + "score": 1.0, + "content": "(the same as the behavior policy). In summary, the policy improves the supply from 0.14 to 1.63", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 580, + 507, + 593 + ], + "spans": [ + { + "bbox": [ + 104, + 580, + 507, + 593 + ], + "score": 1.0, + "content": "percentage points to the behavior policies in the 6 cities. The details of these results are in Appx. H.6.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5, + "bbox_fs": [ + 104, + 525, + 507, + 593 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 599, + 297, + 612 + ], + "lines": [ + { + "bbox": [ + 105, + 597, + 299, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 299, + 614 + ], + "score": 1.0, + "content": "6 DISCUSSION AND FUTURE WORK", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 616, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "In this work, we propose AWRM which handles the generalization challenges of the counterfactual", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "environment model learning. By theoretical modeling, we give a tractable solution to handle AWRM", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 638, + 507, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 507, + 651 + ], + "score": 1.0, + "content": "and propose GALILEO. GALILEO is verified in synthetic environments, complex robot control tasks,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 649, + 472, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 472, + 662 + ], + "score": 1.0, + "content": "and a real-world platform, and shows great generalization ability on counterfactual queries.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 616, + 507, + 662 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 666, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "score": 1.0, + "content": "Giving correct answers to counterfactual queries is important for policy learning. We hope the work", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 104, + 675, + 507, + 691 + ], + "spans": [ + { + "bbox": [ + 104, + 675, + 507, + 691 + ], + "score": 1.0, + "content": "can inspire researchers to develop more powerful tools for counterfactual environment model learning.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "The current limitation lies in: There are several simplifications in the theoretical modeling process", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "(further discussion is in Appx. B), which can be modeled more elaborately . Besides, experiments on", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "MuJoCo indicate that these tasks are still challenging to give correct predictions on counterfactual", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 720, + 356, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 356, + 732 + ], + "score": 1.0, + "content": "data. These should also be further investigated in future work.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 47.5, + "bbox_fs": [ + 104, + 665, + 507, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 175, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 506, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 507, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 507, + 113 + ], + "score": 1.0, + "content": "Ahmed M. Alaa and Mihaela van der Schaar. Limits of estimating heterogeneous treatment effects:", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 111, + 506, + 125 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 506, + 125 + ], + "score": 1.0, + "content": "Guidelines for practical algorithm design. In Proceedings of the 35th International Conference on", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 114, + 122, + 407, + 135 + ], + "spans": [ + { + "bbox": [ + 114, + 122, + 407, + 135 + ], + "score": 1.0, + "content": "Machine Learning, volume 80, pp. 129–138, Stockholm, Sweden, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 142, + 504, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "Syed Mumtaz Ali and Samuel D Silvey. A general class of coefficients of divergence of one", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 505, + 166 + ], + "score": 1.0, + "content": "distribution from another. Journal of the Royal Statistical Society: Series B (Methodological), 28", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 165, + 195, + 176 + ], + "spans": [ + { + "bbox": [ + 116, + 165, + 195, + 176 + ], + "score": 1.0, + "content": "(1):131–142, 1966.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 184, + 505, + 239 + ], + "lines": [ + { + "bbox": [ + 106, + 185, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 506, + 196 + ], + "score": 1.0, + "content": "Serge Assaad, Shuxi Zeng, Chenyang Tao, Shounak Datta, Nikhil Mehta, Ricardo Henao, Fan Li,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 195, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 195, + 505, + 208 + ], + "score": 1.0, + "content": "and Lawrence Carin. Counterfactual representation learning with balancing weights. In Arindam", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 206, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 115, + 206, + 505, + 219 + ], + "score": 1.0, + "content": "Banerjee and Kenji Fukumizu (eds.), The 24th International Conference on Artificial Intelligence", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 115, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "and Statistics, volume 130 of Proceedings of Machine Learning Research, pp. 1972–1980, Virtual", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 228, + 201, + 240 + ], + "spans": [ + { + "bbox": [ + 115, + 228, + 201, + 240 + ], + "score": 1.0, + "content": "Event, 2021. PMLR.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 248, + 505, + 292 + ], + "lines": [ + { + "bbox": [ + 105, + 247, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 506, + 261 + ], + "score": 1.0, + "content": "Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. Analysis of representations for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 258, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 115, + 258, + 506, + 272 + ], + "score": 1.0, + "content": "domain adaptation. In Bernhard Schölkopf, John C. Platt, and Thomas Hofmann (eds.), Advances", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 270, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 114, + 270, + 506, + 282 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems 19, pp. 137–144, British Columbia, Canada, 2006. MIT", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 281, + 143, + 294 + ], + "spans": [ + { + "bbox": [ + 115, + 281, + 143, + 294 + ], + "score": 1.0, + "content": "Press.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 300, + 504, + 324 + ], + "lines": [ + { + "bbox": [ + 107, + 301, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 107, + 301, + 505, + 313 + ], + "score": 1.0, + "content": "Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wortman", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 312, + 495, + 324 + ], + "spans": [ + { + "bbox": [ + 116, + 312, + 495, + 324 + ], + "score": 1.0, + "content": "Vaughan. A theory of learning from different domains. Mach. Learn., 79(1-2):151–175, 2010.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 105, + 332, + 505, + 354 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "Artem Betlei, Eustache Diemert, and Massih-Reza Amini. Treatment targeting by AUUC maximiza-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 343, + 384, + 355 + ], + "spans": [ + { + "bbox": [ + 115, + 343, + 384, + 355 + ], + "score": 1.0, + "content": "tion with generalization guarantees. CoRR, abs/2012.09897, 2020.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 107, + 362, + 505, + 396 + ], + "lines": [ + { + "bbox": [ + 106, + 363, + 504, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 363, + 504, + 375 + ], + "score": 1.0, + "content": "Ioana Bica, James Jordon, and Mihaela van der Schaar. Estimating the effects of continuous-valued", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 373, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 115, + 373, + 506, + 388 + ], + "score": 1.0, + "content": "interventions using generative adversarial networks. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 385, + 244, + 396 + ], + "spans": [ + { + "bbox": [ + 115, + 385, + 244, + 396 + ], + "score": 1.0, + "content": "Systems 33, virtual event, 2020.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 405, + 504, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 505, + 417 + ], + "score": 1.0, + "content": "Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 415, + 375, + 430 + ], + "spans": [ + { + "bbox": [ + 115, + 415, + 375, + 430 + ], + "score": 1.0, + "content": "Wojciech Zaremba. Openai gym. CoRR, abs/1606.01540, 2016.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 506, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "score": 1.0, + "content": "Lars Buesing, Theophane Weber, Yori Zwols, Nicolas Heess, Sébastien Racanière, Arthur Guez, and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 447, + 504, + 460 + ], + "spans": [ + { + "bbox": [ + 116, + 447, + 489, + 460 + ], + "score": 1.0, + "content": "Jean-Baptiste Lespiau. Woulda, coulda, shoulda: Counterfactually-guided policy search. In", + "type": "text" + }, + { + "bbox": [ + 489, + 447, + 504, + 457 + ], + "score": 0.25, + "content": "7 t h", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 458, + 440, + 470 + ], + "spans": [ + { + "bbox": [ + 116, + 458, + 440, + 470 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, New Orleans, LA, 2019.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 478, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 477, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 491 + ], + "score": 1.0, + "content": "Jonathon Byrd and Zachary Chase Lipton. What is the effect of importance weighting in deep", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 489, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 115, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "learning? In Kamalika Chaudhuri and Ruslan Salakhutdinov (eds.), Proceedings of the 36th", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 499, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 115, + 499, + 505, + 514 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 511, + 342, + 523 + ], + "spans": [ + { + "bbox": [ + 115, + 511, + 342, + 523 + ], + "score": 1.0, + "content": "Research, pp. 872–881, Long Beach, CA, 2019. PMLR.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 564 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 506, + 544 + ], + "score": 1.0, + "content": "Xinshi Chen, Shuang Li, Hui Li, Shaohua Jiang, Yuan Qi, and Le Song. Generative adversarial", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 115, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "user model for reinforcement learning based recommendation system. In Proceedings of the 36th", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 552, + 473, + 566 + ], + "spans": [ + { + "bbox": [ + 115, + 552, + 473, + 566 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, pp. 1052–1061, Long Beach, CA, 2019.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 606 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 586 + ], + "score": 1.0, + "content": "Xiong-Hui Chen, Yang Yu, Qingyang Li, Fan-Ming Luo, Zhiwei Tony Qin, Shang Wenjie, and", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 584, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 115, + 584, + 505, + 596 + ], + "score": 1.0, + "content": "Jieping Ye. Offline model-based adaptable policy learning. In Advances in Neural Information", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 595, + 374, + 606 + ], + "spans": [ + { + "bbox": [ + 115, + 595, + 374, + 606 + ], + "score": 1.0, + "content": "Processing Systems 34 (NeurIPS’21), Virtual Conference, 2021.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 106, + 614, + 505, + 659 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 506, + 628 + ], + "score": 1.0, + "content": "Corinna Cortes, Yishay Mansour, and Mehryar Mohri. Learning bounds for importance weighting.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 626, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 115, + 626, + 506, + 638 + ], + "score": 1.0, + "content": "In John D. Lafferty, Christopher K. I. Williams, John Shawe-Taylor, Richard S. Zemel, and Aron", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 636, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 116, + 636, + 505, + 649 + ], + "score": 1.0, + "content": "Culotta (eds.), Advances in Neural Information Processing Systems 23, pp. 442–450, British", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 648, + 316, + 660 + ], + "spans": [ + { + "bbox": [ + 116, + 648, + 316, + 660 + ], + "score": 1.0, + "content": "Columbia, Canada, 2010. Curran Associates, Inc.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 103, + 667, + 504, + 690 + ], + "lines": [ + { + "bbox": [ + 104, + 666, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 505, + 681 + ], + "score": 1.0, + "content": "Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4RL: datasets for deep", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 678, + 383, + 691 + ], + "spans": [ + { + "bbox": [ + 116, + 678, + 383, + 691 + ], + "score": 1.0, + "content": "data-driven reinforcement learning. CoRR, abs/2004.07219, 2020.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Aaron C. Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in Neural", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 721, + 428, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 428, + 733 + ], + "score": 1.0, + "content": "Information Processing Systems 27, pp. 2672–2680, Montréal, Canada, 2014.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 81, + 175, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 506, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 100, + 507, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 100, + 507, + 113 + ], + "score": 1.0, + "content": "Ahmed M. Alaa and Mihaela van der Schaar. Limits of estimating heterogeneous treatment effects:", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 111, + 506, + 125 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 506, + 125 + ], + "score": 1.0, + "content": "Guidelines for practical algorithm design. In Proceedings of the 35th International Conference on", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 114, + 122, + 407, + 135 + ], + "spans": [ + { + "bbox": [ + 114, + 122, + 407, + 135 + ], + "score": 1.0, + "content": "Machine Learning, volume 80, pp. 129–138, Stockholm, Sweden, 2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 100, + 507, + 135 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 142, + 504, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 155 + ], + "score": 1.0, + "content": "Syed Mumtaz Ali and Samuel D Silvey. A general class of coefficients of divergence of one", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 505, + 166 + ], + "score": 1.0, + "content": "distribution from another. Journal of the Royal Statistical Society: Series B (Methodological), 28", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 165, + 195, + 176 + ], + "spans": [ + { + "bbox": [ + 116, + 165, + 195, + 176 + ], + "score": 1.0, + "content": "(1):131–142, 1966.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 106, + 142, + 505, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 184, + 505, + 239 + ], + "lines": [ + { + "bbox": [ + 106, + 185, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 506, + 196 + ], + "score": 1.0, + "content": "Serge Assaad, Shuxi Zeng, Chenyang Tao, Shounak Datta, Nikhil Mehta, Ricardo Henao, Fan Li,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 195, + 505, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 195, + 505, + 208 + ], + "score": 1.0, + "content": "and Lawrence Carin. Counterfactual representation learning with balancing weights. In Arindam", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 206, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 115, + 206, + 505, + 219 + ], + "score": 1.0, + "content": "Banerjee and Kenji Fukumizu (eds.), The 24th International Conference on Artificial Intelligence", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 115, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "and Statistics, volume 130 of Proceedings of Machine Learning Research, pp. 1972–1980, Virtual", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 228, + 201, + 240 + ], + "spans": [ + { + "bbox": [ + 115, + 228, + 201, + 240 + ], + "score": 1.0, + "content": "Event, 2021. PMLR.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9, + "bbox_fs": [ + 106, + 185, + 506, + 240 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 248, + 505, + 292 + ], + "lines": [ + { + "bbox": [ + 105, + 247, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 506, + 261 + ], + "score": 1.0, + "content": "Shai Ben-David, John Blitzer, Koby Crammer, and Fernando Pereira. Analysis of representations for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 258, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 115, + 258, + 506, + 272 + ], + "score": 1.0, + "content": "domain adaptation. In Bernhard Schölkopf, John C. Platt, and Thomas Hofmann (eds.), Advances", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 270, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 114, + 270, + 506, + 282 + ], + "score": 1.0, + "content": "in Neural Information Processing Systems 19, pp. 137–144, British Columbia, Canada, 2006. MIT", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 281, + 143, + 294 + ], + "spans": [ + { + "bbox": [ + 115, + 281, + 143, + 294 + ], + "score": 1.0, + "content": "Press.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 247, + 506, + 294 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 300, + 504, + 324 + ], + "lines": [ + { + "bbox": [ + 107, + 301, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 107, + 301, + 505, + 313 + ], + "score": 1.0, + "content": "Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wortman", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 312, + 495, + 324 + ], + "spans": [ + { + "bbox": [ + 116, + 312, + 495, + 324 + ], + "score": 1.0, + "content": "Vaughan. A theory of learning from different domains. Mach. Learn., 79(1-2):151–175, 2010.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 107, + 301, + 505, + 324 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 332, + 505, + 354 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "Artem Betlei, Eustache Diemert, and Massih-Reza Amini. Treatment targeting by AUUC maximiza-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 343, + 384, + 355 + ], + "spans": [ + { + "bbox": [ + 115, + 343, + 384, + 355 + ], + "score": 1.0, + "content": "tion with generalization guarantees. CoRR, abs/2012.09897, 2020.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 106, + 332, + 506, + 355 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 362, + 505, + 396 + ], + "lines": [ + { + "bbox": [ + 106, + 363, + 504, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 363, + 504, + 375 + ], + "score": 1.0, + "content": "Ioana Bica, James Jordon, and Mihaela van der Schaar. Estimating the effects of continuous-valued", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 373, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 115, + 373, + 506, + 388 + ], + "score": 1.0, + "content": "interventions using generative adversarial networks. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 385, + 244, + 396 + ], + "spans": [ + { + "bbox": [ + 115, + 385, + 244, + 396 + ], + "score": 1.0, + "content": "Systems 33, virtual event, 2020.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 363, + 506, + 396 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 405, + 504, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 505, + 417 + ], + "score": 1.0, + "content": "Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 415, + 375, + 430 + ], + "spans": [ + { + "bbox": [ + 115, + 415, + 375, + 430 + ], + "score": 1.0, + "content": "Wojciech Zaremba. Openai gym. CoRR, abs/1606.01540, 2016.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 405, + 505, + 430 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 506, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "score": 1.0, + "content": "Lars Buesing, Theophane Weber, Yori Zwols, Nicolas Heess, Sébastien Racanière, Arthur Guez, and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 447, + 504, + 460 + ], + "spans": [ + { + "bbox": [ + 116, + 447, + 489, + 460 + ], + "score": 1.0, + "content": "Jean-Baptiste Lespiau. Woulda, coulda, shoulda: Counterfactually-guided policy search. In", + "type": "text" + }, + { + "bbox": [ + 489, + 447, + 504, + 457 + ], + "score": 0.25, + "content": "7 t h", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 458, + 440, + 470 + ], + "spans": [ + { + "bbox": [ + 116, + 458, + 440, + 470 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, New Orleans, LA, 2019.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 435, + 506, + 470 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 478, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 477, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 491 + ], + "score": 1.0, + "content": "Jonathon Byrd and Zachary Chase Lipton. What is the effect of importance weighting in deep", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 489, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 115, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "learning? In Kamalika Chaudhuri and Ruslan Salakhutdinov (eds.), Proceedings of the 36th", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 499, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 115, + 499, + 505, + 514 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 511, + 342, + 523 + ], + "spans": [ + { + "bbox": [ + 115, + 511, + 342, + 523 + ], + "score": 1.0, + "content": "Research, pp. 872–881, Long Beach, CA, 2019. PMLR.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 477, + 506, + 523 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 531, + 505, + 564 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 506, + 544 + ], + "score": 1.0, + "content": "Xinshi Chen, Shuang Li, Hui Li, Shaohua Jiang, Yuan Qi, and Le Song. Generative adversarial", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 115, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "user model for reinforcement learning based recommendation system. In Proceedings of the 36th", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 552, + 473, + 566 + ], + "spans": [ + { + "bbox": [ + 115, + 552, + 473, + 566 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, pp. 1052–1061, Long Beach, CA, 2019.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 531, + 506, + 566 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 606 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 586 + ], + "score": 1.0, + "content": "Xiong-Hui Chen, Yang Yu, Qingyang Li, Fan-Ming Luo, Zhiwei Tony Qin, Shang Wenjie, and", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 584, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 115, + 584, + 505, + 596 + ], + "score": 1.0, + "content": "Jieping Ye. Offline model-based adaptable policy learning. In Advances in Neural Information", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 595, + 374, + 606 + ], + "spans": [ + { + "bbox": [ + 115, + 595, + 374, + 606 + ], + "score": 1.0, + "content": "Processing Systems 34 (NeurIPS’21), Virtual Conference, 2021.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 572, + 506, + 606 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 614, + 505, + 659 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 506, + 628 + ], + "score": 1.0, + "content": "Corinna Cortes, Yishay Mansour, and Mehryar Mohri. Learning bounds for importance weighting.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 626, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 115, + 626, + 506, + 638 + ], + "score": 1.0, + "content": "In John D. Lafferty, Christopher K. I. Williams, John Shawe-Taylor, Richard S. Zemel, and Aron", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 636, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 116, + 636, + 505, + 649 + ], + "score": 1.0, + "content": "Culotta (eds.), Advances in Neural Information Processing Systems 23, pp. 442–450, British", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 648, + 316, + 660 + ], + "spans": [ + { + "bbox": [ + 116, + 648, + 316, + 660 + ], + "score": 1.0, + "content": "Columbia, Canada, 2010. Curran Associates, Inc.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 614, + 506, + 660 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 667, + 504, + 690 + ], + "lines": [ + { + "bbox": [ + 104, + 666, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 505, + 681 + ], + "score": 1.0, + "content": "Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4RL: datasets for deep", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 678, + 383, + 691 + ], + "spans": [ + { + "bbox": [ + 116, + 678, + 383, + 691 + ], + "score": 1.0, + "content": "data-driven reinforcement learning. CoRR, abs/2004.07219, 2020.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 104, + 666, + 505, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Aaron C. Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in Neural", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 721, + 428, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 428, + 733 + ], + "score": 1.0, + "content": "Information Processing Systems 27, pp. 2672–2680, Montréal, Canada, 2014.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 697, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 126 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft Actor-Critic: Off-policy", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "maximum entropy deep reinforcement learning with a stochastic actor. In Proceedings of the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 114, + 104, + 507, + 117 + ], + "spans": [ + { + "bbox": [ + 114, + 104, + 507, + 117 + ], + "score": 1.0, + "content": "35th International Conference on Machine Learning, pp. 1856–1865, Stockholmsmässan, Sweden,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 114, + 115, + 143, + 127 + ], + "spans": [ + { + "bbox": [ + 114, + 115, + 143, + 127 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 107, + 134, + 506, + 189 + ], + "lines": [ + { + "bbox": [ + 106, + 135, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 505, + 146 + ], + "score": 1.0, + "content": "Danijar Hafner, Timothy P. Lillicrap, Ian Fischer, Ruben Villegas, David Ha, Honglak Lee, and James", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 506, + 159 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 506, + 159 + ], + "score": 1.0, + "content": "Davidson. Learning latent dynamics for planning from pixels. In Kamalika Chaudhuri and Ruslan", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 154, + 507, + 171 + ], + "spans": [ + { + "bbox": [ + 115, + 154, + 507, + 171 + ], + "score": 1.0, + "content": "Salakhutdinov (eds.), Proceedings of the 36th International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 166, + 507, + 180 + ], + "spans": [ + { + "bbox": [ + 115, + 166, + 507, + 180 + ], + "score": 1.0, + "content": "volume 97 of Proceedings of Machine Learning Research, pp. 2555–2565, Long Beach, CA, 2019.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 178, + 149, + 191 + ], + "spans": [ + { + "bbox": [ + 115, + 178, + 149, + 191 + ], + "score": 1.0, + "content": "PMLR.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 198, + 506, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "score": 1.0, + "content": "Negar Hassanpour and Russell Greiner. Counterfactual regression with importance sampling weights.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 208, + 507, + 222 + ], + "spans": [ + { + "bbox": [ + 115, + 208, + 507, + 222 + ], + "score": 1.0, + "content": "In Proceedings of the 28th International Joint Conference on Artificial Intelligence, pp. 5880–5887,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 219, + 203, + 231 + ], + "spans": [ + { + "bbox": [ + 116, + 219, + 203, + 231 + ], + "score": 1.0, + "content": "Macao, China, 2019.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 238, + 505, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 238, + 505, + 251 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 505, + 251 + ], + "score": 1.0, + "content": "J.B. Hiriart-Urruty and C. Lemaréchal. Fundamentals of Convex Analysis. 2001. ISBN", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 250, + 186, + 262 + ], + "spans": [ + { + "bbox": [ + 116, + 250, + 186, + 262 + ], + "score": 1.0, + "content": "9783642564680.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 105, + 269, + 504, + 293 + ], + "lines": [ + { + "bbox": [ + 105, + 268, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 282 + ], + "score": 1.0, + "content": "Jonathan Ho and Stefano Ermon. Generative adversarial imitation learning. In Advances in Neural", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 280, + 424, + 293 + ], + "spans": [ + { + "bbox": [ + 116, + 280, + 424, + 293 + ], + "score": 1.0, + "content": "Information Processing Systems 29, pp. 4565–4573, Barcelona, Spain, 2016.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 299, + 504, + 322 + ], + "lines": [ + { + "bbox": [ + 105, + 298, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 506, + 313 + ], + "score": 1.0, + "content": "Guido Imbens. The role of the propensity score in estimating dose-response functions. Econometrics", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 311, + 182, + 322 + ], + "spans": [ + { + "bbox": [ + 116, + 311, + 182, + 322 + ], + "score": 1.0, + "content": "eJournal, 1999.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 106, + 330, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "score": 1.0, + "content": "Edward L Ionides. Truncated importance sampling. Journal of Computational and Graphical", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 340, + 246, + 353 + ], + "spans": [ + { + "bbox": [ + 115, + 340, + 246, + 353 + ], + "score": 1.0, + "content": "Statistics, 17(2):295–311, 2008.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 360, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 359, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 506, + 374 + ], + "score": 1.0, + "content": "Stephen James and Edward Johns. 3D simulation for robot arm control with deep Q-learning. CoRR,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 370, + 210, + 383 + ], + "spans": [ + { + "bbox": [ + 116, + 370, + 210, + 383 + ], + "score": 1.0, + "content": "abs/1609.03759, 2016.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 105, + 390, + 504, + 413 + ], + "lines": [ + { + "bbox": [ + 105, + 389, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 505, + 403 + ], + "score": 1.0, + "content": "Fredrik D Johansson, Nathan Kallus, Uri Shalit, and David Sontag. Learning weighted representations", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 401, + 415, + 414 + ], + "spans": [ + { + "bbox": [ + 116, + 401, + 415, + 414 + ], + "score": 1.0, + "content": "for generalization across designs. arXiv preprint arXiv:1802.08598, 2018.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 420, + 506, + 465 + ], + "lines": [ + { + "bbox": [ + 107, + 421, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 107, + 421, + 506, + 433 + ], + "score": 1.0, + "content": "Yonghan Jung, Jin Tian, and Elias Bareinboim. Learning causal effects via weighted empirical risk", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 114, + 431, + 507, + 444 + ], + "spans": [ + { + "bbox": [ + 114, + 431, + 507, + 444 + ], + "score": 1.0, + "content": "minimization. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 443, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 115, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "and Hsuan-Tien Lin (eds.), Advances in Neural Information Processing Systems 33, Virtual Event,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 453, + 142, + 465 + ], + "spans": [ + { + "bbox": [ + 115, + 453, + 142, + 465 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 472, + 505, + 496 + ], + "lines": [ + { + "bbox": [ + 105, + 472, + 507, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 507, + 485 + ], + "score": 1.0, + "content": "Rahul Kidambi, Aravind Rajeswaran, Praneeth Netrapalli, and Thorsten Joachims. MOReL : Model-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 484, + 388, + 496 + ], + "spans": [ + { + "bbox": [ + 116, + 484, + 388, + 496 + ], + "score": 1.0, + "content": "based offline reinforcement learning. CoRR, abs/2005.05951, 2020.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 106, + 502, + 506, + 526 + ], + "lines": [ + { + "bbox": [ + 106, + 502, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 506, + 516 + ], + "score": 1.0, + "content": "Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tutorial,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 515, + 414, + 526 + ], + "spans": [ + { + "bbox": [ + 116, + 515, + 414, + 526 + ], + "score": 1.0, + "content": "review, and perspectives on open problems. CoRR, abs/2005.01643, 2020.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 505, + 567 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 547 + ], + "score": 1.0, + "content": "Yao Liu, Adith Swaminathan, Alekh Agarwal, and Emma Brunskill. Provably good batch off-policy", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 114, + 543, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 114, + 543, + 505, + 558 + ], + "score": 1.0, + "content": "reinforcement learning without great exploration. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 555, + 244, + 568 + ], + "spans": [ + { + "bbox": [ + 115, + 555, + 244, + 568 + ], + "score": 1.0, + "content": "Systems 33, virtual event, 2020.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 574, + 506, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 574, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 506, + 588 + ], + "score": 1.0, + "content": "XuanLong Nguyen, Martin J. Wainwright, and Michael I. Jordan. Estimating divergence functionals", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 586, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 115, + 586, + 506, + 598 + ], + "score": 1.0, + "content": "and the likelihood ratio by convex risk minimization. IEEE Trans. Inf. Theory, 56(11):5847–5861,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 595, + 143, + 610 + ], + "spans": [ + { + "bbox": [ + 115, + 595, + 143, + 610 + ], + "score": 1.0, + "content": "2010.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 108, + 615, + 504, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "Sebastian Nowozin, Botond Cseke, and Ryota Tomioka. f-GAN: Training generative neural samplers", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 116, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "using variational divergence minimization. In Advances in Neural Information Processing Systems", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 638, + 284, + 650 + ], + "spans": [ + { + "bbox": [ + 116, + 638, + 284, + 650 + ], + "score": 1.0, + "content": "29, pp. 271–279, Barcelona, Spain, 2016.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 657, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 669 + ], + "score": 1.0, + "content": "Silviu Pitis, Elliot Creager, and Animesh Garg. Counterfactual data augmentation using locally", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 668, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 116, + 668, + 505, + 680 + ], + "score": 1.0, + "content": "factored dynamics. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 679, + 507, + 692 + ], + "spans": [ + { + "bbox": [ + 115, + 679, + 507, + 692 + ], + "score": 1.0, + "content": "Balcan, and Hsuan-Tien Lin (eds.), Advances in Neural Information Processing Systems 33, virtual,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 690, + 142, + 702 + ], + "spans": [ + { + "bbox": [ + 115, + 690, + 142, + 702 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5 + }, + { + "type": "text", + "bbox": [ + 109, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 107, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 107, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Dean Pomerleau. Efficient training of artificial neural networks for autonomous navigation. Neural", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 721, + 247, + 731 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 247, + 731 + ], + "score": 1.0, + "content": "Computation, 3(1):88–97, 1991.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 126 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 95 + ], + "score": 1.0, + "content": "Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, and Sergey Levine. Soft Actor-Critic: Off-policy", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "maximum entropy deep reinforcement learning with a stochastic actor. In Proceedings of the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 114, + 104, + 507, + 117 + ], + "spans": [ + { + "bbox": [ + 114, + 104, + 507, + 117 + ], + "score": 1.0, + "content": "35th International Conference on Machine Learning, pp. 1856–1865, Stockholmsmässan, Sweden,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 114, + 115, + 143, + 127 + ], + "spans": [ + { + "bbox": [ + 114, + 115, + 143, + 127 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 81, + 507, + 127 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 134, + 506, + 189 + ], + "lines": [ + { + "bbox": [ + 106, + 135, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 106, + 135, + 505, + 146 + ], + "score": 1.0, + "content": "Danijar Hafner, Timothy P. Lillicrap, Ian Fischer, Ruben Villegas, David Ha, Honglak Lee, and James", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 506, + 159 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 506, + 159 + ], + "score": 1.0, + "content": "Davidson. Learning latent dynamics for planning from pixels. In Kamalika Chaudhuri and Ruslan", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 154, + 507, + 171 + ], + "spans": [ + { + "bbox": [ + 115, + 154, + 507, + 171 + ], + "score": 1.0, + "content": "Salakhutdinov (eds.), Proceedings of the 36th International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 166, + 507, + 180 + ], + "spans": [ + { + "bbox": [ + 115, + 166, + 507, + 180 + ], + "score": 1.0, + "content": "volume 97 of Proceedings of Machine Learning Research, pp. 2555–2565, Long Beach, CA, 2019.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 178, + 149, + 191 + ], + "spans": [ + { + "bbox": [ + 115, + 178, + 149, + 191 + ], + "score": 1.0, + "content": "PMLR.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6, + "bbox_fs": [ + 106, + 135, + 507, + 191 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 198, + 506, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 506, + 211 + ], + "score": 1.0, + "content": "Negar Hassanpour and Russell Greiner. Counterfactual regression with importance sampling weights.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 208, + 507, + 222 + ], + "spans": [ + { + "bbox": [ + 115, + 208, + 507, + 222 + ], + "score": 1.0, + "content": "In Proceedings of the 28th International Joint Conference on Artificial Intelligence, pp. 5880–5887,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 219, + 203, + 231 + ], + "spans": [ + { + "bbox": [ + 116, + 219, + 203, + 231 + ], + "score": 1.0, + "content": "Macao, China, 2019.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 197, + 507, + 231 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 238, + 505, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 238, + 505, + 251 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 505, + 251 + ], + "score": 1.0, + "content": "J.B. Hiriart-Urruty and C. Lemaréchal. Fundamentals of Convex Analysis. 2001. ISBN", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 250, + 186, + 262 + ], + "spans": [ + { + "bbox": [ + 116, + 250, + 186, + 262 + ], + "score": 1.0, + "content": "9783642564680.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 238, + 505, + 262 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 269, + 504, + 293 + ], + "lines": [ + { + "bbox": [ + 105, + 268, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 282 + ], + "score": 1.0, + "content": "Jonathan Ho and Stefano Ermon. Generative adversarial imitation learning. In Advances in Neural", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 280, + 424, + 293 + ], + "spans": [ + { + "bbox": [ + 116, + 280, + 424, + 293 + ], + "score": 1.0, + "content": "Information Processing Systems 29, pp. 4565–4573, Barcelona, Spain, 2016.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 268, + 505, + 293 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 299, + 504, + 322 + ], + "lines": [ + { + "bbox": [ + 105, + 298, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 506, + 313 + ], + "score": 1.0, + "content": "Guido Imbens. The role of the propensity score in estimating dose-response functions. Econometrics", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 311, + 182, + 322 + ], + "spans": [ + { + "bbox": [ + 116, + 311, + 182, + 322 + ], + "score": 1.0, + "content": "eJournal, 1999.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 298, + 506, + 322 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 330, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 344 + ], + "score": 1.0, + "content": "Edward L Ionides. Truncated importance sampling. Journal of Computational and Graphical", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 340, + 246, + 353 + ], + "spans": [ + { + "bbox": [ + 115, + 340, + 246, + 353 + ], + "score": 1.0, + "content": "Statistics, 17(2):295–311, 2008.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 329, + 506, + 353 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 360, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 359, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 506, + 374 + ], + "score": 1.0, + "content": "Stephen James and Edward Johns. 3D simulation for robot arm control with deep Q-learning. CoRR,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 370, + 210, + 383 + ], + "spans": [ + { + "bbox": [ + 116, + 370, + 210, + 383 + ], + "score": 1.0, + "content": "abs/1609.03759, 2016.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 359, + 506, + 383 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 390, + 504, + 413 + ], + "lines": [ + { + "bbox": [ + 105, + 389, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 505, + 403 + ], + "score": 1.0, + "content": "Fredrik D Johansson, Nathan Kallus, Uri Shalit, and David Sontag. Learning weighted representations", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 401, + 415, + 414 + ], + "spans": [ + { + "bbox": [ + 116, + 401, + 415, + 414 + ], + "score": 1.0, + "content": "for generalization across designs. arXiv preprint arXiv:1802.08598, 2018.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 389, + 505, + 414 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 420, + 506, + 465 + ], + "lines": [ + { + "bbox": [ + 107, + 421, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 107, + 421, + 506, + 433 + ], + "score": 1.0, + "content": "Yonghan Jung, Jin Tian, and Elias Bareinboim. Learning causal effects via weighted empirical risk", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 114, + 431, + 507, + 444 + ], + "spans": [ + { + "bbox": [ + 114, + 431, + 507, + 444 + ], + "score": 1.0, + "content": "minimization. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 443, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 115, + 443, + 506, + 455 + ], + "score": 1.0, + "content": "and Hsuan-Tien Lin (eds.), Advances in Neural Information Processing Systems 33, Virtual Event,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 453, + 142, + 465 + ], + "spans": [ + { + "bbox": [ + 115, + 453, + 142, + 465 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 107, + 421, + 507, + 465 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 472, + 505, + 496 + ], + "lines": [ + { + "bbox": [ + 105, + 472, + 507, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 472, + 507, + 485 + ], + "score": 1.0, + "content": "Rahul Kidambi, Aravind Rajeswaran, Praneeth Netrapalli, and Thorsten Joachims. MOReL : Model-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 484, + 388, + 496 + ], + "spans": [ + { + "bbox": [ + 116, + 484, + 388, + 496 + ], + "score": 1.0, + "content": "based offline reinforcement learning. CoRR, abs/2005.05951, 2020.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 472, + 507, + 496 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 502, + 506, + 526 + ], + "lines": [ + { + "bbox": [ + 106, + 502, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 506, + 516 + ], + "score": 1.0, + "content": "Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tutorial,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 515, + 414, + 526 + ], + "spans": [ + { + "bbox": [ + 116, + 515, + 414, + 526 + ], + "score": 1.0, + "content": "review, and perspectives on open problems. CoRR, abs/2005.01643, 2020.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 106, + 502, + 506, + 526 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 505, + 567 + ], + "lines": [ + { + "bbox": [ + 106, + 532, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 547 + ], + "score": 1.0, + "content": "Yao Liu, Adith Swaminathan, Alekh Agarwal, and Emma Brunskill. Provably good batch off-policy", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 114, + 543, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 114, + 543, + 505, + 558 + ], + "score": 1.0, + "content": "reinforcement learning without great exploration. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 555, + 244, + 568 + ], + "spans": [ + { + "bbox": [ + 115, + 555, + 244, + 568 + ], + "score": 1.0, + "content": "Systems 33, virtual event, 2020.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 532, + 505, + 568 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 574, + 506, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 574, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 506, + 588 + ], + "score": 1.0, + "content": "XuanLong Nguyen, Martin J. Wainwright, and Michael I. Jordan. Estimating divergence functionals", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 586, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 115, + 586, + 506, + 598 + ], + "score": 1.0, + "content": "and the likelihood ratio by convex risk minimization. IEEE Trans. Inf. Theory, 56(11):5847–5861,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 595, + 143, + 610 + ], + "spans": [ + { + "bbox": [ + 115, + 595, + 143, + 610 + ], + "score": 1.0, + "content": "2010.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 106, + 574, + 506, + 610 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 615, + 504, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "Sebastian Nowozin, Botond Cseke, and Ryota Tomioka. f-GAN: Training generative neural samplers", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 116, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "using variational divergence minimization. In Advances in Neural Information Processing Systems", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 638, + 284, + 650 + ], + "spans": [ + { + "bbox": [ + 116, + 638, + 284, + 650 + ], + "score": 1.0, + "content": "29, pp. 271–279, Barcelona, Spain, 2016.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 615, + 505, + 650 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 657, + 505, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 505, + 669 + ], + "score": 1.0, + "content": "Silviu Pitis, Elliot Creager, and Animesh Garg. Counterfactual data augmentation using locally", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 668, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 116, + 668, + 505, + 680 + ], + "score": 1.0, + "content": "factored dynamics. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 679, + 507, + 692 + ], + "spans": [ + { + "bbox": [ + 115, + 679, + 507, + 692 + ], + "score": 1.0, + "content": "Balcan, and Hsuan-Tien Lin (eds.), Advances in Neural Information Processing Systems 33, virtual,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 690, + 142, + 702 + ], + "spans": [ + { + "bbox": [ + 115, + 690, + 142, + 702 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5, + "bbox_fs": [ + 106, + 657, + 507, + 702 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 107, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 107, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Dean Pomerleau. Efficient training of artificial neural networks for autonomous navigation. Neural", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 721, + 247, + 731 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 247, + 731 + ], + "score": 1.0, + "content": "Computation, 3(1):88–97, 1991.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 107, + 709, + 505, + 731 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Joaquin Quinonero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D Lawrence. Dataset", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 290, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 290, + 105 + ], + "score": 1.0, + "content": "shift in machine learning. Mit Press, 2008.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 105, + 112, + 504, + 135 + ], + "lines": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "score": 1.0, + "content": "Paul R. Rosenbaum and Donald B. Rubin. The central role of the propensity score in observational", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 117, + 123, + 336, + 135 + ], + "spans": [ + { + "bbox": [ + 117, + 123, + 336, + 135 + ], + "score": 1.0, + "content": "studies for causal effects. Biometrika, 70:41–55, 1983.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 108, + 141, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 104, + 141, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 141, + 505, + 155 + ], + "score": 1.0, + "content": "John Schulman, Sergey Levine, Pieter Abbeel, Michael I. Jordan, and Philipp Moritz. Trust region", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 506, + 167 + ], + "score": 1.0, + "content": "policy optimization. In Proceedings of the 32nd International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 164, + 309, + 177 + ], + "spans": [ + { + "bbox": [ + 116, + 164, + 309, + 177 + ], + "score": 1.0, + "content": "volume 37, pp. 1889–1897, Lille, France, 2015.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 105, + 182, + 504, + 206 + ], + "lines": [ + { + "bbox": [ + 105, + 181, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 505, + 198 + ], + "score": 1.0, + "content": "John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 194, + 340, + 206 + ], + "spans": [ + { + "bbox": [ + 116, + 194, + 340, + 206 + ], + "score": 1.0, + "content": "optimization algorithms. CoRR, abs/1707.06347, 2017.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 107, + 212, + 504, + 247 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 506, + 225 + ], + "score": 1.0, + "content": "Uri Shalit, Fredrik D. Johansson, and David A. Sontag. Estimating individual treatment effect:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 224, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 115, + 224, + 506, + 237 + ], + "score": 1.0, + "content": "generalization bounds and algorithms. In Proceedings of the 34th International Conference on", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 235, + 408, + 247 + ], + "spans": [ + { + "bbox": [ + 115, + 235, + 408, + 247 + ], + "score": 1.0, + "content": "Machine Learning, volume 70, pp. 3076–3085, Sydney, Australia, 2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 253, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 106, + 254, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 506, + 267 + ], + "score": 1.0, + "content": "Wenjie Shang, Yang Yu, Qingyang Li, Zhiwei (Tony) Qin, Yiping Meng, and Jieping Ye. Environment", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 266, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 115, + 266, + 505, + 277 + ], + "score": 1.0, + "content": "reconstruction with hidden confounders for reinforcement learning based recommendation. In", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 115, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 287, + 298, + 299 + ], + "spans": [ + { + "bbox": [ + 115, + 287, + 298, + 299 + ], + "score": 1.0, + "content": "Mining, pp. 566–576, Anchorage, AK, 2019.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 305, + 505, + 339 + ], + "lines": [ + { + "bbox": [ + 105, + 304, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 320 + ], + "score": 1.0, + "content": "Jing-Cheng Shi, Yang Yu, Qing Da, Shi-Yong Chen, and Anxiang Zeng. Virtual-taobao: Virtualizing", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 115, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "real-world online retail environment for reinforcement learning. In The 33th AAAI Conference on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 114, + 327, + 372, + 340 + ], + "spans": [ + { + "bbox": [ + 114, + 327, + 372, + 340 + ], + "score": 1.0, + "content": "Artificial Intelligence, pp. 4902–4909, Honolulu, Hawaii, 2019.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 346, + 504, + 370 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 505, + 361 + ], + "score": 1.0, + "content": "Hidetoshi Shimodaira. Improving predictive inference under covariate shift by weighting the log-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 357, + 469, + 370 + ], + "spans": [ + { + "bbox": [ + 116, + 357, + 469, + 370 + ], + "score": 1.0, + "content": "likelihood function. Journal of statistical planning and inference, 90(2):227–244, 2000.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 375, + 506, + 432 + ], + "lines": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "score": 1.0, + "content": "Sumedh A. Sontakke, Arash Mehrjou, Laurent Itti, and Bernhard Schölkopf. Causal curiosity: RL", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 116, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "agents discovering self-supervised experiments for causal representation learning. In Marina Meila", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 397, + 507, + 412 + ], + "spans": [ + { + "bbox": [ + 115, + 397, + 507, + 412 + ], + "score": 1.0, + "content": "and Tong Zhang (eds.), Proceedings of the 38th International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 409, + 507, + 422 + ], + "spans": [ + { + "bbox": [ + 115, + 409, + 507, + 422 + ], + "score": 1.0, + "content": "volume 139 of Proceedings of Machine Learning Research, pp. 9848–9858, Virtual Event, 2021.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 420, + 149, + 432 + ], + "spans": [ + { + "bbox": [ + 115, + 420, + 149, + 432 + ], + "score": 1.0, + "content": "PMLR.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 105, + 438, + 505, + 462 + ], + "lines": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "score": 1.0, + "content": "Peter Spirtes. Introduction to causal inference. J. Mach. Learn. Res., 11:1643–1662, aug 2010. ISSN", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 449, + 164, + 461 + ], + "spans": [ + { + "bbox": [ + 115, + 449, + 164, + 461 + ], + "score": 1.0, + "content": "1532-4435.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 105, + 468, + 505, + 492 + ], + "lines": [ + { + "bbox": [ + 105, + 468, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 482 + ], + "score": 1.0, + "content": "Richard S. Sutton and Andrew G. Barto. Reinforcement learning - an introduction. Adaptive", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 480, + 331, + 492 + ], + "spans": [ + { + "bbox": [ + 116, + 480, + 331, + 492 + ], + "score": 1.0, + "content": "computation and machine learning. MIT Press, 1998.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 498, + 504, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 498, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 506, + 512 + ], + "score": 1.0, + "content": "Emanuel Todorov, Tom Erez, and Yuval Tassa. MuJoCo: A physics engine for model-based control.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 509, + 507, + 523 + ], + "spans": [ + { + "bbox": [ + 115, + 509, + 507, + 523 + ], + "score": 1.0, + "content": "In Proceedings of the 24th IEEE/RSJ International Conference on Intelligent Robots and Systems,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 521, + 293, + 533 + ], + "spans": [ + { + "bbox": [ + 115, + 521, + 293, + 533 + ], + "score": 1.0, + "content": "pp. 5026–5033, Vilamoura, Portugal, 2012.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 105, + 539, + 505, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 538, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 554 + ], + "score": 1.0, + "content": "Stefan Wager and Susan Athey. Estimation and inference of heterogeneous treatment effects using", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 550, + 489, + 563 + ], + "spans": [ + { + "bbox": [ + 116, + 550, + 489, + 563 + ], + "score": 1.0, + "content": "random forests. Journal of the American Statistical Association, 113(523):1228–1242, 2018.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 107, + 569, + 505, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 568, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 505, + 582 + ], + "score": 1.0, + "content": "Jeremy C. Weiss, Finn Kuusisto, Kendrick Boyd, Jie Liu, and David Page. Machine learning for", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 581, + 505, + 593 + ], + "spans": [ + { + "bbox": [ + 116, + 581, + 505, + 593 + ], + "score": 1.0, + "content": "treatment assignment: Improving individualized risk attribution. In American Medical Informatics", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 591, + 352, + 604 + ], + "spans": [ + { + "bbox": [ + 114, + 591, + 352, + 604 + ], + "score": 1.0, + "content": "Association Annual Symposium, San Francisco, CA, 2015.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 610, + 505, + 644 + ], + "lines": [ + { + "bbox": [ + 106, + 610, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 506, + 622 + ], + "score": 1.0, + "content": "Jinsung Yoon, James Jordon, and Mihaela van der Schaar. GANITE: estimation of individualized", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 620, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 115, + 620, + 505, + 635 + ], + "score": 1.0, + "content": "treatment effects using generative adversarial nets. In 6th International Conference on Learning", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 632, + 210, + 644 + ], + "spans": [ + { + "bbox": [ + 115, + 632, + 210, + 644 + ], + "score": 1.0, + "content": "Representations, 2018.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 651, + 505, + 685 + ], + "lines": [ + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "score": 1.0, + "content": "Tianhe Yu, Garrett Thomas, Lantao Yu, Stefano Ermon, James Y. Zou, Sergey Levine, Chelsea", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 115, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "Finn, and Tengyu Ma. MOPO: model-based offline policy optimization. In Advances in Neural", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 673, + 342, + 686 + ], + "spans": [ + { + "bbox": [ + 115, + 673, + 342, + 686 + ], + "score": 1.0, + "content": "Information Processing Systems 33, virtual event, 2020.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Joaquin Quinonero-Candela, Masashi Sugiyama, Anton Schwaighofer, and Neil D Lawrence. Dataset", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 290, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 290, + 105 + ], + "score": 1.0, + "content": "shift in machine learning. Mit Press, 2008.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 505, + 105 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 112, + 504, + 135 + ], + "lines": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 112, + 505, + 125 + ], + "score": 1.0, + "content": "Paul R. Rosenbaum and Donald B. Rubin. The central role of the propensity score in observational", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 117, + 123, + 336, + 135 + ], + "spans": [ + { + "bbox": [ + 117, + 123, + 336, + 135 + ], + "score": 1.0, + "content": "studies for causal effects. Biometrika, 70:41–55, 1983.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 106, + 112, + 505, + 135 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 141, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 104, + 141, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 141, + 505, + 155 + ], + "score": 1.0, + "content": "John Schulman, Sergey Levine, Pieter Abbeel, Michael I. Jordan, and Philipp Moritz. Trust region", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 506, + 167 + ], + "score": 1.0, + "content": "policy optimization. In Proceedings of the 32nd International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 164, + 309, + 177 + ], + "spans": [ + { + "bbox": [ + 116, + 164, + 309, + 177 + ], + "score": 1.0, + "content": "volume 37, pp. 1889–1897, Lille, France, 2015.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 104, + 141, + 506, + 177 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 182, + 504, + 206 + ], + "lines": [ + { + "bbox": [ + 105, + 181, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 505, + 198 + ], + "score": 1.0, + "content": "John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 194, + 340, + 206 + ], + "spans": [ + { + "bbox": [ + 116, + 194, + 340, + 206 + ], + "score": 1.0, + "content": "optimization algorithms. CoRR, abs/1707.06347, 2017.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 181, + 505, + 206 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 212, + 504, + 247 + ], + "lines": [ + { + "bbox": [ + 106, + 213, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 506, + 225 + ], + "score": 1.0, + "content": "Uri Shalit, Fredrik D. Johansson, and David A. Sontag. Estimating individual treatment effect:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 224, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 115, + 224, + 506, + 237 + ], + "score": 1.0, + "content": "generalization bounds and algorithms. In Proceedings of the 34th International Conference on", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 235, + 408, + 247 + ], + "spans": [ + { + "bbox": [ + 115, + 235, + 408, + 247 + ], + "score": 1.0, + "content": "Machine Learning, volume 70, pp. 3076–3085, Sydney, Australia, 2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 213, + 506, + 247 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 253, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 106, + 254, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 506, + 267 + ], + "score": 1.0, + "content": "Wenjie Shang, Yang Yu, Qingyang Li, Zhiwei (Tony) Qin, Yiping Meng, and Jieping Ye. Environment", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 266, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 115, + 266, + 505, + 277 + ], + "score": 1.0, + "content": "reconstruction with hidden confounders for reinforcement learning based recommendation. In", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 115, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 287, + 298, + 299 + ], + "spans": [ + { + "bbox": [ + 115, + 287, + 298, + 299 + ], + "score": 1.0, + "content": "Mining, pp. 566–576, Anchorage, AK, 2019.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5, + "bbox_fs": [ + 106, + 254, + 506, + 299 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 305, + 505, + 339 + ], + "lines": [ + { + "bbox": [ + 105, + 304, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 320 + ], + "score": 1.0, + "content": "Jing-Cheng Shi, Yang Yu, Qing Da, Shi-Yong Chen, and Anxiang Zeng. Virtual-taobao: Virtualizing", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 115, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "real-world online retail environment for reinforcement learning. In The 33th AAAI Conference on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 114, + 327, + 372, + 340 + ], + "spans": [ + { + "bbox": [ + 114, + 327, + 372, + 340 + ], + "score": 1.0, + "content": "Artificial Intelligence, pp. 4902–4909, Honolulu, Hawaii, 2019.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 304, + 506, + 340 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 346, + 504, + 370 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 505, + 361 + ], + "score": 1.0, + "content": "Hidetoshi Shimodaira. Improving predictive inference under covariate shift by weighting the log-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 357, + 469, + 370 + ], + "spans": [ + { + "bbox": [ + 116, + 357, + 469, + 370 + ], + "score": 1.0, + "content": "likelihood function. Journal of statistical planning and inference, 90(2):227–244, 2000.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 344, + 505, + 370 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 375, + 506, + 432 + ], + "lines": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 505, + 388 + ], + "score": 1.0, + "content": "Sumedh A. Sontakke, Arash Mehrjou, Laurent Itti, and Bernhard Schölkopf. Causal curiosity: RL", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 116, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "agents discovering self-supervised experiments for causal representation learning. In Marina Meila", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 115, + 397, + 507, + 412 + ], + "spans": [ + { + "bbox": [ + 115, + 397, + 507, + 412 + ], + "score": 1.0, + "content": "and Tong Zhang (eds.), Proceedings of the 38th International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 409, + 507, + 422 + ], + "spans": [ + { + "bbox": [ + 115, + 409, + 507, + 422 + ], + "score": 1.0, + "content": "volume 139 of Proceedings of Machine Learning Research, pp. 9848–9858, Virtual Event, 2021.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 420, + 149, + 432 + ], + "spans": [ + { + "bbox": [ + 115, + 420, + 149, + 432 + ], + "score": 1.0, + "content": "PMLR.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 376, + 507, + 432 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 438, + 505, + 462 + ], + "lines": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "score": 1.0, + "content": "Peter Spirtes. Introduction to causal inference. J. Mach. Learn. Res., 11:1643–1662, aug 2010. ISSN", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 449, + 164, + 461 + ], + "spans": [ + { + "bbox": [ + 115, + 449, + 164, + 461 + ], + "score": 1.0, + "content": "1532-4435.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 438, + 505, + 461 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 468, + 505, + 492 + ], + "lines": [ + { + "bbox": [ + 105, + 468, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 505, + 482 + ], + "score": 1.0, + "content": "Richard S. Sutton and Andrew G. Barto. Reinforcement learning - an introduction. Adaptive", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 480, + 331, + 492 + ], + "spans": [ + { + "bbox": [ + 116, + 480, + 331, + 492 + ], + "score": 1.0, + "content": "computation and machine learning. MIT Press, 1998.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 468, + 505, + 492 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 498, + 504, + 533 + ], + "lines": [ + { + "bbox": [ + 105, + 498, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 506, + 512 + ], + "score": 1.0, + "content": "Emanuel Todorov, Tom Erez, and Yuval Tassa. MuJoCo: A physics engine for model-based control.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 509, + 507, + 523 + ], + "spans": [ + { + "bbox": [ + 115, + 509, + 507, + 523 + ], + "score": 1.0, + "content": "In Proceedings of the 24th IEEE/RSJ International Conference on Intelligent Robots and Systems,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 521, + 293, + 533 + ], + "spans": [ + { + "bbox": [ + 115, + 521, + 293, + 533 + ], + "score": 1.0, + "content": "pp. 5026–5033, Vilamoura, Portugal, 2012.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 498, + 507, + 533 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 539, + 505, + 563 + ], + "lines": [ + { + "bbox": [ + 105, + 538, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 554 + ], + "score": 1.0, + "content": "Stefan Wager and Susan Athey. Estimation and inference of heterogeneous treatment effects using", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 550, + 489, + 563 + ], + "spans": [ + { + "bbox": [ + 116, + 550, + 489, + 563 + ], + "score": 1.0, + "content": "random forests. Journal of the American Statistical Association, 113(523):1228–1242, 2018.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 538, + 506, + 563 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 569, + 505, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 568, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 505, + 582 + ], + "score": 1.0, + "content": "Jeremy C. Weiss, Finn Kuusisto, Kendrick Boyd, Jie Liu, and David Page. Machine learning for", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 581, + 505, + 593 + ], + "spans": [ + { + "bbox": [ + 116, + 581, + 505, + 593 + ], + "score": 1.0, + "content": "treatment assignment: Improving individualized risk attribution. In American Medical Informatics", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 591, + 352, + 604 + ], + "spans": [ + { + "bbox": [ + 114, + 591, + 352, + 604 + ], + "score": 1.0, + "content": "Association Annual Symposium, San Francisco, CA, 2015.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 568, + 505, + 604 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 610, + 505, + 644 + ], + "lines": [ + { + "bbox": [ + 106, + 610, + 506, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 506, + 622 + ], + "score": 1.0, + "content": "Jinsung Yoon, James Jordon, and Mihaela van der Schaar. GANITE: estimation of individualized", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 620, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 115, + 620, + 505, + 635 + ], + "score": 1.0, + "content": "treatment effects using generative adversarial nets. In 6th International Conference on Learning", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 632, + 210, + 644 + ], + "spans": [ + { + "bbox": [ + 115, + 632, + 210, + 644 + ], + "score": 1.0, + "content": "Representations, 2018.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 106, + 610, + 506, + 644 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 651, + 505, + 685 + ], + "lines": [ + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "score": 1.0, + "content": "Tianhe Yu, Garrett Thomas, Lantao Yu, Stefano Ermon, James Y. Zou, Sergey Levine, Chelsea", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 115, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "Finn, and Tengyu Ma. MOPO: model-based offline policy optimization. In Advances in Neural", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 673, + 342, + 686 + ], + "spans": [ + { + "bbox": [ + 115, + 673, + 342, + 686 + ], + "score": 1.0, + "content": "Information Processing Systems 33, virtual event, 2020.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 650, + 505, + 686 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 104, + 193, + 125 + ], + "lines": [ + { + "bbox": [ + 103, + 101, + 196, + 128 + ], + "spans": [ + { + "bbox": [ + 103, + 101, + 196, + 128 + ], + "score": 1.0, + "content": "Appendix", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 146, + 216, + 161 + ], + "lines": [ + { + "bbox": [ + 106, + 144, + 217, + 163 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 217, + 163 + ], + "score": 1.0, + "content": "Table of Contents", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 127, + 168, + 482, + 177 + ], + "lines": [ + { + "bbox": [ + 130, + 166, + 482, + 178 + ], + "spans": [ + { + "bbox": [ + 130, + 167, + 254, + 178 + ], + "score": 1.0, + "content": "A Proof of Theoretical Results", + "type": "text" + }, + { + "bbox": [ + 469, + 166, + 482, + 178 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 143, + 179, + 482, + 225 + ], + "lines": [ + { + "bbox": [ + 143, + 178, + 482, + 189 + ], + "spans": [ + { + "bbox": [ + 143, + 179, + 240, + 189 + ], + "score": 1.0, + "content": "A.1 Proof of Lemma 4.3", + "type": "text" + }, + { + "bbox": [ + 470, + 178, + 482, + 189 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 144, + 190, + 482, + 202 + ], + "spans": [ + { + "bbox": [ + 144, + 191, + 161, + 200 + ], + "score": 1.0, + "content": "A.2", + "type": "text" + }, + { + "bbox": [ + 162, + 190, + 219, + 202 + ], + "score": 1.0, + "content": "Proof of Eq. 6", + "type": "text" + }, + { + "bbox": [ + 469, + 190, + 482, + 201 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 143, + 201, + 482, + 213 + ], + "spans": [ + { + "bbox": [ + 143, + 202, + 231, + 213 + ], + "score": 1.0, + "content": "A.3 Proof of Thm. 4.4", + "type": "text" + }, + { + "bbox": [ + 469, + 201, + 482, + 213 + ], + "score": 1.0, + "content": "17", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 142, + 213, + 482, + 225 + ], + "spans": [ + { + "bbox": [ + 142, + 214, + 279, + 225 + ], + "score": 1.0, + "content": "A.4 Proof of the Tractable Solution", + "type": "text" + }, + { + "bbox": [ + 468, + 213, + 482, + 225 + ], + "score": 1.0, + "content": "20", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 131, + 234, + 482, + 287 + ], + "lines": [ + { + "bbox": [ + 128, + 234, + 482, + 248 + ], + "spans": [ + { + "bbox": [ + 128, + 234, + 287, + 246 + ], + "score": 1.0, + "content": "B Discussion of the Theoretical Results", + "type": "text" + }, + { + "bbox": [ + 469, + 234, + 482, + 248 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 129, + 254, + 483, + 270 + ], + "spans": [ + { + "bbox": [ + 129, + 255, + 207, + 267 + ], + "score": 1.0, + "content": "C Societal Impact", + "type": "text" + }, + { + "bbox": [ + 468, + 254, + 483, + 270 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 128, + 274, + 482, + 288 + ], + "spans": [ + { + "bbox": [ + 128, + 275, + 249, + 287 + ], + "score": 1.0, + "content": "D AWRM-oracle Pseudocode", + "type": "text" + }, + { + "bbox": [ + 468, + 274, + 482, + 288 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 131, + 297, + 480, + 307 + ], + "lines": [ + { + "bbox": [ + 129, + 296, + 482, + 308 + ], + "spans": [ + { + "bbox": [ + 129, + 297, + 208, + 308 + ], + "score": 1.0, + "content": "E Implementation", + "type": "text" + }, + { + "bbox": [ + 468, + 296, + 482, + 308 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 142, + 308, + 480, + 331 + ], + "lines": [ + { + "bbox": [ + 142, + 308, + 482, + 319 + ], + "spans": [ + { + "bbox": [ + 142, + 308, + 317, + 319 + ], + "score": 1.0, + "content": "E.1 Details of the GALILEO Implementation", + "type": "text" + }, + { + "bbox": [ + 468, + 308, + 482, + 319 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 318, + 482, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 318, + 348, + 332 + ], + "score": 1.0, + "content": "E.2 Connection with Previous Adversarial Algorithms", + "type": "text" + }, + { + "bbox": [ + 468, + 320, + 482, + 331 + ], + "score": 1.0, + "content": "25", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 131, + 341, + 478, + 351 + ], + "lines": [ + { + "bbox": [ + 129, + 340, + 482, + 353 + ], + "spans": [ + { + "bbox": [ + 129, + 341, + 243, + 351 + ], + "score": 1.0, + "content": "F Additional Related Work", + "type": "text" + }, + { + "bbox": [ + 468, + 340, + 482, + 353 + ], + "score": 1.0, + "content": "26", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 469, + 341, + 481, + 351 + ], + "lines": [ + { + "bbox": [ + 469, + 341, + 481, + 351 + ], + "spans": [], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 129, + 362, + 482, + 372 + ], + "lines": [ + { + "bbox": [ + 141, + 360, + 481, + 373 + ], + "spans": [ + { + "bbox": [ + 141, + 361, + 221, + 373 + ], + "score": 1.0, + "content": "Experiment Details", + "type": "text" + }, + { + "bbox": [ + 469, + 360, + 481, + 373 + ], + "score": 1.0, + "content": "27", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 144, + 373, + 482, + 420 + ], + "lines": [ + { + "bbox": [ + 144, + 372, + 482, + 385 + ], + "spans": [ + { + "bbox": [ + 144, + 373, + 196, + 385 + ], + "score": 1.0, + "content": "G.1 Settings", + "type": "text" + }, + { + "bbox": [ + 468, + 372, + 482, + 385 + ], + "score": 1.0, + "content": "27", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 144, + 383, + 482, + 397 + ], + "spans": [ + { + "bbox": [ + 144, + 385, + 160, + 394 + ], + "score": 1.0, + "content": "G.2", + "type": "text" + }, + { + "bbox": [ + 162, + 385, + 241, + 396 + ], + "score": 1.0, + "content": "Baseline Algorithms", + "type": "text" + }, + { + "bbox": [ + 468, + 383, + 482, + 397 + ], + "score": 1.0, + "content": "29", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 144, + 396, + 482, + 408 + ], + "spans": [ + { + "bbox": [ + 144, + 397, + 160, + 406 + ], + "score": 1.0, + "content": "G.3", + "type": "text" + }, + { + "bbox": [ + 163, + 397, + 232, + 408 + ], + "score": 1.0, + "content": "Hyper-parameters", + "type": "text" + }, + { + "bbox": [ + 468, + 396, + 482, + 408 + ], + "score": 1.0, + "content": "30", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 144, + 407, + 482, + 420 + ], + "spans": [ + { + "bbox": [ + 144, + 407, + 253, + 420 + ], + "score": 1.0, + "content": "G.4 Computation Resources", + "type": "text" + }, + { + "bbox": [ + 468, + 408, + 482, + 420 + ], + "score": 1.0, + "content": "30", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 129, + 428, + 482, + 439 + ], + "lines": [ + { + "bbox": [ + 129, + 428, + 482, + 440 + ], + "spans": [ + { + "bbox": [ + 129, + 429, + 218, + 439 + ], + "score": 1.0, + "content": "H Additional Results", + "type": "text" + }, + { + "bbox": [ + 468, + 428, + 482, + 440 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 144, + 441, + 482, + 511 + ], + "lines": [ + { + "bbox": [ + 142, + 440, + 481, + 452 + ], + "spans": [ + { + "bbox": [ + 142, + 440, + 252, + 452 + ], + "score": 1.0, + "content": "H.1 All of the Result Table .", + "type": "text" + }, + { + "bbox": [ + 469, + 440, + 481, + 452 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 144, + 451, + 482, + 464 + ], + "spans": [ + { + "bbox": [ + 144, + 452, + 160, + 462 + ], + "score": 1.0, + "content": "H.2", + "type": "text" + }, + { + "bbox": [ + 163, + 452, + 265, + 464 + ], + "score": 1.0, + "content": "Averaged Response Curves", + "type": "text" + }, + { + "bbox": [ + 468, + 451, + 482, + 464 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 144, + 463, + 482, + 476 + ], + "spans": [ + { + "bbox": [ + 144, + 464, + 160, + 474 + ], + "score": 1.0, + "content": "H.3", + "type": "text" + }, + { + "bbox": [ + 164, + 463, + 227, + 475 + ], + "score": 1.0, + "content": "Ablation Studies", + "type": "text" + }, + { + "bbox": [ + 468, + 463, + 482, + 476 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 144, + 474, + 482, + 487 + ], + "spans": [ + { + "bbox": [ + 144, + 475, + 160, + 486 + ], + "score": 1.0, + "content": "H.4", + "type": "text" + }, + { + "bbox": [ + 163, + 475, + 270, + 486 + ], + "score": 1.0, + "content": "Worst-Case Prediction Error", + "type": "text" + }, + { + "bbox": [ + 469, + 474, + 482, + 487 + ], + "score": 1.0, + "content": "32", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 144, + 486, + 482, + 499 + ], + "spans": [ + { + "bbox": [ + 144, + 488, + 162, + 497 + ], + "score": 1.0, + "content": "H.5", + "type": "text" + }, + { + "bbox": [ + 163, + 487, + 304, + 498 + ], + "score": 1.0, + "content": "Detailed Results in the MuJoCo Tasks", + "type": "text" + }, + { + "bbox": [ + 469, + 486, + 482, + 499 + ], + "score": 1.0, + "content": "32", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 143, + 498, + 482, + 510 + ], + "spans": [ + { + "bbox": [ + 143, + 499, + 287, + 509 + ], + "score": 1.0, + "content": "H.6 Detailed Results in the BAT Task", + "type": "text" + }, + { + "bbox": [ + 469, + 498, + 482, + 510 + ], + "score": 1.0, + "content": "34", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23.5 + }, + { + "type": "title", + "bbox": [ + 107, + 538, + 307, + 552 + ], + "lines": [ + { + "bbox": [ + 106, + 538, + 308, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 308, + 553 + ], + "score": 1.0, + "content": "A PROOF OF THEORETICAL RESULTS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 428, + 576 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 429, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 429, + 579 + ], + "score": 1.0, + "content": "The overall pipeline to model the tractable solution to AWRM is given in Fig. 6.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 580, + 502, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 579, + 504, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 299, + 594 + ], + "score": 1.0, + "content": "In the proof section, we replace the notation of", + "type": "text" + }, + { + "bbox": [ + 299, + 581, + 307, + 591 + ], + "score": 0.71, + "content": "\\mathbb { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 579, + 504, + 594 + ], + "score": 1.0, + "content": "with an integral for brevity. Now we rewrite the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 591, + 243, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 178, + 607 + ], + "score": 1.0, + "content": "original objective", + "type": "text" + }, + { + "bbox": [ + 179, + 591, + 227, + 606 + ], + "score": 0.94, + "content": "\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , \\bar { M } )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 592, + 243, + 607 + ], + "score": 1.0, + "content": "as:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 608, + 504, + 635 + ], + "lines": [ + { + "bbox": [ + 111, + 608, + 504, + 635 + ], + "spans": [ + { + "bbox": [ + 111, + 608, + 504, + 635 + ], + "score": 0.89, + "content": "\\operatorname* { m i n } _ { \\substack { M \\in M } } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\int _ { \\mathcal { X } , \\mathcal { A } } \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) \\omega ( x , a ) \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\left( - \\log M ( x ^ { \\prime } | x , a ) \\right) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } ,", + "type": "interline_equation", + "image_path": "1c0ae0f1ffd4e5eca34900d7c57ddb97f2a842bdeece077cf33f4bbb2a5b6d08.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 111, + 608, + 504, + 617.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 111, + 617.0, + 504, + 626.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 111, + 626.0, + 504, + 635.0 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 657, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 104, + 653, + 508, + 679 + ], + "spans": [ + { + "bbox": [ + 104, + 653, + 132, + 679 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 657, + 212, + 678 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\omega ( x , a ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 653, + 230, + 679 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 230, + 659, + 385, + 676 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } = \\int _ { \\mathcal { X } , \\mathcal { A } } \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ^ { 2 } \\mathrm { d } a \\mathrm { d } x } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 653, + 470, + 679 + ], + "score": 1.0, + "content": ", which is the squared", + "type": "text" + }, + { + "bbox": [ + 471, + 662, + 479, + 673 + ], + "score": 0.87, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 653, + 508, + 679 + ], + "score": 1.0, + "content": "-norm.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 675, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 104, + 675, + 222, + 690 + ], + "score": 1.0, + "content": "In an MDP, given any policy", + "type": "text" + }, + { + "bbox": [ + 222, + 680, + 229, + 687 + ], + "score": 0.58, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 675, + 232, + 690 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 232, + 676, + 411, + 689 + ], + "score": 0.86, + "content": "\\rho _ { M ^ { * } } ^ { \\pi } ( x , a , x ^ { \\prime } ) = \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 675, + 438, + 690 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 439, + 677, + 471, + 689 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\pi } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 675, + 506, + 690 + ], + "score": 1.0, + "content": "denotes", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 211, + 700 + ], + "score": 1.0, + "content": "the occupancy measure of", + "type": "text" + }, + { + "bbox": [ + 212, + 690, + 218, + 698 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 687, + 260, + 700 + ], + "score": 1.0, + "content": "for policy", + "type": "text" + }, + { + "bbox": [ + 261, + 691, + 267, + 698 + ], + "score": 0.7, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 687, + 388, + 700 + ], + "score": 1.0, + "content": ", which can be defined (Sutton", + "type": "text" + }, + { + "bbox": [ + 389, + 690, + 397, + 698 + ], + "score": 0.39, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "Barto, 1998; Ho & Ermon,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 696, + 507, + 714 + ], + "spans": [ + { + "bbox": [ + 104, + 696, + 145, + 714 + ], + "score": 1.0, + "content": "2016) as", + "type": "text" + }, + { + "bbox": [ + 145, + 699, + 379, + 711 + ], + "score": 0.84, + "content": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\pi } ( x ) : = ( 1 - \\gamma ) \\mathbb { E } _ { x _ { 0 } \\sim \\rho _ { 0 } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 696, + 408, + 714 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 409, + 699, + 494, + 711 + ], + "score": 0.92, + "content": " { \\mathrm { P r } } ^ { \\pi } \\left[ x _ { t } = x | x _ { 0 } , M ^ { * } \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 696, + 507, + 714 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 247, + 722 + ], + "score": 1.0, + "content": "the state visitation probability that", + "type": "text" + }, + { + "bbox": [ + 247, + 713, + 254, + 720 + ], + "score": 0.78, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 709, + 310, + 722 + ], + "score": 1.0, + "content": "starts at state", + "type": "text" + }, + { + "bbox": [ + 310, + 712, + 321, + 721 + ], + "score": 0.83, + "content": "x _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 709, + 361, + 722 + ], + "score": 1.0, + "content": "in model", + "type": "text" + }, + { + "bbox": [ + 361, + 711, + 377, + 720 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 709, + 427, + 722 + ], + "score": 1.0, + "content": "and receive", + "type": "text" + }, + { + "bbox": [ + 427, + 712, + 434, + 720 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 709, + 482, + 722 + ], + "score": 1.0, + "content": "at timestep", + "type": "text" + }, + { + "bbox": [ + 482, + 711, + 487, + 720 + ], + "score": 0.76, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 720, + 235, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 145, + 732 + ], + "score": 0.92, + "content": "\\gamma \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 721, + 235, + 732 + ], + "score": 1.0, + "content": "is the discount factor.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 104, + 193, + 125 + ], + "lines": [ + { + "bbox": [ + 103, + 101, + 196, + 128 + ], + "spans": [ + { + "bbox": [ + 103, + 101, + 196, + 128 + ], + "score": 1.0, + "content": "Appendix", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 146, + 216, + 161 + ], + "lines": [ + { + "bbox": [ + 106, + 144, + 217, + 163 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 217, + 163 + ], + "score": 1.0, + "content": "Table of Contents", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 127, + 168, + 482, + 177 + ], + "lines": [ + { + "bbox": [ + 130, + 166, + 482, + 178 + ], + "spans": [ + { + "bbox": [ + 130, + 167, + 254, + 178 + ], + "score": 1.0, + "content": "A Proof of Theoretical Results", + "type": "text" + }, + { + "bbox": [ + 469, + 166, + 482, + 178 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "index", + "bbox": [ + 143, + 179, + 482, + 225 + ], + "lines": [ + { + "bbox": [ + 143, + 178, + 482, + 189 + ], + "spans": [ + { + "bbox": [ + 143, + 179, + 240, + 189 + ], + "score": 1.0, + "content": "A.1 Proof of Lemma 4.3", + "type": "text" + }, + { + "bbox": [ + 470, + 178, + 482, + 189 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 144, + 190, + 482, + 202 + ], + "spans": [ + { + "bbox": [ + 144, + 191, + 161, + 200 + ], + "score": 1.0, + "content": "A.2", + "type": "text" + }, + { + "bbox": [ + 162, + 190, + 219, + 202 + ], + "score": 1.0, + "content": "Proof of Eq. 6", + "type": "text" + }, + { + "bbox": [ + 469, + 190, + 482, + 201 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 143, + 201, + 482, + 213 + ], + "spans": [ + { + "bbox": [ + 143, + 202, + 231, + 213 + ], + "score": 1.0, + "content": "A.3 Proof of Thm. 4.4", + "type": "text" + }, + { + "bbox": [ + 469, + 201, + 482, + 213 + ], + "score": 1.0, + "content": "17", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 213, + 482, + 225 + ], + "spans": [ + { + "bbox": [ + 142, + 214, + 279, + 225 + ], + "score": 1.0, + "content": "A.4 Proof of the Tractable Solution", + "type": "text" + }, + { + "bbox": [ + 468, + 213, + 482, + 225 + ], + "score": 1.0, + "content": "20", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 234, + 482, + 248 + ], + "spans": [ + { + "bbox": [ + 128, + 234, + 287, + 246 + ], + "score": 1.0, + "content": "B Discussion of the Theoretical Results", + "type": "text" + }, + { + "bbox": [ + 469, + 234, + 482, + 248 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 129, + 254, + 483, + 270 + ], + "spans": [ + { + "bbox": [ + 129, + 255, + 207, + 267 + ], + "score": 1.0, + "content": "C Societal Impact", + "type": "text" + }, + { + "bbox": [ + 468, + 254, + 483, + 270 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 274, + 482, + 288 + ], + "spans": [ + { + "bbox": [ + 128, + 275, + 249, + 287 + ], + "score": 1.0, + "content": "D AWRM-oracle Pseudocode", + "type": "text" + }, + { + "bbox": [ + 468, + 274, + 482, + 288 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + } + ], + "index": 4.5, + "bbox_fs": [ + 142, + 178, + 482, + 225 + ] + }, + { + "type": "index", + "bbox": [ + 131, + 234, + 482, + 287 + ], + "lines": [], + "index": 8, + "bbox_fs": [ + 128, + 234, + 483, + 288 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 131, + 297, + 480, + 307 + ], + "lines": [ + { + "bbox": [ + 129, + 296, + 482, + 308 + ], + "spans": [ + { + "bbox": [ + 129, + 297, + 208, + 308 + ], + "score": 1.0, + "content": "E Implementation", + "type": "text" + }, + { + "bbox": [ + 468, + 296, + 482, + 308 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "index", + "bbox": [ + 142, + 308, + 480, + 331 + ], + "lines": [ + { + "bbox": [ + 142, + 308, + 482, + 319 + ], + "spans": [ + { + "bbox": [ + 142, + 308, + 317, + 319 + ], + "score": 1.0, + "content": "E.1 Details of the GALILEO Implementation", + "type": "text" + }, + { + "bbox": [ + 468, + 308, + 482, + 319 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 318, + 482, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 318, + 348, + 332 + ], + "score": 1.0, + "content": "E.2 Connection with Previous Adversarial Algorithms", + "type": "text" + }, + { + "bbox": [ + 468, + 320, + 482, + 331 + ], + "score": 1.0, + "content": "25", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + } + ], + "index": 11.5, + "bbox_fs": [ + 141, + 308, + 482, + 332 + ] + }, + { + "type": "title", + "bbox": [ + 131, + 341, + 478, + 351 + ], + "lines": [ + { + "bbox": [ + 129, + 340, + 482, + 353 + ], + "spans": [ + { + "bbox": [ + 129, + 341, + 243, + 351 + ], + "score": 1.0, + "content": "F Additional Related Work", + "type": "text" + }, + { + "bbox": [ + 468, + 340, + 482, + 353 + ], + "score": 1.0, + "content": "26", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 469, + 341, + 481, + 351 + ], + "lines": [ + { + "bbox": [ + 469, + 341, + 481, + 351 + ], + "spans": [], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 469, + 341, + 481, + 351 + ] + }, + { + "type": "title", + "bbox": [ + 129, + 362, + 482, + 372 + ], + "lines": [ + { + "bbox": [ + 141, + 360, + 481, + 373 + ], + "spans": [ + { + "bbox": [ + 141, + 361, + 221, + 373 + ], + "score": 1.0, + "content": "Experiment Details", + "type": "text" + }, + { + "bbox": [ + 469, + 360, + 481, + 373 + ], + "score": 1.0, + "content": "27", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "index", + "bbox": [ + 144, + 373, + 482, + 420 + ], + "lines": [ + { + "bbox": [ + 144, + 372, + 482, + 385 + ], + "spans": [ + { + "bbox": [ + 144, + 373, + 196, + 385 + ], + "score": 1.0, + "content": "G.1 Settings", + "type": "text" + }, + { + "bbox": [ + 468, + 372, + 482, + 385 + ], + "score": 1.0, + "content": "27", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 144, + 383, + 482, + 397 + ], + "spans": [ + { + "bbox": [ + 144, + 385, + 160, + 394 + ], + "score": 1.0, + "content": "G.2", + "type": "text" + }, + { + "bbox": [ + 162, + 385, + 241, + 396 + ], + "score": 1.0, + "content": "Baseline Algorithms", + "type": "text" + }, + { + "bbox": [ + 468, + 383, + 482, + 397 + ], + "score": 1.0, + "content": "29", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 144, + 396, + 482, + 408 + ], + "spans": [ + { + "bbox": [ + 144, + 397, + 160, + 406 + ], + "score": 1.0, + "content": "G.3", + "type": "text" + }, + { + "bbox": [ + 163, + 397, + 232, + 408 + ], + "score": 1.0, + "content": "Hyper-parameters", + "type": "text" + }, + { + "bbox": [ + 468, + 396, + 482, + 408 + ], + "score": 1.0, + "content": "30", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 144, + 407, + 482, + 420 + ], + "spans": [ + { + "bbox": [ + 144, + 407, + 253, + 420 + ], + "score": 1.0, + "content": "G.4 Computation Resources", + "type": "text" + }, + { + "bbox": [ + 468, + 408, + 482, + 420 + ], + "score": 1.0, + "content": "30", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + } + ], + "index": 17.5, + "bbox_fs": [ + 144, + 372, + 482, + 420 + ] + }, + { + "type": "title", + "bbox": [ + 129, + 428, + 482, + 439 + ], + "lines": [ + { + "bbox": [ + 129, + 428, + 482, + 440 + ], + "spans": [ + { + "bbox": [ + 129, + 429, + 218, + 439 + ], + "score": 1.0, + "content": "H Additional Results", + "type": "text" + }, + { + "bbox": [ + 468, + 428, + 482, + 440 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "index", + "bbox": [ + 144, + 441, + 482, + 511 + ], + "lines": [ + { + "bbox": [ + 142, + 440, + 481, + 452 + ], + "spans": [ + { + "bbox": [ + 142, + 440, + 252, + 452 + ], + "score": 1.0, + "content": "H.1 All of the Result Table .", + "type": "text" + }, + { + "bbox": [ + 469, + 440, + 481, + 452 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 144, + 451, + 482, + 464 + ], + "spans": [ + { + "bbox": [ + 144, + 452, + 160, + 462 + ], + "score": 1.0, + "content": "H.2", + "type": "text" + }, + { + "bbox": [ + 163, + 452, + 265, + 464 + ], + "score": 1.0, + "content": "Averaged Response Curves", + "type": "text" + }, + { + "bbox": [ + 468, + 451, + 482, + 464 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 144, + 463, + 482, + 476 + ], + "spans": [ + { + "bbox": [ + 144, + 464, + 160, + 474 + ], + "score": 1.0, + "content": "H.3", + "type": "text" + }, + { + "bbox": [ + 164, + 463, + 227, + 475 + ], + "score": 1.0, + "content": "Ablation Studies", + "type": "text" + }, + { + "bbox": [ + 468, + 463, + 482, + 476 + ], + "score": 1.0, + "content": "31", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 144, + 474, + 482, + 487 + ], + "spans": [ + { + "bbox": [ + 144, + 475, + 160, + 486 + ], + "score": 1.0, + "content": "H.4", + "type": "text" + }, + { + "bbox": [ + 163, + 475, + 270, + 486 + ], + "score": 1.0, + "content": "Worst-Case Prediction Error", + "type": "text" + }, + { + "bbox": [ + 469, + 474, + 482, + 487 + ], + "score": 1.0, + "content": "32", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 144, + 486, + 482, + 499 + ], + "spans": [ + { + "bbox": [ + 144, + 488, + 162, + 497 + ], + "score": 1.0, + "content": "H.5", + "type": "text" + }, + { + "bbox": [ + 163, + 487, + 304, + 498 + ], + "score": 1.0, + "content": "Detailed Results in the MuJoCo Tasks", + "type": "text" + }, + { + "bbox": [ + 469, + 486, + 482, + 499 + ], + "score": 1.0, + "content": "32", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 143, + 498, + 482, + 510 + ], + "spans": [ + { + "bbox": [ + 143, + 499, + 287, + 509 + ], + "score": 1.0, + "content": "H.6 Detailed Results in the BAT Task", + "type": "text" + }, + { + "bbox": [ + 469, + 498, + 482, + 510 + ], + "score": 1.0, + "content": "34", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + } + ], + "index": 23.5, + "bbox_fs": [ + 142, + 440, + 482, + 510 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 538, + 307, + 552 + ], + "lines": [ + { + "bbox": [ + 106, + 538, + 308, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 308, + 553 + ], + "score": 1.0, + "content": "A PROOF OF THEORETICAL RESULTS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 428, + 576 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 429, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 429, + 579 + ], + "score": 1.0, + "content": "The overall pipeline to model the tractable solution to AWRM is given in Fig. 6.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 561, + 429, + 579 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 580, + 502, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 579, + 504, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 299, + 594 + ], + "score": 1.0, + "content": "In the proof section, we replace the notation of", + "type": "text" + }, + { + "bbox": [ + 299, + 581, + 307, + 591 + ], + "score": 0.71, + "content": "\\mathbb { E }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 579, + 504, + 594 + ], + "score": 1.0, + "content": "with an integral for brevity. Now we rewrite the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 591, + 243, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 178, + 607 + ], + "score": 1.0, + "content": "original objective", + "type": "text" + }, + { + "bbox": [ + 179, + 591, + 227, + 606 + ], + "score": 0.94, + "content": "\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , \\bar { M } )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 592, + 243, + 607 + ], + "score": 1.0, + "content": "as:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 579, + 504, + 607 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 608, + 504, + 635 + ], + "lines": [ + { + "bbox": [ + 111, + 608, + 504, + 635 + ], + "spans": [ + { + "bbox": [ + 111, + 608, + 504, + 635 + ], + "score": 0.89, + "content": "\\operatorname* { m i n } _ { \\substack { M \\in M } } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\int _ { \\mathcal { X } , \\mathcal { A } } \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) \\omega ( x , a ) \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\left( - \\log M ( x ^ { \\prime } | x , a ) \\right) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } ,", + "type": "interline_equation", + "image_path": "1c0ae0f1ffd4e5eca34900d7c57ddb97f2a842bdeece077cf33f4bbb2a5b6d08.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 111, + 608, + 504, + 617.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 111, + 617.0, + 504, + 626.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 111, + 626.0, + 504, + 635.0 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 657, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 104, + 653, + 508, + 679 + ], + "spans": [ + { + "bbox": [ + 104, + 653, + 132, + 679 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 657, + 212, + 678 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\omega ( x , a ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 653, + 230, + 679 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 230, + 659, + 385, + 676 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } = \\int _ { \\mathcal { X } , \\mathcal { A } } \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ^ { 2 } \\mathrm { d } a \\mathrm { d } x } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 653, + 470, + 679 + ], + "score": 1.0, + "content": ", which is the squared", + "type": "text" + }, + { + "bbox": [ + 471, + 662, + 479, + 673 + ], + "score": 0.87, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 653, + 508, + 679 + ], + "score": 1.0, + "content": "-norm.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 675, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 104, + 675, + 222, + 690 + ], + "score": 1.0, + "content": "In an MDP, given any policy", + "type": "text" + }, + { + "bbox": [ + 222, + 680, + 229, + 687 + ], + "score": 0.58, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 675, + 232, + 690 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 232, + 676, + 411, + 689 + ], + "score": 0.86, + "content": "\\rho _ { M ^ { * } } ^ { \\pi } ( x , a , x ^ { \\prime } ) = \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 675, + 438, + 690 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 439, + 677, + 471, + 689 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\pi } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 675, + 506, + 690 + ], + "score": 1.0, + "content": "denotes", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 211, + 700 + ], + "score": 1.0, + "content": "the occupancy measure of", + "type": "text" + }, + { + "bbox": [ + 212, + 690, + 218, + 698 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 687, + 260, + 700 + ], + "score": 1.0, + "content": "for policy", + "type": "text" + }, + { + "bbox": [ + 261, + 691, + 267, + 698 + ], + "score": 0.7, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 687, + 388, + 700 + ], + "score": 1.0, + "content": ", which can be defined (Sutton", + "type": "text" + }, + { + "bbox": [ + 389, + 690, + 397, + 698 + ], + "score": 0.39, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "Barto, 1998; Ho & Ermon,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 696, + 507, + 714 + ], + "spans": [ + { + "bbox": [ + 104, + 696, + 145, + 714 + ], + "score": 1.0, + "content": "2016) as", + "type": "text" + }, + { + "bbox": [ + 145, + 699, + 379, + 711 + ], + "score": 0.84, + "content": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\pi } ( x ) : = ( 1 - \\gamma ) \\mathbb { E } _ { x _ { 0 } \\sim \\rho _ { 0 } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 696, + 408, + 714 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 409, + 699, + 494, + 711 + ], + "score": 0.92, + "content": " { \\mathrm { P r } } ^ { \\pi } \\left[ x _ { t } = x | x _ { 0 } , M ^ { * } \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 696, + 507, + 714 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 247, + 722 + ], + "score": 1.0, + "content": "the state visitation probability that", + "type": "text" + }, + { + "bbox": [ + 247, + 713, + 254, + 720 + ], + "score": 0.78, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 709, + 310, + 722 + ], + "score": 1.0, + "content": "starts at state", + "type": "text" + }, + { + "bbox": [ + 310, + 712, + 321, + 721 + ], + "score": 0.83, + "content": "x _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 709, + 361, + 722 + ], + "score": 1.0, + "content": "in model", + "type": "text" + }, + { + "bbox": [ + 361, + 711, + 377, + 720 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 709, + 427, + 722 + ], + "score": 1.0, + "content": "and receive", + "type": "text" + }, + { + "bbox": [ + 427, + 712, + 434, + 720 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 709, + 482, + 722 + ], + "score": 1.0, + "content": "at timestep", + "type": "text" + }, + { + "bbox": [ + 482, + 711, + 487, + 720 + ], + "score": 0.76, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 720, + 235, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 145, + 732 + ], + "score": 0.92, + "content": "\\gamma \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 721, + 235, + 732 + ], + "score": 1.0, + "content": "is the discount factor.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5, + "bbox_fs": [ + 104, + 653, + 508, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 173, + 80, + 438, + 268 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 173, + 80, + 438, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 173, + 80, + 438, + 268 + ], + "spans": [ + { + "bbox": [ + 173, + 80, + 438, + 268 + ], + "score": 0.966, + "type": "image", + "image_path": "b8d8ca5df420de7f53155d64557ee7e5645c8c107cdf9e997b057385c9255a79.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 173, + 80, + 438, + 142.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 173, + 142.66666666666666, + 438, + 205.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 173, + 205.33333333333331, + 438, + 268.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 279, + 506, + 313 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 278, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 403, + 292 + ], + "score": 1.0, + "content": "Figure 6: The overall pipeline to model the tractable solution to AWRM.", + "type": "text" + }, + { + "bbox": [ + 403, + 279, + 411, + 291 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 278, + 505, + 292 + ], + "score": 1.0, + "content": "is a generator function", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 289, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 154, + 303 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + }, + { + "bbox": [ + 154, + 291, + 161, + 302 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 289, + 313, + 303 + ], + "score": 1.0, + "content": "-divergence (Nowozin et al., 2016).", + "type": "text" + }, + { + "bbox": [ + 314, + 292, + 321, + 300 + ], + "score": 0.69, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 289, + 505, + 303 + ], + "score": 1.0, + "content": "is an intermediary policy introduced in the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 300, + 153, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 153, + 313 + ], + "score": 1.0, + "content": "estimation.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 333, + 229, + 345 + ], + "lines": [ + { + "bbox": [ + 105, + 331, + 230, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 230, + 347 + ], + "score": 1.0, + "content": "A.1 PROOF OF LEMMA 4.3", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 351, + 367 + ], + "lines": [ + { + "bbox": [ + 106, + 354, + 351, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 351, + 367 + ], + "score": 1.0, + "content": "For better readability, we first rewrite Lemma 4.3 as follows:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 105, + 371, + 503, + 396 + ], + "lines": [ + { + "bbox": [ + 105, + 371, + 504, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 211, + 388 + ], + "score": 1.0, + "content": "Lemma A.1. Given any", + "type": "text" + }, + { + "bbox": [ + 211, + 374, + 223, + 384 + ], + "score": 0.73, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 371, + 236, + 388 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 236, + 371, + 284, + 386 + ], + "score": 0.93, + "content": "\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M )", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 371, + 492, + 388 + ], + "score": 1.0, + "content": ", the distribution of the ideal best-response policy", + "type": "text" + }, + { + "bbox": [ + 492, + 372, + 504, + 385 + ], + "score": 0.84, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 382, + 145, + 399 + ], + "spans": [ + { + "bbox": [ + 104, + 382, + 145, + 399 + ], + "score": 1.0, + "content": "satisfies:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "interline_equation", + "bbox": [ + 176, + 397, + 434, + 423 + ], + "lines": [ + { + "bbox": [ + 176, + 397, + 434, + 423 + ], + "spans": [ + { + "bbox": [ + 176, + 397, + 434, + 423 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ,", + "type": "interline_equation", + "image_path": "55d6a8f2e6526f3bd013bea3c86052f10a04d6c901b66db1ff872d32038fd387.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 176, + 397, + 434, + 423 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 428, + 506, + 475 + ], + "lines": [ + { + "bbox": [ + 106, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 134, + 442 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 429, + 252, + 441 + ], + "score": 0.9, + "content": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 429, + 353, + 442 + ], + "score": 1.0, + "content": "is the Kullback-Leibler", + "type": "text" + }, + { + "bbox": [ + 354, + 430, + 374, + 440 + ], + "score": 0.72, + "content": "( K L )", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 429, + 460, + 442 + ], + "score": 1.0, + "content": "divergence between", + "type": "text" + }, + { + "bbox": [ + 460, + 429, + 505, + 441 + ], + "score": 0.89, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 439, + 504, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 124, + 453 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 125, + 441, + 165, + 452 + ], + "score": 0.57, + "content": "M ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 439, + 169, + 453 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 170, + 441, + 215, + 452 + ], + "score": 0.74, + "content": "H _ { M ^ { \\ast } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 439, + 310, + 453 + ], + "score": 1.0, + "content": "denotes the entropy of", + "type": "text" + }, + { + "bbox": [ + 310, + 440, + 355, + 452 + ], + "score": 0.9, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 439, + 387, + 453 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 387, + 441, + 504, + 452 + ], + "score": 0.89, + "content": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 451, + 506, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 200, + 464 + ], + "score": 1.0, + "content": "is the Kullback-Leibler", + "type": "text" + }, + { + "bbox": [ + 201, + 452, + 220, + 462 + ], + "score": 0.79, + "content": "( K L )", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 451, + 302, + 464 + ], + "score": 1.0, + "content": "divergence between", + "type": "text" + }, + { + "bbox": [ + 303, + 452, + 347, + 463 + ], + "score": 0.95, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 451, + 366, + 464 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 366, + 452, + 406, + 463 + ], + "score": 0.82, + "content": "M ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 451, + 411, + 464 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 411, + 452, + 456, + 463 + ], + "score": 0.83, + "content": "H _ { M ^ { \\ast } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 451, + 506, + 464 + ], + "score": 1.0, + "content": "denotes the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 462, + 502, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 149, + 475 + ], + "score": 1.0, + "content": "entropy of", + "type": "text" + }, + { + "bbox": [ + 149, + 462, + 194, + 474 + ], + "score": 0.93, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 462, + 215, + 475 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 215, + 463, + 231, + 473 + ], + "score": 0.87, + "content": "\\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 462, + 357, + 475 + ], + "score": 1.0, + "content": "is the regularization coefficient", + "type": "text" + }, + { + "bbox": [ + 358, + 465, + 365, + 472 + ], + "score": 0.72, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 462, + 502, + 475 + ], + "score": 1.0, + "content": "in Eq. 9 and also as a normalizer.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 505, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 493, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 249, + 508 + ], + "score": 1.0, + "content": "Proof. Given a transition function", + "type": "text" + }, + { + "bbox": [ + 250, + 495, + 261, + 505 + ], + "score": 0.83, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 493, + 492, + 508 + ], + "score": 1.0, + "content": "of an MDP, the distribution of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 492, + 495, + 504, + 506 + ], + "score": 0.86, + "content": "\\beta ^ { * }", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 504, + 145, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 145, + 518 + ], + "score": 1.0, + "content": "satisfies:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "interline_equation", + "bbox": [ + 102, + 539, + 500, + 735 + ], + "lines": [ + { + "bbox": [ + 102, + 539, + 500, + 735 + ], + "spans": [ + { + "bbox": [ + 102, + 539, + 500, + 735 + ], + "score": 0.93, + "content": "\\begin{array} { r l } { \\rho _ { \\mu \\mu ^ { * } } ^ { * } } & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\omega ( x , \\omega ) \\int _ { x } d x ^ { * } ( x ^ { * } | x , a ) \\left( - \\log { \\mathcal { A } ( x ^ { * } | x , a ) } \\right) \\mathrm { d } x ^ { * } \\mathrm { d } \\alpha \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { \\nu \\mu } ^ { 2 } ( z , \\omega ) } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\int _ { x } \\underline { { \\mathcal { A } } } ^ { \\mathcal { A } ^ { * } } ( x ^ { * } | x , a ) \\left( - \\log { \\mathcal { A } ( x ^ { * } | x , a ) } \\right) \\mathrm { d } x ^ { * } \\mathrm { d } \\alpha \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { \\mu \\nu } ^ { 3 } d \\cdot ( x ^ { * } ) \\| _ { 2 } ^ { 2 } } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\frac { 2 } { \\alpha } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\| \\phi ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\mu \\nu } ^ { 3 } d \\cdot ( x ^ { * } ) \\| _ { 2 } ^ { 2 } } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\alpha \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) g ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\nu \\mu ^ { * } } ^ { 3 } ( z , \\nu ) \\| _ { 2 } ^ { 2 } } \\\\ & = \\operatorname* { m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\alpha \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( x , a ) g ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\mu \\nu } ^ { 3 } ( z , \\omega ) \\| _ { 2 } ^ { 2 } - \\frac \\end{array}", + "type": "interline_equation", + "image_path": "42a5873062af4a9b91baad6a11c887370a4afab3a398ab7be1cec5015ac74585.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 102, + 539, + 500, + 604.3333333333334 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 102, + 604.3333333333334, + 500, + 669.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 102, + 669.6666666666667, + 500, + 735.0000000000001 + ], + "spans": [], + "index": 19 + } + ] + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 173, + 80, + 438, + 268 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 173, + 80, + 438, + 268 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 173, + 80, + 438, + 268 + ], + "spans": [ + { + "bbox": [ + 173, + 80, + 438, + 268 + ], + "score": 0.966, + "type": "image", + "image_path": "b8d8ca5df420de7f53155d64557ee7e5645c8c107cdf9e997b057385c9255a79.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 173, + 80, + 438, + 142.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 173, + 142.66666666666666, + 438, + 205.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 173, + 205.33333333333331, + 438, + 268.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 279, + 506, + 313 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 278, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 403, + 292 + ], + "score": 1.0, + "content": "Figure 6: The overall pipeline to model the tractable solution to AWRM.", + "type": "text" + }, + { + "bbox": [ + 403, + 279, + 411, + 291 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 278, + 505, + 292 + ], + "score": 1.0, + "content": "is a generator function", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 289, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 154, + 303 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + }, + { + "bbox": [ + 154, + 291, + 161, + 302 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 289, + 313, + 303 + ], + "score": 1.0, + "content": "-divergence (Nowozin et al., 2016).", + "type": "text" + }, + { + "bbox": [ + 314, + 292, + 321, + 300 + ], + "score": 0.69, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 289, + 505, + 303 + ], + "score": 1.0, + "content": "is an intermediary policy introduced in the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 300, + 153, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 153, + 313 + ], + "score": 1.0, + "content": "estimation.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 333, + 229, + 345 + ], + "lines": [ + { + "bbox": [ + 105, + 331, + 230, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 230, + 347 + ], + "score": 1.0, + "content": "A.1 PROOF OF LEMMA 4.3", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 351, + 367 + ], + "lines": [ + { + "bbox": [ + 106, + 354, + 351, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 351, + 367 + ], + "score": 1.0, + "content": "For better readability, we first rewrite Lemma 4.3 as follows:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 354, + 351, + 367 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 371, + 503, + 396 + ], + "lines": [ + { + "bbox": [ + 105, + 371, + 504, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 211, + 388 + ], + "score": 1.0, + "content": "Lemma A.1. Given any", + "type": "text" + }, + { + "bbox": [ + 211, + 374, + 223, + 384 + ], + "score": 0.73, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 371, + 236, + 388 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 236, + 371, + 284, + 386 + ], + "score": 0.93, + "content": "\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M )", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 371, + 492, + 388 + ], + "score": 1.0, + "content": ", the distribution of the ideal best-response policy", + "type": "text" + }, + { + "bbox": [ + 492, + 372, + 504, + 385 + ], + "score": 0.84, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 382, + 145, + 399 + ], + "spans": [ + { + "bbox": [ + 104, + 382, + 145, + 399 + ], + "score": 1.0, + "content": "satisfies:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 104, + 371, + 504, + 399 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 176, + 397, + 434, + 423 + ], + "lines": [ + { + "bbox": [ + 176, + 397, + 434, + 423 + ], + "spans": [ + { + "bbox": [ + 176, + 397, + 434, + 423 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ,", + "type": "interline_equation", + "image_path": "55d6a8f2e6526f3bd013bea3c86052f10a04d6c901b66db1ff872d32038fd387.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 176, + 397, + 434, + 423 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 428, + 506, + 475 + ], + "lines": [ + { + "bbox": [ + 106, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 429, + 134, + 442 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 429, + 252, + 441 + ], + "score": 0.9, + "content": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 429, + 353, + 442 + ], + "score": 1.0, + "content": "is the Kullback-Leibler", + "type": "text" + }, + { + "bbox": [ + 354, + 430, + 374, + 440 + ], + "score": 0.72, + "content": "( K L )", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 429, + 460, + 442 + ], + "score": 1.0, + "content": "divergence between", + "type": "text" + }, + { + "bbox": [ + 460, + 429, + 505, + 441 + ], + "score": 0.89, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 439, + 504, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 124, + 453 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 125, + 441, + 165, + 452 + ], + "score": 0.57, + "content": "M ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 439, + 169, + 453 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 170, + 441, + 215, + 452 + ], + "score": 0.74, + "content": "H _ { M ^ { \\ast } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 439, + 310, + 453 + ], + "score": 1.0, + "content": "denotes the entropy of", + "type": "text" + }, + { + "bbox": [ + 310, + 440, + 355, + 452 + ], + "score": 0.9, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 439, + 387, + 453 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 387, + 441, + 504, + 452 + ], + "score": 0.89, + "content": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 451, + 506, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 200, + 464 + ], + "score": 1.0, + "content": "is the Kullback-Leibler", + "type": "text" + }, + { + "bbox": [ + 201, + 452, + 220, + 462 + ], + "score": 0.79, + "content": "( K L )", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 451, + 302, + 464 + ], + "score": 1.0, + "content": "divergence between", + "type": "text" + }, + { + "bbox": [ + 303, + 452, + 347, + 463 + ], + "score": 0.95, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 451, + 366, + 464 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 366, + 452, + 406, + 463 + ], + "score": 0.82, + "content": "M ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 451, + 411, + 464 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 411, + 452, + 456, + 463 + ], + "score": 0.83, + "content": "H _ { M ^ { \\ast } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 451, + 506, + 464 + ], + "score": 1.0, + "content": "denotes the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 462, + 502, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 149, + 475 + ], + "score": 1.0, + "content": "entropy of", + "type": "text" + }, + { + "bbox": [ + 149, + 462, + 194, + 474 + ], + "score": 0.93, + "content": "M ^ { * } ( \\cdot | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 462, + 215, + 475 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 215, + 463, + 231, + 473 + ], + "score": 0.87, + "content": "\\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 462, + 357, + 475 + ], + "score": 1.0, + "content": "is the regularization coefficient", + "type": "text" + }, + { + "bbox": [ + 358, + 465, + 365, + 472 + ], + "score": 0.72, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 462, + 502, + 475 + ], + "score": 1.0, + "content": "in Eq. 9 and also as a normalizer.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 429, + 506, + 475 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 505, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 493, + 504, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 249, + 508 + ], + "score": 1.0, + "content": "Proof. Given a transition function", + "type": "text" + }, + { + "bbox": [ + 250, + 495, + 261, + 505 + ], + "score": 0.83, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 493, + 492, + 508 + ], + "score": 1.0, + "content": "of an MDP, the distribution of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 492, + 495, + 504, + 506 + ], + "score": 0.86, + "content": "\\beta ^ { * }", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 504, + 145, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 145, + 518 + ], + "score": 1.0, + "content": "satisfies:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 493, + 504, + 518 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 102, + 539, + 500, + 735 + ], + "lines": [ + { + "bbox": [ + 102, + 539, + 500, + 735 + ], + "spans": [ + { + "bbox": [ + 102, + 539, + 500, + 735 + ], + "score": 0.93, + "content": "\\begin{array} { r l } { \\rho _ { \\mu \\mu ^ { * } } ^ { * } } & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\omega ( x , \\omega ) \\int _ { x } d x ^ { * } ( x ^ { * } | x , a ) \\left( - \\log { \\mathcal { A } ( x ^ { * } | x , a ) } \\right) \\mathrm { d } x ^ { * } \\mathrm { d } \\alpha \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { \\nu \\mu } ^ { 2 } ( z , \\omega ) } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\int _ { x } \\underline { { \\mathcal { A } } } ^ { \\mathcal { A } ^ { * } } ( x ^ { * } | x , a ) \\left( - \\log { \\mathcal { A } ( x ^ { * } | x , a ) } \\right) \\mathrm { d } x ^ { * } \\mathrm { d } \\alpha \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { \\mu \\nu } ^ { 3 } d \\cdot ( x ^ { * } ) \\| _ { 2 } ^ { 2 } } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\frac { 2 } { \\alpha } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\| \\phi ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\mu \\nu } ^ { 3 } d \\cdot ( x ^ { * } ) \\| _ { 2 } ^ { 2 } } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\alpha \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) g ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\nu \\mu ^ { * } } ^ { 3 } ( z , \\nu ) \\| _ { 2 } ^ { 2 } } \\\\ & = \\operatorname* { m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\alpha \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( x , a ) g ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\mu \\nu } ^ { 3 } ( z , \\omega ) \\| _ { 2 } ^ { 2 } - \\frac \\end{array}", + "type": "interline_equation", + "image_path": "42a5873062af4a9b91baad6a11c887370a4afab3a398ab7be1cec5015ac74585.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 102, + 539, + 500, + 604.3333333333334 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 102, + 604.3333333333334, + 500, + 669.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 102, + 669.6666666666667, + 500, + 735.0000000000001 + ], + "spans": [], + "index": 19 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 505, + 144 + ], + "lines": [ + { + "bbox": [ + 102, + 75, + 509, + 100 + ], + "spans": [ + { + "bbox": [ + 102, + 75, + 292, + 100 + ], + "score": 1.0, + "content": "We know that the occupancy measure", + "type": "text" + }, + { + "bbox": [ + 293, + 80, + 312, + 95 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 75, + 509, + 100 + ], + "score": 1.0, + "content": "is a density function with a constraint", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 506, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 219, + 109 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\int _ { \\mathcal { X } } \\int _ { A } \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) \\mathrm { d } a \\mathrm { d } x = 1 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 94, + 370, + 111 + ], + "score": 1.0, + "content": ". Assuming the occupancy measure", + "type": "text" + }, + { + "bbox": [ + 370, + 94, + 389, + 109 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 94, + 477, + 111 + ], + "score": 1.0, + "content": "has an upper bound", + "type": "text" + }, + { + "bbox": [ + 477, + 99, + 482, + 106 + ], + "score": 0.6, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 94, + 506, + 111 + ], + "score": 1.0, + "content": ", that", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 109, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 104, + 109, + 117, + 124 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 117, + 109, + 296, + 122 + ], + "score": 0.88, + "content": "0 ~ \\leq ~ \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ~ \\leq ~ c , \\forall a ~ \\in ~ { \\mathcal { A } } , \\forall x ~ \\in ~ \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 109, + 473, + 124 + ], + "score": 1.0, + "content": ", constructing a regularization coefficient", + "type": "text" + }, + { + "bbox": [ + 473, + 110, + 505, + 122 + ], + "score": 0.84, + "content": "\\alpha _ { M } =", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 108, + 119, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 108, + 122, + 338, + 134 + ], + "score": 0.81, + "content": "\\begin{array} { r } { \\int _ { \\mathcal { X } } \\int _ { \\mathcal { A } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) \\mathrm { d } x \\mathrm { d } a } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 119, + 468, + 136 + ], + "score": 1.0, + "content": "as a constant value given any", + "type": "text" + }, + { + "bbox": [ + 468, + 122, + 480, + 132 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 119, + 506, + 136 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 132, + 142, + 144 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 142, + 144 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "interline_equation", + "bbox": [ + 132, + 155, + 479, + 263 + ], + "lines": [ + { + "bbox": [ + 132, + 155, + 479, + 263 + ], + "spans": [ + { + "bbox": [ + 132, + 155, + 479, + 263 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\rho _ { M ^ { * } } ^ { \\beta ^ { * } } ( x , a ) = \\frac { g ( x , a ) } { \\alpha _ { M } } } \\\\ & { \\qquad = \\frac { \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\log \\frac { M ^ { * } ( x ^ { \\prime } | x , a ) } { M ( x ^ { \\prime } | x , a ) } \\mathrm { d } x - \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\log M ^ { * } ( x ^ { \\prime } | x , a ) \\mathrm { d } x } { \\alpha _ { M } } } \\\\ & { \\qquad = \\frac { D _ { K L } \\left( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) \\right) + H _ { M ^ { * } } ( x , a ) } { \\alpha _ { M } } } \\\\ & { \\qquad \\propto \\Big ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) \\Big ) , } \\end{array}", + "type": "interline_equation", + "image_path": "1da27a213d4fe3fa509876db24b86328155cda21a6b3900e45f8355fa7c7d562.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 132, + 155, + 479, + 191.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 132, + 191.0, + 479, + 227.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 132, + 227.0, + 479, + 263.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 351, + 287 + ], + "lines": [ + { + "bbox": [ + 105, + 273, + 352, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 311, + 290 + ], + "score": 1.0, + "content": "which is the optimal density function of Eq. 9 with", + "type": "text" + }, + { + "bbox": [ + 311, + 277, + 347, + 286 + ], + "score": 0.87, + "content": "\\alpha = \\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 273, + 352, + 290 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 292, + 505, + 326 + ], + "lines": [ + { + "bbox": [ + 105, + 290, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 227, + 306 + ], + "score": 1.0, + "content": "Note that in some particular", + "type": "text" + }, + { + "bbox": [ + 227, + 293, + 243, + 303 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 290, + 362, + 306 + ], + "score": 1.0, + "content": ", we still cannot construct a", + "type": "text" + }, + { + "bbox": [ + 362, + 293, + 370, + 304 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 290, + 505, + 306 + ], + "score": 1.0, + "content": "that can generate an occupancy", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 158, + 316 + ], + "score": 1.0, + "content": "specified by", + "type": "text" + }, + { + "bbox": [ + 158, + 303, + 207, + 316 + ], + "score": 0.93, + "content": "g ( x , a ) / \\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 303, + 240, + 316 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 240, + 304, + 252, + 314 + ], + "score": 0.78, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 303, + 505, + 316 + ], + "score": 1.0, + "content": ". We can only claim the distribution of the ideal best-response", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 314, + 185, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 134, + 327 + ], + "score": 1.0, + "content": "policy", + "type": "text" + }, + { + "bbox": [ + 134, + 314, + 145, + 326 + ], + "score": 0.89, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 314, + 185, + 327 + ], + "score": 1.0, + "content": "satisfies:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "interline_equation", + "bbox": [ + 176, + 339, + 435, + 365 + ], + "lines": [ + { + "bbox": [ + 176, + 339, + 435, + 365 + ], + "spans": [ + { + "bbox": [ + 176, + 339, + 435, + 365 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ,", + "type": "interline_equation", + "image_path": "131c9bcd79334176a65e290aa8869ff310b09911deb3d3a4a099405db6f0b10f.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 176, + 339, + 435, + 365 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 378, + 505, + 415 + ], + "lines": [ + { + "bbox": [ + 104, + 378, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 104, + 378, + 132, + 393 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 381, + 149, + 390 + ], + "score": 0.86, + "content": "\\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 378, + 227, + 393 + ], + "score": 1.0, + "content": "is a normalizer that", + "type": "text" + }, + { + "bbox": [ + 227, + 379, + 486, + 393 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\alpha _ { M } = \\int _ { \\mathcal { X } } \\int _ { \\mathcal { A } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | \\underline { { x } } , a ) ) + H _ { M ^ { * } } ( x , a ) ) \\mathrm { d } x \\mathrm { d } a } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 378, + 506, + 393 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 379, + 406 + ], + "score": 1.0, + "content": "give a discussion of the rationality of the ideal best-response policy", + "type": "text" + }, + { + "bbox": [ + 379, + 392, + 391, + 404 + ], + "score": 0.89, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 392, + 506, + 406 + ], + "score": 1.0, + "content": "as a replacement of the real", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 403, + 267, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 190, + 416 + ], + "score": 1.0, + "content": "best-response policy", + "type": "text" + }, + { + "bbox": [ + 191, + 404, + 202, + 415 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 403, + 267, + 416 + ], + "score": 1.0, + "content": "in Remark A.2.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 465, + 506, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 313, + 479 + ], + "score": 1.0, + "content": "Remark A.2. The optimal solution Eq. 11 relies on", + "type": "text" + }, + { + "bbox": [ + 313, + 467, + 342, + 479 + ], + "score": 0.92, + "content": "g ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 466, + 422, + 479 + ], + "score": 1.0, + "content": ". In some particular", + "type": "text" + }, + { + "bbox": [ + 423, + 467, + 438, + 477 + ], + "score": 0.85, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 466, + 506, + 479 + ], + "score": 1.0, + "content": ", it is intractable", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 477, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 104, + 477, + 154, + 491 + ], + "score": 1.0, + "content": "to derive a", + "type": "text" + }, + { + "bbox": [ + 155, + 478, + 162, + 489 + ], + "score": 0.82, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 477, + 352, + 491 + ], + "score": 1.0, + "content": "that can generate an occupancy specified by", + "type": "text" + }, + { + "bbox": [ + 352, + 478, + 401, + 489 + ], + "score": 0.95, + "content": "g ( x , a ) / \\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 477, + 506, + 491 + ], + "score": 1.0, + "content": ". Consider the following", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 489, + 504, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 165, + 501 + ], + "score": 1.0, + "content": "case: a state", + "type": "text" + }, + { + "bbox": [ + 165, + 490, + 177, + 500 + ], + "score": 0.83, + "content": "x _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 489, + 191, + 501 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 191, + 489, + 208, + 499 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 489, + 400, + 501 + ], + "score": 1.0, + "content": "might be harder to reach than another state", + "type": "text" + }, + { + "bbox": [ + 400, + 491, + 411, + 500 + ], + "score": 0.82, + "content": "x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 489, + 438, + 501 + ], + "score": 1.0, + "content": ", e.g.,", + "type": "text" + }, + { + "bbox": [ + 438, + 489, + 504, + 501 + ], + "score": 0.91, + "content": "M ^ { \\ast } ( x _ { 1 } | x , a ) \\ <", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 498, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 107, + 500, + 242, + 511 + ], + "score": 0.87, + "content": "M ^ { * } ( x _ { 2 } | x , a ) , \\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 498, + 376, + 512 + ], + "score": 1.0, + "content": ", then it is impossible to find a", + "type": "text" + }, + { + "bbox": [ + 377, + 500, + 384, + 511 + ], + "score": 0.79, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 498, + 505, + 512 + ], + "score": 1.0, + "content": "that the occupancy satisfies", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 510, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 107, + 511, + 215, + 525 + ], + "score": 0.89, + "content": "\\rho _ { M ^ { * } } ^ { \\beta } ( x _ { 1 } , a ) > \\rho _ { M ^ { * } } ^ { \\beta } ( x _ { 2 } , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 510, + 506, + 527 + ], + "score": 1.0, + "content": ". In this case, Eq. 11 can be a sub-optimal solution. Since this work", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 523, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 506, + 536 + ], + "score": 1.0, + "content": "focuses on task-agnostic solution derivation while the solution to the above problem should rely on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 534, + 504, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 214, + 548 + ], + "score": 1.0, + "content": "the specific description of", + "type": "text" + }, + { + "bbox": [ + 214, + 535, + 230, + 545 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 534, + 497, + 548 + ], + "score": 1.0, + "content": ", we leave it as future work. However, we point out that Eq. 11 is", + "type": "text" + }, + { + "bbox": [ + 498, + 539, + 504, + 545 + ], + "score": 0.27, + "content": "a", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 545, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 104, + 546, + 323, + 563 + ], + "score": 1.0, + "content": "reasonable re-weighting term even as a sub-optimum:", + "type": "text" + }, + { + "bbox": [ + 324, + 545, + 343, + 561 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 546, + 506, + 563 + ], + "score": 1.0, + "content": "gives larger densities on the data where", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 559, + 503, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 559, + 440, + 571 + ], + "score": 1.0, + "content": "the distribution distance between the approximation model and the real model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 441, + 559, + 503, + 571 + ], + "score": 0.92, + "content": "D _ { K L } ( M ^ { * } , M ) )", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 570, + 375, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 311, + 583 + ], + "score": 1.0, + "content": "is larger or the stochasticity of the real model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 311, + 570, + 333, + 582 + ], + "score": 0.88, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 570, + 375, + 583 + ], + "score": 1.0, + "content": ") is larger.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 20.5 + }, + { + "type": "title", + "bbox": [ + 108, + 604, + 203, + 616 + ], + "lines": [ + { + "bbox": [ + 105, + 602, + 205, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 205, + 618 + ], + "score": 1.0, + "content": "A.2 PROOF OF EQ. 6", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 628, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 201, + 642 + ], + "score": 1.0, + "content": "The integral process of", + "type": "text" + }, + { + "bbox": [ + 201, + 630, + 223, + 641 + ], + "score": 0.91, + "content": "D _ { K L }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 628, + 505, + 642 + ], + "score": 1.0, + "content": "in Eq. 5 is intractable in the offline setting as it explicitly requires the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 640, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 251, + 653 + ], + "score": 1.0, + "content": "conditional probability function of", + "type": "text" + }, + { + "bbox": [ + 252, + 641, + 267, + 651 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 640, + 506, + 653 + ], + "score": 1.0, + "content": ". Our motivation for the tractable solution is utilizing the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 650, + 506, + 666 + ], + "spans": [ + { + "bbox": [ + 104, + 650, + 163, + 666 + ], + "score": 1.0, + "content": "offline dataset", + "type": "text" + }, + { + "bbox": [ + 163, + 652, + 185, + 662 + ], + "score": 0.91, + "content": "\\mathcal { D } _ { \\mathrm { r e a l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 650, + 316, + 666 + ], + "score": 1.0, + "content": "as the empirical joint distribution", + "type": "text" + }, + { + "bbox": [ + 317, + 651, + 371, + 663 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 650, + 506, + 666 + ], + "score": 1.0, + "content": "and adopting practical techniques", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 662, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 505, + 675 + ], + "score": 1.0, + "content": "for distance estimation on two joint distributions, like GAN (Goodfellow et al., 2014; Nowozin et al.,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "score": 1.0, + "content": "2016), to approximate Eq. 5. To adopt that solution, we should first transform Eq. 5 into a form under", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 684, + 506, + 697 + ], + "spans": [ + { + "bbox": [ + 104, + 684, + 432, + 697 + ], + "score": 1.0, + "content": "joint distributions. Without loss of generality, we introduce an intermediary policy", + "type": "text" + }, + { + "bbox": [ + 433, + 686, + 439, + 694 + ], + "score": 0.73, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 684, + 480, + 697 + ], + "score": 1.0, + "content": ", of which", + "type": "text" + }, + { + "bbox": [ + 480, + 686, + 488, + 696 + ], + "score": 0.82, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 684, + 506, + 697 + ], + "score": 1.0, + "content": "can", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 694, + 507, + 709 + ], + "spans": [ + { + "bbox": [ + 104, + 694, + 300, + 709 + ], + "score": 1.0, + "content": "be regarded as a specific instance. Then we have", + "type": "text" + }, + { + "bbox": [ + 300, + 695, + 451, + 708 + ], + "score": 0.91, + "content": "M ( x ^ { \\prime } | x , a ) = \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) / \\rho _ { M } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 694, + 483, + 709 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 484, + 696, + 495, + 705 + ], + "score": 0.8, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 496, + 694, + 507, + 709 + ], + "score": 1.0, + "content": "if", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 708, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 107, + 709, + 163, + 722 + ], + "score": 0.91, + "content": "\\rho _ { M } ^ { \\kappa } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 708, + 209, + 724 + ], + "score": 1.0, + "content": ". Assuming", + "type": "text" + }, + { + "bbox": [ + 210, + 709, + 342, + 722 + ], + "score": 0.91, + "content": "\\forall x \\in \\mathcal { X } , \\forall a \\in A , \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 708, + 352, + 724 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 353, + 708, + 413, + 722 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 708, + 506, + 724 + ], + "score": 1.0, + "content": ", which will hold when", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 107, + 720, + 295, + 732 + ], + "spans": [ + { + "bbox": [ + 107, + 723, + 114, + 730 + ], + "score": 0.72, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 720, + 171, + 732 + ], + "score": 1.0, + "content": "overlaps with", + "type": "text" + }, + { + "bbox": [ + 171, + 722, + 178, + 732 + ], + "score": 0.8, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 720, + 295, + 732 + ], + "score": 1.0, + "content": ", then Eq. 5 can transform to:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 31 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 420, + 504, + 431 + ], + "lines": [ + { + "bbox": [ + 497, + 422, + 504, + 430 + ], + "spans": [ + { + "bbox": [ + 497, + 422, + 504, + 430 + ], + "score": 0.996, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 505, + 144 + ], + "lines": [ + { + "bbox": [ + 102, + 75, + 509, + 100 + ], + "spans": [ + { + "bbox": [ + 102, + 75, + 292, + 100 + ], + "score": 1.0, + "content": "We know that the occupancy measure", + "type": "text" + }, + { + "bbox": [ + 293, + 80, + 312, + 95 + ], + "score": 0.9, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 75, + 509, + 100 + ], + "score": 1.0, + "content": "is a density function with a constraint", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 506, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 219, + 109 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\int _ { \\mathcal { X } } \\int _ { A } \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) \\mathrm { d } a \\mathrm { d } x = 1 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 94, + 370, + 111 + ], + "score": 1.0, + "content": ". Assuming the occupancy measure", + "type": "text" + }, + { + "bbox": [ + 370, + 94, + 389, + 109 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\beta }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 94, + 477, + 111 + ], + "score": 1.0, + "content": "has an upper bound", + "type": "text" + }, + { + "bbox": [ + 477, + 99, + 482, + 106 + ], + "score": 0.6, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 94, + 506, + 111 + ], + "score": 1.0, + "content": ", that", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 109, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 104, + 109, + 117, + 124 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 117, + 109, + 296, + 122 + ], + "score": 0.88, + "content": "0 ~ \\leq ~ \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ~ \\leq ~ c , \\forall a ~ \\in ~ { \\mathcal { A } } , \\forall x ~ \\in ~ \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 109, + 473, + 124 + ], + "score": 1.0, + "content": ", constructing a regularization coefficient", + "type": "text" + }, + { + "bbox": [ + 473, + 110, + 505, + 122 + ], + "score": 0.84, + "content": "\\alpha _ { M } =", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 108, + 119, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 108, + 122, + 338, + 134 + ], + "score": 0.81, + "content": "\\begin{array} { r } { \\int _ { \\mathcal { X } } \\int _ { \\mathcal { A } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) \\mathrm { d } x \\mathrm { d } a } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 119, + 468, + 136 + ], + "score": 1.0, + "content": "as a constant value given any", + "type": "text" + }, + { + "bbox": [ + 468, + 122, + 480, + 132 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 119, + 506, + 136 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 132, + 142, + 144 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 142, + 144 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 102, + 75, + 509, + 144 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 132, + 155, + 479, + 263 + ], + "lines": [ + { + "bbox": [ + 132, + 155, + 479, + 263 + ], + "spans": [ + { + "bbox": [ + 132, + 155, + 479, + 263 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\rho _ { M ^ { * } } ^ { \\beta ^ { * } } ( x , a ) = \\frac { g ( x , a ) } { \\alpha _ { M } } } \\\\ & { \\qquad = \\frac { \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\log \\frac { M ^ { * } ( x ^ { \\prime } | x , a ) } { M ( x ^ { \\prime } | x , a ) } \\mathrm { d } x - \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\log M ^ { * } ( x ^ { \\prime } | x , a ) \\mathrm { d } x } { \\alpha _ { M } } } \\\\ & { \\qquad = \\frac { D _ { K L } \\left( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) \\right) + H _ { M ^ { * } } ( x , a ) } { \\alpha _ { M } } } \\\\ & { \\qquad \\propto \\Big ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) \\Big ) , } \\end{array}", + "type": "interline_equation", + "image_path": "1da27a213d4fe3fa509876db24b86328155cda21a6b3900e45f8355fa7c7d562.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 132, + 155, + 479, + 191.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 132, + 191.0, + 479, + 227.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 132, + 227.0, + 479, + 263.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 351, + 287 + ], + "lines": [ + { + "bbox": [ + 105, + 273, + 352, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 311, + 290 + ], + "score": 1.0, + "content": "which is the optimal density function of Eq. 9 with", + "type": "text" + }, + { + "bbox": [ + 311, + 277, + 347, + 286 + ], + "score": 0.87, + "content": "\\alpha = \\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 273, + 352, + 290 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 273, + 352, + 290 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 292, + 505, + 326 + ], + "lines": [ + { + "bbox": [ + 105, + 290, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 227, + 306 + ], + "score": 1.0, + "content": "Note that in some particular", + "type": "text" + }, + { + "bbox": [ + 227, + 293, + 243, + 303 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 290, + 362, + 306 + ], + "score": 1.0, + "content": ", we still cannot construct a", + "type": "text" + }, + { + "bbox": [ + 362, + 293, + 370, + 304 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 290, + 505, + 306 + ], + "score": 1.0, + "content": "that can generate an occupancy", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 158, + 316 + ], + "score": 1.0, + "content": "specified by", + "type": "text" + }, + { + "bbox": [ + 158, + 303, + 207, + 316 + ], + "score": 0.93, + "content": "g ( x , a ) / \\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 303, + 240, + 316 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 240, + 304, + 252, + 314 + ], + "score": 0.78, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 303, + 505, + 316 + ], + "score": 1.0, + "content": ". We can only claim the distribution of the ideal best-response", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 314, + 185, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 134, + 327 + ], + "score": 1.0, + "content": "policy", + "type": "text" + }, + { + "bbox": [ + 134, + 314, + 145, + 326 + ], + "score": 0.89, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 314, + 185, + 327 + ], + "score": 1.0, + "content": "satisfies:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 290, + 505, + 327 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 176, + 339, + 435, + 365 + ], + "lines": [ + { + "bbox": [ + 176, + 339, + 435, + 365 + ], + "spans": [ + { + "bbox": [ + 176, + 339, + 435, + 365 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ,", + "type": "interline_equation", + "image_path": "131c9bcd79334176a65e290aa8869ff310b09911deb3d3a4a099405db6f0b10f.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 176, + 339, + 435, + 365 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 378, + 505, + 415 + ], + "lines": [ + { + "bbox": [ + 104, + 378, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 104, + 378, + 132, + 393 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 381, + 149, + 390 + ], + "score": 0.86, + "content": "\\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 378, + 227, + 393 + ], + "score": 1.0, + "content": "is a normalizer that", + "type": "text" + }, + { + "bbox": [ + 227, + 379, + 486, + 393 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\alpha _ { M } = \\int _ { \\mathcal { X } } \\int _ { \\mathcal { A } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | \\underline { { x } } , a ) ) + H _ { M ^ { * } } ( x , a ) ) \\mathrm { d } x \\mathrm { d } a } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 378, + 506, + 393 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 379, + 406 + ], + "score": 1.0, + "content": "give a discussion of the rationality of the ideal best-response policy", + "type": "text" + }, + { + "bbox": [ + 379, + 392, + 391, + 404 + ], + "score": 0.89, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 392, + 506, + 406 + ], + "score": 1.0, + "content": "as a replacement of the real", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 403, + 267, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 190, + 416 + ], + "score": 1.0, + "content": "best-response policy", + "type": "text" + }, + { + "bbox": [ + 191, + 404, + 202, + 415 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 403, + 267, + 416 + ], + "score": 1.0, + "content": "in Remark A.2.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 378, + 506, + 416 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 465, + 506, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 313, + 479 + ], + "score": 1.0, + "content": "Remark A.2. The optimal solution Eq. 11 relies on", + "type": "text" + }, + { + "bbox": [ + 313, + 467, + 342, + 479 + ], + "score": 0.92, + "content": "g ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 466, + 422, + 479 + ], + "score": 1.0, + "content": ". In some particular", + "type": "text" + }, + { + "bbox": [ + 423, + 467, + 438, + 477 + ], + "score": 0.85, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 466, + 506, + 479 + ], + "score": 1.0, + "content": ", it is intractable", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 477, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 104, + 477, + 154, + 491 + ], + "score": 1.0, + "content": "to derive a", + "type": "text" + }, + { + "bbox": [ + 155, + 478, + 162, + 489 + ], + "score": 0.82, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 477, + 352, + 491 + ], + "score": 1.0, + "content": "that can generate an occupancy specified by", + "type": "text" + }, + { + "bbox": [ + 352, + 478, + 401, + 489 + ], + "score": 0.95, + "content": "g ( x , a ) / \\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 477, + 506, + 491 + ], + "score": 1.0, + "content": ". Consider the following", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 489, + 504, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 165, + 501 + ], + "score": 1.0, + "content": "case: a state", + "type": "text" + }, + { + "bbox": [ + 165, + 490, + 177, + 500 + ], + "score": 0.83, + "content": "x _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 489, + 191, + 501 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 191, + 489, + 208, + 499 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 489, + 400, + 501 + ], + "score": 1.0, + "content": "might be harder to reach than another state", + "type": "text" + }, + { + "bbox": [ + 400, + 491, + 411, + 500 + ], + "score": 0.82, + "content": "x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 489, + 438, + 501 + ], + "score": 1.0, + "content": ", e.g.,", + "type": "text" + }, + { + "bbox": [ + 438, + 489, + 504, + 501 + ], + "score": 0.91, + "content": "M ^ { \\ast } ( x _ { 1 } | x , a ) \\ <", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 498, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 107, + 500, + 242, + 511 + ], + "score": 0.87, + "content": "M ^ { * } ( x _ { 2 } | x , a ) , \\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 498, + 376, + 512 + ], + "score": 1.0, + "content": ", then it is impossible to find a", + "type": "text" + }, + { + "bbox": [ + 377, + 500, + 384, + 511 + ], + "score": 0.79, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 498, + 505, + 512 + ], + "score": 1.0, + "content": "that the occupancy satisfies", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 510, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 107, + 511, + 215, + 525 + ], + "score": 0.89, + "content": "\\rho _ { M ^ { * } } ^ { \\beta } ( x _ { 1 } , a ) > \\rho _ { M ^ { * } } ^ { \\beta } ( x _ { 2 } , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 510, + 506, + 527 + ], + "score": 1.0, + "content": ". In this case, Eq. 11 can be a sub-optimal solution. Since this work", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 523, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 506, + 536 + ], + "score": 1.0, + "content": "focuses on task-agnostic solution derivation while the solution to the above problem should rely on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 534, + 504, + 548 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 214, + 548 + ], + "score": 1.0, + "content": "the specific description of", + "type": "text" + }, + { + "bbox": [ + 214, + 535, + 230, + 545 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 534, + 497, + 548 + ], + "score": 1.0, + "content": ", we leave it as future work. However, we point out that Eq. 11 is", + "type": "text" + }, + { + "bbox": [ + 498, + 539, + 504, + 545 + ], + "score": 0.27, + "content": "a", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 545, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 104, + 546, + 323, + 563 + ], + "score": 1.0, + "content": "reasonable re-weighting term even as a sub-optimum:", + "type": "text" + }, + { + "bbox": [ + 324, + 545, + 343, + 561 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 546, + 506, + 563 + ], + "score": 1.0, + "content": "gives larger densities on the data where", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 559, + 503, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 559, + 440, + 571 + ], + "score": 1.0, + "content": "the distribution distance between the approximation model and the real model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 441, + 559, + 503, + 571 + ], + "score": 0.92, + "content": "D _ { K L } ( M ^ { * } , M ) )", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 570, + 375, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 311, + 583 + ], + "score": 1.0, + "content": "is larger or the stochasticity of the real model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 311, + 570, + 333, + 582 + ], + "score": 0.88, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 570, + 375, + 583 + ], + "score": 1.0, + "content": ") is larger.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 20.5, + "bbox_fs": [ + 104, + 466, + 506, + 583 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 604, + 203, + 616 + ], + "lines": [ + { + "bbox": [ + 105, + 602, + 205, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 205, + 618 + ], + "score": 1.0, + "content": "A.2 PROOF OF EQ. 6", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 628, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 201, + 642 + ], + "score": 1.0, + "content": "The integral process of", + "type": "text" + }, + { + "bbox": [ + 201, + 630, + 223, + 641 + ], + "score": 0.91, + "content": "D _ { K L }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 628, + 505, + 642 + ], + "score": 1.0, + "content": "in Eq. 5 is intractable in the offline setting as it explicitly requires the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 640, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 251, + 653 + ], + "score": 1.0, + "content": "conditional probability function of", + "type": "text" + }, + { + "bbox": [ + 252, + 641, + 267, + 651 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 640, + 506, + 653 + ], + "score": 1.0, + "content": ". Our motivation for the tractable solution is utilizing the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 650, + 506, + 666 + ], + "spans": [ + { + "bbox": [ + 104, + 650, + 163, + 666 + ], + "score": 1.0, + "content": "offline dataset", + "type": "text" + }, + { + "bbox": [ + 163, + 652, + 185, + 662 + ], + "score": 0.91, + "content": "\\mathcal { D } _ { \\mathrm { r e a l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 650, + 316, + 666 + ], + "score": 1.0, + "content": "as the empirical joint distribution", + "type": "text" + }, + { + "bbox": [ + 317, + 651, + 371, + 663 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 650, + 506, + 666 + ], + "score": 1.0, + "content": "and adopting practical techniques", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 662, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 505, + 675 + ], + "score": 1.0, + "content": "for distance estimation on two joint distributions, like GAN (Goodfellow et al., 2014; Nowozin et al.,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "score": 1.0, + "content": "2016), to approximate Eq. 5. To adopt that solution, we should first transform Eq. 5 into a form under", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 684, + 506, + 697 + ], + "spans": [ + { + "bbox": [ + 104, + 684, + 432, + 697 + ], + "score": 1.0, + "content": "joint distributions. Without loss of generality, we introduce an intermediary policy", + "type": "text" + }, + { + "bbox": [ + 433, + 686, + 439, + 694 + ], + "score": 0.73, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 684, + 480, + 697 + ], + "score": 1.0, + "content": ", of which", + "type": "text" + }, + { + "bbox": [ + 480, + 686, + 488, + 696 + ], + "score": 0.82, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 684, + 506, + 697 + ], + "score": 1.0, + "content": "can", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 694, + 507, + 709 + ], + "spans": [ + { + "bbox": [ + 104, + 694, + 300, + 709 + ], + "score": 1.0, + "content": "be regarded as a specific instance. Then we have", + "type": "text" + }, + { + "bbox": [ + 300, + 695, + 451, + 708 + ], + "score": 0.91, + "content": "M ( x ^ { \\prime } | x , a ) = \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) / \\rho _ { M } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 694, + 483, + 709 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 484, + 696, + 495, + 705 + ], + "score": 0.8, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 496, + 694, + 507, + 709 + ], + "score": 1.0, + "content": "if", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 708, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 107, + 709, + 163, + 722 + ], + "score": 0.91, + "content": "\\rho _ { M } ^ { \\kappa } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 708, + 209, + 724 + ], + "score": 1.0, + "content": ". Assuming", + "type": "text" + }, + { + "bbox": [ + 210, + 709, + 342, + 722 + ], + "score": 0.91, + "content": "\\forall x \\in \\mathcal { X } , \\forall a \\in A , \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 708, + 352, + 724 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 353, + 708, + 413, + 722 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 708, + 506, + 724 + ], + "score": 1.0, + "content": ", which will hold when", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 107, + 720, + 295, + 732 + ], + "spans": [ + { + "bbox": [ + 107, + 723, + 114, + 730 + ], + "score": 0.72, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 720, + 171, + 732 + ], + "score": 1.0, + "content": "overlaps with", + "type": "text" + }, + { + "bbox": [ + 171, + 722, + 178, + 732 + ], + "score": 0.8, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 720, + 295, + 732 + ], + "score": 1.0, + "content": ", then Eq. 5 can transform to:", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 31, + "bbox_fs": [ + 104, + 628, + 507, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "interline_equation", + "bbox": [ + 111, + 95, + 514, + 338 + ], + "lines": [ + { + "bbox": [ + 111, + 95, + 514, + 338 + ], + "spans": [ + { + "bbox": [ + 111, + 95, + 514, + 338 + ], + "score": 0.8, + "content": "\\begin{array} { r l } { \\frac { d } { d t } } & { ( z , z , n ) = \\frac { 1 } { n + 1 } ( 2 \\frac { n + 1 } { n + 1 } , \\frac { 1 } { n + 1 } ; \\frac { 1 } { n + 2 } , z ) ! \\log ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } ) \\log ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) } \\\\ & { - \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , z , n ) ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) \\log \\log ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n ) ) \\mathrm { d } x ^ { \\prime } \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n ) ) \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } \\exp ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } + \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } - \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) \\mathrm { d } y ^ { \\prime } } \\\\ & = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\exp ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } + \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } - \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , \\end{array}", + "type": "interline_equation", + "image_path": "603aa333161db1e721e5232fd0062b09eeebf1530eef988de2c1ca9fa3679555.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 95, + 514, + 176.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 176.0, + 514, + 257.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 257.0, + 514, + 338.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 347, + 241, + 360 + ], + "lines": [ + { + "bbox": [ + 105, + 345, + 238, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 133, + 363 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 348, + 238, + 360 + ], + "score": 0.91, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M } \\rho _ { M ^ { \\ast } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 362, + 505, + 396 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 171, + 376 + ], + "score": 1.0, + "content": "Definition A.3 (", + "type": "text" + }, + { + "bbox": [ + 172, + 363, + 178, + 375 + ], + "score": 0.8, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 362, + 329, + 376 + ], + "score": 1.0, + "content": "-divergence). Given two distributions", + "type": "text" + }, + { + "bbox": [ + 329, + 363, + 338, + 373 + ], + "score": 0.83, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 362, + 356, + 376 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 356, + 363, + 365, + 374 + ], + "score": 0.83, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 362, + 505, + 376 + ], + "score": 1.0, + "content": ", two absolutely continuous density", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 147, + 386 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 147, + 376, + 154, + 385 + ], + "score": 0.72, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 374, + 175, + 386 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 176, + 376, + 182, + 385 + ], + "score": 0.76, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 374, + 432, + 386 + ], + "score": 1.0, + "content": "with respect to a base measure dx defined on the domain", + "type": "text" + }, + { + "bbox": [ + 432, + 375, + 441, + 383 + ], + "score": 0.79, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 374, + 505, + 386 + ], + "score": 1.0, + "content": ", we define the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 384, + 255, + 397 + ], + "spans": [ + { + "bbox": [ + 107, + 385, + 113, + 396 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 384, + 255, + 397 + ], + "score": 1.0, + "content": "-divergence (Nowozin et al., 2016),", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 401, + 380, + 429 + ], + "lines": [ + { + "bbox": [ + 231, + 401, + 380, + 429 + ], + "spans": [ + { + "bbox": [ + 231, + 401, + 380, + 429 + ], + "score": 0.94, + "content": "D _ { f } ( P \\| Q ) = \\int _ { \\mathcal X } q ( x ) f \\left( \\frac { p ( x ) } { q ( x ) } \\right) \\mathrm { d } x ,", + "type": "interline_equation", + "image_path": "35c46f2fda671505c0299be9d24dc9d673a9f6045c71c952c2e221eab3d4810b.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 231, + 401, + 380, + 429 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 433, + 452, + 445 + ], + "lines": [ + { + "bbox": [ + 105, + 432, + 453, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 225, + 447 + ], + "score": 1.0, + "content": "where the generator function", + "type": "text" + }, + { + "bbox": [ + 225, + 433, + 276, + 445 + ], + "score": 0.92, + "content": "f : \\mathbb { R } _ { + } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 432, + 453, + 447 + ], + "score": 1.0, + "content": "is a convex, lower-semicontinuous function.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 453, + 507, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 204, + 474 + ], + "score": 1.0, + "content": "We notice that the terms", + "type": "text" + }, + { + "bbox": [ + 204, + 455, + 320, + 473 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 453, + 339, + 474 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 339, + 455, + 433, + 472 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } { \\rho _ { M } ^ { \\kappa } ( x , a ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 453, + 506, + 474 + ], + "score": 1.0, + "content": "are the integrated", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 470, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 342, + 484 + ], + "score": 1.0, + "content": "functions in reverse KL divergence, which is an instance of", + "type": "text" + }, + { + "bbox": [ + 342, + 473, + 349, + 483 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 470, + 396, + 484 + ], + "score": 1.0, + "content": "function in", + "type": "text" + }, + { + "bbox": [ + 396, + 473, + 402, + 483 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 470, + 506, + 484 + ], + "score": 1.0, + "content": "-divergence (See Reverse-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 480, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 455, + 495 + ], + "score": 1.0, + "content": "KL divergence of Tab.1 in (Nowozin et al., 2016) for more details). Replacing that form", + "type": "text" + }, + { + "bbox": [ + 456, + 481, + 483, + 495 + ], + "score": 0.91, + "content": "q \\log { \\frac { q } { p } }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 480, + 506, + 495 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 494, + 327, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 133, + 512 + ], + "score": 0.92, + "content": "q f ( { \\underline { { p } } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 494, + 303, + 513 + ], + "score": 1.0, + "content": ", we obtain a generalized representation of", + "type": "text" + }, + { + "bbox": [ + 304, + 495, + 322, + 510 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 494, + 327, + 513 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 516, + 514, + 547 + ], + "lines": [ + { + "bbox": [ + 111, + 516, + 514, + 547 + ], + "spans": [ + { + "bbox": [ + 111, + 516, + 514, + 547 + ], + "score": 0.89, + "content": "\\tilde { \\bar { \\nu } } _ { M ^ { * } } ^ { * } : = \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\Bigg ( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) - H _ { M ^ { * } } ( x , a ) \\right)", + "type": "interline_equation", + "image_path": "beac113262e3629c20136f81ade1f37648176e45e901733b49b520980dec2112.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 111, + 516, + 514, + 526.3333333333334 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 111, + 526.3333333333334, + 514, + 536.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 111, + 536.6666666666667, + 514, + 547.0000000000001 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 569, + 219, + 581 + ], + "lines": [ + { + "bbox": [ + 106, + 568, + 221, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 221, + 582 + ], + "score": 1.0, + "content": "A.3 PROOF OF THM. 4.4", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 590, + 326, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 326, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 326, + 603 + ], + "score": 1.0, + "content": "We first introduce several useful lemmas for the proof.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 105, + 605, + 504, + 628 + ], + "lines": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "Lemma A.4. Rearrangement inequality The rearrangement inequality states that, for two sequences", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 615, + 356, + 628 + ], + "spans": [ + { + "bbox": [ + 107, + 616, + 189, + 627 + ], + "score": 0.91, + "content": "a _ { 1 } \\geq a _ { 2 } \\geq . . . \\geq a _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 615, + 208, + 628 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 208, + 616, + 288, + 627 + ], + "score": 0.91, + "content": "b _ { 1 } \\geq b _ { 2 } \\geq . . . \\geq b _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 615, + 356, + 628 + ], + "score": 1.0, + "content": ", the inequalities", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 632, + 501, + 647 + ], + "lines": [ + { + "bbox": [ + 111, + 632, + 501, + 647 + ], + "spans": [ + { + "bbox": [ + 111, + 632, + 501, + 647 + ], + "score": 0.89, + "content": "a _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \\cdot \\cdot \\cdot + a _ { n } b _ { n } \\geq a _ { 1 } b _ { \\pi ( 1 ) } + a _ { 2 } b _ { \\pi ( 2 ) } + \\cdot \\cdot \\cdot + a _ { n } b _ { \\pi ( n ) } \\geq a _ { 1 } b _ { n } + a _ { 2 } b _ { n - 1 } + \\cdot \\cdot \\cdot + a _ { n } b _ { 1 }", + "type": "interline_equation", + "image_path": "9ebcb262c29a009d3359e82dcd3b1f83c045ef4f7c6892244fad87fa1d23b207.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 111, + 632, + 501, + 647 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 651, + 378, + 665 + ], + "lines": [ + { + "bbox": [ + 106, + 651, + 376, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 155, + 666 + ], + "score": 1.0, + "content": "hold, where", + "type": "text" + }, + { + "bbox": [ + 156, + 652, + 241, + 665 + ], + "score": 0.93, + "content": "\\pi ( 1 ) , \\pi ( 2 ) , \\ldots , \\pi ( n )", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 651, + 330, + 666 + ], + "score": 1.0, + "content": "is any permutation of", + "type": "text" + }, + { + "bbox": [ + 330, + 653, + 373, + 664 + ], + "score": 0.89, + "content": "1 , 2 , \\ldots , n", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 651, + 376, + 666 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 105, + 667, + 489, + 680 + ], + "lines": [ + { + "bbox": [ + 104, + 665, + 490, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 665, + 240, + 681 + ], + "score": 1.0, + "content": "Lemma A.5. For two sequences", + "type": "text" + }, + { + "bbox": [ + 240, + 667, + 322, + 678 + ], + "score": 0.89, + "content": "a _ { 1 } \\geq a _ { 2 } \\geq . . . \\geq a _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 665, + 342, + 681 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 342, + 667, + 421, + 678 + ], + "score": 0.91, + "content": "b _ { 1 } \\geq b _ { 2 } \\geq . . . \\geq b _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 665, + 490, + 681 + ], + "score": 1.0, + "content": ", the inequalities", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 683, + 367, + 716 + ], + "lines": [ + { + "bbox": [ + 243, + 683, + 367, + 716 + ], + "spans": [ + { + "bbox": [ + 243, + 683, + 367, + 716 + ], + "score": 0.94, + "content": "\\sum _ { i = 1 } ^ { n } { \\frac { 1 } { n } } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } { \\frac { 1 } { n } } a _ { i } \\sum { \\frac { 1 } { n } } b _ { i }", + "type": "interline_equation", + "image_path": "a9c1c632cb3afc7394107043c403f4aeb12562063d12e4044609df809363804a.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 243, + 683, + 367, + 699.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 243, + 699.5, + 367, + 716.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 128, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 720, + 129, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 129, + 733 + ], + "score": 1.0, + "content": "hold.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "17", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "interline_equation", + "bbox": [ + 111, + 95, + 514, + 338 + ], + "lines": [ + { + "bbox": [ + 111, + 95, + 514, + 338 + ], + "spans": [ + { + "bbox": [ + 111, + 95, + 514, + 338 + ], + "score": 0.8, + "content": "\\begin{array} { r l } { \\frac { d } { d t } } & { ( z , z , n ) = \\frac { 1 } { n + 1 } ( 2 \\frac { n + 1 } { n + 1 } , \\frac { 1 } { n + 1 } ; \\frac { 1 } { n + 2 } , z ) ! \\log ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } ) \\log ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) } \\\\ & { - \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , z , n ) ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) \\log \\log ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n ) ) \\mathrm { d } x ^ { \\prime } \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n ) ) \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } \\exp ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } + \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } - \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) \\mathrm { d } y ^ { \\prime } } \\\\ & = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\exp ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } + \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } - \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , \\end{array}", + "type": "interline_equation", + "image_path": "603aa333161db1e721e5232fd0062b09eeebf1530eef988de2c1ca9fa3679555.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 95, + 514, + 176.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 176.0, + 514, + 257.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 257.0, + 514, + 338.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 347, + 241, + 360 + ], + "lines": [ + { + "bbox": [ + 105, + 345, + 238, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 133, + 363 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 348, + 238, + 360 + ], + "score": 0.91, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M } \\rho _ { M ^ { \\ast } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 345, + 238, + 363 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 362, + 505, + 396 + ], + "lines": [ + { + "bbox": [ + 105, + 362, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 171, + 376 + ], + "score": 1.0, + "content": "Definition A.3 (", + "type": "text" + }, + { + "bbox": [ + 172, + 363, + 178, + 375 + ], + "score": 0.8, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 362, + 329, + 376 + ], + "score": 1.0, + "content": "-divergence). Given two distributions", + "type": "text" + }, + { + "bbox": [ + 329, + 363, + 338, + 373 + ], + "score": 0.83, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 362, + 356, + 376 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 356, + 363, + 365, + 374 + ], + "score": 0.83, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 362, + 505, + 376 + ], + "score": 1.0, + "content": ", two absolutely continuous density", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 147, + 386 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 147, + 376, + 154, + 385 + ], + "score": 0.72, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 374, + 175, + 386 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 176, + 376, + 182, + 385 + ], + "score": 0.76, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 374, + 432, + 386 + ], + "score": 1.0, + "content": "with respect to a base measure dx defined on the domain", + "type": "text" + }, + { + "bbox": [ + 432, + 375, + 441, + 383 + ], + "score": 0.79, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 374, + 505, + 386 + ], + "score": 1.0, + "content": ", we define the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 384, + 255, + 397 + ], + "spans": [ + { + "bbox": [ + 107, + 385, + 113, + 396 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 384, + 255, + 397 + ], + "score": 1.0, + "content": "-divergence (Nowozin et al., 2016),", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 362, + 505, + 397 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 401, + 380, + 429 + ], + "lines": [ + { + "bbox": [ + 231, + 401, + 380, + 429 + ], + "spans": [ + { + "bbox": [ + 231, + 401, + 380, + 429 + ], + "score": 0.94, + "content": "D _ { f } ( P \\| Q ) = \\int _ { \\mathcal X } q ( x ) f \\left( \\frac { p ( x ) } { q ( x ) } \\right) \\mathrm { d } x ,", + "type": "interline_equation", + "image_path": "35c46f2fda671505c0299be9d24dc9d673a9f6045c71c952c2e221eab3d4810b.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 231, + 401, + 380, + 429 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 433, + 452, + 445 + ], + "lines": [ + { + "bbox": [ + 105, + 432, + 453, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 225, + 447 + ], + "score": 1.0, + "content": "where the generator function", + "type": "text" + }, + { + "bbox": [ + 225, + 433, + 276, + 445 + ], + "score": 0.92, + "content": "f : \\mathbb { R } _ { + } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 432, + 453, + 447 + ], + "score": 1.0, + "content": "is a convex, lower-semicontinuous function.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 432, + 453, + 447 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 453, + 507, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 204, + 474 + ], + "score": 1.0, + "content": "We notice that the terms", + "type": "text" + }, + { + "bbox": [ + 204, + 455, + 320, + 473 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 453, + 339, + 474 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 339, + 455, + 433, + 472 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } { \\rho _ { M } ^ { \\kappa } ( x , a ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 453, + 506, + 474 + ], + "score": 1.0, + "content": "are the integrated", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 470, + 506, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 342, + 484 + ], + "score": 1.0, + "content": "functions in reverse KL divergence, which is an instance of", + "type": "text" + }, + { + "bbox": [ + 342, + 473, + 349, + 483 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 470, + 396, + 484 + ], + "score": 1.0, + "content": "function in", + "type": "text" + }, + { + "bbox": [ + 396, + 473, + 402, + 483 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 470, + 506, + 484 + ], + "score": 1.0, + "content": "-divergence (See Reverse-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 480, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 455, + 495 + ], + "score": 1.0, + "content": "KL divergence of Tab.1 in (Nowozin et al., 2016) for more details). Replacing that form", + "type": "text" + }, + { + "bbox": [ + 456, + 481, + 483, + 495 + ], + "score": 0.91, + "content": "q \\log { \\frac { q } { p } }", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 480, + 506, + 495 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 494, + 327, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 133, + 512 + ], + "score": 0.92, + "content": "q f ( { \\underline { { p } } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 494, + 303, + 513 + ], + "score": 1.0, + "content": ", we obtain a generalized representation of", + "type": "text" + }, + { + "bbox": [ + 304, + 495, + 322, + 510 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 494, + 327, + 513 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 453, + 506, + 513 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 516, + 514, + 547 + ], + "lines": [ + { + "bbox": [ + 111, + 516, + 514, + 547 + ], + "spans": [ + { + "bbox": [ + 111, + 516, + 514, + 547 + ], + "score": 0.89, + "content": "\\tilde { \\bar { \\nu } } _ { M ^ { * } } ^ { * } : = \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\Bigg ( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) - H _ { M ^ { * } } ( x , a ) \\right)", + "type": "interline_equation", + "image_path": "beac113262e3629c20136f81ade1f37648176e45e901733b49b520980dec2112.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 111, + 516, + 514, + 526.3333333333334 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 111, + 526.3333333333334, + 514, + 536.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 111, + 536.6666666666667, + 514, + 547.0000000000001 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 569, + 219, + 581 + ], + "lines": [ + { + "bbox": [ + 106, + 568, + 221, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 221, + 582 + ], + "score": 1.0, + "content": "A.3 PROOF OF THM. 4.4", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 590, + 326, + 602 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 326, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 326, + 603 + ], + "score": 1.0, + "content": "We first introduce several useful lemmas for the proof.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 588, + 326, + 603 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 605, + 504, + 628 + ], + "lines": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "Lemma A.4. Rearrangement inequality The rearrangement inequality states that, for two sequences", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 615, + 356, + 628 + ], + "spans": [ + { + "bbox": [ + 107, + 616, + 189, + 627 + ], + "score": 0.91, + "content": "a _ { 1 } \\geq a _ { 2 } \\geq . . . \\geq a _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 615, + 208, + 628 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 208, + 616, + 288, + 627 + ], + "score": 0.91, + "content": "b _ { 1 } \\geq b _ { 2 } \\geq . . . \\geq b _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 615, + 356, + 628 + ], + "score": 1.0, + "content": ", the inequalities", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 604, + 505, + 628 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 632, + 501, + 647 + ], + "lines": [ + { + "bbox": [ + 111, + 632, + 501, + 647 + ], + "spans": [ + { + "bbox": [ + 111, + 632, + 501, + 647 + ], + "score": 0.89, + "content": "a _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \\cdot \\cdot \\cdot + a _ { n } b _ { n } \\geq a _ { 1 } b _ { \\pi ( 1 ) } + a _ { 2 } b _ { \\pi ( 2 ) } + \\cdot \\cdot \\cdot + a _ { n } b _ { \\pi ( n ) } \\geq a _ { 1 } b _ { n } + a _ { 2 } b _ { n - 1 } + \\cdot \\cdot \\cdot + a _ { n } b _ { 1 }", + "type": "interline_equation", + "image_path": "9ebcb262c29a009d3359e82dcd3b1f83c045ef4f7c6892244fad87fa1d23b207.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 111, + 632, + 501, + 647 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 651, + 378, + 665 + ], + "lines": [ + { + "bbox": [ + 106, + 651, + 376, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 155, + 666 + ], + "score": 1.0, + "content": "hold, where", + "type": "text" + }, + { + "bbox": [ + 156, + 652, + 241, + 665 + ], + "score": 0.93, + "content": "\\pi ( 1 ) , \\pi ( 2 ) , \\ldots , \\pi ( n )", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 651, + 330, + 666 + ], + "score": 1.0, + "content": "is any permutation of", + "type": "text" + }, + { + "bbox": [ + 330, + 653, + 373, + 664 + ], + "score": 0.89, + "content": "1 , 2 , \\ldots , n", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 651, + 376, + 666 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 651, + 376, + 666 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 667, + 489, + 680 + ], + "lines": [ + { + "bbox": [ + 104, + 665, + 490, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 665, + 240, + 681 + ], + "score": 1.0, + "content": "Lemma A.5. For two sequences", + "type": "text" + }, + { + "bbox": [ + 240, + 667, + 322, + 678 + ], + "score": 0.89, + "content": "a _ { 1 } \\geq a _ { 2 } \\geq . . . \\geq a _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 665, + 342, + 681 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 342, + 667, + 421, + 678 + ], + "score": 0.91, + "content": "b _ { 1 } \\geq b _ { 2 } \\geq . . . \\geq b _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 665, + 490, + 681 + ], + "score": 1.0, + "content": ", the inequalities", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 104, + 665, + 490, + 681 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 683, + 367, + 716 + ], + "lines": [ + { + "bbox": [ + 243, + 683, + 367, + 716 + ], + "spans": [ + { + "bbox": [ + 243, + 683, + 367, + 716 + ], + "score": 0.94, + "content": "\\sum _ { i = 1 } ^ { n } { \\frac { 1 } { n } } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } { \\frac { 1 } { n } } a _ { i } \\sum { \\frac { 1 } { n } } b _ { i }", + "type": "interline_equation", + "image_path": "a9c1c632cb3afc7394107043c403f4aeb12562063d12e4044609df809363804a.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 243, + 683, + 367, + 699.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 243, + 699.5, + 367, + 716.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 128, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 720, + 129, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 129, + 733 + ], + "score": 1.0, + "content": "hold.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 720, + 129, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 290, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 290, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 290, + 97 + ], + "score": 1.0, + "content": "Proof. By rearrangement inequality, we have", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 97, + 387, + 253 + ], + "lines": [ + { + "bbox": [ + 221, + 97, + 385, + 253 + ], + "spans": [ + { + "bbox": [ + 221, + 97, + 385, + 253 + ], + "score": 0.85, + "content": "\\begin{array} { l } { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \\cdots + a _ { n } b _ { n } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 2 } + a _ { 2 } b _ { 3 } + \\cdots + a _ { n } b _ { 1 } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \\cdots + a _ { n } b _ { 2 } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \\cdots + a _ { n } b _ { 2 } } } \\\\ { { \\displaystyle \\qquad \\vdots \\qquad } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { n } + a _ { 2 } b _ { 1 } + \\cdots + a _ { n } b _ { n - 1 } } } \\end{array}", + "type": "interline_equation", + "image_path": "54f0537a515b0381540541728acfc06db53a99a7806d3346c8b48a13ebe7b7cf.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 221, + 97, + 387, + 112.6 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 221, + 112.6, + 387, + 128.2 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 221, + 128.2, + 387, + 143.79999999999998 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 221, + 143.79999999999998, + 387, + 159.39999999999998 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 221, + 159.39999999999998, + 387, + 174.99999999999997 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 221, + 174.99999999999997, + 387, + 190.59999999999997 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 221, + 190.59999999999997, + 387, + 206.19999999999996 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 221, + 206.19999999999996, + 387, + 221.79999999999995 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 221, + 221.79999999999995, + 387, + 237.39999999999995 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 221, + 237.39999999999995, + 387, + 252.99999999999994 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 253, + 164, + 264 + ], + "lines": [ + { + "bbox": [ + 105, + 252, + 164, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 164, + 265 + ], + "score": 1.0, + "content": "Then we have", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 265, + 367, + 333 + ], + "lines": [ + { + "bbox": [ + 243, + 265, + 367, + 333 + ], + "spans": [ + { + "bbox": [ + 243, + 265, + 367, + 333 + ], + "score": 0.92, + "content": "\\begin{array} { l } { { \\displaystyle n \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } a _ { i } \\sum _ { i = 1 } ^ { n } b _ { i } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } \\frac { 1 } { n } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } \\frac { 1 } { n } a _ { i } \\sum _ { n } \\frac { 1 } { n } b _ { i } } } \\end{array}", + "type": "interline_equation", + "image_path": "118a0a4436f20ba33227d29825c64f45d04c5eec5a70f85acc6854cdfac4dad9.jpg" + } + ] + } + ], + "index": 12.5, + "virtual_lines": [ + { + "bbox": [ + 243, + 265, + 367, + 299.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 243, + 299.0, + 367, + 333.0 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 356, + 371, + 369 + ], + "lines": [ + { + "bbox": [ + 105, + 355, + 371, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 371, + 370 + ], + "score": 1.0, + "content": "Now we extend Lemma A.5 into the continuous integral scenario:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 505, + 394 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 190, + 383 + ], + "score": 1.0, + "content": "Lemma A.6. Given", + "type": "text" + }, + { + "bbox": [ + 190, + 370, + 221, + 381 + ], + "score": 0.88, + "content": "\\mathcal { X } \\subset \\mathbb { R } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 369, + 295, + 383 + ], + "score": 1.0, + "content": ", for two functions", + "type": "text" + }, + { + "bbox": [ + 295, + 370, + 342, + 382 + ], + "score": 0.91, + "content": "f : \\mathcal { X } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 369, + 360, + 383 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 360, + 371, + 406, + 382 + ], + "score": 0.91, + "content": "g : \\mathcal { X } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 369, + 426, + 383 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 426, + 370, + 478, + 383 + ], + "score": 0.92, + "content": "f \\left( x \\right) \\geq f \\left( y \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "if and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 381, + 294, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 134, + 394 + ], + "score": 1.0, + "content": "only if", + "type": "text" + }, + { + "bbox": [ + 134, + 381, + 185, + 394 + ], + "score": 0.74, + "content": "g ( x ) \\geq g ( y )", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 381, + 190, + 394 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 191, + 382, + 234, + 393 + ], + "score": 0.6, + "content": "\\forall x , y \\in { \\mathcal { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 381, + 294, + 394 + ], + "score": 1.0, + "content": ", the inequality", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 396, + 419, + 422 + ], + "lines": [ + { + "bbox": [ + 192, + 396, + 419, + 422 + ], + "spans": [ + { + "bbox": [ + 192, + 396, + 419, + 422 + ], + "score": 0.93, + "content": "\\int _ { \\mathcal { X } } p ( x ) f ( x ) g ( x ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } p ( x ) f ( x ) \\mathrm { d } x \\int _ { \\mathcal { X } } p ( x ) g ( x ) \\mathrm { d } x", + "type": "interline_equation", + "image_path": "072e0bb93a1ea14e45290006942c024fa487618183a385738ddaa285f223fa11.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 192, + 396, + 419, + 422 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 425, + 382, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 383, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 159, + 440 + ], + "score": 1.0, + "content": "holds, where", + "type": "text" + }, + { + "bbox": [ + 160, + 426, + 205, + 437 + ], + "score": 0.93, + "content": "p : \\mathcal { X } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 423, + 224, + 440 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 225, + 425, + 298, + 438 + ], + "score": 0.92, + "content": "p ( x ) > 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 423, + 317, + 440 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 317, + 425, + 380, + 439 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\int _ { \\mathcal { X } } p ( x ) \\mathrm { d } x = 1 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 423, + 383, + 440 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 449, + 380, + 463 + ], + "lines": [ + { + "bbox": [ + 106, + 448, + 380, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 161, + 464 + ], + "score": 1.0, + "content": "Proof. Since", + "type": "text" + }, + { + "bbox": [ + 162, + 450, + 341, + 462 + ], + "score": 0.91, + "content": "( f ( x ) - f ( y ) ) ( g ( x ) - g ( y ) ) \\geq 0 , \\forall x , y \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 448, + 380, + 464 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 106, + 480, + 514, + 734 + ], + "lines": [ + { + "bbox": [ + 111, + 480, + 514, + 734 + ], + "spans": [ + { + "bbox": [ + 111, + 480, + 514, + 734 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { d } \\xi \\Bigr ( x \\mathrm { S p } ( \\xi ) + \\mathrm { i } \\omega ( 2 \\xi ) \\mathrm { S p } ( \\xi ) + \\mathrm { i } \\omega \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { d } \\xi \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ \\end{array}", + "type": "interline_equation", + "image_path": "5bcefd60596a1023d02424dcb1c80888eb8012c6a9d062c03269c6330e84701f.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 106, + 480, + 514, + 564.6666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 106, + 564.6666666666666, + 514, + 649.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 106, + 649.3333333333333, + 514, + 733.9999999999999 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 333, + 505, + 344 + ], + "lines": [ + { + "bbox": [ + 496, + 335, + 504, + 343 + ], + "spans": [ + { + "bbox": [ + 496, + 335, + 504, + 343 + ], + "score": 0.994, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 290, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 290, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 290, + 97 + ], + "score": 1.0, + "content": "Proof. By rearrangement inequality, we have", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 105, + 81, + 290, + 97 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 221, + 97, + 387, + 253 + ], + "lines": [ + { + "bbox": [ + 221, + 97, + 385, + 253 + ], + "spans": [ + { + "bbox": [ + 221, + 97, + 385, + 253 + ], + "score": 0.85, + "content": "\\begin{array} { l } { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \\cdots + a _ { n } b _ { n } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 2 } + a _ { 2 } b _ { 3 } + \\cdots + a _ { n } b _ { 1 } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \\cdots + a _ { n } b _ { 2 } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \\cdots + a _ { n } b _ { 2 } } } \\\\ { { \\displaystyle \\qquad \\vdots \\qquad } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { n } + a _ { 2 } b _ { 1 } + \\cdots + a _ { n } b _ { n - 1 } } } \\end{array}", + "type": "interline_equation", + "image_path": "54f0537a515b0381540541728acfc06db53a99a7806d3346c8b48a13ebe7b7cf.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 221, + 97, + 387, + 112.6 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 221, + 112.6, + 387, + 128.2 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 221, + 128.2, + 387, + 143.79999999999998 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 221, + 143.79999999999998, + 387, + 159.39999999999998 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 221, + 159.39999999999998, + 387, + 174.99999999999997 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 221, + 174.99999999999997, + 387, + 190.59999999999997 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 221, + 190.59999999999997, + 387, + 206.19999999999996 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 221, + 206.19999999999996, + 387, + 221.79999999999995 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 221, + 221.79999999999995, + 387, + 237.39999999999995 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 221, + 237.39999999999995, + 387, + 252.99999999999994 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 253, + 164, + 264 + ], + "lines": [ + { + "bbox": [ + 105, + 252, + 164, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 164, + 265 + ], + "score": 1.0, + "content": "Then we have", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 252, + 164, + 265 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 265, + 367, + 333 + ], + "lines": [ + { + "bbox": [ + 243, + 265, + 367, + 333 + ], + "spans": [ + { + "bbox": [ + 243, + 265, + 367, + 333 + ], + "score": 0.92, + "content": "\\begin{array} { l } { { \\displaystyle n \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } a _ { i } \\sum _ { i = 1 } ^ { n } b _ { i } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } \\frac { 1 } { n } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } \\frac { 1 } { n } a _ { i } \\sum _ { n } \\frac { 1 } { n } b _ { i } } } \\end{array}", + "type": "interline_equation", + "image_path": "118a0a4436f20ba33227d29825c64f45d04c5eec5a70f85acc6854cdfac4dad9.jpg" + } + ] + } + ], + "index": 12.5, + "virtual_lines": [ + { + "bbox": [ + 243, + 265, + 367, + 299.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 243, + 299.0, + 367, + 333.0 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 356, + 371, + 369 + ], + "lines": [ + { + "bbox": [ + 105, + 355, + 371, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 371, + 370 + ], + "score": 1.0, + "content": "Now we extend Lemma A.5 into the continuous integral scenario:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 355, + 371, + 370 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 505, + 394 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 190, + 383 + ], + "score": 1.0, + "content": "Lemma A.6. Given", + "type": "text" + }, + { + "bbox": [ + 190, + 370, + 221, + 381 + ], + "score": 0.88, + "content": "\\mathcal { X } \\subset \\mathbb { R } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 369, + 295, + 383 + ], + "score": 1.0, + "content": ", for two functions", + "type": "text" + }, + { + "bbox": [ + 295, + 370, + 342, + 382 + ], + "score": 0.91, + "content": "f : \\mathcal { X } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 369, + 360, + 383 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 360, + 371, + 406, + 382 + ], + "score": 0.91, + "content": "g : \\mathcal { X } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 369, + 426, + 383 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 426, + 370, + 478, + 383 + ], + "score": 0.92, + "content": "f \\left( x \\right) \\geq f \\left( y \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "if and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 381, + 294, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 134, + 394 + ], + "score": 1.0, + "content": "only if", + "type": "text" + }, + { + "bbox": [ + 134, + 381, + 185, + 394 + ], + "score": 0.74, + "content": "g ( x ) \\geq g ( y )", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 381, + 190, + 394 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 191, + 382, + 234, + 393 + ], + "score": 0.6, + "content": "\\forall x , y \\in { \\mathcal { X } }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 381, + 294, + 394 + ], + "score": 1.0, + "content": ", the inequality", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 369, + 506, + 394 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 192, + 396, + 419, + 422 + ], + "lines": [ + { + "bbox": [ + 192, + 396, + 419, + 422 + ], + "spans": [ + { + "bbox": [ + 192, + 396, + 419, + 422 + ], + "score": 0.93, + "content": "\\int _ { \\mathcal { X } } p ( x ) f ( x ) g ( x ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } p ( x ) f ( x ) \\mathrm { d } x \\int _ { \\mathcal { X } } p ( x ) g ( x ) \\mathrm { d } x", + "type": "interline_equation", + "image_path": "072e0bb93a1ea14e45290006942c024fa487618183a385738ddaa285f223fa11.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 192, + 396, + 419, + 422 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 425, + 382, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 423, + 383, + 440 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 159, + 440 + ], + "score": 1.0, + "content": "holds, where", + "type": "text" + }, + { + "bbox": [ + 160, + 426, + 205, + 437 + ], + "score": 0.93, + "content": "p : \\mathcal { X } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 423, + 224, + 440 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 225, + 425, + 298, + 438 + ], + "score": 0.92, + "content": "p ( x ) > 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 423, + 317, + 440 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 317, + 425, + 380, + 439 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\int _ { \\mathcal { X } } p ( x ) \\mathrm { d } x = 1 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 423, + 383, + 440 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 423, + 383, + 440 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 449, + 380, + 463 + ], + "lines": [ + { + "bbox": [ + 106, + 448, + 380, + 464 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 161, + 464 + ], + "score": 1.0, + "content": "Proof. Since", + "type": "text" + }, + { + "bbox": [ + 162, + 450, + 341, + 462 + ], + "score": 0.91, + "content": "( f ( x ) - f ( y ) ) ( g ( x ) - g ( y ) ) \\geq 0 , \\forall x , y \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 448, + 380, + 464 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 106, + 448, + 380, + 464 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 106, + 480, + 514, + 734 + ], + "lines": [ + { + "bbox": [ + 111, + 480, + 514, + 734 + ], + "spans": [ + { + "bbox": [ + 111, + 480, + 514, + 734 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { d } \\xi \\Bigr ( x \\mathrm { S p } ( \\xi ) + \\mathrm { i } \\omega ( 2 \\xi ) \\mathrm { S p } ( \\xi ) + \\mathrm { i } \\omega \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { d } \\xi \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ \\end{array}", + "type": "interline_equation", + "image_path": "5bcefd60596a1023d02424dcb1c80888eb8012c6a9d062c03269c6330e84701f.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 106, + 480, + 514, + 564.6666666666666 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 106, + 564.6666666666666, + 514, + 649.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 106, + 649.3333333333333, + 514, + 733.9999999999999 + ], + "spans": [], + "index": 22 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 100, + 506, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 98, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 221, + 118 + ], + "score": 1.0, + "content": "Corollary A.7. Let g( p(x)q(x) )", + "type": "text" + }, + { + "bbox": [ + 191, + 100, + 279, + 117 + ], + "score": 0.95, + "content": "\\begin{array} { r } { g ( \\frac { p ( x ) } { q ( x ) } ) = - \\log \\frac { p ( x ) } { q ( x ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 101, + 308, + 115 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 308, + 102, + 387, + 115 + ], + "score": 0.91, + "content": "p ( x ) > 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 101, + 407, + 115 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 408, + 102, + 486, + 115 + ], + "score": 0.92, + "content": "q ( x ) > 0 , \\forall x \\in \\mathcal { X } _ { }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 101, + 505, + 115 + ], + "score": 1.0, + "content": ", for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 115, + 192, + 128 + ], + "spans": [ + { + "bbox": [ + 107, + 116, + 132, + 126 + ], + "score": 0.87, + "content": "\\upsilon > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 115, + 192, + 128 + ], + "score": 1.0, + "content": ", the inequality", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 154, + 132, + 454, + 160 + ], + "lines": [ + { + "bbox": [ + 154, + 132, + 454, + 160 + ], + "spans": [ + { + "bbox": [ + 154, + 132, + 454, + 160 + ], + "score": 0.92, + "content": "\\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x ,", + "type": "interline_equation", + "image_path": "be15a1d06758081716993cce8840fa46cc7b03706dd260cab3e53516d187e7a5.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 154, + 132, + 454, + 160 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 505, + 189 + ], + "lines": [ + { + "bbox": [ + 105, + 164, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 140, + 180 + ], + "score": 1.0, + "content": "holds if", + "type": "text" + }, + { + "bbox": [ + 141, + 165, + 222, + 178 + ], + "score": 0.9, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 164, + 346, + 180 + ], + "score": 1.0, + "content": ". It is not always satisfied for", + "type": "text" + }, + { + "bbox": [ + 346, + 167, + 353, + 178 + ], + "score": 0.82, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 164, + 406, + 180 + ], + "score": 1.0, + "content": "functions of", + "type": "text" + }, + { + "bbox": [ + 407, + 167, + 413, + 178 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 164, + 506, + 180 + ], + "score": 1.0, + "content": "-divergence. We list a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 176, + 285, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 166, + 189 + ], + "score": 1.0, + "content": "comparison of", + "type": "text" + }, + { + "bbox": [ + 167, + 178, + 173, + 189 + ], + "score": 0.82, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 176, + 285, + 189 + ], + "score": 1.0, + "content": "on that condition in Tab. 2.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 506, + 259 + ], + "lines": [ + { + "bbox": [ + 104, + 199, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 104, + 199, + 136, + 217 + ], + "score": 1.0, + "content": "Proof.", + "type": "text" + }, + { + "bbox": [ + 136, + 202, + 289, + 216 + ], + "score": 0.92, + "content": "\\begin{array} { r } { g ^ { \\prime } ( x ) = - \\log x = - \\frac { 1 } { x } < 0 , \\forall x \\in \\mathcal { X } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 199, + 331, + 217 + ], + "score": 1.0, + "content": ". Suppose", + "type": "text" + }, + { + "bbox": [ + 331, + 203, + 410, + 215 + ], + "score": 0.91, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 199, + 451, + 217 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 452, + 202, + 505, + 215 + ], + "score": 0.92, + "content": "f \\left( x \\right) \\geq f \\left( y \\right)", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 101, + 209, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 101, + 209, + 163, + 239 + ], + "score": 1.0, + "content": "if and only if", + "type": "text" + }, + { + "bbox": [ + 106, + 231, + 187, + 248 + ], + "score": 0.45, + "content": "g ( \\frac { p ( y ) } { q ( y ) } ) , \\forall x , y \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 217, + 263, + 230 + ], + "score": 0.63, + "content": "g ( x ) \\geq g ( y ) , \\forall x , y \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 221, + 241, + 257 + ], + "score": 1.0, + "content": "holds for all", + "type": "text" + }, + { + "bbox": [ + 241, + 234, + 267, + 244 + ], + "score": 0.89, + "content": "\\upsilon > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 209, + 316, + 239 + ], + "score": 1.0, + "content": "holds. Thus", + "type": "text" + }, + { + "bbox": [ + 268, + 221, + 323, + 257 + ], + "score": 1.0, + "content": ". By defining", + "type": "text" + }, + { + "bbox": [ + 317, + 215, + 404, + 232 + ], + "score": 0.92, + "content": "\\begin{array} { r } { f ( v \\frac { p ( x ) } { q ( x ) } ) \\geq f ( v \\frac { p ( y ) } { q ( y ) } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 232, + 400, + 248 + ], + "score": 0.92, + "content": "\\begin{array} { r } { F ( x ) = f ( v \\frac { p ( x ) } { q ( x ) } ) ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 221, + 420, + 257 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 404, + 209, + 461, + 239 + ], + "score": 1.0, + "content": "if and only if", + "type": "text" + }, + { + "bbox": [ + 420, + 232, + 486, + 248 + ], + "score": 0.93, + "content": "\\begin{array} { r } { G ( x ) = g ( \\frac { p ( x ) } { q ( x ) } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 215, + 505, + 232 + ], + "score": 0.91, + "content": "g \\big ( \\frac { p ( x ) } { q ( x ) } \\big ) \\geq", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 233, + 505, + 245 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 246, + 221, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 221, + 258 + ], + "score": 1.0, + "content": "using Lemma A.6, we have:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 264, + 423, + 290 + ], + "lines": [ + { + "bbox": [ + 186, + 264, + 423, + 290 + ], + "spans": [ + { + "bbox": [ + 186, + 264, + 423, + 290 + ], + "score": 0.92, + "content": "\\int _ { \\mathcal { X } } q ( x ) F ( x ) G ( x ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) F ( x ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) G ( x ) \\mathrm { d } x .", + "type": "interline_equation", + "image_path": "c3950b114a613cb86264afb680420911ff7351973e16a1ff704469cd03c1a77b.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 186, + 264, + 423, + 290 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 167, + 306 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 168, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 168, + 307 + ], + "score": 1.0, + "content": "Then we know", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 311, + 454, + 338 + ], + "lines": [ + { + "bbox": [ + 156, + 311, + 454, + 338 + ], + "spans": [ + { + "bbox": [ + 156, + 311, + 454, + 338 + ], + "score": 0.91, + "content": "\\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x", + "type": "interline_equation", + "image_path": "fcb2558754c0452cee69dad339c5d1b39548a5bfe344d27640856e105bdf7c58.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 156, + 311, + 454, + 338 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 343, + 132, + 354 + ], + "lines": [ + { + "bbox": [ + 104, + 342, + 134, + 356 + ], + "spans": [ + { + "bbox": [ + 104, + 342, + 134, + 356 + ], + "score": 1.0, + "content": "holds.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "table", + "bbox": [ + 147, + 394, + 461, + 485 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 183, + 382, + 427, + 394 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 183, + 380, + 428, + 397 + ], + "spans": [ + { + "bbox": [ + 183, + 380, + 273, + 397 + ], + "score": 1.0, + "content": "Table 2: Properties of", + "type": "text" + }, + { + "bbox": [ + 273, + 382, + 350, + 394 + ], + "score": 0.91, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 380, + 366, + 397 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 367, + 383, + 374, + 394 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 380, + 428, + 397 + ], + "score": 1.0, + "content": "-divergences.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "table_body", + "bbox": [ + 147, + 394, + 461, + 485 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 394, + 461, + 485 + ], + "spans": [ + { + "bbox": [ + 147, + 394, + 461, + 485 + ], + "score": 0.982, + "html": "
NameGenerator function f(x)If f'(x)≤0,∀x∈X
Kullback-LeiblerReverse KLPearson x²Squared HellingerJensen-ShannonGANx logxFalseTrueFalseFalseFalseTrue
-logx(x-1)²(√x-1)²-(x+1)log1++xlog x2xlogx- (𝑥+1)log(𝑥+1)
", + "type": "table", + "image_path": "59d020305dc954f7d90ff9b3ad9b6413b635a52d1e47ee126153a040b69b4ab4.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 147, + 394, + 461, + 424.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 147, + 424.3333333333333, + 461, + 454.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 147, + 454.66666666666663, + 461, + 484.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 13.0 + }, + { + "type": "text", + "bbox": [ + 106, + 498, + 446, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 447, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 447, + 511 + ], + "score": 1.0, + "content": "Now, we prove Thm. 4.4. For better readability, we first rewrite Thm. 4.4 as follows:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 508, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 513, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 189, + 532 + ], + "score": 1.0, + "content": "Theorem A.8. Let", + "type": "text" + }, + { + "bbox": [ + 189, + 513, + 208, + 530 + ], + "score": 0.91, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 514, + 426, + 532 + ], + "score": 1.0, + "content": "as the data distribution of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 427, + 516, + 439, + 528 + ], + "score": 0.87, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 514, + 506, + 532 + ], + "score": 1.0, + "content": "in Eq. 4 under", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 528, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 133, + 545 + ], + "score": 1.0, + "content": "model", + "type": "text" + }, + { + "bbox": [ + 133, + 530, + 148, + 542 + ], + "score": 0.65, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 528, + 219, + 545 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 219, + 531, + 225, + 540 + ], + "score": 0.69, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 528, + 341, + 545 + ], + "score": 1.0, + "content": ", then we can find the optimal", + "type": "text" + }, + { + "bbox": [ + 342, + 531, + 352, + 541 + ], + "score": 0.81, + "content": "\\theta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 528, + 360, + 545 + ], + "score": 1.0, + "content": "o", + "type": "text" + }, + { + "bbox": [ + 360, + 529, + 476, + 542 + ], + "score": 0.71, + "content": "\\begin{array} { r } \\mathsf { \\Gamma } _ { \\mathsf { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 528, + 506, + 545 + ], + "score": 1.0, + "content": "(Eq. 4)", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 542, + 507, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 270, + 560 + ], + "score": 1.0, + "content": "via iteratively optimizing the objective", + "type": "text" + }, + { + "bbox": [ + 270, + 542, + 381, + 558 + ], + "score": 0.91, + "content": "\\theta _ { t + 1 } = \\mathrm { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 542, + 414, + 560 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 414, + 542, + 433, + 558 + ], + "score": 0.9, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 542, + 507, + 560 + ], + "score": 1.0, + "content": "is approximated", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 221, + 569 + ], + "score": 1.0, + "content": "via the last-iteration model", + "type": "text" + }, + { + "bbox": [ + 221, + 556, + 238, + 568 + ], + "score": 0.9, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 556, + 506, + 569 + ], + "score": 1.0, + "content": ". Based on Corollary A.7, we have an upper bound objective for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 567, + 322, + 584 + ], + "spans": [ + { + "bbox": [ + 107, + 567, + 181, + 583 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 568, + 322, + 584 + ], + "score": 1.0, + "content": "and derive the following objective", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 588, + 514, + 637 + ], + "lines": [ + { + "bbox": [ + 111, + 588, + 514, + 637 + ], + "spans": [ + { + "bbox": [ + 111, + 588, + 514, + 637 + ], + "score": 0.82, + "content": "\\mathfrak { t } _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\underbrace { \\left( f \\left( \\frac { \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f \\left( \\frac { \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) + H _ { M ^ { * } } ( x , a ) \\right) } _ { W ( x , a , x ^ { \\prime } ) } \\right] .", + "type": "interline_equation", + "image_path": "529b1ff22fa183e3b27afd2d293ac1cfa96e26643dba363fca5f8a97233c05cf.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 111, + 588, + 514, + 604.3333333333334 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 111, + 604.3333333333334, + 514, + 620.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 111, + 620.6666666666667, + 514, + 637.0000000000001 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 642, + 505, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 642, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 133, + 658 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 644, + 281, + 657 + ], + "score": 0.86, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) , \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } [ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 642, + 318, + 658 + ], + "score": 1.0, + "content": "] denotes", + "type": "text" + }, + { + "bbox": [ + 318, + 644, + 368, + 657 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\kappa } }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 642, + 383, + 658 + ], + "score": 1.0, + "content": "[·],", + "type": "text" + }, + { + "bbox": [ + 383, + 644, + 390, + 655 + ], + "score": 0.82, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 642, + 506, + 658 + ], + "score": 1.0, + "content": "is the generator function in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 656, + 504, + 668 + ], + "spans": [ + { + "bbox": [ + 107, + 657, + 114, + 668 + ], + "score": 0.82, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 656, + 226, + 668 + ], + "score": 1.0, + "content": "-divergence which satisfies", + "type": "text" + }, + { + "bbox": [ + 226, + 657, + 308, + 668 + ], + "score": 0.91, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 656, + 331, + 668 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 331, + 657, + 338, + 666 + ], + "score": 0.75, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 656, + 425, + 668 + ], + "score": 1.0, + "content": "is the parameters of", + "type": "text" + }, + { + "bbox": [ + 425, + 657, + 437, + 666 + ], + "score": 0.59, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 656, + 443, + 668 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 443, + 656, + 461, + 668 + ], + "score": 0.86, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 656, + 497, + 668 + ], + "score": 1.0, + "content": "denotes", + "type": "text" + }, + { + "bbox": [ + 497, + 659, + 504, + 666 + ], + "score": 0.26, + "content": "a", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 104, + 667, + 394, + 680 + ], + "score": 1.0, + "content": "probability function with the same parameters as the learned model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 395, + 667, + 419, + 677 + ], + "score": 0.88, + "content": "\\bar { \\theta } = \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 667, + 506, + 680 + ], + "score": 1.0, + "content": ") but the parameter is", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 678, + 240, + 691 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 240, + 691 + ], + "score": 1.0, + "content": "fixed and only used for sampling.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 107, + 704, + 505, + 734 + ], + "lines": [ + { + "bbox": [ + 104, + 704, + 504, + 720 + ], + "spans": [ + { + "bbox": [ + 104, + 704, + 153, + 720 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 153, + 704, + 172, + 720 + ], + "score": 0.92, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 704, + 384, + 720 + ], + "score": 1.0, + "content": "as the data distribution of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 384, + 706, + 396, + 718 + ], + "score": 0.88, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 704, + 488, + 720 + ], + "score": 1.0, + "content": "in Eq. 4 under model", + "type": "text" + }, + { + "bbox": [ + 488, + 707, + 504, + 718 + ], + "score": 0.86, + "content": "M _ { \\theta }", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 180, + 734 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 180, + 721, + 186, + 730 + ], + "score": 0.77, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 719, + 312, + 734 + ], + "score": 1.0, + "content": ", then we can find the optimal", + "type": "text" + }, + { + "bbox": [ + 312, + 721, + 331, + 732 + ], + "score": 0.91, + "content": "\\theta _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 719, + 345, + 734 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 345, + 719, + 456, + 733 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "(Eq. 4) via", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "19", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 82, + 505, + 93 + ], + "lines": [ + { + "bbox": [ + 496, + 84, + 504, + 93 + ], + "spans": [ + { + "bbox": [ + 496, + 84, + 504, + 93 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 343, + 505, + 354 + ], + "lines": [ + { + "bbox": [ + 495, + 344, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 495, + 344, + 505, + 354 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 100, + 506, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 98, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 98, + 221, + 118 + ], + "score": 1.0, + "content": "Corollary A.7. Let g( p(x)q(x) )", + "type": "text" + }, + { + "bbox": [ + 191, + 100, + 279, + 117 + ], + "score": 0.95, + "content": "\\begin{array} { r } { g ( \\frac { p ( x ) } { q ( x ) } ) = - \\log \\frac { p ( x ) } { q ( x ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 101, + 308, + 115 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 308, + 102, + 387, + 115 + ], + "score": 0.91, + "content": "p ( x ) > 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 101, + 407, + 115 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 408, + 102, + 486, + 115 + ], + "score": 0.92, + "content": "q ( x ) > 0 , \\forall x \\in \\mathcal { X } _ { }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 101, + 505, + 115 + ], + "score": 1.0, + "content": ", for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 115, + 192, + 128 + ], + "spans": [ + { + "bbox": [ + 107, + 116, + 132, + 126 + ], + "score": 0.87, + "content": "\\upsilon > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 115, + 192, + 128 + ], + "score": 1.0, + "content": ", the inequality", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 98, + 505, + 128 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 154, + 132, + 454, + 160 + ], + "lines": [ + { + "bbox": [ + 154, + 132, + 454, + 160 + ], + "spans": [ + { + "bbox": [ + 154, + 132, + 454, + 160 + ], + "score": 0.92, + "content": "\\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x ,", + "type": "interline_equation", + "image_path": "be15a1d06758081716993cce8840fa46cc7b03706dd260cab3e53516d187e7a5.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 154, + 132, + 454, + 160 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 165, + 505, + 189 + ], + "lines": [ + { + "bbox": [ + 105, + 164, + 506, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 140, + 180 + ], + "score": 1.0, + "content": "holds if", + "type": "text" + }, + { + "bbox": [ + 141, + 165, + 222, + 178 + ], + "score": 0.9, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 164, + 346, + 180 + ], + "score": 1.0, + "content": ". It is not always satisfied for", + "type": "text" + }, + { + "bbox": [ + 346, + 167, + 353, + 178 + ], + "score": 0.82, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 164, + 406, + 180 + ], + "score": 1.0, + "content": "functions of", + "type": "text" + }, + { + "bbox": [ + 407, + 167, + 413, + 178 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 164, + 506, + 180 + ], + "score": 1.0, + "content": "-divergence. We list a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 176, + 285, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 166, + 189 + ], + "score": 1.0, + "content": "comparison of", + "type": "text" + }, + { + "bbox": [ + 167, + 178, + 173, + 189 + ], + "score": 0.82, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 176, + 285, + 189 + ], + "score": 1.0, + "content": "on that condition in Tab. 2.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 164, + 506, + 189 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 200, + 506, + 259 + ], + "lines": [ + { + "bbox": [ + 104, + 199, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 104, + 199, + 136, + 217 + ], + "score": 1.0, + "content": "Proof.", + "type": "text" + }, + { + "bbox": [ + 136, + 202, + 289, + 216 + ], + "score": 0.92, + "content": "\\begin{array} { r } { g ^ { \\prime } ( x ) = - \\log x = - \\frac { 1 } { x } < 0 , \\forall x \\in \\mathcal { X } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 199, + 331, + 217 + ], + "score": 1.0, + "content": ". Suppose", + "type": "text" + }, + { + "bbox": [ + 331, + 203, + 410, + 215 + ], + "score": 0.91, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 199, + 451, + 217 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 452, + 202, + 505, + 215 + ], + "score": 0.92, + "content": "f \\left( x \\right) \\geq f \\left( y \\right)", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 101, + 209, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 101, + 209, + 163, + 239 + ], + "score": 1.0, + "content": "if and only if", + "type": "text" + }, + { + "bbox": [ + 106, + 231, + 187, + 248 + ], + "score": 0.45, + "content": "g ( \\frac { p ( y ) } { q ( y ) } ) , \\forall x , y \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 217, + 263, + 230 + ], + "score": 0.63, + "content": "g ( x ) \\geq g ( y ) , \\forall x , y \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 221, + 241, + 257 + ], + "score": 1.0, + "content": "holds for all", + "type": "text" + }, + { + "bbox": [ + 241, + 234, + 267, + 244 + ], + "score": 0.89, + "content": "\\upsilon > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 209, + 316, + 239 + ], + "score": 1.0, + "content": "holds. Thus", + "type": "text" + }, + { + "bbox": [ + 268, + 221, + 323, + 257 + ], + "score": 1.0, + "content": ". By defining", + "type": "text" + }, + { + "bbox": [ + 317, + 215, + 404, + 232 + ], + "score": 0.92, + "content": "\\begin{array} { r } { f ( v \\frac { p ( x ) } { q ( x ) } ) \\geq f ( v \\frac { p ( y ) } { q ( y ) } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 232, + 400, + 248 + ], + "score": 0.92, + "content": "\\begin{array} { r } { F ( x ) = f ( v \\frac { p ( x ) } { q ( x ) } ) ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 221, + 420, + 257 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 404, + 209, + 461, + 239 + ], + "score": 1.0, + "content": "if and only if", + "type": "text" + }, + { + "bbox": [ + 420, + 232, + 486, + 248 + ], + "score": 0.93, + "content": "\\begin{array} { r } { G ( x ) = g ( \\frac { p ( x ) } { q ( x ) } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 215, + 505, + 232 + ], + "score": 0.91, + "content": "g \\big ( \\frac { p ( x ) } { q ( x ) } \\big ) \\geq", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 233, + 505, + 245 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 246, + 221, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 221, + 258 + ], + "score": 1.0, + "content": "using Lemma A.6, we have:", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 101, + 199, + 505, + 258 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 264, + 423, + 290 + ], + "lines": [ + { + "bbox": [ + 186, + 264, + 423, + 290 + ], + "spans": [ + { + "bbox": [ + 186, + 264, + 423, + 290 + ], + "score": 0.92, + "content": "\\int _ { \\mathcal { X } } q ( x ) F ( x ) G ( x ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) F ( x ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) G ( x ) \\mathrm { d } x .", + "type": "interline_equation", + "image_path": "c3950b114a613cb86264afb680420911ff7351973e16a1ff704469cd03c1a77b.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 186, + 264, + 423, + 290 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 167, + 306 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 168, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 168, + 307 + ], + "score": 1.0, + "content": "Then we know", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 294, + 168, + 307 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 311, + 454, + 338 + ], + "lines": [ + { + "bbox": [ + 156, + 311, + 454, + 338 + ], + "spans": [ + { + "bbox": [ + 156, + 311, + 454, + 338 + ], + "score": 0.91, + "content": "\\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x", + "type": "interline_equation", + "image_path": "fcb2558754c0452cee69dad339c5d1b39548a5bfe344d27640856e105bdf7c58.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 156, + 311, + 454, + 338 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 343, + 132, + 354 + ], + "lines": [ + { + "bbox": [ + 104, + 342, + 134, + 356 + ], + "spans": [ + { + "bbox": [ + 104, + 342, + 134, + 356 + ], + "score": 1.0, + "content": "holds.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 104, + 342, + 134, + 356 + ] + }, + { + "type": "table", + "bbox": [ + 147, + 394, + 461, + 485 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 183, + 382, + 427, + 394 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 183, + 380, + 428, + 397 + ], + "spans": [ + { + "bbox": [ + 183, + 380, + 273, + 397 + ], + "score": 1.0, + "content": "Table 2: Properties of", + "type": "text" + }, + { + "bbox": [ + 273, + 382, + 350, + 394 + ], + "score": 0.91, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 380, + 366, + 397 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 367, + 383, + 374, + 394 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 380, + 428, + 397 + ], + "score": 1.0, + "content": "-divergences.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "table_body", + "bbox": [ + 147, + 394, + 461, + 485 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 394, + 461, + 485 + ], + "spans": [ + { + "bbox": [ + 147, + 394, + 461, + 485 + ], + "score": 0.982, + "html": "
NameGenerator function f(x)If f'(x)≤0,∀x∈X
Kullback-LeiblerReverse KLPearson x²Squared HellingerJensen-ShannonGANx logxFalseTrueFalseFalseFalseTrue
-logx(x-1)²(√x-1)²-(x+1)log1++xlog x2xlogx- (𝑥+1)log(𝑥+1)
", + "type": "table", + "image_path": "59d020305dc954f7d90ff9b3ad9b6413b635a52d1e47ee126153a040b69b4ab4.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 147, + 394, + 461, + 424.3333333333333 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 147, + 424.3333333333333, + 461, + 454.66666666666663 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 147, + 454.66666666666663, + 461, + 484.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 13.0 + }, + { + "type": "text", + "bbox": [ + 106, + 498, + 446, + 511 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 447, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 447, + 511 + ], + "score": 1.0, + "content": "Now, we prove Thm. 4.4. For better readability, we first rewrite Thm. 4.4 as follows:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 497, + 447, + 511 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 508, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 513, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 189, + 532 + ], + "score": 1.0, + "content": "Theorem A.8. Let", + "type": "text" + }, + { + "bbox": [ + 189, + 513, + 208, + 530 + ], + "score": 0.91, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 514, + 426, + 532 + ], + "score": 1.0, + "content": "as the data distribution of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 427, + 516, + 439, + 528 + ], + "score": 0.87, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 514, + 506, + 532 + ], + "score": 1.0, + "content": "in Eq. 4 under", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 528, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 133, + 545 + ], + "score": 1.0, + "content": "model", + "type": "text" + }, + { + "bbox": [ + 133, + 530, + 148, + 542 + ], + "score": 0.65, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 528, + 219, + 545 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 219, + 531, + 225, + 540 + ], + "score": 0.69, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 528, + 341, + 545 + ], + "score": 1.0, + "content": ", then we can find the optimal", + "type": "text" + }, + { + "bbox": [ + 342, + 531, + 352, + 541 + ], + "score": 0.81, + "content": "\\theta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 528, + 360, + 545 + ], + "score": 1.0, + "content": "o", + "type": "text" + }, + { + "bbox": [ + 360, + 529, + 476, + 542 + ], + "score": 0.71, + "content": "\\begin{array} { r } \\mathsf { \\Gamma } _ { \\mathsf { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 528, + 506, + 545 + ], + "score": 1.0, + "content": "(Eq. 4)", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 542, + 507, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 270, + 560 + ], + "score": 1.0, + "content": "via iteratively optimizing the objective", + "type": "text" + }, + { + "bbox": [ + 270, + 542, + 381, + 558 + ], + "score": 0.91, + "content": "\\theta _ { t + 1 } = \\mathrm { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 542, + 414, + 560 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 414, + 542, + 433, + 558 + ], + "score": 0.9, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 542, + 507, + 560 + ], + "score": 1.0, + "content": "is approximated", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 221, + 569 + ], + "score": 1.0, + "content": "via the last-iteration model", + "type": "text" + }, + { + "bbox": [ + 221, + 556, + 238, + 568 + ], + "score": 0.9, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 556, + 506, + 569 + ], + "score": 1.0, + "content": ". Based on Corollary A.7, we have an upper bound objective for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 567, + 322, + 584 + ], + "spans": [ + { + "bbox": [ + 107, + 567, + 181, + 583 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 568, + 322, + 584 + ], + "score": 1.0, + "content": "and derive the following objective", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 513, + 507, + 584 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 588, + 514, + 637 + ], + "lines": [ + { + "bbox": [ + 111, + 588, + 514, + 637 + ], + "spans": [ + { + "bbox": [ + 111, + 588, + 514, + 637 + ], + "score": 0.82, + "content": "\\mathfrak { t } _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\underbrace { \\left( f \\left( \\frac { \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f \\left( \\frac { \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) + H _ { M ^ { * } } ( x , a ) \\right) } _ { W ( x , a , x ^ { \\prime } ) } \\right] .", + "type": "interline_equation", + "image_path": "529b1ff22fa183e3b27afd2d293ac1cfa96e26643dba363fca5f8a97233c05cf.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 111, + 588, + 514, + 604.3333333333334 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 111, + 604.3333333333334, + 514, + 620.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 111, + 620.6666666666667, + 514, + 637.0000000000001 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 642, + 505, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 642, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 133, + 658 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 644, + 281, + 657 + ], + "score": 0.86, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) , \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } [ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 642, + 318, + 658 + ], + "score": 1.0, + "content": "] denotes", + "type": "text" + }, + { + "bbox": [ + 318, + 644, + 368, + 657 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\kappa } }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 642, + 383, + 658 + ], + "score": 1.0, + "content": "[·],", + "type": "text" + }, + { + "bbox": [ + 383, + 644, + 390, + 655 + ], + "score": 0.82, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 642, + 506, + 658 + ], + "score": 1.0, + "content": "is the generator function in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 656, + 504, + 668 + ], + "spans": [ + { + "bbox": [ + 107, + 657, + 114, + 668 + ], + "score": 0.82, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 656, + 226, + 668 + ], + "score": 1.0, + "content": "-divergence which satisfies", + "type": "text" + }, + { + "bbox": [ + 226, + 657, + 308, + 668 + ], + "score": 0.91, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 656, + 331, + 668 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 331, + 657, + 338, + 666 + ], + "score": 0.75, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 656, + 425, + 668 + ], + "score": 1.0, + "content": "is the parameters of", + "type": "text" + }, + { + "bbox": [ + 425, + 657, + 437, + 666 + ], + "score": 0.59, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 656, + 443, + 668 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 443, + 656, + 461, + 668 + ], + "score": 0.86, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 656, + 497, + 668 + ], + "score": 1.0, + "content": "denotes", + "type": "text" + }, + { + "bbox": [ + 497, + 659, + 504, + 666 + ], + "score": 0.26, + "content": "a", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 104, + 667, + 394, + 680 + ], + "score": 1.0, + "content": "probability function with the same parameters as the learned model (i.e.,", + "type": "text" + }, + { + "bbox": [ + 395, + 667, + 419, + 677 + ], + "score": 0.88, + "content": "\\bar { \\theta } = \\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 667, + 506, + 680 + ], + "score": 1.0, + "content": ") but the parameter is", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 678, + 240, + 691 + ], + "spans": [ + { + "bbox": [ + 104, + 678, + 240, + 691 + ], + "score": 1.0, + "content": "fixed and only used for sampling.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 104, + 642, + 506, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 704, + 505, + 734 + ], + "lines": [ + { + "bbox": [ + 104, + 704, + 504, + 720 + ], + "spans": [ + { + "bbox": [ + 104, + 704, + 153, + 720 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 153, + 704, + 172, + 720 + ], + "score": 0.92, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 704, + 384, + 720 + ], + "score": 1.0, + "content": "as the data distribution of the best-response policy", + "type": "text" + }, + { + "bbox": [ + 384, + 706, + 396, + 718 + ], + "score": 0.88, + "content": "\\bar { \\beta } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 704, + 488, + 720 + ], + "score": 1.0, + "content": "in Eq. 4 under model", + "type": "text" + }, + { + "bbox": [ + 488, + 707, + 504, + 718 + ], + "score": 0.86, + "content": "M _ { \\theta }", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 180, + 734 + ], + "score": 1.0, + "content": "parameterized by", + "type": "text" + }, + { + "bbox": [ + 180, + 721, + 186, + 730 + ], + "score": 0.77, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 719, + 312, + 734 + ], + "score": 1.0, + "content": ", then we can find the optimal", + "type": "text" + }, + { + "bbox": [ + 312, + 721, + 331, + 732 + ], + "score": 0.91, + "content": "\\theta _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 719, + 345, + 734 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 345, + 719, + 456, + 733 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "(Eq. 4) via", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 78, + 507, + 98 + ], + "spans": [ + { + "bbox": [ + 104, + 78, + 250, + 98 + ], + "score": 1.0, + "content": "iteratively optimizing the objective", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 250, + 80, + 356, + 96 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\theta _ { t + 1 } = \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 357, + 78, + 387, + 98 + ], + "score": 1.0, + "content": ", where", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 388, + 80, + 406, + 96 + ], + "score": 0.91, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 407, + 78, + 507, + 98 + ], + "score": 1.0, + "content": "is approximated via the", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 91, + 209, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 187, + 106 + ], + "score": 1.0, + "content": "last-iteration model", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 187, + 94, + 205, + 105 + ], + "score": 0.9, + "content": "\\bar { M } _ { \\theta _ { t } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 205, + 91, + 209, + 106 + ], + "score": 1.0, + "content": ":", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 29.5, + "bbox_fs": [ + 104, + 704, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 104, + 78, + 507, + 98 + ], + "spans": [ + { + "bbox": [ + 104, + 78, + 250, + 98 + ], + "score": 1.0, + "content": "iteratively optimizing the objective", + "type": "text" + }, + { + "bbox": [ + 250, + 80, + 356, + 96 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\theta _ { t + 1 } = \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 78, + 387, + 98 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 388, + 80, + 406, + 96 + ], + "score": 0.91, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 78, + 507, + 98 + ], + "score": 1.0, + "content": "is approximated via the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 91, + 209, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 187, + 106 + ], + "score": 1.0, + "content": "last-iteration model", + "type": "text" + }, + { + "bbox": [ + 187, + 94, + 205, + 105 + ], + "score": 0.9, + "content": "\\bar { M } _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 91, + 209, + 106 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 101, + 114, + 500, + 291 + ], + "lines": [], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 101, + 114, + 500, + 173.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 101, + 173.0, + 500, + 232.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 101, + 232.0, + 500, + 291.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 142, + 527, + 439 + ], + "lines": [ + { + "bbox": [ + 125, + 142, + 527, + 439 + ], + "spans": [ + { + "bbox": [ + 125, + 142, + 527, + 439 + ], + "score": 0.61, + "content": "\\begin{array} { r l } & \\quad \\mathrm { w r e r } \\int _ { - \\infty } ^ { \\infty } \\frac { 1 } { \\mathrm { d } { x } } \\frac { \\mathrm { d } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } } \\left( \\int _ { - \\infty } ^ { \\infty } \\mathrm { d } { x } \\omega \\mathrm { e r } \\omega ^ { 2 } \\mathrm { d } { x } ^ { 2 } \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } } \\right) + \\int _ { - \\infty } ^ { \\infty } \\mathrm { d } { x } ^ { 2 } \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 3 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } } \\\\ & \\qquad - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\left( \\frac { \\sqrt { 3 } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 2 } \\right) } - \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { y } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } } \\\\ & \\qquad - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\left( \\frac { \\sqrt { 3 } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 2 } \\right) } - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\mathrm y \\end{array}", + "type": "interline_equation", + "image_path": "55ba95f92ddc0d86b32bbc4155ee24719fbebd61dd6aca09bbcaba8307bf06aa.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 124, + 142, + 527, + 241.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 124, + 241.0, + 527, + 340.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 124, + 340.0, + 527, + 439.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 457, + 506, + 528 + ], + "lines": [ + { + "bbox": [ + 104, + 457, + 507, + 476 + ], + "spans": [ + { + "bbox": [ + 104, + 457, + 134, + 476 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 460, + 152, + 471 + ], + "score": 0.88, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 457, + 314, + 476 + ], + "score": 1.0, + "content": "is introduced to approximate the term", + "type": "text" + }, + { + "bbox": [ + 314, + 457, + 333, + 472 + ], + "score": 0.92, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 457, + 449, + 476 + ], + "score": 1.0, + "content": "and fixed when optimizing", + "type": "text" + }, + { + "bbox": [ + 450, + 460, + 456, + 470 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 457, + 507, + 476 + ], + "score": 1.0, + "content": ". In Eq. 16,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 471, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 107, + 471, + 158, + 486 + ], + "score": 0.91, + "content": "\\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 472, + 424, + 487 + ], + "score": 1.0, + "content": "for Eq. 9 is eliminated as it does not contribute to the gradient of", + "type": "text" + }, + { + "bbox": [ + 424, + 474, + 430, + 483 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 472, + 470, + 487 + ], + "score": 1.0, + "content": ". Assume", + "type": "text" + }, + { + "bbox": [ + 470, + 473, + 506, + 486 + ], + "score": 0.92, + "content": "f ^ { \\prime } ( x ) \\leq", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 481, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 107, + 491, + 150, + 502 + ], + "score": 0.89, + "content": "0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 481, + 166, + 506 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 167, + 485, + 270, + 506 + ], + "score": 0.89, + "content": "\\begin{array} { r } { v ( x , a ) : = \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } > 0 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 481, + 274, + 506 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 275, + 490, + 377, + 503 + ], + "score": 0.69, + "content": "p ( x ^ { \\prime } | x , a ) = M _ { \\theta } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 481, + 398, + 506 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 399, + 490, + 502, + 503 + ], + "score": 0.92, + "content": "q ( x ^ { \\prime } | x , a ) = M ^ { * } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 481, + 506, + 506 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 504, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 450, + 518 + ], + "score": 1.0, + "content": "the first inequality can be derived by adopting Corollary A.7 and eliminating the first", + "type": "text" + }, + { + "bbox": [ + 450, + 505, + 472, + 516 + ], + "score": 0.91, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 504, + 506, + 518 + ], + "score": 1.0, + "content": "since it", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 515, + 266, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 256, + 528 + ], + "score": 1.0, + "content": "does not contribute to the gradient of", + "type": "text" + }, + { + "bbox": [ + 256, + 516, + 262, + 525 + ], + "score": 0.82, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 515, + 266, + 528 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 107, + 561, + 298, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 299, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 299, + 574 + ], + "score": 1.0, + "content": "A.4 PROOF OF THE TRACTABLE SOLUTION", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 583, + 304, + 596 + ], + "lines": [ + { + "bbox": [ + 106, + 583, + 305, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 305, + 597 + ], + "score": 1.0, + "content": "Now we are ready to prove the tractable solution:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 619, + 506, + 663 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 625, + 277, + 638 + ], + "score": 1.0, + "content": "Proof. The core challenge is that the term", + "type": "text" + }, + { + "bbox": [ + 278, + 621, + 403, + 642 + ], + "score": 0.94, + "content": "\\begin{array} { r } { f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\big ) - f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\big ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 625, + 505, + 638 + ], + "score": 1.0, + "content": "is still intractable. In the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 638, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 506, + 653 + ], + "score": 1.0, + "content": "following, we give a tractable solution to Thm. 4.4. First, we resort to the first-order approximation.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 651, + 299, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 156, + 663 + ], + "score": 1.0, + "content": "Given some", + "type": "text" + }, + { + "bbox": [ + 157, + 651, + 259, + 663 + ], + "score": 0.92, + "content": "u \\bar { \\in } ( 1 - \\xi , 1 + \\xi ) , \\xi > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 651, + 299, + 663 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 245, + 671, + 366, + 686 + ], + "lines": [ + { + "bbox": [ + 245, + 671, + 366, + 686 + ], + "spans": [ + { + "bbox": [ + 245, + 671, + 366, + 686 + ], + "score": 0.9, + "content": "f ( u ) \\approx f ( 1 ) + f ^ { \\prime } ( u ) ( u - 1 ) ,", + "type": "interline_equation", + "image_path": "6242018bc8440a3db08d4d5a227d5ee5d932127bd457b4e7187ada26e7bb0585.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 245, + 671, + 366, + 686 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 694, + 506, + 735 + ], + "lines": [ + { + "bbox": [ + 105, + 694, + 506, + 708 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 132, + 708 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 695, + 142, + 707 + ], + "score": 0.88, + "content": "f ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 694, + 259, + 708 + ], + "score": 1.0, + "content": "is the first-order derivative of", + "type": "text" + }, + { + "bbox": [ + 259, + 696, + 266, + 707 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 694, + 416, + 708 + ], + "score": 1.0, + "content": ". By Taylor’s formula and the fact that", + "type": "text" + }, + { + "bbox": [ + 417, + 695, + 439, + 707 + ], + "score": 0.89, + "content": "f ^ { \\prime } ( u )", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 694, + 506, + 708 + ], + "score": 1.0, + "content": "of the generator", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 703, + 504, + 721 + ], + "spans": [ + { + "bbox": [ + 104, + 703, + 141, + 721 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 141, + 707, + 149, + 718 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 703, + 204, + 721 + ], + "score": 1.0, + "content": "is bounded in", + "type": "text" + }, + { + "bbox": [ + 204, + 707, + 257, + 719 + ], + "score": 0.9, + "content": "( 1 - \\xi , 1 + \\xi )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 703, + 417, + 721 + ], + "score": 1.0, + "content": ", the approximation error is no more than", + "type": "text" + }, + { + "bbox": [ + 417, + 707, + 443, + 719 + ], + "score": 0.9, + "content": "\\mathcal { O } ( \\xi ^ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 703, + 497, + 721 + ], + "score": 1.0, + "content": ". Substituting", + "type": "text" + }, + { + "bbox": [ + 497, + 711, + 504, + 717 + ], + "score": 0.7, + "content": "u", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 714, + 507, + 737 + ], + "spans": [ + { + "bbox": [ + 104, + 714, + 126, + 737 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 718, + 144, + 735 + ], + "score": 0.92, + "content": "\\textstyle { \\frac { p ( x ) } { q ( x ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 714, + 229, + 737 + ], + "score": 1.0, + "content": "in Eq. 18, the pattern", + "type": "text" + }, + { + "bbox": [ + 229, + 719, + 261, + 735 + ], + "score": 0.93, + "content": "f ( { \\frac { p ( x ) } { q ( x ) } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 714, + 378, + 737 + ], + "score": 1.0, + "content": "in Eq. 17 can be converted to", + "type": "text" + }, + { + "bbox": [ + 378, + 718, + 502, + 735 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\frac { p ( x ) } { q ( x ) } f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) - f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) + f ( 1 ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 719, + 507, + 735 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 533, + 505, + 543 + ], + "lines": [ + { + "bbox": [ + 495, + 534, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 495, + 534, + 505, + 544 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 504, + 106 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 104, + 78, + 507, + 106 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 101, + 114, + 500, + 291 + ], + "lines": [], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 101, + 114, + 500, + 173.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 101, + 173.0, + 500, + 232.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 101, + 232.0, + 500, + 291.0 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 142, + 527, + 439 + ], + "lines": [ + { + "bbox": [ + 125, + 142, + 527, + 439 + ], + "spans": [ + { + "bbox": [ + 125, + 142, + 527, + 439 + ], + "score": 0.61, + "content": "\\begin{array} { r l } & \\quad \\mathrm { w r e r } \\int _ { - \\infty } ^ { \\infty } \\frac { 1 } { \\mathrm { d } { x } } \\frac { \\mathrm { d } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } } \\left( \\int _ { - \\infty } ^ { \\infty } \\mathrm { d } { x } \\omega \\mathrm { e r } \\omega ^ { 2 } \\mathrm { d } { x } ^ { 2 } \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } } \\right) + \\int _ { - \\infty } ^ { \\infty } \\mathrm { d } { x } ^ { 2 } \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 3 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } } \\\\ & \\qquad - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\left( \\frac { \\sqrt { 3 } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 2 } \\right) } - \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { y } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } } \\\\ & \\qquad - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\left( \\frac { \\sqrt { 3 } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 2 } \\right) } - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\mathrm y \\end{array}", + "type": "interline_equation", + "image_path": "55ba95f92ddc0d86b32bbc4155ee24719fbebd61dd6aca09bbcaba8307bf06aa.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 124, + 142, + 527, + 241.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 124, + 241.0, + 527, + 340.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 124, + 340.0, + 527, + 439.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 457, + 506, + 528 + ], + "lines": [ + { + "bbox": [ + 104, + 457, + 507, + 476 + ], + "spans": [ + { + "bbox": [ + 104, + 457, + 134, + 476 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 460, + 152, + 471 + ], + "score": 0.88, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 457, + 314, + 476 + ], + "score": 1.0, + "content": "is introduced to approximate the term", + "type": "text" + }, + { + "bbox": [ + 314, + 457, + 333, + 472 + ], + "score": 0.92, + "content": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 457, + 449, + 476 + ], + "score": 1.0, + "content": "and fixed when optimizing", + "type": "text" + }, + { + "bbox": [ + 450, + 460, + 456, + 470 + ], + "score": 0.76, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 457, + 507, + 476 + ], + "score": 1.0, + "content": ". In Eq. 16,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 471, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 107, + 471, + 158, + 486 + ], + "score": 0.91, + "content": "\\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 472, + 424, + 487 + ], + "score": 1.0, + "content": "for Eq. 9 is eliminated as it does not contribute to the gradient of", + "type": "text" + }, + { + "bbox": [ + 424, + 474, + 430, + 483 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 472, + 470, + 487 + ], + "score": 1.0, + "content": ". Assume", + "type": "text" + }, + { + "bbox": [ + 470, + 473, + 506, + 486 + ], + "score": 0.92, + "content": "f ^ { \\prime } ( x ) \\leq", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 481, + 506, + 506 + ], + "spans": [ + { + "bbox": [ + 107, + 491, + 150, + 502 + ], + "score": 0.89, + "content": "0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 481, + 166, + 506 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 167, + 485, + 270, + 506 + ], + "score": 0.89, + "content": "\\begin{array} { r } { v ( x , a ) : = \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } > 0 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 481, + 274, + 506 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 275, + 490, + 377, + 503 + ], + "score": 0.69, + "content": "p ( x ^ { \\prime } | x , a ) = M _ { \\theta } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 481, + 398, + 506 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 399, + 490, + 502, + 503 + ], + "score": 0.92, + "content": "q ( x ^ { \\prime } | x , a ) = M ^ { * } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 481, + 506, + 506 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 504, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 450, + 518 + ], + "score": 1.0, + "content": "the first inequality can be derived by adopting Corollary A.7 and eliminating the first", + "type": "text" + }, + { + "bbox": [ + 450, + 505, + 472, + 516 + ], + "score": 0.91, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 473, + 504, + 506, + 518 + ], + "score": 1.0, + "content": "since it", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 515, + 266, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 515, + 256, + 528 + ], + "score": 1.0, + "content": "does not contribute to the gradient of", + "type": "text" + }, + { + "bbox": [ + 256, + 516, + 262, + 525 + ], + "score": 0.82, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 515, + 266, + 528 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10, + "bbox_fs": [ + 104, + 457, + 507, + 528 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 561, + 298, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 299, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 299, + 574 + ], + "score": 1.0, + "content": "A.4 PROOF OF THE TRACTABLE SOLUTION", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 583, + 304, + 596 + ], + "lines": [ + { + "bbox": [ + 106, + 583, + 305, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 305, + 597 + ], + "score": 1.0, + "content": "Now we are ready to prove the tractable solution:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 106, + 583, + 305, + 597 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 619, + 506, + 663 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 625, + 277, + 638 + ], + "score": 1.0, + "content": "Proof. The core challenge is that the term", + "type": "text" + }, + { + "bbox": [ + 278, + 621, + 403, + 642 + ], + "score": 0.94, + "content": "\\begin{array} { r } { f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\big ) - f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\big ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 625, + 505, + 638 + ], + "score": 1.0, + "content": "is still intractable. In the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 638, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 506, + 653 + ], + "score": 1.0, + "content": "following, we give a tractable solution to Thm. 4.4. First, we resort to the first-order approximation.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 651, + 299, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 156, + 663 + ], + "score": 1.0, + "content": "Given some", + "type": "text" + }, + { + "bbox": [ + 157, + 651, + 259, + 663 + ], + "score": 0.92, + "content": "u \\bar { \\in } ( 1 - \\xi , 1 + \\xi ) , \\xi > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 651, + 299, + 663 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 621, + 506, + 663 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 245, + 671, + 366, + 686 + ], + "lines": [ + { + "bbox": [ + 245, + 671, + 366, + 686 + ], + "spans": [ + { + "bbox": [ + 245, + 671, + 366, + 686 + ], + "score": 0.9, + "content": "f ( u ) \\approx f ( 1 ) + f ^ { \\prime } ( u ) ( u - 1 ) ,", + "type": "interline_equation", + "image_path": "6242018bc8440a3db08d4d5a227d5ee5d932127bd457b4e7187ada26e7bb0585.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 245, + 671, + 366, + 686 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 694, + 506, + 735 + ], + "lines": [ + { + "bbox": [ + 105, + 694, + 506, + 708 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 132, + 708 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 695, + 142, + 707 + ], + "score": 0.88, + "content": "f ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 694, + 259, + 708 + ], + "score": 1.0, + "content": "is the first-order derivative of", + "type": "text" + }, + { + "bbox": [ + 259, + 696, + 266, + 707 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 694, + 416, + 708 + ], + "score": 1.0, + "content": ". By Taylor’s formula and the fact that", + "type": "text" + }, + { + "bbox": [ + 417, + 695, + 439, + 707 + ], + "score": 0.89, + "content": "f ^ { \\prime } ( u )", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 694, + 506, + 708 + ], + "score": 1.0, + "content": "of the generator", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 703, + 504, + 721 + ], + "spans": [ + { + "bbox": [ + 104, + 703, + 141, + 721 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 141, + 707, + 149, + 718 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 703, + 204, + 721 + ], + "score": 1.0, + "content": "is bounded in", + "type": "text" + }, + { + "bbox": [ + 204, + 707, + 257, + 719 + ], + "score": 0.9, + "content": "( 1 - \\xi , 1 + \\xi )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 703, + 417, + 721 + ], + "score": 1.0, + "content": ", the approximation error is no more than", + "type": "text" + }, + { + "bbox": [ + 417, + 707, + 443, + 719 + ], + "score": 0.9, + "content": "\\mathcal { O } ( \\xi ^ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 703, + 497, + 721 + ], + "score": 1.0, + "content": ". Substituting", + "type": "text" + }, + { + "bbox": [ + 497, + 711, + 504, + 717 + ], + "score": 0.7, + "content": "u", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 104, + 714, + 507, + 737 + ], + "spans": [ + { + "bbox": [ + 104, + 714, + 126, + 737 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 127, + 718, + 144, + 735 + ], + "score": 0.92, + "content": "\\textstyle { \\frac { p ( x ) } { q ( x ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 714, + 229, + 737 + ], + "score": 1.0, + "content": "in Eq. 18, the pattern", + "type": "text" + }, + { + "bbox": [ + 229, + 719, + 261, + 735 + ], + "score": 0.93, + "content": "f ( { \\frac { p ( x ) } { q ( x ) } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 714, + 378, + 737 + ], + "score": 1.0, + "content": "in Eq. 17 can be converted to", + "type": "text" + }, + { + "bbox": [ + 378, + 718, + 502, + 735 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\frac { p ( x ) } { q ( x ) } f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) - f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) + f ( 1 ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 719, + 507, + 735 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 104, + 694, + 507, + 737 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 163, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 164, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 164, + 95 + ], + "score": 1.0, + "content": "then we have:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 131, + 112, + 524, + 425 + ], + "lines": [ + { + "bbox": [ + 131, + 112, + 524, + 425 + ], + "spans": [ + { + "bbox": [ + 131, + 112, + 524, + 425 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\gamma _ { 1 } = \\gamma _ { 2 } \\gamma _ { 2 } \\gamma _ { 3 } = \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } } & { \\gamma _ { 2 } \\gamma _ { 3 } = \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } , } \\\\ & { \\qquad \\quad + \\frac { \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 2 } } { \\mu _ { 1 } \\mu _ { 2 } \\mu _ { 3 } } \\gamma _ { 1 } ( \\frac { \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } } { \\mu _ { 2 } } + \\frac { \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } } { \\mu _ { 3 } } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad + \\frac { \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 3 } } { \\mu _ { 2 } \\mu _ { 3 } } \\gamma _ { 1 } ( \\frac { \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } } { \\mu _ { 2 } } + \\frac { \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } } { \\mu _ { 3 } } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad \\times ( \\gamma _ { 2 } + \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 3 } ) \\gamma _ { 4 } \\exp ( - \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad \\times \\exp ( - \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 4 } ) \\gamma _ { 4 } \\exp ( - \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 2 } + \\gamma _ { 3 } \\gamma _ { 4 } ) } \\\\ & \\qquad \\quad \\times \\exp ( - \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } ) \\gamma _ { 5 } ( \\gamma _ 3 \\end{array}", + "type": "interline_equation", + "image_path": "a1bccf439183ce51f0a5ff2f3520c4d9de04322e74aaf72ed2fe05144d6d5327.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 131, + 112, + 524, + 216.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 131, + 216.33333333333331, + 524, + 320.66666666666663 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 131, + 320.66666666666663, + 524, + 424.99999999999994 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 403, + 507, + 468 + ], + "lines": [ + { + "bbox": [ + 107, + 402, + 131, + 417 + ], + "spans": [ + { + "bbox": [ + 107, + 402, + 131, + 417 + ], + "score": 1.0, + "content": "Note", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 415, + 249, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 249, + 430 + ], + "score": 1.0, + "content": "but we choose to keep it and ignore", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 426, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 327, + 438 + ], + "score": 1.0, + "content": "an empirical data distribution through data collected by", + "type": "text" + }, + { + "bbox": [ + 327, + 429, + 334, + 436 + ], + "score": 0.72, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 426, + 345, + 438 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 345, + 428, + 361, + 436 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 426, + 506, + 438 + ], + "score": 1.0, + "content": "directly, rather than from a uniform", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 435, + 507, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 356, + 450 + ], + "score": 1.0, + "content": "distribution which is harder to be generated. Although keeping", + "type": "text" + }, + { + "bbox": [ + 357, + 438, + 398, + 448 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 435, + 507, + 450 + ], + "score": 1.0, + "content": "incurs extra bias in theory,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 446, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 506, + 458 + ], + "score": 1.0, + "content": "the results in our experiments show that it has not made significant negative effects in practice. We", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 456, + 457, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 380, + 469 + ], + "score": 1.0, + "content": "leave this part of modeling in future work. In particular, by ignoring", + "type": "text" + }, + { + "bbox": [ + 381, + 457, + 415, + 468 + ], + "score": 0.91, + "content": "\\bar { \\alpha _ { 0 } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 456, + 457, + 469 + ], + "score": 1.0, + "content": ", we have:", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6.5 + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 470, + 524, + 500 + ], + "lines": [ + { + "bbox": [ + 124, + 470, + 524, + 500 + ], + "spans": [ + { + "bbox": [ + 124, + 470, + 524, + 500 + ], + "score": 0.83, + "content": "= \\arg \\operatorname* { m a x } _ { \\theta } \\int _ { \\chi , A , \\chi } \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ) \\left( f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { \\mathbb { M } ^ { + } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { \\mathbb { M } ^ { + } } ^ { \\kappa } ( x , a ) } \\right) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + ", + "type": "interline_equation", + "image_path": "831acb50e54dfa42d165c4f521955a089e6c78344de0fd7cb2167e4d717c64d8.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 124, + 470, + 524, + 500 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 166, + 511, + 567, + 542 + ], + "lines": [ + { + "bbox": [ + 166, + 511, + 567, + 542 + ], + "spans": [ + { + "bbox": [ + 166, + 511, + 567, + 542 + ], + "score": 0.81, + "content": "\\int _ { \\boldsymbol { x } , A , \\boldsymbol { \\tilde { x } } } \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) \\left( f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathrm { M } \\theta _ { t } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } ) } { \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } ) } \\right) - f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathrm { M } \\theta _ { t } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) } { \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) } \\right) + H _ { \\mathsf { M } ^ { * } } ( \\boldsymbol { x } , \\boldsymbol { a } ) \\right) \\log M _ { \\theta } ( \\boldsymbol { x } ^ { \\prime } | \\boldsymbol { x } , \\boldsymbol { a } ) \\mathrm { d } \\boldsymbol { x } ^ { \\prime } \\mathrm { d } \\boldsymbol { a } \\mathrm { d } \\boldsymbol { x }", + "type": "interline_equation", + "image_path": "f3421198220e4fe5c5c7ce878300a2090d39b0826d8aac72a940954a9e24c7da.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 166, + 511, + 567, + 521.3333333333334 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 166, + 521.3333333333334, + 567, + 531.6666666666667 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 166, + 531.6666666666667, + 567, + 542.0000000000001 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 560, + 417, + 587 + ], + "lines": [ + { + "bbox": [ + 104, + 555, + 419, + 587 + ], + "spans": [ + { + "bbox": [ + 104, + 555, + 173, + 587 + ], + "score": 1.0, + "content": "We can estimate", + "type": "text" + }, + { + "bbox": [ + 174, + 561, + 237, + 587 + ], + "score": 0.95, + "content": "f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 555, + 256, + 587 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 256, + 561, + 330, + 587 + ], + "score": 0.94, + "content": "f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 563, + 419, + 583 + ], + "score": 1.0, + "content": "through Lemma A.9.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 507, + 627 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 160, + 606 + ], + "score": 1.0, + "content": "Lemma A.9", + "type": "text" + }, + { + "bbox": [ + 160, + 591, + 185, + 605 + ], + "score": 0.87, + "content": "( f ^ { \\prime } ( { \\frac { p } { q } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 590, + 392, + 606 + ], + "score": 1.0, + "content": "estimation (Nguyen et al., 2010)). Given a function", + "type": "text" + }, + { + "bbox": [ + 393, + 592, + 444, + 604 + ], + "score": 0.91, + "content": "T _ { \\varphi } : \\mathcal { X } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 590, + 506, + 606 + ], + "score": 1.0, + "content": "parameterized", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 119, + 617 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 119, + 605, + 149, + 616 + ], + "score": 0.89, + "content": "\\varphi \\in \\Phi", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 604, + 163, + 617 + ], + "score": 1.0, + "content": ", if", + "type": "text" + }, + { + "bbox": [ + 163, + 605, + 171, + 616 + ], + "score": 0.8, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 604, + 469, + 617 + ], + "score": 1.0, + "content": "is convex and lower semi-continuous, by finding the maximum point of", + "type": "text" + }, + { + "bbox": [ + 469, + 606, + 477, + 616 + ], + "score": 0.78, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "in the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 616, + 188, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 188, + 628 + ], + "score": 1.0, + "content": "following objective:", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 191, + 629, + 417, + 647 + ], + "lines": [ + { + "bbox": [ + 191, + 629, + 417, + 647 + ], + "spans": [ + { + "bbox": [ + 191, + 629, + 417, + 647 + ], + "score": 0.9, + "content": "\\varphi ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi } \\mathbb { E } _ { x \\sim p ( x ) } \\left[ T _ { \\varphi } ( x ) \\right] - \\mathbb { E } _ { x \\sim q ( x ) } \\left[ f ^ { * } ( T _ { \\varphi } ( x ) ) \\right] ,", + "type": "interline_equation", + "image_path": "b48098841ee718edf4032c358c28ad998611761eb4b579b8043f492fb1df5bfe.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 191, + 629, + 417, + 647 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 651, + 493, + 667 + ], + "lines": [ + { + "bbox": [ + 102, + 647, + 498, + 672 + ], + "spans": [ + { + "bbox": [ + 102, + 647, + 141, + 672 + ], + "score": 1.0, + "content": "we have", + "type": "text" + }, + { + "bbox": [ + 142, + 650, + 218, + 668 + ], + "score": 0.92, + "content": "f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) = T _ { \\varphi ^ { * } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 647, + 222, + 672 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 223, + 653, + 235, + 664 + ], + "score": 0.78, + "content": "f ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 647, + 331, + 672 + ], + "score": 1.0, + "content": "is Fenchel conjugate of", + "type": "text" + }, + { + "bbox": [ + 331, + 653, + 338, + 664 + ], + "score": 0.81, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 647, + 498, + 672 + ], + "score": 1.0, + "content": "(Hiriart-Urruty & Lemaréchal, 2001).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 678, + 159, + 689 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 161, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 161, + 691 + ], + "score": 1.0, + "content": "In particular,", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 690, + 463, + 731 + ], + "lines": [ + { + "bbox": [ + 147, + 690, + 463, + 731 + ], + "spans": [ + { + "bbox": [ + 147, + 690, + 463, + 731 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\varphi _ { 0 } ^ { * } = \\underset { \\varphi _ { 0 } } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { * } } \\left[ T _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] - \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { * } } \\left[ f ^ { * } ( T _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } \\\\ & { \\varphi _ { 1 } ^ { * } = \\underset { \\varphi _ { 1 } } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { * } } ^ { * } } \\left[ T _ { \\varphi _ { 1 } } ( x , a ) \\right] - \\mathbb { E } _ { x , a \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ f ^ { * } ( T _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "281b23db492dda2f3a6b7786173739c9c5791e5c7e320a031be4d318ee9706eb.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 147, + 690, + 463, + 703.6666666666666 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 147, + 703.6666666666666, + 463, + 717.3333333333333 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 147, + 717.3333333333333, + 463, + 730.9999999999999 + ], + "spans": [], + "index": 23 + } + ] + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 761 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "21", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 163, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 164, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 164, + 95 + ], + "score": 1.0, + "content": "then we have:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 106, + 82, + 164, + 95 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 131, + 112, + 524, + 425 + ], + "lines": [ + { + "bbox": [ + 131, + 112, + 524, + 425 + ], + "spans": [ + { + "bbox": [ + 131, + 112, + 524, + 425 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\gamma _ { 1 } = \\gamma _ { 2 } \\gamma _ { 2 } \\gamma _ { 3 } = \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } } & { \\gamma _ { 2 } \\gamma _ { 3 } = \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } , } \\\\ & { \\qquad \\quad + \\frac { \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 2 } } { \\mu _ { 1 } \\mu _ { 2 } \\mu _ { 3 } } \\gamma _ { 1 } ( \\frac { \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } } { \\mu _ { 2 } } + \\frac { \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } } { \\mu _ { 3 } } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad + \\frac { \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 3 } } { \\mu _ { 2 } \\mu _ { 3 } } \\gamma _ { 1 } ( \\frac { \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } } { \\mu _ { 2 } } + \\frac { \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } } { \\mu _ { 3 } } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad \\times ( \\gamma _ { 2 } + \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 3 } ) \\gamma _ { 4 } \\exp ( - \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad \\times \\exp ( - \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 4 } ) \\gamma _ { 4 } \\exp ( - \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 2 } + \\gamma _ { 3 } \\gamma _ { 4 } ) } \\\\ & \\qquad \\quad \\times \\exp ( - \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } ) \\gamma _ { 5 } ( \\gamma _ 3 \\end{array}", + "type": "interline_equation", + "image_path": "a1bccf439183ce51f0a5ff2f3520c4d9de04322e74aaf72ed2fe05144d6d5327.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 131, + 112, + 524, + 216.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 131, + 216.33333333333331, + 524, + 320.66666666666663 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 131, + 320.66666666666663, + 524, + 424.99999999999994 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "list", + "bbox": [ + 106, + 403, + 507, + 468 + ], + "lines": [ + { + "bbox": [ + 107, + 402, + 131, + 417 + ], + "spans": [ + { + "bbox": [ + 107, + 402, + 131, + 417 + ], + "score": 1.0, + "content": "Note", + "type": "text" + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 415, + 249, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 249, + 430 + ], + "score": 1.0, + "content": "but we choose to keep it and ignore", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 426, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 327, + 438 + ], + "score": 1.0, + "content": "an empirical data distribution through data collected by", + "type": "text" + }, + { + "bbox": [ + 327, + 429, + 334, + 436 + ], + "score": 0.72, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 426, + 345, + 438 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 345, + 428, + 361, + 436 + ], + "score": 0.86, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 426, + 506, + 438 + ], + "score": 1.0, + "content": "directly, rather than from a uniform", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 435, + 507, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 356, + 450 + ], + "score": 1.0, + "content": "distribution which is harder to be generated. Although keeping", + "type": "text" + }, + { + "bbox": [ + 357, + 438, + 398, + 448 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 435, + 507, + 450 + ], + "score": 1.0, + "content": "incurs extra bias in theory,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 446, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 506, + 458 + ], + "score": 1.0, + "content": "the results in our experiments show that it has not made significant negative effects in practice. We", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 456, + 457, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 380, + 469 + ], + "score": 1.0, + "content": "leave this part of modeling in future work. In particular, by ignoring", + "type": "text" + }, + { + "bbox": [ + 381, + 457, + 415, + 468 + ], + "score": 0.91, + "content": "\\bar { \\alpha _ { 0 } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 456, + 457, + 469 + ], + "score": 1.0, + "content": ", we have:", + "type": "text" + } + ], + "index": 9, + "is_list_end_line": true + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 402, + 507, + 469 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 470, + 524, + 500 + ], + "lines": [ + { + "bbox": [ + 124, + 470, + 524, + 500 + ], + "spans": [ + { + "bbox": [ + 124, + 470, + 524, + 500 + ], + "score": 0.83, + "content": "= \\arg \\operatorname* { m a x } _ { \\theta } \\int _ { \\chi , A , \\chi } \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ) \\left( f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { \\mathbb { M } ^ { + } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { \\mathbb { M } ^ { + } } ^ { \\kappa } ( x , a ) } \\right) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + ", + "type": "interline_equation", + "image_path": "831acb50e54dfa42d165c4f521955a089e6c78344de0fd7cb2167e4d717c64d8.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 124, + 470, + 524, + 500 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 166, + 511, + 567, + 542 + ], + "lines": [ + { + "bbox": [ + 166, + 511, + 567, + 542 + ], + "spans": [ + { + "bbox": [ + 166, + 511, + 567, + 542 + ], + "score": 0.81, + "content": "\\int _ { \\boldsymbol { x } , A , \\boldsymbol { \\tilde { x } } } \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) \\left( f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathrm { M } \\theta _ { t } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } ) } { \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } ) } \\right) - f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathrm { M } \\theta _ { t } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) } { \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) } \\right) + H _ { \\mathsf { M } ^ { * } } ( \\boldsymbol { x } , \\boldsymbol { a } ) \\right) \\log M _ { \\theta } ( \\boldsymbol { x } ^ { \\prime } | \\boldsymbol { x } , \\boldsymbol { a } ) \\mathrm { d } \\boldsymbol { x } ^ { \\prime } \\mathrm { d } \\boldsymbol { a } \\mathrm { d } \\boldsymbol { x }", + "type": "interline_equation", + "image_path": "f3421198220e4fe5c5c7ce878300a2090d39b0826d8aac72a940954a9e24c7da.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 166, + 511, + 567, + 521.3333333333334 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 166, + 521.3333333333334, + 567, + 531.6666666666667 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 166, + 531.6666666666667, + 567, + 542.0000000000001 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 560, + 417, + 587 + ], + "lines": [ + { + "bbox": [ + 104, + 555, + 419, + 587 + ], + "spans": [ + { + "bbox": [ + 104, + 555, + 173, + 587 + ], + "score": 1.0, + "content": "We can estimate", + "type": "text" + }, + { + "bbox": [ + 174, + 561, + 237, + 587 + ], + "score": 0.95, + "content": "f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 555, + 256, + 587 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 256, + 561, + 330, + 587 + ], + "score": 0.94, + "content": "f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 563, + 419, + 583 + ], + "score": 1.0, + "content": "through Lemma A.9.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 555, + 419, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 507, + 627 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 160, + 606 + ], + "score": 1.0, + "content": "Lemma A.9", + "type": "text" + }, + { + "bbox": [ + 160, + 591, + 185, + 605 + ], + "score": 0.87, + "content": "( f ^ { \\prime } ( { \\frac { p } { q } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 590, + 392, + 606 + ], + "score": 1.0, + "content": "estimation (Nguyen et al., 2010)). Given a function", + "type": "text" + }, + { + "bbox": [ + 393, + 592, + 444, + 604 + ], + "score": 0.91, + "content": "T _ { \\varphi } : \\mathcal { X } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 590, + 506, + 606 + ], + "score": 1.0, + "content": "parameterized", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 119, + 617 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 119, + 605, + 149, + 616 + ], + "score": 0.89, + "content": "\\varphi \\in \\Phi", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 604, + 163, + 617 + ], + "score": 1.0, + "content": ", if", + "type": "text" + }, + { + "bbox": [ + 163, + 605, + 171, + 616 + ], + "score": 0.8, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 604, + 469, + 617 + ], + "score": 1.0, + "content": "is convex and lower semi-continuous, by finding the maximum point of", + "type": "text" + }, + { + "bbox": [ + 469, + 606, + 477, + 616 + ], + "score": 0.78, + "content": "\\varphi", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "in the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 616, + 188, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 188, + 628 + ], + "score": 1.0, + "content": "following objective:", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 590, + 506, + 628 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 191, + 629, + 417, + 647 + ], + "lines": [ + { + "bbox": [ + 191, + 629, + 417, + 647 + ], + "spans": [ + { + "bbox": [ + 191, + 629, + 417, + 647 + ], + "score": 0.9, + "content": "\\varphi ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi } \\mathbb { E } _ { x \\sim p ( x ) } \\left[ T _ { \\varphi } ( x ) \\right] - \\mathbb { E } _ { x \\sim q ( x ) } \\left[ f ^ { * } ( T _ { \\varphi } ( x ) ) \\right] ,", + "type": "interline_equation", + "image_path": "b48098841ee718edf4032c358c28ad998611761eb4b579b8043f492fb1df5bfe.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 191, + 629, + 417, + 647 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 651, + 493, + 667 + ], + "lines": [ + { + "bbox": [ + 102, + 647, + 498, + 672 + ], + "spans": [ + { + "bbox": [ + 102, + 647, + 141, + 672 + ], + "score": 1.0, + "content": "we have", + "type": "text" + }, + { + "bbox": [ + 142, + 650, + 218, + 668 + ], + "score": 0.92, + "content": "f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) = T _ { \\varphi ^ { * } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 647, + 222, + 672 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 223, + 653, + 235, + 664 + ], + "score": 0.78, + "content": "f ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 647, + 331, + 672 + ], + "score": 1.0, + "content": "is Fenchel conjugate of", + "type": "text" + }, + { + "bbox": [ + 331, + 653, + 338, + 664 + ], + "score": 0.81, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 647, + 498, + 672 + ], + "score": 1.0, + "content": "(Hiriart-Urruty & Lemaréchal, 2001).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 102, + 647, + 498, + 672 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 678, + 159, + 689 + ], + "lines": [ + { + "bbox": [ + 105, + 676, + 161, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 161, + 691 + ], + "score": 1.0, + "content": "In particular,", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 676, + 161, + 691 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 690, + 463, + 731 + ], + "lines": [ + { + "bbox": [ + 147, + 690, + 463, + 731 + ], + "spans": [ + { + "bbox": [ + 147, + 690, + 463, + 731 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\varphi _ { 0 } ^ { * } = \\underset { \\varphi _ { 0 } } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { * } } \\left[ T _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] - \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { * } } \\left[ f ^ { * } ( T _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } \\\\ & { \\varphi _ { 1 } ^ { * } = \\underset { \\varphi _ { 1 } } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { * } } ^ { * } } \\left[ T _ { \\varphi _ { 1 } } ( x , a ) \\right] - \\mathbb { E } _ { x , a \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ f ^ { * } ( T _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "281b23db492dda2f3a6b7786173739c9c5791e5c7e320a031be4d318ee9706eb.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 147, + 690, + 463, + 703.6666666666666 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 147, + 703.6666666666666, + 463, + 717.3333333333333 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 147, + 717.3333333333333, + 463, + 730.9999999999999 + ], + "spans": [], + "index": 23 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 79, + 507, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 85, + 163, + 100 + ], + "score": 1.0, + "content": "then we have", + "type": "text" + }, + { + "bbox": [ + 164, + 80, + 303, + 105 + ], + "score": 0.93, + "content": "\\begin{array} { r } { f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\approx T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 83, + 323, + 102 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 324, + 80, + 441, + 106 + ], + "score": 0.94, + "content": "\\begin{array} { r } { f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) \\approx T _ { \\varphi _ { 1 } ^ { * } } ( x , a ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 84, + 474, + 102 + ], + "score": 1.0, + "content": ". Given", + "type": "text" + }, + { + "bbox": [ + 474, + 88, + 486, + 99 + ], + "score": 0.88, + "content": "\\varphi _ { 0 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 84, + 506, + 102 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 103, + 434, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 119, + 117 + ], + "score": 0.88, + "content": "\\varphi _ { 1 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 103, + 134, + 119 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 135, + 105, + 318, + 119 + ], + "score": 0.91, + "content": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , \\dot { a } , x ^ { \\prime } ) = T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) - T _ { \\varphi _ { 1 } ^ { * } } ( x , a ) ,", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 103, + 408, + 119 + ], + "score": 1.0, + "content": ", then we can optimize", + "type": "text" + }, + { + "bbox": [ + 409, + 106, + 415, + 115 + ], + "score": 0.77, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 103, + 434, + 119 + ], + "score": 1.0, + "content": "via:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 124, + 514, + 236 + ], + "lines": [ + { + "bbox": [ + 111, + 124, + 514, + 236 + ], + "spans": [ + { + "bbox": [ + 111, + 124, + 514, + 236 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\mathsf { \\Gamma } _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ) \\left( T _ { \\varphi _ { 0 } ^ { \\star } } ( x , a , x ^ { \\prime } ) - T _ { \\varphi _ { 1 } ^ { \\star } } ( x , a ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + } \\\\ & { \\qquad \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M ^ { \\star } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\left( T _ { \\varphi _ { 1 } ^ { \\star } } ( x , a ) - T _ { \\varphi _ { 0 } ^ { \\star } } ( x , a , x ^ { \\prime } ) + H _ { M ^ { \\star } } ( x , a ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\arg \\operatorname* { m a x } _ { \\theta } \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ) A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + } \\\\ & { \\qquad \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M ^ { \\star } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) ( - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) + H _ { M ^ { \\star } } ( x , a ) ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x . } \\end{array}", + "type": "interline_equation", + "image_path": "316a26a7d4a9506aaa976710d2906b3c44f9f15b0634f1cf4aa65219802c1038.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 111, + 124, + 514, + 161.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 161.33333333333334, + 514, + 198.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 111, + 198.66666666666669, + 514, + 236.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 506, + 290 + ], + "lines": [ + { + "bbox": [ + 106, + 245, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 198, + 258 + ], + "score": 1.0, + "content": "Based on the specific", + "type": "text" + }, + { + "bbox": [ + 198, + 246, + 205, + 257 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 245, + 330, + 258 + ], + "score": 1.0, + "content": "-divergence, we can represent", + "type": "text" + }, + { + "bbox": [ + 331, + 246, + 339, + 255 + ], + "score": 0.84, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 245, + 359, + 258 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 359, + 246, + 385, + 257 + ], + "score": 0.91, + "content": "f ^ { * } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 245, + 474, + 258 + ], + "score": 1.0, + "content": "with a discriminator", + "type": "text" + }, + { + "bbox": [ + 475, + 246, + 489, + 258 + ], + "score": 0.88, + "content": "D _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 245, + 506, + 258 + ], + "score": 1.0, + "content": ". It", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 257, + 506, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 187, + 270 + ], + "score": 1.0, + "content": "can be verified that", + "type": "text" + }, + { + "bbox": [ + 188, + 257, + 339, + 269 + ], + "score": 0.78, + "content": "f ( u ) = u \\log u - ( u + 1 ) \\log ( u + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 257, + 344, + 270 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 344, + 257, + 426, + 269 + ], + "score": 0.88, + "content": "T _ { \\varphi } ( u ) = \\log D _ { \\varphi } ( u )", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 257, + 448, + 270 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 449, + 257, + 506, + 269 + ], + "score": 0.88, + "content": "f ^ { * } ( T _ { \\varphi } ( u ) ) =", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 267, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 182, + 281 + ], + "score": 0.89, + "content": "- \\log ( 1 - D _ { \\varphi } ( u ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 267, + 408, + 281 + ], + "score": 1.0, + "content": "proposed in Nowozin et al. (2016) satisfies the condition", + "type": "text" + }, + { + "bbox": [ + 408, + 268, + 485, + 280 + ], + "score": 0.89, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 267, + 506, + 281 + ], + "score": 1.0, + "content": "(see", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 278, + 462, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 462, + 290 + ], + "score": 1.0, + "content": "Tab. 2). We select the former in the implementation and convert the tractable solution to:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "interline_equation", + "bbox": [ + 102, + 293, + 549, + 351 + ], + "lines": [ + { + "bbox": [ + 111, + 293, + 514, + 351 + ], + "spans": [ + { + "bbox": [ + 111, + 293, + 514, + 351 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\left( H _ { M ^ { \\star } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } \\\\ { \\therefore } & { _ { \\varphi _ { 0 } ^ { \\star } } = \\arg \\operatorname* { m a x } _ { \\theta _ { 0 } } \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } \\\\ & { _ { \\varphi _ { 1 } ^ { \\star } } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 1 } } \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "913cd2a525614fb7f942e419a24666f6fec2c71503368bd7a24a6117750e8fd9.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 102, + 293, + 549, + 312.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 102, + 312.3333333333333, + 549, + 331.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 102, + 331.66666666666663, + 549, + 350.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 366, + 506, + 398 + ], + "lines": [ + { + "bbox": [ + 106, + 358, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 407, + 387 + ], + "score": 1.0, + "content": "where Aφ∗0 ,φ∗1 (x, a, x′) = log Dφ∗0 (x, a, x′) − log Dφ∗1 (x, a), EρκMθt [·", + "type": "text" + }, + { + "bbox": [ + 411, + 365, + 506, + 380 + ], + "score": 1.0, + "content": "is a simplification of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 379, + 176, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 382, + 166, + 399 + ], + "score": 0.82, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } [ \\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 379, + 176, + 402 + ], + "score": 1.0, + "content": "].", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "title", + "bbox": [ + 107, + 430, + 358, + 443 + ], + "lines": [ + { + "bbox": [ + 104, + 429, + 360, + 445 + ], + "spans": [ + { + "bbox": [ + 104, + 429, + 360, + 445 + ], + "score": 1.0, + "content": "B DISCUSSION OF THE THEORETICAL RESULTS", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 455, + 453, + 467 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 453, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 453, + 468 + ], + "score": 1.0, + "content": "We summarize the limitations of current theoretical results and future work as follows:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 129, + 475, + 505, + 659 + ], + "lines": [ + { + "bbox": [ + 128, + 474, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 128, + 474, + 370, + 491 + ], + "score": 1.0, + "content": "1. As discussed in Remark A.2, the solution Eq. 11 relies on", + "type": "text" + }, + { + "bbox": [ + 370, + 474, + 506, + 489 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) \\in [ 0 , c ] , \\forall a \\in \\mathcal { A } , \\forall x \\in", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 142, + 488, + 152, + 497 + ], + "score": 0.75, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 487, + 235, + 500 + ], + "score": 1.0, + "content": ". In some particular", + "type": "text" + }, + { + "bbox": [ + 235, + 487, + 251, + 497 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 487, + 366, + 500 + ], + "score": 1.0, + "content": ", it is intractable to derive a", + "type": "text" + }, + { + "bbox": [ + 366, + 488, + 373, + 498 + ], + "score": 0.85, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "that can generate an occupancy", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 498, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 142, + 498, + 196, + 511 + ], + "score": 1.0, + "content": "specified by", + "type": "text" + }, + { + "bbox": [ + 196, + 498, + 246, + 510 + ], + "score": 0.94, + "content": "g ( x , a ) / \\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 498, + 350, + 511 + ], + "score": 1.0, + "content": ". If more knowledge of", + "type": "text" + }, + { + "bbox": [ + 350, + 499, + 366, + 508 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 498, + 381, + 511 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 381, + 499, + 393, + 510 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 498, + 505, + 511 + ], + "score": 1.0, + "content": "is provided or some mild", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 141, + 509, + 339, + 522 + ], + "score": 1.0, + "content": "assumptions can be made on the properties of", + "type": "text" + }, + { + "bbox": [ + 339, + 509, + 356, + 519 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 509, + 370, + 522 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 370, + 510, + 382, + 520 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 509, + 453, + 522 + ], + "score": 1.0, + "content": ", we may model", + "type": "text" + }, + { + "bbox": [ + 453, + 511, + 460, + 521 + ], + "score": 0.79, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "in a more", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 520, + 369, + 533 + ], + "spans": [ + { + "bbox": [ + 142, + 520, + 369, + 533 + ], + "score": 1.0, + "content": "sophisticated approach to alleviating the above problem.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 128, + 533, + 506, + 549 + ], + "spans": [ + { + "bbox": [ + 128, + 533, + 370, + 549 + ], + "score": 1.0, + "content": "2. In the tractable solution derivation, we ignore the term", + "type": "text" + }, + { + "bbox": [ + 370, + 535, + 483, + 549 + ], + "score": 0.93, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 533, + 506, + 549 + ], + "score": 1.0, + "content": "(See", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 547, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 141, + 547, + 251, + 561 + ], + "score": 1.0, + "content": "Eq. 20). The benefit is that", + "type": "text" + }, + { + "bbox": [ + 252, + 547, + 306, + 560 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 547, + 505, + 561 + ], + "score": 1.0, + "content": "in the tractable solution can be estimated through", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 559, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 142, + 559, + 506, + 570 + ], + "score": 1.0, + "content": "offline datasets directly. Although the results in our experiments show that it does not", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 569, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 141, + 569, + 382, + 583 + ], + "score": 1.0, + "content": "produce significant negative effects in these tasks, ignoring", + "type": "text" + }, + { + "bbox": [ + 382, + 569, + 424, + 581 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 569, + 506, + 583 + ], + "score": 1.0, + "content": "indeed incurs extra", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 579, + 506, + 594 + ], + "spans": [ + { + "bbox": [ + 141, + 579, + 371, + 594 + ], + "score": 1.0, + "content": "bias in theory. In future work, techniques for estimating", + "type": "text" + }, + { + "bbox": [ + 371, + 582, + 414, + 593 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 579, + 506, + 594 + ], + "score": 1.0, + "content": "(Liu et al., 2020) can", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 140, + 590, + 507, + 605 + ], + "spans": [ + { + "bbox": [ + 140, + 590, + 361, + 605 + ], + "score": 1.0, + "content": "be incorporated to correct the bias. On the other hand,", + "type": "text" + }, + { + "bbox": [ + 362, + 594, + 384, + 604 + ], + "score": 0.89, + "content": "\\alpha _ { M _ { \\theta _ { t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 590, + 507, + 605 + ], + "score": 1.0, + "content": "is also ignored in the process.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 142, + 602, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 142, + 604, + 164, + 615 + ], + "score": 0.89, + "content": "\\alpha _ { M _ { \\theta _ { t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 602, + 506, + 616 + ], + "score": 1.0, + "content": "can be regarded as a global rescaling term of the final objective Eq. 20. Intuitively, it", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 613, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 141, + 613, + 506, + 625 + ], + "score": 1.0, + "content": "constructs an adaptive learning rate for Eq. 20, which increases the step size when the model", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 141, + 625, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 141, + 625, + 505, + 636 + ], + "score": 1.0, + "content": "is better fitted and decreases the step size otherwise. It can be considered to further improve", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 141, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 141, + 635, + 505, + 648 + ], + "score": 1.0, + "content": "the learning process in future work, e.g., cooperating with empirical risk minimization by", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 140, + 644, + 378, + 662 + ], + "spans": [ + { + "bbox": [ + 140, + 644, + 350, + 662 + ], + "score": 1.0, + "content": "balancing the weights of the two objectives through", + "type": "text" + }, + { + "bbox": [ + 351, + 648, + 372, + 659 + ], + "score": 0.9, + "content": "\\alpha _ { M _ { \\theta _ { t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 644, + 378, + 662 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 23.5 + }, + { + "type": "title", + "bbox": [ + 107, + 673, + 222, + 686 + ], + "lines": [ + { + "bbox": [ + 105, + 672, + 224, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 224, + 689 + ], + "score": 1.0, + "content": "C SOCIETAL IMPACT", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "This work studies a method toward counterfactual environment model learning. Reconstructing an", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "accurate environment of the real world will promote the wide adoption of decision-making policy", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "optimization methods in real life, enhancing our daily experience. We are aware that decision-making", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + } + ], + "page_idx": 21, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 403, + 505, + 413 + ], + "lines": [ + { + "bbox": [ + 495, + 405, + 505, + 413 + ], + "spans": [ + { + "bbox": [ + 495, + 405, + 505, + 413 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 79, + 507, + 118 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 85, + 163, + 100 + ], + "score": 1.0, + "content": "then we have", + "type": "text" + }, + { + "bbox": [ + 164, + 80, + 303, + 105 + ], + "score": 0.93, + "content": "\\begin{array} { r } { f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\approx T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 83, + 323, + 102 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 324, + 80, + 441, + 106 + ], + "score": 0.94, + "content": "\\begin{array} { r } { f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) \\approx T _ { \\varphi _ { 1 } ^ { * } } ( x , a ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 84, + 474, + 102 + ], + "score": 1.0, + "content": ". Given", + "type": "text" + }, + { + "bbox": [ + 474, + 88, + 486, + 99 + ], + "score": 0.88, + "content": "\\varphi _ { 0 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 84, + 506, + 102 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 103, + 434, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 119, + 117 + ], + "score": 0.88, + "content": "\\varphi _ { 1 } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 103, + 134, + 119 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 135, + 105, + 318, + 119 + ], + "score": 0.91, + "content": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , \\dot { a } , x ^ { \\prime } ) = T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) - T _ { \\varphi _ { 1 } ^ { * } } ( x , a ) ,", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 103, + 408, + 119 + ], + "score": 1.0, + "content": ", then we can optimize", + "type": "text" + }, + { + "bbox": [ + 409, + 106, + 415, + 115 + ], + "score": 0.77, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 103, + 434, + 119 + ], + "score": 1.0, + "content": "via:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 80, + 506, + 119 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 124, + 514, + 236 + ], + "lines": [ + { + "bbox": [ + 111, + 124, + 514, + 236 + ], + "spans": [ + { + "bbox": [ + 111, + 124, + 514, + 236 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\mathsf { \\Gamma } _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ) \\left( T _ { \\varphi _ { 0 } ^ { \\star } } ( x , a , x ^ { \\prime } ) - T _ { \\varphi _ { 1 } ^ { \\star } } ( x , a ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + } \\\\ & { \\qquad \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M ^ { \\star } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\left( T _ { \\varphi _ { 1 } ^ { \\star } } ( x , a ) - T _ { \\varphi _ { 0 } ^ { \\star } } ( x , a , x ^ { \\prime } ) + H _ { M ^ { \\star } } ( x , a ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\arg \\operatorname* { m a x } _ { \\theta } \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ) A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + } \\\\ & { \\qquad \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M ^ { \\star } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) ( - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) + H _ { M ^ { \\star } } ( x , a ) ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x . } \\end{array}", + "type": "interline_equation", + "image_path": "316a26a7d4a9506aaa976710d2906b3c44f9f15b0634f1cf4aa65219802c1038.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 111, + 124, + 514, + 161.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 161.33333333333334, + 514, + 198.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 111, + 198.66666666666669, + 514, + 236.00000000000003 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 506, + 290 + ], + "lines": [ + { + "bbox": [ + 106, + 245, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 106, + 245, + 198, + 258 + ], + "score": 1.0, + "content": "Based on the specific", + "type": "text" + }, + { + "bbox": [ + 198, + 246, + 205, + 257 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 245, + 330, + 258 + ], + "score": 1.0, + "content": "-divergence, we can represent", + "type": "text" + }, + { + "bbox": [ + 331, + 246, + 339, + 255 + ], + "score": 0.84, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 245, + 359, + 258 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 359, + 246, + 385, + 257 + ], + "score": 0.91, + "content": "f ^ { * } ( T )", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 245, + 474, + 258 + ], + "score": 1.0, + "content": "with a discriminator", + "type": "text" + }, + { + "bbox": [ + 475, + 246, + 489, + 258 + ], + "score": 0.88, + "content": "D _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 245, + 506, + 258 + ], + "score": 1.0, + "content": ". It", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 257, + 506, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 187, + 270 + ], + "score": 1.0, + "content": "can be verified that", + "type": "text" + }, + { + "bbox": [ + 188, + 257, + 339, + 269 + ], + "score": 0.78, + "content": "f ( u ) = u \\log u - ( u + 1 ) \\log ( u + 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 257, + 344, + 270 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 344, + 257, + 426, + 269 + ], + "score": 0.88, + "content": "T _ { \\varphi } ( u ) = \\log D _ { \\varphi } ( u )", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 257, + 448, + 270 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 449, + 257, + 506, + 269 + ], + "score": 0.88, + "content": "f ^ { * } ( T _ { \\varphi } ( u ) ) =", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 267, + 506, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 182, + 281 + ], + "score": 0.89, + "content": "- \\log ( 1 - D _ { \\varphi } ( u ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 267, + 408, + 281 + ], + "score": 1.0, + "content": "proposed in Nowozin et al. (2016) satisfies the condition", + "type": "text" + }, + { + "bbox": [ + 408, + 268, + 485, + 280 + ], + "score": 0.89, + "content": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 267, + 506, + 281 + ], + "score": 1.0, + "content": "(see", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 278, + 462, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 462, + 290 + ], + "score": 1.0, + "content": "Tab. 2). We select the former in the implementation and convert the tractable solution to:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 245, + 506, + 290 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 102, + 293, + 549, + 351 + ], + "lines": [ + { + "bbox": [ + 111, + 293, + 514, + 351 + ], + "spans": [ + { + "bbox": [ + 111, + 293, + 514, + 351 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\left( H _ { M ^ { \\star } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } \\\\ { \\therefore } & { _ { \\varphi _ { 0 } ^ { \\star } } = \\arg \\operatorname* { m a x } _ { \\theta _ { 0 } } \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } \\\\ & { _ { \\varphi _ { 1 } ^ { \\star } } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 1 } } \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "913cd2a525614fb7f942e419a24666f6fec2c71503368bd7a24a6117750e8fd9.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 102, + 293, + 549, + 312.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 102, + 312.3333333333333, + 549, + 331.66666666666663 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 102, + 331.66666666666663, + 549, + 350.99999999999994 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 366, + 506, + 398 + ], + "lines": [ + { + "bbox": [ + 106, + 358, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 407, + 387 + ], + "score": 1.0, + "content": "where Aφ∗0 ,φ∗1 (x, a, x′) = log Dφ∗0 (x, a, x′) − log Dφ∗1 (x, a), EρκMθt [·", + "type": "text" + }, + { + "bbox": [ + 411, + 365, + 506, + 380 + ], + "score": 1.0, + "content": "is a simplification of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 379, + 176, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 382, + 166, + 399 + ], + "score": 0.82, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } [ \\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 379, + 176, + 402 + ], + "score": 1.0, + "content": "].", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 106, + 358, + 506, + 402 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 430, + 358, + 443 + ], + "lines": [ + { + "bbox": [ + 104, + 429, + 360, + 445 + ], + "spans": [ + { + "bbox": [ + 104, + 429, + 360, + 445 + ], + "score": 1.0, + "content": "B DISCUSSION OF THE THEORETICAL RESULTS", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 455, + 453, + 467 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 453, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 453, + 468 + ], + "score": 1.0, + "content": "We summarize the limitations of current theoretical results and future work as follows:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 106, + 455, + 453, + 468 + ] + }, + { + "type": "list", + "bbox": [ + 129, + 475, + 505, + 659 + ], + "lines": [ + { + "bbox": [ + 128, + 474, + 506, + 491 + ], + "spans": [ + { + "bbox": [ + 128, + 474, + 370, + 491 + ], + "score": 1.0, + "content": "1. As discussed in Remark A.2, the solution Eq. 11 relies on", + "type": "text" + }, + { + "bbox": [ + 370, + 474, + 506, + 489 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) \\in [ 0 , c ] , \\forall a \\in \\mathcal { A } , \\forall x \\in", + "type": "inline_equation" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 142, + 488, + 152, + 497 + ], + "score": 0.75, + "content": "\\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 487, + 235, + 500 + ], + "score": 1.0, + "content": ". In some particular", + "type": "text" + }, + { + "bbox": [ + 235, + 487, + 251, + 497 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 487, + 366, + 500 + ], + "score": 1.0, + "content": ", it is intractable to derive a", + "type": "text" + }, + { + "bbox": [ + 366, + 488, + 373, + 498 + ], + "score": 0.85, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "that can generate an occupancy", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 498, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 142, + 498, + 196, + 511 + ], + "score": 1.0, + "content": "specified by", + "type": "text" + }, + { + "bbox": [ + 196, + 498, + 246, + 510 + ], + "score": 0.94, + "content": "g ( x , a ) / \\alpha _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 498, + 350, + 511 + ], + "score": 1.0, + "content": ". If more knowledge of", + "type": "text" + }, + { + "bbox": [ + 350, + 499, + 366, + 508 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 498, + 381, + 511 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 381, + 499, + 393, + 510 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 498, + 505, + 511 + ], + "score": 1.0, + "content": "is provided or some mild", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 141, + 509, + 339, + 522 + ], + "score": 1.0, + "content": "assumptions can be made on the properties of", + "type": "text" + }, + { + "bbox": [ + 339, + 509, + 356, + 519 + ], + "score": 0.88, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 509, + 370, + 522 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 370, + 510, + 382, + 520 + ], + "score": 0.88, + "content": "\\beta ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 509, + 453, + 522 + ], + "score": 1.0, + "content": ", we may model", + "type": "text" + }, + { + "bbox": [ + 453, + 511, + 460, + 521 + ], + "score": 0.79, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "in a more", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 142, + 520, + 369, + 533 + ], + "spans": [ + { + "bbox": [ + 142, + 520, + 369, + 533 + ], + "score": 1.0, + "content": "sophisticated approach to alleviating the above problem.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 533, + 506, + 549 + ], + "spans": [ + { + "bbox": [ + 128, + 533, + 370, + 549 + ], + "score": 1.0, + "content": "2. In the tractable solution derivation, we ignore the term", + "type": "text" + }, + { + "bbox": [ + 370, + 535, + 483, + 549 + ], + "score": 0.93, + "content": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 533, + 506, + 549 + ], + "score": 1.0, + "content": "(See", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 547, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 141, + 547, + 251, + 561 + ], + "score": 1.0, + "content": "Eq. 20). The benefit is that", + "type": "text" + }, + { + "bbox": [ + 252, + 547, + 306, + 560 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 547, + 505, + 561 + ], + "score": 1.0, + "content": "in the tractable solution can be estimated through", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 559, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 142, + 559, + 506, + 570 + ], + "score": 1.0, + "content": "offline datasets directly. Although the results in our experiments show that it does not", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 569, + 506, + 583 + ], + "spans": [ + { + "bbox": [ + 141, + 569, + 382, + 583 + ], + "score": 1.0, + "content": "produce significant negative effects in these tasks, ignoring", + "type": "text" + }, + { + "bbox": [ + 382, + 569, + 424, + 581 + ], + "score": 0.93, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 569, + 506, + 583 + ], + "score": 1.0, + "content": "indeed incurs extra", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 579, + 506, + 594 + ], + "spans": [ + { + "bbox": [ + 141, + 579, + 371, + 594 + ], + "score": 1.0, + "content": "bias in theory. In future work, techniques for estimating", + "type": "text" + }, + { + "bbox": [ + 371, + 582, + 414, + 593 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 579, + 506, + 594 + ], + "score": 1.0, + "content": "(Liu et al., 2020) can", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 140, + 590, + 507, + 605 + ], + "spans": [ + { + "bbox": [ + 140, + 590, + 361, + 605 + ], + "score": 1.0, + "content": "be incorporated to correct the bias. On the other hand,", + "type": "text" + }, + { + "bbox": [ + 362, + 594, + 384, + 604 + ], + "score": 0.89, + "content": "\\alpha _ { M _ { \\theta _ { t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 590, + 507, + 605 + ], + "score": 1.0, + "content": "is also ignored in the process.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 142, + 602, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 142, + 604, + 164, + 615 + ], + "score": 0.89, + "content": "\\alpha _ { M _ { \\theta _ { t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 602, + 506, + 616 + ], + "score": 1.0, + "content": "can be regarded as a global rescaling term of the final objective Eq. 20. Intuitively, it", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 613, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 141, + 613, + 506, + 625 + ], + "score": 1.0, + "content": "constructs an adaptive learning rate for Eq. 20, which increases the step size when the model", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 141, + 625, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 141, + 625, + 505, + 636 + ], + "score": 1.0, + "content": "is better fitted and decreases the step size otherwise. It can be considered to further improve", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 141, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 141, + 635, + 505, + 648 + ], + "score": 1.0, + "content": "the learning process in future work, e.g., cooperating with empirical risk minimization by", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 140, + 644, + 378, + 662 + ], + "spans": [ + { + "bbox": [ + 140, + 644, + 350, + 662 + ], + "score": 1.0, + "content": "balancing the weights of the two objectives through", + "type": "text" + }, + { + "bbox": [ + 351, + 648, + 372, + 659 + ], + "score": 0.9, + "content": "\\alpha _ { M _ { \\theta _ { t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 644, + 378, + 662 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31, + "is_list_end_line": true + } + ], + "index": 23.5, + "bbox_fs": [ + 128, + 474, + 507, + 662 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 673, + 222, + 686 + ], + "lines": [ + { + "bbox": [ + 105, + 672, + 224, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 224, + 689 + ], + "score": 1.0, + "content": "C SOCIETAL IMPACT", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "This work studies a method toward counterfactual environment model learning. Reconstructing an", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "accurate environment of the real world will promote the wide adoption of decision-making policy", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "optimization methods in real life, enhancing our daily experience. We are aware that decision-making", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "policy in some domains like recommendation systems that interact with customers may have risks of", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "causing price discrimination and misleading customers if inappropriately used. A promising way to", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "reduce the risk is to introduce fairness into policy optimization and rules to constrain the actions (Also", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "see our policy design in Sec. G.1.3). We are involved in and advocating research in such directions.", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "score": 1.0, + "content": "We believe that business organizations would like to embrace fair systems that can ultimately bring", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 371, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 371, + 150 + ], + "score": 1.0, + "content": "long-term financial benefits by providing a better user experience.", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 698, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "policy in some domains like recommendation systems that interact with customers may have risks of", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "causing price discrimination and misleading customers if inappropriately used. A promising way to", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "reduce the risk is to introduce fairness into policy optimization and rules to constrain the actions (Also", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "see our policy design in Sec. G.1.3). We are involved in and advocating research in such directions.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "score": 1.0, + "content": "We believe that business organizations would like to embrace fair systems that can ultimately bring", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 371, + 150 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 371, + 150 + ], + "score": 1.0, + "content": "long-term financial benefits by providing a better user experience.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 164, + 288, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 163, + 290, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 290, + 180 + ], + "score": 1.0, + "content": "D AWRM-ORACLE PSEUDOCODE", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 189, + 311, + 201 + ], + "lines": [ + { + "bbox": [ + 106, + 189, + 312, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 312, + 202 + ], + "score": 1.0, + "content": "We list the pseudocode of AWRM-oracle in Alg. 1.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 107, + 212, + 340, + 224 + ], + "lines": [ + { + "bbox": [ + 106, + 212, + 342, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 342, + 226 + ], + "score": 1.0, + "content": "Algorithm 1 AWRM with Oracle Counterfactual Datasets", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 227, + 249, + 249 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 136, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 136, + 239 + ], + "score": 1.0, + "content": "Input:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 237, + 249, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 114, + 247 + ], + "score": 0.76, + "content": "\\Phi", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 237, + 174, + 250 + ], + "score": 1.0, + "content": ": policy space;", + "type": "text" + }, + { + "bbox": [ + 174, + 238, + 184, + 247 + ], + "score": 0.73, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 237, + 249, + 250 + ], + "score": 1.0, + "content": ": total iterations", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "title", + "bbox": [ + 107, + 249, + 142, + 259 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 144, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 144, + 261 + ], + "score": 1.0, + "content": "Process:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 110, + 263, + 435, + 330 + ], + "lines": [ + { + "bbox": [ + 109, + 262, + 433, + 277 + ], + "spans": [ + { + "bbox": [ + 109, + 262, + 255, + 277 + ], + "score": 1.0, + "content": "1: Generate counterfactual datasets", + "type": "text" + }, + { + "bbox": [ + 255, + 264, + 284, + 276 + ], + "score": 0.9, + "content": "\\{ \\mathcal { D } _ { \\pi _ { \\phi } } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 262, + 391, + 277 + ], + "score": 1.0, + "content": "for all adversarial policies", + "type": "text" + }, + { + "bbox": [ + 391, + 264, + 433, + 276 + ], + "score": 0.87, + "content": "\\pi _ { \\phi } , \\phi \\in \\Phi", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 109, + 274, + 268, + 287 + ], + "spans": [ + { + "bbox": [ + 109, + 274, + 253, + 287 + ], + "score": 1.0, + "content": "2: Initialize an environment model", + "type": "text" + }, + { + "bbox": [ + 253, + 276, + 268, + 286 + ], + "score": 0.77, + "content": "M _ { \\theta }", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 109, + 285, + 183, + 298 + ], + "spans": [ + { + "bbox": [ + 109, + 285, + 137, + 298 + ], + "score": 1.0, + "content": "3: for", + "type": "text" + }, + { + "bbox": [ + 138, + 286, + 168, + 296 + ], + "score": 0.86, + "content": "\\mathrm { i } = 1 { : } \\mathrm { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 285, + 183, + 298 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 110, + 294, + 399, + 311 + ], + "spans": [ + { + "bbox": [ + 110, + 297, + 122, + 308 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 136, + 294, + 165, + 311 + ], + "score": 1.0, + "content": "Select", + "type": "text" + }, + { + "bbox": [ + 165, + 297, + 183, + 309 + ], + "score": 0.89, + "content": "\\mathcal { D } _ { \\pi _ { \\phi } }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 294, + 332, + 311 + ], + "score": 1.0, + "content": "with worst prediction errors through", + "type": "text" + }, + { + "bbox": [ + 332, + 297, + 347, + 308 + ], + "score": 0.88, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 294, + 371, + 311 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 371, + 297, + 399, + 310 + ], + "score": 0.83, + "content": "\\{ \\mathcal { D } _ { \\pi _ { \\phi } } \\}", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 110, + 304, + 387, + 324 + ], + "spans": [ + { + "bbox": [ + 110, + 308, + 122, + 320 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 135, + 304, + 178, + 324 + ], + "score": 1.0, + "content": "Optimize", + "type": "text" + }, + { + "bbox": [ + 178, + 308, + 193, + 319 + ], + "score": 0.87, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 304, + 369, + 324 + ], + "score": 1.0, + "content": "with standard supervised learning based on", + "type": "text" + }, + { + "bbox": [ + 369, + 309, + 387, + 321 + ], + "score": 0.66, + "content": "\\mathcal { D } _ { \\pi _ { \\phi } }", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 110, + 319, + 157, + 331 + ], + "spans": [ + { + "bbox": [ + 110, + 319, + 157, + 331 + ], + "score": 1.0, + "content": "6: end for", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 107, + 354, + 221, + 367 + ], + "lines": [ + { + "bbox": [ + 105, + 354, + 222, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 222, + 369 + ], + "score": 1.0, + "content": "E IMPLEMENTATION", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 107, + 378, + 333, + 391 + ], + "lines": [ + { + "bbox": [ + 105, + 379, + 334, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 334, + 392 + ], + "score": 1.0, + "content": "E.1 DETAILS OF THE GALILEO IMPLEMENTATION", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 105, + 397, + 504, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 298, + 412 + ], + "score": 1.0, + "content": "The approximation of Eq. 18 holds only when", + "type": "text" + }, + { + "bbox": [ + 298, + 397, + 315, + 415 + ], + "score": 0.93, + "content": "\\textstyle { \\frac { p ( x ) } { q ( x ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 397, + 506, + 412 + ], + "score": 1.0, + "content": "is close to 1, which might not be satisfied. To", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 412, + 369, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 369, + 425 + ], + "score": 1.0, + "content": "handle the problem, we inject a standard supervised learning loss", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 428, + 371, + 446 + ], + "lines": [ + { + "bbox": [ + 239, + 428, + 371, + 446 + ], + "spans": [ + { + "bbox": [ + 239, + 428, + 371, + 446 + ], + "score": 0.93, + "content": "\\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right]", + "type": "interline_equation", + "image_path": "8ba42c7940c2ea542bde8b4314180594bc5b9ef185f5b6ac16999cb1a17f4b2f.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 239, + 428, + 371, + 446 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 451, + 505, + 474 + ], + "lines": [ + { + "bbox": [ + 105, + 450, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 426, + 464 + ], + "score": 1.0, + "content": "to replace the second term of the above objective when the output probability of", + "type": "text" + }, + { + "bbox": [ + 427, + 452, + 436, + 461 + ], + "score": 0.85, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 450, + 505, + 464 + ], + "score": 1.0, + "content": "is far away from", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 462, + 195, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 124, + 475 + ], + "score": 1.0, + "content": "0.5", + "type": "text" + }, + { + "bbox": [ + 124, + 462, + 190, + 474 + ], + "score": 0.57, + "content": "\\bar { f } ^ { \\prime } ( 1 ) = \\log { 0 . 5 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 462, + 195, + 475 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 477, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 392, + 489 + ], + "score": 1.0, + "content": "In the offline model-learning setting, we only have a real-world dataset", + "type": "text" + }, + { + "bbox": [ + 392, + 478, + 401, + 488 + ], + "score": 0.8, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 478, + 505, + 489 + ], + "score": 1.0, + "content": "collected by the behavior", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 134, + 500 + ], + "score": 1.0, + "content": "policy", + "type": "text" + }, + { + "bbox": [ + 135, + 489, + 142, + 499 + ], + "score": 0.77, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 487, + 222, + 500 + ], + "score": 1.0, + "content": ". We learn a policy", + "type": "text" + }, + { + "bbox": [ + 222, + 488, + 249, + 499 + ], + "score": 0.91, + "content": "\\hat { \\mu } \\approx \\bar { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 487, + 358, + 500 + ], + "score": 1.0, + "content": "via behavior cloning with", + "type": "text" + }, + { + "bbox": [ + 359, + 488, + 368, + 498 + ], + "score": 0.81, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "(Pomerleau, 1991; Ho & Ermon,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 496, + 507, + 513 + ], + "spans": [ + { + "bbox": [ + 104, + 496, + 164, + 513 + ], + "score": 1.0, + "content": "2016) and let", + "type": "text" + }, + { + "bbox": [ + 164, + 499, + 172, + 509 + ], + "score": 0.84, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 496, + 230, + 513 + ], + "score": 1.0, + "content": "be the policy", + "type": "text" + }, + { + "bbox": [ + 231, + 501, + 237, + 508 + ], + "score": 0.78, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 496, + 289, + 513 + ], + "score": 1.0, + "content": ". We regard", + "type": "text" + }, + { + "bbox": [ + 289, + 499, + 299, + 507 + ], + "score": 0.83, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 496, + 451, + 513 + ], + "score": 1.0, + "content": "as the empirical data distribution of", + "type": "text" + }, + { + "bbox": [ + 451, + 499, + 470, + 509 + ], + "score": 0.87, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 496, + 507, + 513 + ], + "score": 1.0, + "content": "and the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 504, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 104, + 504, + 208, + 525 + ], + "score": 1.0, + "content": "trajectories collected by", + "type": "text" + }, + { + "bbox": [ + 208, + 509, + 216, + 520 + ], + "score": 0.84, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 504, + 273, + 525 + ], + "score": 1.0, + "content": "in the model", + "type": "text" + }, + { + "bbox": [ + 273, + 509, + 291, + 520 + ], + "score": 0.91, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 504, + 444, + 525 + ], + "score": 1.0, + "content": "as the empirical data distribution of", + "type": "text" + }, + { + "bbox": [ + 444, + 509, + 465, + 523 + ], + "score": 0.9, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 504, + 506, + 525 + ], + "score": 1.0, + "content": ". But the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 521, + 504, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 155, + 534 + ], + "score": 1.0, + "content": "assumption", + "type": "text" + }, + { + "bbox": [ + 155, + 521, + 275, + 533 + ], + "score": 0.83, + "content": "\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\mu ( a | x ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 521, + 496, + 534 + ], + "score": 1.0, + "content": "might not be satisfied. In behavior cloning, we model", + "type": "text" + }, + { + "bbox": [ + 497, + 522, + 504, + 533 + ], + "score": 0.82, + "content": "\\hat { \\mu }", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 531, + 504, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 475, + 543 + ], + "score": 1.0, + "content": "with a Gaussian distribution and constrain the lower bound of the variance with a small value", + "type": "text" + }, + { + "bbox": [ + 475, + 533, + 504, + 543 + ], + "score": 0.91, + "content": "\\epsilon _ { \\mu } > 0", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 542, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 393, + 555 + ], + "score": 1.0, + "content": "to keep the assumption holding. Besides, we add small Gaussian noises", + "type": "text" + }, + { + "bbox": [ + 393, + 542, + 452, + 555 + ], + "score": 0.93, + "content": "\\mathbf { u } \\sim \\mathcal { N } ( 0 , \\epsilon _ { D } )", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 542, + 505, + 555 + ], + "score": 1.0, + "content": "to the inputs", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 554, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 554, + 117, + 570 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 117, + 556, + 132, + 568 + ], + "score": 0.9, + "content": "D _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 554, + 261, + 570 + ], + "score": 1.0, + "content": "to handle the mismatch between", + "type": "text" + }, + { + "bbox": [ + 261, + 555, + 280, + 568 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 554, + 298, + 570 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 298, + 554, + 317, + 568 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\hat { \\mu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 554, + 345, + 570 + ], + "score": 1.0, + "content": "due to", + "type": "text" + }, + { + "bbox": [ + 345, + 557, + 356, + 568 + ], + "score": 0.85, + "content": "\\epsilon _ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 554, + 426, + 570 + ], + "score": 1.0, + "content": ". In particular, for", + "type": "text" + }, + { + "bbox": [ + 426, + 558, + 438, + 568 + ], + "score": 0.87, + "content": "\\varphi _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 554, + 456, + 570 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 456, + 557, + 468, + 568 + ], + "score": 0.86, + "content": "\\varphi _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 554, + 506, + 570 + ], + "score": 1.0, + "content": "learning,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 567, + 144, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 144, + 578 + ], + "score": 1.0, + "content": "we have:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 579, + 514, + 618 + ], + "lines": [ + { + "bbox": [ + 111, + 579, + 514, + 618 + ], + "spans": [ + { + "bbox": [ + 111, + 579, + 514, + 618 + ], + "score": 0.87, + "content": "\\begin{array} { r l } & { \\gamma _ { 0 } ^ { * } = \\arg \\underset { \\varphi _ { 0 } } { \\operatorname* { m a x } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) ) \\right] } \\\\ & { \\gamma _ { 1 } ^ { * } = \\arg \\underset { \\varphi _ { 1 } } { \\operatorname* { m a x } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "78d40402d4911316f7024794630711f2a1bed4b1c9fce5c830274a47b7d591a6.jpg" + } + ] + } + ], + "index": 35, + "virtual_lines": [ + { + "bbox": [ + 111, + 579, + 514, + 592.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 111, + 592.0, + 514, + 605.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 111, + 605.0, + 514, + 618.0 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 622, + 457, + 640 + ], + "lines": [ + { + "bbox": [ + 108, + 618, + 454, + 644 + ], + "spans": [ + { + "bbox": [ + 108, + 618, + 164, + 641 + ], + "score": 1.0, + "content": "where EρκMθ ,", + "type": "text" + }, + { + "bbox": [ + 177, + 619, + 262, + 644 + ], + "score": 1.0, + "content": "is a simplification of", + "type": "text" + }, + { + "bbox": [ + 262, + 623, + 366, + 640 + ], + "score": 0.76, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\mathbf { u } \\sim \\mathcal { N } ( 0 , \\epsilon _ { D } ) } [ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 619, + 384, + 644 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 384, + 623, + 454, + 635 + ], + "score": 0.84, + "content": "\\mathbf { u } = [ u _ { x } , u _ { a } , u _ { x ^ { \\prime } } ] .", + "type": "inline_equation" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 643, + 506, + 734 + ], + "lines": [ + { + "bbox": [ + 106, + 643, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 643, + 506, + 655 + ], + "score": 1.0, + "content": "On the other hand, we notice that the first term in Eq. 21 is similar to the objective of", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 653, + 507, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 291, + 667 + ], + "score": 1.0, + "content": "GAIL (Ho & Ermon, 2016) by regarding", + "type": "text" + }, + { + "bbox": [ + 291, + 655, + 307, + 666 + ], + "score": 0.89, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 653, + 428, + 667 + ], + "score": 1.0, + "content": "as the policy to learn and", + "type": "text" + }, + { + "bbox": [ + 429, + 657, + 436, + 664 + ], + "score": 0.76, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 653, + 507, + 667 + ], + "score": 1.0, + "content": "as the environ-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "ment to generate data. For better capability in sequential environment model learning, here", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "we introduce some practical tricks inspired by GAIL for model learning (Shi et al., 2019;", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 686, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 316, + 700 + ], + "score": 1.0, + "content": "Shang et al., 2019): we introduce an MDP for", + "type": "text" + }, + { + "bbox": [ + 316, + 690, + 324, + 698 + ], + "score": 0.75, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 686, + 345, + 700 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 346, + 688, + 361, + 698 + ], + "score": 0.89, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 686, + 506, + 700 + ], + "score": 1.0, + "content": ", where the reward is defined by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 697, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 182, + 712 + ], + "score": 1.0, + "content": "the discriminator", + "type": "text" + }, + { + "bbox": [ + 182, + 699, + 192, + 708 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 697, + 219, + 712 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + }, + { + "bbox": [ + 219, + 698, + 340, + 711 + ], + "score": 0.91, + "content": "r ( x , a , x ^ { \\prime } ) ~ = ~ \\log D ( x , a , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 697, + 352, + 712 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 352, + 699, + 367, + 709 + ], + "score": 0.86, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 697, + 507, + 712 + ], + "score": 1.0, + "content": "is learned to maximize the cu-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 708, + 506, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 721 + ], + "score": 1.0, + "content": "mulative rewards. With advanced policy gradient methods (Schulman et al., 2015; 2017), the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 719, + 507, + 737 + ], + "spans": [ + { + "bbox": [ + 104, + 719, + 239, + 734 + ], + "score": 1.0, + "content": "objective is converted to maxθ", + "type": "text" + }, + { + "bbox": [ + 240, + 720, + 380, + 734 + ], + "score": 0.91, + "content": "\\left[ \\bar { A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 719, + 415, + 734 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 415, + 720, + 502, + 735 + ], + "score": 0.92, + "content": "{ \\cal A } \\ = \\ Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } \\ - \\ V _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 720, + 507, + 737 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5 + } + ], + "page_idx": 22, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [], + "index": 2.5, + "bbox_fs": [ + 105, + 82, + 506, + 150 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 164, + 288, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 163, + 290, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 290, + 180 + ], + "score": 1.0, + "content": "D AWRM-ORACLE PSEUDOCODE", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 189, + 311, + 201 + ], + "lines": [ + { + "bbox": [ + 106, + 189, + 312, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 312, + 202 + ], + "score": 1.0, + "content": "We list the pseudocode of AWRM-oracle in Alg. 1.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 189, + 312, + 202 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 212, + 340, + 224 + ], + "lines": [ + { + "bbox": [ + 106, + 212, + 342, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 212, + 342, + 226 + ], + "score": 1.0, + "content": "Algorithm 1 AWRM with Oracle Counterfactual Datasets", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 227, + 249, + 249 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 136, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 136, + 239 + ], + "score": 1.0, + "content": "Input:", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 237, + 249, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 114, + 247 + ], + "score": 0.76, + "content": "\\Phi", + "type": "inline_equation" + }, + { + "bbox": [ + 115, + 237, + 174, + 250 + ], + "score": 1.0, + "content": ": policy space;", + "type": "text" + }, + { + "bbox": [ + 174, + 238, + 184, + 247 + ], + "score": 0.73, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 237, + 249, + 250 + ], + "score": 1.0, + "content": ": total iterations", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 225, + 249, + 250 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 249, + 142, + 259 + ], + "lines": [ + { + "bbox": [ + 105, + 248, + 144, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 144, + 261 + ], + "score": 1.0, + "content": "Process:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "index", + "bbox": [ + 110, + 263, + 435, + 330 + ], + "lines": [ + { + "bbox": [ + 109, + 262, + 433, + 277 + ], + "spans": [ + { + "bbox": [ + 109, + 262, + 255, + 277 + ], + "score": 1.0, + "content": "1: Generate counterfactual datasets", + "type": "text" + }, + { + "bbox": [ + 255, + 264, + 284, + 276 + ], + "score": 0.9, + "content": "\\{ \\mathcal { D } _ { \\pi _ { \\phi } } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 262, + 391, + 277 + ], + "score": 1.0, + "content": "for all adversarial policies", + "type": "text" + }, + { + "bbox": [ + 391, + 264, + 433, + 276 + ], + "score": 0.87, + "content": "\\pi _ { \\phi } , \\phi \\in \\Phi", + "type": "inline_equation" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 274, + 268, + 287 + ], + "spans": [ + { + "bbox": [ + 109, + 274, + 253, + 287 + ], + "score": 1.0, + "content": "2: Initialize an environment model", + "type": "text" + }, + { + "bbox": [ + 253, + 276, + 268, + 286 + ], + "score": 0.77, + "content": "M _ { \\theta }", + "type": "inline_equation" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 285, + 183, + 298 + ], + "spans": [ + { + "bbox": [ + 109, + 285, + 137, + 298 + ], + "score": 1.0, + "content": "3: for", + "type": "text" + }, + { + "bbox": [ + 138, + 286, + 168, + 296 + ], + "score": 0.86, + "content": "\\mathrm { i } = 1 { : } \\mathrm { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 285, + 183, + 298 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 294, + 399, + 311 + ], + "spans": [ + { + "bbox": [ + 110, + 297, + 122, + 308 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 136, + 294, + 165, + 311 + ], + "score": 1.0, + "content": "Select", + "type": "text" + }, + { + "bbox": [ + 165, + 297, + 183, + 309 + ], + "score": 0.89, + "content": "\\mathcal { D } _ { \\pi _ { \\phi } }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 294, + 332, + 311 + ], + "score": 1.0, + "content": "with worst prediction errors through", + "type": "text" + }, + { + "bbox": [ + 332, + 297, + 347, + 308 + ], + "score": 0.88, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 294, + 371, + 311 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 371, + 297, + 399, + 310 + ], + "score": 0.83, + "content": "\\{ \\mathcal { D } _ { \\pi _ { \\phi } } \\}", + "type": "inline_equation" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 304, + 387, + 324 + ], + "spans": [ + { + "bbox": [ + 110, + 308, + 122, + 320 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 135, + 304, + 178, + 324 + ], + "score": 1.0, + "content": "Optimize", + "type": "text" + }, + { + "bbox": [ + 178, + 308, + 193, + 319 + ], + "score": 0.87, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 304, + 369, + 324 + ], + "score": 1.0, + "content": "with standard supervised learning based on", + "type": "text" + }, + { + "bbox": [ + 369, + 309, + 387, + 321 + ], + "score": 0.66, + "content": "\\mathcal { D } _ { \\pi _ { \\phi } }", + "type": "inline_equation" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 319, + 157, + 331 + ], + "spans": [ + { + "bbox": [ + 110, + 319, + 157, + 331 + ], + "score": 1.0, + "content": "6: end for", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + } + ], + "index": 14.5, + "bbox_fs": [ + 109, + 262, + 433, + 331 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 354, + 221, + 367 + ], + "lines": [ + { + "bbox": [ + 105, + 354, + 222, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 354, + 222, + 369 + ], + "score": 1.0, + "content": "E IMPLEMENTATION", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 107, + 378, + 333, + 391 + ], + "lines": [ + { + "bbox": [ + 105, + 379, + 334, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 334, + 392 + ], + "score": 1.0, + "content": "E.1 DETAILS OF THE GALILEO IMPLEMENTATION", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 105, + 397, + 504, + 424 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 506, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 298, + 412 + ], + "score": 1.0, + "content": "The approximation of Eq. 18 holds only when", + "type": "text" + }, + { + "bbox": [ + 298, + 397, + 315, + 415 + ], + "score": 0.93, + "content": "\\textstyle { \\frac { p ( x ) } { q ( x ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 397, + 506, + 412 + ], + "score": 1.0, + "content": "is close to 1, which might not be satisfied. To", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 412, + 369, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 369, + 425 + ], + "score": 1.0, + "content": "handle the problem, we inject a standard supervised learning loss", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 397, + 506, + 425 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 428, + 371, + 446 + ], + "lines": [ + { + "bbox": [ + 239, + 428, + 371, + 446 + ], + "spans": [ + { + "bbox": [ + 239, + 428, + 371, + 446 + ], + "score": 0.93, + "content": "\\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right]", + "type": "interline_equation", + "image_path": "8ba42c7940c2ea542bde8b4314180594bc5b9ef185f5b6ac16999cb1a17f4b2f.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 239, + 428, + 371, + 446 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 451, + 505, + 474 + ], + "lines": [ + { + "bbox": [ + 105, + 450, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 426, + 464 + ], + "score": 1.0, + "content": "to replace the second term of the above objective when the output probability of", + "type": "text" + }, + { + "bbox": [ + 427, + 452, + 436, + 461 + ], + "score": 0.85, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 450, + 505, + 464 + ], + "score": 1.0, + "content": "is far away from", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 462, + 195, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 124, + 475 + ], + "score": 1.0, + "content": "0.5", + "type": "text" + }, + { + "bbox": [ + 124, + 462, + 190, + 474 + ], + "score": 0.57, + "content": "\\bar { f } ^ { \\prime } ( 1 ) = \\log { 0 . 5 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 462, + 195, + 475 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 450, + 505, + 475 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 477, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 106, + 478, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 478, + 392, + 489 + ], + "score": 1.0, + "content": "In the offline model-learning setting, we only have a real-world dataset", + "type": "text" + }, + { + "bbox": [ + 392, + 478, + 401, + 488 + ], + "score": 0.8, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 478, + 505, + 489 + ], + "score": 1.0, + "content": "collected by the behavior", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 134, + 500 + ], + "score": 1.0, + "content": "policy", + "type": "text" + }, + { + "bbox": [ + 135, + 489, + 142, + 499 + ], + "score": 0.77, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 487, + 222, + 500 + ], + "score": 1.0, + "content": ". We learn a policy", + "type": "text" + }, + { + "bbox": [ + 222, + 488, + 249, + 499 + ], + "score": 0.91, + "content": "\\hat { \\mu } \\approx \\bar { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 487, + 358, + 500 + ], + "score": 1.0, + "content": "via behavior cloning with", + "type": "text" + }, + { + "bbox": [ + 359, + 488, + 368, + 498 + ], + "score": 0.81, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "(Pomerleau, 1991; Ho & Ermon,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 496, + 507, + 513 + ], + "spans": [ + { + "bbox": [ + 104, + 496, + 164, + 513 + ], + "score": 1.0, + "content": "2016) and let", + "type": "text" + }, + { + "bbox": [ + 164, + 499, + 172, + 509 + ], + "score": 0.84, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 496, + 230, + 513 + ], + "score": 1.0, + "content": "be the policy", + "type": "text" + }, + { + "bbox": [ + 231, + 501, + 237, + 508 + ], + "score": 0.78, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 496, + 289, + 513 + ], + "score": 1.0, + "content": ". We regard", + "type": "text" + }, + { + "bbox": [ + 289, + 499, + 299, + 507 + ], + "score": 0.83, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 496, + 451, + 513 + ], + "score": 1.0, + "content": "as the empirical data distribution of", + "type": "text" + }, + { + "bbox": [ + 451, + 499, + 470, + 509 + ], + "score": 0.87, + "content": "\\rho _ { M ^ { * } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 496, + 507, + 513 + ], + "score": 1.0, + "content": "and the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 504, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 104, + 504, + 208, + 525 + ], + "score": 1.0, + "content": "trajectories collected by", + "type": "text" + }, + { + "bbox": [ + 208, + 509, + 216, + 520 + ], + "score": 0.84, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 504, + 273, + 525 + ], + "score": 1.0, + "content": "in the model", + "type": "text" + }, + { + "bbox": [ + 273, + 509, + 291, + 520 + ], + "score": 0.91, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 504, + 444, + 525 + ], + "score": 1.0, + "content": "as the empirical data distribution of", + "type": "text" + }, + { + "bbox": [ + 444, + 509, + 465, + 523 + ], + "score": 0.9, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 504, + 506, + 525 + ], + "score": 1.0, + "content": ". But the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 521, + 504, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 155, + 534 + ], + "score": 1.0, + "content": "assumption", + "type": "text" + }, + { + "bbox": [ + 155, + 521, + 275, + 533 + ], + "score": 0.83, + "content": "\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\mu ( a | x ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 521, + 496, + 534 + ], + "score": 1.0, + "content": "might not be satisfied. In behavior cloning, we model", + "type": "text" + }, + { + "bbox": [ + 497, + 522, + 504, + 533 + ], + "score": 0.82, + "content": "\\hat { \\mu }", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 531, + 504, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 475, + 543 + ], + "score": 1.0, + "content": "with a Gaussian distribution and constrain the lower bound of the variance with a small value", + "type": "text" + }, + { + "bbox": [ + 475, + 533, + 504, + 543 + ], + "score": 0.91, + "content": "\\epsilon _ { \\mu } > 0", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 542, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 393, + 555 + ], + "score": 1.0, + "content": "to keep the assumption holding. Besides, we add small Gaussian noises", + "type": "text" + }, + { + "bbox": [ + 393, + 542, + 452, + 555 + ], + "score": 0.93, + "content": "\\mathbf { u } \\sim \\mathcal { N } ( 0 , \\epsilon _ { D } )", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 542, + 505, + 555 + ], + "score": 1.0, + "content": "to the inputs", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 554, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 104, + 554, + 117, + 570 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 117, + 556, + 132, + 568 + ], + "score": 0.9, + "content": "D _ { \\varphi }", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 554, + 261, + 570 + ], + "score": 1.0, + "content": "to handle the mismatch between", + "type": "text" + }, + { + "bbox": [ + 261, + 555, + 280, + 568 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 554, + 298, + 570 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 298, + 554, + 317, + 568 + ], + "score": 0.92, + "content": "\\rho _ { M ^ { * } } ^ { \\hat { \\mu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 554, + 345, + 570 + ], + "score": 1.0, + "content": "due to", + "type": "text" + }, + { + "bbox": [ + 345, + 557, + 356, + 568 + ], + "score": 0.85, + "content": "\\epsilon _ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 554, + 426, + 570 + ], + "score": 1.0, + "content": ". In particular, for", + "type": "text" + }, + { + "bbox": [ + 426, + 558, + 438, + 568 + ], + "score": 0.87, + "content": "\\varphi _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 554, + 456, + 570 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 456, + 557, + 468, + 568 + ], + "score": 0.86, + "content": "\\varphi _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 554, + 506, + 570 + ], + "score": 1.0, + "content": "learning,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 567, + 144, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 144, + 578 + ], + "score": 1.0, + "content": "we have:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29, + "bbox_fs": [ + 104, + 478, + 507, + 578 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 579, + 514, + 618 + ], + "lines": [ + { + "bbox": [ + 111, + 579, + 514, + 618 + ], + "spans": [ + { + "bbox": [ + 111, + 579, + 514, + 618 + ], + "score": 0.87, + "content": "\\begin{array} { r l } & { \\gamma _ { 0 } ^ { * } = \\arg \\underset { \\varphi _ { 0 } } { \\operatorname* { m a x } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) ) \\right] } \\\\ & { \\gamma _ { 1 } ^ { * } = \\arg \\underset { \\varphi _ { 1 } } { \\operatorname* { m a x } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "78d40402d4911316f7024794630711f2a1bed4b1c9fce5c830274a47b7d591a6.jpg" + } + ] + } + ], + "index": 35, + "virtual_lines": [ + { + "bbox": [ + 111, + 579, + 514, + 592.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 111, + 592.0, + 514, + 605.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 111, + 605.0, + 514, + 618.0 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 622, + 457, + 640 + ], + "lines": [ + { + "bbox": [ + 108, + 618, + 454, + 644 + ], + "spans": [ + { + "bbox": [ + 108, + 618, + 164, + 641 + ], + "score": 1.0, + "content": "where EρκMθ ,", + "type": "text" + }, + { + "bbox": [ + 177, + 619, + 262, + 644 + ], + "score": 1.0, + "content": "is a simplification of", + "type": "text" + }, + { + "bbox": [ + 262, + 623, + 366, + 640 + ], + "score": 0.76, + "content": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\mathbf { u } \\sim \\mathcal { N } ( 0 , \\epsilon _ { D } ) } [ \\cdot ]", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 619, + 384, + 644 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 384, + 623, + 454, + 635 + ], + "score": 0.84, + "content": "\\mathbf { u } = [ u _ { x } , u _ { a } , u _ { x ^ { \\prime } } ] .", + "type": "inline_equation" + } + ], + "index": 37 + } + ], + "index": 37, + "bbox_fs": [ + 108, + 618, + 454, + 644 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 643, + 506, + 734 + ], + "lines": [ + { + "bbox": [ + 106, + 643, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 643, + 506, + 655 + ], + "score": 1.0, + "content": "On the other hand, we notice that the first term in Eq. 21 is similar to the objective of", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 653, + 507, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 291, + 667 + ], + "score": 1.0, + "content": "GAIL (Ho & Ermon, 2016) by regarding", + "type": "text" + }, + { + "bbox": [ + 291, + 655, + 307, + 666 + ], + "score": 0.89, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 653, + 428, + 667 + ], + "score": 1.0, + "content": "as the policy to learn and", + "type": "text" + }, + { + "bbox": [ + 429, + 657, + 436, + 664 + ], + "score": 0.76, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 653, + 507, + 667 + ], + "score": 1.0, + "content": "as the environ-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "ment to generate data. For better capability in sequential environment model learning, here", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "we introduce some practical tricks inspired by GAIL for model learning (Shi et al., 2019;", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 686, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 316, + 700 + ], + "score": 1.0, + "content": "Shang et al., 2019): we introduce an MDP for", + "type": "text" + }, + { + "bbox": [ + 316, + 690, + 324, + 698 + ], + "score": 0.75, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 686, + 345, + 700 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 346, + 688, + 361, + 698 + ], + "score": 0.89, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 686, + 506, + 700 + ], + "score": 1.0, + "content": ", where the reward is defined by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 697, + 507, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 182, + 712 + ], + "score": 1.0, + "content": "the discriminator", + "type": "text" + }, + { + "bbox": [ + 182, + 699, + 192, + 708 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 697, + 219, + 712 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + }, + { + "bbox": [ + 219, + 698, + 340, + 711 + ], + "score": 0.91, + "content": "r ( x , a , x ^ { \\prime } ) ~ = ~ \\log D ( x , a , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 697, + 352, + 712 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 352, + 699, + 367, + 709 + ], + "score": 0.86, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 697, + 507, + 712 + ], + "score": 1.0, + "content": "is learned to maximize the cu-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 708, + 506, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 506, + 721 + ], + "score": 1.0, + "content": "mulative rewards. With advanced policy gradient methods (Schulman et al., 2015; 2017), the", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 719, + 507, + 737 + ], + "spans": [ + { + "bbox": [ + 104, + 719, + 239, + 734 + ], + "score": 1.0, + "content": "objective is converted to maxθ", + "type": "text" + }, + { + "bbox": [ + 240, + 720, + 380, + 734 + ], + "score": 0.91, + "content": "\\left[ \\bar { A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 719, + 415, + 734 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 415, + 720, + 502, + 735 + ], + "score": 0.92, + "content": "{ \\cal A } \\ = \\ Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } \\ - \\ V _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 720, + 507, + 737 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5, + "bbox_fs": [ + 104, + 643, + 507, + 737 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 254, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 256, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 256, + 96 + ], + "score": 1.0, + "content": "Algorithm 2 GALILEO pseudocode", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 96, + 134, + 106 + ], + "lines": [ + { + "bbox": [ + 107, + 96, + 134, + 106 + ], + "spans": [], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 106, + 396, + 128 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 397, + 120 + ], + "spans": [ + { + "bbox": [ + 106, + 107, + 128, + 118 + ], + "score": 0.88, + "content": "\\mathcal { D } _ { \\mathrm { r e a l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 105, + 249, + 120 + ], + "score": 1.0, + "content": ": offline dataset sampled from", + "type": "text" + }, + { + "bbox": [ + 249, + 106, + 268, + 119 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 105, + 297, + 120 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 297, + 109, + 304, + 118 + ], + "score": 0.82, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 105, + 397, + 120 + ], + "score": 1.0, + "content": "is the behavior policy;", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 116, + 184, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 117, + 127 + ], + "score": 0.78, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 116, + 184, + 129 + ], + "score": 1.0, + "content": ": total iterations;", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 106, + 129, + 142, + 139 + ], + "lines": [ + { + "bbox": [ + 105, + 128, + 144, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 144, + 141 + ], + "score": 1.0, + "content": "Process:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 110, + 145, + 506, + 343 + ], + "lines": [ + { + "bbox": [ + 111, + 143, + 342, + 157 + ], + "spans": [ + { + "bbox": [ + 111, + 143, + 250, + 157 + ], + "score": 1.0, + "content": "1: Approximate a behavior policy", + "type": "text" + }, + { + "bbox": [ + 250, + 145, + 257, + 156 + ], + "score": 0.84, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 143, + 342, + 157 + ], + "score": 1.0, + "content": "via behavior cloning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 110, + 154, + 272, + 168 + ], + "spans": [ + { + "bbox": [ + 110, + 154, + 253, + 168 + ], + "score": 1.0, + "content": "2: Initialize an environment model", + "type": "text" + }, + { + "bbox": [ + 253, + 156, + 272, + 167 + ], + "score": 0.9, + "content": "M _ { \\theta _ { 1 } }", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 110, + 166, + 194, + 178 + ], + "spans": [ + { + "bbox": [ + 110, + 166, + 138, + 178 + ], + "score": 1.0, + "content": "3: for", + "type": "text" + }, + { + "bbox": [ + 138, + 167, + 179, + 177 + ], + "score": 0.89, + "content": "t = 1 : N", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 166, + 194, + 178 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 110, + 176, + 349, + 191 + ], + "spans": [ + { + "bbox": [ + 110, + 177, + 122, + 189 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 137, + 176, + 156, + 191 + ], + "score": 1.0, + "content": "Use", + "type": "text" + }, + { + "bbox": [ + 156, + 178, + 164, + 189 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 176, + 248, + 191 + ], + "score": 1.0, + "content": "to generate a dataset", + "type": "text" + }, + { + "bbox": [ + 249, + 178, + 268, + 190 + ], + "score": 0.91, + "content": "\\mathcal { D } _ { \\mathrm { g e n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 176, + 331, + 191 + ], + "score": 1.0, + "content": "with the model", + "type": "text" + }, + { + "bbox": [ + 332, + 177, + 349, + 189 + ], + "score": 0.9, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 111, + 185, + 504, + 208 + ], + "spans": [ + { + "bbox": [ + 111, + 191, + 122, + 203 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 135, + 185, + 245, + 208 + ], + "score": 1.0, + "content": "Update the discriminators", + "type": "text" + }, + { + "bbox": [ + 245, + 191, + 264, + 203 + ], + "score": 0.91, + "content": "D _ { \\varphi _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 185, + 282, + 208 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 283, + 191, + 302, + 203 + ], + "score": 0.91, + "content": "D _ { \\varphi _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 185, + 473, + 208 + ], + "score": 1.0, + "content": "via Eq. 26 and Eq. 27 respectively, where", + "type": "text" + }, + { + "bbox": [ + 473, + 189, + 493, + 206 + ], + "score": 0.92, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\hat { \\mu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 185, + 504, + 208 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 122, + 203, + 321, + 219 + ], + "spans": [ + { + "bbox": [ + 122, + 203, + 176, + 219 + ], + "score": 1.0, + "content": "estimated by", + "type": "text" + }, + { + "bbox": [ + 177, + 205, + 198, + 217 + ], + "score": 0.91, + "content": "\\mathcal { D } _ { \\mathrm { g e n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 203, + 216, + 219 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 216, + 204, + 235, + 218 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 203, + 299, + 219 + ], + "score": 1.0, + "content": "is estimated by", + "type": "text" + }, + { + "bbox": [ + 299, + 205, + 321, + 216 + ], + "score": 0.89, + "content": "\\mathcal { D } _ { \\mathrm { r e a l } }", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 111, + 213, + 413, + 231 + ], + "spans": [ + { + "bbox": [ + 111, + 217, + 122, + 228 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 136, + 213, + 169, + 231 + ], + "score": 1.0, + "content": "Update", + "type": "text" + }, + { + "bbox": [ + 170, + 217, + 179, + 227 + ], + "score": 0.86, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 213, + 197, + 231 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 197, + 217, + 207, + 226 + ], + "score": 0.77, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 213, + 329, + 231 + ], + "score": 1.0, + "content": "via Eq. 24 and Eq. 25 through", + "type": "text" + }, + { + "bbox": [ + 329, + 217, + 351, + 228 + ], + "score": 0.88, + "content": "\\mathcal { D } _ { \\mathrm { g e n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 213, + 354, + 231 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 354, + 216, + 373, + 228 + ], + "score": 0.89, + "content": "D _ { \\varphi _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 213, + 394, + 231 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 394, + 217, + 413, + 228 + ], + "score": 0.9, + "content": "D _ { \\varphi _ { 1 } }", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 111, + 226, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 111, + 227, + 122, + 239 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 138, + 226, + 213, + 239 + ], + "score": 1.0, + "content": "Update the model", + "type": "text" + }, + { + "bbox": [ + 214, + 227, + 232, + 239 + ], + "score": 0.92, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 226, + 506, + 239 + ], + "score": 1.0, + "content": "via the first term of Eq. 23, which is implemented with a standard", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 122, + 238, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 122, + 238, + 506, + 250 + ], + "score": 1.0, + "content": "policy gradient method like TRPO (Schulman et al., 2015) or PPO (Schulman et al., 2017).", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 122, + 247, + 249, + 263 + ], + "spans": [ + { + "bbox": [ + 122, + 247, + 249, + 263 + ], + "score": 1.0, + "content": "Record the policy gradient gpg", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 110, + 259, + 320, + 275 + ], + "spans": [ + { + "bbox": [ + 110, + 260, + 122, + 272 + ], + "score": 1.0, + "content": "8:", + "type": "text" + }, + { + "bbox": [ + 136, + 259, + 146, + 275 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 146, + 260, + 296, + 273 + ], + "score": 0.88, + "content": "p _ { 0 } < \\mathbb { E } _ { \\mathcal { D } _ { \\mathrm { g e n } } } \\left[ D _ { \\varphi _ { 0 } } ( x _ { t } , \\hat { a _ { t } } , x _ { t + 1 } ) \\right] < p _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 259, + 320, + 275 + ], + "score": 1.0, + "content": "then", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 110, + 268, + 468, + 285 + ], + "spans": [ + { + "bbox": [ + 110, + 271, + 122, + 282 + ], + "score": 1.0, + "content": "9:", + "type": "text" + }, + { + "bbox": [ + 151, + 268, + 253, + 285 + ], + "score": 1.0, + "content": "Compute the gradient of", + "type": "text" + }, + { + "bbox": [ + 253, + 272, + 271, + 283 + ], + "score": 0.9, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 268, + 456, + 285 + ], + "score": 1.0, + "content": "via the second term of Eq. 23 and record it as", + "type": "text" + }, + { + "bbox": [ + 456, + 273, + 468, + 282 + ], + "score": 0.81, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 281, + 158, + 294 + ], + "spans": [ + { + "bbox": [ + 107, + 281, + 122, + 294 + ], + "score": 1.0, + "content": "10:", + "type": "text" + }, + { + "bbox": [ + 137, + 281, + 158, + 294 + ], + "score": 1.0, + "content": "else", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 291, + 392, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 122, + 304 + ], + "score": 1.0, + "content": "11:", + "type": "text" + }, + { + "bbox": [ + 152, + 291, + 253, + 307 + ], + "score": 1.0, + "content": "Compute the gradient of", + "type": "text" + }, + { + "bbox": [ + 253, + 293, + 271, + 305 + ], + "score": 0.92, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 291, + 379, + 307 + ], + "score": 1.0, + "content": "via Eq. 22 and record it as", + "type": "text" + }, + { + "bbox": [ + 379, + 295, + 392, + 304 + ], + "score": 0.76, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 303, + 167, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 122, + 315 + ], + "score": 1.0, + "content": "12:", + "type": "text" + }, + { + "bbox": [ + 138, + 303, + 167, + 315 + ], + "score": 1.0, + "content": "end if", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 315, + 228, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 122, + 327 + ], + "score": 1.0, + "content": "13:", + "type": "text" + }, + { + "bbox": [ + 137, + 315, + 171, + 327 + ], + "score": 1.0, + "content": "Rescale", + "type": "text" + }, + { + "bbox": [ + 172, + 316, + 184, + 326 + ], + "score": 0.84, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 315, + 228, + 327 + ], + "score": 1.0, + "content": "via Eq. 28", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 321, + 380, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 122, + 338 + ], + "score": 1.0, + "content": "14:", + "type": "text" + }, + { + "bbox": [ + 136, + 321, + 212, + 342 + ], + "score": 1.0, + "content": "Update the model", + "type": "text" + }, + { + "bbox": [ + 212, + 326, + 230, + 337 + ], + "score": 0.91, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 321, + 296, + 342 + ], + "score": 1.0, + "content": "via the gradient", + "type": "text" + }, + { + "bbox": [ + 296, + 327, + 308, + 337 + ], + "score": 0.84, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 321, + 353, + 342 + ], + "score": 1.0, + "content": "and obtain", + "type": "text" + }, + { + "bbox": [ + 354, + 326, + 380, + 339 + ], + "score": 0.91, + "content": "M _ { \\theta _ { t + 1 } }", + "type": "inline_equation" + } + ], + "index": 21 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 390, + 506, + 521 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 391, + 429, + 405 + ], + "score": 0.8, + "content": "\\begin{array} { r } { Q _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = { \\mathbb E } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r ( x _ { t } , a _ { t } , x _ { t + 1 } ) \\mid ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 389, + 450, + 407 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 451, + 391, + 506, + 405 + ], + "score": 0.91, + "content": "{ \\cal V } _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a ) =", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 405, + 507, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 193, + 424 + ], + "score": 0.86, + "content": "\\mathbb { E } _ { M _ { \\bar { \\theta } } } \\left[ Q _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 405, + 198, + 424 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 198, + 408, + 207, + 419 + ], + "score": 0.64, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 405, + 507, + 424 + ], + "score": 1.0, + "content": "in Eq. 21 can also be constructed similarly. Although it looks unnecessary in", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 421, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 266, + 436 + ], + "score": 1.0, + "content": "theory since the one-step optimal model", + "type": "text" + }, + { + "bbox": [ + 266, + 423, + 281, + 434 + ], + "score": 0.89, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 421, + 506, + 436 + ], + "score": 1.0, + "content": "is the global optimal model in this setting, the technique", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 433, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 237, + 446 + ], + "score": 1.0, + "content": "is helpful in practice as it makes", + "type": "text" + }, + { + "bbox": [ + 238, + 434, + 246, + 443 + ], + "score": 0.75, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 433, + 505, + 446 + ], + "score": 1.0, + "content": "more sensitive to the compounding effect of one-step prediction", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "errors: we would consider the cumulative effects of prediction errors induced by multi-step transitions", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 102, + 453, + 509, + 485 + ], + "spans": [ + { + "bbox": [ + 102, + 453, + 360, + 485 + ], + "score": 1.0, + "content": "in environments. In particular, to consider the cumulative effmulti-step of transitions in environments, we overwrite function", + "type": "text" + }, + { + "bbox": [ + 360, + 467, + 390, + 480 + ], + "score": 0.93, + "content": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 453, + 402, + 485 + ], + "score": 1.0, + "content": "preas", + "type": "text" + }, + { + "bbox": [ + 403, + 466, + 502, + 481 + ], + "score": 0.9, + "content": "A _ { \\varphi _ { 0 } ^ { \\ast } , \\varphi _ { 1 } ^ { \\ast } } = Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } - V _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 453, + 509, + 485 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 479, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 104, + 479, + 131, + 495 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 480, + 484, + 494 + ], + "score": 0.91, + "content": "\\begin{array} { r } { Q _ { M _ { \\theta _ { \\star } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 0 } ^ { \\star } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , \\bar { M } _ { \\theta _ { t } } \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 479, + 506, + 495 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 108, + 492, + 507, + 510 + ], + "spans": [ + { + "bbox": [ + 108, + 496, + 378, + 509 + ], + "score": 0.83, + "content": "\\begin{array} { r } { V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \\kappa , M _ { \\theta _ { t } } \\right] , } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 492, + 507, + 510 + ], + "score": 1.0, + "content": ". To give an algorithm for single-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 508, + 380, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 303, + 521 + ], + "score": 1.0, + "content": "step environment model learning, we can just set", + "type": "text" + }, + { + "bbox": [ + 303, + 511, + 311, + 520 + ], + "score": 0.83, + "content": "\\gamma", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 508, + 321, + 521 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 322, + 509, + 331, + 520 + ], + "score": 0.85, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 508, + 349, + 521 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 349, + 509, + 358, + 519 + ], + "score": 0.8, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 508, + 380, + 521 + ], + "score": 1.0, + "content": "to 0.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27 + }, + { + "type": "image", + "bbox": [ + 110, + 556, + 503, + 700 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 108, + 338, + 155, + 348 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 336, + 157, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 157, + 349 + ], + "score": 1.0, + "content": "15: end for", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "image_body", + "bbox": [ + 110, + 556, + 503, + 700 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 556, + 503, + 700 + ], + "spans": [ + { + "bbox": [ + 110, + 556, + 503, + 700 + ], + "score": 0.971, + "type": "image", + "image_path": "250eec592a0278e03dcb9c611ad65d4003d340434d2a84445af36733f46969b9.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 110, + 556, + 503, + 604.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 110, + 604.0, + 503, + 652.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 110, + 652.0, + 503, + 700.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 171, + 718, + 438, + 730 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 172, + 717, + 437, + 730 + ], + "spans": [ + { + "bbox": [ + 172, + 717, + 437, + 730 + ], + "score": 1.0, + "content": "Figure 7: Illustration of the workflow of the GALILEO algorithm.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + } + ], + "index": 33 + } + ], + "page_idx": 23, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 254, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 256, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 256, + 96 + ], + "score": 1.0, + "content": "Algorithm 2 GALILEO pseudocode", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 96, + 134, + 106 + ], + "lines": [ + { + "bbox": [ + 107, + 96, + 134, + 106 + ], + "spans": [], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "list", + "bbox": [ + 106, + 106, + 396, + 128 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 397, + 120 + ], + "spans": [ + { + "bbox": [ + 106, + 107, + 128, + 118 + ], + "score": 0.88, + "content": "\\mathcal { D } _ { \\mathrm { r e a l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 105, + 249, + 120 + ], + "score": 1.0, + "content": ": offline dataset sampled from", + "type": "text" + }, + { + "bbox": [ + 249, + 106, + 268, + 119 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 105, + 297, + 120 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 297, + 109, + 304, + 118 + ], + "score": 0.82, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 105, + 397, + 120 + ], + "score": 1.0, + "content": "is the behavior policy;", + "type": "text" + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 116, + 184, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 117, + 127 + ], + "score": 0.78, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 116, + 184, + 129 + ], + "score": 1.0, + "content": ": total iterations;", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 2.5, + "bbox_fs": [ + 106, + 105, + 397, + 129 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 129, + 142, + 139 + ], + "lines": [ + { + "bbox": [ + 105, + 128, + 144, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 144, + 141 + ], + "score": 1.0, + "content": "Process:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "index", + "bbox": [ + 110, + 145, + 506, + 343 + ], + "lines": [ + { + "bbox": [ + 111, + 143, + 342, + 157 + ], + "spans": [ + { + "bbox": [ + 111, + 143, + 250, + 157 + ], + "score": 1.0, + "content": "1: Approximate a behavior policy", + "type": "text" + }, + { + "bbox": [ + 250, + 145, + 257, + 156 + ], + "score": 0.84, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 143, + 342, + 157 + ], + "score": 1.0, + "content": "via behavior cloning", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 154, + 272, + 168 + ], + "spans": [ + { + "bbox": [ + 110, + 154, + 253, + 168 + ], + "score": 1.0, + "content": "2: Initialize an environment model", + "type": "text" + }, + { + "bbox": [ + 253, + 156, + 272, + 167 + ], + "score": 0.9, + "content": "M _ { \\theta _ { 1 } }", + "type": "inline_equation" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 166, + 194, + 178 + ], + "spans": [ + { + "bbox": [ + 110, + 166, + 138, + 178 + ], + "score": 1.0, + "content": "3: for", + "type": "text" + }, + { + "bbox": [ + 138, + 167, + 179, + 177 + ], + "score": 0.89, + "content": "t = 1 : N", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 166, + 194, + 178 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 176, + 349, + 191 + ], + "spans": [ + { + "bbox": [ + 110, + 177, + 122, + 189 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 137, + 176, + 156, + 191 + ], + "score": 1.0, + "content": "Use", + "type": "text" + }, + { + "bbox": [ + 156, + 178, + 164, + 189 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 176, + 248, + 191 + ], + "score": 1.0, + "content": "to generate a dataset", + "type": "text" + }, + { + "bbox": [ + 249, + 178, + 268, + 190 + ], + "score": 0.91, + "content": "\\mathcal { D } _ { \\mathrm { g e n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 176, + 331, + 191 + ], + "score": 1.0, + "content": "with the model", + "type": "text" + }, + { + "bbox": [ + 332, + 177, + 349, + 189 + ], + "score": 0.9, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 185, + 504, + 208 + ], + "spans": [ + { + "bbox": [ + 111, + 191, + 122, + 203 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 135, + 185, + 245, + 208 + ], + "score": 1.0, + "content": "Update the discriminators", + "type": "text" + }, + { + "bbox": [ + 245, + 191, + 264, + 203 + ], + "score": 0.91, + "content": "D _ { \\varphi _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 185, + 282, + 208 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 283, + 191, + 302, + 203 + ], + "score": 0.91, + "content": "D _ { \\varphi _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 185, + 473, + 208 + ], + "score": 1.0, + "content": "via Eq. 26 and Eq. 27 respectively, where", + "type": "text" + }, + { + "bbox": [ + 473, + 189, + 493, + 206 + ], + "score": 0.92, + "content": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\hat { \\mu } }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 185, + 504, + 208 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 122, + 203, + 321, + 219 + ], + "spans": [ + { + "bbox": [ + 122, + 203, + 176, + 219 + ], + "score": 1.0, + "content": "estimated by", + "type": "text" + }, + { + "bbox": [ + 177, + 205, + 198, + 217 + ], + "score": 0.91, + "content": "\\mathcal { D } _ { \\mathrm { g e n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 203, + 216, + 219 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 216, + 204, + 235, + 218 + ], + "score": 0.91, + "content": "\\rho _ { M ^ { * } } ^ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 203, + 299, + 219 + ], + "score": 1.0, + "content": "is estimated by", + "type": "text" + }, + { + "bbox": [ + 299, + 205, + 321, + 216 + ], + "score": 0.89, + "content": "\\mathcal { D } _ { \\mathrm { r e a l } }", + "type": "inline_equation" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 213, + 413, + 231 + ], + "spans": [ + { + "bbox": [ + 111, + 217, + 122, + 228 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 136, + 213, + 169, + 231 + ], + "score": 1.0, + "content": "Update", + "type": "text" + }, + { + "bbox": [ + 170, + 217, + 179, + 227 + ], + "score": 0.86, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 213, + 197, + 231 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 197, + 217, + 207, + 226 + ], + "score": 0.77, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 213, + 329, + 231 + ], + "score": 1.0, + "content": "via Eq. 24 and Eq. 25 through", + "type": "text" + }, + { + "bbox": [ + 329, + 217, + 351, + 228 + ], + "score": 0.88, + "content": "\\mathcal { D } _ { \\mathrm { g e n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 213, + 354, + 231 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 354, + 216, + 373, + 228 + ], + "score": 0.89, + "content": "D _ { \\varphi _ { 0 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 213, + 394, + 231 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 394, + 217, + 413, + 228 + ], + "score": 0.9, + "content": "D _ { \\varphi _ { 1 } }", + "type": "inline_equation" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 226, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 111, + 227, + 122, + 239 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 138, + 226, + 213, + 239 + ], + "score": 1.0, + "content": "Update the model", + "type": "text" + }, + { + "bbox": [ + 214, + 227, + 232, + 239 + ], + "score": 0.92, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 226, + 506, + 239 + ], + "score": 1.0, + "content": "via the first term of Eq. 23, which is implemented with a standard", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 122, + 238, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 122, + 238, + 506, + 250 + ], + "score": 1.0, + "content": "policy gradient method like TRPO (Schulman et al., 2015) or PPO (Schulman et al., 2017).", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 122, + 247, + 249, + 263 + ], + "spans": [ + { + "bbox": [ + 122, + 247, + 249, + 263 + ], + "score": 1.0, + "content": "Record the policy gradient gpg", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 259, + 320, + 275 + ], + "spans": [ + { + "bbox": [ + 110, + 260, + 122, + 272 + ], + "score": 1.0, + "content": "8:", + "type": "text" + }, + { + "bbox": [ + 136, + 259, + 146, + 275 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 146, + 260, + 296, + 273 + ], + "score": 0.88, + "content": "p _ { 0 } < \\mathbb { E } _ { \\mathcal { D } _ { \\mathrm { g e n } } } \\left[ D _ { \\varphi _ { 0 } } ( x _ { t } , \\hat { a _ { t } } , x _ { t + 1 } ) \\right] < p _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 259, + 320, + 275 + ], + "score": 1.0, + "content": "then", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 268, + 468, + 285 + ], + "spans": [ + { + "bbox": [ + 110, + 271, + 122, + 282 + ], + "score": 1.0, + "content": "9:", + "type": "text" + }, + { + "bbox": [ + 151, + 268, + 253, + 285 + ], + "score": 1.0, + "content": "Compute the gradient of", + "type": "text" + }, + { + "bbox": [ + 253, + 272, + 271, + 283 + ], + "score": 0.9, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 268, + 456, + 285 + ], + "score": 1.0, + "content": "via the second term of Eq. 23 and record it as", + "type": "text" + }, + { + "bbox": [ + 456, + 273, + 468, + 282 + ], + "score": 0.81, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 107, + 281, + 158, + 294 + ], + "spans": [ + { + "bbox": [ + 107, + 281, + 122, + 294 + ], + "score": 1.0, + "content": "10:", + "type": "text" + }, + { + "bbox": [ + 137, + 281, + 158, + 294 + ], + "score": 1.0, + "content": "else", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 291, + 392, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 122, + 304 + ], + "score": 1.0, + "content": "11:", + "type": "text" + }, + { + "bbox": [ + 152, + 291, + 253, + 307 + ], + "score": 1.0, + "content": "Compute the gradient of", + "type": "text" + }, + { + "bbox": [ + 253, + 293, + 271, + 305 + ], + "score": 0.92, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 291, + 379, + 307 + ], + "score": 1.0, + "content": "via Eq. 22 and record it as", + "type": "text" + }, + { + "bbox": [ + 379, + 295, + 392, + 304 + ], + "score": 0.76, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 303, + 167, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 122, + 315 + ], + "score": 1.0, + "content": "12:", + "type": "text" + }, + { + "bbox": [ + 138, + 303, + 167, + 315 + ], + "score": 1.0, + "content": "end if", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 315, + 228, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 122, + 327 + ], + "score": 1.0, + "content": "13:", + "type": "text" + }, + { + "bbox": [ + 137, + 315, + 171, + 327 + ], + "score": 1.0, + "content": "Rescale", + "type": "text" + }, + { + "bbox": [ + 172, + 316, + 184, + 326 + ], + "score": 0.84, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 315, + 228, + 327 + ], + "score": 1.0, + "content": "via Eq. 28", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 321, + 380, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 122, + 338 + ], + "score": 1.0, + "content": "14:", + "type": "text" + }, + { + "bbox": [ + 136, + 321, + 212, + 342 + ], + "score": 1.0, + "content": "Update the model", + "type": "text" + }, + { + "bbox": [ + 212, + 326, + 230, + 337 + ], + "score": 0.91, + "content": "M _ { \\theta _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 321, + 296, + 342 + ], + "score": 1.0, + "content": "via the gradient", + "type": "text" + }, + { + "bbox": [ + 296, + 327, + 308, + 337 + ], + "score": 0.84, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 321, + 353, + 342 + ], + "score": 1.0, + "content": "and obtain", + "type": "text" + }, + { + "bbox": [ + 354, + 326, + 380, + 339 + ], + "score": 0.91, + "content": "M _ { \\theta _ { t + 1 } }", + "type": "inline_equation" + } + ], + "index": 21, + "is_list_start_line": true + } + ], + "index": 13, + "bbox_fs": [ + 106, + 143, + 506, + 342 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 390, + 506, + 521 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 391, + 429, + 405 + ], + "score": 0.8, + "content": "\\begin{array} { r } { Q _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = { \\mathbb E } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r ( x _ { t } , a _ { t } , x _ { t + 1 } ) \\mid ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 389, + 450, + 407 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 451, + 391, + 506, + 405 + ], + "score": 0.91, + "content": "{ \\cal V } _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a ) =", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 405, + 507, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 193, + 424 + ], + "score": 0.86, + "content": "\\mathbb { E } _ { M _ { \\bar { \\theta } } } \\left[ Q _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 405, + 198, + 424 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 198, + 408, + 207, + 419 + ], + "score": 0.64, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 405, + 507, + 424 + ], + "score": 1.0, + "content": "in Eq. 21 can also be constructed similarly. Although it looks unnecessary in", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 421, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 266, + 436 + ], + "score": 1.0, + "content": "theory since the one-step optimal model", + "type": "text" + }, + { + "bbox": [ + 266, + 423, + 281, + 434 + ], + "score": 0.89, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 421, + 506, + 436 + ], + "score": 1.0, + "content": "is the global optimal model in this setting, the technique", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 433, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 237, + 446 + ], + "score": 1.0, + "content": "is helpful in practice as it makes", + "type": "text" + }, + { + "bbox": [ + 238, + 434, + 246, + 443 + ], + "score": 0.75, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 433, + 505, + 446 + ], + "score": 1.0, + "content": "more sensitive to the compounding effect of one-step prediction", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "errors: we would consider the cumulative effects of prediction errors induced by multi-step transitions", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 102, + 453, + 509, + 485 + ], + "spans": [ + { + "bbox": [ + 102, + 453, + 360, + 485 + ], + "score": 1.0, + "content": "in environments. In particular, to consider the cumulative effmulti-step of transitions in environments, we overwrite function", + "type": "text" + }, + { + "bbox": [ + 360, + 467, + 390, + 480 + ], + "score": 0.93, + "content": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 453, + 402, + 485 + ], + "score": 1.0, + "content": "preas", + "type": "text" + }, + { + "bbox": [ + 403, + 466, + 502, + 481 + ], + "score": 0.9, + "content": "A _ { \\varphi _ { 0 } ^ { \\ast } , \\varphi _ { 1 } ^ { \\ast } } = Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } - V _ { M _ { \\theta _ { t } } } ^ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 453, + 509, + 485 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 479, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 104, + 479, + 131, + 495 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 480, + 484, + 494 + ], + "score": 0.91, + "content": "\\begin{array} { r } { Q _ { M _ { \\theta _ { \\star } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 0 } ^ { \\star } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , \\bar { M } _ { \\theta _ { t } } \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 479, + 506, + 495 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 108, + 492, + 507, + 510 + ], + "spans": [ + { + "bbox": [ + 108, + 496, + 378, + 509 + ], + "score": 0.83, + "content": "\\begin{array} { r } { V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \\kappa , M _ { \\theta _ { t } } \\right] , } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 492, + 507, + 510 + ], + "score": 1.0, + "content": ". To give an algorithm for single-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 508, + 380, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 303, + 521 + ], + "score": 1.0, + "content": "step environment model learning, we can just set", + "type": "text" + }, + { + "bbox": [ + 303, + 511, + 311, + 520 + ], + "score": 0.83, + "content": "\\gamma", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 508, + 321, + 521 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 322, + 509, + 331, + 520 + ], + "score": 0.85, + "content": "Q", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 508, + 349, + 521 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 349, + 509, + 358, + 519 + ], + "score": 0.8, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 508, + 380, + 521 + ], + "score": 1.0, + "content": "to 0.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27, + "bbox_fs": [ + 102, + 389, + 509, + 521 + ] + }, + { + "type": "image", + "bbox": [ + 110, + 556, + 503, + 700 + ], + "blocks": [ + { + "type": "image_caption", + "bbox": [ + 108, + 338, + 155, + 348 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 336, + 157, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 157, + 349 + ], + "score": 1.0, + "content": "15: end for", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "image_body", + "bbox": [ + 110, + 556, + 503, + 700 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 556, + 503, + 700 + ], + "spans": [ + { + "bbox": [ + 110, + 556, + 503, + 700 + ], + "score": 0.971, + "type": "image", + "image_path": "250eec592a0278e03dcb9c611ad65d4003d340434d2a84445af36733f46969b9.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 110, + 556, + 503, + 604.0 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 110, + 604.0, + 503, + 652.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 110, + 652.0, + 503, + 700.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 171, + 718, + 438, + 730 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 172, + 717, + 437, + 730 + ], + "spans": [ + { + "bbox": [ + 172, + 717, + 437, + 730 + ], + "score": 1.0, + "content": "Figure 7: Illustration of the workflow of the GALILEO algorithm.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + } + ], + "index": 33 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 506, + 97 + ], + "score": 1.0, + "content": "By adopting the above implementation techniques, we convert the objective into the following", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 91, + 157, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 157, + 105 + ], + "score": 1.0, + "content": "formulation", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 112, + 527, + 129 + ], + "lines": [ + { + "bbox": [ + 124, + 112, + 527, + 129 + ], + "spans": [ + { + "bbox": [ + 124, + 112, + 527, + 129 + ], + "score": 0.83, + "content": "t + 1 = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\star } } \\left[ A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\star } } \\left[ \\left( H _ { M ^ { \\star } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] .", + "type": "interline_equation", + "image_path": "6a73c9667b1b03e9f317568a2a21065bca1ae266f00327e9f119736f1381d336.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 124, + 112, + 527, + 129 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 157, + 120, + 166 + ], + "lines": [ + { + "bbox": [ + 105, + 156, + 123, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 123, + 168 + ], + "score": 1.0, + "content": "s.t.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 146, + 540, + 258 + ], + "lines": [ + { + "bbox": [ + 144, + 146, + 540, + 258 + ], + "spans": [ + { + "bbox": [ + 144, + 146, + 540, + 258 + ], + "score": 0.87, + "content": "\\begin{array} { r l } & { Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb E [ \\displaystyle \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 0 } ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } ] \\qquad ( 2 4 ) } \\\\ & { V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb E [ \\displaystyle \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \\kappa , M _ { \\theta _ { t } } ] } \\\\ & { \\varphi _ { 0 } ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 0 } ^ { * } } \\mathbb E _ { \\rho _ { M ^ { * } } ^ { \\kappa } , \\textbf { u } } [ \\log D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) ] + \\mathbb E _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\textbf { u } } [ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } ) ] ] } \\\\ & { \\varphi _ { 1 } ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 1 } ^ { * } } \\mathbb E _ { \\rho _ { M ^ { * } } ^ { \\kappa } , \\textbf { u } } [ \\log D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ] + \\mathbb E _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\textbf { u } } [ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) ] , } \\end{array}", + "type": "interline_equation", + "image_path": "c0d58b24005654f7799b4139385c1b768d7080cb3287d2e0e505d2fca3235f30.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 144, + 146, + 540, + 183.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 144, + 183.33333333333334, + 540, + 220.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 144, + 220.66666666666669, + 540, + 258.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 276, + 506, + 420 + ], + "lines": [ + { + "bbox": [ + 108, + 274, + 507, + 291 + ], + "spans": [ + { + "bbox": [ + 108, + 274, + 137, + 291 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 137, + 277, + 328, + 291 + ], + "score": 0.92, + "content": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , \\underline { { a } } , x ^ { \\prime } ) = Q _ { M _ { \\theta } } ^ { \\kappa } ( x , a , \\underline { { x } } ^ { \\prime } ) - V _ { M _ { \\theta } } ^ { \\kappa } ( x , \\underline { { a } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 274, + 507, + 291 + ], + "score": 1.0, + "content": ". In practice, GALILEO optimizes the first", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "score": 1.0, + "content": "term of Eq. 23 with conservative policy gradient algorithms (e.g., PPO (Schulman et al., 2017)", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 298, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 506, + 311 + ], + "score": 1.0, + "content": "or TRPO (Schulman et al., 2015)) to avoid unreliable gradients for model improvements. Eq. 26", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 310, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 505, + 322 + ], + "score": 1.0, + "content": "and Eq. 27 are optimized with supervised learning. The second term of Eq. 23 is optimized with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 294, + 336 + ], + "score": 1.0, + "content": "supervised learning with a re-weighting term", + "type": "text" + }, + { + "bbox": [ + 294, + 321, + 366, + 334 + ], + "score": 0.92, + "content": "- A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } + H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 320, + 397, + 336 + ], + "score": 1.0, + "content": ". Since", + "type": "text" + }, + { + "bbox": [ + 397, + 321, + 419, + 332 + ], + "score": 0.91, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 320, + 506, + 336 + ], + "score": 1.0, + "content": "is unknown, we use", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 107, + 331, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 107, + 332, + 128, + 344 + ], + "score": 0.9, + "content": "H _ { M _ { \\theta } }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 331, + 506, + 344 + ], + "score": 1.0, + "content": "to estimate it. When the mean output probability of a batch of data is larger than 0.6 or", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 343, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 354 + ], + "score": 1.0, + "content": "small than 0.4, we replace the second term of Eq. 23 with a standard supervised learning in Eq. 22.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "Besides, unreliable gradients also exist in the process of optimizing the second term of Eq. 23. In", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 364, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 506, + 377 + ], + "score": 1.0, + "content": "our implementation, we use the scale of policy gradients to constrain the gradients of the second", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 306, + 388 + ], + "score": 1.0, + "content": "term of Eq. 23. In particular, we first compute the", + "type": "text" + }, + { + "bbox": [ + 307, + 376, + 315, + 387 + ], + "score": 0.87, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "-norm of the gradient of the first term of Eq. 23", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 386, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 317, + 400 + ], + "score": 1.0, + "content": "via conservative policy gradient algorithms, named", + "type": "text" + }, + { + "bbox": [ + 318, + 387, + 348, + 399 + ], + "score": 0.91, + "content": "| | g _ { \\mathrm { p g } } | | _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 386, + 443, + 400 + ], + "score": 1.0, + "content": ". Then we compute the", + "type": "text" + }, + { + "bbox": [ + 444, + 387, + 452, + 398 + ], + "score": 0.87, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 386, + 506, + 400 + ], + "score": 1.0, + "content": "-norm of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 285, + 411 + ], + "score": 1.0, + "content": "gradient of the second term of Eq. 23, name", + "type": "text" + }, + { + "bbox": [ + 286, + 397, + 312, + 410 + ], + "score": 0.91, + "content": "| | g _ { \\mathrm { s l } } | | _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 397, + 506, + 411 + ], + "score": 1.0, + "content": ". Finally, we rescale the gradients of the second", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 407, + 153, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 127, + 422 + ], + "score": 1.0, + "content": "term", + "type": "text" + }, + { + "bbox": [ + 127, + 410, + 139, + 420 + ], + "score": 0.84, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 407, + 153, + 422 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 429, + 372, + 456 + ], + "lines": [ + { + "bbox": [ + 239, + 429, + 372, + 456 + ], + "spans": [ + { + "bbox": [ + 239, + 429, + 372, + 456 + ], + "score": 0.94, + "content": "g _ { \\mathrm { s l } } g _ { \\mathrm { s l } } \\frac { \\vert \\vert g _ { \\mathrm { p g } } \\vert \\vert _ { 2 } } { \\operatorname* { m a x } \\{ \\vert \\vert g _ { \\mathrm { p g } } \\vert \\vert _ { 2 } , \\vert \\vert g _ { \\mathrm { s l } } \\vert \\vert _ { 2 } \\} } .", + "type": "interline_equation", + "image_path": "b8ea05dedc4d1c7fb1f382c89b84db33367cdb5848c41c2fddafbb413d0abf1c.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 239, + 429, + 372, + 442.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 239, + 442.5, + 372, + 456.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 464, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 463, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 506, + 478 + ], + "score": 1.0, + "content": "For each iteration, Eq. 23, Eq. 26, and Eq. 27 are trained with certain steps (See Tab. 5) following", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "the same framework as GAIL. Based on the above techniques, we summarize the pseudocode of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 486, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 221, + 499 + ], + "score": 1.0, + "content": "GALILEO in Alg. 2, where", + "type": "text" + }, + { + "bbox": [ + 221, + 488, + 232, + 498 + ], + "score": 0.85, + "content": "p _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 486, + 250, + 499 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 251, + 488, + 262, + 498 + ], + "score": 0.86, + "content": "p _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 486, + 505, + 499 + ], + "score": 1.0, + "content": "are set to 0.4 and 0.6 in all of our experiments. The overall", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 497, + 232, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 232, + 510 + ], + "score": 1.0, + "content": "architecture is shown in Fig. 7.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 513, + 498, + 584, + 509 + ], + "lines": [ + { + "bbox": [ + 513, + 497, + 586, + 510 + ], + "spans": [ + { + "bbox": [ + 513, + 497, + 586, + 510 + ], + "score": 1.0, + "content": "to Reviewer bcJS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "title", + "bbox": [ + 106, + 527, + 389, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 526, + 390, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 390, + 540 + ], + "score": 1.0, + "content": "E.2 CONNECTION WITH PREVIOUS ADVERSARIAL ALGORITHMS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 108, + 549, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 548, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 548, + 505, + 562 + ], + "score": 1.0, + "content": "Standard GAN (Goodfellow et al., 2014) can be regarded as a partial implementation including the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 560, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 573 + ], + "score": 1.0, + "content": "first term of Eq. 23 and Eq. 26 by degrading them into the single-step scenario. In the context of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 571, + 253, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 253, + 583 + ], + "score": 1.0, + "content": "GALILEO, the objective of GAN is", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 591, + 465, + 631 + ], + "lines": [ + { + "bbox": [ + 145, + 591, + 465, + 631 + ], + "spans": [ + { + "bbox": [ + 145, + 591, + 465, + 631 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { \\theta _ { t + 1 } = \\underset { \\theta } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } \\\\ & { s . t . \\quad \\varphi * = \\underset { \\varphi } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ \\log ( 1 - D _ { \\varphi } ( x , a , x ^ { \\prime } ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "d97b5374a96162d6f13436b835ca976cd3c134e79756321353bb6471be8747ec.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 145, + 591, + 465, + 604.3333333333334 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 145, + 604.3333333333334, + 465, + 617.6666666666667 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 145, + 617.6666666666667, + 465, + 631.0000000000001 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 639, + 505, + 707 + ], + "lines": [ + { + "bbox": [ + 104, + 638, + 503, + 655 + ], + "spans": [ + { + "bbox": [ + 104, + 638, + 136, + 654 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 137, + 640, + 287, + 653 + ], + "score": 0.92, + "content": "\\begin{array} { r c l } { { A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) } } & { { = } } & { { \\log D _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 638, + 429, + 654 + ], + "score": 1.0, + "content": ". In the single-step scenario,", + "type": "text" + }, + { + "bbox": [ + 429, + 640, + 503, + 655 + ], + "score": 0.88, + "content": "\\begin{array} { r l } { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } & { { } = } \\end{array}", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 652, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 210, + 667 + ], + "score": 0.9, + "content": "\\rho _ { 0 } ( x ) \\kappa ( a | x ) M _ { \\theta _ { t } } ( x ^ { \\prime } | a , x )", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 652, + 277, + 670 + ], + "score": 1.0, + "content": ". The term", + "type": "text" + }, + { + "bbox": [ + 277, + 653, + 430, + 670 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 652, + 506, + 670 + ], + "score": 1.0, + "content": "can convert to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 107, + 667, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 107, + 669, + 207, + 684 + ], + "score": 0.86, + "content": "\\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 667, + 348, + 683 + ], + "score": 1.0, + "content": "by replacing the gradient of", + "type": "text" + }, + { + "bbox": [ + 348, + 669, + 479, + 682 + ], + "score": 0.89, + "content": "M _ { \\theta _ { t } } ( x ^ { \\prime } | x , a ) \\nabla _ { \\theta } \\log M _ { \\theta } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 667, + 506, + 683 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 682, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 107, + 684, + 169, + 695 + ], + "score": 0.89, + "content": "\\nabla _ { \\boldsymbol { \\theta } } M _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } ^ { \\prime } | \\boldsymbol { x } , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 682, + 506, + 696 + ], + "score": 1.0, + "content": "(Sutton & Barto, 1998). Previous algorithms like GANITE (Yoon et al., 2018)", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 694, + 481, + 707 + ], + "spans": [ + { + "bbox": [ + 106, + 694, + 481, + 707 + ], + "score": 1.0, + "content": "and SCIGAN (Bica et al., 2020) can be regarded as variants of the above training framework.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 105, + 710, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 711, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 461, + 722 + ], + "score": 1.0, + "content": "The first term of Eq. 23 and Eq. 26 are similar to the objective of GAIL by regarding", + "type": "text" + }, + { + "bbox": [ + 461, + 711, + 477, + 721 + ], + "score": 0.88, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 711, + 505, + 722 + ], + "score": 1.0, + "content": "as the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 104, + 720, + 203, + 733 + ], + "score": 1.0, + "content": "“policy” to imitate and", + "type": "text" + }, + { + "bbox": [ + 203, + 721, + 210, + 732 + ], + "score": 0.84, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 720, + 506, + 733 + ], + "score": 1.0, + "content": "as the “environment” to collect data. In the context of GALILEO, the", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + } + ], + "page_idx": 24, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 16, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 506, + 97 + ], + "score": 1.0, + "content": "By adopting the above implementation techniques, we convert the objective into the following", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 91, + 157, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 157, + 105 + ], + "score": 1.0, + "content": "formulation", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 80, + 506, + 105 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 112, + 527, + 129 + ], + "lines": [ + { + "bbox": [ + 124, + 112, + 527, + 129 + ], + "spans": [ + { + "bbox": [ + 124, + 112, + 527, + 129 + ], + "score": 0.83, + "content": "t + 1 = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\star } } \\left[ A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\star } } \\left[ \\left( H _ { M ^ { \\star } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] .", + "type": "interline_equation", + "image_path": "6a73c9667b1b03e9f317568a2a21065bca1ae266f00327e9f119736f1381d336.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 124, + 112, + 527, + 129 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 157, + 120, + 166 + ], + "lines": [ + { + "bbox": [ + 105, + 156, + 123, + 168 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 123, + 168 + ], + "score": 1.0, + "content": "s.t.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 156, + 123, + 168 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 144, + 146, + 540, + 258 + ], + "lines": [ + { + "bbox": [ + 144, + 146, + 540, + 258 + ], + "spans": [ + { + "bbox": [ + 144, + 146, + 540, + 258 + ], + "score": 0.87, + "content": "\\begin{array} { r l } & { Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb E [ \\displaystyle \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 0 } ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } ] \\qquad ( 2 4 ) } \\\\ & { V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb E [ \\displaystyle \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \\kappa , M _ { \\theta _ { t } } ] } \\\\ & { \\varphi _ { 0 } ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 0 } ^ { * } } \\mathbb E _ { \\rho _ { M ^ { * } } ^ { \\kappa } , \\textbf { u } } [ \\log D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) ] + \\mathbb E _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\textbf { u } } [ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } ) ] ] } \\\\ & { \\varphi _ { 1 } ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 1 } ^ { * } } \\mathbb E _ { \\rho _ { M ^ { * } } ^ { \\kappa } , \\textbf { u } } [ \\log D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ] + \\mathbb E _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\textbf { u } } [ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) ] , } \\end{array}", + "type": "interline_equation", + "image_path": "c0d58b24005654f7799b4139385c1b768d7080cb3287d2e0e505d2fca3235f30.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 144, + 146, + 540, + 183.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 144, + 183.33333333333334, + 540, + 220.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 144, + 220.66666666666669, + 540, + 258.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 276, + 506, + 420 + ], + "lines": [ + { + "bbox": [ + 108, + 274, + 507, + 291 + ], + "spans": [ + { + "bbox": [ + 108, + 274, + 137, + 291 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 137, + 277, + 328, + 291 + ], + "score": 0.92, + "content": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , \\underline { { a } } , x ^ { \\prime } ) = Q _ { M _ { \\theta } } ^ { \\kappa } ( x , a , \\underline { { x } } ^ { \\prime } ) - V _ { M _ { \\theta } } ^ { \\kappa } ( x , \\underline { { a } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 274, + 507, + 291 + ], + "score": 1.0, + "content": ". In practice, GALILEO optimizes the first", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "score": 1.0, + "content": "term of Eq. 23 with conservative policy gradient algorithms (e.g., PPO (Schulman et al., 2017)", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 298, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 506, + 311 + ], + "score": 1.0, + "content": "or TRPO (Schulman et al., 2015)) to avoid unreliable gradients for model improvements. Eq. 26", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 310, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 505, + 322 + ], + "score": 1.0, + "content": "and Eq. 27 are optimized with supervised learning. The second term of Eq. 23 is optimized with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 320, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 294, + 336 + ], + "score": 1.0, + "content": "supervised learning with a re-weighting term", + "type": "text" + }, + { + "bbox": [ + 294, + 321, + 366, + 334 + ], + "score": 0.92, + "content": "- A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } + H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 320, + 397, + 336 + ], + "score": 1.0, + "content": ". Since", + "type": "text" + }, + { + "bbox": [ + 397, + 321, + 419, + 332 + ], + "score": 0.91, + "content": "H _ { M ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 320, + 506, + 336 + ], + "score": 1.0, + "content": "is unknown, we use", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 107, + 331, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 107, + 332, + 128, + 344 + ], + "score": 0.9, + "content": "H _ { M _ { \\theta } }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 331, + 506, + 344 + ], + "score": 1.0, + "content": "to estimate it. When the mean output probability of a batch of data is larger than 0.6 or", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 343, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 354 + ], + "score": 1.0, + "content": "small than 0.4, we replace the second term of Eq. 23 with a standard supervised learning in Eq. 22.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "Besides, unreliable gradients also exist in the process of optimizing the second term of Eq. 23. In", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 364, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 506, + 377 + ], + "score": 1.0, + "content": "our implementation, we use the scale of policy gradients to constrain the gradients of the second", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 375, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 306, + 388 + ], + "score": 1.0, + "content": "term of Eq. 23. In particular, we first compute the", + "type": "text" + }, + { + "bbox": [ + 307, + 376, + 315, + 387 + ], + "score": 0.87, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 375, + 505, + 388 + ], + "score": 1.0, + "content": "-norm of the gradient of the first term of Eq. 23", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 386, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 317, + 400 + ], + "score": 1.0, + "content": "via conservative policy gradient algorithms, named", + "type": "text" + }, + { + "bbox": [ + 318, + 387, + 348, + 399 + ], + "score": 0.91, + "content": "| | g _ { \\mathrm { p g } } | | _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 386, + 443, + 400 + ], + "score": 1.0, + "content": ". Then we compute the", + "type": "text" + }, + { + "bbox": [ + 444, + 387, + 452, + 398 + ], + "score": 0.87, + "content": "l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 386, + 506, + 400 + ], + "score": 1.0, + "content": "-norm of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 397, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 285, + 411 + ], + "score": 1.0, + "content": "gradient of the second term of Eq. 23, name", + "type": "text" + }, + { + "bbox": [ + 286, + 397, + 312, + 410 + ], + "score": 0.91, + "content": "| | g _ { \\mathrm { s l } } | | _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 397, + 506, + 411 + ], + "score": 1.0, + "content": ". Finally, we rescale the gradients of the second", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 407, + 153, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 127, + 422 + ], + "score": 1.0, + "content": "term", + "type": "text" + }, + { + "bbox": [ + 127, + 410, + 139, + 420 + ], + "score": 0.84, + "content": "g _ { \\mathrm { s l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 407, + 153, + 422 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 274, + 507, + 422 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 429, + 372, + 456 + ], + "lines": [ + { + "bbox": [ + 239, + 429, + 372, + 456 + ], + "spans": [ + { + "bbox": [ + 239, + 429, + 372, + 456 + ], + "score": 0.94, + "content": "g _ { \\mathrm { s l } } g _ { \\mathrm { s l } } \\frac { \\vert \\vert g _ { \\mathrm { p g } } \\vert \\vert _ { 2 } } { \\operatorname* { m a x } \\{ \\vert \\vert g _ { \\mathrm { p g } } \\vert \\vert _ { 2 } , \\vert \\vert g _ { \\mathrm { s l } } \\vert \\vert _ { 2 } \\} } .", + "type": "interline_equation", + "image_path": "b8ea05dedc4d1c7fb1f382c89b84db33367cdb5848c41c2fddafbb413d0abf1c.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 239, + 429, + 372, + 442.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 239, + 442.5, + 372, + 456.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 464, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 463, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 506, + 478 + ], + "score": 1.0, + "content": "For each iteration, Eq. 23, Eq. 26, and Eq. 27 are trained with certain steps (See Tab. 5) following", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 476, + 505, + 488 + ], + "score": 1.0, + "content": "the same framework as GAIL. Based on the above techniques, we summarize the pseudocode of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 486, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 221, + 499 + ], + "score": 1.0, + "content": "GALILEO in Alg. 2, where", + "type": "text" + }, + { + "bbox": [ + 221, + 488, + 232, + 498 + ], + "score": 0.85, + "content": "p _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 486, + 250, + 499 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 251, + 488, + 262, + 498 + ], + "score": 0.86, + "content": "p _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 486, + 505, + 499 + ], + "score": 1.0, + "content": "are set to 0.4 and 0.6 in all of our experiments. The overall", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 497, + 232, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 232, + 510 + ], + "score": 1.0, + "content": "architecture is shown in Fig. 7.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 463, + 506, + 510 + ] + }, + { + "type": "text", + "bbox": [ + 513, + 498, + 584, + 509 + ], + "lines": [ + { + "bbox": [ + 513, + 497, + 586, + 510 + ], + "spans": [ + { + "bbox": [ + 513, + 497, + 586, + 510 + ], + "score": 1.0, + "content": "to Reviewer bcJS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 513, + 497, + 586, + 510 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 527, + 389, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 526, + 390, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 390, + 540 + ], + "score": 1.0, + "content": "E.2 CONNECTION WITH PREVIOUS ADVERSARIAL ALGORITHMS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 108, + 549, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 548, + 505, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 548, + 505, + 562 + ], + "score": 1.0, + "content": "Standard GAN (Goodfellow et al., 2014) can be regarded as a partial implementation including the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 560, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 506, + 573 + ], + "score": 1.0, + "content": "first term of Eq. 23 and Eq. 26 by degrading them into the single-step scenario. In the context of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 571, + 253, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 571, + 253, + 583 + ], + "score": 1.0, + "content": "GALILEO, the objective of GAN is", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 548, + 506, + 583 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 591, + 465, + 631 + ], + "lines": [ + { + "bbox": [ + 145, + 591, + 465, + 631 + ], + "spans": [ + { + "bbox": [ + 145, + 591, + 465, + 631 + ], + "score": 0.9, + "content": "\\begin{array} { r l } & { \\theta _ { t + 1 } = \\underset { \\theta } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } \\\\ & { s . t . \\quad \\varphi * = \\underset { \\varphi } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ \\log ( 1 - D _ { \\varphi } ( x , a , x ^ { \\prime } ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "d97b5374a96162d6f13436b835ca976cd3c134e79756321353bb6471be8747ec.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 145, + 591, + 465, + 604.3333333333334 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 145, + 604.3333333333334, + 465, + 617.6666666666667 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 145, + 617.6666666666667, + 465, + 631.0000000000001 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 639, + 505, + 707 + ], + "lines": [ + { + "bbox": [ + 104, + 638, + 503, + 655 + ], + "spans": [ + { + "bbox": [ + 104, + 638, + 136, + 654 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 137, + 640, + 287, + 653 + ], + "score": 0.92, + "content": "\\begin{array} { r c l } { { A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) } } & { { = } } & { { \\log D _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 638, + 429, + 654 + ], + "score": 1.0, + "content": ". In the single-step scenario,", + "type": "text" + }, + { + "bbox": [ + 429, + 640, + 503, + 655 + ], + "score": 0.88, + "content": "\\begin{array} { r l } { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } & { { } = } \\end{array}", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 652, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 210, + 667 + ], + "score": 0.9, + "content": "\\rho _ { 0 } ( x ) \\kappa ( a | x ) M _ { \\theta _ { t } } ( x ^ { \\prime } | a , x )", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 652, + 277, + 670 + ], + "score": 1.0, + "content": ". The term", + "type": "text" + }, + { + "bbox": [ + 277, + 653, + 430, + 670 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 652, + 506, + 670 + ], + "score": 1.0, + "content": "can convert to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 107, + 667, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 107, + 669, + 207, + 684 + ], + "score": 0.86, + "content": "\\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 667, + 348, + 683 + ], + "score": 1.0, + "content": "by replacing the gradient of", + "type": "text" + }, + { + "bbox": [ + 348, + 669, + 479, + 682 + ], + "score": 0.89, + "content": "M _ { \\theta _ { t } } ( x ^ { \\prime } | x , a ) \\nabla _ { \\theta } \\log M _ { \\theta } ( x ^ { \\prime } | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 667, + 506, + 683 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 682, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 107, + 684, + 169, + 695 + ], + "score": 0.89, + "content": "\\nabla _ { \\boldsymbol { \\theta } } M _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } ^ { \\prime } | \\boldsymbol { x } , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 682, + 506, + 696 + ], + "score": 1.0, + "content": "(Sutton & Barto, 1998). Previous algorithms like GANITE (Yoon et al., 2018)", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 694, + 481, + 707 + ], + "spans": [ + { + "bbox": [ + 106, + 694, + 481, + 707 + ], + "score": 1.0, + "content": "and SCIGAN (Bica et al., 2020) can be regarded as variants of the above training framework.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36, + "bbox_fs": [ + 104, + 638, + 506, + 707 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 710, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 711, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 461, + 722 + ], + "score": 1.0, + "content": "The first term of Eq. 23 and Eq. 26 are similar to the objective of GAIL by regarding", + "type": "text" + }, + { + "bbox": [ + 461, + 711, + 477, + 721 + ], + "score": 0.88, + "content": "M _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 711, + 505, + 722 + ], + "score": 1.0, + "content": "as the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 104, + 720, + 203, + 733 + ], + "score": 1.0, + "content": "“policy” to imitate and", + "type": "text" + }, + { + "bbox": [ + 203, + 721, + 210, + 732 + ], + "score": 0.84, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 720, + 506, + 733 + ], + "score": 1.0, + "content": "as the “environment” to collect data. In the context of GALILEO, the", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 104, + 711, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 193, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 194, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 194, + 96 + ], + "score": 1.0, + "content": "objective of GAIL is:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 101, + 478, + 171 + ], + "lines": [ + { + "bbox": [ + 133, + 101, + 478, + 171 + ], + "spans": [ + { + "bbox": [ + 133, + 101, + 478, + 171 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\displaystyle \\theta _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } & { } \\\\ { \\displaystyle s . t . } & { Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } \\right] } \\\\ { \\displaystyle \\varphi ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ \\log ( 1 - D _ { \\varphi } ( x , a , x ^ { \\prime } ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "dcc7cb0b1e1425d35c6360b351bcb8b8deac7e8830c1f531755e2409a6b68632.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 133, + 101, + 478, + 124.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 133, + 124.33333333333333, + 478, + 147.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 133, + 147.66666666666666, + 478, + 171.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 177, + 488, + 193 + ], + "lines": [ + { + "bbox": [ + 108, + 177, + 486, + 194 + ], + "spans": [ + { + "bbox": [ + 108, + 177, + 136, + 194 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 179, + 313, + 192 + ], + "score": 0.93, + "content": "A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) = Q _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) - V _ { M _ { \\theta } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 177, + 331, + 194 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 332, + 179, + 486, + 194 + ], + "score": 0.9, + "content": "V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb { E } _ { M _ { \\theta _ { t } } ( x , a ) } \\left[ Q ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\right] ,", + "type": "inline_equation" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "title", + "bbox": [ + 108, + 210, + 281, + 223 + ], + "lines": [ + { + "bbox": [ + 105, + 210, + 283, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 283, + 225 + ], + "score": 1.0, + "content": "F ADDITIONAL RELATED WORK", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 238, + 505, + 540 + ], + "lines": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "Our primitive objective is inspired by weighted empirical risk minimization (WERM) based on", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "inverse propensity score (IPS). WERM is originally proposed to solve the generalization problem of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 260, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 505, + 271 + ], + "score": 1.0, + "content": "domain adaptation in machine learning literature. For instance, we would like to train a predictor", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 270, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 107, + 272, + 140, + 283 + ], + "score": 0.94, + "content": "M ( y | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 270, + 259, + 283 + ], + "score": 1.0, + "content": "in a domain with distribution", + "type": "text" + }, + { + "bbox": [ + 259, + 272, + 297, + 283 + ], + "score": 0.93, + "content": "P _ { \\mathrm { t r a i n } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 270, + 505, + 283 + ], + "score": 1.0, + "content": "to minimize the prediction risks in the domain with", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 280, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 155, + 295 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 156, + 283, + 189, + 294 + ], + "score": 0.92, + "content": "P _ { \\mathrm { { t e s t } } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 280, + 221, + 295 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 221, + 284, + 274, + 293 + ], + "score": 0.92, + "content": "P _ { \\mathrm { t e s t } } \\neq P _ { \\mathrm { t e s t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 280, + 506, + 295 + ], + "score": 1.0, + "content": ". To solve the problem, we can train a weighted objective", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 128, + 291, + 506, + 310 + ], + "spans": [ + { + "bbox": [ + 128, + 294, + 282, + 310 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\operatorname* { m a x } _ { M } \\mathbb { E } _ { x \\sim P _ { \\mathrm { t r a i n } } } [ \\frac { P _ { \\mathrm { t e s t } } ( x ) } { P _ { \\mathrm { t r a i n } } ( x ) } \\log M ( y | x ) ] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 291, + 506, + 309 + ], + "score": 1.0, + "content": ", which is called weighted empirical risk minimization", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "methods (Ben-David et al., 2006; 2010; Cortes et al., 2010; Byrd & Lipton, 2019; Quinonero-Candela", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "et al., 2008). These results have been extended and applied to causal inference, where the predictor", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 329, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 342 + ], + "score": 1.0, + "content": "is required to be generalized from the data distribution in observational studies (source domain)", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 341, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 505, + 352 + ], + "score": 1.0, + "content": "to the data distribution in randomized controlled trials (target domain) (Shimodaira, 2000; Assaad", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 350, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 209, + 365 + ], + "score": 1.0, + "content": "et al., 2021; Hassanpour", + "type": "text" + }, + { + "bbox": [ + 209, + 354, + 217, + 362 + ], + "score": 0.63, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 350, + 506, + 365 + ], + "score": 1.0, + "content": "Greiner, 2019; Jung et al., 2020; Johansson et al., 2018). In this case,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 363, + 244, + 375 + ], + "score": 1.0, + "content": "the input features include a state", + "type": "text" + }, + { + "bbox": [ + 244, + 368, + 250, + 373 + ], + "score": 0.87, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 363, + 388, + 375 + ], + "score": 1.0, + "content": "(a.k.a. covariates) and an action", + "type": "text" + }, + { + "bbox": [ + 388, + 368, + 394, + 372 + ], + "score": 0.87, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "(a.k.a. treatment variable)", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 373, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 383, + 386 + ], + "score": 1.0, + "content": "which is sampled from a policy. We often assume the distribution of", + "type": "text" + }, + { + "bbox": [ + 383, + 379, + 389, + 383 + ], + "score": 0.83, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 373, + 394, + 386 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 394, + 375, + 415, + 386 + ], + "score": 0.92, + "content": "P ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 373, + 505, + 386 + ], + "score": 1.0, + "content": "is consistent between", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 103, + 383, + 506, + 404 + ], + "spans": [ + { + "bbox": [ + 103, + 384, + 413, + 404 + ], + "score": 1.0, + "content": "the source domain and the test domain, then we have Ptest(x)Ptrain(x) = P (x)β(a|x)P (x)µ(a|x)", + "type": "text" + }, + { + "bbox": [ + 449, + 383, + 480, + 403 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 480, + 393, + 486, + 399 + ], + "score": 0.87, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 383, + 506, + 403 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 400, + 506, + 413 + ], + "spans": [ + { + "bbox": [ + 107, + 402, + 114, + 412 + ], + "score": 0.86, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 400, + 506, + 413 + ], + "score": 1.0, + "content": "are the policies in source and target domains respectively. In Shimodaira (2000); Assaad et al.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 411, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 190, + 424 + ], + "score": 1.0, + "content": "(2021); Hassanpour", + "type": "text" + }, + { + "bbox": [ + 190, + 414, + 199, + 421 + ], + "score": 0.64, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 411, + 506, + 424 + ], + "score": 1.0, + "content": "Greiner (2019), the policy in randomized controlled trials is modeled as a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 421, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 194, + 438 + ], + "score": 1.0, + "content": "uniform policy, then", + "type": "text" + }, + { + "bbox": [ + 195, + 424, + 367, + 439 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\frac { P _ { \\mathrm { t e s t } } ( x ) } { P _ { \\mathrm { t r a i n } } ( x ) } = \\frac { P ( x ) \\beta ( a | x ) } { P ( x ) \\mu ( a | x ) } = \\frac { \\beta ( a | x ) } { \\mu ( a | x ) } \\propto \\frac { 1 } { \\mu ( a | x ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 427, + 374, + 440 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 374, + 425, + 399, + 440 + ], + "score": 0.79, + "content": "{ \\frac { 1 } { \\mu ( a | x ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 425, + 505, + 437 + ], + "score": 1.0, + "content": "is also known as inverse", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "score": 1.0, + "content": "propensity score (IPS). In Johansson et al. (2018), it assumes that the policy in the target domain is", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 448, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 104, + 448, + 162, + 463 + ], + "score": 1.0, + "content": "predefined as", + "type": "text" + }, + { + "bbox": [ + 163, + 452, + 190, + 462 + ], + "score": 0.94, + "content": "\\beta ( a | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 448, + 386, + 463 + ], + "score": 1.0, + "content": "before environment model learning, then it uses", + "type": "text" + }, + { + "bbox": [ + 387, + 450, + 393, + 465 + ], + "score": 0.87, + "content": "\\frac { \\beta } { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 448, + 506, + 463 + ], + "score": 1.0, + "content": "as the IPS. The differences", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 463, + 504, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 504, + 474 + ], + "score": 1.0, + "content": "between AWRM and previous works are fallen in two aspects: (1) We consider the distribution-shift", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 474, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 505, + 485 + ], + "score": 1.0, + "content": "problem in the sequential decision-making scenario. In this scenario, we not only consider the action", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 485, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 324, + 496 + ], + "score": 1.0, + "content": "distribution mismatching between the behavior policy", + "type": "text" + }, + { + "bbox": [ + 325, + 489, + 331, + 496 + ], + "score": 0.88, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 485, + 446, + 496 + ], + "score": 1.0, + "content": "and the policy to evaluation", + "type": "text" + }, + { + "bbox": [ + 446, + 487, + 452, + 496 + ], + "score": 0.88, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 485, + 505, + 496 + ], + "score": 1.0, + "content": ", but also the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 495, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 508 + ], + "score": 1.0, + "content": "follow-up effects of policies to the state distribution; (2) For faithful offline policy optimization, we", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 506, + 504, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 504, + 517 + ], + "score": 1.0, + "content": "require the environment model to have generalization ability in numerous different policies. The", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 517, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 403, + 529 + ], + "score": 1.0, + "content": "objective of AWRM is proposed to guarantee the generalization ability of", + "type": "text" + }, + { + "bbox": [ + 403, + 520, + 415, + 527 + ], + "score": 0.9, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 517, + 506, + 529 + ], + "score": 1.0, + "content": "in numerous different", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 527, + 250, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 250, + 541 + ], + "score": 1.0, + "content": "policies instead of a specific policy.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "On a different thread, there are also studies that bring counterfactual inference techniques of causal", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 555, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 506, + 568 + ], + "score": 1.0, + "content": "inference into model-based RL (Buesing et al., 2019; Pitis et al., 2020; Sontakke et al., 2021).", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "These works consider that the transition function is relevant to some hidden noise variables and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "use Pearl-style structural causal models (SCMs), which is a directed acyclic graphs to define the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 589, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 506, + 601 + ], + "score": 1.0, + "content": "causality of nodes in an environment, to handle the problem. SCMs can help RL in different ways:", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "Buesing et al. (2019) approximate the posterior of the noise variables based on the observation of", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "data, and environment models are learned based on the inferred noises. The generalization ability is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "improved if we can infer the correct value of the noise variables. Pitis et al. (2020) discover several", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 633, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 505, + 644 + ], + "score": 1.0, + "content": "local causal structural models of a global environment model, then data augmentation strategies", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "by leveraging these local structures to generate counterfactual experiences. Sontakke et al. (2021)", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "proposes a representation learning technique for causal factors, which is an instance of the hidden", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "noise variables, in partially observable Markov decision processes (POMDPs). With the learned", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "representation of causal factors, the performance of policy learning and transfer in downstream", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "tasks will be improved. Instead of considering the hidden noise variables in the environments, our", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "study considers the environment model learning problem in the fully observed setting and focuses", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "on unbiased causal effect estimation in the offline dataset under behavior policies collected with", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 721, + 165, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 165, + 732 + ], + "score": 1.0, + "content": "selection bias.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 40 + } + ], + "page_idx": 25, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 513, + 221, + 584, + 243 + ], + "lines": [ + { + "bbox": [ + 512, + 219, + 586, + 234 + ], + "spans": [ + { + "bbox": [ + 512, + 219, + 586, + 234 + ], + "score": 1.0, + "content": "to Reviewer j5p5", + "type": "text" + } + ] + }, + { + "bbox": [ + 512, + 230, + 554, + 244 + ], + "spans": [ + { + "bbox": [ + 512, + 230, + 554, + 244 + ], + "score": 1.0, + "content": "and rQ79", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 514, + 688, + 583, + 699 + ], + "lines": [ + { + "bbox": [ + 513, + 686, + 584, + 701 + ], + "spans": [ + { + "bbox": [ + 513, + 686, + 584, + 701 + ], + "score": 1.0, + "content": "to Reviewer j5p5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 193, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 194, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 194, + 96 + ], + "score": 1.0, + "content": "objective of GAIL is:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 106, + 81, + 194, + 96 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 101, + 478, + 171 + ], + "lines": [ + { + "bbox": [ + 133, + 101, + 478, + 171 + ], + "spans": [ + { + "bbox": [ + 133, + 101, + 478, + 171 + ], + "score": 0.94, + "content": "\\begin{array} { r l } { \\displaystyle \\theta _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } & { } \\\\ { \\displaystyle s . t . } & { Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } \\right] } \\\\ { \\displaystyle \\varphi ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ \\log ( 1 - D _ { \\varphi } ( x , a , x ^ { \\prime } ) ) \\right] , } \\end{array}", + "type": "interline_equation", + "image_path": "dcc7cb0b1e1425d35c6360b351bcb8b8deac7e8830c1f531755e2409a6b68632.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 133, + 101, + 478, + 124.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 133, + 124.33333333333333, + 478, + 147.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 133, + 147.66666666666666, + 478, + 171.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 177, + 488, + 193 + ], + "lines": [ + { + "bbox": [ + 108, + 177, + 486, + 194 + ], + "spans": [ + { + "bbox": [ + 108, + 177, + 136, + 194 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 179, + 313, + 192 + ], + "score": 0.93, + "content": "A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) = Q _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) - V _ { M _ { \\theta } } ^ { \\kappa } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 177, + 331, + 194 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 332, + 179, + 486, + 194 + ], + "score": 0.9, + "content": "V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb { E } _ { M _ { \\theta _ { t } } ( x , a ) } \\left[ Q ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\right] ,", + "type": "inline_equation" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 108, + 177, + 486, + 194 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 210, + 281, + 223 + ], + "lines": [ + { + "bbox": [ + 105, + 210, + 283, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 283, + 225 + ], + "score": 1.0, + "content": "F ADDITIONAL RELATED WORK", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 238, + 505, + 540 + ], + "lines": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "Our primitive objective is inspired by weighted empirical risk minimization (WERM) based on", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "inverse propensity score (IPS). WERM is originally proposed to solve the generalization problem of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 260, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 505, + 271 + ], + "score": 1.0, + "content": "domain adaptation in machine learning literature. For instance, we would like to train a predictor", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 270, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 107, + 272, + 140, + 283 + ], + "score": 0.94, + "content": "M ( y | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 270, + 259, + 283 + ], + "score": 1.0, + "content": "in a domain with distribution", + "type": "text" + }, + { + "bbox": [ + 259, + 272, + 297, + 283 + ], + "score": 0.93, + "content": "P _ { \\mathrm { t r a i n } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 270, + 505, + 283 + ], + "score": 1.0, + "content": "to minimize the prediction risks in the domain with", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 280, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 155, + 295 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 156, + 283, + 189, + 294 + ], + "score": 0.92, + "content": "P _ { \\mathrm { { t e s t } } } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 280, + 221, + 295 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 221, + 284, + 274, + 293 + ], + "score": 0.92, + "content": "P _ { \\mathrm { t e s t } } \\neq P _ { \\mathrm { t e s t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 280, + 506, + 295 + ], + "score": 1.0, + "content": ". To solve the problem, we can train a weighted objective", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 128, + 291, + 506, + 310 + ], + "spans": [ + { + "bbox": [ + 128, + 294, + 282, + 310 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\operatorname* { m a x } _ { M } \\mathbb { E } _ { x \\sim P _ { \\mathrm { t r a i n } } } [ \\frac { P _ { \\mathrm { t e s t } } ( x ) } { P _ { \\mathrm { t r a i n } } ( x ) } \\log M ( y | x ) ] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 291, + 506, + 309 + ], + "score": 1.0, + "content": ", which is called weighted empirical risk minimization", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "methods (Ben-David et al., 2006; 2010; Cortes et al., 2010; Byrd & Lipton, 2019; Quinonero-Candela", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "et al., 2008). These results have been extended and applied to causal inference, where the predictor", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 329, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 342 + ], + "score": 1.0, + "content": "is required to be generalized from the data distribution in observational studies (source domain)", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 341, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 505, + 352 + ], + "score": 1.0, + "content": "to the data distribution in randomized controlled trials (target domain) (Shimodaira, 2000; Assaad", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 350, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 209, + 365 + ], + "score": 1.0, + "content": "et al., 2021; Hassanpour", + "type": "text" + }, + { + "bbox": [ + 209, + 354, + 217, + 362 + ], + "score": 0.63, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 350, + 506, + 365 + ], + "score": 1.0, + "content": "Greiner, 2019; Jung et al., 2020; Johansson et al., 2018). In this case,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 363, + 244, + 375 + ], + "score": 1.0, + "content": "the input features include a state", + "type": "text" + }, + { + "bbox": [ + 244, + 368, + 250, + 373 + ], + "score": 0.87, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 363, + 388, + 375 + ], + "score": 1.0, + "content": "(a.k.a. covariates) and an action", + "type": "text" + }, + { + "bbox": [ + 388, + 368, + 394, + 372 + ], + "score": 0.87, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "(a.k.a. treatment variable)", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 373, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 383, + 386 + ], + "score": 1.0, + "content": "which is sampled from a policy. We often assume the distribution of", + "type": "text" + }, + { + "bbox": [ + 383, + 379, + 389, + 383 + ], + "score": 0.83, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 373, + 394, + 386 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 394, + 375, + 415, + 386 + ], + "score": 0.92, + "content": "P ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 373, + 505, + 386 + ], + "score": 1.0, + "content": "is consistent between", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 103, + 383, + 506, + 404 + ], + "spans": [ + { + "bbox": [ + 103, + 384, + 413, + 404 + ], + "score": 1.0, + "content": "the source domain and the test domain, then we have Ptest(x)Ptrain(x) = P (x)β(a|x)P (x)µ(a|x)", + "type": "text" + }, + { + "bbox": [ + 449, + 383, + 480, + 403 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 480, + 393, + 486, + 399 + ], + "score": 0.87, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 383, + 506, + 403 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 107, + 400, + 506, + 413 + ], + "spans": [ + { + "bbox": [ + 107, + 402, + 114, + 412 + ], + "score": 0.86, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 400, + 506, + 413 + ], + "score": 1.0, + "content": "are the policies in source and target domains respectively. In Shimodaira (2000); Assaad et al.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 411, + 506, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 190, + 424 + ], + "score": 1.0, + "content": "(2021); Hassanpour", + "type": "text" + }, + { + "bbox": [ + 190, + 414, + 199, + 421 + ], + "score": 0.64, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 411, + 506, + 424 + ], + "score": 1.0, + "content": "Greiner (2019), the policy in randomized controlled trials is modeled as a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 421, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 194, + 438 + ], + "score": 1.0, + "content": "uniform policy, then", + "type": "text" + }, + { + "bbox": [ + 195, + 424, + 367, + 439 + ], + "score": 0.88, + "content": "\\begin{array} { r } { \\frac { P _ { \\mathrm { t e s t } } ( x ) } { P _ { \\mathrm { t r a i n } } ( x ) } = \\frac { P ( x ) \\beta ( a | x ) } { P ( x ) \\mu ( a | x ) } = \\frac { \\beta ( a | x ) } { \\mu ( a | x ) } \\propto \\frac { 1 } { \\mu ( a | x ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 427, + 374, + 440 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 374, + 425, + 399, + 440 + ], + "score": 0.79, + "content": "{ \\frac { 1 } { \\mu ( a | x ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 425, + 505, + 437 + ], + "score": 1.0, + "content": "is also known as inverse", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 450 + ], + "score": 1.0, + "content": "propensity score (IPS). In Johansson et al. (2018), it assumes that the policy in the target domain is", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 448, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 104, + 448, + 162, + 463 + ], + "score": 1.0, + "content": "predefined as", + "type": "text" + }, + { + "bbox": [ + 163, + 452, + 190, + 462 + ], + "score": 0.94, + "content": "\\beta ( a | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 448, + 386, + 463 + ], + "score": 1.0, + "content": "before environment model learning, then it uses", + "type": "text" + }, + { + "bbox": [ + 387, + 450, + 393, + 465 + ], + "score": 0.87, + "content": "\\frac { \\beta } { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 448, + 506, + 463 + ], + "score": 1.0, + "content": "as the IPS. The differences", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 463, + 504, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 504, + 474 + ], + "score": 1.0, + "content": "between AWRM and previous works are fallen in two aspects: (1) We consider the distribution-shift", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 474, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 505, + 485 + ], + "score": 1.0, + "content": "problem in the sequential decision-making scenario. In this scenario, we not only consider the action", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 485, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 324, + 496 + ], + "score": 1.0, + "content": "distribution mismatching between the behavior policy", + "type": "text" + }, + { + "bbox": [ + 325, + 489, + 331, + 496 + ], + "score": 0.88, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 485, + 446, + 496 + ], + "score": 1.0, + "content": "and the policy to evaluation", + "type": "text" + }, + { + "bbox": [ + 446, + 487, + 452, + 496 + ], + "score": 0.88, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 485, + 505, + 496 + ], + "score": 1.0, + "content": ", but also the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 495, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 508 + ], + "score": 1.0, + "content": "follow-up effects of policies to the state distribution; (2) For faithful offline policy optimization, we", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 506, + 504, + 517 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 504, + 517 + ], + "score": 1.0, + "content": "require the environment model to have generalization ability in numerous different policies. The", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 517, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 403, + 529 + ], + "score": 1.0, + "content": "objective of AWRM is proposed to guarantee the generalization ability of", + "type": "text" + }, + { + "bbox": [ + 403, + 520, + 415, + 527 + ], + "score": 0.9, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 517, + 506, + 529 + ], + "score": 1.0, + "content": "in numerous different", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 527, + 250, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 527, + 250, + 541 + ], + "score": 1.0, + "content": "policies instead of a specific policy.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 18.5, + "bbox_fs": [ + 103, + 237, + 506, + 541 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "On a different thread, there are also studies that bring counterfactual inference techniques of causal", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 555, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 506, + 568 + ], + "score": 1.0, + "content": "inference into model-based RL (Buesing et al., 2019; Pitis et al., 2020; Sontakke et al., 2021).", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "These works consider that the transition function is relevant to some hidden noise variables and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "use Pearl-style structural causal models (SCMs), which is a directed acyclic graphs to define the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 589, + 506, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 506, + 601 + ], + "score": 1.0, + "content": "causality of nodes in an environment, to handle the problem. SCMs can help RL in different ways:", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "Buesing et al. (2019) approximate the posterior of the noise variables based on the observation of", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "data, and environment models are learned based on the inferred noises. The generalization ability is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 505, + 634 + ], + "score": 1.0, + "content": "improved if we can infer the correct value of the noise variables. Pitis et al. (2020) discover several", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 633, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 505, + 644 + ], + "score": 1.0, + "content": "local causal structural models of a global environment model, then data augmentation strategies", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 642, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 106, + 642, + 505, + 655 + ], + "score": 1.0, + "content": "by leveraging these local structures to generate counterfactual experiences. Sontakke et al. (2021)", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "proposes a representation learning technique for causal factors, which is an instance of the hidden", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 505, + 677 + ], + "score": 1.0, + "content": "noise variables, in partially observable Markov decision processes (POMDPs). With the learned", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "representation of causal factors, the performance of policy learning and transfer in downstream", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "tasks will be improved. Instead of considering the hidden noise variables in the environments, our", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "study considers the environment model learning problem in the fully observed setting and focuses", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 721 + ], + "score": 1.0, + "content": "on unbiased causal effect estimation in the offline dataset under behavior policies collected with", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 721, + 165, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 165, + 732 + ], + "score": 1.0, + "content": "selection bias.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 545, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "In offline model-based RL, the problem is called distribution shift (Yu et al., 2020; Levine et al.,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "2020; Chen et al., 2021) which has received great attentions. However, previous algorithms do not", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "handle the model learning challenge directly but propose techniques to suppress policy sampling and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "learning in risky regions (Yu et al., 2020; Kidambi et al., 2020). Although these algorithms have made", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 506, + 138 + ], + "score": 1.0, + "content": "great progress in offline policy optimization in many tasks, so far, how to learn a better environment", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 138, + 303, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 303, + 149 + ], + "score": 1.0, + "content": "model in this scenario has rarely been discussed.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 164, + 245, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 163, + 246, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 246, + 180 + ], + "score": 1.0, + "content": "G EXPERIMENT DETAILS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "title", + "bbox": [ + 107, + 189, + 177, + 200 + ], + "lines": [ + { + "bbox": [ + 106, + 189, + 178, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 178, + 201 + ], + "score": 1.0, + "content": "G.1 SETTINGS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 209, + 364, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 208, + 365, + 224 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 365, + 224 + ], + "score": 1.0, + "content": "G.1.1 GENERAL NEGATIVE FEEDBACK CONTROL (GNFC)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 229, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 106, + 228, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 437, + 242 + ], + "score": 1.0, + "content": "The design of GNFC is inspired by a classic type of scenario that behavior policies", + "type": "text" + }, + { + "bbox": [ + 438, + 231, + 445, + 241 + ], + "score": 0.81, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 228, + 505, + 242 + ], + "score": 1.0, + "content": "have selection", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 240, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 505, + 253 + ], + "score": 1.0, + "content": "bias and easily lead to counterfactual risks: For some internet platforms, we would like to allocate", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "score": 1.0, + "content": "budgets to a set of targets (e.g., customers or cities) to increase the engagement of the targets in the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 262, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 506, + 275 + ], + "score": 1.0, + "content": "platforms. Our task is to train a model to predict targets’ feedback on engagement given targets’", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 272, + 231, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 231, + 286 + ], + "score": 1.0, + "content": "features and allocated budgets.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 290, + 505, + 368 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 506, + 304 + ], + "score": 1.0, + "content": "In these tasks, for better benefits, the online working policy (i.e., the behavior policy) will tend to cut", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 300, + 504, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 504, + 313 + ], + "score": 1.0, + "content": "down the budgets if targets have better engagement, otherwise, the budgets might be increased. The", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 312, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 505, + 324 + ], + "score": 1.0, + "content": "risk of counterfactual environment model learning in the task is that: the object with better historical", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 322, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 505, + 336 + ], + "score": 1.0, + "content": "engagement will be sent to smaller budgets because of the selection bias of the behavior policies, then", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 333, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 347 + ], + "score": 1.0, + "content": "the model might exploit this correlation for learning and get a conclusion that: increasing budgets", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 345, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 506, + 358 + ], + "score": 1.0, + "content": "will reduce the targets’ engagement, which violates the real causality. We construct an environment", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 356, + 486, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 441, + 369 + ], + "score": 1.0, + "content": "and a behavior policy to mimic the above process. In particular, the behavior policy", + "type": "text" + }, + { + "bbox": [ + 442, + 358, + 475, + 367 + ], + "score": 0.34, + "content": "\\mu _ { G N F C }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 356, + 486, + 369 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 371, + 383, + 397 + ], + "lines": [ + { + "bbox": [ + 227, + 371, + 383, + 397 + ], + "spans": [ + { + "bbox": [ + 227, + 371, + 383, + 397 + ], + "score": 0.94, + "content": "\\mu _ { G N F C } ( x ) = \\frac { ( 6 2 . 5 - \\mathrm { m e a n } ( x ) ) } { 1 5 } + \\epsilon ,", + "type": "interline_equation", + "image_path": "7366a38ced3d0e1a19ebf10b54b8fb13efb412b10f6569aee91727b2cee843e6.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 227, + 371, + 383, + 397 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 400, + 479, + 412 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 481, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 133, + 413 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 402, + 138, + 410 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 399, + 481, + 413 + ], + "score": 1.0, + "content": "is a sample noise, which will be discussed later. The environment includes two parts:", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 416, + 243, + 428 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 244, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 193, + 430 + ], + "score": 1.0, + "content": "(1) response function", + "type": "text" + }, + { + "bbox": [ + 194, + 416, + 240, + 429 + ], + "score": 0.93, + "content": "M _ { 1 } ( y | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 416, + 244, + 430 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 433, + 377, + 447 + ], + "lines": [ + { + "bbox": [ + 234, + 433, + 377, + 447 + ], + "spans": [ + { + "bbox": [ + 234, + 433, + 377, + 447 + ], + "score": 0.9, + "content": "M _ { 1 } ( y | x , a ) = \\mathcal { N } ( \\mathrm { m e a n } ( x ) + a , 2 )", + "type": "interline_equation", + "image_path": "aeabc49d314cce4db29df3c871644034fe0c1b6bc9e8e50cbea5b151dbafa052.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 234, + 433, + 377, + 447 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 457, + 246, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 243, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 193, + 471 + ], + "score": 1.0, + "content": "(2) mapping function", + "type": "text" + }, + { + "bbox": [ + 194, + 457, + 243, + 469 + ], + "score": 0.93, + "content": "M _ { 2 } ( x ^ { \\prime } | x , y )", + "type": "inline_equation" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 473, + 379, + 487 + ], + "lines": [ + { + "bbox": [ + 232, + 473, + 379, + 487 + ], + "spans": [ + { + "bbox": [ + 232, + 473, + 379, + 487 + ], + "score": 0.92, + "content": "M _ { 2 } ( x ^ { \\prime } | x , a , y ) = y - \\mathrm { m e a n } ( x ) + x", + "type": "interline_equation", + "image_path": "be029771be3f70c87d8809389293e5a4aa3ecb5df20f4ec1c492ac7a5ce34ddf.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 232, + 473, + 379, + 487 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 491, + 506, + 624 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 198, + 504 + ], + "score": 1.0, + "content": "The transition function", + "type": "text" + }, + { + "bbox": [ + 198, + 492, + 214, + 501 + ], + "score": 0.89, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 491, + 285, + 504 + ], + "score": 1.0, + "content": "is a composite of", + "type": "text" + }, + { + "bbox": [ + 285, + 491, + 447, + 504 + ], + "score": 0.91, + "content": "M ^ { * } ( x ^ { \\prime } | x , a ) = M _ { 2 } ( x ^ { \\prime } | x , a , M _ { 1 } ( y | x , a ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 491, + 506, + 504 + ], + "score": 1.0, + "content": ". The behavior", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "policies have selection bias: the actions taken are negatively correlated with the states, as illustrated", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 514, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 467, + 526 + ], + "score": 1.0, + "content": "in Fig. 8(a) and Fig. 8(b). We control the difficulty of distinguishing the correct causality of", + "type": "text" + }, + { + "bbox": [ + 468, + 515, + 485, + 524 + ], + "score": 0.44, + "content": "x , a ,", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 514, + 505, + 526 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 107, + 525, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 107, + 527, + 113, + 536 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 113, + 525, + 327, + 537 + ], + "score": 1.0, + "content": "by designing different strategies of noise sampling on", + "type": "text" + }, + { + "bbox": [ + 327, + 527, + 333, + 534 + ], + "score": 0.56, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 525, + 505, + 537 + ], + "score": 1.0, + "content": ". In principle, with a larger number or more", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 535, + 506, + 548 + ], + "spans": [ + { + "bbox": [ + 104, + 535, + 448, + 548 + ], + "score": 1.0, + "content": "pronounced disturbances, there are more samples violating the correlation between", + "type": "text" + }, + { + "bbox": [ + 448, + 538, + 456, + 545 + ], + "score": 0.69, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 535, + 474, + 548 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 474, + 537, + 481, + 545 + ], + "score": 0.71, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 535, + 506, + 548 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 546, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 104, + 546, + 506, + 559 + ], + "score": 1.0, + "content": "more samples can be used to find the correct causality. Therefore, we can control the difficulty of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 558, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 506, + 570 + ], + "score": 1.0, + "content": "counterfactual environment model learning by controlling the strength of disturbance. In particular,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 150, + 581 + ], + "score": 1.0, + "content": "we sample", + "type": "text" + }, + { + "bbox": [ + 150, + 570, + 156, + 578 + ], + "score": 0.77, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 567, + 267, + 581 + ], + "score": 1.0, + "content": "from a uniform distribution", + "type": "text" + }, + { + "bbox": [ + 268, + 568, + 305, + 580 + ], + "score": 0.92, + "content": "U ( - e , e )", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 567, + 372, + 581 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 372, + 570, + 379, + 580 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 567, + 415, + 581 + ], + "score": 1.0, + "content": ". That is,", + "type": "text" + }, + { + "bbox": [ + 415, + 569, + 439, + 578 + ], + "score": 0.89, + "content": "\\epsilon = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 567, + 506, + 581 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 579, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 130, + 591 + ], + "score": 0.88, + "content": "1 - p", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 579, + 147, + 592 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 147, + 579, + 203, + 591 + ], + "score": 0.94, + "content": "\\epsilon \\sim U ( - e , e )", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 579, + 269, + 592 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 270, + 582, + 276, + 591 + ], + "score": 0.77, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 579, + 347, + 592 + ], + "score": 1.0, + "content": ". Then with larger", + "type": "text" + }, + { + "bbox": [ + 348, + 581, + 354, + 591 + ], + "score": 0.76, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 579, + 506, + 592 + ], + "score": 1.0, + "content": ", there are more samples in the dataset", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 590, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 265, + 604 + ], + "score": 1.0, + "content": "violating the negative correlation (i.e.,", + "type": "text" + }, + { + "bbox": [ + 265, + 592, + 299, + 602 + ], + "score": 0.76, + "content": "\\mu _ { G N F C } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 590, + 370, + 604 + ], + "score": 1.0, + "content": "), and with larger", + "type": "text" + }, + { + "bbox": [ + 371, + 592, + 376, + 600 + ], + "score": 0.67, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 590, + 506, + 604 + ], + "score": 1.0, + "content": ", the difference of the feedback", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 601, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 288, + 614 + ], + "score": 1.0, + "content": "will be more obvious. By selecting different", + "type": "text" + }, + { + "bbox": [ + 288, + 604, + 294, + 611 + ], + "score": 0.77, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 601, + 312, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 312, + 603, + 318, + 613 + ], + "score": 0.78, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 601, + 506, + 614 + ], + "score": 1.0, + "content": ", we can construct different tasks to verify the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 612, + 448, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 448, + 624 + ], + "score": 1.0, + "content": "effectiveness and ability of the counterfactual environment model learning algorithm.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 32.5 + }, + { + "type": "title", + "bbox": [ + 107, + 635, + 315, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 316, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 316, + 649 + ], + "score": 1.0, + "content": "G.1.2 THE CANCER GENOMIC ATLAS (TCGA)", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 506, + 667 + ], + "score": 1.0, + "content": "The Cancer Genomic Atlas (TCGA) is a project that has profiled and analyzed large numbers of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "human tumors to discover molecular aberrations at the DNA, RNA, protein, and epigenetic levels. The", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "resulting rich data provide a significant opportunity to accelerate our understanding of the molecular", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 262, + 699 + ], + "score": 1.0, + "content": "basis of cancer. We obtain features,", + "type": "text" + }, + { + "bbox": [ + 262, + 690, + 270, + 698 + ], + "score": 0.33, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 688, + 505, + 699 + ], + "score": 1.0, + "content": ", from the TCGA dataset and consider three continuous", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 386, + 711 + ], + "score": 1.0, + "content": "treatments as done in SCIGAN (Bica et al., 2020). Each treatment,", + "type": "text" + }, + { + "bbox": [ + 386, + 701, + 393, + 709 + ], + "score": 0.69, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 698, + 506, + 711 + ], + "score": 1.0, + "content": ", is associated with a set of", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 156, + 722 + ], + "score": 1.0, + "content": "parameters,", + "type": "text" + }, + { + "bbox": [ + 157, + 712, + 168, + 721 + ], + "score": 0.54, + "content": "\\mathbf { v } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 709, + 173, + 722 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 173, + 712, + 185, + 721 + ], + "score": 0.29, + "content": "\\mathbf { v } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 709, + 188, + 722 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 189, + 711, + 200, + 721 + ], + "score": 0.35, + "content": "\\mathbf { v } _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 709, + 506, + 722 + ], + "score": 1.0, + "content": ", that are sampled randomly by sampling a vector from a standard normal", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 721, + 504, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 472, + 732 + ], + "score": 1.0, + "content": "distribution and scaling it with its norm. We assign interventions by sampling a treatment,", + "type": "text" + }, + { + "bbox": [ + 473, + 723, + 479, + 731 + ], + "score": 0.69, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 721, + 504, + 732 + ], + "score": 1.0, + "content": ", from", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43 + } + ], + "page_idx": 26, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "In offline model-based RL, the problem is called distribution shift (Yu et al., 2020; Levine et al.,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "2020; Chen et al., 2021) which has received great attentions. However, previous algorithms do not", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "handle the model learning challenge directly but propose techniques to suppress policy sampling and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 505, + 127 + ], + "score": 1.0, + "content": "learning in risky regions (Yu et al., 2020; Kidambi et al., 2020). Although these algorithms have made", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 506, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 506, + 138 + ], + "score": 1.0, + "content": "great progress in offline policy optimization in many tasks, so far, how to learn a better environment", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 138, + 303, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 303, + 149 + ], + "score": 1.0, + "content": "model in this scenario has rarely been discussed.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 82, + 506, + 149 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 164, + 245, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 163, + 246, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 246, + 180 + ], + "score": 1.0, + "content": "G EXPERIMENT DETAILS", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "title", + "bbox": [ + 107, + 189, + 177, + 200 + ], + "lines": [ + { + "bbox": [ + 106, + 189, + 178, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 189, + 178, + 201 + ], + "score": 1.0, + "content": "G.1 SETTINGS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 209, + 364, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 208, + 365, + 224 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 365, + 224 + ], + "score": 1.0, + "content": "G.1.1 GENERAL NEGATIVE FEEDBACK CONTROL (GNFC)", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 208, + 365, + 224 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 229, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 106, + 228, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 106, + 228, + 437, + 242 + ], + "score": 1.0, + "content": "The design of GNFC is inspired by a classic type of scenario that behavior policies", + "type": "text" + }, + { + "bbox": [ + 438, + 231, + 445, + 241 + ], + "score": 0.81, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 228, + 505, + 242 + ], + "score": 1.0, + "content": "have selection", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 240, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 505, + 253 + ], + "score": 1.0, + "content": "bias and easily lead to counterfactual risks: For some internet platforms, we would like to allocate", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "score": 1.0, + "content": "budgets to a set of targets (e.g., customers or cities) to increase the engagement of the targets in the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 262, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 506, + 275 + ], + "score": 1.0, + "content": "platforms. Our task is to train a model to predict targets’ feedback on engagement given targets’", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 272, + 231, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 272, + 231, + 286 + ], + "score": 1.0, + "content": "features and allocated budgets.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 228, + 506, + 286 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 290, + 505, + 368 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 506, + 304 + ], + "score": 1.0, + "content": "In these tasks, for better benefits, the online working policy (i.e., the behavior policy) will tend to cut", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 300, + 504, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 504, + 313 + ], + "score": 1.0, + "content": "down the budgets if targets have better engagement, otherwise, the budgets might be increased. The", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 312, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 505, + 324 + ], + "score": 1.0, + "content": "risk of counterfactual environment model learning in the task is that: the object with better historical", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 322, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 505, + 336 + ], + "score": 1.0, + "content": "engagement will be sent to smaller budgets because of the selection bias of the behavior policies, then", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 333, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 505, + 347 + ], + "score": 1.0, + "content": "the model might exploit this correlation for learning and get a conclusion that: increasing budgets", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 345, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 506, + 358 + ], + "score": 1.0, + "content": "will reduce the targets’ engagement, which violates the real causality. We construct an environment", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 356, + 486, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 441, + 369 + ], + "score": 1.0, + "content": "and a behavior policy to mimic the above process. In particular, the behavior policy", + "type": "text" + }, + { + "bbox": [ + 442, + 358, + 475, + 367 + ], + "score": 0.34, + "content": "\\mu _ { G N F C }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 356, + 486, + 369 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 289, + 506, + 369 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 227, + 371, + 383, + 397 + ], + "lines": [ + { + "bbox": [ + 227, + 371, + 383, + 397 + ], + "spans": [ + { + "bbox": [ + 227, + 371, + 383, + 397 + ], + "score": 0.94, + "content": "\\mu _ { G N F C } ( x ) = \\frac { ( 6 2 . 5 - \\mathrm { m e a n } ( x ) ) } { 1 5 } + \\epsilon ,", + "type": "interline_equation", + "image_path": "7366a38ced3d0e1a19ebf10b54b8fb13efb412b10f6569aee91727b2cee843e6.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 227, + 371, + 383, + 397 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 400, + 479, + 412 + ], + "lines": [ + { + "bbox": [ + 105, + 399, + 481, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 133, + 413 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 402, + 138, + 410 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 399, + 481, + 413 + ], + "score": 1.0, + "content": "is a sample noise, which will be discussed later. The environment includes two parts:", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 399, + 481, + 413 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 416, + 243, + 428 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 244, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 193, + 430 + ], + "score": 1.0, + "content": "(1) response function", + "type": "text" + }, + { + "bbox": [ + 194, + 416, + 240, + 429 + ], + "score": 0.93, + "content": "M _ { 1 } ( y | x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 416, + 244, + 430 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 416, + 244, + 430 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 433, + 377, + 447 + ], + "lines": [ + { + "bbox": [ + 234, + 433, + 377, + 447 + ], + "spans": [ + { + "bbox": [ + 234, + 433, + 377, + 447 + ], + "score": 0.9, + "content": "M _ { 1 } ( y | x , a ) = \\mathcal { N } ( \\mathrm { m e a n } ( x ) + a , 2 )", + "type": "interline_equation", + "image_path": "aeabc49d314cce4db29df3c871644034fe0c1b6bc9e8e50cbea5b151dbafa052.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 234, + 433, + 377, + 447 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 457, + 246, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 243, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 193, + 471 + ], + "score": 1.0, + "content": "(2) mapping function", + "type": "text" + }, + { + "bbox": [ + 194, + 457, + 243, + 469 + ], + "score": 0.93, + "content": "M _ { 2 } ( x ^ { \\prime } | x , y )", + "type": "inline_equation" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 456, + 243, + 471 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 473, + 379, + 487 + ], + "lines": [ + { + "bbox": [ + 232, + 473, + 379, + 487 + ], + "spans": [ + { + "bbox": [ + 232, + 473, + 379, + 487 + ], + "score": 0.92, + "content": "M _ { 2 } ( x ^ { \\prime } | x , a , y ) = y - \\mathrm { m e a n } ( x ) + x", + "type": "interline_equation", + "image_path": "be029771be3f70c87d8809389293e5a4aa3ecb5df20f4ec1c492ac7a5ce34ddf.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 232, + 473, + 379, + 487 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 491, + 506, + 624 + ], + "lines": [ + { + "bbox": [ + 105, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 198, + 504 + ], + "score": 1.0, + "content": "The transition function", + "type": "text" + }, + { + "bbox": [ + 198, + 492, + 214, + 501 + ], + "score": 0.89, + "content": "M ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 491, + 285, + 504 + ], + "score": 1.0, + "content": "is a composite of", + "type": "text" + }, + { + "bbox": [ + 285, + 491, + 447, + 504 + ], + "score": 0.91, + "content": "M ^ { * } ( x ^ { \\prime } | x , a ) = M _ { 2 } ( x ^ { \\prime } | x , a , M _ { 1 } ( y | x , a ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 491, + 506, + 504 + ], + "score": 1.0, + "content": ". The behavior", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "policies have selection bias: the actions taken are negatively correlated with the states, as illustrated", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 514, + 505, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 467, + 526 + ], + "score": 1.0, + "content": "in Fig. 8(a) and Fig. 8(b). We control the difficulty of distinguishing the correct causality of", + "type": "text" + }, + { + "bbox": [ + 468, + 515, + 485, + 524 + ], + "score": 0.44, + "content": "x , a ,", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 514, + 505, + 526 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 107, + 525, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 107, + 527, + 113, + 536 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 113, + 525, + 327, + 537 + ], + "score": 1.0, + "content": "by designing different strategies of noise sampling on", + "type": "text" + }, + { + "bbox": [ + 327, + 527, + 333, + 534 + ], + "score": 0.56, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 525, + 505, + 537 + ], + "score": 1.0, + "content": ". In principle, with a larger number or more", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 535, + 506, + 548 + ], + "spans": [ + { + "bbox": [ + 104, + 535, + 448, + 548 + ], + "score": 1.0, + "content": "pronounced disturbances, there are more samples violating the correlation between", + "type": "text" + }, + { + "bbox": [ + 448, + 538, + 456, + 545 + ], + "score": 0.69, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 535, + 474, + 548 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 474, + 537, + 481, + 545 + ], + "score": 0.71, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 535, + 506, + 548 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 546, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 104, + 546, + 506, + 559 + ], + "score": 1.0, + "content": "more samples can be used to find the correct causality. Therefore, we can control the difficulty of", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 558, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 506, + 570 + ], + "score": 1.0, + "content": "counterfactual environment model learning by controlling the strength of disturbance. In particular,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 567, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 567, + 150, + 581 + ], + "score": 1.0, + "content": "we sample", + "type": "text" + }, + { + "bbox": [ + 150, + 570, + 156, + 578 + ], + "score": 0.77, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 567, + 267, + 581 + ], + "score": 1.0, + "content": "from a uniform distribution", + "type": "text" + }, + { + "bbox": [ + 268, + 568, + 305, + 580 + ], + "score": 0.92, + "content": "U ( - e , e )", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 567, + 372, + 581 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 372, + 570, + 379, + 580 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 567, + 415, + 581 + ], + "score": 1.0, + "content": ". That is,", + "type": "text" + }, + { + "bbox": [ + 415, + 569, + 439, + 578 + ], + "score": 0.89, + "content": "\\epsilon = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 567, + 506, + 581 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 579, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 130, + 591 + ], + "score": 0.88, + "content": "1 - p", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 579, + 147, + 592 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 147, + 579, + 203, + 591 + ], + "score": 0.94, + "content": "\\epsilon \\sim U ( - e , e )", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 579, + 269, + 592 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 270, + 582, + 276, + 591 + ], + "score": 0.77, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 579, + 347, + 592 + ], + "score": 1.0, + "content": ". Then with larger", + "type": "text" + }, + { + "bbox": [ + 348, + 581, + 354, + 591 + ], + "score": 0.76, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 579, + 506, + 592 + ], + "score": 1.0, + "content": ", there are more samples in the dataset", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 590, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 265, + 604 + ], + "score": 1.0, + "content": "violating the negative correlation (i.e.,", + "type": "text" + }, + { + "bbox": [ + 265, + 592, + 299, + 602 + ], + "score": 0.76, + "content": "\\mu _ { G N F C } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 590, + 370, + 604 + ], + "score": 1.0, + "content": "), and with larger", + "type": "text" + }, + { + "bbox": [ + 371, + 592, + 376, + 600 + ], + "score": 0.67, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 590, + 506, + 604 + ], + "score": 1.0, + "content": ", the difference of the feedback", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 601, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 288, + 614 + ], + "score": 1.0, + "content": "will be more obvious. By selecting different", + "type": "text" + }, + { + "bbox": [ + 288, + 604, + 294, + 611 + ], + "score": 0.77, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 601, + 312, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 312, + 603, + 318, + 613 + ], + "score": 0.78, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 601, + 506, + 614 + ], + "score": 1.0, + "content": ", we can construct different tasks to verify the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 612, + 448, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 448, + 624 + ], + "score": 1.0, + "content": "effectiveness and ability of the counterfactual environment model learning algorithm.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 32.5, + "bbox_fs": [ + 104, + 491, + 506, + 624 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 635, + 315, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 633, + 316, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 316, + 649 + ], + "score": 1.0, + "content": "G.1.2 THE CANCER GENOMIC ATLAS (TCGA)", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 655, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 506, + 667 + ], + "score": 1.0, + "content": "The Cancer Genomic Atlas (TCGA) is a project that has profiled and analyzed large numbers of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "human tumors to discover molecular aberrations at the DNA, RNA, protein, and epigenetic levels. The", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "resulting rich data provide a significant opportunity to accelerate our understanding of the molecular", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 262, + 699 + ], + "score": 1.0, + "content": "basis of cancer. We obtain features,", + "type": "text" + }, + { + "bbox": [ + 262, + 690, + 270, + 698 + ], + "score": 0.33, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 688, + 505, + 699 + ], + "score": 1.0, + "content": ", from the TCGA dataset and consider three continuous", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 386, + 711 + ], + "score": 1.0, + "content": "treatments as done in SCIGAN (Bica et al., 2020). Each treatment,", + "type": "text" + }, + { + "bbox": [ + 386, + 701, + 393, + 709 + ], + "score": 0.69, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 698, + 506, + 711 + ], + "score": 1.0, + "content": ", is associated with a set of", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 156, + 722 + ], + "score": 1.0, + "content": "parameters,", + "type": "text" + }, + { + "bbox": [ + 157, + 712, + 168, + 721 + ], + "score": 0.54, + "content": "\\mathbf { v } _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 709, + 173, + 722 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 173, + 712, + 185, + 721 + ], + "score": 0.29, + "content": "\\mathbf { v } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 709, + 188, + 722 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 189, + 711, + 200, + 721 + ], + "score": 0.35, + "content": "\\mathbf { v } _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 709, + 506, + 722 + ], + "score": 1.0, + "content": ", that are sampled randomly by sampling a vector from a standard normal", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 721, + 504, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 472, + 732 + ], + "score": 1.0, + "content": "distribution and scaling it with its norm. We assign interventions by sampling a treatment,", + "type": "text" + }, + { + "bbox": [ + 473, + 723, + 479, + 731 + ], + "score": 0.69, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 721, + 504, + 732 + ], + "score": 1.0, + "content": ", from", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 104, + 276, + 503, + 293 + ], + "spans": [ + { + "bbox": [ + 104, + 276, + 185, + 293 + ], + "score": 1.0, + "content": "a beta distribution,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 185, + 279, + 267, + 291 + ], + "score": 0.85, + "content": "a \\mid \\mathbf { x } \\sim \\operatorname { B e t a } \\left( \\alpha , \\beta \\right)", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 268, + 276, + 272, + 293 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 272, + 280, + 299, + 290 + ], + "score": 0.87, + "content": "\\alpha \\geq 1", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 299, + 276, + 427, + 293 + ], + "score": 1.0, + "content": "controls the sampling bias and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 427, + 278, + 503, + 291 + ], + "score": 0.9, + "content": "\\textstyle { \\beta = { \\frac { \\alpha - 1 } { a ^ { * } } } + 2 - \\alpha }", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 289, + 488, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 133, + 303 + ], + "score": 1.0, + "content": "where", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 133, + 291, + 144, + 300 + ], + "score": 0.86, + "content": "a ^ { * }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 144, + 289, + 305, + 303 + ], + "score": 1.0, + "content": "is the optimal treatment. This setting of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 306, + 291, + 313, + 302 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 313, + 289, + 436, + 303 + ], + "score": 1.0, + "content": "ensures that the mode of Beta", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 437, + 291, + 461, + 302 + ], + "score": 0.88, + "content": "( \\alpha , \\tilde { \\beta } )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 462, + 289, + 472, + 303 + ], + "score": 1.0, + "content": "is", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 473, + 291, + 483, + 300 + ], + "score": 0.85, + "content": "a ^ { * }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 483, + 289, + 488, + 303 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 6 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 655, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 145, + 98, + 454, + 219 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 145, + 98, + 454, + 219 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 145, + 98, + 454, + 219 + ], + "spans": [ + { + "bbox": [ + 145, + 98, + 454, + 219 + ], + "score": 0.959, + "type": "image", + "image_path": "55abe1b08792484f5a786bb75b1b3aab1b1d67d16735214a33fde14020f480e8.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 145, + 98, + 454, + 138.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 145, + 138.33333333333334, + 454, + 178.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 145, + 178.66666666666669, + 454, + 219.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 231, + 505, + 255 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 231, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 244 + ], + "score": 1.0, + "content": "Figure 8: Illustration of information about the collected dataset in GNFC. Each color of the line", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 242, + 461, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 288, + 257 + ], + "score": 1.0, + "content": "denotes one of the collected trajectories. The", + "type": "text" + }, + { + "bbox": [ + 288, + 243, + 297, + 253 + ], + "score": 0.46, + "content": "\\mathrm { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 242, + 461, + 257 + ], + "score": 1.0, + "content": "-axis denotes the timestep of a trajectory.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 279, + 505, + 302 + ], + "lines": [ + { + "bbox": [ + 104, + 276, + 503, + 293 + ], + "spans": [ + { + "bbox": [ + 104, + 276, + 185, + 293 + ], + "score": 1.0, + "content": "a beta distribution,", + "type": "text" + }, + { + "bbox": [ + 185, + 279, + 267, + 291 + ], + "score": 0.85, + "content": "a \\mid \\mathbf { x } \\sim \\operatorname { B e t a } \\left( \\alpha , \\beta \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 276, + 272, + 293 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 272, + 280, + 299, + 290 + ], + "score": 0.87, + "content": "\\alpha \\geq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 276, + 427, + 293 + ], + "score": 1.0, + "content": "controls the sampling bias and", + "type": "text" + }, + { + "bbox": [ + 427, + 278, + 503, + 291 + ], + "score": 0.9, + "content": "\\textstyle { \\beta = { \\frac { \\alpha - 1 } { a ^ { * } } } + 2 - \\alpha }", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 289, + 488, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 133, + 303 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 291, + 144, + 300 + ], + "score": 0.86, + "content": "a ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 289, + 305, + 303 + ], + "score": 1.0, + "content": "is the optimal treatment. This setting of", + "type": "text" + }, + { + "bbox": [ + 306, + 291, + 313, + 302 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 289, + 436, + 303 + ], + "score": 1.0, + "content": "ensures that the mode of Beta", + "type": "text" + }, + { + "bbox": [ + 437, + 291, + 461, + 302 + ], + "score": 0.88, + "content": "( \\alpha , \\tilde { \\beta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 289, + 472, + 303 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 473, + 291, + 483, + 300 + ], + "score": 0.85, + "content": "a ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 289, + 488, + 303 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 108, + 307, + 435, + 318 + ], + "lines": [ + { + "bbox": [ + 105, + 306, + 437, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 437, + 320 + ], + "score": 1.0, + "content": "The calculation of treatment response and optimal treatment are shown in Table 3.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "table", + "bbox": [ + 106, + 364, + 502, + 453 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 342, + 503, + 363 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 340, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 104, + 340, + 468, + 355 + ], + "score": 1.0, + "content": "Table 3: Treatment response used to generate semi-synthetic outcomes for patient features", + "type": "text" + }, + { + "bbox": [ + 468, + 344, + 475, + 352 + ], + "score": 0.49, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 340, + 505, + 355 + ], + "score": 1.0, + "content": ". In the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 353, + 223, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 187, + 364 + ], + "score": 1.0, + "content": "experiments, we set", + "type": "text" + }, + { + "bbox": [ + 188, + 353, + 220, + 363 + ], + "score": 0.88, + "content": "C = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 354, + 223, + 364 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 364, + 502, + 453 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 364, + 502, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 502, + 453 + ], + "score": 0.98, + "html": "
TreatmentTreatment ResponseOptimal treatment
1f1(x,a1)=C((v)Tx+12(ν)T xa1-12(ν)T x²)(v)x a= 2(v3)x
2a2))(v)x 2(v)x
3(v2)x) f3(x,a3)=C((vi)Tx+12a3(a3 -b)²,where b=0.75
", + "type": "table", + "image_path": "c7b8468ffd79d11fa3b999eb55c75879cc969434cf2cccead0c129b634700ed5.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 106, + 364, + 502, + 393.6666666666667 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 106, + 393.6666666666667, + 502, + 423.33333333333337 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 106, + 423.33333333333337, + 502, + 453.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 9.75 + }, + { + "type": "text", + "bbox": [ + 106, + 468, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 486, + 480 + ], + "score": 1.0, + "content": "We conduct experiments on three different treatments separately and change the value of bias", + "type": "text" + }, + { + "bbox": [ + 486, + 470, + 493, + 478 + ], + "score": 0.77, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 478, + 507, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 507, + 492 + ], + "score": 1.0, + "content": "assess the robustness of different methods to treatment bias. When the bias of treatment is large,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 488, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 161, + 502 + ], + "score": 1.0, + "content": "which means", + "type": "text" + }, + { + "bbox": [ + 161, + 491, + 169, + 500 + ], + "score": 0.8, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 488, + 505, + 502 + ], + "score": 1.0, + "content": "is large, the training set contains data with a strong bias on treatment so it would be", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 499, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 104, + 499, + 506, + 515 + ], + "score": 1.0, + "content": "difficult for models to appropriately predict the treatment responses out of the distribution of training", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 510, + 129, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 129, + 525 + ], + "score": 1.0, + "content": "data.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 540, + 386, + 551 + ], + "lines": [ + { + "bbox": [ + 106, + 539, + 386, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 386, + 553 + ], + "score": 1.0, + "content": "G.1.3 BUDGET ALLOCATION TASK TO THE TIME PERIOD (BAT)", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 560, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "We deploy GALILEO in a real-world large-scale food-delivery platform. The platform contains", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 572, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 583 + ], + "score": 1.0, + "content": "various food stores, and food delivery clerks. The overall workflow is as follows: the platform", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "presents the nearby food stores to the customers and the customers make orders, i.e., purchase", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "take-out foods from some stores on the platform. The food delivery clerks can select orders from", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "the platform to fulfill. After an order is selected to fulfill, the delivery clerks will take the ordered", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "take-out foods from the stores and then send the food to the customers. The platform will pay the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "score": 1.0, + "content": "delivery clerks (mainly in proportion to the distance between the store and the customers’ location)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 637, + 440, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 440, + 650 + ], + "score": 1.0, + "content": "once the orders are fulfilled. An illustration of the workflow can be found in Fig. 9.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "However, there is an imbalance problem between the demanded orders from customers and the supply", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "of delivery clerks to fulfill these orders. For example, at peak times like lunchtime, there will be", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "many more demanded orders than at other times, and the existed delivery clerks might not be able to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "fulfill all of these orders timely. The goal of the Budget Allocation task to the Time period (BAT) is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "to handle the imbalance problem in time periods by sending reasonable allowances to different time", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "periods. More precisely, the goal of BAT is to make all orders (i.e., the demand) sent in different time", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 721, + 298, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 298, + 733 + ], + "score": 1.0, + "content": "periods can be fulfilled (i.e., the supply) timely.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30 + } + ], + "page_idx": 27, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "28", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 145, + 98, + 454, + 219 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 145, + 98, + 454, + 219 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 145, + 98, + 454, + 219 + ], + "spans": [ + { + "bbox": [ + 145, + 98, + 454, + 219 + ], + "score": 0.959, + "type": "image", + "image_path": "55abe1b08792484f5a786bb75b1b3aab1b1d67d16735214a33fde14020f480e8.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 145, + 98, + 454, + 138.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 145, + 138.33333333333334, + 454, + 178.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 145, + 178.66666666666669, + 454, + 219.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 231, + 505, + 255 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 231, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 244 + ], + "score": 1.0, + "content": "Figure 8: Illustration of information about the collected dataset in GNFC. Each color of the line", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 242, + 461, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 288, + 257 + ], + "score": 1.0, + "content": "denotes one of the collected trajectories. The", + "type": "text" + }, + { + "bbox": [ + 288, + 243, + 297, + 253 + ], + "score": 0.46, + "content": "\\mathrm { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 242, + 461, + 257 + ], + "score": 1.0, + "content": "-axis denotes the timestep of a trajectory.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 279, + 505, + 302 + ], + "lines": [], + "index": 5.5, + "bbox_fs": [ + 104, + 276, + 503, + 303 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 108, + 307, + 435, + 318 + ], + "lines": [ + { + "bbox": [ + 105, + 306, + 437, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 437, + 320 + ], + "score": 1.0, + "content": "The calculation of treatment response and optimal treatment are shown in Table 3.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 306, + 437, + 320 + ] + }, + { + "type": "table", + "bbox": [ + 106, + 364, + 502, + 453 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 342, + 503, + 363 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 340, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 104, + 340, + 468, + 355 + ], + "score": 1.0, + "content": "Table 3: Treatment response used to generate semi-synthetic outcomes for patient features", + "type": "text" + }, + { + "bbox": [ + 468, + 344, + 475, + 352 + ], + "score": 0.49, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 340, + 505, + 355 + ], + "score": 1.0, + "content": ". In the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 353, + 223, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 187, + 364 + ], + "score": 1.0, + "content": "experiments, we set", + "type": "text" + }, + { + "bbox": [ + 188, + 353, + 220, + 363 + ], + "score": 0.88, + "content": "C = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 354, + 223, + 364 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "table_body", + "bbox": [ + 106, + 364, + 502, + 453 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 364, + 502, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 502, + 453 + ], + "score": 0.98, + "html": "
TreatmentTreatment ResponseOptimal treatment
1f1(x,a1)=C((v)Tx+12(ν)T xa1-12(ν)T x²)(v)x a= 2(v3)x
2a2))(v)x 2(v)x
3(v2)x) f3(x,a3)=C((vi)Tx+12a3(a3 -b)²,where b=0.75
", + "type": "table", + "image_path": "c7b8468ffd79d11fa3b999eb55c75879cc969434cf2cccead0c129b634700ed5.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 106, + 364, + 502, + 393.6666666666667 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 106, + 393.6666666666667, + 502, + 423.33333333333337 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 106, + 423.33333333333337, + 502, + 453.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 9.75 + }, + { + "type": "text", + "bbox": [ + 106, + 468, + 505, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 486, + 480 + ], + "score": 1.0, + "content": "We conduct experiments on three different treatments separately and change the value of bias", + "type": "text" + }, + { + "bbox": [ + 486, + 470, + 493, + 478 + ], + "score": 0.77, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 478, + 507, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 507, + 492 + ], + "score": 1.0, + "content": "assess the robustness of different methods to treatment bias. When the bias of treatment is large,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 488, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 161, + 502 + ], + "score": 1.0, + "content": "which means", + "type": "text" + }, + { + "bbox": [ + 161, + 491, + 169, + 500 + ], + "score": 0.8, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 488, + 505, + 502 + ], + "score": 1.0, + "content": "is large, the training set contains data with a strong bias on treatment so it would be", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 499, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 104, + 499, + 506, + 515 + ], + "score": 1.0, + "content": "difficult for models to appropriately predict the treatment responses out of the distribution of training", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 510, + 129, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 129, + 525 + ], + "score": 1.0, + "content": "data.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15, + "bbox_fs": [ + 104, + 467, + 507, + 525 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 540, + 386, + 551 + ], + "lines": [ + { + "bbox": [ + 106, + 539, + 386, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 386, + 553 + ], + "score": 1.0, + "content": "G.1.3 BUDGET ALLOCATION TASK TO THE TIME PERIOD (BAT)", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 560, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "We deploy GALILEO in a real-world large-scale food-delivery platform. The platform contains", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 572, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 583 + ], + "score": 1.0, + "content": "various food stores, and food delivery clerks. The overall workflow is as follows: the platform", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "presents the nearby food stores to the customers and the customers make orders, i.e., purchase", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "take-out foods from some stores on the platform. The food delivery clerks can select orders from", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "the platform to fulfill. After an order is selected to fulfill, the delivery clerks will take the ordered", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "take-out foods from the stores and then send the food to the customers. The platform will pay the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 639 + ], + "score": 1.0, + "content": "delivery clerks (mainly in proportion to the distance between the store and the customers’ location)", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 637, + 440, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 440, + 650 + ], + "score": 1.0, + "content": "once the orders are fulfilled. An illustration of the workflow can be found in Fig. 9.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 561, + 506, + 650 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "However, there is an imbalance problem between the demanded orders from customers and the supply", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "of delivery clerks to fulfill these orders. For example, at peak times like lunchtime, there will be", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "many more demanded orders than at other times, and the existed delivery clerks might not be able to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "fulfill all of these orders timely. The goal of the Budget Allocation task to the Time period (BAT) is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "to handle the imbalance problem in time periods by sending reasonable allowances to different time", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "periods. More precisely, the goal of BAT is to make all orders (i.e., the demand) sent in different time", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 721, + 298, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 298, + 733 + ], + "score": 1.0, + "content": "periods can be fulfilled (i.e., the supply) timely.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 653, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 164, + 82, + 445, + 252 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 164, + 82, + 445, + 252 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 164, + 82, + 445, + 252 + ], + "spans": [ + { + "bbox": [ + 164, + 82, + 445, + 252 + ], + "score": 0.973, + "type": "image", + "image_path": "b1e7c25ec14554cb2a78317429793502eb19c5fd3212ebf07b308c2649c646bd.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 164, + 82, + 445, + 138.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 164, + 138.66666666666666, + 445, + 195.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 164, + 195.33333333333331, + 445, + 251.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 169, + 265, + 441, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 168, + 263, + 442, + 279 + ], + "spans": [ + { + "bbox": [ + 168, + 263, + 442, + 279 + ], + "score": 1.0, + "content": "Figure 9: Illustration of the workflow of the food-delivery platform.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 298, + 505, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "To handle the imbalance problem in different time periods, in the platform, the orders in different", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 308, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 159, + 322 + ], + "score": 1.0, + "content": "time periods", + "type": "text" + }, + { + "bbox": [ + 159, + 309, + 227, + 321 + ], + "score": 0.83, + "content": "t \\in [ 0 , 1 , 2 . . . , 2 3 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 308, + 402, + 322 + ], + "score": 1.0, + "content": "will be allocated with different allowances", + "type": "text" + }, + { + "bbox": [ + 402, + 309, + 435, + 320 + ], + "score": 0.91, + "content": "c \\in \\mathcal { N } ^ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 308, + 506, + 322 + ], + "score": 1.0, + "content": ". For example, at", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 139, + 331 + ], + "score": 0.33, + "content": "1 0 \\mathrm { A . M }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 320, + 163, + 332 + ], + "score": 1.0, + "content": ". (i.e.,", + "type": "text" + }, + { + "bbox": [ + 163, + 321, + 192, + 331 + ], + "score": 0.87, + "content": "t = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 320, + 230, + 332 + ], + "score": 1.0, + "content": "), we add", + "type": "text" + }, + { + "bbox": [ + 231, + 321, + 250, + 331 + ], + "score": 0.82, + "content": "0 . 5 \\ S", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 320, + 271, + 332 + ], + "score": 1.0, + "content": "(i.e.,", + "type": "text" + }, + { + "bbox": [ + 272, + 321, + 304, + 331 + ], + "score": 0.87, + "content": "c = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 320, + 505, + 332 + ], + "score": 1.0, + "content": ") allowances to all of the demanded orders. From", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 330, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 139, + 342 + ], + "score": 0.31, + "content": "1 0 \\mathrm { \\ A . M }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 330, + 505, + 344 + ], + "score": 1.0, + "content": ". to 11 A.M., the delivery clerks who take orders and send food to customers will receive", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 383, + 355 + ], + "score": 1.0, + "content": "extra allowances. Specifically, if the platform pays the delivery clerks", + "type": "text" + }, + { + "bbox": [ + 383, + 342, + 394, + 353 + ], + "score": 0.67, + "content": "2 \\mathfrak { H }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 342, + 506, + 355 + ], + "score": 1.0, + "content": "for fulfilling the order, now", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 184, + 365 + ], + "score": 1.0, + "content": "he/she will receive", + "type": "text" + }, + { + "bbox": [ + 184, + 353, + 203, + 364 + ], + "score": 0.74, + "content": "2 . 5 \\ S", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 353, + 362, + 365 + ], + "score": 1.0, + "content": ". For each day, the budget of allowance", + "type": "text" + }, + { + "bbox": [ + 363, + 354, + 372, + 363 + ], + "score": 0.84, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "is fixed. We should find the best", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 363, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 205, + 378 + ], + "score": 1.0, + "content": "budget allocation policy", + "type": "text" + }, + { + "bbox": [ + 205, + 364, + 235, + 376 + ], + "score": 0.95, + "content": "\\pi ^ { * } ( c | t )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 363, + 323, + 378 + ], + "score": 1.0, + "content": "of the limited budget", + "type": "text" + }, + { + "bbox": [ + 323, + 365, + 332, + 374 + ], + "score": 0.82, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 363, + 506, + 378 + ], + "score": 1.0, + "content": "to make as many orders as possible can be", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 374, + 160, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 160, + 389 + ], + "score": 1.0, + "content": "taken timely.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 391, + 506, + 484 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 406 + ], + "score": 1.0, + "content": "To find the policy, we first learn a model to reconstruct the response of allowance for each delivery", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 403, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 130, + 418 + ], + "score": 1.0, + "content": "clerk", + "type": "text" + }, + { + "bbox": [ + 130, + 403, + 204, + 417 + ], + "score": 0.93, + "content": "\\hat { M } \\big ( y _ { t + 1 } \\vert s _ { t } , p _ { t } , c _ { t } \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 403, + 235, + 418 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 236, + 406, + 255, + 416 + ], + "score": 0.89, + "content": "y _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 403, + 456, + 418 + ], + "score": 1.0, + "content": "is the taken orders of the delivery clerks in state", + "type": "text" + }, + { + "bbox": [ + 456, + 406, + 479, + 416 + ], + "score": 0.27, + "content": "s _ { t } , c _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 403, + 506, + 418 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 154, + 428 + ], + "score": 1.0, + "content": "allowances,", + "type": "text" + }, + { + "bbox": [ + 154, + 418, + 164, + 427 + ], + "score": 0.84, + "content": "p _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 415, + 325, + 428 + ], + "score": 1.0, + "content": "denotes static features of the time period", + "type": "text" + }, + { + "bbox": [ + 325, + 416, + 330, + 425 + ], + "score": 0.67, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 415, + 420, + 428 + ], + "score": 1.0, + "content": ". In particular, the state", + "type": "text" + }, + { + "bbox": [ + 421, + 417, + 430, + 426 + ], + "score": 0.86, + "content": "s _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 415, + 506, + 428 + ], + "score": 1.0, + "content": "includes historical", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "order-taken information of the delivery clerks, current orders information, the feature of weather,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 398, + 449 + ], + "score": 1.0, + "content": "city information, and so on. Then we use a rule-based mapping function", + "type": "text" + }, + { + "bbox": [ + 399, + 438, + 406, + 449 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 438, + 505, + 449 + ], + "score": 1.0, + "content": "to fill the complete next", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 447, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 204, + 463 + ], + "score": 1.0, + "content": "time-period states, i.e.,", + "type": "text" + }, + { + "bbox": [ + 204, + 448, + 309, + 461 + ], + "score": 0.91, + "content": "s _ { t + 1 } = f ( s _ { t } , p _ { t } , c _ { t } , y _ { t + 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 447, + 506, + 463 + ], + "score": 1.0, + "content": ". Here we define the composition of the above", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 460, + 507, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 146, + 474 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 146, + 460, + 159, + 471 + ], + "score": 0.87, + "content": "\\hat { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 460, + 176, + 474 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 177, + 461, + 184, + 473 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 460, + 196, + 474 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 196, + 460, + 212, + 474 + ], + "score": 0.9, + "content": "\\hat { M } _ { f }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 460, + 507, + 474 + ], + "score": 1.0, + "content": ". Finally, we learn a budget allocation policy based on the learned model.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 473, + 303, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 303, + 484 + ], + "score": 1.0, + "content": "For each day, the policy we would like to find is:", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "interline_equation", + "bbox": [ + 245, + 490, + 366, + 553 + ], + "lines": [ + { + "bbox": [ + 245, + 490, + 366, + 553 + ], + "spans": [ + { + "bbox": [ + 245, + 490, + 366, + 553 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\underset { \\pi } { \\operatorname* { m a x } } \\mathbb { E } _ { s _ { 0 } \\sim { \\cal S } } \\left[ \\displaystyle \\sum _ { t = 0 } ^ { 2 3 } y _ { t } \\vert \\hat { M } _ { f } , \\pi \\right] , } \\\\ { \\mathrm { s . t . , } \\displaystyle \\sum _ { t , s \\in { \\cal S } } c _ { t } y _ { t } \\leq C } \\end{array}", + "type": "interline_equation", + "image_path": "4ed7580e27e1590c372d6def6b135f8cbcccb6b39699b0f23d3340c6516875e9.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 245, + 490, + 366, + 521.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 245, + 521.5, + 366, + 553.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 557, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "score": 1.0, + "content": "In our experiment, we evaluate the degree of balancing between demand and supply by computing the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 569, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 569, + 506, + 581 + ], + "score": 1.0, + "content": "averaged five-minute order-taken rate, that is the percentage of orders picked up within five minutes.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "Note that the behavior policy is fixed for the long term in this application. So we directly use the data", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 590, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 506, + 605 + ], + "score": 1.0, + "content": "replay with a small scale of noise (See Tab. 5) to reconstruct the behavior policy for model learning", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 601, + 164, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 164, + 614 + ], + "score": 1.0, + "content": "in GALILEO.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 618, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 105, + 618, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 505, + 632 + ], + "score": 1.0, + "content": "Also note that although we model the response for each delivery clerk, for fairness, the budget", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 630, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 630, + 396, + 642 + ], + "score": 1.0, + "content": "allocation policy is just determining the allowance of each time period", + "type": "text" + }, + { + "bbox": [ + 397, + 631, + 402, + 640 + ], + "score": 0.67, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 630, + 505, + 642 + ], + "score": 1.0, + "content": "and keeps the allowance", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 641, + 250, + 654 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 199, + 654 + ], + "score": 1.0, + "content": "to each delivery clerk", + "type": "text" + }, + { + "bbox": [ + 200, + 643, + 206, + 651 + ], + "score": 0.68, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 641, + 250, + 654 + ], + "score": 1.0, + "content": "the same.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "title", + "bbox": [ + 107, + 667, + 239, + 678 + ], + "lines": [ + { + "bbox": [ + 106, + 666, + 240, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 240, + 679 + ], + "score": 1.0, + "content": "G.2 BASELINE ALGORITHMS", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 734 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "The algorithm we compared are: (1) Supervised Learning (SL): training a environment model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "to minimize the expectation of prediction error, without considering the counterfactual risks; (2)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "inverse propensity weighting (IPW) (Spirtes, 2010): a practical way to balance the selection bias", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 720, + 505, + 735 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 266, + 735 + ], + "score": 1.0, + "content": "by re-weighting. It can be regarded as", + "type": "text" + }, + { + "bbox": [ + 267, + 720, + 295, + 735 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\dot { \\omega } = \\frac { 1 } { \\hat { \\mu } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 720, + 328, + 735 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 328, + 721, + 335, + 732 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 720, + 505, + 735 + ], + "score": 1.0, + "content": "is another model learned to approximate", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + } + ], + "page_idx": 28, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 164, + 82, + 445, + 252 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 164, + 82, + 445, + 252 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 164, + 82, + 445, + 252 + ], + "spans": [ + { + "bbox": [ + 164, + 82, + 445, + 252 + ], + "score": 0.973, + "type": "image", + "image_path": "b1e7c25ec14554cb2a78317429793502eb19c5fd3212ebf07b308c2649c646bd.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 164, + 82, + 445, + 138.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 164, + 138.66666666666666, + 445, + 195.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 164, + 195.33333333333331, + 445, + 251.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 169, + 265, + 441, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 168, + 263, + 442, + 279 + ], + "spans": [ + { + "bbox": [ + 168, + 263, + 442, + 279 + ], + "score": 1.0, + "content": "Figure 9: Illustration of the workflow of the food-delivery platform.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 106, + 298, + 505, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "To handle the imbalance problem in different time periods, in the platform, the orders in different", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 308, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 159, + 322 + ], + "score": 1.0, + "content": "time periods", + "type": "text" + }, + { + "bbox": [ + 159, + 309, + 227, + 321 + ], + "score": 0.83, + "content": "t \\in [ 0 , 1 , 2 . . . , 2 3 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 308, + 402, + 322 + ], + "score": 1.0, + "content": "will be allocated with different allowances", + "type": "text" + }, + { + "bbox": [ + 402, + 309, + 435, + 320 + ], + "score": 0.91, + "content": "c \\in \\mathcal { N } ^ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 308, + 506, + 322 + ], + "score": 1.0, + "content": ". For example, at", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 139, + 331 + ], + "score": 0.33, + "content": "1 0 \\mathrm { A . M }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 320, + 163, + 332 + ], + "score": 1.0, + "content": ". (i.e.,", + "type": "text" + }, + { + "bbox": [ + 163, + 321, + 192, + 331 + ], + "score": 0.87, + "content": "t = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 320, + 230, + 332 + ], + "score": 1.0, + "content": "), we add", + "type": "text" + }, + { + "bbox": [ + 231, + 321, + 250, + 331 + ], + "score": 0.82, + "content": "0 . 5 \\ S", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 320, + 271, + 332 + ], + "score": 1.0, + "content": "(i.e.,", + "type": "text" + }, + { + "bbox": [ + 272, + 321, + 304, + 331 + ], + "score": 0.87, + "content": "c = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 320, + 505, + 332 + ], + "score": 1.0, + "content": ") allowances to all of the demanded orders. From", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 330, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 139, + 342 + ], + "score": 0.31, + "content": "1 0 \\mathrm { \\ A . M }", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 330, + 505, + 344 + ], + "score": 1.0, + "content": ". to 11 A.M., the delivery clerks who take orders and send food to customers will receive", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 383, + 355 + ], + "score": 1.0, + "content": "extra allowances. Specifically, if the platform pays the delivery clerks", + "type": "text" + }, + { + "bbox": [ + 383, + 342, + 394, + 353 + ], + "score": 0.67, + "content": "2 \\mathfrak { H }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 342, + 506, + 355 + ], + "score": 1.0, + "content": "for fulfilling the order, now", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 184, + 365 + ], + "score": 1.0, + "content": "he/she will receive", + "type": "text" + }, + { + "bbox": [ + 184, + 353, + 203, + 364 + ], + "score": 0.74, + "content": "2 . 5 \\ S", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 353, + 362, + 365 + ], + "score": 1.0, + "content": ". For each day, the budget of allowance", + "type": "text" + }, + { + "bbox": [ + 363, + 354, + 372, + 363 + ], + "score": 0.84, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "is fixed. We should find the best", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 363, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 205, + 378 + ], + "score": 1.0, + "content": "budget allocation policy", + "type": "text" + }, + { + "bbox": [ + 205, + 364, + 235, + 376 + ], + "score": 0.95, + "content": "\\pi ^ { * } ( c | t )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 363, + 323, + 378 + ], + "score": 1.0, + "content": "of the limited budget", + "type": "text" + }, + { + "bbox": [ + 323, + 365, + 332, + 374 + ], + "score": 0.82, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 363, + 506, + 378 + ], + "score": 1.0, + "content": "to make as many orders as possible can be", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 374, + 160, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 160, + 389 + ], + "score": 1.0, + "content": "taken timely.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 298, + 506, + 389 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 391, + 506, + 484 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 406 + ], + "score": 1.0, + "content": "To find the policy, we first learn a model to reconstruct the response of allowance for each delivery", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 403, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 130, + 418 + ], + "score": 1.0, + "content": "clerk", + "type": "text" + }, + { + "bbox": [ + 130, + 403, + 204, + 417 + ], + "score": 0.93, + "content": "\\hat { M } \\big ( y _ { t + 1 } \\vert s _ { t } , p _ { t } , c _ { t } \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 403, + 235, + 418 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 236, + 406, + 255, + 416 + ], + "score": 0.89, + "content": "y _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 403, + 456, + 418 + ], + "score": 1.0, + "content": "is the taken orders of the delivery clerks in state", + "type": "text" + }, + { + "bbox": [ + 456, + 406, + 479, + 416 + ], + "score": 0.27, + "content": "s _ { t } , c _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 403, + 506, + 418 + ], + "score": 1.0, + "content": "is the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 415, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 154, + 428 + ], + "score": 1.0, + "content": "allowances,", + "type": "text" + }, + { + "bbox": [ + 154, + 418, + 164, + 427 + ], + "score": 0.84, + "content": "p _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 415, + 325, + 428 + ], + "score": 1.0, + "content": "denotes static features of the time period", + "type": "text" + }, + { + "bbox": [ + 325, + 416, + 330, + 425 + ], + "score": 0.67, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 415, + 420, + 428 + ], + "score": 1.0, + "content": ". In particular, the state", + "type": "text" + }, + { + "bbox": [ + 421, + 417, + 430, + 426 + ], + "score": 0.86, + "content": "s _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 415, + 506, + 428 + ], + "score": 1.0, + "content": "includes historical", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "order-taken information of the delivery clerks, current orders information, the feature of weather,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 398, + 449 + ], + "score": 1.0, + "content": "city information, and so on. Then we use a rule-based mapping function", + "type": "text" + }, + { + "bbox": [ + 399, + 438, + 406, + 449 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 438, + 505, + 449 + ], + "score": 1.0, + "content": "to fill the complete next", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 447, + 506, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 204, + 463 + ], + "score": 1.0, + "content": "time-period states, i.e.,", + "type": "text" + }, + { + "bbox": [ + 204, + 448, + 309, + 461 + ], + "score": 0.91, + "content": "s _ { t + 1 } = f ( s _ { t } , p _ { t } , c _ { t } , y _ { t + 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 447, + 506, + 463 + ], + "score": 1.0, + "content": ". Here we define the composition of the above", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 460, + 507, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 146, + 474 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 146, + 460, + 159, + 471 + ], + "score": 0.87, + "content": "\\hat { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 460, + 176, + 474 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 177, + 461, + 184, + 473 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 460, + 196, + 474 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 196, + 460, + 212, + 474 + ], + "score": 0.9, + "content": "\\hat { M } _ { f }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 460, + 507, + 474 + ], + "score": 1.0, + "content": ". Finally, we learn a budget allocation policy based on the learned model.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 473, + 303, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 473, + 303, + 484 + ], + "score": 1.0, + "content": "For each day, the policy we would like to find is:", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 390, + 507, + 484 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 245, + 490, + 366, + 553 + ], + "lines": [ + { + "bbox": [ + 245, + 490, + 366, + 553 + ], + "spans": [ + { + "bbox": [ + 245, + 490, + 366, + 553 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\underset { \\pi } { \\operatorname* { m a x } } \\mathbb { E } _ { s _ { 0 } \\sim { \\cal S } } \\left[ \\displaystyle \\sum _ { t = 0 } ^ { 2 3 } y _ { t } \\vert \\hat { M } _ { f } , \\pi \\right] , } \\\\ { \\mathrm { s . t . , } \\displaystyle \\sum _ { t , s \\in { \\cal S } } c _ { t } y _ { t } \\leq C } \\end{array}", + "type": "interline_equation", + "image_path": "4ed7580e27e1590c372d6def6b135f8cbcccb6b39699b0f23d3340c6516875e9.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 245, + 490, + 366, + 521.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 245, + 521.5, + 366, + 553.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 557, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "score": 1.0, + "content": "In our experiment, we evaluate the degree of balancing between demand and supply by computing the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 569, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 569, + 506, + 581 + ], + "score": 1.0, + "content": "averaged five-minute order-taken rate, that is the percentage of orders picked up within five minutes.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "Note that the behavior policy is fixed for the long term in this application. So we directly use the data", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 590, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 506, + 605 + ], + "score": 1.0, + "content": "replay with a small scale of noise (See Tab. 5) to reconstruct the behavior policy for model learning", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 601, + 164, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 164, + 614 + ], + "score": 1.0, + "content": "in GALILEO.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 558, + 506, + 614 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 618, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 105, + 618, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 505, + 632 + ], + "score": 1.0, + "content": "Also note that although we model the response for each delivery clerk, for fairness, the budget", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 630, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 630, + 396, + 642 + ], + "score": 1.0, + "content": "allocation policy is just determining the allowance of each time period", + "type": "text" + }, + { + "bbox": [ + 397, + 631, + 402, + 640 + ], + "score": 0.67, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 630, + 505, + 642 + ], + "score": 1.0, + "content": "and keeps the allowance", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 641, + 250, + 654 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 199, + 654 + ], + "score": 1.0, + "content": "to each delivery clerk", + "type": "text" + }, + { + "bbox": [ + 200, + 643, + 206, + 651 + ], + "score": 0.68, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 641, + 250, + 654 + ], + "score": 1.0, + "content": "the same.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 618, + 505, + 654 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 667, + 239, + 678 + ], + "lines": [ + { + "bbox": [ + 106, + 666, + 240, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 666, + 240, + 679 + ], + "score": 1.0, + "content": "G.2 BASELINE ALGORITHMS", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 734 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 700 + ], + "score": 1.0, + "content": "The algorithm we compared are: (1) Supervised Learning (SL): training a environment model", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "to minimize the expectation of prediction error, without considering the counterfactual risks; (2)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "inverse propensity weighting (IPW) (Spirtes, 2010): a practical way to balance the selection bias", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 720, + 505, + 735 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 266, + 735 + ], + "score": 1.0, + "content": "by re-weighting. It can be regarded as", + "type": "text" + }, + { + "bbox": [ + 267, + 720, + 295, + 735 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\dot { \\omega } = \\frac { 1 } { \\hat { \\mu } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 720, + 328, + 735 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 328, + 721, + 335, + 732 + ], + "score": 0.85, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 720, + 505, + 735 + ], + "score": 1.0, + "content": "is another model learned to approximate", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "the behavior policy; (3) SCIGAN: a recent proposed adversarial algorithm for model learning for", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "continuous-valued interventions (Bica et al., 2020). All of the baselines algorithms are implemented", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 329, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 329, + 117 + ], + "score": 1.0, + "content": "with the same capacity of neural networks (See Tab. 5).", + "type": "text", + "cross_page": true + } + ], + "index": 2 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 688, + 506, + 735 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "the behavior policy; (3) SCIGAN: a recent proposed adversarial algorithm for model learning for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 505, + 105 + ], + "score": 1.0, + "content": "continuous-valued interventions (Bica et al., 2020). All of the baselines algorithms are implemented", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 105, + 329, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 329, + 117 + ], + "score": 1.0, + "content": "with the same capacity of neural networks (See Tab. 5).", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 108, + 132, + 267, + 144 + ], + "lines": [ + { + "bbox": [ + 105, + 130, + 268, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 268, + 147 + ], + "score": 1.0, + "content": "G.2.1 SUPERVISED LEARNING (SL)", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 154, + 506, + 222 + ], + "lines": [ + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "As a baseline, we train a multilayer perceptron model to directly predict the response of different", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 166, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 505, + 177 + ], + "score": 1.0, + "content": "treatments, without considering the counterfactual risks. We use mean square error to estimate the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 101, + 173, + 502, + 197 + ], + "spans": [ + { + "bbox": [ + 101, + 173, + 132, + 197 + ], + "score": 1.0, + "content": "perfor", + "type": "text" + }, + { + "bbox": [ + 141, + 173, + 387, + 197 + ], + "score": 1.0, + "content": "ance of our model so that the loss function can be expressed as", + "type": "text" + }, + { + "bbox": [ + 387, + 176, + 502, + 191 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\boldsymbol { M } \\boldsymbol { \\bar { S } } \\boldsymbol { E } = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\left( y _ { i } - \\hat { y } _ { i } \\right) ^ { 2 } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 133, + 189, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 133, + 191, + 140, + 199 + ], + "score": 0.74, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 190, + 390, + 200 + ], + "score": 0.84, + "content": "\\hat { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 189, + 505, + 201 + ], + "score": 1.0, + "content": "is the predicted response. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 199, + 506, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 448, + 212 + ], + "score": 1.0, + "content": "practice, we train our SL models using Adam optimizer and the initial learning rate", + "type": "text" + }, + { + "bbox": [ + 449, + 199, + 470, + 210 + ], + "score": 0.9, + "content": "3 e ^ { - \\bar { 4 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 199, + 506, + 212 + ], + "score": 1.0, + "content": "on both", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 210, + 450, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 450, + 222 + ], + "score": 1.0, + "content": "datasets TCGA and GNFC. The architecture of the neural networks is listed in Tab. 5.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6.5 + }, + { + "type": "title", + "bbox": [ + 107, + 239, + 321, + 250 + ], + "lines": [ + { + "bbox": [ + 105, + 238, + 322, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 322, + 253 + ], + "score": 1.0, + "content": "G.2.2 INVERSE PROPENSITY WEIGHTING (IPW)", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 505, + 377 + ], + "lines": [ + { + "bbox": [ + 106, + 260, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 506, + 273 + ], + "score": 1.0, + "content": "Inverse propensity weighting (Spirtes, 2010) is an approach where the treatment outcome model", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "uses sample weights to balance the selection bias by re-weighting. The weights are defined as the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 281, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 104, + 281, + 415, + 298 + ], + "score": 1.0, + "content": "inverse propensity of actually getting the treatment, which can be expressed as", + "type": "text" + }, + { + "bbox": [ + 415, + 282, + 440, + 297 + ], + "score": 0.93, + "content": "\\dot { \\overline { { { \\hat { \\mu } } ( a | x ) } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 281, + 470, + 298 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 470, + 284, + 477, + 292 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 281, + 505, + 298 + ], + "score": 1.0, + "content": "stands", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 295, + 504, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 252, + 308 + ], + "score": 1.0, + "content": "for the feature vectors in a dataset,", + "type": "text" + }, + { + "bbox": [ + 253, + 298, + 259, + 306 + ], + "score": 0.7, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 295, + 392, + 308 + ], + "score": 1.0, + "content": "is the corresponding action and", + "type": "text" + }, + { + "bbox": [ + 393, + 296, + 421, + 308 + ], + "score": 0.91, + "content": "{ \\hat { \\mu } } ( a | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 295, + 504, + 308 + ], + "score": 1.0, + "content": "indicates the action", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 307, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 188, + 319 + ], + "score": 1.0, + "content": "taken probability of", + "type": "text" + }, + { + "bbox": [ + 188, + 309, + 195, + 317 + ], + "score": 0.71, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 307, + 268, + 319 + ], + "score": 1.0, + "content": "given the features", + "type": "text" + }, + { + "bbox": [ + 269, + 309, + 276, + 317 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 307, + 353, + 319 + ], + "score": 1.0, + "content": "within the dataset.", + "type": "text" + }, + { + "bbox": [ + 353, + 308, + 361, + 319 + ], + "score": 0.83, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 307, + 505, + 319 + ], + "score": 1.0, + "content": "is learned with standard supervised", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 318, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 506, + 330 + ], + "score": 1.0, + "content": "learning. Standard IPW leads to large weights for the points with small sampling probabilities and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 327, + 507, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 507, + 342 + ], + "score": 1.0, + "content": "finally makes the learning process unstable. We solve the problem by clipping the propensity score:", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 339, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 107, + 340, + 181, + 352 + ], + "score": 0.91, + "content": "\\hat { \\mu } \\gets \\operatorname* { m i n } ( \\hat { \\mu } , 0 . 0 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 339, + 506, + 353 + ], + "score": 1.0, + "content": ", which is common used in existing studies (Ionides, 2008). The loss function can", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 103, + 349, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 103, + 349, + 260, + 369 + ], + "score": 1.0, + "content": "thus be expressed as 1n Pni=1 1µˆ(ai|xi)", + "type": "text" + }, + { + "bbox": [ + 191, + 351, + 299, + 367 + ], + "score": 0.93, + "content": "\\begin{array} { r } { { \\frac { 1 } { n } } \\sum _ { i = 1 } ^ { n } { \\frac { 1 } { { \\hat { \\mu } } \\left( a _ { i } | x _ { i } \\right) } } \\left( y _ { i } - \\hat { y } _ { i } \\right) ^ { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 351, + 506, + 365 + ], + "score": 1.0, + "content": ". The architecture of the neural networks is listed in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 363, + 136, + 378 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 136, + 378 + ], + "score": 1.0, + "content": "Tab. 5.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 107, + 393, + 182, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 184, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 184, + 407 + ], + "score": 1.0, + "content": "G.2.3 SCIGAN", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 415, + 505, + 525 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "SCIGAN (Bica et al., 2020) is a model that uses generative adversarial networks to learn the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 425, + 507, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 507, + 439 + ], + "score": 1.0, + "content": "data distribution of the counterfactual outcomes and thus generate individualized response curves.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "score": 1.0, + "content": "SCIGAN does not place any restrictions on the form of the treatment-does response functions", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 448, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 506, + 461 + ], + "score": 1.0, + "content": "and is capable of estimating patient outcomes for multiple treatments, each with an associated", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 459, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 472 + ], + "score": 1.0, + "content": "parameter. SCIGAN first trains a generator to generate response curves for each sample within", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 469, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 484 + ], + "score": 1.0, + "content": "the training dataset. The learned generator can then be used to train an inference network using", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "score": 1.0, + "content": "standard supervised methods. For fair comparison, we increase the number of parameters for the the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "score": 1.0, + "content": "open-source version of SCIGAN so that the SCIGAN model can have same order of magnitude of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 503, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 506, + 516 + ], + "score": 1.0, + "content": "network parameters as GALILEO. In addition, we also finetune the hyperparameters (Tab. 4) of the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 514, + 465, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 431, + 527 + ], + "score": 1.0, + "content": "enlarged SCIGAN to realize its full strength. We set num_dosage_samples 9 and", + "type": "text" + }, + { + "bbox": [ + 432, + 514, + 461, + 524 + ], + "score": 0.9, + "content": "\\lambda = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 514, + 465, + 527 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26.5 + }, + { + "type": "table", + "bbox": [ + 225, + 561, + 383, + 606 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 205, + 549, + 402, + 561 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 204, + 549, + 403, + 563 + ], + "spans": [ + { + "bbox": [ + 204, + 549, + 403, + 563 + ], + "score": 1.0, + "content": "Table 4: Table of hyper-parameters for SCIGAN.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "table_body", + "bbox": [ + 225, + 561, + 383, + 606 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 225, + 561, + 383, + 606 + ], + "spans": [ + { + "bbox": [ + 225, + 561, + 383, + 606 + ], + "score": 0.975, + "html": "
ParameterValues
Number of samples3,5,7,9, 11
入0.1,1,10, 20
", + "type": "table", + "image_path": "8ec5dc6c111badbd1e205f3b3f1e840f45d4e74ac912ed430f8883e738032a52.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 225, + 561, + 383, + 576.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 225, + 576.0, + 383, + 591.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 225, + 591.0, + 383, + 606.0 + ], + "spans": [], + "index": 35 + } + ] + } + ], + "index": 33.0 + }, + { + "type": "title", + "bbox": [ + 107, + 634, + 225, + 646 + ], + "lines": [ + { + "bbox": [ + 106, + 634, + 226, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 226, + 646 + ], + "score": 1.0, + "content": "G.3 HYPER-PARAMETERS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 313, + 669 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 315, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 315, + 669 + ], + "score": 1.0, + "content": "We list the hyper-parameter of GALILEO in Tab. 5.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "title", + "bbox": [ + 108, + 686, + 252, + 698 + ], + "lines": [ + { + "bbox": [ + 106, + 685, + 253, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 685, + 253, + 699 + ], + "score": 1.0, + "content": "G.4 COMPUTATION RESOURCES", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "We use one Tesla V100 PCIe 32GB GPU and a 32-core Intel(R) Xeon(R) Gold 5118 CPU @ 2.30GHz", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 721, + 206, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 206, + 732 + ], + "score": 1.0, + "content": "to train all of our model.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + } + ], + "page_idx": 29, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "30", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 116 + ], + "lines": [], + "index": 1, + "bbox_fs": [ + 105, + 82, + 505, + 117 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 132, + 267, + 144 + ], + "lines": [ + { + "bbox": [ + 105, + 130, + 268, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 268, + 147 + ], + "score": 1.0, + "content": "G.2.1 SUPERVISED LEARNING (SL)", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 154, + 506, + 222 + ], + "lines": [ + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "As a baseline, we train a multilayer perceptron model to directly predict the response of different", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 166, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 166, + 505, + 177 + ], + "score": 1.0, + "content": "treatments, without considering the counterfactual risks. We use mean square error to estimate the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 101, + 173, + 502, + 197 + ], + "spans": [ + { + "bbox": [ + 101, + 173, + 132, + 197 + ], + "score": 1.0, + "content": "perfor", + "type": "text" + }, + { + "bbox": [ + 141, + 173, + 387, + 197 + ], + "score": 1.0, + "content": "ance of our model so that the loss function can be expressed as", + "type": "text" + }, + { + "bbox": [ + 387, + 176, + 502, + 191 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\boldsymbol { M } \\boldsymbol { \\bar { S } } \\boldsymbol { E } = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\left( y _ { i } - \\hat { y } _ { i } \\right) ^ { 2 } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 133, + 189, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 133, + 191, + 140, + 199 + ], + "score": 0.74, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 190, + 390, + 200 + ], + "score": 0.84, + "content": "\\hat { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 189, + 505, + 201 + ], + "score": 1.0, + "content": "is the predicted response. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 199, + 506, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 448, + 212 + ], + "score": 1.0, + "content": "practice, we train our SL models using Adam optimizer and the initial learning rate", + "type": "text" + }, + { + "bbox": [ + 449, + 199, + 470, + 210 + ], + "score": 0.9, + "content": "3 e ^ { - \\bar { 4 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 199, + 506, + 212 + ], + "score": 1.0, + "content": "on both", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 210, + 450, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 450, + 222 + ], + "score": 1.0, + "content": "datasets TCGA and GNFC. The architecture of the neural networks is listed in Tab. 5.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6.5, + "bbox_fs": [ + 101, + 154, + 506, + 222 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 239, + 321, + 250 + ], + "lines": [ + { + "bbox": [ + 105, + 238, + 322, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 238, + 322, + 253 + ], + "score": 1.0, + "content": "G.2.2 INVERSE PROPENSITY WEIGHTING (IPW)", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 505, + 377 + ], + "lines": [ + { + "bbox": [ + 106, + 260, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 506, + 273 + ], + "score": 1.0, + "content": "Inverse propensity weighting (Spirtes, 2010) is an approach where the treatment outcome model", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "uses sample weights to balance the selection bias by re-weighting. The weights are defined as the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 281, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 104, + 281, + 415, + 298 + ], + "score": 1.0, + "content": "inverse propensity of actually getting the treatment, which can be expressed as", + "type": "text" + }, + { + "bbox": [ + 415, + 282, + 440, + 297 + ], + "score": 0.93, + "content": "\\dot { \\overline { { { \\hat { \\mu } } ( a | x ) } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 281, + 470, + 298 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 470, + 284, + 477, + 292 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 281, + 505, + 298 + ], + "score": 1.0, + "content": "stands", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 295, + 504, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 295, + 252, + 308 + ], + "score": 1.0, + "content": "for the feature vectors in a dataset,", + "type": "text" + }, + { + "bbox": [ + 253, + 298, + 259, + 306 + ], + "score": 0.7, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 295, + 392, + 308 + ], + "score": 1.0, + "content": "is the corresponding action and", + "type": "text" + }, + { + "bbox": [ + 393, + 296, + 421, + 308 + ], + "score": 0.91, + "content": "{ \\hat { \\mu } } ( a | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 295, + 504, + 308 + ], + "score": 1.0, + "content": "indicates the action", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 307, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 188, + 319 + ], + "score": 1.0, + "content": "taken probability of", + "type": "text" + }, + { + "bbox": [ + 188, + 309, + 195, + 317 + ], + "score": 0.71, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 307, + 268, + 319 + ], + "score": 1.0, + "content": "given the features", + "type": "text" + }, + { + "bbox": [ + 269, + 309, + 276, + 317 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 307, + 353, + 319 + ], + "score": 1.0, + "content": "within the dataset.", + "type": "text" + }, + { + "bbox": [ + 353, + 308, + 361, + 319 + ], + "score": 0.83, + "content": "\\hat { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 307, + 505, + 319 + ], + "score": 1.0, + "content": "is learned with standard supervised", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 318, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 506, + 330 + ], + "score": 1.0, + "content": "learning. Standard IPW leads to large weights for the points with small sampling probabilities and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 327, + 507, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 507, + 342 + ], + "score": 1.0, + "content": "finally makes the learning process unstable. We solve the problem by clipping the propensity score:", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 339, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 107, + 340, + 181, + 352 + ], + "score": 0.91, + "content": "\\hat { \\mu } \\gets \\operatorname* { m i n } ( \\hat { \\mu } , 0 . 0 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 339, + 506, + 353 + ], + "score": 1.0, + "content": ", which is common used in existing studies (Ionides, 2008). The loss function can", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 103, + 349, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 103, + 349, + 260, + 369 + ], + "score": 1.0, + "content": "thus be expressed as 1n Pni=1 1µˆ(ai|xi)", + "type": "text" + }, + { + "bbox": [ + 191, + 351, + 299, + 367 + ], + "score": 0.93, + "content": "\\begin{array} { r } { { \\frac { 1 } { n } } \\sum _ { i = 1 } ^ { n } { \\frac { 1 } { { \\hat { \\mu } } \\left( a _ { i } | x _ { i } \\right) } } \\left( y _ { i } - \\hat { y } _ { i } \\right) ^ { 2 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 351, + 506, + 365 + ], + "score": 1.0, + "content": ". The architecture of the neural networks is listed in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 363, + 136, + 378 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 136, + 378 + ], + "score": 1.0, + "content": "Tab. 5.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 15.5, + "bbox_fs": [ + 103, + 260, + 507, + 378 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 393, + 182, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 184, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 184, + 407 + ], + "score": 1.0, + "content": "G.2.3 SCIGAN", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 415, + 505, + 525 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "SCIGAN (Bica et al., 2020) is a model that uses generative adversarial networks to learn the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 425, + 507, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 507, + 439 + ], + "score": 1.0, + "content": "data distribution of the counterfactual outcomes and thus generate individualized response curves.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 449 + ], + "score": 1.0, + "content": "SCIGAN does not place any restrictions on the form of the treatment-does response functions", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 448, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 506, + 461 + ], + "score": 1.0, + "content": "and is capable of estimating patient outcomes for multiple treatments, each with an associated", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 459, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 472 + ], + "score": 1.0, + "content": "parameter. SCIGAN first trains a generator to generate response curves for each sample within", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 469, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 484 + ], + "score": 1.0, + "content": "the training dataset. The learned generator can then be used to train an inference network using", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 505, + 493 + ], + "score": 1.0, + "content": "standard supervised methods. For fair comparison, we increase the number of parameters for the the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "score": 1.0, + "content": "open-source version of SCIGAN so that the SCIGAN model can have same order of magnitude of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 503, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 506, + 516 + ], + "score": 1.0, + "content": "network parameters as GALILEO. In addition, we also finetune the hyperparameters (Tab. 4) of the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 514, + 465, + 527 + ], + "spans": [ + { + "bbox": [ + 106, + 514, + 431, + 527 + ], + "score": 1.0, + "content": "enlarged SCIGAN to realize its full strength. We set num_dosage_samples 9 and", + "type": "text" + }, + { + "bbox": [ + 432, + 514, + 461, + 524 + ], + "score": 0.9, + "content": "\\lambda = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 514, + 465, + 527 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 414, + 507, + 527 + ] + }, + { + "type": "table", + "bbox": [ + 225, + 561, + 383, + 606 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 205, + 549, + 402, + 561 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 204, + 549, + 403, + 563 + ], + "spans": [ + { + "bbox": [ + 204, + 549, + 403, + 563 + ], + "score": 1.0, + "content": "Table 4: Table of hyper-parameters for SCIGAN.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "table_body", + "bbox": [ + 225, + 561, + 383, + 606 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 225, + 561, + 383, + 606 + ], + "spans": [ + { + "bbox": [ + 225, + 561, + 383, + 606 + ], + "score": 0.975, + "html": "
ParameterValues
Number of samples3,5,7,9, 11
入0.1,1,10, 20
", + "type": "table", + "image_path": "8ec5dc6c111badbd1e205f3b3f1e840f45d4e74ac912ed430f8883e738032a52.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 225, + 561, + 383, + 576.0 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 225, + 576.0, + 383, + 591.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 225, + 591.0, + 383, + 606.0 + ], + "spans": [], + "index": 35 + } + ] + } + ], + "index": 33.0 + }, + { + "type": "title", + "bbox": [ + 107, + 634, + 225, + 646 + ], + "lines": [ + { + "bbox": [ + 106, + 634, + 226, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 226, + 646 + ], + "score": 1.0, + "content": "G.3 HYPER-PARAMETERS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 313, + 669 + ], + "lines": [ + { + "bbox": [ + 106, + 657, + 315, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 657, + 315, + 669 + ], + "score": 1.0, + "content": "We list the hyper-parameter of GALILEO in Tab. 5.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37, + "bbox_fs": [ + 106, + 657, + 315, + 669 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 686, + 252, + 698 + ], + "lines": [ + { + "bbox": [ + 106, + 685, + 253, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 685, + 253, + 699 + ], + "score": 1.0, + "content": "G.4 COMPUTATION RESOURCES", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "We use one Tesla V100 PCIe 32GB GPU and a 32-core Intel(R) Xeon(R) Gold 5118 CPU @ 2.30GHz", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 721, + 206, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 206, + 732 + ], + "score": 1.0, + "content": "to train all of our model.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 709, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 102, + 567, + 277 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 194, + 90, + 413, + 101 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 194, + 89, + 413, + 102 + ], + "spans": [ + { + "bbox": [ + 194, + 89, + 413, + 102 + ], + "score": 1.0, + "content": "Table 5: Table of hyper-parameters for all of the tasks.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 108, + 102, + 567, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 102, + 567, + 277 + ], + "spans": [ + { + "bbox": [ + 108, + 102, + 567, + 277 + ], + "score": 0.983, + "html": "
ParameterGNFCTAGCMuJoCoBAT
hidden layers of all neural networks4455
hidden units of all neural networks256256512512
collect samples for each time of model update500050004000096000
batch size of discriminators500050004000080000
horizon50110048 (half-hours)
∈μ(also ∈D)0.0050.010.05 (0.1 for walker2d)0.05
times for discriminator update2215
times for model update11220
times for supervised learning update11420
learning rate for supervised learning1e-51e-53e-41e-5
Y0.990.00.990.99
clip-ratioNANNANNAN0.1
max DkL0.0010.0010.001NAN
optimization algorithm (the first term of Eq.23)TRPOTRPOTRPOPPO
", + "type": "table", + "image_path": "61ef1cf2a80485a84bf0e9655791dc7d13baf5a81d841666a2976ef65838b5ec.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 108, + 102, + 567, + 160.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 160.33333333333334, + 567, + 218.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 218.66666666666669, + 567, + 277.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "title", + "bbox": [ + 108, + 298, + 246, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 298, + 246, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 246, + 313 + ], + "score": 1.0, + "content": "H ADDITIONAL RESULTS", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "title", + "bbox": [ + 108, + 324, + 253, + 335 + ], + "lines": [ + { + "bbox": [ + 106, + 324, + 254, + 337 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 254, + 337 + ], + "score": 1.0, + "content": "H.1 ALL OF THE RESULT TABLE", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 105, + 345, + 480, + 357 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 482, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 482, + 358 + ], + "score": 1.0, + "content": "We give the result of CNFC in Tab. 8, TCGA in Tab. 9, BAT in Tab. 7, and MuJoCo in Tab. 6.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "title", + "bbox": [ + 108, + 371, + 267, + 383 + ], + "lines": [ + { + "bbox": [ + 106, + 371, + 267, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 267, + 384 + ], + "score": 1.0, + "content": "H.2 AVERAGED RESPONSE CURVES", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 393, + 506, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 393, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 393, + 505, + 405 + ], + "score": 1.0, + "content": "We give the averaged responses for all of the tasks and the algorithms in Fig. 16 to Fig. 23. We", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 174, + 416 + ], + "score": 1.0, + "content": "randomly select", + "type": "text" + }, + { + "bbox": [ + 174, + 404, + 194, + 415 + ], + "score": 0.88, + "content": "20 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "of the states in the dataset and equidistantly sample actions from the action", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 414, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 506, + 428 + ], + "score": 1.0, + "content": "space for each sampled state, and plot the averaged predicted feedback of each action. The real", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "response is slightly different among different figure as the randomly-selected states for testing is", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 436, + 455, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 455, + 451 + ], + "score": 1.0, + "content": "different. We sample 9 points in GNFC tasks and 33 points in TAGC tasks for plotting.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 108, + 463, + 218, + 474 + ], + "lines": [ + { + "bbox": [ + 106, + 463, + 220, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 220, + 475 + ], + "score": 1.0, + "content": "H.3 ABLATION STUDIES", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 484, + 337, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 337, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 337, + 497 + ], + "score": 1.0, + "content": "In Sec. 4.3 and Appx. E.1, we introduce several techniques", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 495, + 337, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 337, + 507 + ], + "score": 1.0, + "content": "to develop a practical GALILEO algorithm. Based on task", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 506, + 337, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 167, + 518 + ], + "score": 0.87, + "content": "\\mathtt { e 0 } . 2 _ { - \\mathtt { p } 0 . 0 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 506, + 337, + 518 + ], + "score": 1.0, + "content": "in GNFC, we give the ablation studies to", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 517, + 338, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 338, + 529 + ], + "score": 1.0, + "content": "investigate the effects of these techniques. We first com-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 528, + 338, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 338, + 541 + ], + "score": 1.0, + "content": "pare two variants that do not handle the assumptions viola-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 540, + 335, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 293, + 552 + ], + "score": 1.0, + "content": "tion problems: (1) NO_INJECT_NOISE: set", + "type": "text" + }, + { + "bbox": [ + 293, + 541, + 304, + 552 + ], + "score": 0.87, + "content": "\\epsilon _ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 540, + 322, + 552 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 323, + 541, + 335, + 551 + ], + "score": 0.81, + "content": "\\epsilon _ { D }", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 550, + 338, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 338, + 563 + ], + "score": 1.0, + "content": "to zero, which makes the overlap assumption not satisfied;;", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 561, + 337, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 337, + 573 + ], + "score": 1.0, + "content": "(2) SINGLE_SL: without replacing the second term in", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 572, + 337, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 337, + 585 + ], + "score": 1.0, + "content": "Eq. 8 with standard supervised learning even when the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 583, + 338, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 195, + 595 + ], + "score": 1.0, + "content": "output probability of", + "type": "text" + }, + { + "bbox": [ + 196, + 583, + 205, + 593 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 583, + 338, + 595 + ], + "score": 1.0, + "content": "is far away from 0.5. Besides,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 595, + 337, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 337, + 606 + ], + "score": 1.0, + "content": "we introduced several tricks inspired by GAIL and give", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 605, + 338, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 338, + 616 + ], + "score": 1.0, + "content": "a comparison of these tricks and GAIL: (3) ONE_STEP:", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 617, + 337, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 337, + 628 + ], + "score": 1.0, + "content": "use one-step reward instead of cumulative rewards (i.e.,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 627, + 337, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 337, + 639 + ], + "score": 1.0, + "content": "Q and V; see Eq. 24 and Eq. 25) for re-weighting, which", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 639, + 337, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 199, + 650 + ], + "score": 1.0, + "content": "is implemented by set", + "type": "text" + }, + { + "bbox": [ + 199, + 640, + 207, + 649 + ], + "score": 0.81, + "content": "\\gamma", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 639, + 337, + 650 + ], + "score": 1.0, + "content": "to 0; (4) SINGE_DIS: remove", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 107, + 648, + 337, + 662 + ], + "spans": [ + { + "bbox": [ + 107, + 649, + 146, + 662 + ], + "score": 0.92, + "content": "T _ { \\varphi _ { 1 } ^ { * } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 649, + 224, + 662 + ], + "score": 1.0, + "content": "and replace it with", + "type": "text" + }, + { + "bbox": [ + 225, + 648, + 306, + 662 + ], + "score": 0.88, + "content": "\\mathbb { E } _ { M _ { \\theta } } \\left[ T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 649, + 337, + 662 + ], + "score": 1.0, + "content": ", which", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 659, + 336, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 336, + 672 + ], + "score": 1.0, + "content": "is inspired by GAIL that uses a value function as a baseline", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 671, + 338, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 338, + 682 + ], + "score": 1.0, + "content": "instead of using another discriminator; (5) PURE_GAIL:", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 22.5 + }, + { + "type": "image", + "bbox": [ + 345, + 497, + 501, + 634 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 345, + 497, + 501, + 634 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 345, + 497, + 501, + 634 + ], + "spans": [ + { + "bbox": [ + 345, + 497, + 501, + 634 + ], + "score": 0.958, + "type": "image", + "image_path": "bd32c544f33fa909bf10ef2d6bc1392ce066275899883232e0b00bb79348eae2.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 345, + 497, + 501, + 509.45454545454544 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 345, + 509.45454545454544, + 501, + 521.9090909090909 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 345, + 521.9090909090909, + 501, + 534.3636363636364 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 345, + 534.3636363636364, + 501, + 546.8181818181819 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 345, + 546.8181818181819, + 501, + 559.2727272727274 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 345, + 559.2727272727274, + 501, + 571.7272727272729 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 345, + 571.7272727272729, + 501, + 584.1818181818184 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 345, + 584.1818181818184, + 501, + 596.6363636363639 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 345, + 596.6363636363639, + 501, + 609.0909090909093 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 345, + 609.0909090909093, + 501, + 621.5454545454548 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 345, + 621.5454545454548, + 501, + 634.0000000000003 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 344, + 648, + 504, + 681 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 343, + 648, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 343, + 648, + 505, + 659 + ], + "score": 1.0, + "content": "Figure 10: Illustration of the ablation", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 344, + 660, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 344, + 660, + 505, + 671 + ], + "score": 1.0, + "content": "studies. The error bars are the standard", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 343, + 672, + 368, + 681 + ], + "spans": [ + { + "bbox": [ + 343, + 672, + 368, + 681 + ], + "score": 1.0, + "content": "error.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 108, + 682, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 681, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 694 + ], + "score": 1.0, + "content": "remove the second term in Eq. 8. It can be regarded as a naive adoption of GAIL and a partial", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 693, + 231, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 231, + 704 + ], + "score": 1.0, + "content": "implementation of GALILEO.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "We summarize the results in Fig. 10. Based on the results of NO_INJECT_NOISE and SINGLE_SL,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "we can see that handling the assumption violation problems is important and will increase the ability", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48.5 + } + ], + "page_idx": 30, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 102, + 567, + 277 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 194, + 90, + 413, + 101 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 194, + 89, + 413, + 102 + ], + "spans": [ + { + "bbox": [ + 194, + 89, + 413, + 102 + ], + "score": 1.0, + "content": "Table 5: Table of hyper-parameters for all of the tasks.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 108, + 102, + 567, + 277 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 102, + 567, + 277 + ], + "spans": [ + { + "bbox": [ + 108, + 102, + 567, + 277 + ], + "score": 0.983, + "html": "
ParameterGNFCTAGCMuJoCoBAT
hidden layers of all neural networks4455
hidden units of all neural networks256256512512
collect samples for each time of model update500050004000096000
batch size of discriminators500050004000080000
horizon50110048 (half-hours)
∈μ(also ∈D)0.0050.010.05 (0.1 for walker2d)0.05
times for discriminator update2215
times for model update11220
times for supervised learning update11420
learning rate for supervised learning1e-51e-53e-41e-5
Y0.990.00.990.99
clip-ratioNANNANNAN0.1
max DkL0.0010.0010.001NAN
optimization algorithm (the first term of Eq.23)TRPOTRPOTRPOPPO
", + "type": "table", + "image_path": "61ef1cf2a80485a84bf0e9655791dc7d13baf5a81d841666a2976ef65838b5ec.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 108, + 102, + 567, + 160.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 160.33333333333334, + 567, + 218.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 218.66666666666669, + 567, + 277.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "title", + "bbox": [ + 108, + 298, + 246, + 311 + ], + "lines": [ + { + "bbox": [ + 106, + 298, + 246, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 246, + 313 + ], + "score": 1.0, + "content": "H ADDITIONAL RESULTS", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "title", + "bbox": [ + 108, + 324, + 253, + 335 + ], + "lines": [ + { + "bbox": [ + 106, + 324, + 254, + 337 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 254, + 337 + ], + "score": 1.0, + "content": "H.1 ALL OF THE RESULT TABLE", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 105, + 345, + 480, + 357 + ], + "lines": [ + { + "bbox": [ + 105, + 344, + 482, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 482, + 358 + ], + "score": 1.0, + "content": "We give the result of CNFC in Tab. 8, TCGA in Tab. 9, BAT in Tab. 7, and MuJoCo in Tab. 6.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 344, + 482, + 358 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 371, + 267, + 383 + ], + "lines": [ + { + "bbox": [ + 106, + 371, + 267, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 267, + 384 + ], + "score": 1.0, + "content": "H.2 AVERAGED RESPONSE CURVES", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 393, + 506, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 393, + 505, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 393, + 505, + 405 + ], + "score": 1.0, + "content": "We give the averaged responses for all of the tasks and the algorithms in Fig. 16 to Fig. 23. We", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 174, + 416 + ], + "score": 1.0, + "content": "randomly select", + "type": "text" + }, + { + "bbox": [ + 174, + 404, + 194, + 415 + ], + "score": 0.88, + "content": "20 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "of the states in the dataset and equidistantly sample actions from the action", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 414, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 506, + 428 + ], + "score": 1.0, + "content": "space for each sampled state, and plot the averaged predicted feedback of each action. The real", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "response is slightly different among different figure as the randomly-selected states for testing is", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 436, + 455, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 436, + 455, + 451 + ], + "score": 1.0, + "content": "different. We sample 9 points in GNFC tasks and 33 points in TAGC tasks for plotting.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 393, + 506, + 451 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 463, + 218, + 474 + ], + "lines": [ + { + "bbox": [ + 106, + 463, + 220, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 463, + 220, + 475 + ], + "score": 1.0, + "content": "H.3 ABLATION STUDIES", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 484, + 337, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 337, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 337, + 497 + ], + "score": 1.0, + "content": "In Sec. 4.3 and Appx. E.1, we introduce several techniques", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 495, + 337, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 337, + 507 + ], + "score": 1.0, + "content": "to develop a practical GALILEO algorithm. Based on task", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 506, + 337, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 167, + 518 + ], + "score": 0.87, + "content": "\\mathtt { e 0 } . 2 _ { - \\mathtt { p } 0 . 0 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 506, + 337, + 518 + ], + "score": 1.0, + "content": "in GNFC, we give the ablation studies to", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 517, + 338, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 338, + 529 + ], + "score": 1.0, + "content": "investigate the effects of these techniques. We first com-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 528, + 338, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 338, + 541 + ], + "score": 1.0, + "content": "pare two variants that do not handle the assumptions viola-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 540, + 335, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 293, + 552 + ], + "score": 1.0, + "content": "tion problems: (1) NO_INJECT_NOISE: set", + "type": "text" + }, + { + "bbox": [ + 293, + 541, + 304, + 552 + ], + "score": 0.87, + "content": "\\epsilon _ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 540, + 322, + 552 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 323, + 541, + 335, + 551 + ], + "score": 0.81, + "content": "\\epsilon _ { D }", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 550, + 338, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 338, + 563 + ], + "score": 1.0, + "content": "to zero, which makes the overlap assumption not satisfied;;", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 561, + 337, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 337, + 573 + ], + "score": 1.0, + "content": "(2) SINGLE_SL: without replacing the second term in", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 572, + 337, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 337, + 585 + ], + "score": 1.0, + "content": "Eq. 8 with standard supervised learning even when the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 583, + 338, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 195, + 595 + ], + "score": 1.0, + "content": "output probability of", + "type": "text" + }, + { + "bbox": [ + 196, + 583, + 205, + 593 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 583, + 338, + 595 + ], + "score": 1.0, + "content": "is far away from 0.5. Besides,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 595, + 337, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 337, + 606 + ], + "score": 1.0, + "content": "we introduced several tricks inspired by GAIL and give", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 605, + 338, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 338, + 616 + ], + "score": 1.0, + "content": "a comparison of these tricks and GAIL: (3) ONE_STEP:", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 617, + 337, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 337, + 628 + ], + "score": 1.0, + "content": "use one-step reward instead of cumulative rewards (i.e.,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 627, + 337, + 639 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 337, + 639 + ], + "score": 1.0, + "content": "Q and V; see Eq. 24 and Eq. 25) for re-weighting, which", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 639, + 337, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 199, + 650 + ], + "score": 1.0, + "content": "is implemented by set", + "type": "text" + }, + { + "bbox": [ + 199, + 640, + 207, + 649 + ], + "score": 0.81, + "content": "\\gamma", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 639, + 337, + 650 + ], + "score": 1.0, + "content": "to 0; (4) SINGE_DIS: remove", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 107, + 648, + 337, + 662 + ], + "spans": [ + { + "bbox": [ + 107, + 649, + 146, + 662 + ], + "score": 0.92, + "content": "T _ { \\varphi _ { 1 } ^ { * } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 649, + 224, + 662 + ], + "score": 1.0, + "content": "and replace it with", + "type": "text" + }, + { + "bbox": [ + 225, + 648, + 306, + 662 + ], + "score": 0.88, + "content": "\\mathbb { E } _ { M _ { \\theta } } \\left[ T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 649, + 337, + 662 + ], + "score": 1.0, + "content": ", which", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 659, + 336, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 336, + 672 + ], + "score": 1.0, + "content": "is inspired by GAIL that uses a value function as a baseline", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 671, + 338, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 338, + 682 + ], + "score": 1.0, + "content": "instead of using another discriminator; (5) PURE_GAIL:", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 484, + 338, + 682 + ] + }, + { + "type": "image", + "bbox": [ + 345, + 497, + 501, + 634 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 345, + 497, + 501, + 634 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 345, + 497, + 501, + 634 + ], + "spans": [ + { + "bbox": [ + 345, + 497, + 501, + 634 + ], + "score": 0.958, + "type": "image", + "image_path": "bd32c544f33fa909bf10ef2d6bc1392ce066275899883232e0b00bb79348eae2.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 345, + 497, + 501, + 509.45454545454544 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 345, + 509.45454545454544, + 501, + 521.9090909090909 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 345, + 521.9090909090909, + 501, + 534.3636363636364 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 345, + 534.3636363636364, + 501, + 546.8181818181819 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 345, + 546.8181818181819, + 501, + 559.2727272727274 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 345, + 559.2727272727274, + 501, + 571.7272727272729 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 345, + 571.7272727272729, + 501, + 584.1818181818184 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 345, + 584.1818181818184, + 501, + 596.6363636363639 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 345, + 596.6363636363639, + 501, + 609.0909090909093 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 345, + 609.0909090909093, + 501, + 621.5454545454548 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 345, + 621.5454545454548, + 501, + 634.0000000000003 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 344, + 648, + 504, + 681 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 343, + 648, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 343, + 648, + 505, + 659 + ], + "score": 1.0, + "content": "Figure 10: Illustration of the ablation", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 344, + 660, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 344, + 660, + 505, + 671 + ], + "score": 1.0, + "content": "studies. The error bars are the standard", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 343, + 672, + 368, + 681 + ], + "spans": [ + { + "bbox": [ + 343, + 672, + 368, + 681 + ], + "score": 1.0, + "content": "error.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 108, + 682, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 681, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 694 + ], + "score": 1.0, + "content": "remove the second term in Eq. 8. It can be regarded as a naive adoption of GAIL and a partial", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 693, + 231, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 231, + 704 + ], + "score": 1.0, + "content": "implementation of GALILEO.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46.5, + "bbox_fs": [ + 105, + 681, + 505, + 704 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "We summarize the results in Fig. 10. Based on the results of NO_INJECT_NOISE and SINGLE_SL,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "we can see that handling the assumption violation problems is important and will increase the ability", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 289, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 506, + 302 + ], + "score": 1.0, + "content": "on counterfactual queries. The results of PURE_GAIL tell us that the partial implementation of", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "score": 1.0, + "content": "GALILEO is not enough to give stable predictions on counterfactual data; On the other hand, the", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 312, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 505, + 324 + ], + "score": 1.0, + "content": "result of ONE_STEP also demonstrates that embedding the cumulative error of one-step prediction", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 323, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 505, + 334 + ], + "score": 1.0, + "content": "is helpful for GALILEO training; Finally, we also found that SINGLE_DIS nearly has almost no", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 331, + 508, + 351 + ], + "spans": [ + { + "bbox": [ + 104, + 331, + 352, + 351 + ], + "score": 1.0, + "content": "effect on the results. It suggests that, empirically, we can use", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 356, + 333, + 434, + 347 + ], + "score": 0.58, + "content": "\\mathbb { E } _ { M _ { \\theta } } \\left[ T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 434, + 331, + 508, + 351 + ], + "score": 1.0, + "content": "as a replacement", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 345, + 465, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 120, + 359 + ], + "score": 1.0, + "content": "for", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 120, + 345, + 160, + 359 + ], + "score": 0.93, + "content": "T _ { \\varphi _ { 1 } ^ { * } } ( x , a )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 161, + 345, + 465, + 359 + ], + "score": 1.0, + "content": ", which can reduce the computation costs of the extra discriminator training.", + "type": "text", + "cross_page": true + } + ], + "index": 11 + } + ], + "index": 48.5, + "bbox_fs": [ + 105, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 155, + 94, + 453, + 278 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 60, + 506, + 94 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 61, + 506, + 73 + ], + "spans": [ + { + "bbox": [ + 105, + 61, + 506, + 73 + ], + "score": 1.0, + "content": "Table 6: The root mean square errors on MuJoCo tasks. We bold the lowest error for each task.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 71, + 505, + 84 + ], + "spans": [ + { + "bbox": [ + 105, + 71, + 505, + 84 + ], + "score": 1.0, + "content": "“medium” dataset is used for training, while “expert” and “medium-replay” datasets are just used", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 83, + 347, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 154, + 95 + ], + "score": 1.0, + "content": "for testing.", + "type": "text" + }, + { + "bbox": [ + 155, + 83, + 165, + 94 + ], + "score": 0.32, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 83, + 347, + 95 + ], + "score": 1.0, + "content": "follows the standard deviation of three seeds.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 155, + 94, + 453, + 278 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 155, + 94, + 453, + 278 + ], + "spans": [ + { + "bbox": [ + 155, + 94, + 453, + 278 + ], + "score": 0.929, + "html": "
TASKHalfCheetah
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.378 ± 0.0032.287± 0.0051.411 ± 0.037
OFF-SL0.404 ± 0.0013.311 ± 0.0552.246 ± 0.016
TASKWalker2d
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.49 ± 0.001.514 ± 0.0020.968 ± 0.004
OFF-SL0.467 ± 0.0041.825 ± 0.0611.239 ± 0.004
TASKHopper
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.037 ± 0.0020.322 ± 0.0360.408 ± 0.003
OFF-SL0.034 ± 0.0010.464 ± 0.0210.574 ± 0.008
", + "type": "table", + "image_path": "38ea1f1978b9e9b624ac794e28682b08e25655d901960c5c6e9f60cfe4af3b80.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 155, + 94, + 453, + 155.33333333333334 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 155, + 155.33333333333334, + 453, + 216.66666666666669 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 155, + 216.66666666666669, + 453, + 278.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 289, + 506, + 357 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 506, + 302 + ], + "score": 1.0, + "content": "on counterfactual queries. The results of PURE_GAIL tell us that the partial implementation of", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "score": 1.0, + "content": "GALILEO is not enough to give stable predictions on counterfactual data; On the other hand, the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 312, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 505, + 324 + ], + "score": 1.0, + "content": "result of ONE_STEP also demonstrates that embedding the cumulative error of one-step prediction", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 323, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 505, + 334 + ], + "score": 1.0, + "content": "is helpful for GALILEO training; Finally, we also found that SINGLE_DIS nearly has almost no", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 331, + 508, + 351 + ], + "spans": [ + { + "bbox": [ + 104, + 331, + 352, + 351 + ], + "score": 1.0, + "content": "effect on the results. It suggests that, empirically, we can use", + "type": "text" + }, + { + "bbox": [ + 356, + 333, + 434, + 347 + ], + "score": 0.58, + "content": "\\mathbb { E } _ { M _ { \\theta } } \\left[ T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 331, + 508, + 351 + ], + "score": 1.0, + "content": "as a replacement", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 345, + 465, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 120, + 359 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 120, + 345, + 160, + 359 + ], + "score": 0.93, + "content": "T _ { \\varphi _ { 1 } ^ { * } } ( x , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 345, + 465, + 359 + ], + "score": 1.0, + "content": ", which can reduce the computation costs of the extra discriminator training.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 108, + 370, + 281, + 382 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 282, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 282, + 383 + ], + "score": 1.0, + "content": "H.4 WORST-CASE PREDICTION ERROR", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 391, + 506, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 392, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 505, + 403 + ], + "score": 1.0, + "content": "In theory, GALILEO increases the generalization ability by focusing on the worst-case samples’", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "training to achieve AWRM. To demonstrate the property, we propose a new metric named Mean-Max", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 413, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 208, + 431 + ], + "score": 1.0, + "content": "Square Error (MMSE):", + "type": "text" + }, + { + "bbox": [ + 208, + 413, + 378, + 433 + ], + "score": 0.8, + "content": "\\begin{array} { r } { \\bar { \\mathfrak { z } } \\left[ \\operatorname* { m a x } _ { a \\in \\mathcal { A } } \\left( M ^ { * } ( x ^ { \\prime } | x , a ) - \\bar { M } ( x ^ { \\prime } | x , a ) \\right) ^ { 2 } \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 415, + 506, + 431 + ], + "score": 1.0, + "content": "and give the results of MMSE", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 430, + 296, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 296, + 442 + ], + "score": 1.0, + "content": "for GNFC in Tab. 10 and for TCGA in Tab. 11.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 107, + 455, + 324, + 468 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 324, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 324, + 469 + ], + "score": 1.0, + "content": "H.5 DETAILED RESULTS IN THE MUJOCO TASKS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 476, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 106, + 477, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 505, + 489 + ], + "score": 1.0, + "content": "We select 3 environments from D4RL (Fu et al., 2020) to construct our model learning tasks. We", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 487, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 506, + 499 + ], + "score": 1.0, + "content": "compare it with a typical transition model learning algorithm used in the previous offline model-based", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "RL algorithms (Yu et al., 2020; Kidambi et al., 2020), which is a variant of standard supervised", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 509, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 506, + 521 + ], + "score": 1.0, + "content": "learning. We name the method OFF-SL. We train models in datasets HalfCheetah-medium, Walker2d-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 521, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 532 + ], + "score": 1.0, + "content": "medium, and Hopper-medium, which are collected by a behavior policy with 1/3 performance to", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 531, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 544 + ], + "score": 1.0, + "content": "the expert policy, then we test them in the corresponding expert dataset. We plot the converged", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "results and learning curves of GALILEO and OFF-SL in three MuJoCo tasks in Tab. 6 and Fig. 11", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 553, + 159, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 159, + 567 + ], + "score": 1.0, + "content": "respectively.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 570, + 505, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 570, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 506, + 582 + ], + "score": 1.0, + "content": "In Fig. 11, we can see that both OFF-SL and GALILEO perform well in the training datasets.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 580, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 505, + 594 + ], + "score": 1.0, + "content": "OFF-SL can even reach a bit lower error in halfcheetah and walker2d. However, when we verify", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 591, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 506, + 605 + ], + "score": 1.0, + "content": "the models through “expert” and “medium-replay” datasets, which are collected by other policies,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 602, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 506, + 617 + ], + "score": 1.0, + "content": "the performance of GALILEO is significantly more stable and better than OFF-SL. As the training", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 614, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 506, + 627 + ], + "score": 1.0, + "content": "continues, OFF-SL even gets worse and worse. In summary, GALILEO reaches significantly better", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 625, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 506, + 638 + ], + "score": 1.0, + "content": "performances in the expert dataset: the averaged declines of root MSE in three environments are", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 634, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 133, + 647 + ], + "score": 0.85, + "content": "5 6 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 634, + 137, + 650 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 137, + 636, + 164, + 647 + ], + "score": 0.86, + "content": "4 9 . 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 634, + 184, + 650 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 184, + 636, + 211, + 647 + ], + "score": 0.86, + "content": "3 4 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 634, + 505, + 650 + ], + "score": 1.0, + "content": ". However, whether in GALILEO or OFF-SL, the performance for testing", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 145, + 659 + ], + "score": 1.0, + "content": "is at least", + "type": "text" + }, + { + "bbox": [ + 146, + 648, + 158, + 657 + ], + "score": 0.32, + "content": "2 \\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "worse than in the training dataset. The phenomenon indicates that although GALILEO", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 657, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 672 + ], + "score": 1.0, + "content": "can make better performances for counterfactual queries, the risks of using the models are still large", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 669, + 272, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 272, + 680 + ], + "score": 1.0, + "content": "and still challenging to be further solved.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5 + } + ], + "page_idx": 31, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 155, + 94, + 453, + 278 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 60, + 506, + 94 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 61, + 506, + 73 + ], + "spans": [ + { + "bbox": [ + 105, + 61, + 506, + 73 + ], + "score": 1.0, + "content": "Table 6: The root mean square errors on MuJoCo tasks. We bold the lowest error for each task.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 71, + 505, + 84 + ], + "spans": [ + { + "bbox": [ + 105, + 71, + 505, + 84 + ], + "score": 1.0, + "content": "“medium” dataset is used for training, while “expert” and “medium-replay” datasets are just used", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 83, + 347, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 154, + 95 + ], + "score": 1.0, + "content": "for testing.", + "type": "text" + }, + { + "bbox": [ + 155, + 83, + 165, + 94 + ], + "score": 0.32, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 83, + 347, + 95 + ], + "score": 1.0, + "content": "follows the standard deviation of three seeds.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 155, + 94, + 453, + 278 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 155, + 94, + 453, + 278 + ], + "spans": [ + { + "bbox": [ + 155, + 94, + 453, + 278 + ], + "score": 0.929, + "html": "
TASKHalfCheetah
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.378 ± 0.0032.287± 0.0051.411 ± 0.037
OFF-SL0.404 ± 0.0013.311 ± 0.0552.246 ± 0.016
TASKWalker2d
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.49 ± 0.001.514 ± 0.0020.968 ± 0.004
OFF-SL0.467 ± 0.0041.825 ± 0.0611.239 ± 0.004
TASKHopper
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.037 ± 0.0020.322 ± 0.0360.408 ± 0.003
OFF-SL0.034 ± 0.0010.464 ± 0.0210.574 ± 0.008
", + "type": "table", + "image_path": "38ea1f1978b9e9b624ac794e28682b08e25655d901960c5c6e9f60cfe4af3b80.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 155, + 94, + 453, + 155.33333333333334 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 155, + 155.33333333333334, + 453, + 216.66666666666669 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 155, + 216.66666666666669, + 453, + 278.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 289, + 506, + 357 + ], + "lines": [], + "index": 8.5, + "bbox_fs": [ + 104, + 289, + 508, + 359 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 370, + 281, + 382 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 282, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 282, + 383 + ], + "score": 1.0, + "content": "H.4 WORST-CASE PREDICTION ERROR", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 391, + 506, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 392, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 505, + 403 + ], + "score": 1.0, + "content": "In theory, GALILEO increases the generalization ability by focusing on the worst-case samples’", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 402, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 106, + 402, + 505, + 415 + ], + "score": 1.0, + "content": "training to achieve AWRM. To demonstrate the property, we propose a new metric named Mean-Max", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 413, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 208, + 431 + ], + "score": 1.0, + "content": "Square Error (MMSE):", + "type": "text" + }, + { + "bbox": [ + 208, + 413, + 378, + 433 + ], + "score": 0.8, + "content": "\\begin{array} { r } { \\bar { \\mathfrak { z } } \\left[ \\operatorname* { m a x } _ { a \\in \\mathcal { A } } \\left( M ^ { * } ( x ^ { \\prime } | x , a ) - \\bar { M } ( x ^ { \\prime } | x , a ) \\right) ^ { 2 } \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 415, + 506, + 431 + ], + "score": 1.0, + "content": "and give the results of MMSE", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 430, + 296, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 296, + 442 + ], + "score": 1.0, + "content": "for GNFC in Tab. 10 and for TCGA in Tab. 11.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 392, + 506, + 442 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 455, + 324, + 468 + ], + "lines": [ + { + "bbox": [ + 106, + 455, + 324, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 324, + 469 + ], + "score": 1.0, + "content": "H.5 DETAILED RESULTS IN THE MUJOCO TASKS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 476, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 106, + 477, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 505, + 489 + ], + "score": 1.0, + "content": "We select 3 environments from D4RL (Fu et al., 2020) to construct our model learning tasks. We", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 487, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 506, + 499 + ], + "score": 1.0, + "content": "compare it with a typical transition model learning algorithm used in the previous offline model-based", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "RL algorithms (Yu et al., 2020; Kidambi et al., 2020), which is a variant of standard supervised", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 509, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 506, + 521 + ], + "score": 1.0, + "content": "learning. We name the method OFF-SL. We train models in datasets HalfCheetah-medium, Walker2d-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 521, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 532 + ], + "score": 1.0, + "content": "medium, and Hopper-medium, which are collected by a behavior policy with 1/3 performance to", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 531, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 544 + ], + "score": 1.0, + "content": "the expert policy, then we test them in the corresponding expert dataset. We plot the converged", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "results and learning curves of GALILEO and OFF-SL in three MuJoCo tasks in Tab. 6 and Fig. 11", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 553, + 159, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 159, + 567 + ], + "score": 1.0, + "content": "respectively.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 477, + 506, + 567 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 570, + 505, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 570, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 506, + 582 + ], + "score": 1.0, + "content": "In Fig. 11, we can see that both OFF-SL and GALILEO perform well in the training datasets.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 580, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 505, + 594 + ], + "score": 1.0, + "content": "OFF-SL can even reach a bit lower error in halfcheetah and walker2d. However, when we verify", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 591, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 506, + 605 + ], + "score": 1.0, + "content": "the models through “expert” and “medium-replay” datasets, which are collected by other policies,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 602, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 506, + 617 + ], + "score": 1.0, + "content": "the performance of GALILEO is significantly more stable and better than OFF-SL. As the training", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 614, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 506, + 627 + ], + "score": 1.0, + "content": "continues, OFF-SL even gets worse and worse. In summary, GALILEO reaches significantly better", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 625, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 506, + 638 + ], + "score": 1.0, + "content": "performances in the expert dataset: the averaged declines of root MSE in three environments are", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 634, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 133, + 647 + ], + "score": 0.85, + "content": "5 6 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 634, + 137, + 650 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 137, + 636, + 164, + 647 + ], + "score": 0.86, + "content": "4 9 . 2 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 634, + 184, + 650 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 184, + 636, + 211, + 647 + ], + "score": 0.86, + "content": "3 4 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 634, + 505, + 650 + ], + "score": 1.0, + "content": ". However, whether in GALILEO or OFF-SL, the performance for testing", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 145, + 659 + ], + "score": 1.0, + "content": "is at least", + "type": "text" + }, + { + "bbox": [ + 146, + 648, + 158, + 657 + ], + "score": 0.32, + "content": "2 \\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "worse than in the training dataset. The phenomenon indicates that although GALILEO", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 657, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 672 + ], + "score": 1.0, + "content": "can make better performances for counterfactual queries, the risks of using the models are still large", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 669, + 272, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 272, + 680 + ], + "score": 1.0, + "content": "and still challenging to be further solved.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 570, + 506, + 680 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 113, + 185, + 496, + 563 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 185, + 496, + 563 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 185, + 496, + 563 + ], + "spans": [ + { + "bbox": [ + 113, + 185, + 496, + 563 + ], + "score": 0.977, + "type": "image", + "image_path": "5547f6091a98561bc79b830999d27857990e83ebd0ac546794410f3a7f6be8af.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 113, + 185, + 496, + 311.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 113, + 311.0, + 496, + 437.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 113, + 437.0, + 496, + 563.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 575, + 505, + 631 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 576, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 505, + 587 + ], + "score": 1.0, + "content": "Figure 11: Illustration of learning curves of the MuJoCo Tasks. The X-axis record the steps of the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 586, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 599 + ], + "score": 1.0, + "content": "environment model update, and the Y-axis is the corresponding prediction error. The figures with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 596, + 506, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 506, + 610 + ], + "score": 1.0, + "content": "titles ending in “(train)” means the dataset is used for training while the titles ending in “(test)” means", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "the dataset is just used for testing. The solid curves are the mean reward and the shadow is the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 620, + 223, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 223, + 631 + ], + "score": 1.0, + "content": "standard error of three seeds.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + } + ], + "page_idx": 32, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "33", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 113, + 185, + 496, + 563 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 185, + 496, + 563 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 185, + 496, + 563 + ], + "spans": [ + { + "bbox": [ + 113, + 185, + 496, + 563 + ], + "score": 0.977, + "type": "image", + "image_path": "5547f6091a98561bc79b830999d27857990e83ebd0ac546794410f3a7f6be8af.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 113, + 185, + 496, + 311.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 113, + 311.0, + 496, + 437.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 113, + 437.0, + 496, + 563.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 575, + 505, + 631 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 576, + 505, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 505, + 587 + ], + "score": 1.0, + "content": "Figure 11: Illustration of learning curves of the MuJoCo Tasks. The X-axis record the steps of the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 586, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 599 + ], + "score": 1.0, + "content": "environment model update, and the Y-axis is the corresponding prediction error. The figures with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 596, + 506, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 506, + 610 + ], + "score": 1.0, + "content": "titles ending in “(train)” means the dataset is used for training while the titles ending in “(test)” means", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 608, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 505, + 620 + ], + "score": 1.0, + "content": "the dataset is just used for testing. The solid curves are the mean reward and the shadow is the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 620, + 223, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 223, + 631 + ], + "score": 1.0, + "content": "standard error of three seeds.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 226, + 82, + 383, + 200 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 226, + 82, + 383, + 200 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 226, + 82, + 383, + 200 + ], + "spans": [ + { + "bbox": [ + 226, + 82, + 383, + 200 + ], + "score": 0.966, + "type": "image", + "image_path": "7fc39a335c2cd42ff20c40f5c27345da54078256b068d1373bfc35f96bdfa28c.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 226, + 82, + 383, + 95.11111111111111 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 226, + 95.11111111111111, + 383, + 108.22222222222223 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 226, + 108.22222222222223, + 383, + 121.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 226, + 121.33333333333334, + 383, + 134.44444444444446 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 226, + 134.44444444444446, + 383, + 147.55555555555557 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 226, + 147.55555555555557, + 383, + 160.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 226, + 160.66666666666669, + 383, + 173.7777777777778 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 226, + 173.7777777777778, + 383, + 186.8888888888889 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 226, + 186.8888888888889, + 383, + 200.00000000000003 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 213, + 504, + 236 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "Figure 12: Illustration of relationship between user feedback and the actions of the offline dataset in", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 225, + 261, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 261, + 236 + ], + "score": 1.0, + "content": "the real-world food-delivery platform.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + } + ], + "index": 6.75 + }, + { + "type": "title", + "bbox": [ + 108, + 257, + 298, + 269 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 300, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 300, + 270 + ], + "score": 1.0, + "content": "H.6 DETAILED RESULTS IN THE BAT TASK", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 279, + 505, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 279, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 505, + 292 + ], + "score": 1.0, + "content": "The core challenge of the environment model learning in BAT tasks is similar to the challenge in", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 290, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 506, + 303 + ], + "score": 1.0, + "content": "Fig. 1. Specifically, the behavior policy in BAT tasks is a human-expert policy, which will tend to", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 302, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 505, + 313 + ], + "score": 1.0, + "content": "increase the budget of allowance in the time periods with a lower supply of delivery clerks, otherwise", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 313, + 458, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 458, + 324 + ], + "score": 1.0, + "content": "will decrease the budget (Fig. 12 gives an instance of this phenomenon in the real data).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 107, + 329, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 106, + 328, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 106, + 328, + 506, + 341 + ], + "score": 1.0, + "content": "Since there is no oracle environment model for querying, we have to describe the results with other", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 339, + 141, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 141, + 352 + ], + "score": 1.0, + "content": "metrics.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "image", + "bbox": [ + 116, + 366, + 495, + 588 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 366, + 495, + 588 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 116, + 366, + 495, + 588 + ], + "spans": [ + { + "bbox": [ + 116, + 366, + 495, + 588 + ], + "score": 0.977, + "type": "image", + "image_path": "b869771cf06ee02caa745de9087c18bd06b71c1215b64025a89e0777903965ba.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 116, + 366, + 495, + 440.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 116, + 440.0, + 495, + 514.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 116, + 514.0, + 495, + 588.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 600, + 504, + 634 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 599, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 506, + 613 + ], + "score": 1.0, + "content": "Figure 13: Illustration of the response curves in the 6 cities. Although the ground-truth curves", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 612, + 504, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 504, + 624 + ], + "score": 1.0, + "content": "are unknown, through human expert knowledge, we know that it is expected to be monotonically", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 621, + 157, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 157, + 635 + ], + "score": 1.0, + "content": "increasing .", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "First, we review whether the tendency of the response curve is consistent. In this application, with a", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "score": 1.0, + "content": "larger budget of allowance, the supply will not be decreased. As can be seen in Fig. 13, the tendency", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "of GALILEO’s response is valid in 6 cities but almost all of the models of SL give opposite directions", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 681, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 694 + ], + "score": 1.0, + "content": "to the response. If we learn a policy through the model of SL, the optimal solution is canceling all of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 693, + 332, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 332, + 705 + ], + "score": 1.0, + "content": "the allowances, which is obviously incorrect in practice.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 107, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 107, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "Second, we conduct randomized controlled trials (RCT) in one of the testing cities. Using the RCT", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "samples, we can evaluate the correctness of the sort order of the model predictions via Area Under", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + } + ], + "page_idx": 33, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 226, + 82, + 383, + 200 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 226, + 82, + 383, + 200 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 226, + 82, + 383, + 200 + ], + "spans": [ + { + "bbox": [ + 226, + 82, + 383, + 200 + ], + "score": 0.966, + "type": "image", + "image_path": "7fc39a335c2cd42ff20c40f5c27345da54078256b068d1373bfc35f96bdfa28c.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 226, + 82, + 383, + 95.11111111111111 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 226, + 95.11111111111111, + 383, + 108.22222222222223 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 226, + 108.22222222222223, + 383, + 121.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 226, + 121.33333333333334, + 383, + 134.44444444444446 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 226, + 134.44444444444446, + 383, + 147.55555555555557 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 226, + 147.55555555555557, + 383, + 160.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 226, + 160.66666666666669, + 383, + 173.7777777777778 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 226, + 173.7777777777778, + 383, + 186.8888888888889 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 226, + 186.8888888888889, + 383, + 200.00000000000003 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 213, + 504, + 236 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "Figure 12: Illustration of relationship between user feedback and the actions of the offline dataset in", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 225, + 261, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 261, + 236 + ], + "score": 1.0, + "content": "the real-world food-delivery platform.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + } + ], + "index": 6.75 + }, + { + "type": "title", + "bbox": [ + 108, + 257, + 298, + 269 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 300, + 270 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 300, + 270 + ], + "score": 1.0, + "content": "H.6 DETAILED RESULTS IN THE BAT TASK", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 279, + 505, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 279, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 505, + 292 + ], + "score": 1.0, + "content": "The core challenge of the environment model learning in BAT tasks is similar to the challenge in", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 290, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 290, + 506, + 303 + ], + "score": 1.0, + "content": "Fig. 1. Specifically, the behavior policy in BAT tasks is a human-expert policy, which will tend to", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 302, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 505, + 313 + ], + "score": 1.0, + "content": "increase the budget of allowance in the time periods with a lower supply of delivery clerks, otherwise", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 313, + 458, + 324 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 458, + 324 + ], + "score": 1.0, + "content": "will decrease the budget (Fig. 12 gives an instance of this phenomenon in the real data).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 279, + 506, + 324 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 329, + 504, + 352 + ], + "lines": [ + { + "bbox": [ + 106, + 328, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 106, + 328, + 506, + 341 + ], + "score": 1.0, + "content": "Since there is no oracle environment model for querying, we have to describe the results with other", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 339, + 141, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 141, + 352 + ], + "score": 1.0, + "content": "metrics.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 328, + 506, + 352 + ] + }, + { + "type": "image", + "bbox": [ + 116, + 366, + 495, + 588 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 366, + 495, + 588 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 116, + 366, + 495, + 588 + ], + "spans": [ + { + "bbox": [ + 116, + 366, + 495, + 588 + ], + "score": 0.977, + "type": "image", + "image_path": "b869771cf06ee02caa745de9087c18bd06b71c1215b64025a89e0777903965ba.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 116, + 366, + 495, + 440.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 116, + 440.0, + 495, + 514.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 116, + 514.0, + 495, + 588.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 600, + 504, + 634 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 599, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 506, + 613 + ], + "score": 1.0, + "content": "Figure 13: Illustration of the response curves in the 6 cities. Although the ground-truth curves", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 612, + 504, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 504, + 624 + ], + "score": 1.0, + "content": "are unknown, through human expert knowledge, we know that it is expected to be monotonically", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 621, + 157, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 157, + 635 + ], + "score": 1.0, + "content": "increasing .", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "First, we review whether the tendency of the response curve is consistent. In this application, with a", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 673 + ], + "score": 1.0, + "content": "larger budget of allowance, the supply will not be decreased. As can be seen in Fig. 13, the tendency", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 671, + 505, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 683 + ], + "score": 1.0, + "content": "of GALILEO’s response is valid in 6 cities but almost all of the models of SL give opposite directions", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 681, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 506, + 694 + ], + "score": 1.0, + "content": "to the response. If we learn a policy through the model of SL, the optimal solution is canceling all of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 693, + 332, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 332, + 705 + ], + "score": 1.0, + "content": "the allowances, which is obviously incorrect in practice.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 648, + 506, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 107, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 107, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "Second, we conduct randomized controlled trials (RCT) in one of the testing cities. Using the RCT", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "samples, we can evaluate the correctness of the sort order of the model predictions via Area Under", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 254, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 506, + 267 + ], + "score": 1.0, + "content": "the Uplift Curve (AUUC) (Betlei et al., 2020). To plot AUUC, we first sort the RCT samples based", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 264, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 280 + ], + "score": 1.0, + "content": "on the predicted treatment effects. Then the cumulative treatment effects are computed by scanning", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 275, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 506, + 290 + ], + "score": 1.0, + "content": "the sorted sample list. If the sort order of the model predictions is better, the sample with larger", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "score": 1.0, + "content": "treatment effects will be computed early. Then the area of AUUC will be larger than the one via a", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 298, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 506, + 311 + ], + "score": 1.0, + "content": "random sorting strategy. The result of AUUC show GALILEO gives a reasonable sorting to the RCT", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 309, + 196, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 196, + 323 + ], + "score": 1.0, + "content": "samples (see Fig. 14).", + "type": "text", + "cross_page": true + } + ], + "index": 15 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 710, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 226, + 82, + 384, + 209 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 226, + 82, + 384, + 209 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 226, + 82, + 384, + 209 + ], + "spans": [ + { + "bbox": [ + 226, + 82, + 384, + 209 + ], + "score": 0.97, + "type": "image", + "image_path": "eb37f5b144b993f4cd093333bca2ea018eace1f580556caafc4eb6ea290aeb5e.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 226, + 82, + 384, + 96.11111111111111 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 226, + 96.11111111111111, + 384, + 110.22222222222223 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 226, + 110.22222222222223, + 384, + 124.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 226, + 124.33333333333334, + 384, + 138.44444444444446 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 226, + 138.44444444444446, + 384, + 152.55555555555557 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 226, + 152.55555555555557, + 384, + 166.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 226, + 166.66666666666669, + 384, + 180.7777777777778 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 226, + 180.7777777777778, + 384, + 194.8888888888889 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 226, + 194.8888888888889, + 384, + 209.00000000000003 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 201, + 222, + 409, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 201, + 222, + 410, + 235 + ], + "spans": [ + { + "bbox": [ + 201, + 222, + 410, + 235 + ], + "score": 1.0, + "content": "Figure 14: Illustration of the AUUC result for BAT.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 254, + 506, + 321 + ], + "lines": [ + { + "bbox": [ + 106, + 254, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 506, + 267 + ], + "score": 1.0, + "content": "the Uplift Curve (AUUC) (Betlei et al., 2020). To plot AUUC, we first sort the RCT samples based", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 264, + 506, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 280 + ], + "score": 1.0, + "content": "on the predicted treatment effects. Then the cumulative treatment effects are computed by scanning", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 275, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 506, + 290 + ], + "score": 1.0, + "content": "the sorted sample list. If the sort order of the model predictions is better, the sample with larger", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 506, + 300 + ], + "score": 1.0, + "content": "treatment effects will be computed early. Then the area of AUUC will be larger than the one via a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 298, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 506, + 311 + ], + "score": 1.0, + "content": "random sorting strategy. The result of AUUC show GALILEO gives a reasonable sorting to the RCT", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 309, + 196, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 196, + 323 + ], + "score": 1.0, + "content": "samples (see Fig. 14).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 326, + 506, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 505, + 339 + ], + "score": 1.0, + "content": "Finally, we search for the optimal policy via the cross-entropy method planner (Hafner et al., 2019)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 337, + 505, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 350 + ], + "score": 1.0, + "content": "based on the learned model. We test the online supply improvement in 6 cities. The algorithm", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 348, + 506, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 506, + 361 + ], + "score": 1.0, + "content": "compared is a human-expert policy, which is also the behavior policy of the offline datasets. We", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "score": 1.0, + "content": "conduct online A/B tests for each of the cities. For each test, we randomly split a city into two", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 370, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 506, + 382 + ], + "score": 1.0, + "content": "partitions, one is for deploying the optimal policy learned from the GALILEO model, and the other is", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "score": 1.0, + "content": "as a control group, which keeps the human-expert policy as before. Before the intervention, we collect", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "score": 1.0, + "content": "10 days’ observation data and compute the averaged five-minute order-taken rates as the baselines", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 282, + 416 + ], + "score": 1.0, + "content": "of the treatment and control group, named", + "type": "text" + }, + { + "bbox": [ + 283, + 403, + 292, + 413 + ], + "score": 0.85, + "content": "b ^ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 403, + 311, + 416 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 311, + 403, + 321, + 414 + ], + "score": 0.86, + "content": "b ^ { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "respectively. Then we start intervention and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 413, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 505, + 427 + ], + "score": 1.0, + "content": "observe the five-minute order-taken rate in the following 14 days for the two groups. The results of the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 425, + 507, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 233, + 439 + ], + "score": 1.0, + "content": "treatment and control groups are", + "type": "text" + }, + { + "bbox": [ + 234, + 425, + 243, + 437 + ], + "score": 0.9, + "content": "y _ { i } ^ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 425, + 260, + 439 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 261, + 425, + 271, + 437 + ], + "score": 0.89, + "content": "y _ { i } ^ { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 425, + 348, + 439 + ], + "score": 1.0, + "content": "respectively, where", + "type": "text" + }, + { + "bbox": [ + 348, + 426, + 353, + 435 + ], + "score": 0.75, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 425, + 399, + 439 + ], + "score": 1.0, + "content": "denotes the", + "type": "text" + }, + { + "bbox": [ + 399, + 426, + 404, + 435 + ], + "score": 0.74, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 425, + 507, + 439 + ], + "score": 1.0, + "content": "-th day of the deployment.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "score": 1.0, + "content": "The percentage points of the supply improvement are computed via difference-in-difference (DID):", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 237, + 452, + 374, + 480 + ], + "lines": [ + { + "bbox": [ + 237, + 452, + 374, + 480 + ], + "spans": [ + { + "bbox": [ + 237, + 452, + 374, + 480 + ], + "score": 0.94, + "content": "\\frac { \\sum _ { i } ^ { T } ( y _ { i } ^ { t } - b ^ { t } ) - ( y _ { i } ^ { c } - b ^ { c } ) } { T } \\times 1 0 0 ,", + "type": "interline_equation", + "image_path": "2cb9d2f90f71f871223efc3b44a5dd77d13e117cc1308406538d937414424e6d.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 237, + 452, + 374, + 466.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 237, + 466.0, + 374, + 480.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 484, + 411, + 496 + ], + "lines": [ + { + "bbox": [ + 105, + 483, + 413, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 133, + 497 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 485, + 141, + 494 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 483, + 300, + 497 + ], + "score": 1.0, + "content": "is the total days of the intervention and", + "type": "text" + }, + { + "bbox": [ + 300, + 484, + 332, + 495 + ], + "score": 0.9, + "content": "T = 1 4", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 483, + 413, + 497 + ], + "score": 1.0, + "content": "in our experiments.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "table", + "bbox": [ + 187, + 538, + 420, + 606 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 515, + 502, + 537 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 265, + 528 + ], + "score": 1.0, + "content": "Table 7: Results on BAT. We use City-", + "type": "text" + }, + { + "bbox": [ + 265, + 516, + 274, + 525 + ], + "score": 0.35, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 514, + 505, + 528 + ], + "score": 1.0, + "content": "to denote the experiments on different cities. “pp” is an", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 526, + 356, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 356, + 539 + ], + "score": 1.0, + "content": "abbreviation of percentage points on the supply improvement.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "table_body", + "bbox": [ + 187, + 538, + 420, + 606 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 187, + 538, + 420, + 606 + ], + "spans": [ + { + "bbox": [ + 187, + 538, + 420, + 606 + ], + "score": 0.98, + "html": "
targetCity-ACity-BCity-C
supply improvement+1.63pp+0.79pp+0.27pp
targetCity-DCity-ECity-F
supply improvement+0.2pp+0.14pp+0.41pp
", + "type": "table", + "image_path": "f1acd02301fdd4d8b62e1604fd4e54f1d26dc6146602254dcedde6d5575f914d.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 187, + 538, + 420, + 551.6 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 187, + 551.6, + 420, + 565.2 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 187, + 565.2, + 420, + 578.8000000000001 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 187, + 578.8000000000001, + 420, + 592.4000000000001 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 187, + 592.4000000000001, + 420, + 606.0000000000001 + ], + "spans": [], + "index": 36 + } + ] + } + ], + "index": 32.25 + }, + { + "type": "text", + "bbox": [ + 106, + 617, + 505, + 674 + ], + "lines": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "The results are summarized in Tab. 7. The online experiment is conducted in 14 days and the results", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 628, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 506, + 641 + ], + "score": 1.0, + "content": "show that the policy learned with GALILEO can make better (the supply improvements are from 0.14", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 639, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 653 + ], + "score": 1.0, + "content": "to 1.63 percentage points) budget allocation than the behavior policies in all the testing cities. We", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 650, + 506, + 664 + ], + "spans": [ + { + "bbox": [ + 104, + 650, + 506, + 664 + ], + "score": 1.0, + "content": "give detailed results which record the supply difference between the treatment group and the control", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 662, + 176, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 176, + 674 + ], + "score": 1.0, + "content": "group in Fig. 15.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 39 + } + ], + "page_idx": 34, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 226, + 82, + 384, + 209 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 226, + 82, + 384, + 209 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 226, + 82, + 384, + 209 + ], + "spans": [ + { + "bbox": [ + 226, + 82, + 384, + 209 + ], + "score": 0.97, + "type": "image", + "image_path": "eb37f5b144b993f4cd093333bca2ea018eace1f580556caafc4eb6ea290aeb5e.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 226, + 82, + 384, + 96.11111111111111 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 226, + 96.11111111111111, + 384, + 110.22222222222223 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 226, + 110.22222222222223, + 384, + 124.33333333333334 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 226, + 124.33333333333334, + 384, + 138.44444444444446 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 226, + 138.44444444444446, + 384, + 152.55555555555557 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 226, + 152.55555555555557, + 384, + 166.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 226, + 166.66666666666669, + 384, + 180.7777777777778 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 226, + 180.7777777777778, + 384, + 194.8888888888889 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 226, + 194.8888888888889, + 384, + 209.00000000000003 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 201, + 222, + 409, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 201, + 222, + 410, + 235 + ], + "spans": [ + { + "bbox": [ + 201, + 222, + 410, + 235 + ], + "score": 1.0, + "content": "Figure 14: Illustration of the AUUC result for BAT.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 254, + 506, + 321 + ], + "lines": [], + "index": 12.5, + "bbox_fs": [ + 105, + 254, + 506, + 323 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 326, + 506, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 505, + 339 + ], + "score": 1.0, + "content": "Finally, we search for the optimal policy via the cross-entropy method planner (Hafner et al., 2019)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 337, + 505, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 350 + ], + "score": 1.0, + "content": "based on the learned model. We test the online supply improvement in 6 cities. The algorithm", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 348, + 506, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 506, + 361 + ], + "score": 1.0, + "content": "compared is a human-expert policy, which is also the behavior policy of the offline datasets. We", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "score": 1.0, + "content": "conduct online A/B tests for each of the cities. For each test, we randomly split a city into two", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 370, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 506, + 382 + ], + "score": 1.0, + "content": "partitions, one is for deploying the optimal policy learned from the GALILEO model, and the other is", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 506, + 394 + ], + "score": 1.0, + "content": "as a control group, which keeps the human-expert policy as before. Before the intervention, we collect", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "score": 1.0, + "content": "10 days’ observation data and compute the averaged five-minute order-taken rates as the baselines", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 282, + 416 + ], + "score": 1.0, + "content": "of the treatment and control group, named", + "type": "text" + }, + { + "bbox": [ + 283, + 403, + 292, + 413 + ], + "score": 0.85, + "content": "b ^ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 403, + 311, + 416 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 311, + 403, + 321, + 414 + ], + "score": 0.86, + "content": "b ^ { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "respectively. Then we start intervention and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 413, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 505, + 427 + ], + "score": 1.0, + "content": "observe the five-minute order-taken rate in the following 14 days for the two groups. The results of the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 425, + 507, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 425, + 233, + 439 + ], + "score": 1.0, + "content": "treatment and control groups are", + "type": "text" + }, + { + "bbox": [ + 234, + 425, + 243, + 437 + ], + "score": 0.9, + "content": "y _ { i } ^ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 425, + 260, + 439 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 261, + 425, + 271, + 437 + ], + "score": 0.89, + "content": "y _ { i } ^ { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 425, + 348, + 439 + ], + "score": 1.0, + "content": "respectively, where", + "type": "text" + }, + { + "bbox": [ + 348, + 426, + 353, + 435 + ], + "score": 0.75, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 425, + 399, + 439 + ], + "score": 1.0, + "content": "denotes the", + "type": "text" + }, + { + "bbox": [ + 399, + 426, + 404, + 435 + ], + "score": 0.74, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 425, + 507, + 439 + ], + "score": 1.0, + "content": "-th day of the deployment.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 505, + 450 + ], + "score": 1.0, + "content": "The percentage points of the supply improvement are computed via difference-in-difference (DID):", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 326, + 507, + 450 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 237, + 452, + 374, + 480 + ], + "lines": [ + { + "bbox": [ + 237, + 452, + 374, + 480 + ], + "spans": [ + { + "bbox": [ + 237, + 452, + 374, + 480 + ], + "score": 0.94, + "content": "\\frac { \\sum _ { i } ^ { T } ( y _ { i } ^ { t } - b ^ { t } ) - ( y _ { i } ^ { c } - b ^ { c } ) } { T } \\times 1 0 0 ,", + "type": "interline_equation", + "image_path": "2cb9d2f90f71f871223efc3b44a5dd77d13e117cc1308406538d937414424e6d.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 237, + 452, + 374, + 466.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 237, + 466.0, + 374, + 480.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 484, + 411, + 496 + ], + "lines": [ + { + "bbox": [ + 105, + 483, + 413, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 133, + 497 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 485, + 141, + 494 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 483, + 300, + 497 + ], + "score": 1.0, + "content": "is the total days of the intervention and", + "type": "text" + }, + { + "bbox": [ + 300, + 484, + 332, + 495 + ], + "score": 0.9, + "content": "T = 1 4", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 483, + 413, + 497 + ], + "score": 1.0, + "content": "in our experiments.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 483, + 413, + 497 + ] + }, + { + "type": "table", + "bbox": [ + 187, + 538, + 420, + 606 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 515, + 502, + 537 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 265, + 528 + ], + "score": 1.0, + "content": "Table 7: Results on BAT. We use City-", + "type": "text" + }, + { + "bbox": [ + 265, + 516, + 274, + 525 + ], + "score": 0.35, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 514, + 505, + 528 + ], + "score": 1.0, + "content": "to denote the experiments on different cities. “pp” is an", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 526, + 356, + 539 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 356, + 539 + ], + "score": 1.0, + "content": "abbreviation of percentage points on the supply improvement.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "table_body", + "bbox": [ + 187, + 538, + 420, + 606 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 187, + 538, + 420, + 606 + ], + "spans": [ + { + "bbox": [ + 187, + 538, + 420, + 606 + ], + "score": 0.98, + "html": "
targetCity-ACity-BCity-C
supply improvement+1.63pp+0.79pp+0.27pp
targetCity-DCity-ECity-F
supply improvement+0.2pp+0.14pp+0.41pp
", + "type": "table", + "image_path": "f1acd02301fdd4d8b62e1604fd4e54f1d26dc6146602254dcedde6d5575f914d.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 187, + 538, + 420, + 551.6 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 187, + 551.6, + 420, + 565.2 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 187, + 565.2, + 420, + 578.8000000000001 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 187, + 578.8000000000001, + 420, + 592.4000000000001 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 187, + 592.4000000000001, + 420, + 606.0000000000001 + ], + "spans": [], + "index": 36 + } + ] + } + ], + "index": 32.25 + }, + { + "type": "text", + "bbox": [ + 106, + 617, + 505, + 674 + ], + "lines": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "The results are summarized in Tab. 7. The online experiment is conducted in 14 days and the results", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 628, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 506, + 641 + ], + "score": 1.0, + "content": "show that the policy learned with GALILEO can make better (the supply improvements are from 0.14", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 639, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 653 + ], + "score": 1.0, + "content": "to 1.63 percentage points) budget allocation than the behavior policies in all the testing cities. We", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 650, + 506, + 664 + ], + "spans": [ + { + "bbox": [ + 104, + 650, + 506, + 664 + ], + "score": 1.0, + "content": "give detailed results which record the supply difference between the treatment group and the control", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 662, + 176, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 176, + 674 + ], + "score": 1.0, + "content": "group in Fig. 15.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 39, + "bbox_fs": [ + 104, + 618, + 506, + 674 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 117, + 278, + 494, + 513 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 278, + 494, + 513 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 278, + 494, + 513 + ], + "spans": [ + { + "bbox": [ + 117, + 278, + 494, + 513 + ], + "score": 0.977, + "type": "image", + "image_path": "1a2796121e8ea865a6cc920895fd3318ac583274799e2549cdb6430e7ce60dd6.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 117, + 278, + 494, + 356.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 117, + 356.3333333333333, + 494, + 434.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 117, + 434.66666666666663, + 494, + 513.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 156, + 525, + 454, + 538 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 156, + 524, + 455, + 539 + ], + "spans": [ + { + "bbox": [ + 156, + 524, + 455, + 539 + ], + "score": 1.0, + "content": "Figure 15: Illustration of the daily responses in the A/B test in the 6 cities.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 35, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "36", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 117, + 278, + 494, + 513 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 278, + 494, + 513 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 278, + 494, + 513 + ], + "spans": [ + { + "bbox": [ + 117, + 278, + 494, + 513 + ], + "score": 0.977, + "type": "image", + "image_path": "1a2796121e8ea865a6cc920895fd3318ac583274799e2549cdb6430e7ce60dd6.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 117, + 278, + 494, + 356.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 117, + 356.3333333333333, + 494, + 434.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 117, + 434.66666666666663, + 494, + 513.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 156, + 525, + 454, + 538 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 156, + 524, + 455, + 539 + ], + "spans": [ + { + "bbox": [ + 156, + 524, + 455, + 539 + ], + "score": 1.0, + "content": "Figure 15: Illustration of the daily responses in the A/B test in the 6 cities.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 177, + 113, + 434, + 314 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 91, + 504, + 113 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 89, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 104, + 89, + 144, + 105 + ], + "score": 1.0, + "content": "Table 8:", + "type": "text" + }, + { + "bbox": [ + 145, + 90, + 185, + 102 + ], + "score": 0.87, + "content": "\\sqrt { M I S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 89, + 430, + 105 + ], + "score": 1.0, + "content": "results on GNFC. We bold the lowest error for each task.", + "type": "text" + }, + { + "bbox": [ + 430, + 92, + 441, + 102 + ], + "score": 0.58, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 89, + 506, + 105 + ], + "score": 1.0, + "content": "is the standard", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 102, + 237, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 237, + 114 + ], + "score": 1.0, + "content": "deviation of three random seeds.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 177, + 113, + 434, + 314 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 177, + 113, + 434, + 314 + ], + "spans": [ + { + "bbox": [ + 177, + 113, + 434, + 314 + ], + "score": 0.981, + "html": "
e1_p1e0.2_p1e0.05_p1
GALILEO5.17 ± 0.064.73 ± 0.134.70 ± 0.02
SL5.15 ± 0.234.73 ± 0.3123.64 ± 4.86
IPW5.22 ± 0.095.50 ± 0.015.02 ± 0.07
SCIGAN7.05 ± 0.526.58 ± 0.5818.55 ± 3.50
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO5.03 ± 0.094.72 ± 0.054.87 ± 0.15
SL5.21 ± 0.636.74 ± 0.1533.52 ±1.32
IPW5.27 ± 0.055.69 ± 0.0020.23 ± 0.45
SCIGAN16.07 ± 0.2712.07 ± 1.9319.27 ± 10.72
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO5.23 ± 0.415.01 ± 0.086.17 ± 0.33
SL5.89 ± 0.8814.25 ± 3.4837.50 ± 2.29
IPW5.21 ± 0.015.52 ± 0.4431.95 ± 0.05
SCIGAN11.50 ± 7.7613.05 ± 4.1925.74 ± 8.30
", + "type": "table", + "image_path": "94d6e978af6f49e3cb43fe74b880b37bd0df7a36aceed850b0753e61885b8634.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 177, + 113, + 434, + 127.35714285714286 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 177, + 127.35714285714286, + 434, + 141.71428571428572 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 177, + 141.71428571428572, + 434, + 156.07142857142858 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 177, + 156.07142857142858, + 434, + 170.42857142857144 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 177, + 170.42857142857144, + 434, + 184.7857142857143 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 177, + 184.7857142857143, + 434, + 199.14285714285717 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 177, + 199.14285714285717, + 434, + 213.50000000000003 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 177, + 213.50000000000003, + 434, + 227.8571428571429 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 177, + 227.8571428571429, + 434, + 242.21428571428575 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 177, + 242.21428571428575, + 434, + 256.5714285714286 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 177, + 256.5714285714286, + 434, + 270.92857142857144 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 177, + 270.92857142857144, + 434, + 285.2857142857143 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 177, + 285.2857142857143, + 434, + 299.6428571428571 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 177, + 299.6428571428571, + 434, + 313.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 4.5 + }, + { + "type": "table", + "bbox": [ + 183, + 357, + 427, + 557 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 335, + 502, + 356 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 104, + 333, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 104, + 333, + 144, + 348 + ], + "score": 1.0, + "content": "Table 9:", + "type": "text" + }, + { + "bbox": [ + 144, + 334, + 185, + 346 + ], + "score": 0.89, + "content": "\\sqrt { M I S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 333, + 430, + 348 + ], + "score": 1.0, + "content": "results on TCGA. We bold the lowest error for each task.", + "type": "text" + }, + { + "bbox": [ + 430, + 336, + 441, + 346 + ], + "score": 0.33, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 333, + 505, + 348 + ], + "score": 1.0, + "content": "is the standard", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 346, + 238, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 238, + 358 + ], + "score": 1.0, + "content": "deviation of three random seeds.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "table_body", + "bbox": [ + 183, + 357, + 427, + 557 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 183, + 357, + 427, + 557 + ], + "spans": [ + { + "bbox": [ + 183, + 357, + 427, + 557 + ], + "score": 0.979, + "html": "
t0_bias_2.0 t0_bias_s_20.0 t0_bias_50.0
GALILEO SL0.34 ± 0.05 0.67 ± 0.13 0.38 ± 0.13 1.50 ± 0.312.04 ± 0.12 3.06 ± 0.65
IPW SCIGAN6.57 ± 1.16 0.74 ± 0.056.88 ± 0.30 5.84 ± 0.71 2.74 ± 0.35 3.19 ± 0.09
t1_bias_2.0t1_bias_6.0 t1_bias_8.0
GALILEO SL0.43 ± 0.05 0.47 ± 0.050.25 ± 0.02 0.21 ± 0.04 1.33 ± 0.97 1.18 ± 0.73
IPW3.67 ± 2.370.54 ± 0.13 2.69 ± 1.17
SCIGAN0.45 ± 0.251.08 ± 1.04 1.01 ± 0.77
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO
1.46 ± 0.090.85 ± 0.04 0.46 ± 0.01
SL IPW0.81 ± 0.143.74 ± 2.04 3.59 ± 0.14
", + "type": "table", + "image_path": "44b6289cb60800febc0eeed8fbf9b650658828a2122f74672b8671c30af9c4d0.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 183, + 357, + 427, + 371.2857142857143 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 183, + 371.2857142857143, + 427, + 385.57142857142856 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 183, + 385.57142857142856, + 427, + 399.85714285714283 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 183, + 399.85714285714283, + 427, + 414.1428571428571 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 183, + 414.1428571428571, + 427, + 428.4285714285714 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 183, + 428.4285714285714, + 427, + 442.71428571428567 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 183, + 442.71428571428567, + 427, + 456.99999999999994 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 183, + 456.99999999999994, + 427, + 471.2857142857142 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 183, + 471.2857142857142, + 427, + 485.5714285714285 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 183, + 485.5714285714285, + 427, + 499.8571428571428 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 183, + 499.8571428571428, + 427, + 514.1428571428571 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 183, + 514.1428571428571, + 427, + 528.4285714285714 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 183, + 528.4285714285714, + 427, + 542.7142857142858 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 183, + 542.7142857142858, + 427, + 557.0000000000001 + ], + "spans": [], + "index": 31 + } + ] + } + ], + "index": 20.5 + } + ], + "page_idx": 36, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 177, + 113, + 434, + 314 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 105, + 91, + 504, + 113 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 89, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 104, + 89, + 144, + 105 + ], + "score": 1.0, + "content": "Table 8:", + "type": "text" + }, + { + "bbox": [ + 145, + 90, + 185, + 102 + ], + "score": 0.87, + "content": "\\sqrt { M I S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 89, + 430, + 105 + ], + "score": 1.0, + "content": "results on GNFC. We bold the lowest error for each task.", + "type": "text" + }, + { + "bbox": [ + 430, + 92, + 441, + 102 + ], + "score": 0.58, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 89, + 506, + 105 + ], + "score": 1.0, + "content": "is the standard", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 102, + 237, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 102, + 237, + 114 + ], + "score": 1.0, + "content": "deviation of three random seeds.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 177, + 113, + 434, + 314 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 177, + 113, + 434, + 314 + ], + "spans": [ + { + "bbox": [ + 177, + 113, + 434, + 314 + ], + "score": 0.981, + "html": "
e1_p1e0.2_p1e0.05_p1
GALILEO5.17 ± 0.064.73 ± 0.134.70 ± 0.02
SL5.15 ± 0.234.73 ± 0.3123.64 ± 4.86
IPW5.22 ± 0.095.50 ± 0.015.02 ± 0.07
SCIGAN7.05 ± 0.526.58 ± 0.5818.55 ± 3.50
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO5.03 ± 0.094.72 ± 0.054.87 ± 0.15
SL5.21 ± 0.636.74 ± 0.1533.52 ±1.32
IPW5.27 ± 0.055.69 ± 0.0020.23 ± 0.45
SCIGAN16.07 ± 0.2712.07 ± 1.9319.27 ± 10.72
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO5.23 ± 0.415.01 ± 0.086.17 ± 0.33
SL5.89 ± 0.8814.25 ± 3.4837.50 ± 2.29
IPW5.21 ± 0.015.52 ± 0.4431.95 ± 0.05
SCIGAN11.50 ± 7.7613.05 ± 4.1925.74 ± 8.30
", + "type": "table", + "image_path": "94d6e978af6f49e3cb43fe74b880b37bd0df7a36aceed850b0753e61885b8634.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 177, + 113, + 434, + 127.35714285714286 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 177, + 127.35714285714286, + 434, + 141.71428571428572 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 177, + 141.71428571428572, + 434, + 156.07142857142858 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 177, + 156.07142857142858, + 434, + 170.42857142857144 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 177, + 170.42857142857144, + 434, + 184.7857142857143 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 177, + 184.7857142857143, + 434, + 199.14285714285717 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 177, + 199.14285714285717, + 434, + 213.50000000000003 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 177, + 213.50000000000003, + 434, + 227.8571428571429 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 177, + 227.8571428571429, + 434, + 242.21428571428575 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 177, + 242.21428571428575, + 434, + 256.5714285714286 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 177, + 256.5714285714286, + 434, + 270.92857142857144 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 177, + 270.92857142857144, + 434, + 285.2857142857143 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 177, + 285.2857142857143, + 434, + 299.6428571428571 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 177, + 299.6428571428571, + 434, + 313.99999999999994 + ], + "spans": [], + "index": 15 + } + ] + } + ], + "index": 4.5 + }, + { + "type": "table", + "bbox": [ + 183, + 357, + 427, + 557 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 335, + 502, + 356 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 104, + 333, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 104, + 333, + 144, + 348 + ], + "score": 1.0, + "content": "Table 9:", + "type": "text" + }, + { + "bbox": [ + 144, + 334, + 185, + 346 + ], + "score": 0.89, + "content": "\\sqrt { M I S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 333, + 430, + 348 + ], + "score": 1.0, + "content": "results on TCGA. We bold the lowest error for each task.", + "type": "text" + }, + { + "bbox": [ + 430, + 336, + 441, + 346 + ], + "score": 0.33, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 333, + 505, + 348 + ], + "score": 1.0, + "content": "is the standard", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 346, + 238, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 238, + 358 + ], + "score": 1.0, + "content": "deviation of three random seeds.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "table_body", + "bbox": [ + 183, + 357, + 427, + 557 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 183, + 357, + 427, + 557 + ], + "spans": [ + { + "bbox": [ + 183, + 357, + 427, + 557 + ], + "score": 0.979, + "html": "
t0_bias_2.0 t0_bias_s_20.0 t0_bias_50.0
GALILEO SL0.34 ± 0.05 0.67 ± 0.13 0.38 ± 0.13 1.50 ± 0.312.04 ± 0.12 3.06 ± 0.65
IPW SCIGAN6.57 ± 1.16 0.74 ± 0.056.88 ± 0.30 5.84 ± 0.71 2.74 ± 0.35 3.19 ± 0.09
t1_bias_2.0t1_bias_6.0 t1_bias_8.0
GALILEO SL0.43 ± 0.05 0.47 ± 0.050.25 ± 0.02 0.21 ± 0.04 1.33 ± 0.97 1.18 ± 0.73
IPW3.67 ± 2.370.54 ± 0.13 2.69 ± 1.17
SCIGAN0.45 ± 0.251.08 ± 1.04 1.01 ± 0.77
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO
1.46 ± 0.090.85 ± 0.04 0.46 ± 0.01
SL IPW0.81 ± 0.143.74 ± 2.04 3.59 ± 0.14
", + "type": "table", + "image_path": "44b6289cb60800febc0eeed8fbf9b650658828a2122f74672b8671c30af9c4d0.jpg" + } + ] + } + ], + "index": 24.5, + "virtual_lines": [ + { + "bbox": [ + 183, + 357, + 427, + 371.2857142857143 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 183, + 371.2857142857143, + 427, + 385.57142857142856 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 183, + 385.57142857142856, + 427, + 399.85714285714283 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 183, + 399.85714285714283, + 427, + 414.1428571428571 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 183, + 414.1428571428571, + 427, + 428.4285714285714 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 183, + 428.4285714285714, + 427, + 442.71428571428567 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 183, + 442.71428571428567, + 427, + 456.99999999999994 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 183, + 456.99999999999994, + 427, + 471.2857142857142 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 183, + 471.2857142857142, + 427, + 485.5714285714285 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 183, + 485.5714285714285, + 427, + 499.8571428571428 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 183, + 499.8571428571428, + 427, + 514.1428571428571 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 183, + 514.1428571428571, + 427, + 528.4285714285714 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 183, + 528.4285714285714, + 427, + 542.7142857142858 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 183, + 542.7142857142858, + 427, + 557.0000000000001 + ], + "spans": [], + "index": 31 + } + ] + } + ], + "index": 20.5 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 180, + 158, + 432, + 358 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 104, + 135, + 504, + 157 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 134, + 505, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 147, + 147 + ], + "score": 1.0, + "content": "Table 10:", + "type": "text" + }, + { + "bbox": [ + 148, + 134, + 194, + 147 + ], + "score": 0.91, + "content": "\\sqrt { M M S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 134, + 432, + 147 + ], + "score": 1.0, + "content": "results on GNFC. We bold the lowest error for each task.", + "type": "text" + }, + { + "bbox": [ + 432, + 136, + 442, + 146 + ], + "score": 0.68, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 134, + 505, + 147 + ], + "score": 1.0, + "content": "is the standard", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 147, + 238, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 238, + 158 + ], + "score": 1.0, + "content": "deviation of three random seeds.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 180, + 158, + 432, + 358 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 158, + 432, + 358 + ], + "spans": [ + { + "bbox": [ + 180, + 158, + 432, + 358 + ], + "score": 0.982, + "html": "
e1_p1e0.2_p1e0.05_p1
GALILEO3.86 ± 0.033.99 ± 0.014.07 ± 0.03
SL5.73 ± 0.335.80 ± 0.2818.78 ± 3.13
IPW4.02 ± 0.054.15 ± 0.1222.66 ± 0.33
SCIGAN8.84 ± 0.5412.62 ± 2.1724.21 ± 5.20
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO4.13 ± 0.104.11 ± 0.154.21 ± 0.15
SL5.87 ± 0.437.44 ± 1.1329.13 ± 3.44
IPW4.12 ± 0.026.12 ± 0.4830.96 ± 0.17
SCIGAN12.87 ± 3.0214.59 ± 2.1324.57 ± 3.00
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO4.39 ± 0.204.34 ± 0.205.26 ± 0.29
SL6.12 ± 0.4314.88 ± 4.4130.81 ± 1.69
IPW13.60 ± 7.8326.27 ± 2.6732.55 ± 0.12
SCIGAN9.19 ± 1.0415.08 ± 1.2617.52 ± 0.02
", + "type": "table", + "image_path": "7c570ae7c458cd9f5d3077904feab0408ba171a8ed0836f36684ec0bea2ee2c6.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 180, + 158, + 432, + 170.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 180, + 170.5, + 432, + 183.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 180, + 183.0, + 432, + 195.5 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 180, + 195.5, + 432, + 208.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 180, + 208.0, + 432, + 220.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 180, + 220.5, + 432, + 233.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 180, + 233.0, + 432, + 245.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 180, + 245.5, + 432, + 258.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 180, + 258.0, + 432, + 270.5 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 180, + 270.5, + 432, + 283.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 180, + 283.0, + 432, + 295.5 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 180, + 295.5, + 432, + 308.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 180, + 308.0, + 432, + 320.5 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 180, + 320.5, + 432, + 333.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 180, + 333.0, + 432, + 345.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 180, + 345.5, + 432, + 358.0 + ], + "spans": [], + "index": 17 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "table", + "bbox": [ + 183, + 484, + 427, + 686 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 104, + 462, + 503, + 485 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 461, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 148, + 476 + ], + "score": 1.0, + "content": "Table 11:", + "type": "text" + }, + { + "bbox": [ + 148, + 462, + 194, + 475 + ], + "score": 0.91, + "content": "\\sqrt { M M S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 461, + 432, + 476 + ], + "score": 1.0, + "content": "results on TCGA. We bold the lowest error for each task.", + "type": "text" + }, + { + "bbox": [ + 432, + 464, + 442, + 474 + ], + "score": 0.64, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 461, + 505, + 476 + ], + "score": 1.0, + "content": "is the standard", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 475, + 237, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 237, + 486 + ], + "score": 1.0, + "content": "deviation of three random seeds.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "table_body", + "bbox": [ + 183, + 484, + 427, + 686 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 183, + 484, + 427, + 686 + ], + "spans": [ + { + "bbox": [ + 183, + 484, + 427, + 686 + ], + "score": 0.981, + "html": "
t0_bias_2.0t0_bias_20.0 tO_bias_50.0
GALILEO1.56 ± 0.041.96 ± 0.53 3.16 ± 0.13
SL1.92 ± 0.672.31 ± 0.19 5.11 ± 0.66
IPW7.42 ± 0.465.36 ± 0.96 5.38 ± 1.24
SCIGAN2.11 ± 0.475.23 ±0.27 5.59 ± 1.02
tl_bias_2.0tl_bias_6.0 tl_bias_8.0
GALILEO1.43 ± 0.061.09 ± 0.05 1.36 ± 0.36
SL1.12 ± 0.153.65 ± 1.91 3.96 ± 1.81
IPW1.14 ± 0.11 0.90 ± 0.092.04 ± 0.99
SCIGAN3.32 ± 0.88 4.74 ± 2.125.17 ± 2.42
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO3.77 ± 0.35 3.99 ± 0.402.08 ± 0.60
SL2.70 ± 0.67 8.33 ± 5.059.70 ± 3.12
IPW2.92 ± 0.15 3.90 ± 0.174.47 ± 2.16
SCIGAN3.82 ± 2.12 1.83 ± 1.493.62 ± 4.9
", + "type": "table", + "image_path": "7f0576f822a959d36f6b1cef1e840ad20fc9bb83231043ef7d8337d604ea2f7a.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 183, + 484, + 427, + 496.625 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 183, + 496.625, + 427, + 509.25 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 183, + 509.25, + 427, + 521.875 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 183, + 521.875, + 427, + 534.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 183, + 534.5, + 427, + 547.125 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 183, + 547.125, + 427, + 559.75 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 183, + 559.75, + 427, + 572.375 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 183, + 572.375, + 427, + 585.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 183, + 585.0, + 427, + 597.625 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 183, + 597.625, + 427, + 610.25 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 183, + 610.25, + 427, + 622.875 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 183, + 622.875, + 427, + 635.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 183, + 635.5, + 427, + 648.125 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 183, + 648.125, + 427, + 660.75 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 183, + 660.75, + 427, + 673.375 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 183, + 673.375, + 427, + 686.0 + ], + "spans": [], + "index": 35 + } + ] + } + ], + "index": 23.0 + } + ], + "page_idx": 37, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 26, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "38", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 180, + 158, + 432, + 358 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 104, + 135, + 504, + 157 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 134, + 505, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 147, + 147 + ], + "score": 1.0, + "content": "Table 10:", + "type": "text" + }, + { + "bbox": [ + 148, + 134, + 194, + 147 + ], + "score": 0.91, + "content": "\\sqrt { M M S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 134, + 432, + 147 + ], + "score": 1.0, + "content": "results on GNFC. We bold the lowest error for each task.", + "type": "text" + }, + { + "bbox": [ + 432, + 136, + 442, + 146 + ], + "score": 0.68, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 134, + 505, + 147 + ], + "score": 1.0, + "content": "is the standard", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 147, + 238, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 238, + 158 + ], + "score": 1.0, + "content": "deviation of three random seeds.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 180, + 158, + 432, + 358 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 158, + 432, + 358 + ], + "spans": [ + { + "bbox": [ + 180, + 158, + 432, + 358 + ], + "score": 0.982, + "html": "
e1_p1e0.2_p1e0.05_p1
GALILEO3.86 ± 0.033.99 ± 0.014.07 ± 0.03
SL5.73 ± 0.335.80 ± 0.2818.78 ± 3.13
IPW4.02 ± 0.054.15 ± 0.1222.66 ± 0.33
SCIGAN8.84 ± 0.5412.62 ± 2.1724.21 ± 5.20
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO4.13 ± 0.104.11 ± 0.154.21 ± 0.15
SL5.87 ± 0.437.44 ± 1.1329.13 ± 3.44
IPW4.12 ± 0.026.12 ± 0.4830.96 ± 0.17
SCIGAN12.87 ± 3.0214.59 ± 2.1324.57 ± 3.00
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO4.39 ± 0.204.34 ± 0.205.26 ± 0.29
SL6.12 ± 0.4314.88 ± 4.4130.81 ± 1.69
IPW13.60 ± 7.8326.27 ± 2.6732.55 ± 0.12
SCIGAN9.19 ± 1.0415.08 ± 1.2617.52 ± 0.02
", + "type": "table", + "image_path": "7c570ae7c458cd9f5d3077904feab0408ba171a8ed0836f36684ec0bea2ee2c6.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 180, + 158, + 432, + 170.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 180, + 170.5, + 432, + 183.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 180, + 183.0, + 432, + 195.5 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 180, + 195.5, + 432, + 208.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 180, + 208.0, + 432, + 220.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 180, + 220.5, + 432, + 233.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 180, + 233.0, + 432, + 245.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 180, + 245.5, + 432, + 258.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 180, + 258.0, + 432, + 270.5 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 180, + 270.5, + 432, + 283.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 180, + 283.0, + 432, + 295.5 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 180, + 295.5, + 432, + 308.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 180, + 308.0, + 432, + 320.5 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 180, + 320.5, + 432, + 333.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 180, + 333.0, + 432, + 345.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 180, + 345.5, + 432, + 358.0 + ], + "spans": [], + "index": 17 + } + ] + } + ], + "index": 5.0 + }, + { + "type": "table", + "bbox": [ + 183, + 484, + 427, + 686 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 104, + 462, + 503, + 485 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 461, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 148, + 476 + ], + "score": 1.0, + "content": "Table 11:", + "type": "text" + }, + { + "bbox": [ + 148, + 462, + 194, + 475 + ], + "score": 0.91, + "content": "\\sqrt { M M S E }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 461, + 432, + 476 + ], + "score": 1.0, + "content": "results on TCGA. We bold the lowest error for each task.", + "type": "text" + }, + { + "bbox": [ + 432, + 464, + 442, + 474 + ], + "score": 0.64, + "content": "\\pm", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 461, + 505, + 476 + ], + "score": 1.0, + "content": "is the standard", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 475, + 237, + 486 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 237, + 486 + ], + "score": 1.0, + "content": "deviation of three random seeds.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "table_body", + "bbox": [ + 183, + 484, + 427, + 686 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 183, + 484, + 427, + 686 + ], + "spans": [ + { + "bbox": [ + 183, + 484, + 427, + 686 + ], + "score": 0.981, + "html": "
t0_bias_2.0t0_bias_20.0 tO_bias_50.0
GALILEO1.56 ± 0.041.96 ± 0.53 3.16 ± 0.13
SL1.92 ± 0.672.31 ± 0.19 5.11 ± 0.66
IPW7.42 ± 0.465.36 ± 0.96 5.38 ± 1.24
SCIGAN2.11 ± 0.475.23 ±0.27 5.59 ± 1.02
tl_bias_2.0tl_bias_6.0 tl_bias_8.0
GALILEO1.43 ± 0.061.09 ± 0.05 1.36 ± 0.36
SL1.12 ± 0.153.65 ± 1.91 3.96 ± 1.81
IPW1.14 ± 0.11 0.90 ± 0.092.04 ± 0.99
SCIGAN3.32 ± 0.88 4.74 ± 2.125.17 ± 2.42
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO3.77 ± 0.35 3.99 ± 0.402.08 ± 0.60
SL2.70 ± 0.67 8.33 ± 5.059.70 ± 3.12
IPW2.92 ± 0.15 3.90 ± 0.174.47 ± 2.16
SCIGAN3.82 ± 2.12 1.83 ± 1.493.62 ± 4.9
", + "type": "table", + "image_path": "7f0576f822a959d36f6b1cef1e840ad20fc9bb83231043ef7d8337d604ea2f7a.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 183, + 484, + 427, + 496.625 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 183, + 496.625, + 427, + 509.25 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 183, + 509.25, + 427, + 521.875 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 183, + 521.875, + 427, + 534.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 183, + 534.5, + 427, + 547.125 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 183, + 547.125, + 427, + 559.75 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 183, + 559.75, + 427, + 572.375 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 183, + 572.375, + 427, + 585.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 183, + 585.0, + 427, + 597.625 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 183, + 597.625, + 427, + 610.25 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 183, + 610.25, + 427, + 622.875 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 183, + 622.875, + 427, + 635.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 183, + 635.5, + 427, + 648.125 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 183, + 648.125, + 427, + 660.75 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 183, + 660.75, + 427, + 673.375 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 183, + 673.375, + 427, + 686.0 + ], + "spans": [], + "index": 35 + } + ] + } + ], + "index": 23.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 117, + 241, + 494, + 552 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 241, + 494, + 552 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 241, + 494, + 552 + ], + "spans": [ + { + "bbox": [ + 117, + 241, + 494, + 552 + ], + "score": 0.976, + "type": "image", + "image_path": "0f4a66ef4baf70e73f3c6d49b3d583ae88a89e44b95edef40bd86d6ec7c7718d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 117, + 241, + 494, + 344.6666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 117, + 344.6666666666667, + 494, + 448.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 117, + 448.33333333333337, + 494, + 552.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 114, + 564, + 493, + 576 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 563, + 494, + 577 + ], + "spans": [ + { + "bbox": [ + 114, + 563, + 494, + 577 + ], + "score": 1.0, + "content": "Figure 16: Illustration of the averaged response curves of Supervised Learning (SL) in TCGA.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 38, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "39", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 117, + 241, + 494, + 552 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 241, + 494, + 552 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 117, + 241, + 494, + 552 + ], + "spans": [ + { + "bbox": [ + 117, + 241, + 494, + 552 + ], + "score": 0.976, + "type": "image", + "image_path": "0f4a66ef4baf70e73f3c6d49b3d583ae88a89e44b95edef40bd86d6ec7c7718d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 117, + 241, + 494, + 344.6666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 117, + 344.6666666666667, + 494, + 448.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 117, + 448.33333333333337, + 494, + 552.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 114, + 564, + 493, + 576 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 563, + 494, + 577 + ], + "spans": [ + { + "bbox": [ + 114, + 563, + 494, + 577 + ], + "score": 1.0, + "content": "Figure 16: Illustration of the averaged response curves of Supervised Learning (SL) in TCGA.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 240, + 493, + 553 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 240, + 493, + 553 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 240, + 493, + 553 + ], + "spans": [ + { + "bbox": [ + 116, + 240, + 493, + 553 + ], + "score": 0.976, + "type": "image", + "image_path": "2abae9ea1c75f9d0de485507d99f7e3fc9e0798cab13f63e1ba494cd5a148ac7.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 240, + 493, + 344.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 344.3333333333333, + 493, + 448.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 448.66666666666663, + 493, + 553.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 113, + 564, + 492, + 577 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 564, + 493, + 578 + ], + "spans": [ + { + "bbox": [ + 114, + 564, + 493, + 578 + ], + "score": 1.0, + "content": "Figure 17: Illustration of the averaged response curves of Supervised Learning (SL) in GNFC.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 39, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "40", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 240, + 493, + 553 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 240, + 493, + 553 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 240, + 493, + 553 + ], + "spans": [ + { + "bbox": [ + 116, + 240, + 493, + 553 + ], + "score": 0.976, + "type": "image", + "image_path": "2abae9ea1c75f9d0de485507d99f7e3fc9e0798cab13f63e1ba494cd5a148ac7.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 240, + 493, + 344.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 344.3333333333333, + 493, + 448.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 448.66666666666663, + 493, + 553.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 113, + 564, + 492, + 577 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 564, + 493, + 578 + ], + "spans": [ + { + "bbox": [ + 114, + 564, + 493, + 578 + ], + "score": 1.0, + "content": "Figure 17: Illustration of the averaged response curves of Supervised Learning (SL) in GNFC.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 223, + 496, + 560 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 223, + 496, + 560 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 223, + 496, + 560 + ], + "spans": [ + { + "bbox": [ + 115, + 223, + 496, + 560 + ], + "score": 0.976, + "type": "image", + "image_path": "e95af2c71f12750c1e7632c99166d6e0f6065e19f98d35100fc5980516143630.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 223, + 496, + 335.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 335.3333333333333, + 496, + 447.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 447.66666666666663, + 496, + 560.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 571, + 506, + 594 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 569, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 104, + 569, + 506, + 586 + ], + "score": 1.0, + "content": "Figure 18: Illustration of the averaged response curves of Inverse Propensity Weighting (IPW) in", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 581, + 139, + 594 + ], + "spans": [ + { + "bbox": [ + 104, + 581, + 139, + 594 + ], + "score": 1.0, + "content": "TCGA.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + } + ], + "page_idx": 40, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 223, + 496, + 560 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 223, + 496, + 560 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 223, + 496, + 560 + ], + "spans": [ + { + "bbox": [ + 115, + 223, + 496, + 560 + ], + "score": 0.976, + "type": "image", + "image_path": "e95af2c71f12750c1e7632c99166d6e0f6065e19f98d35100fc5980516143630.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 223, + 496, + 335.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 335.3333333333333, + 496, + 447.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 447.66666666666663, + 496, + 560.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 571, + 506, + 594 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 569, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 104, + 569, + 506, + 586 + ], + "score": 1.0, + "content": "Figure 18: Illustration of the averaged response curves of Inverse Propensity Weighting (IPW) in", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 581, + 139, + 594 + ], + "spans": [ + { + "bbox": [ + 104, + 581, + 139, + 594 + ], + "score": 1.0, + "content": "TCGA.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 225, + 495, + 559 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 225, + 495, + 559 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 225, + 495, + 559 + ], + "spans": [ + { + "bbox": [ + 115, + 225, + 495, + 559 + ], + "score": 0.976, + "type": "image", + "image_path": "cef37522bd881e3dbc177930bf4e98a750e94226e9d0ed7fc4238330fda9b47a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 225, + 495, + 336.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 336.3333333333333, + 495, + 447.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 447.66666666666663, + 495, + 559.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 569, + 505, + 592 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 568, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 104, + 568, + 505, + 583 + ], + "score": 1.0, + "content": "Figure 19: Illustration of the averaged response curves of Inverse Propensity Weighting (IPW) in", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 579, + 139, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 139, + 593 + ], + "score": 1.0, + "content": "GNFC.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + } + ], + "page_idx": 41, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "42", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 225, + 495, + 559 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 225, + 495, + 559 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 225, + 495, + 559 + ], + "spans": [ + { + "bbox": [ + 115, + 225, + 495, + 559 + ], + "score": 0.976, + "type": "image", + "image_path": "cef37522bd881e3dbc177930bf4e98a750e94226e9d0ed7fc4238330fda9b47a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 225, + 495, + 336.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 336.3333333333333, + 495, + 447.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 447.66666666666663, + 495, + 559.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 105, + 569, + 505, + 592 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 568, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 104, + 568, + 505, + 583 + ], + "score": 1.0, + "content": "Figure 19: Illustration of the averaged response curves of Inverse Propensity Weighting (IPW) in", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 579, + 139, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 139, + 593 + ], + "score": 1.0, + "content": "GNFC.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 241, + 494, + 552 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 241, + 494, + 552 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 241, + 494, + 552 + ], + "spans": [ + { + "bbox": [ + 116, + 241, + 494, + 552 + ], + "score": 0.976, + "type": "image", + "image_path": "fa953ff9d5f43062141219c074a774984b11e3b159955de3d0d2558401086b92.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 241, + 494, + 344.6666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 344.6666666666667, + 494, + 448.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 448.33333333333337, + 494, + 552.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 148, + 564, + 460, + 576 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 563, + 461, + 577 + ], + "spans": [ + { + "bbox": [ + 147, + 563, + 461, + 577 + ], + "score": 1.0, + "content": "Figure 20: Illustration of the averaged response curves of SCIGAN in TCGA.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 42, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "43", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 241, + 494, + 552 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 241, + 494, + 552 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 241, + 494, + 552 + ], + "spans": [ + { + "bbox": [ + 116, + 241, + 494, + 552 + ], + "score": 0.976, + "type": "image", + "image_path": "fa953ff9d5f43062141219c074a774984b11e3b159955de3d0d2558401086b92.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 241, + 494, + 344.6666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 344.6666666666667, + 494, + 448.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 448.33333333333337, + 494, + 552.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 148, + 564, + 460, + 576 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 563, + 461, + 577 + ], + "spans": [ + { + "bbox": [ + 147, + 563, + 461, + 577 + ], + "score": 1.0, + "content": "Figure 20: Illustration of the averaged response curves of SCIGAN in TCGA.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 240, + 493, + 554 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 240, + 493, + 554 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 240, + 493, + 554 + ], + "spans": [ + { + "bbox": [ + 116, + 240, + 493, + 554 + ], + "score": 0.975, + "type": "image", + "image_path": "f7f3fe5f9e09c0042241c841726d9b8c5004cb2f7cb0769552686dc09c60b91f.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 240, + 493, + 344.6666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 344.6666666666667, + 493, + 449.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 449.33333333333337, + 493, + 554.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 149, + 565, + 460, + 578 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 564, + 460, + 579 + ], + "spans": [ + { + "bbox": [ + 147, + 564, + 460, + 579 + ], + "score": 1.0, + "content": "Figure 21: Illustration of the averaged response curves of SCIGAN in GNFC.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 43, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "44", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 240, + 493, + 554 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 240, + 493, + 554 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 240, + 493, + 554 + ], + "spans": [ + { + "bbox": [ + 116, + 240, + 493, + 554 + ], + "score": 0.975, + "type": "image", + "image_path": "f7f3fe5f9e09c0042241c841726d9b8c5004cb2f7cb0769552686dc09c60b91f.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 240, + 493, + 344.6666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 344.6666666666667, + 493, + 449.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 449.33333333333337, + 493, + 554.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 149, + 565, + 460, + 578 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 564, + 460, + 579 + ], + "spans": [ + { + "bbox": [ + 147, + 564, + 460, + 579 + ], + "score": 1.0, + "content": "Figure 21: Illustration of the averaged response curves of SCIGAN in GNFC.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 114, + 232, + 495, + 560 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 114, + 232, + 495, + 560 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 232, + 495, + 560 + ], + "spans": [ + { + "bbox": [ + 114, + 232, + 495, + 560 + ], + "score": 0.975, + "type": "image", + "image_path": "f4bdbf1e7453b14d1e9a36552e1a5b2ea53c70db26825936362204bac36ee869.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 114, + 232, + 495, + 341.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 114, + 341.3333333333333, + 495, + 450.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 114, + 450.66666666666663, + 495, + 560.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 144, + 571, + 463, + 584 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 144, + 570, + 464, + 586 + ], + "spans": [ + { + "bbox": [ + 144, + 570, + 464, + 586 + ], + "score": 1.0, + "content": "Figure 22: Illustration of the averaged response curves of GALILEO in TCGA.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 44, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "45", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 114, + 232, + 495, + 560 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 114, + 232, + 495, + 560 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 232, + 495, + 560 + ], + "spans": [ + { + "bbox": [ + 114, + 232, + 495, + 560 + ], + "score": 0.975, + "type": "image", + "image_path": "f4bdbf1e7453b14d1e9a36552e1a5b2ea53c70db26825936362204bac36ee869.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 114, + 232, + 495, + 341.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 114, + 341.3333333333333, + 495, + 450.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 114, + 450.66666666666663, + 495, + 560.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 144, + 571, + 463, + 584 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 144, + 570, + 464, + 586 + ], + "spans": [ + { + "bbox": [ + 144, + 570, + 464, + 586 + ], + "score": 1.0, + "content": "Figure 22: Illustration of the averaged response curves of GALILEO in TCGA.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 231, + 495, + 562 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 231, + 495, + 562 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 231, + 495, + 562 + ], + "spans": [ + { + "bbox": [ + 115, + 231, + 495, + 562 + ], + "score": 0.975, + "type": "image", + "image_path": "2290b75ec5c556d9f23c1eec2979a5a76f05ef08e2d626f32314d745cbc6a3c0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 231, + 495, + 341.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 341.3333333333333, + 495, + 451.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 451.66666666666663, + 495, + 562.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 144, + 573, + 463, + 586 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 144, + 572, + 464, + 587 + ], + "spans": [ + { + "bbox": [ + 144, + 572, + 464, + 587 + ], + "score": 1.0, + "content": "Figure 23: Illustration of the averaged response curves of GALILEO in GNFC.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 45, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 27, + 308, + 38 + ], + "lines": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 107, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2023", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "46", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 231, + 495, + 562 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 231, + 495, + 562 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 231, + 495, + 562 + ], + "spans": [ + { + "bbox": [ + 115, + 231, + 495, + 562 + ], + "score": 0.975, + "type": "image", + "image_path": "2290b75ec5c556d9f23c1eec2979a5a76f05ef08e2d626f32314d745cbc6a3c0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 231, + 495, + 341.3333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 341.3333333333333, + 495, + 451.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 451.66666666666663, + 495, + 562.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 144, + 573, + 463, + 586 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 144, + 572, + 464, + 587 + ], + "spans": [ + { + "bbox": [ + 144, + 572, + 464, + 587 + ], + "score": 1.0, + "content": "Figure 23: Illustration of the averaged response curves of GALILEO in GNFC.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/dev/zT5T9gHpGI/zT5T9gHpGI_model.json b/parse/dev/zT5T9gHpGI/zT5T9gHpGI_model.json new file mode 100644 index 0000000000000000000000000000000000000000..e8f2202e2f826e86e34bf4aff8af953fd37d4775 --- /dev/null +++ b/parse/dev/zT5T9gHpGI/zT5T9gHpGI_model.json @@ -0,0 +1,86809 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 535, + 1303, + 535, + 1303, + 1115, + 398, + 1115 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1668, + 1404, + 1668, + 1404, + 2033, + 298, + 2033 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1377, + 1404, + 1377, + 1404, + 1651, + 298, + 1651 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1179, + 1404, + 1179, + 1404, + 1361, + 299, + 1361 + ], + "score": 0.979 + }, + { + "category_id": 0, + "poly": [ + 301, + 220, + 1400, + 220, + 1400, + 323, + 301, + 323 + ], + "score": 0.962 + }, + { + "category_id": 0, + "poly": [ + 301, + 1139, + 574, + 1139, + 574, + 1172, + 301, + 1172 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 313, + 377, + 680, + 377, + 680, + 438, + 313, + 438 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 856, + 75, + 856, + 105, + 298, + 105 + ], + "score": 0.894 + }, + { + "category_id": 0, + "poly": [ + 773, + 464, + 927, + 464, + 927, + 498, + 773, + 498 + ], + "score": 0.871 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 856, + 2088, + 856, + 2112, + 841, + 2112 + ], + "score": 0.727 + }, + { + "category_id": 15, + "poly": [ + 292.0, + 217.0, + 1406.0, + 217.0, + 1406.0, + 274.0, + 292.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 274.0, + 692.0, + 274.0, + 692.0, + 330.0, + 294.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1135.0, + 579.0, + 1135.0, + 579.0, + 1180.0, + 293.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 463.0, + 932.0, + 463.0, + 932.0, + 501.0, + 770.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2088.0, + 857.0, + 2088.0, + 857.0, + 2116.0, + 840.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 536.0, + 1304.0, + 536.0, + 1304.0, + 570.0, + 394.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 568.0, + 1304.0, + 568.0, + 1304.0, + 601.0, + 394.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 594.0, + 1305.0, + 594.0, + 1305.0, + 635.0, + 392.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 626.0, + 1307.0, + 626.0, + 1307.0, + 666.0, + 392.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 657.0, + 1305.0, + 657.0, + 1305.0, + 692.0, + 394.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 689.0, + 1306.0, + 689.0, + 1306.0, + 722.0, + 394.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 719.0, + 1307.0, + 719.0, + 1307.0, + 755.0, + 393.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 748.0, + 1305.0, + 748.0, + 1305.0, + 787.0, + 393.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 780.0, + 1306.0, + 780.0, + 1306.0, + 813.0, + 394.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 810.0, + 1305.0, + 810.0, + 1305.0, + 846.0, + 394.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 841.0, + 1307.0, + 841.0, + 1307.0, + 877.0, + 394.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 870.0, + 1307.0, + 870.0, + 1307.0, + 907.0, + 392.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 903.0, + 1304.0, + 903.0, + 1304.0, + 936.0, + 394.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 932.0, + 1306.0, + 932.0, + 1306.0, + 965.0, + 394.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 963.0, + 1305.0, + 963.0, + 1305.0, + 996.0, + 394.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 993.0, + 1306.0, + 993.0, + 1306.0, + 1026.0, + 394.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1024.0, + 1306.0, + 1024.0, + 1306.0, + 1057.0, + 394.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1054.0, + 1306.0, + 1054.0, + 1306.0, + 1087.0, + 395.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1082.0, + 1237.0, + 1082.0, + 1237.0, + 1121.0, + 394.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1666.0, + 1406.0, + 1666.0, + 1406.0, + 1704.0, + 294.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1698.0, + 1405.0, + 1698.0, + 1405.0, + 1732.0, + 294.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1728.0, + 1405.0, + 1728.0, + 1405.0, + 1762.0, + 293.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1758.0, + 1406.0, + 1758.0, + 1406.0, + 1792.0, + 293.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1789.0, + 1405.0, + 1789.0, + 1405.0, + 1823.0, + 294.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1820.0, + 1405.0, + 1820.0, + 1405.0, + 1854.0, + 294.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1849.0, + 1405.0, + 1849.0, + 1405.0, + 1883.0, + 293.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1881.0, + 1405.0, + 1881.0, + 1405.0, + 1915.0, + 294.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1910.0, + 1404.0, + 1910.0, + 1404.0, + 1947.0, + 295.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1939.0, + 1409.0, + 1939.0, + 1409.0, + 1979.0, + 292.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1970.0, + 1405.0, + 1970.0, + 1405.0, + 2006.0, + 293.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2004.0, + 1160.0, + 2004.0, + 1160.0, + 2035.0, + 296.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1379.0, + 1404.0, + 1379.0, + 1404.0, + 1411.0, + 296.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1406.0, + 1408.0, + 1406.0, + 1408.0, + 1444.0, + 293.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1436.0, + 1405.0, + 1436.0, + 1405.0, + 1474.0, + 293.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1467.0, + 1406.0, + 1467.0, + 1406.0, + 1504.0, + 294.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1498.0, + 1406.0, + 1498.0, + 1406.0, + 1535.0, + 293.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1531.0, + 1402.0, + 1531.0, + 1402.0, + 1563.0, + 296.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1559.0, + 1406.0, + 1559.0, + 1406.0, + 1596.0, + 294.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1589.0, + 1406.0, + 1589.0, + 1406.0, + 1626.0, + 292.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1621.0, + 1409.0, + 1621.0, + 1409.0, + 1654.0, + 293.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1177.0, + 1408.0, + 1177.0, + 1408.0, + 1211.0, + 293.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1207.0, + 1406.0, + 1207.0, + 1406.0, + 1241.0, + 293.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1239.0, + 1406.0, + 1239.0, + 1406.0, + 1274.0, + 295.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1266.0, + 1406.0, + 1266.0, + 1406.0, + 1303.0, + 294.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1299.0, + 1406.0, + 1299.0, + 1406.0, + 1334.0, + 293.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1329.0, + 1336.0, + 1329.0, + 1336.0, + 1364.0, + 295.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 377.0, + 560.0, + 377.0, + 560.0, + 410.0, + 315.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 407.0, + 682.0, + 407.0, + 682.0, + 440.0, + 311.0, + 440.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 1165, + 962, + 1165, + 962, + 1925, + 299, + 1925 + ], + "score": 0.98 + }, + { + "category_id": 3, + "poly": [ + 333, + 117, + 1341, + 117, + 1341, + 445, + 333, + 445 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 301, + 1944, + 1402, + 1944, + 1402, + 2034, + 301, + 2034 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 297, + 873, + 1403, + 873, + 1403, + 997, + 297, + 997 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 297, + 998, + 963, + 998, + 963, + 1150, + 297, + 1150 + ], + "score": 0.927 + }, + { + "category_id": 4, + "poly": [ + 985, + 1335, + 1403, + 1335, + 1403, + 1939, + 985, + 1939 + ], + "score": 0.892 + }, + { + "category_id": 3, + "poly": [ + 988, + 1015, + 1371, + 1015, + 1371, + 1297, + 988, + 1297 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 856, + 75, + 856, + 104, + 297, + 104 + ], + "score": 0.886 + }, + { + "category_id": 4, + "poly": [ + 295, + 456, + 1407, + 456, + 1407, + 856, + 295, + 856 + ], + "score": 0.87 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.697 + }, + { + "category_id": 2, + "poly": [ + 1427, + 967, + 1626, + 967, + 1626, + 996, + 1427, + 996 + ], + "score": 0.57 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.236 + }, + { + "category_id": 13, + "poly": [ + 926, + 731, + 1077, + 731, + 1077, + 765, + 926, + 765 + ], + "score": 0.93, + "latex": "\\Delta a \\in [ - 1 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 488, + 581, + 559, + 581, + 559, + 612, + 488, + 612 + ], + "score": 0.92, + "latex": "y = \\phi" + }, + { + "category_id": 13, + "poly": [ + 498, + 519, + 638, + 519, + 638, + 552, + 498, + 552 + ], + "score": 0.92, + "latex": "( x _ { t + 1 } , y _ { t + 1 } )" + }, + { + "category_id": 13, + "poly": [ + 741, + 732, + 841, + 732, + 841, + 763, + 741, + 763 + ], + "score": 0.91, + "latex": "a _ { t } + \\Delta a" + }, + { + "category_id": 13, + "poly": [ + 599, + 549, + 908, + 549, + 908, + 583, + 599, + 583 + ], + "score": 0.91, + "latex": "a _ { t } \\sim \\mathcal { N } ( ( \\phi - y _ { t } ) / 1 5 , 0 . 0 5 )" + }, + { + "category_id": 13, + "poly": [ + 1013, + 519, + 1264, + 519, + 1264, + 551, + 1013, + 551 + ], + "score": 0.91, + "latex": "y _ { t + 1 } \\sim \\mathcal { N } ( y _ { t } + a _ { t } , 2 )" + }, + { + "category_id": 13, + "poly": [ + 791, + 520, + 961, + 520, + 961, + 551, + 791, + 551 + ], + "score": 0.9, + "latex": "x _ { t + 1 } = x _ { t } + 1" + }, + { + "category_id": 13, + "poly": [ + 1159, + 489, + 1243, + 489, + 1243, + 520, + 1159, + 520 + ], + "score": 0.9, + "latex": "( x _ { t } , y _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 1062, + 674, + 1116, + 674, + 1116, + 704, + 1062, + 704 + ], + "score": 0.87, + "latex": "y _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 1351, + 1456, + 1406, + 1456, + 1406, + 1487, + 1351, + 1487 + ], + "score": 0.86, + "latex": "\\phi =" + }, + { + "category_id": 13, + "poly": [ + 1346, + 523, + 1374, + 523, + 1374, + 549, + 1346, + 549 + ], + "score": 0.85, + "latex": "a _ { t }" + }, + { + "category_id": 13, + "poly": [ + 505, + 736, + 532, + 736, + 532, + 763, + 505, + 763 + ], + "score": 0.84, + "latex": "a _ { t }" + }, + { + "category_id": 13, + "poly": [ + 759, + 583, + 779, + 583, + 779, + 612, + 759, + 612 + ], + "score": 0.84, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 829, + 675, + 857, + 675, + 857, + 702, + 829, + 702 + ], + "score": 0.84, + "latex": "a _ { t }" + }, + { + "category_id": 13, + "poly": [ + 366, + 704, + 394, + 704, + 394, + 732, + 366, + 732 + ], + "score": 0.84, + "latex": "a _ { t }" + }, + { + "category_id": 13, + "poly": [ + 466, + 705, + 493, + 705, + 493, + 733, + 466, + 733 + ], + "score": 0.83, + "latex": "y _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1114, + 552, + 1134, + 552, + 1134, + 581, + 1114, + 581 + ], + "score": 0.83, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 1327, + 706, + 1354, + 706, + 1354, + 734, + 1327, + 734 + ], + "score": 0.82, + "latex": "y _ { t }" + }, + { + "category_id": 13, + "poly": [ + 972, + 645, + 992, + 645, + 992, + 673, + 972, + 673 + ], + "score": 0.78, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1284, + 615, + 1302, + 615, + 1302, + 642, + 1284, + 642 + ], + "score": 0.76, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1329, + 491, + 1343, + 491, + 1343, + 516, + 1329, + 516 + ], + "score": 0.61, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 1195, + 1280, + 1208, + 1280, + 1208, + 1296, + 1195, + 1296 + ], + "score": 0.44, + "latex": "\\boldsymbol { \\Phi }" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 115.0, + 1127.0, + 115.0, + 1127.0, + 377.0, + 1094.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 124.0, + 815.0, + 124.0, + 815.0, + 156.0, + 498.0, + 156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 129.0, + 385.0, + 129.0, + 385.0, + 147.0, + 366.0, + 147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 129.0, + 1144.0, + 129.0, + 1144.0, + 144.0, + 1131.0, + 144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 136.0, + 883.0, + 136.0, + 883.0, + 163.0, + 854.0, + 163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 137.0, + 1262.0, + 137.0, + 1262.0, + 157.0, + 1226.0, + 157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 140.0, + 868.0, + 140.0, + 868.0, + 339.0, + 831.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 154.0, + 1319.0, + 154.0, + 1319.0, + 176.0, + 1227.0, + 176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 156.0, + 885.0, + 156.0, + 885.0, + 219.0, + 853.0, + 219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 156.0, + 1145.0, + 156.0, + 1145.0, + 173.0, + 1131.0, + 173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 169.0, + 473.0, + 169.0, + 473.0, + 200.0, + 353.0, + 200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 176.0, + 668.0, + 176.0, + 668.0, + 191.0, + 653.0, + 191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 170.0, + 717.0, + 170.0, + 717.0, + 187.0, + 699.0, + 187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1203.0, + 172.0, + 1291.0, + 172.0, + 1291.0, + 194.0, + 1203.0, + 194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 184.0, + 1145.0, + 184.0, + 1145.0, + 199.0, + 1131.0, + 199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 196.0, + 661.0, + 196.0, + 661.0, + 242.0, + 498.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 211.0, + 791.0, + 211.0, + 791.0, + 252.0, + 686.0, + 252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 214.0, + 882.0, + 214.0, + 882.0, + 238.0, + 855.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 213.0, + 1143.0, + 213.0, + 1143.0, + 224.0, + 1132.0, + 224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 191.0, + 1253.0, + 191.0, + 1253.0, + 212.0, + 1207.0, + 212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 245.0, + 615.0, + 245.0, + 615.0, + 283.0, + 443.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 247.0, + 779.0, + 247.0, + 779.0, + 279.0, + 683.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 242.0, + 883.0, + 242.0, + 883.0, + 264.0, + 855.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 239.0, + 1145.0, + 239.0, + 1145.0, + 253.0, + 1131.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 248.0, + 1199.0, + 248.0, + 1199.0, + 255.0, + 1186.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 268.0, + 882.0, + 268.0, + 882.0, + 291.0, + 857.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 261.0, + 973.0, + 261.0, + 973.0, + 284.0, + 933.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 267.0, + 1262.0, + 267.0, + 1262.0, + 272.0, + 1256.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 279.0, + 1043.0, + 279.0, + 1043.0, + 307.0, + 932.0, + 307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 294.0, + 882.0, + 294.0, + 882.0, + 318.0, + 855.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 297.0, + 1142.0, + 297.0, + 1142.0, + 307.0, + 1130.0, + 307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 311.0, + 560.0, + 311.0, + 560.0, + 397.0, + 355.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 320.0, + 883.0, + 320.0, + 883.0, + 367.0, + 857.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 302.0, + 1008.0, + 302.0, + 1008.0, + 349.0, + 932.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 322.0, + 1145.0, + 322.0, + 1145.0, + 366.0, + 1127.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 358.0, + 806.0, + 358.0, + 806.0, + 388.0, + 777.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 361.0, + 944.0, + 361.0, + 944.0, + 381.0, + 926.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 361.0, + 1003.0, + 361.0, + 1003.0, + 380.0, + 985.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 361.0, + 1062.0, + 361.0, + 1062.0, + 381.0, + 1045.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 357.0, + 1326.0, + 357.0, + 1326.0, + 384.0, + 1147.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 376.0, + 1008.0, + 376.0, + 1008.0, + 401.0, + 942.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 378.0, + 1346.0, + 378.0, + 1346.0, + 402.0, + 1128.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 413.0, + 723.0, + 413.0, + 723.0, + 446.0, + 426.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 414.0, + 1087.0, + 414.0, + 1087.0, + 446.0, + 821.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 417.0, + 1317.0, + 417.0, + 1317.0, + 446.0, + 1124.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 331.0, + 1063.0, + 331.0, + 1063.0, + 348.0, + 1045.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1333.0, + 1406.0, + 1333.0, + 1406.0, + 1365.0, + 984.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1365.0, + 1407.0, + 1365.0, + 1407.0, + 1396.0, + 983.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 1394.0, + 1404.0, + 1394.0, + 1404.0, + 1425.0, + 982.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1426.0, + 1404.0, + 1426.0, + 1404.0, + 1455.0, + 981.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1454.0, + 1350.0, + 1454.0, + 1350.0, + 1488.0, + 983.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1484.0, + 1404.0, + 1484.0, + 1404.0, + 1517.0, + 983.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 1513.0, + 1406.0, + 1513.0, + 1406.0, + 1551.0, + 982.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1547.0, + 1403.0, + 1547.0, + 1403.0, + 1577.0, + 984.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1579.0, + 1406.0, + 1579.0, + 1406.0, + 1607.0, + 983.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1608.0, + 1404.0, + 1608.0, + 1404.0, + 1639.0, + 984.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1639.0, + 1404.0, + 1639.0, + 1404.0, + 1670.0, + 983.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 1667.0, + 1406.0, + 1667.0, + 1406.0, + 1703.0, + 982.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1699.0, + 1407.0, + 1699.0, + 1407.0, + 1731.0, + 983.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1732.0, + 1407.0, + 1732.0, + 1407.0, + 1760.0, + 984.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1760.0, + 1404.0, + 1760.0, + 1404.0, + 1791.0, + 984.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1791.0, + 1403.0, + 1791.0, + 1403.0, + 1822.0, + 984.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 1819.0, + 1404.0, + 1819.0, + 1404.0, + 1853.0, + 982.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1850.0, + 1408.0, + 1850.0, + 1408.0, + 1885.0, + 983.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1882.0, + 1404.0, + 1882.0, + 1404.0, + 1913.0, + 984.0, + 1913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1913.0, + 1316.0, + 1913.0, + 1316.0, + 1944.0, + 983.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1008.0, + 1011.0, + 1008.0, + 1011.0, + 1276.0, + 986.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 1024.0, + 1252.0, + 1024.0, + 1252.0, + 1044.0, + 1172.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 1031.0, + 1035.0, + 1031.0, + 1035.0, + 1054.0, + 1004.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1037.0, + 1232.0, + 1037.0, + 1232.0, + 1060.0, + 1171.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 1060.0, + 1035.0, + 1060.0, + 1035.0, + 1081.0, + 1004.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1054.0, + 1199.0, + 1054.0, + 1199.0, + 1076.0, + 1169.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 1086.0, + 1035.0, + 1086.0, + 1035.0, + 1109.0, + 1004.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 1114.0, + 1035.0, + 1114.0, + 1035.0, + 1137.0, + 1004.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1142.0, + 1035.0, + 1142.0, + 1035.0, + 1164.0, + 1005.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1170.0, + 1035.0, + 1170.0, + 1035.0, + 1192.0, + 1012.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 1198.0, + 1035.0, + 1198.0, + 1035.0, + 1218.0, + 1011.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1228.0, + 1033.0, + 1228.0, + 1033.0, + 1244.0, + 1012.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1256.0, + 1033.0, + 1256.0, + 1033.0, + 1270.0, + 1020.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 1262.0, + 1354.0, + 1262.0, + 1354.0, + 1281.0, + 1047.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1274.0, + 1194.0, + 1274.0, + 1194.0, + 1299.0, + 994.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 1274.0, + 1377.0, + 1274.0, + 1377.0, + 1299.0, + 1209.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 455.0, + 1407.0, + 455.0, + 1407.0, + 493.0, + 293.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 485.0, + 1158.0, + 485.0, + 1158.0, + 525.0, + 293.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 485.0, + 1328.0, + 485.0, + 1328.0, + 525.0, + 1244.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 485.0, + 1407.0, + 485.0, + 1407.0, + 525.0, + 1344.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 517.0, + 497.0, + 517.0, + 497.0, + 557.0, + 293.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 517.0, + 790.0, + 517.0, + 790.0, + 557.0, + 639.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 517.0, + 1012.0, + 517.0, + 1012.0, + 557.0, + 962.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 517.0, + 1345.0, + 517.0, + 1345.0, + 557.0, + 1265.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 517.0, + 1408.0, + 517.0, + 1408.0, + 557.0, + 1375.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 545.0, + 598.0, + 545.0, + 598.0, + 587.0, + 293.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 545.0, + 1113.0, + 545.0, + 1113.0, + 587.0, + 909.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 545.0, + 1407.0, + 545.0, + 1407.0, + 587.0, + 1135.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 575.0, + 487.0, + 575.0, + 487.0, + 618.0, + 292.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 575.0, + 758.0, + 575.0, + 758.0, + 618.0, + 560.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 575.0, + 1407.0, + 575.0, + 1407.0, + 618.0, + 780.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 609.0, + 1283.0, + 609.0, + 1283.0, + 646.0, + 294.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 609.0, + 1405.0, + 609.0, + 1405.0, + 646.0, + 1303.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 638.0, + 971.0, + 638.0, + 971.0, + 678.0, + 292.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 638.0, + 1408.0, + 638.0, + 1408.0, + 678.0, + 993.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 669.0, + 828.0, + 669.0, + 828.0, + 708.0, + 293.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 669.0, + 1061.0, + 669.0, + 1061.0, + 708.0, + 858.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 669.0, + 1407.0, + 669.0, + 1407.0, + 708.0, + 1117.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 702.0, + 365.0, + 702.0, + 365.0, + 738.0, + 294.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 702.0, + 465.0, + 702.0, + 465.0, + 738.0, + 395.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 702.0, + 1326.0, + 702.0, + 1326.0, + 738.0, + 494.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 702.0, + 1406.0, + 702.0, + 1406.0, + 738.0, + 1355.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 732.0, + 504.0, + 732.0, + 504.0, + 768.0, + 293.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 732.0, + 740.0, + 732.0, + 740.0, + 768.0, + 533.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 732.0, + 925.0, + 732.0, + 925.0, + 768.0, + 842.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 732.0, + 1408.0, + 732.0, + 1408.0, + 768.0, + 1078.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 761.0, + 1406.0, + 761.0, + 1406.0, + 798.0, + 294.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 793.0, + 1406.0, + 793.0, + 1406.0, + 827.0, + 293.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 823.0, + 1124.0, + 823.0, + 1124.0, + 859.0, + 296.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1424.0, + 962.0, + 1629.0, + 962.0, + 1629.0, + 1001.0, + 1424.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1163.0, + 965.0, + 1163.0, + 965.0, + 1199.0, + 294.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1194.0, + 963.0, + 1194.0, + 963.0, + 1228.0, + 296.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1225.0, + 964.0, + 1225.0, + 964.0, + 1259.0, + 296.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1255.0, + 964.0, + 1255.0, + 964.0, + 1289.0, + 294.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1285.0, + 964.0, + 1285.0, + 964.0, + 1321.0, + 295.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1316.0, + 965.0, + 1316.0, + 965.0, + 1350.0, + 296.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1348.0, + 966.0, + 1348.0, + 966.0, + 1382.0, + 296.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1378.0, + 963.0, + 1378.0, + 963.0, + 1409.0, + 297.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1406.0, + 965.0, + 1406.0, + 965.0, + 1443.0, + 295.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1439.0, + 964.0, + 1439.0, + 964.0, + 1470.0, + 296.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1469.0, + 964.0, + 1469.0, + 964.0, + 1500.0, + 296.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1501.0, + 964.0, + 1501.0, + 964.0, + 1530.0, + 295.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1531.0, + 965.0, + 1531.0, + 965.0, + 1559.0, + 295.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1562.0, + 964.0, + 1562.0, + 964.0, + 1593.0, + 296.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1592.0, + 967.0, + 1592.0, + 967.0, + 1622.0, + 294.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1622.0, + 963.0, + 1622.0, + 963.0, + 1653.0, + 297.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1653.0, + 965.0, + 1653.0, + 965.0, + 1684.0, + 295.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1682.0, + 964.0, + 1682.0, + 964.0, + 1713.0, + 295.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1714.0, + 964.0, + 1714.0, + 964.0, + 1745.0, + 295.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1743.0, + 964.0, + 1743.0, + 964.0, + 1774.0, + 296.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1773.0, + 966.0, + 1773.0, + 966.0, + 1806.0, + 294.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1805.0, + 963.0, + 1805.0, + 963.0, + 1836.0, + 296.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1830.0, + 965.0, + 1830.0, + 965.0, + 1872.0, + 294.0, + 1872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1861.0, + 964.0, + 1861.0, + 964.0, + 1902.0, + 294.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1894.0, + 626.0, + 1894.0, + 626.0, + 1926.0, + 297.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1943.0, + 1403.0, + 1943.0, + 1403.0, + 1976.0, + 296.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2003.0, + 1404.0, + 2003.0, + 1404.0, + 2035.0, + 296.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 871.0, + 1406.0, + 871.0, + 1406.0, + 906.0, + 291.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 903.0, + 1407.0, + 903.0, + 1407.0, + 936.0, + 296.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 931.0, + 1405.0, + 931.0, + 1405.0, + 968.0, + 296.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 962.0, + 1394.0, + 962.0, + 1394.0, + 1002.0, + 294.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 994.0, + 965.0, + 994.0, + 965.0, + 1029.0, + 295.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1026.0, + 964.0, + 1026.0, + 964.0, + 1058.0, + 295.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1056.0, + 964.0, + 1056.0, + 964.0, + 1090.0, + 295.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1086.0, + 962.0, + 1086.0, + 962.0, + 1116.0, + 296.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1116.0, + 965.0, + 1116.0, + 965.0, + 1151.0, + 297.0, + 1151.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 406, + 1405, + 406, + 1405, + 955, + 298, + 955 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 1122, + 1405, + 1122, + 1405, + 1806, + 297, + 1806 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1820, + 1404, + 1820, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 300, + 229, + 1402, + 229, + 1402, + 322, + 300, + 322 + ], + "score": 0.962 + }, + { + "category_id": 0, + "poly": [ + 300, + 343, + 588, + 343, + 588, + 379, + 300, + 379 + ], + "score": 0.904 + }, + { + "category_id": 0, + "poly": [ + 305, + 1045, + 1356, + 1045, + 1356, + 1107, + 305, + 1107 + ], + "score": 0.886 + }, + { + "category_id": 0, + "poly": [ + 301, + 991, + 578, + 991, + 578, + 1026, + 301, + 1026 + ], + "score": 0.884 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 856, + 76, + 856, + 104, + 298, + 104 + ], + "score": 0.883 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2112, + 841, + 2112 + ], + "score": 0.594 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 859, + 2088, + 859, + 2112, + 842, + 2112 + ], + "score": 0.41 + }, + { + "category_id": 13, + "poly": [ + 621, + 1307, + 692, + 1307, + 692, + 1340, + 621, + 1340 + ], + "score": 0.93, + "latex": "\\{ 0 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 896, + 1911, + 1242, + 1911, + 1242, + 1945, + 896, + 1945 + ], + "score": 0.92, + "latex": "\\mu ( a | x ) > 0 , \\forall a \\stackrel { - } { \\in } \\mathcal { \\bar { A } } , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 534, + 1739, + 984, + 1739, + 984, + 1776, + 534, + 1776 + ], + "score": 0.92, + "latex": "\\hat { I T E } ( x ) = \\mathbb { E } [ M ( y | x , 1 ) ] - \\mathbb { E } [ M ( y | x , 0 ) ]" + }, + { + "category_id": 13, + "poly": [ + 1004, + 1648, + 1278, + 1648, + 1278, + 1681, + 1004, + 1681 + ], + "score": 0.92, + "latex": "P _ { C F } = \\{ ( x _ { i } , 1 - a _ { i } ) \\} _ { i } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1141, + 1617, + 1350, + 1617, + 1350, + 1650, + 1141, + 1650 + ], + "score": 0.92, + "latex": "\\bar { P _ { F } } = \\{ ( x _ { i } , a _ { i } ) \\} _ { i } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 900, + 1368, + 989, + 1368, + 989, + 1396, + 900, + 1396 + ], + "score": 0.91, + "latex": "A = 1" + }, + { + "category_id": 13, + "poly": [ + 440, + 1588, + 543, + 1588, + 543, + 1617, + 440, + 1617 + ], + "score": 0.91, + "latex": "M \\in \\mathcal { M }" + }, + { + "category_id": 13, + "poly": [ + 1046, + 1368, + 1136, + 1368, + 1136, + 1396, + 1046, + 1396 + ], + "score": 0.9, + "latex": "A = 0" + }, + { + "category_id": 13, + "poly": [ + 565, + 1185, + 1398, + 1185, + 1398, + 1218, + 565, + 1218 + ], + "score": 0.9, + "latex": "I T E ( x ) : = \\mathbb { E } [ M ^ { * } ( y | x , 1 ) | A = 1 , X = x ] - \\mathbb { E } [ M ^ { * } ( y | x , 0 ) | A = 0 , X = x ]" + }, + { + "category_id": 13, + "poly": [ + 1259, + 1681, + 1315, + 1681, + 1315, + 1710, + 1259, + 1710 + ], + "score": 0.9, + "latex": "P _ { C F }" + }, + { + "category_id": 13, + "poly": [ + 767, + 1556, + 803, + 1556, + 803, + 1590, + 767, + 1590 + ], + "score": 0.9, + "latex": "y _ { i } ^ { F }" + }, + { + "category_id": 13, + "poly": [ + 815, + 1217, + 945, + 1217, + 945, + 1248, + 815, + 1248 + ], + "score": 0.89, + "latex": "M ^ { * } ( y | x , a )" + }, + { + "category_id": 13, + "poly": [ + 1040, + 1399, + 1101, + 1399, + 1101, + 1430, + 1040, + 1430 + ], + "score": 0.89, + "latex": "X , A" + }, + { + "category_id": 13, + "poly": [ + 376, + 1521, + 1169, + 1521, + 1169, + 1558, + 376, + 1558 + ], + "score": 0.88, + "latex": "\\hat { I T E } ( x _ { i } ) = a _ { i } ( y _ { i } ^ { F } - M ( x _ { i } , 1 - a _ { i } ) ) + ( 1 - a _ { i } ) ( M ( x _ { i } , 1 - a _ { i } ) - y _ { i } ^ { F } )" + }, + { + "category_id": 13, + "poly": [ + 1289, + 1399, + 1352, + 1399, + 1352, + 1430, + 1289, + 1430 + ], + "score": 0.88, + "latex": "x , A" + }, + { + "category_id": 13, + "poly": [ + 464, + 1403, + 511, + 1403, + 511, + 1430, + 464, + 1430 + ], + "score": 0.87, + "latex": "x , a" + }, + { + "category_id": 13, + "poly": [ + 1167, + 1681, + 1206, + 1681, + 1206, + 1709, + 1167, + 1709 + ], + "score": 0.86, + "latex": "P _ { F }" + }, + { + "category_id": 13, + "poly": [ + 686, + 1562, + 716, + 1562, + 716, + 1587, + 686, + 1587 + ], + "score": 0.85, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 565, + 1404, + 584, + 1404, + 584, + 1430, + 565, + 1430 + ], + "score": 0.81, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 650, + 1278, + 677, + 1278, + 677, + 1305, + 650, + 1305 + ], + "score": 0.81, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 1035, + 1588, + 1071, + 1588, + 1071, + 1616, + 1035, + 1616 + ], + "score": 0.8, + "latex": "\\mathcal { M }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1429, + 323, + 1429, + 323, + 1459, + 297, + 1459 + ], + "score": 0.8, + "latex": "\\mathcal { V }" + }, + { + "category_id": 13, + "poly": [ + 374, + 1221, + 394, + 1221, + 394, + 1248, + 374, + 1248 + ], + "score": 0.79, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1247, + 1369, + 1276, + 1369, + 1276, + 1396, + 1247, + 1396 + ], + "score": 0.77, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 774, + 1917, + 793, + 1917, + 793, + 1944, + 774, + 1944 + ], + "score": 0.76, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 742, + 1278, + 766, + 1278, + 766, + 1304, + 742, + 1304 + ], + "score": 0.76, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 1155, + 1399, + 1181, + 1399, + 1181, + 1427, + 1155, + 1427 + ], + "score": 0.74, + "latex": "Y" + }, + { + "category_id": 13, + "poly": [ + 528, + 1308, + 552, + 1308, + 552, + 1334, + 528, + 1334 + ], + "score": 0.7, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 1356, + 1559, + 1369, + 1559, + 1369, + 1584, + 1356, + 1584 + ], + "score": 0.69, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 1151, + 1978, + 1173, + 1978, + 1173, + 2000, + 1151, + 2000 + ], + "score": 0.69, + "latex": "\\omega" + }, + { + "category_id": 13, + "poly": [ + 1323, + 1654, + 1341, + 1654, + 1341, + 1676, + 1323, + 1676 + ], + "score": 0.69, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 961, + 1218, + 990, + 1218, + 990, + 1244, + 961, + 1244 + ], + "score": 0.68, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 879, + 1249, + 904, + 1249, + 904, + 1275, + 879, + 1275 + ], + "score": 0.66, + "latex": "A" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 339.0, + 594.0, + 339.0, + 594.0, + 385.0, + 290.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1043.0, + 1357.0, + 1043.0, + 1357.0, + 1080.0, + 298.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1075.0, + 771.0, + 1075.0, + 771.0, + 1109.0, + 362.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 987.0, + 583.0, + 987.0, + 583.0, + 1033.0, + 291.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 405.0, + 1406.0, + 405.0, + 1406.0, + 441.0, + 295.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 435.0, + 1407.0, + 435.0, + 1407.0, + 469.0, + 293.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 466.0, + 1405.0, + 466.0, + 1405.0, + 500.0, + 293.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 498.0, + 1403.0, + 498.0, + 1403.0, + 530.0, + 296.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 527.0, + 1407.0, + 527.0, + 1407.0, + 563.0, + 295.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 556.0, + 1408.0, + 556.0, + 1408.0, + 594.0, + 292.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 588.0, + 1408.0, + 588.0, + 1408.0, + 623.0, + 295.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 618.0, + 1405.0, + 618.0, + 1405.0, + 654.0, + 293.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 649.0, + 1405.0, + 649.0, + 1405.0, + 684.0, + 295.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 678.0, + 1405.0, + 678.0, + 1405.0, + 713.0, + 295.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 710.0, + 1405.0, + 710.0, + 1405.0, + 745.0, + 295.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 738.0, + 1408.0, + 738.0, + 1408.0, + 778.0, + 291.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 770.0, + 1406.0, + 770.0, + 1406.0, + 806.0, + 296.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 802.0, + 1405.0, + 802.0, + 1405.0, + 835.0, + 292.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 831.0, + 1405.0, + 831.0, + 1405.0, + 867.0, + 293.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 858.0, + 1406.0, + 858.0, + 1406.0, + 900.0, + 291.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 892.0, + 1406.0, + 892.0, + 1406.0, + 926.0, + 293.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 923.0, + 1279.0, + 923.0, + 1279.0, + 957.0, + 293.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1121.0, + 1407.0, + 1121.0, + 1407.0, + 1158.0, + 294.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1152.0, + 1406.0, + 1152.0, + 1406.0, + 1188.0, + 294.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1183.0, + 564.0, + 1183.0, + 564.0, + 1223.0, + 292.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1183.0, + 1410.0, + 1183.0, + 1410.0, + 1223.0, + 1399.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1213.0, + 373.0, + 1213.0, + 373.0, + 1253.0, + 292.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1213.0, + 814.0, + 1213.0, + 814.0, + 1253.0, + 395.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 1213.0, + 960.0, + 1213.0, + 960.0, + 1253.0, + 946.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 1213.0, + 1406.0, + 1213.0, + 1406.0, + 1253.0, + 991.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1246.0, + 878.0, + 1246.0, + 878.0, + 1281.0, + 292.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 1246.0, + 1407.0, + 1246.0, + 1407.0, + 1281.0, + 905.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1276.0, + 649.0, + 1276.0, + 649.0, + 1312.0, + 294.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 1276.0, + 741.0, + 1276.0, + 741.0, + 1312.0, + 678.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1276.0, + 1408.0, + 1276.0, + 1408.0, + 1312.0, + 767.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1305.0, + 527.0, + 1305.0, + 527.0, + 1342.0, + 294.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1305.0, + 620.0, + 1305.0, + 620.0, + 1342.0, + 553.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1305.0, + 1406.0, + 1305.0, + 1406.0, + 1342.0, + 693.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1335.0, + 1406.0, + 1335.0, + 1406.0, + 1372.0, + 292.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1366.0, + 899.0, + 1366.0, + 899.0, + 1402.0, + 294.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1366.0, + 1045.0, + 1366.0, + 1045.0, + 1402.0, + 990.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1366.0, + 1246.0, + 1366.0, + 1246.0, + 1402.0, + 1137.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 1366.0, + 1407.0, + 1366.0, + 1407.0, + 1402.0, + 1277.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1398.0, + 463.0, + 1398.0, + 463.0, + 1432.0, + 292.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1398.0, + 564.0, + 1398.0, + 564.0, + 1432.0, + 512.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1398.0, + 1039.0, + 1398.0, + 1039.0, + 1432.0, + 585.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1398.0, + 1154.0, + 1398.0, + 1154.0, + 1432.0, + 1102.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 1398.0, + 1288.0, + 1398.0, + 1288.0, + 1432.0, + 1182.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 1398.0, + 1405.0, + 1398.0, + 1405.0, + 1432.0, + 1353.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1429.0, + 1405.0, + 1429.0, + 1405.0, + 1461.0, + 324.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1459.0, + 1406.0, + 1459.0, + 1406.0, + 1495.0, + 294.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1489.0, + 1405.0, + 1489.0, + 1405.0, + 1525.0, + 294.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1518.0, + 375.0, + 1518.0, + 375.0, + 1562.0, + 291.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 1518.0, + 1407.0, + 1518.0, + 1407.0, + 1562.0, + 1170.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1553.0, + 685.0, + 1553.0, + 685.0, + 1593.0, + 292.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1553.0, + 766.0, + 1553.0, + 766.0, + 1593.0, + 717.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1553.0, + 1355.0, + 1553.0, + 1355.0, + 1593.0, + 804.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 1553.0, + 1407.0, + 1553.0, + 1407.0, + 1593.0, + 1370.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1586.0, + 439.0, + 1586.0, + 439.0, + 1623.0, + 295.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 1586.0, + 1034.0, + 1586.0, + 1034.0, + 1623.0, + 544.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1586.0, + 1405.0, + 1586.0, + 1405.0, + 1623.0, + 1072.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1611.0, + 1140.0, + 1611.0, + 1140.0, + 1657.0, + 291.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1611.0, + 1408.0, + 1611.0, + 1408.0, + 1657.0, + 1351.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1643.0, + 1003.0, + 1643.0, + 1003.0, + 1687.0, + 292.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1643.0, + 1322.0, + 1643.0, + 1322.0, + 1687.0, + 1279.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1643.0, + 1408.0, + 1643.0, + 1408.0, + 1687.0, + 1342.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1677.0, + 1166.0, + 1677.0, + 1166.0, + 1713.0, + 292.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 1677.0, + 1258.0, + 1677.0, + 1258.0, + 1713.0, + 1207.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 1677.0, + 1406.0, + 1677.0, + 1406.0, + 1713.0, + 1316.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1708.0, + 1405.0, + 1708.0, + 1405.0, + 1744.0, + 295.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1740.0, + 533.0, + 1740.0, + 533.0, + 1780.0, + 294.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 1740.0, + 1408.0, + 1740.0, + 1408.0, + 1780.0, + 985.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1771.0, + 1023.0, + 1771.0, + 1023.0, + 1811.0, + 294.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1819.0, + 1405.0, + 1819.0, + 1405.0, + 1853.0, + 296.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1847.0, + 1408.0, + 1847.0, + 1408.0, + 1888.0, + 293.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1405.0, + 1882.0, + 1405.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 773.0, + 1912.0, + 773.0, + 1944.0, + 293.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1912.0, + 895.0, + 1912.0, + 895.0, + 1944.0, + 794.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 1912.0, + 1406.0, + 1912.0, + 1406.0, + 1944.0, + 1243.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 1406.0, + 1942.0, + 1406.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1150.0, + 1972.0, + 1150.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 1972.0, + 1408.0, + 1972.0, + 1408.0, + 2007.0, + 1174.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1999.0, + 1061.0, + 1999.0, + 1061.0, + 2036.0, + 295.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 228.0, + 1406.0, + 228.0, + 1406.0, + 266.0, + 294.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 259.0, + 1406.0, + 259.0, + 1406.0, + 298.0, + 293.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 294.0, + 669.0, + 294.0, + 669.0, + 324.0, + 296.0, + 324.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 531, + 1405, + 531, + 1405, + 814, + 296, + 814 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 1309, + 1406, + 1309, + 1406, + 1586, + 297, + 1586 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 868, + 1405, + 868, + 1405, + 1083, + 297, + 1083 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1148, + 1405, + 1148, + 1405, + 1241, + 299, + 1241 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1939, + 1404, + 1939, + 1404, + 2039, + 297, + 2039 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 299, + 415, + 1402, + 415, + 1402, + 509, + 299, + 509 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 299, + 1594, + 1404, + 1594, + 1404, + 1688, + 299, + 1688 + ], + "score": 0.968 + }, + { + "category_id": 8, + "poly": [ + 517, + 1698, + 1182, + 1698, + 1182, + 1837, + 517, + 1837 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 290, + 1852, + 1404, + 1852, + 1404, + 1920, + 290, + 1920 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 517, + 272, + 1182, + 272, + 1182, + 402, + 517, + 402 + ], + "score": 0.944 + }, + { + "category_id": 0, + "poly": [ + 298, + 826, + 870, + 826, + 870, + 857, + 298, + 857 + ], + "score": 0.929 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 106, + 297, + 106 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 298, + 1102, + 482, + 1102, + 482, + 1137, + 298, + 1137 + ], + "score": 0.905 + }, + { + "category_id": 0, + "poly": [ + 299, + 1262, + 669, + 1262, + 669, + 1293, + 299, + 1293 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1367, + 324, + 1402, + 324, + 1402, + 354, + 1367, + 354 + ], + "score": 0.88 + }, + { + "category_id": 1, + "poly": [ + 292, + 228, + 1173, + 228, + 1173, + 263, + 292, + 263 + ], + "score": 0.855 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.794 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1753, + 1402, + 1753, + 1402, + 1782, + 1367, + 1782 + ], + "score": 0.792 + }, + { + "category_id": 9, + "poly": [ + 1368, + 1753, + 1402, + 1753, + 1402, + 1782, + 1368, + 1782 + ], + "score": 0.114 + }, + { + "category_id": 13, + "poly": [ + 1075, + 1021, + 1264, + 1021, + 1264, + 1054, + 1075, + 1054 + ], + "score": 0.93, + "latex": "M ^ { \\ast } ( x _ { t + 1 } | x _ { t } , a _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 496, + 959, + 672, + 959, + 672, + 992, + 496, + 992 + ], + "score": 0.92, + "latex": "t \\in \\{ 0 , 1 , 2 , \\ldots \\}" + }, + { + "category_id": 14, + "poly": [ + 516, + 272, + 1178, + 272, + 1178, + 406, + 516, + 406 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } { \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } L ( M ) = \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } \\mathbb { E } _ { { x } , { a } , { y } \\sim { p } _ { M ^ { * } } } [ \\omega ( { x } , { a } ) \\ell ( M ( { y } | { x } , { a } ) , { y } ) ] , } & { } \\\\ { { s . t . } \\quad \\omega ( { x } , { a } ) = \\frac { \\beta ( a | { x } ) } { \\mu ( a | { x } ) } , } & { } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 293, + 2001, + 805, + 2001, + 805, + 2039, + 293, + 2039 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { ( 1 - \\gamma ) \\mathbb { E } _ { x _ { 0 } \\sim \\rho _ { 0 } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\dot { \\gamma } ^ { t } \\mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 859, + 447, + 926, + 447, + 926, + 480, + 859, + 480 + ], + "score": 0.92, + "latex": "\\rho _ { 0 } ( x )" + }, + { + "category_id": 13, + "poly": [ + 904, + 655, + 1040, + 655, + 1040, + 689, + 904, + 689 + ], + "score": 0.92, + "latex": "\\hat { \\omega } ( x , a ) = 1" + }, + { + "category_id": 13, + "poly": [ + 967, + 960, + 1052, + 960, + 1052, + 990, + 967, + 990 + ], + "score": 0.92, + "latex": "a _ { t } \\in { \\mathcal { A } }" + }, + { + "category_id": 13, + "poly": [ + 1094, + 415, + 1145, + 415, + 1145, + 450, + 1094, + 450 + ], + "score": 0.92, + "latex": "p _ { M ^ { * } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 1040, + 1595, + 1407, + 1595, + 1407, + 1628, + 1040, + 1628 + ], + "score": 0.92, + "latex": "M ^ { * } ( x ^ { \\prime } | x , a ) > 0 , \\forall x \\in \\mathcal { X } , \\forall a \\in" + }, + { + "category_id": 14, + "poly": [ + 514, + 1697, + 1182, + 1697, + 1182, + 1839, + 514, + 1839 + ], + "score": 0.91, + "latex": "\\begin{array} { r l } & { \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } L ( M ) = \\underset { M \\in \\mathcal { M } } { \\operatorname* { m i n } } \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\mu } } [ \\omega ( x , a , x ^ { \\prime } ) \\ell _ { M } ( x , a , x ^ { \\prime } ) ] , } \\\\ & { ~ s . t . ~ \\omega ( x , a , x ^ { \\prime } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } ) } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1197, + 929, + 1329, + 929, + 1329, + 961, + 1197, + 961 + ], + "score": 0.91, + "latex": "x _ { 0 } \\sim \\rho _ { 0 } ( x )" + }, + { + "category_id": 13, + "poly": [ + 506, + 994, + 593, + 994, + 593, + 1021, + 506, + 1021 + ], + "score": 0.91, + "latex": "x _ { t } \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 618, + 1626, + 964, + 1626, + 964, + 1659, + 618, + 1659 + ], + "score": 0.91, + "latex": "\\mu ( a | x ) > 0 , \\forall a \\in \\mathcal { A } , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 1248, + 960, + 1401, + 960, + 1401, + 992, + 1248, + 992 + ], + "score": 0.91, + "latex": "\\pi ( a _ { t } | x _ { t } ) \\in \\Pi" + }, + { + "category_id": 13, + "poly": [ + 1242, + 683, + 1320, + 683, + 1320, + 725, + 1242, + 725 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\omega = \\frac { \\mathrm { i } } { \\hat { \\mu } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 446, + 755, + 446, + 755, + 480, + 297, + 480 + ], + "score": 0.9, + "latex": "p _ { M ^ { * } } ^ { \\mu } ( x , a , y ) : = \\rho _ { 0 } ( x ) \\mu ( a | x ) M ^ { * } ( y | x , a )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1626, + 438, + 1626, + 438, + 1656, + 298, + 1656 + ], + "score": 0.9, + "latex": "\\mathcal { A } , \\forall x ^ { \\prime } \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 700, + 1941, + 1220, + 1941, + 1220, + 1976, + 700, + 1976 + ], + "score": 0.9, + "latex": "\\rho _ { M ^ { * } } ^ { \\pi } ( x , a , x ^ { \\prime } ) ~ = ~ \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a )" + }, + { + "category_id": 13, + "poly": [ + 1274, + 992, + 1403, + 992, + 1403, + 1023, + 1274, + 1023 + ], + "score": 0.89, + "latex": "r ( x _ { t } , a _ { t } ) \\in" + }, + { + "category_id": 13, + "poly": [ + 369, + 1854, + 420, + 1854, + 420, + 1889, + 369, + 1889 + ], + "score": 0.89, + "latex": "\\rho _ { M ^ { * } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 661, + 1025, + 718, + 1025, + 718, + 1053, + 661, + 1053 + ], + "score": 0.89, + "latex": "x _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 674, + 1886, + 1053, + 1886, + 1053, + 1919, + 674, + 1919 + ], + "score": 0.88, + "latex": "\\ell _ { M } ( x , a , x ^ { \\prime } ) : = \\ell ( M ( x ^ { \\prime } | x , a ) , x ^ { \\prime } ) ^ { - }" + }, + { + "category_id": 13, + "poly": [ + 1312, + 1942, + 1403, + 1942, + 1403, + 1975, + 1312, + 1975 + ], + "score": 0.88, + "latex": "\\rho _ { M ^ { \\ast } } ^ { \\pi } ( x )" + }, + { + "category_id": 13, + "poly": [ + 1020, + 1848, + 1072, + 1848, + 1072, + 1888, + 1020, + 1888 + ], + "score": 0.88, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 851, + 1597, + 896, + 1597, + 896, + 1623, + 851, + 1623 + ], + "score": 0.87, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1256, + 1975, + 1405, + 1975, + 1405, + 2007, + 1256, + 2007 + ], + "score": 0.87, + "latex": "\\rho _ { M ^ { \\ast } } ^ { \\pi } ( x ) : =" + }, + { + "category_id": 13, + "poly": [ + 1030, + 722, + 1050, + 722, + 1050, + 752, + 1030, + 752 + ], + "score": 0.85, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 894, + 1435, + 915, + 1435, + 915, + 1465, + 894, + 1465 + ], + "score": 0.85, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 975, + 995, + 1001, + 995, + 1001, + 1021, + 975, + 1021 + ], + "score": 0.84, + "latex": "r _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1380, + 686, + 1401, + 686, + 1401, + 718, + 1380, + 718 + ], + "score": 0.82, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 733, + 1980, + 752, + 1980, + 752, + 2000, + 733, + 2000 + ], + "score": 0.81, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 447, + 1435, + 466, + 1435, + 466, + 1465, + 447, + 1465 + ], + "score": 0.8, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 494, + 1629, + 515, + 1629, + 515, + 1658, + 494, + 1658 + ], + "score": 0.8, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 372, + 417, + 392, + 417, + 392, + 446, + 372, + 446 + ], + "score": 0.79, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 297, + 1021, + 321, + 1021, + 321, + 1048, + 297, + 1048 + ], + "score": 0.79, + "latex": "\\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 1379, + 565, + 1401, + 565, + 1401, + 591, + 1379, + 591 + ], + "score": 0.79, + "latex": "\\hat { \\omega }" + }, + { + "category_id": 13, + "poly": [ + 346, + 538, + 368, + 538, + 368, + 562, + 346, + 562 + ], + "score": 0.79, + "latex": "\\omega" + }, + { + "category_id": 13, + "poly": [ + 571, + 1052, + 597, + 1052, + 597, + 1078, + 571, + 1078 + ], + "score": 0.78, + "latex": "\\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 1339, + 1373, + 1372, + 1373, + 1372, + 1401, + 1339, + 1401 + ], + "score": 0.77, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 897, + 1981, + 917, + 1981, + 917, + 2001, + 897, + 2001 + ], + "score": 0.77, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 715, + 596, + 748, + 596, + 748, + 623, + 715, + 623 + ], + "score": 0.77, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1366, + 1343, + 1398, + 1343, + 1398, + 1370, + 1366, + 1370 + ], + "score": 0.76, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 519, + 1887, + 540, + 1887, + 540, + 1917, + 519, + 1917 + ], + "score": 0.76, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 475, + 600, + 498, + 600, + 498, + 622, + 475, + 622 + ], + "score": 0.75, + "latex": "\\omega" + }, + { + "category_id": 13, + "poly": [ + 415, + 569, + 438, + 569, + 438, + 591, + 415, + 591 + ], + "score": 0.74, + "latex": "\\omega" + }, + { + "category_id": 13, + "poly": [ + 1226, + 448, + 1262, + 448, + 1262, + 474, + 1226, + 474 + ], + "score": 0.74, + "latex": "\\mathcal { M }" + }, + { + "category_id": 13, + "poly": [ + 375, + 756, + 395, + 756, + 395, + 783, + 375, + 783 + ], + "score": 0.73, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 444, + 421, + 464, + 421, + 464, + 447, + 444, + 447 + ], + "score": 0.7, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 486, + 1052, + 513, + 1052, + 513, + 1078, + 486, + 1078 + ], + "score": 0.65, + "latex": "\\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 664, + 1948, + 684, + 1948, + 684, + 1971, + 664, + 1971 + ], + "score": 0.56, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 375, + 481, + 391, + 481, + 391, + 504, + 375, + 504 + ], + "score": 0.53, + "latex": "\\ell" + }, + { + "category_id": 13, + "poly": [ + 904, + 2006, + 929, + 2006, + 929, + 2031, + 904, + 2031 + ], + "score": 0.49, + "latex": "\\&" + }, + { + "category_id": 13, + "poly": [ + 983, + 1861, + 1001, + 1861, + 1001, + 1886, + 983, + 1886 + ], + "score": 0.33, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 452, + 1051, + 476, + 1051, + 476, + 1078, + 452, + 1078 + ], + "score": 0.26, + "latex": "\\Pi" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 825.0, + 872.0, + 825.0, + 872.0, + 861.0, + 295.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1098.0, + 485.0, + 1098.0, + 485.0, + 1144.0, + 293.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1259.0, + 670.0, + 1259.0, + 670.0, + 1297.0, + 294.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 837.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 531.0, + 345.0, + 531.0, + 345.0, + 567.0, + 294.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 531.0, + 1403.0, + 531.0, + 1403.0, + 567.0, + 369.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 564.0, + 414.0, + 564.0, + 414.0, + 597.0, + 296.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 564.0, + 1378.0, + 564.0, + 1378.0, + 597.0, + 439.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 564.0, + 1405.0, + 564.0, + 1405.0, + 597.0, + 1402.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 595.0, + 474.0, + 595.0, + 474.0, + 628.0, + 294.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 595.0, + 714.0, + 595.0, + 714.0, + 628.0, + 499.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 595.0, + 1405.0, + 595.0, + 1405.0, + 628.0, + 749.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 625.0, + 1405.0, + 625.0, + 1405.0, + 658.0, + 294.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 656.0, + 903.0, + 656.0, + 903.0, + 689.0, + 295.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 656.0, + 1405.0, + 656.0, + 1405.0, + 689.0, + 1041.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 681.0, + 1241.0, + 681.0, + 1241.0, + 724.0, + 290.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 681.0, + 1379.0, + 681.0, + 1379.0, + 724.0, + 1321.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 681.0, + 1407.0, + 681.0, + 1407.0, + 724.0, + 1402.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 719.0, + 1029.0, + 719.0, + 1029.0, + 756.0, + 293.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 719.0, + 1406.0, + 719.0, + 1406.0, + 756.0, + 1051.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 753.0, + 374.0, + 753.0, + 374.0, + 784.0, + 294.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 753.0, + 1405.0, + 753.0, + 1405.0, + 784.0, + 396.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 779.0, + 1223.0, + 779.0, + 1223.0, + 816.0, + 294.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1310.0, + 1406.0, + 1310.0, + 1406.0, + 1347.0, + 295.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1341.0, + 1365.0, + 1341.0, + 1365.0, + 1377.0, + 292.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1341.0, + 1409.0, + 1341.0, + 1409.0, + 1377.0, + 1399.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1370.0, + 1338.0, + 1370.0, + 1338.0, + 1408.0, + 294.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 1370.0, + 1406.0, + 1370.0, + 1406.0, + 1408.0, + 1373.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1403.0, + 1404.0, + 1403.0, + 1404.0, + 1436.0, + 295.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1433.0, + 446.0, + 1433.0, + 446.0, + 1467.0, + 292.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1433.0, + 893.0, + 1433.0, + 893.0, + 1467.0, + 467.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1433.0, + 1404.0, + 1433.0, + 1404.0, + 1467.0, + 916.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1462.0, + 1406.0, + 1462.0, + 1406.0, + 1500.0, + 292.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1493.0, + 1406.0, + 1493.0, + 1406.0, + 1530.0, + 295.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1522.0, + 1407.0, + 1522.0, + 1407.0, + 1560.0, + 292.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1552.0, + 734.0, + 1552.0, + 734.0, + 1590.0, + 295.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 866.0, + 1406.0, + 866.0, + 1406.0, + 903.0, + 292.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 897.0, + 1406.0, + 897.0, + 1406.0, + 932.0, + 294.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 929.0, + 1196.0, + 929.0, + 1196.0, + 963.0, + 295.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 929.0, + 1405.0, + 929.0, + 1405.0, + 963.0, + 1330.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 958.0, + 495.0, + 958.0, + 495.0, + 994.0, + 292.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 958.0, + 966.0, + 958.0, + 966.0, + 994.0, + 673.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 958.0, + 1247.0, + 958.0, + 1247.0, + 994.0, + 1053.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 958.0, + 1406.0, + 958.0, + 1406.0, + 994.0, + 1402.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 990.0, + 505.0, + 990.0, + 505.0, + 1025.0, + 295.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 990.0, + 974.0, + 990.0, + 974.0, + 1025.0, + 594.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 990.0, + 1273.0, + 990.0, + 1273.0, + 1025.0, + 1002.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1017.0, + 660.0, + 1017.0, + 660.0, + 1057.0, + 322.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1017.0, + 1074.0, + 1017.0, + 1074.0, + 1057.0, + 719.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 1017.0, + 1405.0, + 1017.0, + 1405.0, + 1057.0, + 1265.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1051.0, + 451.0, + 1051.0, + 451.0, + 1085.0, + 295.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 1051.0, + 485.0, + 1051.0, + 485.0, + 1085.0, + 477.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1051.0, + 570.0, + 1051.0, + 570.0, + 1085.0, + 514.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1051.0, + 1074.0, + 1051.0, + 1074.0, + 1085.0, + 598.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1143.0, + 1405.0, + 1143.0, + 1405.0, + 1185.0, + 292.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1175.0, + 1409.0, + 1175.0, + 1409.0, + 1215.0, + 293.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1208.0, + 915.0, + 1208.0, + 915.0, + 1245.0, + 294.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1938.0, + 663.0, + 1938.0, + 663.0, + 1980.0, + 292.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1938.0, + 699.0, + 1938.0, + 699.0, + 1980.0, + 685.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 1938.0, + 1311.0, + 1938.0, + 1311.0, + 1980.0, + 1221.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1938.0, + 1407.0, + 1938.0, + 1407.0, + 1980.0, + 1404.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1966.0, + 732.0, + 1966.0, + 732.0, + 2013.0, + 290.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 1966.0, + 896.0, + 1966.0, + 896.0, + 2013.0, + 753.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1966.0, + 1255.0, + 1966.0, + 1255.0, + 2013.0, + 918.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1997.0, + 292.0, + 1997.0, + 292.0, + 2044.0, + 287.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 1997.0, + 903.0, + 1997.0, + 903.0, + 2044.0, + 806.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1997.0, + 1409.0, + 1997.0, + 1409.0, + 2044.0, + 930.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 412.0, + 371.0, + 412.0, + 371.0, + 454.0, + 293.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 412.0, + 443.0, + 412.0, + 443.0, + 454.0, + 393.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 412.0, + 1093.0, + 412.0, + 1093.0, + 454.0, + 465.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 412.0, + 1407.0, + 412.0, + 1407.0, + 454.0, + 1146.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 444.0, + 858.0, + 444.0, + 858.0, + 485.0, + 756.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 444.0, + 1225.0, + 444.0, + 1225.0, + 485.0, + 927.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 444.0, + 1407.0, + 444.0, + 1407.0, + 485.0, + 1263.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 477.0, + 374.0, + 477.0, + 374.0, + 514.0, + 294.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 477.0, + 809.0, + 477.0, + 809.0, + 514.0, + 392.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1592.0, + 850.0, + 1592.0, + 850.0, + 1630.0, + 294.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1592.0, + 1039.0, + 1592.0, + 1039.0, + 1630.0, + 897.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1624.0, + 493.0, + 1624.0, + 493.0, + 1662.0, + 439.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1624.0, + 617.0, + 1624.0, + 617.0, + 1662.0, + 516.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1624.0, + 1406.0, + 1624.0, + 1406.0, + 1662.0, + 965.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1654.0, + 708.0, + 1654.0, + 708.0, + 1692.0, + 294.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1847.0, + 368.0, + 1847.0, + 368.0, + 1895.0, + 292.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 1847.0, + 982.0, + 1847.0, + 982.0, + 1895.0, + 421.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1847.0, + 1019.0, + 1847.0, + 1019.0, + 1895.0, + 1002.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1847.0, + 1408.0, + 1847.0, + 1408.0, + 1895.0, + 1073.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1885.0, + 518.0, + 1885.0, + 518.0, + 1921.0, + 293.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1885.0, + 673.0, + 1885.0, + 673.0, + 1921.0, + 541.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1885.0, + 1180.0, + 1885.0, + 1180.0, + 1921.0, + 1054.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 223.0, + 1176.0, + 223.0, + 1176.0, + 271.0, + 293.0, + 271.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 227, + 1406, + 227, + 1406, + 455, + 295, + 455 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 1274, + 1405, + 1274, + 1405, + 1448, + 296, + 1448 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 467, + 1405, + 467, + 1405, + 622, + 297, + 622 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1492, + 1406, + 1492, + 1406, + 1617, + 297, + 1617 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 942, + 1405, + 942, + 1405, + 1074, + 298, + 1074 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 300, + 1942, + 1405, + 1942, + 1405, + 2036, + 300, + 2036 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 298, + 1727, + 1407, + 1727, + 1407, + 1815, + 298, + 1815 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 296, + 1141, + 1402, + 1141, + 1402, + 1205, + 296, + 1205 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 296, + 848, + 1401, + 848, + 1401, + 926, + 296, + 926 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 294, + 627, + 1404, + 627, + 1404, + 690, + 294, + 690 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 511, + 1882, + 1182, + 1882, + 1182, + 1938, + 511, + 1938 + ], + "score": 0.931 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 106, + 297, + 106 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 299, + 1092, + 741, + 1092, + 741, + 1124, + 299, + 1124 + ], + "score": 0.912 + }, + { + "category_id": 1, + "poly": [ + 290, + 1820, + 1401, + 1820, + 1401, + 1883, + 290, + 1883 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 299, + 1454, + 1150, + 1454, + 1150, + 1486, + 299, + 1486 + ], + "score": 0.897 + }, + { + "category_id": 8, + "poly": [ + 296, + 1624, + 1429, + 1624, + 1429, + 1685, + 296, + 1685 + ], + "score": 0.881 + }, + { + "category_id": 8, + "poly": [ + 475, + 1212, + 1221, + 1212, + 1221, + 1270, + 475, + 1270 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1367, + 754, + 1402, + 754, + 1402, + 784, + 1367, + 784 + ], + "score": 0.872 + }, + { + "category_id": 9, + "poly": [ + 1370, + 1896, + 1402, + 1896, + 1402, + 1924, + 1370, + 1924 + ], + "score": 0.863 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1686, + 1403, + 1686, + 1403, + 1715, + 1367, + 1715 + ], + "score": 0.828 + }, + { + "category_id": 8, + "poly": [ + 375, + 699, + 1320, + 699, + 1320, + 753, + 375, + 753 + ], + "score": 0.785 + }, + { + "category_id": 2, + "poly": [ + 840, + 2087, + 859, + 2087, + 859, + 2112, + 840, + 2112 + ], + "score": 0.767 + }, + { + "category_id": 8, + "poly": [ + 421, + 762, + 798, + 762, + 798, + 839, + 421, + 839 + ], + "score": 0.565 + }, + { + "category_id": 8, + "poly": [ + 377, + 698, + 1322, + 698, + 1322, + 840, + 377, + 840 + ], + "score": 0.141 + }, + { + "category_id": 1, + "poly": [ + 421, + 762, + 798, + 762, + 798, + 839, + 421, + 839 + ], + "score": 0.09 + }, + { + "category_id": 13, + "poly": [ + 653, + 1818, + 788, + 1818, + 788, + 1859, + 653, + 1859 + ], + "score": 0.93, + "latex": "\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M )" + }, + { + "category_id": 13, + "poly": [ + 952, + 1727, + 1401, + 1727, + 1401, + 1774, + 952, + 1774 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } = \\int _ { \\mathcal { X } , \\mathcal { A } } ( \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ) ^ { 2 } \\mathrm { d } a \\mathrm { d } x . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 722, + 944, + 968, + 944, + 968, + 979, + 722, + 979 + ], + "score": 0.93, + "latex": "\\omega ( x , a , x ^ { \\prime } ) = \\omega ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 1168, + 1488, + 1284, + 1488, + 1284, + 1528, + 1168, + 1528 + ], + "score": 0.93, + "latex": "\\rho _ { M ^ { \\ast } } ^ { \\beta } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 621, + 847, + 763, + 847, + 763, + 888, + 621, + 888 + ], + "score": 0.93, + "latex": "\\omega ( x , a | \\rho _ { M ^ { * } } ^ { \\beta } )" + }, + { + "category_id": 13, + "poly": [ + 838, + 1974, + 960, + 1974, + 960, + 2006, + 838, + 2006 + ], + "score": 0.93, + "latex": "M ^ { * } ( \\cdot | x , a )" + }, + { + "category_id": 13, + "poly": [ + 571, + 1378, + 623, + 1378, + 623, + 1419, + 571, + 1419 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 1203, + 847, + 1254, + 847, + 1254, + 888, + 1203, + 888 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 856, + 1772, + 908, + 1772, + 908, + 1814, + 856, + 1814 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 979, + 890, + 1063, + 890, + 1063, + 924, + 979, + 924 + ], + "score": 0.92, + "latex": "\\omega ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 830, + 885, + 881, + 885, + 881, + 925, + 830, + 925 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta }" + }, + { + "category_id": 14, + "poly": [ + 310, + 1621, + 1427, + 1621, + 1427, + 1686, + 310, + 1686 + ], + "score": 0.92, + "latex": "\\operatorname* { m i n } _ { d \\in M } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M ) = \\operatorname* { m i n } _ { M \\in M } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { * } } ^ { \\mu } } \\Big [ \\omega ( x , a ) \\mathbb { E } _ { M ^ { * } } \\big [ - \\log M ( x ^ { \\prime } | x , a ) \\big ] \\Big ] - \\frac { \\alpha } { 2 } \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } ." + }, + { + "category_id": 13, + "poly": [ + 297, + 230, + 535, + 230, + 535, + 264, + 297, + 264 + ], + "score": 0.92, + "latex": " { \\mathrm { P r } } ^ { \\pi } \\left[ x _ { t } = x | x _ { 0 } , M ^ { * } \\right]" + }, + { + "category_id": 14, + "poly": [ + 516, + 1879, + 1182, + 1879, + 1182, + 1939, + 516, + 1939 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ," + }, + { + "category_id": 13, + "poly": [ + 298, + 386, + 488, + 386, + 488, + 420, + 298, + 420 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\pi } ( x ) = \\rho _ { 0 } ( x )" + }, + { + "category_id": 13, + "poly": [ + 1045, + 262, + 1155, + 262, + 1155, + 295, + 1045, + 295 + ], + "score": 0.92, + "latex": "\\gamma \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 662, + 379, + 1043, + 379, + 1043, + 426, + 662, + 426 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\omega ( x , a , x ^ { \\prime } ) = \\frac { \\rho _ { 0 } ( x ) \\beta ( a | x ) } { \\rho _ { 0 } ( x ) \\mu ( a | x ) } = \\frac { \\beta ( a | x ) } { \\mu ( a | x ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 847, + 1728, + 898, + 1728, + 898, + 1768, + 847, + 1768 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 448, + 1006, + 690, + 1006, + 690, + 1045, + 448, + 1045 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { \\ast } } ^ { \\beta } ( x , a , x ^ { \\prime } ) , \\beta \\in \\Pi" + }, + { + "category_id": 13, + "poly": [ + 551, + 1278, + 719, + 1278, + 719, + 1313, + 551, + 1313 + ], + "score": 0.91, + "latex": "\\mathbb { E } _ { x ^ { \\prime } \\sim M ^ { * } \\left( x ^ { \\prime } \\mid x , a \\right) }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1310, + 349, + 1310, + 349, + 1352, + 298, + 1352 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }" + }, + { + "category_id": 14, + "poly": [ + 475, + 1210, + 1219, + 1210, + 1219, + 1270, + 475, + 1270 + ], + "score": 0.91, + "latex": "L ( \\rho _ { M ^ { \\ast } } ^ { \\beta } , M ) = \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { \\ast } } ^ { \\mu } } \\left[ \\omega ( x , a | \\rho _ { M ^ { \\ast } } ^ { \\beta } ) \\mathbb { E } _ { M ^ { \\ast } } \\left( - \\log M ( x ^ { \\prime } | x , a ) \\right) \\right] ," + }, + { + "category_id": 13, + "poly": [ + 1279, + 1944, + 1403, + 1944, + 1403, + 1976, + 1279, + 1976 + ], + "score": 0.91, + "latex": "M ^ { * } ( \\cdot | x , a )" + }, + { + "category_id": 13, + "poly": [ + 574, + 1175, + 613, + 1175, + 613, + 1204, + 574, + 1204 + ], + "score": 0.9, + "latex": "\\ell _ { M }" + }, + { + "category_id": 13, + "poly": [ + 518, + 292, + 844, + 292, + 844, + 325, + 518, + 325 + ], + "score": 0.9, + "latex": "\\rho _ { M ^ { * } } ^ { \\pi } ( x , a ) : = \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x )" + }, + { + "category_id": 13, + "poly": [ + 384, + 1943, + 709, + 1943, + 709, + 1975, + 384, + 1975 + ], + "score": 0.89, + "latex": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )" + }, + { + "category_id": 13, + "poly": [ + 1368, + 1826, + 1399, + 1826, + 1399, + 1857, + 1368, + 1857 + ], + "score": 0.89, + "latex": "\\bar { \\beta } ^ { * }" + }, + { + "category_id": 14, + "poly": [ + 374, + 695, + 1321, + 695, + 1321, + 841, + 374, + 841 + ], + "score": 0.89, + "latex": "\\begin{array} { r l } & { \\hat { \\boldsymbol M } ^ { * } = \\underset { M \\in \\mathcal { M } } { \\mathrm { \\operatorname* { m i n } } } \\underset { \\beta \\in \\Pi } { \\mathrm { m a x } } \\boldsymbol L ( \\rho _ { M ^ { * } } ^ { \\beta } , M ) = \\underset { M \\in \\mathcal { M } } { \\mathrm { m i n } } \\underset { \\beta \\in \\Pi } { \\mathrm { m a x } } \\mathbb { E } _ { \\boldsymbol x , \\boldsymbol a , \\boldsymbol x ^ { \\prime } \\sim \\boldsymbol \\rho _ { M ^ { * } } ^ { \\mu } } [ \\omega ( \\boldsymbol x , \\boldsymbol a | \\rho _ { M ^ { * } } ^ { \\beta } ) \\ell _ { M } ( \\boldsymbol x , \\boldsymbol a , \\boldsymbol x ^ { \\prime } ) ] , } \\\\ & { } \\\\ & { s . t . \\quad \\omega ( \\boldsymbol x , \\boldsymbol a | \\rho _ { M ^ { * } } ^ { \\beta } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( \\boldsymbol x , \\boldsymbol a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( \\boldsymbol x , \\boldsymbol a ) } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1368, + 1561, + 1400, + 1561, + 1400, + 1585, + 1368, + 1585 + ], + "score": 0.88, + "latex": "L _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 827, + 1349, + 859, + 1349, + 859, + 1380, + 827, + 1380 + ], + "score": 0.88, + "latex": "\\beta ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 338, + 325, + 882, + 325, + 882, + 381, + 338, + 381 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { \\omega ( x , a , x ^ { \\prime } ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x ) \\beta ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x ) \\mu ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a ) } = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 667, + 1318, + 700, + 1318, + 700, + 1349, + 667, + 1349 + ], + "score": 0.88, + "latex": "\\beta ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 635, + 262, + 679, + 262, + 679, + 289, + 635, + 289 + ], + "score": 0.88, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 735, + 1318, + 780, + 1318, + 780, + 1346, + 735, + 1346 + ], + "score": 0.87, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 961, + 1416, + 1006, + 1416, + 1006, + 1443, + 961, + 1443 + ], + "score": 0.87, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1198, + 1318, + 1243, + 1318, + 1243, + 1346, + 1198, + 1346 + ], + "score": 0.87, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 848, + 630, + 890, + 630, + 890, + 656, + 848, + 656 + ], + "score": 0.86, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1018, + 1976, + 1062, + 1976, + 1062, + 2003, + 1018, + 2003 + ], + "score": 0.85, + "latex": "\\alpha _ { M }" + }, + { + "category_id": 13, + "poly": [ + 813, + 1385, + 846, + 1385, + 846, + 1412, + 813, + 1412 + ], + "score": 0.84, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 954, + 266, + 984, + 266, + 984, + 293, + 954, + 293 + ], + "score": 0.84, + "latex": "x _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 891, + 1494, + 912, + 1494, + 912, + 1525, + 891, + 1525 + ], + "score": 0.84, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 364, + 470, + 385, + 470, + 385, + 500, + 364, + 500 + ], + "score": 0.83, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 509, + 1494, + 542, + 1494, + 542, + 1522, + 509, + 1522 + ], + "score": 0.79, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 372, + 1740, + 393, + 1740, + 393, + 1761, + 372, + 1761 + ], + "score": 0.78, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 467, + 1976, + 591, + 1976, + 591, + 2006, + 467, + 2006 + ], + "score": 0.78, + "latex": "H _ { M ^ { \\ast } } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 370, + 1276, + 456, + 1276, + 456, + 1311, + 370, + 1311 + ], + "score": 0.78, + "latex": "\\mathbb { E } _ { M ^ { * } } \\left[ \\cdot \\right]" + }, + { + "category_id": 13, + "poly": [ + 1184, + 237, + 1204, + 237, + 1204, + 258, + 1184, + 258 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 401, + 1416, + 435, + 1416, + 435, + 1443, + 401, + 1443 + ], + "score": 0.75, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 583, + 1824, + 616, + 1824, + 616, + 1853, + 583, + 1853 + ], + "score": 0.75, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1199, + 297, + 1221, + 297, + 1221, + 319, + 1199, + 319 + ], + "score": 0.75, + "latex": "\\omega" + }, + { + "category_id": 13, + "poly": [ + 1351, + 234, + 1365, + 234, + 1365, + 258, + 1351, + 258 + ], + "score": 0.74, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 1090, + 237, + 1111, + 237, + 1111, + 258, + 1090, + 258 + ], + "score": 0.74, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 544, + 1046, + 567, + 1046, + 567, + 1069, + 544, + 1069 + ], + "score": 0.72, + "latex": "\\Pi" + }, + { + "category_id": 13, + "poly": [ + 1378, + 344, + 1397, + 344, + 1397, + 364, + 1378, + 364 + ], + "score": 0.72, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 988, + 1945, + 1041, + 1945, + 1041, + 1973, + 988, + 1973 + ], + "score": 0.68, + "latex": "( K L )" + }, + { + "category_id": 13, + "poly": [ + 443, + 892, + 464, + 892, + 464, + 922, + 443, + 922 + ], + "score": 0.68, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 298, + 2010, + 318, + 2010, + 318, + 2030, + 298, + 2030 + ], + "score": 0.67, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 346, + 1975, + 456, + 1975, + 456, + 2006, + 346, + 2006 + ], + "score": 0.66, + "latex": "M ( \\cdot | x , a )" + }, + { + "category_id": 13, + "poly": [ + 395, + 2007, + 413, + 2007, + 413, + 2029, + 395, + 2029 + ], + "score": 0.28, + "latex": "^ { 4 }" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1088.0, + 745.0, + 1088.0, + 745.0, + 1130.0, + 293.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1452.0, + 1150.0, + 1452.0, + 1150.0, + 1491.0, + 295.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 861.0, + 2086.0, + 861.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 224.0, + 296.0, + 224.0, + 296.0, + 268.0, + 293.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 224.0, + 1089.0, + 224.0, + 1089.0, + 268.0, + 536.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 224.0, + 1183.0, + 224.0, + 1183.0, + 268.0, + 1112.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 224.0, + 1350.0, + 224.0, + 1350.0, + 268.0, + 1205.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 224.0, + 1407.0, + 224.0, + 1407.0, + 268.0, + 1366.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 258.0, + 634.0, + 258.0, + 634.0, + 297.0, + 294.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 258.0, + 953.0, + 258.0, + 953.0, + 297.0, + 680.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 258.0, + 1044.0, + 258.0, + 1044.0, + 297.0, + 985.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 258.0, + 1409.0, + 258.0, + 1409.0, + 297.0, + 1156.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 290.0, + 517.0, + 290.0, + 517.0, + 329.0, + 294.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 290.0, + 1198.0, + 290.0, + 1198.0, + 329.0, + 845.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 290.0, + 1407.0, + 290.0, + 1407.0, + 329.0, + 1222.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 322.0, + 337.0, + 322.0, + 337.0, + 369.0, + 292.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 313.0, + 1377.0, + 313.0, + 1377.0, + 376.0, + 883.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 313.0, + 1411.0, + 313.0, + 1411.0, + 376.0, + 1398.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 350.0, + 886.0, + 350.0, + 886.0, + 384.0, + 883.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 376.0, + 1404.0, + 376.0, + 1404.0, + 431.0, + 1044.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 420.0, + 1054.0, + 420.0, + 1054.0, + 456.0, + 294.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 374.0, + 947.0, + 374.0, + 947.0, + 426.0, + 290.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1272.0, + 369.0, + 1272.0, + 369.0, + 1316.0, + 293.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1272.0, + 550.0, + 1272.0, + 550.0, + 1316.0, + 457.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 1272.0, + 1406.0, + 1272.0, + 1406.0, + 1316.0, + 720.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1311.0, + 297.0, + 1311.0, + 297.0, + 1352.0, + 290.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1311.0, + 666.0, + 1311.0, + 666.0, + 1352.0, + 350.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1311.0, + 734.0, + 1311.0, + 734.0, + 1352.0, + 701.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1311.0, + 1197.0, + 1311.0, + 1197.0, + 1352.0, + 781.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 1311.0, + 1408.0, + 1311.0, + 1408.0, + 1352.0, + 1244.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1346.0, + 826.0, + 1346.0, + 826.0, + 1382.0, + 291.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1346.0, + 1406.0, + 1346.0, + 1406.0, + 1382.0, + 860.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1376.0, + 570.0, + 1376.0, + 570.0, + 1427.0, + 290.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 1376.0, + 812.0, + 1376.0, + 812.0, + 1427.0, + 624.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1376.0, + 1408.0, + 1376.0, + 1408.0, + 1427.0, + 847.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1415.0, + 400.0, + 1415.0, + 400.0, + 1449.0, + 295.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1415.0, + 960.0, + 1415.0, + 960.0, + 1449.0, + 436.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 1415.0, + 1018.0, + 1415.0, + 1018.0, + 1449.0, + 1007.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 469.0, + 363.0, + 469.0, + 363.0, + 502.0, + 296.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 469.0, + 1403.0, + 469.0, + 1403.0, + 502.0, + 386.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 497.0, + 1403.0, + 497.0, + 1403.0, + 531.0, + 294.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 529.0, + 1406.0, + 529.0, + 1406.0, + 563.0, + 292.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 560.0, + 1406.0, + 560.0, + 1406.0, + 594.0, + 296.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 590.0, + 757.0, + 590.0, + 757.0, + 623.0, + 296.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1490.0, + 508.0, + 1490.0, + 508.0, + 1530.0, + 292.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1490.0, + 890.0, + 1490.0, + 890.0, + 1530.0, + 543.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 1490.0, + 1167.0, + 1490.0, + 1167.0, + 1530.0, + 913.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 1490.0, + 1406.0, + 1490.0, + 1406.0, + 1530.0, + 1285.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1523.0, + 1403.0, + 1523.0, + 1403.0, + 1556.0, + 295.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1556.0, + 1367.0, + 1556.0, + 1367.0, + 1588.0, + 294.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1584.0, + 767.0, + 1584.0, + 767.0, + 1620.0, + 292.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 943.0, + 721.0, + 943.0, + 721.0, + 980.0, + 295.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 943.0, + 1405.0, + 943.0, + 1405.0, + 980.0, + 969.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 974.0, + 1406.0, + 974.0, + 1406.0, + 1009.0, + 293.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1007.0, + 447.0, + 1007.0, + 447.0, + 1052.0, + 292.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1007.0, + 1408.0, + 1007.0, + 1408.0, + 1052.0, + 691.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1043.0, + 543.0, + 1043.0, + 543.0, + 1074.0, + 294.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1043.0, + 578.0, + 1043.0, + 578.0, + 1074.0, + 568.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1936.0, + 383.0, + 1936.0, + 383.0, + 1981.0, + 302.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1936.0, + 987.0, + 1936.0, + 987.0, + 1981.0, + 710.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 1936.0, + 1278.0, + 1936.0, + 1278.0, + 1981.0, + 1042.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1936.0, + 1407.0, + 1936.0, + 1407.0, + 1981.0, + 1404.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1968.0, + 345.0, + 1968.0, + 345.0, + 2011.0, + 294.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1968.0, + 466.0, + 1968.0, + 466.0, + 2011.0, + 457.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1968.0, + 837.0, + 1968.0, + 837.0, + 2011.0, + 592.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1968.0, + 1017.0, + 1968.0, + 1017.0, + 2011.0, + 961.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1968.0, + 1405.0, + 1968.0, + 1405.0, + 2011.0, + 1063.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 297.0, + 2001.0, + 297.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 2001.0, + 394.0, + 2001.0, + 394.0, + 2037.0, + 319.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 2001.0, + 801.0, + 2001.0, + 801.0, + 2037.0, + 414.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1718.0, + 371.0, + 1718.0, + 371.0, + 1777.0, + 288.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1718.0, + 846.0, + 1718.0, + 846.0, + 1777.0, + 394.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1718.0, + 951.0, + 1718.0, + 951.0, + 1777.0, + 899.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1718.0, + 1413.0, + 1718.0, + 1413.0, + 1777.0, + 1402.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 1752.0, + 855.0, + 1752.0, + 855.0, + 1830.0, + 281.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 1752.0, + 1116.0, + 1752.0, + 1116.0, + 1830.0, + 909.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1141.0, + 1404.0, + 1141.0, + 1404.0, + 1177.0, + 294.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1174.0, + 573.0, + 1174.0, + 573.0, + 1206.0, + 298.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1174.0, + 1392.0, + 1174.0, + 1392.0, + 1206.0, + 614.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 842.0, + 620.0, + 842.0, + 620.0, + 895.0, + 292.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 842.0, + 1202.0, + 842.0, + 1202.0, + 895.0, + 764.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 842.0, + 1408.0, + 842.0, + 1408.0, + 895.0, + 1255.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 884.0, + 442.0, + 884.0, + 442.0, + 932.0, + 294.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 884.0, + 829.0, + 884.0, + 829.0, + 932.0, + 465.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 884.0, + 978.0, + 884.0, + 978.0, + 932.0, + 882.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 884.0, + 1195.0, + 884.0, + 1195.0, + 932.0, + 1064.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 625.0, + 847.0, + 625.0, + 847.0, + 664.0, + 294.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 625.0, + 1406.0, + 625.0, + 1406.0, + 664.0, + 891.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 659.0, + 936.0, + 659.0, + 936.0, + 694.0, + 296.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1817.0, + 582.0, + 1817.0, + 582.0, + 1864.0, + 290.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 1817.0, + 652.0, + 1817.0, + 652.0, + 1864.0, + 617.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 1817.0, + 1367.0, + 1817.0, + 1367.0, + 1864.0, + 789.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1817.0, + 1404.0, + 1817.0, + 1404.0, + 1864.0, + 1400.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1849.0, + 403.0, + 1849.0, + 403.0, + 1891.0, + 292.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.25, + 770.0, + 691.25, + 770.0, + 691.25, + 829.0, + 408.25, + 829.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 224, + 1406, + 224, + 1406, + 563, + 296, + 563 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 1849, + 1407, + 1849, + 1407, + 2037, + 296, + 2037 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 1482, + 1405, + 1482, + 1405, + 1615, + 296, + 1615 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 677, + 1406, + 677, + 1406, + 792, + 299, + 792 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 1693, + 1407, + 1693, + 1407, + 1786, + 297, + 1786 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 296, + 1052, + 1406, + 1052, + 1406, + 1241, + 296, + 1241 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 297, + 936, + 1407, + 936, + 1407, + 1043, + 297, + 1043 + ], + "score": 0.965 + }, + { + "category_id": 8, + "poly": [ + 682, + 1796, + 1018, + 1796, + 1018, + 1838, + 682, + 1838 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 300, + 1389, + 1403, + 1389, + 1403, + 1464, + 300, + 1464 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 296, + 1252, + 1408, + 1252, + 1408, + 1376, + 296, + 1376 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 292, + 810, + 1454, + 810, + 1454, + 887, + 292, + 887 + ], + "score": 0.927 + }, + { + "category_id": 8, + "poly": [ + 326, + 579, + 1371, + 579, + 1371, + 656, + 326, + 656 + ], + "score": 0.92 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 105, + 297, + 105 + ], + "score": 0.918 + }, + { + "category_id": 0, + "poly": [ + 299, + 1644, + 821, + 1644, + 821, + 1678, + 299, + 1678 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 1427, + 457, + 1624, + 457, + 1624, + 488, + 1427, + 488 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1370, + 889, + 1403, + 889, + 1403, + 915, + 1370, + 915 + ], + "score": 0.837 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2113, + 840, + 2113 + ], + "score": 0.8 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1801, + 1402, + 1801, + 1402, + 1832, + 1367, + 1832 + ], + "score": 0.576 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1801, + 1402, + 1801, + 1402, + 1832, + 1367, + 1832 + ], + "score": 0.41 + }, + { + "category_id": 13, + "poly": [ + 1111, + 1390, + 1341, + 1390, + 1341, + 1425, + 1111, + 1425 + ], + "score": 0.94, + "latex": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 382, + 680, + 673, + 680, + 673, + 716, + 382, + 716 + ], + "score": 0.94, + "latex": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 543, + 1697, + 891, + 1697, + 891, + 1755, + 543, + 1755 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\big ) - f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\big ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 639, + 1919, + 726, + 1919, + 726, + 1963, + 639, + 1963 + ], + "score": 0.94, + "latex": "f ( { \\frac { p ( x ) } { q ( x ) } } )" + }, + { + "category_id": 13, + "poly": [ + 1098, + 1916, + 1398, + 1916, + 1398, + 1963, + 1098, + 1963 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) ( \\frac { p ( x ) } { q ( x ) } - 1 ) + f ( 1 ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 1915, + 402, + 1915, + 402, + 1963, + 298, + 1963 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { u = \\frac { p ( x ) } { q ( x ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 971, + 1962, + 1067, + 1962, + 1067, + 2008, + 971, + 2008 + ], + "score": 0.93, + "latex": "f ^ { \\prime } \\big ( \\frac { p ( x ) } { q ( x ) } \\big )" + }, + { + "category_id": 13, + "poly": [ + 711, + 488, + 869, + 488, + 869, + 523, + 711, + 523 + ], + "score": 0.93, + "latex": "\\rho _ { M } ^ { \\kappa } ( x , a ) > 0" + }, + { + "category_id": 13, + "poly": [ + 1159, + 1851, + 1223, + 1851, + 1223, + 1885, + 1159, + 1885 + ], + "score": 0.93, + "latex": "f ^ { \\prime } ( u )" + }, + { + "category_id": 13, + "poly": [ + 1134, + 934, + 1186, + 934, + 1186, + 978, + 1134, + 978 + ], + "score": 0.93, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 1175, + 750, + 1226, + 750, + 1226, + 794, + 1175, + 794 + ], + "score": 0.93, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 378, + 944, + 521, + 944, + 521, + 976, + 378, + 976 + ], + "score": 0.92, + "latex": "f : \\mathbb { R } _ { + } \\to \\mathbb { R }" + }, + { + "category_id": 14, + "poly": [ + 680, + 1796, + 1017, + 1796, + 1017, + 1836, + 680, + 1836 + ], + "score": 0.92, + "latex": "f ( u ) \\approx f ( 1 ) + f ^ { \\prime } ( u ) ( u - 1 ) ," + }, + { + "category_id": 13, + "poly": [ + 1191, + 1517, + 1248, + 1517, + 1248, + 1556, + 1191, + 1556 + ], + "score": 0.92, + "latex": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 1341, + 1003, + 1392, + 1003, + 1392, + 1045, + 1341, + 1045 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 427, + 224, + 479, + 224, + 479, + 267, + 427, + 267 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 962, + 1751, + 1247, + 1751, + 1247, + 1786, + 962, + 1786 + ], + "score": 0.91, + "latex": "u \\in ( 1 - \\xi , 1 + \\xi ) , \\xi > 0" + }, + { + "category_id": 13, + "poly": [ + 1118, + 1133, + 1169, + 1133, + 1169, + 1174, + 1118, + 1174 + ], + "score": 0.91, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 577, + 1169, + 625, + 1169, + 625, + 1201, + 577, + 1201 + ], + "score": 0.91, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 508, + 1052, + 559, + 1052, + 559, + 1096, + 508, + 1096 + ], + "score": 0.91, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 963, + 674, + 1110, + 674, + 1110, + 725, + 963, + 725 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } } { \\rho _ { M } ^ { \\kappa } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1201, + 347, + 1353, + 347, + 1353, + 381, + 1201, + 381 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1429, + 606, + 1429, + 606, + 1464, + 298, + 1464 + ], + "score": 0.91, + "latex": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 300, + 1961, + 348, + 1961, + 348, + 2008, + 300, + 2008 + ], + "score": 0.91, + "latex": "\\frac { p ( x ) } { q ( x ) }" + }, + { + "category_id": 13, + "poly": [ + 763, + 349, + 824, + 349, + 824, + 378, + 763, + 378 + ], + "score": 0.91, + "latex": "\\mathcal { D } _ { \\mathrm { r e a l } }" + }, + { + "category_id": 13, + "poly": [ + 959, + 294, + 1020, + 294, + 1020, + 322, + 959, + 322 + ], + "score": 0.9, + "latex": "D _ { K L }" + }, + { + "category_id": 14, + "poly": [ + 331, + 574, + 1367, + 574, + 1367, + 661, + 331, + 661 + ], + "score": 0.9, + "latex": "\\displaystyle \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\Bigg ( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } { \\rho _ { M } ^ { \\kappa } ( x , a ) } - H _ { M ^ { * } } ( x , a ) \\right) \\Bigg ) ," + }, + { + "category_id": 13, + "poly": [ + 736, + 1132, + 1031, + 1132, + 1031, + 1173, + 736, + 1173 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\theta _ { t + 1 } = \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 1198, + 505, + 1198, + 505, + 1242, + 298, + 1242 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1264, + 1883, + 1336, + 1883, + 1336, + 1917, + 1264, + 1917 + ], + "score": 0.9, + "latex": "\\mathcal { O } ( \\xi ^ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1244, + 457, + 1404, + 457, + 1404, + 490, + 1244, + 490 + ], + "score": 0.9, + "latex": "M ( x ^ { \\prime } | x , a ) =" + }, + { + "category_id": 13, + "poly": [ + 600, + 1885, + 766, + 1885, + 766, + 1918, + 600, + 1918 + ], + "score": 0.89, + "latex": "( 1 - \\xi , 1 + \\xi )" + }, + { + "category_id": 13, + "poly": [ + 1003, + 489, + 1375, + 489, + 1375, + 522, + 1003, + 522 + ], + "score": 0.89, + "latex": "\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) > 0" + }, + { + "category_id": 14, + "poly": [ + 310, + 805, + 1428, + 805, + 1428, + 890, + 310, + 890 + ], + "score": 0.89, + "latex": "\\tilde { \\bar { \\nu } } _ { M ^ { * } } ^ { * } : = \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\left( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) - H _ { M ^ { * } } ( x , a ) \\right) \\right." + }, + { + "category_id": 13, + "poly": [ + 1095, + 321, + 1139, + 321, + 1139, + 347, + 1095, + 347 + ], + "score": 0.89, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1301, + 1517, + 1353, + 1517, + 1353, + 1550, + 1301, + 1550 + ], + "score": 0.89, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa }" + }, + { + "category_id": 14, + "poly": [ + 310, + 1253, + 1408, + 1253, + 1408, + 1378, + 310, + 1378 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { \\iota _ { t + 1 } = \\underset { \\theta } { \\operatorname* { m i n } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\frac { - 1 } { \\alpha _ { 0 } ( x , a ) } \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\left( f \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) + H _ { M ^ { * } } ( x , a ) \\right) \\right] , } \\\\ { W ( x , a , x ^ { \\prime } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 638, + 263, + 975, + 263, + 975, + 298, + 638, + 298 + ], + "score": 0.88, + "latex": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) )" + }, + { + "category_id": 13, + "poly": [ + 369, + 1852, + 397, + 1852, + 397, + 1884, + 369, + 1884 + ], + "score": 0.88, + "latex": "f ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 298, + 522, + 465, + 522, + 465, + 563, + 298, + 563 + ], + "score": 0.88, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) > 0" + }, + { + "category_id": 13, + "poly": [ + 428, + 293, + 490, + 293, + 490, + 322, + 428, + 322 + ], + "score": 0.88, + "latex": "H _ { M ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 1129, + 1059, + 1161, + 1059, + 1161, + 1093, + 1129, + 1093 + ], + "score": 0.88, + "latex": "\\bar { \\beta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1154, + 1011, + 1199, + 1011, + 1199, + 1039, + 1154, + 1039 + ], + "score": 0.87, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 757, + 2005, + 801, + 2005, + 801, + 2038, + 757, + 2038 + ], + "score": 0.87, + "latex": "T _ { \\varphi ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 792, + 1757, + 809, + 1757, + 809, + 1785, + 792, + 1785 + ], + "score": 0.86, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 579, + 1392, + 716, + 1392, + 716, + 1430, + 579, + 1430 + ], + "score": 0.86, + "latex": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\kappa } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 488, + 556, + 488, + 556, + 522, + 298, + 522 + ], + "score": 0.86, + "latex": "\\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) / \\rho _ { M } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 857, + 721, + 877, + 721, + 877, + 753, + 857, + 753 + ], + "score": 0.86, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 582, + 760, + 602, + 760, + 602, + 791, + 582, + 791 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 999, + 1095, + 1311, + 1095, + 1311, + 1134, + 999, + 1134 + ], + "score": 0.85, + "latex": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 402, + 1885, + 422, + 1885, + 422, + 1916, + 402, + 1916 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 383, + 1392, + 446, + 1392, + 446, + 1429, + 383, + 1429 + ], + "score": 0.84, + "latex": "\\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } }" + }, + { + "category_id": 13, + "poly": [ + 1010, + 723, + 1029, + 723, + 1029, + 753, + 1010, + 753 + ], + "score": 0.83, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 931, + 1100, + 961, + 1100, + 961, + 1128, + 931, + 1128 + ], + "score": 0.83, + "latex": "\\theta ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 647, + 464, + 667, + 464, + 667, + 490, + 647, + 490 + ], + "score": 0.82, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 649, + 490, + 682, + 490, + 682, + 517, + 649, + 517 + ], + "score": 0.82, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 722, + 1853, + 740, + 1853, + 740, + 1883, + 722, + 1883 + ], + "score": 0.81, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 761, + 1392, + 781, + 1392, + 781, + 1425, + 761, + 1425 + ], + "score": 0.81, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 903, + 533, + 923, + 533, + 923, + 561, + 903, + 561 + ], + "score": 0.79, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 842, + 1491, + 862, + 1491, + 862, + 1512, + 842, + 1512 + ], + "score": 0.75, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 725, + 534, + 744, + 534, + 744, + 556, + 725, + 556 + ], + "score": 0.74, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 517, + 470, + 534, + 470, + 534, + 485, + 517, + 485 + ], + "score": 0.7, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 556, + 1101, + 572, + 1101, + 572, + 1127, + 556, + 1127 + ], + "score": 0.69, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 298, + 1099, + 339, + 1099, + 339, + 1131, + 298, + 1131 + ], + "score": 0.67, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1238, + 1485, + 1270, + 1485, + 1270, + 1513, + 1238, + 1513 + ], + "score": 0.46, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 413, + 721, + 455, + 721, + 455, + 750, + 413, + 750 + ], + "score": 0.32, + "latex": "\\mathrm { K L }" + }, + { + "category_id": 13, + "poly": [ + 1284, + 1485, + 1316, + 1485, + 1316, + 1513, + 1284, + 1513 + ], + "score": 0.31, + "latex": "W" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1643.0, + 825.0, + 1643.0, + 825.0, + 1680.0, + 295.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2117.0, + 840.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 226.0, + 426.0, + 226.0, + 426.0, + 273.0, + 291.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 226.0, + 1409.0, + 226.0, + 1409.0, + 273.0, + 480.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 264.0, + 637.0, + 264.0, + 637.0, + 299.0, + 295.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 264.0, + 1406.0, + 264.0, + 1406.0, + 299.0, + 976.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 291.0, + 427.0, + 291.0, + 427.0, + 326.0, + 294.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 291.0, + 958.0, + 291.0, + 958.0, + 326.0, + 491.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 291.0, + 1406.0, + 291.0, + 1406.0, + 326.0, + 1021.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 319.0, + 1094.0, + 319.0, + 1094.0, + 354.0, + 294.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 319.0, + 1407.0, + 319.0, + 1407.0, + 354.0, + 1140.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 342.0, + 762.0, + 342.0, + 762.0, + 385.0, + 293.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 342.0, + 1200.0, + 342.0, + 1200.0, + 385.0, + 825.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 342.0, + 1408.0, + 342.0, + 1408.0, + 385.0, + 1354.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 375.0, + 1406.0, + 375.0, + 1406.0, + 406.0, + 294.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 401.0, + 1407.0, + 401.0, + 1407.0, + 436.0, + 294.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 430.0, + 1404.0, + 430.0, + 1404.0, + 465.0, + 294.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 457.0, + 516.0, + 457.0, + 516.0, + 491.0, + 293.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 457.0, + 646.0, + 457.0, + 646.0, + 491.0, + 535.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 457.0, + 1243.0, + 457.0, + 1243.0, + 491.0, + 668.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 486.0, + 297.0, + 486.0, + 297.0, + 523.0, + 294.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 486.0, + 648.0, + 486.0, + 648.0, + 523.0, + 557.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 486.0, + 710.0, + 486.0, + 710.0, + 523.0, + 683.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 486.0, + 1002.0, + 486.0, + 1002.0, + 523.0, + 870.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 486.0, + 1408.0, + 486.0, + 1408.0, + 523.0, + 1376.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 513.0, + 297.0, + 513.0, + 297.0, + 572.0, + 290.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 513.0, + 724.0, + 513.0, + 724.0, + 572.0, + 466.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 513.0, + 902.0, + 513.0, + 902.0, + 572.0, + 745.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 513.0, + 1256.0, + 513.0, + 1256.0, + 572.0, + 924.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 368.0, + 1849.0, + 368.0, + 1885.0, + 294.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1849.0, + 721.0, + 1849.0, + 721.0, + 1885.0, + 398.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1849.0, + 1158.0, + 1849.0, + 1158.0, + 1885.0, + 741.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 1849.0, + 1407.0, + 1849.0, + 1407.0, + 1885.0, + 1224.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1881.0, + 401.0, + 1881.0, + 401.0, + 1917.0, + 293.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1881.0, + 599.0, + 1881.0, + 599.0, + 1917.0, + 423.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1881.0, + 1263.0, + 1881.0, + 1263.0, + 1917.0, + 767.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 1881.0, + 1407.0, + 1881.0, + 1407.0, + 1917.0, + 1337.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1921.0, + 297.0, + 1921.0, + 297.0, + 1956.0, + 294.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1903.0, + 638.0, + 1903.0, + 638.0, + 1971.0, + 403.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 1903.0, + 1097.0, + 1903.0, + 1097.0, + 1971.0, + 727.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1903.0, + 1408.0, + 1903.0, + 1408.0, + 1971.0, + 1399.0, + 1971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1955.0, + 299.0, + 1955.0, + 299.0, + 2011.0, + 293.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1955.0, + 970.0, + 1955.0, + 970.0, + 2011.0, + 349.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1955.0, + 1405.0, + 1955.0, + 1405.0, + 2011.0, + 1068.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 756.0, + 2001.0, + 756.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 2001.0, + 1405.0, + 2001.0, + 1405.0, + 2037.0, + 802.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1483.0, + 841.0, + 1483.0, + 841.0, + 1521.0, + 295.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 1483.0, + 1237.0, + 1483.0, + 1237.0, + 1521.0, + 863.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1483.0, + 1283.0, + 1483.0, + 1283.0, + 1521.0, + 1271.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1483.0, + 1405.0, + 1483.0, + 1405.0, + 1521.0, + 1317.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1512.0, + 1190.0, + 1512.0, + 1190.0, + 1558.0, + 291.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 1512.0, + 1300.0, + 1512.0, + 1300.0, + 1558.0, + 1249.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1512.0, + 1407.0, + 1512.0, + 1407.0, + 1558.0, + 1354.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1552.0, + 1405.0, + 1552.0, + 1405.0, + 1586.0, + 296.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1582.0, + 1368.0, + 1582.0, + 1368.0, + 1616.0, + 294.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 671.0, + 381.0, + 671.0, + 381.0, + 730.0, + 297.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 671.0, + 962.0, + 671.0, + 962.0, + 730.0, + 674.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 718.0, + 412.0, + 718.0, + 412.0, + 756.0, + 296.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 718.0, + 856.0, + 718.0, + 856.0, + 756.0, + 456.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 718.0, + 1009.0, + 718.0, + 1009.0, + 756.0, + 878.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 718.0, + 1408.0, + 718.0, + 1408.0, + 756.0, + 1030.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 747.0, + 581.0, + 747.0, + 581.0, + 802.0, + 291.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 747.0, + 1174.0, + 747.0, + 1174.0, + 802.0, + 603.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 747.0, + 1244.0, + 747.0, + 1244.0, + 802.0, + 1227.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 670.0, + 1407.0, + 670.0, + 1407.0, + 721.0, + 1050.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1690.0, + 542.0, + 1690.0, + 542.0, + 1762.0, + 293.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1709.0, + 1409.0, + 1709.0, + 1409.0, + 1747.0, + 892.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1752.0, + 791.0, + 1752.0, + 791.0, + 1785.0, + 294.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1752.0, + 961.0, + 1752.0, + 961.0, + 1785.0, + 810.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 1752.0, + 1355.0, + 1752.0, + 1355.0, + 1785.0, + 1248.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1689.5, + 621.0, + 1689.5, + 621.0, + 1733.5, + 576.0, + 1733.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.75, + 1686.0, + 832.75, + 1686.0, + 832.75, + 1734.0, + 770.75, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1055.0, + 507.0, + 1055.0, + 507.0, + 1103.0, + 293.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1055.0, + 1128.0, + 1055.0, + 1128.0, + 1103.0, + 560.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 1055.0, + 1408.0, + 1055.0, + 1408.0, + 1103.0, + 1162.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1096.0, + 555.0, + 1096.0, + 555.0, + 1136.0, + 340.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1096.0, + 930.0, + 1096.0, + 930.0, + 1136.0, + 573.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1096.0, + 998.0, + 1096.0, + 998.0, + 1136.0, + 962.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 1096.0, + 1407.0, + 1096.0, + 1407.0, + 1136.0, + 1312.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1130.0, + 735.0, + 1130.0, + 735.0, + 1183.0, + 290.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1130.0, + 1117.0, + 1130.0, + 1117.0, + 1183.0, + 1032.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 1130.0, + 1411.0, + 1130.0, + 1411.0, + 1183.0, + 1170.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1167.0, + 576.0, + 1167.0, + 576.0, + 1203.0, + 295.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 1167.0, + 1406.0, + 1167.0, + 1406.0, + 1203.0, + 626.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1201.0, + 297.0, + 1201.0, + 297.0, + 1246.0, + 293.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1201.0, + 522.0, + 1201.0, + 522.0, + 1246.0, + 506.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 930.0, + 377.0, + 930.0, + 377.0, + 987.0, + 299.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 930.0, + 1133.0, + 930.0, + 1133.0, + 987.0, + 522.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 930.0, + 1409.0, + 930.0, + 1409.0, + 987.0, + 1187.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 973.0, + 1405.0, + 973.0, + 1405.0, + 1006.0, + 292.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1004.0, + 1153.0, + 1004.0, + 1153.0, + 1053.0, + 291.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1004.0, + 1340.0, + 1004.0, + 1340.0, + 1053.0, + 1200.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1393.0, + 1004.0, + 1408.0, + 1004.0, + 1408.0, + 1053.0, + 1393.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 1002.5, + 1387.0, + 1002.5, + 1387.0, + 1023.5, + 1350.0, + 1023.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1383.0, + 382.0, + 1383.0, + 382.0, + 1433.0, + 300.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 1383.0, + 578.0, + 1383.0, + 578.0, + 1433.0, + 447.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1383.0, + 760.0, + 1383.0, + 760.0, + 1433.0, + 717.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 1383.0, + 1110.0, + 1383.0, + 1110.0, + 1433.0, + 782.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1383.0, + 1409.0, + 1383.0, + 1409.0, + 1433.0, + 1342.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1424.0, + 297.0, + 1424.0, + 297.0, + 1466.0, + 294.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 1424.0, + 616.0, + 1424.0, + 616.0, + 1466.0, + 607.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1423.0, + 454.0, + 1627.0, + 454.0, + 1627.0, + 491.0, + 1423.0, + 491.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1112, + 1405, + 1112, + 1405, + 1440, + 296, + 1440 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 296, + 618, + 1407, + 618, + 1407, + 874, + 296, + 874 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 1778, + 1406, + 1778, + 1406, + 1985, + 296, + 1985 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1591, + 1403, + 1591, + 1403, + 1715, + 298, + 1715 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 296, + 228, + 1406, + 228, + 1406, + 443, + 296, + 443 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 298, + 1452, + 1403, + 1452, + 1403, + 1575, + 298, + 1575 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 298, + 1005, + 1404, + 1005, + 1404, + 1110, + 298, + 1110 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 296, + 941, + 1401, + 941, + 1401, + 1002, + 296, + 1002 + ], + "score": 0.94 + }, + { + "category_id": 2, + "poly": [ + 333, + 2005, + 873, + 2005, + 873, + 2034, + 333, + 2034 + ], + "score": 0.922 + }, + { + "category_id": 0, + "poly": [ + 299, + 1727, + 559, + 1727, + 559, + 1761, + 299, + 1761 + ], + "score": 0.902 + }, + { + "category_id": 0, + "poly": [ + 300, + 892, + 726, + 892, + 726, + 924, + 300, + 924 + ], + "score": 0.895 + }, + { + "category_id": 1, + "poly": [ + 1427, + 1544, + 1623, + 1544, + 1623, + 1574, + 1427, + 1574 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.857 + }, + { + "category_id": 1, + "poly": [ + 1427, + 1240, + 1623, + 1240, + 1623, + 1270, + 1427, + 1270 + ], + "score": 0.841 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.771 + }, + { + "category_id": 9, + "poly": [ + 1370, + 511, + 1403, + 511, + 1403, + 539, + 1370, + 539 + ], + "score": 0.763 + }, + { + "category_id": 8, + "poly": [ + 306, + 506, + 1150, + 506, + 1150, + 606, + 306, + 606 + ], + "score": 0.713 + }, + { + "category_id": 1, + "poly": [ + 293, + 454, + 1488, + 454, + 1488, + 498, + 293, + 498 + ], + "score": 0.488 + }, + { + "category_id": 8, + "poly": [ + 293, + 454, + 1488, + 454, + 1488, + 498, + 293, + 498 + ], + "score": 0.409 + }, + { + "category_id": 8, + "poly": [ + 340, + 561, + 1101, + 561, + 1101, + 605, + 340, + 605 + ], + "score": 0.127 + }, + { + "category_id": 8, + "poly": [ + 314, + 507, + 1160, + 507, + 1160, + 551, + 314, + 551 + ], + "score": 0.117 + }, + { + "category_id": 1, + "poly": [ + 306, + 506, + 1150, + 506, + 1150, + 606, + 306, + 606 + ], + "score": 0.108 + }, + { + "category_id": 13, + "poly": [ + 443, + 384, + 548, + 384, + 548, + 444, + 443, + 444 + ], + "score": 0.94, + "latex": "f ^ { \\prime } \\Bigl ( \\frac { { { \\rho } _ { M _ { \\theta _ { t } } } ^ { \\kappa } } } { { { \\rho } _ { M ^ { * } } ^ { \\kappa } } } \\Bigr )" + }, + { + "category_id": 13, + "poly": [ + 853, + 1003, + 901, + 1003, + 901, + 1050, + 853, + 1050 + ], + "score": 0.93, + "latex": "\\textstyle { \\frac { p ( x ) } { q ( x ) } }" + }, + { + "category_id": 13, + "poly": [ + 763, + 692, + 1111, + 692, + 1111, + 751, + 763, + 751 + ], + "score": 0.93, + "latex": "f ^ { \\prime } ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } ) \\approx \\log D _ { \\varphi _ { 1 } ^ { * } } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 1263, + 1368, + 1371, + 1368, + 1371, + 1402, + 1263, + 1402 + ], + "score": 0.93, + "latex": "\\mathcal { N } ( 0 , \\epsilon _ { D } )" + }, + { + "category_id": 13, + "poly": [ + 1107, + 841, + 1282, + 841, + 1282, + 876, + 1107, + 876 + ], + "score": 0.93, + "latex": "H _ { M ^ { * } } - A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 624, + 1271, + 673, + 1271, + 673, + 1303, + 624, + 1303 + ], + "score": 0.92, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 691, + 708, + 691, + 708, + 751, + 297, + 751 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { f ^ { \\prime } ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } ) \\approx \\log D _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 779, + 394, + 779, + 394, + 812, + 298, + 812 + ], + "score": 0.92, + "latex": "\\alpha _ { 0 } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 1193, + 355, + 1369, + 355, + 1369, + 387, + 1193, + 387 + ], + "score": 0.92, + "latex": "T _ { \\varphi ^ { * } } = \\log { \\hat { D } _ { \\varphi ^ { * } } }" + }, + { + "category_id": 13, + "poly": [ + 364, + 1371, + 445, + 1371, + 445, + 1403, + 364, + 1403 + ], + "score": 0.92, + "latex": "\\epsilon _ { \\mu } > 0" + }, + { + "category_id": 13, + "poly": [ + 1098, + 1273, + 1156, + 1273, + 1156, + 1310, + 1098, + 1310 + ], + "score": 0.91, + "latex": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 957, + 1402, + 1008, + 1402, + 1008, + 1440, + 957, + 1440 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\hat { \\mu } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 657, + 450, + 657, + 450, + 692, + 298, + 692 + ], + "score": 0.91, + "latex": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M } ^ { \\kappa } } [ \\cdot ]" + }, + { + "category_id": 13, + "poly": [ + 978, + 658, + 1022, + 658, + 1022, + 690, + 978, + 690 + ], + "score": 0.9, + "latex": "T _ { \\varphi _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 443, + 1406, + 484, + 1406, + 484, + 1440, + 443, + 1440 + ], + "score": 0.9, + "latex": "D _ { \\varphi }" + }, + { + "category_id": 13, + "poly": [ + 879, + 658, + 923, + 658, + 923, + 690, + 879, + 690 + ], + "score": 0.9, + "latex": "T _ { \\varphi _ { 0 } }" + }, + { + "category_id": 13, + "poly": [ + 827, + 1077, + 884, + 1077, + 884, + 1116, + 827, + 1116 + ], + "score": 0.9, + "latex": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 853, + 1405, + 904, + 1405, + 904, + 1440, + 853, + 1440 + ], + "score": 0.9, + "latex": "\\rho _ { M ^ { * } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 1117, + 1241, + 1169, + 1241, + 1169, + 1273, + 1117, + 1273 + ], + "score": 0.9, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 925, + 1144, + 983, + 1144, + 983, + 1183, + 925, + 1183 + ], + "score": 0.9, + "latex": "\\cdot" + }, + { + "category_id": 13, + "poly": [ + 507, + 749, + 604, + 749, + 604, + 781, + 507, + 781 + ], + "score": 0.9, + "latex": "\\alpha _ { 0 } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 472, + 1145, + 523, + 1145, + 523, + 1177, + 472, + 1177 + ], + "score": 0.9, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 1132, + 232, + 1168, + 232, + 1168, + 265, + 1132, + 265 + ], + "score": 0.89, + "latex": "T _ { \\varphi }" + }, + { + "category_id": 13, + "poly": [ + 742, + 1077, + 793, + 1077, + 793, + 1111, + 742, + 1111 + ], + "score": 0.89, + "latex": "D _ { \\varphi _ { 0 } }" + }, + { + "category_id": 13, + "poly": [ + 299, + 1212, + 376, + 1212, + 376, + 1241, + 299, + 1241 + ], + "score": 0.89, + "latex": "\\cdot" + }, + { + "category_id": 13, + "poly": [ + 296, + 1870, + 734, + 1870, + 734, + 1926, + 296, + 1926 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\bar { \\mathbb { E } } \\left[ \\int _ { \\mathcal { A } } \\left( M ^ { * } ( y | x , a ) - \\bar { M } ( y | x , a ) \\right) ^ { 2 } \\mathrm { d } a \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 729, + 294, + 956, + 294, + 956, + 328, + 729, + 328 + ], + "score": 0.89, + "latex": "T _ { \\varphi } ( u ) = \\log D _ { \\varphi } ( u )" + }, + { + "category_id": 13, + "poly": [ + 395, + 621, + 1141, + 621, + 1141, + 658, + 395, + 658 + ], + "score": 0.89, + "latex": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ = \\ \\log D _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ - \\ \\log D _ { \\varphi _ { 1 } ^ { * } } ( x , a ) , \\ \\mathbb { E } _ { \\rho _ { M } ^ { \\kappa } } [ \\cdot ]" + }, + { + "category_id": 13, + "poly": [ + 1296, + 324, + 1343, + 324, + 1343, + 356, + 1296, + 356 + ], + "score": 0.89, + "latex": "f ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 942, + 1456, + 1000, + 1456, + 1000, + 1486, + 942, + 1486 + ], + "score": 0.89, + "latex": "H _ { M _ { \\theta } }" + }, + { + "category_id": 13, + "poly": [ + 1284, + 1515, + 1343, + 1515, + 1343, + 1544, + 1284, + 1544 + ], + "score": 0.89, + "latex": "H _ { M ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 505, + 1485, + 547, + 1485, + 547, + 1514, + 505, + 1514 + ], + "score": 0.88, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 297, + 293, + 716, + 293, + 716, + 327, + 297, + 327 + ], + "score": 0.88, + "latex": "f ( u ) = u \\log u - ( u + 1 ) \\log ( u + 1 )" + }, + { + "category_id": 14, + "poly": [ + 310, + 451, + 1428, + 451, + 1428, + 608, + 310, + 608 + ], + "score": 0.88, + "latex": "\\begin{array} { r l } { { \\iota _ { + 1 } } = \\mathop { \\operatorname* { m a x } } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\log M \\theta ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\left( H _ { M ^ { * } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] , } & { } \\\\ { { s . t . } } & { \\varphi _ { 0 } ^ { * } = \\mathop { \\operatorname* { a r g m a x } } _ { \\varphi _ { 0 } ^ { * } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } & { ( 8 ) } \\\\ & { \\varphi _ { 1 } ^ { * } = \\mathop { \\operatorname { a r g m a x } } _ { \\varphi _ { 1 } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 399, + 1453, + 460, + 1453, + 460, + 1484, + 399, + 1484 + ], + "score": 0.88, + "latex": "H _ { M ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 410, + 841, + 451, + 841, + 451, + 870, + 410, + 870 + ], + "score": 0.88, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1129, + 1075, + 1311, + 1075, + 1311, + 1110, + 1129, + 1110 + ], + "score": 0.87, + "latex": "f ^ { \\prime } ( 1 ) = \\log 0 . 5 )" + }, + { + "category_id": 13, + "poly": [ + 298, + 263, + 338, + 263, + 338, + 295, + 298, + 295 + ], + "score": 0.86, + "latex": "D _ { \\varphi }" + }, + { + "category_id": 13, + "poly": [ + 789, + 1519, + 811, + 1519, + 811, + 1541, + 789, + 1541 + ], + "score": 0.85, + "latex": "\\Sigma" + }, + { + "category_id": 13, + "poly": [ + 803, + 232, + 822, + 232, + 822, + 262, + 803, + 262 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 592, + 661, + 625, + 661, + 625, + 688, + 592, + 688 + ], + "score": 0.85, + "latex": "\\varphi _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 412, + 1341, + 431, + 1341, + 431, + 1369, + 412, + 1369 + ], + "score": 0.85, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 899, + 1549, + 920, + 1549, + 920, + 1571, + 899, + 1571 + ], + "score": 0.84, + "latex": "\\Sigma" + }, + { + "category_id": 13, + "poly": [ + 1089, + 1411, + 1117, + 1411, + 1117, + 1440, + 1089, + 1440 + ], + "score": 0.83, + "latex": "\\epsilon _ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 296, + 1306, + 635, + 1306, + 635, + 1340, + 296, + 1340 + ], + "score": 0.83, + "latex": "\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\mu ( a | x ) > 0" + }, + { + "category_id": 13, + "poly": [ + 659, + 1240, + 686, + 1240, + 686, + 1267, + 659, + 1267 + ], + "score": 0.82, + "latex": "\\mathcal { D }" + }, + { + "category_id": 13, + "poly": [ + 1303, + 1210, + 1324, + 1210, + 1324, + 1241, + 1303, + 1241 + ], + "score": 0.82, + "latex": "\\cdot" + }, + { + "category_id": 13, + "poly": [ + 505, + 661, + 537, + 661, + 537, + 688, + 505, + 688 + ], + "score": 0.82, + "latex": "\\varphi _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 445, + 1271, + 465, + 1271, + 465, + 1302, + 445, + 1302 + ], + "score": 0.82, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 696, + 1180, + 722, + 1180, + 722, + 1206, + 696, + 1206 + ], + "score": 0.81, + "latex": "\\cdot" + }, + { + "category_id": 13, + "poly": [ + 731, + 1210, + 757, + 1210, + 757, + 1236, + 731, + 1236 + ], + "score": 0.78, + "latex": "\\mathcal { D }" + }, + { + "category_id": 13, + "poly": [ + 775, + 263, + 870, + 263, + 870, + 292, + 775, + 292 + ], + "score": 0.78, + "latex": "\\forall x \\in { \\mathcal { X } }" + }, + { + "category_id": 13, + "poly": [ + 1107, + 1184, + 1127, + 1184, + 1127, + 1211, + 1107, + 1211 + ], + "score": 0.75, + "latex": "\\cdot" + }, + { + "category_id": 13, + "poly": [ + 1085, + 779, + 1119, + 779, + 1119, + 807, + 1085, + 807 + ], + "score": 0.72, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 881, + 262, + 999, + 262, + 999, + 294, + 881, + 294 + ], + "score": 0.71, + "latex": "f ^ { \\prime } ( x ) \\leq 0" + }, + { + "category_id": 13, + "poly": [ + 699, + 1118, + 718, + 1118, + 718, + 1139, + 699, + 1139 + ], + "score": 0.7, + "latex": "\\cdot" + }, + { + "category_id": 13, + "poly": [ + 378, + 1248, + 396, + 1248, + 396, + 1267, + 378, + 1267 + ], + "score": 0.7, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 1371, + 1842, + 1404, + 1842, + 1404, + 1871, + 1371, + 1871 + ], + "score": 0.66, + "latex": "=" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 2000.0, + 876.0, + 2000.0, + 876.0, + 2038.0, + 333.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1722.0, + 562.0, + 1722.0, + 562.0, + 1767.0, + 291.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 893.0, + 729.0, + 893.0, + 729.0, + 926.0, + 296.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1108.0, + 698.0, + 1108.0, + 698.0, + 1146.0, + 294.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1108.0, + 1407.0, + 1108.0, + 1407.0, + 1146.0, + 719.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1137.0, + 471.0, + 1137.0, + 471.0, + 1181.0, + 293.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1137.0, + 924.0, + 1137.0, + 924.0, + 1181.0, + 524.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1137.0, + 1410.0, + 1137.0, + 1410.0, + 1181.0, + 984.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1176.0, + 695.0, + 1176.0, + 695.0, + 1213.0, + 295.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 1176.0, + 1106.0, + 1176.0, + 1106.0, + 1213.0, + 723.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1176.0, + 1403.0, + 1176.0, + 1403.0, + 1213.0, + 1128.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1205.0, + 298.0, + 1205.0, + 298.0, + 1243.0, + 295.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 1205.0, + 730.0, + 1205.0, + 730.0, + 1243.0, + 377.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 1205.0, + 1302.0, + 1205.0, + 1302.0, + 1243.0, + 758.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1205.0, + 1405.0, + 1205.0, + 1405.0, + 1243.0, + 1325.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1236.0, + 377.0, + 1236.0, + 377.0, + 1276.0, + 294.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1236.0, + 658.0, + 1236.0, + 658.0, + 1276.0, + 397.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 1236.0, + 1116.0, + 1236.0, + 1116.0, + 1276.0, + 687.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 1236.0, + 1405.0, + 1236.0, + 1405.0, + 1276.0, + 1170.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1264.0, + 444.0, + 1264.0, + 444.0, + 1314.0, + 290.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 1264.0, + 623.0, + 1264.0, + 623.0, + 1314.0, + 466.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1264.0, + 1097.0, + 1264.0, + 1097.0, + 1314.0, + 674.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 1264.0, + 1408.0, + 1264.0, + 1408.0, + 1314.0, + 1157.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1303.0, + 1406.0, + 1303.0, + 1406.0, + 1343.0, + 636.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1337.0, + 411.0, + 1337.0, + 411.0, + 1370.0, + 294.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1337.0, + 1405.0, + 1337.0, + 1405.0, + 1370.0, + 432.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1366.0, + 363.0, + 1366.0, + 363.0, + 1404.0, + 294.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1366.0, + 1262.0, + 1366.0, + 1262.0, + 1404.0, + 446.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 1366.0, + 1405.0, + 1366.0, + 1405.0, + 1404.0, + 1372.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1398.0, + 442.0, + 1398.0, + 442.0, + 1445.0, + 292.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 1398.0, + 852.0, + 1398.0, + 852.0, + 1445.0, + 485.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 1398.0, + 956.0, + 1398.0, + 956.0, + 1445.0, + 905.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 1398.0, + 1088.0, + 1398.0, + 1088.0, + 1445.0, + 1009.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 1398.0, + 1130.0, + 1398.0, + 1130.0, + 1445.0, + 1118.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 616.0, + 394.0, + 616.0, + 394.0, + 662.0, + 308.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 616.0, + 1410.0, + 616.0, + 1410.0, + 662.0, + 1142.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 651.0, + 297.0, + 651.0, + 297.0, + 697.0, + 291.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 651.0, + 504.0, + 651.0, + 504.0, + 697.0, + 451.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 651.0, + 591.0, + 651.0, + 591.0, + 697.0, + 538.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 651.0, + 878.0, + 651.0, + 878.0, + 697.0, + 626.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 651.0, + 977.0, + 651.0, + 977.0, + 697.0, + 924.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 651.0, + 1409.0, + 651.0, + 1409.0, + 697.0, + 1023.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 680.0, + 762.0, + 680.0, + 762.0, + 753.0, + 709.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 680.0, + 1409.0, + 680.0, + 1409.0, + 753.0, + 1112.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 744.0, + 506.0, + 744.0, + 506.0, + 788.0, + 290.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 744.0, + 1408.0, + 744.0, + 1408.0, + 788.0, + 605.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 776.0, + 297.0, + 776.0, + 297.0, + 814.0, + 293.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 776.0, + 1084.0, + 776.0, + 1084.0, + 814.0, + 395.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 776.0, + 1407.0, + 776.0, + 1407.0, + 814.0, + 1120.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 808.0, + 1405.0, + 808.0, + 1405.0, + 844.0, + 294.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 830.0, + 409.0, + 830.0, + 409.0, + 886.0, + 290.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 830.0, + 1106.0, + 830.0, + 1106.0, + 886.0, + 452.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 830.0, + 1298.0, + 830.0, + 1298.0, + 886.0, + 1283.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1777.0, + 1406.0, + 1777.0, + 1406.0, + 1814.0, + 294.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1810.0, + 1409.0, + 1810.0, + 1409.0, + 1843.0, + 295.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1841.0, + 1370.0, + 1841.0, + 1370.0, + 1874.0, + 295.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1872.0, + 295.0, + 1872.0, + 295.0, + 1926.0, + 290.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1872.0, + 1413.0, + 1872.0, + 1413.0, + 1926.0, + 735.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1920.0, + 1404.0, + 1920.0, + 1404.0, + 1953.0, + 293.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1951.0, + 1405.0, + 1951.0, + 1405.0, + 1986.0, + 294.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1590.0, + 1408.0, + 1590.0, + 1408.0, + 1626.0, + 294.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1621.0, + 1404.0, + 1621.0, + 1404.0, + 1654.0, + 294.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1653.0, + 1404.0, + 1653.0, + 1404.0, + 1686.0, + 296.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1682.0, + 1015.0, + 1682.0, + 1015.0, + 1718.0, + 294.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 227.0, + 802.0, + 227.0, + 802.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 227.0, + 1131.0, + 227.0, + 1131.0, + 267.0, + 823.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 227.0, + 1407.0, + 227.0, + 1407.0, + 267.0, + 1169.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 260.0, + 774.0, + 260.0, + 774.0, + 299.0, + 339.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 260.0, + 880.0, + 260.0, + 880.0, + 299.0, + 871.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 260.0, + 1407.0, + 260.0, + 1407.0, + 299.0, + 1000.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 290.0, + 296.0, + 290.0, + 296.0, + 332.0, + 293.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 290.0, + 728.0, + 290.0, + 728.0, + 332.0, + 717.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 290.0, + 1407.0, + 290.0, + 1407.0, + 332.0, + 957.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 324.0, + 1295.0, + 324.0, + 1295.0, + 359.0, + 294.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 324.0, + 1404.0, + 324.0, + 1404.0, + 359.0, + 1344.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 351.0, + 1192.0, + 351.0, + 1192.0, + 390.0, + 293.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 351.0, + 1407.0, + 351.0, + 1407.0, + 390.0, + 1370.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 401.0, + 442.0, + 401.0, + 442.0, + 435.0, + 294.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 401.0, + 1190.0, + 401.0, + 1190.0, + 435.0, + 549.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 378.0, + 540.0, + 378.0, + 540.0, + 423.5, + 479.0, + 423.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 414.5, + 527.0, + 414.5, + 527.0, + 449.5, + 484.0, + 449.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1451.0, + 398.0, + 1451.0, + 398.0, + 1487.0, + 293.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 1451.0, + 941.0, + 1451.0, + 941.0, + 1487.0, + 461.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1451.0, + 1407.0, + 1451.0, + 1407.0, + 1487.0, + 1001.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1483.0, + 504.0, + 1483.0, + 504.0, + 1516.0, + 293.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1483.0, + 1405.0, + 1483.0, + 1405.0, + 1516.0, + 548.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1511.0, + 788.0, + 1511.0, + 788.0, + 1547.0, + 294.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1511.0, + 1283.0, + 1511.0, + 1283.0, + 1547.0, + 812.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 1511.0, + 1404.0, + 1511.0, + 1404.0, + 1547.0, + 1344.0, + 1547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1541.0, + 898.0, + 1541.0, + 898.0, + 1577.0, + 294.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 1541.0, + 931.0, + 1541.0, + 931.0, + 1577.0, + 921.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1003.0, + 852.0, + 1003.0, + 852.0, + 1054.0, + 293.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 1003.0, + 1406.0, + 1003.0, + 1406.0, + 1054.0, + 902.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1043.0, + 1405.0, + 1043.0, + 1405.0, + 1081.0, + 293.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1072.0, + 741.0, + 1072.0, + 741.0, + 1119.0, + 292.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1072.0, + 826.0, + 1072.0, + 826.0, + 1119.0, + 794.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 1072.0, + 1128.0, + 1072.0, + 1128.0, + 1119.0, + 885.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 1072.0, + 1330.0, + 1072.0, + 1330.0, + 1119.0, + 1312.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 939.0, + 1405.0, + 939.0, + 1405.0, + 979.0, + 292.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 973.0, + 654.0, + 973.0, + 654.0, + 1004.0, + 295.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 1542.0, + 1627.0, + 1542.0, + 1627.0, + 1577.0, + 1425.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 1238.0, + 1627.0, + 1238.0, + 1627.0, + 1273.0, + 1425.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 449.0, + 309.0, + 449.0, + 309.0, + 505.0, + 290.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1429.0, + 449.0, + 1491.0, + 449.0, + 1491.0, + 505.0, + 1429.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.25, + 494.5, + 894.25, + 494.5, + 894.25, + 555.0, + 302.25, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 543.0, + 868.0, + 543.0, + 868.0, + 615.5, + 360.0, + 615.5 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 848, + 1405, + 848, + 1405, + 1460, + 296, + 1460 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 520, + 1405, + 520, + 1405, + 769, + 297, + 769 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1485, + 937, + 1485, + 937, + 1820, + 298, + 1820 + ], + "score": 0.972 + }, + { + "category_id": 3, + "poly": [ + 332, + 123, + 1354, + 123, + 1354, + 432, + 332, + 432 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 297, + 1822, + 1404, + 1822, + 1404, + 2036, + 297, + 2036 + ], + "score": 0.964 + }, + { + "category_id": 3, + "poly": [ + 969, + 1451, + 1389, + 1451, + 1389, + 1725, + 969, + 1725 + ], + "score": 0.961 + }, + { + "category_id": 4, + "poly": [ + 955, + 1760, + 1404, + 1760, + 1404, + 1821, + 955, + 1821 + ], + "score": 0.946 + }, + { + "category_id": 4, + "poly": [ + 296, + 437, + 1402, + 437, + 1402, + 500, + 296, + 500 + ], + "score": 0.944 + }, + { + "category_id": 0, + "poly": [ + 301, + 803, + 807, + 803, + 807, + 834, + 301, + 834 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 296, + 75, + 857, + 75, + 857, + 105, + 296, + 105 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.788 + }, + { + "category_id": 2, + "poly": [ + 1428, + 736, + 1625, + 736, + 1625, + 766, + 1428, + 766 + ], + "score": 0.603 + }, + { + "category_id": 1, + "poly": [ + 1428, + 736, + 1625, + 736, + 1625, + 766, + 1428, + 766 + ], + "score": 0.376 + }, + { + "category_id": 13, + "poly": [ + 750, + 1002, + 855, + 1002, + 855, + 1035, + 750, + 1035 + ], + "score": 0.93, + "latex": "U ( - e , e )" + }, + { + "category_id": 13, + "poly": [ + 1195, + 1064, + 1301, + 1064, + 1301, + 1097, + 1195, + 1097 + ], + "score": 0.92, + "latex": "U ( - 1 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1033, + 519, + 1033, + 519, + 1066, + 298, + 1066 + ], + "score": 0.91, + "latex": "e \\in \\{ 1 . 0 , 0 . 2 , 0 . 0 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 366, + 1216, + 535, + 1216, + 535, + 1247, + 366, + 1247 + ], + "score": 0.91, + "latex": "0 . 2 \\leq e \\leq 1 . 0" + }, + { + "category_id": 13, + "poly": [ + 569, + 1033, + 792, + 1033, + 792, + 1066, + 569, + 1066 + ], + "score": 0.91, + "latex": "p \\in \\{ 1 . 0 , 0 . 2 , 0 . 0 5 \\}" + }, + { + "category_id": 13, + "poly": [ + 841, + 1248, + 1069, + 1248, + 1069, + 1278, + 841, + 1278 + ], + "score": 0.9, + "latex": "e = 0 . 0 5 , p < = 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 953, + 1156, + 1056, + 1156, + 1056, + 1185, + 953, + 1185 + ], + "score": 0.89, + "latex": "e = 0 . 0 5" + }, + { + "category_id": 13, + "poly": [ + 1269, + 1822, + 1339, + 1822, + 1339, + 1849, + 1269, + 1849 + ], + "score": 0.88, + "latex": "c = 2" + }, + { + "category_id": 13, + "poly": [ + 708, + 1217, + 937, + 1217, + 937, + 1247, + 708, + 1247 + ], + "score": 0.86, + "latex": "e = 0 . 0 5 , p < = 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 555, + 1067, + 674, + 1067, + 674, + 1095, + 555, + 1095 + ], + "score": 0.86, + "latex": "\\Theta ^ { 1 } \\mathrm { \\mathrm { - } } \\Theta ^ { 0 } \\cdot 2" + }, + { + "category_id": 13, + "poly": [ + 1114, + 2001, + 1135, + 2001, + 1135, + 2043, + 1114, + 2043 + ], + "score": 0.86, + "latex": "\\stackrel { \\mathrm { ~ \\tiny ~ 1 ~ } } { \\hat { \\mu } }" + }, + { + "category_id": 13, + "poly": [ + 296, + 1066, + 388, + 1066, + 388, + 1096, + 296, + 1096 + ], + "score": 0.85, + "latex": "^ { \\ast } \\mathrm { e } ^ { \\ast } \\mathrm { \\_ p ^ { \\ast } } ^ { \\ast }" + }, + { + "category_id": 13, + "poly": [ + 368, + 469, + 428, + 469, + 428, + 499, + 368, + 499 + ], + "score": 0.85, + "latex": "\\mathrm { \\times 0 . 3 }" + }, + { + "category_id": 13, + "poly": [ + 1256, + 1974, + 1277, + 1974, + 1277, + 2005, + 1256, + 2005 + ], + "score": 0.85, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 332, + 1733, + 353, + 1733, + 353, + 1761, + 332, + 1761 + ], + "score": 0.8, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 297, + 1008, + 318, + 1008, + 318, + 1034, + 297, + 1034 + ], + "score": 0.76, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 1166, + 1007, + 1185, + 1007, + 1185, + 1035, + 1166, + 1035 + ], + "score": 0.75, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1251, + 1130, + 1268, + 1130, + 1268, + 1156, + 1251, + 1156 + ], + "score": 0.72, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 505, + 1765, + 523, + 1765, + 523, + 1787, + 505, + 1787 + ], + "score": 0.64, + "latex": "c" + }, + { + "category_id": 13, + "poly": [ + 1184, + 1130, + 1201, + 1130, + 1201, + 1152, + 1184, + 1152 + ], + "score": 0.63, + "latex": "e" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 140.0, + 388.0, + 140.0, + 388.0, + 229.0, + 352.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 130.0, + 505.0, + 130.0, + 505.0, + 178.0, + 400.0, + 178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 130.0, + 907.0, + 130.0, + 907.0, + 179.0, + 890.0, + 179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 167.0, + 1032.0, + 167.0, + 1032.0, + 179.0, + 1022.0, + 179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 136.0, + 1336.0, + 136.0, + 1336.0, + 181.0, + 1224.0, + 181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 181.0, + 363.0, + 181.0, + 363.0, + 281.0, + 332.0, + 281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 181.0, + 420.0, + 181.0, + 420.0, + 190.0, + 407.0, + 190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 172.0, + 473.0, + 172.0, + 473.0, + 197.0, + 432.0, + 197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 176.0, + 1302.0, + 176.0, + 1302.0, + 201.0, + 1260.0, + 201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 192.0, + 501.0, + 192.0, + 501.0, + 218.0, + 432.0, + 218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 863.0, + 186.0, + 907.0, + 186.0, + 907.0, + 287.0, + 863.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 185.0, + 1247.0, + 185.0, + 1247.0, + 194.0, + 1233.0, + 194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 196.0, + 1330.0, + 196.0, + 1330.0, + 223.0, + 1260.0, + 223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 225.0, + 384.0, + 225.0, + 384.0, + 254.0, + 356.0, + 254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 229.0, + 1084.0, + 229.0, + 1084.0, + 244.0, + 1071.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 257.0, + 384.0, + 257.0, + 384.0, + 281.0, + 357.0, + 281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 241.0, + 906.0, + 241.0, + 906.0, + 267.0, + 890.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 257.0, + 1181.0, + 257.0, + 1181.0, + 273.0, + 1168.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 253.0, + 1230.0, + 253.0, + 1230.0, + 264.0, + 1221.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 272.0, + 906.0, + 272.0, + 906.0, + 296.0, + 890.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 287.0, + 385.0, + 287.0, + 385.0, + 310.0, + 358.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 302.0, + 905.0, + 302.0, + 905.0, + 319.0, + 892.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 311.0, + 1279.0, + 311.0, + 1279.0, + 319.0, + 1270.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 315.0, + 384.0, + 315.0, + 384.0, + 338.0, + 365.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 318.0, + 508.0, + 318.0, + 508.0, + 332.0, + 494.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 322.0, + 652.0, + 322.0, + 652.0, + 331.0, + 644.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 322.0, + 704.0, + 322.0, + 704.0, + 332.0, + 693.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 325.0, + 799.0, + 325.0, + 799.0, + 333.0, + 789.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 329.0, + 907.0, + 329.0, + 907.0, + 351.0, + 890.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 355.0, + 447.0, + 355.0, + 447.0, + 379.0, + 422.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 359.0, + 494.0, + 359.0, + 494.0, + 381.0, + 467.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 366.0, + 965.0, + 366.0, + 965.0, + 389.0, + 940.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 366.0, + 1015.0, + 366.0, + 1015.0, + 389.0, + 990.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 365.0, + 1066.0, + 365.0, + 1066.0, + 384.0, + 1044.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 366.0, + 1113.0, + 366.0, + 1113.0, + 384.0, + 1094.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 366.0, + 1164.0, + 366.0, + 1164.0, + 384.0, + 1142.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1242.0, + 365.0, + 1264.0, + 365.0, + 1264.0, + 384.0, + 1242.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 363.0, + 1316.0, + 363.0, + 1316.0, + 386.0, + 1289.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 400.0, + 647.0, + 400.0, + 647.0, + 434.0, + 528.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 400.0, + 1172.0, + 400.0, + 1172.0, + 434.0, + 1051.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.25, + 345.0, + 435.25, + 345.0, + 435.25, + 374.0, + 373.25, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 341.0, + 492.0, + 341.0, + 492.0, + 371.0, + 434.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.75, + 342.0, + 546.75, + 342.0, + 546.75, + 373.0, + 476.75, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 341.5, + 592.0, + 341.5, + 592.0, + 378.0, + 508.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.25, + 341.0, + 645.25, + 341.0, + 645.25, + 370.5, + 585.25, + 370.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 355.5, + 607.0, + 355.5, + 607.0, + 381.0, + 555.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.25, + 346.5, + 701.25, + 346.5, + 701.25, + 381.5, + 594.25, + 381.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.25, + 344.5, + 741.25, + 344.5, + 741.25, + 379.5, + 650.25, + 379.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 346.5, + 798.0, + 346.5, + 798.0, + 380.0, + 692.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 351.0, + 970.0, + 351.0, + 970.0, + 382.5, + 886.0, + 382.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 351.0, + 1022.0, + 351.0, + 1022.0, + 379.0, + 948.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.75, + 351.5, + 1071.75, + 351.5, + 1071.75, + 380.0, + 999.75, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 350.5, + 1115.0, + 350.5, + 1115.0, + 378.0, + 1054.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.75, + 351.0, + 1163.75, + 351.0, + 1163.75, + 376.0, + 1105.75, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 349.0, + 1216.0, + 349.0, + 1216.0, + 378.0, + 1152.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.25, + 352.5, + 1264.25, + 352.5, + 1264.25, + 381.0, + 1182.25, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 350.0, + 1313.0, + 350.0, + 1313.0, + 378.0, + 1249.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 349.0, + 850.0, + 349.0, + 850.0, + 383.0, + 738.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 350.5, + 1362.0, + 350.5, + 1362.0, + 378.0, + 1299.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1457.0, + 1009.0, + 1457.0, + 1009.0, + 1476.0, + 986.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1456.0, + 1106.0, + 1456.0, + 1106.0, + 1477.0, + 1070.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1468.0, + 993.0, + 1468.0, + 993.0, + 1626.0, + 964.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 1471.0, + 1233.0, + 1471.0, + 1233.0, + 1490.0, + 1208.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 1472.0, + 1135.0, + 1472.0, + 1135.0, + 1495.0, + 1071.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1472.0, + 1218.0, + 1472.0, + 1218.0, + 1625.0, + 1191.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 1475.0, + 1355.0, + 1475.0, + 1355.0, + 1495.0, + 1293.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1479.0, + 1009.0, + 1479.0, + 1009.0, + 1498.0, + 986.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1502.0, + 1009.0, + 1502.0, + 1009.0, + 1523.0, + 986.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1491.0, + 1099.0, + 1491.0, + 1099.0, + 1513.0, + 1070.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1502.0, + 1232.0, + 1502.0, + 1232.0, + 1518.0, + 1210.0, + 1518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 1493.0, + 1320.0, + 1493.0, + 1320.0, + 1515.0, + 1292.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1509.0, + 1110.0, + 1509.0, + 1110.0, + 1532.0, + 1070.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1509.0, + 1161.0, + 1509.0, + 1161.0, + 1523.0, + 1132.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 1511.0, + 1328.0, + 1511.0, + 1328.0, + 1532.0, + 1294.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1526.0, + 1009.0, + 1526.0, + 1009.0, + 1545.0, + 986.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1526.0, + 1129.0, + 1526.0, + 1129.0, + 1549.0, + 1070.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 1531.0, + 1232.0, + 1531.0, + 1232.0, + 1549.0, + 1209.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 1528.0, + 1351.0, + 1528.0, + 1351.0, + 1550.0, + 1293.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 1547.0, + 1010.0, + 1547.0, + 1010.0, + 1570.0, + 985.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 1562.0, + 1232.0, + 1562.0, + 1232.0, + 1580.0, + 1209.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 1572.0, + 1009.0, + 1572.0, + 1009.0, + 1592.0, + 985.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 1594.0, + 1010.0, + 1594.0, + 1010.0, + 1617.0, + 985.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 1593.0, + 1231.0, + 1593.0, + 1231.0, + 1609.0, + 1211.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1619.0, + 1010.0, + 1619.0, + 1010.0, + 1639.0, + 986.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1624.0, + 1231.0, + 1624.0, + 1231.0, + 1638.0, + 1218.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1644.0, + 1169.0, + 1644.0, + 1169.0, + 1666.0, + 1003.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1646.0, + 1392.0, + 1646.0, + 1392.0, + 1665.0, + 1238.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 1661.0, + 1158.0, + 1661.0, + 1158.0, + 1682.0, + 1015.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1661.0, + 1378.0, + 1661.0, + 1378.0, + 1682.0, + 1239.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1691.0, + 1149.0, + 1691.0, + 1149.0, + 1724.0, + 988.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1692.0, + 1365.0, + 1692.0, + 1365.0, + 1722.0, + 1218.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 1454.0, + 1387.0, + 1454.0, + 1387.0, + 1478.5, + 1265.0, + 1478.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 1473.0, + 1264.0, + 1473.0, + 1264.0, + 1486.0, + 1230.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1755.0, + 1409.0, + 1755.0, + 1409.0, + 1797.0, + 954.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 1794.0, + 1126.0, + 1794.0, + 1126.0, + 1822.0, + 955.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 435.0, + 1403.0, + 435.0, + 1403.0, + 471.0, + 295.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 469.0, + 367.0, + 469.0, + 367.0, + 502.0, + 294.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 469.0, + 781.0, + 469.0, + 781.0, + 502.0, + 429.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 802.0, + 810.0, + 802.0, + 810.0, + 837.0, + 295.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 2085.0, + 859.0, + 2085.0, + 859.0, + 2116.0, + 836.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1424.0, + 733.0, + 1630.0, + 733.0, + 1630.0, + 771.0, + 1424.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 846.0, + 1407.0, + 846.0, + 1407.0, + 887.0, + 294.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 882.0, + 1405.0, + 882.0, + 1405.0, + 915.0, + 294.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 909.0, + 1405.0, + 909.0, + 1405.0, + 946.0, + 293.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 941.0, + 1403.0, + 941.0, + 1403.0, + 976.0, + 294.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 970.0, + 1405.0, + 970.0, + 1405.0, + 1007.0, + 294.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1001.0, + 296.0, + 1001.0, + 296.0, + 1038.0, + 291.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1001.0, + 749.0, + 1001.0, + 749.0, + 1038.0, + 319.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1001.0, + 1165.0, + 1001.0, + 1165.0, + 1038.0, + 856.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 1001.0, + 1405.0, + 1001.0, + 1405.0, + 1038.0, + 1186.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1033.0, + 297.0, + 1033.0, + 297.0, + 1067.0, + 293.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 1033.0, + 568.0, + 1033.0, + 568.0, + 1067.0, + 520.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 1033.0, + 1405.0, + 1033.0, + 1405.0, + 1067.0, + 793.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 1063.0, + 554.0, + 1063.0, + 554.0, + 1097.0, + 389.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 1063.0, + 1194.0, + 1063.0, + 1194.0, + 1097.0, + 675.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1063.0, + 1406.0, + 1063.0, + 1406.0, + 1097.0, + 1302.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1092.0, + 1406.0, + 1092.0, + 1406.0, + 1129.0, + 291.0, + 1129.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1122.0, + 1183.0, + 1122.0, + 1183.0, + 1159.0, + 293.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1122.0, + 1250.0, + 1122.0, + 1250.0, + 1159.0, + 1202.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1122.0, + 1405.0, + 1122.0, + 1405.0, + 1159.0, + 1269.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1155.0, + 952.0, + 1155.0, + 952.0, + 1190.0, + 295.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1155.0, + 1405.0, + 1155.0, + 1405.0, + 1190.0, + 1057.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1186.0, + 1405.0, + 1186.0, + 1405.0, + 1218.0, + 294.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1215.0, + 365.0, + 1215.0, + 365.0, + 1248.0, + 294.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1215.0, + 707.0, + 1215.0, + 707.0, + 1248.0, + 536.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1215.0, + 1405.0, + 1215.0, + 1405.0, + 1248.0, + 938.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1245.0, + 840.0, + 1245.0, + 840.0, + 1279.0, + 293.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1245.0, + 1405.0, + 1245.0, + 1405.0, + 1279.0, + 1070.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1275.0, + 1406.0, + 1275.0, + 1406.0, + 1311.0, + 294.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1307.0, + 1406.0, + 1307.0, + 1406.0, + 1341.0, + 293.0, + 1341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1339.0, + 1405.0, + 1339.0, + 1405.0, + 1370.0, + 296.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1363.0, + 1406.0, + 1363.0, + 1406.0, + 1406.0, + 293.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1396.0, + 1407.0, + 1396.0, + 1407.0, + 1433.0, + 293.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1427.0, + 658.0, + 1427.0, + 658.0, + 1466.0, + 292.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 1451.5, + 1026.0, + 1451.5, + 1026.0, + 1460.5, + 1007.0, + 1460.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 521.0, + 1405.0, + 521.0, + 1405.0, + 555.0, + 295.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 554.0, + 1405.0, + 554.0, + 1405.0, + 584.0, + 295.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 580.0, + 1408.0, + 580.0, + 1408.0, + 618.0, + 295.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 611.0, + 1406.0, + 611.0, + 1406.0, + 649.0, + 292.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 643.0, + 1405.0, + 643.0, + 1405.0, + 677.0, + 295.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 675.0, + 1403.0, + 675.0, + 1403.0, + 705.0, + 297.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 706.0, + 1402.0, + 706.0, + 1402.0, + 737.0, + 296.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 733.0, + 1376.0, + 733.0, + 1376.0, + 769.0, + 292.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1484.0, + 937.0, + 1484.0, + 937.0, + 1519.0, + 295.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1517.0, + 936.0, + 1517.0, + 936.0, + 1546.0, + 295.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1546.0, + 939.0, + 1546.0, + 939.0, + 1579.0, + 294.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1575.0, + 939.0, + 1575.0, + 939.0, + 1609.0, + 295.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1608.0, + 937.0, + 1608.0, + 937.0, + 1638.0, + 295.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1636.0, + 937.0, + 1636.0, + 937.0, + 1668.0, + 294.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1668.0, + 938.0, + 1668.0, + 938.0, + 1701.0, + 295.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1697.0, + 937.0, + 1697.0, + 937.0, + 1730.0, + 294.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1729.0, + 331.0, + 1729.0, + 331.0, + 1758.0, + 295.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1729.0, + 938.0, + 1729.0, + 938.0, + 1758.0, + 354.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1760.0, + 504.0, + 1760.0, + 504.0, + 1789.0, + 295.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1760.0, + 936.0, + 1760.0, + 936.0, + 1789.0, + 524.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1789.0, + 937.0, + 1789.0, + 937.0, + 1820.0, + 296.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1817.0, + 1268.0, + 1817.0, + 1268.0, + 1853.0, + 292.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1340.0, + 1817.0, + 1405.0, + 1817.0, + 1405.0, + 1853.0, + 1340.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1848.0, + 1404.0, + 1848.0, + 1404.0, + 1885.0, + 294.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1879.0, + 1405.0, + 1879.0, + 1405.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1906.0, + 1405.0, + 1906.0, + 1405.0, + 1950.0, + 291.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1977.0, + 295.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 1255.0, + 1973.0, + 1255.0, + 2008.0, + 295.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2008.0, + 1278.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2003.0, + 1113.0, + 2003.0, + 1113.0, + 2037.0, + 295.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 2003.0, + 1404.0, + 2003.0, + 1404.0, + 2037.0, + 1136.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1424.0, + 733.0, + 1630.0, + 733.0, + 1630.0, + 771.0, + 1424.0, + 771.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1452, + 1405, + 1452, + 1405, + 2035, + 297, + 2035 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 872, + 1405, + 872, + 1405, + 1179, + 298, + 1179 + ], + "score": 0.983 + }, + { + "category_id": 5, + "poly": [ + 300, + 286, + 1398, + 286, + 1398, + 468, + 300, + 468 + ], + "score": 0.979, + "html": "
TaskHopperWalker2dHalfCheetah|avg. norm.
HorizonH=10H=20H=40H=10H=20H=40H=10H=20H=40/
GALILEO13.0 ± 0.133.2 ± 0.153.5± 1.211.7±0.229.9 ± 0.361.2 ± 3.40.7±0.2-1.1 ± 0.2-14.2 ± 1.451.1
SL4.8±0.53.0±0.24.6±0.210.7±0.220.1±0.837.5±6.70.4±0.5-1.1± 0.6-13.2 ± 0.321.1
IPW5.9 ±0.74.1 ± 0.65.9±0.24.7 ± 1.12.8 ±3.914.5 ± 1.41.6± 0.20.5 ± 0.8-11.3 ± 0.919.7
SCIGAN12.7 ± 0.129.2 ± 0.646.2 ±5.28.4 ± 0.59.1 ± 1.71.0 ± 5.81.2 ± 0.3-0.3 ±1.0-11.4 ± 0.341.8
MAX-RETURN13.2 ±0.033.3±0.271.0±0.514.9 ± 1.360.7 ± 11.1221.1 ± 8.92.6±0.113.3 ± 1.149.1 ± 2.3100.0
" + }, + { + "category_id": 1, + "poly": [ + 297, + 561, + 1404, + 561, + 1404, + 715, + 297, + 715 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1193, + 1403, + 1193, + 1403, + 1439, + 297, + 1439 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 301, + 724, + 1401, + 724, + 1401, + 819, + 301, + 819 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 299, + 488, + 1398, + 488, + 1398, + 551, + 299, + 551 + ], + "score": 0.937 + }, + { + "category_id": 0, + "poly": [ + 301, + 832, + 788, + 832, + 788, + 864, + 301, + 864 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 856, + 75, + 856, + 105, + 298, + 105 + ], + "score": 0.848 + }, + { + "category_id": 6, + "poly": [ + 296, + 130, + 1403, + 130, + 1403, + 285, + 296, + 285 + ], + "score": 0.835 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2111, + 840, + 2111 + ], + "score": 0.788 + }, + { + "category_id": 2, + "poly": [ + 1426, + 1149, + 1628, + 1149, + 1628, + 1210, + 1426, + 1210 + ], + "score": 0.579 + }, + { + "category_id": 1, + "poly": [ + 1426, + 1149, + 1628, + 1149, + 1628, + 1210, + 1426, + 1210 + ], + "score": 0.507 + }, + { + "category_id": 13, + "poly": [ + 1150, + 1409, + 1346, + 1409, + 1346, + 1442, + 1150, + 1442 + ], + "score": 0.91, + "latex": "H \\in \\{ 1 0 , 2 0 , 4 0 \\}" + }, + { + "category_id": 13, + "poly": [ + 813, + 1516, + 882, + 1516, + 882, + 1545, + 813, + 1545 + ], + "score": 0.87, + "latex": "161 \\%" + }, + { + "category_id": 13, + "poly": [ + 720, + 1516, + 777, + 1516, + 777, + 1545, + 720, + 1545 + ], + "score": 0.86, + "latex": "24 \\%" + }, + { + "category_id": 13, + "poly": [ + 589, + 757, + 758, + 757, + 758, + 788, + 589, + 788 + ], + "score": 0.85, + "latex": "\\mathtt { e 0 } . 2 _ { - \\mathtt { p } 0 . 0 5 }" + }, + { + "category_id": 13, + "poly": [ + 753, + 491, + 774, + 491, + 774, + 521, + 753, + 521 + ], + "score": 0.82, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 794, + 520, + 815, + 520, + 815, + 551, + 794, + 551 + ], + "score": 0.81, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 459, + 490, + 479, + 490, + 479, + 520, + 459, + 520 + ], + "score": 0.78, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 830.0, + 790.0, + 830.0, + 790.0, + 867.0, + 295.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 126.0, + 1406.0, + 126.0, + 1406.0, + 165.0, + 293.0, + 165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 157.0, + 1405.0, + 157.0, + 1405.0, + 194.0, + 294.0, + 194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 190.0, + 1405.0, + 190.0, + 1405.0, + 227.0, + 294.0, + 227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 222.0, + 1405.0, + 222.0, + 1405.0, + 255.0, + 295.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 250.0, + 1381.0, + 250.0, + 1381.0, + 287.0, + 294.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 1146.0, + 1633.0, + 1146.0, + 1633.0, + 1178.0, + 1425.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1421.0, + 1177.0, + 1604.0, + 1177.0, + 1604.0, + 1212.0, + 1421.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1455.0, + 1403.0, + 1455.0, + 1403.0, + 1486.0, + 296.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1484.0, + 1405.0, + 1484.0, + 1405.0, + 1519.0, + 295.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1514.0, + 719.0, + 1514.0, + 719.0, + 1546.0, + 295.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 1514.0, + 812.0, + 1514.0, + 812.0, + 1546.0, + 778.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1514.0, + 1403.0, + 1514.0, + 1403.0, + 1546.0, + 883.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1546.0, + 1405.0, + 1546.0, + 1405.0, + 1579.0, + 292.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1573.0, + 1407.0, + 1573.0, + 1407.0, + 1614.0, + 292.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1608.0, + 1402.0, + 1608.0, + 1402.0, + 1639.0, + 296.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1636.0, + 1407.0, + 1636.0, + 1407.0, + 1672.0, + 294.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1669.0, + 1403.0, + 1669.0, + 1403.0, + 1701.0, + 296.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1696.0, + 1403.0, + 1696.0, + 1403.0, + 1732.0, + 294.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1728.0, + 1403.0, + 1728.0, + 1403.0, + 1760.0, + 296.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1760.0, + 1405.0, + 1760.0, + 1405.0, + 1791.0, + 295.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1789.0, + 1403.0, + 1789.0, + 1403.0, + 1820.0, + 296.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1819.0, + 1405.0, + 1819.0, + 1405.0, + 1854.0, + 295.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1851.0, + 1403.0, + 1851.0, + 1403.0, + 1883.0, + 296.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1880.0, + 1405.0, + 1880.0, + 1405.0, + 1916.0, + 295.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1909.0, + 1405.0, + 1909.0, + 1405.0, + 1946.0, + 294.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1939.0, + 1405.0, + 1939.0, + 1405.0, + 1978.0, + 295.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1973.0, + 1403.0, + 1973.0, + 1403.0, + 2005.0, + 296.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2000.0, + 1405.0, + 2000.0, + 1405.0, + 2037.0, + 295.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 873.0, + 1405.0, + 873.0, + 1405.0, + 905.0, + 296.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 902.0, + 1405.0, + 902.0, + 1405.0, + 940.0, + 293.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 928.0, + 1407.0, + 928.0, + 1407.0, + 975.0, + 290.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 964.0, + 1406.0, + 964.0, + 1406.0, + 1000.0, + 295.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 994.0, + 1407.0, + 994.0, + 1407.0, + 1029.0, + 295.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1026.0, + 1405.0, + 1026.0, + 1405.0, + 1060.0, + 293.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1052.0, + 1410.0, + 1052.0, + 1410.0, + 1094.0, + 292.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1084.0, + 1407.0, + 1084.0, + 1407.0, + 1122.0, + 293.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1116.0, + 1407.0, + 1116.0, + 1407.0, + 1153.0, + 292.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1146.0, + 1319.0, + 1146.0, + 1319.0, + 1181.0, + 293.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 563.0, + 1404.0, + 563.0, + 1404.0, + 596.0, + 296.0, + 596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 590.0, + 1405.0, + 590.0, + 1405.0, + 627.0, + 295.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 622.0, + 1406.0, + 622.0, + 1406.0, + 659.0, + 295.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 650.0, + 1406.0, + 650.0, + 1406.0, + 690.0, + 292.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 680.0, + 952.0, + 680.0, + 952.0, + 717.0, + 294.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1193.0, + 1405.0, + 1193.0, + 1405.0, + 1227.0, + 296.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1223.0, + 1403.0, + 1223.0, + 1403.0, + 1258.0, + 294.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1255.0, + 1405.0, + 1255.0, + 1405.0, + 1288.0, + 295.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1285.0, + 1404.0, + 1285.0, + 1404.0, + 1322.0, + 292.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1316.0, + 1403.0, + 1316.0, + 1403.0, + 1350.0, + 295.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1346.0, + 1405.0, + 1346.0, + 1405.0, + 1380.0, + 295.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1376.0, + 1404.0, + 1376.0, + 1404.0, + 1411.0, + 294.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1405.0, + 1149.0, + 1405.0, + 1149.0, + 1442.0, + 293.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1405.0, + 1356.0, + 1405.0, + 1356.0, + 1442.0, + 1347.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 723.0, + 1404.0, + 723.0, + 1404.0, + 759.0, + 295.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 756.0, + 588.0, + 756.0, + 588.0, + 790.0, + 297.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 756.0, + 1405.0, + 756.0, + 1405.0, + 790.0, + 759.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 785.0, + 1379.0, + 785.0, + 1379.0, + 823.0, + 296.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 489.0, + 458.0, + 489.0, + 458.0, + 521.0, + 295.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 489.0, + 752.0, + 489.0, + 752.0, + 521.0, + 480.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 489.0, + 1402.0, + 489.0, + 1402.0, + 521.0, + 775.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 518.0, + 793.0, + 518.0, + 793.0, + 554.0, + 295.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 518.0, + 933.0, + 518.0, + 933.0, + 554.0, + 816.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 1146.0, + 1633.0, + 1146.0, + 1633.0, + 1178.0, + 1425.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1421.0, + 1177.0, + 1604.0, + 1177.0, + 1604.0, + 1212.0, + 1421.0, + 1212.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 679, + 1404, + 679, + 1404, + 1016, + 298, + 1016 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1031, + 1404, + 1031, + 1404, + 1277, + 297, + 1277 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1850, + 1404, + 1850, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1460, + 1404, + 1460, + 1404, + 1644, + 297, + 1644 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1290, + 1402, + 1290, + 1402, + 1446, + 298, + 1446 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1712, + 1404, + 1712, + 1404, + 1835, + 298, + 1835 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 331, + 123, + 1360, + 123, + 1360, + 429, + 331, + 429 + ], + "score": 0.972 + }, + { + "category_id": 4, + "poly": [ + 297, + 435, + 1408, + 435, + 1408, + 588, + 297, + 588 + ], + "score": 0.969 + }, + { + "category_id": 2, + "poly": [ + 1427, + 1245, + 1625, + 1245, + 1625, + 1309, + 1427, + 1309 + ], + "score": 0.93 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 857, + 75, + 857, + 105, + 297, + 105 + ], + "score": 0.875 + }, + { + "category_id": 0, + "poly": [ + 300, + 1665, + 826, + 1665, + 826, + 1700, + 300, + 1700 + ], + "score": 0.861 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2112, + 836, + 2112 + ], + "score": 0.849 + }, + { + "category_id": 1, + "poly": [ + 299, + 599, + 1399, + 599, + 1399, + 661, + 299, + 661 + ], + "score": 0.682 + }, + { + "category_id": 4, + "poly": [ + 299, + 599, + 1399, + 599, + 1399, + 661, + 299, + 661 + ], + "score": 0.192 + }, + { + "category_id": 15, + "poly": [ + 679.0, + 120.0, + 704.0, + 120.0, + 704.0, + 379.0, + 679.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 120.0, + 1058.0, + 120.0, + 1058.0, + 379.0, + 1029.0, + 379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 146.0, + 381.0, + 146.0, + 381.0, + 166.0, + 346.0, + 166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 146.0, + 1082.0, + 146.0, + 1082.0, + 166.0, + 1050.0, + 166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 147.0, + 357.0, + 147.0, + 357.0, + 354.0, + 328.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 147.0, + 731.0, + 147.0, + 731.0, + 165.0, + 707.0, + 165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 602.0, + 149.0, + 654.0, + 149.0, + 654.0, + 168.0, + 602.0, + 168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 149.0, + 803.0, + 149.0, + 803.0, + 169.0, + 760.0, + 169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 153.0, + 604.0, + 153.0, + 604.0, + 164.0, + 583.0, + 164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 181.0, + 382.0, + 181.0, + 382.0, + 206.0, + 345.0, + 206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 163.0, + 624.0, + 163.0, + 624.0, + 181.0, + 603.0, + 181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 185.0, + 733.0, + 185.0, + 733.0, + 211.0, + 705.0, + 211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 162.0, + 809.0, + 162.0, + 809.0, + 193.0, + 760.0, + 193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 169.0, + 1082.0, + 169.0, + 1082.0, + 213.0, + 1048.0, + 213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 218.0, + 382.0, + 218.0, + 382.0, + 244.0, + 345.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 229.0, + 733.0, + 229.0, + 733.0, + 252.0, + 704.0, + 252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 213.0, + 1084.0, + 213.0, + 1084.0, + 263.0, + 1045.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 258.0, + 382.0, + 258.0, + 382.0, + 282.0, + 345.0, + 282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 268.0, + 732.0, + 268.0, + 732.0, + 295.0, + 700.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 261.0, + 1082.0, + 261.0, + 1082.0, + 283.0, + 1048.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 283.0, + 1084.0, + 283.0, + 1084.0, + 307.0, + 1045.0, + 307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 298.0, + 381.0, + 298.0, + 381.0, + 319.0, + 347.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 314.0, + 733.0, + 314.0, + 733.0, + 336.0, + 698.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 306.0, + 1083.0, + 306.0, + 1083.0, + 331.0, + 1045.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 303.0, + 1356.0, + 303.0, + 1356.0, + 336.0, + 1228.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 335.0, + 382.0, + 335.0, + 382.0, + 356.0, + 346.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 331.0, + 1082.0, + 331.0, + 1082.0, + 352.0, + 1047.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 330.0, + 1338.0, + 330.0, + 1338.0, + 349.0, + 1230.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 351.0, + 408.0, + 351.0, + 408.0, + 372.0, + 377.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 351.0, + 459.0, + 351.0, + 459.0, + 372.0, + 428.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 353.0, + 507.0, + 353.0, + 507.0, + 371.0, + 479.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 353.0, + 557.0, + 353.0, + 557.0, + 371.0, + 529.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 351.0, + 609.0, + 351.0, + 609.0, + 373.0, + 577.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 351.0, + 659.0, + 351.0, + 659.0, + 373.0, + 628.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 351.0, + 760.0, + 351.0, + 760.0, + 372.0, + 729.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 351.0, + 809.0, + 351.0, + 809.0, + 372.0, + 778.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 351.0, + 860.0, + 351.0, + 860.0, + 372.0, + 829.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 351.0, + 909.0, + 351.0, + 909.0, + 372.0, + 878.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 351.0, + 960.0, + 351.0, + 960.0, + 372.0, + 929.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 353.0, + 1007.0, + 353.0, + 1007.0, + 371.0, + 981.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 354.0, + 1102.0, + 354.0, + 1102.0, + 371.0, + 1087.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 352.0, + 1157.0, + 352.0, + 1157.0, + 372.0, + 1140.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 352.0, + 1215.0, + 352.0, + 1215.0, + 371.0, + 1192.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 350.0, + 1271.0, + 350.0, + 1271.0, + 373.0, + 1245.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 350.0, + 1327.0, + 350.0, + 1327.0, + 373.0, + 1298.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 367.0, + 648.0, + 367.0, + 648.0, + 389.0, + 388.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 364.0, + 1008.0, + 364.0, + 1008.0, + 389.0, + 730.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 363.0, + 1272.0, + 363.0, + 1272.0, + 392.0, + 1167.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 398.0, + 646.0, + 398.0, + 646.0, + 431.0, + 345.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 398.0, + 942.0, + 398.0, + 942.0, + 432.0, + 753.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 398.0, + 1329.0, + 398.0, + 1329.0, + 432.0, + 1065.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 434.0, + 1402.0, + 434.0, + 1402.0, + 467.0, + 296.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 464.0, + 1406.0, + 464.0, + 1406.0, + 500.0, + 292.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 494.0, + 1407.0, + 494.0, + 1407.0, + 529.0, + 292.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 525.0, + 1406.0, + 525.0, + 1406.0, + 562.0, + 295.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 555.0, + 1312.0, + 555.0, + 1312.0, + 592.0, + 295.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 1242.0, + 1628.0, + 1242.0, + 1628.0, + 1278.0, + 1425.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1423.0, + 1270.0, + 1534.0, + 1270.0, + 1534.0, + 1314.0, + 1423.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1661.0, + 832.0, + 1661.0, + 832.0, + 1707.0, + 292.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 868.0, + 2084.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 593.0, + 1405.0, + 593.0, + 1405.0, + 635.0, + 293.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 626.0, + 1168.0, + 626.0, + 1168.0, + 664.0, + 296.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 676.0, + 1404.0, + 676.0, + 1404.0, + 716.0, + 293.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 710.0, + 1404.0, + 710.0, + 1404.0, + 744.0, + 294.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 739.0, + 1405.0, + 739.0, + 1405.0, + 774.0, + 294.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 771.0, + 1404.0, + 771.0, + 1404.0, + 805.0, + 294.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 800.0, + 1404.0, + 800.0, + 1404.0, + 835.0, + 294.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 830.0, + 1405.0, + 830.0, + 1405.0, + 866.0, + 293.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 861.0, + 1405.0, + 861.0, + 1405.0, + 896.0, + 296.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 895.0, + 1405.0, + 895.0, + 1405.0, + 926.0, + 296.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 921.0, + 1408.0, + 921.0, + 1408.0, + 959.0, + 294.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 953.0, + 1405.0, + 953.0, + 1405.0, + 988.0, + 293.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 984.0, + 460.0, + 984.0, + 460.0, + 1016.0, + 294.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1030.0, + 1405.0, + 1030.0, + 1405.0, + 1064.0, + 296.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1059.0, + 1404.0, + 1059.0, + 1404.0, + 1095.0, + 294.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1090.0, + 1409.0, + 1090.0, + 1409.0, + 1128.0, + 292.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1120.0, + 1406.0, + 1120.0, + 1406.0, + 1156.0, + 295.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1152.0, + 1405.0, + 1152.0, + 1405.0, + 1186.0, + 294.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1180.0, + 1404.0, + 1180.0, + 1404.0, + 1216.0, + 294.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1215.0, + 1402.0, + 1215.0, + 1402.0, + 1245.0, + 296.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1244.0, + 1221.0, + 1244.0, + 1221.0, + 1278.0, + 295.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 1406.0, + 1849.0, + 1406.0, + 1885.0, + 294.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1877.0, + 1409.0, + 1877.0, + 1409.0, + 1920.0, + 291.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1909.0, + 1405.0, + 1909.0, + 1405.0, + 1947.0, + 292.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1941.0, + 1405.0, + 1941.0, + 1405.0, + 1978.0, + 293.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1970.0, + 1405.0, + 1970.0, + 1405.0, + 2008.0, + 292.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 989.0, + 2002.0, + 989.0, + 2036.0, + 295.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1459.0, + 1406.0, + 1459.0, + 1406.0, + 1495.0, + 295.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1490.0, + 1406.0, + 1490.0, + 1406.0, + 1525.0, + 295.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1516.0, + 1408.0, + 1516.0, + 1408.0, + 1560.0, + 292.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1551.0, + 1406.0, + 1551.0, + 1406.0, + 1588.0, + 292.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1580.0, + 1406.0, + 1580.0, + 1406.0, + 1618.0, + 294.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1612.0, + 1410.0, + 1612.0, + 1410.0, + 1648.0, + 290.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1291.0, + 1404.0, + 1291.0, + 1404.0, + 1324.0, + 297.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1323.0, + 1404.0, + 1323.0, + 1404.0, + 1356.0, + 296.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1351.0, + 1406.0, + 1351.0, + 1406.0, + 1388.0, + 294.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1380.0, + 1408.0, + 1380.0, + 1408.0, + 1420.0, + 292.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1412.0, + 1362.0, + 1412.0, + 1362.0, + 1448.0, + 293.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1713.0, + 1405.0, + 1713.0, + 1405.0, + 1746.0, + 294.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1742.0, + 1405.0, + 1742.0, + 1405.0, + 1777.0, + 293.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1773.0, + 1409.0, + 1773.0, + 1409.0, + 1810.0, + 294.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1803.0, + 1313.0, + 1803.0, + 1313.0, + 1840.0, + 292.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 593.0, + 1405.0, + 593.0, + 1405.0, + 635.0, + 293.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 626.0, + 1168.0, + 626.0, + 1168.0, + 664.0, + 296.0, + 664.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 689, + 1405, + 689, + 1405, + 813, + 299, + 813 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 298, + 512, + 1405, + 512, + 1405, + 666, + 298, + 666 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 297, + 1708, + 1405, + 1708, + 1405, + 1832, + 297, + 1832 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 298, + 396, + 1400, + 396, + 1400, + 489, + 298, + 489 + ], + "score": 0.906 + }, + { + "category_id": 1, + "poly": [ + 298, + 1008, + 1405, + 1008, + 1405, + 1102, + 298, + 1102 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 856, + 75, + 856, + 105, + 298, + 105 + ], + "score": 0.894 + }, + { + "category_id": 1, + "poly": [ + 298, + 1591, + 1404, + 1591, + 1404, + 1685, + 298, + 1685 + ], + "score": 0.884 + }, + { + "category_id": 1, + "poly": [ + 297, + 1328, + 1405, + 1328, + 1405, + 1453, + 297, + 1453 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 298, + 1475, + 1405, + 1475, + 1405, + 1569, + 298, + 1569 + ], + "score": 0.877 + }, + { + "category_id": 1, + "poly": [ + 297, + 1125, + 1402, + 1125, + 1402, + 1190, + 297, + 1190 + ], + "score": 0.874 + }, + { + "category_id": 1, + "poly": [ + 299, + 1941, + 1404, + 1941, + 1404, + 2035, + 299, + 2035 + ], + "score": 0.872 + }, + { + "category_id": 1, + "poly": [ + 297, + 1210, + 1406, + 1210, + 1406, + 1305, + 297, + 1305 + ], + "score": 0.868 + }, + { + "category_id": 1, + "poly": [ + 295, + 835, + 1401, + 835, + 1401, + 901, + 295, + 901 + ], + "score": 0.857 + }, + { + "category_id": 1, + "poly": [ + 293, + 923, + 1403, + 923, + 1403, + 986, + 293, + 986 + ], + "score": 0.848 + }, + { + "category_id": 1, + "poly": [ + 288, + 1854, + 1402, + 1854, + 1402, + 1919, + 288, + 1919 + ], + "score": 0.838 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 861, + 2088, + 861, + 2113, + 835, + 2113 + ], + "score": 0.82 + }, + { + "category_id": 1, + "poly": [ + 299, + 280, + 1406, + 280, + 1406, + 372, + 299, + 372 + ], + "score": 0.81 + }, + { + "category_id": 0, + "poly": [ + 300, + 227, + 488, + 227, + 488, + 261, + 300, + 261 + ], + "score": 0.798 + }, + { + "category_id": 13, + "poly": [ + 1361, + 1244, + 1402, + 1244, + 1402, + 1272, + 1361, + 1272 + ], + "score": 0.25, + "latex": "7 t h" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 227.0, + 490.0, + 227.0, + 490.0, + 264.0, + 296.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 688.0, + 1406.0, + 688.0, + 1406.0, + 727.0, + 293.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 719.0, + 1406.0, + 719.0, + 1406.0, + 757.0, + 321.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 751.0, + 1407.0, + 751.0, + 1407.0, + 786.0, + 319.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 781.0, + 399.0, + 781.0, + 399.0, + 818.0, + 321.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 514.0, + 1407.0, + 514.0, + 1407.0, + 547.0, + 297.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 543.0, + 1405.0, + 543.0, + 1405.0, + 580.0, + 322.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 573.0, + 1403.0, + 573.0, + 1403.0, + 609.0, + 322.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 604.0, + 1403.0, + 604.0, + 1403.0, + 641.0, + 322.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 635.0, + 561.0, + 635.0, + 561.0, + 668.0, + 322.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1706.0, + 1408.0, + 1706.0, + 1408.0, + 1747.0, + 294.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1739.0, + 1406.0, + 1739.0, + 1406.0, + 1773.0, + 320.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1769.0, + 1405.0, + 1769.0, + 1405.0, + 1805.0, + 323.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1802.0, + 880.0, + 1802.0, + 880.0, + 1834.0, + 324.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 395.0, + 1404.0, + 395.0, + 1404.0, + 433.0, + 296.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 426.0, + 1405.0, + 426.0, + 1405.0, + 463.0, + 322.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 459.0, + 543.0, + 459.0, + 543.0, + 491.0, + 324.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1011.0, + 1402.0, + 1011.0, + 1402.0, + 1042.0, + 297.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1037.0, + 1406.0, + 1037.0, + 1406.0, + 1078.0, + 320.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1071.0, + 680.0, + 1071.0, + 680.0, + 1102.0, + 321.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1591.0, + 1406.0, + 1591.0, + 1406.0, + 1628.0, + 294.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1623.0, + 1405.0, + 1623.0, + 1405.0, + 1657.0, + 322.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1653.0, + 1041.0, + 1653.0, + 1041.0, + 1686.0, + 322.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1326.0, + 1405.0, + 1326.0, + 1405.0, + 1366.0, + 292.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1359.0, + 1406.0, + 1359.0, + 1406.0, + 1395.0, + 321.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1388.0, + 1405.0, + 1388.0, + 1405.0, + 1428.0, + 320.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1421.0, + 950.0, + 1421.0, + 950.0, + 1454.0, + 321.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1475.0, + 1406.0, + 1475.0, + 1406.0, + 1512.0, + 295.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1507.0, + 1405.0, + 1507.0, + 1405.0, + 1540.0, + 322.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1535.0, + 1314.0, + 1535.0, + 1314.0, + 1573.0, + 322.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1125.0, + 1404.0, + 1125.0, + 1404.0, + 1161.0, + 297.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1155.0, + 1044.0, + 1155.0, + 1044.0, + 1195.0, + 322.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1937.0, + 1408.0, + 1937.0, + 1408.0, + 1980.0, + 293.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2006.0, + 322.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 2003.0, + 1190.0, + 2003.0, + 1190.0, + 2037.0, + 323.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1210.0, + 1406.0, + 1210.0, + 1406.0, + 1248.0, + 294.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1244.0, + 1360.0, + 1244.0, + 1360.0, + 1278.0, + 323.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1273.0, + 1224.0, + 1273.0, + 1224.0, + 1307.0, + 323.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 838.0, + 1403.0, + 838.0, + 1403.0, + 871.0, + 298.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 868.0, + 1376.0, + 868.0, + 1376.0, + 902.0, + 323.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 924.0, + 1406.0, + 924.0, + 1406.0, + 960.0, + 296.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 953.0, + 1067.0, + 953.0, + 1067.0, + 988.0, + 322.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1850.0, + 1405.0, + 1850.0, + 1405.0, + 1893.0, + 291.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1885.0, + 1065.0, + 1885.0, + 1065.0, + 1921.0, + 324.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 279.0, + 1409.0, + 279.0, + 1409.0, + 316.0, + 294.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 310.0, + 1406.0, + 310.0, + 1406.0, + 348.0, + 323.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 340.0, + 1132.0, + 340.0, + 1132.0, + 376.0, + 319.0, + 376.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 229, + 1404, + 229, + 1404, + 352, + 299, + 352 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 298, + 373, + 1406, + 373, + 1406, + 527, + 298, + 527 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 298, + 1597, + 1406, + 1597, + 1406, + 1690, + 298, + 1690 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 298, + 1826, + 1405, + 1826, + 1405, + 1950, + 298, + 1950 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 299, + 1481, + 1405, + 1481, + 1405, + 1575, + 299, + 1575 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 299, + 1168, + 1406, + 1168, + 1406, + 1292, + 299, + 1292 + ], + "score": 0.913 + }, + { + "category_id": 1, + "poly": [ + 297, + 550, + 1408, + 550, + 1408, + 644, + 297, + 644 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 294, + 748, + 1402, + 748, + 1402, + 814, + 294, + 814 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 300, + 1710, + 1401, + 1710, + 1401, + 1805, + 300, + 1805 + ], + "score": 0.902 + }, + { + "category_id": 1, + "poly": [ + 303, + 1972, + 1401, + 1972, + 1401, + 2034, + 303, + 2034 + ], + "score": 0.9 + }, + { + "category_id": 1, + "poly": [ + 296, + 1001, + 1404, + 1001, + 1404, + 1064, + 296, + 1064 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 296, + 832, + 1401, + 832, + 1401, + 896, + 296, + 896 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 297, + 1397, + 1406, + 1397, + 1406, + 1463, + 297, + 1463 + ], + "score": 0.895 + }, + { + "category_id": 1, + "poly": [ + 294, + 1085, + 1402, + 1085, + 1402, + 1149, + 294, + 1149 + ], + "score": 0.889 + }, + { + "category_id": 1, + "poly": [ + 296, + 917, + 1402, + 917, + 1402, + 980, + 296, + 980 + ], + "score": 0.888 + }, + { + "category_id": 1, + "poly": [ + 295, + 663, + 1403, + 663, + 1403, + 728, + 295, + 728 + ], + "score": 0.871 + }, + { + "category_id": 1, + "poly": [ + 296, + 1313, + 1403, + 1313, + 1403, + 1378, + 296, + 1378 + ], + "score": 0.865 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 856, + 75, + 856, + 105, + 298, + 105 + ], + "score": 0.865 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2112, + 836, + 2112 + ], + "score": 0.836 + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 227.0, + 1405.0, + 227.0, + 1405.0, + 266.0, + 294.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 261.0, + 1404.0, + 261.0, + 1404.0, + 296.0, + 321.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 289.0, + 1409.0, + 289.0, + 1409.0, + 327.0, + 318.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 320.0, + 398.0, + 320.0, + 398.0, + 354.0, + 319.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 375.0, + 1403.0, + 375.0, + 1403.0, + 408.0, + 296.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 405.0, + 1406.0, + 405.0, + 1406.0, + 442.0, + 321.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 430.0, + 1411.0, + 430.0, + 1411.0, + 477.0, + 320.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 463.0, + 1409.0, + 463.0, + 1409.0, + 502.0, + 320.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 496.0, + 415.0, + 496.0, + 415.0, + 531.0, + 321.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1596.0, + 1406.0, + 1596.0, + 1406.0, + 1634.0, + 295.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1629.0, + 1408.0, + 1629.0, + 1408.0, + 1663.0, + 322.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1654.0, + 398.0, + 1654.0, + 398.0, + 1695.0, + 320.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1826.0, + 1403.0, + 1826.0, + 1403.0, + 1861.0, + 295.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1858.0, + 1405.0, + 1858.0, + 1405.0, + 1890.0, + 323.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1887.0, + 1409.0, + 1887.0, + 1409.0, + 1923.0, + 321.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1917.0, + 395.0, + 1917.0, + 395.0, + 1950.0, + 321.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1478.0, + 1405.0, + 1478.0, + 1405.0, + 1521.0, + 296.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1509.0, + 1405.0, + 1509.0, + 1405.0, + 1552.0, + 319.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1544.0, + 680.0, + 1544.0, + 680.0, + 1578.0, + 322.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1170.0, + 1406.0, + 1170.0, + 1406.0, + 1203.0, + 298.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1198.0, + 1409.0, + 1198.0, + 1409.0, + 1236.0, + 319.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1232.0, + 1408.0, + 1232.0, + 1408.0, + 1265.0, + 322.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1260.0, + 395.0, + 1260.0, + 395.0, + 1293.0, + 321.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 548.0, + 1408.0, + 548.0, + 1408.0, + 587.0, + 294.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 580.0, + 1409.0, + 580.0, + 1409.0, + 618.0, + 321.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 610.0, + 564.0, + 610.0, + 564.0, + 644.0, + 323.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 746.0, + 1405.0, + 746.0, + 1405.0, + 784.0, + 294.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 780.0, + 1180.0, + 780.0, + 1180.0, + 816.0, + 323.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1709.0, + 1405.0, + 1709.0, + 1405.0, + 1749.0, + 293.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1744.0, + 1405.0, + 1744.0, + 1405.0, + 1778.0, + 323.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1773.0, + 790.0, + 1773.0, + 790.0, + 1807.0, + 323.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2007.0, + 298.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 2004.0, + 688.0, + 2004.0, + 688.0, + 2033.0, + 324.0, + 2033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 998.0, + 1407.0, + 998.0, + 1407.0, + 1039.0, + 294.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1030.0, + 584.0, + 1030.0, + 584.0, + 1066.0, + 323.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 828.0, + 1406.0, + 828.0, + 1406.0, + 870.0, + 294.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 864.0, + 506.0, + 864.0, + 506.0, + 897.0, + 324.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1396.0, + 1408.0, + 1396.0, + 1408.0, + 1436.0, + 295.0, + 1436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1431.0, + 1152.0, + 1431.0, + 1152.0, + 1462.0, + 323.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1082.0, + 1404.0, + 1082.0, + 1404.0, + 1122.0, + 294.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1116.0, + 1155.0, + 1116.0, + 1155.0, + 1152.0, + 323.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 914.0, + 1406.0, + 914.0, + 1406.0, + 956.0, + 293.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 947.0, + 686.0, + 947.0, + 686.0, + 982.0, + 321.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 663.0, + 1404.0, + 663.0, + 1404.0, + 699.0, + 294.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 695.0, + 519.0, + 695.0, + 519.0, + 728.0, + 323.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1313.0, + 1409.0, + 1313.0, + 1409.0, + 1349.0, + 294.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1345.0, + 1080.0, + 1345.0, + 1080.0, + 1378.0, + 324.0, + 1378.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2112, + 835, + 2112 + ], + "score": 0.838 + }, + { + "category_id": 1, + "poly": [ + 297, + 962, + 1400, + 962, + 1400, + 1028, + 297, + 1028 + ], + "score": 0.836 + }, + { + "category_id": 1, + "poly": [ + 297, + 848, + 1405, + 848, + 1405, + 944, + 297, + 944 + ], + "score": 0.817 + }, + { + "category_id": 1, + "poly": [ + 294, + 1218, + 1403, + 1218, + 1403, + 1285, + 294, + 1285 + ], + "score": 0.809 + }, + { + "category_id": 1, + "poly": [ + 299, + 1385, + 1402, + 1385, + 1402, + 1481, + 299, + 1481 + ], + "score": 0.797 + }, + { + "category_id": 1, + "poly": [ + 292, + 1302, + 1404, + 1302, + 1404, + 1368, + 292, + 1368 + ], + "score": 0.794 + }, + { + "category_id": 1, + "poly": [ + 297, + 1044, + 1407, + 1044, + 1407, + 1200, + 297, + 1200 + ], + "score": 0.791 + }, + { + "category_id": 1, + "poly": [ + 292, + 1499, + 1403, + 1499, + 1403, + 1564, + 292, + 1564 + ], + "score": 0.767 + }, + { + "category_id": 1, + "poly": [ + 298, + 704, + 1405, + 704, + 1405, + 830, + 298, + 830 + ], + "score": 0.755 + }, + { + "category_id": 1, + "poly": [ + 298, + 1809, + 1405, + 1809, + 1405, + 1904, + 298, + 1904 + ], + "score": 0.735 + }, + { + "category_id": 1, + "poly": [ + 299, + 591, + 1402, + 591, + 1402, + 687, + 299, + 687 + ], + "score": 0.726 + }, + { + "category_id": 1, + "poly": [ + 294, + 508, + 1402, + 508, + 1402, + 573, + 294, + 573 + ], + "score": 0.696 + }, + { + "category_id": 1, + "poly": [ + 291, + 229, + 1402, + 229, + 1402, + 294, + 291, + 294 + ], + "score": 0.671 + }, + { + "category_id": 1, + "poly": [ + 298, + 1696, + 1405, + 1696, + 1405, + 1790, + 298, + 1790 + ], + "score": 0.665 + }, + { + "category_id": 1, + "poly": [ + 292, + 312, + 1402, + 312, + 1402, + 377, + 292, + 377 + ], + "score": 0.653 + }, + { + "category_id": 1, + "poly": [ + 300, + 394, + 1403, + 394, + 1403, + 490, + 300, + 490 + ], + "score": 0.65 + }, + { + "category_id": 1, + "poly": [ + 299, + 1581, + 1403, + 1581, + 1403, + 1677, + 299, + 1677 + ], + "score": 0.571 + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 868.0, + 2084.0, + 868.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 957.0, + 1405.0, + 957.0, + 1405.0, + 1003.0, + 292.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 994.0, + 1305.0, + 994.0, + 1305.0, + 1030.0, + 323.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 845.0, + 1406.0, + 845.0, + 1406.0, + 890.0, + 292.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 881.0, + 1405.0, + 881.0, + 1405.0, + 916.0, + 321.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 911.0, + 1035.0, + 911.0, + 1035.0, + 945.0, + 319.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1218.0, + 1404.0, + 1218.0, + 1404.0, + 1255.0, + 293.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1249.0, + 458.0, + 1249.0, + 458.0, + 1283.0, + 322.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1385.0, + 1408.0, + 1385.0, + 1408.0, + 1423.0, + 294.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1415.0, + 1409.0, + 1415.0, + 1409.0, + 1455.0, + 322.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1448.0, + 814.0, + 1448.0, + 814.0, + 1482.0, + 320.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1301.0, + 1405.0, + 1301.0, + 1405.0, + 1340.0, + 294.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1335.0, + 920.0, + 1335.0, + 920.0, + 1367.0, + 323.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1046.0, + 1405.0, + 1046.0, + 1405.0, + 1079.0, + 297.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1077.0, + 1405.0, + 1077.0, + 1405.0, + 1111.0, + 323.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1105.0, + 1409.0, + 1105.0, + 1409.0, + 1145.0, + 321.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1137.0, + 1409.0, + 1137.0, + 1409.0, + 1174.0, + 321.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1167.0, + 416.0, + 1167.0, + 416.0, + 1202.0, + 321.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1495.0, + 1406.0, + 1495.0, + 1406.0, + 1539.0, + 294.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1530.0, + 1360.0, + 1530.0, + 1360.0, + 1566.0, + 323.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 706.0, + 1406.0, + 706.0, + 1406.0, + 742.0, + 296.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 739.0, + 1403.0, + 739.0, + 1403.0, + 772.0, + 321.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 767.0, + 1405.0, + 767.0, + 1405.0, + 802.0, + 321.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 799.0, + 828.0, + 799.0, + 828.0, + 832.0, + 321.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1807.0, + 1405.0, + 1807.0, + 1405.0, + 1847.0, + 293.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1842.0, + 1405.0, + 1842.0, + 1405.0, + 1876.0, + 322.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1870.0, + 950.0, + 1870.0, + 950.0, + 1908.0, + 322.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 592.0, + 1407.0, + 592.0, + 1407.0, + 627.0, + 297.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 623.0, + 1407.0, + 623.0, + 1407.0, + 660.0, + 322.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 654.0, + 1134.0, + 654.0, + 1134.0, + 688.0, + 322.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 504.0, + 1405.0, + 504.0, + 1405.0, + 550.0, + 293.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 541.0, + 946.0, + 541.0, + 946.0, + 573.0, + 323.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 265.0, + 295.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 263.0, + 807.0, + 263.0, + 807.0, + 292.0, + 323.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1696.0, + 1406.0, + 1696.0, + 1406.0, + 1730.0, + 295.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1724.0, + 1405.0, + 1724.0, + 1405.0, + 1765.0, + 320.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1758.0, + 585.0, + 1758.0, + 585.0, + 1791.0, + 321.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 312.0, + 1404.0, + 312.0, + 1404.0, + 348.0, + 295.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 344.0, + 936.0, + 344.0, + 936.0, + 377.0, + 325.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 392.0, + 1404.0, + 392.0, + 1404.0, + 433.0, + 291.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 425.0, + 1408.0, + 425.0, + 1408.0, + 465.0, + 321.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 458.0, + 860.0, + 458.0, + 860.0, + 492.0, + 324.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1580.0, + 1405.0, + 1580.0, + 1405.0, + 1619.0, + 294.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1614.0, + 1404.0, + 1614.0, + 1404.0, + 1649.0, + 323.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1644.0, + 978.0, + 1644.0, + 978.0, + 1678.0, + 319.0, + 1678.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1827, + 1407, + 1827, + 1407, + 2036, + 296, + 2036 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 401, + 1227, + 1339, + 1227, + 1339, + 1420, + 401, + 1420 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 298, + 1612, + 1395, + 1612, + 1395, + 1683, + 298, + 1683 + ], + "score": 0.936 + }, + { + "category_id": 0, + "poly": [ + 298, + 1497, + 854, + 1497, + 854, + 1536, + 298, + 1536 + ], + "score": 0.924 + }, + { + "category_id": 0, + "poly": [ + 299, + 290, + 537, + 290, + 537, + 349, + 299, + 349 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 298, + 1565, + 1190, + 1565, + 1190, + 1600, + 298, + 1600 + ], + "score": 0.909 + }, + { + "category_id": 1, + "poly": [ + 400, + 1038, + 1340, + 1038, + 1340, + 1167, + 400, + 1167 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 299, + 407, + 601, + 407, + 601, + 449, + 299, + 449 + ], + "score": 0.906 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 106, + 297, + 106 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2113, + 835, + 2113 + ], + "score": 0.859 + }, + { + "category_id": 8, + "poly": [ + 298, + 1693, + 1397, + 1693, + 1397, + 1765, + 298, + 1765 + ], + "score": 0.85 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1765, + 1402, + 1765, + 1402, + 1794, + 1367, + 1794 + ], + "score": 0.779 + }, + { + "category_id": 1, + "poly": [ + 398, + 498, + 1339, + 498, + 1339, + 626, + 398, + 626 + ], + "score": 0.756 + }, + { + "category_id": 0, + "poly": [ + 355, + 467, + 1340, + 467, + 1340, + 493, + 355, + 493 + ], + "score": 0.643 + }, + { + "category_id": 1, + "poly": [ + 396, + 858, + 1335, + 858, + 1335, + 921, + 396, + 921 + ], + "score": 0.633 + }, + { + "category_id": 0, + "poly": [ + 361, + 1009, + 1339, + 1009, + 1339, + 1034, + 361, + 1034 + ], + "score": 0.601 + }, + { + "category_id": 0, + "poly": [ + 360, + 1191, + 1339, + 1191, + 1339, + 1221, + 360, + 1221 + ], + "score": 0.545 + }, + { + "category_id": 1, + "poly": [ + 1305, + 949, + 1337, + 949, + 1337, + 976, + 1305, + 976 + ], + "score": 0.482 + }, + { + "category_id": 1, + "poly": [ + 364, + 652, + 1339, + 652, + 1339, + 798, + 364, + 798 + ], + "score": 0.469 + }, + { + "category_id": 1, + "poly": [ + 364, + 948, + 672, + 948, + 672, + 976, + 364, + 976 + ], + "score": 0.44 + }, + { + "category_id": 0, + "poly": [ + 388, + 827, + 1336, + 827, + 1336, + 852, + 388, + 852 + ], + "score": 0.341 + }, + { + "category_id": 0, + "poly": [ + 365, + 827, + 576, + 827, + 576, + 854, + 365, + 854 + ], + "score": 0.183 + }, + { + "category_id": 0, + "poly": [ + 407, + 948, + 1328, + 948, + 1328, + 976, + 407, + 976 + ], + "score": 0.179 + }, + { + "category_id": 0, + "poly": [ + 364, + 1007, + 611, + 1007, + 611, + 1035, + 364, + 1035 + ], + "score": 0.177 + }, + { + "category_id": 0, + "poly": [ + 374, + 653, + 1335, + 653, + 1335, + 682, + 374, + 682 + ], + "score": 0.165 + }, + { + "category_id": 13, + "poly": [ + 498, + 1644, + 633, + 1644, + 633, + 1684, + 498, + 1684 + ], + "score": 0.94, + "latex": "\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , \\bar { M } )" + }, + { + "category_id": 13, + "poly": [ + 370, + 1826, + 590, + 1826, + 590, + 1884, + 370, + 1884 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\omega ( x , a ) = \\frac { \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 2002, + 405, + 2002, + 405, + 2036, + 297, + 2036 + ], + "score": 0.92, + "latex": "\\gamma \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 641, + 1832, + 1071, + 1832, + 1071, + 1880, + 641, + 1880 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } = \\int _ { \\mathcal { X } , \\mathcal { A } } \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ^ { 2 } \\mathrm { d } a \\mathrm { d } x } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1220, + 1881, + 1309, + 1881, + 1309, + 1915, + 1220, + 1915 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\pi } ( x )" + }, + { + "category_id": 13, + "poly": [ + 1137, + 1942, + 1373, + 1942, + 1373, + 1976, + 1137, + 1976 + ], + "score": 0.92, + "latex": " { \\mathrm { P r } } ^ { \\pi } \\left[ x _ { t } = x | x _ { 0 } , M ^ { * } \\right]" + }, + { + "category_id": 14, + "poly": [ + 310, + 1691, + 1402, + 1691, + 1402, + 1765, + 310, + 1765 + ], + "score": 0.89, + "latex": "\\operatorname* { m i n } _ { \\substack { M \\in M } } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\int _ { \\mathcal { X } , \\mathcal { A } } \\rho _ { M ^ { * } } ^ { \\mu } ( x , a ) \\omega ( x , a ) \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\left( - \\log M ( x ^ { \\prime } | x , a ) \\right) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 } ," + }, + { + "category_id": 13, + "poly": [ + 1309, + 1841, + 1332, + 1841, + 1332, + 1871, + 1309, + 1871 + ], + "score": 0.87, + "latex": "l _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 646, + 1880, + 1142, + 1880, + 1142, + 1915, + 646, + 1915 + ], + "score": 0.86, + "latex": "\\rho _ { M ^ { * } } ^ { \\pi } ( x , a , x ^ { \\prime } ) = \\rho _ { M ^ { * } } ^ { \\pi } ( x ) \\pi ( a | x ) M ^ { * } ( x ^ { \\prime } | x , a )" + }, + { + "category_id": 13, + "poly": [ + 1005, + 1975, + 1049, + 1975, + 1049, + 2000, + 1005, + 2000 + ], + "score": 0.86, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 405, + 1942, + 1053, + 1942, + 1053, + 1977, + 405, + 1977 + ], + "score": 0.84, + "latex": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\pi } ( x ) : = ( 1 - \\gamma ) \\mathbb { E } _ { x _ { 0 } \\sim \\rho _ { 0 } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } \\mathrm { P r } ( x _ { t } = x | x _ { 0 } , M ^ { * } ) \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 863, + 1979, + 894, + 1979, + 894, + 2003, + 863, + 2003 + ], + "score": 0.83, + "latex": "x _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1188, + 1980, + 1207, + 1980, + 1207, + 2000, + 1188, + 2000 + ], + "score": 0.8, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 688, + 1981, + 707, + 1981, + 707, + 2000, + 688, + 2000 + ], + "score": 0.78, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1340, + 1977, + 1354, + 1977, + 1354, + 2000, + 1340, + 2000 + ], + "score": 0.76, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 589, + 1918, + 607, + 1918, + 607, + 1939, + 589, + 1939 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 833, + 1616, + 855, + 1616, + 855, + 1642, + 833, + 1642 + ], + "score": 0.71, + "latex": "\\mathbb { E }" + }, + { + "category_id": 13, + "poly": [ + 725, + 1921, + 744, + 1921, + 744, + 1939, + 725, + 1939 + ], + "score": 0.7, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 618, + 1891, + 637, + 1891, + 637, + 1909, + 618, + 1909 + ], + "score": 0.58, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1081, + 1917, + 1104, + 1917, + 1104, + 1939, + 1081, + 1939 + ], + "score": 0.39, + "latex": "\\&" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1497.0, + 856.0, + 1497.0, + 856.0, + 1538.0, + 295.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 282.0, + 545.0, + 282.0, + 545.0, + 356.0, + 288.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 400.0, + 603.0, + 400.0, + 603.0, + 455.0, + 296.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 464.0, + 706.0, + 464.0, + 706.0, + 497.0, + 362.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 463.0, + 1340.0, + 463.0, + 1340.0, + 497.0, + 1303.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1008.0, + 392.0, + 1008.0, + 392.0, + 1033.0, + 364.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1005.0, + 614.0, + 1005.0, + 614.0, + 1037.0, + 394.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1002.0, + 1338.0, + 1002.0, + 1338.0, + 1038.0, + 1303.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1193.0, + 606.0, + 1193.0, + 606.0, + 1221.0, + 361.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1191.0, + 1340.0, + 1191.0, + 1340.0, + 1224.0, + 1301.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 825.0, + 579.0, + 825.0, + 579.0, + 856.0, + 399.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 823.0, + 1340.0, + 823.0, + 1340.0, + 856.0, + 1302.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 825.0, + 578.0, + 825.0, + 578.0, + 856.0, + 360.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 948.0, + 676.0, + 948.0, + 676.0, + 976.0, + 403.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 945.0, + 1337.0, + 945.0, + 1337.0, + 978.0, + 1301.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1006.0, + 612.0, + 1006.0, + 612.0, + 1035.0, + 362.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 651.0, + 798.0, + 651.0, + 798.0, + 683.0, + 364.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 650.0, + 1341.0, + 650.0, + 1341.0, + 684.0, + 1302.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1814.0, + 369.0, + 1814.0, + 369.0, + 1888.0, + 291.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 1814.0, + 640.0, + 1814.0, + 640.0, + 1888.0, + 591.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1814.0, + 1308.0, + 1814.0, + 1308.0, + 1888.0, + 1072.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 1814.0, + 1413.0, + 1814.0, + 1413.0, + 1888.0, + 1333.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1877.0, + 617.0, + 1877.0, + 617.0, + 1918.0, + 291.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1877.0, + 645.0, + 1877.0, + 645.0, + 1918.0, + 638.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 1877.0, + 1219.0, + 1877.0, + 1219.0, + 1918.0, + 1143.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 1877.0, + 1408.0, + 1877.0, + 1408.0, + 1918.0, + 1310.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1910.0, + 588.0, + 1910.0, + 588.0, + 1947.0, + 293.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 1910.0, + 724.0, + 1910.0, + 724.0, + 1947.0, + 608.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1910.0, + 1080.0, + 1910.0, + 1080.0, + 1947.0, + 745.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1910.0, + 1408.0, + 1910.0, + 1408.0, + 1947.0, + 1105.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1936.0, + 404.0, + 1936.0, + 404.0, + 1986.0, + 289.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1936.0, + 1136.0, + 1936.0, + 1136.0, + 1986.0, + 1054.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1936.0, + 1410.0, + 1936.0, + 1410.0, + 1986.0, + 1374.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1970.0, + 687.0, + 1970.0, + 687.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 1970.0, + 862.0, + 1970.0, + 862.0, + 2008.0, + 708.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1970.0, + 1004.0, + 1970.0, + 1004.0, + 2008.0, + 895.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 1970.0, + 1187.0, + 1970.0, + 1187.0, + 2008.0, + 1050.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 1970.0, + 1339.0, + 1970.0, + 1339.0, + 2008.0, + 1208.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1970.0, + 1407.0, + 1970.0, + 1407.0, + 2008.0, + 1355.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2003.0, + 296.0, + 2003.0, + 296.0, + 2035.0, + 293.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 2003.0, + 653.0, + 2003.0, + 653.0, + 2035.0, + 406.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1224.0, + 702.0, + 1224.0, + 702.0, + 1256.0, + 396.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 1224.0, + 1338.0, + 1224.0, + 1338.0, + 1256.0, + 1304.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1257.0, + 447.0, + 1257.0, + 447.0, + 1286.0, + 400.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1257.0, + 738.0, + 1257.0, + 738.0, + 1289.0, + 455.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1253.0, + 1339.0, + 1253.0, + 1339.0, + 1289.0, + 1301.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1290.0, + 446.0, + 1290.0, + 446.0, + 1319.0, + 400.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 1288.0, + 632.0, + 1288.0, + 632.0, + 1320.0, + 456.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1287.0, + 1339.0, + 1287.0, + 1339.0, + 1323.0, + 1301.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1322.0, + 447.0, + 1322.0, + 447.0, + 1350.0, + 400.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1321.0, + 750.0, + 1321.0, + 750.0, + 1352.0, + 455.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1319.0, + 1340.0, + 1319.0, + 1340.0, + 1354.0, + 1303.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1356.0, + 452.0, + 1356.0, + 452.0, + 1383.0, + 400.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1353.0, + 846.0, + 1353.0, + 846.0, + 1385.0, + 455.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1351.0, + 1340.0, + 1351.0, + 1340.0, + 1388.0, + 1303.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1388.0, + 798.0, + 1388.0, + 798.0, + 1416.0, + 399.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 1385.0, + 1339.0, + 1385.0, + 1339.0, + 1417.0, + 1304.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1609.0, + 832.0, + 1609.0, + 832.0, + 1652.0, + 293.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1609.0, + 1401.0, + 1609.0, + 1401.0, + 1652.0, + 856.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1646.0, + 497.0, + 1646.0, + 497.0, + 1688.0, + 294.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1646.0, + 676.0, + 1646.0, + 676.0, + 1688.0, + 634.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1559.0, + 1194.0, + 1559.0, + 1194.0, + 1609.0, + 292.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1037.0, + 547.0, + 1037.0, + 547.0, + 1070.0, + 400.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1036.0, + 1339.0, + 1036.0, + 1339.0, + 1070.0, + 1302.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1071.0, + 446.0, + 1071.0, + 446.0, + 1097.0, + 402.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1070.0, + 672.0, + 1070.0, + 672.0, + 1102.0, + 452.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1066.0, + 1341.0, + 1066.0, + 1341.0, + 1104.0, + 1301.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1105.0, + 446.0, + 1105.0, + 446.0, + 1130.0, + 402.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 1105.0, + 647.0, + 1105.0, + 647.0, + 1134.0, + 454.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1100.0, + 1341.0, + 1100.0, + 1341.0, + 1134.0, + 1302.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1133.0, + 704.0, + 1133.0, + 704.0, + 1168.0, + 400.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1134.0, + 1341.0, + 1134.0, + 1341.0, + 1167.0, + 1302.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 498.0, + 668.0, + 498.0, + 668.0, + 527.0, + 399.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 497.0, + 1339.0, + 497.0, + 1339.0, + 527.0, + 1306.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 532.0, + 449.0, + 532.0, + 449.0, + 558.0, + 400.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 529.0, + 609.0, + 529.0, + 609.0, + 562.0, + 451.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 529.0, + 1341.0, + 529.0, + 1341.0, + 561.0, + 1303.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 563.0, + 644.0, + 563.0, + 644.0, + 592.0, + 399.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 561.0, + 1341.0, + 561.0, + 1341.0, + 594.0, + 1303.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 595.0, + 775.0, + 595.0, + 775.0, + 626.0, + 397.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 594.0, + 1341.0, + 594.0, + 1341.0, + 626.0, + 1301.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 856.0, + 882.0, + 856.0, + 882.0, + 887.0, + 397.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 856.0, + 1340.0, + 856.0, + 1340.0, + 888.0, + 1302.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 886.0, + 967.0, + 886.0, + 967.0, + 923.0, + 394.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 890.0, + 1340.0, + 890.0, + 1340.0, + 921.0, + 1302.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 946.0, + 1341.0, + 946.0, + 1341.0, + 983.0, + 1302.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 650.0, + 798.0, + 650.0, + 798.0, + 684.0, + 357.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 650.0, + 1341.0, + 650.0, + 1341.0, + 689.0, + 1303.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 709.0, + 576.0, + 709.0, + 576.0, + 744.0, + 360.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 706.0, + 1342.0, + 706.0, + 1342.0, + 751.0, + 1301.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 765.0, + 694.0, + 765.0, + 694.0, + 799.0, + 358.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 763.0, + 1341.0, + 763.0, + 1341.0, + 801.0, + 1302.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 948.0, + 675.0, + 948.0, + 675.0, + 976.0, + 360.0, + 976.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1190, + 1406, + 1190, + 1406, + 1320, + 297, + 1320 + ], + "score": 0.975 + }, + { + "category_id": 8, + "poly": [ + 296, + 1500, + 1455, + 1500, + 1455, + 2037, + 296, + 2037 + ], + "score": 0.967 + }, + { + "category_id": 3, + "poly": [ + 482, + 223, + 1218, + 223, + 1218, + 745, + 482, + 745 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 297, + 1372, + 1403, + 1372, + 1403, + 1438, + 297, + 1438 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 485, + 1108, + 1212, + 1108, + 1212, + 1177, + 485, + 1177 + ], + "score": 0.942 + }, + { + "category_id": 4, + "poly": [ + 295, + 775, + 1407, + 775, + 1407, + 870, + 295, + 870 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 294, + 1033, + 1399, + 1033, + 1399, + 1102, + 294, + 1102 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 296, + 986, + 975, + 986, + 975, + 1021, + 296, + 1021 + ], + "score": 0.921 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 857, + 75, + 857, + 105, + 297, + 105 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 300, + 927, + 637, + 927, + 637, + 959, + 300, + 959 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1125, + 1401, + 1125, + 1401, + 1157, + 1353, + 1157 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 864, + 2087, + 864, + 2113, + 835, + 2113 + ], + "score": 0.855 + }, + { + "category_id": 13, + "poly": [ + 842, + 1256, + 966, + 1256, + 966, + 1288, + 842, + 1288 + ], + "score": 0.95, + "latex": "M ^ { * } ( \\cdot | x , a )" + }, + { + "category_id": 14, + "poly": [ + 489, + 1105, + 1207, + 1105, + 1207, + 1176, + 489, + 1176 + ], + "score": 0.93, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ," + }, + { + "category_id": 13, + "poly": [ + 416, + 1285, + 540, + 1285, + 540, + 1319, + 416, + 1319 + ], + "score": 0.93, + "latex": "M ^ { * } ( \\cdot | x , a )" + }, + { + "category_id": 13, + "poly": [ + 657, + 1032, + 791, + 1032, + 791, + 1073, + 657, + 1073 + ], + "score": 0.93, + "latex": "\\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M )" + }, + { + "category_id": 14, + "poly": [ + 286, + 1499, + 1389, + 1499, + 1389, + 2044, + 286, + 2044 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } { \\rho _ { \\mu \\mu ^ { * } } ^ { * } } & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\omega ( x , \\omega ) \\int _ { x } d x ^ { * } ( x ^ { * } | x , a ) \\left( - \\log { \\mathcal { A } ( x ^ { * } | x , a ) } \\right) \\mathrm { d } x ^ { * } \\mathrm { d } \\alpha \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { \\nu \\mu } ^ { 2 } ( z , \\omega ) } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\int _ { x } \\underline { { \\mathcal { A } } } ^ { \\mathcal { A } ^ { * } } ( x ^ { * } | x , a ) \\left( - \\log { \\mathcal { A } ( x ^ { * } | x , a ) } \\right) \\mathrm { d } x ^ { * } \\mathrm { d } \\alpha \\mathrm { d } x - \\frac { \\alpha } { 2 } \\| \\rho _ { \\mu \\nu } ^ { 3 } d \\cdot ( x ^ { * } ) \\| _ { 2 } ^ { 2 } } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\frac { 2 } { \\alpha } \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) \\| \\phi ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\mu \\nu } ^ { 3 } d \\cdot ( x ^ { * } ) \\| _ { 2 } ^ { 2 } } \\\\ & { = \\operatorname* { s u g m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\alpha \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( z , \\omega ) g ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\nu \\mu ^ { * } } ^ { 3 } ( z , \\nu ) \\| _ { 2 } ^ { 2 } } \\\\ & = \\operatorname* { m a x } _ { \\hat { \\rho } _ { \\nu \\mu ^ { * } } ^ { 0 } , } \\alpha \\int _ { x , z } \\rho _ { \\mu \\nu } ^ { * } ( x , a ) g ( x , a ) \\mathrm { d } \\alpha \\mathrm { d } x - \\| \\rho _ { \\mu \\nu } ^ { 3 } ( z , \\omega ) \\| _ { 2 } ^ { 2 } - \\frac \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 375, + 1192, + 702, + 1192, + 702, + 1226, + 375, + 1226 + ], + "score": 0.9, + "latex": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )" + }, + { + "category_id": 13, + "poly": [ + 863, + 1224, + 987, + 1224, + 987, + 1257, + 863, + 1257 + ], + "score": 0.9, + "latex": "M ^ { * } ( \\cdot | x , a )" + }, + { + "category_id": 13, + "poly": [ + 1279, + 1194, + 1403, + 1194, + 1403, + 1226, + 1279, + 1226 + ], + "score": 0.89, + "latex": "M ^ { * } ( \\cdot | x , a )" + }, + { + "category_id": 13, + "poly": [ + 1076, + 1225, + 1401, + 1225, + 1401, + 1257, + 1076, + 1257 + ], + "score": 0.89, + "latex": "D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) )" + }, + { + "category_id": 13, + "poly": [ + 713, + 454, + 752, + 454, + 752, + 488, + 713, + 488 + ], + "score": 0.88, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 599, + 1288, + 644, + 1288, + 644, + 1316, + 599, + 1316 + ], + "score": 0.87, + "latex": "\\alpha _ { M }" + }, + { + "category_id": 13, + "poly": [ + 1368, + 1377, + 1400, + 1377, + 1400, + 1408, + 1368, + 1408 + ], + "score": 0.86, + "latex": "\\beta ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 429, + 809, + 448, + 809, + 448, + 840, + 429, + 840 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1368, + 1036, + 1400, + 1036, + 1400, + 1071, + 1368, + 1071 + ], + "score": 0.84, + "latex": "\\bar { \\beta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1121, + 777, + 1142, + 777, + 1142, + 809, + 1121, + 809 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 695, + 1377, + 727, + 1377, + 727, + 1403, + 695, + 1403 + ], + "score": 0.83, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1143, + 1258, + 1267, + 1258, + 1267, + 1288, + 1143, + 1288 + ], + "score": 0.83, + "latex": "H _ { M ^ { \\ast } } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 1103, + 635, + 1139, + 635, + 1139, + 669, + 1103, + 669 + ], + "score": 0.82, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 1019, + 1256, + 1130, + 1256, + 1130, + 1288, + 1019, + 1288 + ], + "score": 0.82, + "latex": "M ( \\cdot | x , a )" + }, + { + "category_id": 13, + "poly": [ + 885, + 588, + 1073, + 588, + 1073, + 621, + 885, + 621 + ], + "score": 0.79, + "latex": "\\mathrm { m i n } _ { M \\in \\mathcal { M } } L ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M )" + }, + { + "category_id": 13, + "poly": [ + 559, + 1257, + 612, + 1257, + 612, + 1285, + 559, + 1285 + ], + "score": 0.79, + "latex": "( K L )" + }, + { + "category_id": 13, + "poly": [ + 662, + 331, + 701, + 331, + 701, + 365, + 662, + 365 + ], + "score": 0.76, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 990, + 416, + 1015, + 416, + 1015, + 443, + 990, + 443 + ], + "score": 0.75, + "latex": "\\bar { \\beta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 473, + 1227, + 598, + 1227, + 598, + 1257, + 473, + 1257 + ], + "score": 0.74, + "latex": "H _ { M ^ { \\ast } } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 588, + 1039, + 621, + 1039, + 621, + 1067, + 588, + 1067 + ], + "score": 0.73, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 984, + 1195, + 1039, + 1195, + 1039, + 1224, + 984, + 1224 + ], + "score": 0.72, + "latex": "( K L )" + }, + { + "category_id": 13, + "poly": [ + 781, + 530, + 819, + 530, + 819, + 562, + 781, + 562 + ], + "score": 0.72, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 995, + 1292, + 1016, + 1292, + 1016, + 1313, + 995, + 1313 + ], + "score": 0.72, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 636, + 426, + 671, + 426, + 671, + 457, + 636, + 457 + ], + "score": 0.71, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 873, + 813, + 894, + 813, + 894, + 835, + 873, + 835 + ], + "score": 0.69, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 348, + 1226, + 459, + 1226, + 459, + 1257, + 348, + 1257 + ], + "score": 0.57, + "latex": "M ( \\cdot | x , a )" + }, + { + "category_id": 13, + "poly": [ + 835, + 559, + 1013, + 559, + 1013, + 590, + 835, + 590 + ], + "score": 0.52, + "latex": "\\mathrm { m a x } _ { \\beta \\in \\Pi } L ( \\rho _ { M ^ { * } } ^ { \\beta } , M )" + }, + { + "category_id": 13, + "poly": [ + 924, + 382, + 938, + 382, + 938, + 406, + 924, + 406 + ], + "score": 0.51, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 656, + 389, + 670, + 389, + 670, + 402, + 656, + 402 + ], + "score": 0.39, + "latex": "\\kappa" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 233.0, + 1091.0, + 233.0, + 1091.0, + 269.0, + 820.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 308.0, + 858.0, + 308.0, + 858.0, + 334.0, + 579.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 334.0, + 661.0, + 334.0, + 661.0, + 368.0, + 555.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 334.0, + 883.0, + 334.0, + 883.0, + 368.0, + 702.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 347.0, + 1150.0, + 347.0, + 1150.0, + 376.0, + 987.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 376.0, + 655.0, + 376.0, + 655.0, + 409.0, + 485.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 376.0, + 677.0, + 376.0, + 677.0, + 409.0, + 671.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 381.0, + 923.0, + 381.0, + 923.0, + 406.0, + 759.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 371.0, + 1197.0, + 371.0, + 1197.0, + 397.0, + 988.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 391.0, + 1203.0, + 391.0, + 1203.0, + 425.0, + 985.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 420.0, + 635.0, + 420.0, + 635.0, + 467.0, + 569.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 420.0, + 868.0, + 420.0, + 868.0, + 467.0, + 672.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 412.0, + 989.0, + 412.0, + 989.0, + 445.0, + 986.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 412.0, + 1020.0, + 412.0, + 1020.0, + 445.0, + 1016.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 452.0, + 712.0, + 452.0, + 712.0, + 492.0, + 582.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 452.0, + 853.0, + 452.0, + 853.0, + 492.0, + 753.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 529.0, + 780.0, + 529.0, + 780.0, + 567.0, + 740.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 529.0, + 1168.0, + 529.0, + 1168.0, + 567.0, + 820.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 560.0, + 834.0, + 560.0, + 834.0, + 590.0, + 739.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 560.0, + 1171.0, + 560.0, + 1171.0, + 590.0, + 1014.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 589.0, + 884.0, + 589.0, + 884.0, + 622.0, + 736.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 589.0, + 1171.0, + 589.0, + 1171.0, + 622.0, + 1074.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 620.0, + 1102.0, + 620.0, + 1102.0, + 683.0, + 978.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 620.0, + 1203.0, + 620.0, + 1203.0, + 683.0, + 1140.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 665.0, + 1208.0, + 665.0, + 1208.0, + 691.0, + 988.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 711.0, + 1085.0, + 711.0, + 1085.0, + 740.0, + 826.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 453.5, + 749.0, + 453.5, + 749.0, + 469.5, + 721.0, + 469.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 796.0, + 526.0, + 809.0, + 526.0, + 809.0, + 548.0, + 796.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 773.0, + 1120.0, + 773.0, + 1120.0, + 812.0, + 294.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 773.0, + 1404.0, + 773.0, + 1404.0, + 812.0, + 1143.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 805.0, + 428.0, + 805.0, + 428.0, + 842.0, + 294.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 805.0, + 872.0, + 805.0, + 872.0, + 842.0, + 449.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 805.0, + 1404.0, + 805.0, + 1404.0, + 842.0, + 895.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 836.0, + 427.0, + 836.0, + 427.0, + 871.0, + 293.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 922.0, + 641.0, + 922.0, + 641.0, + 964.0, + 294.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1192.0, + 374.0, + 1192.0, + 374.0, + 1229.0, + 295.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1192.0, + 983.0, + 1192.0, + 983.0, + 1229.0, + 703.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 1192.0, + 1278.0, + 1192.0, + 1278.0, + 1229.0, + 1040.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1192.0, + 1407.0, + 1192.0, + 1407.0, + 1229.0, + 1404.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1221.0, + 347.0, + 1221.0, + 347.0, + 1260.0, + 292.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1221.0, + 472.0, + 1221.0, + 472.0, + 1260.0, + 460.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 1221.0, + 862.0, + 1221.0, + 862.0, + 1260.0, + 599.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1221.0, + 1075.0, + 1221.0, + 1075.0, + 1260.0, + 988.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1221.0, + 1406.0, + 1221.0, + 1406.0, + 1260.0, + 1402.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1253.0, + 558.0, + 1253.0, + 558.0, + 1289.0, + 292.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 1253.0, + 841.0, + 1253.0, + 841.0, + 1289.0, + 613.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 1253.0, + 1018.0, + 1253.0, + 1018.0, + 1289.0, + 967.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1253.0, + 1142.0, + 1253.0, + 1142.0, + 1289.0, + 1131.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1253.0, + 1406.0, + 1253.0, + 1406.0, + 1289.0, + 1268.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1286.0, + 415.0, + 1286.0, + 415.0, + 1320.0, + 296.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1286.0, + 598.0, + 1286.0, + 598.0, + 1320.0, + 541.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1286.0, + 994.0, + 1286.0, + 994.0, + 1320.0, + 645.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1286.0, + 1396.0, + 1286.0, + 1396.0, + 1320.0, + 1017.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1370.0, + 694.0, + 1370.0, + 694.0, + 1412.0, + 292.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1370.0, + 1367.0, + 1370.0, + 1367.0, + 1412.0, + 728.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1402.0, + 403.0, + 1402.0, + 403.0, + 1440.0, + 292.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1031.0, + 587.0, + 1031.0, + 587.0, + 1079.0, + 292.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 1031.0, + 656.0, + 1031.0, + 656.0, + 1079.0, + 622.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 1031.0, + 1367.0, + 1031.0, + 1367.0, + 1079.0, + 792.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1031.0, + 1405.0, + 1031.0, + 1405.0, + 1079.0, + 1401.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1062.0, + 404.0, + 1062.0, + 404.0, + 1110.0, + 291.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 986.0, + 976.0, + 986.0, + 976.0, + 1022.0, + 295.0, + 1022.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1746, + 1407, + 1746, + 1407, + 2036, + 296, + 2036 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 296, + 1294, + 1406, + 1294, + 1406, + 1619, + 296, + 1619 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 296, + 224, + 1405, + 224, + 1405, + 401, + 296, + 401 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1051, + 1404, + 1051, + 1404, + 1154, + 298, + 1154 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 812, + 1404, + 812, + 1404, + 907, + 298, + 907 + ], + "score": 0.973 + }, + { + "category_id": 8, + "poly": [ + 368, + 433, + 1329, + 433, + 1329, + 732, + 368, + 732 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 487, + 945, + 1211, + 945, + 1211, + 1015, + 487, + 1015 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 295, + 764, + 975, + 764, + 975, + 799, + 295, + 799 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 105, + 297, + 105 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 300, + 1680, + 566, + 1680, + 566, + 1712, + 300, + 1712 + ], + "score": 0.916 + }, + { + "category_id": 9, + "poly": [ + 1354, + 963, + 1402, + 963, + 1402, + 995, + 1354, + 995 + ], + "score": 0.899 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2113, + 835, + 2113 + ], + "score": 0.858 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1169, + 1402, + 1169, + 1402, + 1198, + 1374, + 1198 + ], + "score": 0.791 + }, + { + "category_id": 14, + "poly": [ + 368, + 431, + 1331, + 431, + 1331, + 733, + 368, + 733 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\rho _ { M ^ { * } } ^ { \\beta ^ { * } } ( x , a ) = \\frac { g ( x , a ) } { \\alpha _ { M } } } \\\\ & { \\qquad = \\frac { \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\log \\frac { M ^ { * } ( x ^ { \\prime } | x , a ) } { M ( x ^ { \\prime } | x , a ) } \\mathrm { d } x - \\int _ { \\mathcal { X } } M ^ { * } ( x ^ { \\prime } | x , a ) \\log M ^ { * } ( x ^ { \\prime } | x , a ) \\mathrm { d } x } { \\alpha _ { M } } } \\\\ & { \\qquad = \\frac { D _ { K L } \\left( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) \\right) + H _ { M ^ { * } } ( x , a ) } { \\alpha _ { M } } } \\\\ & { \\qquad \\propto \\Big ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) \\Big ) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 979, + 1329, + 1114, + 1329, + 1114, + 1361, + 979, + 1361 + ], + "score": 0.95, + "latex": "g ( x , a ) / \\alpha _ { M }" + }, + { + "category_id": 13, + "poly": [ + 881, + 1810, + 1033, + 1810, + 1033, + 1844, + 881, + 1844 + ], + "score": 0.93, + "latex": "\\rho _ { M ^ { * } } ^ { \\mu } ( x , a , x ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 441, + 844, + 577, + 844, + 577, + 878, + 441, + 878 + ], + "score": 0.93, + "latex": "g ( x , a ) / \\alpha _ { M }" + }, + { + "category_id": 13, + "poly": [ + 872, + 1298, + 952, + 1298, + 952, + 1331, + 872, + 1331 + ], + "score": 0.92, + "latex": "g ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 1030, + 262, + 1083, + 262, + 1083, + 304, + 1030, + 304 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 1225, + 1555, + 1399, + 1555, + 1399, + 1588, + 1225, + 1588 + ], + "score": 0.92, + "latex": "D _ { K L } ( M ^ { * } , M ) )" + }, + { + "category_id": 13, + "poly": [ + 900, + 1516, + 953, + 1516, + 953, + 1559, + 900, + 1559 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 981, + 1967, + 1149, + 1967, + 1149, + 2007, + 981, + 2007 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) > 0" + }, + { + "category_id": 13, + "poly": [ + 298, + 1971, + 455, + 1971, + 455, + 2007, + 298, + 2007 + ], + "score": 0.91, + "latex": "\\rho _ { M } ^ { \\kappa } ( x , a ) > 0" + }, + { + "category_id": 13, + "poly": [ + 835, + 1933, + 1255, + 1933, + 1255, + 1967, + 835, + 1967 + ], + "score": 0.91, + "latex": "M ( x ^ { \\prime } | x , a ) = \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) / \\rho _ { M } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 584, + 1972, + 952, + 1972, + 952, + 2007, + 584, + 2007 + ], + "score": 0.91, + "latex": "\\forall x \\in \\mathcal { X } , \\forall a \\in A , \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) > 0" + }, + { + "category_id": 14, + "poly": [ + 490, + 943, + 1209, + 943, + 1209, + 1014, + 490, + 1014 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } ( x , a ) = \\frac { 1 } { \\alpha _ { M } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) ," + }, + { + "category_id": 13, + "poly": [ + 454, + 1812, + 516, + 1812, + 516, + 1841, + 454, + 1841 + ], + "score": 0.91, + "latex": "\\mathcal { D } _ { \\mathrm { r e a l } }" + }, + { + "category_id": 13, + "poly": [ + 1219, + 1359, + 1402, + 1359, + 1402, + 1392, + 1219, + 1392 + ], + "score": 0.91, + "latex": "M ^ { \\ast } ( x _ { 1 } | x , a ) \\ <" + }, + { + "category_id": 13, + "poly": [ + 297, + 261, + 610, + 261, + 610, + 303, + 297, + 303 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\int _ { \\mathcal { X } } \\int _ { A } \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) \\mathrm { d } a \\mathrm { d } x = 1 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 560, + 1751, + 622, + 1751, + 622, + 1781, + 560, + 1781 + ], + "score": 0.91, + "latex": "D _ { K L }" + }, + { + "category_id": 13, + "poly": [ + 814, + 224, + 868, + 224, + 868, + 266, + 814, + 266 + ], + "score": 0.9, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1422, + 598, + 1422, + 598, + 1460, + 298, + 1460 + ], + "score": 0.89, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta } ( x _ { 1 } , a ) > \\rho _ { M ^ { * } } ^ { \\beta } ( x _ { 2 } , a )" + }, + { + "category_id": 13, + "poly": [ + 1055, + 1090, + 1087, + 1090, + 1087, + 1123, + 1055, + 1123 + ], + "score": 0.89, + "latex": "\\bar { \\beta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 633, + 1054, + 1351, + 1054, + 1351, + 1093, + 633, + 1093 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\alpha _ { M } = \\int _ { \\mathcal { X } } \\int _ { \\mathcal { A } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | \\underline { { x } } , a ) ) + H _ { M ^ { * } } ( x , a ) ) \\mathrm { d } x \\mathrm { d } a } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 374, + 873, + 405, + 873, + 405, + 907, + 374, + 907 + ], + "score": 0.89, + "latex": "\\bar { \\beta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 326, + 304, + 823, + 304, + 823, + 339, + 326, + 339 + ], + "score": 0.88, + "latex": "0 ~ \\leq ~ \\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) ~ \\leq ~ c , \\forall a ~ \\in ~ { \\mathcal { A } } , \\forall x ~ \\in ~ \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 700, + 1781, + 744, + 1781, + 744, + 1809, + 700, + 1809 + ], + "score": 0.88, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 531, + 1123, + 563, + 1123, + 563, + 1153, + 531, + 1153 + ], + "score": 0.88, + "latex": "\\beta ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 866, + 1586, + 926, + 1586, + 926, + 1617, + 866, + 1617 + ], + "score": 0.88, + "latex": "H _ { M ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 633, + 815, + 677, + 815, + 677, + 842, + 633, + 842 + ], + "score": 0.88, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 866, + 771, + 965, + 771, + 965, + 797, + 866, + 797 + ], + "score": 0.87, + "latex": "\\alpha = \\alpha _ { M }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1389, + 673, + 1389, + 673, + 1421, + 298, + 1421 + ], + "score": 0.87, + "latex": "M ^ { * } ( x _ { 2 } | x , a ) , \\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 370, + 1061, + 414, + 1061, + 414, + 1086, + 370, + 1086 + ], + "score": 0.86, + "latex": "\\alpha _ { M }" + }, + { + "category_id": 13, + "poly": [ + 533, + 1360, + 578, + 1360, + 578, + 1387, + 533, + 1387 + ], + "score": 0.86, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 596, + 1487, + 639, + 1487, + 639, + 1515, + 596, + 1515 + ], + "score": 0.86, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1175, + 1299, + 1219, + 1299, + 1219, + 1326, + 1175, + 1326 + ], + "score": 0.85, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1008, + 815, + 1030, + 815, + 1030, + 845, + 1008, + 845 + ], + "score": 0.84, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 1316, + 308, + 1403, + 308, + 1403, + 339, + 1316, + 339 + ], + "score": 0.84, + "latex": "\\alpha _ { M } =" + }, + { + "category_id": 13, + "poly": [ + 461, + 1363, + 493, + 1363, + 493, + 1389, + 461, + 1389 + ], + "score": 0.83, + "latex": "x _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1113, + 1365, + 1143, + 1365, + 1143, + 1389, + 1113, + 1389 + ], + "score": 0.82, + "latex": "x _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 431, + 1329, + 452, + 1329, + 452, + 1360, + 431, + 1360 + ], + "score": 0.82, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 1336, + 1908, + 1356, + 1908, + 1356, + 1934, + 1336, + 1934 + ], + "score": 0.82, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 301, + 340, + 940, + 340, + 940, + 374, + 301, + 374 + ], + "score": 0.81, + "latex": "\\begin{array} { r } { \\int _ { \\mathcal { X } } \\int _ { \\mathcal { A } } ( D _ { K L } ( M ^ { * } ( \\cdot | x , a ) , M ( \\cdot | x , a ) ) + H _ { M ^ { * } } ( x , a ) ) \\mathrm { d } x \\mathrm { d } a } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1345, + 1934, + 1377, + 1934, + 1377, + 1961, + 1345, + 1961 + ], + "score": 0.8, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 476, + 2007, + 496, + 2007, + 496, + 2035, + 476, + 2035 + ], + "score": 0.8, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 1301, + 340, + 1334, + 340, + 1334, + 367, + 1301, + 367 + ], + "score": 0.79, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1048, + 1390, + 1069, + 1390, + 1069, + 1421, + 1048, + 1421 + ], + "score": 0.79, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 669, + 846, + 702, + 846, + 702, + 873, + 669, + 873 + ], + "score": 0.78, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1203, + 1908, + 1222, + 1908, + 1222, + 1930, + 1203, + 1930 + ], + "score": 0.73, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 298, + 2010, + 317, + 2010, + 317, + 2030, + 298, + 2030 + ], + "score": 0.72, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 1326, + 275, + 1341, + 275, + 1341, + 297, + 1326, + 297 + ], + "score": 0.6, + "latex": "c" + }, + { + "category_id": 13, + "poly": [ + 1384, + 1498, + 1401, + 1498, + 1401, + 1514, + 1384, + 1514 + ], + "score": 0.27, + "latex": "a" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1674.0, + 572.0, + 1674.0, + 572.0, + 1719.0, + 293.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1381.0, + 1174.0, + 1401.0, + 1174.0, + 1401.0, + 1196.0, + 1381.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1747.0, + 559.0, + 1747.0, + 559.0, + 1785.0, + 294.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 1747.0, + 1405.0, + 1747.0, + 1405.0, + 1785.0, + 623.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1778.0, + 699.0, + 1778.0, + 699.0, + 1815.0, + 294.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1778.0, + 1407.0, + 1778.0, + 1407.0, + 1815.0, + 745.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1806.0, + 453.0, + 1806.0, + 453.0, + 1850.0, + 291.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1806.0, + 880.0, + 1806.0, + 880.0, + 1850.0, + 517.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1806.0, + 1408.0, + 1806.0, + 1408.0, + 1850.0, + 1034.0, + 1850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1841.0, + 1405.0, + 1841.0, + 1405.0, + 1875.0, + 294.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1868.0, + 1407.0, + 1868.0, + 1407.0, + 1907.0, + 293.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1901.0, + 1202.0, + 1901.0, + 1202.0, + 1937.0, + 291.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 1901.0, + 1335.0, + 1901.0, + 1335.0, + 1937.0, + 1223.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1357.0, + 1901.0, + 1407.0, + 1901.0, + 1407.0, + 1937.0, + 1357.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1929.0, + 834.0, + 1929.0, + 834.0, + 1970.0, + 291.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 1929.0, + 1344.0, + 1929.0, + 1344.0, + 1970.0, + 1256.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1929.0, + 1409.0, + 1929.0, + 1409.0, + 1970.0, + 1378.0, + 1970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1967.0, + 297.0, + 1967.0, + 297.0, + 2012.0, + 293.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 1967.0, + 583.0, + 1967.0, + 583.0, + 2012.0, + 456.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1967.0, + 980.0, + 1967.0, + 980.0, + 2012.0, + 953.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 1967.0, + 1407.0, + 1967.0, + 1407.0, + 2012.0, + 1150.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2002.0, + 297.0, + 2002.0, + 297.0, + 2036.0, + 294.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 2002.0, + 475.0, + 2002.0, + 475.0, + 2036.0, + 318.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 2002.0, + 822.0, + 2002.0, + 822.0, + 2036.0, + 497.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1295.0, + 871.0, + 1295.0, + 871.0, + 1332.0, + 294.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1295.0, + 1174.0, + 1295.0, + 1174.0, + 1332.0, + 953.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 1295.0, + 1406.0, + 1295.0, + 1406.0, + 1332.0, + 1220.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1325.0, + 430.0, + 1325.0, + 430.0, + 1364.0, + 291.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1325.0, + 978.0, + 1325.0, + 978.0, + 1364.0, + 453.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 1325.0, + 1406.0, + 1325.0, + 1406.0, + 1364.0, + 1115.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1359.0, + 460.0, + 1359.0, + 460.0, + 1393.0, + 294.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1359.0, + 532.0, + 1359.0, + 532.0, + 1393.0, + 494.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 1359.0, + 1112.0, + 1359.0, + 1112.0, + 1393.0, + 579.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1359.0, + 1218.0, + 1359.0, + 1218.0, + 1393.0, + 1144.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1386.0, + 1047.0, + 1386.0, + 1047.0, + 1423.0, + 674.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1386.0, + 1404.0, + 1386.0, + 1404.0, + 1423.0, + 1070.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1419.0, + 297.0, + 1419.0, + 297.0, + 1465.0, + 293.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 1419.0, + 1408.0, + 1419.0, + 1408.0, + 1465.0, + 599.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1455.0, + 1406.0, + 1455.0, + 1406.0, + 1490.0, + 293.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1485.0, + 595.0, + 1485.0, + 595.0, + 1523.0, + 293.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 1485.0, + 1383.0, + 1485.0, + 1383.0, + 1523.0, + 640.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1485.0, + 1407.0, + 1485.0, + 1407.0, + 1523.0, + 1402.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1517.0, + 899.0, + 1517.0, + 899.0, + 1565.0, + 291.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1517.0, + 1408.0, + 1517.0, + 1408.0, + 1565.0, + 954.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1554.0, + 1224.0, + 1554.0, + 1224.0, + 1588.0, + 295.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1554.0, + 1406.0, + 1554.0, + 1406.0, + 1588.0, + 1400.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1584.0, + 865.0, + 1584.0, + 865.0, + 1621.0, + 294.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1584.0, + 1043.0, + 1584.0, + 1043.0, + 1621.0, + 927.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 211.0, + 813.0, + 211.0, + 813.0, + 280.0, + 284.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 211.0, + 1416.0, + 211.0, + 1416.0, + 280.0, + 869.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 263.0, + 1029.0, + 263.0, + 1029.0, + 309.0, + 611.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 263.0, + 1325.0, + 263.0, + 1325.0, + 309.0, + 1084.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 263.0, + 1408.0, + 263.0, + 1408.0, + 309.0, + 1342.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 303.0, + 325.0, + 303.0, + 325.0, + 345.0, + 291.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 303.0, + 1315.0, + 303.0, + 1315.0, + 345.0, + 824.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 332.0, + 300.0, + 332.0, + 300.0, + 378.0, + 293.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 332.0, + 1300.0, + 332.0, + 1300.0, + 378.0, + 941.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 332.0, + 1408.0, + 332.0, + 1408.0, + 378.0, + 1335.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 369.0, + 397.0, + 369.0, + 397.0, + 402.0, + 293.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1052.0, + 369.0, + 1052.0, + 369.0, + 1092.0, + 291.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1052.0, + 632.0, + 1052.0, + 632.0, + 1092.0, + 415.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 1052.0, + 1406.0, + 1052.0, + 1406.0, + 1092.0, + 1352.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1089.0, + 1054.0, + 1089.0, + 1054.0, + 1128.0, + 294.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1089.0, + 1406.0, + 1089.0, + 1406.0, + 1128.0, + 1088.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1120.0, + 530.0, + 1120.0, + 530.0, + 1156.0, + 294.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 1120.0, + 743.0, + 1120.0, + 743.0, + 1156.0, + 564.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 808.0, + 632.0, + 808.0, + 632.0, + 852.0, + 292.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 808.0, + 1007.0, + 808.0, + 1007.0, + 852.0, + 678.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 808.0, + 1405.0, + 808.0, + 1405.0, + 852.0, + 1031.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 844.0, + 440.0, + 844.0, + 440.0, + 880.0, + 293.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 844.0, + 668.0, + 844.0, + 668.0, + 880.0, + 578.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 844.0, + 1405.0, + 844.0, + 1405.0, + 880.0, + 703.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 873.0, + 373.0, + 873.0, + 373.0, + 909.0, + 293.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 873.0, + 514.0, + 873.0, + 514.0, + 909.0, + 406.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 759.0, + 865.0, + 759.0, + 865.0, + 807.0, + 294.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 759.0, + 978.0, + 759.0, + 978.0, + 807.0, + 966.0, + 807.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 1260, + 1410, + 1260, + 1410, + 1419, + 295, + 1419 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 295, + 1008, + 1405, + 1008, + 1405, + 1101, + 295, + 1101 + ], + "score": 0.971 + }, + { + "category_id": 8, + "poly": [ + 678, + 1903, + 1019, + 1903, + 1019, + 1990, + 678, + 1990 + ], + "score": 0.961 + }, + { + "category_id": 8, + "poly": [ + 641, + 1116, + 1058, + 1116, + 1058, + 1193, + 641, + 1193 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 295, + 265, + 1515, + 265, + 1515, + 922, + 295, + 922 + ], + "score": 0.945 + }, + { + "category_id": 0, + "poly": [ + 299, + 1582, + 609, + 1582, + 609, + 1615, + 299, + 1615 + ], + "score": 0.931 + }, + { + "category_id": 2, + "poly": [ + 296, + 73, + 858, + 73, + 858, + 106, + 296, + 106 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 297, + 1203, + 1256, + 1203, + 1256, + 1238, + 297, + 1238 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 297, + 966, + 672, + 966, + 672, + 1002, + 297, + 1002 + ], + "score": 0.918 + }, + { + "category_id": 8, + "poly": [ + 290, + 1440, + 1456, + 1440, + 1456, + 1518, + 290, + 1518 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 296, + 1639, + 906, + 1639, + 906, + 1673, + 296, + 1673 + ], + "score": 0.91 + }, + { + "category_id": 1, + "poly": [ + 292, + 1681, + 1400, + 1681, + 1400, + 1746, + 292, + 1746 + ], + "score": 0.902 + }, + { + "category_id": 1, + "poly": [ + 296, + 2002, + 357, + 2002, + 357, + 2034, + 296, + 2034 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 294, + 1853, + 1360, + 1853, + 1360, + 1890, + 294, + 1890 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1138, + 1402, + 1138, + 1402, + 1169, + 1353, + 1169 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 864, + 2087, + 864, + 2113, + 835, + 2113 + ], + "score": 0.865 + }, + { + "category_id": 1, + "poly": [ + 295, + 1811, + 1051, + 1811, + 1051, + 1848, + 295, + 1848 + ], + "score": 0.855 + }, + { + "category_id": 9, + "poly": [ + 1353, + 918, + 1403, + 918, + 1403, + 948, + 1353, + 948 + ], + "score": 0.847 + }, + { + "category_id": 9, + "poly": [ + 1358, + 1518, + 1403, + 1518, + 1403, + 1546, + 1358, + 1546 + ], + "score": 0.776 + }, + { + "category_id": 8, + "poly": [ + 301, + 1758, + 1390, + 1758, + 1390, + 1799, + 301, + 1799 + ], + "score": 0.682 + }, + { + "category_id": 9, + "poly": [ + 1353, + 496, + 1403, + 496, + 1403, + 525, + 1353, + 525 + ], + "score": 0.495 + }, + { + "category_id": 1, + "poly": [ + 301, + 1758, + 1390, + 1758, + 1390, + 1799, + 301, + 1799 + ], + "score": 0.192 + }, + { + "category_id": 14, + "poly": [ + 642, + 1115, + 1056, + 1115, + 1056, + 1192, + 642, + 1192 + ], + "score": 0.94, + "latex": "D _ { f } ( P \\| Q ) = \\int _ { \\mathcal X } q ( x ) f \\left( \\frac { p ( x ) } { q ( x ) } \\right) \\mathrm { d } x ," + }, + { + "category_id": 14, + "poly": [ + 675, + 1899, + 1020, + 1899, + 1020, + 1990, + 675, + 1990 + ], + "score": 0.94, + "latex": "\\sum _ { i = 1 } ^ { n } { \\frac { 1 } { n } } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } { \\frac { 1 } { n } } a _ { i } \\sum { \\frac { 1 } { n } } b _ { i }" + }, + { + "category_id": 13, + "poly": [ + 434, + 1812, + 670, + 1812, + 670, + 1848, + 434, + 1848 + ], + "score": 0.93, + "latex": "\\pi ( 1 ) , \\pi ( 2 ) , \\ldots , \\pi ( n )" + }, + { + "category_id": 13, + "poly": [ + 944, + 1264, + 1204, + 1264, + 1204, + 1313, + 944, + 1313 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } { \\rho _ { M } ^ { \\kappa } ( x , a ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1380, + 370, + 1380, + 370, + 1423, + 297, + 1423 + ], + "score": 0.92, + "latex": "q f ( { \\underline { { p } } } )" + }, + { + "category_id": 13, + "poly": [ + 845, + 1375, + 896, + 1375, + 896, + 1418, + 845, + 1418 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 626, + 1205, + 769, + 1205, + 769, + 1237, + 626, + 1237 + ], + "score": 0.92, + "latex": "f : \\mathbb { R } _ { + } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 569, + 1264, + 891, + 1264, + 891, + 1314, + 569, + 1314 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\log \\frac { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 579, + 1712, + 800, + 1712, + 800, + 1744, + 579, + 1744 + ], + "score": 0.91, + "latex": "b _ { 1 } \\geq b _ { 2 } \\geq . . . \\geq b _ { n }" + }, + { + "category_id": 13, + "poly": [ + 371, + 967, + 663, + 967, + 663, + 1002, + 371, + 1002 + ], + "score": 0.91, + "latex": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M } \\rho _ { M ^ { \\ast } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 951, + 1854, + 1172, + 1854, + 1172, + 1886, + 951, + 1886 + ], + "score": 0.91, + "latex": "b _ { 1 } \\geq b _ { 2 } \\geq . . . \\geq b _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1267, + 1338, + 1344, + 1338, + 1344, + 1377, + 1267, + 1377 + ], + "score": 0.91, + "latex": "q \\log { \\frac { q } { p } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1713, + 526, + 1713, + 526, + 1744, + 298, + 1744 + ], + "score": 0.91, + "latex": "a _ { 1 } \\geq a _ { 2 } \\geq . . . \\geq a _ { n }" + }, + { + "category_id": 14, + "poly": [ + 310, + 1435, + 1428, + 1435, + 1428, + 1520, + 310, + 1520 + ], + "score": 0.89, + "latex": "\\tilde { \\bar { \\nu } } _ { M ^ { * } } ^ { * } : = \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\Bigg ( \\int _ { \\mathcal { X } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\mathrm { d } x ^ { \\prime } - \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) \\left( f \\left( \\frac { \\rho _ { M } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) - H _ { M ^ { * } } ( x , a ) \\right)" + }, + { + "category_id": 13, + "poly": [ + 669, + 1855, + 896, + 1855, + 896, + 1886, + 669, + 1886 + ], + "score": 0.89, + "latex": "a _ { 1 } \\geq a _ { 2 } \\geq . . . \\geq a _ { n }" + }, + { + "category_id": 14, + "poly": [ + 310, + 1758, + 1392, + 1758, + 1392, + 1798, + 310, + 1798 + ], + "score": 0.89, + "latex": "a _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \\cdot \\cdot \\cdot + a _ { n } b _ { n } \\geq a _ { 1 } b _ { \\pi ( 1 ) } + a _ { 2 } b _ { \\pi ( 2 ) } + \\cdot \\cdot \\cdot + a _ { n } b _ { \\pi ( n ) } \\geq a _ { 1 } b _ { n } + a _ { 2 } b _ { n - 1 } + \\cdot \\cdot \\cdot + a _ { n } b _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 918, + 1815, + 1037, + 1815, + 1037, + 1846, + 918, + 1846 + ], + "score": 0.89, + "latex": "1 , 2 , \\ldots , n" + }, + { + "category_id": 13, + "poly": [ + 1101, + 1314, + 1119, + 1314, + 1119, + 1342, + 1101, + 1342 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 952, + 1314, + 970, + 1314, + 970, + 1342, + 952, + 1342 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 991, + 1011, + 1015, + 1011, + 1015, + 1041, + 991, + 1041 + ], + "score": 0.83, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 298, + 1072, + 316, + 1072, + 316, + 1102, + 298, + 1102 + ], + "score": 0.83, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 915, + 1011, + 939, + 1011, + 939, + 1037, + 915, + 1037 + ], + "score": 0.83, + "latex": "P" + }, + { + "category_id": 14, + "poly": [ + 310, + 265, + 1428, + 265, + 1428, + 940, + 310, + 940 + ], + "score": 0.8, + "latex": "\\begin{array} { r l } { \\frac { d } { d t } } & { ( z , z , n ) = \\frac { 1 } { n + 1 } ( 2 \\frac { n + 1 } { n + 1 } , \\frac { 1 } { n + 1 } ; \\frac { 1 } { n + 2 } , z ) ! \\log ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } ) \\log ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) } \\\\ & { - \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , z , n ) ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) \\log \\log ( \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n ) ) \\mathrm { d } x ^ { \\prime } \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n ) ) \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\mathrm { d } x ^ { \\prime } \\exp ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } + \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } - \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , z , n ) ) \\mathrm { d } y ^ { \\prime } } \\\\ & = \\frac { 1 } { n + 1 } \\int _ { x } d \\gamma \\exp ( z , \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) ) \\exp ( \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } + \\log \\frac { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } { \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , z ) } - \\log \\mathcal { N } _ { \\mathrm { e f f } } ( z , n , \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 478, + 1011, + 496, + 1011, + 496, + 1042, + 478, + 1042 + ], + "score": 0.8, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1201, + 1043, + 1226, + 1043, + 1226, + 1066, + 1201, + 1066 + ], + "score": 0.79, + "latex": "\\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 489, + 1045, + 507, + 1045, + 507, + 1072, + 489, + 1072 + ], + "score": 0.76, + "latex": "q" + }, + { + "category_id": 13, + "poly": [ + 411, + 1046, + 430, + 1046, + 430, + 1072, + 411, + 1072 + ], + "score": 0.72, + "latex": "p" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1579.0, + 614.0, + 1579.0, + 614.0, + 1618.0, + 295.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1259.0, + 568.0, + 1259.0, + 568.0, + 1318.0, + 292.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1259.0, + 943.0, + 1259.0, + 943.0, + 1318.0, + 892.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1259.0, + 1406.0, + 1259.0, + 1406.0, + 1318.0, + 1205.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1306.0, + 951.0, + 1306.0, + 951.0, + 1345.0, + 294.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1306.0, + 1100.0, + 1306.0, + 1100.0, + 1345.0, + 971.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1306.0, + 1408.0, + 1306.0, + 1408.0, + 1345.0, + 1120.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1334.0, + 1266.0, + 1334.0, + 1266.0, + 1376.0, + 294.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 1334.0, + 1407.0, + 1334.0, + 1407.0, + 1376.0, + 1345.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1373.0, + 296.0, + 1373.0, + 296.0, + 1426.0, + 293.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1373.0, + 844.0, + 1373.0, + 844.0, + 1426.0, + 371.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1373.0, + 910.0, + 1373.0, + 910.0, + 1426.0, + 897.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1006.0, + 477.0, + 1006.0, + 477.0, + 1045.0, + 294.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 1006.0, + 914.0, + 1006.0, + 914.0, + 1045.0, + 497.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1006.0, + 990.0, + 1006.0, + 990.0, + 1045.0, + 940.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 1006.0, + 1403.0, + 1006.0, + 1403.0, + 1045.0, + 1016.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1040.0, + 410.0, + 1040.0, + 410.0, + 1074.0, + 293.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 1040.0, + 488.0, + 1040.0, + 488.0, + 1074.0, + 431.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1040.0, + 1200.0, + 1040.0, + 1200.0, + 1074.0, + 508.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 1040.0, + 1403.0, + 1040.0, + 1403.0, + 1074.0, + 1227.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1069.0, + 297.0, + 1069.0, + 297.0, + 1103.0, + 293.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1069.0, + 709.0, + 1069.0, + 709.0, + 1103.0, + 317.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1200.0, + 625.0, + 1200.0, + 625.0, + 1242.0, + 294.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1200.0, + 1261.0, + 1200.0, + 1261.0, + 1242.0, + 770.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 959.0, + 370.0, + 959.0, + 370.0, + 1009.0, + 294.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 959.0, + 674.0, + 959.0, + 674.0, + 1009.0, + 664.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1635.0, + 906.0, + 1635.0, + 906.0, + 1677.0, + 293.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1679.0, + 1403.0, + 1679.0, + 1403.0, + 1717.0, + 294.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1711.0, + 297.0, + 1711.0, + 297.0, + 1747.0, + 294.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1711.0, + 578.0, + 1711.0, + 578.0, + 1747.0, + 527.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1711.0, + 989.0, + 1711.0, + 989.0, + 1747.0, + 801.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 360.0, + 2000.0, + 360.0, + 2037.0, + 294.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1849.0, + 668.0, + 1849.0, + 668.0, + 1894.0, + 290.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 1849.0, + 950.0, + 1849.0, + 950.0, + 1894.0, + 897.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1849.0, + 1363.0, + 1849.0, + 1363.0, + 1894.0, + 1173.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1810.0, + 433.0, + 1810.0, + 433.0, + 1851.0, + 295.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 1810.0, + 917.0, + 1810.0, + 917.0, + 1851.0, + 671.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 1810.0, + 1047.0, + 1810.0, + 1047.0, + 1851.0, + 1038.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1756.0, + 309.0, + 1756.0, + 309.0, + 1802.0, + 304.0, + 1802.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 8, + "poly": [ + 676, + 738, + 1019, + 738, + 1019, + 923, + 676, + 923 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 536, + 1103, + 1164, + 1103, + 1164, + 1175, + 536, + 1175 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 299, + 1029, + 1403, + 1029, + 1403, + 1095, + 299, + 1095 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 295, + 1248, + 1058, + 1248, + 1058, + 1288, + 295, + 1288 + ], + "score": 0.933 + }, + { + "category_id": 8, + "poly": [ + 615, + 273, + 1079, + 273, + 1079, + 698, + 615, + 698 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 299, + 229, + 806, + 229, + 806, + 264, + 299, + 264 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 703, + 457, + 703, + 457, + 736, + 297, + 736 + ], + "score": 0.924 + }, + { + "category_id": 1, + "poly": [ + 300, + 1181, + 1063, + 1181, + 1063, + 1219, + 300, + 1219 + ], + "score": 0.92 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 858, + 73, + 858, + 107, + 297, + 107 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 295, + 991, + 1033, + 991, + 1033, + 1025, + 295, + 1025 + ], + "score": 0.915 + }, + { + "category_id": 8, + "poly": [ + 303, + 1331, + 1671, + 1331, + 1671, + 1830, + 303, + 1830 + ], + "score": 0.866 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 864, + 2087, + 864, + 2114, + 835, + 2114 + ], + "score": 0.838 + }, + { + "category_id": 2, + "poly": [ + 1373, + 926, + 1404, + 926, + 1404, + 957, + 1373, + 957 + ], + "score": 0.831 + }, + { + "category_id": 8, + "poly": [ + 294, + 1904, + 985, + 1904, + 985, + 2042, + 294, + 2042 + ], + "score": 0.785 + }, + { + "category_id": 1, + "poly": [ + 295, + 1833, + 978, + 1833, + 978, + 1900, + 295, + 1900 + ], + "score": 0.739 + }, + { + "category_id": 8, + "poly": [ + 293, + 1327, + 1625, + 1327, + 1625, + 2045, + 293, + 2045 + ], + "score": 0.207 + }, + { + "category_id": 8, + "poly": [ + 299, + 1331, + 1656, + 1331, + 1656, + 2042, + 299, + 2042 + ], + "score": 0.152 + }, + { + "category_id": 1, + "poly": [ + 294, + 1904, + 985, + 1904, + 985, + 2042, + 294, + 2042 + ], + "score": 0.127 + }, + { + "category_id": 8, + "poly": [ + 675, + 738, + 1020, + 738, + 1020, + 922, + 675, + 922 + ], + "score": 0.106 + }, + { + "category_id": 14, + "poly": [ + 310, + 1334, + 1428, + 1334, + 1428, + 2039, + 310, + 2039 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { d } \\xi \\Bigr ( x \\mathrm { S p } ( \\xi ) + \\mathrm { i } \\omega ( 2 \\xi ) \\mathrm { S p } ( \\xi ) + \\mathrm { i } \\omega \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { d } \\xi \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & { \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\mathcal { \\bar { E } } _ { x } ^ { x } \\Bigl ( \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) \\mathrm { S p } ( \\xi ) } \\\\ & \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ { \\infty } \\int _ { \\infty } ^ \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 445, + 1184, + 572, + 1184, + 572, + 1216, + 445, + 1216 + ], + "score": 0.93, + "latex": "p : \\mathcal { X } \\mathbb { R }" + }, + { + "category_id": 14, + "poly": [ + 534, + 1101, + 1164, + 1101, + 1164, + 1174, + 534, + 1174 + ], + "score": 0.93, + "latex": "\\int _ { \\mathcal { X } } p ( x ) f ( x ) g ( x ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } p ( x ) f ( x ) \\mathrm { d } x \\int _ { \\mathcal { X } } p ( x ) g ( x ) \\mathrm { d } x" + }, + { + "category_id": 13, + "poly": [ + 1185, + 1029, + 1329, + 1029, + 1329, + 1064, + 1185, + 1064 + ], + "score": 0.92, + "latex": "f \\left( x \\right) \\geq f \\left( y \\right)" + }, + { + "category_id": 13, + "poly": [ + 625, + 1182, + 830, + 1182, + 830, + 1217, + 625, + 1217 + ], + "score": 0.92, + "latex": "p ( x ) > 0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 14, + "poly": [ + 676, + 738, + 1020, + 738, + 1020, + 926, + 676, + 926 + ], + "score": 0.92, + "latex": "\\begin{array} { l } { { \\displaystyle n \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } a _ { i } \\sum _ { i = 1 } ^ { n } b _ { i } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } \\frac { 1 } { n } a _ { i } b _ { i } \\geq \\sum _ { i = 1 } ^ { n } \\frac { 1 } { n } a _ { i } \\sum _ { n } \\frac { 1 } { n } b _ { i } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 450, + 1250, + 949, + 1250, + 949, + 1286, + 450, + 1286 + ], + "score": 0.91, + "latex": "( f ( x ) - f ( y ) ) ( g ( x ) - g ( y ) ) \\geq 0 , \\forall x , y \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 821, + 1030, + 950, + 1030, + 950, + 1063, + 821, + 1063 + ], + "score": 0.91, + "latex": "f : \\mathcal { X } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 883, + 1181, + 1056, + 1181, + 1056, + 1220, + 883, + 1220 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\int _ { \\mathcal { X } } p ( x ) \\mathrm { d } x = 1 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1002, + 1031, + 1130, + 1031, + 1130, + 1063, + 1002, + 1063 + ], + "score": 0.91, + "latex": "g : \\mathcal { X } \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 530, + 1030, + 615, + 1030, + 615, + 1060, + 530, + 1060 + ], + "score": 0.88, + "latex": "\\mathcal { X } \\subset \\mathbb { R } ," + }, + { + "category_id": 14, + "poly": [ + 614, + 270, + 1071, + 270, + 1071, + 704, + 614, + 704 + ], + "score": 0.85, + "latex": "\\begin{array} { l } { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 1 } + a _ { 2 } b _ { 2 } + \\cdots + a _ { n } b _ { n } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 2 } + a _ { 2 } b _ { 3 } + \\cdots + a _ { n } b _ { 1 } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \\cdots + a _ { n } b _ { 2 } } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { 3 } + a _ { 2 } b _ { 4 } + \\cdots + a _ { n } b _ { 2 } } } \\\\ { { \\displaystyle \\qquad \\vdots \\qquad } } \\\\ { { \\displaystyle \\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { n } + a _ { 2 } b _ { 1 } + \\cdots + a _ { n } b _ { n - 1 } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 374, + 1061, + 515, + 1061, + 515, + 1095, + 374, + 1095 + ], + "score": 0.74, + "latex": "g ( x ) \\geq g ( y )" + }, + { + "category_id": 13, + "poly": [ + 531, + 1062, + 650, + 1062, + 650, + 1094, + 531, + 1094 + ], + "score": 0.6, + "latex": "\\forall x , y \\in { \\mathcal { X } }" + }, + { + "category_id": 14, + "poly": [ + 619, + 609, + 1076, + 609, + 1076, + 699, + 619, + 699 + ], + "score": 0.45, + "latex": "\\sum _ { i = 1 } ^ { n } a _ { i } b _ { i } \\geq a _ { 1 } b _ { n } + a _ { 2 } b _ { 1 } + \\cdots + a _ { n } b _ { n - 1 }" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2121.0, + 831.0, + 2121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 932.0, + 1402.0, + 932.0, + 1402.0, + 955.0, + 1379.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1026.0, + 529.0, + 1026.0, + 529.0, + 1066.0, + 294.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1026.0, + 820.0, + 1026.0, + 820.0, + 1066.0, + 616.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 1026.0, + 1001.0, + 1026.0, + 1001.0, + 1066.0, + 951.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1026.0, + 1184.0, + 1026.0, + 1184.0, + 1066.0, + 1131.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 1026.0, + 1407.0, + 1026.0, + 1407.0, + 1066.0, + 1330.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1059.0, + 373.0, + 1059.0, + 373.0, + 1096.0, + 295.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1059.0, + 530.0, + 1059.0, + 530.0, + 1096.0, + 516.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 1059.0, + 818.0, + 1059.0, + 818.0, + 1096.0, + 651.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1247.0, + 449.0, + 1247.0, + 449.0, + 1289.0, + 296.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 1247.0, + 1057.0, + 1247.0, + 1057.0, + 1289.0, + 950.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 225.0, + 806.0, + 225.0, + 806.0, + 271.0, + 294.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 701.0, + 458.0, + 701.0, + 458.0, + 737.0, + 294.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1176.0, + 444.0, + 1176.0, + 444.0, + 1223.0, + 293.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1176.0, + 624.0, + 1176.0, + 624.0, + 1223.0, + 573.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1176.0, + 882.0, + 1176.0, + 882.0, + 1223.0, + 831.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1176.0, + 1066.0, + 1176.0, + 1066.0, + 1223.0, + 1057.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 988.0, + 1032.0, + 988.0, + 1032.0, + 1029.0, + 293.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1850.0, + 309.0, + 1850.0, + 309.0, + 1874.0, + 299.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1897.0, + 309.0, + 1897.0, + 309.0, + 1964.0, + 288.0, + 1964.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 410, + 1097, + 1281, + 1097, + 1281, + 1349, + 410, + 1349 + ], + "score": 0.982, + "html": "
NameGenerator function f(x)If f'(x)≤0,∀x∈X
Kullback-LeiblerReverse KLPearson x²Squared HellingerJensen-ShannonGANx logxFalseTrueFalseFalseFalseTrue
-logx(x-1)²(√x-1)²-(x+1)log1++xlog x2xlogx- (𝑥+1)log(𝑥+1)
" + }, + { + "category_id": 1, + "poly": [ + 296, + 1428, + 1413, + 1428, + 1413, + 1620, + 296, + 1620 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 558, + 1406, + 558, + 1406, + 720, + 296, + 720 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 1785, + 1405, + 1785, + 1405, + 1919, + 296, + 1919 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 298, + 279, + 1406, + 279, + 1406, + 355, + 298, + 355 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 297, + 461, + 1405, + 461, + 1405, + 526, + 297, + 526 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 521, + 734, + 1177, + 734, + 1177, + 808, + 521, + 808 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 431, + 369, + 1268, + 369, + 1268, + 447, + 431, + 447 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 435, + 862, + 1263, + 862, + 1263, + 942, + 435, + 942 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 299, + 1957, + 1403, + 1957, + 1403, + 2039, + 299, + 2039 + ], + "score": 0.936 + }, + { + "category_id": 8, + "poly": [ + 295, + 1634, + 1537, + 1634, + 1537, + 1771, + 295, + 1771 + ], + "score": 0.929 + }, + { + "category_id": 2, + "poly": [ + 297, + 73, + 858, + 73, + 858, + 106, + 297, + 106 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 821, + 465, + 821, + 465, + 852, + 297, + 852 + ], + "score": 0.926 + }, + { + "category_id": 6, + "poly": [ + 509, + 1062, + 1188, + 1062, + 1188, + 1095, + 509, + 1095 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 295, + 954, + 368, + 954, + 368, + 985, + 295, + 985 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 864, + 2087, + 864, + 2113, + 835, + 2113 + ], + "score": 0.866 + }, + { + "category_id": 2, + "poly": [ + 1373, + 230, + 1404, + 230, + 1404, + 261, + 1373, + 261 + ], + "score": 0.826 + }, + { + "category_id": 2, + "poly": [ + 1373, + 953, + 1404, + 953, + 1404, + 984, + 1373, + 984 + ], + "score": 0.82 + }, + { + "category_id": 1, + "poly": [ + 295, + 1384, + 1240, + 1384, + 1240, + 1420, + 295, + 1420 + ], + "score": 0.805 + }, + { + "category_id": 13, + "poly": [ + 533, + 279, + 775, + 279, + 775, + 327, + 533, + 327 + ], + "score": 0.95, + "latex": "\\begin{array} { r } { g ( \\frac { p ( x ) } { q ( x ) } ) = - \\log \\frac { p ( x ) } { q ( x ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1168, + 645, + 1352, + 645, + 1352, + 691, + 1168, + 691 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { G ( x ) = g ( \\frac { p ( x ) } { q ( x ) } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 1576, + 505, + 1576, + 505, + 1620, + 298, + 1620 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1256, + 563, + 1403, + 563, + 1403, + 598, + 1256, + 598 + ], + "score": 0.92, + "latex": "f \\left( x \\right) \\geq f \\left( y \\right)" + }, + { + "category_id": 13, + "poly": [ + 426, + 1956, + 478, + 1956, + 478, + 2000, + 426, + 2000 + ], + "score": 0.92, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 380, + 562, + 803, + 562, + 803, + 601, + 380, + 601 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { g ^ { \\prime } ( x ) = - \\log x = - \\frac { 1 } { x } < 0 , \\forall x \\in \\mathcal { X } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1134, + 286, + 1352, + 286, + 1352, + 320, + 1134, + 320 + ], + "score": 0.92, + "latex": "q ( x ) > 0 , \\forall x \\in \\mathcal { X } _ { }" + }, + { + "category_id": 13, + "poly": [ + 881, + 599, + 1123, + 599, + 1123, + 645, + 881, + 645 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { f ( v \\frac { p ( x ) } { q ( x ) } ) \\geq f ( v \\frac { p ( y ) } { q ( y ) } ) } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 517, + 734, + 1177, + 734, + 1177, + 807, + 517, + 807 + ], + "score": 0.92, + "latex": "\\int _ { \\mathcal { X } } q ( x ) F ( x ) G ( x ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) F ( x ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) G ( x ) \\mathrm { d } x ." + }, + { + "category_id": 14, + "poly": [ + 430, + 369, + 1263, + 369, + 1263, + 446, + 430, + 446 + ], + "score": 0.92, + "latex": "\\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x ," + }, + { + "category_id": 13, + "poly": [ + 900, + 645, + 1113, + 645, + 1113, + 691, + 900, + 691 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { F ( x ) = f ( v \\frac { p ( x ) } { q ( x ) } ) ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 921, + 564, + 1141, + 564, + 1141, + 598, + 921, + 598 + ], + "score": 0.91, + "latex": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 1284, + 599, + 1404, + 599, + 1404, + 645, + 1284, + 645 + ], + "score": 0.91, + "latex": "g \\big ( \\frac { p ( x ) } { q ( x ) } \\big ) \\geq" + }, + { + "category_id": 13, + "poly": [ + 752, + 1507, + 1059, + 1507, + 1059, + 1550, + 752, + 1550 + ], + "score": 0.91, + "latex": "\\theta _ { t + 1 } = \\mathrm { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } )" + }, + { + "category_id": 14, + "poly": [ + 434, + 864, + 1263, + 864, + 1263, + 939, + 434, + 939 + ], + "score": 0.91, + "latex": "\\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\geq \\int _ { \\mathcal { X } } q ( x ) f ( v \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x \\int _ { \\mathcal { X } } q ( x ) g ( \\frac { p ( x ) } { q ( x ) } ) \\mathrm { d } x" + }, + { + "category_id": 13, + "poly": [ + 526, + 1427, + 579, + 1427, + 579, + 1473, + 526, + 1473 + ], + "score": 0.91, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 858, + 285, + 1077, + 285, + 1077, + 320, + 858, + 320 + ], + "score": 0.91, + "latex": "p ( x ) > 0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 868, + 2003, + 921, + 2003, + 921, + 2036, + 868, + 2036 + ], + "score": 0.91, + "latex": "\\theta _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 761, + 1063, + 974, + 1063, + 974, + 1097, + 761, + 1097 + ], + "score": 0.91, + "latex": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 629, + 1825, + 856, + 1825, + 856, + 1858, + 629, + 1858 + ], + "score": 0.91, + "latex": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 392, + 461, + 619, + 461, + 619, + 495, + 392, + 495 + ], + "score": 0.9, + "latex": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 1152, + 1508, + 1205, + 1508, + 1205, + 1550, + 1152, + 1550 + ], + "score": 0.9, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 615, + 1547, + 663, + 1547, + 663, + 1579, + 615, + 1579 + ], + "score": 0.9, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 886, + 1789, + 1024, + 1789, + 1024, + 1827, + 886, + 1827 + ], + "score": 0.89, + "latex": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { \\kappa } }" + }, + { + "category_id": 13, + "poly": [ + 671, + 652, + 744, + 652, + 744, + 680, + 671, + 680 + ], + "score": 0.89, + "latex": "\\upsilon > 0" + }, + { + "category_id": 13, + "poly": [ + 960, + 1998, + 1268, + 1998, + 1268, + 2038, + 960, + 2038 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\operatorname* { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1098, + 1853, + 1165, + 1853, + 1165, + 1883, + 1098, + 1883 + ], + "score": 0.88, + "latex": "\\bar { \\theta } = \\theta" + }, + { + "category_id": 13, + "poly": [ + 1069, + 1963, + 1101, + 1963, + 1101, + 1997, + 1069, + 1997 + ], + "score": 0.88, + "latex": "\\bar { \\beta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 298, + 323, + 367, + 323, + 367, + 352, + 298, + 352 + ], + "score": 0.87, + "latex": "\\upsilon > 0" + }, + { + "category_id": 13, + "poly": [ + 1187, + 1434, + 1220, + 1434, + 1220, + 1469, + 1187, + 1469 + ], + "score": 0.87, + "latex": "\\bar { \\beta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1358, + 1966, + 1400, + 1966, + 1400, + 1996, + 1358, + 1996 + ], + "score": 0.86, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 373, + 1789, + 783, + 1789, + 783, + 1826, + 373, + 1826 + ], + "score": 0.86, + "latex": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) , \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } [ \\cdot ]" + }, + { + "category_id": 13, + "poly": [ + 1232, + 1824, + 1282, + 1824, + 1282, + 1858, + 1232, + 1858 + ], + "score": 0.86, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 1131, + 464, + 1149, + 464, + 1149, + 495, + 1131, + 495 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1020, + 1065, + 1039, + 1065, + 1039, + 1096, + 1020, + 1096 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 666, + 1307, + 989, + 1307, + 989, + 1339, + 666, + 1339 + ], + "score": 0.84, + "latex": "x \\log x - ( x + \\bar { 1 } ) \\log ( x + 1 )" + }, + { + "category_id": 13, + "poly": [ + 883, + 1105, + 939, + 1105, + 939, + 1139, + 883, + 1139 + ], + "score": 0.83, + "latex": "f ( x )" + }, + { + "category_id": 13, + "poly": [ + 1066, + 1790, + 1086, + 1790, + 1086, + 1822, + 1066, + 1822 + ], + "score": 0.82, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 298, + 1825, + 317, + 1825, + 317, + 1857, + 298, + 1857 + ], + "score": 0.82, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 464, + 496, + 483, + 496, + 483, + 525, + 464, + 525 + ], + "score": 0.82, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 963, + 464, + 982, + 464, + 982, + 495, + 963, + 495 + ], + "score": 0.82, + "latex": "f" + }, + { + "category_id": 14, + "poly": [ + 310, + 1634, + 1428, + 1634, + 1428, + 1771, + 310, + 1771 + ], + "score": 0.82, + "latex": "\\mathfrak { t } _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\times } } \\left[ \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\underbrace { \\left( f \\left( \\frac { \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f \\left( \\frac { \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) + H _ { M ^ { * } } ( x , a ) \\right) } _ { W ( x , a , x ^ { \\prime } ) } \\right] ." + }, + { + "category_id": 13, + "poly": [ + 950, + 1475, + 979, + 1475, + 979, + 1503, + 950, + 1503 + ], + "score": 0.81, + "latex": "\\theta ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 665, + 1274, + 983, + 1274, + 983, + 1308, + 665, + 1308 + ], + "score": 0.79, + "latex": "- ( x + 1 ) \\log { \\frac { 1 + x } { 2 } } + x \\log x" + }, + { + "category_id": 13, + "poly": [ + 502, + 2004, + 519, + 2004, + 519, + 2030, + 502, + 2030 + ], + "score": 0.77, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 665, + 1244, + 787, + 1244, + 787, + 1276, + 665, + 1276 + ], + "score": 0.76, + "latex": "( \\sqrt { x } - 1 ) ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 921, + 1827, + 939, + 1827, + 939, + 1852, + 921, + 1852 + ], + "score": 0.75, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 297, + 642, + 383, + 642, + 383, + 691, + 297, + 691 + ], + "score": 0.72, + "latex": "g ( \\textstyle { \\frac { p ( y ) } { q ( y ) } } )" + }, + { + "category_id": 13, + "poly": [ + 523, + 1210, + 556, + 1210, + 556, + 1243, + 523, + 1243 + ], + "score": 0.72, + "latex": "\\chi ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1002, + 1470, + 1324, + 1470, + 1324, + 1508, + 1002, + 1508 + ], + "score": 0.71, + "latex": "\\begin{array} { r } \\mathsf { \\Gamma } _ { \\mathsf { m i n } _ { \\theta } \\operatorname* { m a x } _ { \\beta \\in \\Pi } \\bar { L } ( \\rho _ { M ^ { * } } ^ { \\beta } , M _ { \\theta } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 399, + 648, + 522, + 648, + 522, + 684, + 399, + 684 + ], + "score": 0.69, + "latex": "\\forall x , y \\in { \\mathcal { X } }" + }, + { + "category_id": 13, + "poly": [ + 611, + 1476, + 627, + 1476, + 627, + 1502, + 611, + 1502 + ], + "score": 0.69, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 666, + 1211, + 765, + 1211, + 765, + 1244, + 666, + 1244 + ], + "score": 0.68, + "latex": "( x - 1 ) ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 371, + 1474, + 412, + 1474, + 412, + 1506, + 371, + 1506 + ], + "score": 0.65, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 611, + 606, + 733, + 606, + 733, + 638, + 611, + 638 + ], + "score": 0.63, + "latex": "\\forall x , y \\in { \\mathcal { X } }" + }, + { + "category_id": 13, + "poly": [ + 454, + 605, + 732, + 605, + 732, + 640, + 454, + 640 + ], + "score": 0.63, + "latex": "g ( x ) \\geq g ( y ) , \\forall x , y \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 1182, + 1825, + 1214, + 1825, + 1214, + 1852, + 1182, + 1852 + ], + "score": 0.59, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 1046, + 1102, + 1262, + 1102, + 1262, + 1139, + 1046, + 1139 + ], + "score": 0.47, + "latex": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 297, + 642, + 521, + 642, + 521, + 690, + 297, + 690 + ], + "score": 0.45, + "latex": "g ( \\frac { p ( y ) } { q ( y ) } ) , \\forall x , y \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 665, + 1183, + 750, + 1183, + 750, + 1209, + 665, + 1209 + ], + "score": 0.4, + "latex": "- \\log x" + }, + { + "category_id": 13, + "poly": [ + 665, + 1150, + 745, + 1150, + 745, + 1180, + 665, + 1180 + ], + "score": 0.39, + "latex": "x \\log x" + }, + { + "category_id": 13, + "poly": [ + 1383, + 1831, + 1402, + 1831, + 1402, + 1852, + 1383, + 1852 + ], + "score": 0.26, + "latex": "a" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 1057.0, + 760.0, + 1057.0, + 760.0, + 1104.0, + 509.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 1057.0, + 1019.0, + 1057.0, + 1019.0, + 1104.0, + 975.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 1057.0, + 1189.0, + 1057.0, + 1189.0, + 1104.0, + 1040.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 236.0, + 1402.0, + 236.0, + 1402.0, + 259.0, + 1379.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 958.0, + 1403.0, + 958.0, + 1403.0, + 985.0, + 1377.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1430.0, + 525.0, + 1430.0, + 525.0, + 1480.0, + 292.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1430.0, + 1186.0, + 1430.0, + 1186.0, + 1480.0, + 580.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 1430.0, + 1408.0, + 1430.0, + 1408.0, + 1480.0, + 1221.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1469.0, + 370.0, + 1469.0, + 370.0, + 1514.0, + 293.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1469.0, + 610.0, + 1469.0, + 610.0, + 1514.0, + 413.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1469.0, + 949.0, + 1469.0, + 949.0, + 1514.0, + 628.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1469.0, + 1001.0, + 1469.0, + 1001.0, + 1514.0, + 980.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1469.0, + 1408.0, + 1469.0, + 1408.0, + 1514.0, + 1325.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1508.0, + 751.0, + 1508.0, + 751.0, + 1558.0, + 292.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 1508.0, + 1151.0, + 1508.0, + 1151.0, + 1558.0, + 1060.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1508.0, + 1410.0, + 1508.0, + 1410.0, + 1558.0, + 1206.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1545.0, + 614.0, + 1545.0, + 614.0, + 1581.0, + 295.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 1545.0, + 1406.0, + 1545.0, + 1406.0, + 1581.0, + 664.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1580.0, + 297.0, + 1580.0, + 297.0, + 1624.0, + 293.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1580.0, + 896.0, + 1580.0, + 896.0, + 1624.0, + 506.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 554.0, + 379.0, + 554.0, + 379.0, + 604.0, + 291.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 554.0, + 920.0, + 554.0, + 920.0, + 604.0, + 804.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 554.0, + 1255.0, + 554.0, + 1255.0, + 604.0, + 1142.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 554.0, + 1408.0, + 554.0, + 1408.0, + 604.0, + 1404.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 582.0, + 453.0, + 582.0, + 453.0, + 664.0, + 281.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 582.0, + 880.0, + 582.0, + 880.0, + 664.0, + 734.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 582.0, + 1283.0, + 582.0, + 1283.0, + 664.0, + 1124.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 582.0, + 1418.0, + 582.0, + 1418.0, + 664.0, + 1405.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 616.0, + 296.0, + 616.0, + 296.0, + 715.0, + 276.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 616.0, + 670.0, + 616.0, + 670.0, + 715.0, + 523.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 616.0, + 899.0, + 616.0, + 899.0, + 715.0, + 745.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 616.0, + 1167.0, + 616.0, + 1167.0, + 715.0, + 1114.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 648.0, + 1405.0, + 648.0, + 1405.0, + 683.0, + 1353.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 685.0, + 615.0, + 685.0, + 615.0, + 719.0, + 294.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1785.0, + 372.0, + 1785.0, + 372.0, + 1829.0, + 293.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1785.0, + 885.0, + 1785.0, + 885.0, + 1829.0, + 784.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1785.0, + 1065.0, + 1785.0, + 1065.0, + 1829.0, + 1025.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1785.0, + 1406.0, + 1785.0, + 1406.0, + 1829.0, + 1087.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1824.0, + 628.0, + 1824.0, + 628.0, + 1858.0, + 318.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 1824.0, + 920.0, + 1824.0, + 920.0, + 1858.0, + 857.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 1824.0, + 1181.0, + 1824.0, + 1181.0, + 1858.0, + 940.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1824.0, + 1231.0, + 1824.0, + 1231.0, + 1858.0, + 1215.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 1824.0, + 1382.0, + 1824.0, + 1382.0, + 1858.0, + 1283.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1853.0, + 1097.0, + 1853.0, + 1097.0, + 1890.0, + 290.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 1853.0, + 1407.0, + 1853.0, + 1407.0, + 1890.0, + 1166.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1885.0, + 669.0, + 1885.0, + 669.0, + 1922.0, + 290.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 281.0, + 857.0, + 281.0, + 857.0, + 320.0, + 778.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 281.0, + 1133.0, + 281.0, + 1133.0, + 320.0, + 1078.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 281.0, + 1404.0, + 281.0, + 1404.0, + 320.0, + 1353.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 320.0, + 297.0, + 320.0, + 297.0, + 358.0, + 292.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 320.0, + 535.0, + 320.0, + 535.0, + 358.0, + 368.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 274.0, + 615.0, + 274.0, + 615.0, + 330.0, + 292.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 457.0, + 391.0, + 457.0, + 391.0, + 500.0, + 292.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 457.0, + 962.0, + 457.0, + 962.0, + 500.0, + 620.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 457.0, + 1130.0, + 457.0, + 1130.0, + 500.0, + 983.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 457.0, + 1406.0, + 457.0, + 1406.0, + 500.0, + 1150.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 491.0, + 463.0, + 491.0, + 463.0, + 526.0, + 293.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 491.0, + 793.0, + 491.0, + 793.0, + 526.0, + 484.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1956.0, + 425.0, + 1956.0, + 425.0, + 2002.0, + 290.0, + 2002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1956.0, + 1068.0, + 1956.0, + 1068.0, + 2002.0, + 479.0, + 2002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1956.0, + 1357.0, + 1956.0, + 1357.0, + 2002.0, + 1102.0, + 2002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1956.0, + 1405.0, + 1956.0, + 1405.0, + 2002.0, + 1401.0, + 2002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1998.0, + 501.0, + 1998.0, + 501.0, + 2040.0, + 292.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 1998.0, + 867.0, + 1998.0, + 867.0, + 2040.0, + 520.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 1998.0, + 959.0, + 1998.0, + 959.0, + 2040.0, + 922.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1998.0, + 1408.0, + 1998.0, + 1408.0, + 2040.0, + 1269.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 817.0, + 469.0, + 817.0, + 469.0, + 855.0, + 294.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 951.0, + 374.0, + 951.0, + 374.0, + 989.0, + 291.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1383.0, + 1243.0, + 1383.0, + 1243.0, + 1422.0, + 294.0, + 1422.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1270, + 1408, + 1270, + 1408, + 1467, + 296, + 1467 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1722, + 1407, + 1722, + 1407, + 1843, + 297, + 1843 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1930, + 1406, + 1930, + 1406, + 2044, + 297, + 2044 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 307, + 321, + 1633, + 321, + 1633, + 1205, + 307, + 1205 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 682, + 1866, + 1018, + 1866, + 1018, + 1909, + 682, + 1909 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 296, + 224, + 1401, + 224, + 1401, + 295, + 296, + 295 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 297, + 1622, + 847, + 1622, + 847, + 1656, + 297, + 1656 + ], + "score": 0.929 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 106, + 297, + 106 + ], + "score": 0.925 + }, + { + "category_id": 9, + "poly": [ + 1354, + 1870, + 1401, + 1870, + 1401, + 1902, + 1354, + 1902 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 866, + 2087, + 866, + 2113, + 834, + 2113 + ], + "score": 0.867 + }, + { + "category_id": 0, + "poly": [ + 299, + 1560, + 828, + 1560, + 828, + 1594, + 299, + 1594 + ], + "score": 0.857 + }, + { + "category_id": 9, + "poly": [ + 1357, + 335, + 1403, + 335, + 1403, + 365, + 1357, + 365 + ], + "score": 0.76 + }, + { + "category_id": 9, + "poly": [ + 1357, + 1195, + 1403, + 1195, + 1403, + 1223, + 1357, + 1223 + ], + "score": 0.683 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1481, + 1403, + 1481, + 1403, + 1510, + 1374, + 1510 + ], + "score": 0.586 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1481, + 1403, + 1481, + 1403, + 1511, + 1374, + 1511 + ], + "score": 0.503 + }, + { + "category_id": 13, + "poly": [ + 773, + 1725, + 1121, + 1725, + 1121, + 1784, + 773, + 1784 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\big ) - f \\big ( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\big ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1051, + 1996, + 1397, + 1996, + 1397, + 2043, + 1051, + 2043 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\frac { p ( x ) } { q ( x ) } f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) - f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) + f ( 1 ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 638, + 1998, + 725, + 1998, + 725, + 2043, + 638, + 2043 + ], + "score": 0.93, + "latex": "f ( { \\frac { p ( x ) } { q ( x ) } } )" + }, + { + "category_id": 13, + "poly": [ + 1109, + 1363, + 1397, + 1363, + 1397, + 1398, + 1109, + 1398 + ], + "score": 0.92, + "latex": "q ( x ^ { \\prime } | x , a ) = M ^ { * } ( x ^ { \\prime } | x , a )" + }, + { + "category_id": 13, + "poly": [ + 353, + 1996, + 401, + 1996, + 401, + 2043, + 353, + 2043 + ], + "score": 0.92, + "latex": "\\textstyle { \\frac { p ( x ) } { q ( x ) } }" + }, + { + "category_id": 13, + "poly": [ + 1307, + 1314, + 1406, + 1314, + 1406, + 1350, + 1307, + 1350 + ], + "score": 0.92, + "latex": "f ^ { \\prime } ( x ) \\leq" + }, + { + "category_id": 13, + "poly": [ + 437, + 1809, + 721, + 1809, + 721, + 1843, + 437, + 1843 + ], + "score": 0.92, + "latex": "u \\bar { \\in } ( 1 - \\xi , 1 + \\xi ) , \\xi > 0" + }, + { + "category_id": 13, + "poly": [ + 874, + 1270, + 926, + 1270, + 926, + 1313, + 874, + 1313 + ], + "score": 0.92, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1311, + 440, + 1311, + 440, + 1352, + 298, + 1352 + ], + "score": 0.91, + "latex": "\\| \\rho _ { M ^ { * } } ^ { \\beta } ( \\cdot , \\cdot ) \\| _ { 2 } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 696, + 224, + 991, + 224, + 991, + 267, + 696, + 267 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\theta _ { t + 1 } = \\operatorname* { m i n } _ { \\theta } \\bar { L } ( \\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } } , M _ { \\theta } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1078, + 224, + 1130, + 224, + 1130, + 267, + 1078, + 267 + ], + "score": 0.91, + "latex": "\\bar { \\rho } _ { M ^ { * } } ^ { \\bar { \\beta } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 1252, + 1404, + 1313, + 1404, + 1313, + 1434, + 1252, + 1434 + ], + "score": 0.91, + "latex": "H _ { M ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 521, + 262, + 570, + 262, + 570, + 294, + 521, + 294 + ], + "score": 0.9, + "latex": "\\bar { M } _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 1161, + 1966, + 1231, + 1966, + 1231, + 1998, + 1161, + 1998 + ], + "score": 0.9, + "latex": "\\mathcal { O } ( \\xi ^ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 569, + 1964, + 714, + 1964, + 714, + 1998, + 569, + 1998 + ], + "score": 0.9, + "latex": "( 1 - \\xi , 1 + \\xi )" + }, + { + "category_id": 14, + "poly": [ + 681, + 1865, + 1017, + 1865, + 1017, + 1906, + 681, + 1906 + ], + "score": 0.9, + "latex": "f ( u ) \\approx f ( 1 ) + f ^ { \\prime } ( u ) ( u - 1 ) ," + }, + { + "category_id": 13, + "poly": [ + 1159, + 1932, + 1222, + 1932, + 1222, + 1965, + 1159, + 1965 + ], + "score": 0.89, + "latex": "f ^ { \\prime } ( u )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1364, + 417, + 1364, + 417, + 1397, + 298, + 1397 + ], + "score": 0.89, + "latex": "0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 464, + 1348, + 752, + 1348, + 752, + 1408, + 464, + 1408 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { v ( x , a ) : = \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } > 0 } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 347, + 859, + 1465, + 859, + 1465, + 1202, + 347, + 1202 + ], + "score": 0.89, + "latex": "\\begin{array} { r l r } & { } & { - \\rho _ { \\mathcal { M } ^ { * } } ^ { \\kappa } ( x , a ) ( f ( \\frac { \\rho _ { \\mathcal { M } _ { \\theta } } ^ { \\kappa } ( x , a ) } { \\rho _ { \\mathcal { M } ^ { * } } ^ { \\kappa } ( x , a ) } ) - H _ { \\mathcal { M } ^ { * } } ( x , a ) ) \\displaystyle \\int _ { \\mathcal { X } } { \\cal M } ^ { * } ( x ^ { \\prime } | x , a ) ( - \\log { \\cal M } _ { \\theta } ( x ^ { \\prime } | x , a ) } \\\\ & { } & { = \\operatorname* { m i n } _ { \\theta } \\displaystyle \\int _ { \\mathcal { X } , \\mathcal { A } } \\displaystyle \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } ( \\rho _ { \\mathcal { M } ^ { * } } ^ { \\kappa } ( x , a ) \\int _ { \\mathcal { X } } ( { \\cal M } ^ { * } ( x ^ { \\prime } | x , a ) f ( \\frac { \\rho _ { \\mathcal { M } _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { \\mathcal { M } ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } ) ( - \\log { \\cal M } _ { \\theta } ( x ^ { \\prime } | x , a ) ) ) \\mathrm { d } x ^ { \\prime } } \\\\ & { } & { - \\rho _ { \\mathcal { M } ^ { * } } ^ { \\kappa } ( x , a ) ( f ( \\frac { \\rho _ { \\mathcal { M } _ { \\theta } } ^ { \\kappa } ( x , a ) } { \\rho _ { \\mathcal { M } ^ { * } } ^ { \\kappa } ( x , a ) } ) - H _ { \\mathcal { M } ^ { * } } ( x , a ) ) \\displaystyle \\int _ { \\mathcal { X } } { \\cal M } ^ { * } ( x ^ { \\prime } | x , a ) ( - \\log { \\cal M } _ { \\theta } ( x ^ { \\prime } | x , a ) } \\\\ & { } & = \\operatorname* { m a x } _ { \\theta } \\displaystyle \\int _ { \\mathcal { X } , \\mathcal { A } , \\mathcal { X } } \\displaystyle \\frac { 1 } { \\alpha _ { 0 } ( x , a ) } \\rho _ { \\mathcal { M } ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\log { \\cal M } _ { \\theta } ( x ^ { \\prime } | x , a ) ( f ( \\frac { \\rho _ { \\mathcal { M } _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\rho _ { \\mathcal { M } ^ { * } } ^ { \\kappa } ( x , a \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 374, + 1280, + 424, + 1280, + 424, + 1311, + 374, + 1311 + ], + "score": 0.88, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 14, + "poly": [ + 282, + 317, + 1391, + 317, + 1391, + 810, + 282, + 810 + ], + "score": 0.88, + "latex": "\\begin{array} { r l } { \\theta _ { i + 1 } = } & { \\operatorname* { m i n } _ { \\theta } \\int _ { x , A } \\overline { { \\rho } } _ { n + } ^ { \\infty } ( z , \\boldsymbol { \\alpha } ) \\int _ { x } M ^ { * } ( z ^ { * } | \\mathbf { x } , \\boldsymbol { \\alpha } ) ( - \\log M _ { \\theta } ( z ^ { * } | \\mathbf { x } , \\boldsymbol { \\alpha } ) ) \\mathrm { d } z ^ { * } \\mathrm { d } \\mathrm { e d } z } \\\\ & { = \\operatorname* { m i n } _ { \\theta } \\int _ { x , A } \\frac { 1 } { \\alpha \\alpha ( | x , \\boldsymbol { \\alpha } ) } \\Bigg ( \\int _ { x } p _ { A + } ^ { \\infty } ( x , a , \\boldsymbol { \\alpha } ) \\cdot p _ { i } \\int _ { x } \\int _ { x } \\frac { \\rho _ { A + } ^ { \\infty } ( x , a , z ^ { * } ) } { p _ { A + } ^ { * } ( x , a , x ^ { * } ) } \\Bigg ) \\mathrm { d } x ^ { * } \\int _ { x } M ^ { * } ( x ^ { * } | x , a ) ( - \\log M _ { \\theta } ( z ^ { * } | x , a ) } \\\\ & { \\qquad - p _ { A + } ^ { * } ( x , a ) \\Bigg ( \\int _ { x } \\frac { \\rho _ { A + } ^ { \\infty } ( x , a ) } { p _ { A + } ^ { * } ( x , a ) } \\Bigg ) - R _ { x } ( x , a ) \\Bigg ) \\int _ { x } M ^ { * } ( x ^ { * } | x , a ) ( - \\log M _ { \\theta } ( z ^ { * } | x , a ) ) } \\\\ & { = \\operatorname* { m i n } _ { \\theta } \\int _ { x , A } \\frac { 1 } { \\alpha ( | x , \\boldsymbol { \\alpha } ) } \\Bigg ( \\int _ { x } p _ { A + } ^ { \\infty } ( x , a , x ^ { * } ) \\int _ { x } \\frac { \\rho _ { A + } ^ { * } ( x , a , z ^ { * } ) } { p _ { A + } ^ { * } ( x , a , x ^ { * } ) } \\Bigg ) \\mathrm { d } x ^ { * } ( \\int _ { x } M ^ { * } ( z ^ { * } | x , a ) ( - \\log \\frac { \\mathrm { d } \\mathrm { d } \\theta } { \\mathrm { d } x ^ { * } ( z ^ { * } | x , a ) } } \\\\ & \\qquad - p _ { A + } ^ { * } ( x , a ) ( \\int _ { x } \\frac { \\rho _ { A + } ^ { * } ( x , a ) } { p _ { A + } ^ { * } ( x , a ) } ) - R _ M ^ \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 369, + 1933, + 397, + 1933, + 397, + 1964, + 369, + 1964 + ], + "score": 0.88, + "latex": "f ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 394, + 1966, + 414, + 1966, + 414, + 1996, + 394, + 1996 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 722, + 1934, + 740, + 1934, + 740, + 1965, + 722, + 1965 + ], + "score": 0.83, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 713, + 1435, + 729, + 1435, + 729, + 1461, + 713, + 1461 + ], + "score": 0.82, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 1179, + 1317, + 1195, + 1317, + 1195, + 1344, + 1179, + 1344 + ], + "score": 0.79, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 1250, + 1280, + 1267, + 1280, + 1267, + 1306, + 1250, + 1306 + ], + "score": 0.76, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 1383, + 1975, + 1401, + 1975, + 1401, + 1992, + 1383, + 1992 + ], + "score": 0.7, + "latex": "u" + }, + { + "category_id": 13, + "poly": [ + 764, + 1362, + 1048, + 1362, + 1048, + 1399, + 764, + 1399 + ], + "score": 0.69, + "latex": "p ( x ^ { \\prime } | x , a ) = M _ { \\theta } ( x ^ { \\prime } | x , a )" + }, + { + "category_id": 14, + "poly": [ + 348, + 396, + 1465, + 396, + 1465, + 1222, + 348, + 1222 + ], + "score": 0.61, + "latex": "\\begin{array} { r l } & \\quad \\mathrm { w r e r } \\int _ { - \\infty } ^ { \\infty } \\frac { 1 } { \\mathrm { d } { x } } \\frac { \\mathrm { d } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } } \\left( \\int _ { - \\infty } ^ { \\infty } \\mathrm { d } { x } \\omega \\mathrm { e r } \\omega ^ { 2 } \\mathrm { d } { x } ^ { 2 } \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } } \\right) + \\int _ { - \\infty } ^ { \\infty } \\mathrm { d } { x } ^ { 2 } \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 3 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 2 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } } \\\\ & \\qquad - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\left( \\frac { \\sqrt { 3 } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 2 } \\right) } - \\mathrm { e } ^ { \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { y } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } } \\\\ & \\qquad - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } \\mathrm { a n } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\left( \\frac { \\sqrt { 3 } } { \\mathrm { d } { x } ^ { 2 } \\mathrm { x } ^ { 3 } \\mathrm { x } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 2 } \\right) } - \\mathrm { e } ^ \\mathrm { i } \\omega _ { 3 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\mathrm { x } ^ { 3 } \\mathrm { y } ^ { 4 } \\mathrm y \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 2083.0, + 872.0, + 2083.0, + 872.0, + 2124.0, + 828.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1558.0, + 831.0, + 1558.0, + 831.0, + 1596.0, + 295.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1485.0, + 1403.0, + 1485.0, + 1403.0, + 1511.0, + 1378.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1486.0, + 1403.0, + 1486.0, + 1403.0, + 1512.0, + 1377.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1272.0, + 373.0, + 1272.0, + 373.0, + 1323.0, + 291.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1272.0, + 873.0, + 1272.0, + 873.0, + 1323.0, + 425.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1272.0, + 1249.0, + 1272.0, + 1249.0, + 1323.0, + 927.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1272.0, + 1410.0, + 1272.0, + 1410.0, + 1323.0, + 1268.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 1313.0, + 1178.0, + 1313.0, + 1178.0, + 1354.0, + 441.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 1313.0, + 1306.0, + 1313.0, + 1306.0, + 1354.0, + 1196.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1338.0, + 297.0, + 1338.0, + 297.0, + 1408.0, + 293.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1338.0, + 463.0, + 1338.0, + 463.0, + 1408.0, + 418.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 1338.0, + 763.0, + 1338.0, + 763.0, + 1408.0, + 753.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.0, + 1338.0, + 1108.0, + 1338.0, + 1108.0, + 1408.0, + 1049.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1338.0, + 1407.0, + 1338.0, + 1407.0, + 1408.0, + 1398.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1402.0, + 1251.0, + 1402.0, + 1251.0, + 1439.0, + 295.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 1402.0, + 1406.0, + 1402.0, + 1406.0, + 1439.0, + 1314.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1433.0, + 712.0, + 1433.0, + 712.0, + 1467.0, + 295.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1433.0, + 740.0, + 1433.0, + 740.0, + 1467.0, + 730.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1738.0, + 772.0, + 1738.0, + 772.0, + 1774.0, + 295.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1738.0, + 1404.0, + 1738.0, + 1404.0, + 1773.0, + 1122.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1774.0, + 1408.0, + 1774.0, + 1408.0, + 1816.0, + 294.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1809.0, + 436.0, + 1809.0, + 436.0, + 1843.0, + 296.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1809.0, + 831.0, + 1809.0, + 831.0, + 1843.0, + 722.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.75, + 1714.0, + 856.75, + 1714.0, + 856.75, + 1762.5, + 798.75, + 1762.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1715.0, + 1061.0, + 1715.0, + 1061.0, + 1762.5, + 1005.0, + 1762.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1928.0, + 368.0, + 1928.0, + 368.0, + 1967.0, + 294.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1928.0, + 721.0, + 1928.0, + 721.0, + 1967.0, + 398.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1928.0, + 1158.0, + 1928.0, + 1158.0, + 1967.0, + 741.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 1928.0, + 1407.0, + 1928.0, + 1407.0, + 1967.0, + 1223.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1954.0, + 393.0, + 1954.0, + 393.0, + 2003.0, + 291.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1954.0, + 568.0, + 1954.0, + 568.0, + 2003.0, + 415.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1954.0, + 1160.0, + 1954.0, + 1160.0, + 2003.0, + 715.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 1954.0, + 1382.0, + 1954.0, + 1382.0, + 2003.0, + 1232.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1954.0, + 1411.0, + 1954.0, + 1411.0, + 2003.0, + 1402.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1986.0, + 352.0, + 1986.0, + 352.0, + 2048.0, + 291.0, + 2048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1986.0, + 637.0, + 1986.0, + 637.0, + 2048.0, + 402.0, + 2048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 1986.0, + 1050.0, + 1986.0, + 1050.0, + 2048.0, + 726.0, + 2048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1998.0, + 1409.0, + 1998.0, + 1409.0, + 2042.0, + 1398.0, + 2042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 2015.0, + 1050.0, + 2015.0, + 1050.0, + 2046.0, + 1046.0, + 2046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1994.0, + 1273.0, + 1994.0, + 1273.0, + 2044.5, + 1179.0, + 2044.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 1994.5, + 1303.0, + 1994.5, + 1303.0, + 2023.0, + 1248.0, + 2023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 219.0, + 695.0, + 219.0, + 695.0, + 274.0, + 290.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 219.0, + 1077.0, + 219.0, + 1077.0, + 274.0, + 992.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 219.0, + 1410.0, + 219.0, + 1410.0, + 274.0, + 1131.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 255.0, + 520.0, + 255.0, + 520.0, + 297.0, + 293.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 255.0, + 583.0, + 255.0, + 583.0, + 297.0, + 571.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1622.0, + 848.0, + 1622.0, + 848.0, + 1659.0, + 296.0, + 1659.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 1644, + 1409, + 1644, + 1409, + 1743, + 295, + 1743 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 297, + 1122, + 1409, + 1122, + 1409, + 1301, + 297, + 1301 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 297, + 1557, + 1159, + 1557, + 1159, + 1631, + 297, + 1631 + ], + "score": 0.936 + }, + { + "category_id": 8, + "poly": [ + 407, + 1920, + 1287, + 1920, + 1287, + 2028, + 407, + 2028 + ], + "score": 0.934 + }, + { + "category_id": 8, + "poly": [ + 530, + 1748, + 1166, + 1748, + 1166, + 1797, + 530, + 1797 + ], + "score": 0.928 + }, + { + "category_id": 2, + "poly": [ + 296, + 73, + 858, + 73, + 858, + 107, + 296, + 107 + ], + "score": 0.923 + }, + { + "category_id": 8, + "poly": [ + 461, + 1421, + 1587, + 1421, + 1587, + 1509, + 461, + 1509 + ], + "score": 0.874 + }, + { + "category_id": 1, + "poly": [ + 294, + 1811, + 1371, + 1811, + 1371, + 1853, + 294, + 1853 + ], + "score": 0.874 + }, + { + "category_id": 8, + "poly": [ + 291, + 1306, + 1461, + 1306, + 1461, + 1393, + 291, + 1393 + ], + "score": 0.858 + }, + { + "category_id": 8, + "poly": [ + 355, + 255, + 1669, + 255, + 1669, + 1118, + 355, + 1118 + ], + "score": 0.829 + }, + { + "category_id": 9, + "poly": [ + 1358, + 1388, + 1403, + 1388, + 1403, + 1416, + 1358, + 1416 + ], + "score": 0.795 + }, + { + "category_id": 1, + "poly": [ + 296, + 230, + 455, + 230, + 455, + 261, + 296, + 261 + ], + "score": 0.786 + }, + { + "category_id": 9, + "poly": [ + 1358, + 1505, + 1403, + 1505, + 1403, + 1532, + 1358, + 1532 + ], + "score": 0.758 + }, + { + "category_id": 1, + "poly": [ + 296, + 1884, + 443, + 1884, + 443, + 1916, + 296, + 1916 + ], + "score": 0.616 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 864, + 2087, + 864, + 2114, + 834, + 2114 + ], + "score": 0.6 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 863, + 2087, + 863, + 2114, + 834, + 2114 + ], + "score": 0.495 + }, + { + "category_id": 8, + "poly": [ + 296, + 262, + 1372, + 262, + 1372, + 513, + 296, + 513 + ], + "score": 0.268 + }, + { + "category_id": 0, + "poly": [ + 296, + 230, + 455, + 230, + 455, + 261, + 296, + 261 + ], + "score": 0.142 + }, + { + "category_id": 13, + "poly": [ + 484, + 1559, + 661, + 1559, + 661, + 1633, + 484, + 1633 + ], + "score": 0.95, + "latex": "f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right)" + }, + { + "category_id": 13, + "poly": [ + 713, + 1559, + 917, + 1559, + 917, + 1633, + 713, + 1633 + ], + "score": 0.94, + "latex": "f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right)" + }, + { + "category_id": 14, + "poly": [ + 366, + 312, + 1458, + 312, + 1458, + 1182, + 366, + 1182 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } { \\gamma _ { 1 } = \\gamma _ { 2 } \\gamma _ { 2 } \\gamma _ { 3 } = \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } } & { \\gamma _ { 2 } \\gamma _ { 3 } = \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } , } \\\\ & { \\qquad \\quad + \\frac { \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 2 } } { \\mu _ { 1 } \\mu _ { 2 } \\mu _ { 3 } } \\gamma _ { 1 } ( \\frac { \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } } { \\mu _ { 2 } } + \\frac { \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } } { \\mu _ { 3 } } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad + \\frac { \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 3 } } { \\mu _ { 2 } \\mu _ { 3 } } \\gamma _ { 1 } ( \\frac { \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } } { \\mu _ { 2 } } + \\frac { \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } } { \\mu _ { 3 } } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad \\times ( \\gamma _ { 2 } + \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } + \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 3 } ) \\gamma _ { 4 } \\exp ( - \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } + \\gamma _ { 2 } \\gamma _ { 3 } ) } \\\\ & { \\qquad \\quad \\times \\exp ( - \\gamma _ { 1 } \\gamma _ { 1 } \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 3 } ) \\gamma _ { 3 } ( \\gamma _ { 3 } + \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 4 } ) \\gamma _ { 4 } \\exp ( - \\gamma _ { 2 } \\gamma _ { 3 } \\gamma _ { 2 } + \\gamma _ { 3 } \\gamma _ { 4 } ) } \\\\ & \\qquad \\quad \\times \\exp ( - \\gamma _ { 1 } \\gamma _ { 3 } \\gamma _ { 3 } \\gamma _ { 4 } ) \\gamma _ { 5 } ( \\gamma _ 3 \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 992, + 1217, + 1108, + 1217, + 1108, + 1247, + 992, + 1247 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 14, + "poly": [ + 409, + 1919, + 1288, + 1919, + 1288, + 2031, + 409, + 2031 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { \\varphi _ { 0 } ^ { * } = \\underset { \\varphi _ { 0 } } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M ^ { * } } ^ { * } } \\left[ T _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] - \\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { * } } \\left[ f ^ { * } ( T _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } \\\\ & { \\varphi _ { 1 } ^ { * } = \\underset { \\varphi _ { 1 } } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { x , a \\sim \\rho _ { M ^ { * } } ^ { * } } \\left[ T _ { \\varphi _ { 1 } } ( x , a ) \\right] - \\mathbb { E } _ { x , a \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ f ^ { * } ( T _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 494, + 1125, + 609, + 1125, + 609, + 1158, + 494, + 1158 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 395, + 1808, + 608, + 1808, + 608, + 1856, + 395, + 1856 + ], + "score": 0.92, + "latex": "f ^ { \\prime } ( \\frac { p ( x ) } { q ( x ) } ) = T _ { \\varphi ^ { * } } ( x )" + }, + { + "category_id": 13, + "poly": [ + 1092, + 1646, + 1236, + 1646, + 1236, + 1680, + 1092, + 1680 + ], + "score": 0.91, + "latex": "T _ { \\varphi } : \\mathcal { X } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 1059, + 1271, + 1154, + 1271, + 1154, + 1302, + 1059, + 1302 + ], + "score": 0.91, + "latex": "\\bar { \\alpha _ { 0 } } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 1188, + 1160, + 1340, + 1160, + 1340, + 1192, + 1188, + 1192 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 640, + 1124, + 792, + 1124, + 792, + 1158, + 640, + 1158 + ], + "score": 0.9, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 693, + 1160, + 788, + 1160, + 788, + 1191, + 693, + 1191 + ], + "score": 0.9, + "latex": "\\alpha _ { 0 } ( x , a )" + }, + { + "category_id": 14, + "poly": [ + 533, + 1748, + 1161, + 1748, + 1161, + 1799, + 533, + 1799 + ], + "score": 0.9, + "latex": "\\varphi ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi } \\mathbb { E } _ { x \\sim p ( x ) } \\left[ T _ { \\varphi } ( x ) \\right] - \\mathbb { E } _ { x \\sim q ( x ) } \\left[ f ^ { * } ( T _ { \\varphi } ( x ) ) \\right] ," + }, + { + "category_id": 13, + "poly": [ + 333, + 1681, + 415, + 1681, + 415, + 1713, + 333, + 1713 + ], + "score": 0.89, + "latex": "\\varphi \\in \\Phi" + }, + { + "category_id": 13, + "poly": [ + 1089, + 1124, + 1397, + 1124, + 1397, + 1160, + 1089, + 1160 + ], + "score": 0.89, + "latex": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 446, + 1644, + 516, + 1644, + 516, + 1683, + 446, + 1683 + ], + "score": 0.87, + "latex": "( f ^ { \\prime } ( { \\frac { p } { q } } )" + }, + { + "category_id": 13, + "poly": [ + 961, + 1189, + 1005, + 1189, + 1005, + 1213, + 961, + 1213 + ], + "score": 0.86, + "latex": "M ^ { * }" + }, + { + "category_id": 14, + "poly": [ + 347, + 1307, + 1456, + 1307, + 1456, + 1391, + 347, + 1391 + ], + "score": 0.83, + "latex": "= \\arg \\operatorname* { m a x } _ { \\theta } \\int _ { \\chi , A , \\chi } \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ) \\left( f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { \\mathbb { M } ^ { + } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) - f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathbb { M } _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { \\mathbb { M } ^ { + } } ^ { \\kappa } ( x , a ) } \\right) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + " + }, + { + "category_id": 14, + "poly": [ + 462, + 1421, + 1577, + 1421, + 1577, + 1508, + 462, + 1508 + ], + "score": 0.81, + "latex": "\\int _ { \\boldsymbol { x } , A , \\boldsymbol { \\tilde { x } } } \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) \\left( f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathrm { M } \\theta _ { t } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } ) } { \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } ) } \\right) - f ^ { \\prime } \\left( \\frac { \\rho _ { \\mathrm { M } \\theta _ { t } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) } { \\rho _ { \\mathrm { M } ^ { * } } ^ { \\kappa } ( \\boldsymbol { x } , \\boldsymbol { a } , \\boldsymbol { x } ^ { \\prime } ) } \\right) + H _ { \\mathsf { M } ^ { * } } ( \\boldsymbol { x } , \\boldsymbol { a } ) \\right) \\log M _ { \\theta } ( \\boldsymbol { x } ^ { \\prime } | \\boldsymbol { x } , \\boldsymbol { a } ) \\mathrm { d } \\boldsymbol { x } ^ { \\prime } \\mathrm { d } \\boldsymbol { a } \\mathrm { d } \\boldsymbol { x }" + }, + { + "category_id": 13, + "poly": [ + 922, + 1816, + 941, + 1816, + 941, + 1847, + 922, + 1847 + ], + "score": 0.81, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 455, + 1683, + 475, + 1683, + 475, + 1713, + 455, + 1713 + ], + "score": 0.8, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1305, + 1686, + 1327, + 1686, + 1327, + 1713, + 1305, + 1713 + ], + "score": 0.78, + "latex": "\\varphi" + }, + { + "category_id": 13, + "poly": [ + 620, + 1814, + 654, + 1814, + 654, + 1847, + 620, + 1847 + ], + "score": 0.78, + "latex": "f ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 910, + 1194, + 928, + 1194, + 928, + 1213, + 910, + 1213 + ], + "score": 0.72, + "latex": "\\kappa" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 457.0, + 228.0, + 457.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1641.0, + 445.0, + 1641.0, + 445.0, + 1684.0, + 293.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1641.0, + 1091.0, + 1641.0, + 1091.0, + 1684.0, + 517.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1641.0, + 1408.0, + 1641.0, + 1408.0, + 1684.0, + 1237.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1679.0, + 332.0, + 1679.0, + 332.0, + 1716.0, + 293.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1679.0, + 454.0, + 1679.0, + 454.0, + 1716.0, + 416.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1679.0, + 1304.0, + 1679.0, + 1304.0, + 1716.0, + 476.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 1679.0, + 1403.0, + 1679.0, + 1403.0, + 1716.0, + 1328.0, + 1716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1713.0, + 524.0, + 1713.0, + 524.0, + 1746.0, + 293.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1119.0, + 365.0, + 1119.0, + 365.0, + 1160.0, + 298.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1154.0, + 692.0, + 1154.0, + 692.0, + 1197.0, + 292.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 1154.0, + 1187.0, + 1154.0, + 1187.0, + 1197.0, + 789.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1341.0, + 1154.0, + 1408.0, + 1154.0, + 1408.0, + 1197.0, + 1341.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1184.0, + 909.0, + 1184.0, + 909.0, + 1219.0, + 294.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 1184.0, + 960.0, + 1184.0, + 960.0, + 1219.0, + 929.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 1184.0, + 1406.0, + 1184.0, + 1406.0, + 1219.0, + 1006.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1210.0, + 991.0, + 1210.0, + 991.0, + 1251.0, + 292.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1210.0, + 1409.0, + 1210.0, + 1409.0, + 1251.0, + 1109.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1240.0, + 1407.0, + 1240.0, + 1407.0, + 1274.0, + 294.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1269.0, + 1058.0, + 1269.0, + 1058.0, + 1304.0, + 295.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 1269.0, + 1272.0, + 1269.0, + 1272.0, + 1304.0, + 1155.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1544.0, + 483.0, + 1544.0, + 483.0, + 1633.0, + 289.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 1544.0, + 712.0, + 1544.0, + 712.0, + 1633.0, + 662.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1565.0, + 1164.0, + 1565.0, + 1164.0, + 1620.0, + 918.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1798.0, + 394.0, + 1798.0, + 394.0, + 1867.0, + 284.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1798.0, + 619.0, + 1798.0, + 619.0, + 1867.0, + 609.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 1798.0, + 921.0, + 1798.0, + 921.0, + 1867.0, + 655.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1798.0, + 1384.0, + 1798.0, + 1384.0, + 1867.0, + 942.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 457.0, + 228.0, + 457.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1879.0, + 448.0, + 1879.0, + 448.0, + 1922.0, + 294.0, + 1922.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 300, + 1942, + 1402, + 1942, + 1402, + 2035, + 300, + 2035 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 297, + 682, + 1408, + 682, + 1408, + 807, + 297, + 807 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 298, + 1017, + 1408, + 1017, + 1408, + 1108, + 298, + 1108 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 296, + 221, + 1409, + 221, + 1409, + 330, + 296, + 330 + ], + "score": 0.939 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 106, + 297, + 106 + ], + "score": 0.921 + }, + { + "category_id": 8, + "poly": [ + 296, + 344, + 1483, + 344, + 1483, + 654, + 296, + 654 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 295, + 1265, + 1259, + 1265, + 1259, + 1298, + 295, + 1298 + ], + "score": 0.912 + }, + { + "category_id": 0, + "poly": [ + 298, + 1196, + 996, + 1196, + 996, + 1233, + 298, + 1233 + ], + "score": 0.904 + }, + { + "category_id": 0, + "poly": [ + 299, + 1871, + 618, + 1871, + 618, + 1908, + 299, + 1908 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2113, + 835, + 2113 + ], + "score": 0.854 + }, + { + "category_id": 9, + "poly": [ + 1358, + 972, + 1402, + 972, + 1402, + 1000, + 1358, + 1000 + ], + "score": 0.83 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1121, + 1403, + 1121, + 1403, + 1149, + 1374, + 1149 + ], + "score": 0.796 + }, + { + "category_id": 1, + "poly": [ + 360, + 1320, + 1405, + 1320, + 1405, + 1833, + 360, + 1833 + ], + "score": 0.596 + }, + { + "category_id": 8, + "poly": [ + 294, + 871, + 1157, + 871, + 1157, + 972, + 294, + 972 + ], + "score": 0.574 + }, + { + "category_id": 8, + "poly": [ + 295, + 819, + 1530, + 819, + 1530, + 974, + 295, + 974 + ], + "score": 0.221 + }, + { + "category_id": 1, + "poly": [ + 286, + 819, + 1526, + 819, + 1526, + 863, + 286, + 863 + ], + "score": 0.191 + }, + { + "category_id": 13, + "poly": [ + 547, + 1385, + 684, + 1385, + 684, + 1419, + 547, + 1419 + ], + "score": 0.94, + "latex": "g ( x , a ) / \\alpha _ { M }" + }, + { + "category_id": 13, + "poly": [ + 900, + 224, + 1226, + 224, + 1226, + 297, + 900, + 297 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a ) } \\right) \\approx T _ { \\varphi _ { 1 } ^ { * } } ( x , a ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 456, + 224, + 844, + 224, + 844, + 294, + 456, + 294 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { f ^ { \\prime } \\left( \\frac { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } { \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } \\right) \\approx T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 700, + 1522, + 851, + 1522, + 851, + 1556, + 700, + 1556 + ], + "score": 0.93, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a , x ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 1029, + 1487, + 1342, + 1487, + 1342, + 1525, + 1029, + 1525 + ], + "score": 0.93, + "latex": "\\alpha _ { 0 } ( x , a ) = \\alpha _ { M _ { \\theta _ { t } } } \\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 1063, + 1583, + 1180, + 1583, + 1180, + 1616, + 1063, + 1616 + ], + "score": 0.93, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 14, + "poly": [ + 310, + 345, + 1428, + 345, + 1428, + 656, + 310, + 656 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { \\mathsf { \\Gamma } _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ) \\left( T _ { \\varphi _ { 0 } ^ { \\star } } ( x , a , x ^ { \\prime } ) - T _ { \\varphi _ { 1 } ^ { \\star } } ( x , a ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + } \\\\ & { \\qquad \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M ^ { \\star } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\left( T _ { \\varphi _ { 1 } ^ { \\star } } ( x , a ) - T _ { \\varphi _ { 0 } ^ { \\star } } ( x , a , x ^ { \\prime } ) + H _ { M ^ { \\star } } ( x , a ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x ^ { \\prime } } \\\\ & { = \\arg \\operatorname* { m a x } _ { \\theta } \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ) A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x + } \\\\ & { \\qquad \\displaystyle \\int _ { \\mathcal X , A , \\mathcal X } \\rho _ { M ^ { \\star } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) ( - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) + H _ { M ^ { \\star } } ( x , a ) ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\mathrm { d } x ^ { \\prime } \\mathrm { d } a \\mathrm { d } x . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 999, + 684, + 1072, + 684, + 1072, + 716, + 999, + 716 + ], + "score": 0.91, + "latex": "f ^ { * } ( T )" + }, + { + "category_id": 13, + "poly": [ + 375, + 293, + 885, + 293, + 885, + 331, + 375, + 331 + ], + "score": 0.91, + "latex": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , \\dot { a } , x ^ { \\prime } ) = T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) - T _ { \\varphi _ { 1 } ^ { * } } ( x , a ) ," + }, + { + "category_id": 13, + "poly": [ + 1033, + 1617, + 1150, + 1617, + 1150, + 1648, + 1033, + 1648 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 1029, + 1319, + 1407, + 1319, + 1407, + 1359, + 1029, + 1359 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\beta } ( x , a ) \\in [ 0 , c ] , \\forall a \\in \\mathcal { A } , \\forall x \\in" + }, + { + "category_id": 14, + "poly": [ + 310, + 816, + 1428, + 816, + 1428, + 976, + 310, + 976 + ], + "score": 0.9, + "latex": "\\begin{array} { r l } & { _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\left( H _ { M ^ { \\star } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } \\\\ { \\therefore } & { _ { \\varphi _ { 0 } ^ { \\star } } = \\arg \\operatorname* { m a x } _ { \\theta _ { 0 } } \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x , a , x ^ { \\prime } ) ) \\right] } \\\\ & { _ { \\varphi _ { 1 } ^ { \\star } } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 1 } } \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\times } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\times } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x , a ) ) \\right] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 975, + 1801, + 1036, + 1801, + 1036, + 1833, + 975, + 1833 + ], + "score": 0.9, + "latex": "\\alpha _ { M _ { \\theta _ { t } } }" + }, + { + "category_id": 13, + "poly": [ + 295, + 746, + 506, + 746, + 506, + 781, + 295, + 781 + ], + "score": 0.89, + "latex": "- \\log ( 1 - D _ { \\varphi } ( u ) )" + }, + { + "category_id": 13, + "poly": [ + 1135, + 747, + 1348, + 747, + 1348, + 779, + 1135, + 779 + ], + "score": 0.89, + "latex": "f ^ { \\prime } ( x ) \\leq 0 , \\forall x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 1006, + 1650, + 1067, + 1650, + 1067, + 1680, + 1006, + 1680 + ], + "score": 0.89, + "latex": "\\alpha _ { M _ { \\theta _ { t } } }" + }, + { + "category_id": 13, + "poly": [ + 397, + 1680, + 457, + 1680, + 457, + 1711, + 397, + 1711 + ], + "score": 0.89, + "latex": "\\alpha _ { M _ { \\theta _ { t } } }" + }, + { + "category_id": 13, + "poly": [ + 1320, + 685, + 1360, + 685, + 1360, + 717, + 1320, + 717 + ], + "score": 0.88, + "latex": "D _ { \\varphi }" + }, + { + "category_id": 13, + "poly": [ + 1030, + 1417, + 1062, + 1417, + 1062, + 1447, + 1030, + 1447 + ], + "score": 0.88, + "latex": "\\beta ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 957, + 716, + 1186, + 716, + 1186, + 749, + 957, + 749 + ], + "score": 0.88, + "latex": "T _ { \\varphi } ( u ) = \\log D _ { \\varphi } ( u )" + }, + { + "category_id": 13, + "poly": [ + 391, + 1019, + 1141, + 1019, + 1141, + 1059, + 391, + 1059 + ], + "score": 0.88, + "latex": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ = \\ \\log D _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\ - \\ \\log D _ { \\varphi _ { 1 } ^ { * } } ( x , a ) , \\ \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { \\star } } } ^ { \\kappa } } [ \\cdot ]" + }, + { + "category_id": 13, + "poly": [ + 1248, + 716, + 1406, + 716, + 1406, + 749, + 1248, + 749 + ], + "score": 0.88, + "latex": "f ^ { * } ( T _ { \\varphi } ( u ) ) =" + }, + { + "category_id": 13, + "poly": [ + 974, + 1387, + 1019, + 1387, + 1019, + 1413, + 974, + 1413 + ], + "score": 0.88, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 944, + 1416, + 989, + 1416, + 989, + 1444, + 944, + 1444 + ], + "score": 0.88, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1318, + 245, + 1352, + 245, + 1352, + 277, + 1318, + 277 + ], + "score": 0.88, + "latex": "\\varphi _ { 0 } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 655, + 1355, + 699, + 1355, + 699, + 1382, + 655, + 1382 + ], + "score": 0.88, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1061, + 1387, + 1094, + 1387, + 1094, + 1417, + 1061, + 1417 + ], + "score": 0.88, + "latex": "\\beta ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 297, + 294, + 331, + 294, + 331, + 326, + 297, + 326 + ], + "score": 0.88, + "latex": "\\varphi _ { 1 } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1018, + 1357, + 1038, + 1357, + 1038, + 1386, + 1018, + 1386 + ], + "score": 0.85, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 552, + 686, + 571, + 686, + 571, + 715, + 552, + 715 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 920, + 685, + 943, + 685, + 943, + 711, + 920, + 711 + ], + "score": 0.84, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 297, + 1062, + 462, + 1062, + 462, + 1109, + 297, + 1109 + ], + "score": 0.82, + "latex": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } [ \\cdot" + }, + { + "category_id": 13, + "poly": [ + 1260, + 1420, + 1279, + 1420, + 1279, + 1448, + 1260, + 1448 + ], + "score": 0.79, + "latex": "\\rho" + }, + { + "category_id": 13, + "poly": [ + 523, + 715, + 943, + 715, + 943, + 749, + 523, + 749 + ], + "score": 0.78, + "latex": "f ( u ) = u \\log u - ( u + 1 ) \\log ( u + 1 )" + }, + { + "category_id": 13, + "poly": [ + 1137, + 296, + 1154, + 296, + 1154, + 320, + 1137, + 320 + ], + "score": 0.77, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 397, + 1356, + 424, + 1356, + 424, + 1382, + 397, + 1382 + ], + "score": 0.75, + "latex": "\\mathcal { X }" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1193.0, + 1000.0, + 1193.0, + 1000.0, + 1238.0, + 291.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1867.0, + 624.0, + 1867.0, + 624.0, + 1914.0, + 294.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1125.0, + 1403.0, + 1125.0, + 1403.0, + 1149.0, + 1377.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1939.0, + 1406.0, + 1939.0, + 1406.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1970.0, + 1406.0, + 1970.0, + 1406.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 1406.0, + 2002.0, + 1406.0, + 2039.0, + 295.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 681.0, + 551.0, + 681.0, + 551.0, + 718.0, + 295.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 681.0, + 919.0, + 681.0, + 919.0, + 718.0, + 572.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 681.0, + 998.0, + 681.0, + 998.0, + 718.0, + 944.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 681.0, + 1319.0, + 681.0, + 1319.0, + 718.0, + 1073.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 681.0, + 1408.0, + 681.0, + 1408.0, + 718.0, + 1361.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 714.0, + 522.0, + 714.0, + 522.0, + 750.0, + 295.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 714.0, + 956.0, + 714.0, + 956.0, + 750.0, + 944.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 714.0, + 1247.0, + 714.0, + 1247.0, + 750.0, + 1187.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 743.0, + 1134.0, + 743.0, + 1134.0, + 781.0, + 507.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 743.0, + 1406.0, + 743.0, + 1406.0, + 781.0, + 1349.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 773.0, + 1285.0, + 773.0, + 1285.0, + 808.0, + 292.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1016.0, + 1408.0, + 1016.0, + 1408.0, + 1058.0, + 1142.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1053.0, + 296.0, + 1053.0, + 296.0, + 1117.0, + 284.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1053.0, + 489.0, + 1053.0, + 489.0, + 1117.0, + 463.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.75, + 997.0, + 1131.75, + 997.0, + 1131.75, + 1077.0, + 295.75, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 238.0, + 455.0, + 238.0, + 455.0, + 279.0, + 294.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 231.0, + 899.0, + 231.0, + 899.0, + 286.0, + 845.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 234.0, + 1317.0, + 234.0, + 1317.0, + 286.0, + 1227.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 234.0, + 1407.0, + 234.0, + 1407.0, + 286.0, + 1353.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 287.0, + 374.0, + 287.0, + 374.0, + 332.0, + 332.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 287.0, + 1136.0, + 287.0, + 1136.0, + 332.0, + 886.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 287.0, + 1208.0, + 287.0, + 1208.0, + 332.0, + 1155.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1265.0, + 1261.0, + 1265.0, + 1261.0, + 1300.0, + 295.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1317.0, + 1028.0, + 1317.0, + 1028.0, + 1366.0, + 358.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1353.0, + 396.0, + 1353.0, + 396.0, + 1390.0, + 392.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1353.0, + 654.0, + 1353.0, + 654.0, + 1390.0, + 425.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1353.0, + 1017.0, + 1353.0, + 1017.0, + 1390.0, + 700.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1353.0, + 1404.0, + 1353.0, + 1404.0, + 1390.0, + 1039.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1386.0, + 546.0, + 1386.0, + 546.0, + 1420.0, + 396.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1386.0, + 973.0, + 1386.0, + 973.0, + 1420.0, + 685.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1386.0, + 1060.0, + 1386.0, + 1060.0, + 1420.0, + 1020.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1386.0, + 1405.0, + 1386.0, + 1405.0, + 1420.0, + 1095.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1416.0, + 943.0, + 1416.0, + 943.0, + 1450.0, + 393.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1416.0, + 1029.0, + 1416.0, + 1029.0, + 1450.0, + 990.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1416.0, + 1259.0, + 1416.0, + 1259.0, + 1450.0, + 1063.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1280.0, + 1416.0, + 1405.0, + 1416.0, + 1405.0, + 1450.0, + 1280.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1447.0, + 1027.0, + 1447.0, + 1027.0, + 1481.0, + 395.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1483.0, + 1028.0, + 1483.0, + 1028.0, + 1526.0, + 357.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.0, + 1483.0, + 1407.0, + 1483.0, + 1407.0, + 1526.0, + 1343.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1521.0, + 699.0, + 1521.0, + 699.0, + 1559.0, + 393.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1521.0, + 1405.0, + 1521.0, + 1405.0, + 1559.0, + 852.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1553.0, + 1406.0, + 1553.0, + 1406.0, + 1586.0, + 396.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1581.0, + 1062.0, + 1581.0, + 1062.0, + 1622.0, + 392.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1581.0, + 1408.0, + 1581.0, + 1408.0, + 1622.0, + 1181.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1611.0, + 1032.0, + 1611.0, + 1032.0, + 1652.0, + 392.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 1611.0, + 1408.0, + 1611.0, + 1408.0, + 1652.0, + 1151.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1639.0, + 1005.0, + 1639.0, + 1005.0, + 1683.0, + 391.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1639.0, + 1410.0, + 1639.0, + 1410.0, + 1683.0, + 1068.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1674.0, + 396.0, + 1674.0, + 396.0, + 1713.0, + 393.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 1674.0, + 1407.0, + 1674.0, + 1407.0, + 1713.0, + 458.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1705.0, + 1406.0, + 1705.0, + 1406.0, + 1737.0, + 392.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1737.0, + 1405.0, + 1737.0, + 1405.0, + 1767.0, + 393.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1764.0, + 1404.0, + 1764.0, + 1404.0, + 1800.0, + 394.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1789.0, + 974.0, + 1789.0, + 974.0, + 1839.0, + 390.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1789.0, + 1051.0, + 1789.0, + 1051.0, + 1839.0, + 1037.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 815.0, + 309.0, + 815.0, + 309.0, + 871.0, + 290.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1429.0, + 815.0, + 1536.0, + 815.0, + 1536.0, + 871.0, + 1429.0, + 871.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 21, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1327, + 1405, + 1327, + 1405, + 1606, + 296, + 1606 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1405, + 228, + 1405, + 415, + 298, + 415 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1787, + 1407, + 1787, + 1407, + 2041, + 297, + 2041 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 294, + 1105, + 1402, + 1105, + 1402, + 1179, + 294, + 1179 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 298, + 1253, + 1403, + 1253, + 1403, + 1317, + 298, + 1317 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 667, + 1190, + 1034, + 1190, + 1034, + 1238, + 667, + 1238 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 297, + 526, + 866, + 526, + 866, + 559, + 297, + 559 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 298, + 1052, + 925, + 1052, + 925, + 1087, + 298, + 1087 + ], + "score": 0.922 + }, + { + "category_id": 0, + "poly": [ + 299, + 985, + 614, + 985, + 614, + 1021, + 299, + 1021 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 300, + 457, + 802, + 457, + 802, + 494, + 300, + 494 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1354, + 1195, + 1401, + 1195, + 1401, + 1225, + 1354, + 1225 + ], + "score": 0.902 + }, + { + "category_id": 8, + "poly": [ + 297, + 1613, + 1483, + 1613, + 1483, + 1714, + 297, + 1714 + ], + "score": 0.879 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.861 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 856, + 75, + 856, + 105, + 298, + 105 + ], + "score": 0.848 + }, + { + "category_id": 0, + "poly": [ + 298, + 591, + 947, + 591, + 947, + 624, + 298, + 624 + ], + "score": 0.825 + }, + { + "category_id": 1, + "poly": [ + 302, + 1729, + 1271, + 1729, + 1271, + 1778, + 302, + 1778 + ], + "score": 0.756 + }, + { + "category_id": 0, + "poly": [ + 298, + 694, + 397, + 694, + 397, + 720, + 298, + 720 + ], + "score": 0.343 + }, + { + "category_id": 8, + "poly": [ + 296, + 1613, + 1483, + 1613, + 1483, + 1714, + 296, + 1714 + ], + "score": 0.34 + }, + { + "category_id": 1, + "poly": [ + 308, + 732, + 1210, + 732, + 1210, + 919, + 308, + 919 + ], + "score": 0.273 + }, + { + "category_id": 1, + "poly": [ + 297, + 632, + 692, + 632, + 692, + 692, + 297, + 692 + ], + "score": 0.219 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 856, + 75, + 856, + 105, + 298, + 105 + ], + "score": 0.101 + }, + { + "category_id": 13, + "poly": [ + 379, + 1730, + 493, + 1730, + 493, + 1777, + 379, + 1777 + ], + "score": 0.93, + "latex": "\\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\mathbf { u } \\left[ \\cdot \\right] }" + }, + { + "category_id": 14, + "poly": [ + 666, + 1190, + 1033, + 1190, + 1033, + 1241, + 666, + 1241 + ], + "score": 0.93, + "latex": "\\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 1094, + 1508, + 1256, + 1508, + 1256, + 1542, + 1094, + 1542 + ], + "score": 0.93, + "latex": "\\mathbf { u } \\sim \\mathcal { N } ( 0 , \\epsilon _ { D } )" + }, + { + "category_id": 13, + "poly": [ + 829, + 1105, + 877, + 1105, + 877, + 1153, + 829, + 1153 + ], + "score": 0.93, + "latex": "\\textstyle { \\frac { p ( x ) } { q ( x ) } }" + }, + { + "category_id": 13, + "poly": [ + 1155, + 2002, + 1395, + 2002, + 1395, + 2043, + 1155, + 2043 + ], + "score": 0.92, + "latex": "{ \\cal A } \\ = \\ Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } \\ - \\ V _ { M _ { \\theta _ { t } } } ^ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 830, + 1540, + 882, + 1540, + 882, + 1580, + 830, + 1580 + ], + "score": 0.92, + "latex": "\\rho _ { M ^ { * } } ^ { \\hat { \\mu } }" + }, + { + "category_id": 13, + "poly": [ + 760, + 1414, + 809, + 1414, + 809, + 1445, + 760, + 1445 + ], + "score": 0.91, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 727, + 1544, + 779, + 1544, + 779, + 1580, + 727, + 1580 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 618, + 1358, + 694, + 1358, + 694, + 1388, + 618, + 1388 + ], + "score": 0.91, + "latex": "\\hat { \\mu } \\approx \\bar { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 667, + 2001, + 1057, + 2001, + 1057, + 2039, + 667, + 2039 + ], + "score": 0.91, + "latex": "\\left[ \\bar { A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x , a , x ^ { \\prime } ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 610, + 1941, + 947, + 1941, + 947, + 1975, + 610, + 1975 + ], + "score": 0.91, + "latex": "r ( x , a , x ^ { \\prime } ) ~ = ~ \\log D ( x , a , x ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 1321, + 1481, + 1402, + 1481, + 1402, + 1511, + 1321, + 1511 + ], + "score": 0.91, + "latex": "\\epsilon _ { \\mu } > 0" + }, + { + "category_id": 13, + "poly": [ + 1235, + 1416, + 1293, + 1416, + 1293, + 1453, + 1235, + 1453 + ], + "score": 0.9, + "latex": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 327, + 1546, + 368, + 1546, + 368, + 1579, + 327, + 1579 + ], + "score": 0.9, + "latex": "D _ { \\varphi }" + }, + { + "category_id": 13, + "poly": [ + 711, + 734, + 790, + 734, + 790, + 769, + 711, + 769 + ], + "score": 0.9, + "latex": "\\{ \\mathcal { D } _ { \\pi _ { \\phi } } \\}" + }, + { + "category_id": 13, + "poly": [ + 460, + 827, + 511, + 827, + 511, + 860, + 460, + 860 + ], + "score": 0.89, + "latex": "\\mathcal { D } _ { \\pi _ { \\phi } }" + }, + { + "category_id": 13, + "poly": [ + 962, + 1912, + 1004, + 1912, + 1004, + 1941, + 962, + 1941 + ], + "score": 0.89, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 811, + 1820, + 854, + 1820, + 854, + 1850, + 811, + 1850 + ], + "score": 0.89, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 924, + 827, + 966, + 827, + 966, + 857, + 924, + 857 + ], + "score": 0.88, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 496, + 858, + 537, + 858, + 537, + 887, + 496, + 887 + ], + "score": 0.87, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1185, + 1550, + 1218, + 1550, + 1218, + 1578, + 1185, + 1578 + ], + "score": 0.87, + "latex": "\\varphi _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1255, + 1387, + 1306, + 1387, + 1306, + 1416, + 1255, + 1416 + ], + "score": 0.87, + "latex": "\\rho _ { M ^ { * } } ^ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 1088, + 735, + 1205, + 735, + 1205, + 769, + 1088, + 769 + ], + "score": 0.87, + "latex": "\\pi _ { \\phi } , \\phi \\in \\Phi" + }, + { + "category_id": 14, + "poly": [ + 310, + 1611, + 1428, + 1611, + 1428, + 1718, + 310, + 1718 + ], + "score": 0.87, + "latex": "\\begin{array} { r l } & { \\gamma _ { 0 } ^ { * } = \\arg \\underset { \\varphi _ { 0 } } { \\operatorname* { m a x } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) ) \\right] } \\\\ & { \\gamma _ { 1 } ^ { * } = \\arg \\underset { \\varphi _ { 1 } } { \\operatorname* { m a x } } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\mathrm { s } } , \\mathrm { ~ u } } \\left[ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) \\right] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 384, + 796, + 469, + 796, + 469, + 824, + 384, + 824 + ], + "score": 0.86, + "latex": "\\mathrm { i } = 1 { : } \\mathrm { N }" + }, + { + "category_id": 13, + "poly": [ + 979, + 1943, + 1021, + 1943, + 1021, + 1972, + 979, + 1972 + ], + "score": 0.86, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1268, + 1549, + 1301, + 1549, + 1301, + 1578, + 1268, + 1578 + ], + "score": 0.86, + "latex": "\\varphi _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 960, + 1549, + 989, + 1549, + 989, + 1580, + 960, + 1580 + ], + "score": 0.85, + "latex": "\\epsilon _ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 1187, + 1256, + 1213, + 1256, + 1213, + 1282, + 1187, + 1282 + ], + "score": 0.85, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1069, + 1731, + 1263, + 1731, + 1263, + 1766, + 1069, + 1766 + ], + "score": 0.84, + "latex": "\\mathbf { u } = [ u _ { x } , u _ { a } , u _ { x ^ { \\prime } } ] ." + }, + { + "category_id": 13, + "poly": [ + 458, + 1387, + 479, + 1387, + 479, + 1416, + 458, + 1416 + ], + "score": 0.84, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 580, + 1415, + 600, + 1415, + 600, + 1445, + 580, + 1445 + ], + "score": 0.84, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 432, + 1449, + 766, + 1449, + 766, + 1482, + 432, + 1482 + ], + "score": 0.83, + "latex": "\\forall x \\in \\mathcal { X } , \\forall a \\in \\mathcal { A } , \\mu ( a | x ) > 0" + }, + { + "category_id": 13, + "poly": [ + 1032, + 826, + 1110, + 826, + 1110, + 862, + 1032, + 862 + ], + "score": 0.83, + "latex": "\\{ \\mathcal { D } _ { \\pi _ { \\phi } } \\}" + }, + { + "category_id": 13, + "poly": [ + 805, + 1387, + 832, + 1387, + 832, + 1411, + 805, + 1411 + ], + "score": 0.83, + "latex": "\\mathcal { D }" + }, + { + "category_id": 13, + "poly": [ + 1381, + 1451, + 1401, + 1451, + 1401, + 1481, + 1381, + 1481 + ], + "score": 0.82, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 998, + 1358, + 1024, + 1358, + 1024, + 1384, + 998, + 1384 + ], + "score": 0.81, + "latex": "\\mathcal { D }" + }, + { + "category_id": 13, + "poly": [ + 1090, + 1330, + 1116, + 1330, + 1116, + 1356, + 1090, + 1356 + ], + "score": 0.8, + "latex": "\\mathcal { D }" + }, + { + "category_id": 13, + "poly": [ + 508, + 1943, + 535, + 1943, + 535, + 1969, + 508, + 1969 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 642, + 1392, + 661, + 1392, + 661, + 1412, + 642, + 1412 + ], + "score": 0.78, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 375, + 1361, + 395, + 1361, + 395, + 1388, + 375, + 1388 + ], + "score": 0.77, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 705, + 768, + 747, + 768, + 747, + 796, + 705, + 796 + ], + "score": 0.77, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1192, + 1825, + 1213, + 1825, + 1213, + 1847, + 1192, + 1847 + ], + "score": 0.76, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 296, + 662, + 319, + 662, + 319, + 688, + 296, + 688 + ], + "score": 0.76, + "latex": "\\Phi" + }, + { + "category_id": 13, + "poly": [ + 730, + 1731, + 1019, + 1731, + 1019, + 1778, + 730, + 1778 + ], + "score": 0.76, + "latex": "\\mathbb { E } _ { x , a , x ^ { \\prime } \\sim \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\mathbf { u } \\sim \\mathcal { N } ( 0 , \\epsilon _ { D } ) } [ \\cdot ]" + }, + { + "category_id": 13, + "poly": [ + 880, + 1917, + 900, + 1917, + 900, + 1939, + 880, + 1939 + ], + "score": 0.75, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 485, + 662, + 513, + 662, + 513, + 688, + 485, + 688 + ], + "score": 0.73, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 1026, + 860, + 1077, + 860, + 1077, + 893, + 1026, + 893 + ], + "score": 0.66, + "latex": "\\mathcal { D } _ { \\pi _ { \\phi } }" + }, + { + "category_id": 13, + "poly": [ + 347, + 1284, + 529, + 1284, + 529, + 1319, + 347, + 1319 + ], + "score": 0.57, + "latex": "\\bar { f } ^ { \\prime } ( 1 ) = \\log { 0 . 5 } )" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1054.0, + 929.0, + 1054.0, + 929.0, + 1090.0, + 294.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 984.0, + 617.0, + 984.0, + 617.0, + 1025.0, + 293.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 454.0, + 807.0, + 454.0, + 807.0, + 501.0, + 293.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 589.0, + 950.0, + 589.0, + 950.0, + 628.0, + 295.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 689.0, + 402.0, + 689.0, + 402.0, + 725.0, + 294.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1328.0, + 1089.0, + 1328.0, + 1089.0, + 1361.0, + 295.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 1328.0, + 1405.0, + 1328.0, + 1405.0, + 1361.0, + 1117.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1353.0, + 374.0, + 1353.0, + 374.0, + 1390.0, + 294.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1353.0, + 617.0, + 1353.0, + 617.0, + 1390.0, + 396.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1353.0, + 997.0, + 1353.0, + 997.0, + 1390.0, + 695.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1353.0, + 1405.0, + 1353.0, + 1405.0, + 1390.0, + 1025.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1378.0, + 457.0, + 1378.0, + 457.0, + 1427.0, + 290.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 1378.0, + 641.0, + 1378.0, + 641.0, + 1427.0, + 480.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 1378.0, + 804.0, + 1378.0, + 804.0, + 1427.0, + 662.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 1378.0, + 1254.0, + 1378.0, + 1254.0, + 1427.0, + 833.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1378.0, + 1411.0, + 1378.0, + 1411.0, + 1427.0, + 1307.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1401.0, + 579.0, + 1401.0, + 579.0, + 1459.0, + 289.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 1401.0, + 759.0, + 1401.0, + 759.0, + 1459.0, + 601.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1401.0, + 1234.0, + 1401.0, + 1234.0, + 1459.0, + 810.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 1401.0, + 1406.0, + 1401.0, + 1406.0, + 1459.0, + 1294.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1448.0, + 431.0, + 1448.0, + 431.0, + 1485.0, + 294.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1448.0, + 1380.0, + 1448.0, + 1380.0, + 1485.0, + 767.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1448.0, + 1407.0, + 1448.0, + 1407.0, + 1485.0, + 1402.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1476.0, + 1320.0, + 1476.0, + 1320.0, + 1511.0, + 294.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1476.0, + 1406.0, + 1476.0, + 1406.0, + 1511.0, + 1403.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1507.0, + 1093.0, + 1507.0, + 1093.0, + 1544.0, + 294.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1507.0, + 1405.0, + 1507.0, + 1405.0, + 1544.0, + 1257.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1540.0, + 326.0, + 1540.0, + 326.0, + 1584.0, + 291.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1540.0, + 726.0, + 1540.0, + 726.0, + 1584.0, + 369.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1540.0, + 829.0, + 1540.0, + 829.0, + 1584.0, + 780.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1540.0, + 959.0, + 1540.0, + 959.0, + 1584.0, + 883.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1540.0, + 1184.0, + 1540.0, + 1184.0, + 1584.0, + 990.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 1540.0, + 1267.0, + 1540.0, + 1267.0, + 1584.0, + 1219.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1540.0, + 1408.0, + 1540.0, + 1408.0, + 1584.0, + 1302.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1576.0, + 401.0, + 1576.0, + 401.0, + 1606.0, + 294.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 229.0, + 1406.0, + 229.0, + 1406.0, + 264.0, + 293.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 261.0, + 1406.0, + 261.0, + 1406.0, + 297.0, + 295.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 289.0, + 1406.0, + 289.0, + 1406.0, + 327.0, + 292.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 320.0, + 1408.0, + 320.0, + 1408.0, + 356.0, + 293.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 350.0, + 1405.0, + 350.0, + 1405.0, + 389.0, + 293.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 383.0, + 1031.0, + 383.0, + 1031.0, + 419.0, + 295.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1787.0, + 1407.0, + 1787.0, + 1407.0, + 1822.0, + 295.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1816.0, + 810.0, + 1816.0, + 810.0, + 1854.0, + 294.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1816.0, + 1191.0, + 1816.0, + 1191.0, + 1854.0, + 855.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 1816.0, + 1409.0, + 1816.0, + 1409.0, + 1854.0, + 1214.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1851.0, + 1405.0, + 1851.0, + 1405.0, + 1882.0, + 295.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1880.0, + 1407.0, + 1880.0, + 1407.0, + 1914.0, + 294.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1907.0, + 879.0, + 1907.0, + 879.0, + 1947.0, + 292.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 1907.0, + 961.0, + 1907.0, + 961.0, + 1947.0, + 901.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1907.0, + 1407.0, + 1907.0, + 1407.0, + 1947.0, + 1005.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1937.0, + 507.0, + 1937.0, + 507.0, + 1978.0, + 294.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1937.0, + 609.0, + 1937.0, + 609.0, + 1978.0, + 536.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 1937.0, + 978.0, + 1937.0, + 978.0, + 1978.0, + 948.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 1937.0, + 1409.0, + 1937.0, + 1409.0, + 1978.0, + 1022.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1969.0, + 1407.0, + 1969.0, + 1407.0, + 2004.0, + 292.0, + 2004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1999.0, + 666.0, + 1999.0, + 666.0, + 2039.0, + 291.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1999.0, + 1154.0, + 1999.0, + 1154.0, + 2039.0, + 1058.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1396.0, + 2002.0, + 1410.0, + 2002.0, + 1410.0, + 2048.0, + 1396.0, + 2048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.75, + 1991.0, + 1302.75, + 1991.0, + 1302.75, + 2048.5, + 1224.75, + 2048.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1104.0, + 828.0, + 1104.0, + 828.0, + 1147.0, + 293.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 1104.0, + 1407.0, + 1104.0, + 1407.0, + 1147.0, + 878.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1124.0, + 828.0, + 1124.0, + 828.0, + 1157.0, + 825.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1147.0, + 1027.0, + 1147.0, + 1027.0, + 1181.0, + 296.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1250.0, + 1186.0, + 1250.0, + 1186.0, + 1291.0, + 292.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 1250.0, + 1404.0, + 1250.0, + 1404.0, + 1291.0, + 1214.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1284.0, + 346.0, + 1284.0, + 346.0, + 1320.0, + 294.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 1284.0, + 543.0, + 1284.0, + 543.0, + 1320.0, + 530.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 526.0, + 868.0, + 526.0, + 868.0, + 562.0, + 296.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1720.0, + 729.0, + 1720.0, + 729.0, + 1790.0, + 494.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1720.0, + 1068.0, + 1720.0, + 1068.0, + 1790.0, + 1020.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 1720.0, + 1280.0, + 1720.0, + 1280.0, + 1790.0, + 1264.0, + 1790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1719.0, + 458.0, + 1719.0, + 458.0, + 1781.0, + 300.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 729.0, + 710.0, + 729.0, + 710.0, + 772.0, + 305.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 729.0, + 1087.0, + 729.0, + 1087.0, + 772.0, + 791.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 729.0, + 1212.0, + 729.0, + 1212.0, + 772.0, + 1206.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 762.0, + 704.0, + 762.0, + 704.0, + 799.0, + 305.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 762.0, + 751.0, + 762.0, + 751.0, + 799.0, + 748.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 794.0, + 383.0, + 794.0, + 383.0, + 828.0, + 305.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 794.0, + 511.0, + 794.0, + 511.0, + 828.0, + 470.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 827.0, + 340.0, + 827.0, + 340.0, + 857.0, + 308.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 817.0, + 459.0, + 817.0, + 459.0, + 866.0, + 378.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 817.0, + 923.0, + 817.0, + 923.0, + 866.0, + 512.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 817.0, + 1031.0, + 817.0, + 1031.0, + 866.0, + 967.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 817.0, + 1115.0, + 817.0, + 1115.0, + 866.0, + 1111.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 856.0, + 340.0, + 856.0, + 340.0, + 889.0, + 306.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 846.0, + 495.0, + 846.0, + 495.0, + 900.0, + 377.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 846.0, + 1025.0, + 846.0, + 1025.0, + 900.0, + 538.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 846.0, + 1085.0, + 846.0, + 1085.0, + 900.0, + 1078.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 887.0, + 438.0, + 887.0, + 438.0, + 920.0, + 306.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 627.0, + 380.0, + 627.0, + 380.0, + 664.0, + 293.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 659.0, + 484.0, + 659.0, + 484.0, + 695.0, + 320.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 659.0, + 693.0, + 659.0, + 693.0, + 695.0, + 514.0, + 695.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 22, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 1084, + 1407, + 1084, + 1407, + 1448, + 295, + 1448 + ], + "score": 0.98 + }, + { + "category_id": 3, + "poly": [ + 308, + 1545, + 1399, + 1545, + 1399, + 1945, + 308, + 1945 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 306, + 403, + 1407, + 403, + 1407, + 953, + 306, + 953 + ], + "score": 0.965 + }, + { + "category_id": 4, + "poly": [ + 477, + 1996, + 1218, + 1996, + 1218, + 2029, + 477, + 2029 + ], + "score": 0.93 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2112, + 834, + 2112 + ], + "score": 0.878 + }, + { + "category_id": 0, + "poly": [ + 300, + 228, + 708, + 228, + 708, + 262, + 300, + 262 + ], + "score": 0.865 + }, + { + "category_id": 1, + "poly": [ + 295, + 297, + 1100, + 297, + 1100, + 358, + 295, + 358 + ], + "score": 0.749 + }, + { + "category_id": 0, + "poly": [ + 299, + 269, + 374, + 269, + 374, + 297, + 299, + 297 + ], + "score": 0.725 + }, + { + "category_id": 0, + "poly": [ + 297, + 360, + 397, + 360, + 397, + 388, + 297, + 388 + ], + "score": 0.68 + }, + { + "category_id": 1, + "poly": [ + 297, + 360, + 397, + 360, + 397, + 388, + 297, + 388 + ], + "score": 0.131 + }, + { + "category_id": 7, + "poly": [ + 300, + 939, + 433, + 939, + 433, + 967, + 300, + 967 + ], + "score": 0.127 + }, + { + "category_id": 1, + "poly": [ + 300, + 939, + 433, + 939, + 433, + 967, + 300, + 967 + ], + "score": 0.126 + }, + { + "category_id": 13, + "poly": [ + 1002, + 1298, + 1085, + 1298, + 1085, + 1334, + 1002, + 1334 + ], + "score": 0.93, + "latex": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 595, + 633, + 645, + 633, + 645, + 664, + 595, + 664 + ], + "score": 0.92, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 1315, + 526, + 1372, + 526, + 1372, + 573, + 1315, + 573 + ], + "score": 0.92, + "latex": "\\rho _ { M _ { \\theta _ { t } } } ^ { \\hat { \\mu } }" + }, + { + "category_id": 13, + "poly": [ + 705, + 816, + 754, + 816, + 754, + 848, + 705, + 848 + ], + "score": 0.92, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 984, + 907, + 1057, + 907, + 1057, + 942, + 984, + 942 + ], + "score": 0.91, + "latex": "M _ { \\theta _ { t + 1 } }" + }, + { + "category_id": 13, + "poly": [ + 602, + 569, + 653, + 569, + 653, + 606, + 602, + 606 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 367, + 1335, + 1346, + 1335, + 1346, + 1374, + 367, + 1374 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { Q _ { M _ { \\theta _ { \\star } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 0 } ^ { \\star } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , \\bar { M } _ { \\theta _ { t } } \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 787, + 532, + 839, + 532, + 839, + 566, + 787, + 566 + ], + "score": 0.91, + "latex": "D _ { \\varphi _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 692, + 495, + 746, + 495, + 746, + 529, + 692, + 529 + ], + "score": 0.91, + "latex": "\\mathcal { D } _ { \\mathrm { g e n } }" + }, + { + "category_id": 13, + "poly": [ + 1253, + 1088, + 1406, + 1088, + 1406, + 1127, + 1253, + 1127 + ], + "score": 0.91, + "latex": "{ \\cal V } _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a ) =" + }, + { + "category_id": 13, + "poly": [ + 683, + 533, + 734, + 533, + 734, + 566, + 683, + 566 + ], + "score": 0.91, + "latex": "D _ { \\varphi _ { 0 } }" + }, + { + "category_id": 13, + "poly": [ + 590, + 907, + 639, + 907, + 639, + 938, + 590, + 938 + ], + "score": 0.91, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 694, + 297, + 747, + 297, + 747, + 332, + 694, + 332 + ], + "score": 0.91, + "latex": "\\rho _ { M ^ { * } } ^ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 492, + 571, + 551, + 571, + 551, + 604, + 492, + 604 + ], + "score": 0.91, + "latex": "\\mathcal { D } _ { \\mathrm { g e n } }" + }, + { + "category_id": 13, + "poly": [ + 1097, + 603, + 1149, + 603, + 1149, + 636, + 1097, + 636 + ], + "score": 0.9, + "latex": "D _ { \\varphi _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 705, + 756, + 754, + 756, + 754, + 787, + 705, + 787 + ], + "score": 0.9, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 705, + 434, + 756, + 434, + 756, + 466, + 705, + 466 + ], + "score": 0.9, + "latex": "M _ { \\theta _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 923, + 494, + 972, + 494, + 972, + 527, + 923, + 527 + ], + "score": 0.9, + "latex": "M _ { \\theta _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 1120, + 1296, + 1396, + 1296, + 1396, + 1337, + 1120, + 1337 + ], + "score": 0.9, + "latex": "A _ { \\varphi _ { 0 } ^ { \\ast } , \\varphi _ { 1 } ^ { \\ast } } = Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } - V _ { M _ { \\theta _ { t } } } ^ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 832, + 571, + 894, + 571, + 894, + 602, + 832, + 602 + ], + "score": 0.89, + "latex": "\\mathcal { D } _ { \\mathrm { r e a l } }" + }, + { + "category_id": 13, + "poly": [ + 740, + 1176, + 783, + 1176, + 783, + 1206, + 740, + 1206 + ], + "score": 0.89, + "latex": "M _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 986, + 602, + 1038, + 602, + 1038, + 636, + 986, + 636 + ], + "score": 0.89, + "latex": "D _ { \\varphi _ { 0 } }" + }, + { + "category_id": 13, + "poly": [ + 385, + 464, + 499, + 464, + 499, + 492, + 385, + 492 + ], + "score": 0.89, + "latex": "t = 1 : N" + }, + { + "category_id": 13, + "poly": [ + 297, + 298, + 358, + 298, + 358, + 328, + 297, + 328 + ], + "score": 0.88, + "latex": "\\mathcal { D } _ { \\mathrm { r e a l } }" + }, + { + "category_id": 13, + "poly": [ + 916, + 603, + 975, + 603, + 975, + 634, + 916, + 634 + ], + "score": 0.88, + "latex": "\\mathcal { D } _ { \\mathrm { g e n } }" + }, + { + "category_id": 13, + "poly": [ + 408, + 724, + 824, + 724, + 824, + 759, + 408, + 759 + ], + "score": 0.88, + "latex": "p _ { 0 } < \\mathbb { E } _ { \\mathcal { D } _ { \\mathrm { g e n } } } \\left[ D _ { \\varphi _ { 0 } } ( x _ { t } , \\hat { a _ { t } } , x _ { t + 1 } ) \\right] < p _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1128, + 538, + 1128, + 538, + 1180, + 297, + 1180 + ], + "score": 0.86, + "latex": "\\mathbb { E } _ { M _ { \\bar { \\theta } } } \\left[ Q _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 473, + 604, + 498, + 604, + 498, + 633, + 473, + 633 + ], + "score": 0.86, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 895, + 1416, + 921, + 1416, + 921, + 1446, + 895, + 1446 + ], + "score": 0.85, + "latex": "Q" + }, + { + "category_id": 13, + "poly": [ + 435, + 496, + 456, + 496, + 456, + 526, + 435, + 526 + ], + "score": 0.85, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 478, + 880, + 512, + 880, + 512, + 908, + 478, + 908 + ], + "score": 0.84, + "latex": "g _ { \\mathrm { s l } }" + }, + { + "category_id": 13, + "poly": [ + 824, + 911, + 857, + 911, + 857, + 938, + 824, + 938 + ], + "score": 0.84, + "latex": "g _ { \\mathrm { s l } }" + }, + { + "category_id": 13, + "poly": [ + 696, + 404, + 716, + 404, + 716, + 434, + 696, + 434 + ], + "score": 0.84, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 843, + 1420, + 865, + 1420, + 865, + 1447, + 843, + 1447 + ], + "score": 0.83, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 300, + 1378, + 1051, + 1378, + 1051, + 1415, + 300, + 1415 + ], + "score": 0.83, + "latex": "\\begin{array} { r } { V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \\kappa , M _ { \\theta _ { t } } \\right] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 826, + 303, + 847, + 303, + 847, + 330, + 826, + 330 + ], + "score": 0.82, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 1268, + 759, + 1301, + 759, + 1301, + 786, + 1268, + 786 + ], + "score": 0.81, + "latex": "g _ { \\mathrm { s l } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1087, + 1194, + 1087, + 1194, + 1126, + 297, + 1126 + ], + "score": 0.8, + "latex": "\\begin{array} { r } { Q _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = { \\mathbb E } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r ( x _ { t } , a _ { t } , x _ { t + 1 } ) \\mid ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 971, + 1416, + 997, + 1416, + 997, + 1443, + 971, + 1443 + ], + "score": 0.8, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 645, + 698, + 687, + 698, + 687, + 725, + 645, + 725 + ], + "score": 0.79, + "latex": "g _ { \\mathrm { p g } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 330, + 325, + 330, + 325, + 355, + 297, + 355 + ], + "score": 0.78, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 549, + 603, + 575, + 603, + 575, + 629, + 549, + 629 + ], + "score": 0.77, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 1055, + 821, + 1089, + 821, + 1089, + 847, + 1055, + 847 + ], + "score": 0.76, + "latex": "g _ { \\mathrm { s l } }" + }, + { + "category_id": 13, + "poly": [ + 662, + 1207, + 686, + 1207, + 686, + 1233, + 662, + 1233 + ], + "score": 0.75, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 551, + 1136, + 575, + 1136, + 575, + 1164, + 551, + 1164 + ], + "score": 0.64, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 851, + 1569, + 890, + 1569, + 890, + 1592, + 851, + 1592 + ], + "score": 0.48, + "latex": "\\mathcal { D } _ { \\mathrm { g e n } }" + }, + { + "category_id": 14, + "poly": [ + 296, + 1086, + 1196, + 1086, + 1196, + 1126, + 296, + 1126 + ], + "score": 0.47, + "latex": "\\begin{array} { r } { Q _ { M _ { \\bar { \\theta } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = { \\mathbb E } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } r ( x _ { t } , a _ { t } , x _ { t + 1 } ) \\mid ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } \\right] , } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1542.0, + 922.0, + 1542.0, + 922.0, + 1572.0, + 822.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1584.0, + 659.0, + 1584.0, + 659.0, + 1610.0, + 440.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 1654.0, + 868.0, + 1654.0, + 868.0, + 1685.0, + 551.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 1676.0, + 822.0, + 1676.0, + 822.0, + 1703.0, + 593.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 1699.0, + 442.0, + 1699.0, + 442.0, + 1756.0, + 314.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 1703.0, + 599.0, + 1703.0, + 599.0, + 1758.0, + 470.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1693.0, + 763.0, + 1693.0, + 763.0, + 1727.0, + 659.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 1692.0, + 1160.0, + 1692.0, + 1160.0, + 1767.0, + 902.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 1689.0, + 1401.0, + 1689.0, + 1401.0, + 1761.0, + 1182.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1815.0, + 728.0, + 1815.0, + 728.0, + 1841.0, + 460.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1834.0, + 648.0, + 1834.0, + 648.0, + 1862.0, + 540.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 1830.0, + 942.0, + 1830.0, + 942.0, + 1861.0, + 827.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 1901.0, + 655.0, + 1901.0, + 655.0, + 1928.0, + 528.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1898.0, + 959.0, + 1898.0, + 959.0, + 1929.0, + 856.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1898.0, + 1258.0, + 1898.0, + 1258.0, + 1929.0, + 1140.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1564.5, + 894.0, + 1564.5, + 894.0, + 1599.5, + 847.0, + 1599.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 1851.5, + 909.0, + 1851.5, + 909.0, + 1885.0, + 858.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 1994.0, + 1216.0, + 1994.0, + 1216.0, + 2030.0, + 480.0, + 2030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 872.0, + 2084.0, + 872.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 224.0, + 712.0, + 224.0, + 712.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 263.0, + 382.0, + 263.0, + 382.0, + 313.0, + 289.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 357.0, + 402.0, + 357.0, + 402.0, + 393.0, + 294.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 934.0, + 438.0, + 934.0, + 438.0, + 970.0, + 297.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1081.0, + 1252.0, + 1081.0, + 1252.0, + 1132.0, + 1197.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1126.0, + 296.0, + 1126.0, + 296.0, + 1180.0, + 290.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1126.0, + 550.0, + 1126.0, + 550.0, + 1180.0, + 539.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1126.0, + 1409.0, + 1126.0, + 1409.0, + 1180.0, + 576.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1170.0, + 739.0, + 1170.0, + 739.0, + 1212.0, + 293.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1170.0, + 1407.0, + 1170.0, + 1407.0, + 1212.0, + 784.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1203.0, + 661.0, + 1203.0, + 661.0, + 1241.0, + 293.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 687.0, + 1203.0, + 1405.0, + 1203.0, + 1405.0, + 1241.0, + 687.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1234.0, + 1405.0, + 1234.0, + 1405.0, + 1270.0, + 293.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1261.0, + 1001.0, + 1261.0, + 1001.0, + 1349.0, + 285.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1261.0, + 1119.0, + 1261.0, + 1119.0, + 1349.0, + 1086.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1261.0, + 1414.0, + 1261.0, + 1414.0, + 1349.0, + 1397.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1332.0, + 366.0, + 1332.0, + 366.0, + 1377.0, + 290.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1332.0, + 1408.0, + 1332.0, + 1408.0, + 1377.0, + 1347.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1369.0, + 299.0, + 1369.0, + 299.0, + 1418.0, + 290.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1369.0, + 1410.0, + 1369.0, + 1410.0, + 1418.0, + 1052.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1412.0, + 842.0, + 1412.0, + 842.0, + 1449.0, + 294.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 1412.0, + 894.0, + 1412.0, + 894.0, + 1449.0, + 866.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 1412.0, + 970.0, + 1412.0, + 970.0, + 1449.0, + 922.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1412.0, + 1057.0, + 1412.0, + 1057.0, + 1449.0, + 998.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 399.0, + 695.0, + 399.0, + 695.0, + 438.0, + 309.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 399.0, + 952.0, + 399.0, + 952.0, + 438.0, + 717.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 428.0, + 704.0, + 428.0, + 704.0, + 469.0, + 306.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 463.0, + 384.0, + 463.0, + 384.0, + 495.0, + 307.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 463.0, + 539.0, + 463.0, + 539.0, + 495.0, + 500.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 494.0, + 341.0, + 494.0, + 341.0, + 526.0, + 307.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 491.0, + 434.0, + 491.0, + 434.0, + 531.0, + 382.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 491.0, + 691.0, + 491.0, + 691.0, + 531.0, + 457.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 491.0, + 922.0, + 491.0, + 922.0, + 531.0, + 747.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 532.0, + 340.0, + 532.0, + 340.0, + 565.0, + 309.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 514.0, + 682.0, + 514.0, + 682.0, + 580.0, + 376.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 514.0, + 786.0, + 514.0, + 786.0, + 580.0, + 735.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 514.0, + 1314.0, + 514.0, + 1314.0, + 580.0, + 840.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 514.0, + 1401.0, + 514.0, + 1401.0, + 580.0, + 1373.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 565.0, + 491.0, + 565.0, + 491.0, + 611.0, + 339.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 565.0, + 601.0, + 565.0, + 601.0, + 611.0, + 552.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 565.0, + 831.0, + 565.0, + 831.0, + 611.0, + 654.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 565.0, + 902.0, + 565.0, + 902.0, + 611.0, + 895.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 603.0, + 339.0, + 603.0, + 339.0, + 635.0, + 309.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 593.0, + 472.0, + 593.0, + 472.0, + 643.0, + 380.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 593.0, + 548.0, + 593.0, + 548.0, + 643.0, + 499.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 593.0, + 915.0, + 593.0, + 915.0, + 643.0, + 576.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 593.0, + 985.0, + 593.0, + 985.0, + 643.0, + 976.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 593.0, + 1096.0, + 593.0, + 1096.0, + 643.0, + 1039.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 631.0, + 339.0, + 631.0, + 339.0, + 664.0, + 309.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 630.0, + 594.0, + 630.0, + 594.0, + 664.0, + 384.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 630.0, + 1406.0, + 630.0, + 1406.0, + 664.0, + 646.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 662.0, + 1408.0, + 662.0, + 1408.0, + 696.0, + 341.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 724.0, + 339.0, + 724.0, + 339.0, + 757.0, + 307.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 720.0, + 407.0, + 720.0, + 407.0, + 764.0, + 379.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 720.0, + 890.0, + 720.0, + 890.0, + 764.0, + 825.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 753.0, + 339.0, + 753.0, + 339.0, + 786.0, + 306.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 747.0, + 704.0, + 747.0, + 704.0, + 794.0, + 422.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 747.0, + 1267.0, + 747.0, + 1267.0, + 794.0, + 755.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 747.0, + 1307.0, + 747.0, + 1307.0, + 794.0, + 1302.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 782.0, + 439.0, + 782.0, + 439.0, + 818.0, + 382.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 783.0, + 340.0, + 783.0, + 340.0, + 818.0, + 299.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 814.0, + 340.0, + 814.0, + 340.0, + 847.0, + 297.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 809.0, + 704.0, + 809.0, + 704.0, + 853.0, + 424.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 809.0, + 1054.0, + 809.0, + 1054.0, + 853.0, + 755.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 809.0, + 1093.0, + 809.0, + 1093.0, + 853.0, + 1090.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 844.0, + 340.0, + 844.0, + 340.0, + 877.0, + 297.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 844.0, + 465.0, + 844.0, + 465.0, + 876.0, + 384.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 875.0, + 340.0, + 875.0, + 340.0, + 909.0, + 296.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 875.0, + 477.0, + 875.0, + 477.0, + 910.0, + 382.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 875.0, + 636.0, + 875.0, + 636.0, + 910.0, + 513.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 905.0, + 340.0, + 905.0, + 340.0, + 940.0, + 296.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 894.0, + 589.0, + 894.0, + 589.0, + 950.0, + 379.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 894.0, + 823.0, + 894.0, + 823.0, + 950.0, + 640.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 894.0, + 983.0, + 894.0, + 983.0, + 950.0, + 858.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 894.0, + 1062.0, + 894.0, + 1062.0, + 950.0, + 1058.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 936.0, + 435.0, + 936.0, + 435.0, + 962.0, + 299.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 686.5, + 693.0, + 686.5, + 693.0, + 731.5, + 339.0, + 731.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 292.0, + 693.0, + 292.0, + 693.0, + 336.0, + 359.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 292.0, + 825.0, + 292.0, + 825.0, + 336.0, + 748.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 292.0, + 1104.0, + 292.0, + 1104.0, + 336.0, + 848.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 324.0, + 513.0, + 324.0, + 513.0, + 361.0, + 326.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 357.0, + 402.0, + 357.0, + 402.0, + 393.0, + 294.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 934.0, + 438.0, + 934.0, + 438.0, + 970.0, + 297.0, + 970.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 23, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 767, + 1406, + 767, + 1406, + 1169, + 297, + 1169 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 296, + 1777, + 1405, + 1777, + 1405, + 1964, + 296, + 1964 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 300, + 1525, + 1404, + 1525, + 1404, + 1621, + 300, + 1621 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1290, + 1404, + 1290, + 1404, + 1417, + 297, + 1417 + ], + "score": 0.976 + }, + { + "category_id": 8, + "poly": [ + 665, + 1195, + 1032, + 1195, + 1032, + 1272, + 665, + 1272 + ], + "score": 0.964 + }, + { + "category_id": 8, + "poly": [ + 412, + 1645, + 1292, + 1645, + 1292, + 1751, + 412, + 1751 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 292, + 1974, + 1405, + 1974, + 1405, + 2035, + 292, + 2035 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 294, + 228, + 1403, + 228, + 1403, + 293, + 294, + 293 + ], + "score": 0.936 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 106, + 297, + 106 + ], + "score": 0.922 + }, + { + "category_id": 8, + "poly": [ + 404, + 496, + 1101, + 496, + 1101, + 576, + 404, + 576 + ], + "score": 0.912 + }, + { + "category_id": 8, + "poly": [ + 351, + 410, + 1290, + 410, + 1290, + 491, + 351, + 491 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 297, + 1465, + 1083, + 1465, + 1083, + 1499, + 297, + 1499 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1354, + 1214, + 1401, + 1214, + 1401, + 1245, + 1354, + 1245 + ], + "score": 0.887 + }, + { + "category_id": 8, + "poly": [ + 406, + 663, + 1399, + 663, + 1399, + 710, + 406, + 710 + ], + "score": 0.885 + }, + { + "category_id": 9, + "poly": [ + 1359, + 630, + 1402, + 630, + 1402, + 656, + 1359, + 656 + ], + "score": 0.878 + }, + { + "category_id": 9, + "poly": [ + 1358, + 522, + 1402, + 522, + 1402, + 550, + 1358, + 550 + ], + "score": 0.875 + }, + { + "category_id": 9, + "poly": [ + 1358, + 372, + 1402, + 372, + 1402, + 400, + 1358, + 400 + ], + "score": 0.868 + }, + { + "category_id": 1, + "poly": [ + 1427, + 1384, + 1623, + 1384, + 1623, + 1416, + 1427, + 1416 + ], + "score": 0.865 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.862 + }, + { + "category_id": 9, + "poly": [ + 1359, + 435, + 1403, + 435, + 1403, + 463, + 1359, + 463 + ], + "score": 0.845 + }, + { + "category_id": 8, + "poly": [ + 318, + 312, + 1557, + 312, + 1557, + 359, + 318, + 359 + ], + "score": 0.762 + }, + { + "category_id": 1, + "poly": [ + 298, + 437, + 336, + 437, + 336, + 463, + 298, + 463 + ], + "score": 0.654 + }, + { + "category_id": 9, + "poly": [ + 1358, + 711, + 1401, + 711, + 1401, + 739, + 1358, + 739 + ], + "score": 0.582 + }, + { + "category_id": 8, + "poly": [ + 1398, + 581, + 1596, + 581, + 1596, + 618, + 1398, + 618 + ], + "score": 0.524 + }, + { + "category_id": 8, + "poly": [ + 409, + 581, + 1414, + 581, + 1414, + 627, + 409, + 627 + ], + "score": 0.41 + }, + { + "category_id": 9, + "poly": [ + 1359, + 711, + 1402, + 711, + 1402, + 739, + 1359, + 739 + ], + "score": 0.341 + }, + { + "category_id": 1, + "poly": [ + 318, + 312, + 1557, + 312, + 1557, + 359, + 318, + 359 + ], + "score": 0.179 + }, + { + "category_id": 9, + "poly": [ + 298, + 437, + 336, + 437, + 336, + 463, + 298, + 463 + ], + "score": 0.101 + }, + { + "category_id": 14, + "poly": [ + 664, + 1192, + 1035, + 1192, + 1035, + 1269, + 664, + 1269 + ], + "score": 0.94, + "latex": "g _ { \\mathrm { s l } } g _ { \\mathrm { s l } } \\frac { \\vert \\vert g _ { \\mathrm { p g } } \\vert \\vert _ { 2 } } { \\operatorname* { m a x } \\{ \\vert \\vert g _ { \\mathrm { p g } } \\vert \\vert _ { 2 } , \\vert \\vert g _ { \\mathrm { s l } } \\vert \\vert _ { 2 } \\} } ." + }, + { + "category_id": 13, + "poly": [ + 381, + 1779, + 799, + 1779, + 799, + 1815, + 381, + 1815 + ], + "score": 0.92, + "latex": "\\begin{array} { r c l } { { A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) } } & { { = } } & { { \\log D _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 383, + 770, + 912, + 770, + 912, + 809, + 383, + 809 + ], + "score": 0.92, + "latex": "A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } ( x , \\underline { { a } } , x ^ { \\prime } ) = Q _ { M _ { \\theta } } ^ { \\kappa } ( x , a , \\underline { { x } } ^ { \\prime } ) - V _ { M _ { \\theta } } ^ { \\kappa } ( x , \\underline { { a } } )" + }, + { + "category_id": 13, + "poly": [ + 819, + 893, + 1017, + 893, + 1017, + 929, + 819, + 929 + ], + "score": 0.92, + "latex": "- A _ { \\varphi _ { 0 } ^ { * } , \\varphi _ { 1 } ^ { * } } + H _ { M ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 884, + 1075, + 967, + 1075, + 967, + 1109, + 884, + 1109 + ], + "score": 0.91, + "latex": "| | g _ { \\mathrm { p g } } | | _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 795, + 1105, + 869, + 1105, + 869, + 1139, + 795, + 1139 + ], + "score": 0.91, + "latex": "| | g _ { \\mathrm { s l } } | | _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1105, + 893, + 1166, + 893, + 1166, + 923, + 1105, + 923 + ], + "score": 0.91, + "latex": "H _ { M ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 923, + 356, + 923, + 356, + 956, + 298, + 956 + ], + "score": 0.9, + "latex": "H _ { M _ { \\theta } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1817, + 584, + 1817, + 584, + 1853, + 297, + 1853 + ], + "score": 0.9, + "latex": "\\rho _ { 0 } ( x ) \\kappa ( a | x ) M _ { \\theta _ { t } } ( x ^ { \\prime } | a , x )" + }, + { + "category_id": 13, + "poly": [ + 772, + 1816, + 1197, + 1816, + 1197, + 1863, + 772, + 1863 + ], + "score": 0.9, + "latex": "\\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right]" + }, + { + "category_id": 14, + "poly": [ + 403, + 1642, + 1294, + 1642, + 1294, + 1754, + 403, + 1754 + ], + "score": 0.9, + "latex": "\\begin{array} { r l } & { \\theta _ { t + 1 } = \\underset { \\theta } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } \\\\ & { s . t . \\quad \\varphi * = \\underset { \\varphi } { \\arg \\operatorname* { m a x } } ~ \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ \\log ( 1 - D _ { \\varphi } ( x , a , x ^ { \\prime } ) ) \\right] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 1902, + 470, + 1902, + 470, + 1933, + 298, + 1933 + ], + "score": 0.89, + "latex": "\\nabla _ { \\boldsymbol { \\theta } } M _ { \\boldsymbol { \\theta } } ( \\boldsymbol { x } ^ { \\prime } | \\boldsymbol { x } , a )" + }, + { + "category_id": 13, + "poly": [ + 968, + 1860, + 1331, + 1860, + 1331, + 1896, + 968, + 1896 + ], + "score": 0.89, + "latex": "M _ { \\theta _ { t } } ( x ^ { \\prime } | x , a ) \\nabla _ { \\theta } \\log M _ { \\theta } ( x ^ { \\prime } | x , a )" + }, + { + "category_id": 13, + "poly": [ + 1193, + 1779, + 1399, + 1779, + 1399, + 1820, + 1193, + 1820 + ], + "score": 0.88, + "latex": "\\begin{array} { r l } { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) } & { { } = } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1283, + 1976, + 1326, + 1976, + 1326, + 2005, + 1283, + 2005 + ], + "score": 0.88, + "latex": "M _ { \\theta }" + }, + { + "category_id": 14, + "poly": [ + 401, + 408, + 1502, + 408, + 1502, + 719, + 401, + 719 + ], + "score": 0.87, + "latex": "\\begin{array} { r l } & { Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb E [ \\displaystyle \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 0 } ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } ] \\qquad ( 2 4 ) } \\\\ & { V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb E [ \\displaystyle \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi _ { 1 } ^ { * } } ( x _ { t } , a _ { t } ) | ( x _ { t } , a _ { t } ) = ( x , a ) , \\kappa , M _ { \\theta _ { t } } ] } \\\\ & { \\varphi _ { 0 } ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 0 } ^ { * } } \\mathbb E _ { \\rho _ { M ^ { * } } ^ { \\kappa } , \\textbf { u } } [ \\log D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } + u _ { x ^ { \\prime } } ) ] + \\mathbb E _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\textbf { u } } [ \\log ( 1 - D _ { \\varphi _ { 0 } } ( x + u _ { x } , a + u _ { a } , x ^ { \\prime } ) ] ] } \\\\ & { \\varphi _ { 1 } ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi _ { 1 } ^ { * } } \\mathbb E _ { \\rho _ { M ^ { * } } ^ { \\kappa } , \\textbf { u } } [ \\log D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ] + \\mathbb E _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } , \\textbf { u } } [ \\log ( 1 - D _ { \\varphi _ { 1 } } ( x + u _ { x } , a + u _ { a } ) ) ] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1234, + 1076, + 1258, + 1076, + 1258, + 1106, + 1234, + 1106 + ], + "score": 0.87, + "latex": "l _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 853, + 1046, + 877, + 1046, + 877, + 1075, + 853, + 1075 + ], + "score": 0.87, + "latex": "l _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1860, + 575, + 1860, + 575, + 1901, + 298, + 1901 + ], + "score": 0.86, + "latex": "\\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 698, + 1357, + 728, + 1357, + 728, + 1386, + 698, + 1386 + ], + "score": 0.86, + "latex": "p _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 616, + 1358, + 647, + 1358, + 647, + 1386, + 616, + 1386 + ], + "score": 0.85, + "latex": "p _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 565, + 2004, + 586, + 2004, + 586, + 2035, + 565, + 2035 + ], + "score": 0.84, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 355, + 1139, + 388, + 1139, + 388, + 1168, + 355, + 1168 + ], + "score": 0.84, + "latex": "g _ { \\mathrm { s l } }" + }, + { + "category_id": 14, + "poly": [ + 347, + 313, + 1465, + 313, + 1465, + 361, + 347, + 361 + ], + "score": 0.83, + "latex": "t + 1 = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta } } ^ { \\star } } \\left[ A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] + \\mathbb { E } _ { \\rho _ { M ^ { \\star } } ^ { \\star } } \\left[ \\left( H _ { M ^ { \\star } } ( x , a ) - A _ { \\varphi _ { 0 } ^ { \\star } , \\varphi _ { 1 } ^ { \\star } } ( x , a , x ^ { \\prime } ) \\right) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] ." + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1463.0, + 1086.0, + 1463.0, + 1086.0, + 1502.0, + 293.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 871.0, + 2083.0, + 871.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 763.0, + 382.0, + 763.0, + 382.0, + 811.0, + 300.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 763.0, + 1409.0, + 763.0, + 1409.0, + 811.0, + 913.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 799.0, + 1406.0, + 799.0, + 1406.0, + 836.0, + 294.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 829.0, + 1406.0, + 829.0, + 1406.0, + 866.0, + 295.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 862.0, + 1403.0, + 862.0, + 1403.0, + 895.0, + 296.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 890.0, + 818.0, + 890.0, + 818.0, + 934.0, + 292.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 890.0, + 1104.0, + 890.0, + 1104.0, + 934.0, + 1018.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 890.0, + 1408.0, + 890.0, + 1408.0, + 934.0, + 1167.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 920.0, + 1406.0, + 920.0, + 1406.0, + 956.0, + 357.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 954.0, + 1404.0, + 954.0, + 1404.0, + 986.0, + 295.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 981.0, + 1406.0, + 981.0, + 1406.0, + 1016.0, + 294.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1012.0, + 1406.0, + 1012.0, + 1406.0, + 1048.0, + 295.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1043.0, + 852.0, + 1043.0, + 852.0, + 1078.0, + 294.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 1043.0, + 1404.0, + 1043.0, + 1404.0, + 1078.0, + 878.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1073.0, + 883.0, + 1073.0, + 883.0, + 1113.0, + 294.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 1073.0, + 1233.0, + 1073.0, + 1233.0, + 1113.0, + 968.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1073.0, + 1406.0, + 1073.0, + 1406.0, + 1113.0, + 1259.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1103.0, + 794.0, + 1103.0, + 794.0, + 1143.0, + 294.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1103.0, + 1407.0, + 1103.0, + 1407.0, + 1143.0, + 870.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1131.0, + 354.0, + 1131.0, + 354.0, + 1174.0, + 293.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 1131.0, + 427.0, + 1131.0, + 427.0, + 1174.0, + 389.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1773.0, + 380.0, + 1773.0, + 380.0, + 1819.0, + 291.0, + 1819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 1773.0, + 1192.0, + 1773.0, + 1192.0, + 1819.0, + 800.0, + 1819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1773.0, + 1406.0, + 1773.0, + 1406.0, + 1819.0, + 1400.0, + 1819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1812.0, + 296.0, + 1812.0, + 296.0, + 1862.0, + 291.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1812.0, + 771.0, + 1812.0, + 771.0, + 1862.0, + 585.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1812.0, + 1408.0, + 1812.0, + 1408.0, + 1862.0, + 1198.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1855.0, + 297.0, + 1855.0, + 297.0, + 1899.0, + 290.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1855.0, + 967.0, + 1855.0, + 967.0, + 1899.0, + 576.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 1855.0, + 1406.0, + 1855.0, + 1406.0, + 1899.0, + 1332.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1895.0, + 297.0, + 1895.0, + 297.0, + 1936.0, + 294.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 1895.0, + 1407.0, + 1895.0, + 1407.0, + 1936.0, + 471.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1928.0, + 1337.0, + 1928.0, + 1337.0, + 1964.0, + 295.0, + 1964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1523.0, + 1405.0, + 1523.0, + 1405.0, + 1563.0, + 293.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1556.0, + 1408.0, + 1556.0, + 1408.0, + 1593.0, + 293.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1588.0, + 703.0, + 1588.0, + 703.0, + 1622.0, + 296.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1287.0, + 1408.0, + 1287.0, + 1408.0, + 1328.0, + 292.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1323.0, + 1405.0, + 1323.0, + 1405.0, + 1356.0, + 295.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1350.0, + 615.0, + 1350.0, + 615.0, + 1388.0, + 294.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 1350.0, + 697.0, + 1350.0, + 697.0, + 1388.0, + 648.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 1350.0, + 1405.0, + 1350.0, + 1405.0, + 1388.0, + 729.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1381.0, + 647.0, + 1381.0, + 647.0, + 1419.0, + 294.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1976.0, + 1282.0, + 1976.0, + 1282.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1327.0, + 1976.0, + 1403.0, + 1976.0, + 1403.0, + 2007.0, + 1327.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 2000.0, + 564.0, + 2000.0, + 564.0, + 2038.0, + 291.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 2000.0, + 1406.0, + 2000.0, + 1406.0, + 2038.0, + 587.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 224.0, + 1406.0, + 224.0, + 1406.0, + 270.0, + 293.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 255.0, + 438.0, + 255.0, + 438.0, + 292.0, + 294.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 1382.0, + 1628.0, + 1382.0, + 1628.0, + 1419.0, + 1425.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 434.0, + 342.0, + 434.0, + 342.0, + 467.0, + 292.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 308.0, + 346.0, + 308.0, + 346.0, + 366.0, + 325.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1466.0, + 308.0, + 1564.0, + 308.0, + 1564.0, + 366.0, + 1466.0, + 366.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 24, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1515, + 1404, + 1515, + 1404, + 2034, + 297, + 2034 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 663, + 1405, + 663, + 1405, + 1502, + 297, + 1502 + ], + "score": 0.98 + }, + { + "category_id": 8, + "poly": [ + 372, + 282, + 1336, + 282, + 1336, + 471, + 372, + 471 + ], + "score": 0.964 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 857, + 74, + 857, + 106, + 297, + 106 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 538, + 229, + 538, + 262, + 297, + 262 + ], + "score": 0.913 + }, + { + "category_id": 0, + "poly": [ + 300, + 584, + 783, + 584, + 783, + 621, + 300, + 621 + ], + "score": 0.901 + }, + { + "category_id": 1, + "poly": [ + 296, + 494, + 1358, + 494, + 1358, + 537, + 296, + 537 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2113, + 835, + 2113 + ], + "score": 0.847 + }, + { + "category_id": 2, + "poly": [ + 1427, + 614, + 1624, + 614, + 1624, + 677, + 1427, + 677 + ], + "score": 0.633 + }, + { + "category_id": 2, + "poly": [ + 1428, + 1912, + 1621, + 1912, + 1621, + 1943, + 1428, + 1943 + ], + "score": 0.489 + }, + { + "category_id": 1, + "poly": [ + 1428, + 1912, + 1621, + 1912, + 1621, + 1943, + 1428, + 1943 + ], + "score": 0.464 + }, + { + "category_id": 1, + "poly": [ + 1427, + 614, + 1624, + 614, + 1624, + 677, + 1427, + 677 + ], + "score": 0.29 + }, + { + "category_id": 13, + "poly": [ + 299, + 758, + 389, + 758, + 389, + 788, + 299, + 788 + ], + "score": 0.94, + "latex": "M ( y | x )" + }, + { + "category_id": 13, + "poly": [ + 453, + 1256, + 530, + 1256, + 530, + 1285, + 453, + 1285 + ], + "score": 0.94, + "latex": "\\beta ( a | x )" + }, + { + "category_id": 14, + "poly": [ + 372, + 282, + 1329, + 282, + 1329, + 475, + 372, + 475 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } { \\displaystyle \\theta _ { t + 1 } = \\arg \\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) \\log M _ { \\theta } ( x ^ { \\prime } | x , a ) \\right] } & { } \\\\ { \\displaystyle s . t . } & { Q _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) = \\mathbb { E } \\left[ \\sum _ { t } ^ { \\infty } \\gamma ^ { t } \\log D _ { \\varphi ^ { * } } ( x _ { t } , a _ { t } , x _ { t + 1 } ) | ( x _ { t } , a _ { t } , x _ { t + 1 } ) = ( x , a , x ^ { \\prime } ) , \\kappa , M _ { \\theta _ { t } } \\right] } \\\\ { \\displaystyle \\varphi ^ { * } = \\arg \\operatorname* { m a x } _ { \\varphi } \\mathbb { E } _ { \\rho _ { M ^ { * } } ^ { \\kappa } } \\left[ \\log D _ { \\varphi } ( x , a , x ^ { \\prime } ) \\right] + \\mathbb { E } _ { \\rho _ { M _ { \\theta _ { t } } } ^ { \\kappa } } \\left[ \\log ( 1 - D _ { \\varphi } ( x , a , x ^ { \\prime } ) ) \\right] , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 897, + 1074, + 1247, + 1074, + 1247, + 1118, + 897, + 1118 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\frac { P _ { \\mathrm { t e s t } } ( x ) } { P _ { \\mathrm { t r a i n } } ( x ) } = \\frac { P ( x ) \\beta ( a | x ) } { P ( x ) \\mu ( a | x ) } = \\frac { \\beta ( a | x ) } { \\mu ( a | x ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 722, + 758, + 826, + 758, + 826, + 788, + 722, + 788 + ], + "score": 0.93, + "latex": "P _ { \\mathrm { t r a i n } } ( x )" + }, + { + "category_id": 13, + "poly": [ + 380, + 498, + 871, + 498, + 871, + 535, + 380, + 535 + ], + "score": 0.93, + "latex": "A _ { \\varphi ^ { * } } ( x , a , x ^ { \\prime } ) = Q _ { M _ { \\theta } } ^ { \\kappa } ( x , a , x ^ { \\prime } ) - V _ { M _ { \\theta } } ^ { \\kappa } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 1096, + 1044, + 1155, + 1044, + 1155, + 1073, + 1096, + 1073 + ], + "score": 0.92, + "latex": "P ( x )" + }, + { + "category_id": 13, + "poly": [ + 615, + 790, + 763, + 790, + 763, + 816, + 615, + 816 + ], + "score": 0.92, + "latex": "P _ { \\mathrm { t e s t } } \\neq P _ { \\mathrm { t e s t } }" + }, + { + "category_id": 13, + "poly": [ + 434, + 788, + 526, + 788, + 526, + 817, + 434, + 817 + ], + "score": 0.92, + "latex": "P _ { \\mathrm { { t e s t } } } ( x )" + }, + { + "category_id": 13, + "poly": [ + 356, + 819, + 784, + 819, + 784, + 862, + 356, + 862 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\operatorname* { m a x } _ { M } \\mathbb { E } _ { x \\sim P _ { \\mathrm { t r a i n } } } [ \\frac { P _ { \\mathrm { t e s t } } ( x ) } { P _ { \\mathrm { t r a i n } } ( x ) } \\log M ( y | x ) ] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 923, + 498, + 1352, + 498, + 1352, + 539, + 923, + 539 + ], + "score": 0.9, + "latex": "V _ { M _ { \\theta _ { t } } } ^ { \\kappa } ( x , a ) = \\mathbb { E } _ { M _ { \\theta _ { t } } ( x , a ) } \\left[ Q ^ { \\kappa } ( x , a , x ^ { \\prime } ) \\right] ," + }, + { + "category_id": 13, + "poly": [ + 1122, + 1445, + 1153, + 1445, + 1153, + 1466, + 1122, + 1466 + ], + "score": 0.9, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 903, + 1361, + 921, + 1361, + 921, + 1380, + 903, + 1380 + ], + "score": 0.88, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 542, + 1178, + 1022, + 1178, + 1022, + 1222, + 542, + 1222 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { \\frac { P _ { \\mathrm { t e s t } } ( x ) } { P _ { \\mathrm { t r a i n } } ( x ) } = \\frac { P ( x ) \\beta ( a | x ) } { P ( x ) \\mu ( a | x ) } = \\frac { \\beta ( a | x ) } { \\mu ( a | x ) } \\propto \\frac { 1 } { \\mu ( a | x ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1241, + 1354, + 1258, + 1354, + 1258, + 1380, + 1241, + 1380 + ], + "score": 0.88, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 1080, + 1023, + 1095, + 1023, + 1095, + 1036, + 1080, + 1036 + ], + "score": 0.87, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 679, + 1023, + 696, + 1023, + 696, + 1037, + 679, + 1037 + ], + "score": 0.87, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1075, + 1251, + 1094, + 1251, + 1094, + 1292, + 1075, + 1292 + ], + "score": 0.87, + "latex": "\\frac { \\beta } { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 1335, + 1092, + 1352, + 1092, + 1352, + 1111, + 1335, + 1111 + ], + "score": 0.87, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 299, + 1119, + 317, + 1119, + 317, + 1146, + 299, + 1146 + ], + "score": 0.86, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 1066, + 1053, + 1082, + 1053, + 1082, + 1066, + 1066, + 1066 + ], + "score": 0.83, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1041, + 1183, + 1111, + 1183, + 1111, + 1223, + 1041, + 1223 + ], + "score": 0.79, + "latex": "{ \\frac { 1 } { \\mu ( a | x ) } }" + }, + { + "category_id": 13, + "poly": [ + 530, + 1152, + 553, + 1152, + 553, + 1172, + 530, + 1172 + ], + "score": 0.64, + "latex": "\\&" + }, + { + "category_id": 13, + "poly": [ + 583, + 986, + 605, + 986, + 605, + 1006, + 583, + 1006 + ], + "score": 0.63, + "latex": "\\&" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 858.0, + 72.0, + 858.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 584.0, + 787.0, + 584.0, + 787.0, + 625.0, + 294.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 871.0, + 2083.0, + 871.0, + 2123.0, + 829.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1423.0, + 611.0, + 1629.0, + 611.0, + 1629.0, + 650.0, + 1423.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1423.0, + 640.0, + 1541.0, + 640.0, + 1541.0, + 680.0, + 1423.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 1908.0, + 1623.0, + 1908.0, + 1623.0, + 1948.0, + 1425.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1515.0, + 1404.0, + 1515.0, + 1404.0, + 1548.0, + 297.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1544.0, + 1407.0, + 1544.0, + 1407.0, + 1580.0, + 295.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1577.0, + 1405.0, + 1577.0, + 1405.0, + 1609.0, + 296.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1608.0, + 1405.0, + 1608.0, + 1405.0, + 1640.0, + 296.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1638.0, + 1407.0, + 1638.0, + 1407.0, + 1670.0, + 296.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1669.0, + 1405.0, + 1669.0, + 1405.0, + 1701.0, + 296.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1698.0, + 1405.0, + 1698.0, + 1405.0, + 1731.0, + 296.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1725.0, + 1404.0, + 1725.0, + 1404.0, + 1762.0, + 295.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1759.0, + 1405.0, + 1759.0, + 1405.0, + 1791.0, + 296.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1786.0, + 1405.0, + 1786.0, + 1405.0, + 1822.0, + 295.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1820.0, + 1405.0, + 1820.0, + 1405.0, + 1854.0, + 292.0, + 1854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1851.0, + 1404.0, + 1851.0, + 1404.0, + 1883.0, + 295.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1880.0, + 1406.0, + 1880.0, + 1406.0, + 1915.0, + 292.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1910.0, + 1405.0, + 1910.0, + 1405.0, + 1945.0, + 294.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1941.0, + 1406.0, + 1941.0, + 1406.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1971.0, + 1404.0, + 1971.0, + 1404.0, + 2005.0, + 294.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2003.0, + 461.0, + 2003.0, + 461.0, + 2035.0, + 295.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 659.0, + 1405.0, + 659.0, + 1405.0, + 693.0, + 294.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 691.0, + 1405.0, + 691.0, + 1405.0, + 727.0, + 295.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 723.0, + 1403.0, + 723.0, + 1403.0, + 755.0, + 297.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 752.0, + 721.0, + 752.0, + 721.0, + 788.0, + 390.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 752.0, + 1403.0, + 752.0, + 1403.0, + 788.0, + 827.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 779.0, + 433.0, + 779.0, + 433.0, + 820.0, + 294.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 779.0, + 614.0, + 779.0, + 614.0, + 820.0, + 527.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 779.0, + 1406.0, + 779.0, + 1406.0, + 820.0, + 764.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 809.0, + 1407.0, + 809.0, + 1407.0, + 859.0, + 785.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 855.0, + 1405.0, + 855.0, + 1405.0, + 891.0, + 295.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 886.0, + 1407.0, + 886.0, + 1407.0, + 922.0, + 295.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 915.0, + 1406.0, + 915.0, + 1406.0, + 952.0, + 292.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 949.0, + 1403.0, + 949.0, + 1403.0, + 980.0, + 296.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 974.0, + 582.0, + 974.0, + 582.0, + 1014.0, + 294.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 974.0, + 1407.0, + 974.0, + 1407.0, + 1014.0, + 606.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1010.0, + 678.0, + 1010.0, + 678.0, + 1042.0, + 297.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1010.0, + 1079.0, + 1010.0, + 1079.0, + 1042.0, + 697.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1010.0, + 1405.0, + 1010.0, + 1405.0, + 1042.0, + 1096.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1038.0, + 1065.0, + 1038.0, + 1065.0, + 1074.0, + 295.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 1038.0, + 1095.0, + 1038.0, + 1095.0, + 1074.0, + 1083.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 1038.0, + 1405.0, + 1038.0, + 1405.0, + 1074.0, + 1156.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1066.0, + 1006.0, + 1066.0, + 1006.0, + 1099.0, + 898.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 1065.0, + 1143.0, + 1065.0, + 1143.0, + 1098.0, + 1021.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 1064.0, + 1334.0, + 1064.0, + 1334.0, + 1121.0, + 1248.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 1064.0, + 1407.0, + 1064.0, + 1407.0, + 1121.0, + 1353.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1112.0, + 1408.0, + 1112.0, + 1408.0, + 1148.0, + 318.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1144.0, + 529.0, + 1144.0, + 529.0, + 1179.0, + 296.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 1144.0, + 1407.0, + 1144.0, + 1407.0, + 1179.0, + 554.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1171.0, + 541.0, + 1171.0, + 541.0, + 1217.0, + 295.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1167.0, + 911.0, + 1167.0, + 911.0, + 1205.0, + 836.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1181.0, + 1405.0, + 1181.0, + 1405.0, + 1216.0, + 1112.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1192.0, + 541.0, + 1192.0, + 541.0, + 1225.0, + 538.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1187.0, + 1040.0, + 1187.0, + 1040.0, + 1224.0, + 1023.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1187.0, + 1115.0, + 1187.0, + 1115.0, + 1224.0, + 1112.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1215.0, + 1405.0, + 1215.0, + 1405.0, + 1250.0, + 294.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1245.0, + 452.0, + 1245.0, + 452.0, + 1288.0, + 291.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1245.0, + 1074.0, + 1245.0, + 1074.0, + 1288.0, + 531.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1245.0, + 1407.0, + 1245.0, + 1407.0, + 1288.0, + 1095.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1287.0, + 1402.0, + 1287.0, + 1402.0, + 1319.0, + 296.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1317.0, + 1403.0, + 1317.0, + 1403.0, + 1349.0, + 296.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1348.0, + 902.0, + 1348.0, + 902.0, + 1380.0, + 296.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 1348.0, + 1240.0, + 1348.0, + 1240.0, + 1380.0, + 922.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1348.0, + 1403.0, + 1348.0, + 1403.0, + 1380.0, + 1259.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1375.0, + 1405.0, + 1375.0, + 1405.0, + 1412.0, + 294.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1407.0, + 1402.0, + 1407.0, + 1402.0, + 1438.0, + 296.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1437.0, + 1121.0, + 1437.0, + 1121.0, + 1472.0, + 295.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 1437.0, + 1406.0, + 1437.0, + 1406.0, + 1472.0, + 1154.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1466.0, + 695.0, + 1466.0, + 695.0, + 1505.0, + 295.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.25, + 820.0, + 644.25, + 820.0, + 644.25, + 864.0, + 294.25, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.25, + 1067.5, + 1148.25, + 1067.5, + 1148.25, + 1124.0, + 288.25, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 227.0, + 541.0, + 227.0, + 541.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 492.0, + 379.0, + 492.0, + 379.0, + 541.0, + 300.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 492.0, + 922.0, + 492.0, + 922.0, + 541.0, + 872.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 492.0, + 1362.0, + 492.0, + 1362.0, + 541.0, + 1353.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1425.0, + 1908.0, + 1623.0, + 1908.0, + 1623.0, + 1948.0, + 1425.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1423.0, + 611.0, + 1629.0, + 611.0, + 1629.0, + 650.0, + 1423.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1423.0, + 640.0, + 1541.0, + 640.0, + 1541.0, + 680.0, + 1423.0, + 680.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 25, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1364, + 1406, + 1364, + 1406, + 1734, + 297, + 1734 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 806, + 1404, + 806, + 1404, + 1023, + 297, + 1023 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1818, + 1404, + 1818, + 1404, + 2035, + 298, + 2035 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1405, + 228, + 1405, + 414, + 298, + 414 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 637, + 1404, + 637, + 1404, + 792, + 298, + 792 + ], + "score": 0.978 + }, + { + "category_id": 8, + "poly": [ + 649, + 1317, + 1053, + 1317, + 1053, + 1355, + 649, + 1355 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 631, + 1034, + 1070, + 1034, + 1070, + 1103, + 631, + 1103 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 655, + 1205, + 1047, + 1205, + 1047, + 1243, + 655, + 1243 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 298, + 1270, + 685, + 1270, + 685, + 1303, + 298, + 1303 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 295, + 1112, + 1333, + 1112, + 1333, + 1145, + 295, + 1145 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 298, + 1764, + 876, + 1764, + 876, + 1797, + 298, + 1797 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 298, + 1158, + 677, + 1158, + 677, + 1191, + 298, + 1191 + ], + "score": 0.913 + }, + { + "category_id": 0, + "poly": [ + 299, + 525, + 494, + 525, + 494, + 558, + 299, + 558 + ], + "score": 0.897 + }, + { + "category_id": 0, + "poly": [ + 300, + 458, + 681, + 458, + 681, + 493, + 300, + 493 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 856, + 75, + 856, + 104, + 298, + 104 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2112, + 835, + 2112 + ], + "score": 0.855 + }, + { + "category_id": 1, + "poly": [ + 298, + 582, + 1013, + 582, + 1013, + 616, + 298, + 616 + ], + "score": 0.474 + }, + { + "category_id": 0, + "poly": [ + 298, + 582, + 1013, + 582, + 1013, + 616, + 298, + 616 + ], + "score": 0.472 + }, + { + "category_id": 13, + "poly": [ + 411, + 1610, + 564, + 1610, + 564, + 1644, + 411, + 1644 + ], + "score": 0.94, + "latex": "\\epsilon \\sim U ( - e , e )" + }, + { + "category_id": 14, + "poly": [ + 631, + 1033, + 1066, + 1033, + 1066, + 1103, + 631, + 1103 + ], + "score": 0.94, + "latex": "\\mu _ { G N F C } ( x ) = \\frac { ( 6 2 . 5 - \\mathrm { m e a n } ( x ) ) } { 1 5 } + \\epsilon ," + }, + { + "category_id": 13, + "poly": [ + 539, + 1158, + 667, + 1158, + 667, + 1193, + 539, + 1193 + ], + "score": 0.93, + "latex": "M _ { 1 } ( y | x , a )" + }, + { + "category_id": 13, + "poly": [ + 539, + 1271, + 676, + 1271, + 676, + 1305, + 539, + 1305 + ], + "score": 0.93, + "latex": "M _ { 2 } ( x ^ { \\prime } | x , y )" + }, + { + "category_id": 13, + "poly": [ + 745, + 1580, + 849, + 1580, + 849, + 1613, + 745, + 1613 + ], + "score": 0.92, + "latex": "U ( - e , e )" + }, + { + "category_id": 14, + "poly": [ + 647, + 1315, + 1053, + 1315, + 1053, + 1355, + 647, + 1355 + ], + "score": 0.92, + "latex": "M _ { 2 } ( x ^ { \\prime } | x , a , y ) = y - \\mathrm { m e a n } ( x ) + x" + }, + { + "category_id": 13, + "poly": [ + 793, + 1365, + 1243, + 1365, + 1243, + 1400, + 793, + 1400 + ], + "score": 0.91, + "latex": "M ^ { * } ( x ^ { \\prime } | x , a ) = M _ { 2 } ( x ^ { \\prime } | x , a , M _ { 1 } ( y | x , a ) )" + }, + { + "category_id": 14, + "poly": [ + 652, + 1204, + 1048, + 1204, + 1048, + 1243, + 652, + 1243 + ], + "score": 0.9, + "latex": "M _ { 1 } ( y | x , a ) = \\mathcal { N } ( \\mathrm { m e a n } ( x ) + a , 2 )" + }, + { + "category_id": 13, + "poly": [ + 1154, + 1581, + 1220, + 1581, + 1220, + 1608, + 1154, + 1608 + ], + "score": 0.89, + "latex": "\\epsilon = 0" + }, + { + "category_id": 13, + "poly": [ + 552, + 1367, + 597, + 1367, + 597, + 1394, + 552, + 1394 + ], + "score": 0.89, + "latex": "M ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 296, + 1611, + 362, + 1611, + 362, + 1642, + 296, + 1642 + ], + "score": 0.88, + "latex": "1 - p" + }, + { + "category_id": 13, + "poly": [ + 1217, + 643, + 1238, + 643, + 1238, + 670, + 1217, + 670 + ], + "score": 0.81, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 1036, + 1584, + 1054, + 1584, + 1054, + 1612, + 1036, + 1612 + ], + "score": 0.79, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 868, + 1677, + 886, + 1677, + 886, + 1703, + 868, + 1703 + ], + "score": 0.78, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 750, + 1618, + 767, + 1618, + 767, + 1642, + 750, + 1642 + ], + "score": 0.77, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 801, + 1678, + 818, + 1678, + 818, + 1699, + 801, + 1699 + ], + "score": 0.77, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 419, + 1585, + 435, + 1585, + 435, + 1607, + 419, + 1607 + ], + "score": 0.77, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 967, + 1616, + 985, + 1616, + 985, + 1642, + 967, + 1642 + ], + "score": 0.76, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 298, + 1464, + 315, + 1464, + 315, + 1491, + 298, + 1491 + ], + "score": 0.76, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 738, + 1645, + 832, + 1645, + 832, + 1673, + 738, + 1673 + ], + "score": 0.76, + "latex": "\\mu _ { G N F C } ," + }, + { + "category_id": 13, + "poly": [ + 371, + 1119, + 386, + 1119, + 386, + 1140, + 371, + 1140 + ], + "score": 0.74, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 1319, + 1494, + 1337, + 1494, + 1337, + 1516, + 1319, + 1516 + ], + "score": 0.71, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 1247, + 1495, + 1267, + 1495, + 1267, + 1516, + 1247, + 1516 + ], + "score": 0.69, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1074, + 1949, + 1092, + 1949, + 1092, + 1970, + 1074, + 1970 + ], + "score": 0.69, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 1314, + 2009, + 1332, + 2009, + 1332, + 2031, + 1314, + 2031 + ], + "score": 0.69, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 1031, + 1647, + 1047, + 1647, + 1047, + 1668, + 1031, + 1668 + ], + "score": 0.67, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 911, + 1465, + 925, + 1465, + 925, + 1486, + 911, + 1486 + ], + "score": 0.56, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 437, + 1978, + 468, + 1978, + 468, + 2003, + 437, + 2003 + ], + "score": 0.54, + "latex": "\\mathbf { v } _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1301, + 1432, + 1349, + 1432, + 1349, + 1457, + 1301, + 1457 + ], + "score": 0.44, + "latex": "x , a ," + }, + { + "category_id": 13, + "poly": [ + 525, + 1977, + 558, + 1977, + 558, + 2004, + 525, + 2004 + ], + "score": 0.35, + "latex": "\\mathbf { v } _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1228, + 995, + 1320, + 995, + 1320, + 1022, + 1228, + 1022 + ], + "score": 0.34, + "latex": "\\mu _ { G N F C }" + }, + { + "category_id": 13, + "poly": [ + 729, + 1917, + 751, + 1917, + 751, + 1940, + 729, + 1940 + ], + "score": 0.33, + "latex": "\\mathbf { x }" + }, + { + "category_id": 13, + "poly": [ + 482, + 1978, + 514, + 1978, + 514, + 2003, + 482, + 2003 + ], + "score": 0.29, + "latex": "\\mathbf { v } _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1331, + 1435, + 1348, + 1435, + 1348, + 1455, + 1331, + 1455 + ], + "score": 0.28, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 1300, + 1434, + 1319, + 1434, + 1319, + 1455, + 1300, + 1455 + ], + "score": 0.26, + "latex": "x" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1759.0, + 879.0, + 1759.0, + 879.0, + 1804.0, + 293.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 525.0, + 495.0, + 525.0, + 495.0, + 561.0, + 295.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 454.0, + 685.0, + 454.0, + 685.0, + 500.0, + 294.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 858.0, + 72.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 578.0, + 1016.0, + 578.0, + 1016.0, + 623.0, + 293.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1364.0, + 551.0, + 1364.0, + 551.0, + 1402.0, + 294.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1364.0, + 792.0, + 1364.0, + 792.0, + 1402.0, + 598.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 1364.0, + 1407.0, + 1364.0, + 1407.0, + 1402.0, + 1244.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1397.0, + 1406.0, + 1397.0, + 1406.0, + 1432.0, + 295.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1428.0, + 1299.0, + 1428.0, + 1299.0, + 1462.0, + 294.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 1428.0, + 1403.0, + 1428.0, + 1403.0, + 1462.0, + 1350.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1460.0, + 297.0, + 1460.0, + 297.0, + 1494.0, + 294.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1460.0, + 910.0, + 1460.0, + 910.0, + 1494.0, + 316.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1460.0, + 1404.0, + 1460.0, + 1404.0, + 1494.0, + 926.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1488.0, + 1246.0, + 1488.0, + 1246.0, + 1523.0, + 291.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1488.0, + 1318.0, + 1488.0, + 1318.0, + 1523.0, + 1268.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1488.0, + 1407.0, + 1488.0, + 1407.0, + 1523.0, + 1338.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1517.0, + 1407.0, + 1517.0, + 1407.0, + 1554.0, + 291.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1550.0, + 1408.0, + 1550.0, + 1408.0, + 1584.0, + 295.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1577.0, + 418.0, + 1577.0, + 418.0, + 1616.0, + 292.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1577.0, + 744.0, + 1577.0, + 744.0, + 1616.0, + 436.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1577.0, + 1035.0, + 1577.0, + 1035.0, + 1616.0, + 850.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1577.0, + 1153.0, + 1577.0, + 1153.0, + 1616.0, + 1055.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 1577.0, + 1406.0, + 1577.0, + 1406.0, + 1616.0, + 1221.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1609.0, + 295.0, + 1609.0, + 295.0, + 1647.0, + 292.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1609.0, + 410.0, + 1609.0, + 410.0, + 1647.0, + 363.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 1609.0, + 749.0, + 1609.0, + 749.0, + 1647.0, + 565.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1609.0, + 966.0, + 1609.0, + 966.0, + 1647.0, + 768.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1609.0, + 1408.0, + 1609.0, + 1408.0, + 1647.0, + 986.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1640.0, + 737.0, + 1640.0, + 737.0, + 1678.0, + 292.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 1640.0, + 1030.0, + 1640.0, + 1030.0, + 1678.0, + 833.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 1640.0, + 1407.0, + 1640.0, + 1407.0, + 1678.0, + 1048.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1670.0, + 800.0, + 1670.0, + 800.0, + 1708.0, + 292.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 1670.0, + 867.0, + 1670.0, + 867.0, + 1708.0, + 819.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 1670.0, + 1406.0, + 1670.0, + 1406.0, + 1708.0, + 887.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1701.0, + 1246.0, + 1701.0, + 1246.0, + 1735.0, + 295.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 803.0, + 1408.0, + 803.0, + 1408.0, + 845.0, + 292.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 836.0, + 1402.0, + 836.0, + 1402.0, + 871.0, + 295.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 867.0, + 1405.0, + 867.0, + 1405.0, + 902.0, + 294.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 897.0, + 1405.0, + 897.0, + 1405.0, + 934.0, + 292.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 927.0, + 1405.0, + 927.0, + 1405.0, + 964.0, + 294.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 960.0, + 1406.0, + 960.0, + 1406.0, + 995.0, + 295.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 989.0, + 1227.0, + 989.0, + 1227.0, + 1026.0, + 294.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 989.0, + 1351.0, + 989.0, + 1351.0, + 1026.0, + 1321.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1820.0, + 1406.0, + 1820.0, + 1406.0, + 1855.0, + 294.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1851.0, + 1404.0, + 1851.0, + 1404.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1406.0, + 1882.0, + 1406.0, + 1917.0, + 294.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1913.0, + 728.0, + 1913.0, + 728.0, + 1944.0, + 296.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 1913.0, + 1405.0, + 1913.0, + 1405.0, + 1944.0, + 752.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1941.0, + 1073.0, + 1941.0, + 1073.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1941.0, + 1406.0, + 1941.0, + 1406.0, + 1976.0, + 1093.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 436.0, + 1972.0, + 436.0, + 2006.0, + 293.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1972.0, + 481.0, + 1972.0, + 481.0, + 2006.0, + 469.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1972.0, + 524.0, + 1972.0, + 524.0, + 2006.0, + 515.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2006.0, + 559.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2005.0, + 1313.0, + 2005.0, + 1313.0, + 2035.0, + 296.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 2005.0, + 1402.0, + 2005.0, + 1402.0, + 2035.0, + 1333.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 1407.0, + 228.0, + 1407.0, + 264.0, + 295.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 260.0, + 1405.0, + 260.0, + 1405.0, + 294.0, + 295.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 290.0, + 1405.0, + 290.0, + 1405.0, + 326.0, + 295.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 320.0, + 1405.0, + 320.0, + 1405.0, + 355.0, + 295.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 353.0, + 1406.0, + 353.0, + 1406.0, + 386.0, + 293.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 384.0, + 843.0, + 384.0, + 843.0, + 415.0, + 296.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 635.0, + 1216.0, + 635.0, + 1216.0, + 673.0, + 295.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 635.0, + 1404.0, + 635.0, + 1404.0, + 673.0, + 1239.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 667.0, + 1403.0, + 667.0, + 1403.0, + 703.0, + 295.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 699.0, + 1405.0, + 699.0, + 1405.0, + 736.0, + 294.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 729.0, + 1406.0, + 729.0, + 1406.0, + 766.0, + 294.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 758.0, + 644.0, + 758.0, + 644.0, + 797.0, + 293.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1267.0, + 538.0, + 1267.0, + 538.0, + 1310.0, + 293.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1267.0, + 689.0, + 1267.0, + 689.0, + 1310.0, + 677.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1109.0, + 370.0, + 1109.0, + 370.0, + 1149.0, + 294.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 1109.0, + 1337.0, + 1109.0, + 1337.0, + 1149.0, + 387.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1156.0, + 538.0, + 1156.0, + 538.0, + 1196.0, + 294.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 1156.0, + 680.0, + 1156.0, + 680.0, + 1196.0, + 668.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 578.0, + 1016.0, + 578.0, + 1016.0, + 623.0, + 293.0, + 623.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 26, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1558, + 1403, + 1558, + 1403, + 1805, + 298, + 1805 + ], + "score": 0.985 + }, + { + "category_id": 5, + "poly": [ + 297, + 1012, + 1397, + 1012, + 1397, + 1259, + 297, + 1259 + ], + "score": 0.98, + "html": "
TreatmentTreatment ResponseOptimal treatment
1f1(x,a1)=C((v)Tx+12(ν)T xa1-12(ν)T x²)(v)x a= 2(v3)x
2a2))(v)x 2(v)x
3(v2)x) f3(x,a3)=C((vi)Tx+12a3(a3 -b)²,where b=0.75
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1819, + 1403, + 1819, + 1403, + 2035, + 298, + 2035 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1300, + 1405, + 1300, + 1405, + 1454, + 297, + 1454 + ], + "score": 0.975 + }, + { + "category_id": 3, + "poly": [ + 403, + 274, + 1262, + 274, + 1262, + 611, + 403, + 611 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 295, + 775, + 1404, + 775, + 1404, + 840, + 295, + 840 + ], + "score": 0.941 + }, + { + "category_id": 6, + "poly": [ + 292, + 951, + 1399, + 951, + 1399, + 1011, + 292, + 1011 + ], + "score": 0.925 + }, + { + "category_id": 4, + "poly": [ + 293, + 644, + 1405, + 644, + 1405, + 709, + 293, + 709 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 300, + 854, + 1210, + 854, + 1210, + 886, + 300, + 886 + ], + "score": 0.91 + }, + { + "category_id": 2, + "poly": [ + 297, + 76, + 857, + 76, + 857, + 104, + 297, + 104 + ], + "score": 0.883 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2112, + 835, + 2112 + ], + "score": 0.856 + }, + { + "category_id": 0, + "poly": [ + 299, + 1500, + 1073, + 1500, + 1073, + 1533, + 299, + 1533 + ], + "score": 0.843 + }, + { + "category_id": 13, + "poly": [ + 1188, + 774, + 1398, + 774, + 1398, + 809, + 1188, + 809 + ], + "score": 0.9, + "latex": "\\textstyle { \\beta = { \\frac { \\alpha - 1 } { a ^ { * } } } + 2 - \\alpha }" + }, + { + "category_id": 13, + "poly": [ + 523, + 983, + 613, + 983, + 613, + 1011, + 523, + 1011 + ], + "score": 0.88, + "latex": "C = 1 0" + }, + { + "category_id": 13, + "poly": [ + 1214, + 809, + 1283, + 809, + 1283, + 840, + 1214, + 840 + ], + "score": 0.88, + "latex": "( \\alpha , \\tilde { \\beta } )" + }, + { + "category_id": 13, + "poly": [ + 757, + 778, + 832, + 778, + 832, + 808, + 757, + 808 + ], + "score": 0.87, + "latex": "\\alpha \\geq 1" + }, + { + "category_id": 13, + "poly": [ + 371, + 809, + 401, + 809, + 401, + 835, + 371, + 835 + ], + "score": 0.86, + "latex": "a ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 516, + 776, + 744, + 776, + 744, + 810, + 516, + 810 + ], + "score": 0.85, + "latex": "a \\mid \\mathbf { x } \\sim \\operatorname { B e t a } \\left( \\alpha , \\beta \\right)" + }, + { + "category_id": 13, + "poly": [ + 1314, + 811, + 1343, + 811, + 1343, + 835, + 1314, + 835 + ], + "score": 0.85, + "latex": "a ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 850, + 809, + 871, + 809, + 871, + 839, + 850, + 839 + ], + "score": 0.84, + "latex": "\\beta" + }, + { + "category_id": 14, + "poly": [ + 531, + 1126, + 998, + 1126, + 998, + 1179, + 531, + 1179 + ], + "score": 0.84, + "latex": "\\begin{array} { r } { f _ { 2 } ( \\mathbf { x } , a _ { 2 } ) = C \\left( \\left( \\mathbf { v } _ { 1 } ^ { 2 } \\right) ^ { T } \\mathbf { x } + \\sin \\left( \\pi \\left( \\frac { \\mathbf { v } _ { 2 } ^ { 2 T } \\mathbf { x } } { \\mathbf { v } _ { 3 } ^ { 2 T } \\mathbf { x } } \\right) a _ { 2 } \\right) \\right) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 449, + 1366, + 471, + 1366, + 471, + 1389, + 449, + 1389 + ], + "score": 0.8, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 1351, + 1306, + 1372, + 1306, + 1372, + 1328, + 1351, + 1328 + ], + "score": 0.77, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 506, + 585, + 523, + 585, + 523, + 609, + 506, + 609 + ], + "score": 0.73, + "latex": "_ y" + }, + { + "category_id": 13, + "poly": [ + 967, + 585, + 984, + 585, + 984, + 605, + 967, + 605 + ], + "score": 0.68, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 1292, + 1206, + 1379, + 1206, + 1379, + 1234, + 1292, + 1234 + ], + "score": 0.66, + "latex": "b < 0 . 7 5" + }, + { + "category_id": 13, + "poly": [ + 1119, + 1208, + 1134, + 1208, + 1134, + 1238, + 1119, + 1238 + ], + "score": 0.6, + "latex": "\\frac { 3 } { b }" + }, + { + "category_id": 14, + "poly": [ + 481, + 1059, + 1322, + 1059, + 1322, + 1114, + 481, + 1114 + ], + "score": 0.6, + "latex": "f _ { 1 } ( \\mathbf { x } , a _ { 1 } ) = C \\left( \\left( \\mathbf { v _ { 1 } ^ { 1 } } \\right) ^ { T } \\mathbf { x } + 1 2 \\left( \\mathbf { v _ { 2 } ^ { 1 } } \\right) ^ { T } \\mathbf { x } a _ { 1 } - 1 2 \\left( \\mathbf { v _ { 3 } ^ { 1 } } \\right) ^ { T } \\mathbf { x } a _ { 1 } ^ { 2 } \\right) \\qquad a _ { 1 } ^ { * } = \\frac { \\left( \\mathbf { v _ { 2 } ^ { 1 } } \\right) ^ { T } \\mathbf { x } } { 2 \\left( \\mathbf { v _ { 3 } ^ { 1 } } \\right) ^ { T } \\mathbf { x } }" + }, + { + "category_id": 13, + "poly": [ + 437, + 1195, + 844, + 1195, + 844, + 1246, + 437, + 1246 + ], + "score": 0.52, + "latex": "f _ { 3 } ( \\mathbf { x } , a _ { 3 } ) = C \\left( \\left( \\mathbf { v } _ { 1 } ^ { 3 } \\right) ^ { T } \\mathbf { x } + 1 2 a _ { 3 } ( a _ { 3 } - b ) ^ { 2 } \\right." + }, + { + "category_id": 13, + "poly": [ + 1301, + 957, + 1321, + 957, + 1321, + 979, + 1301, + 979 + ], + "score": 0.49, + "latex": "\\mathbf { x }" + }, + { + "category_id": 13, + "poly": [ + 921, + 1192, + 1092, + 1192, + 1092, + 1252, + 921, + 1252 + ], + "score": 0.46, + "latex": "b = 0 . 7 5 \\frac { { ( { { \\bf { v } } _ { 2 } ^ { 3 } } ) ^ { T } } { { \\bf { x } } } } { { ( { { \\bf { v } } _ { 3 } ^ { 3 } } ) ^ { T } } { \\bf { x } } } )" + }, + { + "category_id": 13, + "poly": [ + 801, + 677, + 825, + 677, + 825, + 705, + 801, + 705 + ], + "score": 0.46, + "latex": "\\mathrm { X }" + }, + { + "category_id": 13, + "poly": [ + 1160, + 1205, + 1268, + 1205, + 1268, + 1236, + 1160, + 1236 + ], + "score": 0.32, + "latex": "b \\geq 0 . 7 5 , 1" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 291.0, + 909.0, + 291.0, + 909.0, + 299.0, + 900.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 307.0, + 454.0, + 307.0, + 454.0, + 331.0, + 427.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 344.0, + 455.0, + 344.0, + 455.0, + 371.0, + 425.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 342.0, + 911.0, + 342.0, + 911.0, + 355.0, + 898.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 363.0, + 429.0, + 363.0, + 429.0, + 453.0, + 406.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 375.0, + 911.0, + 375.0, + 911.0, + 442.0, + 875.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 416.0, + 455.0, + 416.0, + 455.0, + 453.0, + 425.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 459.0, + 455.0, + 459.0, + 455.0, + 489.0, + 426.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 501.0, + 454.0, + 501.0, + 454.0, + 525.0, + 426.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 502.0, + 911.0, + 502.0, + 911.0, + 515.0, + 898.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 536.0, + 476.0, + 536.0, + 476.0, + 550.0, + 464.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 532.0, + 548.0, + 532.0, + 548.0, + 554.0, + 520.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 531.0, + 675.0, + 531.0, + 675.0, + 575.0, + 581.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 532.0, + 738.0, + 532.0, + 738.0, + 554.0, + 710.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 533.0, + 799.0, + 533.0, + 799.0, + 553.0, + 774.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 536.0, + 937.0, + 536.0, + 937.0, + 550.0, + 925.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 533.0, + 1007.0, + 533.0, + 1007.0, + 553.0, + 982.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 531.0, + 1135.0, + 531.0, + 1135.0, + 575.0, + 1041.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 532.0, + 1198.0, + 532.0, + 1198.0, + 554.0, + 1170.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 533.0, + 1260.0, + 533.0, + 1260.0, + 553.0, + 1235.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 576.0, + 505.0, + 576.0, + 505.0, + 611.0, + 467.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 576.0, + 771.0, + 576.0, + 771.0, + 611.0, + 524.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 577.0, + 966.0, + 577.0, + 966.0, + 611.0, + 929.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 577.0, + 1232.0, + 577.0, + 1232.0, + 611.0, + 985.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 381.5, + 447.0, + 381.5, + 447.0, + 411.0, + 432.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 947.0, + 1300.0, + 947.0, + 1300.0, + 987.0, + 291.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 947.0, + 1405.0, + 947.0, + 1405.0, + 987.0, + 1322.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 985.0, + 522.0, + 985.0, + 522.0, + 1013.0, + 295.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 985.0, + 621.0, + 985.0, + 621.0, + 1013.0, + 614.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 644.0, + 1404.0, + 644.0, + 1404.0, + 680.0, + 295.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 673.0, + 800.0, + 673.0, + 800.0, + 714.0, + 294.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 673.0, + 1281.0, + 673.0, + 1281.0, + 714.0, + 826.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2121.0, + 829.0, + 2121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1499.0, + 1074.0, + 1499.0, + 1074.0, + 1537.0, + 296.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1559.0, + 1404.0, + 1559.0, + 1404.0, + 1595.0, + 293.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1591.0, + 1403.0, + 1591.0, + 1403.0, + 1621.0, + 296.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1620.0, + 1405.0, + 1620.0, + 1405.0, + 1655.0, + 293.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1651.0, + 1405.0, + 1651.0, + 1405.0, + 1685.0, + 294.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1681.0, + 1405.0, + 1681.0, + 1405.0, + 1715.0, + 294.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1712.0, + 1404.0, + 1712.0, + 1404.0, + 1745.0, + 294.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1743.0, + 1407.0, + 1743.0, + 1407.0, + 1777.0, + 294.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1772.0, + 1223.0, + 1772.0, + 1223.0, + 1808.0, + 293.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1816.0, + 1407.0, + 1816.0, + 1407.0, + 1858.0, + 292.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 1404.0, + 1849.0, + 1404.0, + 1884.0, + 294.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1881.0, + 1405.0, + 1881.0, + 1405.0, + 1916.0, + 293.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1910.0, + 1405.0, + 1910.0, + 1405.0, + 1947.0, + 293.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1939.0, + 1407.0, + 1939.0, + 1407.0, + 1979.0, + 292.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2006.0, + 293.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2004.0, + 828.0, + 2004.0, + 828.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1298.0, + 1350.0, + 1298.0, + 1350.0, + 1336.0, + 294.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 1298.0, + 1406.0, + 1298.0, + 1406.0, + 1336.0, + 1373.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1328.0, + 1410.0, + 1328.0, + 1410.0, + 1369.0, + 294.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1358.0, + 448.0, + 1358.0, + 448.0, + 1397.0, + 294.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 1358.0, + 1405.0, + 1358.0, + 1405.0, + 1397.0, + 472.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1388.0, + 1406.0, + 1388.0, + 1406.0, + 1431.0, + 291.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1417.0, + 360.0, + 1417.0, + 360.0, + 1459.0, + 293.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 768.0, + 515.0, + 768.0, + 515.0, + 815.0, + 290.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 768.0, + 756.0, + 768.0, + 756.0, + 815.0, + 745.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 768.0, + 1187.0, + 768.0, + 1187.0, + 815.0, + 833.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 768.0, + 1411.0, + 768.0, + 1411.0, + 815.0, + 1399.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 804.0, + 370.0, + 804.0, + 370.0, + 842.0, + 292.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 804.0, + 849.0, + 804.0, + 849.0, + 842.0, + 402.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 804.0, + 1213.0, + 804.0, + 1213.0, + 842.0, + 872.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 804.0, + 1313.0, + 804.0, + 1313.0, + 842.0, + 1284.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 804.0, + 1357.0, + 804.0, + 1357.0, + 842.0, + 1344.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 851.0, + 1214.0, + 851.0, + 1214.0, + 890.0, + 294.0, + 890.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 27, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1088, + 1406, + 1088, + 1406, + 1346, + 297, + 1346 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 828, + 1405, + 828, + 1405, + 1076, + 297, + 1076 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1549, + 1405, + 1549, + 1405, + 1705, + 298, + 1705 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1910, + 1404, + 1910, + 1404, + 2040, + 299, + 2040 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 456, + 228, + 1237, + 228, + 1237, + 701, + 456, + 701 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 1718, + 1403, + 1718, + 1403, + 1814, + 298, + 1814 + ], + "score": 0.971 + }, + { + "category_id": 8, + "poly": [ + 683, + 1362, + 1017, + 1362, + 1017, + 1533, + 683, + 1533 + ], + "score": 0.953 + }, + { + "category_id": 4, + "poly": [ + 470, + 738, + 1227, + 738, + 1227, + 772, + 470, + 772 + ], + "score": 0.924 + }, + { + "category_id": 0, + "poly": [ + 299, + 1853, + 664, + 1853, + 664, + 1884, + 299, + 1884 + ], + "score": 0.908 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.859 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 856, + 76, + 856, + 104, + 298, + 104 + ], + "score": 0.553 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 855, + 76, + 855, + 103, + 300, + 103 + ], + "score": 0.546 + }, + { + "category_id": 13, + "poly": [ + 572, + 1013, + 655, + 1013, + 655, + 1046, + 572, + 1046 + ], + "score": 0.95, + "latex": "\\pi ^ { * } ( c | t )" + }, + { + "category_id": 13, + "poly": [ + 363, + 1121, + 567, + 1121, + 567, + 1159, + 363, + 1159 + ], + "score": 0.93, + "latex": "\\hat { M } \\big ( y _ { t + 1 } \\vert s _ { t } , p _ { t } , c _ { t } \\big )" + }, + { + "category_id": 13, + "poly": [ + 742, + 2002, + 822, + 2002, + 822, + 2043, + 742, + 2043 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\dot { \\omega } = \\frac { 1 } { \\hat { \\mu } } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 682, + 1363, + 1017, + 1363, + 1017, + 1537, + 682, + 1537 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\underset { \\pi } { \\operatorname* { m a x } } \\mathbb { E } _ { s _ { 0 } \\sim { \\cal S } } \\left[ \\displaystyle \\sum _ { t = 0 } ^ { 2 3 } y _ { t } \\vert \\hat { M } _ { f } , \\pi \\right] , } \\\\ { \\mathrm { s . t . , } \\displaystyle \\sum _ { t , s \\in { \\cal S } } c _ { t } y _ { t } \\leq C } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 569, + 1247, + 860, + 1247, + 860, + 1281, + 569, + 1281 + ], + "score": 0.91, + "latex": "s _ { t + 1 } = f ( s _ { t } , p _ { t } , c _ { t } , y _ { t + 1 } )" + }, + { + "category_id": 13, + "poly": [ + 1118, + 860, + 1211, + 860, + 1211, + 889, + 1118, + 889 + ], + "score": 0.91, + "latex": "c \\in \\mathcal { N } ^ { + }" + }, + { + "category_id": 13, + "poly": [ + 546, + 1279, + 590, + 1279, + 590, + 1317, + 546, + 1317 + ], + "score": 0.9, + "latex": "\\hat { M } _ { f }" + }, + { + "category_id": 13, + "poly": [ + 656, + 1129, + 711, + 1129, + 711, + 1158, + 656, + 1158 + ], + "score": 0.89, + "latex": "y _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 756, + 893, + 845, + 893, + 845, + 922, + 756, + 922 + ], + "score": 0.87, + "latex": "c = 0 . 5" + }, + { + "category_id": 13, + "poly": [ + 408, + 1278, + 442, + 1278, + 442, + 1311, + 408, + 1311 + ], + "score": 0.87, + "latex": "\\hat { M }" + }, + { + "category_id": 13, + "poly": [ + 455, + 894, + 536, + 894, + 536, + 921, + 455, + 921 + ], + "score": 0.87, + "latex": "t = 1 0" + }, + { + "category_id": 13, + "poly": [ + 1170, + 1159, + 1197, + 1159, + 1197, + 1186, + 1170, + 1186 + ], + "score": 0.86, + "latex": "s _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1109, + 1218, + 1129, + 1218, + 1129, + 1249, + 1109, + 1249 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 913, + 2005, + 932, + 2005, + 932, + 2035, + 913, + 2035 + ], + "score": 0.85, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 1009, + 984, + 1034, + 984, + 1034, + 1010, + 1009, + 1010 + ], + "score": 0.84, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 430, + 1162, + 457, + 1162, + 457, + 1187, + 430, + 1187 + ], + "score": 0.84, + "latex": "p _ { t }" + }, + { + "category_id": 13, + "poly": [ + 443, + 861, + 632, + 861, + 632, + 893, + 443, + 893 + ], + "score": 0.83, + "latex": "t \\in [ 0 , 1 , 2 . . . , 2 3 ]" + }, + { + "category_id": 13, + "poly": [ + 492, + 1281, + 513, + 1281, + 513, + 1315, + 492, + 1315 + ], + "score": 0.83, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 899, + 1014, + 924, + 1014, + 924, + 1041, + 899, + 1041 + ], + "score": 0.82, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 642, + 892, + 696, + 892, + 696, + 921, + 642, + 921 + ], + "score": 0.82, + "latex": "0 . 5 \\ S" + }, + { + "category_id": 13, + "poly": [ + 513, + 982, + 566, + 982, + 566, + 1012, + 513, + 1012 + ], + "score": 0.74, + "latex": "2 . 5 \\ S" + }, + { + "category_id": 13, + "poly": [ + 1306, + 1131, + 1332, + 1131, + 1332, + 1156, + 1306, + 1156 + ], + "score": 0.71, + "latex": "c _ { t }" + }, + { + "category_id": 13, + "poly": [ + 556, + 1787, + 573, + 1787, + 573, + 1809, + 556, + 1809 + ], + "score": 0.68, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 1065, + 951, + 1097, + 951, + 1097, + 981, + 1065, + 981 + ], + "score": 0.67, + "latex": "2 \\mathfrak { H }" + }, + { + "category_id": 13, + "poly": [ + 905, + 1158, + 918, + 1158, + 918, + 1183, + 905, + 1183 + ], + "score": 0.67, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 1103, + 1753, + 1118, + 1753, + 1118, + 1778, + 1103, + 1778 + ], + "score": 0.67, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 1268, + 1132, + 1295, + 1132, + 1295, + 1156, + 1268, + 1156 + ], + "score": 0.58, + "latex": "s _ { t }" + }, + { + "category_id": 13, + "poly": [ + 296, + 892, + 387, + 892, + 387, + 920, + 296, + 920 + ], + "score": 0.33, + "latex": "1 0 \\mathrm { A . M }" + }, + { + "category_id": 13, + "poly": [ + 295, + 922, + 388, + 922, + 388, + 951, + 295, + 951 + ], + "score": 0.31, + "latex": "1 0 \\mathrm { \\ A . M }" + }, + { + "category_id": 13, + "poly": [ + 1268, + 1129, + 1332, + 1129, + 1332, + 1157, + 1268, + 1157 + ], + "score": 0.27, + "latex": "s _ { t } , c _ { t }" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 227.0, + 763.0, + 227.0, + 763.0, + 252.0, + 629.0, + 252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 246.0, + 803.0, + 246.0, + 803.0, + 273.0, + 588.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 269.0, + 778.0, + 269.0, + 778.0, + 290.0, + 617.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 261.0, + 1177.0, + 261.0, + 1177.0, + 285.0, + 1033.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 300.0, + 1133.0, + 300.0, + 1133.0, + 310.0, + 1124.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 315.0, + 726.0, + 315.0, + 726.0, + 325.0, + 719.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 357.0, + 928.0, + 357.0, + 928.0, + 380.0, + 875.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 395.0, + 1099.0, + 395.0, + 1099.0, + 450.0, + 1052.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 409.0, + 1240.0, + 409.0, + 1240.0, + 438.0, + 1112.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 435.0, + 796.0, + 435.0, + 796.0, + 458.0, + 672.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 433.0, + 1214.0, + 433.0, + 1214.0, + 459.0, + 1137.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 464.0, + 556.0, + 464.0, + 556.0, + 486.0, + 475.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 456.0, + 788.0, + 456.0, + 788.0, + 478.0, + 681.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 450.0, + 1235.0, + 450.0, + 1235.0, + 479.0, + 1119.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 474.0, + 799.0, + 474.0, + 799.0, + 499.0, + 673.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 507.0, + 777.0, + 507.0, + 777.0, + 571.0, + 705.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 576.0, + 1023.0, + 576.0, + 1023.0, + 600.0, + 978.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 624.0, + 1085.0, + 624.0, + 1085.0, + 657.0, + 911.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 653.0, + 798.0, + 653.0, + 798.0, + 665.0, + 785.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 650.0, + 1077.0, + 650.0, + 1077.0, + 674.0, + 918.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 671.0, + 826.0, + 671.0, + 826.0, + 695.0, + 754.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 667.0, + 1066.0, + 667.0, + 1066.0, + 695.0, + 930.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.25, + 307.5, + 722.25, + 307.5, + 722.25, + 340.5, + 684.25, + 340.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 339.0, + 642.0, + 339.0, + 642.0, + 368.0, + 613.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 408.5, + 532.0, + 408.5, + 532.0, + 447.5, + 500.0, + 447.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 554.0, + 1030.0, + 554.0, + 1030.0, + 580.0, + 966.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 733.0, + 1228.0, + 733.0, + 1228.0, + 777.0, + 469.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1850.0, + 668.0, + 1850.0, + 668.0, + 1888.0, + 296.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 107.0, + 297.0, + 107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1086.0, + 1404.0, + 1086.0, + 1404.0, + 1128.0, + 294.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1121.0, + 362.0, + 1121.0, + 362.0, + 1162.0, + 292.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1121.0, + 655.0, + 1121.0, + 655.0, + 1162.0, + 568.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 1121.0, + 1267.0, + 1121.0, + 1267.0, + 1162.0, + 712.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 1121.0, + 1407.0, + 1121.0, + 1407.0, + 1162.0, + 1333.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1154.0, + 429.0, + 1154.0, + 429.0, + 1190.0, + 294.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 1154.0, + 904.0, + 1154.0, + 904.0, + 1190.0, + 458.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 1154.0, + 1169.0, + 1154.0, + 1169.0, + 1190.0, + 919.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1154.0, + 1406.0, + 1154.0, + 1406.0, + 1190.0, + 1198.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1186.0, + 1408.0, + 1186.0, + 1408.0, + 1221.0, + 295.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1218.0, + 1108.0, + 1218.0, + 1108.0, + 1249.0, + 296.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1218.0, + 1404.0, + 1218.0, + 1404.0, + 1249.0, + 1130.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1244.0, + 568.0, + 1244.0, + 568.0, + 1287.0, + 292.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 1244.0, + 1408.0, + 1244.0, + 1408.0, + 1287.0, + 861.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1279.0, + 407.0, + 1279.0, + 407.0, + 1318.0, + 294.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 1279.0, + 491.0, + 1279.0, + 491.0, + 1318.0, + 443.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1279.0, + 545.0, + 1279.0, + 545.0, + 1318.0, + 514.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 1279.0, + 1409.0, + 1279.0, + 1409.0, + 1318.0, + 591.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1315.0, + 844.0, + 1315.0, + 844.0, + 1346.0, + 296.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 830.0, + 1405.0, + 830.0, + 1405.0, + 864.0, + 294.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 858.0, + 442.0, + 858.0, + 442.0, + 895.0, + 294.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 858.0, + 1117.0, + 858.0, + 1117.0, + 895.0, + 633.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 858.0, + 1407.0, + 858.0, + 1407.0, + 895.0, + 1212.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 889.0, + 295.0, + 889.0, + 295.0, + 924.0, + 292.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 889.0, + 454.0, + 889.0, + 454.0, + 924.0, + 388.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 889.0, + 641.0, + 889.0, + 641.0, + 924.0, + 537.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 889.0, + 755.0, + 889.0, + 755.0, + 924.0, + 697.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 889.0, + 1405.0, + 889.0, + 1405.0, + 924.0, + 846.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 919.0, + 1405.0, + 919.0, + 1405.0, + 956.0, + 389.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 950.0, + 1064.0, + 950.0, + 1064.0, + 987.0, + 294.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 950.0, + 1407.0, + 950.0, + 1407.0, + 987.0, + 1098.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 982.0, + 512.0, + 982.0, + 512.0, + 1016.0, + 294.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 982.0, + 1008.0, + 982.0, + 1008.0, + 1016.0, + 567.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 982.0, + 1406.0, + 982.0, + 1406.0, + 1016.0, + 1035.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1010.0, + 571.0, + 1010.0, + 571.0, + 1051.0, + 292.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1010.0, + 898.0, + 1010.0, + 898.0, + 1051.0, + 656.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1010.0, + 1407.0, + 1010.0, + 1407.0, + 1051.0, + 925.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1039.0, + 447.0, + 1039.0, + 447.0, + 1082.0, + 294.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1550.0, + 1405.0, + 1550.0, + 1405.0, + 1587.0, + 295.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1582.0, + 1406.0, + 1582.0, + 1406.0, + 1615.0, + 296.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1611.0, + 1405.0, + 1611.0, + 1405.0, + 1647.0, + 295.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1639.0, + 1406.0, + 1639.0, + 1406.0, + 1683.0, + 292.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1672.0, + 457.0, + 1672.0, + 457.0, + 1706.0, + 295.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1912.0, + 1405.0, + 1912.0, + 1405.0, + 1945.0, + 295.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1943.0, + 1405.0, + 1943.0, + 1405.0, + 1976.0, + 294.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1971.0, + 1407.0, + 1971.0, + 1407.0, + 2006.0, + 293.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2000.0, + 741.0, + 2000.0, + 741.0, + 2044.0, + 295.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 2000.0, + 912.0, + 2000.0, + 912.0, + 2044.0, + 823.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 2000.0, + 1403.0, + 2000.0, + 1403.0, + 2044.0, + 933.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1717.0, + 1405.0, + 1717.0, + 1405.0, + 1756.0, + 293.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1751.0, + 1102.0, + 1751.0, + 1102.0, + 1786.0, + 296.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1751.0, + 1404.0, + 1751.0, + 1404.0, + 1786.0, + 1119.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1781.0, + 555.0, + 1781.0, + 555.0, + 1817.0, + 295.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1781.0, + 695.0, + 1781.0, + 695.0, + 1817.0, + 574.0, + 1817.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 28, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1154, + 1405, + 1154, + 1405, + 1461, + 298, + 1461 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 723, + 1405, + 723, + 1405, + 1049, + 297, + 1049 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 428, + 1406, + 428, + 1406, + 617, + 298, + 617 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 627, + 1560, + 1064, + 1560, + 1064, + 1684, + 627, + 1684 + ], + "score": 0.975, + "html": "
ParameterValues
Number of samples3,5,7,9, 11
入0.1,1,10, 20
" + }, + { + "category_id": 1, + "poly": [ + 299, + 229, + 1404, + 229, + 1404, + 323, + 299, + 323 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 298, + 1971, + 1401, + 1971, + 1401, + 2033, + 298, + 2033 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 301, + 1825, + 870, + 1825, + 870, + 1859, + 301, + 1859 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 299, + 1763, + 625, + 1763, + 625, + 1796, + 299, + 1796 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 299, + 1094, + 508, + 1094, + 508, + 1127, + 299, + 1127 + ], + "score": 0.905 + }, + { + "category_id": 0, + "poly": [ + 301, + 1907, + 700, + 1907, + 700, + 1941, + 301, + 1941 + ], + "score": 0.903 + }, + { + "category_id": 6, + "poly": [ + 571, + 1527, + 1119, + 1527, + 1119, + 1559, + 571, + 1559 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 298, + 665, + 893, + 665, + 893, + 697, + 298, + 697 + ], + "score": 0.896 + }, + { + "category_id": 0, + "poly": [ + 301, + 369, + 743, + 369, + 743, + 402, + 301, + 402 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.853 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 856, + 75, + 856, + 105, + 298, + 105 + ], + "score": 0.846 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.099 + }, + { + "category_id": 13, + "poly": [ + 531, + 976, + 832, + 976, + 832, + 1022, + 531, + 1022 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { { \\frac { 1 } { n } } \\sum _ { i = 1 } ^ { n } { \\frac { 1 } { { \\hat { \\mu } } \\left( a _ { i } | x _ { i } \\right) } } \\left( y _ { i } - \\hat { y } _ { i } \\right) ^ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1154, + 784, + 1223, + 784, + 1223, + 826, + 1154, + 826 + ], + "score": 0.93, + "latex": "\\dot { \\overline { { { \\hat { \\mu } } ( a | x ) } } }" + }, + { + "category_id": 13, + "poly": [ + 1092, + 824, + 1171, + 824, + 1171, + 857, + 1092, + 857 + ], + "score": 0.91, + "latex": "{ \\hat { \\mu } } ( a | x )" + }, + { + "category_id": 13, + "poly": [ + 1077, + 490, + 1397, + 490, + 1397, + 531, + 1077, + 531 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\boldsymbol { M } \\boldsymbol { \\bar { S } } \\boldsymbol { E } = \\frac { 1 } { n } \\sum _ { i = 1 } ^ { n } \\left( y _ { i } - \\hat { y } _ { i } \\right) ^ { 2 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 945, + 504, + 945, + 504, + 979, + 298, + 979 + ], + "score": 0.91, + "latex": "\\hat { \\mu } \\gets \\operatorname* { m i n } ( \\hat { \\mu } , 0 . 0 5 )" + }, + { + "category_id": 13, + "poly": [ + 1248, + 555, + 1307, + 555, + 1307, + 585, + 1248, + 585 + ], + "score": 0.9, + "latex": "3 e ^ { - \\bar { 4 } }" + }, + { + "category_id": 13, + "poly": [ + 1200, + 1430, + 1283, + 1430, + 1283, + 1458, + 1200, + 1458 + ], + "score": 0.9, + "latex": "\\lambda = 1 0" + }, + { + "category_id": 13, + "poly": [ + 1066, + 528, + 1084, + 528, + 1084, + 557, + 1066, + 557 + ], + "score": 0.84, + "latex": "\\hat { y }" + }, + { + "category_id": 13, + "poly": [ + 982, + 856, + 1003, + 856, + 1003, + 887, + 982, + 887 + ], + "score": 0.83, + "latex": "\\hat { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 683, + 530, + 702, + 530, + 702, + 557, + 683, + 557 + ], + "score": 0.79, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1307, + 790, + 1326, + 790, + 1326, + 813, + 1307, + 813 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 370, + 531, + 391, + 531, + 391, + 553, + 370, + 553 + ], + "score": 0.74, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 748, + 860, + 768, + 860, + 768, + 882, + 748, + 882 + ], + "score": 0.73, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 524, + 860, + 542, + 860, + 542, + 882, + 524, + 882 + ], + "score": 0.71, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 703, + 829, + 721, + 829, + 721, + 851, + 703, + 851 + ], + "score": 0.7, + "latex": "a" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1762.0, + 629.0, + 1762.0, + 629.0, + 1797.0, + 295.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1090.0, + 513.0, + 1090.0, + 513.0, + 1133.0, + 294.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1905.0, + 703.0, + 1905.0, + 703.0, + 1944.0, + 295.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 1525.0, + 1122.0, + 1525.0, + 1122.0, + 1565.0, + 569.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 662.0, + 896.0, + 662.0, + 896.0, + 703.0, + 294.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 363.0, + 747.0, + 363.0, + 747.0, + 410.0, + 293.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2120.0, + 830.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1152.0, + 1405.0, + 1152.0, + 1405.0, + 1189.0, + 293.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1182.0, + 1410.0, + 1182.0, + 1410.0, + 1222.0, + 293.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1217.0, + 1405.0, + 1217.0, + 1405.0, + 1249.0, + 296.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1246.0, + 1406.0, + 1246.0, + 1406.0, + 1282.0, + 295.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1277.0, + 1405.0, + 1277.0, + 1405.0, + 1313.0, + 292.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1304.0, + 1405.0, + 1304.0, + 1405.0, + 1346.0, + 293.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1339.0, + 1405.0, + 1339.0, + 1405.0, + 1371.0, + 296.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1367.0, + 1407.0, + 1367.0, + 1407.0, + 1403.0, + 292.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1399.0, + 1406.0, + 1399.0, + 1406.0, + 1435.0, + 293.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1430.0, + 1199.0, + 1430.0, + 1199.0, + 1465.0, + 295.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1430.0, + 1293.0, + 1430.0, + 1293.0, + 1465.0, + 1284.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 723.0, + 1406.0, + 723.0, + 1406.0, + 760.0, + 295.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 755.0, + 1405.0, + 755.0, + 1405.0, + 789.0, + 295.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 781.0, + 1153.0, + 781.0, + 1153.0, + 829.0, + 291.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 781.0, + 1306.0, + 781.0, + 1306.0, + 829.0, + 1224.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1327.0, + 781.0, + 1405.0, + 781.0, + 1405.0, + 829.0, + 1327.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 822.0, + 702.0, + 822.0, + 702.0, + 856.0, + 296.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 822.0, + 1091.0, + 822.0, + 1091.0, + 856.0, + 722.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 822.0, + 1402.0, + 822.0, + 1402.0, + 856.0, + 1172.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 853.0, + 523.0, + 853.0, + 523.0, + 887.0, + 295.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 853.0, + 747.0, + 853.0, + 747.0, + 887.0, + 543.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 853.0, + 981.0, + 853.0, + 981.0, + 887.0, + 769.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 853.0, + 1403.0, + 853.0, + 1403.0, + 887.0, + 1004.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 885.0, + 1406.0, + 885.0, + 1406.0, + 919.0, + 295.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 911.0, + 1410.0, + 911.0, + 1410.0, + 951.0, + 292.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 943.0, + 297.0, + 943.0, + 297.0, + 983.0, + 292.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 943.0, + 1407.0, + 943.0, + 1407.0, + 983.0, + 505.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 977.0, + 1408.0, + 977.0, + 1408.0, + 1015.0, + 833.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1011.0, + 380.0, + 1011.0, + 380.0, + 1052.0, + 291.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.25, + 972.0, + 724.25, + 972.0, + 724.25, + 1025.5, + 288.25, + 1025.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 428.0, + 1406.0, + 428.0, + 1406.0, + 464.0, + 295.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 462.0, + 1404.0, + 462.0, + 1404.0, + 494.0, + 296.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 483.0, + 369.0, + 483.0, + 369.0, + 549.0, + 282.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 483.0, + 1076.0, + 483.0, + 1076.0, + 549.0, + 392.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 483.0, + 1415.0, + 483.0, + 1415.0, + 549.0, + 1398.0, + 549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 525.0, + 369.0, + 525.0, + 369.0, + 561.0, + 295.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 525.0, + 682.0, + 525.0, + 682.0, + 561.0, + 392.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 525.0, + 1065.0, + 525.0, + 1065.0, + 561.0, + 703.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 525.0, + 1404.0, + 525.0, + 1404.0, + 561.0, + 1085.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 555.0, + 1247.0, + 555.0, + 1247.0, + 591.0, + 295.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1308.0, + 555.0, + 1406.0, + 555.0, + 1406.0, + 591.0, + 1308.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 584.0, + 1251.0, + 584.0, + 1251.0, + 617.0, + 296.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 230.0, + 1405.0, + 230.0, + 1405.0, + 264.0, + 294.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 261.0, + 1405.0, + 261.0, + 1405.0, + 294.0, + 293.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 292.0, + 916.0, + 292.0, + 916.0, + 326.0, + 295.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1970.0, + 1405.0, + 1970.0, + 1405.0, + 2006.0, + 294.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 574.0, + 2003.0, + 574.0, + 2034.0, + 294.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1825.0, + 875.0, + 1825.0, + 875.0, + 1861.0, + 297.0, + 1861.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 29, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 301, + 285, + 1577, + 285, + 1577, + 770, + 301, + 770 + ], + "score": 0.983, + "html": "
ParameterGNFCTAGCMuJoCoBAT
hidden layers of all neural networks4455
hidden units of all neural networks256256512512
collect samples for each time of model update500050004000096000
batch size of discriminators500050004000080000
horizon50110048 (half-hours)
∈μ(also ∈D)0.0050.010.05 (0.1 for walker2d)0.05
times for discriminator update2215
times for model update11220
times for supervised learning update11420
learning rate for supervised learning1e-51e-53e-41e-5
Y0.990.00.990.99
clip-ratioNANNANNAN0.1
max DkL0.0010.0010.001NAN
optimization algorithm (the first term of Eq.23)TRPOTRPOTRPOPPO
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1092, + 1406, + 1092, + 1406, + 1247, + 298, + 1247 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1345, + 938, + 1345, + 938, + 1894, + 298, + 1894 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 961, + 1382, + 1394, + 1382, + 1394, + 1763, + 961, + 1763 + ], + "score": 0.958 + }, + { + "category_id": 4, + "poly": [ + 956, + 1802, + 1402, + 1802, + 1402, + 1893, + 956, + 1893 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 299, + 1972, + 1399, + 1972, + 1399, + 2035, + 299, + 2035 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 300, + 1895, + 1401, + 1895, + 1401, + 1956, + 300, + 1956 + ], + "score": 0.922 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.913 + }, + { + "category_id": 1, + "poly": [ + 293, + 959, + 1334, + 959, + 1334, + 993, + 293, + 993 + ], + "score": 0.912 + }, + { + "category_id": 6, + "poly": [ + 539, + 250, + 1149, + 250, + 1149, + 281, + 539, + 281 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 301, + 830, + 684, + 830, + 684, + 866, + 301, + 866 + ], + "score": 0.903 + }, + { + "category_id": 0, + "poly": [ + 300, + 1288, + 608, + 1288, + 608, + 1318, + 300, + 1318 + ], + "score": 0.883 + }, + { + "category_id": 0, + "poly": [ + 301, + 1033, + 742, + 1033, + 742, + 1065, + 301, + 1065 + ], + "score": 0.855 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 862, + 2088, + 862, + 2113, + 835, + 2113 + ], + "score": 0.821 + }, + { + "category_id": 0, + "poly": [ + 300, + 901, + 705, + 901, + 705, + 933, + 300, + 933 + ], + "score": 0.608 + }, + { + "category_id": 1, + "poly": [ + 300, + 901, + 705, + 901, + 705, + 933, + 300, + 933 + ], + "score": 0.354 + }, + { + "category_id": 13, + "poly": [ + 298, + 1804, + 408, + 1804, + 408, + 1840, + 298, + 1840 + ], + "score": 0.92, + "latex": "T _ { \\varphi _ { 1 } ^ { * } } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 626, + 1802, + 850, + 1802, + 850, + 1840, + 626, + 1840 + ], + "score": 0.88, + "latex": "\\mathbb { E } _ { M _ { \\theta } } \\left[ T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 485, + 1124, + 541, + 1124, + 541, + 1153, + 485, + 1153 + ], + "score": 0.88, + "latex": "20 \\%" + }, + { + "category_id": 13, + "poly": [ + 815, + 1504, + 846, + 1504, + 846, + 1534, + 815, + 1534 + ], + "score": 0.87, + "latex": "\\epsilon _ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 296, + 1409, + 466, + 1409, + 466, + 1440, + 296, + 1440 + ], + "score": 0.87, + "latex": "\\mathtt { e 0 } . 2 _ { - \\mathtt { p } 0 . 0 5 }" + }, + { + "category_id": 13, + "poly": [ + 545, + 1622, + 572, + 1622, + 572, + 1648, + 545, + 1648 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 898, + 1504, + 932, + 1504, + 932, + 1531, + 898, + 1531 + ], + "score": 0.81, + "latex": "\\epsilon _ { D }" + }, + { + "category_id": 13, + "poly": [ + 555, + 1778, + 576, + 1778, + 576, + 1805, + 555, + 1805 + ], + "score": 0.81, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 321, + 490, + 350, + 490, + 350, + 521, + 321, + 521 + ], + "score": 0.8, + "latex": "\\epsilon _ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 414, + 490, + 449, + 490, + 449, + 517, + 414, + 517 + ], + "score": 0.72, + "latex": "\\epsilon _ { D }" + }, + { + "category_id": 13, + "poly": [ + 375, + 699, + 438, + 699, + 438, + 731, + 375, + 731 + ], + "score": 0.64, + "latex": "D _ { K L }" + }, + { + "category_id": 13, + "poly": [ + 320, + 645, + 341, + 645, + 341, + 671, + 320, + 671 + ], + "score": 0.52, + "latex": "\\gamma" + }, + { + "category_id": 13, + "poly": [ + 624, + 1805, + 678, + 1805, + 678, + 1837, + 624, + 1837 + ], + "score": 0.37, + "latex": "\\mathbb { E } _ { M _ { \\theta } }" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1394.0, + 1009.0, + 1394.0, + 1009.0, + 1417.0, + 984.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1454.0, + 1008.0, + 1454.0, + 1008.0, + 1474.0, + 992.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1499.0, + 988.0, + 1499.0, + 988.0, + 1592.0, + 956.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1514.0, + 1006.0, + 1514.0, + 1006.0, + 1530.0, + 994.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1574.0, + 1005.0, + 1574.0, + 1005.0, + 1588.0, + 993.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1630.0, + 1007.0, + 1630.0, + 1007.0, + 1649.0, + 992.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1691.0, + 1005.0, + 1691.0, + 1005.0, + 1704.0, + 994.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.25, + 1705.0, + 1088.25, + 1705.0, + 1088.25, + 1731.5, + 1008.25, + 1731.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1708.0, + 1157.0, + 1708.0, + 1157.0, + 1738.0, + 1051.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1704.5, + 1212.0, + 1704.5, + 1212.0, + 1741.0, + 1106.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.75, + 1707.0, + 1266.75, + 1707.0, + 1266.75, + 1737.0, + 1164.75, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 1709.0, + 1321.0, + 1709.0, + 1321.0, + 1736.0, + 1220.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1717.0, + 1401.0, + 1717.0, + 1401.0, + 1746.0, + 1251.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1801.0, + 1405.0, + 1801.0, + 1405.0, + 1833.0, + 954.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1834.0, + 1404.0, + 1834.0, + 1404.0, + 1864.0, + 956.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 1867.0, + 1023.0, + 1867.0, + 1023.0, + 1894.0, + 955.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 248.0, + 1148.0, + 248.0, + 1148.0, + 284.0, + 540.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 830.0, + 685.0, + 830.0, + 685.0, + 870.0, + 295.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1287.0, + 612.0, + 1287.0, + 612.0, + 1322.0, + 296.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1033.0, + 744.0, + 1033.0, + 744.0, + 1069.0, + 295.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 901.0, + 708.0, + 901.0, + 708.0, + 937.0, + 295.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1092.0, + 1403.0, + 1092.0, + 1403.0, + 1126.0, + 297.0, + 1126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1124.0, + 484.0, + 1124.0, + 484.0, + 1158.0, + 296.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 1124.0, + 1403.0, + 1124.0, + 1403.0, + 1158.0, + 542.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1152.0, + 1407.0, + 1152.0, + 1407.0, + 1191.0, + 292.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1184.0, + 1406.0, + 1184.0, + 1406.0, + 1220.0, + 293.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1212.0, + 1266.0, + 1212.0, + 1266.0, + 1253.0, + 295.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1345.0, + 937.0, + 1345.0, + 937.0, + 1381.0, + 294.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1376.0, + 938.0, + 1376.0, + 938.0, + 1410.0, + 294.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 1407.0, + 937.0, + 1407.0, + 937.0, + 1441.0, + 467.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1438.0, + 940.0, + 1438.0, + 940.0, + 1472.0, + 295.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1468.0, + 940.0, + 1468.0, + 940.0, + 1503.0, + 293.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1500.0, + 814.0, + 1500.0, + 814.0, + 1534.0, + 294.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1500.0, + 897.0, + 1500.0, + 897.0, + 1534.0, + 847.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 1500.0, + 937.0, + 1500.0, + 937.0, + 1534.0, + 933.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1529.0, + 939.0, + 1529.0, + 939.0, + 1564.0, + 293.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1561.0, + 937.0, + 1561.0, + 937.0, + 1594.0, + 294.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1590.0, + 938.0, + 1590.0, + 938.0, + 1625.0, + 294.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1622.0, + 544.0, + 1622.0, + 544.0, + 1653.0, + 295.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1622.0, + 940.0, + 1622.0, + 940.0, + 1653.0, + 573.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1653.0, + 937.0, + 1653.0, + 937.0, + 1684.0, + 295.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1683.0, + 940.0, + 1683.0, + 940.0, + 1713.0, + 294.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1714.0, + 938.0, + 1714.0, + 938.0, + 1745.0, + 294.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1742.0, + 937.0, + 1742.0, + 937.0, + 1777.0, + 295.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1775.0, + 554.0, + 1775.0, + 554.0, + 1806.0, + 295.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1775.0, + 937.0, + 1775.0, + 937.0, + 1806.0, + 577.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1803.0, + 297.0, + 1803.0, + 297.0, + 1840.0, + 294.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 1803.0, + 623.0, + 1803.0, + 623.0, + 1840.0, + 409.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1803.0, + 937.0, + 1803.0, + 937.0, + 1840.0, + 851.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1833.0, + 936.0, + 1833.0, + 936.0, + 1868.0, + 294.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1866.0, + 939.0, + 1866.0, + 939.0, + 1897.0, + 296.0, + 1897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1970.0, + 1404.0, + 1970.0, + 1404.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2002.0, + 1403.0, + 2002.0, + 1403.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1892.0, + 1405.0, + 1892.0, + 1405.0, + 1929.0, + 293.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1926.0, + 643.0, + 1926.0, + 643.0, + 1957.0, + 296.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 958.0, + 1341.0, + 958.0, + 1341.0, + 996.0, + 293.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 901.0, + 708.0, + 901.0, + 708.0, + 937.0, + 295.0, + 937.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 30, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1584, + 1405, + 1584, + 1405, + 1891, + 297, + 1891 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1323, + 1405, + 1323, + 1405, + 1571, + 297, + 1571 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1087, + 1406, + 1087, + 1406, + 1231, + 298, + 1231 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 805, + 1406, + 805, + 1406, + 994, + 296, + 994 + ], + "score": 0.97 + }, + { + "category_id": 6, + "poly": [ + 294, + 169, + 1408, + 169, + 1408, + 262, + 294, + 262 + ], + "score": 0.937 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 857, + 75, + 857, + 105, + 298, + 105 + ], + "score": 0.917 + }, + { + "category_id": 0, + "poly": [ + 302, + 1029, + 782, + 1029, + 782, + 1063, + 302, + 1063 + ], + "score": 0.904 + }, + { + "category_id": 0, + "poly": [ + 298, + 1266, + 900, + 1266, + 900, + 1300, + 298, + 1300 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2113, + 835, + 2113 + ], + "score": 0.858 + }, + { + "category_id": 13, + "poly": [ + 336, + 960, + 447, + 960, + 447, + 999, + 336, + 999 + ], + "score": 0.93, + "latex": "T _ { \\varphi _ { 1 } ^ { * } } ( x , a )" + }, + { + "category_id": 13, + "poly": [ + 513, + 1769, + 588, + 1769, + 588, + 1798, + 513, + 1798 + ], + "score": 0.86, + "latex": "3 4 . 8 \\%" + }, + { + "category_id": 13, + "poly": [ + 382, + 1768, + 456, + 1768, + 456, + 1798, + 382, + 1798 + ], + "score": 0.86, + "latex": "4 9 . 2 \\%" + }, + { + "category_id": 13, + "poly": [ + 297, + 1768, + 371, + 1768, + 371, + 1798, + 297, + 1798 + ], + "score": 0.85, + "latex": "5 6 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 579, + 1148, + 1052, + 1148, + 1052, + 1204, + 579, + 1204 + ], + "score": 0.8, + "latex": "\\begin{array} { r } { \\bar { \\mathfrak { z } } \\left[ \\operatorname* { m a x } _ { a \\in \\mathcal { A } } \\left( M ^ { * } ( x ^ { \\prime } | x , a ) - \\bar { M } ( x ^ { \\prime } | x , a ) \\right) ^ { 2 } \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 989, + 926, + 1206, + 926, + 1206, + 965, + 989, + 965 + ], + "score": 0.58, + "latex": "\\mathbb { E } _ { M _ { \\theta } } \\left[ T _ { \\varphi _ { 0 } ^ { * } } ( x , a , x ^ { \\prime } ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 980, + 928, + 1035, + 928, + 1035, + 962, + 980, + 962 + ], + "score": 0.54, + "latex": "\\mathbb { E } _ { M _ { \\theta } }" + }, + { + "category_id": 13, + "poly": [ + 406, + 1801, + 439, + 1801, + 439, + 1827, + 406, + 1827 + ], + "score": 0.32, + "latex": "2 \\mathbf { x }" + }, + { + "category_id": 13, + "poly": [ + 431, + 233, + 459, + 233, + 459, + 262, + 431, + 262 + ], + "score": 0.32, + "latex": "\\pm" + }, + { + "category_id": 5, + "poly": [ + 432, + 262, + 1259, + 262, + 1259, + 774, + 432, + 774 + ], + "score": 0.929, + "html": "
TASKHalfCheetah
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.378 ± 0.0032.287± 0.0051.411 ± 0.037
OFF-SL0.404 ± 0.0013.311 ± 0.0552.246 ± 0.016
TASKWalker2d
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.49 ± 0.001.514 ± 0.0020.968 ± 0.004
OFF-SL0.467 ± 0.0041.825 ± 0.0611.239 ± 0.004
TASKHopper
DATASETmedium (train)expert (test)medium-replay (test)
GALILEO0.037 ± 0.0020.322 ± 0.0360.408 ± 0.003
OFF-SL0.034 ± 0.0010.464 ± 0.0210.574 ± 0.008
" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 170.0, + 1406.0, + 170.0, + 1406.0, + 203.0, + 293.0, + 203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 199.0, + 1405.0, + 199.0, + 1405.0, + 236.0, + 293.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 231.0, + 430.0, + 231.0, + 430.0, + 265.0, + 295.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 231.0, + 966.0, + 231.0, + 966.0, + 265.0, + 460.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1028.0, + 786.0, + 1028.0, + 786.0, + 1066.0, + 294.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1266.0, + 902.0, + 1266.0, + 902.0, + 1303.0, + 295.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2125.0, + 830.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1586.0, + 1407.0, + 1586.0, + 1407.0, + 1618.0, + 295.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1612.0, + 1405.0, + 1612.0, + 1405.0, + 1651.0, + 294.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1643.0, + 1408.0, + 1643.0, + 1408.0, + 1682.0, + 294.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1673.0, + 1407.0, + 1673.0, + 1407.0, + 1715.0, + 292.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1707.0, + 1407.0, + 1707.0, + 1407.0, + 1743.0, + 295.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1738.0, + 1406.0, + 1738.0, + 1406.0, + 1773.0, + 294.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1763.0, + 296.0, + 1763.0, + 296.0, + 1806.0, + 292.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1763.0, + 381.0, + 1763.0, + 381.0, + 1806.0, + 372.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1763.0, + 512.0, + 1763.0, + 512.0, + 1806.0, + 457.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1763.0, + 1405.0, + 1763.0, + 1405.0, + 1806.0, + 589.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1797.0, + 405.0, + 1797.0, + 405.0, + 1832.0, + 292.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1797.0, + 1405.0, + 1797.0, + 1405.0, + 1832.0, + 440.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1827.0, + 1406.0, + 1827.0, + 1406.0, + 1867.0, + 292.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1860.0, + 757.0, + 1860.0, + 757.0, + 1891.0, + 295.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1325.0, + 1405.0, + 1325.0, + 1405.0, + 1359.0, + 295.0, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1355.0, + 1406.0, + 1355.0, + 1406.0, + 1388.0, + 292.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1385.0, + 1405.0, + 1385.0, + 1405.0, + 1419.0, + 295.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1416.0, + 1407.0, + 1416.0, + 1407.0, + 1449.0, + 295.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1449.0, + 1403.0, + 1449.0, + 1403.0, + 1480.0, + 296.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1475.0, + 1406.0, + 1475.0, + 1406.0, + 1513.0, + 292.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1507.0, + 1405.0, + 1507.0, + 1405.0, + 1541.0, + 292.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1538.0, + 443.0, + 1538.0, + 443.0, + 1576.0, + 292.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1089.0, + 1403.0, + 1089.0, + 1403.0, + 1121.0, + 296.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1118.0, + 1404.0, + 1118.0, + 1404.0, + 1154.0, + 295.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1154.0, + 578.0, + 1154.0, + 578.0, + 1199.0, + 294.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1154.0, + 1406.0, + 1154.0, + 1406.0, + 1199.0, + 1053.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1197.0, + 823.0, + 1197.0, + 823.0, + 1229.0, + 295.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 805.0, + 1406.0, + 805.0, + 1406.0, + 840.0, + 293.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 834.0, + 1406.0, + 834.0, + 1406.0, + 871.0, + 294.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 869.0, + 1404.0, + 869.0, + 1404.0, + 901.0, + 295.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 898.0, + 1404.0, + 898.0, + 1404.0, + 930.0, + 294.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 920.0, + 979.0, + 920.0, + 979.0, + 976.0, + 291.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 920.0, + 1412.0, + 920.0, + 1412.0, + 976.0, + 1207.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 959.0, + 335.0, + 959.0, + 335.0, + 999.0, + 294.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 959.0, + 1293.0, + 959.0, + 1293.0, + 999.0, + 448.0, + 999.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 31, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 316, + 514, + 1380, + 514, + 1380, + 1566, + 316, + 1566 + ], + "score": 0.977 + }, + { + "category_id": 4, + "poly": [ + 294, + 1598, + 1405, + 1598, + 1405, + 1753, + 294, + 1753 + ], + "score": 0.963 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 857, + 75, + 857, + 105, + 297, + 105 + ], + "score": 0.907 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2113, + 835, + 2113 + ], + "score": 0.849 + }, + { + "category_id": 15, + "poly": [ + 341.0, + 534.0, + 375.0, + 534.0, + 375.0, + 560.0, + 341.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 529.0, + 654.0, + 529.0, + 654.0, + 570.0, + 592.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 531.0, + 760.0, + 531.0, + 760.0, + 744.0, + 664.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 536.0, + 1008.0, + 536.0, + 1008.0, + 576.0, + 949.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 537.0, + 1088.0, + 537.0, + 1088.0, + 591.0, + 1053.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 529.0, + 1367.0, + 529.0, + 1367.0, + 570.0, + 1305.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 583.0, + 375.0, + 583.0, + 375.0, + 688.0, + 322.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 630.0, + 736.0, + 630.0, + 736.0, + 655.0, + 696.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 592.0, + 1091.0, + 592.0, + 1091.0, + 693.0, + 1015.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 684.0, + 375.0, + 684.0, + 375.0, + 711.0, + 341.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 678.0, + 738.0, + 678.0, + 738.0, + 703.0, + 698.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 676.0, + 1086.0, + 676.0, + 1086.0, + 700.0, + 1055.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 703.0, + 1088.0, + 703.0, + 1088.0, + 729.0, + 1053.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 734.0, + 375.0, + 734.0, + 375.0, + 760.0, + 340.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 724.0, + 738.0, + 724.0, + 738.0, + 751.0, + 696.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 738.0, + 909.0, + 738.0, + 909.0, + 753.0, + 853.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 741.0, + 1003.0, + 741.0, + 1003.0, + 757.0, + 956.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 730.0, + 1088.0, + 730.0, + 1088.0, + 757.0, + 1053.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 763.0, + 401.0, + 763.0, + 401.0, + 786.0, + 368.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 763.0, + 452.0, + 763.0, + 452.0, + 786.0, + 420.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 763.0, + 505.0, + 763.0, + 505.0, + 786.0, + 472.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 763.0, + 557.0, + 763.0, + 557.0, + 786.0, + 522.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 763.0, + 609.0, + 763.0, + 609.0, + 786.0, + 575.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 757.0, + 669.0, + 757.0, + 669.0, + 805.0, + 616.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 762.0, + 764.0, + 762.0, + 764.0, + 788.0, + 729.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 762.0, + 814.0, + 762.0, + 814.0, + 788.0, + 779.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 764.0, + 865.0, + 764.0, + 865.0, + 786.0, + 832.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 764.0, + 916.0, + 764.0, + 916.0, + 786.0, + 881.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 764.0, + 966.0, + 764.0, + 966.0, + 787.0, + 934.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 758.0, + 1023.0, + 758.0, + 1023.0, + 804.0, + 974.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 763.0, + 1114.0, + 763.0, + 1114.0, + 786.0, + 1080.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 763.0, + 1164.0, + 763.0, + 1164.0, + 786.0, + 1133.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 763.0, + 1217.0, + 763.0, + 1217.0, + 786.0, + 1184.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 763.0, + 1269.0, + 763.0, + 1269.0, + 786.0, + 1235.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1288.0, + 763.0, + 1321.0, + 763.0, + 1321.0, + 786.0, + 1288.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 757.0, + 1380.0, + 757.0, + 1380.0, + 805.0, + 1328.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 777.0, + 569.0, + 777.0, + 569.0, + 804.0, + 460.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 777.0, + 928.0, + 777.0, + 928.0, + 804.0, + 819.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 777.0, + 1282.0, + 777.0, + 1282.0, + 804.0, + 1173.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 810.0, + 653.0, + 810.0, + 653.0, + 842.0, + 334.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 810.0, + 987.0, + 810.0, + 987.0, + 843.0, + 713.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 810.0, + 1355.0, + 810.0, + 1355.0, + 842.0, + 1055.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 888.0, + 375.0, + 888.0, + 375.0, + 914.0, + 341.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 877.0, + 738.0, + 877.0, + 738.0, + 902.0, + 696.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 884.0, + 1009.0, + 884.0, + 1009.0, + 923.0, + 950.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 888.0, + 1087.0, + 888.0, + 1087.0, + 910.0, + 1054.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 877.0, + 1366.0, + 877.0, + 1366.0, + 918.0, + 1303.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 918.0, + 1088.0, + 918.0, + 1088.0, + 944.0, + 1053.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 937.0, + 374.0, + 937.0, + 374.0, + 1035.0, + 322.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 974.0, + 652.0, + 974.0, + 652.0, + 1012.0, + 592.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 940.0, + 756.0, + 940.0, + 756.0, + 1087.0, + 664.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 998.0, + 789.0, + 998.0, + 789.0, + 1011.0, + 773.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 936.0, + 1101.0, + 936.0, + 1101.0, + 1052.0, + 1023.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1009.0, + 1337.0, + 1009.0, + 1337.0, + 1021.0, + 1317.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1028.0, + 375.0, + 1028.0, + 375.0, + 1056.0, + 341.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1038.0, + 736.0, + 1038.0, + 736.0, + 1065.0, + 696.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1044.0, + 1087.0, + 1044.0, + 1087.0, + 1071.0, + 1053.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1026.0, + 1134.0, + 1026.0, + 1134.0, + 1042.0, + 1109.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 1020.0, + 1200.0, + 1020.0, + 1200.0, + 1040.0, + 1152.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 1018.0, + 1234.0, + 1018.0, + 1234.0, + 1032.0, + 1208.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1075.0, + 376.0, + 1075.0, + 376.0, + 1102.0, + 341.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1082.0, + 736.0, + 1082.0, + 736.0, + 1104.0, + 698.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1077.0, + 1087.0, + 1077.0, + 1087.0, + 1100.0, + 1054.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 1108.0, + 404.0, + 1108.0, + 404.0, + 1135.0, + 367.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1108.0, + 453.0, + 1108.0, + 453.0, + 1135.0, + 418.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 1111.0, + 506.0, + 1111.0, + 506.0, + 1132.0, + 472.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1110.0, + 557.0, + 1110.0, + 557.0, + 1132.0, + 523.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1108.0, + 610.0, + 1108.0, + 610.0, + 1135.0, + 574.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 1104.0, + 669.0, + 1104.0, + 669.0, + 1152.0, + 615.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1111.0, + 763.0, + 1111.0, + 763.0, + 1134.0, + 730.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1110.0, + 814.0, + 1110.0, + 814.0, + 1136.0, + 779.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1111.0, + 865.0, + 1111.0, + 865.0, + 1132.0, + 832.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 1111.0, + 916.0, + 1111.0, + 916.0, + 1134.0, + 882.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1111.0, + 966.0, + 1111.0, + 966.0, + 1134.0, + 934.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1106.0, + 1023.0, + 1106.0, + 1023.0, + 1152.0, + 974.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1108.0, + 1116.0, + 1108.0, + 1116.0, + 1135.0, + 1080.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1108.0, + 1166.0, + 1108.0, + 1166.0, + 1135.0, + 1132.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 1111.0, + 1218.0, + 1111.0, + 1218.0, + 1132.0, + 1184.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 1110.0, + 1270.0, + 1110.0, + 1270.0, + 1132.0, + 1235.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 1108.0, + 1322.0, + 1108.0, + 1322.0, + 1135.0, + 1287.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1106.0, + 1378.0, + 1106.0, + 1378.0, + 1149.0, + 1325.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 1126.0, + 568.0, + 1126.0, + 568.0, + 1149.0, + 461.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1126.0, + 926.0, + 1126.0, + 926.0, + 1150.0, + 821.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 1126.0, + 1280.0, + 1126.0, + 1280.0, + 1149.0, + 1174.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1153.0, + 1385.0, + 1153.0, + 1385.0, + 1194.0, + 307.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 1183.0, + 379.0, + 1183.0, + 379.0, + 1220.0, + 310.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1249.0, + 375.0, + 1249.0, + 375.0, + 1276.0, + 341.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 1254.0, + 654.0, + 1254.0, + 654.0, + 1293.0, + 442.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1261.0, + 736.0, + 1261.0, + 736.0, + 1286.0, + 696.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 1257.0, + 1009.0, + 1257.0, + 1009.0, + 1278.0, + 951.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1254.0, + 1086.0, + 1254.0, + 1086.0, + 1276.0, + 1054.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 1272.0, + 1003.0, + 1272.0, + 1003.0, + 1296.0, + 950.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1282.0, + 375.0, + 1282.0, + 375.0, + 1311.0, + 341.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1296.0, + 709.0, + 1296.0, + 709.0, + 1444.0, + 681.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1296.0, + 738.0, + 1296.0, + 738.0, + 1321.0, + 696.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1298.0, + 1087.0, + 1298.0, + 1087.0, + 1321.0, + 1055.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1317.0, + 375.0, + 1317.0, + 375.0, + 1410.0, + 322.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1346.0, + 652.0, + 1346.0, + 652.0, + 1367.0, + 592.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1330.0, + 736.0, + 1330.0, + 736.0, + 1356.0, + 698.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1313.0, + 1006.0, + 1313.0, + 1006.0, + 1347.0, + 854.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 1317.0, + 1089.0, + 1317.0, + 1089.0, + 1421.0, + 1015.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 1346.0, + 1366.0, + 1346.0, + 1366.0, + 1367.0, + 1305.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1363.0, + 646.0, + 1363.0, + 646.0, + 1387.0, + 592.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1364.0, + 736.0, + 1364.0, + 736.0, + 1389.0, + 696.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 1369.0, + 791.0, + 1369.0, + 791.0, + 1377.0, + 778.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 1363.0, + 1358.0, + 1363.0, + 1358.0, + 1387.0, + 1305.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1383.0, + 375.0, + 1383.0, + 375.0, + 1411.0, + 341.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1399.0, + 736.0, + 1399.0, + 736.0, + 1424.0, + 698.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1417.0, + 375.0, + 1417.0, + 375.0, + 1444.0, + 341.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1434.0, + 738.0, + 1434.0, + 738.0, + 1459.0, + 696.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1431.0, + 1087.0, + 1431.0, + 1087.0, + 1454.0, + 1055.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1450.0, + 375.0, + 1450.0, + 375.0, + 1478.0, + 341.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1467.0, + 738.0, + 1467.0, + 738.0, + 1494.0, + 698.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1484.0, + 402.0, + 1484.0, + 402.0, + 1507.0, + 368.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 1484.0, + 452.0, + 1484.0, + 452.0, + 1507.0, + 420.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 1484.0, + 505.0, + 1484.0, + 505.0, + 1506.0, + 471.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1484.0, + 557.0, + 1484.0, + 557.0, + 1506.0, + 523.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1484.0, + 609.0, + 1484.0, + 609.0, + 1507.0, + 575.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 1479.0, + 666.0, + 1479.0, + 666.0, + 1525.0, + 618.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1484.0, + 762.0, + 1484.0, + 762.0, + 1507.0, + 730.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1484.0, + 813.0, + 1484.0, + 813.0, + 1507.0, + 781.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1485.0, + 864.0, + 1485.0, + 864.0, + 1507.0, + 832.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 1485.0, + 915.0, + 1485.0, + 915.0, + 1507.0, + 882.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 1483.0, + 967.0, + 1483.0, + 967.0, + 1509.0, + 933.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1485.0, + 1017.0, + 1485.0, + 1017.0, + 1507.0, + 984.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 1484.0, + 1114.0, + 1484.0, + 1114.0, + 1507.0, + 1081.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1484.0, + 1164.0, + 1484.0, + 1164.0, + 1507.0, + 1133.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 1484.0, + 1218.0, + 1484.0, + 1218.0, + 1506.0, + 1184.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1484.0, + 1269.0, + 1484.0, + 1269.0, + 1507.0, + 1236.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1288.0, + 1484.0, + 1321.0, + 1484.0, + 1321.0, + 1507.0, + 1288.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 1478.0, + 1380.0, + 1478.0, + 1380.0, + 1526.0, + 1328.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1498.0, + 569.0, + 1498.0, + 569.0, + 1525.0, + 460.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1500.0, + 926.0, + 1500.0, + 926.0, + 1524.0, + 821.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1496.0, + 1019.0, + 1496.0, + 1019.0, + 1521.0, + 978.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1498.0, + 1281.0, + 1498.0, + 1281.0, + 1525.0, + 1173.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1531.0, + 639.0, + 1531.0, + 639.0, + 1564.0, + 348.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1531.0, + 972.0, + 1531.0, + 972.0, + 1566.0, + 728.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 1530.0, + 1338.0, + 1530.0, + 1338.0, + 1564.0, + 1076.0, + 1564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 716.5, + 824.0, + 716.5, + 824.0, + 744.5, + 756.0, + 744.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.75, + 919.5, + 1008.75, + 919.5, + 1008.75, + 967.5, + 867.75, + 967.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 915.0, + 756.0, + 915.0, + 756.0, + 962.5, + 646.0, + 962.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.25, + 965.5, + 869.25, + 965.5, + 869.25, + 976.5, + 849.25, + 976.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 975.5, + 865.0, + 975.5, + 865.0, + 992.0, + 819.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.75, + 995.5, + 837.75, + 995.5, + 837.75, + 1022.5, + 783.75, + 1022.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1009.0, + 1306.0, + 1009.0, + 1306.0, + 1029.0, + 1238.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 1292.5, + 1348.0, + 1292.5, + 1348.0, + 1306.0, + 1330.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.25, + 1336.0, + 873.25, + 1336.0, + 873.25, + 1359.0, + 827.25, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.25, + 1321.0, + 1311.25, + 1321.0, + 1311.25, + 1390.0, + 1060.25, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1436.0, + 1337.0, + 1436.0, + 1337.0, + 1456.5, + 1269.0, + 1456.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1600.0, + 1403.0, + 1600.0, + 1403.0, + 1633.0, + 296.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1630.0, + 1403.0, + 1630.0, + 1403.0, + 1664.0, + 294.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1657.0, + 1406.0, + 1657.0, + 1406.0, + 1697.0, + 293.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1690.0, + 1405.0, + 1690.0, + 1405.0, + 1723.0, + 296.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1723.0, + 622.0, + 1723.0, + 622.0, + 1753.0, + 297.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 32, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 323, + 1018, + 1375, + 1018, + 1375, + 1634, + 323, + 1634 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1802, + 1404, + 1802, + 1404, + 1957, + 298, + 1957 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 776, + 1404, + 776, + 1404, + 900, + 297, + 900 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 628, + 228, + 1065, + 228, + 1065, + 556, + 628, + 556 + ], + "score": 0.966 + }, + { + "category_id": 4, + "poly": [ + 297, + 1668, + 1402, + 1668, + 1402, + 1762, + 297, + 1762 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 297, + 1971, + 1400, + 1971, + 1400, + 2034, + 297, + 2034 + ], + "score": 0.942 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 857, + 75, + 857, + 105, + 297, + 105 + ], + "score": 0.908 + }, + { + "category_id": 4, + "poly": [ + 296, + 593, + 1402, + 593, + 1402, + 657, + 296, + 657 + ], + "score": 0.88 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2112, + 835, + 2112 + ], + "score": 0.858 + }, + { + "category_id": 0, + "poly": [ + 301, + 716, + 830, + 716, + 830, + 748, + 301, + 748 + ], + "score": 0.851 + }, + { + "category_id": 1, + "poly": [ + 299, + 915, + 1401, + 915, + 1401, + 978, + 299, + 978 + ], + "score": 0.84 + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1023.0, + 379.0, + 1023.0, + 379.0, + 1236.0, + 322.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1030.0, + 469.0, + 1030.0, + 469.0, + 1039.0, + 448.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1001.0, + 755.0, + 1001.0, + 755.0, + 1255.0, + 584.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 1027.0, + 1011.0, + 1027.0, + 1011.0, + 1047.0, + 941.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 1001.0, + 1114.0, + 1001.0, + 1114.0, + 1254.0, + 1015.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 1026.0, + 1369.0, + 1026.0, + 1369.0, + 1049.0, + 1298.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1050.0, + 549.0, + 1050.0, + 549.0, + 1059.0, + 536.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1042.0, + 626.0, + 1042.0, + 626.0, + 1061.0, + 581.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1049.0, + 735.0, + 1049.0, + 735.0, + 1074.0, + 695.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 1040.0, + 983.0, + 1040.0, + 983.0, + 1063.0, + 933.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 1042.0, + 1337.0, + 1042.0, + 1337.0, + 1061.0, + 1294.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1060.0, + 379.0, + 1060.0, + 379.0, + 1087.0, + 339.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1069.0, + 971.0, + 1069.0, + 971.0, + 1079.0, + 957.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1054.0, + 1091.0, + 1054.0, + 1091.0, + 1079.0, + 1053.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1076.0, + 599.0, + 1076.0, + 599.0, + 1086.0, + 589.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1074.0, + 735.0, + 1074.0, + 735.0, + 1101.0, + 695.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 1074.0, + 980.0, + 1074.0, + 980.0, + 1083.0, + 970.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1079.0, + 994.0, + 1079.0, + 994.0, + 1088.0, + 981.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1311.0, + 1078.0, + 1327.0, + 1078.0, + 1327.0, + 1090.0, + 1311.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1083.0, + 1091.0, + 1083.0, + 1091.0, + 1109.0, + 1053.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1100.0, + 379.0, + 1100.0, + 379.0, + 1125.0, + 339.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 1120.0, + 511.0, + 1120.0, + 511.0, + 1129.0, + 500.0, + 1129.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1101.0, + 548.0, + 1101.0, + 548.0, + 1110.0, + 536.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1101.0, + 735.0, + 1101.0, + 735.0, + 1127.0, + 695.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1128.0, + 735.0, + 1128.0, + 735.0, + 1154.0, + 695.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 1131.0, + 943.0, + 1131.0, + 943.0, + 1139.0, + 931.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1138.0, + 379.0, + 1138.0, + 379.0, + 1163.0, + 340.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 1147.0, + 880.0, + 1147.0, + 880.0, + 1155.0, + 868.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1154.0, + 734.0, + 1154.0, + 734.0, + 1180.0, + 697.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1176.0, + 379.0, + 1176.0, + 379.0, + 1202.0, + 340.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1181.0, + 735.0, + 1181.0, + 735.0, + 1207.0, + 695.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1176.0, + 1091.0, + 1176.0, + 1091.0, + 1201.0, + 1052.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1214.0, + 379.0, + 1214.0, + 379.0, + 1239.0, + 340.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1221.0, + 1106.0, + 1221.0, + 1106.0, + 1231.0, + 1096.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1233.0, + 404.0, + 1233.0, + 404.0, + 1255.0, + 374.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1234.0, + 455.0, + 1234.0, + 455.0, + 1255.0, + 424.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1234.0, + 507.0, + 1234.0, + 507.0, + 1255.0, + 476.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1234.0, + 557.0, + 1234.0, + 557.0, + 1255.0, + 526.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1234.0, + 609.0, + 1234.0, + 609.0, + 1255.0, + 577.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 1234.0, + 659.0, + 1234.0, + 659.0, + 1255.0, + 627.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1234.0, + 761.0, + 1234.0, + 761.0, + 1255.0, + 730.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1233.0, + 812.0, + 1233.0, + 812.0, + 1255.0, + 781.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1234.0, + 862.0, + 1234.0, + 862.0, + 1255.0, + 831.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1234.0, + 915.0, + 1234.0, + 915.0, + 1255.0, + 883.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1234.0, + 965.0, + 1234.0, + 965.0, + 1255.0, + 934.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1234.0, + 1016.0, + 1234.0, + 1016.0, + 1255.0, + 983.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1234.0, + 1118.0, + 1234.0, + 1118.0, + 1255.0, + 1087.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1233.0, + 1168.0, + 1233.0, + 1168.0, + 1255.0, + 1137.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 1234.0, + 1220.0, + 1234.0, + 1220.0, + 1255.0, + 1189.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1234.0, + 1270.0, + 1234.0, + 1270.0, + 1255.0, + 1239.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 1233.0, + 1322.0, + 1233.0, + 1322.0, + 1255.0, + 1289.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1234.0, + 1371.0, + 1234.0, + 1371.0, + 1255.0, + 1342.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1248.0, + 651.0, + 1248.0, + 651.0, + 1274.0, + 382.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1248.0, + 1006.0, + 1248.0, + 1006.0, + 1274.0, + 739.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1248.0, + 1363.0, + 1248.0, + 1363.0, + 1274.0, + 1095.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1279.0, + 554.0, + 1279.0, + 554.0, + 1318.0, + 433.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 1280.0, + 909.0, + 1280.0, + 909.0, + 1315.0, + 791.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1279.0, + 1267.0, + 1279.0, + 1267.0, + 1318.0, + 1147.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1342.0, + 379.0, + 1342.0, + 379.0, + 1554.0, + 322.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1347.0, + 457.0, + 1347.0, + 457.0, + 1367.0, + 385.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 1350.0, + 629.0, + 1350.0, + 629.0, + 1361.0, + 593.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 1320.0, + 758.0, + 1320.0, + 758.0, + 1574.0, + 658.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 1321.0, + 1114.0, + 1321.0, + 1114.0, + 1576.0, + 1015.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1350.0, + 1148.0, + 1350.0, + 1148.0, + 1361.0, + 1119.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 1345.0, + 1370.0, + 1345.0, + 1370.0, + 1368.0, + 1298.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1368.0, + 380.0, + 1368.0, + 380.0, + 1394.0, + 339.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1361.0, + 428.0, + 1361.0, + 428.0, + 1381.0, + 405.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1370.0, + 1091.0, + 1370.0, + 1091.0, + 1394.0, + 1053.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 1360.0, + 1197.0, + 1360.0, + 1197.0, + 1368.0, + 1186.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 1360.0, + 1339.0, + 1360.0, + 1339.0, + 1383.0, + 1298.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1395.0, + 379.0, + 1395.0, + 379.0, + 1421.0, + 339.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1399.0, + 956.0, + 1399.0, + 956.0, + 1408.0, + 944.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1397.0, + 1091.0, + 1397.0, + 1091.0, + 1421.0, + 1052.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1421.0, + 380.0, + 1421.0, + 380.0, + 1447.0, + 339.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 1431.0, + 400.0, + 1431.0, + 400.0, + 1439.0, + 386.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 1420.0, + 433.0, + 1420.0, + 433.0, + 1430.0, + 421.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 1418.0, + 981.0, + 1418.0, + 981.0, + 1428.0, + 969.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1450.0, + 379.0, + 1450.0, + 379.0, + 1474.0, + 339.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1451.0, + 1091.0, + 1451.0, + 1091.0, + 1476.0, + 1052.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1476.0, + 379.0, + 1476.0, + 379.0, + 1501.0, + 340.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1470.0, + 734.0, + 1470.0, + 734.0, + 1496.0, + 695.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1487.0, + 787.0, + 1487.0, + 787.0, + 1493.0, + 781.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1478.0, + 1091.0, + 1478.0, + 1091.0, + 1503.0, + 1053.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 1481.0, + 1183.0, + 1481.0, + 1183.0, + 1491.0, + 1173.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1501.0, + 380.0, + 1501.0, + 380.0, + 1527.0, + 339.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1504.0, + 1091.0, + 1504.0, + 1091.0, + 1531.0, + 1052.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 1505.0, + 1360.0, + 1505.0, + 1360.0, + 1514.0, + 1352.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1523.0, + 382.0, + 1523.0, + 382.0, + 1556.0, + 337.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1510.0, + 736.0, + 1510.0, + 736.0, + 1540.0, + 697.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 1518.0, + 1011.0, + 1518.0, + 1011.0, + 1553.0, + 932.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1529.0, + 1092.0, + 1529.0, + 1092.0, + 1559.0, + 1054.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1542.0, + 1106.0, + 1542.0, + 1106.0, + 1550.0, + 1096.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1520.0, + 1132.0, + 1520.0, + 1132.0, + 1531.0, + 1121.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1553.0, + 405.0, + 1553.0, + 405.0, + 1574.0, + 374.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1553.0, + 457.0, + 1553.0, + 457.0, + 1574.0, + 424.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1553.0, + 507.0, + 1553.0, + 507.0, + 1574.0, + 476.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1553.0, + 557.0, + 1553.0, + 557.0, + 1574.0, + 526.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1553.0, + 609.0, + 1553.0, + 609.0, + 1574.0, + 577.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1553.0, + 659.0, + 1553.0, + 659.0, + 1574.0, + 628.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 1553.0, + 761.0, + 1553.0, + 761.0, + 1574.0, + 730.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1553.0, + 812.0, + 1553.0, + 812.0, + 1574.0, + 781.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 1553.0, + 863.0, + 1553.0, + 863.0, + 1574.0, + 831.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1553.0, + 914.0, + 1553.0, + 914.0, + 1574.0, + 883.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1552.0, + 965.0, + 1552.0, + 965.0, + 1574.0, + 934.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 1551.0, + 1017.0, + 1551.0, + 1017.0, + 1577.0, + 982.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1553.0, + 1117.0, + 1553.0, + 1117.0, + 1574.0, + 1087.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1553.0, + 1168.0, + 1553.0, + 1168.0, + 1574.0, + 1137.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 1553.0, + 1220.0, + 1553.0, + 1220.0, + 1574.0, + 1189.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1553.0, + 1270.0, + 1553.0, + 1270.0, + 1574.0, + 1239.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1289.0, + 1553.0, + 1322.0, + 1553.0, + 1322.0, + 1574.0, + 1289.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1553.0, + 1371.0, + 1553.0, + 1371.0, + 1574.0, + 1342.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1567.0, + 651.0, + 1567.0, + 651.0, + 1593.0, + 382.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1567.0, + 1006.0, + 1567.0, + 1006.0, + 1593.0, + 739.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1567.0, + 1363.0, + 1567.0, + 1363.0, + 1593.0, + 1095.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1598.0, + 555.0, + 1598.0, + 555.0, + 1637.0, + 433.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 1598.0, + 910.0, + 1598.0, + 910.0, + 1636.0, + 789.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1598.0, + 1265.0, + 1598.0, + 1265.0, + 1636.0, + 1147.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.25, + 1030.5, + 1134.25, + 1030.5, + 1134.25, + 1046.0, + 1093.25, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.25, + 1087.5, + 1361.25, + 1087.5, + 1361.25, + 1108.5, + 1325.25, + 1108.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1121.5, + 1003.0, + 1121.5, + 1003.0, + 1136.5, + 938.0, + 1136.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.75, + 1195.5, + 792.75, + 1195.5, + 792.75, + 1203.5, + 774.75, + 1203.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 1382.5, + 512.0, + 1382.5, + 512.0, + 1392.0, + 497.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.25, + 1409.0, + 964.25, + 1409.0, + 964.25, + 1414.5, + 960.25, + 1414.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1428.0, + 991.0, + 1428.0, + 991.0, + 1438.0, + 983.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 1473.0, + 1340.0, + 1473.0, + 1340.0, + 1488.0, + 1323.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 1487.5, + 1351.0, + 1487.5, + 1351.0, + 1500.0, + 1336.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 225.0, + 658.0, + 225.0, + 658.0, + 524.0, + 629.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 236.0, + 681.0, + 236.0, + 681.0, + 255.0, + 653.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 247.0, + 774.0, + 247.0, + 774.0, + 255.0, + 767.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 263.0, + 750.0, + 263.0, + 750.0, + 273.0, + 742.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 284.0, + 728.0, + 284.0, + 728.0, + 291.0, + 718.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 283.0, + 804.0, + 283.0, + 804.0, + 291.0, + 795.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 279.0, + 926.0, + 279.0, + 926.0, + 288.0, + 918.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 289.0, + 683.0, + 289.0, + 683.0, + 312.0, + 652.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 291.0, + 766.0, + 291.0, + 766.0, + 301.0, + 757.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 288.0, + 792.0, + 288.0, + 792.0, + 298.0, + 783.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 291.0, + 809.0, + 291.0, + 809.0, + 299.0, + 801.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 297.0, + 773.0, + 297.0, + 773.0, + 306.0, + 766.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 297.0, + 796.0, + 297.0, + 796.0, + 308.0, + 787.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 304.0, + 811.0, + 304.0, + 811.0, + 312.0, + 802.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 295.0, + 831.0, + 295.0, + 831.0, + 309.0, + 820.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 304.0, + 864.0, + 304.0, + 864.0, + 312.0, + 856.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 302.0, + 888.0, + 302.0, + 888.0, + 311.0, + 881.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 302.0, + 928.0, + 302.0, + 928.0, + 317.0, + 918.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 302.0, + 947.0, + 302.0, + 947.0, + 315.0, + 935.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 304.0, + 1020.0, + 304.0, + 1020.0, + 312.0, + 1012.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 312.0, + 722.0, + 312.0, + 722.0, + 321.0, + 714.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 309.0, + 845.0, + 309.0, + 845.0, + 317.0, + 837.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 320.0, + 763.0, + 320.0, + 763.0, + 327.0, + 755.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 321.0, + 801.0, + 321.0, + 801.0, + 333.0, + 789.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 322.0, + 833.0, + 322.0, + 833.0, + 331.0, + 823.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 322.0, + 956.0, + 322.0, + 956.0, + 334.0, + 944.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 333.0, + 762.0, + 333.0, + 762.0, + 342.0, + 753.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 338.0, + 780.0, + 338.0, + 780.0, + 348.0, + 772.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 335.0, + 827.0, + 335.0, + 827.0, + 348.0, + 814.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 338.0, + 865.0, + 338.0, + 865.0, + 348.0, + 856.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 332.0, + 899.0, + 332.0, + 899.0, + 345.0, + 887.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 331.0, + 985.0, + 331.0, + 985.0, + 343.0, + 965.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 338.0, + 998.0, + 338.0, + 998.0, + 348.0, + 989.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 344.0, + 683.0, + 344.0, + 683.0, + 366.0, + 652.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 345.0, + 1013.0, + 345.0, + 1013.0, + 357.0, + 1001.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 353.0, + 713.0, + 353.0, + 713.0, + 366.0, + 700.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 352.0, + 792.0, + 352.0, + 792.0, + 366.0, + 763.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 355.0, + 837.0, + 355.0, + 837.0, + 364.0, + 827.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 362.0, + 849.0, + 362.0, + 849.0, + 372.0, + 841.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 361.0, + 909.0, + 361.0, + 909.0, + 371.0, + 900.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 358.0, + 935.0, + 358.0, + 935.0, + 366.0, + 926.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 352.0, + 973.0, + 352.0, + 973.0, + 359.0, + 963.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 354.0, + 1000.0, + 354.0, + 1000.0, + 365.0, + 985.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 366.0, + 754.0, + 366.0, + 754.0, + 377.0, + 745.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 371.0, + 919.0, + 371.0, + 919.0, + 387.0, + 908.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 379.0, + 1046.0, + 379.0, + 1046.0, + 386.0, + 1038.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 866.0, + 391.0, + 878.0, + 391.0, + 878.0, + 403.0, + 866.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 398.0, + 682.0, + 398.0, + 682.0, + 421.0, + 651.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 412.0, + 821.0, + 412.0, + 821.0, + 422.0, + 812.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 412.0, + 864.0, + 412.0, + 864.0, + 421.0, + 854.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 411.0, + 891.0, + 411.0, + 891.0, + 422.0, + 869.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 406.0, + 919.0, + 406.0, + 919.0, + 419.0, + 907.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 405.0, + 958.0, + 405.0, + 958.0, + 420.0, + 947.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 413.0, + 977.0, + 413.0, + 977.0, + 420.0, + 963.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 436.0, + 1030.0, + 436.0, + 1030.0, + 446.0, + 1022.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 440.0, + 1006.0, + 440.0, + 1006.0, + 450.0, + 999.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 453.0, + 683.0, + 453.0, + 683.0, + 475.0, + 652.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 467.0, + 769.0, + 467.0, + 769.0, + 479.0, + 757.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 467.0, + 809.0, + 467.0, + 809.0, + 479.0, + 797.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 468.0, + 843.0, + 468.0, + 843.0, + 479.0, + 820.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 467.0, + 880.0, + 467.0, + 880.0, + 479.0, + 867.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 466.0, + 953.0, + 466.0, + 953.0, + 480.0, + 923.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 468.0, + 1000.0, + 468.0, + 1000.0, + 478.0, + 991.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 469.0, + 1020.0, + 469.0, + 1020.0, + 477.0, + 1010.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 469.0, + 1039.0, + 469.0, + 1039.0, + 477.0, + 1030.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 451.0, + 1049.0, + 451.0, + 1049.0, + 461.0, + 1040.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 484.0, + 826.0, + 484.0, + 826.0, + 496.0, + 814.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 496.0, + 770.0, + 496.0, + 770.0, + 508.0, + 759.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 513.0, + 717.0, + 513.0, + 717.0, + 537.0, + 685.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 513.0, + 786.0, + 513.0, + 786.0, + 537.0, + 755.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 513.0, + 855.0, + 513.0, + 855.0, + 537.0, + 824.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 513.0, + 924.0, + 513.0, + 924.0, + 537.0, + 892.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 514.0, + 994.0, + 514.0, + 994.0, + 537.0, + 961.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 513.0, + 1063.0, + 513.0, + 1063.0, + 537.0, + 1031.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 533.0, + 959.0, + 533.0, + 959.0, + 556.0, + 789.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 270.5, + 888.0, + 270.5, + 888.0, + 284.5, + 865.0, + 284.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 281.0, + 895.0, + 281.0, + 895.0, + 294.5, + 880.0, + 294.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 307.0, + 802.0, + 307.0, + 802.0, + 319.0, + 797.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 322.5, + 931.0, + 322.5, + 931.0, + 330.5, + 924.0, + 330.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.75, + 428.5, + 992.75, + 428.5, + 992.75, + 438.5, + 974.75, + 438.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1666.0, + 1406.0, + 1666.0, + 1406.0, + 1705.0, + 293.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1700.0, + 1402.0, + 1700.0, + 1402.0, + 1734.0, + 295.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1727.0, + 437.0, + 1727.0, + 437.0, + 1766.0, + 294.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 592.0, + 1404.0, + 592.0, + 1404.0, + 628.0, + 295.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 625.0, + 725.0, + 625.0, + 725.0, + 657.0, + 296.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 872.0, + 2083.0, + 872.0, + 2123.0, + 829.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 716.0, + 835.0, + 716.0, + 835.0, + 752.0, + 295.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1802.0, + 1408.0, + 1802.0, + 1408.0, + 1840.0, + 294.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1833.0, + 1404.0, + 1833.0, + 1404.0, + 1872.0, + 293.0, + 1872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1865.0, + 1405.0, + 1865.0, + 1405.0, + 1898.0, + 296.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1893.0, + 1407.0, + 1893.0, + 1407.0, + 1929.0, + 293.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1925.0, + 923.0, + 1925.0, + 923.0, + 1961.0, + 295.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 775.0, + 1405.0, + 775.0, + 1405.0, + 812.0, + 292.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 806.0, + 1406.0, + 806.0, + 1406.0, + 843.0, + 295.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 839.0, + 1404.0, + 839.0, + 1404.0, + 872.0, + 295.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 870.0, + 1273.0, + 870.0, + 1273.0, + 902.0, + 295.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1973.0, + 1404.0, + 1973.0, + 1404.0, + 2005.0, + 298.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 1405.0, + 2001.0, + 1405.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 913.0, + 1406.0, + 913.0, + 1406.0, + 949.0, + 295.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 944.0, + 393.0, + 944.0, + 393.0, + 978.0, + 294.0, + 978.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 33, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 907, + 1406, + 907, + 1406, + 1245, + 297, + 1245 + ], + "score": 0.984 + }, + { + "category_id": 5, + "poly": [ + 521, + 1496, + 1169, + 1496, + 1169, + 1685, + 521, + 1685 + ], + "score": 0.98, + "html": "
targetCity-ACity-BCity-C
supply improvement+1.63pp+0.79pp+0.27pp
targetCity-DCity-ECity-F
supply improvement+0.2pp+0.14pp+0.41pp
" + }, + { + "category_id": 1, + "poly": [ + 297, + 707, + 1406, + 707, + 1406, + 893, + 297, + 893 + ], + "score": 0.978 + }, + { + "category_id": 3, + "poly": [ + 628, + 229, + 1068, + 229, + 1068, + 582, + 628, + 582 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 296, + 1714, + 1405, + 1714, + 1405, + 1873, + 296, + 1873 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 662, + 1259, + 1039, + 1259, + 1039, + 1334, + 662, + 1334 + ], + "score": 0.955 + }, + { + "category_id": 4, + "poly": [ + 560, + 619, + 1137, + 619, + 1137, + 652, + 560, + 652 + ], + "score": 0.934 + }, + { + "category_id": 6, + "poly": [ + 298, + 1431, + 1397, + 1431, + 1397, + 1494, + 298, + 1494 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 297, + 1345, + 1144, + 1345, + 1144, + 1379, + 297, + 1379 + ], + "score": 0.93 + }, + { + "category_id": 2, + "poly": [ + 297, + 76, + 857, + 76, + 857, + 104, + 297, + 104 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2113, + 835, + 2113 + ], + "score": 0.865 + }, + { + "category_id": 14, + "poly": [ + 659, + 1257, + 1039, + 1257, + 1039, + 1334, + 659, + 1334 + ], + "score": 0.94, + "latex": "\\frac { \\sum _ { i } ^ { T } ( y _ { i } ^ { t } - b ^ { t } ) - ( y _ { i } ^ { c } - b ^ { c } ) } { T } \\times 1 0 0 ," + }, + { + "category_id": 13, + "poly": [ + 650, + 1181, + 677, + 1181, + 677, + 1216, + 650, + 1216 + ], + "score": 0.9, + "latex": "y _ { i } ^ { t }" + }, + { + "category_id": 13, + "poly": [ + 835, + 1347, + 924, + 1347, + 924, + 1375, + 835, + 1375 + ], + "score": 0.9, + "latex": "T = 1 4" + }, + { + "category_id": 13, + "poly": [ + 725, + 1183, + 754, + 1183, + 754, + 1216, + 725, + 1216 + ], + "score": 0.89, + "latex": "y _ { i } ^ { c }" + }, + { + "category_id": 13, + "poly": [ + 865, + 1122, + 893, + 1122, + 893, + 1150, + 865, + 1150 + ], + "score": 0.86, + "latex": "b ^ { c }" + }, + { + "category_id": 13, + "poly": [ + 787, + 1121, + 812, + 1121, + 812, + 1149, + 787, + 1149 + ], + "score": 0.85, + "latex": "b ^ { t }" + }, + { + "category_id": 13, + "poly": [ + 371, + 1348, + 394, + 1348, + 394, + 1374, + 371, + 1374 + ], + "score": 0.81, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 969, + 1185, + 982, + 1185, + 982, + 1210, + 969, + 1210 + ], + "score": 0.75, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 1111, + 1185, + 1123, + 1185, + 1123, + 1210, + 1111, + 1210 + ], + "score": 0.74, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 809, + 1643, + 893, + 1643, + 893, + 1676, + 809, + 1676 + ], + "score": 0.58, + "latex": "+ 0 . 2 \\mathrm { p p }" + }, + { + "category_id": 13, + "poly": [ + 738, + 1434, + 763, + 1434, + 763, + 1461, + 738, + 1461 + ], + "score": 0.35, + "latex": "\\mathbf { X }" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 227.0, + 657.0, + 227.0, + 657.0, + 565.0, + 629.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 262.0, + 694.0, + 262.0, + 694.0, + 284.0, + 665.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 267.0, + 788.0, + 267.0, + 788.0, + 284.0, + 738.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 278.0, + 800.0, + 278.0, + 800.0, + 304.0, + 734.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 298.0, + 763.0, + 298.0, + 763.0, + 322.0, + 736.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 316.0, + 696.0, + 316.0, + 696.0, + 342.0, + 662.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 371.0, + 696.0, + 371.0, + 696.0, + 398.0, + 662.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 426.0, + 697.0, + 426.0, + 697.0, + 456.0, + 656.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 485.0, + 696.0, + 485.0, + 696.0, + 509.0, + 657.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 532.0, + 732.0, + 532.0, + 732.0, + 557.0, + 695.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 532.0, + 799.0, + 532.0, + 799.0, + 558.0, + 762.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 532.0, + 868.0, + 532.0, + 868.0, + 558.0, + 827.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 530.0, + 940.0, + 530.0, + 940.0, + 558.0, + 894.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 533.0, + 1001.0, + 533.0, + 1001.0, + 558.0, + 955.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 532.0, + 1065.0, + 532.0, + 1065.0, + 558.0, + 1030.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 551.0, + 1063.0, + 551.0, + 1063.0, + 585.0, + 695.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 617.0, + 1139.0, + 617.0, + 1139.0, + 654.0, + 559.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1429.0, + 737.0, + 1429.0, + 737.0, + 1468.0, + 293.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1429.0, + 1403.0, + 1429.0, + 1403.0, + 1468.0, + 764.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1463.0, + 990.0, + 1463.0, + 990.0, + 1498.0, + 295.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 871.0, + 2083.0, + 871.0, + 2123.0, + 829.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 908.0, + 1404.0, + 908.0, + 1404.0, + 943.0, + 295.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 937.0, + 1404.0, + 937.0, + 1404.0, + 974.0, + 294.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 968.0, + 1406.0, + 968.0, + 1406.0, + 1003.0, + 294.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 998.0, + 1406.0, + 998.0, + 1406.0, + 1035.0, + 294.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1030.0, + 1406.0, + 1030.0, + 1406.0, + 1062.0, + 294.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1061.0, + 1406.0, + 1061.0, + 1406.0, + 1096.0, + 294.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1091.0, + 1406.0, + 1091.0, + 1406.0, + 1126.0, + 294.0, + 1126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1121.0, + 786.0, + 1121.0, + 786.0, + 1156.0, + 295.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1121.0, + 864.0, + 1121.0, + 864.0, + 1156.0, + 813.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1121.0, + 1404.0, + 1121.0, + 1404.0, + 1156.0, + 894.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1149.0, + 1404.0, + 1149.0, + 1404.0, + 1187.0, + 294.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1182.0, + 649.0, + 1182.0, + 649.0, + 1220.0, + 294.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 1182.0, + 724.0, + 1182.0, + 724.0, + 1220.0, + 678.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1182.0, + 968.0, + 1182.0, + 968.0, + 1220.0, + 755.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1182.0, + 1110.0, + 1182.0, + 1110.0, + 1220.0, + 983.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1182.0, + 1409.0, + 1182.0, + 1409.0, + 1220.0, + 1124.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1212.0, + 1404.0, + 1212.0, + 1404.0, + 1250.0, + 294.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 707.0, + 1406.0, + 707.0, + 1406.0, + 743.0, + 295.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 736.0, + 1407.0, + 736.0, + 1407.0, + 778.0, + 292.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 764.0, + 1407.0, + 764.0, + 1407.0, + 808.0, + 292.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 799.0, + 1408.0, + 799.0, + 1408.0, + 836.0, + 294.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 829.0, + 1407.0, + 829.0, + 1407.0, + 865.0, + 294.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 861.0, + 547.0, + 861.0, + 547.0, + 898.0, + 295.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1717.0, + 1403.0, + 1717.0, + 1403.0, + 1750.0, + 295.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1745.0, + 1406.0, + 1745.0, + 1406.0, + 1783.0, + 294.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1776.0, + 1405.0, + 1776.0, + 1405.0, + 1814.0, + 294.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1806.0, + 1407.0, + 1806.0, + 1407.0, + 1845.0, + 291.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1840.0, + 490.0, + 1840.0, + 490.0, + 1873.0, + 292.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1342.0, + 370.0, + 1342.0, + 370.0, + 1383.0, + 294.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1342.0, + 834.0, + 1342.0, + 834.0, + 1383.0, + 395.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1342.0, + 1148.0, + 1342.0, + 1148.0, + 1383.0, + 925.0, + 1383.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 34, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 325, + 774, + 1373, + 774, + 1373, + 1427, + 325, + 1427 + ], + "score": 0.977 + }, + { + "category_id": 4, + "poly": [ + 435, + 1461, + 1262, + 1461, + 1262, + 1495, + 435, + 1495 + ], + "score": 0.915 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 858, + 75, + 858, + 106, + 297, + 106 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2113, + 835, + 2113 + ], + "score": 0.842 + }, + { + "category_id": 15, + "poly": [ + 323.0, + 773.0, + 349.0, + 773.0, + 349.0, + 1033.0, + 323.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 773.0, + 705.0, + 773.0, + 705.0, + 1033.0, + 679.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 777.0, + 1064.0, + 777.0, + 1064.0, + 1036.0, + 1037.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 798.0, + 377.0, + 798.0, + 377.0, + 819.0, + 343.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 798.0, + 733.0, + 798.0, + 733.0, + 819.0, + 700.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 804.0, + 1095.0, + 804.0, + 1095.0, + 824.0, + 1057.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 822.0, + 377.0, + 822.0, + 377.0, + 843.0, + 342.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 819.0, + 735.0, + 819.0, + 735.0, + 850.0, + 696.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 830.0, + 1095.0, + 830.0, + 1095.0, + 851.0, + 1053.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 843.0, + 378.0, + 843.0, + 378.0, + 869.0, + 340.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 849.0, + 734.0, + 849.0, + 734.0, + 876.0, + 696.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 856.0, + 1095.0, + 856.0, + 1095.0, + 877.0, + 1053.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 866.0, + 378.0, + 866.0, + 378.0, + 913.0, + 340.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 877.0, + 733.0, + 877.0, + 733.0, + 899.0, + 698.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 883.0, + 1095.0, + 883.0, + 1095.0, + 904.0, + 1055.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 916.0, + 377.0, + 916.0, + 377.0, + 960.0, + 341.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 904.0, + 734.0, + 904.0, + 734.0, + 953.0, + 697.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 920.0, + 975.0, + 920.0, + 975.0, + 958.0, + 959.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 909.0, + 1095.0, + 909.0, + 1095.0, + 957.0, + 1053.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 962.0, + 378.0, + 962.0, + 378.0, + 986.0, + 340.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 958.0, + 654.0, + 958.0, + 654.0, + 977.0, + 527.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 955.0, + 734.0, + 955.0, + 734.0, + 980.0, + 697.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 958.0, + 1011.0, + 958.0, + 1011.0, + 978.0, + 881.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 963.0, + 1095.0, + 963.0, + 1095.0, + 984.0, + 1053.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 960.0, + 1367.0, + 960.0, + 1367.0, + 979.0, + 1228.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 972.0, + 593.0, + 972.0, + 593.0, + 991.0, + 527.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 970.0, + 951.0, + 970.0, + 951.0, + 993.0, + 883.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1242.0, + 975.0, + 1307.0, + 975.0, + 1307.0, + 993.0, + 1242.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 986.0, + 377.0, + 986.0, + 377.0, + 1008.0, + 341.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 984.0, + 640.0, + 984.0, + 640.0, + 1006.0, + 526.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 984.0, + 733.0, + 984.0, + 733.0, + 1006.0, + 698.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 984.0, + 995.0, + 984.0, + 995.0, + 1006.0, + 881.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 990.0, + 1095.0, + 990.0, + 1095.0, + 1010.0, + 1053.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 985.0, + 1352.0, + 985.0, + 1352.0, + 1008.0, + 1240.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1010.0, + 396.0, + 1010.0, + 396.0, + 1025.0, + 381.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1010.0, + 452.0, + 1010.0, + 452.0, + 1026.0, + 436.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1009.0, + 512.0, + 1009.0, + 512.0, + 1028.0, + 489.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1006.0, + 569.0, + 1006.0, + 569.0, + 1029.0, + 543.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1006.0, + 624.0, + 1006.0, + 624.0, + 1029.0, + 598.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1010.0, + 752.0, + 1010.0, + 752.0, + 1026.0, + 738.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 1008.0, + 810.0, + 1008.0, + 810.0, + 1028.0, + 792.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1006.0, + 869.0, + 1006.0, + 869.0, + 1029.0, + 843.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1005.0, + 926.0, + 1005.0, + 926.0, + 1029.0, + 899.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1006.0, + 982.0, + 1006.0, + 982.0, + 1029.0, + 954.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1011.0, + 1114.0, + 1011.0, + 1114.0, + 1028.0, + 1099.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1009.0, + 1171.0, + 1009.0, + 1171.0, + 1029.0, + 1153.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1009.0, + 1227.0, + 1009.0, + 1227.0, + 1028.0, + 1205.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1006.0, + 1283.0, + 1006.0, + 1283.0, + 1030.0, + 1258.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 1006.0, + 1338.0, + 1006.0, + 1338.0, + 1030.0, + 1312.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 1020.0, + 569.0, + 1020.0, + 569.0, + 1048.0, + 464.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1020.0, + 926.0, + 1020.0, + 926.0, + 1048.0, + 820.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1020.0, + 1283.0, + 1020.0, + 1283.0, + 1048.0, + 1180.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1052.0, + 554.0, + 1052.0, + 554.0, + 1091.0, + 434.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1052.0, + 910.0, + 1052.0, + 910.0, + 1091.0, + 790.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1051.0, + 1267.0, + 1051.0, + 1267.0, + 1091.0, + 1144.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1110.0, + 1064.0, + 1110.0, + 1064.0, + 1313.0, + 1034.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1111.0, + 350.0, + 1111.0, + 350.0, + 1371.0, + 324.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1111.0, + 705.0, + 1111.0, + 705.0, + 1371.0, + 679.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1136.0, + 733.0, + 1136.0, + 733.0, + 1158.0, + 700.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1136.0, + 1089.0, + 1136.0, + 1089.0, + 1158.0, + 1057.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1137.0, + 377.0, + 1137.0, + 377.0, + 1158.0, + 343.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1159.0, + 1089.0, + 1159.0, + 1089.0, + 1185.0, + 1055.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1177.0, + 377.0, + 1177.0, + 377.0, + 1198.0, + 342.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1176.0, + 734.0, + 1176.0, + 734.0, + 1202.0, + 697.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1185.0, + 1091.0, + 1185.0, + 1091.0, + 1210.0, + 1052.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1176.0, + 1257.0, + 1176.0, + 1257.0, + 1224.0, + 1238.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1217.0, + 377.0, + 1217.0, + 377.0, + 1238.0, + 342.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1221.0, + 733.0, + 1221.0, + 733.0, + 1243.0, + 698.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1211.0, + 1088.0, + 1211.0, + 1088.0, + 1233.0, + 1053.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1237.0, + 1089.0, + 1237.0, + 1089.0, + 1259.0, + 1055.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1257.0, + 375.0, + 1257.0, + 375.0, + 1279.0, + 341.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1261.0, + 734.0, + 1261.0, + 734.0, + 1286.0, + 696.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1263.0, + 1089.0, + 1263.0, + 1089.0, + 1284.0, + 1053.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1297.0, + 377.0, + 1297.0, + 377.0, + 1318.0, + 341.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1295.0, + 655.0, + 1295.0, + 655.0, + 1330.0, + 526.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1305.0, + 733.0, + 1305.0, + 733.0, + 1328.0, + 698.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1295.0, + 890.0, + 1295.0, + 890.0, + 1330.0, + 761.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 1288.0, + 1091.0, + 1288.0, + 1091.0, + 1371.0, + 1034.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 1297.0, + 1245.0, + 1297.0, + 1245.0, + 1330.0, + 1118.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1323.0, + 640.0, + 1323.0, + 640.0, + 1345.0, + 526.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1324.0, + 873.0, + 1324.0, + 873.0, + 1343.0, + 762.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 1324.0, + 1229.0, + 1324.0, + 1229.0, + 1343.0, + 1118.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1349.0, + 396.0, + 1349.0, + 396.0, + 1364.0, + 381.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1346.0, + 453.0, + 1346.0, + 453.0, + 1366.0, + 435.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1346.0, + 512.0, + 1346.0, + 512.0, + 1366.0, + 490.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1344.0, + 569.0, + 1344.0, + 569.0, + 1368.0, + 543.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1344.0, + 624.0, + 1344.0, + 624.0, + 1368.0, + 598.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1346.0, + 753.0, + 1346.0, + 753.0, + 1366.0, + 735.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 1345.0, + 810.0, + 1345.0, + 810.0, + 1366.0, + 792.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1344.0, + 869.0, + 1344.0, + 869.0, + 1366.0, + 843.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1345.0, + 926.0, + 1345.0, + 926.0, + 1368.0, + 899.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 1345.0, + 981.0, + 1345.0, + 981.0, + 1368.0, + 954.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1346.0, + 1111.0, + 1346.0, + 1111.0, + 1366.0, + 1093.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 1345.0, + 1166.0, + 1345.0, + 1166.0, + 1366.0, + 1148.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1346.0, + 1224.0, + 1346.0, + 1224.0, + 1366.0, + 1202.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 1346.0, + 1279.0, + 1346.0, + 1279.0, + 1366.0, + 1257.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1311.0, + 1345.0, + 1337.0, + 1345.0, + 1337.0, + 1368.0, + 1311.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1356.0, + 571.0, + 1356.0, + 571.0, + 1389.0, + 462.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1356.0, + 927.0, + 1356.0, + 927.0, + 1389.0, + 818.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 1356.0, + 1283.0, + 1356.0, + 1283.0, + 1389.0, + 1175.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1391.0, + 555.0, + 1391.0, + 555.0, + 1430.0, + 434.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1391.0, + 909.0, + 1391.0, + 909.0, + 1429.0, + 790.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1389.0, + 1265.0, + 1389.0, + 1265.0, + 1431.0, + 1149.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1458.0, + 1265.0, + 1458.0, + 1265.0, + 1498.0, + 434.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2121.0, + 830.0, + 2121.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 35, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 492, + 315, + 1207, + 315, + 1207, + 873, + 492, + 873 + ], + "score": 0.981, + "html": "
e1_p1e0.2_p1e0.05_p1
GALILEO5.17 ± 0.064.73 ± 0.134.70 ± 0.02
SL5.15 ± 0.234.73 ± 0.3123.64 ± 4.86
IPW5.22 ± 0.095.50 ± 0.015.02 ± 0.07
SCIGAN7.05 ± 0.526.58 ± 0.5818.55 ± 3.50
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO5.03 ± 0.094.72 ± 0.054.87 ± 0.15
SL5.21 ± 0.636.74 ± 0.1533.52 ±1.32
IPW5.27 ± 0.055.69 ± 0.0020.23 ± 0.45
SCIGAN16.07 ± 0.2712.07 ± 1.9319.27 ± 10.72
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO5.23 ± 0.415.01 ± 0.086.17 ± 0.33
SL5.89 ± 0.8814.25 ± 3.4837.50 ± 2.29
IPW5.21 ± 0.015.52 ± 0.4431.95 ± 0.05
SCIGAN11.50 ± 7.7613.05 ± 4.1925.74 ± 8.30
" + }, + { + "category_id": 5, + "poly": [ + 510, + 993, + 1187, + 993, + 1187, + 1549, + 510, + 1549 + ], + "score": 0.979, + "html": "
t0_bias_2.0 t0_bias_s_20.0 t0_bias_50.0
GALILEO SL0.34 ± 0.05 0.67 ± 0.13 0.38 ± 0.13 1.50 ± 0.312.04 ± 0.12 3.06 ± 0.65
IPW SCIGAN6.57 ± 1.16 0.74 ± 0.056.88 ± 0.30 5.84 ± 0.71 2.74 ± 0.35 3.19 ± 0.09
t1_bias_2.0t1_bias_6.0 t1_bias_8.0
GALILEO SL0.43 ± 0.05 0.47 ± 0.050.25 ± 0.02 0.21 ± 0.04 1.33 ± 0.97 1.18 ± 0.73
IPW3.67 ± 2.370.54 ± 0.13 2.69 ± 1.17
SCIGAN0.45 ± 0.251.08 ± 1.04 1.01 ± 0.77
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO
1.46 ± 0.090.85 ± 0.04 0.46 ± 0.01
SL IPW0.81 ± 0.143.74 ± 2.04 3.59 ± 0.14
" + }, + { + "category_id": 6, + "poly": [ + 292, + 254, + 1400, + 254, + 1400, + 315, + 292, + 315 + ], + "score": 0.932 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 857, + 75, + 857, + 105, + 297, + 105 + ], + "score": 0.917 + }, + { + "category_id": 6, + "poly": [ + 295, + 931, + 1397, + 931, + 1397, + 991, + 295, + 991 + ], + "score": 0.914 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 864, + 2088, + 864, + 2112, + 835, + 2112 + ], + "score": 0.851 + }, + { + "category_id": 13, + "poly": [ + 402, + 929, + 515, + 929, + 515, + 963, + 402, + 963 + ], + "score": 0.89, + "latex": "\\sqrt { M I S E }" + }, + { + "category_id": 13, + "poly": [ + 403, + 252, + 515, + 252, + 515, + 286, + 403, + 286 + ], + "score": 0.87, + "latex": "\\sqrt { M I S E }" + }, + { + "category_id": 13, + "poly": [ + 1196, + 258, + 1227, + 258, + 1227, + 286, + 1196, + 286 + ], + "score": 0.58, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 1196, + 935, + 1227, + 935, + 1227, + 963, + 1196, + 963 + ], + "score": 0.33, + "latex": "\\pm" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 249.0, + 402.0, + 249.0, + 402.0, + 292.0, + 291.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 249.0, + 1195.0, + 249.0, + 1195.0, + 292.0, + 516.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 249.0, + 1406.0, + 249.0, + 1406.0, + 292.0, + 1228.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 286.0, + 661.0, + 286.0, + 661.0, + 317.0, + 296.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 927.0, + 401.0, + 927.0, + 401.0, + 969.0, + 291.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 927.0, + 1195.0, + 927.0, + 1195.0, + 969.0, + 516.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 927.0, + 1404.0, + 927.0, + 1404.0, + 969.0, + 1228.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 963.0, + 662.0, + 963.0, + 662.0, + 995.0, + 296.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 36, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 500, + 440, + 1200, + 440, + 1200, + 996, + 500, + 996 + ], + "score": 0.982, + "html": "
e1_p1e0.2_p1e0.05_p1
GALILEO3.86 ± 0.033.99 ± 0.014.07 ± 0.03
SL5.73 ± 0.335.80 ± 0.2818.78 ± 3.13
IPW4.02 ± 0.054.15 ± 0.1222.66 ± 0.33
SCIGAN8.84 ± 0.5412.62 ± 2.1724.21 ± 5.20
e1_p0.2e0.2_p0.2e0.05_p0.2
GALILEO4.13 ± 0.104.11 ± 0.154.21 ± 0.15
SL5.87 ± 0.437.44 ± 1.1329.13 ± 3.44
IPW4.12 ± 0.026.12 ± 0.4830.96 ± 0.17
SCIGAN12.87 ± 3.0214.59 ± 2.1324.57 ± 3.00
e1_p0.05e0.2_p0.05e0.05_p0.05
GALILEO4.39 ± 0.204.34 ± 0.205.26 ± 0.29
SL6.12 ± 0.4314.88 ± 4.4130.81 ± 1.69
IPW13.60 ± 7.8326.27 ± 2.6732.55 ± 0.12
SCIGAN9.19 ± 1.0415.08 ± 1.2617.52 ± 0.02
" + }, + { + "category_id": 5, + "poly": [ + 510, + 1347, + 1187, + 1347, + 1187, + 1907, + 510, + 1907 + ], + "score": 0.981, + "html": "
t0_bias_2.0t0_bias_20.0 tO_bias_50.0
GALILEO1.56 ± 0.041.96 ± 0.53 3.16 ± 0.13
SL1.92 ± 0.672.31 ± 0.19 5.11 ± 0.66
IPW7.42 ± 0.465.36 ± 0.96 5.38 ± 1.24
SCIGAN2.11 ± 0.475.23 ±0.27 5.59 ± 1.02
tl_bias_2.0tl_bias_6.0 tl_bias_8.0
GALILEO1.43 ± 0.061.09 ± 0.05 1.36 ± 0.36
SL1.12 ± 0.153.65 ± 1.91 3.96 ± 1.81
IPW1.14 ± 0.11 0.90 ± 0.092.04 ± 0.99
SCIGAN3.32 ± 0.88 4.74 ± 2.125.17 ± 2.42
t2_bias_2.0t2_bias_6.0 t2_bias_8.0
GALILEO3.77 ± 0.35 3.99 ± 0.402.08 ± 0.60
SL2.70 ± 0.67 8.33 ± 5.059.70 ± 3.12
IPW2.92 ± 0.15 3.90 ± 0.174.47 ± 2.16
SCIGAN3.82 ± 2.12 1.83 ± 1.493.62 ± 4.9
" + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 858, + 74, + 858, + 106, + 297, + 106 + ], + "score": 0.913 + }, + { + "category_id": 6, + "poly": [ + 291, + 375, + 1402, + 375, + 1402, + 437, + 291, + 437 + ], + "score": 0.895 + }, + { + "category_id": 6, + "poly": [ + 289, + 1286, + 1399, + 1286, + 1399, + 1349, + 289, + 1349 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 865, + 2087, + 865, + 2113, + 834, + 2113 + ], + "score": 0.862 + }, + { + "category_id": 13, + "poly": [ + 413, + 1285, + 540, + 1285, + 540, + 1320, + 413, + 1320 + ], + "score": 0.91, + "latex": "\\sqrt { M M S E }" + }, + { + "category_id": 13, + "poly": [ + 412, + 374, + 540, + 374, + 540, + 409, + 412, + 409 + ], + "score": 0.91, + "latex": "\\sqrt { M M S E }" + }, + { + "category_id": 13, + "poly": [ + 1201, + 380, + 1230, + 380, + 1230, + 408, + 1201, + 408 + ], + "score": 0.68, + "latex": "\\pm" + }, + { + "category_id": 13, + "poly": [ + 1202, + 1291, + 1230, + 1291, + 1230, + 1319, + 1202, + 1319 + ], + "score": 0.64, + "latex": "\\pm" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 374.0, + 411.0, + 374.0, + 411.0, + 410.0, + 293.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 374.0, + 1200.0, + 374.0, + 1200.0, + 410.0, + 541.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 374.0, + 1405.0, + 374.0, + 1405.0, + 410.0, + 1231.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 409.0, + 662.0, + 409.0, + 662.0, + 440.0, + 295.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1283.0, + 412.0, + 1283.0, + 412.0, + 1324.0, + 292.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1283.0, + 1201.0, + 1283.0, + 1201.0, + 1324.0, + 541.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 1283.0, + 1405.0, + 1283.0, + 1405.0, + 1324.0, + 1231.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1320.0, + 661.0, + 1320.0, + 661.0, + 1352.0, + 296.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2121.0, + 830.0, + 2121.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 37, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 325, + 671, + 1374, + 671, + 1374, + 1534, + 325, + 1534 + ], + "score": 0.976 + }, + { + "category_id": 4, + "poly": [ + 317, + 1568, + 1370, + 1568, + 1370, + 1602, + 317, + 1602 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 858, + 75, + 858, + 106, + 297, + 106 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2113, + 835, + 2113 + ], + "score": 0.845 + }, + { + "category_id": 15, + "poly": [ + 344.0, + 672.0, + 371.0, + 672.0, + 371.0, + 696.0, + 344.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 674.0, + 726.0, + 674.0, + 726.0, + 694.0, + 703.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 672.0, + 1083.0, + 672.0, + 1083.0, + 696.0, + 1058.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 685.0, + 371.0, + 685.0, + 371.0, + 850.0, + 325.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 682.0, + 731.0, + 682.0, + 731.0, + 851.0, + 676.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 687.0, + 1007.0, + 687.0, + 1007.0, + 755.0, + 992.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 687.0, + 1085.0, + 687.0, + 1085.0, + 771.0, + 1041.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 709.0, + 375.0, + 709.0, + 375.0, + 762.0, + 344.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 741.0, + 411.0, + 741.0, + 411.0, + 802.0, + 387.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 733.0, + 639.0, + 733.0, + 639.0, + 744.0, + 629.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 716.0, + 727.0, + 716.0, + 727.0, + 762.0, + 701.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 717.0, + 1084.0, + 717.0, + 1084.0, + 763.0, + 1058.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 746.0, + 1346.0, + 746.0, + 1346.0, + 770.0, + 1334.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 762.0, + 370.0, + 762.0, + 370.0, + 786.0, + 344.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 764.0, + 725.0, + 764.0, + 725.0, + 785.0, + 702.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 758.0, + 1084.0, + 758.0, + 1084.0, + 848.0, + 1036.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.0, + 759.0, + 1344.0, + 759.0, + 1344.0, + 798.0, + 1329.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 784.0, + 370.0, + 784.0, + 370.0, + 809.0, + 344.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 785.0, + 727.0, + 785.0, + 727.0, + 809.0, + 702.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 786.0, + 1084.0, + 786.0, + 1084.0, + 809.0, + 1058.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 810.0, + 368.0, + 810.0, + 368.0, + 827.0, + 354.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 811.0, + 453.0, + 811.0, + 453.0, + 834.0, + 410.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 809.0, + 727.0, + 809.0, + 727.0, + 828.0, + 708.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 812.0, + 810.0, + 812.0, + 810.0, + 832.0, + 765.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 810.0, + 1082.0, + 810.0, + 1082.0, + 827.0, + 1066.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 812.0, + 1165.0, + 812.0, + 1165.0, + 834.0, + 1125.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 805.0, + 1325.0, + 805.0, + 1325.0, + 824.0, + 1310.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 829.0, + 371.0, + 829.0, + 371.0, + 852.0, + 353.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 830.0, + 450.0, + 830.0, + 450.0, + 852.0, + 411.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 829.0, + 727.0, + 829.0, + 727.0, + 852.0, + 709.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 830.0, + 805.0, + 830.0, + 805.0, + 852.0, + 767.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 830.0, + 1083.0, + 830.0, + 1083.0, + 852.0, + 1065.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 830.0, + 1162.0, + 830.0, + 1162.0, + 852.0, + 1125.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 857.0, + 397.0, + 857.0, + 397.0, + 883.0, + 362.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 857.0, + 450.0, + 857.0, + 450.0, + 883.0, + 415.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 858.0, + 502.0, + 858.0, + 502.0, + 881.0, + 470.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 858.0, + 556.0, + 858.0, + 556.0, + 881.0, + 523.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 858.0, + 609.0, + 858.0, + 609.0, + 881.0, + 577.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 858.0, + 662.0, + 858.0, + 662.0, + 881.0, + 628.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 857.0, + 779.0, + 857.0, + 779.0, + 883.0, + 744.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 858.0, + 838.0, + 858.0, + 838.0, + 881.0, + 805.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 857.0, + 899.0, + 857.0, + 899.0, + 883.0, + 864.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 858.0, + 958.0, + 858.0, + 958.0, + 881.0, + 926.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 858.0, + 1018.0, + 858.0, + 1018.0, + 881.0, + 987.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 858.0, + 1374.0, + 858.0, + 1374.0, + 881.0, + 1086.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 873.0, + 545.0, + 873.0, + 545.0, + 901.0, + 481.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 871.0, + 901.0, + 871.0, + 901.0, + 900.0, + 839.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 873.0, + 1258.0, + 873.0, + 1258.0, + 900.0, + 1194.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 905.0, + 563.0, + 905.0, + 563.0, + 941.0, + 423.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 905.0, + 925.0, + 905.0, + 925.0, + 940.0, + 774.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 907.0, + 1280.0, + 907.0, + 1280.0, + 939.0, + 1132.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 965.0, + 372.0, + 965.0, + 372.0, + 989.0, + 343.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 984.0, + 371.0, + 984.0, + 371.0, + 1148.0, + 322.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 981.0, + 452.0, + 981.0, + 452.0, + 1022.0, + 386.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 982.0, + 729.0, + 982.0, + 729.0, + 1148.0, + 681.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 985.0, + 962.0, + 985.0, + 962.0, + 993.0, + 945.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 995.0, + 1004.0, + 995.0, + 1004.0, + 1021.0, + 963.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 985.0, + 1083.0, + 985.0, + 1083.0, + 1147.0, + 1035.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1022.0, + 727.0, + 1022.0, + 727.0, + 1047.0, + 702.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1016.0, + 1084.0, + 1016.0, + 1084.0, + 1039.0, + 1058.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1034.0, + 371.0, + 1034.0, + 371.0, + 1058.0, + 344.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1042.0, + 727.0, + 1042.0, + 727.0, + 1069.0, + 701.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1041.0, + 1083.0, + 1041.0, + 1083.0, + 1065.0, + 1058.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1069.0, + 371.0, + 1069.0, + 371.0, + 1093.0, + 344.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1064.0, + 727.0, + 1064.0, + 727.0, + 1089.0, + 701.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1068.0, + 1083.0, + 1068.0, + 1083.0, + 1092.0, + 1058.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1104.0, + 371.0, + 1104.0, + 371.0, + 1127.0, + 344.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1110.0, + 404.0, + 1110.0, + 404.0, + 1157.0, + 383.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1088.0, + 726.0, + 1088.0, + 726.0, + 1131.0, + 701.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 1095.0, + 1083.0, + 1095.0, + 1083.0, + 1118.0, + 1060.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1110.0, + 1261.0, + 1110.0, + 1261.0, + 1132.0, + 1216.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1122.0, + 730.0, + 1122.0, + 730.0, + 1152.0, + 702.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1136.0, + 999.0, + 1136.0, + 999.0, + 1142.0, + 993.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1123.0, + 1082.0, + 1123.0, + 1082.0, + 1142.0, + 1057.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1127.0, + 1259.0, + 1127.0, + 1259.0, + 1153.0, + 1194.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1134.0, + 1351.0, + 1134.0, + 1351.0, + 1152.0, + 1339.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1139.0, + 371.0, + 1139.0, + 371.0, + 1160.0, + 353.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1141.0, + 1004.0, + 1141.0, + 1004.0, + 1153.0, + 993.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1157.0, + 396.0, + 1157.0, + 396.0, + 1180.0, + 365.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1155.0, + 451.0, + 1155.0, + 451.0, + 1182.0, + 416.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 1157.0, + 504.0, + 1157.0, + 504.0, + 1180.0, + 470.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1157.0, + 556.0, + 1157.0, + 556.0, + 1180.0, + 523.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1155.0, + 610.0, + 1155.0, + 610.0, + 1182.0, + 575.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1157.0, + 662.0, + 1157.0, + 662.0, + 1180.0, + 629.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1154.0, + 785.0, + 1154.0, + 785.0, + 1181.0, + 751.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 1154.0, + 848.0, + 1154.0, + 848.0, + 1181.0, + 814.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1154.0, + 912.0, + 1154.0, + 912.0, + 1181.0, + 876.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1154.0, + 974.0, + 1154.0, + 974.0, + 1181.0, + 938.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1157.0, + 1374.0, + 1157.0, + 1374.0, + 1180.0, + 1092.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1170.0, + 543.0, + 1170.0, + 543.0, + 1198.0, + 478.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 1167.0, + 904.0, + 1167.0, + 904.0, + 1199.0, + 839.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1172.0, + 1253.0, + 1172.0, + 1253.0, + 1196.0, + 1194.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1204.0, + 563.0, + 1204.0, + 563.0, + 1239.0, + 423.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1205.0, + 918.0, + 1205.0, + 918.0, + 1240.0, + 784.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1205.0, + 1271.0, + 1205.0, + 1271.0, + 1238.0, + 1140.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 1265.0, + 1097.0, + 1265.0, + 1097.0, + 1461.0, + 1021.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1270.0, + 371.0, + 1270.0, + 371.0, + 1293.0, + 344.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1271.0, + 1084.0, + 1271.0, + 1084.0, + 1295.0, + 1058.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1273.0, + 738.0, + 1273.0, + 738.0, + 1299.0, + 700.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1279.0, + 354.0, + 1279.0, + 354.0, + 1443.0, + 324.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1283.0, + 711.0, + 1283.0, + 711.0, + 1443.0, + 681.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1296.0, + 737.0, + 1296.0, + 737.0, + 1323.0, + 700.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1298.0, + 1082.0, + 1298.0, + 1082.0, + 1318.0, + 1059.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1308.0, + 371.0, + 1308.0, + 371.0, + 1331.0, + 344.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1318.0, + 739.0, + 1318.0, + 739.0, + 1346.0, + 700.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1322.0, + 1084.0, + 1322.0, + 1084.0, + 1344.0, + 1058.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1346.0, + 371.0, + 1346.0, + 371.0, + 1369.0, + 344.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1343.0, + 738.0, + 1343.0, + 738.0, + 1369.0, + 700.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 1360.0, + 779.0, + 1360.0, + 779.0, + 1370.0, + 771.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1347.0, + 1084.0, + 1347.0, + 1084.0, + 1371.0, + 1058.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 1359.0, + 1124.0, + 1359.0, + 1124.0, + 1370.0, + 1115.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1367.0, + 420.0, + 1367.0, + 420.0, + 1379.0, + 411.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1366.0, + 738.0, + 1366.0, + 738.0, + 1394.0, + 700.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1367.0, + 775.0, + 1367.0, + 775.0, + 1379.0, + 766.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 1367.0, + 1119.0, + 1367.0, + 1119.0, + 1379.0, + 1110.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1373.0, + 1082.0, + 1373.0, + 1082.0, + 1394.0, + 1059.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1384.0, + 371.0, + 1384.0, + 371.0, + 1446.0, + 343.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1420.0, + 393.0, + 1420.0, + 393.0, + 1432.0, + 383.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 1405.0, + 549.0, + 1405.0, + 549.0, + 1448.0, + 483.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1390.0, + 739.0, + 1390.0, + 739.0, + 1440.0, + 699.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1396.0, + 760.0, + 1396.0, + 760.0, + 1418.0, + 747.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 1407.0, + 913.0, + 1407.0, + 913.0, + 1448.0, + 840.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1397.0, + 1083.0, + 1397.0, + 1083.0, + 1420.0, + 1057.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1405.0, + 1261.0, + 1405.0, + 1261.0, + 1447.0, + 1191.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1438.0, + 388.0, + 1438.0, + 388.0, + 1448.0, + 379.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1437.0, + 739.0, + 1437.0, + 739.0, + 1464.0, + 699.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1440.0, + 1005.0, + 1440.0, + 1005.0, + 1449.0, + 996.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1452.0, + 396.0, + 1452.0, + 396.0, + 1474.0, + 365.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1449.0, + 450.0, + 1449.0, + 450.0, + 1476.0, + 416.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1449.0, + 505.0, + 1449.0, + 505.0, + 1476.0, + 469.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1449.0, + 557.0, + 1449.0, + 557.0, + 1476.0, + 522.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1449.0, + 610.0, + 1449.0, + 610.0, + 1476.0, + 575.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1449.0, + 663.0, + 1449.0, + 663.0, + 1476.0, + 628.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1452.0, + 792.0, + 1452.0, + 792.0, + 1476.0, + 762.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1452.0, + 853.0, + 1452.0, + 853.0, + 1474.0, + 821.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1449.0, + 915.0, + 1449.0, + 915.0, + 1477.0, + 881.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1453.0, + 974.0, + 1453.0, + 974.0, + 1476.0, + 944.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 1449.0, + 1304.0, + 1449.0, + 1304.0, + 1476.0, + 1091.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1452.0, + 1339.0, + 1452.0, + 1339.0, + 1473.0, + 1307.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1341.0, + 1452.0, + 1373.0, + 1452.0, + 1373.0, + 1473.0, + 1341.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 1466.0, + 543.0, + 1466.0, + 543.0, + 1494.0, + 480.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 1466.0, + 906.0, + 1466.0, + 906.0, + 1494.0, + 844.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 1466.0, + 1255.0, + 1466.0, + 1255.0, + 1494.0, + 1193.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1499.0, + 563.0, + 1499.0, + 563.0, + 1536.0, + 424.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1499.0, + 920.0, + 1499.0, + 920.0, + 1536.0, + 780.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1499.0, + 1273.0, + 1499.0, + 1273.0, + 1536.0, + 1139.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 758.0, + 539.0, + 758.0, + 539.0, + 768.0, + 481.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 763.5, + 879.0, + 763.5, + 879.0, + 781.5, + 818.0, + 781.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.25, + 710.5, + 428.25, + 710.5, + 428.25, + 715.5, + 404.25, + 715.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.25, + 776.5, + 1222.25, + 776.5, + 1222.25, + 787.5, + 1193.25, + 787.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 783.5, + 993.0, + 783.5, + 993.0, + 789.5, + 953.0, + 789.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 976.5, + 1008.0, + 976.5, + 1008.0, + 1002.0, + 961.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1069.5, + 633.0, + 1069.5, + 633.0, + 1084.0, + 576.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.25, + 1049.5, + 968.25, + 1049.5, + 968.25, + 1066.5, + 941.25, + 1066.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.25, + 1030.5, + 1167.25, + 1030.5, + 1167.25, + 1040.5, + 1121.25, + 1040.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.75, + 1042.0, + 818.75, + 1042.0, + 818.75, + 1063.0, + 724.75, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1091.5, + 1130.0, + 1091.5, + 1130.0, + 1100.0, + 1084.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 1312.5, + 934.0, + 1312.5, + 934.0, + 1321.0, + 923.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1375.0, + 972.0, + 1375.0, + 972.0, + 1381.5, + 964.0, + 1381.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.25, + 1337.0, + 1152.25, + 1337.0, + 1152.25, + 1346.0, + 1113.25, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1565.0, + 1373.0, + 1565.0, + 1373.0, + 1605.0, + 319.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2122.0, + 830.0, + 2122.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 38, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 324, + 669, + 1372, + 669, + 1372, + 1538, + 324, + 1538 + ], + "score": 0.976 + }, + { + "category_id": 4, + "poly": [ + 316, + 1569, + 1369, + 1569, + 1369, + 1604, + 316, + 1604 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 858, + 75, + 858, + 106, + 297, + 106 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2113, + 834, + 2113 + ], + "score": 0.848 + }, + { + "category_id": 15, + "poly": [ + 324.0, + 680.0, + 371.0, + 680.0, + 371.0, + 847.0, + 324.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 678.0, + 414.0, + 678.0, + 414.0, + 699.0, + 379.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 680.0, + 727.0, + 680.0, + 727.0, + 772.0, + 681.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 682.0, + 761.0, + 682.0, + 761.0, + 695.0, + 739.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 678.0, + 809.0, + 678.0, + 809.0, + 701.0, + 766.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 672.0, + 1008.0, + 672.0, + 1008.0, + 697.0, + 980.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 669.0, + 1096.0, + 669.0, + 1096.0, + 860.0, + 1021.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 682.0, + 1316.0, + 682.0, + 1316.0, + 695.0, + 1296.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 700.0, + 371.0, + 700.0, + 371.0, + 749.0, + 343.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 695.0, + 452.0, + 695.0, + 452.0, + 721.0, + 382.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 709.0, + 730.0, + 709.0, + 730.0, + 732.0, + 701.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 695.0, + 807.0, + 695.0, + 807.0, + 721.0, + 737.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 710.0, + 1084.0, + 710.0, + 1084.0, + 733.0, + 1056.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 695.0, + 1360.0, + 695.0, + 1360.0, + 734.0, + 1286.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 745.0, + 373.0, + 745.0, + 373.0, + 774.0, + 343.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 738.0, + 737.0, + 738.0, + 737.0, + 859.0, + 665.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 747.0, + 1084.0, + 747.0, + 1084.0, + 770.0, + 1056.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 766.0, + 372.0, + 766.0, + 372.0, + 796.0, + 343.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 766.0, + 728.0, + 766.0, + 728.0, + 790.0, + 700.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 781.0, + 1263.0, + 781.0, + 1263.0, + 790.0, + 1253.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 793.0, + 371.0, + 793.0, + 371.0, + 819.0, + 345.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 795.0, + 727.0, + 795.0, + 727.0, + 819.0, + 701.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 785.0, + 1084.0, + 785.0, + 1084.0, + 808.0, + 1057.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 815.0, + 372.0, + 815.0, + 372.0, + 843.0, + 341.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 830.0, + 391.0, + 830.0, + 391.0, + 841.0, + 379.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 841.0, + 372.0, + 841.0, + 372.0, + 864.0, + 345.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 841.0, + 392.0, + 841.0, + 392.0, + 855.0, + 378.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 859.0, + 407.0, + 859.0, + 407.0, + 878.0, + 382.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 856.0, + 443.0, + 856.0, + 443.0, + 879.0, + 410.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 858.0, + 473.0, + 858.0, + 473.0, + 878.0, + 452.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 858.0, + 508.0, + 858.0, + 508.0, + 877.0, + 489.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 858.0, + 544.0, + 858.0, + 544.0, + 878.0, + 523.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 858.0, + 580.0, + 858.0, + 580.0, + 878.0, + 560.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 858.0, + 617.0, + 858.0, + 617.0, + 878.0, + 594.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 858.0, + 651.0, + 858.0, + 651.0, + 878.0, + 632.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 859.0, + 776.0, + 859.0, + 776.0, + 878.0, + 752.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 858.0, + 818.0, + 858.0, + 818.0, + 878.0, + 798.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 860.0, + 860.0, + 860.0, + 860.0, + 876.0, + 844.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 858.0, + 905.0, + 858.0, + 905.0, + 878.0, + 886.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 858.0, + 952.0, + 858.0, + 952.0, + 878.0, + 931.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 858.0, + 995.0, + 858.0, + 995.0, + 878.0, + 977.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 859.0, + 1131.0, + 859.0, + 1131.0, + 878.0, + 1106.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 858.0, + 1173.0, + 858.0, + 1173.0, + 878.0, + 1154.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 858.0, + 1219.0, + 858.0, + 1219.0, + 877.0, + 1200.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 858.0, + 1264.0, + 858.0, + 1264.0, + 878.0, + 1245.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 858.0, + 1311.0, + 858.0, + 1311.0, + 878.0, + 1292.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 859.0, + 1357.0, + 859.0, + 1357.0, + 878.0, + 1338.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 871.0, + 545.0, + 871.0, + 545.0, + 899.0, + 482.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 870.0, + 903.0, + 870.0, + 903.0, + 899.0, + 837.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 871.0, + 1258.0, + 871.0, + 1258.0, + 899.0, + 1195.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 901.0, + 550.0, + 901.0, + 550.0, + 945.0, + 435.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 902.0, + 916.0, + 902.0, + 916.0, + 942.0, + 783.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 902.0, + 1276.0, + 902.0, + 1276.0, + 942.0, + 1135.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 971.0, + 372.0, + 971.0, + 372.0, + 1146.0, + 324.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 981.0, + 404.0, + 981.0, + 404.0, + 994.0, + 384.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 976.0, + 455.0, + 976.0, + 455.0, + 1000.0, + 407.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 976.0, + 728.0, + 976.0, + 728.0, + 1148.0, + 677.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 977.0, + 810.0, + 977.0, + 810.0, + 999.0, + 765.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 973.0, + 1005.0, + 973.0, + 1005.0, + 987.0, + 988.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 969.0, + 1099.0, + 969.0, + 1099.0, + 1080.0, + 1026.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 977.0, + 1362.0, + 977.0, + 1362.0, + 999.0, + 1317.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 993.0, + 371.0, + 993.0, + 371.0, + 1019.0, + 345.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 994.0, + 452.0, + 994.0, + 452.0, + 1020.0, + 385.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 994.0, + 807.0, + 994.0, + 807.0, + 1020.0, + 744.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 994.0, + 1360.0, + 994.0, + 1360.0, + 1020.0, + 1294.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1007.0, + 728.0, + 1007.0, + 728.0, + 1030.0, + 701.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1009.0, + 1084.0, + 1009.0, + 1084.0, + 1033.0, + 1055.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1019.0, + 371.0, + 1019.0, + 371.0, + 1043.0, + 343.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1042.0, + 372.0, + 1042.0, + 372.0, + 1067.0, + 345.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1037.0, + 727.0, + 1037.0, + 727.0, + 1061.0, + 701.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1057.0, + 745.0, + 1057.0, + 745.0, + 1066.0, + 736.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1043.0, + 1097.0, + 1043.0, + 1097.0, + 1155.0, + 1023.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1067.0, + 371.0, + 1067.0, + 371.0, + 1091.0, + 345.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1066.0, + 728.0, + 1066.0, + 728.0, + 1090.0, + 701.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1082.0, + 844.0, + 1082.0, + 844.0, + 1090.0, + 832.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 1082.0, + 940.0, + 1082.0, + 940.0, + 1091.0, + 927.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1093.0, + 371.0, + 1093.0, + 371.0, + 1117.0, + 343.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1096.0, + 728.0, + 1096.0, + 728.0, + 1120.0, + 701.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1099.0, + 811.0, + 1099.0, + 811.0, + 1110.0, + 801.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1101.0, + 974.0, + 1101.0, + 974.0, + 1110.0, + 962.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1117.0, + 372.0, + 1117.0, + 372.0, + 1141.0, + 343.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1120.0, + 1329.0, + 1120.0, + 1329.0, + 1129.0, + 1319.0, + 1129.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1125.0, + 728.0, + 1125.0, + 728.0, + 1150.0, + 700.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1141.0, + 372.0, + 1141.0, + 372.0, + 1165.0, + 345.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1154.0, + 398.0, + 1154.0, + 398.0, + 1177.0, + 368.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1156.0, + 435.0, + 1156.0, + 435.0, + 1177.0, + 402.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1156.0, + 469.0, + 1156.0, + 469.0, + 1179.0, + 446.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1157.0, + 504.0, + 1157.0, + 504.0, + 1177.0, + 484.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1156.0, + 542.0, + 1156.0, + 542.0, + 1177.0, + 522.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1154.0, + 583.0, + 1154.0, + 583.0, + 1179.0, + 557.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1157.0, + 618.0, + 1157.0, + 618.0, + 1177.0, + 596.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1156.0, + 657.0, + 1156.0, + 657.0, + 1179.0, + 634.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1157.0, + 771.0, + 1157.0, + 771.0, + 1176.0, + 748.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1156.0, + 815.0, + 1156.0, + 815.0, + 1177.0, + 794.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 1158.0, + 858.0, + 1158.0, + 858.0, + 1175.0, + 842.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 1156.0, + 908.0, + 1156.0, + 908.0, + 1179.0, + 885.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 1156.0, + 953.0, + 1156.0, + 953.0, + 1177.0, + 933.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1156.0, + 1001.0, + 1156.0, + 1001.0, + 1179.0, + 978.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1154.0, + 1128.0, + 1154.0, + 1128.0, + 1177.0, + 1100.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1156.0, + 1170.0, + 1156.0, + 1170.0, + 1177.0, + 1149.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1157.0, + 1216.0, + 1157.0, + 1216.0, + 1177.0, + 1197.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 1156.0, + 1264.0, + 1156.0, + 1264.0, + 1177.0, + 1244.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 1156.0, + 1312.0, + 1156.0, + 1312.0, + 1177.0, + 1292.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1157.0, + 1359.0, + 1157.0, + 1359.0, + 1176.0, + 1338.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 1169.0, + 545.0, + 1169.0, + 545.0, + 1197.0, + 481.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1169.0, + 903.0, + 1169.0, + 903.0, + 1197.0, + 838.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1169.0, + 1258.0, + 1169.0, + 1258.0, + 1197.0, + 1195.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1199.0, + 562.0, + 1199.0, + 562.0, + 1240.0, + 424.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 1201.0, + 926.0, + 1201.0, + 926.0, + 1238.0, + 774.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 1200.0, + 1287.0, + 1200.0, + 1287.0, + 1239.0, + 1127.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1266.0, + 383.0, + 1266.0, + 383.0, + 1458.0, + 308.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1280.0, + 407.0, + 1280.0, + 407.0, + 1292.0, + 385.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1274.0, + 457.0, + 1274.0, + 457.0, + 1302.0, + 410.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1272.0, + 648.0, + 1272.0, + 648.0, + 1283.0, + 636.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1264.0, + 746.0, + 1264.0, + 746.0, + 1459.0, + 669.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1272.0, + 1084.0, + 1272.0, + 1084.0, + 1367.0, + 1035.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1272.0, + 1105.0, + 1272.0, + 1105.0, + 1285.0, + 1090.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1276.0, + 1363.0, + 1276.0, + 1363.0, + 1298.0, + 1317.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1301.0, + 372.0, + 1301.0, + 372.0, + 1325.0, + 345.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1292.0, + 451.0, + 1292.0, + 451.0, + 1318.0, + 379.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 1294.0, + 616.0, + 1294.0, + 616.0, + 1305.0, + 604.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1305.0, + 728.0, + 1305.0, + 728.0, + 1329.0, + 701.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1306.0, + 975.0, + 1306.0, + 975.0, + 1315.0, + 960.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1297.0, + 1003.0, + 1297.0, + 1003.0, + 1307.0, + 994.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1294.0, + 1084.0, + 1294.0, + 1084.0, + 1318.0, + 1056.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 1292.0, + 1360.0, + 1292.0, + 1360.0, + 1318.0, + 1294.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1315.0, + 583.0, + 1315.0, + 583.0, + 1324.0, + 573.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1317.0, + 1084.0, + 1317.0, + 1084.0, + 1341.0, + 1056.0, + 1341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1320.0, + 1362.0, + 1320.0, + 1362.0, + 1334.0, + 1347.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1329.0, + 373.0, + 1329.0, + 373.0, + 1357.0, + 342.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1329.0, + 727.0, + 1329.0, + 727.0, + 1354.0, + 701.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1336.0, + 846.0, + 1336.0, + 846.0, + 1351.0, + 830.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1336.0, + 1086.0, + 1336.0, + 1086.0, + 1365.0, + 1056.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 1337.0, + 1200.0, + 1337.0, + 1200.0, + 1347.0, + 1185.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1360.0, + 371.0, + 1360.0, + 371.0, + 1385.0, + 345.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1353.0, + 727.0, + 1353.0, + 727.0, + 1377.0, + 701.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1342.0, + 1098.0, + 1342.0, + 1098.0, + 1457.0, + 1025.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 1380.0, + 727.0, + 1380.0, + 727.0, + 1399.0, + 703.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1389.0, + 371.0, + 1389.0, + 371.0, + 1415.0, + 345.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1391.0, + 941.0, + 1391.0, + 941.0, + 1404.0, + 926.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 1399.0, + 452.0, + 1399.0, + 452.0, + 1408.0, + 443.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1400.0, + 728.0, + 1400.0, + 728.0, + 1426.0, + 701.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1407.0, + 810.0, + 1407.0, + 810.0, + 1427.0, + 766.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1416.0, + 972.0, + 1416.0, + 972.0, + 1424.0, + 962.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1405.0, + 1085.0, + 1405.0, + 1085.0, + 1434.0, + 1055.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1427.0, + 647.0, + 1427.0, + 647.0, + 1438.0, + 636.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1424.0, + 728.0, + 1424.0, + 728.0, + 1449.0, + 701.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1428.0, + 762.0, + 1428.0, + 762.0, + 1441.0, + 739.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1423.0, + 807.0, + 1423.0, + 807.0, + 1449.0, + 767.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1440.0, + 1003.0, + 1440.0, + 1003.0, + 1450.0, + 994.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1431.0, + 1084.0, + 1431.0, + 1084.0, + 1455.0, + 1057.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1421.0, + 1329.0, + 1421.0, + 1329.0, + 1432.0, + 1317.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1441.0, + 1360.0, + 1441.0, + 1360.0, + 1451.0, + 1348.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1454.0, + 423.0, + 1454.0, + 423.0, + 1477.0, + 397.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 1455.0, + 464.0, + 1455.0, + 464.0, + 1475.0, + 444.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1455.0, + 508.0, + 1455.0, + 508.0, + 1474.0, + 489.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 1455.0, + 555.0, + 1455.0, + 555.0, + 1475.0, + 535.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1455.0, + 600.0, + 1455.0, + 600.0, + 1475.0, + 580.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1455.0, + 646.0, + 1455.0, + 646.0, + 1475.0, + 628.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1454.0, + 775.0, + 1454.0, + 775.0, + 1477.0, + 746.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1455.0, + 817.0, + 1455.0, + 817.0, + 1475.0, + 795.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1455.0, + 862.0, + 1455.0, + 862.0, + 1475.0, + 843.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1455.0, + 909.0, + 1455.0, + 909.0, + 1475.0, + 890.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 1455.0, + 956.0, + 1455.0, + 956.0, + 1475.0, + 935.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1455.0, + 1001.0, + 1455.0, + 1001.0, + 1475.0, + 983.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1456.0, + 1130.0, + 1456.0, + 1130.0, + 1475.0, + 1106.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1455.0, + 1173.0, + 1455.0, + 1173.0, + 1475.0, + 1153.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1455.0, + 1220.0, + 1455.0, + 1220.0, + 1475.0, + 1200.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 1455.0, + 1265.0, + 1455.0, + 1265.0, + 1475.0, + 1245.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 1455.0, + 1312.0, + 1455.0, + 1312.0, + 1475.0, + 1293.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1455.0, + 1357.0, + 1455.0, + 1357.0, + 1475.0, + 1339.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1467.0, + 547.0, + 1467.0, + 547.0, + 1496.0, + 482.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1467.0, + 902.0, + 1467.0, + 902.0, + 1496.0, + 838.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1467.0, + 1259.0, + 1467.0, + 1259.0, + 1496.0, + 1195.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 1500.0, + 566.0, + 1500.0, + 566.0, + 1538.0, + 421.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1500.0, + 931.0, + 1500.0, + 931.0, + 1536.0, + 768.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1500.0, + 1291.0, + 1500.0, + 1291.0, + 1537.0, + 1120.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 676.5, + 456.0, + 676.5, + 456.0, + 702.0, + 408.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 676.5, + 1366.0, + 676.5, + 1366.0, + 702.0, + 1317.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 715.5, + 597.0, + 715.5, + 597.0, + 727.0, + 568.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.75, + 736.5, + 1203.75, + 736.5, + 1203.75, + 751.0, + 1184.75, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 801.0, + 481.0, + 801.0, + 481.0, + 821.0, + 397.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 996.5, + 976.0, + 996.5, + 976.0, + 1004.5, + 957.0, + 1004.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.25, + 1016.0, + 660.25, + 1016.0, + 660.25, + 1041.0, + 483.25, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 1077.0, + 498.0, + 1077.0, + 498.0, + 1090.0, + 468.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1111.5, + 443.0, + 1111.5, + 443.0, + 1129.0, + 398.0, + 1129.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1341.0, + 560.0, + 1341.0, + 560.0, + 1357.5, + 513.0, + 1357.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1377.5, + 622.0, + 1377.5, + 622.0, + 1398.0, + 600.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1568.0, + 1372.0, + 1568.0, + 1372.0, + 1608.0, + 318.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2121.0, + 829.0, + 2121.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 39, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 320, + 620, + 1378, + 620, + 1378, + 1557, + 320, + 1557 + ], + "score": 0.976 + }, + { + "category_id": 4, + "poly": [ + 297, + 1588, + 1408, + 1588, + 1408, + 1650, + 297, + 1650 + ], + "score": 0.93 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 857, + 75, + 857, + 105, + 297, + 105 + ], + "score": 0.902 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 862, + 2088, + 862, + 2112, + 834, + 2112 + ], + "score": 0.839 + }, + { + "category_id": 15, + "poly": [ + 341.0, + 624.0, + 368.0, + 624.0, + 368.0, + 648.0, + 341.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 637.0, + 605.0, + 637.0, + 605.0, + 647.0, + 596.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 632.0, + 654.0, + 632.0, + 654.0, + 653.0, + 612.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 636.0, + 1113.0, + 636.0, + 1113.0, + 648.0, + 1091.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 632.0, + 1157.0, + 632.0, + 1157.0, + 653.0, + 1115.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 648.0, + 368.0, + 648.0, + 368.0, + 816.0, + 317.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 654.0, + 613.0, + 654.0, + 613.0, + 667.0, + 590.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 648.0, + 654.0, + 648.0, + 654.0, + 673.0, + 614.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 646.0, + 724.0, + 646.0, + 724.0, + 815.0, + 673.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 643.0, + 1079.0, + 643.0, + 1079.0, + 818.0, + 1027.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 648.0, + 1155.0, + 648.0, + 1155.0, + 673.0, + 1091.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 673.0, + 368.0, + 673.0, + 368.0, + 698.0, + 341.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 671.0, + 648.0, + 671.0, + 648.0, + 685.0, + 638.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 678.0, + 1079.0, + 678.0, + 1079.0, + 702.0, + 1053.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 698.0, + 367.0, + 698.0, + 367.0, + 724.0, + 341.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 725.0, + 367.0, + 725.0, + 367.0, + 749.0, + 340.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 713.0, + 722.0, + 713.0, + 722.0, + 737.0, + 697.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 714.0, + 1011.0, + 714.0, + 1011.0, + 767.0, + 994.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 712.0, + 1079.0, + 712.0, + 1079.0, + 736.0, + 1053.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 713.0, + 1363.0, + 713.0, + 1363.0, + 724.0, + 1354.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 749.0, + 367.0, + 749.0, + 367.0, + 773.0, + 340.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 758.0, + 993.0, + 758.0, + 993.0, + 775.0, + 979.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 748.0, + 1079.0, + 748.0, + 1079.0, + 773.0, + 1052.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 775.0, + 364.0, + 775.0, + 364.0, + 794.0, + 350.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 768.0, + 623.0, + 768.0, + 623.0, + 780.0, + 615.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 782.0, + 722.0, + 782.0, + 722.0, + 806.0, + 696.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 773.0, + 985.0, + 773.0, + 985.0, + 784.0, + 977.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 784.0, + 1079.0, + 784.0, + 1079.0, + 806.0, + 1053.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 798.0, + 367.0, + 798.0, + 367.0, + 821.0, + 348.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 799.0, + 802.0, + 799.0, + 802.0, + 820.0, + 758.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 827.0, + 365.0, + 827.0, + 365.0, + 842.0, + 351.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 818.0, + 722.0, + 818.0, + 722.0, + 839.0, + 704.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 815.0, + 800.0, + 815.0, + 800.0, + 840.0, + 733.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 821.0, + 1078.0, + 821.0, + 1078.0, + 838.0, + 1063.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 845.0, + 395.0, + 845.0, + 395.0, + 868.0, + 364.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 844.0, + 451.0, + 844.0, + 451.0, + 870.0, + 416.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 845.0, + 505.0, + 845.0, + 505.0, + 868.0, + 473.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 845.0, + 558.0, + 845.0, + 558.0, + 868.0, + 527.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 845.0, + 614.0, + 845.0, + 614.0, + 868.0, + 580.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 844.0, + 667.0, + 844.0, + 667.0, + 870.0, + 633.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 844.0, + 765.0, + 844.0, + 765.0, + 870.0, + 731.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 796.0, + 845.0, + 828.0, + 845.0, + 828.0, + 868.0, + 796.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 845.0, + 894.0, + 845.0, + 894.0, + 868.0, + 861.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 845.0, + 959.0, + 845.0, + 959.0, + 868.0, + 926.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 845.0, + 1023.0, + 845.0, + 1023.0, + 868.0, + 991.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 845.0, + 1380.0, + 845.0, + 1380.0, + 869.0, + 1083.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 859.0, + 534.0, + 859.0, + 534.0, + 884.0, + 470.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 857.0, + 902.0, + 857.0, + 902.0, + 886.0, + 839.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 858.0, + 1263.0, + 858.0, + 1263.0, + 886.0, + 1199.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 882.0, + 562.0, + 882.0, + 562.0, + 918.0, + 424.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 882.0, + 926.0, + 882.0, + 926.0, + 918.0, + 773.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 881.0, + 1282.0, + 881.0, + 1282.0, + 918.0, + 1130.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 941.0, + 368.0, + 941.0, + 368.0, + 964.0, + 341.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 956.0, + 648.0, + 956.0, + 648.0, + 976.0, + 605.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 951.0, + 1007.0, + 951.0, + 1007.0, + 974.0, + 963.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 955.0, + 1366.0, + 955.0, + 1366.0, + 976.0, + 1324.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 972.0, + 368.0, + 972.0, + 368.0, + 1135.0, + 321.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 972.0, + 646.0, + 972.0, + 646.0, + 997.0, + 608.0, + 997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 965.0, + 720.0, + 965.0, + 720.0, + 1134.0, + 673.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 967.0, + 1005.0, + 967.0, + 1005.0, + 994.0, + 965.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 970.0, + 1081.0, + 970.0, + 1081.0, + 1136.0, + 1030.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 971.0, + 1364.0, + 971.0, + 1364.0, + 996.0, + 1324.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 996.0, + 720.0, + 996.0, + 720.0, + 1016.0, + 697.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1006.0, + 1082.0, + 1006.0, + 1082.0, + 1030.0, + 1055.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1019.0, + 367.0, + 1019.0, + 367.0, + 1042.0, + 340.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1025.0, + 721.0, + 1025.0, + 721.0, + 1049.0, + 696.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 1039.0, + 435.0, + 1039.0, + 435.0, + 1063.0, + 426.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1044.0, + 1081.0, + 1044.0, + 1081.0, + 1068.0, + 1054.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1057.0, + 367.0, + 1057.0, + 367.0, + 1081.0, + 341.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1054.0, + 721.0, + 1054.0, + 721.0, + 1080.0, + 696.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 1073.0, + 974.0, + 1073.0, + 974.0, + 1079.0, + 968.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1085.0, + 720.0, + 1085.0, + 720.0, + 1109.0, + 695.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1080.0, + 1082.0, + 1080.0, + 1082.0, + 1104.0, + 1054.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1097.0, + 368.0, + 1097.0, + 368.0, + 1120.0, + 341.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 1097.0, + 633.0, + 1097.0, + 633.0, + 1108.0, + 625.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1114.0, + 721.0, + 1114.0, + 721.0, + 1138.0, + 695.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1120.0, + 1079.0, + 1120.0, + 1079.0, + 1140.0, + 1056.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1138.0, + 367.0, + 1138.0, + 367.0, + 1159.0, + 347.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1139.0, + 1002.0, + 1139.0, + 1002.0, + 1145.0, + 996.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 1149.0, + 645.0, + 1149.0, + 645.0, + 1161.0, + 619.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1147.0, + 720.0, + 1147.0, + 720.0, + 1167.0, + 696.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1163.0, + 393.0, + 1163.0, + 393.0, + 1186.0, + 363.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1164.0, + 449.0, + 1164.0, + 449.0, + 1187.0, + 417.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 1163.0, + 558.0, + 1163.0, + 558.0, + 1205.0, + 471.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1164.0, + 613.0, + 1164.0, + 613.0, + 1187.0, + 581.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1163.0, + 667.0, + 1163.0, + 667.0, + 1186.0, + 637.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1161.0, + 1021.0, + 1161.0, + 1021.0, + 1203.0, + 711.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 1165.0, + 1128.0, + 1165.0, + 1128.0, + 1186.0, + 1094.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1164.0, + 1318.0, + 1164.0, + 1318.0, + 1203.0, + 1130.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 1165.0, + 1356.0, + 1165.0, + 1356.0, + 1187.0, + 1323.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1200.0, + 564.0, + 1200.0, + 564.0, + 1236.0, + 423.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1199.0, + 920.0, + 1199.0, + 920.0, + 1237.0, + 780.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1199.0, + 1274.0, + 1199.0, + 1274.0, + 1236.0, + 1137.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1267.0, + 370.0, + 1267.0, + 370.0, + 1291.0, + 341.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 1275.0, + 446.0, + 1275.0, + 446.0, + 1297.0, + 387.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1265.0, + 724.0, + 1265.0, + 724.0, + 1289.0, + 696.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1272.0, + 1006.0, + 1272.0, + 1006.0, + 1294.0, + 963.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1263.0, + 1079.0, + 1263.0, + 1079.0, + 1288.0, + 1052.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 1267.0, + 1368.0, + 1267.0, + 1368.0, + 1293.0, + 1324.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1294.0, + 348.0, + 1294.0, + 348.0, + 1452.0, + 321.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1291.0, + 444.0, + 1291.0, + 444.0, + 1314.0, + 376.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 1288.0, + 722.0, + 1288.0, + 722.0, + 1453.0, + 673.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1290.0, + 1005.0, + 1290.0, + 1005.0, + 1312.0, + 942.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 1287.0, + 1060.0, + 1287.0, + 1060.0, + 1452.0, + 1029.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 1285.0, + 1364.0, + 1285.0, + 1364.0, + 1311.0, + 1323.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1299.0, + 1079.0, + 1299.0, + 1079.0, + 1325.0, + 1053.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1314.0, + 368.0, + 1314.0, + 368.0, + 1335.0, + 345.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1313.0, + 722.0, + 1313.0, + 722.0, + 1339.0, + 697.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1339.0, + 722.0, + 1339.0, + 722.0, + 1363.0, + 697.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1336.0, + 1079.0, + 1336.0, + 1079.0, + 1360.0, + 1053.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1355.0, + 369.0, + 1355.0, + 369.0, + 1379.0, + 344.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1362.0, + 724.0, + 1362.0, + 724.0, + 1393.0, + 696.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1371.0, + 1078.0, + 1371.0, + 1078.0, + 1397.0, + 1053.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1386.0, + 722.0, + 1386.0, + 722.0, + 1413.0, + 696.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1401.0, + 369.0, + 1401.0, + 369.0, + 1425.0, + 344.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 1401.0, + 627.0, + 1401.0, + 627.0, + 1420.0, + 615.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 1404.0, + 1338.0, + 1404.0, + 1338.0, + 1410.0, + 1332.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1410.0, + 724.0, + 1410.0, + 724.0, + 1437.0, + 697.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1408.0, + 1078.0, + 1408.0, + 1078.0, + 1433.0, + 1053.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1410.0, + 1116.0, + 1410.0, + 1116.0, + 1462.0, + 1097.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.0, + 1413.0, + 1342.0, + 1413.0, + 1342.0, + 1433.0, + 1329.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1445.0, + 370.0, + 1445.0, + 370.0, + 1469.0, + 342.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 1457.0, + 389.0, + 1457.0, + 389.0, + 1468.0, + 380.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1438.0, + 724.0, + 1438.0, + 724.0, + 1462.0, + 697.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1445.0, + 1079.0, + 1445.0, + 1079.0, + 1468.0, + 1052.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1462.0, + 724.0, + 1462.0, + 724.0, + 1486.0, + 696.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 1482.0, + 397.0, + 1482.0, + 397.0, + 1505.0, + 365.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1481.0, + 452.0, + 1481.0, + 452.0, + 1507.0, + 417.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 1482.0, + 506.0, + 1482.0, + 506.0, + 1505.0, + 473.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1478.0, + 562.0, + 1478.0, + 562.0, + 1508.0, + 527.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1481.0, + 616.0, + 1481.0, + 616.0, + 1507.0, + 581.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1480.0, + 669.0, + 1480.0, + 669.0, + 1506.0, + 634.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 1480.0, + 1025.0, + 1480.0, + 1025.0, + 1507.0, + 725.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1479.0, + 1202.0, + 1479.0, + 1202.0, + 1506.0, + 1089.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1480.0, + 1243.0, + 1480.0, + 1243.0, + 1505.0, + 1205.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1480.0, + 1322.0, + 1480.0, + 1322.0, + 1505.0, + 1246.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 1481.0, + 1357.0, + 1481.0, + 1357.0, + 1504.0, + 1323.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1495.0, + 543.0, + 1495.0, + 543.0, + 1520.0, + 478.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1494.0, + 896.0, + 1494.0, + 896.0, + 1522.0, + 832.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1493.0, + 1254.0, + 1493.0, + 1254.0, + 1522.0, + 1192.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1520.0, + 562.0, + 1520.0, + 562.0, + 1554.0, + 424.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1517.0, + 920.0, + 1517.0, + 920.0, + 1554.0, + 780.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 1518.0, + 1272.0, + 1518.0, + 1272.0, + 1555.0, + 1138.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 628.0, + 1006.0, + 628.0, + 1006.0, + 643.0, + 950.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 677.0, + 726.0, + 677.0, + 726.0, + 704.0, + 693.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 663.5, + 1229.0, + 663.5, + 1229.0, + 671.5, + 1202.0, + 671.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 757.5, + 540.0, + 757.5, + 540.0, + 765.5, + 521.0, + 765.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1177.0, + 696.5, + 1218.0, + 696.5, + 1218.0, + 703.5, + 1177.0, + 703.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.25, + 747.5, + 725.25, + 747.5, + 725.25, + 771.5, + 691.25, + 771.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 738.0, + 1364.0, + 738.0, + 1364.0, + 743.0, + 1339.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.25, + 978.0, + 928.25, + 978.0, + 928.25, + 993.0, + 899.25, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 983.5, + 844.0, + 983.5, + 844.0, + 994.0, + 795.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.25, + 1034.0, + 961.25, + 1034.0, + 961.25, + 1047.0, + 952.25, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1050.5, + 611.0, + 1050.5, + 611.0, + 1057.0, + 596.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.75, + 1365.5, + 976.75, + 1365.5, + 976.75, + 1375.5, + 969.75, + 1375.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 1363.5, + 1324.0, + 1363.5, + 1324.0, + 1375.0, + 1316.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1370.0, + 799.0, + 1370.0, + 799.0, + 1386.5, + 744.0, + 1386.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 1374.0, + 1150.0, + 1374.0, + 1150.0, + 1389.0, + 1083.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1417.5, + 1007.0, + 1417.5, + 1007.0, + 1439.0, + 987.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1582.0, + 1406.0, + 1582.0, + 1406.0, + 1628.0, + 291.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1614.0, + 387.0, + 1614.0, + 387.0, + 1652.0, + 291.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 858.0, + 73.0, + 858.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 867.0, + 2084.0, + 867.0, + 2123.0, + 829.0, + 2123.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 40, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 321, + 626, + 1376, + 626, + 1376, + 1554, + 321, + 1554 + ], + "score": 0.976 + }, + { + "category_id": 4, + "poly": [ + 292, + 1582, + 1404, + 1582, + 1404, + 1646, + 292, + 1646 + ], + "score": 0.935 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 858, + 75, + 858, + 106, + 297, + 106 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2113, + 834, + 2113 + ], + "score": 0.851 + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 626.0, + 1104.0, + 626.0, + 1104.0, + 837.0, + 1024.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 627.0, + 733.0, + 627.0, + 733.0, + 655.0, + 702.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 634.0, + 448.0, + 634.0, + 448.0, + 656.0, + 403.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 635.0, + 406.0, + 635.0, + 406.0, + 653.0, + 377.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 635.0, + 650.0, + 635.0, + 650.0, + 646.0, + 640.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 635.0, + 1159.0, + 635.0, + 1159.0, + 656.0, + 1116.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 640.0, + 370.0, + 640.0, + 370.0, + 669.0, + 340.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 641.0, + 807.0, + 641.0, + 807.0, + 662.0, + 764.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 647.0, + 759.0, + 647.0, + 759.0, + 656.0, + 745.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 650.0, + 447.0, + 650.0, + 447.0, + 676.0, + 403.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 651.0, + 1159.0, + 651.0, + 1159.0, + 677.0, + 1117.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 655.0, + 350.0, + 655.0, + 350.0, + 818.0, + 319.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 655.0, + 407.0, + 655.0, + 407.0, + 671.0, + 377.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 656.0, + 775.0, + 656.0, + 775.0, + 680.0, + 737.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 656.0, + 1117.0, + 656.0, + 1117.0, + 669.0, + 1093.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 659.0, + 713.0, + 659.0, + 713.0, + 819.0, + 681.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 659.0, + 731.0, + 659.0, + 731.0, + 683.0, + 703.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 659.0, + 805.0, + 659.0, + 805.0, + 681.0, + 765.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 666.0, + 370.0, + 666.0, + 370.0, + 697.0, + 340.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 697.0, + 370.0, + 697.0, + 370.0, + 726.0, + 340.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 708.0, + 554.0, + 708.0, + 554.0, + 719.0, + 543.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 688.0, + 729.0, + 688.0, + 729.0, + 713.0, + 704.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 694.0, + 1081.0, + 694.0, + 1081.0, + 719.0, + 1053.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 727.0, + 368.0, + 727.0, + 368.0, + 752.0, + 341.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 733.0, + 519.0, + 733.0, + 519.0, + 743.0, + 509.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 719.0, + 731.0, + 719.0, + 731.0, + 743.0, + 704.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 732.0, + 1082.0, + 732.0, + 1082.0, + 759.0, + 1051.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 751.0, + 372.0, + 751.0, + 372.0, + 785.0, + 342.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 751.0, + 728.0, + 751.0, + 728.0, + 771.0, + 705.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 783.0, + 370.0, + 783.0, + 370.0, + 812.0, + 340.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 778.0, + 729.0, + 778.0, + 729.0, + 803.0, + 704.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 772.0, + 1082.0, + 772.0, + 1082.0, + 800.0, + 1051.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 812.0, + 371.0, + 812.0, + 371.0, + 839.0, + 340.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 808.0, + 731.0, + 808.0, + 731.0, + 832.0, + 704.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 813.0, + 1083.0, + 813.0, + 1083.0, + 840.0, + 1052.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 846.0, + 460.0, + 846.0, + 460.0, + 869.0, + 395.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 847.0, + 494.0, + 847.0, + 494.0, + 866.0, + 473.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 847.0, + 531.0, + 847.0, + 531.0, + 866.0, + 512.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 847.0, + 571.0, + 847.0, + 571.0, + 866.0, + 550.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 847.0, + 608.0, + 847.0, + 608.0, + 868.0, + 589.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 847.0, + 648.0, + 847.0, + 648.0, + 868.0, + 627.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 847.0, + 792.0, + 847.0, + 792.0, + 870.0, + 764.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 850.0, + 834.0, + 850.0, + 834.0, + 869.0, + 815.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 851.0, + 881.0, + 851.0, + 881.0, + 866.0, + 864.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 848.0, + 930.0, + 848.0, + 930.0, + 869.0, + 911.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 848.0, + 980.0, + 848.0, + 980.0, + 869.0, + 959.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 847.0, + 1124.0, + 847.0, + 1124.0, + 870.0, + 1098.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 848.0, + 1173.0, + 848.0, + 1173.0, + 869.0, + 1153.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 839.0, + 1280.0, + 839.0, + 1280.0, + 890.0, + 1185.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 848.0, + 1333.0, + 848.0, + 1333.0, + 869.0, + 1314.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 857.0, + 543.0, + 857.0, + 543.0, + 885.0, + 478.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 857.0, + 894.0, + 857.0, + 894.0, + 885.0, + 832.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 881.0, + 551.0, + 881.0, + 551.0, + 925.0, + 435.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 879.0, + 918.0, + 879.0, + 918.0, + 926.0, + 780.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 881.0, + 1279.0, + 881.0, + 1279.0, + 923.0, + 1132.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 953.0, + 372.0, + 953.0, + 372.0, + 981.0, + 341.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 952.0, + 450.0, + 952.0, + 450.0, + 977.0, + 378.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 947.0, + 804.0, + 947.0, + 804.0, + 973.0, + 757.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 951.0, + 1006.0, + 951.0, + 1006.0, + 962.0, + 995.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 954.0, + 1101.0, + 954.0, + 1101.0, + 965.0, + 1092.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 955.0, + 1367.0, + 955.0, + 1367.0, + 976.0, + 1321.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 972.0, + 352.0, + 972.0, + 352.0, + 1135.0, + 320.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 970.0, + 447.0, + 970.0, + 447.0, + 996.0, + 378.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 964.0, + 726.0, + 964.0, + 726.0, + 991.0, + 695.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 972.0, + 755.0, + 972.0, + 755.0, + 981.0, + 740.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 965.0, + 803.0, + 965.0, + 803.0, + 991.0, + 759.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 963.0, + 1083.0, + 963.0, + 1083.0, + 1134.0, + 1034.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 971.0, + 1363.0, + 971.0, + 1363.0, + 996.0, + 1323.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 985.0, + 371.0, + 985.0, + 371.0, + 1045.0, + 340.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 986.0, + 723.0, + 986.0, + 723.0, + 1142.0, + 675.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1003.0, + 1083.0, + 1003.0, + 1083.0, + 1028.0, + 1057.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 1031.0, + 1208.0, + 1031.0, + 1208.0, + 1050.0, + 1185.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1014.0, + 1361.0, + 1014.0, + 1361.0, + 1023.0, + 1348.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1050.0, + 370.0, + 1050.0, + 370.0, + 1075.0, + 342.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1044.0, + 723.0, + 1044.0, + 723.0, + 1067.0, + 696.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1046.0, + 1083.0, + 1046.0, + 1083.0, + 1070.0, + 1057.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1082.0, + 368.0, + 1082.0, + 368.0, + 1105.0, + 342.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1084.0, + 723.0, + 1084.0, + 723.0, + 1108.0, + 696.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 796.0, + 1091.0, + 811.0, + 1091.0, + 811.0, + 1108.0, + 796.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1086.0, + 1083.0, + 1086.0, + 1083.0, + 1111.0, + 1057.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1110.0, + 372.0, + 1110.0, + 372.0, + 1138.0, + 340.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1122.0, + 726.0, + 1122.0, + 726.0, + 1146.0, + 697.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1129.0, + 1084.0, + 1129.0, + 1084.0, + 1153.0, + 1055.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1144.0, + 371.0, + 1144.0, + 371.0, + 1167.0, + 342.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1164.0, + 402.0, + 1164.0, + 402.0, + 1186.0, + 376.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 1164.0, + 445.0, + 1164.0, + 445.0, + 1186.0, + 414.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 1157.0, + 569.0, + 1157.0, + 569.0, + 1209.0, + 473.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1165.0, + 613.0, + 1165.0, + 613.0, + 1186.0, + 592.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 1165.0, + 656.0, + 1165.0, + 656.0, + 1185.0, + 636.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1163.0, + 772.0, + 1163.0, + 772.0, + 1182.0, + 747.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 1163.0, + 821.0, + 1163.0, + 821.0, + 1183.0, + 800.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1164.0, + 871.0, + 1164.0, + 871.0, + 1180.0, + 856.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 1163.0, + 928.0, + 1163.0, + 928.0, + 1183.0, + 906.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 1163.0, + 982.0, + 1163.0, + 982.0, + 1185.0, + 959.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1164.0, + 1126.0, + 1164.0, + 1126.0, + 1186.0, + 1098.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1164.0, + 1176.0, + 1164.0, + 1176.0, + 1186.0, + 1153.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1158.0, + 1252.0, + 1158.0, + 1252.0, + 1204.0, + 1180.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 1165.0, + 1280.0, + 1165.0, + 1280.0, + 1185.0, + 1260.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 1164.0, + 1333.0, + 1164.0, + 1333.0, + 1185.0, + 1313.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1174.0, + 893.0, + 1174.0, + 893.0, + 1203.0, + 832.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1197.0, + 564.0, + 1197.0, + 564.0, + 1237.0, + 423.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 1197.0, + 927.0, + 1197.0, + 927.0, + 1237.0, + 772.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1197.0, + 1289.0, + 1197.0, + 1289.0, + 1239.0, + 1124.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1261.0, + 373.0, + 1261.0, + 373.0, + 1451.0, + 322.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 1268.0, + 449.0, + 1268.0, + 449.0, + 1310.0, + 378.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1286.0, + 725.0, + 1286.0, + 725.0, + 1310.0, + 697.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1267.0, + 804.0, + 1267.0, + 804.0, + 1310.0, + 759.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1278.0, + 1086.0, + 1278.0, + 1086.0, + 1449.0, + 1036.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1268.0, + 1369.0, + 1268.0, + 1369.0, + 1311.0, + 1301.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 1303.0, + 703.0, + 1303.0, + 703.0, + 1453.0, + 676.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1303.0, + 1086.0, + 1303.0, + 1086.0, + 1329.0, + 1059.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1333.0, + 371.0, + 1333.0, + 371.0, + 1358.0, + 343.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1324.0, + 723.0, + 1324.0, + 723.0, + 1351.0, + 697.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1332.0, + 1086.0, + 1332.0, + 1086.0, + 1355.0, + 1059.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 1343.0, + 1332.0, + 1343.0, + 1332.0, + 1352.0, + 1318.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1354.0, + 553.0, + 1354.0, + 553.0, + 1362.0, + 543.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1368.0, + 371.0, + 1368.0, + 371.0, + 1393.0, + 343.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1365.0, + 723.0, + 1365.0, + 723.0, + 1391.0, + 697.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1359.0, + 1086.0, + 1359.0, + 1086.0, + 1382.0, + 1059.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1385.0, + 1086.0, + 1385.0, + 1086.0, + 1409.0, + 1059.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1402.0, + 371.0, + 1402.0, + 371.0, + 1430.0, + 341.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1405.0, + 723.0, + 1405.0, + 723.0, + 1429.0, + 697.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1412.0, + 1086.0, + 1412.0, + 1086.0, + 1435.0, + 1059.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1439.0, + 371.0, + 1439.0, + 371.0, + 1464.0, + 343.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1445.0, + 725.0, + 1445.0, + 725.0, + 1468.0, + 697.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1439.0, + 1086.0, + 1439.0, + 1086.0, + 1462.0, + 1059.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1480.0, + 426.0, + 1480.0, + 426.0, + 1503.0, + 397.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1481.0, + 471.0, + 1481.0, + 471.0, + 1503.0, + 450.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1473.0, + 549.0, + 1473.0, + 549.0, + 1522.0, + 476.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1480.0, + 573.0, + 1480.0, + 573.0, + 1502.0, + 553.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 1480.0, + 622.0, + 1480.0, + 622.0, + 1502.0, + 603.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1481.0, + 773.0, + 1481.0, + 773.0, + 1504.0, + 745.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 1481.0, + 821.0, + 1481.0, + 821.0, + 1503.0, + 800.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 1473.0, + 897.0, + 1473.0, + 897.0, + 1522.0, + 827.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1479.0, + 928.0, + 1479.0, + 928.0, + 1505.0, + 904.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 1481.0, + 978.0, + 1481.0, + 978.0, + 1503.0, + 959.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1464.0, + 1088.0, + 1464.0, + 1088.0, + 1491.0, + 1057.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 1478.0, + 1130.0, + 1478.0, + 1130.0, + 1502.0, + 1101.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 1480.0, + 1178.0, + 1480.0, + 1178.0, + 1500.0, + 1156.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 1472.0, + 1283.0, + 1472.0, + 1283.0, + 1521.0, + 1200.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 1479.0, + 1335.0, + 1479.0, + 1335.0, + 1500.0, + 1315.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1513.0, + 567.0, + 1513.0, + 567.0, + 1550.0, + 418.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1511.0, + 933.0, + 1511.0, + 933.0, + 1550.0, + 766.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1513.0, + 1292.0, + 1513.0, + 1292.0, + 1549.0, + 1120.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 639.0, + 1113.0, + 639.0, + 1113.0, + 652.0, + 1092.0, + 652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.75, + 660.5, + 624.75, + 660.5, + 624.75, + 669.0, + 601.75, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.25, + 684.0, + 592.25, + 684.0, + 592.25, + 692.5, + 569.25, + 692.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.75, + 965.0, + 660.75, + 965.0, + 660.75, + 982.0, + 594.75, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 989.5, + 988.0, + 989.5, + 988.0, + 1005.0, + 910.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.25, + 1023.5, + 924.25, + 1023.5, + 924.25, + 1037.0, + 886.25, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.25, + 1048.5, + 878.25, + 1048.5, + 878.25, + 1059.0, + 857.25, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1065.5, + 525.0, + 1065.5, + 525.0, + 1082.5, + 453.0, + 1082.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.75, + 1065.5, + 871.75, + 1065.5, + 871.75, + 1083.5, + 797.75, + 1083.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 1164.5, + 480.0, + 1164.5, + 480.0, + 1186.0, + 465.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1297.0, + 369.0, + 1297.0, + 369.0, + 1324.5, + 344.0, + 1324.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1288.0, + 670.0, + 1288.0, + 670.0, + 1311.0, + 580.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1578.0, + 1405.0, + 1578.0, + 1405.0, + 1620.0, + 291.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1610.0, + 387.0, + 1610.0, + 387.0, + 1649.0, + 293.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2122.0, + 829.0, + 2122.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 41, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 324, + 671, + 1373, + 671, + 1373, + 1534, + 324, + 1534 + ], + "score": 0.976 + }, + { + "category_id": 4, + "poly": [ + 413, + 1568, + 1279, + 1568, + 1279, + 1602, + 413, + 1602 + ], + "score": 0.913 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 858, + 75, + 858, + 105, + 297, + 105 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 864, + 2088, + 864, + 2112, + 834, + 2112 + ], + "score": 0.847 + }, + { + "category_id": 15, + "poly": [ + 345.0, + 672.0, + 371.0, + 672.0, + 371.0, + 694.0, + 345.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 672.0, + 728.0, + 672.0, + 728.0, + 698.0, + 701.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 672.0, + 1083.0, + 672.0, + 1083.0, + 696.0, + 1058.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 683.0, + 372.0, + 683.0, + 372.0, + 850.0, + 322.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 684.0, + 729.0, + 684.0, + 729.0, + 850.0, + 681.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 686.0, + 1007.0, + 686.0, + 1007.0, + 724.0, + 991.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 687.0, + 1083.0, + 687.0, + 1083.0, + 773.0, + 1041.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 715.0, + 371.0, + 715.0, + 371.0, + 741.0, + 345.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 712.0, + 729.0, + 712.0, + 729.0, + 744.0, + 701.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 716.0, + 1084.0, + 716.0, + 1084.0, + 741.0, + 1057.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 734.0, + 372.0, + 734.0, + 372.0, + 764.0, + 343.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 736.0, + 660.0, + 736.0, + 660.0, + 839.0, + 624.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 738.0, + 728.0, + 738.0, + 728.0, + 762.0, + 701.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 738.0, + 1084.0, + 738.0, + 1084.0, + 762.0, + 1057.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 738.0, + 1348.0, + 738.0, + 1348.0, + 750.0, + 1338.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 762.0, + 371.0, + 762.0, + 371.0, + 786.0, + 345.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 762.0, + 726.0, + 762.0, + 726.0, + 786.0, + 701.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 758.0, + 1083.0, + 758.0, + 1083.0, + 850.0, + 1035.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 746.0, + 1348.0, + 746.0, + 1348.0, + 803.0, + 1332.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 784.0, + 371.0, + 784.0, + 371.0, + 809.0, + 346.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 785.0, + 728.0, + 785.0, + 728.0, + 809.0, + 702.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 785.0, + 1084.0, + 785.0, + 1084.0, + 809.0, + 1057.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 792.0, + 1329.0, + 792.0, + 1329.0, + 818.0, + 1316.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 810.0, + 369.0, + 810.0, + 369.0, + 827.0, + 354.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 811.0, + 453.0, + 811.0, + 453.0, + 833.0, + 409.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 808.0, + 728.0, + 808.0, + 728.0, + 829.0, + 708.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 810.0, + 810.0, + 810.0, + 810.0, + 835.0, + 766.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 810.0, + 1082.0, + 810.0, + 1082.0, + 827.0, + 1066.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 810.0, + 1167.0, + 810.0, + 1167.0, + 835.0, + 1123.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 816.0, + 1319.0, + 816.0, + 1319.0, + 826.0, + 1310.0, + 826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 830.0, + 371.0, + 830.0, + 371.0, + 852.0, + 353.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 835.0, + 404.0, + 835.0, + 404.0, + 844.0, + 388.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 829.0, + 452.0, + 829.0, + 452.0, + 855.0, + 409.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 828.0, + 728.0, + 828.0, + 728.0, + 852.0, + 708.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 829.0, + 807.0, + 829.0, + 807.0, + 855.0, + 766.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 828.0, + 1083.0, + 828.0, + 1083.0, + 852.0, + 1065.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 830.0, + 1161.0, + 830.0, + 1161.0, + 852.0, + 1124.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 857.0, + 396.0, + 857.0, + 396.0, + 883.0, + 363.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 857.0, + 451.0, + 857.0, + 451.0, + 883.0, + 415.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 858.0, + 504.0, + 858.0, + 504.0, + 881.0, + 469.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 858.0, + 556.0, + 858.0, + 556.0, + 881.0, + 522.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 858.0, + 609.0, + 858.0, + 609.0, + 881.0, + 577.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 858.0, + 662.0, + 858.0, + 662.0, + 881.0, + 629.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 857.0, + 779.0, + 857.0, + 779.0, + 883.0, + 743.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 857.0, + 839.0, + 857.0, + 839.0, + 883.0, + 804.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 857.0, + 899.0, + 857.0, + 899.0, + 883.0, + 864.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 857.0, + 960.0, + 857.0, + 960.0, + 883.0, + 925.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 857.0, + 1020.0, + 857.0, + 1020.0, + 883.0, + 985.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 857.0, + 1374.0, + 857.0, + 1374.0, + 883.0, + 1084.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 873.0, + 546.0, + 873.0, + 546.0, + 901.0, + 480.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 837.0, + 870.0, + 902.0, + 870.0, + 902.0, + 903.0, + 837.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 873.0, + 1257.0, + 873.0, + 1257.0, + 899.0, + 1193.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 904.0, + 563.0, + 904.0, + 563.0, + 942.0, + 423.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 905.0, + 926.0, + 905.0, + 926.0, + 940.0, + 773.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 905.0, + 1281.0, + 905.0, + 1281.0, + 941.0, + 1130.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 966.0, + 371.0, + 966.0, + 371.0, + 1148.0, + 324.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 983.0, + 728.0, + 983.0, + 728.0, + 1070.0, + 683.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 971.0, + 1099.0, + 971.0, + 1099.0, + 1159.0, + 1023.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1016.0, + 1084.0, + 1016.0, + 1084.0, + 1041.0, + 1057.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1017.0, + 1161.0, + 1017.0, + 1161.0, + 1028.0, + 1153.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1024.0, + 370.0, + 1024.0, + 370.0, + 1051.0, + 345.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1022.0, + 728.0, + 1022.0, + 728.0, + 1047.0, + 702.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1053.0, + 370.0, + 1053.0, + 370.0, + 1080.0, + 345.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1044.0, + 728.0, + 1044.0, + 728.0, + 1146.0, + 681.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1041.0, + 1084.0, + 1041.0, + 1084.0, + 1093.0, + 1056.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1087.0, + 369.0, + 1087.0, + 369.0, + 1107.0, + 346.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1086.0, + 728.0, + 1086.0, + 728.0, + 1111.0, + 702.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1094.0, + 1084.0, + 1094.0, + 1084.0, + 1119.0, + 1057.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1117.0, + 369.0, + 1117.0, + 369.0, + 1135.0, + 354.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 1134.0, + 497.0, + 1134.0, + 497.0, + 1142.0, + 487.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 1109.0, + 549.0, + 1109.0, + 549.0, + 1153.0, + 504.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1106.0, + 726.0, + 1106.0, + 726.0, + 1133.0, + 702.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 1109.0, + 910.0, + 1109.0, + 910.0, + 1153.0, + 839.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1106.0, + 1106.0, + 1106.0, + 1106.0, + 1122.0, + 1090.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1109.0, + 1261.0, + 1109.0, + 1261.0, + 1153.0, + 1191.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1147.0, + 396.0, + 1147.0, + 396.0, + 1180.0, + 355.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1155.0, + 451.0, + 1155.0, + 451.0, + 1182.0, + 415.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 1154.0, + 505.0, + 1154.0, + 505.0, + 1181.0, + 468.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1154.0, + 559.0, + 1154.0, + 559.0, + 1182.0, + 521.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1155.0, + 612.0, + 1155.0, + 612.0, + 1182.0, + 576.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1155.0, + 664.0, + 1155.0, + 664.0, + 1182.0, + 628.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1154.0, + 786.0, + 1154.0, + 786.0, + 1181.0, + 750.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1154.0, + 848.0, + 1154.0, + 848.0, + 1181.0, + 813.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1153.0, + 912.0, + 1153.0, + 912.0, + 1181.0, + 875.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1154.0, + 973.0, + 1154.0, + 973.0, + 1182.0, + 938.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1157.0, + 1373.0, + 1157.0, + 1373.0, + 1180.0, + 1093.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1170.0, + 544.0, + 1170.0, + 544.0, + 1199.0, + 479.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1166.0, + 905.0, + 1166.0, + 905.0, + 1200.0, + 836.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1171.0, + 1255.0, + 1171.0, + 1255.0, + 1199.0, + 1192.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1204.0, + 563.0, + 1204.0, + 563.0, + 1239.0, + 423.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 1205.0, + 917.0, + 1205.0, + 917.0, + 1238.0, + 781.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1202.0, + 1273.0, + 1202.0, + 1273.0, + 1241.0, + 1139.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1270.0, + 371.0, + 1270.0, + 371.0, + 1443.0, + 324.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1273.0, + 738.0, + 1273.0, + 738.0, + 1444.0, + 681.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1280.0, + 1006.0, + 1280.0, + 1006.0, + 1322.0, + 964.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1266.0, + 1101.0, + 1266.0, + 1101.0, + 1462.0, + 1024.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 1293.0, + 1356.0, + 1293.0, + 1356.0, + 1318.0, + 1315.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1317.0, + 738.0, + 1317.0, + 738.0, + 1344.0, + 700.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1320.0, + 1083.0, + 1320.0, + 1083.0, + 1344.0, + 1056.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1346.0, + 370.0, + 1346.0, + 370.0, + 1370.0, + 345.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1341.0, + 738.0, + 1341.0, + 738.0, + 1366.0, + 699.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 1347.0, + 1083.0, + 1347.0, + 1083.0, + 1371.0, + 1058.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 1363.0, + 738.0, + 1363.0, + 738.0, + 1389.0, + 699.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1371.0, + 1083.0, + 1371.0, + 1083.0, + 1395.0, + 1057.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1384.0, + 370.0, + 1384.0, + 370.0, + 1446.0, + 341.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 1390.0, + 411.0, + 1390.0, + 411.0, + 1457.0, + 387.0, + 1457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 1405.0, + 550.0, + 1405.0, + 550.0, + 1447.0, + 480.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1387.0, + 740.0, + 1387.0, + 740.0, + 1459.0, + 700.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1397.0, + 1083.0, + 1397.0, + 1083.0, + 1421.0, + 1056.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1449.0, + 397.0, + 1449.0, + 397.0, + 1476.0, + 364.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1449.0, + 451.0, + 1449.0, + 451.0, + 1476.0, + 415.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1449.0, + 505.0, + 1449.0, + 505.0, + 1476.0, + 469.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1449.0, + 558.0, + 1449.0, + 558.0, + 1476.0, + 521.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1449.0, + 610.0, + 1449.0, + 610.0, + 1476.0, + 576.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1449.0, + 663.0, + 1449.0, + 663.0, + 1476.0, + 628.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 1449.0, + 795.0, + 1449.0, + 795.0, + 1477.0, + 760.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1450.0, + 854.0, + 1450.0, + 854.0, + 1477.0, + 820.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 1449.0, + 916.0, + 1449.0, + 916.0, + 1477.0, + 880.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1450.0, + 975.0, + 1450.0, + 975.0, + 1477.0, + 942.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 1449.0, + 1373.0, + 1449.0, + 1373.0, + 1476.0, + 1091.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1466.0, + 544.0, + 1466.0, + 544.0, + 1494.0, + 479.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 1466.0, + 906.0, + 1466.0, + 906.0, + 1494.0, + 844.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1466.0, + 1255.0, + 1466.0, + 1255.0, + 1494.0, + 1192.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1499.0, + 561.0, + 1499.0, + 561.0, + 1533.0, + 424.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1499.0, + 920.0, + 1499.0, + 920.0, + 1534.0, + 780.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1498.0, + 1271.0, + 1498.0, + 1271.0, + 1533.0, + 1139.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 752.5, + 1226.0, + 752.5, + 1226.0, + 763.5, + 1159.0, + 763.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 724.5, + 492.0, + 724.5, + 492.0, + 734.0, + 485.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 726.5, + 830.0, + 726.5, + 830.0, + 733.5, + 826.0, + 733.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 785.5, + 549.0, + 785.5, + 549.0, + 799.5, + 507.0, + 799.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 787.5, + 887.0, + 787.5, + 887.0, + 797.5, + 847.0, + 797.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 783.5, + 989.0, + 783.5, + 989.0, + 789.5, + 958.0, + 789.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.75, + 1043.0, + 1332.75, + 1043.0, + 1332.75, + 1069.5, + 1273.75, + 1069.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1019.0, + 946.0, + 1019.0, + 946.0, + 1036.5, + 926.0, + 1036.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1071.0, + 634.0, + 1071.0, + 634.0, + 1086.0, + 600.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.75, + 1056.0, + 1164.75, + 1056.0, + 1164.75, + 1068.5, + 1092.75, + 1068.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.75, + 1125.5, + 724.75, + 1125.5, + 724.75, + 1150.5, + 702.75, + 1150.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.25, + 1052.5, + 829.25, + 1052.5, + 829.25, + 1069.5, + 720.25, + 1069.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1126.0, + 1361.0, + 1126.0, + 1361.0, + 1155.0, + 1334.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1055.5, + 502.0, + 1055.5, + 502.0, + 1080.0, + 352.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1353.0, + 994.0, + 1353.0, + 994.0, + 1378.0, + 917.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 1274.0, + 1361.0, + 1274.0, + 1361.0, + 1299.5, + 1313.0, + 1299.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1294.5, + 1087.0, + 1294.5, + 1087.0, + 1320.0, + 1053.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1347.5, + 1330.0, + 1347.5, + 1330.0, + 1371.5, + 1303.0, + 1371.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 1297.0, + 1317.0, + 1297.0, + 1317.0, + 1309.0, + 1293.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.25, + 1373.5, + 631.25, + 1373.5, + 631.25, + 1391.0, + 605.25, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1349.5, + 460.0, + 1349.5, + 460.0, + 1361.0, + 393.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1366.5, + 798.0, + 1366.5, + 798.0, + 1383.0, + 724.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1356.5, + 1165.0, + 1356.5, + 1165.0, + 1369.5, + 1078.0, + 1369.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 1564.0, + 1281.0, + 1564.0, + 1281.0, + 1605.0, + 409.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2120.0, + 829.0, + 2120.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 42, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 324, + 668, + 1372, + 668, + 1372, + 1540, + 324, + 1540 + ], + "score": 0.975 + }, + { + "category_id": 4, + "poly": [ + 415, + 1571, + 1279, + 1571, + 1279, + 1606, + 415, + 1606 + ], + "score": 0.909 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 858, + 75, + 858, + 106, + 297, + 106 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 864, + 2088, + 864, + 2112, + 834, + 2112 + ], + "score": 0.844 + }, + { + "category_id": 15, + "poly": [ + 324.0, + 679.0, + 371.0, + 679.0, + 371.0, + 846.0, + 324.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 678.0, + 453.0, + 678.0, + 453.0, + 699.0, + 385.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 675.0, + 646.0, + 675.0, + 646.0, + 685.0, + 638.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 665.0, + 741.0, + 665.0, + 741.0, + 862.0, + 664.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 684.0, + 760.0, + 684.0, + 760.0, + 692.0, + 745.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 678.0, + 809.0, + 678.0, + 809.0, + 700.0, + 766.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 674.0, + 1003.0, + 674.0, + 1003.0, + 685.0, + 994.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 682.0, + 1067.0, + 682.0, + 1067.0, + 769.0, + 1039.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 675.0, + 1103.0, + 675.0, + 1103.0, + 684.0, + 1093.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 678.0, + 1363.0, + 678.0, + 1363.0, + 700.0, + 1319.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 699.0, + 371.0, + 699.0, + 371.0, + 725.0, + 345.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 695.0, + 449.0, + 695.0, + 449.0, + 718.0, + 384.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 696.0, + 616.0, + 696.0, + 616.0, + 707.0, + 606.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 699.0, + 727.0, + 699.0, + 727.0, + 726.0, + 702.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 701.0, + 761.0, + 701.0, + 761.0, + 713.0, + 742.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 696.0, + 806.0, + 696.0, + 806.0, + 718.0, + 768.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 688.0, + 1085.0, + 688.0, + 1085.0, + 715.0, + 1056.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 695.0, + 1360.0, + 695.0, + 1360.0, + 720.0, + 1293.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 724.0, + 371.0, + 724.0, + 371.0, + 747.0, + 345.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 725.0, + 727.0, + 725.0, + 727.0, + 750.0, + 701.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 725.0, + 1084.0, + 725.0, + 1084.0, + 748.0, + 1057.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 746.0, + 371.0, + 746.0, + 371.0, + 770.0, + 345.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 749.0, + 727.0, + 749.0, + 727.0, + 775.0, + 702.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 755.0, + 1082.0, + 755.0, + 1082.0, + 845.0, + 1037.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 769.0, + 371.0, + 769.0, + 371.0, + 794.0, + 345.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 775.0, + 727.0, + 775.0, + 727.0, + 798.0, + 701.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 792.0, + 370.0, + 792.0, + 370.0, + 818.0, + 345.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 798.0, + 458.0, + 798.0, + 458.0, + 811.0, + 443.0, + 811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 799.0, + 727.0, + 799.0, + 727.0, + 823.0, + 700.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 794.0, + 1084.0, + 794.0, + 1084.0, + 818.0, + 1057.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 806.0, + 1200.0, + 806.0, + 1200.0, + 816.0, + 1185.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 816.0, + 371.0, + 816.0, + 371.0, + 840.0, + 342.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 818.0, + 1167.0, + 818.0, + 1167.0, + 828.0, + 1154.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 825.0, + 779.0, + 825.0, + 779.0, + 834.0, + 769.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 828.0, + 1084.0, + 828.0, + 1084.0, + 852.0, + 1056.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 832.0, + 1131.0, + 832.0, + 1131.0, + 838.0, + 1125.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 839.0, + 372.0, + 839.0, + 372.0, + 862.0, + 345.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 842.0, + 1104.0, + 842.0, + 1104.0, + 852.0, + 1092.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 857.0, + 407.0, + 857.0, + 407.0, + 876.0, + 384.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 857.0, + 441.0, + 857.0, + 441.0, + 876.0, + 412.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 856.0, + 473.0, + 856.0, + 473.0, + 876.0, + 452.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 859.0, + 507.0, + 859.0, + 507.0, + 874.0, + 490.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 856.0, + 544.0, + 856.0, + 544.0, + 877.0, + 524.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 856.0, + 580.0, + 856.0, + 580.0, + 877.0, + 560.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 857.0, + 616.0, + 857.0, + 616.0, + 876.0, + 596.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 856.0, + 651.0, + 856.0, + 651.0, + 877.0, + 632.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 857.0, + 776.0, + 857.0, + 776.0, + 877.0, + 754.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 856.0, + 817.0, + 856.0, + 817.0, + 876.0, + 798.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 859.0, + 860.0, + 859.0, + 860.0, + 874.0, + 844.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 856.0, + 905.0, + 856.0, + 905.0, + 877.0, + 887.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 856.0, + 951.0, + 856.0, + 951.0, + 877.0, + 932.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 859.0, + 994.0, + 859.0, + 994.0, + 876.0, + 978.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1107.0, + 857.0, + 1130.0, + 857.0, + 1130.0, + 876.0, + 1107.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 856.0, + 1173.0, + 856.0, + 1173.0, + 876.0, + 1154.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 859.0, + 1218.0, + 859.0, + 1218.0, + 874.0, + 1202.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 859.0, + 1263.0, + 859.0, + 1263.0, + 876.0, + 1247.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 856.0, + 1311.0, + 856.0, + 1311.0, + 877.0, + 1293.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 857.0, + 1357.0, + 857.0, + 1357.0, + 876.0, + 1337.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 870.0, + 544.0, + 870.0, + 544.0, + 897.0, + 482.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 870.0, + 902.0, + 870.0, + 902.0, + 897.0, + 838.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 870.0, + 1257.0, + 870.0, + 1257.0, + 897.0, + 1195.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 904.0, + 548.0, + 904.0, + 548.0, + 940.0, + 438.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 901.0, + 915.0, + 901.0, + 915.0, + 941.0, + 783.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 902.0, + 1276.0, + 902.0, + 1276.0, + 941.0, + 1135.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 968.0, + 385.0, + 968.0, + 385.0, + 1160.0, + 308.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 974.0, + 649.0, + 974.0, + 649.0, + 987.0, + 634.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 968.0, + 741.0, + 968.0, + 741.0, + 1160.0, + 664.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 978.0, + 810.0, + 978.0, + 810.0, + 1001.0, + 766.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 974.0, + 1005.0, + 974.0, + 1005.0, + 987.0, + 988.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 963.0, + 1105.0, + 963.0, + 1105.0, + 1080.0, + 1027.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 978.0, + 1363.0, + 978.0, + 1363.0, + 1001.0, + 1318.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 992.0, + 371.0, + 992.0, + 371.0, + 1018.0, + 343.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1009.0, + 390.0, + 1009.0, + 390.0, + 1020.0, + 382.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 993.0, + 616.0, + 993.0, + 616.0, + 1004.0, + 606.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 998.0, + 728.0, + 998.0, + 728.0, + 1023.0, + 702.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 997.0, + 806.0, + 997.0, + 806.0, + 1019.0, + 742.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1004.0, + 1085.0, + 1004.0, + 1085.0, + 1029.0, + 1056.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 997.0, + 1359.0, + 997.0, + 1359.0, + 1019.0, + 1290.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1014.0, + 371.0, + 1014.0, + 371.0, + 1040.0, + 343.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1023.0, + 728.0, + 1023.0, + 728.0, + 1047.0, + 701.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1026.0, + 1084.0, + 1026.0, + 1084.0, + 1051.0, + 1057.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1029.0, + 1361.0, + 1029.0, + 1361.0, + 1037.0, + 1348.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1037.0, + 371.0, + 1037.0, + 371.0, + 1061.0, + 345.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 1041.0, + 1330.0, + 1041.0, + 1330.0, + 1054.0, + 1313.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1060.0, + 371.0, + 1060.0, + 371.0, + 1128.0, + 343.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1110.0, + 421.0, + 1110.0, + 421.0, + 1119.0, + 410.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1053.0, + 519.0, + 1053.0, + 519.0, + 1061.0, + 508.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 1051.0, + 646.0, + 1051.0, + 646.0, + 1085.0, + 580.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1048.0, + 728.0, + 1048.0, + 728.0, + 1122.0, + 701.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1088.0, + 908.0, + 1088.0, + 908.0, + 1098.0, + 898.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1063.0, + 1003.0, + 1063.0, + 1003.0, + 1072.0, + 990.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1047.0, + 1084.0, + 1047.0, + 1084.0, + 1147.0, + 1035.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1069.0, + 1265.0, + 1069.0, + 1265.0, + 1080.0, + 1255.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1116.0, + 786.0, + 1116.0, + 786.0, + 1134.0, + 762.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1125.0, + 371.0, + 1125.0, + 371.0, + 1149.0, + 343.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 1128.0, + 390.0, + 1128.0, + 390.0, + 1137.0, + 380.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 1132.0, + 752.0, + 1132.0, + 752.0, + 1147.0, + 734.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1128.0, + 1084.0, + 1128.0, + 1084.0, + 1154.0, + 1057.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1127.0, + 1137.0, + 1127.0, + 1137.0, + 1140.0, + 1122.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1156.0, + 398.0, + 1156.0, + 398.0, + 1178.0, + 368.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1156.0, + 434.0, + 1156.0, + 434.0, + 1178.0, + 402.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1157.0, + 467.0, + 1157.0, + 467.0, + 1178.0, + 446.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1157.0, + 504.0, + 1157.0, + 504.0, + 1177.0, + 484.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1157.0, + 542.0, + 1157.0, + 542.0, + 1178.0, + 522.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 1157.0, + 580.0, + 1157.0, + 580.0, + 1178.0, + 560.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 1157.0, + 617.0, + 1157.0, + 617.0, + 1177.0, + 597.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1157.0, + 654.0, + 1157.0, + 654.0, + 1178.0, + 634.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 1157.0, + 771.0, + 1157.0, + 771.0, + 1177.0, + 748.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1157.0, + 815.0, + 1157.0, + 815.0, + 1178.0, + 795.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1159.0, + 858.0, + 1159.0, + 858.0, + 1176.0, + 843.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1157.0, + 905.0, + 1157.0, + 905.0, + 1178.0, + 886.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 1157.0, + 953.0, + 1157.0, + 953.0, + 1178.0, + 933.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1157.0, + 999.0, + 1157.0, + 999.0, + 1177.0, + 978.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1157.0, + 1127.0, + 1157.0, + 1127.0, + 1177.0, + 1102.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1157.0, + 1170.0, + 1157.0, + 1170.0, + 1178.0, + 1149.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 1157.0, + 1263.0, + 1157.0, + 1263.0, + 1178.0, + 1245.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 1157.0, + 1311.0, + 1157.0, + 1311.0, + 1178.0, + 1292.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1157.0, + 1359.0, + 1157.0, + 1359.0, + 1178.0, + 1338.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 1172.0, + 543.0, + 1172.0, + 543.0, + 1196.0, + 483.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 1172.0, + 901.0, + 1172.0, + 901.0, + 1196.0, + 840.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 1172.0, + 1256.0, + 1172.0, + 1256.0, + 1196.0, + 1196.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 1203.0, + 560.0, + 1203.0, + 560.0, + 1240.0, + 426.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 1202.0, + 925.0, + 1202.0, + 925.0, + 1241.0, + 774.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 1202.0, + 1286.0, + 1202.0, + 1286.0, + 1241.0, + 1127.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1280.0, + 354.0, + 1280.0, + 354.0, + 1444.0, + 324.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 1281.0, + 402.0, + 1281.0, + 402.0, + 1295.0, + 386.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1277.0, + 453.0, + 1277.0, + 453.0, + 1299.0, + 410.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1275.0, + 727.0, + 1275.0, + 727.0, + 1448.0, + 679.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 1272.0, + 1097.0, + 1272.0, + 1097.0, + 1379.0, + 1029.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1275.0, + 1360.0, + 1275.0, + 1360.0, + 1285.0, + 1348.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1289.0, + 371.0, + 1289.0, + 371.0, + 1312.0, + 343.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1296.0, + 449.0, + 1296.0, + 449.0, + 1318.0, + 379.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1301.0, + 728.0, + 1301.0, + 728.0, + 1324.0, + 702.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1313.0, + 1084.0, + 1313.0, + 1084.0, + 1338.0, + 1056.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1324.0, + 371.0, + 1324.0, + 371.0, + 1348.0, + 345.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 1337.0, + 487.0, + 1337.0, + 487.0, + 1346.0, + 477.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1326.0, + 728.0, + 1326.0, + 728.0, + 1349.0, + 702.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1343.0, + 1084.0, + 1343.0, + 1084.0, + 1366.0, + 1057.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1343.0, + 1139.0, + 1343.0, + 1139.0, + 1357.0, + 1122.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1352.0, + 728.0, + 1352.0, + 728.0, + 1375.0, + 701.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1354.0, + 1069.0, + 1354.0, + 1069.0, + 1448.0, + 1037.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1356.0, + 1104.0, + 1356.0, + 1104.0, + 1365.0, + 1092.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1360.0, + 371.0, + 1360.0, + 371.0, + 1383.0, + 343.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1381.0, + 727.0, + 1381.0, + 727.0, + 1400.0, + 702.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1371.0, + 1084.0, + 1371.0, + 1084.0, + 1396.0, + 1057.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1396.0, + 371.0, + 1396.0, + 371.0, + 1419.0, + 345.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1404.0, + 728.0, + 1404.0, + 728.0, + 1427.0, + 701.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 1413.0, + 956.0, + 1413.0, + 956.0, + 1422.0, + 941.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1407.0, + 1007.0, + 1407.0, + 1007.0, + 1432.0, + 964.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1400.0, + 1085.0, + 1400.0, + 1085.0, + 1424.0, + 1055.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1411.0, + 1117.0, + 1411.0, + 1117.0, + 1426.0, + 1098.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1407.0, + 1166.0, + 1407.0, + 1166.0, + 1432.0, + 1124.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1433.0, + 370.0, + 1433.0, + 370.0, + 1453.0, + 346.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1431.0, + 728.0, + 1431.0, + 728.0, + 1454.0, + 701.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1426.0, + 1002.0, + 1426.0, + 1002.0, + 1448.0, + 964.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1428.0, + 1085.0, + 1428.0, + 1085.0, + 1451.0, + 1057.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1425.0, + 1164.0, + 1425.0, + 1164.0, + 1450.0, + 1123.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1456.0, + 422.0, + 1456.0, + 422.0, + 1476.0, + 398.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 1456.0, + 464.0, + 1456.0, + 464.0, + 1477.0, + 445.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1455.0, + 508.0, + 1455.0, + 508.0, + 1476.0, + 489.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 1455.0, + 555.0, + 1455.0, + 555.0, + 1477.0, + 535.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1456.0, + 600.0, + 1456.0, + 600.0, + 1477.0, + 581.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1458.0, + 645.0, + 1458.0, + 645.0, + 1475.0, + 629.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 1456.0, + 774.0, + 1456.0, + 774.0, + 1476.0, + 752.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 1456.0, + 816.0, + 1456.0, + 816.0, + 1477.0, + 798.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1456.0, + 862.0, + 1456.0, + 862.0, + 1476.0, + 843.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1458.0, + 907.0, + 1458.0, + 907.0, + 1475.0, + 892.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 1455.0, + 956.0, + 1455.0, + 956.0, + 1477.0, + 935.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1455.0, + 1002.0, + 1455.0, + 1002.0, + 1477.0, + 983.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1454.0, + 1131.0, + 1454.0, + 1131.0, + 1477.0, + 1105.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1455.0, + 1172.0, + 1455.0, + 1172.0, + 1477.0, + 1153.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1458.0, + 1218.0, + 1458.0, + 1218.0, + 1475.0, + 1202.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1455.0, + 1265.0, + 1455.0, + 1265.0, + 1477.0, + 1246.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 1455.0, + 1311.0, + 1455.0, + 1311.0, + 1477.0, + 1293.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1458.0, + 1356.0, + 1458.0, + 1356.0, + 1475.0, + 1342.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1468.0, + 545.0, + 1468.0, + 545.0, + 1496.0, + 482.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1468.0, + 902.0, + 1468.0, + 902.0, + 1496.0, + 838.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1468.0, + 1257.0, + 1468.0, + 1257.0, + 1496.0, + 1195.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 1503.0, + 566.0, + 1503.0, + 566.0, + 1537.0, + 421.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1502.0, + 931.0, + 1502.0, + 931.0, + 1538.0, + 768.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1503.0, + 1290.0, + 1503.0, + 1290.0, + 1536.0, + 1122.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.75, + 777.0, + 490.75, + 777.0, + 490.75, + 789.0, + 467.75, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 763.0, + 525.0, + 763.0, + 525.0, + 777.5, + 491.0, + 777.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 1041.5, + 653.0, + 1041.5, + 653.0, + 1067.0, + 605.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.25, + 1157.0, + 1218.25, + 1157.0, + 1218.25, + 1177.0, + 1194.25, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 1290.5, + 1133.0, + 1290.5, + 1133.0, + 1299.0, + 1125.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1568.0, + 1280.0, + 1568.0, + 1280.0, + 1610.0, + 410.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2121.0, + 830.0, + 2121.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 43, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 319, + 645, + 1377, + 645, + 1377, + 1558, + 319, + 1558 + ], + "score": 0.975 + }, + { + "category_id": 4, + "poly": [ + 400, + 1588, + 1287, + 1588, + 1287, + 1623, + 400, + 1623 + ], + "score": 0.917 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 858, + 75, + 858, + 106, + 297, + 106 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 864, + 2088, + 864, + 2112, + 834, + 2112 + ], + "score": 0.848 + }, + { + "category_id": 15, + "poly": [ + 341.0, + 647.0, + 363.0, + 647.0, + 363.0, + 668.0, + 341.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 647.0, + 720.0, + 647.0, + 720.0, + 672.0, + 695.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 656.0, + 746.0, + 656.0, + 746.0, + 684.0, + 729.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 647.0, + 1075.0, + 647.0, + 1075.0, + 668.0, + 1053.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 653.0, + 1123.0, + 653.0, + 1123.0, + 674.0, + 1082.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 672.0, + 349.0, + 672.0, + 349.0, + 821.0, + 322.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 679.0, + 651.0, + 679.0, + 651.0, + 706.0, + 639.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 669.0, + 720.0, + 669.0, + 720.0, + 825.0, + 674.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 671.0, + 1075.0, + 671.0, + 1075.0, + 821.0, + 1031.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 696.0, + 364.0, + 696.0, + 364.0, + 722.0, + 340.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 696.0, + 720.0, + 696.0, + 720.0, + 723.0, + 696.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 697.0, + 1074.0, + 697.0, + 1074.0, + 719.0, + 1053.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 720.0, + 364.0, + 720.0, + 364.0, + 747.0, + 341.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 720.0, + 720.0, + 720.0, + 720.0, + 747.0, + 696.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 722.0, + 1074.0, + 722.0, + 1074.0, + 745.0, + 1054.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 745.0, + 363.0, + 745.0, + 363.0, + 772.0, + 340.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 747.0, + 718.0, + 747.0, + 718.0, + 770.0, + 697.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 747.0, + 1074.0, + 747.0, + 1074.0, + 769.0, + 1053.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 773.0, + 364.0, + 773.0, + 364.0, + 796.0, + 341.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 773.0, + 719.0, + 773.0, + 719.0, + 795.0, + 698.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 773.0, + 1075.0, + 773.0, + 1075.0, + 795.0, + 1054.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 800.0, + 362.0, + 800.0, + 362.0, + 817.0, + 347.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 798.0, + 720.0, + 798.0, + 720.0, + 819.0, + 701.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 801.0, + 1074.0, + 801.0, + 1074.0, + 817.0, + 1058.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 813.0, + 543.0, + 813.0, + 543.0, + 835.0, + 493.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 815.0, + 784.0, + 815.0, + 784.0, + 834.0, + 747.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 815.0, + 1140.0, + 815.0, + 1140.0, + 833.0, + 1103.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 825.0, + 363.0, + 825.0, + 363.0, + 842.0, + 347.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 828.0, + 540.0, + 828.0, + 540.0, + 849.0, + 508.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 826.0, + 720.0, + 826.0, + 720.0, + 842.0, + 703.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 829.0, + 783.0, + 829.0, + 783.0, + 851.0, + 732.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 826.0, + 1074.0, + 826.0, + 1074.0, + 842.0, + 1058.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 829.0, + 1139.0, + 829.0, + 1139.0, + 849.0, + 1105.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 851.0, + 391.0, + 851.0, + 391.0, + 873.0, + 361.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 851.0, + 445.0, + 851.0, + 445.0, + 873.0, + 416.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 849.0, + 555.0, + 849.0, + 555.0, + 886.0, + 470.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 851.0, + 610.0, + 851.0, + 610.0, + 873.0, + 581.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 849.0, + 666.0, + 849.0, + 666.0, + 873.0, + 634.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 851.0, + 758.0, + 851.0, + 758.0, + 874.0, + 729.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 852.0, + 824.0, + 852.0, + 824.0, + 874.0, + 795.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 845.0, + 907.0, + 845.0, + 907.0, + 893.0, + 838.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 852.0, + 955.0, + 852.0, + 955.0, + 874.0, + 926.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 853.0, + 1019.0, + 853.0, + 1019.0, + 873.0, + 994.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 853.0, + 1110.0, + 853.0, + 1110.0, + 871.0, + 1082.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 851.0, + 1144.0, + 851.0, + 1144.0, + 873.0, + 1113.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1148.0, + 853.0, + 1176.0, + 853.0, + 1176.0, + 871.0, + 1148.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 846.0, + 1275.0, + 846.0, + 1275.0, + 891.0, + 1182.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 853.0, + 1310.0, + 853.0, + 1310.0, + 871.0, + 1282.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 853.0, + 1344.0, + 853.0, + 1344.0, + 871.0, + 1316.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 853.0, + 1375.0, + 853.0, + 1375.0, + 871.0, + 1349.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 884.0, + 562.0, + 884.0, + 562.0, + 920.0, + 423.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 884.0, + 926.0, + 884.0, + 926.0, + 920.0, + 772.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 884.0, + 1282.0, + 884.0, + 1282.0, + 920.0, + 1130.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 960.0, + 362.0, + 960.0, + 362.0, + 980.0, + 339.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 972.0, + 654.0, + 972.0, + 654.0, + 990.0, + 617.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 970.0, + 1013.0, + 970.0, + 1013.0, + 992.0, + 954.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 966.0, + 1078.0, + 966.0, + 1078.0, + 1135.0, + 1029.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 951.0, + 1368.0, + 951.0, + 1368.0, + 993.0, + 1323.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 989.0, + 363.0, + 989.0, + 363.0, + 1139.0, + 320.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 986.0, + 651.0, + 986.0, + 651.0, + 1008.0, + 619.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 982.0, + 721.0, + 982.0, + 721.0, + 1140.0, + 678.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 985.0, + 1011.0, + 985.0, + 1011.0, + 1007.0, + 957.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1003.0, + 1078.0, + 1003.0, + 1078.0, + 1027.0, + 1053.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1033.0, + 361.0, + 1033.0, + 361.0, + 1054.0, + 340.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1037.0, + 720.0, + 1037.0, + 720.0, + 1064.0, + 697.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1041.0, + 1078.0, + 1041.0, + 1078.0, + 1065.0, + 1053.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1071.0, + 361.0, + 1071.0, + 361.0, + 1092.0, + 341.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1065.0, + 720.0, + 1065.0, + 720.0, + 1092.0, + 696.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1078.0, + 1078.0, + 1078.0, + 1078.0, + 1103.0, + 1053.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1094.0, + 720.0, + 1094.0, + 720.0, + 1121.0, + 697.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1108.0, + 361.0, + 1108.0, + 361.0, + 1130.0, + 340.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1125.0, + 719.0, + 1125.0, + 719.0, + 1145.0, + 698.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1133.0, + 1005.0, + 1133.0, + 1005.0, + 1143.0, + 996.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1115.0, + 1078.0, + 1115.0, + 1078.0, + 1139.0, + 1053.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1121.0, + 1105.0, + 1121.0, + 1105.0, + 1133.0, + 1095.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 1125.0, + 1355.0, + 1125.0, + 1355.0, + 1155.0, + 1338.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1150.0, + 387.0, + 1150.0, + 387.0, + 1190.0, + 349.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 1168.0, + 444.0, + 1168.0, + 444.0, + 1190.0, + 414.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1161.0, + 557.0, + 1161.0, + 557.0, + 1210.0, + 463.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 1168.0, + 614.0, + 1168.0, + 614.0, + 1190.0, + 584.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 1167.0, + 669.0, + 1167.0, + 669.0, + 1190.0, + 639.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1154.0, + 742.0, + 1154.0, + 742.0, + 1188.0, + 698.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1159.0, + 990.0, + 1159.0, + 990.0, + 1211.0, + 745.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1167.0, + 1024.0, + 1167.0, + 1024.0, + 1189.0, + 992.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1164.0, + 1123.0, + 1164.0, + 1123.0, + 1186.0, + 1090.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1164.0, + 1163.0, + 1164.0, + 1163.0, + 1186.0, + 1128.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 1162.0, + 1317.0, + 1162.0, + 1317.0, + 1204.0, + 1168.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 1164.0, + 1355.0, + 1164.0, + 1355.0, + 1186.0, + 1323.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1200.0, + 564.0, + 1200.0, + 564.0, + 1238.0, + 423.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1200.0, + 919.0, + 1200.0, + 919.0, + 1238.0, + 780.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1202.0, + 1274.0, + 1202.0, + 1274.0, + 1238.0, + 1137.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1262.0, + 1081.0, + 1262.0, + 1081.0, + 1288.0, + 1053.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 1269.0, + 1368.0, + 1269.0, + 1368.0, + 1291.0, + 1324.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1289.0, + 364.0, + 1289.0, + 364.0, + 1310.0, + 343.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 1290.0, + 655.0, + 1290.0, + 655.0, + 1316.0, + 615.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 667.0, + 1283.0, + 738.0, + 1283.0, + 738.0, + 1475.0, + 667.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1288.0, + 1012.0, + 1288.0, + 1012.0, + 1307.0, + 974.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1287.0, + 1061.0, + 1287.0, + 1061.0, + 1452.0, + 1030.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 1285.0, + 1364.0, + 1285.0, + 1364.0, + 1311.0, + 1304.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1310.0, + 363.0, + 1310.0, + 363.0, + 1458.0, + 319.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1305.0, + 651.0, + 1305.0, + 651.0, + 1329.0, + 590.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1329.0, + 719.0, + 1329.0, + 719.0, + 1355.0, + 694.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1302.0, + 1008.0, + 1302.0, + 1008.0, + 1324.0, + 976.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1298.0, + 1081.0, + 1298.0, + 1081.0, + 1360.0, + 1053.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1350.0, + 719.0, + 1350.0, + 719.0, + 1379.0, + 695.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1371.0, + 363.0, + 1371.0, + 363.0, + 1392.0, + 344.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 1377.0, + 717.0, + 1377.0, + 717.0, + 1399.0, + 696.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1371.0, + 1081.0, + 1371.0, + 1081.0, + 1395.0, + 1052.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 1375.0, + 1130.0, + 1375.0, + 1130.0, + 1405.0, + 1117.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1412.0, + 363.0, + 1412.0, + 363.0, + 1433.0, + 344.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 1395.0, + 721.0, + 1395.0, + 721.0, + 1472.0, + 692.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1407.0, + 1081.0, + 1407.0, + 1081.0, + 1433.0, + 1053.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1444.0, + 1081.0, + 1444.0, + 1081.0, + 1468.0, + 1053.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1445.0, + 1101.0, + 1445.0, + 1101.0, + 1456.0, + 1089.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1453.0, + 364.0, + 1453.0, + 364.0, + 1473.0, + 343.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1469.0, + 718.0, + 1469.0, + 718.0, + 1490.0, + 695.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1486.0, + 391.0, + 1486.0, + 391.0, + 1508.0, + 361.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1486.0, + 445.0, + 1486.0, + 445.0, + 1509.0, + 416.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 470.0, + 1486.0, + 555.0, + 1486.0, + 555.0, + 1521.0, + 470.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 583.0, + 1486.0, + 613.0, + 1486.0, + 613.0, + 1509.0, + 583.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1485.0, + 668.0, + 1485.0, + 668.0, + 1508.0, + 638.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1486.0, + 754.0, + 1486.0, + 754.0, + 1508.0, + 724.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1486.0, + 820.0, + 1486.0, + 820.0, + 1508.0, + 755.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1486.0, + 888.0, + 1486.0, + 888.0, + 1522.0, + 823.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1486.0, + 924.0, + 1486.0, + 924.0, + 1508.0, + 894.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1486.0, + 958.0, + 1486.0, + 958.0, + 1508.0, + 925.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1489.0, + 990.0, + 1489.0, + 990.0, + 1507.0, + 961.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1486.0, + 1024.0, + 1486.0, + 1024.0, + 1508.0, + 992.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1480.0, + 1166.0, + 1480.0, + 1166.0, + 1506.0, + 1092.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 1480.0, + 1357.0, + 1480.0, + 1357.0, + 1521.0, + 1168.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1521.0, + 563.0, + 1521.0, + 563.0, + 1554.0, + 423.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 1518.0, + 920.0, + 1518.0, + 920.0, + 1558.0, + 782.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 1518.0, + 1273.0, + 1518.0, + 1273.0, + 1555.0, + 1138.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 715.0, + 855.0, + 715.0, + 855.0, + 731.0, + 800.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.75, + 672.5, + 365.75, + 672.5, + 365.75, + 693.5, + 337.75, + 693.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.75, + 696.0, + 1369.75, + 696.0, + 1369.75, + 702.0, + 1329.75, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 800.0, + 985.0, + 800.0, + 985.0, + 808.0, + 946.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 963.5, + 1266.0, + 963.5, + 1266.0, + 976.5, + 1243.0, + 976.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 1048.0, + 1331.0, + 1048.0, + 1331.0, + 1057.5, + 1295.0, + 1057.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1048.5, + 608.0, + 1048.5, + 608.0, + 1058.5, + 585.0, + 1058.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 980.5, + 1220.0, + 980.5, + 1220.0, + 999.5, + 1129.0, + 999.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 1104.5, + 638.0, + 1104.5, + 638.0, + 1115.5, + 625.0, + 1115.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 1054.5, + 1162.0, + 1054.5, + 1162.0, + 1071.5, + 1081.0, + 1071.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.75, + 1363.0, + 447.75, + 1363.0, + 447.75, + 1370.0, + 414.75, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1397.0, + 625.0, + 1397.0, + 625.0, + 1408.0, + 614.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 1585.0, + 1289.0, + 1585.0, + 1289.0, + 1629.0, + 401.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2120.0, + 830.0, + 2120.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 44, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 320, + 643, + 1377, + 643, + 1377, + 1562, + 320, + 1562 + ], + "score": 0.975 + }, + { + "category_id": 4, + "poly": [ + 400, + 1594, + 1287, + 1594, + 1287, + 1628, + 400, + 1628 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 297, + 75, + 858, + 75, + 858, + 106, + 297, + 106 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2113, + 834, + 2113 + ], + "score": 0.85 + }, + { + "category_id": 15, + "poly": [ + 321.0, + 657.0, + 370.0, + 657.0, + 370.0, + 832.0, + 321.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 649.0, + 449.0, + 649.0, + 449.0, + 691.0, + 376.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 649.0, + 652.0, + 649.0, + 652.0, + 676.0, + 638.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 647.0, + 804.0, + 647.0, + 804.0, + 847.0, + 662.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 670.0, + 1076.0, + 670.0, + 1076.0, + 690.0, + 1053.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 671.0, + 1102.0, + 671.0, + 1102.0, + 680.0, + 1082.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 667.0, + 1140.0, + 667.0, + 1140.0, + 686.0, + 1103.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 685.0, + 370.0, + 685.0, + 370.0, + 711.0, + 342.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 679.0, + 1063.0, + 679.0, + 1063.0, + 837.0, + 1032.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 681.0, + 1139.0, + 681.0, + 1139.0, + 703.0, + 1103.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 703.0, + 1076.0, + 703.0, + 1076.0, + 724.0, + 1055.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 714.0, + 369.0, + 714.0, + 369.0, + 740.0, + 342.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 713.0, + 726.0, + 713.0, + 726.0, + 738.0, + 699.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 735.0, + 1075.0, + 735.0, + 1075.0, + 755.0, + 1053.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 744.0, + 369.0, + 744.0, + 369.0, + 769.0, + 342.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 746.0, + 726.0, + 746.0, + 726.0, + 771.0, + 699.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 774.0, + 366.0, + 774.0, + 366.0, + 795.0, + 344.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 778.0, + 724.0, + 778.0, + 724.0, + 805.0, + 699.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 768.0, + 1075.0, + 768.0, + 1075.0, + 789.0, + 1055.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 801.0, + 369.0, + 801.0, + 369.0, + 824.0, + 340.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 810.0, + 726.0, + 810.0, + 726.0, + 834.0, + 699.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 800.0, + 1076.0, + 800.0, + 1076.0, + 823.0, + 1053.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 829.0, + 369.0, + 829.0, + 369.0, + 854.0, + 341.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 833.0, + 1076.0, + 833.0, + 1076.0, + 854.0, + 1053.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 844.0, + 727.0, + 844.0, + 727.0, + 867.0, + 699.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 861.0, + 423.0, + 861.0, + 423.0, + 883.0, + 394.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 861.0, + 460.0, + 861.0, + 460.0, + 883.0, + 428.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 862.0, + 544.0, + 862.0, + 544.0, + 900.0, + 472.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 862.0, + 571.0, + 862.0, + 571.0, + 883.0, + 550.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 862.0, + 610.0, + 862.0, + 610.0, + 883.0, + 588.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 862.0, + 648.0, + 862.0, + 648.0, + 882.0, + 627.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 860.0, + 788.0, + 860.0, + 788.0, + 883.0, + 759.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 860.0, + 893.0, + 860.0, + 893.0, + 900.0, + 810.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 859.0, + 933.0, + 859.0, + 933.0, + 883.0, + 908.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 859.0, + 981.0, + 859.0, + 981.0, + 884.0, + 957.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 869.0, + 1116.0, + 869.0, + 1116.0, + 882.0, + 1099.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 869.0, + 1168.0, + 869.0, + 1168.0, + 883.0, + 1153.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 861.0, + 1249.0, + 861.0, + 1249.0, + 904.0, + 1180.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 869.0, + 1276.0, + 869.0, + 1276.0, + 883.0, + 1261.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 870.0, + 1329.0, + 870.0, + 1329.0, + 882.0, + 1319.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 896.0, + 551.0, + 896.0, + 551.0, + 937.0, + 432.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 894.0, + 917.0, + 894.0, + 917.0, + 938.0, + 781.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 894.0, + 1279.0, + 894.0, + 1279.0, + 938.0, + 1133.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 957.0, + 726.0, + 957.0, + 726.0, + 981.0, + 697.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 962.0, + 804.0, + 962.0, + 804.0, + 983.0, + 758.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 975.0, + 365.0, + 975.0, + 365.0, + 1145.0, + 319.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 975.0, + 430.0, + 975.0, + 430.0, + 1012.0, + 393.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 986.0, + 724.0, + 986.0, + 724.0, + 1148.0, + 677.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 978.0, + 801.0, + 978.0, + 801.0, + 1003.0, + 730.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1002.0, + 1009.0, + 1002.0, + 1009.0, + 1011.0, + 996.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 996.0, + 1077.0, + 996.0, + 1077.0, + 1148.0, + 1032.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 977.0, + 1141.0, + 977.0, + 1141.0, + 1013.0, + 1103.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1014.0, + 724.0, + 1014.0, + 724.0, + 1040.0, + 697.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1034.0, + 365.0, + 1034.0, + 365.0, + 1057.0, + 342.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1038.0, + 1076.0, + 1038.0, + 1076.0, + 1058.0, + 1055.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1063.0, + 365.0, + 1063.0, + 365.0, + 1089.0, + 341.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1069.0, + 726.0, + 1069.0, + 726.0, + 1100.0, + 695.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1077.0, + 1077.0, + 1077.0, + 1077.0, + 1101.0, + 1052.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1093.0, + 365.0, + 1093.0, + 365.0, + 1118.0, + 341.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1098.0, + 724.0, + 1098.0, + 724.0, + 1128.0, + 695.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1125.0, + 364.0, + 1125.0, + 364.0, + 1145.0, + 342.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1127.0, + 726.0, + 1127.0, + 726.0, + 1156.0, + 695.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1118.0, + 1077.0, + 1118.0, + 1077.0, + 1143.0, + 1051.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1155.0, + 364.0, + 1155.0, + 364.0, + 1176.0, + 342.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1158.0, + 724.0, + 1158.0, + 724.0, + 1182.0, + 697.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 1158.0, + 742.0, + 1158.0, + 742.0, + 1169.0, + 732.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1174.0, + 394.0, + 1174.0, + 394.0, + 1193.0, + 371.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1175.0, + 438.0, + 1175.0, + 438.0, + 1193.0, + 413.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1166.0, + 540.0, + 1166.0, + 540.0, + 1215.0, + 459.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1174.0, + 568.0, + 1174.0, + 568.0, + 1193.0, + 547.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 1174.0, + 611.0, + 1174.0, + 611.0, + 1194.0, + 592.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 1174.0, + 656.0, + 1174.0, + 656.0, + 1194.0, + 635.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1174.0, + 774.0, + 1174.0, + 774.0, + 1197.0, + 745.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 1174.0, + 823.0, + 1174.0, + 823.0, + 1197.0, + 800.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 1164.0, + 899.0, + 1164.0, + 899.0, + 1218.0, + 827.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1172.0, + 928.0, + 1172.0, + 928.0, + 1198.0, + 904.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1174.0, + 981.0, + 1174.0, + 981.0, + 1197.0, + 958.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 1163.0, + 1077.0, + 1163.0, + 1077.0, + 1183.0, + 1055.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1176.0, + 1116.0, + 1176.0, + 1116.0, + 1196.0, + 1092.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1167.0, + 1238.0, + 1167.0, + 1238.0, + 1216.0, + 1147.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1176.0, + 1279.0, + 1176.0, + 1279.0, + 1196.0, + 1258.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 1176.0, + 1334.0, + 1176.0, + 1334.0, + 1196.0, + 1315.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1207.0, + 561.0, + 1207.0, + 561.0, + 1244.0, + 425.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 1207.0, + 926.0, + 1207.0, + 926.0, + 1244.0, + 772.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 1207.0, + 1287.0, + 1207.0, + 1287.0, + 1246.0, + 1126.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1275.0, + 448.0, + 1275.0, + 448.0, + 1302.0, + 405.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1275.0, + 650.0, + 1275.0, + 650.0, + 1285.0, + 641.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1294.0, + 369.0, + 1294.0, + 369.0, + 1459.0, + 317.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1294.0, + 444.0, + 1294.0, + 444.0, + 1317.0, + 382.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1290.0, + 785.0, + 1290.0, + 785.0, + 1480.0, + 672.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1295.0, + 1102.0, + 1295.0, + 1102.0, + 1479.0, + 1020.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1290.0, + 1140.0, + 1290.0, + 1140.0, + 1327.0, + 1103.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 1318.0, + 1075.0, + 1318.0, + 1075.0, + 1343.0, + 1050.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1332.0, + 369.0, + 1332.0, + 369.0, + 1356.0, + 340.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 694.0, + 1351.0, + 721.0, + 1351.0, + 721.0, + 1378.0, + 694.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1369.0, + 368.0, + 1369.0, + 368.0, + 1395.0, + 340.0, + 1395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1367.0, + 1074.0, + 1367.0, + 1074.0, + 1388.0, + 1052.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 1385.0, + 718.0, + 1385.0, + 718.0, + 1406.0, + 695.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1392.0, + 1075.0, + 1392.0, + 1075.0, + 1415.0, + 1052.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 1407.0, + 368.0, + 1407.0, + 368.0, + 1432.0, + 340.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1416.0, + 720.0, + 1416.0, + 720.0, + 1439.0, + 697.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1416.0, + 1075.0, + 1416.0, + 1075.0, + 1439.0, + 1052.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 1432.0, + 1151.0, + 1432.0, + 1151.0, + 1445.0, + 1134.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1442.0, + 370.0, + 1442.0, + 370.0, + 1471.0, + 342.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1445.0, + 720.0, + 1445.0, + 720.0, + 1469.0, + 697.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1465.0, + 1075.0, + 1465.0, + 1075.0, + 1486.0, + 1052.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 1479.0, + 548.0, + 1479.0, + 548.0, + 1531.0, + 471.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1479.0, + 718.0, + 1479.0, + 718.0, + 1498.0, + 697.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1488.0, + 424.0, + 1488.0, + 424.0, + 1510.0, + 395.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1488.0, + 471.0, + 1488.0, + 471.0, + 1512.0, + 448.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 1488.0, + 571.0, + 1488.0, + 571.0, + 1509.0, + 550.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1488.0, + 623.0, + 1488.0, + 623.0, + 1512.0, + 600.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1492.0, + 761.0, + 1492.0, + 761.0, + 1510.0, + 738.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 1492.0, + 812.0, + 1492.0, + 812.0, + 1512.0, + 793.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1497.0, + 864.0, + 1497.0, + 864.0, + 1507.0, + 851.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 1492.0, + 921.0, + 1492.0, + 921.0, + 1512.0, + 902.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 1494.0, + 974.0, + 1494.0, + 974.0, + 1509.0, + 959.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1492.0, + 1121.0, + 1492.0, + 1121.0, + 1510.0, + 1099.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 1493.0, + 1170.0, + 1493.0, + 1170.0, + 1509.0, + 1155.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 1488.0, + 1251.0, + 1488.0, + 1251.0, + 1529.0, + 1187.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 1496.0, + 1278.0, + 1496.0, + 1278.0, + 1507.0, + 1266.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1494.0, + 1334.0, + 1494.0, + 1334.0, + 1508.0, + 1319.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1501.0, + 886.0, + 1501.0, + 886.0, + 1526.0, + 830.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1523.0, + 568.0, + 1523.0, + 568.0, + 1560.0, + 418.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1521.0, + 933.0, + 1521.0, + 933.0, + 1560.0, + 766.0, + 1560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1522.0, + 1293.0, + 1522.0, + 1293.0, + 1562.0, + 1119.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 685.0, + 1106.0, + 685.0, + 1106.0, + 697.0, + 1086.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 693.5, + 643.0, + 693.5, + 643.0, + 721.5, + 523.0, + 721.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 702.0, + 1017.0, + 702.0, + 1017.0, + 729.0, + 828.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 774.5, + 502.0, + 774.5, + 502.0, + 793.5, + 429.0, + 793.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 822.0, + 793.0, + 822.0, + 793.0, + 843.0, + 719.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1042.0, + 719.0, + 1042.0, + 719.0, + 1070.0, + 701.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.25, + 1054.5, + 566.25, + 1054.5, + 566.25, + 1073.5, + 490.25, + 1073.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1090.0, + 849.0, + 1090.0, + 849.0, + 1104.5, + 820.0, + 1104.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1087.0, + 1205.0, + 1087.0, + 1205.0, + 1101.0, + 1179.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1113.5, + 473.0, + 1113.5, + 473.0, + 1129.5, + 397.0, + 1129.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.25, + 1118.0, + 825.25, + 1118.0, + 825.25, + 1139.5, + 744.25, + 1139.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1343.0, + 1068.0, + 1343.0, + 1068.0, + 1364.0, + 1057.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1393.5, + 851.0, + 1393.5, + 851.0, + 1406.5, + 821.0, + 1406.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.25, + 1413.0, + 1186.25, + 1413.0, + 1186.25, + 1423.0, + 1165.25, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 1434.0, + 803.0, + 1434.0, + 803.0, + 1450.0, + 749.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1590.0, + 1289.0, + 1590.0, + 1289.0, + 1633.0, + 400.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 298.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 870.0, + 2084.0, + 870.0, + 2121.0, + 829.0, + 2121.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 45, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/0lz69oI5iZP/0lz69oI5iZP_layout.pdf b/parse/train/0lz69oI5iZP/0lz69oI5iZP_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..da9cdafd525b7f49a2816a7ccbbf8a97022fe8ef --- /dev/null +++ b/parse/train/0lz69oI5iZP/0lz69oI5iZP_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4da5eab435c1498e9a263edb261c03b2301173711a14d77b25a8ab717d9d7de +size 697779 diff --git a/parse/train/0lz69oI5iZP/0lz69oI5iZP_origin.pdf b/parse/train/0lz69oI5iZP/0lz69oI5iZP_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e39b6f4ac935226b3d14cb8a7101dfb87679cb5b --- /dev/null +++ b/parse/train/0lz69oI5iZP/0lz69oI5iZP_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7ec1567e860d46b98e69750b14a9a37f678a427316363cb10621a6e9dc097fa8 +size 521207 diff --git a/parse/train/0lz69oI5iZP/0lz69oI5iZP_span.pdf b/parse/train/0lz69oI5iZP/0lz69oI5iZP_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..afc7e551f41f45553822f01f953bf41edd31faca --- /dev/null +++ b/parse/train/0lz69oI5iZP/0lz69oI5iZP_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ececa5aba2948876f46c6f21c0cb7c772c933e180d546f619d8f429d38837e7 +size 709851 diff --git a/parse/train/B1IzH7cxl/B1IzH7cxl_layout.pdf b/parse/train/B1IzH7cxl/B1IzH7cxl_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4be4b9f0288453b99e78a07f5c08f340b9049bf1 --- /dev/null +++ b/parse/train/B1IzH7cxl/B1IzH7cxl_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8e67ee472b89653ab801927ef91940b1baec16bda06490df448b172558b9dfff +size 1711122 diff --git a/parse/train/B1IzH7cxl/B1IzH7cxl_origin.pdf b/parse/train/B1IzH7cxl/B1IzH7cxl_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..d473fbba866f6d800c871d421b5be2b4ce1caee9 --- /dev/null +++ b/parse/train/B1IzH7cxl/B1IzH7cxl_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f428229027ce0d236a308c92c6e21753eac681dc7e795e7dcedb49147b7e1be9 +size 1503170 diff --git a/parse/train/B1IzH7cxl/B1IzH7cxl_span.pdf b/parse/train/B1IzH7cxl/B1IzH7cxl_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..dc5ad439fdc61e4412af42051267803c9d3e26e5 --- /dev/null +++ b/parse/train/B1IzH7cxl/B1IzH7cxl_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a57d20b1cd2e4d3a9a9e86a3d265ebbd76785f5dc30e5c3cae7fbca6678c291 +size 1716291 diff --git a/parse/train/BJgcwh4FwS/BJgcwh4FwS_layout.pdf b/parse/train/BJgcwh4FwS/BJgcwh4FwS_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..7b8e63fedf95c613d269b66f93c7200e08f5b55b --- /dev/null +++ b/parse/train/BJgcwh4FwS/BJgcwh4FwS_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:15e1a6cf413f823f5ef9de0f5082f4a814019dc35225b0b124a0e58ae0a98eba +size 4751317 diff --git a/parse/train/BJgcwh4FwS/BJgcwh4FwS_origin.pdf b/parse/train/BJgcwh4FwS/BJgcwh4FwS_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..52c0acb4757b8c66fe160eb1c00552b5b45bd1cd --- /dev/null +++ b/parse/train/BJgcwh4FwS/BJgcwh4FwS_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06cf4ed493bb1d842d91e16edcffaa53837cc0c7ab11949e7dc58df67f5d3ba7 +size 4559866 diff --git a/parse/train/BJgcwh4FwS/BJgcwh4FwS_span.pdf b/parse/train/BJgcwh4FwS/BJgcwh4FwS_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..54ebdecb9cc537478e8ed48e2d469494340f71c0 --- /dev/null +++ b/parse/train/BJgcwh4FwS/BJgcwh4FwS_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e89e9a682ac0bea05e255951d2a226d6805fb590f0063efd21dacb8b6b3422dd +size 4757595 diff --git a/parse/train/FsLTUzZlsgT/FsLTUzZlsgT_layout.pdf b/parse/train/FsLTUzZlsgT/FsLTUzZlsgT_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..28d993d8a53468dde554d93db46cc82781a0adf0 --- /dev/null +++ b/parse/train/FsLTUzZlsgT/FsLTUzZlsgT_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:69ca0718529d300c43f27d4be11aa4efde26819701d1056025548d4d8567185a +size 1570852 diff --git a/parse/train/FsLTUzZlsgT/FsLTUzZlsgT_origin.pdf b/parse/train/FsLTUzZlsgT/FsLTUzZlsgT_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..69b9d6019a73b2e61d0ed7b992fe5100a629e6e9 --- /dev/null +++ b/parse/train/FsLTUzZlsgT/FsLTUzZlsgT_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:233787b94ccad2256da9fc01028f263dba9062d5d8c23505d2d73630d3e59a68 +size 1350795 diff --git a/parse/train/FsLTUzZlsgT/FsLTUzZlsgT_span.pdf b/parse/train/FsLTUzZlsgT/FsLTUzZlsgT_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a84d82811b42a40130102db5835aeb0a668afa0d --- /dev/null +++ b/parse/train/FsLTUzZlsgT/FsLTUzZlsgT_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ba29ca19d74ca909d45eb4af9dcb5c2f14631d5a597339ff4b2cf95299ff67e +size 1581083 diff --git a/parse/train/HJWHIKqgl/HJWHIKqgl_layout.pdf b/parse/train/HJWHIKqgl/HJWHIKqgl_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..fb9a56212108f752e76204f1dace5cf220cbdb1e --- /dev/null +++ b/parse/train/HJWHIKqgl/HJWHIKqgl_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6d4ce696718323aa0093a70d458021d9d0c6e24020bbc814e713930652e234e +size 2387303 diff --git a/parse/train/HJWHIKqgl/HJWHIKqgl_origin.pdf b/parse/train/HJWHIKqgl/HJWHIKqgl_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5a70c111786285c4c1c647142e8895958e727dc6 --- /dev/null +++ b/parse/train/HJWHIKqgl/HJWHIKqgl_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8474dfd03cc80de874261bc7ab1907dba6f8fd7affa69bf4247384f970e23f3f +size 2238893 diff --git a/parse/train/HJWHIKqgl/HJWHIKqgl_span.pdf b/parse/train/HJWHIKqgl/HJWHIKqgl_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..79242a41c906a5913800a048a585115646dd2b54 --- /dev/null +++ b/parse/train/HJWHIKqgl/HJWHIKqgl_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5cd734f57b9479f6114ac82cee294976b6a73649d383922d3030b9700f601746 +size 2395988 diff --git a/parse/train/HJem3yHKwH/HJem3yHKwH_layout.pdf b/parse/train/HJem3yHKwH/HJem3yHKwH_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..85e8fdbc770862f843acb8f3bbe7b852367513ab --- /dev/null +++ b/parse/train/HJem3yHKwH/HJem3yHKwH_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddc3e48d9fcf2ff2988c0a4c353ad5bc161dd437a682d691c12e8cd6feb51343 +size 2430081 diff --git a/parse/train/HJem3yHKwH/HJem3yHKwH_origin.pdf b/parse/train/HJem3yHKwH/HJem3yHKwH_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4fee2c3dfa8c08018ad389d203250bbd39d84407 --- /dev/null +++ b/parse/train/HJem3yHKwH/HJem3yHKwH_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5873d66d2dea7c2e40e6826fc33078d381e2ca047d96541319e6fa40f96a0154 +size 2326269 diff --git a/parse/train/HJem3yHKwH/HJem3yHKwH_span.pdf b/parse/train/HJem3yHKwH/HJem3yHKwH_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8fe71d065299ef5b0b45746e78a3de3642868c5a --- /dev/null +++ b/parse/train/HJem3yHKwH/HJem3yHKwH_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0bde9bbbe548285a4390c501c77d191cd15cc9e7ec15e2e42c87cc7a0900436 +size 2433610 diff --git a/parse/train/S1WRibb0Z/S1WRibb0Z_layout.pdf b/parse/train/S1WRibb0Z/S1WRibb0Z_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ee574f9e7fa61ad04d05b16983fee788c77d55a7 --- /dev/null +++ b/parse/train/S1WRibb0Z/S1WRibb0Z_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c224439eeaffc967cfe864a9064f4d93c029ae9b19ca0b8ed1f89183782c1f16 +size 1039442 diff --git a/parse/train/S1WRibb0Z/S1WRibb0Z_origin.pdf b/parse/train/S1WRibb0Z/S1WRibb0Z_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..f69058dc8177529d65db64e0decfa9e5c895e143 --- /dev/null +++ b/parse/train/S1WRibb0Z/S1WRibb0Z_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22e4466d3048c3ea9353d45ef5893a86422688b1a5008fd23a3a1112a901968f +size 867921 diff --git a/parse/train/S1WRibb0Z/S1WRibb0Z_span.pdf b/parse/train/S1WRibb0Z/S1WRibb0Z_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..412d43968bc8be44c081f3de68cc9c1f3113bbfc --- /dev/null +++ b/parse/train/S1WRibb0Z/S1WRibb0Z_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bccd482b58b30226349af1c1812ca773f9fa4641c97a7e6eda7489663fda984a +size 1038302 diff --git a/parse/train/S1c2cvqee/S1c2cvqee_layout.pdf b/parse/train/S1c2cvqee/S1c2cvqee_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a291548382cd28828660da01377e9728d9e0d33f --- /dev/null +++ b/parse/train/S1c2cvqee/S1c2cvqee_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9686116e3884d1d58f2e391dba5c14ec2d20d89a9e816e4b6a73e1b99ae24981 +size 2676060 diff --git a/parse/train/S1c2cvqee/S1c2cvqee_origin.pdf b/parse/train/S1c2cvqee/S1c2cvqee_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..89fe4b6435157b33da95f863bc1d32fa6bd6155d --- /dev/null +++ b/parse/train/S1c2cvqee/S1c2cvqee_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6794ddda989a0c42430bd4f7732f1923fe32c1afae33245400ed72cd9d128a21 +size 2488804 diff --git a/parse/train/S1c2cvqee/S1c2cvqee_span.pdf b/parse/train/S1c2cvqee/S1c2cvqee_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..7473a312a1ff15e3780929af755ddcbe10fdf6e0 --- /dev/null +++ b/parse/train/S1c2cvqee/S1c2cvqee_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:31f5a36843d5ac23a998f4323d6110131e05eefc3ef389eef55acab2fce68b51 +size 2685541 diff --git a/parse/train/S1lvm305YQ/S1lvm305YQ_layout.pdf b/parse/train/S1lvm305YQ/S1lvm305YQ_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8d7abf3a599159f4c53a271bdb7073fc2c598bc3 --- /dev/null +++ b/parse/train/S1lvm305YQ/S1lvm305YQ_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c52b75a99ebf5cc123a22489adaa088f3979fc0916e8a56d9906bf1359ff8e71 +size 4926928 diff --git a/parse/train/S1lvm305YQ/S1lvm305YQ_origin.pdf b/parse/train/S1lvm305YQ/S1lvm305YQ_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3d31f4aa5175d7e2a976002efd28ab7821be649b --- /dev/null +++ b/parse/train/S1lvm305YQ/S1lvm305YQ_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd57a3d09f51d4b8960898eacbaf12ca94a41bcb0ce8c0d6faa3fe6881e9fcca +size 4767020 diff --git a/parse/train/S1lvm305YQ/S1lvm305YQ_span.pdf b/parse/train/S1lvm305YQ/S1lvm305YQ_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5f414eb38ec51f1627ad24ca574b87995bda67ac --- /dev/null +++ b/parse/train/S1lvm305YQ/S1lvm305YQ_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:915f9f90a6ea61dba5765ac1c6c543ac29ec1f88ce8b463624a63524bed5ed7f +size 4931663 diff --git a/parse/train/S1vuO-bCW/S1vuO-bCW_layout.pdf b/parse/train/S1vuO-bCW/S1vuO-bCW_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..159fd4cab2a8175d971decbb32c394887bddccf3 --- /dev/null +++ b/parse/train/S1vuO-bCW/S1vuO-bCW_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:71fd82e595b86f2488ff47d47213733356208fe7996c203cf08308b75847e841 +size 5560938 diff --git a/parse/train/S1vuO-bCW/S1vuO-bCW_origin.pdf b/parse/train/S1vuO-bCW/S1vuO-bCW_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..de5860e4a6c6e2623e44cab271ccb02fdcf3abbf --- /dev/null +++ b/parse/train/S1vuO-bCW/S1vuO-bCW_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6068f65b470a2424259b72231966e5e1b98c078ed90cc9e7c11aac55c08b011a +size 5385671 diff --git a/parse/train/S1vuO-bCW/S1vuO-bCW_span.pdf b/parse/train/S1vuO-bCW/S1vuO-bCW_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ba915f91c94dd765007af899ba5b3dbe7624671f --- /dev/null +++ b/parse/train/S1vuO-bCW/S1vuO-bCW_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2cab7c14c7b32813ea0ef7eddfeab16588eb27cff08994e969bc8a90e43ce7b2 +size 5561203 diff --git a/parse/train/Sk-oDY9ge/Sk-oDY9ge_layout.pdf b/parse/train/Sk-oDY9ge/Sk-oDY9ge_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..26f24748716de0ed97f15c996beae27d5ebcb3ec --- /dev/null +++ b/parse/train/Sk-oDY9ge/Sk-oDY9ge_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf421e7d1854ab2af3969949d3d1cb7c5120fb5b8c7fa9830f17f7721e45cf35 +size 579966 diff --git a/parse/train/Sk-oDY9ge/Sk-oDY9ge_origin.pdf b/parse/train/Sk-oDY9ge/Sk-oDY9ge_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..204d9c059e5e81552148d54f7fd0ff2dfe4118b0 --- /dev/null +++ b/parse/train/Sk-oDY9ge/Sk-oDY9ge_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f995c5ef4f44980ce1f6c05f648a14cf1d75202baff6ac8f27c20a2924e16adc +size 473408 diff --git a/parse/train/Sk-oDY9ge/Sk-oDY9ge_span.pdf b/parse/train/Sk-oDY9ge/Sk-oDY9ge_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e10d11554cb066eef5e5749ed29927cd9bb42907 --- /dev/null +++ b/parse/train/Sk-oDY9ge/Sk-oDY9ge_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28d36782803de22adfe6992cbadc739340e0f15c987a070d2966f6903497a519 +size 581799 diff --git a/parse/train/SkVhlh09tX/SkVhlh09tX_layout.pdf b/parse/train/SkVhlh09tX/SkVhlh09tX_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3d30535b6249fb0cd38b6732fcbb6797069e604f --- /dev/null +++ b/parse/train/SkVhlh09tX/SkVhlh09tX_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f56814cac314a0b821358ae1ebeb18b624fbb0507323c1f36a78e4da15a62a7 +size 679403 diff --git a/parse/train/SkVhlh09tX/SkVhlh09tX_origin.pdf b/parse/train/SkVhlh09tX/SkVhlh09tX_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..1ab778c66eb4172b3d78d8ea3628881689b7950b --- /dev/null +++ b/parse/train/SkVhlh09tX/SkVhlh09tX_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4431f4dad0508c3162325c0930f486c6a4892eeee9d748bf577f9f9d04b46c00 +size 562057 diff --git a/parse/train/SkVhlh09tX/SkVhlh09tX_span.pdf b/parse/train/SkVhlh09tX/SkVhlh09tX_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4616ad7e679bf631fe64c6dd3018eecbb8e78825 --- /dev/null +++ b/parse/train/SkVhlh09tX/SkVhlh09tX_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:157cc957871938e3ce2824931de17d4060bcd1371f28e438780b7ab410e42ccc +size 678897 diff --git a/parse/train/Skn9Shcxe/Skn9Shcxe_layout.pdf b/parse/train/Skn9Shcxe/Skn9Shcxe_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..bc9e98d5686d785ec75e207b7410226e56ac1c77 --- /dev/null +++ b/parse/train/Skn9Shcxe/Skn9Shcxe_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9a3fa13c29d7f7017c696c6e76e584554020c8d09677c59e5109efff445eccf +size 2069224 diff --git a/parse/train/Skn9Shcxe/Skn9Shcxe_origin.pdf b/parse/train/Skn9Shcxe/Skn9Shcxe_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..aed9be0b615a9fefba965be4c7783b714397d80b --- /dev/null +++ b/parse/train/Skn9Shcxe/Skn9Shcxe_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1f4e711cd25c9f2cd260c3ef8f4ac9757d245887a937c6368c1a8041f9fba73 +size 1923959 diff --git a/parse/train/Skn9Shcxe/Skn9Shcxe_span.pdf b/parse/train/Skn9Shcxe/Skn9Shcxe_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..eb1bd2f6f1e8dc4f36ae0d7911747d6866aee59d --- /dev/null +++ b/parse/train/Skn9Shcxe/Skn9Shcxe_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3358b397567d1507724a058f6aa2127955d32ffd687654bcee2deddfc5c44b94 +size 2071683 diff --git a/parse/train/TR-Nj6nFx42/TR-Nj6nFx42_layout.pdf b/parse/train/TR-Nj6nFx42/TR-Nj6nFx42_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..1498df8a194a08f09104535e342423b3e8403785 --- /dev/null +++ b/parse/train/TR-Nj6nFx42/TR-Nj6nFx42_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8faee94d11671d435105e9065b40bceb2371079a28628e987698714ecbd4300b +size 1120689 diff --git a/parse/train/TR-Nj6nFx42/TR-Nj6nFx42_origin.pdf b/parse/train/TR-Nj6nFx42/TR-Nj6nFx42_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ecf948de84972fb68e2321cb1d5bda8944ce85fc --- /dev/null +++ b/parse/train/TR-Nj6nFx42/TR-Nj6nFx42_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:851b268061021c661b4730f5b00b116c34744f9d9c17b3510d352bbf74715103 +size 797018 diff --git a/parse/train/TR-Nj6nFx42/TR-Nj6nFx42_span.pdf b/parse/train/TR-Nj6nFx42/TR-Nj6nFx42_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..bc6552c703267e0b96512caccbcbdf62cddad7a9 --- /dev/null +++ b/parse/train/TR-Nj6nFx42/TR-Nj6nFx42_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21e6dee2f19bc4798ccceb73a32bac3a3df43be488e1da75d3c3cfaef63f8a0d +size 1132679 diff --git a/parse/train/YUEFlzlG_0c/YUEFlzlG_0c_origin.pdf b/parse/train/YUEFlzlG_0c/YUEFlzlG_0c_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..c793d3e32e460c61fc17ad7ddd4e08aee3449367 --- /dev/null +++ b/parse/train/YUEFlzlG_0c/YUEFlzlG_0c_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18a1ba27994f286724357a68fe5855ebce91e2a66dc968a3d692a1ac2b053315 +size 1718408 diff --git a/parse/train/hm0i-cunzGW/hm0i-cunzGW_layout.pdf b/parse/train/hm0i-cunzGW/hm0i-cunzGW_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..2a353846ba0c48eee110904e7395a93130553298 --- /dev/null +++ b/parse/train/hm0i-cunzGW/hm0i-cunzGW_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e414343ff35014c3123a1e445c3756b871097940675800dca816edf4ef2035a9 +size 10563315 diff --git a/parse/train/hm0i-cunzGW/hm0i-cunzGW_origin.pdf b/parse/train/hm0i-cunzGW/hm0i-cunzGW_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..023a376c5b6c4dc34755ca51cbc2469c6f9f0c80 --- /dev/null +++ b/parse/train/hm0i-cunzGW/hm0i-cunzGW_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ffe8f7a1b683218f4db4c8e3e993cea1cf99092f9481bd86c8b1c61cb8078ea +size 10422235 diff --git a/parse/train/hm0i-cunzGW/hm0i-cunzGW_span.pdf b/parse/train/hm0i-cunzGW/hm0i-cunzGW_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..e7e3effca52026ebb8ac0d78ecde70612a498e69 --- /dev/null +++ b/parse/train/hm0i-cunzGW/hm0i-cunzGW_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:211d861df8039b10f08d88e673327ef1db20041828b74592f63d175268dc511f +size 10573289 diff --git a/parse/train/lM2971LAwV/lM2971LAwV_layout.pdf b/parse/train/lM2971LAwV/lM2971LAwV_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..405d80b7df272a2004ef834c4c90b1feafce5e49 --- /dev/null +++ b/parse/train/lM2971LAwV/lM2971LAwV_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:754aaa57b7b3531bab57117cc87d3893dddd8f55b2550e0703fa05a0f4317dc0 +size 1626420 diff --git a/parse/train/lM2971LAwV/lM2971LAwV_origin.pdf b/parse/train/lM2971LAwV/lM2971LAwV_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..a8c273b484dc47c11c2165cc198c941c322b8db4 --- /dev/null +++ b/parse/train/lM2971LAwV/lM2971LAwV_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f50db93d7e32b419e8c1ef94ebd834c533d0e98f65e6c82e6894f4df90aa51a9 +size 1522297 diff --git a/parse/train/lM2971LAwV/lM2971LAwV_span.pdf b/parse/train/lM2971LAwV/lM2971LAwV_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5c597690af2976b05f16d91c7e0bfd9952513410 --- /dev/null +++ b/parse/train/lM2971LAwV/lM2971LAwV_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67debf5616742529e6843236c56b0172db9a35c965f3695f25a636c0e466b8f1 +size 1630623 diff --git a/parse/train/q8qLAbQBupm/q8qLAbQBupm_layout.pdf b/parse/train/q8qLAbQBupm/q8qLAbQBupm_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..63907e04ba395316dce1356c098f8e6a4ff83351 --- /dev/null +++ b/parse/train/q8qLAbQBupm/q8qLAbQBupm_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c1db3389b029723aa44182d531fe754e3e7086288e046729a4683a09f84dd9e3 +size 13784718 diff --git a/parse/train/q8qLAbQBupm/q8qLAbQBupm_origin.pdf b/parse/train/q8qLAbQBupm/q8qLAbQBupm_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..ac70f229a6c3bb5d0642baad445d6f7c79f8e4f5 --- /dev/null +++ b/parse/train/q8qLAbQBupm/q8qLAbQBupm_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a8b6f00586138c3430793038307fe9c1042b67a95d7153558dc8f2468ffa949 +size 13284407 diff --git a/parse/train/q8qLAbQBupm/q8qLAbQBupm_span.pdf b/parse/train/q8qLAbQBupm/q8qLAbQBupm_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3ee3163ca5ac7d809c3b7096daa5d01b95c9a26a --- /dev/null +++ b/parse/train/q8qLAbQBupm/q8qLAbQBupm_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30e44be3bda9de2171d42a509bccd4ca1a09f4fd677aff61b0df6465d48ad41b +size 13796699 diff --git a/parse/train/r1lZgyBYwS/r1lZgyBYwS_layout.pdf b/parse/train/r1lZgyBYwS/r1lZgyBYwS_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..eec1d95bbacc339e5c2c8e30df143cf2761e8989 --- /dev/null +++ b/parse/train/r1lZgyBYwS/r1lZgyBYwS_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df6483616dafa36de50bb567a821f7ab4a6f16987af6bf00d4562ea203a1e636 +size 1487430 diff --git a/parse/train/r1lZgyBYwS/r1lZgyBYwS_origin.pdf b/parse/train/r1lZgyBYwS/r1lZgyBYwS_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..1d5af3e418bda7a1708f90fce16a9728dde0c4a3 --- /dev/null +++ b/parse/train/r1lZgyBYwS/r1lZgyBYwS_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:548ead34756971a190cb35dfb0c2add8455586e8514f97b6f79f92f5ff257636 +size 1338873 diff --git a/parse/train/r1lZgyBYwS/r1lZgyBYwS_span.pdf b/parse/train/r1lZgyBYwS/r1lZgyBYwS_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..1a125fb3f18572c62dbae8c67eef776354d8a2fa --- /dev/null +++ b/parse/train/r1lZgyBYwS/r1lZgyBYwS_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf1dabb00df0c710458fb8364b243a5db66c0bccca53c495f067d8c0fbf4610e +size 1491374 diff --git a/parse/train/rJe4_xSFDB/rJe4_xSFDB_layout.pdf b/parse/train/rJe4_xSFDB/rJe4_xSFDB_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..bdd8db0608585791761ecada794e4fe3fa5e3744 --- /dev/null +++ b/parse/train/rJe4_xSFDB/rJe4_xSFDB_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afe9fdc5056d37d7465aa2a6c7118442ec67531425c8e6e7141c6bb120fbe603 +size 952371 diff --git a/parse/train/rJe4_xSFDB/rJe4_xSFDB_origin.pdf b/parse/train/rJe4_xSFDB/rJe4_xSFDB_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..db6a3caa3e6d7c1cea4b8b426f1c77c2b742d61f --- /dev/null +++ b/parse/train/rJe4_xSFDB/rJe4_xSFDB_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:796b1756ca583d7b738f04aa14f84a1f32b68b1056dffa8e15994bdc328634ed +size 757985 diff --git a/parse/train/rJe4_xSFDB/rJe4_xSFDB_span.pdf b/parse/train/rJe4_xSFDB/rJe4_xSFDB_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..3fb32075db03a29f178a2d862e5f074bb96ca365 --- /dev/null +++ b/parse/train/rJe4_xSFDB/rJe4_xSFDB_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:38145a9da04e26b550a9463910b1db2e1b1a78bb9d84e76f94bdc3dab2a030b6 +size 961743 diff --git a/parse/train/rJqFGTslg/rJqFGTslg_layout.pdf b/parse/train/rJqFGTslg/rJqFGTslg_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..4e0337abf510138aaeefe8be1ce27ab0e02d262e --- /dev/null +++ b/parse/train/rJqFGTslg/rJqFGTslg_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4452c0bbc8f39264a7080edb70e438cf7e1238aed54de4af3094e2dcd1a27875 +size 4383432 diff --git a/parse/train/rJqFGTslg/rJqFGTslg_origin.pdf b/parse/train/rJqFGTslg/rJqFGTslg_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..5a9eb28711aaaaed2596847863dbe7597c0e2459 --- /dev/null +++ b/parse/train/rJqFGTslg/rJqFGTslg_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7173fdb990e3e4d5a3afc37e61aae8c548c832b34b8b15359e823a6a2098ad18 +size 4257138 diff --git a/parse/train/rJqFGTslg/rJqFGTslg_span.pdf b/parse/train/rJqFGTslg/rJqFGTslg_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..2b6f36afa6ddd93584a223e0ba5de33ee2a6d62e --- /dev/null +++ b/parse/train/rJqFGTslg/rJqFGTslg_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab5ed47a7a090d392e30e1f7824f09486457d21f0aece020185a2c9b9bdbca2b +size 4389518 diff --git a/parse/train/rJxt0JHKvS/rJxt0JHKvS_layout.pdf b/parse/train/rJxt0JHKvS/rJxt0JHKvS_layout.pdf new file mode 100644 index 0000000000000000000000000000000000000000..495054f8e13869aa2707fd95995cc7f7d4ff2f63 --- /dev/null +++ b/parse/train/rJxt0JHKvS/rJxt0JHKvS_layout.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb58a5544dea8f1049414492416f9823739bb4c0b13e28b7f2c3f58c794ea51c +size 692480 diff --git a/parse/train/rJxt0JHKvS/rJxt0JHKvS_origin.pdf b/parse/train/rJxt0JHKvS/rJxt0JHKvS_origin.pdf new file mode 100644 index 0000000000000000000000000000000000000000..8eef2b8a6e287d5f60cc5babbca656b89885c1b5 --- /dev/null +++ b/parse/train/rJxt0JHKvS/rJxt0JHKvS_origin.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d8afa70b776ceac697782432a6f91c660986d88fc697a95979b53d17101d077 +size 469033 diff --git a/parse/train/rJxt0JHKvS/rJxt0JHKvS_span.pdf b/parse/train/rJxt0JHKvS/rJxt0JHKvS_span.pdf new file mode 100644 index 0000000000000000000000000000000000000000..346123a0f5596bd888abb0bc6caceda2d295c755 --- /dev/null +++ b/parse/train/rJxt0JHKvS/rJxt0JHKvS_span.pdf @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65f480d939db4c482d7206641b4cd735587a586cb20138b6f8eea538e062405c +size 705725